MOLECULAR MARKERS OF LYMPH NODE METASTASES IN ORAL CANCER

Transcription

1 Per juni 2008 ving zijn promotieonderzoek aan onder begeleiding van prof. dr. R. Koole en prof. dr. P.J. van Diest, waarvan de resultaten in dit proefschrift beschreven staan. Hij was ruim 1,5 jaar werkzaam als arts-onderzoeker bij de afdeling Pathologie en Mondziekten, Kaak- en Aangezichtschirurgie in het Universitair Medisch Centrum Utrecht. Met de presentatie van de studie beschreven in Hoofdstuk 3 getiteld Validation of a gene expression signature for assessment of lymph node metastasis in oral squamous-cell carcinoma won hij de Marten Hut prijs tijdens de 54ste Najaarsvergadering van de Nederlandse Vereniging voor Mondziekten, Kaak- en Aangezichtschirurgie (2010), de presentatieprijs tijdens de 1e jonge onderzoekers dag van de Nederlandse Werkgroep Hoofd Hals Tumoren (2010), en de Best scientific research paper tijdens de 3rd World Congres of the International Academy of Oral Oncology in Singapore (2011). Laatste prijs resulteerde in een uitnodiging tot het schrijven van een invited review voor The Lancet Oncology, zie Hoofdstuk 4. Van januari 2010 tot december 2013 volgde hij de opleiding Mondziekten, Kaak- en Aangezichtschirurgie in het Universitair Medisch Centrum Utrecht onder leiding van prof. dr. R. Koole. Na een jaar als freelance MKA-chirurg gewerkt te hebben in diverse ziekenhuizen startte hij in januari 2015 met het fellowship Hoofd-Hals Chirurgie in het Nederlands Kanker Instituut Antoni van Leeuwenhoek met als opleider prof. dr. L. Smeele. Simultaan aan dit fellowship startte hij met de Masterstudie Esthetische Aangezichtschirurgie aan de universiteit van Witten/Herdecke, Duitsland en neemt hij deel aan het Global Fellowship Head and Neck Oncology, Memorial Sloan Kettering Cancer Center, New York, VS. MOLECULAR MARKERS OF LYMPH NODE METASTASES IN ORAL CANCER Frank Leusink werd op 7 april 1979 geboren in Hengelo (O). In 1997 behaalde hij zijn Gymnasium diploma aan de O.S.G. Bataafse Kamp. Van 1997 tot 2004 studeerde hij tandheelkunde en van 2004 tot 2008 studeerde hij geneeskunde aan de medische faculteit van de Rijksuniversiteit Groningen. MOLECULAR MARKERS OF LYMPH NODE METASTASES IN ORAL CANCER Frank K.J. Leusink UITNODIGING Voor het bijwonen van de openbare verdediging van het proefschrift MOLECULAR MARKER OF LYMPH NODE METASTASES IN ORAL CANCER Door Frank Leusink Op dinsdag 31 januari 2017 Om uur In het Academiegebouw Domplein 29, te Utrecht Receptie na afloop van de promotie in het St. Eloyen Gasthuis Boterstraat 22, te Utrecht Frank Leusink Traaij GL Driebergen - Rijsenburg frankleusink@hotmail.com In 2017 zal hij zijn loopbaan een vervolg geven als staflid op de afdeling Mondziekten, Kaak- en Aangezichtschirurgie in het VU medisch centrum. Frank Leusink is naast fervent sportliefhebber vooral een trotse vader van Tim en Emma. Frank K.J. Leusink Paranimfen Marijn Rutgers marijnrutgers@gmail.com Michaël Frank m.frank@haaglandenmc.nl

2

3 MOLECULAR MARKERS OF LYMPH NODE METASTASES IN ORAL CANCER Frank Karel Johan Leusink

4 The research in this thesis was performed at the departments of Oral & Maxillofacial Surgery and Pathology, University Medical Center Utrecht, the Netherlands, and at the departments of Pathology and Otolaryngology-Head and Neck Surgery, Section Tumor Biology, VU University Medical Center, Amsterdam, the Netherlands. The research was financially supported by the Dutch Cancer Society (KWF Kankerbestrijding) Grant No. KUN and by the Dutch Society of Oral & Maxillofacial Surgery (NVMKA) BOOA Research Grant No commercial funding was received to perform this research. Printed by Layout by Gildeprint Nicole Nijhuis, Gildeprint Cover design Karel J. Leusink kareljleusink.blogspot.nl ISBN: Copyright 2016, F.K.J. Leusink All rights reserved. No part of this book may be reproduced in any form, by print, photocopy, electronic data transfer or any other means, without prior permission of the author.

5 MOLECULAR MARKERS OF LYMPH NODE METASTASES IN ORAL CANCER Moleculaire markers voor lymfklier metastasen van mondkanker [met een samenvatting in het Nederlands] Proefschrift ter verkrijging van de graad van doctor aan de Universiteit Utrecht op gezag van de rector magnificus, prof. dr. G.J. van der Zwaan, ingevolge het besluit van het college voor promoties in het openbaar te verdedigen op dinsdag 31 januari 2017 des middags 2.30 uur door Frank Karel Johan Leusink geboren 7 april 1979 te Hengelo (O.)

6 Promotoren: Prof. dr. P.J. van Diest Prof. dr. R. Koole Copromotoren: Dr. S.M. Willems Dr. R.J.J. van Es

7 CONTENTS Chapter 1. General introduction and outline of the thesis 7 Chapter 2. Lymphatic vessel density and lymph node metastasis in head and neck 21 squamous-cell carcinoma: a systematic review Chapter 3. Validation of a gene expression signature for assessment of lymph node 43 metastasis in oral squamous-cell carcinoma Chapter 4. Novel diagnostic modalities for the clinically node-negative neck in oral 63 squamous-cell carcinoma Chapter 5. Tumor biological determinants of locoregional recurrence of non-hpv 83 head and neck squamous-cell carcinoma Chapter 6. Nodal metastasis and survival in oral cancer: Association with protein 101 expression of SLPI, not with LCN2, TACSTD2, or THBS2 Chapter 7. The co-expression of kallikrein 5 and kallikrein 7 associates with poor 121 survival in non- HPV oral squamous-cell carcinoma Chapter 8. Cathepsin K associates with lymph node metastasis and poor prognosis 145 in oral squamous-cell carcinoma Chapter 9. General discussion 165 Chapter 10. Future perspectives 177 Chapter 11. Summary 185 Chapter 12. Nederlandse samenvatting 195 Chapter 13. List of publications 205 Chapter 14. Dankwoord 209

8

9 CHAPTER 1 General introduction and outline of the thesis

10 Chapter 1 8

11 General introduction and outline of the thesis GENERAL INTRODUCTION Head and neck squamous-cell carcinoma Head and neck squamous-cell carcinoma (HNSCC) is the seventh most common malignancy globally and accounts for approximately 4% of all malignant tumors. 1,2 HNSCC arises in mucosal linings of the oral cavity, oropharynx, nasopharynx, hypopharynx and larynx (figure 1). 3 1 Figure 1 Head and neck cancer regions. This figure illustrates location of oral cavity, pharynx (including the nasopharynx, oropharynx and hypopharynx) and larynx. (source: 2012, Terese Winslow LLC) The research in this thesis focuses on two subgroups of HNSCC: first, tumors originating in the oral cavity (oral squamous-cell carcinomas, OSCC) and second, tumors originating in the oropharynx (oropharyngeal squamous-cell carcinomas, OPSCC). Risk factors include tobacco smoking, betel nut chewing, excessive alcohol consumption, and human papillomavirus (HPV) infection, 4-6 although the etiologic role of HPV in OSCC is unclear. 7 Together, OSCC and OPSCC account for two thirds of the worldwide incidence of HNSCC with an estimation of cases and deaths in

12 Chapter 1 Oral squamous-cell carcinoma With approximately new cases annually, OSCC has the highest incidence of all HNSCCs. In The Nether1lands, the incidence is around cases and this has slowly been rising over the last two decades. 9,10 For OSCC the male to female ratio is 1.6 and reflects the distribution of smoking between both sexes. Subsites of the oral cavity include the lips, the anterior two-thirds of the tongue, the floor of the mouth, the gums, the lining inside the cheeks and lips, the hard palate and the retromolar trigone. Symptoms can consist of a white or red patch or an ulcer on the mucosal lining of the mouth, a mucosal swelling of the jaw that causes dentures to fit poorly or become uncomfortable and unusual bleeding or pain in the mouth. 3 Staging of the tumor and the neck is performed by clinical examination (i.e. palpation) and imaging, but accuracy of these techniques to detect small lymph node metastases (LNM) is limited. In general, 30 40% of patients will have occult nodal disease and will develop clinically detectable LNM when the neck is left untreated. 11 Surgery is the preferred modality for local treatment of the primary OSCC. Adjuvant treatment (re-resection, radiotherapy, or chemotherapy) may be indicated based on adverse tumor features diagnosed at histopathological examination. For regional treatment of the node-positive neck (cn+) a therapeutic neck dissection (TND) is indicated, but for the clinically node-negative neck (cn0) the choice is either elective neck dissection (END) or watchful waiting (WW) followed by TND in patients who develop manifest metastases. Oropharyngeal squamous-cell carcinoma OPSCC occurs with an annual estimated worldwide incidence of cases. 8,9 Subsites of the oropharynx include the base of tongue, the tonsils, the soft palate and the posterior and lateral walls of the oropharynx. HPV is reported as a causative factor with prevalence rates varying between 15-90%. 12 In The Netherlands the prevalence rate of HPV in OPSCC is around 25% and has increased in the last two decades, 13 in contrast to the HPV prevalence rate in OSCC which is reported to be around 3%. 14 Treatment of early OPSCC (T1-T2) can be curatively performed effectively with primary radiotherapy or surgery. 15 More advanced OPSCC (T3-4 or node positive) calls for multimodal therapy consisting of chemoradiation or primary radiotherapy. 16 Although patients with a HPV-positive tumor usually present with a smaller primary tumor, and show more often nodal disease at time of diagnosis they respond better to therapy and therefore have a better outcome compared to patients with HPV-negative OPSCC. 17 However, clinical management for the HPV-positive OPSCC has not changed. Therefore, as in OSCC, the most important factor for management and prognosis in HPV-positive as well as HPV-negative OPSCC is regional LNM. 10

13 General introduction and outline of the thesis To treat or not to treat? - The dilemma of the clinically node-negative neck The presence of regional neck metastases is a major determinant of both prognosis and treatment decisions in patients with OSCC. 20 However, the low sensitivity of currently available diagnostic modalities is a problem, because a fairly high proportion (30 40%) of LNM are left undetected in this population. These metastases will develop into overt neck disease during follow-up. 1 A policy has been in place during the last decades to treat the neck even when the tumor has been classified as cn0, indicating no disease is detectable in the neck 21 on physical examination and imaging. This strategy prevents disease in the neck becoming more advanced once previously occult metastases become clinically apparent or are detected late during follow-up. Thus, 60 70% of patients receive unnecessary treatment, which in the case of elective neck dissection (END) encompasses a surgical procedure potentially causing disfigurement and associated morbidity The alternative approach of watchful waiting (WW) entails careful monitoring of the neck (eg, by ultrasound-guided fine-needle aspiration cytology during follow-up). 26,27 Treatment (therapeutic neck dissection) will be given only to patients who develop manifest metastasis, which usually arises within 1 2 years. Although some authors report evidence showing convincingly that END should be preferred over WW, 28 others doubt the generalizability of these results and question whether they can be implemented worldwide. 29 Consensus partly based on an often cited but outdated decision analysis model suggests that elective neck treatment is indicated when the chance of occult nodal disease exceeds 20%. 30 Hence, increased accuracy to ascertain the metastatic status of the neck will result in fewer unnecessary elective treatment of the neck. Currently, the neck is staged by palpation and different imaging techniques, including CT, MRI, PET/CT, and ultrasound, which are more accurate than palpation alone Morphological and size criteria are the main determinants of specificity of imaging techniques, whereas sensitivity is limited by the detection threshold. Up to a third of nodal metastases in patients with oral SCC are smaller than the 3 mm detection threshold that limits sensitivity of available imaging techniques, 34 and occult neck disease therefore remains a relevant issue. In a meta-analysis of assessment of cervical LNM, 32 sensitivity of PET/CT was only 50% for patients who were nodenegative on palpation, whereas specificity was 87%. The performance of ultrasound, MRI, and CT was equally disappointing. Later research has shown similar results. 35 The limitation of imaging techniques to detect small metastatic deposits has led to a search for additional characteristics or biomarkers assessable on the primary tumor to predict nodal disease. Histopathological or molecular features of the primary tumor can predict the presence of nodal metastases in the individual patient, irrespective of the actual size of the tumor. 36 Measures such as perineural invasion, vascular invasion, tumor border (infiltrative or pushing), tumor thickness, 11

14 Chapter 1 and depth of invasion have been studied extensively for such correlations. Only depth of invasion is consistently associated with the presence of nodal metastases. 37 However, definitions and methods to measure vary (suggested cutoffs range from 1,5-10 mm) and consensus when to recommend elective neck dissection is lacking Thus, the dilemma in current clinical management of the neck is the choice between possible undertreatment of 30 40% of patients with occult metastases and overtreatment of the remaining 60 70%. Personalized management of the cn0 neck, especially in patients with oral SCCs, would benefit greatly from staging techniques that add accuracy to the assessment of nodal disease, particularly when these methods are non-invasive and are not or only minimally dependent on size of metastases. New diagnostic modalities for the clinically node-negative neck Sentinel lymph-node biopsy In an attempt to more reliably select lymph nodes that potentially contain metastases, sentinel lymph-node (SLN) assessment, which is used extensively in melanoma and breast cancer, has been introduced. The SLN is likely to be the first lymph node to harbor metastases and can be used to provide information on the remainder of the nodal basin. It is usually identified by peritumoral injection of radioactive colloid and blue dye. Preoperative lymphoscintigraphy (figure 1A), intra-operative visualization of blue coloration, and intra-operative radionuclide detection with a gamma probe (figure 1B) allow identification of the SLN(s). After surgical removal, the node or nodes are studied meticulously by histopathological examination, using step sectioning and immunohistochemistry (figure 1C). If the SLN contains metastatic tumor cells, treatment of the neck is recommended, usually in a second procedure. 40 The SLN procedure is deemed more precise than imaging procedures and less invasive than END. Moreover, it is associated with significantly less postoperative morbidity and better shoulder function compared with END, however paralysis of the marginal mandibular branch of the facial nerve has also been observed in the SLN biopsy procedure. Overall, negative predictive values of SLN biopsy procedures are generally above 88% In the American NCCN guidelines, the UK NICE guidelines as well as the guidelines of the Dutch Head and Neck Society, SLN biopsy is already mentioned as an alternative for END However, this technique does require experience and is currently recommended only for centers with the necessary facilities and expertise

15 General introduction and outline of the thesis 1 Figure 2 Sentinel lymph-node biopsy procedure. (A) Lymphoscintigram showing injection site (large white area) and SLN (small white area). (B) Gamma probeguided SLN biopsy. (C) A micrometastasis (red color) is depicted by immunohistochemistry using antibodies against cytokeratin. (source: 2012, Derrek Heuveling, Remco de Bree and Elisabeth Bloemena) Molecular diagnosis and tumor profiling Besides histological variables, many single molecular markers have been studied to predict the presence of nodal metastasis. In view of the complexity of the metastatic process, no consistent and clinically useful correlations have been noted for any markers 36. Gene-expression profiling with DNA microarrays and next generation sequencing approaches (RNAseq) signal a new era of tumor classification and prognostication. RNA isolated from the tumor specimen can be used to ascertain expression levels of all genes simultaneously in one experiment (fig 3). This process has led to novel classifications of lymphomas, breast cancer, and HNSCC Furthermore, prognostic profiles have been identified in many tumor types, including breast cancer and HNSCC. 52,53 With regard to HNSCC, previous studies identified a gene expression signature for distinguishing metastasizing (N+) from N0 OSCC and N0 oropharynx SCC (OPSCC). 54,55 Although promising (negative predictive value of 100%), the independent validation cohort in that study was small (n = 22), with all samples (N = 104) derived from a single clinical center. 54 Reanalysis of the entire data set identified more genes with predictive power. 55 Before applying such a LNM predicting profile in clinical practice, a study is needed to validate this signature in a large multicenter patient cohort. 13

16 Chapter 1 Figure 3 Microarray procedure and identification of a gene expression signature. DNA microarrays are generated by grid-wise fixation of a collection of DNA probes on a glass surface. Each probe represents one human gene. From the analyzed samples, i.e. analyzed tumors, RNA is isolated and subsequently labeled with a fluorescent group, such as cy5 (red) or cy3 (green). The sample target is combined with a reverse labeled reference target and hybridized on a DNA microarray. After quantification of the sample and reference signal channels on the array, a sample reference ratio is calculated by which genes can be identified that are higher or lower expressed in the analyzed samples compared to the reference. By combining the results of multiple samples, a specific set of genes can be identified that is able to discriminate the analyzed samples in specific groups (i.e. N0 vs N+). (source: 2005, Paul Roepman) OUTLINE OF THIS THESIS The aim of the work in this thesis was to investigate additional modalities to improve assessment of the cn0 neck in patients with early OSCC. This has been achieved by applying different approaches with regard to detecting occult metastases. The aim of each individual study was to discover one or more molecular biomarkers associated with LNM or prognosis to provide new insight into and the ability to distinguish between different underlying disease phenotypes, with the goal of providing options to change and improve current management in each patient. This thesis started with general aspects of diagnosis and prognosis of OSCC and OPSCC and the dilemma about the cn0 neck in patients with early OSCC. 14

17 General introduction and outline of the thesis The density of the lymphatic vessels (LVD), both intratumoral and peritumoral, as a variable used for the appraisal of the tumor related lymphatic system is described; LVD has been associated to the presence of nodal metastasis in various malignancies. In chapter 2 a systematic review of the literature is presented about the association of both intratumoral and peritumoral LVD to nodal spread in HNSCC, although OSCC was the main primary malignancy studied. 1 The second study described in chapter 3 is the multicenter validation of a previously published diagnostic gene signature that predicts LNM in OSCC. In light of the imperfect clinical examination, which is the current standard and is especially suboptimal for determining the absence of LNM, a concurrent and independent diagnosis based on expression profiling of the primary tumor can greatly improve the matching of patients with the appropriate treatment. Chapter 4 consists of a review of recent developments in staging of the neck, including expression profiling and sentinel node biopsy, and suggests a new staging algorithm to incorporate both methods, to optimize management of the cn0 neck in patients with early-stage OSCC. Chapter 5 aims to identify tumor characteristics that are associated with the development of locoregional recurrences and a poor prognosis and to reveal their biological basis. With gene expression microarray analysis the biological basis of the determinants of locoregional recurrence was established. Four of the most predictive genes of the LNM predicting signature described in chapter 2 are secretory leukocyte protease inhibitor (SLPI), lipocalin-2 (LCN2), thrombospondin-2 (THBS2), and tumor-associated calcium signal transducer 2 (TACSTD2). In chapter 6 their protein expression is correlated with LNM and survival. Chapter 7 tries to reveal the importance of the balance between the kallikrein 5 and 7 proteins and their inhibitor serine protease inhibitor kazal type 5 with regard to LNM. These three genes are included in the LNM predicting signature described in chapter 2. In chapter 8 the role of the protease cathepsin K is described and its value as a single marker predicting LNM and survival. Finally, the results obtained in the current thesis are discussed in chapter 9, followed by some future perspectives in chapter 10, and a summary in chapter

18 Chapter 1 REFERENCES 1. Ferlay J, Shin HR, Bray F, et al. Estimates of worldwide burden of cancer in 2008: GLOBOCAN Int J Cancer 2010; 127: Jemal A, Bray F, Center MM, et al. Global cancer statistics. CA Cancer J Clin 2011; 61: Blot WJ, McLaughlin JK, Winn DM, et al. Smoking and drinking in relation to oral and pharyngeal cancer. Cancer Res 1988; 48: Argiris A, Karamouzis MV, Raben D, et al. Head and neck cancer. Lancet 2008; 371: Ang KK, Harris J, Wheeler R, et al. Human papillomavirus and survival of patients with oropharyngeal cancer. N Engl J Med 2010; 363: Lingen MW, Xiao W, Schmitt A, et al. Low etiologic fraction for high-risk human papillomavirus in oral cavity squamous cell carcinomas. Oral Oncol 2013; 49: Chaturvedi AK, Anderson WF, Lortet-Tieulent J, et al. Worldwide trends in incidence rates for oral cavity and oropharyngeal cancers. J Clin Oncol 2013; 31: Braakhuis BJ, Visser O, Leemans CR. Oral and oropharyngeal cancer in the Netherlands between 1989 and 2006: Increasing incidence, but not in young adults. Oral Oncol 2009; 45:e85-e Dutch cancer figures [Homepage on the Internet] 2016; cited August 28, Available from: Flach GB, Tenhagen M, de Bree R, et al. Outcome of patients with early stage oral cancer managed by an observation strategy towards the N0 neck using ultrasound guided fine needle aspiration cytology: No survival difference as compared to elective neck dissection. Oral Oncol 2013; 49: Marur S, D Souza G, Westra WH, et al. HPV-associated head and neck cancer: a virus-related cancer epidemic. Lancet Oncol 2010; 11: Rietbergen MM, Leemans CR, Bloemena E et al. Increasing prevalence rates of HPV attributable oropharyngeal squamous cell carcinomas in the Netherlands as assessed by a validated test algorithm. Int J Cancer 2013; 132: Combes JD, Franceschi S. Role of human papillomavirus in non-oropharyngeal head and neck cancers. Oral Oncol 2014; 50: Cohan DM, Popat S, Kaplan SE, et al. Oropharyngeal cancer: current understanding and management. Curr Opin Otolaryngol Head Neck Surg 2009; 17: Lanza L, Rizzi L, Durso D, et al. Integrated treatment in locally advanced carcinoma of the oropharynx. J Surg Oncol 2000; 74: Gillison ML, D Souza G, Westra W, et al. Distinct risk factor profiles for human papillomavirus type 16-positive and human papillomavirus type 16-negative head and neck cancers. J Natl Cancer Inst ; 100: Schwartz SR, Yueh B, McDougall JK, et al. Human papillomavirus infection and survival in oral squamous cell cancer: a population-based study. Otolaryngol Head Neck Surg 2001; 125: Ang KK, Harris J, Wheeler R, et al. Human papillomavirus and survival of patients with oropharyngeal cancer. N Engl J Med 2010; 363: Argiris A, Karamouzis MV, Raben D, et al. Head and neck cancer. Lancet 2008; 371: Ferlito A, Silver CE, Rinaldo A. Elective management of the neck in oral cavity squamous carcinoma: current concepts supported by prospective studies. Br J Oral Maxillofac Surg 2009; 47: van Wilgen CP, Dijkstra PU, van der Laan BF, et al. Morbidity of the neck after head and neck cancer therapy. Head Neck 2004; 26: van Wouwe M, de Bree R, Kuik DJ, et al. Shoulder morbidity after non-surgical treatment of the neck. Radiother Oncol 2009; 90: Bradley PJ, Ferlito A, Silver CE, et al. Neck treatment and shoulder morbidity: still a challenge. Head Neck 2011; 33:

19 General introduction and outline of the thesis 25. Speksnijder CM, van der Bilt A, Slappendel M, et al. Neck and shoulder function in patients treated for oral malignancies: a 1-year prospective cohort study. Head Neck 2013; 35: Nieuwenhuis EJ, Castelijns JA, Pijpers R, et al. Wait-and-see policy for the N0 neck in early-stage oral and oropharyngeal squamous cell carcinoma using ultrasonography-guided cytology: is there a role for identification of the sentinel node? Head Neck 2002; 24: Rodrigo JP, Shah JP, Silver CE, et al. Management of the clinically negative neck in early-stage head and neck cancers after transoral resection. Head Neck 2011; 33: D Cruz AK, Vaish R, Kapre N, et al. Elective versus Therapeutic Neck Dissection in Node-Negative Oral Cancer. N Engl J Med 2015; 373: de Bree R, van den Brekel MWM. Elective neck dissection versus observation in the clinically node negative neck in early oral cancer: Do we have the answer yet? Oral Oncol 2015; 51: Weiss MH, Harrison LB, Isaacs RS. Use of decision analysis in planning a management strategy for the stage N0 neck. Arch Otolaryngol Head Neck Surg 1994; 120: de Bondt RB, Nelemans PJ, Hofman PA, et al. Detection of lymph node metastases in head and neck cancer: a meta-analysis comparing US, USgFNAC, CT and MR imaging. Eur J Radiol 2007; 64: Kyzas PA, Evangelou E, Denaxa-Kyza D, et al. 18F-fluorodeoxyglucose positron emission tomography to evaluate cervical node metastases in patients with head and neck squamous cell carcinoma: a metaanalysis. J Natl Cancer Inst 2008; 100: Ng SH, Yen TC, Chang JT, et al. Prospective study of [18F] fluorodeoxyglucose positron emission tomography and computed tomography and magnetic resonance imaging in oral cavity squamous cell carcinoma with palpably negative neck. J Clin Oncol 2006; 24: Buckley JG, MacLennan K. Cervical node metastases in laryngeal and hypopharyngeal cancer: a prospective analysis of prevalence and distribution. Head Neck 2000; 22: Liao LJ, Lo WC, Hsu WL, et al. Detection of cervical lymph node metastasis in head and neck cancer patients with clinically N0 neck-a meta-analysis comparing different imaging modalities. BMC Cancer 2012; 12: Takes RP, Rinaldo A, Rodrigo JP, et al. Can biomarkers play a role in the decision about treatment of the clinically negative neck in patients with head and neck cancer? Head Neck 2008; 30: Melchers LJ, Schuuring E, van Dijk BA, et al. Tumour infiltration depth >/=4 mm is an indication for an elective neck dissection in pt1cn0 oral squamous cell carcinoma. Oral Oncol 2012; 48: Pentenero M, Gandolfo S, Carrozzo M. Importance of tumor thickness and depth of invasion in nodal involvement and prognosis of oral squamous cell carcinoma: a review of the literature. Head Neck 2005; 27: Huang SH, Hwang D, Lockwood G, et al. Predictive value of tumor thickness for cervical lymph-node involvement in squamous cell carcinoma of the oral cavity: a meta-analysis of reported studies. Cancer 2009; 115: Alkureishi LW, Burak Z, Alvarez JA, et al. Joint practice guidelines for radionuclide lymphoscintigraphy for sentinel node localization in oral/oropharyngeal squamous cell carcinoma. Ann Surg Oncol 2009; 16: Murer K, Huber GF, Haile SR, et al. Comparison of morbidity between sentinel node biopsy and elective neck dissection for treatment of the n0 neck in patients with oral squamous cell carcinoma. Head Neck 2011; 33: Schiefke F, Akdemir M, Weber A, et al. Function, postoperative morbidity, and quality of life after cervical sentinel node biopsy and after selective neck dissection. Head Neck 2009; 31: Civantos FJ, Zitsch RP, Schuller DE, et al. Sentinel lymph node biopsy accurately stages the regional lymph nodes for T1 T2 oral squamous cell carcinomas: results of a prospective multi-institutional trial. J Clin Oncol 2010; 28: Govers TM, Hannink G, Merkx MA, et al. Sentinel node biopsy for squamous cell carcinoma of the oral cavity and oropharynx: a diagnostic meta-analysis. Oral Oncol, 2013; 49:

20 Chapter Flach GB, Bloemena E, Klop WMC, et al. Sentinel lymph node biopsy in clinically N0 T1 T2 staged oral cancer: The Dutch multicenter trial. Oral Oncol 2014; 50: NCCN Guidelines Version head-and-neck.pdf Accesses on November 9, NICE Guidelines. [NG36]1.3.5 Published date February chapter/recommendations. Accessed on November 9, Dutch Guideline Head and Neck Tumors. Hoofd-halstumoren% pdf Accesses on November 9, Alizadeh AA, Eisen MB, Davis RE, et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 2000; 403: Perou CM, Sorlie T, Eisen MB, et al. Molecular portraits of human breast tumors. Nature 2000; 406: Chung CH, Parker JS, Karaca G, et al. Molecular classification of head and neck squamous cell carcinomas using patterns of gene expression. Cancer Cell 2004; 5: van t Veer LJ, Dai H, van de Vijver MJ, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002; 415: Chung CH, Parker JS, Ely K, et al. Gene expression profiles identify epithelial-to-mesenchymal transition and activation of nuclear factor-κb signaling as characteristics of a high-risk head and neck squamous cell carcinoma. Cancer Res 2006; 66: Roepman P, Wessels LF, Kettelarij N, et al. An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas. Nat Genet 2005; 37: Roepman P, Kemmeren P, Wessels LF, et al. Multiple robust signatures for detecting lymph node metastasis in head and neck cancer. Cancer Res 2006; 66:

21 General introduction and outline of the thesis 1 19

22

23 CHAPTER 2 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma: a systematic review E Koudounarakis F.K.J. Leusink S.M. Willems P.A. Bhairosing M.W.M. van den Brekel C.L. Zuur Accepted in European Journal of Surgical Oncology

24 Chapter 2 ABSTRACT INTRODUCTION The density of the lymphatic vessels (LVD), both intratumoral and peritumoral, is a parameter used for the appraisal of the tumor related lymphatic system and has been associated to the presence of nodal metastasis in various malignancies. This study constitutes the first systematic review of the literature on the association of both intratumoral and peritumoral LVD to nodal spread in head and neck squamous cell carcinoma (HNSCC). METHODS A systematic database search of Pubmed and Embase was conducted to identify original articles addressing the association of LVD to lymphatic metastasis in patients with HNSCC. RESULTS The database search yielded 151 articles. Among these, 28 original studies were eligible for inclusion. Oral cancer was the main primary malignancy studied, but also patients with laryngeal, hypopharyngeal and oropharyngeal carcinoma were included. CONCLUSION The majority of the articles reported significantly higher values of LVD in the group of patients with nodal metastasis compared to the non-metastatic group, suggesting that LVD can be potentially used as an additional tool for the prediction of occult metastasis in cases of HNSCC. KEYWORDS Lymphatic vessel density; lymphangiogenesis; head and neck squamous cell carcinoma; lymph node metastasis; biomarkers 22

25 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma INTRODUCTION Squamous-cell carcinoma of the head and neck accounts for less than 5% of the human malignant tumours. 1 Lymph node metastasis can be an early event in disease progression and is associated with a 50% decrease in survival, being one of the most important independent prognostic factors. 2 Prompt diagnosis of occult nodal metastasis holds a crucial role in tumor staging and treatment planning. The area of the head and neck features a rich lymphatic network, including approximately three hundred out of the total eight hundred lymph nodes of the human body. 3 Physical examination and various imaging techniques, including ultrasound along with guided fine needle aspiration cytology (US-FNAC), computed tomography (CT) and magnetic resonance imaging (MRI), are commonly used for the detection of metastasis in regional lymph nodes. The sensitivity of US-FNAC in the N0 neck has been reported to range between 48-73%, whereas the corresponding values for CT and MRI are 40-83% and 55-80%, respectively. 4,5,6 Additionally, US is highly performer dependent, while CT and MRI rely on morphological criteria based on size and contrast enhancement. The use of positron emission tomography CT (PET-CT) in detecting metastasis in the clinically negative neck has shown a sensitivity of 30-79%. 7,8 Currently, the sentinel node procedure provides the best sensitivity and specificity for the detection of occult neck metastasis. However, it is an invasive technique that complicates future neck dissection and has the disadvantage of any surgical intervention. Its role and prognostic impact has been well established in N0 melanomas but is also increasingly used in early staged oral carcinomas in the head and neck region. 9 2 Whereas imaging and sentinel node biopsies aim at early detection, during the last decades, a significant amount of research has focused on the biological information harbored in the primary tumor that predicts the chance of neck node metastasis. Many immunohistochemical markers, gene expression profiles but also features like depth of infiltration, invasive front and perineural growth have been shown to correlate with the risk of metastasis. 10,11,12 In this context, tumor lymphangiogenesis, the formation of new lymphatics related to malignant disease, has become a new field of studying the lymphatic dissemination of various neoplasms. 13 The initial concept of lymphatic metastasis was that tumor cells spread solely through preexisting, peritumoral lymphatic vessels that serve as passive channels. 14 However, the emergence of a number of lymphatic vessel markers has provided new insights into the active process of developing a tumor-associated lymphatic vasculature, analogous to the process of neovascularization. The first markers identified include the homeobox transcription factor Prox-1, the lymphatic vessel endothelial hyaluronan receptor-1 (LYVE-1), desmoplakin, podoplanin and vascular endothelial growth factor receptor-3 (VEGFR-3), offering the advantage to differentiate between the presence of blood and lymphatic vessels. 15 Moreover, research has revealed a number of growth factors and chemokines that are involved in the process of tumor lymphangiogenesis and provide additional information to 23

26 Chapter 2 understand the active role of malignant cells in the actual development of a pathological lymphatic vasculature, which serves as a conduit for regional spread. 16, Interestingly, structural changes have been also observed in non-metastatic sentinel nodes, as demonstrated in mouse models with skin cancer, combined with the overexpression of the lymphangiogenic growth factor VEGF-C. 17 It is assumed that this remodeling creates a favorable environment for the advent and growth of the tumor cells. 18 Nowadays, studies oriented to the prognostic role of tumor lymphangiogenesis in various neoplasms, including head and neck cancer, use immunohistochemistry (IHC) to quantify the density of lymphatic vessels inside the tumor mass, as well as in the peritumoral area in close proximity to the tumor margin. The main interest has been engrossed in the correlation of the lymphatic vessel density (LVD) to regional spread of HNSCC, in an attempt to discover an additional tool to estimate the risk of occult metastasis. The purpose of the present systematic review is to assess the robustness of the correlation of intratumoral and peritumoral LVD to lymph node metastasis. MATERIALS AND METHODS Search strategy A systematic search was conducted with the help of a medical information specialist in the PubMed and Embase databases for original articles published until the 5th of July Search terms used included lymphangiogenesis, lymphatic vessel density, head and neck neoplasms, squamous cell carcinoma and lymph node metastasis their synonyms in title and abstract (Supplemental File 1). Two authors (EK and FKL) independently screened all titles and abstracts of the retrieved search for selection using predefined inclusion and exclusion criteria that are reported below. Subsequently, full text of relevant studies was screened for a more detailed selection. Citations and references of selected articles and reviews were checked to identify potentially missed relevant studies. Discordant judgments were resolved by consensus discussion. Study selection Inclusion criteria Articles were selected on the basis of (1) original full text publication with detailed description of used methods, (2) original reports containing data on the assessment of LVD in HNSCC and its subsites (3) studies performed in human primary tumor treated surgically, (4) containing a correlation between LVD and lymph node metastasis. 24

27 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma Exclusion criteria Exclusion criteria were (1) duplicate articles that contained all or some of the original publication data, (2) reviews, book chapters, case reports, oral and poster presentations (3) experimental studies on animal models (4) articles that included subsites other than the head and neck region, (5) languages other than English. 2 Data extraction and analysis Using a standardized data extraction form we extracted first author, year of publication, sample size, tumor location, distribution of average age, endothelial marker, LVD assessment method and outcome in terms of established regional metastasis (through neck dissection, sentinel node procedure, or US-FNAC). Data extraction was performed independently by 2 authors (EK and FKL) and disagreements were resolved by discussion. Critical appraisal Using the Quality in Prognostic Studies (QUIPS) tool, the included articles were evaluated for the risk of bias. According to the QUIPS tool, risk of bias was scored as low, moderate or high for the following six items: sufficient study population, study attrition, adequate measurement of prognostic factor and outcome of interest, account of potential confounding factors and appropriate statistical analysis. 19 Studies scoring low risk of bias on three or more items were considered to be of high methodological quality, while studies scoring high risk of bias on three or more items were considered to be of low methodological quality. Articles with high and/or low risk of bias in less than three items were classified as of moderate quality. The methodological quality of the studies was appraised by two reviewers (EK and FKL) and disagreements on scoring were resolved by discussion. RESULTS Search results Combined PubMed and Embase searches retrieved 151 articles (Figure 1). After removing duplicates and screening titles and abstracts, 63 articles remained. The remaining articles underwent full text screening for formal review. Twenty-eight articles met the inclusion criteria and were eligible for further analysis The main reason for exclusion was the absence of correlation between LVD and lymphatic metastasis. The characteristics of the included studies are summarized in Table 1. 25

28 Chapter 2 Table 1 Characteristics of the included studies and correlation to lymph node metastasis. First Author Number of patients Tumor location Specimen Lymphatic Marker Studied area Counting method Stage Correlation of LVD to LNM Mashhadiabbas (2012) OC Resection Podoplanin IT-PT vessels per field I-IV S (p<0.001) Miyahara (2007) OC Biopsy Podoplanin IT-PT vessels per field I-IV S (p=0.001) Munoz-Guerra (2004) OC Resection Podoplanin IT present/absent T1-T2 S (regional recurrence, p=0.02) Siriwardena (2007) OC Not defined Podoplanin IT-PT vessels per field I-IV S (IT, p=0.028; PT, p=0.002) Zhao (2008) OC Resection Podoplanin IT-PT vessels per field I-IV S (IT, p<0.001; PT, p<0.001) Sugiura (2009) OC Biopsy Podoplanin Not defined vessels per field I-IV S (p<0.001; OR:1.80, CI: ) Kono (2013) OC Not defined Podoplanin PT vessels per field I-IV S (p<0.0001; OR:10.50) Chung (2010) To Resection Podoplanin IT-PT vessels per field I-IV S (IT, p=0.009) Zhang (2011) To Resection Podoplanin IT-PT vessels per field I-IV S (p<0.05; OR:1.73, CI: ) Sasahira (2010) To Biopsy-Resection LYVE-1 Not defined vessels per field I-IV S (p<0.0001) Baek (2009) La Resection Podoplanin Not defined Chalkley counting II-IV S (p< 0.001) method Lin (2011) La Resection Podoplanin IT-PT vessels per field I-IV S (IT, p<0.01; PT, p=0.004) Bolzoni Villaret (2013) La Resection Podoplanin PT present/absent T1-T2 S (locoregional recurrence, p=0.013) Audet (2005) La (73) OP (62) HP (33) Resection LYVE-1 IT Chalkley counting method III-IV (T3-T4 in larynx) S (La, p=0.001) Beasley (2002) OC (31) OP (23) La (16) Resection LYVE-1 IT Chalkley counting method I-IV S (OP, p=0.027) Frech (2009) OC (15) OP (18) HP (34) La (38) Resection LYVE-1 IT-PT vessels per field I-IV S (IT in OC, p=0.032; PT in HP, p=0.015) 26

29 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma Resection Podoplanin IT-PT vessels per field I-IV S (IT, p=0.002; PT, p=0.023) Garcia-Garracedo (2010) HP(18) OP (31) La (54) Resection Podoplanin IT vessels per field I-IV S (p<0.001; OR:3.2, CI: ) Hinojar-Gutierez (2010) La (104) HP (16) Resection Podoplanin IT-PT vessels per field III-IV (T3-T4) S (IT, p=0.02) Bolzoni Villaret (2010) OC (28) OP (17) I-IV S (p< 0.003) Not defined Podoplanin Not defined Chalkley counting method Chung (2011) OC (29) OP (19) Resection Podoplanin IT-PT vessels per field I-IV S (IT, p=0.003; PT, p=0.01; OR:11.32) Franchi (2004) OC (18) OP (12) La (22) Resection Podoplanin IT-PT vessels per field I-IV S (IT, p<0.001; PT, p=0.007) Kyzas (2005) LL (50) OC (22) La (9) De Sousa (2012) OC Resection-LN Podoplanin IT-PT vessels per field I-IV NS (IT p=0.083; PT p=0.729) Longatto Filho (2007) OC Resection Podoplanin IT-PT vessels per field I-IV NS (PT p=0.722; IT p=0.767) Watanabe (2013) OC Biopsy-Resection Podoplanin PT vessels per field I-IV NS (p=0.608) Okada (2010) To Biopsy Podoplanin Not defined vessels per field I-IV NS Ding (2014) To Resection LYVE-1 IT-PT vessels per field I-IV NS (p= 0.285) Biopsy -Resection LYVE-1 IT-PT vessels per field I-IV NS Koskinen (2005) La (33) HP (8) tonsil (4) OC (15) 2 27

30 Chapter 2 Figure 1 Flowchart of included and excluded studies. Quality assessment According to the QUIPS evaluation tool, the quality of the included studies ranged from low to high. Eight studies were of high quality, whereas six and fourteen articles were classified as of low and moderate quality, respectively (Table 2). The main reasons for lower scoring included the absence of control for confounding factors and logistic regression analysis, as well as the lack of cut-off values of the predictive factor. In some studies, the method of evaluating the lymph node status at presentation and during the follow-up period was not reported. Additionally, non-blinded prognostic factor assessment also led to poor scoring. However, the majority of the studies described thoroughly the procedure of immunostaining evaluation. No studies were excluded based on methodological quality. 28

31 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma Table 2 Quality assessment of the included studies using the QUIPS. Quality level Statistical analysis and reporting Study confounding Outcome measurement Prognostic factor measurement Study attrition Study Study participation Mashhadiabbas (2012) 25 Low Miyahara (2007) 26 Moderate Munoz-Guerra (2004) 27 Low Siriwardena (2007) 30 Low Zhao (2008) 33 Moderate Sugiura (2009) 34 High Kono (2013) 23 Moderate Chung (2010) 20 High Zhang (2011) 32 Moderate Sasahira (2010) 29 Moderate Baek (2009) 35 High Lin (2011) 38 Low Bolzoni Villaret (2013) 39 Moderate Audet (2005) 42 Moderate Beasley (2002) 43 Moderate Frech (2009) 44 High Garcia-Garracedo (2010) 45 High Hinojar-Gutierez (2010) 36 Moderate Bolzoni Villaret (2010) 40 High Chung (2011) 41 Moderate Franchi (2004) 46 High Kyzas (2005) 47 Moderate De Sousa (2012) 21 Moderate Longatto Filho (2007) 24 Low Watanabe (2013) 31 Moderate Okada (2010) 28 Low Ding (2014) 22 High Koskinen (2005) 37 Moderate 2 29

