Applying the power of MetaCore to

Transcription

1 Applying the power of MetaCore to Next Generation Sequencing Data: Uncovering potential mechanisms for Interferon alpha induced Thyroiditis case study Introduction Interferon (IFN) alpha was the first cytokine to be reproduced by recombinant DNA technology and has emerged as a major therapeutic intervention for several disease including melanoma, renal cell carcinomas, hairy cell leukemia, hepatitis B and most commonly, Hepatitis C (Baron S, et al., 1991). However, despite its successful therapeutic actions, IFN alpha therapy is also associated with a common side effect profile that includes the development of autoimmune disease and reports of the development of type I diabetes mellitus, psoriasis, rheumatoid arthritis, systemic lupus-like syndromes, and thyroid disease. In particular IFN alpha induced thyroiditis (IIT) is a major clinical problem and has been reported to develop in a range from percent of patients receiving IFN alpha therapy (Reviewed in Oppenheim Y et al., 2003). A new classification for IIT has been suggested based on the epidemiological characteristics of IIT where IIT is described as either Autoimmune IIT or Nonautoimmune IIT. Autoimmune IIT can manifest by the development of thyroid antibodies with or without clinical disease. Non-autoimmune IIT can manifest as destructive thyroiditis or as hypothyroidism with negative thyroid antibodies. (Reviewed in Menconi and Tomer, 2009). This suggests IFN alpha induces tissue inflammation through atypical immune as well as non-immune mechanisms however, neither of these mechanisms are well understood. To further elucidate the immune and non-immune mechanisms induced by IFN alpha, the current study applied a pathway analysis approach using Thomson Reuters MetaCore to Next Generation Sequencing (RNAseq) data generated from an in vivo mouse model of IFN alpha induced Thyroiditis (IIT). The current study demonstrates one of the logical and powerful workflows available using Thomson Reuters Systems Biology Solutions. Through the application of unique tools and manually curated content, Next Generation Sequencing data was analyzed and interpreted to generate hypothesis based on strong experimentally derived information. These hypotheses could save valuable time and money as they provide a strong foundation to potentially drive fast and accurate decisions for the direction for future research and significant findings. Materials and Methods Next generation sequencing data used for this study was obtained from the transcriptome analysis (using RNAseq) of an in vivo IIT mouse model (Akeno et al., 2011). Briefly, thyroid tissue RNA was dissected, pooled and isolated from the thyroid tissues of 10 transgenic mice with thyroid specific overexpression of IFN alpha at 8 months of age (TG) and a pool of control thyroid tissues obtained from 8 litter mate controls (WT) (Akeno et al., 2011). For the two samples (TG and WT) mrna was extracted, fragmented and used to prepare a cdna library compatible with the Illumina Next Generation Sequencing technology and sequencing was performed on an Illumina Genome Analyzer IIx as described (Akeno et al., 2011). Gene expression in each sample is proportional to the number of reads aligning to a specific gene sequence and reads were aligned to the mouse (mm9) genome where the resulting counts were calculated and normalized to relative abundance in Fragment per kilobase of exon model per Million (FPKM). This data was obtained from the public GEO database under Series GSE25115 ( nlm.nih.gov/geo/query/acc.cgi?acc=gse25115). The FPKM values for samples GSM (mouse_wt_ RNA-seq) and GSM (mouse-tg_rna-seq) were used to generate fold change values for the differentially expressed genes of the mouse-tg_rnaseq sample relative to the mouse_wt_rna-seq sample (DEGs TGvsWT) applying similar methodology outlined in the corresponding publication (Akeno et al., 2011). The samples were read into R and in order to enable quantitative comparisons between the samples, only those transcripts which had values available in both samples were compared. Samples were normalized by scaling to have the same total mean expression value and only genes with > 50 FPKM in at least 1 sample were retained. These settings were slightly more stringent than the authors who accepted genes where one sample had 50 FPKM. The differential expression analysis was then performed using the DEGseq package and the MA method with random sampling. Genes were filtered where a fold change +/- 2 was retained. The fold change values for the resulting 927 mouse RefSeq IDs were converted to a log2 ratio and uploaded into MetaCore for pathway analysis. Of the 927 RefSeq IDs, 926 were recognized by MetaCore.