32 Chapter 2 Study characteristics Patients characteristics The number of patients included across all studies ranged from 30 to 168, while the mean age of the patients was between 45.3 to 65.2 years. Among the included articles, 15 dealt only with oral SCC and, moreover, 5 of these just with tongue cancer Three studies encompassed solely laryngeal SCC, whereas the rest referred to HNSCC from various anatomical sites, in particular oral cavity, oropharynx, larynx and hypopharynx In two papers the results were limited to advanced (stage III-IV) disease and in another two only tumors with early T stage were included. 27,39,40,42 The tissue specimens consisted either of completely resected tumors or biopsy material and formalin-fixed paraffin-embedded specimens were used in all studies. The follow-up period ranged from 1 to 143 months. Biomarkers In order to outline the lymphatic endothelium, immunohistochemistry using the polyclonal anti- LYVE-1 antibody was performed in six studies. 22,29,37,42-44 Podoplanin, traced by the monoclonal D2-40 antibody, was used as a lymphatic marker in all other studies. Staining with the proliferative marker Ki-67, combined with a lymphatic marker, was performed in 4 studies to identify potential proliferative activity in the tumor-associated lymphatic vessels. 26,33,43,47 Additional staining with a blood vessel endothelial marker, such as CD31, CD34 or CD105, was also performed in several studies in order to evaluate the blood vessel density. The association to the growth factor receptor VEGFR-3 was reported in only one of the articles. 21 LVD assessment In most of the studies both intratumoral and peritumoral lymphatics were assessed. In four papers only the intratumoural lymphatic vessels were studied. 27,36,42,43 On the contrary, in three articles the peritumoral lymphatics were exclusively quantified. 23,31,39 Although there is no universally accepted definition of the peritumoral area, it was reported as the region within 500 μm from the tumor margin in all but one study, in which it was defined as within 1 mm. 30 In all reports, evaluation of the LVD was performed in the areas with the highest lymphatic vessel density ( hot spots ), as this was estimated at low magnification. Scoring of the lymphatic vessels was accomplished with the Chalkley point overlap morphometric technique in four studies. 35,41-43 This method involves the use of an eyepiece graticule containing 25 randomly positioned dots, which is rotated until the maximum number of points is on or within the vessels. Thus, instead of counting the individual vessel, the overlaying dots are counted, reflecting the relative area occupied by the lymphatic vasculature. This technique allows for rapid analysis with a relatively low interobserver variability. 48 In the rest of the articles, LVD was assessed as the number of lymphatic vessels per high magnification field and in two papers as present or absent. 27,39 Mean or median values of LVD from 2 to 10 high magnification fields were used as cut-off values. Apart 30

33 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma from nodal metastasis, lymphatic vessel counts were also related to other clinical parameters, such as the tumor size, grade, gender and age, in most of the papers. Correlations to LVD Morphological features of tumor-associated lymphatic vessels A few papers provided morphological details on the intratumoral and peritumoral lymphatic vasculature. Lymphatic vessels were clearly distinguished from the adjacent blood vessels, due to the highly specific endothelial markers (anti-lyve-1, D2-40). Beasley et al described intratumoral lymphatic vessels with a distinctive reticular architecture within sheets of tumor cells in SCCs with a pushing margin and in areas containing leukocyte infiltration in tumors with an invasive margin. 43 In accordance with the findings of other authors, these vessels featured an immature morphology, comprising two or three endothelial cells, with miniscule, ill-defined lumina that could be easily discriminated from the wider peritumoral lymphatics with conventional architecture. 20,21,27,38 Furthermore, peritumoral lymphatics appeared more abundant than the intratumoral vessels. Decreased perimeter of the latter vessels was, interestingly, associated to higher risk for distant metastasis, according to one study. 40 Positive staining with the proliferative marker Ki-67 was observed only in a few small intratumoral lymphatics, but no dividing nuclei were found in the peritumoral area. The distribution of the proliferative marker supports the presence of newly formed intratumoral lymph vessels and not only pre-existing lymphatics. 43,47 A potential interconnection between intratumoral and peritumoral lymphatics in favor of tumor lymphatic spread, was also assumed possible in these articles. 2 LVD and lymph node metastasis in head and neck cancer In five studies that included a mix of SCC from various sites of the head and neck region, a strong association of high LVD values to regional dissemination was demonstrated. 36,40,41,45,46 Franchi et al reported mean intratumoral LVD values of 25.4±16.1 vessels per field for the N0 versus 45.7±19 vessels per field for the N+ group of patients, whereas values of peritumoral LVD were 39±8.9 and 51.8±12.8 vessels per field, respectively. 46 In another article, a 3.2-fold increased risk for lymphatic metastasis was exhibited in patients with increased counts of intratumoral lymphatic vessels. 36 On the contrary, in a study conducted by Koskinen et al on tumors from various subsites of the head and neck, a median LVD value of 7 vessels per field was found, but the correlation between LVD and N status was not significant. 37 LVD and lymph node metastasis in oral cancer A significant correlation to regional dissemination was observed in 12 studies that included oral carcinomas. 20,23,25-27,29,30,32-34,44,47 Two of these studies scored highly according to the quality assessment tool and four were judged as of low methodological level. In particular, according to Chung et al, this association was found only for intratumoral LVD (mean value: 16.2 vessels 31

34 Chapter 2 per field), which was additionally correlated to an increased rate of extracapsular spread of SCC, 20 Although multivariate analysis identified intratumoral LVD as the only variable significantly correlated to regional spread in the study of Chung et al, this was not proven by Frech et al, who showed tumor site as the most important predictive factor for nodal involvement. In 6 other series, both intratumoral and peritumoral LVD were found to be related to higher rates of regional dissemination, whereas higher nodal stage was also connected to higher lymphatic vessel counts. 25,26,29,30,32,34 Mean LVD values of 14.9±2.86 and 34.8±2.51 vessels per field were calculated, according to Miyahara et al, for the N0 and N+ group, respectively. 26 Similar values were found in another study regarding tongue cancer (15.97±0.67 vessels per field for the N0 group versus 29.38±0.57 vessels per field for the N+ group). 32 Odds ratios for lymphatic dissemination were calculated in only 3 studies, with values ranging from 1.73 to for cases with high LVD. 23,32,34 The relationship to regional relapse after surgical treatment of the neck in oral cancer was assessed only in a limited number of studies. An increased number of intratumoral lymphatic vessels was shown to be associated to regional, as well as local, recurrence, in contrast to the peritumoral LVD. 27,33 Sasahira et al demonstrated such an association only for cases of local relapse, which is in disagreement to the findings of another study. 24,29 In six studies, no statistically significant association was found between higher LVD values in the primary tumor and nodal metastasis or regional recurrence. 21,22,24,28,31,43 Only one of these studies was considered of high methodological quality. 22 It is also worth noting that, according to one paper, tumor LVD tended to decrease with higher stages of nodal involvement, although no significant differences of LVD were found in relationship to N status. 31 LVD and lymph node metastasis in oropharyngeal and hypopharyngeal cancer In cases of oropharyngeal and hypopharyngeal SCC, Frech et al did not find any significant correlation between intratumoral LVD and nodal disease, though there was a strong association of high peritumoral LVD to lymph node status in hypopharyngeal SCC (p= 0.015). 44 Additionally, another research did not demonstrate an association of increased intratumoral LVD to lymph node metastasis in oropharyngeal and hypopharyngeal cancer, in contrast to the results of a report considering oropharyngeal carcinoma. 42,43 In all these studies, a polyclonal antibody against LYVE-1 was used in immunostaining, but the latter two performed counting with the Chalkley point grid. LVD and lymph node metastasis in laryngeal cancer Regarding laryngeal SCC, all three studies that encountered solely larynx as the primary site, demonstrated a significant association of increased LVD to regional metastasis, using podoplanin 32

35 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma as a lymphatic marker. 35,37,38 Additionally, Audet et al found that high counts of intratumoral lymphatics are related to nodal metastasis in laryngeal carcinoma (p= 0.001), but not in oropharyngeal or hypopharyngeal cancer. 42 Baek et al reported a median LVD value of 4 (using the Chalkley counting method) and showed an important correlation of high LVD to lymph node metastasis at presentation in supraglottic SCC of the larynx (p< 0.001). 35 A significant association (p= 0.011) to regional relapse was also shown. However, in the logistic regression analysis model, VEGF-C appeared to be the strongest independent predictive factor for regional recurrence (p=0.001), when compared to T stage, N status and LVD, with an odd ratio of Lin et al reported median values of 0 (range 0-26) and 4 (range 0-32) vessels per field for intratumoral and peritumoral LVD, respectively. 38 Locoregional recurrence was not correlated to LVD according to the latter article. In contrast, Bolzoni Villaret et al demonstrated a significant correlation of increased LVD to locoregional relapse (p=0.013). 39 Nonetheless, no correlation was reported in one paper that included a limited number (sixteen) of cases with laryngeal carcinoma.,43 2 LVD in relationship to other parameters In a small number of articles an association between LVD values and primary tumor site was demonstrated. De Sousa et al found that a high peritumoral LVD was significantly associated to tumors localized in the tongue and floor of the mouth in comparison to other subsites of the oral cavity. 21 This is in accordance with the findings of three other researchers studying oral SCC. 23,26,27 Miyahara et al reported a mean value of 28.3±12.5 for tongue SCC versus 16.5±5.6 vessels per field for SCC from other sites of the oral cavity 26. However, it is not confirmed by other studies, that did not find any correlation between LVD and oral subsites. 24,25,33 These differences can be explained on the basis of the lower number of cases in these studies. Furthermore, other authors demonstrated a different density of lymphatic vessels in different subsites of the head and neck region. In particular, Franchi et al found an increased mean value (44.9±28.7 vessels per field) of intratumoral LVD in SCC of the oropharynx, compared to SCC of the oral cavity (26.2±16.4 vessels per field) and larynx (31.5±18.0 vessels per field). 46 Similarly, in another article, higher mean LVD values were observed for pharyngeal in comparison to laryngeal tumor sites. 45 An interesting finding is that a considerable proportion of glottic tumors (approximately 45%) displayed intratumoral lymphatics, even in T1 stage, though the glottic area is known for its absent lymphatic network and T1 glottic cancers rarely metastasize. 45 Moreover, Kyzas et al reported higher intratumoral LVD for oral cavity and laryngeal SCC in comparison to lip cancer, but not significantly higher values for peritumoral LVD. 47 Three authors showed a correlation of high LVD to moderate/poorly differentiated oral carcinomas. This was also reported in one of the studies dealing with laryngeal cancer. 25,28,42 Regarding the T stage, 5 authors reported a strong relationship of advanced T stage to increased intratumoral LVD in oral cancer and one study reported a similar finding in laryngeal and hypopharyngeal cancer. 20,21,26,29,41,45 33

36 Chapter 2 Some studies demonstrated also a correlation of increased LVD to survival. In a study on supraglottic laryngeal SCC, patients with increased LVD showed significantly shorter diseasefree and overall survival rates, a result that was confirmed by Garcia-Carracedo et al not only for laryngeal but also for pharyngeal carcinomas. 38,45 Multivariate analysis using a Cox proportional hazards model demonstrated intratumoral LVD as an independent prognostic factor for both overall and disease-free survival with a relative risk of 3.22 and 3.08 (p=0.008 and 0.005), respectively. 38 In a similar logistic regression model for disease free survival, Baek et al found a hazards ratio of 1.10 in patients with high LVD (p=0.904). 35 In that report, multivariate analysis highlighted VEGF-C expression as the strongest prognostic factor for disease-free survival with a hazards ratio of (p=0.006). Furthermore, there was a strong association between increased VEGF-C expression and positive N status, as well as regional recurrence, but not with LVD values. Disease-free survival was also strongly affected by high LVD in patients with oral carcinoma according to four articles. 26,27,29,33 Multivariate analysis performed by Zhao et al demonstrated intratumoral LVD as the only independent prognostic factor for cumulative survival in cases of oral SCC (p< 0.001), whereas N status was the strongest predictor of disease-free survival. 33 In another paper, intratumoral LVD was the most important prognostic factor for disease-free survival, but nodal status was not incorporated in the Cox hazards model. 27 Kyzas et al found that elevated intratumoral LVD was an independent factor of overall survival (p= 0.004) in cases of oral cavity and laryngeal SCC. 47 Hazard ratios, though, are not provided in the latter two articles. In the rest of the papers, either no association to survival rates was found or the relationship was not assessed. DISCUSSION Recent advances in the field of specific endothelial-type antibodies have provided valuable tools in studying the lymphatic vasculature and its prognostic implications. The monoclonal D2-40 antibody, against podoplanin, has been found to delineate an increased number of intratumoral lymphatics in comparison to other markers, demonstrating a sensitivity and specificity of 98.8% and 97.3%, respectively. 49,50 In addition, extensive analyses in many different laboratories have revealed that LYVE-1 is a reliable marker for distinguishing lymphatic vessels from blood vessels in a range of different human cancers In 2006, the first international consensus on the methodology of lymphangiogenesis quantification in solid tumors was established, providing guidelines for the assessment of the lymphatic vessel counts with the use of specific markers and the expression of lymphangiogenic growth factors. 60 According to these recommendations, studies on LVD should fulfill the following criteria: a. double immunostaining with the D2-40 and Ki-67 monoclonal antibodies, in order to detect 34

37 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma the presence of lymphatic vessels and proliferating lymphatic endothelial cells; b. hot spot selection at low magnification in intratumoral and/or peritumoral areas; c. quantification with the use of the Chalkley point graticule; d. counting both the proliferating and non-proliferating lymphatic endothelial cells; and e. independent evaluation by two pathologists. However, none of the studies, included in this review, adhered completely to the above recommendations, even those published after As detailed above, combined immunostaining with the proliferative marker Ki-67 and a lymphatic marker, which allows accurate detection of proliferating endothelial lymphatic cells, was performed in just four of the included studies. 26,33,43,47 In these studies, Ki-67 was mainly detected in the endothelial cells of the intratumoral lymphatic vessels, suggesting that these are proliferating new vessels and not preexisting lymphatics. However, the proliferative marker was only used to detect the presence of proliferating endothelial cells, and not to count these cells, as recommended by the consensus. Regarding the counting method, only four articles reported the use of the Chalkley point graticule. 35,41-43 LYVE-1 was the sole endothelial marker used in six studies, out of which two did not show any important correlation to regional metastasis. Although, the D2-40 antibody is considered the most reliable marker of the lymphatic endothelium, differences in the correlations between studies using the antibody against podoplanin or LYVE-1 were not observed in the current review, since a limited number of studies using the latter antibody were included. Additionally, while most researchers selected three hot spots for the assessment of LVD, in three studies only two hot spots were used. 31,37,39 Another issue is the evidence of regional disease. A neck dissection is not always performed in cases of HNSCC and US-FNAC or imaging are often the modality of staging the neck. Therefore, the presence of lymphatic metastasis may have been underestimated, contributing to the absence of association between LVD and nodal dissemination in some cases. This might work both ways and makes it difficult to calculate a risk factor associated with LVD. Multivariate analysis was not included in most of the articles and, thus, the relationship between LVD and nodal disease was not fully assessed. 2 Although the vast majority of the studies included in the review showed a significant correlation between high LVD and the presence of regional involvement, adoption of the recommended guidelines in future research will ensure a better comparison between the different studies from a methodological standpoint. In addition, there is an implication of a different effect of intratumoral and peritumoral LVD on regional disease and, thus, it would be optimal to study separately the correlation of these two parameters to lymphatic metastasis. This, however, is only possible in surgical resection specimens, which is difficult for some tumors, such as early stage larynx, nasopharynx and oropharynx, where most often a tissue biopsy is available. Furthermore, newly formed lymphatics, detected by dual staining with a lymphatic marker and the proliferative marker Ki-67, should be counted, according to the recommendations of the consensus. The current review also shows that there are differences in baseline levels of LVD at different (sub) sites in the head and neck, and it is thus not justified to analyze all these (sub)sites as one group. 35

38 Chapter 2 Reports on other types of malignancy support the significant correlation of LVD to nodal metastasis. A meta-analysis on the predictive value of LVD in melanoma, revealed a strong association of increased peritumoral LVD with lymph node metastasis. 55 However, a similar relationship was not proven for intratumoral LVD. High peritumoral LVD was also found to be associated to nodal status in gastric and breast carcinoma. 56,57 Nevertheless, another article on early gastric cancer, indicated a strong correlation between increased intratumoral LVD values and lymphatic metastasis. 58 Such a correlation was also found in a study on papillary thyroid carcinoma. 59 It is worth noting that heterogeneity is commonly found in the methodology (for example type of antibody used, counting method, number of hot spots selected) and the reported values of LVD in studies on other tumor sites. In conclusion, there is strong evidence that LVD is closely associated to lymphatic spread of HNSCC. In this review, most of the included studies had a mix of clinically N+ and N0 patients. However, in order to obtain quantitative data on prediction of occult metastasis in HNSCC, further research should focus on the value of LVD as a predictive marker in the clinically T1-2N0 group for each distinct tumor site. CONFLICT OF INTEREST The authors of this manuscript have no conflicts of interest to disclose and there has been no financial support for this research study. 36

39 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma REFERENCES 1. Marur S, Forastiere AA. Head and neck cancer: changing epidemiology, diagnosis, and treatment. Mayo Clin Proc. 2008; 83: Jemal A, Siegel R, Ward E, et al. Cancer statistics, CA Cancer J Clin 2009; 59: Rouviere H. Lymphatic system of the head and neck. Tobias MJ, Translator. Ann Arbor, MI: Edwards Brothers; van den Brekel MW, Castelijns JA, Stel HV, et al. Modern imaging techniques and ultrasound-guided aspiration cytology for the assessment of neck node metastases: a prospective comparative study. Eur Arch Otorhinolaryngol. 1993; 250: Stern WB, Silver CE, Zeifer BA, et al. Computed tomography of the clinically negative neck. Head Neck. 1990; 12: Friedman M, Mafee MF, Pacella BL Jr, et al. Rationale for elective neck dissection in Laryngoscope. 1990; 100: Myers LL, Wax MK, Nabi H, et al. Positron emission tomography in the evaluation of the N0 neck. Laryngoscope. 1998; 108: Wensing BM, Vogel WV, Marres HA, et al. FDG-PET in the clinically negative neck in oral squamous cell carcinoma. Laryngoscope 2006; 116: de Bree R, Nieweg OE. The history of sentinel node biopsy in head and neck cancer: From visualization of lymphatic vessels to sentinel nodes. Oral Oncol 2015; 51: Leusink FK, van Es RJ, de Bree R, et al. Novel diagnostic modalities for assessment of the clinically nodenegative neck in oral squamous cell carcinoma. Lancet Oncol 2012; 13: e van Hooff SR, Leusink FK, Roepman P, et al. Validation of a gene expression signature for assessment of lymph node metastasis in oral squamous cell carcinoma. J Clin Oncol 2012; 30: Mendez E, Fan W, Choi P, et al. Tumor-specific genetic expression profile of metastatic oral squamous cell carcinoma. Head Neck 2007; 29: Hirakawa S. From tumor lymphangiogenesis to lymphvascular niche. Cancer Sci 2009; 100: Zeidman I., Copeland B. E., Warren S. Experimental studies on the spread of cancer in the lymphatic system. II. Absence of a lymphatic supply in carcinoma. Cancer 1955; 8: Oliver G, Detmar M. The rediscovery of the lymphatic system. Old and new insights into the development and biological function of lymphatic vascular system. Genes Dev 2002; 16: Karatzanis AD, Koudounarakis E, Papadakis I, et al. Molecular pathways of lymphangiogenesis and lymph node metastasis in head and neck cancer. Eur Arch Otorhinolaryngol 2012; 269: Harrell MI, Iritani BM, Ruddell A. Tumor-induced sentinel lymph node lymphangiogenesis and increased lymph flow precede melanoma metastasis. Am J Pathol 2007; 170: Qian CN, Berghuis B, Tsarfaty G, et al. Preparing the soil : the primary tumor induces vasculature reorganization in the sentinel lymph node before the arrival of metastatic cancer cells. Cancer Res 2006; 66: Hayden JA, Côté P, Bombardier C. Evaluation of the quality of prognosis studies in systematic reviews. Ann Intern Med 2006; 144: Chung MK, Min JY, So YK, et al. Correlation between lymphatic vessel density and regional metastasis in squamous cell carcinoma of the tongue. Head Neck 2010; 32: de Sousa SF, Gleber-Netto FO, de Oliveira-Neto HH, et al. Lymphangiogenesis and podoplanin expression in oral squamous cell carcinoma and the associated lymph nodes. Appl Immunohistochem Mol Morphol 2012; 20: Ding L, Zhang Z, Shang D, et al. α-smooth muscle actin-positive myofibroblasts, in association with epithelial-mesenchymal transition and lymphogenesis, is a critical prognostic parameter in patients with oral tongue squamous cell carcinoma. J Oral Pathol Med 2014; 43:

40 Chapter Kono M, Watanabe M, Abukawa H, et al. Cyclo-oxygenase-2 expression is associated with vascular endothelial growth factor C expression and lymph node metastasis in oral squamous cell carcinoma. J Oral Maxillofac Surg 2013; 71: Longatto Filho A, Oliveira TG, Pinheiro C, et al. How useful is the assessment of lymphatic vascular density in oral carcinoma prognosis? World J Surg Oncol 2007; 5: Mashhadiabbas F, Mahjour F, Mahjour SB, et al. The immunohistochemical characterization of MMP-2, MMP-10, TIMP-1, TIMP-2, and podoplanin in oral squamous cell carcinoma. Oral Surg Oral Med Oral Pathol Oral Radiol 2012; 114: Miyahara M, Tanuma J, Sugihara K, et al. Tumor lymphangiogenesis correlates with lymph node metastasis and clinicopathologic parameters in oral squamous cell carcinoma. Cancer 2007; 110: Muñoz-Guerra MF, Marazuela EG, Martín-Villar E, et al. Prognostic significance of intratumoral lymphangiogenesis in squamous cell carcinoma of the oral cavity. Cancer 2004; 100: Okada Y. Relationships of cervical lymph node metastasis to histopathological malignancy grade, tumor angiogenesis, and lymphatic invasion in tongue cancer. Odontology 2010; 98: Sasahira T, Kirita T, Yamamoto K, et al. MIA-dependent angiogenesis and lymphangiogenesis are closely associated with progression, nodal metastasis and poor prognosis in tongue squamous cell carcinoma. Eur J Cancer 2010; 46: Siriwardena BS, Kudo Y, Ogawa I, et al. VEGF-C is associated with lymphatic status and invasion in oral cancer. J Clin Pathol 2008; 61: Watanabe S, Kato M, Ryoke K, Hayashi K. Lymphatic Vessel Density and Vascular Endothelial Growth Factor Expression in Squamous Cell Carcinomas of Lip and Oral Cavity: A Clinicopathological Analysis with Immunohistochemistry Using Antibodies to D2-40, VEGF-C and VEGF-D. Yonago Acta Med 2013; 56: Zhang Z, Pan J, Li L, et al. Survey of risk factors contributed to lymphatic metastasis in patients with oral tongue cancer by immunohistochemistry. J Oral Pathol Med 2011; 40: Zhao D, Pan J, Li XQ, et al. Intratumoral lymphangiogenesis in oral squamous cell carcinoma and its clinicopathological significance. J Oral Pathol Med 2008; 37: Sugiura T, Inoue Y, Matsuki R, et al. VEGF-C and VEGF-D expression is correlated with lymphatic vessel density and lymph node metastasis in oral squamous cell carcinoma: Implications for use as a prognostic marker. Int J Oncol 2009; 34: Baek SK, Jung KY, Lee SH, et al. Prognostic significance of vascular endothelial growth factor-c expression and lymphatic vessel density in supraglottic squamous cell carcinoma. Laryngoscope 2009; 119: Hinojar-Gutiérrez A, Fernández-Contreras ME, Alvarez-Carrillo et al. Role of intratumoral lymphatic vessels in the lymph node dissemination of laryngopharyngeal squamous cell carcinoma. Head Neck 2010; 32: Koskinen WJ, Bono P, Leivo I, et al. Lymphatic vessel density in vocal cord carcinomas assessed with LYVE-1 receptor expression. Radiother Oncol 2005; 77: Lin JY, Li XY, Dong P, et al. Prognostic value of lymphangiogenesis in supraglottic laryngeal carcinoma. J Laryngol Otol 2011; 125: Bolzoni Villaret A, Barbieri D, Peretti G, et al. Angiogenesis and lymphangiogenesis in early-stage laryngeal carcinoma: Prognostic implications. Head Neck 2013; 35: Bolzoni Villaret A, Schreiber A, Facchetti F, et al. Immunostaining patterns of CD31 and podoplanin in previously untreated advanced oral/oropharyngeal cancer: prognostic implications. Head Neck 2010; 32: Chung EJ, Rho YS, Baek SK, et al. Does a growing tumour volume induce lymphangiogenesis? A study of oral/oropharyngeal cancer. J Otolaryngol Head Neck Surg 2011; 40: Audet N, Beasley NJ, MacMillan C, et al. Lymphatic vessel density, nodal metastases, and prognosis in patients with head and neck cancer. Arch Otolaryngol Head Neck Surg 2005; 131:

41 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma 43. Beasley NJ, Prevo R, Banerji S, et al. Intratumoral lymphangiogenesis and lymph node metastasis in head and neck cancer. Cancer Res 2002; 62: Frech S, Hörmann K, Riedel F, et al. Lymphatic vessel density in correlation to lymph node metastasis in head and neck squamous cell carcinoma. Anticancer Res 2009; 29: Garcia-Carracedo D, Rodrigo JP, Astudillo A, et al. Prognostic significance of lymphangiogenesis in pharyngolaryngeal carcinoma patients. BMC Cancer 2010; 10: Franchi A, Gallo O, Massi D, et al. Tumor lymphangiogenesis in head and neck squamous cell carcinoma: a morphometric study with clinical correlations. Cancer 2004; 101: Kyzas PA, Geleff S, Batistatou A, et al. Evidence for lymphangiogenesis and its prognostic implications in head and neck squamous cell carcinoma. J Pathol 2005; 206: Vermeulen PB, Gasparini G, Fox SB, et al. Second international consensus on the methodology and criteria of evaluation of angiogenesis quantification in solid human tumours. Eur J Cancer 2002; 38: Dadras SS, Paul T, Bertoncini J, et al. Tumor lymphangiogenesis: a novel prognostic indicator for cutaneous melanoma metastasis and survival. Am J Pathol 2003; 162: Kane SV, Gupta M, Kakade AC, et al. Depth of invasion is the most significant histological predictor of subclinical cervical lymph node metastasis in early squamous carcinomas of the oral cavity. Eur J Surg Oncol 2006; 32: Wetterwald A, Hoffstetter W, Cecchini MG, et al. Characterization and cloning of the E11 antigen, a marker expressed by rat osteoblasts and osteocytes. Bone 1996; 18: Breiteneder-Geleff S, Soleiman A, Kowalski H, et al. Angiosarcomas express mixed endothelial phenotypes of blood and lymphatic capillaries: podoplanin as a specific marker for lymphatic endothelium. Am J Pathol 1999; 154: Kahn HJ, Bailey D, Marks A. Monoclonal antibody D2-40, a new marker of lymphatic endothelium, reacts with Kaposi s sarcoma and a subset of angiosarcomas. Mod Pathol 2002; 15: Van der Auwera I, Cao Y, Tille JC, et al. First international consensus on the methodology of lymphangiogenesis quantification in solid human tumours. Br J Cancer 2006; 95: Pastushenko I, Vermeulen PB, Carapeto FJ, et al. Blood microvessel density, lymphatic microvessel density and lymphatic invasion in predicting melanoma metastases: systematic review and metaanalysis. Br J Dermatol 2014; 170: Wang XL, Fang JP, Tang RY, et al. Different significance between intratumoral and peritumoral lymphatic vessel density in gastric cancer: a retrospective study of 123 cases. BMC Cancer 2010; 10: Kandemir NO, Barut F, Bektas S, et al. Can lymphatic vascular density be used in determining metastatic spreading potential of tumor in invasive ductal carcinomas? Pathol Oncol Res 2012; 18: Gao P, Zhou GY, Zhang QH, et al. Clinicopathological significance of peritumoral lymphatic vessel density in gastric carcinoma. Cancer Lett 2008, 263: Hall FT, Freeman JL, Asa SL, et al. Intratumoral lymphatics and lymph node metastases in papillary thyroid carcinoma. Arch Otolaryngol Head Neck Surg 2003, 129:

42 Chapter 2 SUPPLEMENTARY FILES Supplementary Table 1 Search strategy for Pubmed. #1 #2 Pubmed ( Carcinoma, Squamous Cell [Mesh:NoExp] OR squamous cell carcinoma* [tiab] OR squamous carcinoma* [tiab] OR epidermoid carcinoma* [tiab] OR Planocellular carcinoma* [tiab] OR Carcinoma, squamous cell of head and neck [Supplementary Concept]) AND ( Head and Neck Neoplasms [MeSH:NoExp] OR ((head [tiab] OR neck [tiab] OR UADT [tiab] OR aerodigestive tract [tiab] OR facial [tiab] OR mouth [tiab] OR gingival [tiab] OR lip [tiab] OR lips [tiab] OR palatal [tiab] OR Salivary Gland* [tiab] OR parotid [tiab] OR Sublingual Gland* [tiab] OR Submandibular Gland [tiab] OR tongue [tiab] OR Otorhinolaryn* [tiab] OR ear [tiab] OR ears [tiab] OR Auricular [tiab] OR laryn* [tiab] OR nose [tiab] OR nasal [tiab] OR Paranasal Sinus [tiab] OR Maxillary Sinus [tiab] OR pharyn* [tiab] OR Hypopharyn* [tiab] OR Nasopharyn* [tiab] OR Oropharyn* [tiab] OR Tonsillar [tiab] OR tonsil* [tiab]) AND (neoplasm* [tiab] OR tumor* [tiab] OR tumour* [tiab] OR cancer* [tiab] OR malign* [tiab] OR oncolog* [tiab] OR carcinom* [tiab] OR carcinogenes* [tiab] OR oncogenes* [tiab]))) OR oral neoplasm* [tiab] OR oral tumor* [tiab] OR oral tumour* [tiab] OR oral cancer* [tiab] OR oral malign* [tiab] OR oral oncolog* [tiab] OR oral carcinom* [tiab] OR oral carcinogens* [tiab] OR oral oncogenes* [tiab] OR ((oral cavit* [tiab] OR oral squamous cell [tiab]) AND (neoplasm* [tiab] OR tumor* [tiab] OR tumour* [tiab] OR cancer* [tiab] OR malign* [tiab] OR oncolog* [tiab] OR carcinom* [tiab] OR carcinogenes* [tiab] OR oncogenes* [tiab]) ( Lymphangiogenesis [Mesh] OR Lymphangiogenes* [tiab] OR lymphatic vessel densit* [tiab]) AND ( Lymphatic Metastasis [Mesh] OR Lymphatic Metastas* [tiab] OR lymph node metastas* [tiab]) Domain Determinant #3 #1 AND #2 Final search Supplementary Table 2 Search strategy for Embase. Embase #1 ( head and neck cancer / OR face cancer/ OR head and neck carcinoma / OR head and neck squamous cell carcinoma / OR head cancer/ OR lip cancer/ OR mouth cancer/ OR neck cancer/ OR nose cancer/ OR paranasal sinus cancer/ OR pharynx cancer/ OR salivary gland cancer/ OR tongue cancer/ OR tonsil cancer/) OR (head OR neck OR UADT OR (aerodigestive adj1 tract) OR facial OR mouth OR gingival OR lip OR lips OR palatal OR (Salivary adj1 Gland*) OR parotid OR (Sublingual adj1 Gland*) OR (Submandibular adj1 Gland) OR tongue OR Otorhinolaryn* OR ear OR ears OR Auricular OR laryn* OR nose or nasal or (Paranasal adj1 Sinus) or (Maxillary adj1 Sinus) Domain or pharyn* or Hypopharyn* or Nasopharyn* or Oropharyn* or Tonsillar or tonsil* or oral) adj3 (neoplasm* or tumor* or tumour* or cancer* or malign* or oncolog* or carcinom* or carcinogenes* or oncogenes*)).ti,ab.and (squamous cell carcinoma/ exp) OR (squamous adj1 cell) OR squamous OR epidermoid OR Planocellular) adj1 carcinoma*).ti,ab). #2 (Lymphangiogenes* OR (lymphatic adj1 vessel adj1 densit*)).ti,ab.or Lymphangiogenesis/exp AND ((Lymphatic adj1 Metastas*) OR (lymph adj1 node adj1 metastas*)).ti,ab OR Lymphatic Metastasis/exp) Determinant #3 #1 AND #2 Final search 40

43 Lymphatic vessel density and lymph node metastasis in head and neck squamous-cell carcinoma 2 41

44

45 CHAPTER 3 Validation of a gene expression signature for the assessment of lymph node metastasis in oral squamous-cell carcinoma F.K.J. Leusink* S.R. van Hooff* P. Roepman R.J. Baatenburg de Jong E.J.M. Speel M.W.M. van den Brekel M.L.F. van Velthuysen P.J. van Diest R.J.J. van Es M.A.W. Merkx J.A. Kummer C.R. Leemans E. Schuuring J.A. Langendijk M. Lacko M.J. De Herdt J.C. Jansen R.H. Brakenhoff P.J. Slootweg R.P. Takes* F.C.P. Holstege* * Equal contribution Journal of Clinical Oncology 2012;30:

46 Chapter 3 ABSTRACT INTRODUCTION Current assessment of lymph node metastasis in patients with head and neck squamouscell carcinoma is not accurate enough to prevent overtreatment. The aim of this study was validation of a gene expression signature for distinguishing metastasizing (N+) from nonmetastasizing (N0) squamous-cell carcinoma of the oral cavity (OSCC) and oropharynx (OPSCC) in a large multicenter cohort, using a diagnostic DNA microarray in a Clinical Laboratory Improvement Amendments/International Organization for Standardization approved laboratory. METHODS A multigene signature, previously reported as predictive for the presence of lymph node metastases in OSCC and OPSCC, was first re-evaluated and trained on 94 samples using generic, whole-genome, DNA microarrays. Signature genes were then transferred to a dedicated diagnostic microarray using the same technology platform. Additional samples (n = 222) were collected from all head and neck oncologic centers in the Netherlands and analyzed with the diagnostic microarray. Human papillomavirus status was determined by real-time quantitative polymerase chain reaction. RESULTS The negative predictive value (NPV) of the diagnostic signature on the entire validation cohort (n = 222) was 72%. The signature performed well on the most relevant subset of early-stage (ct1-t2n0) OSCC (n = 101), with an NPV of 89%. CONCLUSION Combining current clinical assessment with the expression signature would decrease the rate of undetected nodal metastases from 28% to 11% in early-stage OSCC. This should be sufficient to enable clinicians to refrain from elective neck treatment. A new clinical decision model that incorporates the expression signature is therefore proposed for testing in a prospective study, which could substantially improve treatment for this group of patients. 44

47 Validation of a gene expression signature for the assessment of lymph node metastasis in oral squamous-cell carcinoma INTRODUCTION Each year, nearly 500,000 people are diagnosed with head and neck squamous cell cancer (HNSCC), 1,2 making it the sixth most common cancer worldwide. Cervical lymph node metastasis occurs frequently in these patients and is a major determinant of prognosis and treatment planning. Accurate lymph node staging is therefore crucial. Current preoperative clinical assessment of the lymph nodes by physical examination and imaging is suboptimal. Depending on the techniques used, in approximately 20% to 40% of all patients with HNSCC, nodal metastases remain undetected during diagnostic work-up. 3-6 Establishing optimal treatment of patients who are clinically assessed as having nonmetastasizing (N0) disease is therefore challenging. 7 3 If no nodal metastasis is detected, the preferred treatment in many centers is to electively treat the neck, in addition to treating the primary tumor in all patients. An alternative is to refrain from additional neck treatment, closely observe the patient (watchful waiting), and only treat the necks of patients who develop clinically apparent metastases during follow-up. Whichever strategy is chosen, suboptimal treatment may be given. Most patients with HNSCC are treated by surgery, (chemo)radiotherapy, or combinations thereof. The primary treatment of oral cavity squamous cell cancer (OSCC) is surgery, in most cases followed by (chemo)radiotherapy on indication. Neck dissection is therefore also the most frequent mode for treatment of the neck in OSCC. In the case of elective neck dissection, the 60% to 80% of patients who are clinically assessed as N0 (cn0) and are indeed free of metastases receive unnecessary treatment of the neck, causing morbidity, including shoulder dysfunction, pain, lymphedema, contour changes, and lower lip paresis, even in more conservative types of selective neck dissection. 8,9 Conversely, watchful waiting results in undertreatment of the 20% to 40% of patients with occult metastases, which may result in an unfavorable prognosis in the case of delayed treatment. Because there is no conclusive evidence to indicate which approach is best, the management of patients who are clinically assessed as having N0 disease is one of the major issues in the management of HNSCC and OSCC in particular, which is reflected in the different approaches between centers. 10,11 Treatment of the N0 neck, both surgically and by radiotherapy, would greatly benefit from improved preoperative assessment of lymph node metastasis, which is the goal of this study. Gene expression profiling has been shown to be useful for diagnosis and prognosis of several cancers. 12 With regard to HNSCC, a previous study identified a gene expression signature for distinguishing metastasizing (N) from N0 OSCC and N0 oropharynx SCC (OPSCC). 13 Although promising, the independent validation cohort in that study was small (n=22), with all samples (N=104) derived from a single clinical center. Reanalysis of the entire data set identified more 45

48 Chapter 3 genes with predictive power. 14 In the current study, this predictive gene set was first re-evaluated using a different microarray platform on 94 samples, transferred to a dedicated diagnostic microarray, and subsequently tested in a large, multicenter patient cohort (n=222). On the basis of the results, a new clinical decision model that incorporates the gene expression signature is proposed for prospective analysis of early-stage OSCC management. METHODS Platform Transition Previously, a 102-gene profile, 13 and later, a more comprehensive set of genes, 14 were reported as having predictive value for assessment of lymph node status in OSCC and OPSCC. These were identified on self-made, whole-genome, DNA oligonucleotide microarrays using a pool of tumor samples as common reference. The current study consisted of two stages (Fig 1). In the first stage (platform transition), a more generic diagnostic tool was developed for application in a Clinical Laboratory Improvement Amendments/International Organization for Standardization (CLIA/ISO) approved laboratory that also performs a US Food and Drug Administration cleared microarray test. Platform transition was achieved by adopting commercially available technology and by determining classification settings for the new platform on the first cohort as a whole, or on a subset for which improved assessment of metastasis is most relevant (Appendix Fig A1, online only). The second stage consisted of independent multicenter validation of the resulting diagnostic array on a second cohort (Fig 1). Samples from patients Fresh frozen primary tumor samples of OSCC (86% of all samples) and OPSCC were collected from patients with no previous oncologic history. Because the original signature was developed on samples of both OSCC and OPSCC, both were also included here. In the Data Supplement, the characteristics of each patient are described individually, and these data are linked to individual array data. Clinical assessment and histologic examination consisted of palpation, computed tomography and/ or magnetic resonance imaging, and ultrasound with ultrasound-guided, fineneedle aspiration cytology. The reference standard for nodal status was considered positive in the case of cytologically or histologically proven metastases and negative in the case of no detected metastasis in the neck dissection specimen and/or during follow-up of at least 2 years. Because the selection of patients was based on the availability of frozen material, the number of patients who did not receive a neck dissection in this study population reflects clinical practice in the participating centers. Some centers have adopted a watchful waiting strategy with respect to the neck in patients with ct1n0 tumors. Nine such patients were included in the validation cohort and three in the platform transition cohort. 46

49 Validation of a gene expression signature for the assessment of lymph node metastasis in oral squamous-cell carcinoma 3 Figure 1 Schematic representation of the different stages of this study. Previous studies resulted in discovery of a comprehensive multigene signature. 13,14 In this study, the platform transition cohort (n = 94) was used to transfer this signature to a generic diagnostic platform. The resulting diagnostic microarray with trained settings was then validated on the multicenter validation cohort (n = 222). The results of this validation are incorporated in a clinical decision model (see Fig 3). Two separate cohorts were studied. The first, the platform transition cohort, consisted of 94 tumors that were collected from patients who were treated in the University Medical Center Utrecht (Utrecht, the Netherlands) between 1998 and The second, the multicenter validation cohort, consisted of 222 tumors that were collected from patients who were treated in one of the eight centers cooperating in the Dutch Head and Neck Society between 2000 and Table 1 describes the clinical characteristics of both cohorts. Table 2 describes the distribution across the participating centers. The study was approved by the institutional review boards of the participating centers. 47

50 Chapter 3 Table 1 Characteristics of patients in the two study cohorts* Characteristic Platform Transition Cohort (n = 94) Validation Cohort (n = 222) No. % No. % Age.66 Years SD Age group Location of primary tumor.16 Oral cavity Oropharynx Clinical T stage.002 T T T T Clinical N status.21 Positive Negative Pathologic N status.27 Positive Negative Abbreviation: SD, standard deviation. *Percentages may not total 100 because of rounding. P values were calculated with the use of Mann-Whitney test for continuous variables and Fisher s exact test for categorical variables. P Table 2 Participating medical centers and their sample contributions to the multicenter validation set Medical Center Name Abbreviation No. of Samples Netherlands Cancer Institute Antoni van Leeuwenhoek NKI-AvL 64 Radboud University Nijmegen Medical Center RUNMC 50 University Medical Center Utrecht UMCU 49 Vrije Universiteit Medical Center VUMC 18 Maastricht University Medical Center MUMC 16 University Medical Center Groningen UMCG 13 Erasmus Medical Center Erasmus MC 7 Leiden University Medical Center LUMC 5 48

51 Validation of a gene expression signature for the assessment of lymph node metastasis in oral squamous-cell carcinoma Analysis of gene expression Frozen tumor samples were sectioned at the participating institutions, aliquoted in Trizol (Life Technologies, Frederick, MD), and sent to Agendia laboratories (Amsterdam, the Netherlands) for expression profile analysis. Tumor percentage (at least 50%) was assessed on hematoxylin and eosin stained sections taken in parallel. RNA isolation, amplification, and labeling were performed as previously described. 15 Tumor sample RNA was labeled as Cy3, and reference RNA was labeled Cy5. As a reference, the Universal Human Reference RNA (Agilent Technologies, Santa Clara, CA) was used. Samples from the platform transition cohort were hybridized on full-genome Agilent arrays that included the originally identified gene expression signature 13,14 in replicate. Samples from the multicenter validation cohort were hybridized on dedicated diagnostic eight-pack arrays that represented this signature. Raw fluorescence intensities were quantified using Agilent Feature Extraction software and imported into R/Bioconductor ( for normalization (loess normalization using the limma package) and additional analysis. Because of global differences between the cell line reference RNA and tumor sample RNA, the expression values of all genes were centered using the unweighted aggregate mean per gene of pn and pn0 samples from the platform transition cohort. More information on these procedures and array designs is given in the Appendix (online only) and Data Supplement. Microarray data has been deposited in the National Center for Biotechnology Information Gene Expression Omnibus 16 under accession number GSE Gene Signature Readout Analysis of the originally identified comprehensive gene set, 13,14 with updated genome annotation and alongside the probes available on the manufacturers arrays, resulted in 732 unique probes that represent 696 genes. All analyses are based on this 732-probe set. Signature readout was through a mean-based classification, 13 on the basis of an updated risk profile that was derived from the mean expression profile of the signature in all patients with lymph node positive disease (pn) in the platform transition set. This mean expression profile was used as the Nrisk profile for all subsequent analyses. Patients with a cosine correlation coefficient (ie, signature score) of were classified as N0; patients with a signature score of > were classified as N. The threshold was based on the subgroup for which additional nodal assessment is most crucial (early-stage [ct1-t2n0] OSCC) and set to achieve a negative predictive value (NPV) of at least 90%, with the highest possible specificity (Fig A1D). For the multicenter validation, the same procedure was used: the signature score was calculated using the Nrisk profile and the threshold determined on the platform transition cohort was applied. To determine whether any center-specific bias existed, the signature scores for patients from individual centers were compared with the signature scores of patients from the remaining centers. This was performed for patients assessed as pn0 and pn separately using a Mann-Whitney test. 3 49

52 Chapter 3 Detection of human papillomavirus Detection of human papillomavirus (HPV16) was performed by realtime quantitative polymerase chain reaction with 100 ng of total RNA using the iscriptcdnakit (Bio-Rad, Veenendaal, the Netherlands)17 with polymerase chain reactions and controls as described. 18 RESULTS Transition to a dedicated diagnostic platform Previously reported gene signatures 13,14 were based on self-made microarrays using a common reference pool of samples, both with limited availability. To perform analyses with CLIA/ISOapprovable settings, commercially available microarrays and commercially available reference material were adopted (see Methods). A platform transition step was therefore required to determine classification settings with these new technologic features before independent multicenter validation (Fig 1). The platform transition cohort consisted of 94 tumor samples (Table 1). First, the performance of the most comprehensive, previously reported gene set was re-evaluated on whole-genome arrays. No gene set was found (data not shown) that performed better; this was expected, given that this signature is a compilation of many different, robustly performing gene sets. 14 This signature was therefore applied for additional analyses. Signature scores for the platform transition cohort are depicted in Figure 2A. To complete the transition to a diagnostic microarray (Fig 1), an optimal classification threshold was sought (see Methods). Figure A1 shows the trade-off between NPV (pn0 as percentage of all N0 predictions) versus specificity, compared with positive predictive value versus sensitivity, when choosing a cutoff. Because no threshold could be found that was sufficiently optimal for all patients (Fig A1C), the threshold was based on those patients for whom an improved N0 prediction would be most useful, that is, patients with early-stage (ct1-t2n0) OSCC,who are most frequently treated with neck dissection (Fig 2B and Fig A1D). This threshold yields an NPV of 92% for this group (Fig 2B; Table 3). The NPV for the entire platform transition cohort is 75% (Fig 2A; Table 3). The result of the platform transition stage is therefore a set of genes with predictive expression patterns as determined from previous studies, 13,14 with settings trained for application in an independent validation study using a diagnostic microarray (Fig 1). 50

53 Validation of a gene expression signature for the assessment of lymph node metastasis in oral squamous-cell carcinoma 3 Figure 2 Signature scores of the study cohorts and subgroups. Gold triangles indicate patients with pn disease. Blue squares indicate patients with pn0 disease. The dotted line indicates the signature score threshold ( 0.256), below which patients are classified as N0 and above which patients are classified as N. (A) Signature scores for all patients in the platform transition cohort. (B) Signature scores for patients with early-stage (ct1-t2n0) oral cavity squamous cell cancer (OSCC) in the platform transition cohort. (C) Signature scores for all patients in the validation cohort. (D) Signature scores for patients with early-stage (ct1-t2n0) OSCC in the validation cohort. (E) Distribution of signature scores for patients in the validation cohort from all centers combined, the three centers with the highest sample contribution (Netherlands Cancer Institute [NKI-AvL], Radboud University Nijmegen Medical Center [RUNMC], and University Medical Center Utrecht [UMCU]), and the aggregate of the five other centers (Erasmus Medical Center, Leiden University Medical Center, Vrije Universiteit Medical Center, University Medical Center Groningen, and Maastricht University Medical Center). The blue boxplots show the signature score distributions for patients with pn0 disease; the gold boxplots show the signature score distributions for patients with pn disease. The rectangular box represents the interquartile range (IQR); the black horizontal line within the box represents the median. The upper whisker extends to the highest observed value within 1.5 IQR of the upper quartile; the lower whisker extends to the lowest observed value within 1.5 IQR of the lower quartile. Observations outside of the maximum ranges of the whiskers are shown as individual points. The horizontal dotted line indicates the signature score threshold, above which patients are classified as N and below which patients are classified as N0. 51

54 Chapter 3 Table 3 Performance metrics of the gene signature and clinical assessment Platform Transition Cohort, All (n = 94) Platform Transition Cohort, ct1-t2n0 OSCC (n = 39) Validation Cohort, All (n = 222) Validation Cohort, ct1-t2n0 OSCC (n = 101) Gene Signature Clinical Assessment Gene Signature Clinical Assessment Gene Signature Clinical Assessment Gene Signature Clinical Assessment NPV, % % CI* to to to to to to to 80.1 TN TN + FN PPV, % NA NA 95% CI* to to to to to 49.1 TP TP + FP Sensitivity, % NA NA 95% CI* to to to to to 94.3 TP TP + FN Specificity, % NA NA 95% CI* to to to to to 55.2 TN TN + FP Accuracy, % % CI* to to to to to to to 80.1 TP + TN TP + FP + TN + FN Abbreviations: FN, false negative; FP, false positive; NA, not applicable; NPV, negative predictive value; OSCC, oral cavity squamous cell cancer; PPV, positive predictive value; TN, true negative; TP, true positive. *CIs were calculated using the Wilson method. The clinically relevant subset contains only N0 clinical assessments, precluding the presence of TPs and FPs. 52

55 Validation of a gene expression signature for the assessment of lymph node metastasis in oral squamous-cell carcinoma Independent validation on multicenter cohort To validate the signature-based classification, 222 samples were collected from eight institutions (Tables 1 and 2). Gene expression was analyzed on the dedicated diagnostic array and classified on the basis of the settings derived from the platform transition cohort (Figs 2C and 2D). Performance is summarized in Table 3. The NPV of the signature on the entire validation cohort was 72% (Fig 2C; Table 3). This result is similar to the NPV that was achieved by previous clinical assessment (73%; Table 3). Combining clinical assessment with the signature outcome results in an NPV of 89% for patients who have early-stage (ct1-t2n0) OSCC (Fig 2D; Table 3). No significant outliers between the signature scores from the different centers were detected for patients with and without lymph node metastasis (Mann-Whitney, P >.05; Fig 2E). This indicates that the signature performs similarly across clinical centers. Importantly, the independent validation shows that the signature-based prediction has a significant added value when applied in combination with current clinical assessment, increasing the NPV to 89% for early-stage OSCC. 3 Clinical decision model incorporating the gene expression signature On the basis of these results, the following clinical decision model is proposed (Fig 3A). A patient with early-stage (ct1-t2n0) OSCC can be analyzed by gene expression profiling. If applied to resection material, this would imply a two-stage procedure: resection of the primary tumor first and a subsequent neck dissection in patients with a high-risk profile. This would be analogous to sentinel node procedures 19,20 and could cause some delay in treatment and logistical challenges. More optimally, the test could be performed on biopsy material that was obtained at initial diagnosis. This would not alter current timing but first requires a separate evaluation of the similarity in signature outcomes for such biopsies. On the basis of the additional information provided by the signature, a choice of treatment could be made between elective neck dissection when the signature-based prediction is N+, or watchful waiting when the prediction is N0. Current elective treatment with a neck dissection of level I to III results in overtreatment of 72% of patients (Fig 3B, left side). The proposed scheme would prevent overtreatment in almost half of these patients by replacing neck dissection with watchful waiting (Fig 3B, right side). In addition to enhancing the effectiveness of the surgical intervention, this projected improvement will yield significant benefits in terms of morbidity and costs for a large number of patients who are currently being overtreated. 53

56 Chapter 3 Figure 3 Incorporation of the gene signature in a clinical decision model. (A) Clinical decision model that includes the readout of the gene signature. This model is proposed for patients with early-stage (ct1-t2n0) oral cavity squamous cell cancer (OSCC). In the event of a positive classification (N) on the basis of the gene signature, elective neck dissection of level I-III is proposed. For patients with a negative classification (N0), watchful waiting is proposed. (B) The consequences of this model on the basis of the validation cohort. Currently, all patients with early-stage (ct1-t2n0) OSCC are treated with elective neck dissection of level I-III, causing overtreatment in 73 patients (first bar). In the model involving gene signature readout, overtreatment is limited to 41 patients (second bar), and 32 patients now receive the most appropriate treatment (third bar). The group of patients that are actually N and who receive elective neck dissection of level I-III (indicated by gold color) are labeled as receiving mostly appropriate treatment here, given that in subclinical or limited metastatic disease, the metastases are usually confined to level I-III. DISCUSSION Previous studies have indicated that gene expression profiling can potentially help predict lymph node status in HNSCC and OSCC, in particular Such a signature 13,14 is completely independently verified in this large multicenter study that was performed in a CLIA/ISO certified laboratory using a diagnostic array platform. The results indicate that the signature should be prospectively tested and applied alongside current clinical assessment to identify a subgroup of patients with OSCC for whom a watchful waiting strategy would be appropriate. OSCC is predominantly treated surgically. 7 In patients with advanced tumors (ct3-t4), for whom the probability of cervical lymph node metastases is relatively high and for whom free flap reconstruction of the neck is often required, most head and neck surgeons opt for elective neck dissection. 24,25 The decision model (Fig 3A) does not propose that this approach be changed. The issue of elective neck dissection 54

57 Validation of a gene expression signature for the assessment of lymph node metastasis in oral squamous-cell carcinoma versus watchful waiting is most relevant for early-stage (ct1t2n0) OSCC. 24,25 In these patients elective neck dissection is commonly performed. Treatment of the cn0 neck in early-stage OSCC is one of most heavily debated subjects in head and neck oncology. Although it can be argued that performing elective neck dissection provides prognostic information and could identify patients with pn disease at an early stage, this strategy has not been convincingly proven to be superior as regards survival and control of neck disease. 7 The procedure also comes with a price. 3 Currently, approximately 60% to 80% of these patients unnecessarily undergo a procedure that has potentially serious associated morbidity, even in the often performed selective neck dissections. 8,9 Improving the NPV of lymph node staging such that a watchful waiting strategy becomes more appropriate would therefore improve the quality of life for a significant number of patients with OSCC. Reduction in operating time and improved effectiveness of surgical treatment are additional likely advantages. 3 Factors that require consideration when judging the combined 89% NPV for early-stage (ct1- T2N0) OSCC that is reported here include the incidence of occult metastases in general, as well as HPV infection. 26 The false prediction rate of N0 is dependent on the prevalence of occult metastases in the population studied (Bayes theorem). Within the validation cohort, 28% of the patients with early-stage (ct1-t2n0) OSCC show occult metastases. This is in line with the level of incidence reported in other studies (20% to 40%) 3-6 and indicates that 89% is a realistic estimate of performance in this group of patients. HPV infection represents a distinct etiologic factor for HNSCC, with different tumor characteristics and prognosis In studies that distinguish between SCC in the oral cavity (OSCC) and oropharynx (OPSCC), the incidence of HPV in OSCC is less than 4%, 30 which indicates that this is unlikely to influence the results reported here. To confirm this, HPV incidence was determined (see Methods) and was found to be 4% in OSCC samples of the validation cohort and not at all present in any of the patients with early-stage (ct1-t2n0) OSCC. HPV incidence therefore does not interfere with the proposed clinical decision model (Fig 3A), which only includes earlystage (ct1-t2n0) OSCC. OPSCC exhibits a higher incidence of HPV but was excluded from the clinical decision model because these tumors are predominantly treated nonsurgically in the Netherlands; therefore, elective neck dissection is less commonly performed in such patients. Performance of the signature on OPSCC and HPV-positive subgroups from the entire validation cohort was nevertheless analyzed separately and demonstrated poorer performance (Appendix Table A1, online only), in agreement with a distinct etiology and higher rate of occult metastasis. Is a combined NPV of 89% for early stage OSCC sufficient to warrant watchful waiting (Fig 3)? Feasibility of watchful waiting is strongly linked to the probability of occult metastases, the frequency and sensitivity of follow-up, and the success rate of salvage therapy. Decision analysis 55

58 Chapter 3 techniques have recommended watchful waiting below risk thresholds varying from 17% to 44%. 5,31 Such thresholds are much higher than the 11% risk of occult metastases that is projected by the proposed clinical decision model, arguing in favor of implementation of the expression signature alongside current clinical assessment. A logical next step is implementation of the gene expression signature prospectively. The CLIA/ ISO-approvable features that have been incorporated here have been adopted with this goal in mind. The expression signature is not the only relevant development. A prospective study may also benefit from inclusion of sentinel lymph node detection procedures. 19,20 Performing such procedures on patients with a positive expression signature could further reduce the number of unnecessary neck treatments, while at the same time restricting the disadvantages of sentinel node procedures to a smaller group of higher-risk patients (Fig 3B, middle). Regardless of whether these different procedures can be combined, the results presented here indicate that a signature for predicting lymph node metastasis in OSCC, after application in a prospective study, may usefully be implemented in a clinical setting. AUTHORS DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Although all authors completed the disclosure declaration, the following author(s) and/or an author s immediate family member(s) indicated a financial or other interest that is relevant to the subject matter under consideration in this article. Certain relationships marked with a U are those for which no compensation was received; those relationships marked with a C were compensated. For a detailed description of the disclosure categories, or for more information about ASCO s conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors. Employment or Leadership Position: Paul Roepman, Agendia (C) Consultant or Advisory Role: None Stock Ownership: None Honoraria: None Research Funding: None Expert Testimony: None Other Remuneration: None AUTHOR CONTRIBUTIONS Conception and design: Paul Roepman, J. Alain Kummer, Ruud H. Brakenhoff, Piet J. Slootweg, Robert P. Takes, Frank C.P. Holstege Provision of study materials or patients: Piet J. Slootweg Collection and assembly of data: Sander R. van Hooff, Frank K.J. Leusink, Paul Roepman, Robert J. Baatenburg de Jong, Ernst-Jan M. Speel, Michiel W.M. van den Brekel, Marie-Louise F. van Velthuysen, Paul J. van Diest, Robert J.J. van Es, Matthias A.W. Merkx, J. Alain Kummer, Ed 56

59 Validation of a gene expression signature for the assessment of lymph node metastasis in oral squamous-cell carcinoma Schuuring, Johannes A. Langendijk, Martin Lacko, Maria J. De Herdt, Jeroen C. Jansen, Ruud H. Brakenhoff, Piet J. Slootweg, Robert P. Takes, Frank C.P. Holstege Data analysis and interpretation: Sander R. van Hooff, Frank K.J. Leusink, Paul Roepman, C. Rene Leemans, Piet J. Slootweg, Robert P. Takes, Frank C.P. Holstege Manuscript writing: All authors Final approval of manuscript: All authors 3 57

60 Chapter 3 REFERENCES 1. Argiris A, Karamouzis MV, Raben D, et al. Head and neck cancer. Lancet 2008; 371: Kamangar F, Dores GM, Anderson WF. Patterns of cancer incidence, mortality, and prevalence across five continents: Defining priorities to reduce cancer disparities in different geographic regions of the world. J Clin Oncol 2006; 24: Borgemeester MC, van den Brekel MW, van Tinteren H, et al: Ultrasound-guided aspiration cytology for the assessment of the clinically N0 neck: Factors influencing its accuracy. Head Neck 2008; 30: Ganly I, Patel S, Shah J. Early stage squamous cell cancer of the oral tongue: Clinicopathologic features affecting outcome. Cancer 2012; 118: Okura M, Aikawa T, Sawai NY, et al. Decision analysis and treatment threshold in a management for the N0 neck of the oral cavity carcinoma. Oral Oncol 2009; 45: Wensing BM, Merkx MA, De Wilde PC, et al. Assessment of preoperative ultrasonography of the neck and elective neck dissection in patients with oral squamous cell carcinoma. Oral Oncol 2010; 46: Rodrigo JP, Shah JP, Silver CE, et al. Management of the clinically negative neck in early-stage head and neck cancers after transoral resection. Head Neck 2010; 33: Bradley PJ, Ferlito A, Silver CE, et al. Neck treatment and shoulder morbidity: Still a challenge. Head Neck 2011; 33: van Wilgen CP, Dijkstra PU, van der Laan BF, et al. Morbidity of the neck after head and neck cancer therapy. Head Neck 2004; 26: Dünne AA, Folz BJ, Kuropkat C, et al. Extent of surgical intervention in case of N0 neck in head and neck cancer patients: An analysis of data collection of 39 hospitals. Eur Arch Otorhinolaryngol 2004; 261: Werning JW, Heard D, Pagano C, et al. Elective management of the clinically negative neck by otolaryngologists in patients with oral tongue cancer. Arch Otolaryngol Head Neck Surg 2003; 129: van t Veer LJ, Bernards R. Enabling personalized cancer medicine through analysis of gene expression patterns. Nature 2008; 452: Roepman P, Wessels LF, Kettelarij N, et al. An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas. Nat Genet 2005; 37: Roepman P, Kemmeren P, Wessels LF, et al. Multiple robust signatures for detecting lymph node metastasis in head and neck cancer. Cancer Res 2006; 66: Glas AM, Floore A, Delahaye LJ, et al. Converting a breast cancer microarray signature into a highthroughput diagnostic test. BMC Genomics, 2006; 7: National Center for Biotechnology Information: Gene expression omnibus. cgse Shi W, Kato H, Perez-Ordonez B, et al. Comparative prognostic value of HPV16 E6 mrna compared with in situ hybridization for human oropharyngeal squamous carcinoma. J Clin Oncol 2009; 27: Hafkamp HC, Speel EJ, Haesevoets A, et al. A subset of head and neck squamous cell carcinomas exhibits integration of HPV 16/18 DNA and overexpression of p16ink4a and p53 in the absence of mutations in p53 exons 5-8. Int J Cancer 2003; 107: Alkureishi LW, Ross GL, Shoaib T, et al. Sentinel node biopsy in head and neck squamous cell cancer: 5-year follow-up of a European multicenter trial. Ann Surg Oncol 2010; 17: Civantos FJ, Zitsch RP, Schuller DE, et al. Sentinel lymph node biopsy accurately stages the regional lymph nodes for T1 T2 oral squamous cell carcinomas: Results of a prospective multiinstitutional trial. J Clin Oncol 2010; 28:

61 Validation of a gene expression signature for the assessment of lymph node metastasis in oral squamous-cell carcinoma 21. Chung CH, Parker JS, Karaca G, et al. Molecular classification of head and neck squamous cell carcinomas using patterns of gene expression. Cancer Cell 2004; 5: O Donnell RK, Kupferman M, Wei SJ, et al. Gene expression signature predicts lymphatic metastasis in squamous cell carcinoma of the oral cavity. Oncogene 2005; 24: Warner GC, Reis PP, Jurisica I, et al. Molecular classification of oral cancer by cdna microarrays identifies overexpressed genes correlated with nodal metastasis. Int J Cancer 2004; 110: Bar Ad V, Chalian A. Management of clinically negative neck for the patients with head and neck squamous cell carcinomas in the modern era. Oral Oncol 2008; 44: Genden EM, Ferlito A, Silver CE, et al. Contemporary management of cancer of the oral cavity. Eur Arch Otorhinolaryngol 2010; 267: D Souza G, Kreimer AR, Viscidi R, et al. Casecontrol study of human papillomavirus and oropharyngeal cancer. N Engl J Med 2007; 356: Ang KK, Harris J, Wheeler R, et al. Human papillomavirus and survival of patients with oropharyngeal cancer. N Engl J Med 2010; 363: Chung CH, Gillison ML. Human papillomavirus in head and neck cancer: Its role in pathogenesis and clinical implications. Clin Cancer Res 2009; 15: Leemans CR, Braakhuis BJ, Brakenhoff RH. The molecular biology of head and neck cancer. Nat Rev Cancer 2011; 11: Psyrri A, Gouveris P, Vermorken JB. Human papillomavirus-related head and neck tumors: Clinical and research implication. Curr Opin Oncol 2009; 21: Song T, Bi N, Gui L, et al. Elective neck dissection or watchful waiting : Optimal management strategy for early stage N0 tongue carcinoma using decision analysis techniques. Chin Med J (Engl) 2008; 121:

62 Chapter 3 SUPPORTING INFORMATION Supporting methods Arrays The whole-genome array was based on the standard Agilent whole-genome highdensity (4 44K) array (Agilent Technologies, Santa Clara, CA), with default Agilent probes for all genes. The array (ID ) was designed in 2007 using the best (at that time) updated standard Agilent human genome probe set and the Agilent earray database and interface ( The full-genome array was custom made for Agendia (Amsterdam, the Netherlands) to also include the presence of additional probes that represented previously published signature probes in replicate. These probes were selected by mapping probe sequences from the original studies and replicating standard Agilent probes for these genes. Of the 825 previously published predictive gene probes, 732 unique probes representing 696 unique genes could be unambiguously mapped in this way and were therefore applied throughout this study. Other than the full-genome array, a dedicated diagnostic array was designed (high-density eightpack) that contained, in addition to Agilent s default control probe set, all probes necessary for a diagnostic readout of the gene signatures (5 replicate) alongside a set of normalization probes. The normalization probes were selected from the full-genome array by identifying the most stably expressed genes across all samples that were profiled for the platform transition step, covering a complete range of expression levels. These probes were also transferred to the dedicated diagnostic array. The Data Supplement contains sequence information and probe mapping information for all signature probes (for both arrays). Analysis of gene expression Unless otherwise stated, the procedures described here refer to both the full-genome (44K) and diagnostic (eight-pack) arrays. Raw fluorescence intensities were quantified using Agilent Feature Extraction software (v. 9.5) and imported into R/Bioconductor ( and Normalization was performed with the limma Bioconductor package v in R (v ) using the normalizewithinarrays function with a loess span of 0.4 and including local background subtraction. Normalization of the full-genome expression data was performed in a global fashion to include all noncontrol probes on the array. For the dedicated diagnostic arrays the normalization probe set was used. Individual spot measurements with missing or negative values were dismissed from additional analysis. Replicate spots (with a positive value) were averaged, resulting in a single value for every probe. Because of replication, all of the signatures probes could be read out. After normalization, a centering procedure was introduced to increase the dynamic range of the signature score. Given that the reference RNA is derived from cell lines, the difference in expression levels is much larger between sample and reference than between pn0 and pn+ samples. This compresses the range of the signature scores. Centering was therefore performed to alleviate this compression of values. The transformation is 60

63 Validation of a gene expression signature for the assessment of lymph node metastasis in oral squamous-cell carcinoma a modified version of mean centering, that is, an unweighted aggregate mean centering. After normalization, the expression values of all probes were centered using the unweighted aggregate mean per probe of pn+ and pn0 samples from the platform transition cohort. In other words, for every probe, the mean expression of all pn+ samples (n = 54) as well as all pn0 samples (n = 40) was calculated. Next, the mean of the pn+ and pn0 means (the unweighted aggregate mean) was calculated. The resulting unweighted aggregate mean for a particular probe was then subtracted from the individual expression values of this probe in the platform transition and the multicenter validation cohorts. The aggregate mean was not recalculated for the multicenter validation cohort, thereby avoiding the introduction of unwarranted bias. 3 Supporting tables Due to size the tables are not shown here. Supplementary tables 1, 2 & 3 can be found at: jco.ascopubs.org/content/suppl/2012/11/15/jco dc1/online_table_1.xls jco.ascopubs.org/content/suppl/2012/11/15/jco dc1/online_table_2.xls jco.ascopubs.org/content/suppl/2012/11/15/ JCO DC1/Online_table_3.xls 61

64

65 CHAPTER 4 Novel diagnostic modalities for assessment of the clinically node-negative neck in oral squamous-cell carcinoma F.K.J. Leusink R.J.J. van Es R. de Bree R.J. Baatenburg de Jong S.R. van Hooff F.C.P. Holstege P.J. Slootweg R.H. Brakenhoff R.P. Takes Lancet Oncology 2012; 13: e554-61

66 Chapter 4 ABSTRACT Oral squamous-cell carcinomas arise in mucosal linings of the oral cavity and frequently metastasize to regional lymph nodes in the neck. The presence of nodal metastases is a determinant of prognosis and clinical management. The neck is staged by palpation and imaging, but accuracy of these techniques to detect small metastases is low. In general, 30 40% of patients will have occult nodal disease and will develop clinically detectable lymphnode metastases when the neck is left untreated. The choice at present is either elective treatment or careful observation followed by treatment of the neck in patients who develop manifest metastases. These unsatisfying therapeutic options have been the subject of debate for decades. Recent developments in staging of the neck, including expression profiling and sentinel lymph-node biopsy, will allow more personalized management of the neck. 64

67 Novel diagnostic modalities for assessment of the clinically node-negative neck in oral squamous-cell carcinoma INTRODUCTION Head and neck squamous-cell carcinoma (HNSCC) is a common cancer worldwide. 1 Risk factors include tobacco smoking, betel nut chewing, excessive alcohol consumption, and human papillomavirus (HPV) infection. 2,3 Of all HNSCCs, oral squamous-cell carcinoma (SCC) has the highest incidence. Most tumors in oral-cavity subsites have a comparable high propensity to metastasize to regional lymph nodes in the neck. Findings of recent studies seem to refute the traditional belief that maxillary SCCs have a lower rate of metastasis to the neck. 4 6 The presence of regional neck metastases is a major determinant of both prognosis and treatment decisions in patients with oral SCC. 7 However, the low sensitivity of currently available diagnostic modalities is a problem, because a fairly high proportion (30 40%) of lymph node metastases are left undetected in this population. These metastases will develop into overt neck disease during follow-up. 4 A policy is in place to treat the neck even when the tumor has been classified clinically as nodenegative (cn0), indicating no disease is detectable in the neck. 8 This strategy prevents disease in the neck becoming more advanced once previously occult metastases become clinically apparent or are detected late during follow-up. Thus, 60 70% of patients receive unnecessary treatment, which in the case of neck dissection encompasses a surgical procedure potentially causing disfigurement and associated morbidity The alternative approach of watchful waiting entails careful monitoring of the neck (eg, by ultrasound-guided fine-needle aspiration cytology during follow-up). 12,13 Treatment will be given only to patients who develop manifest metastasis, which usually arises within 1 2 years. Although no evidence shows convincingly that one strategy should be preferred over the other, consensus partly based on an often cited but outdated decision analysis model suggests that elective neck treatment is indicated when the chance of occult nodal disease exceeds 20%. 14 Hence, increased accuracy to ascertain the metastatic status of the neck will result in fewer unnecessary elective treatments of the neck. Currently, the neck is staged by palpation and different imaging techniques, including CT, MRI, PET/CT, and ultrasound, which are more accurate than palpation alone Morphological and size criteria are the main determinants of specificity of imaging techniques, whereas sensitivity is limited by the detection threshold. Up to a third of nodal metastases in patients with oral SCC are smaller than the 3 mm detection threshold that limits sensitivity of available imaging techniques, 18 and occult neck disease therefore remains a relevant issue. In a meta-analysis of PET/CT for assessment of cervical lymph-node metastasis, 16 sensitivity of PET/CT was only 50% for patients who were node-negative on palpation, whereas specificity was 87%. The performance of ultrasound, MRI, and CT was equally disappointing. Later research has shown similar results

68 Chapter 4 The limitation of imaging techniques to detect small metastatic deposits has led to a search for additional characteristics or biomarkers assessable on the primary tumor to predict nodal disease. Histopathological or molecular features of the primary tumor can predict the presence of nodal metastases in the individual patient, irrespective of the actual size of the tumour. 20 Measures such as perineural invasion, vascular invasion, tumor border (infiltrative or pushing), tumor thickness, and depth of invasion have been studied extensively for correlations. Only tumor thickness is associated consistently with the presence of nodal metastases. 21 However, suggested cutoff s for thickness range from 1 5 mm to 10 mm, and consensus to recommend elective neck dissection is missing One reason could be that definitions and methods to measure tumor thickness vary in reports. Moreover, tumor thickness is usually ascertained on histopathological examination of the resection specimen, implying a second-stage surgical procedure when elective neck dissection is indicated. Measurement of tumor thick ness before surgery, using an intraoral ultrasound probe, could be more promising in this respect. 24 Besides histological variables, many single molecular markers have been studied to predict the presence of nodal metastasis. In view of the complexity of the metastatic process, no consistent and clinically useful correlations have been noted for any markers. 20 Thus, the dilemma in current clinical management of the neck is the choice between possible undertreatment of 30 40% of patients with occult metastases and overtreatment of the remaining 60 70%. Personalized management of the clinically node-negative (cn0) neck, especially in patients with oral SCCs, would benefit greatly from staging techniques that add accuracy to assessment of nodal disease, particularly when these methods are not or are minimally dependent on size of the metastasis. Methods are being developed to diagnose or predict occult metastases in the neck. In this Personal View, we restrict ourselves to discussion of the two most promising techniques, which are arguably ready for clinical implementation and have a very different but complementary nature: gene-expression profiling and sentinel lymph-node biopsy. We describe various aspects of these two approaches and suggest a new staging algorithm to incorporate both methods, to optimize management of the cn0 neck in patients with early-stage oral SCC. SEARCH STRATEGY AND SELECTION CRITERIA We searched PubMed for studies published in the English language from 2000, until June, 2012, with the terms: head and neck cancer, head and neck squamous cell carcinomas, oral cavity squamous cell carcinomas, lymph node metastasis, occult lymph node metastasis, 66

69 Novel diagnostic modalities for assessment of the clinically node-negative neck in oral squamous-cell carcinoma staging techniques, sentinel lymph node, molecular biology, predictive markers, expression profiling, and management of the N0 neck. One article published before 2000 was used as a historical reference. Selection was based on novelty and relevance to the scope of this Personal View. 4 Figure 1 Sentinel lymph-node procedure (A) Lymphoscintigram showing injection site (large white area) and sentinel lymph node (small white area). (B) Gamma probe-guided sentinel lymph-node biopsy. (C) A micrometastasis (red colour) is depicted by immunohistochemistry using antibodies against cytokeratin. 67

70 Chapter 4 MOLECULAR DIAGNOSIS AND TUMOR PROFILING Gene-expression profiling with DNA microarrays and next generation sequencing approaches (RNAseq) signal a new era of tumor classification and prognostication. RNA isolated from the tumor specimen can be used to ascertain expression levels of all genes simultaneously in one experiment. This process has led to novel classifications of lymphomas, breast cancer, and HNSCC Furthermore, prognostic profiles have been identified in many tumor types, including breast cancer and HNSCC. 28,29 Roepman and colleagues 30,31 used gene-expression profiling to stage the clinically N0 neck in HNSCC patients and, using RNA from the primary tumor specimen, identified particular profiles that could predict N-stage. In a multicenter validation study, undertaken at all head and neck oncological centers in the Netherlands, 32 the expression profile to predict lymph-node metastasis was transferred to a diagnostic platform to facilitate clinical implementation. Subsequently, the profile was validated with an independent series of 222 samples of oral and oropharyngeal SCC. Although the array platform was changed, the profile predicted N-stage as expected. For the group of ct1 T2N0 or early-stage oral SCCs (n=101), a negative predictive value of 89% was recorded (table 1). For these cases the issue of elective neck treatment is most relevant, because early stage oral SCC is treated by transoral surgery and, thus, there is no need to enter the neck for excision of the primary tumor. Some cancers in the validation series were HPV-positive, and the profile worked less well with HPV-positive tumours. 32 However, HPV occurs rarely in tumors of the oral cavity, with a prevalence estimated at less than 5%. 3,34 Hence, in patients with early-stage (ct1 T2N0) oral SCC, gene-expression profiling might reduce greatly the number of unnecessary elective neck dissections, allowing more personalized treatment. Improvements can still be made, however. N-status was ascertained by routine histopathological examination as reference standard, whereas detection of micrometastases (0 2 2 mm) can be increased by stepped serial sectioning and immunohistochemistry of all lymph nodes. Such extensive immunohistopathological examination allows detection of 5 58% (mean 20%) of metastases 35 that escape routine histological detection. Recently, we showed that the predictive power of the profiling approach rises further when histopathological assessment of all lymph nodes is scrutinized (unpublished data). Another improvement of the gene-expression profiling approach might be noted with analysis of multiple biopsy samples. Findings of next-generation sequencing studies indicate that intratumor heterogeneity is present in several cancer types, and this heterogeneity can sometimes be linked to particular areas of the tumor specimen Intratumor genetic heterogeneity might not necessarily be reflected in global gene-expression profiles, but further studies are needed because the predictive power of gene-expression profiles in early-stage oral SCC could be affected by genetic heterogeneity. 68

71 Novel diagnostic modalities for assessment of the clinically node-negative neck in oral squamous-cell carcinoma Table 1 Patient characteristics and performance metrics of two compared studies *, ** Characteristics GEP study (Van Hooff) ct1-2n0 (N=101) SNB study (Alkureishi) ct1-2n0 (N=134) P Value Tumours included from Follow-up yr > 5 > 5 Participating centres no. 8 6 Age yr±sd NA Location Primary Tumor no. (%) oral cavity 101 (100) 122 (91) oropharynx 0 (0) 12 (9) Sub-location Primary Tumor no. (%) Floor of mouth 22 (22) 42 (31) Anterior tongue 58 (57) 50 (37) Other sites 21 (21) 42 (31) Clinical T stage no. (%) <0.001 T1 24 (24) 75 (56) T2 77 (76) 59 (44) Pathological N status no. (%) 0.32 positive 28 (28) 46 (34) negative 73 (72) 88 (65) Performance metrics NPV % (tn/tn+fn) 89 (33/36) 95 (88/92) Sensitivity % (tp/tp+fn) 86 (24/28) 91 (42/46) * Percentages may not total 100 because of rounding. ** NPV denotes negative predictive value, tp true positive, tn true negative, and fn false negative. The positive predictive value and the specificity are dependent on the number of false positive samples. Since SNB assessment precludes the presence of false positives, the positive predictive value and the specificity are left out of this comparison. 4 Encouragingly, the highest predictive power for gene-expression profiling was seen in the clinically relevant group of ct1 T2N0 oral SCCs. Although this association could be related to the lower prevalence of nodal metastasis in this group, it seems to also make sense from a biological perspective. Early lymphatic dissemination of tumors classified as T1 and T2 might be the direct result of intrinsic tumor properties reflected in specific gene-expression profiles, whereas in advanced cancers, destruction of anatomical barriers and invasive growth is likely to become an important factor associated with lymphatic metastasis as well. Gene-expression profiling is best undertaken on fresh or frozen tumor samples; arguably, this strategy might restrict applicability, because biopsy samples or surgical specimens are routinely formalin-fixed and paraffin-embedded (FFPE). However, taking an additional biopsy sample for gene-expression profiling would hardly impose a substantial burden. Moreover, recent developments to extract RNA from FFPE specimens for gene-expression profiling could be helpful. 29,40 An alternative option would be to ascertain genome-wide genetic changes in DNA, using a comparative genomic hybridization microarray or next-generation sequencing platform. 69

72 Chapter 4 Switching from RNA to DNA profiling is not an unrealistic idea; the breast cancer classification ascertained by RNA expression profiling has been reproduced accurately by comparative genomic hybridization microarray DNA profiling, 41 suggesting that differences in gene-expression profiles between tumors might be reflected well in the number and type of DNA changes. Whether or not this idea also holds true for staging of the cn0 neck in patients with oral SCC remains to be established. Gene-expression profiles of tumor biopsy samples assessed by microarray hybridization have proven their potential value for N-staging of oral SCC. However, although DNA or protein profiles might be applied likewise, N-stage predictive profiles have not been published, to the best of our knowledge, let alone validated in prospective multicenter studies. SENTINEL LYMPH-NODE BIOPSY In an attempt to more reliably select lymph nodes that potentially contain metastases, sentinel lymph-node assessment, which is used extensively in melanoma and breast cancer, has been introduced. The sentinel lymph node is likely to be the first lymph node to harbor metastasis and can be used to provide information on the rest of the nodal basin. It is usually identified by peritumoral injection of radioactive colloid and blue dye. Preoperative lymphoscintigraphy (figure 1A), intra-operative visualization of blue coloration, and intra operative radionuclide detection with a gamma probe (figure 1B) allow identification of the sentinel lymph node. After surgical removal, this node is studied meticulously by histopathological examination, using stepped serial sectioning and immunohistochemistry (figure 1C). If the sentinel lymph node contains metastatic tumor cells, treatment of the neck is recommended, usually in a second procedure. 42 The sentinel lymph-node procedure is deemed more precise than imaging procedures and less invasive than elective neck dissection. Moreover, it is associated with significantly less postoperative morbidity and better shoulder function compared with elective neck dissection. 43 Current bestpractice guidelines for provision of sentinel lymph-node biopsy to patients with early-stage oral SCC have been outlined, which provide a framework for the currently evolving recommendations for its use. 42 To safely avoid elective treatment of the neck in as many patients as possible, a high sensitivity and high negative predictive value are both needed. In the American College of Surgeons Oncology Group Z0360 validation study of 140 patients in 25 institutions, sensitivity was 90% and the negative predictive value was 96%; these figures were even better for skilled surgeons. 44 However, in this study, standard histopathological examination of the neck-dissection specimen 70

73 Novel diagnostic modalities for assessment of the clinically node-negative neck in oral squamous-cell carcinoma was used as the gold standard. Therefore, occult micrometastasis might have been missed, contributing to higher figures for sensitivity and negative predictive value. In a meta-analysis of 19 pilot studies with a total of 347 oral and oropharyngeal cancer patients, a pooled sensitivity of 92 6% was reported. 45 After initial studies to validate the sentinel lymph-node approach in patients with early-stage oral SCC, several prospective observational studies have been reported. In two large single-center studies, sensitivities and negative predictive values of at least 90% were noted.46,47 In these studies, neck dissection was undertaken only when the sentinel lymph node contained metastasis, and a watchful-waiting strategy was followed when the sentinel lymph node was tumor-free. In a European multicenter study 33 of 134 patients with ct1/2n0 oral SCC (table 1), 79 patients underwent sentinel lymph-node biopsy as the sole staging method, whereas 55 underwent sentinel lymph-node biopsy followed by elective neck dissection. For the two groups together, using a reference standard of 5 years follow-up after sentinel lymph-node biopsy staging, a sensitivity of 91% and a negative predictive value of 95% were recorded. The better performance seen in patients who underwent both techniques (sensitivity 96%, negative predictive value 97%) compared with those who only had sentinel lymph-node biopsy (87%, 94%) can again be accounted for by use of standard histopathological examination of the neck dissection specimen versus 5 years follow-up as a gold standard for metastasis. 4 Of note, both the sentinel lymph-node identification rate and sensitivity were significantly poorer in patients with floor-of-mouth tumors. Peritumoral injection sites might overshine the sentinel lymph node in these patients, resulting in detection failure by the gamma probes. 33 The observational multicenter European Sentinel Node Trial (SENT), 48 with more than 300 patients enrolled, has completed accrual and long-term follow-up data are awaited. Overall, detection rates, sensitivity, and negative predictive values of sentinel lymph-node biopsy procedures are generally above 90%. Some innovations would improve preoperative imaging of the sentinel lymph node. Hybrid singlephoton emission CT with integrated CT (SPECT/CT) can augment visualization of the relation of sentinel lymph nodes to several vital vascular and neural structures, thus enabling safer removal of these nodes. Furthermore, enhanced topographical orientation and delineation of sentinel lymph nodes against surrounding structures might also reduce surgical time. Although SPECT/CT has the potential to detect more sentinel lymph nodes, it still has some diffculties in visualization of nodes in close spatial relation to the injection site. 49 Promising preclinical results have been reported of 89 Zr-nanocolloidal albumin-based PET/CT lymphoscintigraphy for sentinel lymphnode detection in HNSCC.50 Technical innovations to improve intraoperative localization of sentinel lymph nodes include intraoperative real-time imaging, freehand SPECT, and fluorescence 71

74 Chapter 4 imaging. Intraoperative real-time imaging with the portable gamma camera provides an overview of all radioactive spots and can show sentinel lymph nodes near the injection site by adjustment of its position. Another advantage might be the certainty it can provide about the completeness and accuracy of sentinel lymph node excision, by showing the remaining activity. 51 Freehand SPECT is designed to ascertain the position of the detector relative to the patient, through which three dimensional images are generated. These images provide the surgeon with information about the direction and depth of the sentinel lymph node in relation to the probe. 52 The feasibility of near-infrared fluorescence guided sentinel lymph-node detection has been shown in HNSCC, with indocyanine green as fluorescent tracer. 53 Other tracers with enhanced optical properties have been tested in HNSCC in preclinical settings. 50,54 The real clinical additional value of these techniques has still to be assessed. Developments in detection of metastasis in the sentinel lymph-node specimen have been made. While the biopsy procedure is taking place, immediate frozen section, imprint cytology, or molecular techniques can be done that should reliably show sentinel lymph-node metastasis. 55 Furthermore, a rapid automated quantitative reverse transcriptase PCR assay has been described that took about 35 min to complete and had an accuracy of 94 2% for identification of positive and negative nodes, which might be more accurate than intraoperative pathological analysis. 56 Thus, the surgical procedure including neck dissection in case of a positive test for nodal metastasis could be restricted to one session instead of the two currently needed in sentinelnode procedures. However, this change could interfere with current surgical planning efficiency, because complete neck dissection takes about five times longer to undertake than does sentinel lymph-node biopsy of the neck. These technical developments could raise the rate of sentinel lymph-node detection further and facilitate the procedure, potentially resulting in less morbidity for the patient, reduced operating time, and increased convenience for the head and neck surgeon. TOWARDS A NEW DIAGNOSTIC STRATEGY To compare molecular diagnosis and sentinel lymph node biopsy for staging of the cn0 neck, we selected two studies (table 1). We chose the study by Van Hooff and colleagues 32 because it is the only multicenter gene-expression profiling study available. The sentinel lymph node biopsy study by Alkureishi and colleagues 33 was selected because it is also a multicenter study. Weighing the list of advantages and disadvantages of both methods (table 2), gene-expression profiling might be judged more favorable. Although sentinel lymph-node biopsy has a good negative predictive value, it remains an invasive surgical procedure with (albeit limited) chances 72