2 Figure 1: (b) Metacore Analysis Workflow for mouse Thyroid tissue DEGs TGvs WT. MetaCore is an integrated software suite for functional analysis of experimental data and gene lists including resulting originating from Next Generation Sequencing. MetaCore is based on a high-quality, manually-curated database providing accurate systems biology content and provides an array of sophisticated functional tools to interpret and analyze high throughput data. The current study applies a logical workflow incorporating a selection of these tools to analyze the Next Generation Sequencing data set and elucidate potential mechanisms of IFN alpha in Thyroiditis (refer to Figure 1 for an overview of this workflow). RESULTS Figure 2 MetaCore Canonical Pathway Maps histograms after enrichment analysis with the DEGs from thryoid tissue of Mouse TGvsWT filtered for a fold change +/- 2. The top ten canonical maps for enriched genes are reported as histogram plots as shown. 1. Enrichment analysis: MetaCore Pathway Maps Ontology The manually curated MetaCore Pathway Maps ontology was used to identify canonical Pathway Maps significantly enriched by the TGvsWT dataset. Using the MetaCore variation of the Fisher s exact test and adjusting for multiple sample testing using (FDR) pathway enrichments were more stringent and ensured more accurate results (Figure 2). Consistent with the over expression of IFN alpha in the thyroid of TG mice, the most significantly enriched canonical pathways included Immune response_antiviral actions of interferons and Immune response_ifnalpha/ beta signaling pathway. Other significantly enriched maps included immunological responses such as Antigen presentation by MHC class I and complement system pathways (Figure 2).

3 2. Interactions by Protein Function When exploring transcriptional data it is important to also consider that important regulatory elements may be altered at biological levels other than the transcriptional, which can easily be missed when looking exclusively at transcriptional data. For example, transcription factors often have no detectable change in transcription but change in activity, promoter binding affinity, etc. To further extend the current analysis and try to capture such hidden regulators, the Interactome workflow in MetaCore was utilized using the log2 ratios of the DEGs from mouse thyroid tissue in TGvsWT. The resulting list of 138 potential regulators (Figure 3) was then also used in combination with previous results to get a larger multi-level approach to the analysis. 3. Enrichment analysis: using results from Interactions by Protein Function tool & Pathway Maps Ontology MetaCore allows easy export of all results to either external software such as Excel or as a new file within the tool for further analysis. By saving the interactome analysis results as an experimental list, it could then be used in enrichment analysis of the MetaCore Pathway Maps ontology. This additional analysis compliments the initial enrichment analysis of the DEGs TGvsWT and can also highlight additional canonical pathways that may further elucidate the mechanisms of IFN alpha in inducing Thyroiditis in the current IIT model. Results shown in Figure 4 include the same three Immune pathways identified in Figure 2. And the two lists (1) Figure 3 Results of Interactome analysis for network objects associated with DEGS from thyroid tissue of mouse TGvsWT. The top results for each protein class are shown following filtering by by z-value. Network objects significantly overconnected and also that appears in the original DEG list are highlighted in orange. Columns are as follows: Actual - number of targets in the activated datasets regulated by the chosen network object, n - number of network objects in the activated datasets, R - number of targets in the complete database or background list regulated by the chosen network object N - total number of gene-based objects in the background list, Expected - mean value for hypergeometric distribution (n*r/n), Ratio- connectivity ratio (Actual/Expected), p-value - probability (from the hypergeometric distribution) of seeing the observed value of r or higher, z-score - ((r-mean)/ sqrt(variance)). Figure 4 MetaCore Canonical Pathway Maps histograms after enrichment analysis with Network objects significantly overconnected with DEGs from thryoid tissue of Mouse TGvsWT filtered for a fold change +/- 2. The top thirteen canonical maps for enriched genes are reported as histogram plots as shown.

4 Applying the power of MetaCore to Next Generation Sequencing Data: Uncovering potential mechanisms DEGs and (2) Overconnected network objects can be overlaid onto the MetaCore Canonical Pathway Maps and highlight continuous mechanistic pathways (Figure 5 and 6). Significant up-regulation of the canonical pathway Immune response_antigen presentation by MHC class I demonstrates increased proteosome activity (which is commonly correlated with disease activity in autoimmunity) and antigen processing and presentation by MHC class I (Figure 5). IFN alpha is well known to increase MHC class I antigen expression on cells, including thyroid epithelial cells (Roti, E et al., 1996) and is further confirmed by the Figure 5 MetaCore pathway map Immune response_antigen presentation by MHC class I. The thermometers represent the different datasets overlayed on map: (1) DEGs from thyroid tissue of Mouse TGvsWT (ref). (2) Proteins significantly overconnected to the list of DEGs from thyroid tissue of Mouse TGvsWT. The vertical thermometers represent the log2 ratios for the DEGs (red represents up-regulation) and overconnected network objects are given a default value of +1. Figure 6 MetaCore pathway map Immune response_antiviral actions of Interferons. The thermometers represent the different datasets overlayed on map: (1) DEGs from thyroid tissue of Mouse TGvsWT (ref). (2) Proteins significantly overconnected to the list of DEGs from thyroid tissue of Mouse TGvsWT. The vertical thermometers represent the log2 ratios for the DEGs (red represents up-regulation) and overconnected network objects are given a default value of +1.

5 significantly enriched Immune response _Antiviral actions of interferons pathway in the mouse thyroid tissue of TGvsWT (Figure 6). Results reported from the publication in which the current data set was taken (Akeno et al., 2011) also reported a significant increase in thyroid specific proteins (Tg) when a rat thyroid cell line (PCCL3) is induced with IFN alpha in vitro (data not shown). Combined with an increase in antigen processing and presentation by MHC class I it is reasonable to hypothesis that thyroid self antigens could be the antigens presented to the adaptive immune system in the thyroid tissue of TG mice and potentially be a immune modulated mechanism in which IFN alpha induces autoimmune thyroiditis. Enrichment analysis of the proteins significantly overconnected to the original data set can also highlight additional canonical pathways that may further elucidate the mechanisms of IFN alpha in inducing thyroiditis. As shown from the enrichment results in Figure 4 the IL-6 signaling pathway is also highlighted where IL-6 is known to be induced by IFN alpha and is also associated with autoimmune Thyroiditis (Ajjan et al., 1996). An additional immune response pathway Immune response-il-7 signaling in T lymphocytes is highlighted and is involved in the is an essential cytokine pathway for proliferation, maintenance and survival of T-lymphocytes (Kittipatarin C et al., 2001). 4. Enrichment analysis: Pathway Map Folders Ontology The MetaCore Pathway Map Folders ontology is a collection of manually created pathway maps, grouped hierarchically into folders according to main biological processes. This ontology allows the exploration of the data at a higher level of the hierarchical tree and often can elucidate significant changes to larger categories that can be overlooked when looking at the individual pathways themselves. Using this ontology allows identification of key biological processes from the DEGs of TGvsWT in a single click. This analysis also confirmed a predominant enrichment of Immune system responses (Figure 7). Observations originally published with the current dataset also revealed thyroid cell death both in vivo (during thyroid dissection of the current animals analyzed) and in vitro (from IFN alpha stimulated Rat thyroid cell line (PCCL3)). The authors excluded apoptosis from their in vitro studies by a Vybrant apoptosis assay kit (Akeno et Figure 7 MetaCore Canonical Pathway Map Folders histograms after enrichment analysis with the DEGs from thryoid tissue of Mouse TGvsWT filtered for a fold change +/- 2. The top ten Pathway Map Folders representing biological processes are reported as histogram plots as shown. Bold and faded Orange bars show if the pathway map folder passed or failed the FDR analysis filer, respectively. al., 2011). A similar result for the in vivo IIT model can be hypothesized by the current pathway analysis approach as there is a distinct absence of enrichment in any apoptopic canonical pathway map (Figure 2 and 4) and the apoptoptic pathway map folder (Figure 7). Collectively these results suggest the observed cell death is not mediated by mechanistic pathways of apoptosis. 5. Creating a custom ontology representing toxic pathology types In order to identify potential toxic pathologies associated with IIT in the current model, the DEGs from the thyroid tissue of TGvsWT were enriched using custom built networks based on specific toxic pathologies. The Thomson Reuters Systems Toxicology Module adds manually curated content to MetaCore focusing on specific toxic pathology types annotated specifically to 12 major organs. As many toxic pathology mechanisms and biomarkers overlap between organs (Cobb et al., 1996), manually curated biomarkers associated with the different toxic pathology types were pooled in this study from the 12 major organs annotated. To this end the advanced search tool in MetaCore was utilized and biomarkers associated with different toxic pathology types were extracted from the content knowledgebase add on provided with the Systems Toxicology knowledgebase. These were then used to build custom networks representing biomarkers for 12 main toxic pathology types. This was quickly achieved using the MetaCore network building algorithm: Direct interactions with canonical pathways. The resulting networks were saved and enrichment of the DEGs TGvsWT was performed from the new custom Toxic Pathology types ontology generated from manually curated toxicology biomarkers. 6. Perform Enrichment Analysis using custom ontology for toxic pathology types Inflammation was the top toxic pathology type enriched by the DEGs of thyroid tissue of mouse TGvsWT (Figure 8) and is consistent with non-immune IIT (destructive thyroiditis) (Tomer, Y & Menconi, F. 2009) and is secondary to necrotic cell death. Necrosis was the second most enriched toxic pathology while apoptosis was the least furthering strengthening the findings that thyroid cell death observed in the current IIT in vivo model is by necrosis rather than apoptosis.