75 Novel diagnostic modalities for assessment of the clinically node-negative neck in oral squamous-cell carcinoma of associated morbidity, needs injection of radioactive tracers, and has logistical challenges such as additional second-stage surgery in case of detected metastasis. Would a combination of approaches have additional value? Figure 2A outlines our proposed modification to the management strategy for patients with early-stage (ct1 T2) oral SCC. If the neck is staged cn0 by current imaging techniques, use of gene-expression profiling is indicated, with subsequent sentinel lymph-node biopsy if the signature classification is positive (cn+[gep]). A tumor with a negative gene-expression profile would indicate low risk for metastasis and be classified as cn0(gep). Since occult nodal disease is unlikely, this group could be followed up by watchful waiting eg, surveillance of the neck by ultrasound-guided fine-needle aspiration cytology every 3 4 months during the first year of follow-up. The high-risk group, classified as cn+(gep), would undergo sentinel lymph-node biopsy followed by either watchful waiting in case of pn0(sn) or neck dissection in case of pn+(sn). 42 The main disadvantage of current management is overtreatment, because of the policy to treat the neck in all patients. On the basis of data from the multicenter study by Van Hooff and colleagues (n=101), 32 72% of patients would be overtreated (table 1, figure 2B). 4 Table 2 Advantages and disadvantages of gene expression profiling and sentinel lymph node biopsy* Gene expression profiling Sentinel lymph node biopsy Advantages non-invasive when combined with routine biopsy no morbidity NPV (89%) no general anaesthesia no hospitalization Disadvantages at present fresh or frozen specimen required costs minimally invasive minimal morbidity NPV 90-95% additional 2 nd stage surgery in case of pn+(sn) hospitalization injection of radioactive tracers labour intensive * NPV denotes negative predictive value, PPV positive predictive value and pn+(sn) pathological positive lymph node by sentinel lymph node analysis. With our proposed strategy, the number of unnecessary treatments of the neck could be eliminated. However, 11% of false-negative results after gene-expression profiling imply that four of 36 patients classified as cn0(gep) are undertreated (figure 2B). Furthermore, a negative predictive value of 95% for the sentinel lymph node procedure indicates that two of 43 patients classified as pn0(sn) are also undertreated. Taken together, 6% of patients could be undertreated with our strategy and will need to undergo second-stage treatment. 73

76 Chapter 4 A Patients with early stage (ct1 T2N0) tumour in the oral cavity B Current Appropriate treatment Mostly appropriate treatment Undertreatment Overtreatment Gene-expression profiling pn+ 80 Classified cn0 (gep) Classified cn+ (gep) Sentinel lymph-node biopsy Number of patients pn0 pn+ Proposed model: GEP with or without SLNB pn+ pn0 pn0(sn) pn+(sn) 20 pn0 pn+ Watchful waiting Watchful waiting (Modified) radical neck dissection 0 Selective neck dissection I III Watchful waiting (GEP) Watchful waiting (GEP plus SLNB) (Modified) radical Figure 2 Proposed diagnostic algorithm incorporating gene-signature prediction and sentinel lymph-node biopsy (A) Clinical decision model proposed for early-stage (ct1 T2N0) oral squamous-cell carcinoma patients. In case of a negative classifi cation (cn0[gep]) based on the gene signature, watchful waiting is proposed. For patients with a positive classification (cn+[gep]), sentinel-node biopsy is proposed. In case of a negative sentinel node (pn0[sn]), watchful waiting is proposed, whereas with a positive sentinel node (pn+[sn]), usually a (modifi ed) radical neck dissection is recommended. (B) Currently, all early-stage patients are treated with selective neck dissection (level I III), causing overtreatment in 72% of patients (red). The remaining 28% of patients receive mostly appropriate treatment (light green). In the proposed model with gene-signature readout and sentinel lymph-node biopsy, overtreatment by elective neck dissection does not exist. 94% of patients now receive the most appropriate treatment (dark green) and only 6% (yellow) will undergo second-stage treatment (neck dissection, of which some will also need postoperative irradiation). Numbers of patients for every outcome of this strategy are based on patients enrolled in the gene-expression profi ling validation study by van Hooff and colleagues (table 1).32 Because study populations are not entirely comparable, the numbers shown should be regarded as estimates. GEP=gene-expression profiling SLNB=sentinel lymph-node biopsy. CONSIDERATIONS AND FUTURE PERSPECTIVES Oral SCC is predominantly treated surgically, 13 therefore, the issue of elective neck dissection versus watchful waiting is most relevant for early-stage (ct1 T2N0) tumours. 57,58 In patients with advanced tumors (ct3 T4), who have a fairly high probability of cervical lymph-node metastases and often need neck surgery to access the primary tumor or to reconstruct the surgical defect, most head and neck surgeons will opt for elective neck dissection anyway. 57,58 Our proposed strategy does not intend to change this policy. Morbidity is a relevant issue with current management of early-stage (ct1 T2N0) oral SCC. By applying our proposed algorithm, oncological safety is not so much pursued by over treatment and its associated unnecessary morbidity but rather by further reduction of the rate of occult metastasis and accurate follow-up. Timely treatment is still possible for patients who are under staged

77 Novel diagnostic modalities for assessment of the clinically node-negative neck in oral squamous-cell carcinoma Decision-analysis techniques recommend watchful waiting below risk thresholds that vary from 17% to 44%; 14,60,61 thus, we judge the negative predictive values of gene-expression profiling and sentinel lymph-node biopsy acceptable because such thresholds are much higher than the 6% risk of occult metastases projected by our proposed algorithm. Although a significant survival difference has not been reported between stringent watchful waiting without neck dissection and elective neck dissection, 59,62,63 patients who develop delayed lymph-node metastases can be diagnosed with more advanced disease in the neck, requiring extensive neck dissection or adjuvant (chemo)radiotherapy. With adoption of a watchful-waiting policy, 30 40% of patients will need neck dissection, and under our proposed algorithm this proportion would fall to 6%. 4 In the end, how to weigh the expected 6% of patients needing second-stage (and maybe intensified) treatment against the inevitable morbidity of 60 70% of patients who undergo unnecessary treatment of their neck will remain a matter of subjective judgment. With respect to cost-effectiveness of our proposed algorithm, the price of gene-expression profiles, sentinel lymph-node biopsy procedures, neck dissections, and ultrasound-guided fine-needle aspiration cytology procedures should be weighed against expenses and associated morbidity of elective neck dissections that would take place with current management. Similar to breast cancer, costeffectiveness depends heavily on costs of operation time, hospital stay, 64 rehabilitation programs, and the expected length of sick leave associated with the procedure, which will be reduced substantially with our proposed regimen. Future price developments and cost-comparison studies are awaited to quantify cost precisely. The increase in different prediction profiles and staging modalities will alter future clinical decision making. Categorization of patients into a few prognostic groups such as TNM-derived stages will not be sufficient to further improve personalized cancer treatment. 65 Several variables will have to be integrated into multivariate risk prediction models, using different statistical approaches. 66,67 Another challenge for doctors treating oral SCC will be to communicate these individualized treatment options to the patient. CONCLUSION Currently, physical examination and imaging techniques are the most widely accepted and applied methods for assessment of the neck in oral SCC. Limitations in accuracy of these modalities have not led to altered management of the cn0 neck in early-stage disease. 75

78 Chapter 4 Combining gene-expression profiling and sentinel lymph-node biopsy in addition to current imaging techniques will further reduce the rate of occult metastasis. In view of the accuracy of these techniques, this rate can be reduced to a level acceptable to allow a wait-and-see policy for the neck in patients with oral SCC classified as T1 and T2. Furthermore, restriction of sentinel lymph-node biopsy to individuals who are classified as nodepositive on gene-expression profiling eliminates overtreatment. Prospective clinical trials implementing this staging algorithm are needed to investigate whether gene-expression profiling and sentinel lymph-node biopsy can be combined and whether oncological and functional outcomes in patients with oral SCC will indeed improve. CONTRIBUTORS FKJL, RHB, and RPT designed the report. The literature search was undertaken by FKJL, RHB, RPT, RdB, and RJJvE. Figure 1 was made by RdB and figure 2 by FKJL and SRvH. All authors contributed to interpretation of data and wrote and revised the report. CONFLICT OF INTEREST PJS and FCPH are co-inventors of a patented method (patent owned by University Medical Centre Utrecht) of gene-expression profiling, designed to improve diagnosis of metastases in head and neck squamous-cell carcinoma. No commercial application has been realized to date. All other authors declare that they have no conflicts of interest. ACKNOWLEDGMENTS We thank the Dutch Cancer Society and the Dutch Head and Neck Society for research support. 76

79 Novel diagnostic modalities for assessment of the clinically node-negative neck in oral squamous-cell carcinoma REFERENCES 1. Jemal A, Bray F, Center MM, et al. Global cancer statistics. CA Cancer J Clin 2011; 61: Ang KK, Harris J, Wheeler R, et al. Human papillomavirus and survival of patients with oropharyngeal cancer. N Engl J Med 2010; 363: Leemans CR, Braakhuis BJ, Brakenhoff RH. The molecular biology of head and neck cancer. Nat Rev Cancer 2011; 11: Morris LG, Patel SG, Shah JP, et al. High rates of regional failure in squamous cell carcinoma of the hard palate and maxillary alveolus. Head Neck 2011; 33: Montes DM, Carlson ER, Fernandes R, et al. Oral maxillary squamous carcinoma: an indication for neck dissection in the clinically negative neck. Head Neck 2011; 33: Brown JS, Bekiroglu F, Shaw RJ, et al. Management of the neck and regional recurrence in squamous cell carcinoma of the maxillary alveolus and hard palate compared with other sites in the oral cavity. Head Neck 2012; published online Feb 6. DOI: /hed Argiris A, Karamouzis MV, Raben D, et al. Head and neck cancer. Lancet 2008; 371: Ferlito A, Silver CE, Rinaldo A. Elective management of the neck in oral cavity squamous carcinoma: current concepts supported by prospective studies. Br J Oral Maxillofac Surg 2009; 47: van Wilgen CP, Dijkstra PU, van der Laan BF, et al. Morbidity of the neck after head and neck cancer therapy. Head Neck 2004; 26: van Wouwe M, de Bree R, Kuik DJ, et al. Shoulder morbidity after non-surgical treatment of the neck. Radiother Oncol 2009; 90: Bradley PJ, Ferlito A, Silver CE, et al. Neck treatment and shoulder morbidity: still a challenge. Head Neck 2011; 33: Nieuwenhuis EJ, Castelijns JA, Pijpers R, et al. Wait-and-see policy for the N0 neck in early-stage oral and oropharyngeal squamous cell carcinoma using ultrasonography-guided cytology: is there a role for identification of the sentinel node? Head Neck 2002; 24: Rodrigo JP, Shah JP, Silver CE, et al. Management of the clinically negative neck in early-stage head and neck cancers after transoral resection. Head Neck 2011; 33: Weiss MH, Harrison LB, Isaacs RS. Use of decision analysis in planning a management strategy for the stage N0 neck. Arch Otolaryngol Head Neck Surg 1994; 120: de Bondt RB, Nelemans PJ, Hofman PA, et al. Detection of lymph node metastases in head and neck cancer: a meta-analysis comparing US, USgFNAC, CT and MR imaging. Eur J Radiol 2007; 64: Kyzas PA, Evangelou E, Denaxa-Kyza D, et al. 18F-fl uorodeoxyglucose positron emission tomography to evaluate cervical node metastases in patients with head and neck squamous cell carcinoma: a metaanalysis. J Natl Cancer Inst 2008; 100: Ng SH, Yen TC, Chang JT, et al. Prospective study of [18F] fl uorodeoxyglucose positron emission tomography and computed tomography and magnetic resonance imaging in oral cavity squamous cell carcinoma with palpably negative neck. J Clin Oncol 2006; 24: Buckley JG, MacLennan K. Cervical node metastases in laryngeal and hypopharyngeal cancer: a prospective analysis of prevalence and distribution. Head Neck 2000; 22: Liao LJ, Lo WC, Hsu WL, et al. Detection of cervical lymph node metastasis in head and neck cancer patients with clinically N0 neck-a meta-analysis comparing different imaging modalities. BMC Cancer 2012; 12: Takes RP, Rinaldo A, Rodrigo JP, et al. Can biomarkers play a role in the decision about treatment of the clinically negative neck in patients with head and neck cancer? Head Neck 2008; 30: Melchers LJ, Schuuring E, van Dijk BA, et al. Tumour infi ltration depth >/=4 mm is an indication for an elective neck dissection in pt1cn0 oral squamous cell carcinoma. Oral Oncol 2012; 48:

80 Chapter Pentenero M, Gandolfo S, Carrozzo M. Importance of tumor thickness and depth of invasion in nodal involvement and prognosis of oral squamous cell carcinoma: a review of the literature. Head Neck 2005; 27: Huang SH, Hwang D, Lockwood G, et al. Predictive value of tumor thickness for cervical lymph-node involvement in squamous cell carcinoma of the oral cavity: a meta-analysis of reported studies. Cancer 2009; 115: Lodder WL, Teertstra HJ, Tan IB, et al. Tumour thickness in oral cancer using an intra-oral ultrasound probe. Eur Radiol 2011; 21: Alizadeh AA, Eisen MB, Davis RE, et al. Distinct types of diffuse large B-cell lymphoma identifi ed by gene expression profi ling. Nature 2000; 403: Perou CM, Sorlie T, Eisen MB, et al. Molecular portraits of human breast tumours. Nature 2000; 406: Chung CH, Parker JS, Karaca G, et al. Molecular classifi cation of head and neck squamous cell carcinomas using patterns of gene expression. Cancer Cell 2004; 5: van t Veer LJ, Dai H, van de Vijver MJ, et al. Gene expression profi ling predicts clinical outcome of breast cancer. Nature 2002; 415: Chung CH, Parker JS, Ely K, et al. Gene expression profi les identify epithelial-to-mesenchymal transition and activation of nuclear factor-κb signaling as characteristics of a high-risk head and neck squamous cell carcinoma. Cancer Res 2006; 66: Roepman P, Wessels LF, Kettelarij N, et al. An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas. Nat Genet 2005; 37: Roepman P, Kemmeren P, Wessels LF, et al. Multiple robust signatures for detecting lymph node metastasis in head and neck cancer. Cancer Res 2006; 66: van Hooff SR, Leusink FK, Roepman P, et al. Validation of a gene expression signature for assessment of lymph node metastasis in oral squamous cell carcinoma. J Clin Oncol 2012; published online Oct 8. DOI: /JCO Alkureishi LW, Ross GL, Shoaib T, et al. Sentinel node biopsy in head and neck squamous cell cancer: 5-year follow-up of a European multicenter trial. Ann Surg Oncol 2010; 17: Herrero R, Castellsague X, Pawlita M, et al. Human papillomavirus and oral cancer: the International Agency for Research on Cancer multicenter study. J Natl Cancer Inst 2003; 95: Ferlito A, Rinaldo A, Devaney KO, et al. Detection of lymph node micrometastases in patients with squamous carcinoma of the head and neck. Eur Arch Otorhinolaryngol 2008; 265: Yachida S, Jones S, Bozic I, et al. Distant metastasis occurs late during the genetic evolution of pancreatic cancer. Nature 2010; 467: Gerlinger M, Rowan AJ, Horswell S, et al. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N Engl J Med 2012; 366: Nik-Zainal S, Alexandrov LB, Wedge DC, et al. Mutational processes molding the genomes of 21 breast cancers. Cell 2012; 149: Nik-Zainal S, Van Loo P, Wedge DC, et al. The life history of 21 breast cancers. Cell 2012; 149: Xie Y, Xiao G, Coombes KR, et al. Robust gene expression signature from formalin-fixed paraffinembedded samples predicts prognosis of non-small-cell lung cancer patients. Clin Cancer Res 2011; 17: Smeets SJ, Harjes U, van Wieringen WN, et al. To DNA or not to DNA? That is the question, when it comes to molecular subtyping for the clinic! Clin Cancer Res 2011; 17: Alkureishi LW, Burak Z, Alvarez JA, et al. Joint practice guidelines for radionuclide lymphoscintigraphy for sentinel node localization in oral/oropharyngeal squamous cell carcinoma. Ann Surg Oncol 2009; 16:

81 Novel diagnostic modalities for assessment of the clinically node-negative neck in oral squamous-cell carcinoma 43. Murer K, Huber GF, Haile SR, et al. Comparison of morbidity between sentinel node biopsy and elective neck dissection for treatment of the n0 neck in patients with oral squamous cell carcinoma. Head Neck 2011; 33: Civantos FJ, Zitsch RP, Schuller DE, et al. Sentinel lymph node biopsy accurately stages the regional lymph nodes for T1 T2 oral squamous cell carcinomas: results of a prospective multi-institutional trial. J Clin Oncol 2010; 28: Paleri V, Rees G, Arullendran P, et al. Sentinel node biopsy in squamous cell cancer of the oral cavity and oral pharynx: a diagnostic meta-analysis. Head Neck 2005; 27: Kovacs AF, Stefenelli U, Seitz O, et al. Positive sentinel lymph nodes are a negative prognostic factor for survival in T1-2 oral/ oropharyngeal cancer-a long-term study on 103 patients. Ann Surg Oncol 2009; 16: Broglie MA, Haile SR, Stoeckli SJ. Long-term experience in sentinel node biopsy for early oral and oropharyngeal squamous cell carcinoma. Ann Surg Oncol 2011; 18: Gurney BAS, Schilling C, Putcha V, et al. Implications of a positive sentinel node in oral squamous cell carcinoma. Head Neck 2012; published online Jan 31. DOI: /hed Haerle SK, Hany TF, Strobel K, et al. Is there an additional value of SPECT/CT over planar lymphoscintigraphy for sentinel node mapping in oral/oropharyngeal squamous cell carcinoma? Ann Surg Oncol 2009; 16: Heuveling DA, Visser GW, Baclayon M, et al. 89 Zr-nanocolloidal albumin-based PET/CT lymphoscintigraphy for sentinel node detection in head and neck cancer: preclinical results. J Nucl Med 2011; 52: Vermeeren L, Valdes Olmos RA, Klop WM, et al. A portable gamma-camera for intraoperative detection of sentinel nodes in the head and neck region. J Nucl Med 2010; 51: Heuveling DA, Karagozoglu KH, van Schie A, et al. Sentinel node biopsy using 3D lymphatic mapping by freehand SPECT in early stage oral cancer: a new technique. Clin Otolaryngol 2012; 37: van den Berg NS, Brouwer OR, Klop WM, et al. Concomitant radio- and fluorescence-guided sentinel lymph node biopsy in squamous cell carcinoma of the oral cavity using ICG-(99m)Tc-nanocolloid. Eur J Nucl Med Mol Imaging 2012; 39: Keereweer S, Hutteman M, Kerrebijn JD, et al. Translational optical imaging in diagnosis and treatment of cancer. Curr Pharm Biotechnol 2012; 13: Vorburger MS, Broglie MA, Soltermann A, et al. Validity of frozen section in sentinel lymph node biopsy for the staging in oral and oropharyngeal squamous cell carcinoma. J Surg Oncol 2012; published online May 14. DOI: /jso Ferris RL, Xi L, Seethala RR, et al. Intraoperative qrt-pcr for detection of lymph node metastasis in head and neck cancer. Clin Cancer Res 2011; 17: Bar Ad V, Chalian A. Management of clinically negative neck for the patients with head and neck squamous cell carcinomas in the modern era. Oral Oncol 2008; 44: Genden EM, Ferlito A, Silver CE, et al. Contemporary management of cancer of the oral cavity. Eur Arch Otorhinolaryngol 2010; 267: Yuen AP, Ho CM, Chow TL, et al. Prospective randomized study of selective neck dissection versus observation for N0 neck of early tongue carcinoma. Head Neck 2009; 31: Song T, Bi N, Gui L, et al. Elective neck dissection or watchful waiting : optimal management strategy for early stage N0 tongue carcinoma using decision analysis techniques. Chin Med J (Engl) 2008; 121: Okura M, Aikawa T, Sawai NY, et al. Decision analysis and treatment threshold in a management for the N0 neck of the oral cavity carcinoma. Oral Oncol 2009; 45: Keski-Santti H, Atula T, Tornwall J, et al. Elective neck treatment versus observation in patients with T1/ T2 N0 squamous cell carcinoma of oral tongue. Oral Oncol 2006; 42:

82 Chapter Tsang RK, Chung JC, Howe To VS, et al. Efficacy of salvage neck dissection for isolated nodal recurrences in early carcinoma of oral tongue with watchful waiting management of initial N0 neck. Head Neck 2011; 33: Classe JM, Baff ert S, Sigal-Zafrani B, et al. Cost comparison of axillary sentinel lymph node detection and axillary lymphadenectomy in early breast cancer: a national study based on a prospective multiinstitutional series of 985 patients on behalf of the Group of Surgeons from the French Unicancer Federation. Ann Oncol 2012; 23: Takes RP, Rinaldo A, Silver CE, et al. Future of the TNM classification and staging system in head and neck cancer. Head Neck 2010; 32: Gerds TA, Cai T, Schumacher M. The performance of risk prediction models. Biom J 2008; 50: Stadler WM. Prognosis and prediction in a Facebook world. Cancer 2009; 115:

83 Novel diagnostic modalities for assessment of the clinically node-negative neck in oral squamous-cell carcinoma 4 81

84

85 CHAPTER 5 Tumor biological determinants of locoregional recurrence of non-hpv head and neck squamous cell carcinoma F.K.J. Leusink* B.J.M. Braakhuis* W.N. van Wieringen E.Bloemena F. Rustenburg B. Ylstra J.A. Kummer P.J. van Diest Q.J. Voorham S.J. Smeets P. Roepman R. Koole C.R. Leemans R.H. Brakenhoff * Equal contribution Submitted

86 Chapter 5 ABSTRACT INTRODUCTION Head and neck squamous cell carcinoma (HNSCC) has still a fairly bad prognosis with more than 50% of patients dying within five years after diagnosis. A considerable part of these deaths is caused by locoregionally recurrent cancer. Aim of this study was to identify tumor characteristics that are associated with the development of locoregional recurrences and a poor prognosis and to reveal their biological basis. METHODS Detailed histopathological tumor characteristics were obtained from 170 SCC of the oral cavity and oropharynx and these features were linked to locoregional recurrence-free survival (LRFS) and overall survival (OS). With gene expression array analysis the biological basis of the determinants of locoregional recurrence was established. All patients had been treated with surgery and 120 with adjuvant radiotherapy, if indicated. RESULTS Perineural growth (OS and LRFS) and non-cohesive invasive growth were correlated with worse prognosis (OS and LRFS). The negative effect of these characteristics on survival was maintained in the group that received post-operative radiotherapy. No association was found for degree of differentiation and bone invasion and presence of dysplasia or tumor at the margins. For the pattern of extensive non-cohesive growth, but not for perineural growth, a differential set of 160 genes was established, that included genes involved in extra-cellular matrix modeling CONCLUSION This study confirms the prognostic value of histological features in HNSCC, and reveals biologically relevant genes of tumors that show extensive non-cohesive growth. 84

87 Tumor biological determinants of locoregional recurrence of non-hpv head and neck squamous cell carcinoma INTRODUCTION Head and neck squamous cell carcinoma (HNSCC) contributes to approximately 4% of all cancers worldwide. 1 These carcinomas are largely caused by tobacco and excessive alcohol consumption, which seem to have a synergistic effect. A subgroup of HNSCCs, particularly those of the oropharynx is caused by infection with high-risk-types of human papilloma virus (HPV). 2 There were approximately 130,000 new HNSCC patients in Europe in 2008, more than two-thirds of them with cancer in the oral cavity and pharynx. 3 The percentage of European patients dying from SCC of the oral cavity and pharynx (together coined OPSCC) in that year was estimated at 45%. Unfortunately, the five-year survival rate of OPSCC has not markedly improved over the last decades and an important explanation for this observation is the development of locoregional recurrences. 4 Investigators have tried to unravel the association between histopathological tumor characteristics and the development of locoregionally recurrent cancer and overall survival time. The background of this research is that prognostic features may help to optimize treatment planning. Some publications demonstrate a relation between histopathological characteristics such as the presence of tumor near or in the margins and a shorter locoregional recurrence-free (LRFS) and overall survival (OS). 5-7 In addition, perineural invasion, 8-10 bone involvement have been associated with worse LRFS and OS. In general, the risk associated with these histopathological characteristics is moderate, for some features the results are inconclusive and sometimes OS and not LRFS is reported Gene expression array analysis of HNSCC has also been applied to uncover genes or gene profiles associated with prognosis However, the relation between the development of locoregional recurrent cancer and gene expression profiles has previously hardly been addressed. The present study was designed to identify for a large group of OPSCC the histopathological characteristics that are associated with the development of locoregionally recurrent OPSCC. As a next step we aimed to identify genes associated with these characteristics since they may play a regulatory role. Knowledge on the genes underlying key biological processes could improve diagnostic procedures, and accelerate and facilitate the development of a more targeted treatment in OPSCC. 85

88 Chapter 5 MATERIALS AND METHODS Patients and tumors The study included 170 patients, who had a histopathologically proven HNSCC and were treated by surgery in two large referral University medical centers in The Netherlands University Medical Center Utrecht (UMCU; N=92) and VU University Medical Center (VUmc; N=78) Amsterdam, between 1996 and Written informed consent was obtained from all patients and both Institutional Review Boards approved the study. Patients had a primary HPV-negative HNSCC in the oral cavity (N= 146) or oropharynx (N=24); 70 patients were female and 100 were male. Because HPV-positive tumors belong to a separate group, 21,22 we did not include these in this study. HPV involvement was tested for all carcinomas according to a validated algorithm. 2 All were staged according to the classification of the International Union Against Cancer. 23 In 147 patients a neck dissection was performed allowing pathological neck staging, and in the remaining 23 patients a follow-up period of at least two years was taken to clinically stage the neck. There was a tumor-positive neck in 103 (61%) of the patients. The median age was 61, mean 62 and the range was from 34 to 84 years. At the end of follow-up there were 78 patients alive, with a median survival time of 81 months (range ). Ninety-two patients had died, with a median survival time of 22 months (range 4-119). All patients were treated with curative intent by surgery with (N= 120) or without (N=50) post-operative radiotherapy. Indications for radiotherapy were multiple nodal metastases, extranodal spread and inadequate surgical margins. Histopathology In addition to standard examination of the surgical specimen, detailed histopathological characteristics of each tumor was scored by two experienced pathologists (EB + JAK), according to the guidelines of the Royal College of Pathologists UK ( For this, the routine (formalin fixed, paraffin embedded) slides were used that had been prepared from the surgical specimens. Grading of the differentiation was based on the degree of resemblance of the carcinoma to the normal epithelium. The most aggressively appearing part was scored as well, as moderately or poorly differentiated. The margin status was evaluated and was divided into three groups: tumor-free margins, when the excised carcinoma was > 5 mm from the surgical margin; involved margins when carcinoma was present in or within 1 mm of the margin; and close margins in the remaining cases. 24 The pattern of invasive or infiltrative growth was scored as either cohesive: carcinoma composed of broad cohesive sheets of cells, or non-cohesively invasive: carcinomas composed of narrow strands, spidery small groups or single cells. A distinction was made between focal and extensive infiltrative growth, based on the extent of the growth pattern. In case of extensive, the non-cohesive infiltrative growth pattern was the most predominant one. Perineural growth was scored when tumor 86

89 Tumor biological determinants of locoregional recurrence of non-hpv head and neck squamous cell carcinoma cells were in direct contact with neurons, at or outside the invasive front of the carcinoma, and classified as focal or extensive. Dysplasia in the margins was scored according to the four-tiered WHO categories: 25 no dysplasia, mild, moderate and severe. Bone invasion was scored as erosive or infiltrative as described by Slootweg and Muller. 26 In our patient cohort there were 91 patients with bone in the surgical specimen. Expression arrays A frozen tumor biopsy taken at the time of surgery was the source tissue for expression analysis. RNA was isolated as described before using TRIzol (Life Technologies, Breda, The Netherlands) according to the manufacturer s instructions. The samples were analysed per center on the same platform (4x44K from Agilent Technologies, Santa Clara, CA, USA), but with differences. In Utrecht, custom-made full-genome array (made for Agendia (Amsterdam, The Netherlands) was used to also include the presence of additional probes that represented previously published signature probes in replicate. Raw fluorescence intensities were quantified using Agilent Feature Extraction software (v. 9.5) and imported into R/Bioconductor ( and Further details of the data analysis in Utrecht can be found in the publication of van Hooff et al As for Amsterdam, array hybridization, using the Agilent Low RNA Input Fluorescent Linear Amplification Kit and 4x44K Whole Human Genome Arrays, was carried out according to the manufacturer (Agilent Technologies, Amstelveen, The Netherlands). 43,371 probes were on the array, covering a total of 30,982 different probes. Adhering to the ceteris paribus principle, gene expression data generated in the VUmc was preprocessed in an identical fashion as those from the UMCU. This comprised: 1) extraction of the median signal from the raw data files, 2) no background correction, and 3) median and loess within-array normalization (as implemented in the Limma-package). Next, the preprocessed VUmc and UMCU data set were combined by limiting both data sets to overlapping probes (using their Agilent probe id s). Finally, comparibility of the expression data of both hospitals is ensured by a) joint between-array normalization (A-quantile as implemented in the Limma-package), and b) removal of possible batch (i.e. hospital) effects using the Combatpackage. 28 Ingenuity pathway analysis (IPA from Ingenuity Systems, Redwood City, CA, USA) was used to identify underlying biological processes and pathways. This web-based software program uses an up-to-date database of structured biological and chemical information from the literature. The data are stored at the GEO-internet database as record GSE

90 Chapter 5 Statistical analyis As endpoint overall survival (OS) was calculated, with death (N=97) as event. Locoregional recurrence free survival (LRFS) was determined with the development of a locoregional recurrence (N=33) as event. A locoregional recurrence was defined as tumor regrowth at or near the site of the primary carcinoma within three years after diagnosis of the primary carcinoma. 29 Kaplan-Meier curves were plotted and the log-rank test was performed to test for significance. Univariate Cox regression was used to investigate the influence of histopathological parameters on the Hazard Ratios (HR). Differential gene expression with the arrays was assessed using the hierarchical linear model with empirical Bayes estimation of the variance as implemented in Limma, which also allows for the inclusion of possible confounders. The multiplicity problem (many genes are tested) was addressed by application of the Benjamini-Hochberg procedure to the raw p-values to control the FDR (false discovery rate). The association between the expression levels of (a group of) genes and a clinical variable was assessed using the globaltest. 30 The globaltest evaluates the association at an aggregated level, averaging the evidence (in favor of an association) over all genes. Hence, it does not pinpoint the genes responsible for an observed association. Frequency distribution was analyzed by chi-square (Pearson s) correlation analysis. Statistical tests were taken two-sided and differences were considered significant at α=0.05. For statistical analyses SPSS 20.0 for Windows (IBM, Armonk, NY USA) was used. RESULTS Histopathological characteristics and prognosis Patient and tumor characteristics were analyzed for their association with LRFS (table 1) and OS (table 2). With respect to well known risk factors, a significant association was noted for disease stage and lymph node status, in case of LRFS as well as OS. The development of a locoregional recurrence was associated with the presence of perineural growth (focal and extensive) and a pattern of non-cohesive growth ( focal and extensive) (table 1). Extensive perineural growth and extensive non-cohesive growth, but not the focal features were associated with a worse OS. The presence of dysplasia or carcinoma in the margins, the degree of differentiation and bone invasion of the tumor were not associated with prognosis. 88

91 Tumor biological determinants of locoregional recurrence of non-hpv head and neck squamous cell carcinoma Figure 1 shows the LFRS and OS Kaplan-Meier curves stratified for pattern of invasive growth. The difference between the patient groups with extensive non-cohesive and a cohesive growth pattern was stastistically different and most prominent. Table 1 Association HNSCC characteristics locoregional recurrence free survival (LRFS) Variable Characteristic N HR Value 95% CI P Disease stage I+II III+IV Site oropharynx oral cavity T-stage Lymph node status negative positive Dysplasia margins no mild moderate severe Perineural growth no focal extensive Non-cohesive growth no focal extensive Margin status tumor free close involved Differentiation degree well moderately poorly Bone invasion no invasion invasion The hazard ratio (HR) was calculated with the HNSCC without the particular characteristic as the reference (value 1.0). Significant differences are shown by a p-value in bold. Regarding the degree of differentiation the group of well differentiated HNSCC was chosen to be the reference. The analysis of bone invasion was limited to HNSCC that showed bone in the surgical specimen. In 43 patients the tumor was adjacent or close to bone but did not invade. In 47 patients there was a sign of bone invasive growth: in 4 cases this consisted of bone erosion, in the others bone invasion was observed. 89

92 Chapter 5 A B Figure 1 Kaplan-Meier survival analysis for invasive growth. Stratification of invasive growth: extensive (N=50), focal (N=65) and non (N=55). (A) LRFS: the overall p-value is 0.043, and for the pair-wise comparisons these were (no vs. extensive), (no vs. focal) and (focal vs. extensive). (B) OS: the overall p-value of is shown, for pair-wise comparisons these were (extensive vs. focal), (no vs.focal) and (no vs. extensive). The overall log rank p-value was highly significant for OS (p=0.004) and significant for LFRS (p=0.043). In a separate analysis we stratified for treatment to examine the interaction between the histological characteristics and radiotherapy on survival. An extensive non-cohesive growth pattern and extensive perineural growth had a significant negative effect on OS in this separate patient group (table 4). With regard to the development of a locoregional recurrence, an increased risk for extensive perineural growth and noncohesive growth was observed, but this effect approached statistical significance. The presence of tumor in the margin did not appear to influence the patients prognosis. In the patient group treated with surgery no significant effects were observed, possibly related to the relative low number of patients in the groups (data not shown). 90

93 Tumor biological determinants of locoregional recurrence of non-hpv head and neck squamous cell carcinoma Table 2 Association between tumor characteristics and overall survival Variable Characteristic N HR Value 95% CI P Disease stage I+II III+IV <0.000 Site oropharynx oral cavity T-stage Lymph node status negative positive <0.000 Locoregional recurrence negative positive <0.000 Dysplasia margins no mild moderate severe Perineural growth no focal extensive Non-cohesive growth no focal extensive Margin status tumor free close involved Differentiation degree well moderately poorly Bone invasion no invasion invasion The HR was calculated with the HNSCC without the particular characteristic as the reference (value 1.0). Significant differences are shown by a p-value in bold (see further details in the legends of table 1). 91

94 Chapter 5 Kaplan-Meier analysis was also performed for the pattern of invasive growth when stratified according to different stage groups (table 3). For patients with less advanced carcinomas the relation between type of invasion and survival (OS and LRFS) was not statistically significant. Table 3 Association between extensive non-cohesive gene profile and growth pattern vs. survival. Association Survival type Stage Early All Extensive non-cohesive growth and survival* LRFS OS Extensive non-cohesive gene profile and survival** LRFS OS HNSCC were stratified in two groups: the early stage (stage I and II; N=40), and all HNSCC (N=170). P-values are shown. *overall log-rank p value of the Kaplan-Meier analysis of the three groups **Global test (see materials and methods) with worst pattern of growth 160 gene profile. Significant associations (p<0.05) are shown in bold. Table 4 Association tumor characteristics locoregional recurrence free survival (LRFS) and overall survival (OS) in 120 patients treated with postoperative radiotherapy Variable Characteristic N HR-Value 95% CI P LRFS Perineural growth no 69 focal extensive Invasive growth no focal extensive OS Margin status tumor free close involved Perineural growth no 69 focal extensive Invasive growth no focal extensive Margin status tumor free close involved The HR was calculated with the carcinomas without the particular characteristic as the reference (value 1.0). Significant differences are shown by a p-value in bold. 92

95 Tumor biological determinants of locoregional recurrence of non-hpv head and neck squamous cell carcinoma Gene expression arrays The aim was to identify the genes that were associated with characteristics with prognostic value (table 5). As for non-cohesive growth, 160 different transcripts were found to be differentially expressed, when extensive was compared with no (supplementary table S1 provides the full list of these transcripts). Pathways that stood out were matrix metalloproteinases and HIF1alpha signaling as determined with IPA. In this list many genes related to the extra-cellular matrix component (matrix metalloproteinases and collagen) can be appreciated. No differentially expressed transcripts were observed with regard to perineural growth, bone invasion, degree of differentiation, the presence of dysplasia and carcinoma in or near the surgical margins and the occurrence of a locoregional recurrence. Table 5 Genes differentially expressed between clinically &histologically defined patient groups Variable Characteristic Patients (N) Number of different genes Disease stage I+II vs. III+IV 40 vs ,442 5 Lymph node status Negative vs. positive 67 vs ,769 Locoregional recurrence Negative vs positive 137 vs Invasive growth no vs. focal 55 vs no vs. extensive 55 vs Perineural growth No vs. focal 107 vs No vs. extensive 107 vs Bone invasion Negative vs. positive 43 vs Significantly different genes are shown (p-value <0.05 with FDR-correction). The analysis of bone invasion was limited to the HNSCC that had bone in the surgical specimen. The non-cohesive growth associated gene profiles were also tested for prognostic significance. There was no statistically association between the non-cohesive gene profile and OS or LRFS, for the whole tumor group as well as for the early stage carcinomas (table 3). DISCUSSION This study examined the prognostic value of several histopathological characteristics of a large series of OPSCC, in particular in relation to locoregional recurrence. Because HPV-positive HNSCC are considered as a different class of OPSCC, 31 these were excluded from the study. Disease stage, and some tumor histopathological characteristics were associated with LRFS and OS. A non-cohesive growth pattern was found to be of prognostic significance and moreover, the genes responsible for this phenotype were uncovered. 93

96 Chapter 5 The present results on the correlation of various histopathological characteristics with LRFS and OS confirm those of others. Most studies find a shorter OS and LRFS to be associated with perineural growth, 8-10,16,32 and a non-cohesive growth pattern. 14,16,33,34 Nevertheless, there are publications that do not find such a correlation between growth pattern and the development of locoregional recurrences. 15 The present study revealed that the expression of a certain gene set differs between tumors that showed non-cohesive, spidery invasive growth compared with those with cohesive growth. The group with extensive non-cohesive growth consists of about one-third of the tumor population in this study, showing that this is a relatively frequent phenomenon. Certain genes involved in the homeostasis of the extracellular matrix were overexpressed in tumors that showed a relatively high level of this non-cohesive growth. It makes sense from the biological point of view that enzymes like matrix metalloproteinases, (MMPs) are upregulated in this relatively aggressive group of OPSCC. Previous immunohistochemical studies also noted a correlation between MMPexpression and poor prognosis and MMP-1, -2, -9 and -13 have most frequently be reported to be of significance Our analysis indicates that MMP-13 and -14 stand out most. Furthermore, genes encoding for collagen, e.g. COL5A2 were co-upregulated. It is conceivable that tumor cells synthesize MMPs to enable invasion; the source of collagens is more uncertain and it may be that stromal cells produce these as well. The source cells of collagen overproduction in HNSCC can be stromal as well as tumor cells. 38 Nevertheless, the function of other genes in the non-cohesive gene set, need to be elucidated. The presence of perineural growth was shown to be a poor prognostic factor, but this was not reflected in differences in gene expression between tumor groups with and without this characteristic. A possible explanation is that gene expression is only altered in the relatively small group of cells involved in this process. On the other hand it cannot be excluded that this process is regulated by delicate changes in gene expression. Bone-invasion did not have consequences for prognosis. This is a somewhat unexpected result and is in contrast to an earlier report. 39 Apparently, the surgical and radiation treatment is effective in our series and aggressive biological behavior is sufficiently controlled. No genes associated with bone invasion were uncovered, suggesting that no such genes exists or that these are acting very locally, at the interphase of tumor and bone. Another important finding in the present study is that, despite treatment with radiotherapy, extensive perineural and non-cohesive growth retained their association with worse OS. For LRFS, the HR values were as large as observed for OS, but the significance level of 0.05 was just not reached. Extensive perineural and non-cohesive growth had higher prognostic value than for 94

97 Tumor biological determinants of locoregional recurrence of non-hpv head and neck squamous cell carcinoma tumors with tumor in or near the margin. In fact, there was no association between the presence of tumor or dysplasia in the margin and prognosis (locoregional recurrence or death), and this was found for the total group of patients and for the group that was treated with post-operative radiotherapy. The effectiveness of the radiotherapy is likely the reason that tumor in or near the margins does not lead to a worse prognosis. In this respect the present study confirms the result of others. 16,40 The design of our study was not suited to address the question whether radiotherapy is effective or not. In our centers the administration of radiotherapy is indicated when the margins are compromised. The effect of radiotherapy in this respect is considered to be sufficiently proven. 5 Thus is seems that our study suggests that extensive perineural growth or extensive non-cohesive growth radiotherapy combined with chemotherapy is indicated. The present study could not reveal a distinct association between the non-cohesive gene set and worse LRFS or OS. Thus, based on our patient set, this gene profile has little potential to predict the prognosis of OPSCC patients. This aspect may be particularly important for patients from whom the tumor is not surgically removed. In a small biopsy, that is taken when chemoradiation is planned as primary therapy, it is hard to assess histopathological characteristics, like pattern of invasion. Expression measurement in such biopsy for instance of MMPs could be useful in this respect. The reason for the weak association between a non-cohesive gene profile and the development of locoregional recurrences is subject of speculation. It is possible that the gene profile is more valid in early stage tumors and that in later stage tumors other processes are involved in tumor recurrence as well. 5 In summary, this study confirms the prognostic value of several histological characteristics in OPSCC, and reveals biologically relevant genes of tumors that show a non-cohesive growth pattern. 95

98 Chapter 5 REFERENCES 1. Ferlay J, Soerjomataram I, Ervik M, et al. GLOBOCAN 2012 v1.0, Cancer Incidence and Mortality Worldwide: IARC CancerBase No. 11 [Internet]. Lyon, France: International Agency for Research on Cancer; 2013 Available from: iarc fr, accessed on March 6, Rietbergen MM, Leemans CR, Bloemena E, et al. Increasing prevalence rates of HPV attributable oropharyngeal squamous cell carcinomas in the Netherlands as assessed by a validated test algorithm. Int J Cancer 2013; 132: Ferlay J, Parkin DM, Steliarova-Foucher E. Estimates of cancer incidence and mortality in Europe in Eur J Cancer 2010; 46: Graveland AP, Braakhuis BJM, Eerenstein SEJ et al. Molecular diagnosis of minimal residual disease in head and neck cancer patients. Cell Oncol 2012; 35: Woolgar JA. Histopathological prognosticators in oral and oropharyngeal squamous cell carcinoma. Oral Oncol 2006; 42: Kovacs AF. Relevance of positive margins in case of adjuvant therapy of oral cancer. Int J Oral Maxillofac Surg 2004; 33: Iyer NG, Nixon IJ, Palmer F, et al. Surgical management of squamous cell carcinoma of the soft palate: Factors predictive of outcome. Head and Neck 2012; 34: Woolgar JA, Rogers S, West CR, et al. Survival and patterns of recurrence in 200 oral cancer patients treated by radical surgery and neck dissection. Oral Oncol 1999; 35: Fagan JJ, Collins B, Barnes L, et al. Perineural invasion in squamous cell carcinoma of the head and neck. Arch Otolaryngol Head Neck Surg 1998; 124: Johnston M, Yu E, Kim J. Perineural invasion and spread in head and neck cancer. Expert Rev Anticancer Ther 2012; 12: Brown JS, Lowe D, Kalavrezos N, et al. Patterns of invasion and routes of tumor entry into the mandible by oral squamous cell carcinoma. Head Neck 2002; 24: Bryne M, Koppang HS, Lilleng R, et al. Malignancy grading of the deep invasive margins of oral squamous-cell carcinomas has high prognostic value. J Pathol 1992; 166: Woolgar JA, Scott J. Prediction of cervical lymph-node metastasis in squamous-cell carcinoma of the tongue floor of mouth. Head Neck 1995; 17: Okamoto M, Ozeki S, Watanabe T, et al. Cervical lymph-node metastasis in carcinoma of the tongue - correlation between clinical and histopathological findings and metastasis. J Craniomaxillofac Surg 1988; 16: Janot F, Klijanienko J, Russo A, et al. Prognostic value of clinicopathological parameters in head and neck squamous cell carcinoma: A prospective analysis. Br J Cancer 1996; 73: Brandwein-Gensler M, Teixeira MS, Lewis CM, et al. Oral squamous cell carcinoma: histologic risk assessment, but not margin status, is strongly predictive of local disease-free and overall survival. Am J Surg Pathol 2005; 29: Braakhuis BJ, Bloemena E, Leemans CR, et al. Molecular analysis of surgical margins in head and neck cancer: more than a marginal issue. Oral Oncol 2010; 46: Chung CH, Parker JS, Ely K, et al. Gene Expression Profiles Identify Epithelial-to-Mesenchymal Transition and Activation of Nuclear Factor-{kappa}B Signaling as Characteristics of a High-risk Head and Neck Squamous Cell Carcinoma. Cancer Res 2006; 66: Lohavanichbutr P, Mendez E, Holsinger FC, et al. A 13-gene signature prognostic of HPV-negative OSCC: discovery and external validation. Clin Cancer Res 2013; 19: Jung AC, Job S, Ledrappier S, et al. A poor prognosis subtype of HNSCC is consistently observed across methylome, transcriptome, and mirnome analysis. Clin Cancer Res 2013; 19: Braakhuis BJ, Snijders PJ, Keune WJ, et al. Genetic patterns in head and neck cancers that contain or lack transcriptionally active human papillomavirus. J Natl Cancer Inst 2004; 96:

99 Tumor biological determinants of locoregional recurrence of non-hpv head and neck squamous cell carcinoma 22. Smeets SJ, Braakhuis BJ, Abbas S, et al. Genome-wide DNA copy number alterations in head and neck squamous cell carcinomas with or without oncogene-expressing human papillomavirus. Oncogene 2006; 25: Sobin L, Gospodarowicz M, Wittekind C. TNM Classification of malignant tumours. Seventh edition. TNM classification of malignant tumours, 2009, 7th Edition Wiley Blackwell, Chichester, UK. 24. Graveland AP, de Maaker M, Braakhuis BJ, et al. Molecular detection of minimal residual cancer in surgical margins of head and neck cancer patients. Cell Oncol 2009; 31: Warnakulasuriya S, Reibel J, Bouquot J, et al. Oral epithelial dysplasia classification systems: predictive value, utility, weaknesses and scope for improvement. J Oral Pathol Med 2008; 37: Slootweg PJ, Muller H. Mandibular invasion by oral squamous cell carcinoma. J Craniomaxillofac Surg 1989; 17: van Hooff SR, Leusink FKJ, Roepman P, et al. Validation of a gene expression signature for assessment of lymph node metastasis in oral squamous cell carcinoma. J Clin Oncol 2012; 30: Johnson WE, Li C, Rabinovic A. Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics 2007; 8: Braakhuis BJ, Tabor MP, Leemans CR, et al. Second primary tumors and field cancerization in oral and oropharyngeal cancer: molecular techniques provide new insights and definitions. Head Neck 2002; 24: Goeman JJ, van de Geer SA, de Kort F, et al. A global test for groups of genes: testing association with a clinical outcome. Bioinformatics 2004; 20: Leemans CR, Braakhuis BJ, Brakenhoff RH. The molecular biology of head and neck cancer. Nat Rev Cancer 2011; 11: Dissanayaka WL, Pitiyage G, Kumarasiri PVR, et al. Clinical and histopathologic parameters in survival of oral squamous cell carcinoma. Oral Surg Oral Med Oral Pathol Oral Radiol 2012; 113: Bryne M, Jenssen N, Boysen M. Histological Grading in the Deep Invasive Front of T1 and T2 Glottic Squamous-Cell Carcinomas Has High Prognostic Value. Virchows Arch 1995; 427: Sinha P, Mehrad M, Chernock RD, et al. Histologic and systemic prognosticators for local control and survival in margin-negative transoral laser microsurgery treated oral cavity squamous cell carcinoma. Head and Neck 2015; 37: Al-Azri AR, Gibson RJ, Keefe DMK, et al. Matrix metalloproteinases: do they play a role in mucosal pathology of the oral cavity? Oral Diseases 2013; 19: Jordan RC, Macabeo-Ong M, Shiboski CH, et al. Overexpression of matrix metalloproteinase-1 and -9 mrna is associated with progression of oral dysplasia to cancer. Clin Cancer Res 2004; 10: Katayama A, Bandoh N, Kishibe K, et al. Expressions of matrix metalloproteinases in early-stage oral squamous cell carcinoma as predictive indicators for tumor metastases and prognosis. Clin Cancer Res 2004; 10: Sok JC, Lee JA, Dasari S, et al. Collagen type XI alpha1 facilitates head and neck squamous cell cancer growth and invasion. Br J Cancer 2013; 109: Shaw RJ, Brown JS, Woolgar JA, et al. The influence of the pattern of mandibular invasion on recurrence and survival in oral squamous cell carcinoma. Head Neck 2004; 26: Kademani D, Bell RB, Bagheri S, et al. Prognostic factors in intraoral squamous cell carcinoma: The influence of histologic grade. J Oral Maxillofac Surg 2005; 63:

100 Chapter 5 SUPPLEMENTARY INFORMATION Supplementary Table 1 Invasive growth differentially expressed transcripts (n=160) HNSCC with extensive invasive growth (N=50) vs. without invasive growth (N=55) A_24_P BX GLIS3 PLAU A_32_P C10orf13 GOT1 PORCN A_32_P20040 C10orf77 GPLD1 PRSS23 A_32_P4349 C20orf103 GPM6B PYCR2 A_32_P52268 C20orf175 GRIA3 RPAP1 AA C21orf63 GRP SEDLP ABCA1 C5orf13 HAS2 SGEF ACO2 C6orf35 HEYL SGIP1 ADAM12 C6orf72 HNT SLC16A14 ADAMTS2 C9orf21 HRBL SLC22A15 AF CACNA1G HSD17B7 SLC25A23 AI CALD1 HSPA12A SLC26A10 AI CD248 IFT20 SLC38A2 AK CEBPA IKIP SMCR7L AK CHN1 IL11 STARD13 AK CLEC11A IL7R SYNE1 AK COL11A1 INHBA SYTL2 AK COL1A2 ITGA1 TBL1X AK COL3A1 KCNE4 TBL1Y AK5 COL4A1 KHK TCF8 ALDH5A1 COL5A1 KIF26B TES ANGPT2 COL5A2 LAMA4 TGFBR1 ANKRD38 COL5A3 LOC THC ANTXR1 COL6A1 LOC THC APEX2 COL6A2 LOC THC APLN E2F2 LOC90499 THC ARRDC3 ENSA LOXL2 THC AW ENST MGC23280 THC AW ENST MME THC BC ENST MMP1 TIPARP BC ENST MMP13 TMCC1 BC ENST MMP14 TNFAIP6 BC ENST MMP28 TPST1 BC ENST MSX2 TWIST1 BC FANCD2 MTMR4 UBE2W BF FAP NID2 UFM1 BI FLJ40869 OLFM2 VAMP3 BM GAS1 OSBPL8 WISP1 BQ GGA1 PAPPA ZBTB43 BX GINS4 PARP1 ZNF469 98

101 Tumor biological determinants of locoregional recurrence of non-hpv head and neck squamous cell carcinoma 5 99

102

103 CHAPTER 6 Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS2 R. Noorlag P. van der Groep F.K.J. Leusink S.R. van Hooff M.H. Frank S.M. Willems R.J.J. van Es Head Neck 2015; 37:

104 Chapter 6 ABSTRACT BACKGROUND Gene expression profiling revealed a strong signature predicting lymph node metastases in oral squamous cell carcinoma (OSCC). Four of the most predictive genes are secretory leukocyte protease inhibitor (SLPI), lipocalin-2 (LCN2), thrombospondin-2 (THBS2), and tumor-associated calcium signal transducer 2 (TACSTD2). This study correlates their protein expression with lymph node metastases, overall survival (OS), and disease-specific survival (DSS). METHODS Two hundred twelve patients with OSCC were included for protein expression analysis by immunohistochemistry. RESULTS SLPI expression correlates with lymph node metastases in the whole cohort, not in a subgroup of ct1 to 2N0. SLPI expression correlates with OS (hazard ratio [HR] = 0.61) and DSS (HR = 0.47) in multivariate analysis. LCN2, THBS2, and TACSTD2 show no correlation with lymph node metastases, OS, or DSS. CONCLUSION Although SLPI expression correlates with lymph node metastases, it has no additional value in determining lymph node metastases in early oral cancer. However, it is an independent predictor for both OS and DSS and therefore a relevant prognostic biomarker in OSCC. 102

105 Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS2 INTRODUCTION Head and neck cancer is the sixth most common malignancy worldwide, of which one-third consists of oral squamous cell carcinoma (OSCC). Its incidence in The Netherlands, being 6.2 per in 2010, is rising annually. Despite improvements in therapy, the 5-year survival rate has not changed over the past decades and remains approximately 50%. 1-3 The prognosis depends on numerous clinical and pathological factors, of which cervical lymph node metastases is a major determinant. 4 To perform appropriate treatment, it is therefore pivotal to determine the nodal status of the neck. However, in 30% to 40% of the patients, even optimal imaging is unable to detect nodal disease. 5 To improve the negative predictive value for metastasis detection in OSCC, new diagnostic tools, such as molecular diagnosis and tumor profiling, have been developed.[5] Roepman et al 6 showed that microarray gene expression profiling could be used to predict lymph node metastases for OSCC. Recently, this gene expression signature has been validated in a multicenter study and focused on the prediction of lymph node metastases in early oral cancer. 7 Four of the strongest predictive genes in this signature encode for the proteins lipocalin-2 (LCN2), thrombospondin-2 (THBS2), tumor-associated calcium signal transducer 2 (TACSTD2), and secretory leukocyte protease inhibitor (SLPI). 6 LCN2 participates in carcinogenesis by favoring iron uptake from the extracellular space within the tumor cell, a fundamental process for maintaining neoplastic cell multiplication. Although increased LCN2 plasma levels in patients with OSCC were found, no correlation was established with regional or distant metastases. 8 THBS2 suppresses angiogenesis by inhibiting endothelial cell migration, inducing endothelial cell apoptosis and preventing the interaction of growth factors with the cell surface receptors of the endothelial cell. 9 In supraglottic cancer, THBS2 gene expression seems inversely correlated with nodal metastases. 10 TACSDT2, also known as TROP-2, belongs to a unique family of transmembrane glycoproteins that has a regulatory role in cell-cell adhesion and has a key controlling role in human cancer growth. Tumor development is quantitatively driven by TACSTD2 expression levels in many tumors. 11 Fong et al 12 correlated increased TACSTD2 expression in OSCC with decreased overall survival (OS), but found no correlation with nodal metastases. SLPI, also known as antileukoproteinase, is a protease inhibitor of neutrophil elastase, cathepsin G, chymotrypsin, and trypsin, 13,14 enzymes with extracellular matrix degradative properties, and 103

106 Chapter 6 associated with cancer development, invasiveness, and progression. 15,16 SLPI expression has recently been associated with carcinogenesis and metastasis in various types of cancer, although its role remains controversial. In gastric and prostate cancer, increased SLPI expression is associated with invasiveness, metastases, and a worse survival This is in contrast with the reports of SLPI expression in ovarian cancer, in which SLPI expression is associated with decreased tumor growth and fewer nodal metastases. 20 In head and neck cancer, SLPI mrna and protein levels seem to be increased compared to normal tissue. 21 Reports correlating SLPI expression with lymph node 22, 23 metastases are contradicting. The purpose of this study was to present the correlation of the aforementioned protein expressions with lymph node metastases, OS, and disease specific survival (DSS), and evaluate their potential role as biomarkers for treatment decision and predictors of survival in OSCC. MATERIALS AND METHODS Patient selection Patients with histologically confirmed OSCC, whose primary treatment was by surgery between 1996 and 2005 in our institute were included in this study. Patients who had had a synchronous primary tumor or a previous malignancy in the head and neck region were excluded. Two hundred twelve patients were selected on the availability of both representative formaldehydefixed, paraffin-embedded tissue blocks and frozen tissue samples of the primary tumor. A dedicated head and neck pathologist examined all hematoxylin-eosin stained slides with special attention to the following pathological characteristics: type of tumor, differentiation grade, infiltration depth, invasive pattern, perineural growth, vasoinvasive growth, extracapsular spread, and bone invasion. A tissue microarray was made of the paraffin-embedded tissue. For each tumor block, 3 central tissue cylinders and 3 tissue cylinders at the tumor front with a diameter of 0.6 mm were punched out, avoiding areas of necrosis, and arrayed in a recipient paraffin block. Normal epithelium from the floor of the mouth, gingiva, and tonsil was incorporated in each block to ensure similarity of staining between the different blocks, as described earlier. 24 From each patient, clinical characteristics, clinical TNM classification (based on palpation, ultrasound-guided fine-needle aspiration, MRI or CT, and classified in a multidisciplinary panel), pathological TNM classification, and cause of death were retrieved from the medical records, as listed in Table

107 Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS2 Table 1 Baseline characteristics. Variable No. of patients (%) Sex Female Male 84 (40) 128 (60) Age at diagnosis Median (Range) 61 (26-87) Smoking Never Ceased > 1 year Active smoker or ceased < 1 year Missing Alcohol Never Occasionally 1-4 U/day 5 U/day Missing Clinical N-classification cn0 cn1-3 Clinical T-classification ct1 ct2 ct3 ct4 Pathological N-classification pn0 pn1-3 Pathological T-classification pt1 pt2 pt3 pt4 Infiltration depth < 4.0mm 4.0mm Differentiation grade Good / Moderate Poor / Undifferentiated Vaso-invasion No Yes Missing Bone-invasion No Yes Perineural growth No Yes Missing Invasive pattern Cohesive Non-cohesive Missing Extra capsular spread No Yes No nodal metastasis High risk HPV status Negative Positive 43 (20) 34 (16) 133 (63) 2 (1) 46 (22) 49 (23) 71 (33) 44 (21) 2 (1) 146 (69) 66 (31) 44 (21) 79 (37) 19 (9) 70 (33) 97 (46) 115 (54) 44 (21) 73 (34) 22 (10) 73 (34) 19 (9) 193 (91) 173 (82) 39 (18) 39 (18) 169 (80) 4 (2) 152 (72) 60 (28) 122 (57) 80 (38) 10 (5) 44 (21) 167 (79) 1 (<1) 59 (28) 56 (26) 97 (46) 210 (99) 2 (1) 6 105

108 Chapter 6 Gene expression For a subgroup of 83 tumors, normalized gene expression data were available from an earlier study for which methods has been described in detail earlier. 7 In short, frozen tumor samples were sectioned, aliquoted in Trizol (Life Technologies, Frederick, MD), and sent to Agendia Laboratories (Amsterdam, The Netherlands) for expression profile analysis. Tumor areas with a percentage of at least 50% were assessed on hematoxylin-eosin stained sections and taken in parallel; RNA was isolation and amplification. Tumor sample RNA was labeled as Cy3, and reference RNA was labeled Cy5. As a reference, the Universal Human Reference RNA (Agilent Technologies, Santa Clara, CA) was used. Samples were hybridized on full-genome Agilent arrays. Raw fluorescence intensities were quantified using Agilent Feature Extraction software and imported into R/ Bioconductor ( for normalization (loess normalization using the LIMMA package) and additional analysis. Human papillomavirus type 16 analysis Human papillomavirus type 16 (HPV-16) active tumors were determined by p16 immunohistochemistry (IHC) followed by GP 5+/6+ polymerase chain reaction in positive p16 staining, a reliable algorithm for detection of HPV-16 in paraffin-embedded head and neck cancer specimens, as described by Smeets et al. 25 Immunohistochemistry IHC was performed on 4-µm thick paraffin sections. The tissue sections were deparaffinized with xylene and rehydrated. Endogenous peroxidase activity was blocked for 15 minutes in a 0.3% hydrogen peroxide phosphate-citrate buffer. Then, tissue sections were washed in water and subsequently subjected to antigen retrieval by boiling the slides in ethylenediaminetetraacetic acid buffer, ph 9.0 (SLPI) or citrate buffer, and ph 6.0 (TACSTD2, LCN2, and THBS2) for 20 minutes. Sections were cooled down within the buffers for 30 minutes. After washing with phosphate-buffered saline (PBS) for 5 minutes, tissue slides were incubated with the primary antibody SLPI (clone 31; HyCult biotechnology, Uden, The Netherlands; dilution 1:50), primary antibody TACSTD2 (AF650, R&D Systems, Oxon, England; dilution 1:50), primary antibody LCN2 (MAB1757; R&D Systems, Oxon, England; dilution 1:50), or primary antibody THBS2 (sc-12313, Santa Cruz Biotechnology, Santa Cruz, CA; dilution 1:50) for 60 minutes. After washing with PBS (3 times), incubation with poly-hrp Goat anti-mouse/rabbit/rat (Brightvision, Immunologic, Duiven, The Netherlands; ready to use) for 30 minutes was followed by washing with PBS (3 times). Slides were then developed with diaminobenzidine for 10 minutes, counterstained with hematoxylin, followed by dehydration, and mounted. Evaluation of immunohistochemical staining A core was considered inadequate/lost when the core contained <5% tumor tissue or when >95% of the core contained no tissue. Patients were only included in the study when one or 106

109 Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS2 more tumor cores were available. When two or more cores were available from one patient, the mean (SLPI, THBS2, or LCN2) or maximum (TACSTD2) score was calculated for that patient. The expression of SLPI and THBS2 in the primary tumor was evaluated by scoring the percentage of cytoplasm staining. The percentage of cytoplasm stained was classified as 0 (<5%), 1 (5% to 30%), 2 (31% to 75%), or 3 (>75%), see Figure 1.22, 26 Expression of TACSTD2 was evaluated by scoring the staining intensity of the cell membrane as 0 = no, 1 = weak, 2 = moderate, or 3 = strong staining. For LCN2 expression, both the intensity (0 = no, 1 = weak, 2 = moderate, or 3 = strong) and percentage of cytoplasm staining was scored, multiplying the intensity score with the percentage of staining classified as 1 ( 25%), 2 (26% to 50%), 3 (51% to 75%), or 4 (>75%) was used as a final score for LCN2 expression. Scores 3 were interpreted as negative, and scores >3 as positive.27 A dedicated head and neck pathologist (S.W.) and a researcher (R.N.), both blinded to the clinical characteristics of the patients, evaluated the protein expressions independently. Consensus was reached regarding discordant findings. 6 Figure 1 Scoring system for SLPI, LCN2, TACSTD2 and THBS2. Abbreviations: SLPI, secretory leukocyte protease inhibitor; THBS2, thrombospondin-2; LCN2, lipocalin-2; TACSTD2, tumor-associated calcium signal transducer R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

110 Chapter 6 Statistical analysis An interrater reliability analysis using the Spearman (for continuous data) and Kappa (for categorical data) statistic was performed to determine consistency of IHC scoring among raters. The Mann Whitney test was used to determine differences in gene expression between lymph node-positive and lymph node-negative tumors. Receiver operating characteristic (ROC) curve analysis was used to determine cutoff points for the correlation of gene and protein expression and nodal metastases. Correlations between gene expression or protein expression and lymph node metastases were assessed by the chi-square test. OS was defined as the length of the time interval from surgery to death from any cause. DSS was defined as the time interval from surgery to either death because of or as a recurrence of the disease. ROC curve analysis was used to determine cutoff points for protein expression and survival. The OS and the DSS curves were constructed using the Kaplan Meier method and the log-rank test was used to test for significance. Prognostic value was examined by univariate and multivariate analyses using the Cox proportional hazards regression model. Characteristics with a p <.10 in univariate analysis and potential confounders were included, and the model was created with backward logistic regression. All p values were based on 2-tailed statistical analysis and p <.05 was considered statistically significant. Statistical analysis was performed using the SPSS 20.0 statistical package (SPSS, Chicago, IL). RESULTS Human papillomavirus type 16 analysis Of our 212 tumor samples, 36 showed p16 overexpression on IHC. Of this group, only 2 samples (0.9%) proved to be true HPV-16 positive with polymerase chain reaction, see Table 1. Immunohistochemistry: descriptive analysis A total of 1080 cores (85%) stained with SLPI antibody, 1119 cores (88%) stained with THBS2 antibody, 1077 cores (85%) stained with LCN2 antibody, and 1077 cores (84%) stained with TACSTD2 antibody were available for analysis. There was at least 1 core of each tumor suitable for each staining so no tumors were excluded from analysis. The level of interrater concordance was high, with a Spearman s rank correlation of (p <.001) for continuous data and a Kappa of (95% confidence interval, ; p <.001) for categorical data, scatter plot in Supplemenary Figure 1, online only. The IHC results are given in Table

111 Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS2 Table 2 Immunohistochemistry descriptive results. Variable SLPI THBS2 LCN2 TACSTD2 No. of cores (%) Tumor No Tumor No Core 1080 (85) 50 ( (11) 1119 (88) 86 (7) 67 (5) 1077 (85) 68 (5) 127 (10) 1074 (84) 73 (6) 125 (10) No. of cores per tumor (212 tumors) Score per tumor (212 tumors) Score Score 5% 6-30% 31-75% > 75% > Abbreviations: SLPI, secretory leukocyte protease inhibitor; THBS2, thrombospondin-2; LCN2, lipocalin-2; TACSTD2, tumor-associated calcium signal transducer 2 6 Gene expression and lymph node metastases Analysis of 83 OSCCs show a statistically significant differential gene expression between lymph node-positive and lymph node-negative patients for SLPI (p =.001), LCN2 (p <.001), TACSTD2 (p =.002), and THBS2 (p =.001), see Figure 2. Optimal cutoff points determined with ROC curve analysis (Supplementary Figure 2 and Supplementary Table 1, online only) revealed that gene expression is a significant predictor of lymph node metastasis for all 4 genes. SLPI, LCN2, and TACSTD2 mrna are downregulated and THBS2 mrna is upregulated in lymph node-positive patients, see Table 3. Figure 2 Gene expression and nodal status. Mann-Whitney test, ** p < 0.01, *** p < Abbreviations: SLPI, secretory leukocyte protease inhibitor; THBS2, thrombospondin-2; LCN2, lipocalin-2; TACSTD2, tumor-associated calcium signal transducer 2; pn+, pathologic lymph node positive; pn0, pathologic lymph node negative. 109

112 Chapter 6 Table 3 Gene expressions correlated with lymph node metastasis. Gene expression SLPI M 0,130 M > 0,130 LCN2 M -0,360 M > -0,360 TACSTD2 M 0,027 M > 0,027 THBS2 M > 0,100 M 0,100 No. of patients No 18 (32%) 19 (73%) 11 (26%) 26 (65%) 14 (30%) 23 (64%) 7 (79%) 30 (61%) LNM Yes 39 (68%) 7 (27%) 32 (74%) 14 (35%) 33 (70%) 13 (36%) 27 (21%) 19 (39%) p-value <0.001 < <0.001 Abbreviations: SLPI, secretory leukocyte protease inhibitor; THBS2, thrombospondin-2; LCN2, lipocalin-2; TACSTD2, tumor-associated calcium signal transducer 2; M, log2(sample/reference pool). Figure 3 SLPI expression and survival. Log Rank test, overall survival p = disease-specific survival p < Abbreviations: SLPI, secretory leukocyte protease inhibitor. 110

113 Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS2 Table 4 Biomarkers correlated with lymph node metastasis. Biomarker expression SLPI 30% > 30% Whole cohort (212 tumors) ct1-2n0 (101 tumors) pn0 pn+ p-value pn0 pn+ 83 (43%) 14 (74%) 110 (57%) 5 (26%) (63%) 9 (69%) 33 (37%) 4 (31%) p-value NS LCN2 Score 3 Score > 3 51 (41%) 46 (54%) 75 (59%) 40 (46%) NS 33 (60%) 31 (67%) 22 (40%) 15 (33%) NS TACSTD (41%) 62 (49%) 50 (59%) 65 (51%) NS 31 (65%) 33 (62%) 17 (35%) 20 (38%) NS THBS2 5% > 5% 13 (52%) 84 (45%) 12 (48%) 103 (55%) NS 7 (78%) 57 (62%) 2 (22%) 35 (38%) NS Abbreviations: SLPI, secretory leukocyte protease inhibitor; THBS2, thrombospondin-2; LCN2, lipocalin-2; TACSTD2, tumor-associated calcium signal transducer 2; pn+, pathologic lymph node positive; pn0, pathologic lymph node negative; NS, not significant 6 Table 5 Cox regression analysis of overall survival. Variable HR (95 % CI) p-value SLPI expression 0.56 ( ) Multivariate model SLPI expression Age cn-classification Vaso-invasion Non-cohesive invasive pattern Bone invasion 0.61 ( ) 1.04 ( ) 3.81 ( ) 1.59 ( ) 2.69 ( ) 1.75 ( ) Abbreviations: SLPI, secretory leukocyte protease inhibitor; HR, hazard ratio; CI, confidence interval <0.001 < Table 6 Cox regression analysis of disease-specific survival. Variable HR (95 % CI) p-value SLPI expression 0.43 ( ) <0.001 Multivariate model SLPI expression cn-classification Extra capsular spread 0.47 ( ) 2.14 ( ) 1.93 ( ) Abbreviations: SLPI, secretory leukocyte protease inhibitor; HR, hazard ratio; CI, confidence interval 111

114 Chapter 6 DISCUSSION Biomarkers with diagnostic and prognostic value for determining lymph node metastases and predicting survival in OSCC are crucial for determining treatment planning and possible targets for personalized treatment in the future. Gene expression profiling revealed LCN2, THBS2, TACSTD2, and SLPI as genes with a strong statistically significant differential gene expression between pn+ and pn0 patients with early oral cancer. 7 Although the precise function of these genes is yet not fully understood, an explanation might be their joint role in matrix remodeling. 28 In our cohort of OSCC, LCN2, THBS2, and TACSTD2 showed no correlation with lymph node metastases or survival on protein expression level despite significant differences in staining between tumors on IHC. There are several reasons for the poor correlations between mrna and protein expression levels. First, there is the undervalued role of complicated and varied posttranscriptional and translational mechanisms, which are not yet sufficiently defined. Second, proteins differ substantially in degradation and in vivo half-lives. 29,30 Finally, both protein and mrna experiments contain a significant amount of errors and noise that limits our ability to get a clear picture. 30 A combination of one or more of these factors may explain these poor correlations, which is in line with several studies that report discrepancies between mrna and protein correlations with prognostically relevant outcomes Therefore, mrna levels cannot be used as surrogates for corresponding protein levels without validation. To our knowledge, this is the first study that shows the correlation between SLPI expression by IHC and lymph node metastases, OS, and DSS in a large cohort of patients with OSCC. Despite a significant correlation between SLPI protein expression and lymph node metastases in the whole cohort, SLPI expression has no additional diagnostic value as a predictor for lymph node metastases in a subgroup of early cancers, which are clinically lymph node-negative in this cohort of OSCC. Previous studies report different results correlating SLPI expression with lymph node metastases in head and neck squamous cell carcinoma. Westin et al 23 found no significant correlation, whereas Cordes et al 22 found a strong correlation between lower SLPI protein expression and an increased risk of lymph node metastases (p <.001). However, there are some drawbacks in comparing these studies. First, they did not analyze whether SLPI had additional value as a predictor for lymph node metastases. Second, most cancers in the Cordes study were located in the larynx, oropharynx, and hypopharynx (87.6%). This might explain the difference with our findings in oral cancer, as also for other genes, such as EGFR, pakt, and PTEN, it is known that its expressions vary between oral and oropharyngeal carcinomas

115 Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS2 Although Won et al 34 suggested initially that the difference in HPV-related pathogenesis of the tumors could be the reason for different protein expression in head and neck subsites, a later study by Hoffmann et al 35 identified SLPI expression to be an HPV-independent predictor for lymph node metastases in head and neck cancer. In addition, their cohort contained mainly laryngeal and oropharyngeal carcinomas. Another possibility for the discrepancy could be the amount of tumors with moderate/strong immunoreactivity, which, in our cohort, was 9.0% compared to 31.4% in the Cordes cohort. 22 As a result, the group of tumors with moderate/ strong immunoreactivity in our study could be too small to have additional value as a predictor for lymph node metastases. We identified SLPI as an independent predictor for OS and DSS in OSCC. Patients with low SLPI protein expression had a worse OS and DSS compared with patients with any SLPI expression, see Supplementary Figures S3 and S4. Earlier studies suggested a role for SLPI expression as a prognostic biomarker in head and neck cancer. Westin et al 23 correlated stronger SLPI expression with well-differentiated tumors in a group of 26 head and neck cancers and suggested its use as a prognostic tool, although they found no significant relationship with lymph node metastases and did not correlate its expression with survival. Alkemade et al 36 found the same significant correlation between SLPI expression and tumor differentiation in skin cancer. In addition, Wen et al 37 demonstrated inverse correlations of SLPI expression with multiple tumor invasion parameters, which suggests a protective role of SLPI against OSCC invasion. They also suggested SLPI as a potential biomarker in evaluating prognosis and treatment of the clinically lymph node-negative neck, although they did not correlate SLPI expression with lymph node metastases or survival. 6 In conclusion, this is, to our knowledge, the first study that links SLPI expression with both lymph node metastases and survival in a large cohort of patients with OSCC. Although SLPI expression is correlated with lymph node metastases in the whole cohort, it has no additional value in determining lymph node metastases in early cancers that are clinically lymph node-negative. On the other hand, SLPI seems to be an independent predictor for both OS and DSS. Therefore, SLPI IHC might be relevant as a prognostic biomarker for patients with OSCC. However, its molecular role in progression and metastasis of different head and neck cancer subsites needs further investigation. 113

116 Chapter 6 REFERENCES 1. Jemal A, Bray F, Center MM, et al. Global cancer statistics. CA Cancer J Clin 2011; 61: Incidence of invasive cancer by gender and location per year [Internet]. The Netherlands Cancer Registry Available at: Accessed April 30, Karim-Kos HE, de Vries E, Soerjomataram I, et al. Recent trends of cancer in Europe: a combined approach of incidence, survival and mortality for 17 cancer sites since the 1990s. Eur J Cancer 2008; 44: Mehanna H, West CM, Nutting C, et al. Head and neck cancer Part 2: treatment and prognostic factors. BMJ 2010; 341: c Leusink FK, van Es RJ, de Bree R, et al. Novel diagnostic modalities for assessment of the clinically nodenegative neck in oral squamous-cell carcinoma. Lancet Oncol 2012; 13: e554-e Roepman P, Wessels LF, Kettelarij N, et al. An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas. Nat Genet 2005; 37: van Hooff SR, Leusink FK, Roepman P, et al. Validation of a gene expression signature for assessment of lymph node metastasis in oral squamous cell carcinoma. J Clin Oncol 2012; 30: Lin CW, Tseng SW, Yang SF, et al. Role of lipocalin 2 and its complex with matrix metalloproteinase-9 in oral cancer. Oral Dis 2012; 18: Chong HC, Tan CK, Huang RL, et al. Matricellular proteins: a sticky affair with cancers. J Oncol 2012; 2012: Bai W, Wang L, Ji W, et al. Expression profiling of supraglottic carcinoma: PTEN and thrombospondin 2 are associated with inhibition of lymphatic metastasis. Acta Otolaryngol 2009; 129: Trerotola M, Cantanelli P, Guerra E, et al. Upregulation of Trop-2 quantitatively stimulates human cancer growth. Oncogene 2013; 32: Fong D, Spizzo G, Gostner JM, et al. TROP2: a novel prognostic marker in squamous cell carcinoma of the oral cavity. Mod Pathol 2008; 21: Abe T, Kobayashi N, Yoshimura K, et al. Expression of the secretory leukoprotease inhibitor gene in epithelial cells. J Clin Invest 1991; 87: Boudier C, Cadène M, Bieth JG. Inhibition of neutrophil cathepsin G by oxidized mucus proteinase inhibitor. Effect of heparin. Biochemistry 1999; 38: Del Rosso M, Fibbi G, Pucci M, et al. Multiple pathways of cell invasion are regulated by multiple families of serine proteases. Clin Exp Metastasis 2002; 19: Sun Z, Yang P. Role of imbalance between neutrophil elastase and alpha 1-antitrypsin in cancer development and progression. Lancet Oncol 2004; 5: Cheng WL, Wang CS, Huang YH, et al. Overexpression of a secretory leukocyte protease inhibitor in human gastric cancer. Int J Cancer 2008; 123: Choi BD, Jeong SJ, Wang G, et al. Secretory leukocyte protease inhibitor is associated with MMP-2 and MMP-9 to promote migration and invasion in SNU638 gastric cancer cells. Int J Mol Med 2011; 28: Trojan L, Schaaf A, Steidler A, et al. Identification of metastasis-associated genes in prostate cancer by genetic profiling of human prostate cancer cell lines. Anticancer Res 2005; 25: Nakamura K, Takamoto N, Hongo A, et al. Secretory leukoprotease inhibitor inhibits cell growth through apoptotic pathway on ovarian cancer. Oncol Rep 2008; 19: Dasgupta S, Tripathi PK, Qin H, et al. Identification of molecular targets for immunotherapy of patients with head and neck squamous cell carcinoma. Oral Oncol 2006; 42: Cordes C, Häsler R, Werner C, et al. The level of secretory leukocyte protease inhibitor is decreased in metastatic head and neck squamous cell carcinoma. Int J Oncol 2011; 39:

117 Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS2 23. Westin U, Nyström M, Ljungcrantz I, Eriksson B, Ohlsson K. The presence of elafin, SLPI, IL1-RA and STNFalpha RI in head and neck squamous cell carcinomas and their relation to the degree of tumour differentiation. Mediators Inflamm 2002; 11: Klein Nulent TJ, Van Diest PJ, van der Groep P, et al. Cannabinoid receptor-2 immunoreactivity is associated with survival in squamous cell carcinoma of the head and neck. Br J Oral Maxillofac Surg 2013; 51: Smeets SJ, Hesselink AT, Speel EJ, et al. A novel algorithm for reliable detection of human papillomavirus in paraffin embedded head and neck cancer specimen. Int J Cancer 2007; 121: Kishi M, Nakamura M, Nishimine M, et al. Loss of heterozygosity on chromosome 6q correlates with decreased thrombospondin-2 expression in human salivary gland carcinomas. Cancer Sci 2003; 94: Wang HJ, He XJ, Ma YY, et al. Expressions of neutrophil gelatinase associated lipocalin in gastric cancer: a potential biomarker for prognosis and an ancillary diagnostic test. Anat Rec (Hoboken) 2010; 293: Warde-Farley D, Donaldson SL, Comes O, et al. The GeneMANIA prediction server: biological network integration for gene prioritization and predicting gene function. Nucleic Acids Res 2010;38(Web Server issue): W214 W Vogel C, Marcotte EM. Insights into the regulation of protein abundance from proteomic and transcriptomic analyses. Nat Rev Genet 2012; 13: Dickson BC, Mulligan AM, Zhang H, et al. High-level JAG1 mrna and protein predict poor outcome in breast cancer. Mod Pathol 2007; 20: Hao D, Lau HY, Eliasziw M, et al. Comparing ERCC1 protein expression, mrna levels, and genotype in squamous cell carcinomas of the head and neck treated with concurrent chemoradiation stratified by HPV status. Head Neck 2012; 34: Lichtinghagen R, Musholt PB, Lein M, et al. Different mrna and protein expression of matrix metalloproteinases 2 and 9 and tissue inhibitor of metalloproteinases 1 in benign and malignant prostate tissue. Eur Urol 2002; 42: Greenbaum D, Colangelo C, Williams K, et al. Comparing protein abundance and mrna expression levels on a genomic scale. Genome Biol 2003; 4: Won HS, Jung CK, Chun SH, et al. Difference in expression of EGFR, pakt, and PTEN between oropharyngeal and oral cavity squamous cell carcinoma. Oral Oncol 2012; 48: Hoffmann M, Quabius ES, Tribius S, et al. Human papillomavirus infection in head and neck cancer: the role of the secretory leukocyte protease inhibitor. Oncol Rep 2013; 29: Alkemade HA, van Vlijmen Willems IM, van Haelst UJ, et al. Demonstration of skin-derived antileukoproteinase (SKALP) and its target enzyme human leukocyte elastase in squamous cell carcinoma. J Pathol 1994; 174: Wen J, Nikitakis NG, Chaisuparat R, et al. Secretory leukocyte protease inhibitor (SLPI) expression and tumor invasion in oral squamous cell carcinoma. Am J Pathol 2011; 178:

118 Chapter 6 SUPPLEMENTARY INFORMATION Supplementary Figure 1 Scatter plot for interrator concordance The level of interrater concordance was high, with a Spearman s rank correlation of (p<.001) for continuous data and a Kappa of (95% confidence interval, ; p<.001) for categorical data. Supplementary Figure 2 Cutoff point determination by ROC curve analysis Optimal cutoff points revealed that gene expression is a significant predictor of lymph node metastasis for all 4 genes, see also Table

119 Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS2 6 Supplementary Figure 3 Cutoff point determination by ROC curve to correlate protein expression and LNM in ct1-2n0 OSCC Optimal cut-off points could not be determined by ROC curve analysis. Protein expression of LCN2, TACSTD2, THBS2 and SLPI revealed no significant correlation with lymph node metastases in a subgroup of ct1-t2n0 tumors (see also Table 4). Supplementary Figure 4 Cutoff point determination by ROC curve to correlate SLPI protein expression and survival Determined cutoff points used in Kaplan Meier analysis revealed a significant difference between SLPI expression for both OS and DSS, see also Figure 3 and Supplementary Table

120 Chapter 6 Supplementary Table S1 ROC-curves for gene expression and LNM. ROC curve No. of patients AUC (95% CI) p-value SLPI gene expression ( ) LCN2 gene expression ( ) TACSTD2 gene expression ( ) THBS2 gene expression ( ) Abbreviations: LNM, lymph node metastasis; AUC, area under curve; CI, confidence interval; SLPI, secretory leukocyte protease inhibitor; THBS2, thrombospondin-2; LCN2, lipocalin-2; TACSTD2, tumor-associated calcium signal transducer 2 Supplementary Table S2 ROC-curves for protein expression and LNM. ROC curve No. of patients AUC (95% CI) p-value SLPI protein expression ( ) LCN2 protein expression ( ) TACSTD2 protein expression ( ) THBS2 protein expression ( ) Abbreviations: LNM, lymph node metastasis; AUC, area under curve; CI, confidence interval; SLPI, secretory leukocyte protease inhibitor; THBS2, thrombospondin-2; LCN2, lipocalin-2; TACSTD2, tumor-associated calcium signal transducer 2 Supplementary Table S3 ROC-curves for SLPI expression and survival. ROC curve No. of patients AUC (95% CI) p-value SLPI and OS ( ) SLPI and DSS ( ) Abbreviations: AUC, area under curve; CI, confidence interval; SLPI, secretory leukocyte protease inhibitor; OS, overall survival; DSS, disease specific survival 118

121 Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS

122

123 CHAPTER 7 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma F.K.J. Leusink P.J. van Diest* M.H. Frank* R. Broekhuizen W. Braunius S.R. van Hooff S.M. Willems* R. Koole* * Equal contribution Pathobiology 2015;82:58-67

124 Chapter 7 ABSTRACT OBJECTIVE Oral squamous cell carcinoma (OSCC) has still a poor prognosis. Lymph node metastasis is a major determinant of treatment decisions and prognosis. Serine protease inhibitor Kazaltype 5 (SPINK5) is the inhibitor of kallikrein 5 (KLK5) and kallikrein 7 (KLK5). SPINK5, KLK5, and KLK7 are three of the genes of a recently validated gene expression profile that predicts lymph node metastasis in OSCC. This study evaluates the clinicopathological role and their value as biomarkers in OSCC. METHODS 83 patients with primary OSCC, treated surgically between 1996 and 2000, were included. Gene expression data were acquired in a previous reported study. HPV status was determined by an algorithm for human papillomavirus type 16. Protein expression for KLK5, KLK7 and SPINK5 was semi-quantitatively determined in all 83 tumors by immunohistochemistry (IHC). All expression data were correlated with clinicopathological parameters. RESULTS Concurrent loss of KLK5 and KLK7 correlates with a worse disease specific and overall survival. Multivariate analysis proofed that co-expression is an independent prognostic factor for disease specific (p=0.029) and overall survival (p=0.001). CONCLUSION This report demonstrates that concurrent loss of KLK5 and KLK7 associates with a poor clinical outcome in OSCC and could therefore serve as prognostic marker in OSCC. KEYWORDS Oral squamous-cell carcinoma, KLK5, KLK7, SPINK5, Survival 122

125 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma INTRODUCTION Oral squamous cell carcinoma (OSCC) is the most common head and neck tumor type worldwide 1 with tobacco smoking, betel nut chewing, alcohol consumption, and human papillomavirus (HPV) infection as major risk factors. 2,3 Despite advances in surgery, radiation, and chemotherapy, prognosis of OSCC patients remains poor with a 5-year survival rate of approximately 50%. Lymph node metastasis of OSCC is major determinant of prognosis and treatment but can be difficult to detect with current diagnostic modalities. 4 Approximately 30-40% of lymph node metastases are left undetected in this population and will develop into overt neck disease during follow-up when left untreated. Therefore, treatment of the neck is recommended even when the tumor has been classified as clinically node negative. This results in unnecessary treatment and morbidity in 60-70% of patients with OSCC. Additional biological markers for occult metastasis may improve staging. 4 Human tissue kallikreins, a family of 15 secreted serine proteases encoded by a multigene cluster (KLK genes) on chromosome 19q13.47, may be such biomarkers. 5 These genes play different functional roles like semen liquefaction by digestion of Seminigelin by the most commonly known kallikrein (KLK3), better known as prostate-specific antigen (PSA) 6,7 and skin desquamation by cleaving corneodesmosomes by kallikrein 5 (KLK5) and kallikrein 7 (KLK7). 8 Kallikreins have been implicated in different cancer types 5,6,8-10 and often coexpression of KLK5 and KLK7 has been reported. 9,11,12 Additionally, in OSCC, KLK5, KLK7, KLK8 and KLK10 were reported to be over-expressed, 13 however correlations with clinical and pathological parameters (a.o. nodal disease) were not analyzed. 7 Lymphoepithelial Kazal-type-related inhibitor (LEKTI), the product of the gene encoding for serine protease inhibitor kazal type 5 (SPINK5) inhibits KLK5 and KLK In head and neck squamous cell carcinomas (HNSCC), SPINK5 expression is downregulated both at the mrna and protein level. 17,18 Moreover, SPINK5, KLK5 and KLK7 are all part of the signature genes for predicting lymph node metastases in HNSCC 19,20 including OSCC. 21 Therefore, we further characterized the value of KLK5, KLK7 and SPINK5 in primary OSCCs as biomarkers of lymph node metastases and prognosis. MATERIALS AND METHODS Patients and tissue samples The source population consisted of patients with a histologically confirmed HNSCC, primary treated by surgery and radiotherapy (on indication) between 1996 and 2000 in the University Medical Center Utrecht, as shown in table 1a and as previously described in an earlier reported 123