6 Figure 8 Custom toxicology pathology networks histograms after enrichment analysis with Network objects significantly overconnected with DEGs from thryoid tissue of Mouse TGvsWT filtered for a fold change +/- 2. Twelve networks representing manually curated biomarkers for 12 well annotated toxic pathology types were generartd and used for enrichment in the create your own ontology feature of MetaCore. Bold and faded Orange bars show if the pathway map folder passed or failed the FDR analysis filer, respectively. SUMMARY & CONCLUSION The current study used Next Generation Sequencing data from an in vivo IFN alpha Induced Thyroiditis (IIT) mouse model. The data set was obtained from the GEO database (GSE series 25115) and read counts from the thyroid tissue of transgenic mice overexpressing IFN alpha (TG) and wildtype mice (WT) were processed and uploaded into MetaCore. Various tools within MetaCore were applied including enrichment analysis of the log2 ratios from the DEGs of TGvsWT using the manually curated ontologies, canonical pathway maps and pathway map folders as well as a custom ontology built from networks of various toxic pathology types. A summary of the results is provided in figure 9. Figure 9 Summary of findings for the current case study highlighting the original aim and Metacore tools applied for analysis. STUDY AIM 1. To identify key canonical pathways induced by IFN alpha in the in vivo IIT model THOMSON REUTERS SYSTEMS BIOLOGY SOLUTIONS APPROACH Metacore Pathway Maps Ontology: One click enrichment analysis KEY RESULTING FINDINGS IN THIS STUDY 1. Top significantly enriched pathway maps include: Immune response_antiviral actions of interferons Immune response_ifnalpha/ beta signaling pathway Immune response _Antigen presentation by MHC class I 2. Increased proteosome activity and upregulation of antigen processing and presentation by MHC class I A list of 138 network objects that are significantly overconnected to DEGs TGvsWT that were used to expand enrichment analysis and visualised on Pathway Maps. 1. Further validated results from (1) where same three maps were significantly enriched 2. Highlighted additional pathway maps relevant to in vivo ITT model Immune response-il-7 signaling in T lymphocytes Immune response_il-6 signaling pathway 1. Main biological process enriched was Immunological Responses. 2. Absence of the pathway map folder for Apoptosis. 1. Inflammation was the most significant toxic pathology type enriched by the DEGs of TGvsWT. 2. Necrosis was the second most significant toxic pathology while Apoptosis was the least significant. 2. Identify regulatory components responsible for the biological events induced by IFN alpha in the in vivo IIT model Interactome Analysis 3. To extend analysis and perform enrichment using results from step 2 and further elucidate potential mechanisms of IFN alpha in inducing Thyroiditis in the current IIT model. One click enrichment analysis using results from 2 with Metacore Pathway Maps Ontology 4. Identify the main biological processes induced by IFN alpha in the in vivo IIT model One click enrichment analysis using Pathway Map Folders Ontology 5. Identify key toxic pathology types induced by IFN alpha in the in vivo ITT model One click enrichment analysis using custom Toxic Pathology Ontology consisting of 12 manually built networks from extracted from Thomson Reuters Systems Toxicology Module Knowledge base