126 Chapter 7 gene expression profiling study. 20 Tissues were used in line with the code Proper Secondary Use of Human Tissue as installed by the Dutch Federation of Biomedical Scientific Societies. 22 Table 1a Clinical characteristics of the included OSCC patients. All cases Gender Female Male Age at diagnosis Median (range) No. (%) 83 (100) (43) (57) (44) (56) (37-87) Smoking history Current smoker or ceased < 1 year Ex smoker, ceased > 1 year Never smoker (70) (11) (18) Alcohol consumption 5 U/day 1-4 U/day Occasionally Never Clinical T-stage ct1 ct2 ct3 ct4 Clinical N-Stage cn0 cn1-3 Sub-site tongue floor of mouth buccal cavity gum Mean follow-up (months) 45 (23) (34) (20) (23) (16) (37) (10) (37) (64) (36) (36) (42) (12) (10) 124

127 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma Table 1b Pathological characteristics of the included OSCC patients. All cases Pathological T-Stage pt1 pt2 pt3 pt4 Pathological N-Stage pn0 pn1-3 Stage grouping I II III IVA-IVB Infiltration depth 4.0mm < 4.0mm Differentiation grade Good / Moderate Poor / Undifferentiated Keratinization Present Absent Missing No. (%) 83 (100) (20) (33) (12) (35) (46) (54) (17) (11) (26) (46) (87) (13) (81) (19) (72) (24) (4) 7 Vaso-invasion Present Absent Missing Bone-invasion Present Absent Perineural growth Present Absent Missing Spidery growth Present Absent High risk HPV status positive negative (22) (75) (3) (30) (70) (41) (47) (12) (78) (22) (0) (100) 125

128 Chapter 7 Study inclusion was based on: 1) oral cavity as primary tumor site, 2) HPV-16-status, excluding positive tumors, 3) follow-up time of more than five years and 4) no patients with a synchronous primary tumor or a previous malignancy in the head and neck region. Tumors were selected on the availability of representative formaldehyde-fixed, paraffin-embedded (FFPE) tissue blocks and frozen tissue samples of the primary tumor. Twenty-one patients were excluded due to their primary tumor originating from the orofarynx. HPV-16 positivity was determined in the remaining OSCC patients which resulted in no further exclusions. In total, eighty-three tissue samples met the inclusion criteria. All involved investigators, apart from the study statistician, were blinded to patient outcome throughout all analyses. The workflow of this study is illustrated in figure 1. Figure 1 Schematic representation of the work flow of this study. Previous studies resulted in the discovery and validation of a multi-gene signature In this study, gene expression data 20 were used to correlate the selected genes KLK5, KLK7 and their cognate inhibitor SPINK5 with clinical and histopathological parameters. From the same cohort of tumour samples, a TMA was constructed for immunohistochemical analysis of the selected genes to correlate their protein expression with clinical and histopathological parameters, and outcomes were compared. Representative hematoxylin and eosin (H&E) slides of 83 OSCC were selected and reviewed by two experienced head and neck pathologists (PJvD and SMW), as presented in table 1b. A tissue microarray (TMA) was then made of the FFPE tissue blocks. Of each block, two central and two 126

129 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma peripheral tissue cylinders with a diameter of 0.6 millimetres were punched out, avoiding areas of necrosis, and arrayed in a recipient paraffin block. Normal epithelium from the floor of mouth, gingiva and tonsil tissue was incorporated in each block to ensure similarity of staining between the different blocks, and to compare tumor staining with staining of normal epithelium. KLK5, KLK7 and SPINK5 gene expression analysis Genome-wide gene expression was measured using dual-channel microarrays with a pool of tumor samples as a reference as described earlier. 20 KLK5, KLK7 and SPINK5 were each represented by a single, unique feature on this array. The expression data was normalized for dye and print-tip biases using loess per print-tip normalization and subsequently variance stabilization was applied. 20 The data and protocols used are publicy available (ArrayExpress E-UMCU-11). Detection of Human papillomavirus type 16 Human papillomavirus type 16 (HPV-16) active tumors were determined by a validated algorithm for detection of HPV-16 in paraffin embedded head and neck cancer specimen: 23,24 all tumors were stained for p16 by IHC followed by GP 5+/6+ PCR when p16 staining was positive. Immunohistochemistry TMAs were stained for KLK5, KLK7 and SPINK5 according to the manufacturer s protocol and as previously described. 25 In brief, 4 micrometer thick paraffin sections were deparaffinised. After blocking of endogenous peroxidase and antigen retrieval by pepsin the TMA slides were incubated for 60 minutes with different dilutions of the primary antibodies: KLK5 (AF1108, R&D Systems, Oxon, England; 1:100), KLK7 (AF2624, R&D Systems, Oxon, England; 1:100), and SPINK5 (sc32330, clone nr 1C11G6, Santa Cruz Biotechnology, Santa Cruz, USA; 1:100). Finally slides were counterstained with haematoxylin for 5 minutes, followed by dehydration and mounting. 7 Evaluation of expression was performed independently by 2 observers (PJvD and FKJL), who were both blinded to patient characteristics. In case of disagreement, the observers reanalyzed the staining results until they reached consensus. Cores were considered lost if less than 10% of the core contained tumor ( sampling error ) or when less than 10% of tissue was present ( absent core ). Patients were excluded if more than 2 cores per case were lost. When two or more cores were available from one patient, the mean score was calculated for that patient. To determine the score for each TMA-core, appropriate controls of normal squamous epithelium were used. Protein expression was scored for both its intensity in tumor cells relative to normal epithelium (normal = 2, weaker = 1, total loss = 0) and the percentage of tumor cells in the tissue section with such a specific intensity. The product of these two scoring variables resulted in a scoring range of 0-200, in which a score of 0 represents a complete loss of protein expression in 127

130 Chapter 7 all tumor cells and a score of 200 represents normal expression throughout the tumor, see figure 2. Co-expression of both KLKs was defined as absent (0) if the sample showed complete loss of expression for both KLK5 and KLK7, otherwise as non-absent (1). Figure 2 KLK5, KLK7 and SPINK5 expression in OSCC and normal mucosa. Representative stainings of the TMA, consisting of 83 OSCC cases, are presented. Staining scores were calculated by the product of intensity (normal = 1, strong = 2) and the proportion of stained tumour cells (%). SPINK5 staining: normal mucosa ( a ), no staining ( b ), OSCC score: normal intensity (1) proportion (100%) = 100 ( c ), OSCC score: 2 75% = 150 ( d ). KLK5 staining: normal mucosa ( e ), no staining ( f ), OSCC score: 1 75% = 75 ( g ), OSCC score: 2 50% = 100 ( h ). KLK7 staining: normal mucosa ( i ), no staining ( j ), OSCC score: 1 50% = 50 ( k ), OSCC score: 2 40% = 80 ( l ). Statistical analysis The Mann-Whitney test was used to determine differences in expression between several clinicopathological parameters. OS was defined as the length of the time interval from surgery to death from any cause. DSS was defined as the time interval from surgery to death due to disease. For univariate survival analysis we dichotomized KLK5, KLK7 and SPINK5 expression in absent (0) and non-absent (1-200). Kaplan-Meier curves were plotted to visualize difference in survival between these two groups. Log rank test was used to obtain statistical significance. Prognostic value was examined by univariate and multivariate analyses using the Cox proportional hazards regression model. All p-values were based on two-tailed statistical analysis and p < 0.05 was 128

131 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma considered to be significant. Statistical analysis was performed using the SPSS 20.0 statistical package (SPSS Inc, Chicago, IL). RESULTS Baseline characteristics Clinical data and survival was retrieved from the medical records, with follow-up times in the range of 0 to 15 years, median 45 months, see table 1a and suppl. figure 1a. Kaplan-Meier plots on survival in patient groups stratified by gender, nodal status (pn) and tumor stage are shown in suppl. figures 1b-d. As expected, tumor stage and nodal status represented the clinical factors that exhibited the strongest association with overall survival. Detection of Human papillomavirus type 16 Human papillomavirus type 16 (HPV-16) active tumors were determined likewise a recently validated algorithm. 23,24 Of all 83 included OSCC in the study only 3 tumors showed strong expression for p16. However, in none of these 3 OSCCs HPV-16 positivity was detected by GP 5+/6+ PCR. 7 Gene expression and clinicopathological parameters Analysis of 83 OSCC shows a statistically significant differential gene expression between lymph node positive and lymph node negative patients for KLK7 (p=0.020), not for KLK5 and SPINK5. OSCC lymph node positive patients have significantly less KLK7 mrna, see suppl. figure 2. Analysis of the other parameters in tables 1a and b showed significant less KLK5, KLK7 and SPINK5 mrna in alcohol consuming patients, and less KLK5 and KLK7 mrna in tobacco using patients and in tumors showing absence of keratinization, see table 2. Gene expression and survival To illustrate the prognostic impact of the gene expression of KLK5, KLK7 and SPINK5 we performed a Kaplan-Meier analysis with samples dichotomized into two groups with expression levels < median and levels > median for each respective gene. Supplementary figure 3a shows the results for the 3 genes in the complete OSCC cohort. Furthermore, we performed univariate Cox regression analysis of the three genes, see suppl. table 1. None of the genes had independent prognostic association with survival. 129

132 Chapter 7 Table 2 Correlations between gene (mrna) and protein (IHC) expression of KLK5, KLK7 and SPINK5 and clinical and pathological parameters of the included OSCC cohort (n=83). Cases were stratified according to clinical and pathological characteristics. Smoking history was dichotomized to current smoker or ceased < 1 year versus ex-smoker (ceased > 1 year) and never smoker. Alcohol consumption was dichotomized to 1-4 or > 5 U/day versus occasionally or never. Clinical and pathological nodal status (cn and pn) were dichotomized to cn0 versus cn+ and to pn0 versus pn+. Differentiation was dichotomized to well and moderate versus poor and undifferentiated. Keratinization was dichotomized to present versus absent. P-values represent the Mann-Whitney U test of these comparisons. p-value KLK5 KLK7 SPINK5 Clinical characteristic mrna IHC mrna IHC mrna IHC smoking history * 0.067* alcohol consumption NS cn status NS NS NS NS Pathological characteristic pn status NS NS NS NS NS differentiation NS NS NS keratinization * Abbreviations: NS is non significant, mrna is messenger RNA, IHC is immunohistochemistry. P-values < 0.05 were considered statistically significant. * P-values >0.05 and <0.10 were considered a clinically relevant trend. Immunohistochemistry: descriptive analysis A total of 226 (68%) tumor cores stained with KLK5 antibody, 224 tumor cores (67%) stained with KLK7and 221 tumor cores (67%) stained with SPINK5, were available for analysis. Due to our inclusion criteria (> than 2 tumor cores available per case) 14 cases were missing for KLK5, 15 for KLK7 and 17 for SPINK5. The majority of the OSCCs in this cohort showed (almost) total loss of expression for KLK5 (71%), for KLK7 (59%) and for SPINK5 (59%), see suppl. table 2. Immunohistochemistry and clinicopathological parameters Lower protein expression levels of KLK5 and KLK7 in the TMA cohort were significantly associated with a positive smoking history and alcohol consumption. Furthermore, there was a comparable trend for SPINK5 protein expression and a positive smoking history. Current smokers or patients that ceased smoking < 1 year showed significantly less expression of KLK5, KLK7 versus patients that ceased > 1 year or never smokers (p=0.009;p=0.011). Likewise, patients that consumed 1-4 or > 5 alcohol U/day showed significantly more loss of KLK5 and KLK7 protein expression compared to patients that consumed alcohol occasionally or never (p=0.001; p=0.012). Correlating protein expression with the pathological characteristics (table 1b), statistical analysis showed significant more loss of KLK5, KLK7 and SPINK5 expression in patients with OSCC that were poorly or undifferentiated (p=0.021; p=0.013; p=0.003) and in patients with OSCC that were classified as keratinization absent (p=0.004; p=0.001; p=0.028), see table

133 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma Immunohistochemistry and survival A Cox regression model was applied to identify any prognostic relevance for protein expression of KLK5, KLK7 and SPINK5. Furthermore we added co-expression of KLK5 and KLK7 to this model. All protein expressions were dichotomized into no expression (score 0) versus expression (score 1). Co-expression of KLK5 and KLK7 scored only 0 if both kallikreins scored 0. Univariate analysis indicated a significant prognostic impact in overall survival in OSCC of KLK5 protein expression (HR=0.41, CI , p=0.017) and of KLK7 protein expression (HR=0.43, CI , p=0.016), see table 3a. SPINK5 protein expression did not show any prognostic impact on overall survival. Kaplan-Meier survival plots are shown in suppl. figure 3b. Table 3a Univariate and multivariate OS Cox regression model for protein expression of KLK5, KLK7, co-expression of KLK5 and KLK7. Co-expression was dichotomized into no expression (score 0) versus expression (score 1). univariate HR 95% CI p-value Age* Stage* pn status* KLK5 expression KLK7 expression co-expression KLK5 KLK multivariate co-expression KLK5 KLK corrected for age OS = overall survival, HR = hazard ratio, 95% CI = 95% confidence interval, p-value of the Cox regression model * age: <60 vs >60 years, tumor stage I, II vs III, IV, pn status pn0 vs pn+ 7 Table 3b Univariate and multivariate DSS Cox regression model Protein expression of KLK5, KLK7, co-expression of KLK5 and KLK7 was dichotomized into no expression (score 0) versus expression (score 1). univariate HR 95% CI p-value Age* Stage* pn status* KLK5 expression KLK7 expression co-expression KLK5 KLK multivariate co-expression KLK5 KLK corrected for pn status DSS = disease specific survival, HR = hazard ratio, 95% CI = 95% confidence interval, p-value of the Cox regression model * age: <60 vs >60 years, tumor stage I, II vs III, IV, pn status pn0 vs pn+ 131

134 Chapter 7 The prognostic impact on overall survival of co-expression of KLK5 and KLK7 was even stronger than the impact of KLK5 or KLK7 alone (HR=0.36, CI , p=0.005), reason to apply a multivariate Cox regression model with co-expression of KLK5 and KLK7 together with known prognostic markers like age, stage and pn status. Independent of stage and pn-status but corrected for age co-expression showed significant impact on overall survival (HR=0.32, CI , p=0.002), see table 3a, figure3b. In this TMA cohort co-expression of KLK5 and KLK7 showed 2- and 5- year survival rates of 47% and 24% for patients with complete loss of coexpression (score 0) and 80% and 58% for patients with co-expression (score 1), respectively figure3a. Figure 3 Survival plots of all OSCC patients (n = 83). Patients were stratified according to the co-expression level of KLK5 and KLK7: 0 or no protein expression of both KLKs (red) versus 1 400, the sum of both KLK protein expression (blue). OS plotted by Kaplan-Meier analysis ( a ) and Cox regression analysis ( b ). DSS plotted by Kaplan-Meier analysis ( c ) and Cox regression analysis ( d ). a, c The log-rank test of this group comparison is represented, and the p values therefore differ from the significance levels of the Cox regression analysis (b, d). 132

135 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma With regard to disease specific survival, patients with complete loss of co-expression had a shorter disease specific survival than patients with co-expression (table 3b, figure 3c-d); 2-year survival rate: 53% vs. 82%, 5-year survival rate: 36% vs. 74%, HR=0.34, CI , p=0.011). Multivariate analysis showed independent significant impact of co-expression on disease specific survival corrected for pn status (table 3b; HR=0.36, CI , p=0.022). Follow up time of patients without an event was 80.3 months (median 71.4). DISCUSSION The aim of the present study was to evaluate the role of the proteases KLK5 and KLK7 as biomarkers of lymph node metastases and prognosis in OSCC. We show that only underexpression of KLK7 mrna correlates with lymph node metastasis in OSCC patients which is in line with the gene expression signature predicting lymph node metastasis in OSCC 21 containing a.o. KLK5, KLK7 and SPINK5. However, KLK5 and SPINK5 do not correlate with LNM in this cohort. The majority of the OSCCs had total to almost total loss of expression of KLK5 (71%), KLK7 (59%) and SPINK5 (59%). Our observation concerning SPINK5 expression is consistent with 2 previous studies that demonstrated down-regulation of SPINK5 mrna expression in HNSCC. In the first study a very small panel of 3 oropharynx tumors was investigated, 17 whereas the second study obtained genome-wide transcriptomic profiles of 53 primary oral tongue SCC. 18 The second study however, focused on identifying specific associated genes with oral tongue SCC and did not use and provide complete clinical and pathological data of the patients and samples tested. 7 Next, our observation concerning (almost) total loss of KLK5 and 7 expression seems, at first sight, to be in controversy with a previous study that reported abundant protein expression of KLK5, 7, 8 and 10 in 50 OSCCs. 13 However, analysis of sub-site and differentiation grade reveals that 58% of the included samples by Pettus originate from the oral tongue (vs. 39% in our cohort, table 1a) and that 61% is well, 30% moderately and 9% poorly differentiated (vs. 4%, 77% and 17%, table 1b). Floor of mouth tumors showed lower expression than tongue tumors (data not shown). Furthermore, poorly and undifferentiated tumors had lower expression of both KLK5 and KLK7. These clinical and pathological differences of included samples of both studies point out the influence of the composition of the studied population on the result. From this point of view the conclusion as stated by others 13 that mostly well differentiated OSCCs originating from the tongue have abundant expression of KLK5 and KLK7 is actually in agreement with our observation. 133

136 Chapter 7 In an attempt to investigate the potential functional role of KLK5, KLK7 and SPINK5 in OSCC, we analyzed the expression of all three together in OSCC. KLK5 and 7 play important roles in skin desquamation where shedding of corneocytes happens due to proteolytic processing of junctional structures like corneodesmosin. 5,26 SPINK5 regulates this desquamation by inhibiting KLK5 and 7 and loss of SPINK5, in Netherton syndrome for instance, results in unregulated proteolysis in the outer layers of the epidermis, causing defective barrier function. 26 Normal oral squamous cells have high expression of KLK5, KLK7 and SPINK5 (figure 2d, 2h and 2l) suggesting a similar role for these proteases and their inhibitor in the normal oral mucosa. Our data in table 2 show that OSCCs that are able to produce keratin or keratin whirls (i.e. figure 2f; high KLK7 expression) have a significant higher expression of KLK5, KLK7 and SPINK5. It is interesting to speculate that the role of KLK5 and 7 and the balance with their inhibitor SPINK5 might be related and limited just to the process of keratinization. At least, we did not observe any uncontrolled KLK activity in OSCCs with low SPINK5 expression, nor could we correlate this to tumor progression phenomena like lymph node metastasis. Concurrent loss of KLK5 and KLK7 predicted overall survival in univariate and multivariate age and lymph node status corrected survival analysis. Correction of the OS model by age resulted in a decrease of the HR to 0.29 which means that older patients with a positive co-expression have a better survival than younger patients with complete loss of expression of both KLK5 and KLK7. Correction of the DSS model by pn status resulted in an increase of the HR to 0.39 which means that patients with a positive co-expression survive better than patients without, however, this difference in survival decreased when the patient has a positive pathologic nodal status. To date, no single molecular marker has been able to predict occult lymph node metastasis of OSCC patients reliably. Recently, a gene expression signature predicting lymph node metastasis has been validated 21 and combined with a sentinel lymph node biopsy procedure 4 it could even be more helpful in decision making with regard to the clinically node negative neck. Whether loss of SPINK5 tips the balance to more protease activity by KLK5 and 7 and whether such an imbalance may identify patients with aggressive tumor behavior cannot be confirmed by our data and remains to be elucidated. However, loss of co-expression of KLK5 and 7 does identify OSCC patients with poor clinical outcome and may therefore act as a prognostic marker. ACKNOWLEDGMENTS SMW is funded by the Dutch Cancer Society (clinical fellowship: ). 134

137 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma REFERENCES 1. Jemal A, Bray F, Center MM, Ferlay J, et al. Global cancer statistics. CA Cancer J Clin 2011; 61: Ang KK, Harris J, Wheeler R, et al.human papillomavirus and survival of patients with oropharyngeal cancer. N Engl J Med 2010; 363: Leemans CR, Braakhuis BJ, Brakenhoff RH. The molecular biology of head and neck cancer. Nat Rev Cancer 2011; 11: Leusink FK, van Es RJ, de Bree R, et al. Novel diagnostic modalities for assessment of the clinically nodenegative neck in oral squamous-cell carcinoma. Lancet Oncol 2012; 13: e Borgono CA, Diamandis EP. The emerging roles of human tissue kallikreins in cancer. Nat Rev Cancer 2004; 4: Emami N, Diamandis EP. Utility of kallikrein-related peptidases (klks) as cancer biomarkers. Clin Chem 2008; 54: Lilja H, Ulmert D, Vickers AJ. Prostate-specific antigen and prostate cancer: Prediction, detection and monitoring. Nat Rev Cancer 2008; 8: Sotiropoulou G, Pampalakis G, Diamandis EP. Functional roles of human kallikrein-related peptidases. J Biol Chem 2009; 284: Dong Y, Kaushal A, Brattsand M, et al. Differential splicing of klk5 and klk7 in epithelial ovarian cancer produces novel variants with potential as cancer biomarkers. Clin Cancer Res: 2003; 9: Yousef GM, Polymeris ME, Yacoub GM, et al. Parallel overexpression of seven kallikrein genes in ovarian cancer. Cancer Res 2003; 63: Li X, Liu J, Wang Y, et al. Parallel underexpression of kallikrein 5 and kallikrein 7 mrna in breast malignancies. Cancer Sci 2009; 100: Talieri M, Devetzi M, Scorilas A, et al. Evaluation of kallikrein-related peptidase 5 expression and its significance for breast cancer patients: Association with kallikrein-related peptidase 7 expression. Anticancer Res 2011; 31: Pettus JR, Johnson JJ, Shi Z, et al. Multiple kallikrein (klk 5, 7, 8, and 10) expression in squamous cell carcinoma of the oral cavity. Histol Histopathol 2009; 24: Deraison C, Bonnart C, Lopez F, et al. Lekti fragments specifically inhibit klk5, klk7, and klk14 and control desquamation through a ph-dependent interaction. Mol Biol Cell 2007; 18: Egelrud T, Brattsand M, Kreutzmann P, et al. Hk5 and hk7, two serine proteinases abundant in human skin, are inhibited by lekti domain 6. Br J Dermatol 2005; 153: Schechter NM, Choi EJ, Wang ZM, et al. Inhibition of human kallikreins 5 and 7 by the serine protease inhibitor lympho-epithelial kazal-type inhibitor (lekti). Biol Chem 2005; 386: Gonzalez HE, Gujrati M, Frederick M, et al. Identification of 9 genes differentially expressed in head and neck squamous cell carcinoma. Arch Otolaryngol Head Neck Surg 2003; 129: Ye H, Yu T, Temam S, et al. Transcriptomic dissection of tongue squamous cell carcinoma. BMC genomics 2008; 9: Roepman P, Kemmeren P, Wessels LF, et al. Multiple robust signatures for detecting lymph node metastasis in head and neck cancer. Cancer Res 2006; 66: Roepman P, Wessels LF, Kettelarij N, et al. An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas. Nat Gen 2005; 37: van Hooff SR, Leusink FK, Roepman P, et al. Validation of a gene expression signature for assessment of lymph node metastasis in oral squamous cell carcinoma. J Clin Oncol 2012; 30: van Diest PJ. No consent should be needed for using leftover body material for scientific purposes. BMJ 2002; 325: Rietbergen MM, Leemans CR, Bloemena E, et al. Increasing prevalence rates of hpv attributable oropharyngeal squamous cell carcinomas in the netherlands as assessed by a validated test algorithm. Int J Cancer 2013; 132: Smeets SJ, Hesselink AT, Speel EJ, et al. A novel algorithm for reliable detection of human papillomavirus in paraffin embedded head and neck cancer specimen. Int J Cancer 2007; 121:

138 Chapter Noorlag R, van der Groep P, Leusink FK, et al. Nodal metastasis and survival in oral cancer: Association with protein expression of slpi, not with lcn2, tacstd2, or thbs2. Head Neck 2015; 37: Descargues P, Deraison C, Bonnart C, et al. Spink5-deficient mice mimic netherton syndrome through degradation of desmoglein 1 by epidermal protease hyperactivity. Nat Gen 2005; 37:

139 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma SUPPLEMENTARY INFORMATION Supplementary Figure 1a Survival plot of all OSCC patients (n=83) included in this study. Each mark (+) represents a censored event. The first censored event occurs after 5 years of follow-up. 7 Supplementary Figure 1b gender Kaplan-Meier survival plots of all OSCC patients (n=83) included in the expression analysis in this study. Patients were stratified in accordance to gender, nodal status (pn) and stage (Fig 1b-d). Group comparison was performed using log-rank tests. For supplementary figure 1c and 1d see next page. 137

140 Chapter 7 Supplementary Figure 1c Nodal status Supplementary Figure 1d Stage I/II vs. III/IV 138

141 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma Supplementary Figure 2 Boxplot showing differential expression for nodal status. Cases were stratified according nodal status; pn0 (blue) versus pn+ (red). P-values represent the Logistic regression test of this group comparison and therefore differ from the significance levels of the Mann- Whitney analysis in table 2. 7 Supplementary Figure 3a KLK5 139

142 Chapter 7 Supplementary Figure 3a KLK7 Supplementary Figure 3a SPINK5 Kaplan-Meier survival plots of all OSCC patients (n=83) included in the expression analysis in this study. Cases were stratified according KLK5, KLK7 and SPINK5 mrna expression levels < median (red) versus > median (blue). P-values represent the Log-rank test of this group comparison and therefore differ from the significance levels of the Cox-regression analysis, see suppl. table

143 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma Supplementary Figure 3b KLK5 Supplementary Figure 3b KLK

144 Chapter 7 Supplementary Figure 3b SPINK5 Kaplan-Meier survival plots including all OSCC patients of the TMA cohort (n=83). Cases were stratified according to protein expression of KLK5, KLK7 and SPINK5 dichotomized to the product score (intensity x proportion) 0 versus P-values represent the Log-rank test of this group comparison and therefore differ from the significance levels of the Cox-regression model, see table 3a. Supplementary Table S1 Univariate Cox regression survival analysis for mrna expression of KLK5, KLK7 and SPINK5 in the OSCC cohort (n=83). univariate Cox regression analysis gene HR 95% CI p-value KLK KLK SPINK HR = hazard ratio, p-value of the Cox regression analysis, 95% CI = 95% confidence interval 142

145 The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma Supplementary Table S2 Variable KLK5 KLK7 SPINK5 co-expression KLK5 & KLK7 Number (%) of 332 cores Tumor 226 (68) 224 (67) 221 (67) No tumor 49 (15) 54 (16) 55 (17) No core 57 (17) 54 (16) 56 (17) Number of cores / case (n=83) Score (%) / case (n=83) 0 (total loss) 37 (45) 19 (23) 21 (25) 17 (20) 1-50 (almost total loss) 22 (26) 30 (36) 28 (34) (partial loss) 10 (12) 19 (23) 17 (20) missing 14 (17) 15 (18) 17 (20) 16 (19) >1 50 (60) Immunohistochemical descriptive results of protein expression of KLK5, KLK7 and SPINK5 in the OSCC TMA cohort (n=83). Percentages may not total 100 because of rounding

146

147 CHAPTER 8 Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma F.K.J. Leusink E. Koudounarakis* M.H. Frank* R. Koole P.J. van Diest S.M. Willems * Equal contribution Submitted to the BMC Cancer

148 Chapter 8 ABSTRACT BACKGROUND Lymph node metastasis is a major determinant of prognosis and treatment planning of oral squamous-cell carcinoma (OSCC). Cysteine cathepsins constitute a family of proteolytic enzymes with known role in the degradation of the extracellular matrix. Involvement in pathological processes, such as inflammation and cancer progression, has been proved. The aim of the study was to discover the clinicopathological and prognostic implications of cathepsin K (CTSK) expression in oral squamous cell carcinoma. METHODS Eighty-three patients with primary OSCC, treated surgically between 1996 and 2000, were included. Gene expression data were acquired from a previously reported study. Human papilloma virus (HPV) status was previously determined by an algorithm for HPV-16. Protein expression for CTSK was semi-quantitatively determined by immunohistochemistry. Expression data were correlated with various clinicopathological variables. RESULTS Elevated gene and protein expression of CTSK were strongly associated to lymph node metastasis and perineural invasion (p<0.01). Logistic regression analysis highlighted increased CTSK protein expression as the most significant independent factor of lymphatic metastasis (OR=7.65, CI: , p=0.001). Survival analysis demonstrated both gene and protein CTSK expression as significant indicators of poor 5-year disease specific survival (HR=2.29, CI: , p=0.047 for gene expression; HR=2.79, CI: , p=0.045 for protein expression). CONCLUSION Upregulation of CTSK seems to be associated with high incidence of lymphatic spread and poor survival in OSCC. CTSK could therefore serve as a predictive biomarker for OSCC. KEYWORDS Oral squamous-cell carcinoma; cathepsin K; lymph node metastasis; prognosis 146

149 Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma BACKGROUND Oral squamous-cell carcinoma (OSCC) constitutes the most common malignancy of the head and neck region. 1 Lymph node metastasis has been shown to be the most significant, independent prognostic factor and is related to a decrease of the 5-year survival rate by 50%. 2 Thus, revealing the presence of occult metastasis is of the utmost importance for early and proper management of the neck. Variable imaging studies have been used for this purpose, including ultrasound combined with fine needle aspiration cytology, computed tomography, magnetic resonance imaging and, more recently, positron emission tomography, with variable results. 3 Moreover, the sentinel node procedure has been currently adopted by some oncological centers and embodied in the staging algorithm of early OSCC. However, its greater disadvantage is that the patient undergoes an interventional procedure. In the context of molecular biology, a significant amount of research has been focused during the last decades on biomarkers that may have additional diagnostic value. 4,5 Cathepsin K (also known as cathepsin O2), encoded by the CTSK gene on chromosome 1q21, is one of the 11 lysosomal protein degradation enzymes called cysteine cathepsins, which participate in a considerable number of physiological processes, including MHC-II-mediated antigen presentation, bone remodeling, keratinocyte differentiation and prohormones activation. 6 It is the most potent mammalian collagenase and is highly expressed in osteoclasts and in most epithelial cells. Cathepsin K is the sole matrix-degrading enzyme for which a fundamental role in bone resorption has been unequivocally documented in mice and humans. 6 However, increased expression of this lysosomal enzyme is also observed in various pathological conditions, such as neurological disorders, inflammatory diseases and cancer. Its role in cancer progression and invasion, mainly through degradation of and remodeling in the tumor microenvironment, is supported by several experimental studies and clinical reports in various types of tumors. 7 In the present study, the correlation between the expression of cathepsin K and clinicopathological variables, particularly lymph node metastasis, was examined. 8 METHODS Patients and tissue samples The study included 83 consecutive patients with OSCC who were diagnosed and surgically treated at the University Medical Center in Utrecht, The Netherlands, between 1996 and 2000, described in an earlier reported gene expression profiling study. 4 The work-flow used in this study is summarized in Figure 1. There were 47 men (57%) and 36 women (43%) with a median age of 62 years (range years). The primary tumor was located in the tongue in 30 patients, in 147

150 Chapter 8 the floor of the mouth in 35 patients, in the buccal mucosa in 10 patients and in the gingiva in 8 cases. The primary tumor of 44 (53%) patients was pathologically staged as early (pt1-t2). Fortyfive patients (54%) had histologically confirmed lymph node metastasis at the time of diagnosis, whereas the rest had no clinical or cytological/histological evidence of regional dissemination. The median follow-up period was 45 months (range 0-15 years). Detailed patient characteristics are shown in Table 1. Tissues were used in line with the code Proper Secondary Use of Human Tissue as installed by the Dutch Federation of Biomedical Scientific Societies. Table 1 Clinical characteristics of the included OSCC patients All cases Gender Age at diagnosis Smoking history Female Male Median (range) Current smoker or ceased < 1 year Ex-smoker, ceased > 1 year Never smoker No. (%) 83 (100) (43) (57) (44) (56) (37-87) (70) (11) (18) Alcohol consumption Clinical T-stage Clinical N-Stage Sub-site 5 U/day 1-4 U/day Occasionally Never ct1 ct2 ct3 ct4 cn0 cn1-3 Tongue Floor of mouth Buccal cavity Gum Mean follow-up (months) (23) (34) (20) (23) (16) (37) (10) (37) (64) (36) (36) (42) (12) (10) 148

151 Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma Figure 1 Schematic presentation of the work flow of this study. Previous studies resulted in the discovery and validation of a multi-gene signature. In this study, gene expression data were used to correlate the selected gene CTSK with clinical and histopathological parameters. From the same cohort of tumor samples, a TMA was constructed for immunohistochemical analysis of the selected gene to correlate their protein expression with clinical and histopathological parameters, and outcomes were compared. 8 Representative hematoxylin and eosin (HE) slides of the 83 OSCC were selected and reviewed by 2 experienced head and neck pathologists (SMW, PJvD). Table 2 demonstrates the pathological features of the study population. All tumors had been previously examined for HPV-16 positivity, using immunohistochemistry for p16, as well as GP 5+/6+ polymerase chain reaction (PCR). 8 All oral carcinomas included in the current study were proved to be negative for HPV-16. A tissue microarray (TMA) was constructed out of the formalin-fixed, paraffin-embedded (FFPE) tissue blocks. Two central and two peripheral tissue cylinders with a diameter of 0.6 mm were punched out of the blocks, avoiding areas of necrosis, and then arrayed in a recipient paraffin block, using a manual tissue arrayer (MTA-I, Beecher Instruments Inc., Sun Prairie, WI, USA), which was guided by the MTABooster (Alphelys, Plaisir, France). The distribution and position of the cores was determined in advance with the TMA-designer Software (Alphelys-TMA Designer, Version 1.6.8). Normal epithelium from the floor of the mouth, the gingiva and tonsil tissue was incorporated in each block to ensure similarity of staining in the different blocks, and to compare tumor staining with that of normal epithelium. 149

152 Chapter 8 Table 2 Pathological characteristics of the OSCC patients. All cases Pathological T-Stage pt1 pt2 pt3 pt4 Pathological N-Stage pn0 pn1-3 Stage grouping I II III A-IVB Infiltration depth 4.0mm < 4.0mm Differentiation grade Good / Moderate Poor / Undifferentiated Keratinization Present Absent Missing No. (%) 83 (100) (20) (33) (12) (35) (46) (54) (17) (11) (26) (46) (87) (13) (81) (19) (72) (24) (4) Vaso-invasion Present Absent Missing (22) (75) (3) Bone-invasion Present Absent Perineural growth Present Absent Missing Spidery growth Present Absent High risk HPV status positive negative (30) (70) (41) (47) (12) (78) (22) (0) (100) 150

153 Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma CTSK gene expression analysis Genome-wide gene expression was measured using dual-channel microarrays with a pool of tumor samples, as described in an earlier study. 4 CTSK was represented by a sole, unique feature on this array. The expression data were normalized for dye and print-tip biases using print-tip LOESS normalization, and variance stabilization was then applied. The data and protocol used are publicly available (ArrayExpress E-UMCU-11). Immunohistochemistry Five μm thick sections of FFPE tonsil control tissue and the TMA were cut and mounted on silanecoated glass slides. After deparaffinization the endogenous peroxidase activity was blocked for 30 minutes in a 0.3% H2O2 phosphate-citrate buffer solution of ph5.8 with sodium azide. Then, tissue sections were subjected to antigen retrieval by boiling in sodium citrate solution (ph 6) for 15 minutes at 37 C. Subsequently, the tissue slides were washed with PBS, 0.05% (v/v) Tween-20 and incubated with a dilution of the primary antibody against CTSK (clone CK4, Novocastra, Newcastle, UK) for 1 hour at RT. Slides were washed and incubated with the following species-specific secondary antibodies: 1:250 diluted rabbit anti-mouse (RAMPO, Dako, Glostrup, Denmark) followed by Powervision anti-rabbit/hrp conjugated (Klinipath, Duiven, The Netherlands). All antibodies were diluted in PBS/1%BSA. After washing with PBS, the bound antibodies were visualized using 3,3 -diaminobenzidine (0.6 mg/ml). Slides were counterstained with hematoxylin. Evaluation of protein expression Intensity and percentages of positive tumor cells were semi-quantitatively and independently evaluated by 3 observers (SMW, PJvD and FKL) who were blinded to patient outcome. In case of disagreement, the observers reanalyzed the staining results until they reached consensus. To determine the score for each TMA-core, appropriate controls of normal squamous epithelium were used. Protein expression was scored for both its intensity in tumor cells relative to normal epithelium (strong expression = 2, normal expression = 1, no expression = 0) and the percentage of tumor cells in the tissue section with such a specific intensity. Multiplying of these two scoring variables resulted in a scoring range of 0 up to 200, in which a score of 0 represents a complete loss or no expression of protein in all tumor cells and a score of 200 represents a high expression throughout the tumor (Figure 2a-d). Cores were considered lost if less than 10% of cells contained tumor ( sampling error ) or when less than 10% of tissue was present ( absent core ). Cases were excluded if more than 2 cores were lost per case. When the scores between the cores of a particular case differed, the most frequent score determined the overall score. In case of 4 different scores in one case, the average score was calculated

154 Chapter 8 Figure 2 CTSK expression in OSCC and normal mucosa. Representative stainings of the TMA, consisting of 83 OSCC cases, are presented. Staining scores were calculated by the product of intensity (normal = 1, strong = 2) and the proportion of stained tumor cells (%). Panels A-D represent examples of CTSK staining; a) normal mucosa, b) OSCC with no staining, c) OSCC score = product of intensity normal (=1) x proportion (=50%) = 50, d) OSCC score = 2 x 75% = 150. Statistical nalysis The non-parametric Mann-Whitney U test was used to determine differences in CTSK expression between various clinicopathologically defined groups. Logistic regression techniques were used to assess correlations between CTSK expression and the incidence of neck lymph node metastasis. Overall survival (OS) was defined as the length of the time from surgery to death from any cause. Disease-specific survival (DSS) was defined as the time from surgery to death due to disease. Receiver operating characteristic (ROC) curves were designed to determine optimal cut-off values. The association between CTSK and the primary and secondary outcomes was analysed with crosstabs, chi-square test (or Fisher s Exact Test when appropriate), logistic regression, Kaplan Meier/logrank survival analyses, and Cox-regression. 152

155 Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma All p values were based on two-tailed statistical analysis and p < 0.05 was considered significant. Statistical analysis was performed using the SPSS 25.0 statistical package (IBM Corp. IBM SPSS Statistics for OSx, Version Armonk, NY: IBM Corp.). RESULTS Gene expression and clinicopathological variables A statistically significant association of high CTSK mrna levels to lymphatic metastasis (p<0.01) was observed, as wells as to vaso- and perineural invasion (p< 0.01 in both cases; Table 3). In contrast, no significant correlation was found to other pathological characteristics, such as pt status, depth of invasion and tumor grade. Among the various clinical parameters, a strong correlation of increased gene expression was found only to alcohol consumption (p<0.01). No significant relationship was found to smoking history, age, tumor subsite and clinical T or N stage. Cases were stratified according to clinical and pathological characteristics. Smoking history was dichotomized to current smoker or ceased < 1 year versus ex-smoker (ceased > 1 year) and never smoker. Alcohol consumption was dichotomized to 1-4 or > 5 U/day versus occasionally or never. Clinical and pathological nodal status (cn and pn) were dichotomized to cn0 versus cn+ and to pn0 versus pn+. Infiltration was dichotomized to <4mm versus 4mm. Differentiation was dichotomized to well and moderate versus poor and undifferentiated. P-values represent the Mann-Whitney U test of these comparisons. IHC: immunohistochemistry; NS: non-significant. 8 Table 3 Correlations between gene (mrna) and protein (IHC) expression of CTSK and clinical and pathological parameters of the included OSCC cohort (n=83). CTSK Clinical characteristic mrna IHC Smoking history NS NS Alcohol consumption p<0.01 NS Age NS NS ct status NS NS cn status NS NS Subsite NS NS Pathological characteristic pn status p<0.01 p<0.01 pt-status NS NS Infiltration depth NS NS Differentiation grade NS NS Vaso-invasion p<0.01 NS Bone-invasion NS NS Peri-neural invasion p<0.01 p<0.01 Spidery growth NS NS 153

156 Chapter 8 CTSK gene expression and survival A Cox regression analysis was performed in order to determine the prognostic significance of the CTSK gene expression. Dichotomization was based on the cut-off value of -0.26, determined by ROC analysis. Patients with high CTSK gene expression demonstrated a significantly worse 5-year DSS (HR=2.29, CI: , p=0.047; Table 4). The pathological N status was shown to have the strongest impact for DSS (HR=4.10, CI: , p=0.002). The prognostic significance of CTSK gene expression did not hold for overall survival (OS) (p=0.2). The Kaplan-Meier survival plot is shown in Figure 3a (p=0.040). Table 4 Univariate and multivariate DSS Cox regression model for gene and protein CTSK expression. Univariate HR 95% CI p-value Age a Tumor stage b pn c CSTK protein expression CTSK gene expression Multivariate pn status corrected for CTSK protein expression Dichotomization was made according to the cut-off values into high and low expression. The most important prognostic parameters (age, stage and pn) were added in the regression model. a <60 vs. 60 years; b I, II vs. III, IV; c pn0 vs. pn+. Figure 3 Kaplan Meier disease specific survival (DSS) plots for all patients with OSCC (n=83). Cases were stratified according to differential expression of CTSK, and were dichotomized into low and high expression according to the determined cut-off point in panel A for gene expression (-0.26) and in panel B for protein expression (25). P-values in Figure 3a and 3b represent the Log-rank test of this group comparison and therefore differ from the significance levels of the Cox-regression analysis in Table 4. In both analyses, high CTSK expression was strongly associated with a worse 5-year DSS. 154

157 Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma CTSK protein expression and clinicopathological variables A total of 213 (64%) tumor cores stained with the CTSK antibody were available for analysis. Due to our inclusion criteria ( 2 tumor cores available per case), 19 cases were missing. The majority of the OSCCs in this cohort showed an almost total loss of expression for CTSK (71%), whereas only 5% demonstrated total loss (Additional file 1: Table S1). A cut-off value of 25 was determined by ROC analysis, in order to divide patients into low and high protein expression groups. No statistically significant correlation to clinical variables was found (Table 3). In contrast, there was a significant association of increased CTSK expression with histopathologically proven lymph node metastasis (p<0.01). A similar strong relationship to perineural invasion was also demonstrated (p=0.01). No association to other pathological variables was evident. In logistic regression analysis, factors with known impact to nodal disease were incorporated into the model, including T stage, perineural and vaso-invasion, depth of infiltration and spidery growth pattern, along with CTSK protein expression (Table 5). In univariate analysis, high CTSK expression appeared to be an important independent predictive factor of lymph node involvement (OR=7.65, CI: ; p<0.001). In multivariate analysis, CTSK protein expression, corrected for pathological T stage, remained a strong prognostic factor for regional disease, demonstrating an odds ratio of 9 (CI: ; p<0.01). Table 5 Univariate and multivariate logistic regression model for CTSK protein expression on lymph node metastasis. Univariate OR 95% CI p-value pt* Peri-neural invasion * Vaso-invasion * Depth of invasion Spidery growth CTSK protein expression <0.001 Multivariate CTSK protein expression <0.01 corrected for pt-status 8 Dichotomization was made into low expression (score 0-25) versus high expression (score ). The most important predictive parameters (pt, peri-neural invasion, vaso-invasion, depth of invasion, spidery growth) were added in the model. OR= odds ratio, HR = hazard ratio, 95% CI = 95% confidence interval, p-value of the Cox regression model. * age: <60 vs >60 years, tumor stage I, II vs III, IV, pn status pn0 vs pn+ 155

158 Chapter 8 Next, the predictive value of CTSK as a biomarker of occult metastasis in early stage (ct1-t2n0) OSCC was examined. A total of 24 patients had early T stage without clinically detectable nodal disease. Out of the ten patients with yet occult metastases in the neck dissection specimen, nine had a high protein CTSK expression, whereas only one patient showed a value lower than the cut-off (Table 6). The sensitivity of high protein expression in detecting occult metastasis in early stage OSCC was, thus, calculated at 90%, whereas the specificity was 57%. Additionally, the positive predictive value was found at 60%, with a negative predictive value of 89%. Table 6 Allocation of ct1-t2n0 patients based on their pathological N-status and CTSK protein expression. pn status Total pn0 pn+ CTSK 25 > Total The value of CTSK protein expression in predicting occult metastasis ( 25 predicts pn0, >25 predicts pn+) in ct1-t2n0 patients is calculated as follows: sensitivity (9/9+1) x 100% = 90%; specificity (8+6) x 100% =57%; positive predictive value (9/6+9) x 100%= 60%; negative predictive value (8/8+1) x 100%= 89% CTSK protein expression and survival In Cox regression, CTSK protein expression was dichotomized into low versus high based on the previously reported cut-off value, demonstrating a significantly worse DSS in OSCC subjects with increased CTSK protein expression (HR = 2.79, CI , p = 0.045; Table 4). No prognostic impact on overall survival was found. The Kaplan-Meier survival plot is shown in Figure 3b (p=0.035). In multivariate analysis, pn status was corrected for CTSK protein expression, and pathological N status showed once more showed a strong correlation (HR=3.61, CI: , p=0.03), with a change though of the beta coefficient greater than 10%, confirming the role of CTSK as a significant confounder for DSS. DISCUSSION Cancer metastasis is a complex process that includes a number of different events, referred as the invasion-metastasis cascade. The first critical step of the process is the invasion of the malignant cells into the surrounding extracellular matrix (ECM) and stromal cell layers. 9 The biological role of cathepsins in promoting tumor invasion and migration has been proved ex vivo in cell-based systems. 10,11 Apart from their well-known function of ECM degradation and remodeling, cathepsins are also suggested to participate in the activation cascade of pro- 156

159 Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma urokinase-type plasminogen activator and other proteases, enhancing thus their effect in the dissolution of the tumor matrix and basic membrane. 12 In addition to their extracellular function, there is evidence that intracellular cathepsins promote tumor progression by affecting processes acting both as pro-tumorigenic and anti-tumorigenic. 13 Intracellular collagen degradation is an example of the potential intracellular pro-tumorigenic activity of cathepsins. Recent research has also proposed that cathepsins play a role as mediators in programmed cell death (apoptosis), a process of utmost importance in cancer development and progression. 13 Different hypotheses have been supported in mediating cell death, such as the indirect activation of caspases through degradation of the anti-apoptotic Bcl-2 family members, engaging a mitochondrial apoptotic pathway, and the proteolytic activation of Bid. 14 The pathophysiological role of the cathepsins in cancer metastasis has attracted the interest of studying its value as a biomarker of metastasis and prognosis in various types of malignancy. Increased protein expression of cathepsins V, B and D has been associated with distant metastasis and worse DSS in breast cancer. 15 Similar results have been found for cathepsin A in malignant melanoma and cathepsin B in non-small cell lung carcinoma. 16,17 Overexpression of CTSK has been observed in invasive ductal carcinoma of the breast, lung and prostate adenocarcinoma In all these studies, increased protein expression was related to high metastatic potential. Interestingly, high expression of CTSK was found in the desmoplastic reactive stroma of the lung adenocarcinoma, indicating that stromal production of CTSK can favor or modulate the invasion of tumor cells. 21 Interestingly, de Koning et al reported downregulation of SERPINB13, the cognate inhibitor of CTSK, in oral and oropharyngeal SCC to be associated with lymph node metastasis and poor prognosis. 22 However, only one study exists in the literature regarding the prognostic value of CTSK in tongue SCC. 23 In that study, Bitu et al. reported that CTSK was expressed in both stromal and tumor cells by immunohistochemistry. The only significant finding was that CTSK expression in stromal cells exhibited a potential protective role, since a poorer prognosis in early stage tumors was correlated to weak CTSK expression in the tumor microenvironment front. However, the same study found decreased invasion of HSC-3 tumor cells when cathepsin K silencing was applied. It was, thus, concluded that different prognosis could be exhibited, depending on whether CTSK is expressed more in tumor or stromal cells. 8 The present study is the first conducted to explore the predictive and prognostic value of CTSK in OSCC. Combined evaluation of both gene and protein expression was used to augment the validity of clinicopathological correlations. Although some discrepancies were found in the associations with the clinicopathological parameters between gene and protein expression, the results are likewise. Some variation may be expected since mrna levels are not directly proportional to the protein concentration due to post-translational mechanisms, that control protein turnover and abundance, and different translational rates, which are determined by constants that are not 157

160 Chapter 8 completely known. 24 Another factor could also be that gene expression data were acquired using biopsies taken at the border of the primary tumor and samples were included if they consisted of at least 50% tumor cells. The other part of the sample consisted of stromal cells and epithelial cells adjacent to the tumor. Consequently, gene expression was measured using tumor, stromal and epithelial cells, whereas protein expression was scored by immunoreactivity in tumor cells only. There are various explanations for the disagreement with the results reported by Bitu et al. The different scoring system as well as the different antibody clone used to detect cathepsin K could play a role. In the previous study, there is also insufficient information about the diagnostic approach to the lymph node status and incomplete pathological data, such as infiltration depth and perineural invasion, of the studied cohort. Lack of these data could underestimate the clinicopathological correlations. Third, the findings of the previous study were solely based on immunohistochemistry and were, partly, contradictory with the observation, reported by the same authors, of the diminished invasion potential of the HSC-3 tumor cells, when cathepsin K was silenced or inhibited. The results of the current study suggest that CTSK may be used as a predictive biomarker in patients with OSCC. Its high sensitivity (90%) combined to the high negative predictive value (89%) makes it particularly valuable in excluding occult metastasis in early T1-2N0 OSCC, allowing to perhaps rely on a wait and see policy for the management of the neck. Moreover, it is shown that tumors with up-regulation of CTSK harbored a high potential for perineural invasion. This can be interpreted by the proteolytic action on the nerves epineurium and perineurium, facilitating tumor cell migration into the nerve fasciculus. Hence, CTSK can be a molecular determinant of perineural invasion, apart from the various neurotrophins and chemokines that are involved in this process. 25 The strong relationship of CTSK with both lymphatic spread and perineural invasion is also reflected by its significant impact on DSS. The current study was based on a relatively limited cohort of 83 patients with OSCC. The results should be further validated by studies including higher number of patients with emphasis on predicting occult metastasis in cases of N0 stage. Furthermore, serum levels of CTSK can be also evaluated at different stages of the disease and correlate them to clinicopathological variables. Important prognostic implications of elevated serum levels of cathepsins have been observed in other types of malignancies, such as in prostate cancer. 26 Finally, the emergence of new CTSK inhibitors can provide in the future a new tool for the suppression of tumor progression. 158

161 Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma CONCLUSION In conclusion, our findings provide evidence that increased CTSK expression is associated with lymphatic spread and poor prognosis of OSCC. Due to the high negative predictive value (89%) of CTSK protein expression, this biomarker can be a simple and useful tool in the diagnostic work-up of ct1-t2n0 OSCC. ABBREVIATIONS CTSK, cathepsin K; OSCC, oral squamous cell carcinoma; PCR, polymerase chain reaction; HPV, human papilloma virus; OR, odds ratio; CI, confidence interval; HR, hazard ratio; HE, hematoxylin and eosin; TMA, tissue microarray; FFPE, formalin-fixed, paraffin-embedded; OS, overall survival; DSS, disease-specific survival; ROC, receiver operating characteristic; IHC, immunohistochemistry ACKNOWLEDGMENTS A special thanks to the technicians of the laboratory of pathology for their valuable help. FUNDING 8 No funding was received for this study. AVAILABILITY OF DATASETS AND MATERIALS All data supporting this study s findings are found in the manuscript text, tables, and supplemental files. AUTHORS CONTRIBUTIONS FKL and SMW planned and designed the study. FKL is the principal investigator and performed data analysis. MHF conducted the statistical analyses. FKL and EK wrote the manuscript. PJvD and RK critically reviewed the article. All authors read and approved the final manuscript. 159

162 Chapter 8 COMPETING INTERESTS The authors declare that they have no competing interests. CONSENT FOR PUBLICATION This manuscript does not contain any individual person s data; hence, no consent for publication is needed. ETHICS APPROVAL The study was approved by the institutional review board of the University Medical Centre Utrecht, Utrecht, The Netherlands (RP ). 160

163 Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma REFERENCES 1. Jemal A, Bray F, Center MM, et al. Global cancer statistics. CA Cancer J Clin 2011; 61: RJ Sanderson, JAD Ironside. Squamous cell carcinomas of the head and neck. BMJ 2002; 325: Leusink FK, van Es RJ, de Bree R, et al. Novel diagnostic modalities for assessment of the clinically nodenegative neck in oral squamous-cell carcinoma. Lancet Oncol 2012; 13: e Roepman P, Wessels LF, Kettelarij N, et al. An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas. Nat Genet 2005; 37: van Hooff SR, Leusink FK, Roepman P, et al. Validation of a gene expression signature for assessment of lymph node metastasis in oral squamous cell carcinoma. J Clin Oncol 2012; 30: Turk V, Turk B, Turk D. Lysosomal cysteine proteases: facts and opportunities. EMBO J 2001; 20: Turk V, Stoka V, Vasiljeva O, et al. Cysteine cathepsins: from structure, function and regulation to new frontiers. Biochim Biophys Acta 2012; 1824: Leusink FK, van Diest PJ, Frank MH, et al. The Co-Expression of Kallikrein 5 and Kallikrein 7 Associates with Poor Survival in Non-HPV Oral Squamous-Cell Carcinoma. Pathobiology 2015; 82: Valastyan S, Weinberg RA. Tumor metastasis: molecular insights and evolving paradigms. Cell 2011; 147: Premzl A, Zavasnik-Bergant V, Turk V, et al. Intracellular and extracellular cathepsin B facilitate invasion of MCF-10A neot cells through reconstituted extracellular matrix in vitro. Exp Cell Res 2003; 283: Yang Z, Cox JL. Cathepsin L increases invasion and migration of B16 melanoma. Cancer Cell Int 2007; 7: Guo M, Mathieu PA, Linebaugh B, et al. Phorbol ester activation of a proteolytic cascade capable of activating latent transforming growth factor-betal a process initiated by the exocytosis of cathepsin B. J Biol Chem 2002; 277: Vasiljeva O, Turk B. Dual contrasting roles of cysteine cathepsins in cancer progression: apoptosis versus tumor invasion. Biochimie 2008; 90: Droga-Mazovec G, Bojic L, Petelin A, et al. Cysteine cathepsins trigger caspase-dependent cell death through cleavage of bid and antiapoptotic Bcl-2 homologues. J Biol Chem 2008; 283: Sun T, Jiang D, Zhang L, et al. Expression profile of cathepsins indicates the potential of cathepsins B and D as prognostic factors in breast cancer patients. Oncol Lett 2016; 11: Kozlowski L, Wojtukiewicz MZ, Ostrowska H. Cathepsin A activity in primary and metastatic human melanocytic tumors. Arch Dermatol Res 2000; 292: Chen Q, Fei J, Wu L, et al. Detection of cathepsin B, cathepsin L, cystatin C, urokinase plasminogen activator and urokinase plasminogen activator receptor in the sera of lung cancer patients. Oncol Lett 2011; 2: Brubaker KD, Vessella RL, True LD, et al. Cathepsin K mrna and protein expression in prostate cancer progression. J Bone Miner Res 2003; 18: Littlewood-Evans AJ, Bilbe G, Bowler WB, et al. The osteoclast-associated protease cathepsin K is expressed in human breast carcinoma. Cancer Res 1997; 57: Cordes C, Bartling B, Simm A, et al. Simultaneous expression of Cathepsins B and K in pulmonary adenocarcinomas and squamous cell carcinomas predicts poor recurrence-free and overall survival. Lung Cancer 2009; 64: Rapa I, Volante M, Cappia S, et al. Cathepsin K is selectively expressed in the stroma of lung adenocarcinoma but not in bronchioloalveolar carcinoma. A useful marker of invasive growth. Am J Clin Pathol 2006; 125:

164 Chapter de Koning PJ, Bovenschen N, Leusink FK, et al. Downregulation of SERPINB13 expression in head and neck squamous cell carcinomas associates with poor clinical outcome. Int J Cancer 2009; 125: Bitu CC, Kauppila JH, Bufalino A, et al. Cathepsin K is present in invasive oral tongue squamous cell carcinoma in vivo and in vitro. PLoS One 2013; 8: e Schwanhäusser B, Busse D, Li N, et al. Global quantification of mammalian gene expression control. Nature 2011; 473: Marchesi F, Piemonti L, Mantovani A, Molecular mechanisms of perineural invasion, a forgotten pathway of dissemination and metastasis. Cytokine Growth Factor Rev 2010; 21: Miyake H, Hara I, Eto H. Serum level of cathepsin B and its density in men with prostate cancer as novel markers of disease progression. Anticancer Res 2004; 24:

165 Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma SUPPLEMENTARY INFORMATION Supplementary table S1. Immunohistochemical descriptive results of protein expression of CTSK in the OSCC TMA cohort (n=83) Variable CTSK Number (%) of 332 cores Tumor 213 (64) No tumor 68 (20) No core 51 (15) Number of cores / case (n=83) Score (%) / case (n=83) 0 (total loss) 4 (5) 1-50 (almost total loss) 35 (42) (partial loss) 26 (31) missing 18 (22) Percentages may not total 100 because of rounding

166

167 CHAPTER 9 General discussion

168 Chapter 9 166

169 General discussion GENERAL DISCUSSION Management of the clinically node-negative (cn0) neck has to deal with a problem of risk-benefit evaluation between probability of neck metastases, the problem of complications associated with any type of neck dissection, and the prognostic influence of delayed diagnosis of metastasis during follow-up. 1 The choice in management of the cn0 neck used to be either elective neck dissection (END) with possible overtreatment of 60 70% of patients without occult metastases or watchful waiting (WW) with possible undertreatment of patients with occult metastases in the remaining 30 40%. WW entails careful monitoring of the neck (e.g. by ultrasound-guided fine-needle aspiration cytology - USgFNAC - during follow-up) and therapeutic neck dissection (TND) treatment for patients who develop manifest metastasis. END encompasses a surgical procedure potentially causing disfigurement and associated morbidity but prevents disease in the neck becoming more advanced once previously occult metastases become clinically apparent or are detected late during follow-up. In an attempt to optimize management of the cn0 neck and to make it more personalized, improvement in staging of the neck preferably with methods that are non-invasive and not dependent on the size of the metastasis are needed. Therefore, the main focus in this thesis has been on lymph node metastasis (LNM) predicting biomarkers in clinically T1-T2N0 oral squamous-cell carcinoma (OSCC). The multicenter validation of a gene expression LNM predicting signature has been described and specific signature genes were evaluated on a protein level for their correlation with LNM and prognosis. Considerations in management of the neck in early stage OSCC patients Although consensus partly based on an often cited but outdated decision analysis model still suggests that elective neck treatment is indicated when the chance of occult nodal disease exceeds 20%, 2 there are more reasons besides the above mentioned associated morbidity of END why it is important to prevent unnecessary treatment of the neck. It is the high rate of second primary tumors (SPT) of 1% per year during follow-up 3 and the fact that a LNM of a SPT is more difficult to detect in a previously treated neck due to edema and fibrosis. Moreover, neck dissection removes part of the natural barrier for spread to distant organs of a future SPT, resulting in incurable disease. 4 9 Furthermore, a review by Rodrigo et al. 5 including clinical trials and retrospective studies did not conclude that END is superior to WW, with regard to survival and control of neck disease, but other studies did. Dik et al. reported in a very recent retrospective cohort study that patients managed by WW who turned out to be nodal positive (WWN+) showed significantly more extracapsular spread (ECS) as compared to patients treated by END and being diagnosed as nodal 167

170 Chapter 9 positive (END-N+) (56% vs. 14%, P = 0.016). 6 The WW-N+ patients had a significantly worse 3-year disease specific survival (DSS) than END-N+ patients (56% vs. 82%, P = 0.02). Fasunla et al. reported in the same year as Rodrigo a meta-analysis on the existing randomized controlled clinical trials which compared (a form of) END with watchful waiting with a total of 283 patients. 7 They showed that END significantly reduced the risk of disease specific death (fixed-effects model RR=0.57, 95% CI , p=0.014). However, the time period spanned four decades, T3 tumors were included, half of END patients underwent radical neck dissection, follow-up differed among studies and it is doubtful whether the studies were truly homogenous. 8 The same D Cruz et al. described a prospective randomized clinical trial (RCT) including 500 patients, in which END resulted in higher rates of overall survival (80% vs. 67.5%, P=0.01) and disease-free survival (69.5% vs. 45.9%, P<0.001) than did WW (with TND if WWN+) among patients with early-stage oral squamous-cell cancer. 9 However, according to an editorial about this RCT this study also has some caveats that need careful consideration before the findings can be accepted. 10 In the WW group a neck conversion was reported in 45% which is much higher than the percentages of neck conversions (18%, 21% and 28%) reported by three studies in The Netherlands suggesting a less accurate diagnostic work-up in the RCT which may hamper generalizability of their results. In the RCT the cn0 neck was not clearly defined; the ultrasound scoring criteria were not described and during follow-up only half of the patients had ultrasound without guidance by fine needle aspiration cytology (USgFNAC) which is known to be able to reduce the false negative rate of US alone. 14 Furthermore the meticulousness of the follow-up of the RCT is questioned since 28% of patients had a metastasis larger than 3 cm (18% larger than 6 cm), 93% had extranodal spread and 18% had unresectable neck disease. One can therefore conclude the following: a) there seems to be a substantial difference in pretreatment work-up and follow-up between head and neck centers in the world, b) this RCT clearly shows a prognostic benefit of elective treatment in the population in India, using the quite strict follow-up without routine USgFNAC, and c) if routine very strict follow-up using USgFNAC by a well-trained ultrasonographer cannot be assured, END is the safest strategy. 10 Lymph node metastasis detection by the sentinel node biopsy procedure In an attempt to more reliably select the lymph nodes that potentially contain metastases, the sentinel lymph node (SLN) concept has been introduced. The SLN is likely to be the first lymph node to harbor metastasis and can be used to provide information on the rest of the nodal basin. After surgical removal, this node is studied meticulously by histopathological examination, using step sectioning and immunohistochemistry. If the SLN contains metastatic tumor cells, treatment of the neck is recommended, usually in a second procedure. 15 The SLN procedure is deemed more precise than imaging procedures. Two recently published meta-analyses on SLN biopsy show highly accurate results and negative predictive values ranging between 80 and 100%. 16,17 However, in the majority of the analyzed studies END was used as 168

171 General discussion the reference test for the SLN biopsy. When END is performed irrespective of the SLN biopsy result, the reference is histopathological examination, often considered to be the gold standard. However, if routine techniques and not step sectioning and immunohistochemistry are used, histopathological examination of END specimen will miss minimal disease 18 and is likely to overestimate sensitivity and negative predictive values. Therefore, long-term observation of the untreated neck (no neck dissection or radiotherapy) is the true gold standard for determining the value of diagnostic techniques (e.g. the SLN procedure) for detection of occult lymph node metastases. 19 More recently, two multi-institutional trials on SLN biopsy were published that used observation (WW) for SLN negative patients. Flach et al. reported 20 of a total of 62 patients (32%) had a positive SLN and 5 of the 42 SLN negative patients developed a recurrence resulting in a sensitivity of 80% and a NPV of 88%. 20 Schilling et al. reported 94 of a total of 415 patients (23%) had a positive SLN and a false negative result in 15 of the 109 whom were rescued by salvage TND. 21 Sensitivity was 86% and the NPV was 95%. The latter study, however, had a 3:1 distribution of T1:T2 OSCC and after staging by SLN 23% were nodal positive. In the detection of occult lymph node metastases, if the incidence of metastasis is low the NPV will be high almost regardless of the performance of the diagnostic technique. An upstage by a positive SLN of 30% as reported by Thompson et al. 16 shows that the studied patient population reflects the daily clinical practice and the results of the study can be implemented in routine clinical practice. Compared with END, SLN biopsy is obviously less invasive, SLN biopsy scars are smaller resulting in improved esthetic outcome, and SLN biopsy is associated with significantly less postoperative morbidity and better shoulder function However, SLN biopsy remains a labor intensive invasive procedure for which injection of radioactive tracers and admission is required and additional second stage surgery in case of positive SLN is needed. Damage of the facial and accessory nerve do occur, as well as pain, hematoma, infection, edema and especially fibrosis which spreads through the neck and is likely to hamper a TND in case of a positive SLN. Furthermore, the SLN procedure itself could potentially facilitate dissemination of tumor in the neck. Technical difficulties that remain are a) differentiation between a true SLN and possible second echelon lymph nodes on current imaging methods, and b) visualization of SLNs close to the injection site (known as the shine through phenomenon) resulting in lower accuracy rates for flour of mouth tumors. Although further improvements are needed, several centers have adopted the SLN biopsy as an alternative to END and SLN biopsy is also already mentioned as an alternative for END in the American NCCN guidelines, the UK NICE guidelines, as well as in the guidelines of the Dutch Head and Neck Society

172 Chapter 9 Biomarkers for predicting lymph node metastases and prognosis The limitation of imaging techniques to detect small metastatic deposits has led to a search for additional characteristics or biomarkers assessable on the primary tumor to predict nodal disease. Histopathological features of the primary tumor like lymphatic vessel density (chapter 2), tumor thickness (TT), perineural growth (PG), vascular invasive growth (VG), and infiltrative growth (IG) can predict the presence of nodal metastases in the individual patient, irrespective of the size of the tumor and the metastasis. 28 However, the histologic parameters (PG, VG and IG) in biopsy specimens do not represent the resection specimen. 29 Determination of histologic parameters in routinely taken biopsy specimens of OSCC is not helpful in deciding whether to treat the neck. Furthermore, TT and IG are ascertained on histopathological examination of the resection specimen, implying a second-stage surgical procedure when elective neck dissection is indicated. Measurement of TT before surgery, using an intraoral ultrasound probe, could be more promising in this respect. 30 Molecules involved in several metastasis pathways have been studied, but the complexity of the metastatic process makes it unlikely that a single marker for metastasis can be identified with an acceptable predictive value. Gene expression profiling, allowing the simultaneous study of expression levels of multiple genes, could be the most promising technique. By using RNA from the primary tumor specimen, a particular profile that could predict N-stage was identified. 31 Chapter 3 describes a multicenter validation study, undertaken at all head and neck oncological centers in The Netherlands, in which the expression profile to predict lymph node metastasis was transferred to a diagnostic platform to facilitate clinical implementation. Subsequently, the profile was validated with an independent series of 222 samples of OSCC and oropharyngeal squamous-cell carcinoma. Although the array platform was changed, the profile predicted N-stage as expected. For the group of ct1 T2N0 or early-stage oral SCC (n=101), a NPV of 89% was recorded. For these cases the issue of elective neck treatment is most relevant, because early stage OSCC is treated by transoral surgery and, thus, there is no need to enter the neck for excision of the primary tumor. By applying the microarray predictor in patients with early-stage (ct1 T2N0) OSCC, overtreatment by unnecessary END can potentially be reduced from 73% to 41%. Some remarks should be made however. Although overtreatment is reduced, 41% is still rather high. In 41 out of 65 patients classified as being at risk of LNM, evidence for nodal disease could not be found resulting in a low specificity of the microarray based classification causing still a considerate degree of overtreatment by elective neck dissection. This percentage is likely to be lower since the used routine histopathological examination of END specimen misses minimal disease in 5-58% of cases (mean 20%). 18 Besides overtreatment there is psychological uncertainty of WW for the other 36 node negative classified patients of whom 4 will develop a regional recurrence and risk 170

173 General discussion delayed diagnosis and intensified treatment, however timely treatment is still possible for these 4% understaged patients. 32 Furthermore, the validation study acquired gene expression data out of samples taken from fresh frozen tumor resection specimen. It remains to be determined whether gene expression prediction of LNM using standard pre-operatively taken biopsies is as reliably as gene expression prediction using the samples from the tumor resection specimen. Between pre-op and per-op taken samples certain differences exist: 1. timing, the pre-op sample is taken 4 to 6 weeks before the per-op sample, and 2. genes itself due to intratumor heterogeneity for which solid epithelial tumors (including head and neck tumors) are known. 33 This could mean that a single biopsy specimen might not be representative for the entire tumor. Morbidity remains a relevant issue with current management of early-stage (ct1 T2N0) OSCC. The overtreatment (by END in case of pn+ prediction) in the Van Hooff algorithm could be reduced further by replacing END for SLN biopsy as is proposed by the algorithm described in Chapter 4. By applying this algorithm, oncological safety is not so much pursued by overtreatment and its associated unnecessary morbidity but rather by further reduction of the rate of occult metastasis and accurate follow-up. 32 Decision-analysis techniques recommend WW below risk thresholds that vary from 17% to 44%, 2,34,35 thus, we judge the NPV of gene-expression profiling and SLN biopsy acceptable because such thresholds are much higher than the 6% risk of occult metastases projected by the proposed algorithm. With adoption of WW, 30 40% of patients will need neck dissection, and under the proposed algorithm in chapter 4 this proportion would fall to 6%. In the end, how to weigh the expected 6% of patients needing second-stage (and maybe intensified) treatment against the inevitable morbidity of 60 70% of patients who undergo unnecessary treatment of their neck will remain a matter of subjective judgment. Combining gene-expression profiling and SLN biopsy in addition to current imaging techniques will further reduce the rate of occult metastasis. In view of the accuracy of these techniques, this rate can be reduced to a level acceptable to allow WW for the neck in patients with OSCC classified as T1 and T2. Furthermore, restriction of SLN biopsy to individuals who are classified as node-positive on gene expression profiling leads to further reduction of morbidity of the neck. 9 Another example of the possibilities of gene expression microarray analysis is shown in chapter 5 in which the biological basis of the determinants of locoregional recurrence was established. Perineural growth (OS and LRFS) and non-cohesive invasive growth were correlated with worse prognosis (OS and LRFS). For the pattern of extensive non-cohesive growth, but not for perineural growth, a differential set of 160 genes was established, that included genes involved in extracellular matrix modeling. From the biological point of view it makes sense that certain genes involved in the homeostasis of the extracellular matrix like matrix metalloproteinases (MMPs), were overexpressed in tumors that showed a relatively high level of this non-cohesive growth. Other studies also noted a correlation between MMP-expression and poor prognosis The presence 171

174 Chapter 9 of perineural growth was not reflected in differences in gene expression between tumor groups with and without this characteristic. Possible explanations are that a) standard histopathological examination of the tumor resection specimen misses a substantial amount of perineural growth falsely classifying the tumor having no perineural growth and thus creating incorrect groups or b) that gene expression is only altered in the relatively small group of cells involved in this process. Another important finding in the present study is that, despite treatment with radiotherapy, extensive perineural and non-cohesive growth retained their association with worse OS. Thus it seems that this study suggests that for OSCC with extensive perineural growth or extensive non-cohesive growth radiotherapy combined with chemotherapy might be of additional value. Microarray prediction is costly, and although cost-effectiveness of all different diagnostic modalities for LNM is beyond the scope of this thesis, it is interesting to search for more affordable methods. Therefore, selected genes of the LNM predicting profile were investigated by immunohistochemistry and correlated with clinical and pathological variables (chapter 6-8). In search for more single LNM predicting biomarkers we studied the proteins kallikrein 5 (KLK5), kallikrein 7 (KLK7) and their cognate inhibitor serine protease inhibitor kazal-type 5 (SPINK5) together to investigate the functional role of KLK5, KLK7 and SPINK5 in OSCC, as described in chapter 6. Whether a loss of SPINK5 tips the balance to more protease activity by KLK5 and KLK7, and whether such an imbalance may identify patients with aggressive tumor behavior in OSCC patients cannot be confirmed by our data. However, multivariate survival analyses showed that concurrent complete loss of KLK5 and KLK7 identifies OSCC patients with poor clinical outcome (OS HR 0.29 and DSS HR 0.39) and may therefore act as a prognostic marker. At first glance, implementation of such a biomarker for worse outcome could be performed relatively easy by excising tumors with larger surgical margins and increasing dose of postoperative radiotherapy, resulting in a higher morbidity. On the other hand, de-escalated therapy (less intensive treatment by i.e. smaller margins) for patients that do not have complete concurrent loss of KLK5 and KLK7 could result in lower survival rates. However, how much and in what form this could happen is not clear, although recent data showed there is no evidence for local adjuvant treatment in case of resection margins 3 mm with 2 unfavorable histological features. 39 This suggests resection of early OSCC with smaller margins could result in similar survival rates. Perhaps, KLK5 and KLK7 or other biomarkers like the invasive growth associated gene expression profile (chapter 5) can be helpful in pointing out which tumors can be excised with margins 3 mm instead of 5 mm. Other genes with a strong differential gene expression for LNM that could potentially serve as a single LNM biomarker (chapter 7) are LCN2, THBS2, TACSTD2, and SLPI. Although the precise function of these genes is yet not fully understood, an explanation might be their joint role in matrix remodeling. 40 However, SLPI, also known as antileukoproteinase, is a protease inhibitor of neutrophil elastase, cathepsin G, chymotrypsin, and trypsin, 41 enzymes with extracellular matrix 172

175 General discussion degradative properties, and associated with cancer development, invasiveness, and progression and seems inversely correlated with nodal metastases. Despite a significant correlation between SLPI protein expression and LNM in the entire cohort, SLPI expression had no additional diagnostic value as a predictor for LNM in a subgroup of early cancers, which are clinically lymph node negative in this cohort of OSCC. SLPI was identified as an independent predictor for OS and DSS in OSCC. Patients with low SLPI protein expression had a worse OS and DSS compared with patients with any SLPI expression. Thus, SLPI might be relevant as a prognostic biomarker for patients with OSCC. Cathepsin K (CTSK) is a proteolytic enzyme with a known role in the degradation of the extracellular matrix. Involvement in pathological processes, such as inflammation and cancer progression, has been proved. SerpinB13 (SB13) is its cognate inhibitor, and loss of SB13 expression has been correlated to poor outcome and LNM. 42 Both SB13 and CTSK are on the LNM associated gene list forming the microarray predictor. Therefore, we investigated CTSK expression values as a single predictive marker for disease outcome and LNM (chapter 8). A clinically useful cut-off value was determined and the negative predictive value of CTSK for LNM in the relevant group (ct1-2n0 OSCC) was calculated at 89%. Although these results are promising, the relevant group was small (n=24). Furthermore, the IHC analysis has been performed on a TMA built up with 6 resection specimen cores per tumor, enabling assessment of intratumoral heterogeneity, which is essential for potential biomarkers. It seems relevant to validate the CTSK predictive value on incisional biopsies as well, to confirm its diagnostic value in daily clinical practice. Such a validation study should be prospective and use WW or SLN biopsy combined with therapeutic neck dissection only for SLN-positive patients as treatment of the neck in an attempt to reduce the chance of missing a micrometastasis. 18,19 9 In summary, this thesis shows the potential of molecular markers in solving the debate about the management of cn0 neck in OSCC patients. Better understanding of genes that associate with LNM and other pathological characteristics of OSCC is an important step forward in realizing more personalized cancer treatment and management of the neck. This will eventually improve survival and lower morbidity in OSCC patients in the future. 173

176 Chapter 9 REFERENCES 1. Kowalski LP, Sanabria A. Elective neck dissection in oral carcinoma: a critical review of the evidence. Acta Otorhinolaryngol Ital 2007; 27: Weiss MH, Harrison LB, Isaacs RS. Use of decision analysis in planning a management strategy for the stage N0 neck. Arch Otolaryngol Head Neck Surg 1994; 120: Chuang SC, Scelo G, Tonita JM, et al. Risk of second primary cancer among patients with head and neck cancers: A pooled analysis of 13 cancer registries. Int J Cancer. 2008; 123: Haigentz M Jr, Hartl DM, Silver CE, et al. Distant metastases from head and neck squamous cell carcinoma. Part III. Treatment. Oral Oncol 2012; 48: Rodrigo JP, Shah JP, Silver CE, et al. Management of the clinically negative neck in early-stage head and neck cancers after transoral resection. Head Neck 2011; 33: Dik EA, Willems SM, Ipenburg NA, et al. Watchful waiting of the neck in early stage oral cancer is unfavourable for patients with occult nodal disease. Int J Oral Maxillofac Surg 2016; 45: Fasunla AJ, Greene BH, Timmesfeld N, et al. A meta-analysis of the randomized controlled trials on elective neck dissection versus therapeutic neck dissection in oral cavity cancers with clinically nodenegative neck. Oral Oncol 2011; 47: D Cruz AK, Dandekar MR. Elective versus therapeutic neck dissection in the clinically node negative neck in early oral cavity cancers: Do we have the answer yet? Oral Oncol 2011; 47: D Cruz AK, Vaish R, Kapre N, et al. Elective versus Therapeutic Neck Dissection in Node-Negative Oral Cancer. N Engl J Med 2015; 373: de Bree R, van den Brekel MW. Elective neck dissection versus observation in the clinically node negative neck in early oral cancer: Do we have the answer yet? Oral Oncol 2015; 51: Melchers LJ, Schuuring E, van Dijk BA, et al. Tumour infiltration depth > 4 mm is an indication for an elective neck dissection in pt1cn0 oral squamous cell carcinoma. Oral Oncol 2012; 48: Nieuwenhuis EJ, Castelijns JA, Pijpers R, et al. Wait-and-see policy for the N0 neck in early-stage oral and oropharyngeal squamous cell carcinoma using ultrasonography-guided cytology: is there a role for identification of the sentinel node? Head Neck 2002; 24: Flach GB, Tenhagen M, de Bree R, et al. Outcome of patients with early stage oral cancer managed by an observation strategy towards the N0 neck using ultrasound guided fine needle aspiration cytology: No survival difference as compared to elective neck dissection. Oral Oncol 2013; 49: Borgemeester MC, van den Brekel MW, van Tinteren H, et al. Ultrasound-guided aspiration cytology for the assessment of the clinically N0 neck: factors influencing its accuracy. Head Neck 2008; 30: Alkureishi LW, Burak Z, Alvarez JA, et al. Joint practice guidelines for radionuclide lymphoscintigraphy for sentinel node localization in oral/oropharyngeal squamous cell carcinoma. Ann Surg Oncol 2009; 16: Thompson CF, St John MA, Lawson G, et al. Diagnostic value of sentinel lymph node biopsy in head and neck cancer: a meta-analysis. Eur Arch Otorhinolaryngol 2013; 270: Govers TM, Hannink G, Merkx MA, et al. Sentinel node biopsy for squamous cell carcinoma of the oral cavity and oropharynx: a diagnostic meta-analysis. Oral Oncol, 2013; 49: Ferlito A, Shaha AR, Rinaldo A. The incidence of lymph node micrometastases in patients pathologically staged N0 in cancer of oral cavity and oropharynx. Oral Oncol 2002; 38: de Bree R. How to analyze the diagnostic value of sentinel node biopsy in head and neck cancer. Eur Arch Otorhinolaryngol 2013; 270: Flach GB, Bloemena E, Klop WMC, et al. Sentinel lymph node biopsy in clinically N0 T1 T2 staged oral cancer: The Dutch multicenter trial. Oral Oncol 2014; 50: Schilling C, Stoeckli SJ, Haerle SK, et al. Sentinel European Node Trial (SENT): 3-year results of sentinel node biopsy in oral cancer. Eur J Cancer Dec;51(18):

177 General discussion 22. Murer K, Huber GF, Haile SR, et al. Comparison of morbidity between sentinel node biopsy and elective neck dissection for treatment of the n0 neck in patients with oral squamous cell carcinoma. Head Neck 2011; 33: Schiefke F, Akdemir M, Weber A, et al. Function, postoperative morbidity, and quality of life after cervical sentinel node biopsy and after selective neck dissection. Head Neck 2009; 31: Hernando J, Villarreal P, Alvarez-Marcos F, et al. Comparison of related complications: sentinel node biopsy versus elective neck dissection. Int J Oral Maxillofac Surg 2014; 43: NCCN Guidelines Version head-and-neck.pdf Accesses on November 9, NICE Guidelines. [NG36]1.3.5 Published date February chapter/recommendations. Accessed on November 9, Dutch Guideline Head and Neck Tumors. Hoofd-halstumoren% pdf Accesses on November 9, Takes RP, Rinaldo A, Rodrigo JP, et al. Can biomarkers play a role in the decision about treatment of the clinically negative neck in patients with head and neck cancer? Head Neck 2008; 30: Dik EA, Ipenburg NA, Adriaansens SO, et al. Poor Correlation of Histologic Parameters Between Biopsy and Resection Specimen in Early Stage Oral Squamous Cell Carcinoma. Am J Clin Pathol 2015; 144: Lodder WL, Teertstra HJ, Tan IB, et al. Tumour thickness in oral cancer using an intra-oral ultrasound probe. Eur Radiol 2011; 21: Roepman P, Wessels LF, Kettelarij N, et al. An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas. Nat Genet 2005; 37: Yuen AP, Ho CM, Chow TL, et al. Prospective randomized study of selective neck dissection versus observation for N0 neck of early tongue carcinoma. Head Neck 2009; 31: Marusyk A, Almendro V, Polyak K. Intra-tumour heterogeneity: a looking glass for cancer? Nat Rev Cancer 2012; 12: Song T, Bi N, Gui L, et al. Elective neck dissection or watchful waiting : optimal management strategy for early stage N0 tongue carcinoma using decision analysis techniques. Chin Med J (Engl) 2008; 121: Okura M, Aikawa T, Sawai NY, et al. Decision analysis and treatment threshold in a management for the N0 neck of the oral cavity carcinoma. Oral Oncol 2009; 45: Al-Azri AR, Gibson RJ, Keefe DMK, et al. Matrix metalloproteinases: do they play a role in mucosal pathology of the oral cavity? Oral Diseases 2013; 19: Jordan RC, Macabeo-Ong M, Shiboski CH, et al. Overexpression of matrix metalloproteinase-1 and -9 mrna is associated with progression of oral dysplasia to cancer. Clin Cancer Res 2004; 10: Katayama A, Bandoh N, Kishibe K, et al. Expressions of matrix metalloproteinases in early-stage oral squamous cell carcinoma as predictive indicators for tumor metastases and prognosis. Clin Cancer Res 2004; 10: Dik EA, Willems SM, Ipenburg NA, et al. Resection of early oral squamous cell carcinoma with positive or close margins: relevance of adjuvant treatment in relation to local recurrence: margins of 3 mm as safe as 5 mm. Oral Oncol 2014; 50: Warde-Farley D, Donaldson SL, Comes O, et al. The GeneMANIA prediction server: biological network integration for gene prioritization and predicting gene function. Nucleic Acids Res 2010; 38(Web Server issue): W214 W Boudier C, Cadène M, Bieth JG. Inhibition of neutrophil cathepsin G by oxidized mucus proteinase inhibitor. Effect of heparin. Biochemistry 1999;38: de Koning PJ, Bovenschen N, Leusink FK, et al. Downregulation of SERPINB13 expression in head and neck squamous cell carcinomas associates with poor clinical outcome. Int J Cancer 2009; 125:

178

179 CHAPTER 10 Future perspectives

180 Chapter

181 Future perspectives FUTURE PERSPECTIVES Current state: watchful waiting, sentinel lymph node biopsy and elective neck dissection The 2005 lymph node metastasis (LNM) gene expression profile (GEP) 1 was validated in the Dutch multicenter validation study (Chapter 3) with a negative predictive value (NPV) of 89% in the clinically relevant group (ct1-t2n0 OSCC, n=101). This impressive NPV was achieved despite a change of platform, lab and patient cohort. Often, such a high NPV implies a low positive predictive value (PPV) and indeed, the validation study gave a PPV of only 37%. However, by using the LNM GEP in conjunction with current diagnostic methods the risk of occult metastases can be reduced from 30-40% to 11%. This NPV is approaching that of sentinel lymph node biopsy (SLNB) and exceeds the commonly used standard of a 20%. 2 The question is whether this is low enough for surgeons to refrain from an elective neck dissection (END) and instead counsel the patient towards a watchful waiting (WW) approach. One can argue that SLNB is the current diagnostic gold standard for the N0 neck and features in the guidelines of the Dutch Head & Neck society, the UK NICE guidelines and the American NCCN guidelines as an alternative for END. Indeed the NPV of SLN approaches 95% in some series. SLNB is however not without its disadvantages. Firstly, SLNB is an invasive procedure with its own morbidity. Second, it requires highly specialized equipment which involves considerable costs. Thirdly, SLNB-positive patients are currently treated with a completion modified radical neck dissection (MRND). This procedure needs to be performed without significant delay leading to difficulties with hospital logistics. The MRND itself is also complicated by the prior SLNB causing a non-virgin operative field. This leads us to speculate a completion MRND is neither as oncologically sound nor as functionally good as an END performed on a virgin neck. These disadvantages are the reason that several head and neck centers still consider the SLNB procedure as inferior to an END performed on a virgin neck. Ideally, a diagnostic procedure should be noninvasive and compromise therapeutic procedures as little as possible. 10 Watchful waiting (WW) in the appropriate patient population is also a valid approach. It certainly minimizes any over treatment of patients who are truly pn0 but may compromise the survival outcomes of the minority of patients with occult metastasis (although contradicting results are reported). 3-5 Moreover, many patients tend to conceive this management approach as too hazardous when explained the statistical risk of an occult metastasis. Finally, the costs associated and time investment involved with repeated hospital visits, sonography and fine needle aspiration cytology, also have to be considered. 179