7 Overall the results drew similar findings to the authors where IFN alpha upregulated several immune response pathways in the Thyroid of TG mice consistent with increased IFN alpha levels. There was no significant enrichment of any apoptopic pathway maps or map folders in the in vivo IIT model and inflammation and necrosis were the most significantly enriched toxic pathology types. Apoptosis was the least enriched toxic pathology and reflects similar results found by the authors in vitro where observed cell death was necrosis and not apoptosis. Interestingly, in the current analysis there was no significant upregulation of the Granzyme B signaling pathway referred to during the enrichment analysis by the authors (refer Akeno et al., 2011). Granzyme B is responsible for rapid induction of caspase-dependent apoptosis (Bots et al., 2006) and the 20 percent upregulation of the Granzyme B signaling pathway (upregulated genes were Granzyme B and LaminB1) demonstrated this pathway as significantly upregulated in the authors analysis. In the current analysis Granzyme B was not a significant DEG due to the more stringent FPKM cut off-of >50 rather than a minimum of 50 in one of the two samples. (Granzyme B s FPKM = 50 and 2 for the TG and WT, respectively). Robust methodology in processing read counts is necessary to account for the pooling of RNA tissue and demonstrates the importance of processing FPKM prior to pathway analysis upload. In the current analysis, two DEGs were up-regulated in the MetaCore Granzyme B signaling pathway (Caspase-7 and LaminB1); however this pathway was not significantly enriched and did not appear in the initial and extended enrichment analysis conducted in MetaCore (Refer to figures 2 and 4). Collectively the results of the current study suggest that IFN alpha therapy could be inducing thyroid cell death (by the way of necrosis) and the up-regulation of antigen processing and presentation by the MHC class I molecule events. These results offer a mechanism in which thyroid self antigens are released following necrotic cell death and the increase in proteolysis activity and MHC class I expression could suggest the self antigens are being presented to the adaptive immune system causing the autoimmune response. The current study applied a unique pathway analysis workflow and produce potentially important, robust and actionable conclusions that are a direct result of the high quality, 100 percent manually curated content of the Thomson Reuters Systems Biology Solutions. The combination of a biological interactions knowledgebase built on robust interaction information (quality controlled at the experimental design level) with intuitive yet powerful bioinformatic algorithms allows researchers to make better decisions as to next steps, faster. Using MetaCore, high throughput Next Generation Sequencing data was analyzed and identified key biological pathways and processes involved in IFN alpha Induced Thyroiditis. In addition, when used in combination with the Thomson Reuters Systems Toxicology Module, MetaCore also identified significant toxic pathologies potentially induced by IFN alpha in vivo and were consistent with previous findings in vitro.

8 REFERENCES 1) Baron S, Tyring SK, Fleischmann WR Jr, Coppenhaver DH, Niesel DW, Klimpel GR, Stanton GJ, Hughes TK (1991). The interferons. Mechanisms of action and clinical applications. JAMA. 266(10): ) Oppenheim Y, Ban Y, Tomer Y (2004). Interferon induced Autoimmune Thyroid Disease (AITD): a model for human autoimmunity. Autoimmun Rev. Jul; 3(5): ) Tomer, Y and Menconi, F (2009). Interferon induced thyroiditis. Best Pract Res Clin Endocrinol Metab. 23(6): 703 4) Akeno N., Smith, E.P., Stefan, M., Amanda K. Huber, A.K., Zhang, W., Keddache, M., and Tomer, Y. (2011). Interferon-alpha mediates the development of Autoimmunity both by direct tissue toxicity and through immune-cell recruitment mechanisms. J Immunol. 186(8): ) Roti E, Minelli R, Giuberti T, Marchelli S, Schianchi C, Gardini E, Salvi M, Fiaccadori F, Ugolotti G, Neri TM, Braverman LE. (1996) Multiple changes in thyroid function in patients with chronic active HCV hepatitis treated with recombinant interferon-alpha. Am J Med;101(5): ) Ajjan RA, Watson PF, McIntosh RS, Weetman AP (1996) Intrathyroidal cytokine gene expression in Hashimoto s thyroiditis. Clin Exp Immunol ;105(3): ) Kittipatarin., C. And Khaled AR (2007) Interlinking interleukin-7. Cytokine;39(1): ) Cobb JP, Hotchkiss RS, Karl IE, Buchman TG (1996) Mechanisms of cell injury and death, Buchman Br J Anaesth. 77(1): ) Bots M, Medema JP Granzymes at a glance. Journal of cell science 2006 Dec 15;119(Pt 24): THOMSON REUTERS Regional Offices Contact us to find out more about MetaCore or visit thomsonreuters.com/diseaseinsight North America Philadelphia Latin America Europe, Middle East and Africa Barcelona London Asia Pacific Singapore Tokyo For a complete office list visit: ip-science.thomsonreuters.com/ contact LS SCI-NGSCS Copyright 2012 Thomson Reuters