182 Chapter 10 The perfect work-up Ideally there would exist a safe, non-invasive, readily available and cheap diagnostic work-up for the cn0 neck to better predict the risk of occult neck metastases, guiding patients and surgeons to decide between WW, SLNB and END in each individual case. This diagnostic LNM nomogram will no doubt combine several diagnostic modalities like imaging, cytology, histopathology and molecular analysis. Pathologists already routinely report on tumor size, depth of invasion, differentiation, infiltrative spread, vascular and perineural invasion, which all potentially have therapeutic consequences and the LNM nomogram can be seen as an additional parameter within this work-up. It must however be emphasized that a LNM-nomogram is a diagnostic method still in development. The exact combination of molecular markers (i.e. the LNM GEP) needs to be optimized to give the best compromise between costs, availability and oncological accuracy. In terms of costs and availability, immunohistochemistry (IHC) is widely available and reasonably cheap whereas more advanced techniques such as DNA and RNA sequencing, which allow to examine all genes in one go, are not yet widespread. A future, generally applicable LNMnomogram would therefore ideally use only a few biomarkers, to be tested with IHC. On the other hand, DNA and RNA sequencing could very well be used to discover additional markers for the future LNM-nomogram. Until now, DNA and RNA sequencing studies have used only primary tumor tissues or cancer cell lines. This leaves a void which coming projects can fill; by also sequencing tissue from lymph node metastases to look for possible genetic alterations that could expose the ability of metastasizing. 6 Practical issues also need to be resolved. For example, should testing be performed on a biopsy or after excision and analysis of the whole primary tumor? The first approach would allow simultaneous management of the primary tumor and the neck with either SLNB or END but with the uncertainty of whether the biopsy is truly representative of the whole tumor. 7 Does intratumor genetic heterogeneity influence the accuracy of the LNM GEP? 8 The latter approach would presumably give more reliable results for a future LNM-nomogram (the worst tissue with the worse prognostic indicator could be identified), but would arguably negate the possibility of SLNB and imply a delayed second stage END if the risk for occult metastasis was deemed too high. The presence of intra-tumor genetic heterogeneity in oral cancer was hinted/suggested from data in a yet unpublished study, conducted on tumor specimen of the multicenter study in Chapter 3. This study was conducted to determine whether the standard pre-operative biopsy specimen could be analyzed as reliably as biopsies from the resection specimen. Both biopsy samples were collected in 20 ct1-t2n0 patients. All samples were analyzed using the same platform and laboratory as in the multicenter study (chapter 3). Seven patients were excluded because 180

183 Future perspectives one of the two biopsy samples could not be used. In 9 of the included 13 bridging samples the N-stage was predicted the same, in the other 4 (31%) the prediction differed between biopsy and resection specimen. Possible explanations for the difference in N-stage prediction are intra-tumor genetic heterogeneity or genetic dedifferentiation due to tumor progression in time, because of the delay of ca. 4 weeks between acquiring both samples in each patient. Both phenomena deserve further research. In our opinion, the addition of any molecular test to the standard work-up should not cause undue delays. Many small oral cavity tumors are amenable to excision under local anesthesia in an outpatient setting. Prompt excision of a small primary tumor instead of excision after the usual 4-5 weeks waiting for complete diagnostic work-up of the neck may improve oncological outcomes. 9 This means that molecular analysis of small primaries can occur concomitantly to staging of the neck. An eventual staged neck dissection would also benefit from reduced risk of a fistula to the oral cavity. What is the reference standard? Current publications on management of the cn0 neck often suffer from unclear pathological standards. To properly evaluate the value of any new test or nomogram we need to first be clear what these pathological standards are. Historically, ENDs probably under-staged the neck. This is because pathologists did not have the time nor the techniques to thoroughly examine specimens sent to them by the surgeon. Indeed, using more modern pathological techniques that are better able to detect micrometastases, isolated tumor cells and other minimal changes may lead to upstaging the neck in up to 20% of cases. 10 An example of this effect of changing the reference standard, is shown in the study by de Koning 11 : step sectioning of END specimen of all pn0 staged neck specimen by standard histopathology, resulted in upstaging of the neck in 5 (10%) patients out of 48 patients. Upstaging changed the association of SERPINB13 with LNM from non- significant to significant. 10 The current step-serial sectioning and immunohistochemistry used in SLNB is extremely thorough as compared to routine pathological examination used in END. 12 Although this may represent the new gold standard of pathological examination, it may still not be the best reference standard for staging the N0 neck. The European Sentinel node trial (SENT- EORTC 20421) used long term observation of the untreated neck with the assumption that any neck metastases within a short time period indicated a false negative for SLN. 13 We would argue that long term follow-up of the untreated neck is a more reliable gold standard for staging the neck than pathological examination of either SLN biopsies or END as it negates any inadvertent mistake by the surgeon (e.g. wrong node sampled, non-thorough END) or pathologist (isolated tumor cells missed on pathological examination). 181

184 Chapter 10 Unfortunately, if we wish to use long term follow up as the gold standard for staging the neck rather than pathological examination, it means that much of the published data on biomarkers cannot be used. Our future LNM-nomogram would thus have to be trialed in a prospective manner or failing that, be used retrospectively on a patient group being observed such as the SENT cohort 13 or other cohorts based on a WW approach. 4 Indeed, validation of a nomogram could be performed on such studies before finding its way into a prospective trial. Patient selection A diagnostic test is only useful if it gives the clinician new information which he didn t already have. For example, it is not useful to perform SLNB on a patient with palpable nodal disease. For this, the better diagnostic approach is fine needle aspiration cytology. Similarly, in a patient with a small primary exhibiting few aggressive markers the SLN will more than likely be negative. Such inherent bias due to patient selection can be seen when comparing the results of the SENT trial 13 with the Dutch multicenter trial described in chapter 3. In the former trial the NPV was 95% but the ratio of T1:T2 was 3:1. In our multicenter trial the NPV was 89% but the ratio of T1:T2 was 3:7 i.e. seven times as many T2 patients with a de facto higher risk for occult metastases. It is pivotal that clinicians recognize that the individual patient in their consultation room may not be reflected by published data sets and that the fairly arbitrary cut off of T1 or T2 is too blunt a tool to choose the best approach for management of the neck. In closing This thesis started with the premise that current methods for managing the cn0 neck can be improved upon. We believe that molecular (bio)markers, used as part of an appropriate nomogram in the appropriate patient group, have an important role to play and that this technique will complement the existing options of WW, SLN and END. One possible path forward for research is to design such a nomogram retrospectively by using an existing SLNB dataset being the current pathological gold standard, before prospectively trialing a patient cohort. 182

185 Future perspectives REFERENCES 1. Roepman P, Wessels LF, Kettelarij N, et al. An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas. Nat Genet 2005; 37: Weiss MH, Harrison LB, Isaacs RS. Use of decision analysis in planning a management strategy for the stage N0 neck. Arch Otolaryngol Head Neck Surg 1994; 120: Yuen AP, Ho CM, Chow TL, et al. Prospective randomized study of selective neck dissection versus observation for N0 neck of early tongue carcinoma. Head Neck 2009; 31: Dik EA, Willems SM, Ipenburg NA, et al. Watchful waiting of the neck in early stage oral cancer is unfavourable for patients with occult nodal disease. Int J Oral Maxillofac Surg 2016; 45: D Cruz AK, Vaish R, Kapre N, et al. Elective versus Therapeutic Neck Dissection in Node-Negative Oral Cancer. N Engl J Med 2015; 373: Tabatabaeifar S, Kruse TA, Thomassen M, et al. Use of next generation sequencing in head and neck squamous cell carcinomas: A review. Oral Oncol 2014; 50: Dik EA, Ipenburg NA, Adriaansens SO, et al. Poor Correlation of Histologic Parameters Between Biopsy and Resection Specimen in Early Stage Oral Squamous Cell Carcinoma. Am J Clin Pathol 2015; 144: Gerlinger M, Rowan AJ, Horswell S, et al. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N Engl J Med 2012; 366: van Harten MC, Hoebers FJ,Kross KW, et al. Determinants of treatment waiting times for head and neck cancer in the Netherlands and their relation to survival. Oral Onc 2015; 51: Ferlito A, Rinaldo A, Devaney KO, et al. Detection of lymph node micrometastases in patients with squamous carcinoma of the head and neck. Eur Arch Otorhinolaryngol 2008; 265: de Koning PJ, Bovenschen N, Leusink FK, et al. Downregulation of SERPINB13 expression in head and neck squamous cell carcinomas associates with poor clinical outcome. Int J Cancer 2009; 125: de Bree R. How to analyze the diagnostic value of sentinel node biopsy in head and neck cancer. Eur Arch Otorhinolaryngol 2013; 270: Schilling C, Stoeckli SJ, Haerle SK, et al. Sentinel European Node Trial (SENT): 3-year results of sentinel node biopsy in oral cancer. Eur J Cancer 2015; 51:

186

187 CHAPTER 11 Summary

188 Chapter

189 Summary SUMMARY Oral cancer is the 15th most common cancer worldwide, with more than 300,000 new cases diagnosed (2% of the total) and with more than 145,000 deaths in 2012 (also 2% of the total). Of all oral cancers, around 90% are squamous cell carcinomas, originating in the tissues that line the mouth and lips. Oral squamous-cell carcinomas (OSCC) metastasize frequently to regional lymph nodes in the neck and this occurs in 50% of patients. The presence of nodal metastases is a determinant of prognosis and clinical management. The neck is staged by palpation and imaging, but accuracy of these techniques to detect small metastases is low. Especially in the diagnosis of T1-T2 oral cavity cancers, in which the risk of occult metastases is about 30%, the inaccuracy of lymph node diagnostics is important. In these patients, the primary tumor can often be resected transorally without opening the neck, and in that situation the dilemma exists how to treat the neck: either by elective neck dissection (END) with unnecessary treatment in 70% of patients or watchful waiting (WW) followed by therapeutic neck dissection (TND) in patients who develop manifest metastases with the risk of delayed treatment and worse prognosis in case of metastatic disease. 1,2 These unsatisfying therapeutic options have been the subject of debate for decades and discovery of a better predictor of the nodal status in OSCC will likely be hailed as the Holy Grail in head and neck surgery. The goal of this thesis was to find and develop new methods to diagnose or predict occult metastases in the neck for improved regional staging and prognosis of OSCC patients to allow more personalized management of the neck. To achieve this goal an attempt was made to validate an earlier published lymph node metastasis (LNM) predicting gene expression signature, and to discover other biomarkers associated with LNM and prognosis in OSCC patients. The area of the head and neck features a rich lymphatic network, including approximately three hundred out of the total eight hundred lymph nodes of the human body. 3 Tumor lymphangiogenesis, the formation of new lymphatics related to malignant disease, has become a new field of studying the lymphatic dissemination of various neoplasms. 4 The initial concept of lymphatic metastasis was that tumor cells spread solely through preexisting, peritumoral lymphatic vessels that serve as passive channels. 5 However, the emergence of a number of lymphatic vessel markers has provided new insights into the active process of developing a tumor-associated lymphatic vasculature, analogous to the process of neovascularization. Moreover, research has revealed a number of growth factors and chemokines that are involved in the process of tumor lymphangiogenesis and provide additional information to understand the active role of malignant cells in the actual development of a pathological lymphatic vasculature, which serves as a conduit for regional spread. 6 Studies oriented to the prognostic role of tumor lymphangiogenesis use immunohistochemistry (IHC) to quantify the lymphatic vessels density (LVD) inside the tumor

190 Chapter 11 mass, as well as in the peritumoral area in close proximity to the tumor margin. Chapter 2 summarizes the current knowledge regarding the robustness of the correlation of intratumoral and peritumoral LVD to LNM in an attempt to discover an additional tool to estimate the risk of occult metastasis in HNSCC. Oral cancer was the main primary malignancy studied, but also patients with laryngeal, hypopharyngeal and oropharyngeal carcinoma were included. The majority of the articles reported a significant correlation of increased LVD to the presence of LNM. However, previous studies suffered from several limitations and therefore do not allow drawing definitive conclusions other than that LVD could potentially be used as an additional tool for the prediction of occult metastasis. Another method to predict occult lymph node metastases is gene expression profiling, which has been shown to be useful for diagnosis and prognosis of several cancers. 7,8 With regard to HNSCC, a previous study identified a gene expression signature for distinguishing metastasizing (N+) from N0 OSCC and N0 oropharynx SCC (OPSCC). 9 Although promising, the independent validation cohort in that study was small (n=22), with all samples (n=104) derived from a single clinical center. In Chapter 3, this predictive gene set was first re-evaluated using a different microarray platform on 94 samples, transferred to a dedicated diagnostic microarray, and subsequently tested in a large, multicenter patient cohort (n=222). The negative predictive value (NPV) of the diagnostic signature on the entire validation cohort (n=222) was 72%. The signature performed well on the most relevant subset of early-stage (ct1-t2n0) OSCC (n=101), with an NPV of 89%. Combining current clinical assessment with the expression signature would decrease the rate of undetected nodal metastases from 28% to 11% in early-stage OSCC. Testing the proposed clinical decision model that incorporates the expression signature in a prospective study is needed to assess whether the microarray results are able to guide management of the clinically nodenegative neck in OSCC patients. Presentation of the study described in chapter 3 during the 3 rd World Congress of the International Academy of Oral Oncology in Singapore 2011 prompted an invitation to write a review about novel diagnostic modalities for assessment of the clinically node-negative neck (cn0) in OSCC which is described in Chapter 4. Again the limitations of imaging techniques to detect small metastatic deposits that have led to a search for additional characteristics or biomarkers assessable on the primary tumor to predict nodal disease, are outlined. In this Personal View, we restrict ourselves to discussion of the two most promising techniques, which are arguably ready for clinical implementation and have a very different but complementary nature: gene expression profiling and sentinel lymph node (SLN) biopsy. A new staging algorithm incorporating both methods is suggested, to optimize management of the cn0 neck in patients with early-stage OSCC. By applying our proposed algorithm, oncological safety is not so much pursued by over treatment and its associated unnecessary morbidity but rather by further reduction of the rate of 188

191 Summary occult metastasis and accurate follow-up. Timely treatment is still possible for patients who are under-staged. Combining gene expression profiling and SLN biopsy in addition to current imaging techniques will further reduce the rate of occult metastases to an estimated 6%. Furthermore, restriction of SLN biopsy to individuals who are classified as node-positive on gene expression profiling eliminates overtreatment. Prospective clinical trials implementing this staging algorithm are needed to investigate whether gene expression profiling and SLN biopsy can be combined and whether oncological and functional outcomes in patients with oral SCC will indeed improve. In Chapter 5 another example is described of how gene expression analysis can be used in an attempt to stratify patients who are prone to have locoregional recurrent disease and who are not. This is important since HNSCC has still a fairly bad prognosis with more than 50% of patients dying within five years after diagnosis and a considerable part of these deaths is caused by locoregionally recurrent cancer. 10 This study aimed to identify tumor characteristics that are associated with the development of locoregional recurrences and a poor prognosis and to reveal their biological basis. Perineural growth and non-cohesive invasive growth were correlated with worse prognosis. The negative effect of these characteristics on survival was maintained in the group that received post-operative radiotherapy. No association was found for degree of differentiation and bone invasion and presence of dysplasia or tumor at the margins. For the pattern of extensive non-cohesive growth, but not for perineural growth, a differential set of 160 genes was established, that included genes involved in extra-cellular matrix modeling. The present study could not reveal a distinct association between the non-cohesive gene set and worse prognosis. However, for patients from whom the tumor is not surgically removed and only small biopsies are taken when chemoradiation is planned as primary therapy, it is hard to assess histopathological characteristics, like pattern of invasion. Expression measurement in such biopsies for instance of MMPs could be useful in this respect. In summary we can state that with gene expression array analysis the biological basis of the determinants of locoregional recurrence was established. In chapters 6 to 8 we zoomed in on a few specific genes of the validated LNM predicting profile. We evaluated their correlations with clinical and histopathological variables on a gene and protein level including only OSCC. HPV status was determined by an algorithm for HPV-16 to exclude HPV positive tumors. 11 A tissue microarray (TMA) was made of the paraffin-embedded tissue. For each tumor block, 2-3 central tissue cylinders and 2-3 tissue cylinders at the tumor front with a diameter of 0.6 mm were punched out, avoiding areas of necrosis, and arrayed in a recipient paraffin block. Normal epithelium from the floor of the mouth, gingiva, and tonsil was incorporated in each block to ensure similarity of staining between the different blocks

192 Chapter 11 In Chapter 6 we describe four of the most predictive genes: secretory leukocyte protease inhibitor (SLPI), lipocalin-2 (LCN2), thrombospondin-2 (THBS2), and tumor-associated calcium signal transducer 2 (TACSTD2). Gene and protein expression was correlated with LNM, overall survival (OS), and disease-specific survival (DSS). SLPI protein expression correlated with LNM in the whole cohort, not in a subgroup of ct1 to 2N0. SLPI expression correlated with OS (hazard ratio[hr]=0.61) and DSS (HR=0.47) in multivariate analysis. LCN2, THBS2, and TACSTD2 did not show any correlation with LNM, OS, or DSS. Although SLPI expression correlates with LNM, it had no additional value in determining LNM in early OSCC. However, it is an independent predictor for both OS and DSS and therefore a relevant prognostic biomarker in OSCC In Chapter 7 a similar analysis was performed as in chapter 6 only now the proteases kallikrein 5 (KLK5) and kallikrein 7 (KLK7) and their cognate inhibitor serine protease inhibitor kazal type 5 (SPINK5) were evaluated as predictive or prognostic markers in non-hpv OSCC. Concurrent loss of KLK5 and KLK7 correlated with worse disease-specific and overall survival (DSS and OS). Multivariate analysis proved that co-expression is an independent prognostic factor for DSS (p=0.029) and OS (p=0.001). This report demonstrates that concurrent loss of KLK5 and KLK7 associates with a poor clinical outcome in OSCC and could therefore serve as prognostic marker in this disease. Finally, in Chapter 8 the role of cathepsin K (CTSK) as a biomarker for LNM and prognosis was studied in non-hpv OSCC. Gene expression data were acquired from a previous study 9 and protein expression was semi-quantitatively determined by immunohistochemistry on TMAslides. All expression data were correlated with clinicopathological variables. Elevated gene and protein expression of CTSK were strongly associated to LNM and perineural invasion (p<0.01). Logistic regression analysis highlighted increased CTSK protein expression as the most significant independent factor of lymphatic metastasis (OR=7.65, CI: , p=0.001). Survival analysis demonstrated both gene and protein CTSK expression as significant indicators of poor 5-year disease specific survival (HR=2.29, CI: , p=0.047 for gene expression; HR=2.79, CI: , p=0.045 for protein expression). A cut-off value of 25 was determined by ROC analysis, in order to divide patients into low and high protein expression groups. There was a significant association of increased CTSK expression with histopathologically proven LNM (p<0.01). Next, the predictive value of CTSK as a biomarker of occult metastasis in early stage (ct1-t2n0) OSCC was examined. A total of 24 patients had early T stage without clinically detectable nodal disease. Out of the ten patients with yet occult metastases in the neck dissection specimen, nine had a high protein CTSK expression, whereas only one patient showed a value lower than the cut-off. The sensitivity of high protein expression in detecting occult metastases in early stage OSCC was, thus, calculated at 90%, whereas the specificity was 57%. Additionally, the positive predictive value was found at 60%, with a negative predictive value of 89%. The current study was based 190

193 Summary on a relatively limited cohort of 83 patients with OSCC. The results should therefore be further validated by prospective studies including higher number of patients with emphasis on predicting occult metastases in cases of N0 stage

194 Chapter 11 REFERENCES 1. D Cruz AK, Vaish R, Kapre N, et al. Elective versus Therapeutic Neck Dissection in Node-Negative Oral Cancer. N Engl J Med 2015; 373: Dik EA, Willems SM, Ipenburg NA, et al. Watchful waiting of the neck in early stage oral cancer is unfavourable for patients with occult nodal disease. Int J Oral Maxillofac Surg 2016; 45: Rouviere H. Lymphatic system of the head and neck. Tobias MJ, Translator. Ann Arbor, MI: Edwards Brothers; Hirakawa S. From tumor lymphangiogenesis to lymphvascular niche. Cancer Sci 2009; 100: Zeidman I, Copeland BE, Warren S. Experimental studies on the spread of cancer in the lymphatic system. II. Absence of a lymphatic supply in carcinoma. Cancer 1955; 8: Karatzanis AD, Koudounarakis E, Papadakis I, et al. Molecular pathways of lymphangiogenesis and lymph node metastasis in head and neck cancer. Eur Arch Otorhinolaryngol 2012; 269: van t Veer LJ, Bernards R: Enabling personalized cancer medicine through analysis of gene expression patterns. Nature 2008; 452: Cardoso F, van t Veer LJ, Bogaerts J, et al. 70-Gene Signature as an Aid to Treatment Decisions in Early- Stage Breast Cancer. N Engl J Med 2016; 375: Roepman P, Wessels LF, Kettelarij N, et al. An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas. Nat Genet 2005; 37: Graveland AP, Braakhuis BJM, Eerenstein SEJ, et al. Molecular diagnosis of minimal residual disease in head and neck cancer patients. Cell Oncol 2012; 35: Smeets SJ, Hesselink AT, Speel EJ, et al. A novel algorithm for reliable detection of human papillomavirus in paraffin embedded head and neck cancer specimen. Int J Cancer 2007; 121:

195 Summary

196

197 CHAPTER 12 Nederlandse samenvatting

198 Chapter

199 Nederlandse samenvatting SAMENVATTING Mondkanker is de 15e meest voorkomende kankersoort wereldwijd. In 2012 bedroeg de globale incidentie meer dan (2% van het totaal) en was het aantal sterfgevallen meer dan (ook 2% van het totaal). Ongeveer 90% van alle maligne tumoren van de mondholte behoort tot de plaveiselcarcinomen, welke hun oorsprong vinden in de slijmvliezen van de lippen en de mond. Het orale plaveiselcelcarcinoom (OPCC) metastaseert frequent (ca. 50%) naar de lymfeklieren in de hals. De aanwezigheid van metastasen in de lymfeklieren is bepalend voor de prognose en de behandeling. Diagnostiek van de hals bestaat uit palpatie en beeldvorming, echter de nauwkeurigheid van deze technieken om kleine (<3mm) metastasen te detecteren is laag. Vooral bij patiënten met een vroeg stadium OPCC, gestageerd als ct1-t2n0 (tumoren tot 4 cm zonder aantoonbare metastasen in de cervicale lymfeklieren), waarbij het risico op occulte (niet detecteerbare) metastasen ongeveer 30% bedraagt, is de onnauwkeurigheid van de diagnostiek van de lymfeklieren belangrijk. Indien de hals van deze patiënten niet behandeld wordt dan zullen deze occulte metastasen zich ontwikkelen tot klinisch detecteerbare lymfekliermetastasen met een verdere verslechtering van de prognose als mogelijk gevolg. Behandeling van de primaire tumor kan bij deze patiënten vaak middels transorale resectie zonder de hals te moeten openen. Tot voor kort bestond de keuze voor behandeling van de hals uit electieve halsklierdissectie (EHD) of zorgvuldige observatie (ZO) gevolgd door therapeutische halsklierdissectie (THD) bij patiënten bij wie zich alsnog lymfekliermetastasen manifesteren. Kiest men voor behandeling middels EHD dan krijgt 70% van deze ct1-t2n0 patiënten een onnodige chirurgische behandeling van de hals met als mogelijke bijwerkingen deformiteit van de hals en morbiditeit bestaande uit pijn, schouderdysfunctie en uitval van de ramus marginalis mandibulae, een tak van de aangezichtszenuw, de nervus facialis. Het alternatief, de ZO houdt in dat de hals 3 tot 4 x per jaar echografisch beoordeeld wordt, zonodig aangevuld met dunne naald aspiratie cytologie en dat alleen die patiënten, bij wie zich een manifeste lymfekliermetastase ontwikkelt, behandeld worden met een (therapeutische) gemodificeerde radicale halsklierdissectie. Recente studies 1,2 tonen een slechtere overleving aan bij ZO. Het dilemma bestaat uit enerzijds de keuze voor oncologische veiligheid van 30% van de cn0 patiënten met een occulte metastase, die beter af zijn met een EHD en anderzijds een vermindering van morbiditeit van 70% bij patiënten die geen occulte metastasen hebben en dus gebaat zijn bij ZO. De ontdekking van een optimale voorspeller van occulte metastasen van het OPCC kan worden beschouwd als het vinden van de Heilige Graal in de hoofd-hals chirurgie

200 Chapter 12 Het doel van dit proefschrift bestond uit het vinden van alsook het verder ontwikkelen van nieuwe methoden om occulte metastasen in de hals op te sporen, ten einde een meer gepersonaliseerde behandeling van de hals mogelijk te maken. Om dit doel te bereiken werd een eerder gepubliceerd genexpressieprofiel gevalideerd, dat lymfekliermetastasering bij OPCC voorspelt. Daarnaast werd er getracht andere biomarkers gerelateerd aan lymfekliermetastasering en prognose van patiënten met OPCC te ontdekken. Het hoofd-halsgebied is voorzien van een uitgebreid lymfatisch netwerk en bevat ongeveer driehonderd van de in totaal achthonderd lymfeklieren van het menselijk lichaam. 3 Tumor lymfangiogenese, de vorming van nieuwe lymfevaten rondom maligne tumoren, is een relatief nieuw gebied ter bestudering van lymfogene metastasering van verschillende neoplasmata. 4 Het initiële concept van lymfogene metastasering hield in dat tumorcellen zich alleen verspreiden door reeds bestaande peritumorale lymfevaten, welke dienen als passieve kanalen. 5 Het gebruik van markers waarmee lymfevaten zichtbaar(der) gemaakt kunnen worden leidde tot nieuwe inzichten m.b.t. de actieve ontwikkeling van een tumor-geassocieerde lymfatische vasculatuur, geheel analoog aan het proces van neovascularisatie. Verder onderzoek heeft daarnaast een aantal groeifactoren en chemokinen, betrokken bij het proces van tumor lymfangiogenese, ontdekt en leverde informatie over de actieve rol van tumorcellen bij de ontwikkeling van een pathologische lymfangiogenese, welke dient als voorwaarde voor regionale metastasering. 6 Onderzoek naar de prognostische betekenis van tumorlymfevaten maakt gebruik van immunohistochemie (IHC) om zo de dichtheid van lymfevaten (lymfevatdichtheid, LVD) in en rondom de tumormassa te kwantificeren. Hoofdstuk 2 geeft een overzicht van de huidige kennis over de relatie tussen intratumorale en peritumorale LVD en lymfekliermetastasering met als doel het ontwikkelen van een aanvullend instrument ter voorspelling van het risico op occulte metastasen in patiënten met plaveiselcelcarcinoom van het hoofd-hals gebied (HHPCC). Het OPCC was de belangrijkste primaire maligniteit die onderzocht werd, aangevuld met tumoren afkomstig uit de keelholte en strottenhoofd. De meeste artikelen tonen een significante correlatie tussen verhoogde LVD en de aanwezigheid van lymfekliermetastasering. Een groot deel van dergelijke studies kampte met diverse beperkingen in zowel opzet als uitvoering van de studie. Voorlopige conclusie is daarom dat LVD potentieel zou kunnen dienen als marker ter detectie van occulte lymfekliermetastasen maar dat aanvullend onderzoek nodig is om dit nader uit te werken. Een andere methode ter bepaling van het risico op occulte lymfekliermetastasen is het meten van de expressie van een genprofiel in de tumor. Middels diverse genexpressieprofielen is klinische toepasbaarheid aangetoond voor de diagnose en prognose van verschillende kankers. 7,8 Ook voor plaveiselcarcinomen van het hoofd-halsgebied werd een diagnostisch genexpressieprofiel 198

201 Nederlandse samenvatting geïdentificeerd, waarmee onderscheid tussen metastaserende en niet-metastaserende tumoren afkomstig uit de mond- en keelholte gemaakt kon worden. 9 Hoewel de onafhankelijke set tumoren gebruikt ter validatie klein was (n = 22), scoorde het genexpressieprofiel een veelbelovende negatief voorspellende waarde (NVW) van 100%. In hoofdstuk 3, werd dit genexpressieprofiel 9 opnieuw geëvalueerd in een groter patiënten cohort afkomstig uit 8 ziekenhuizen (n = 222). De NVW van het genexpressieprofiel over het gehele validatie cohort (n = 222) bleek 72%. In een subset met alleen de klinisch meest relevante tumoren (ct1-t2n0) OPCC (n = 101), werd een NVW van 89% vastgesteld. Door toevoeging van het genexpressieprofiel aan de huidige klinische en beeldvormende diagnostiek is men in staat het aantal occulte lymfekliermetastasen van 28% naar 11% te reduceren. Prospectieve studies zijn nodig om te beoordelen of toevoeging van het genexpressieprofiel aan de huidige diagnostiek zal leiden tot meer gepersonaliseerde behandeling van de hals van patiënten met een klinisch vroeg stadium OPCC. Presentatie van de in hoofdstuk 3 beschreven studie tijdens het 3e Wereld Congres van de International Academy of Oral Oncology in Singapore 2011 resulteerde in een uitnodiging tot het schrijven van een review over nieuwe diagnostische methoden voor de beoordeling van de klinisch klier-negatieve hals (cn0) in patiënten met OPSCC (hoofdstuk 4). Beperkingen van beeldvormende technieken om kleine metastasen op te sporen, hebben geleid tot een zoektocht naar additionele kenmerken van, of biomarkers in de primaire tumor, ter voorspelling van (occulte) lymfekliermetastasering. In dit review, beperkten wij ons tot de bespreking van de twee meest belovende technieken: het genexpressieprofiel (GEP) en de schildwachtklier biopsie (SKB). Het SKB is een methode die de lymfeklieren met het hoogste risico op metastasen opspoort en middels chirurgische ingreep verwijderd, waarna de klieren zorgvuldig histopathologisch onderzocht worden. Een nieuw algoritme ter stagering van de cn0 hals waarin beide methoden zijn geïncorporeerd, wordt beschreven met als doel verdere optimalisatie van de behandeling van de hals bij patiënten met een vroeg stadium OPCC. Door toepassing van dit algoritme wordt oncologische veiligheid niet zozeer nagestreefd middels overbehandeling met (EHD) en de bijbehorende onnodige morbiditeit, maar meer door verdere reductie van het percentage occulte metastasen en nauwkeurige follow-up. De combinatie van het GEP en SKB als aanvulling op de huidige beeldvormende technieken, zal het aantal occulte metastasen theoretisch verder reduceren naar ongeveer 6%. Daarnaast zal het gebruik van de SKB alleen bij hoog risico op occulte lymfekliermetastasen bepaald volgens het GEP verdere invasieve overbehandeling reduceren. Prospectieve klinische studies waarin een dergelijk diagnostisch algoritme wordt geïmplementeerd is nodig om te onderzoeken of het GEP en de SKB kunnen worden gecombineerd en of oncologische en functionele uitkomsten bij patiënten met OPCC daadwerkelijk verbeteren

202 Chapter 12 Hoofdstuk 5 beschrijft hoe analyse van genexpressie ook kan worden gebruikt om het risico op het ontwikkelen van lokaal dan wel regionaal recidief te bepalen. Dit is belangrijk omdat de prognose van patiënten met HHPCC nog steeds matig is en meer dan 50% van deze patiënten binnen vijf jaar na diagnose overlijdt. Een aanzienlijk deel van deze sterfgevallen wordt veroorzaakt door lokaal dan wel regionaal recidief. 10 Dit onderzoek was gericht op het identificeren van de biologische basis van tumorkenmerken die zijn geassocieerd met de ontwikkeling van locoregionale recidieven en een slechte prognose. Perineurale- en sprieterige groei waren significant gecorreleerd met een slechtere prognose. Voor het patroon van uitgebreide sprieterige groei, maar niet voor perineurale groei, werd een genexpressieprofiel van 160 genen geïdentificeerd. Er bleek echter geen relatie tussen dit genexpressieprofiel en overleving aantoonbaar. Bij patiënten van wie de tumor middels chemoradiatie wordt behandeld en er slechts kleine diagnostische biopten beschikbaar zijn, kan bepaling van histopathologische karakteristieken als sprieterige groei lastig zijn. Het meten van genexpressie in dergelijke biopsieën kan nuttig zijn in dit opzicht. Samenvattend bleek dat middels genexpressie analyse de biologische basis van de determinanten van locoregionaal recidief werd ontrafeld. In de hoofdstukken 6-8 werd ingezoomd op een aantal specifieke genen van de het gevalideerde genexpressieprofiel ter voorspelling van lymfekliermetastasering. Correlaties met klinische en histopathologische variabelen zowel op gen- als op eiwitniveau werden geëvalueerd in patiënten met OPCC. HPV status werd bepaald middels een algoritme voor HPV-16 om HPVpositieve tumoren te excluderen. 11 Een tissue microarray (TMA) werd vervaardigd en gebruikt om expressie van de verschillende eiwitten middels immunohistochemie te meten. In hoofdstuk 6 werden de eiwitten van 4 genen uit de top 10 van meest voorspellende genen uit het GEP nader geanaliseerd: secretoire leukocyt proteaseremmer (SLPI), lipocaline-2 (LCN2), thrombospondine-2 (THBS2) en tumor geassocieerde calcium signaalomvormer 2 (TACSTD2). Gen- en eiwitexpressie werd gecorreleerd met lymfekliermetastasering, overleving (algehele overleving, AO) en ziektespecifieke overleving (ZSO). SLPI proteïne expressie correleerde wel met lymfekliermetastasering in het gehele cohort, maar niet in de relevante subgroep van de vroeg stadium (ct1-t2n0) OPCCs. SLPI expressie correleerde met AO (hazard ratio [HR] = 0,61) en ZSO (HR = 0,47) in een multivariate analyse. LCN2, THBS2 en TACSTD2 hadden geen correlatie met lymfekliermetastasering, AO of ZSO. Hoewel SLPI expressie correleerde met lymfekliermetastasering, had het geen toegevoegde waarde bij de detectie van occulte lymfekliermetastasen in patiënten met vroeg stadium OPCC. SLPI is wel een onafhankelijke voorspeller van zowel AO en ZSO en derhalve een relevante prognostische biomarker in OPCC. 200

203 Nederlandse samenvatting In hoofdstuk 7 werd een vergelijkbare analyse uitgevoerd zoals in hoofdstuk 6 alleen werden nu de proteasen kallikreïne 5 (KLK5) en kallikreïne 7 (KLK7) en hun aanverwante remmer «serine proteaseinhibitor Kazal type 5» (SPINK5) geëvalueerd als predictieve of prognostische markers in OPCC. Compleet verlies van expressie van zowel KLK5 als KLK7 (co-expressie) correleerde met een slechtere ZSO en AO. Multivariate analyse toonde dat co-expressie een onafhankelijke prognostische factor is voor ZSO(p = 0,029) en AO (p = 0,001) en kan dienen als prognostische marker bij deze ziekte. Tenslotte werd in hoofdstuk 8 de rol van cathepsine K (CTSK) als biomarker voor lymfekliermetastasering en prognose onderzocht in patiënten met OPCC. Genexpressie data werd verkregen uit eerder onderzoek 9 en eiwitexpressie werd semi-kwantitatief bepaald middels immunohistochemie op de vervaardigde TMA. Alle expressie data werd gecorreleerd met klinische en pathologische variabelen. Expressie van CTSK bleek sterk geassocieerd met lymfekliermetastasering en perineurale groei (p <0,01). Logistische regressie-analyse duidde CTSK eiwitexpressie als de belangrijkste onafhankelijke factor van lymfekliermetastasering (odds ratio = 7,65, CI: 2,31-23,31, p = 0,001). Daarnaast had CTSK expressie ook significante invloed op de 5-jaars ZSO (HR = 2,29, CI: 1,01 5,21, p = 0,047 voor genexpressie; HR = 2,79, CI: 1,02-7,64, p = 0,045 voor eiwitexpressie). Middels ROC-analyse werd een afkappunt van 25 bepaald, om patiënten in te delen in groepen met lage en hoge CTSK eiwitexpressie. Hoge CTSK expressie correleerde significant met lymfekliermetastasering (p <0,01). In de klinisch relevante subgroep, bestaande uit patiënten met ct1-t2n0 OPCC (n=24), bleek de voorspellende waarde voor occulte lymfekliermetastasen van CTSK groot. Van de tien patiënten met occulte lymfekliermetastasen in de hals, hadden negen een hoge CTSK eiwitexpressie, terwijl slechts één patiënt een CTSK-waarde lager dan de cut-off bleek te hebben. De sensitiviteit van CTSK eiwitexpressie ter detectie van occulte metastasen in ct1-t2n0 OPCC was 90%, de specificiteit 57%, de positieve voorspellende waarde 60% en de negatief voorspellende waarde 89%. De huidige studie was gebaseerd op een relatief beperkt cohort van 83 patiënten met OSCC. De resultaten dienen verder gevalideerd te worden in bij voorkeur prospectieve studies met een groter aantal patiënten. Eerste stappen zijn reeds gezet t.b.v. een dergelijke studie en zodra bekend zullen de resultaten gepubliceerd worden

204 Chapter 12 REFERENTIES 1. D Cruz AK, Vaish R, Kapre N, et al. Elective versus Therapeutic Neck Dissection in Node-Negative Oral Cancer. N Engl J Med 2015; 373: Dik EA, Willems SM, Ipenburg NA, et al. Watchful waiting of the neck in early stage oral cancer is unfavourable for patients with occult nodal disease. Int J Oral Maxillofac Surg 2016; 45: Rouviere H. Lymphatic system of the head and neck. Tobias MJ, Translator. Ann Arbor, MI: Edwards Brothers; Hirakawa S. From tumor lymphangiogenesis to lymphvascular niche. Cancer Sci 2009; 100: Zeidman I., Copeland B. E., Warren S. Experimental studies on the spread of cancer in the lymphatic system. II. Absence of a lymphatic supply in carcinoma. Cancer 1955; 8: Karatzanis AD, Koudounarakis E, Papadakis I, et al. Molecular pathways of lymphangiogenesis and lymph node metastasis in head and neck cancer. Eur Arch Otorhinolaryngol 2012; 269: van t Veer LJ, Bernards R: Enabling personalized cancer medicine through analysis of gene expression patterns. Nature 2008; 452: Cardoso F, van t Veer LJ, Bogaerts J, et al. 70-Gene Signature as an Aid to Treatment Decisions in Early- Stage Breast Cancer. N Engl J Med 2016; 375: Roepman P, Wessels LF, Kettelarij N, et al. An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas. Nat Genet 2005; 37: Graveland AP, Braakhuis BJM, Eerenstein SEJ, et al. Molecular diagnosis of minimal residual disease in head and neck cancer patients. Cell Oncol 2012; 35: Smeets SJ, Hesselink AT, Speel EJ, et al. A novel algorithm for reliable detection of human papillomavirus in paraffin embedded head and neck cancer specimen. Int J Cancer 2007; 121:

205 Nederlandse samenvatting

206

207 CHAPTER 13 List of publications

208 Chapter 13 LIST OF PUBLICATIONS Lymphatic vessel density as a marker of lymph node metastasis in head and neck squamous-cell carcinoma. A systematic review of the literature. Koudounarakis E, Leusink FKJ, Willems SM, Brekel van den MW, Zuur CL. Accepted in European Journal of Surgical Oncology, Pain as primary symptom of a malignant parotid tumor. Pijn als primair symptoom van een maligne parotis tumor. Kamp van der M, Leusink FKJ, Al Mamgani A, Lohuis PJ, Brekel van den MW. Accepted in Nederlands Tijdschrift voor Tandheelkunde, 2016, A9815, Dutch. Cathepsin K associates with lymph node metastasis and poor prognosis in oral squamous-cell carcinoma. Leusink FKJ, Koudounarakis E, Frank MH, Broekhuizen R, Braunius W, Hooff van SR, Diest van PJ, Koole R, Willems SM. Submitted to BMC Cancer. Tumor biological determinants of locoregional recurrence of non-hpv head and neck squamous-cell carcinoma. Leusink FKJ, Braakhuis BJM, Wieringen van WN, Bloemena E, Rustenburg F, Ylstra B, Kummer JA, Diest van PJ, Voorham Q, Smeets S, Roepman P, Koole R, Leemans CR, Brakenhoff RH. In preparation. The co-expression of kallikrein 5 and kallikrein 7 associates with poor survival in non- HPV oral squamous-cell carcinoma. Leusink FKJ, Diest van PJ, Frank MH, Broekhuizen R, Braunius W, Hooff van SR, Willems SM, Koole R. Pathobiology, 2015; 82: Nodal metastasis and survival in oral cancer: Association with protein expression of SLPI, not with LCN2, TACSTD2, or THBS2. Noorlag R, van der Groep P, Leusink FKJ, Hooff van SR, Frank MH, Willems SM, van Es RJ. Head Neck, 2015; 37: Novel diagnostic methods for detection of occult lymph node metastases in oral cancer. Nieuwe onderzoeksmethoden om occulte halskliermetastasen van mondkanker op te sporen. Leusink FKJ, Takes RP, Brakenhoff RH, Bree de R, Es van RJ. Het Tandheelkundig Jaar 2014, Bohn Stafleu van Loghum, ISBN

209 List of publications Cannabinoid receptor-2 immunoreactivity is associated with survival in squamous-cell carcinoma of the head and neck. Klein Nulent TJ, Van Diest PJ, van der Groep P, Leusink FKJ, Kruitwagen CL, Koole R, Van Cann EM. Br J Oral Maxillofac Surg, 2013; 51: Validation of a gene expression signature for assessment of lymph node metastasis in oral squamous-cell carcinoma. Leusink FKJ, Hooff van SR, Roepman P, Baatenburg de Jong RJ, Speel EM, Brekel van den MW, Velthuysen van MF, Diest van PJ, Es van RJ, Merkx MA, Kummer JA, Leemans CR, Schuuring E, Langendijk JA, Lacko M, De Herdt MJ, Jansen JC, Brakenhoff RH, Slootweg PJ, Takes RP, Holstege FC. Journal of Clinical Oncology, 2012; 30: Novel diagnostic modalities for the clinically node negative neck in oral squamous-cell carcinoma. Leusink FKJ, Es van RJ, Bree de R, Baatenburg de Jong RJ, Hooff van SR, Holstege FC, Slootweg PJ, Brakenhoff RH, Takes RP. Lancet Oncology, 2012; 13: e Downregulation of Serpin B13 expression in head and neck squamous-cell carcinomas associates with poor clinical outcome. Koning de PJ, Bovenschen NA, Leusink FKJ, Broekhuizen R, Quadir R, van Gemert JT, Hordijk GJ, Chang WS, van der Tweel I, Tilanus MG, Kummer JA. International Journal of Cancer, 2009; 125:

210