MICROSCOPY RESEARCH AND TECHNIQUE 51:464 468 (2000) Single-Pinhole Confocal Imaging of Sub-Resolution Sparse Objects Using Experimental Point Spread Function and Image Restoration A. DIASPRO,* S. ANNUNZIATA, AND M. ROBELLO INFM, Biophysical Section, Genoa Research Unit and Department of Physics, University of Genoa, Via Dodecaneso 33, 16146 Genova, Italy KEY WORDS: confocal fluorescence microscopy; subresolution imaging; non linear image restoration; three-dimensional microscopy; cellular biophysics ABSTRACT Sparse fluorescent pointlike subresolution objects have been imaged using a diffraction limited single-pinhole confocal fluorescence microscope. A Maximum likelihood image restoration algorithm has been used in conjunction with a measure of the experimental point spread function for improving the three-dimensional imaging of subresolution sparse objects. The experimental point-spread-function profiles have been improved by a factor of 1.95 in lateral direction and 3.75 in axial direction resulting in full-width half maximum (FWHM) values of 91 11 nm and 160 26 nm. This amounts to 1.43 and 2.15 in optical units, respectively. The lateral and axial FWHM of the sparse pointlike subresolution objects is about 5 and 3 times smaller than the wavelength. This result points to the attractive possibility of utilising a compact confocal architecture for localising punctuate fluorescent objects having subresolution dimensions. The key resides in the utilisation of the measured point spread function coupled to an appropriate image restoration approach, and, of course, in the stability of the confocal system being used. Microsc. Res. Tech. 51:464 468, 2000. 2000 Wiley-Liss, Inc. INTRODUCTION The three-dimensional localisation of various subcellular components is one of the major determinants for understanding the delicate and complex relationship existing between structure and function in biological systems (Shotton, 1989). The confocal microscope provides a practical, non-invasive, and well-established method to obtain microscopic three-dimensional images and to perform both functional and structural studies of biological systems and related biostructures (Pawley, 1995; Wilson, 1990). Two of the most important properties of the confocal microscope are given by the improvement of resolution by a factor of 1.4 over its conventional counterpart (Brakenhoff et al., 1989) and by the optical sectioning ability of fluorescent objects (Wilson, 1990). Because of diffraction, the image of a fluorescent point, which is in focus, is not the very same point but a small patch, called diffraction pattern, whose intensity distribution is more precisely defined as the point spread function or impulse response function of the instrument (Bianco and Diaspro, 1989; Sheppard, 1989). This spreading over a certain spatial volume depends on the geometry of the aperture of the optical system. The related physical constraint imposed to resolution cannot be surpassed in far-field acquisition schemes such as those commonly used in wide field and in confocal optical microscopy (Bertero and Boccacci, 1998; Wilson, 1990). Subdiffraction resolution has been achieved by disabling the fluorescence from the outer part of the focal spot in terms of stimulated-emission depletion (STED) (Klar and Hell, 1999), by using especially calculated image-plane masks (Brand et el., 1999) or structured illumination (Heinzmann and Cremer, 1998) combined with an interferometric technique (I 5 M) in which the sample is observed and/or illuminated from both sides simultaneously using two opposing objective lenses (Gustafsson et al., 1999). Also 4Pi-confocal microscopy can provide a focus that is sharpened up by physical methods (Hänninen et al, 1995; Hell and Stelzer, 1992; Hell et al., 1994). Unfortunately, the above-mentioned methods need to modify the microscope architecture significantly and are not of immediate access for the majority of users. Moreover, any technology, such as I 5 M or 4Pi-confocal, which accesses focal plane within the sample from both sides, is inherently limited to a class of reasonably thin and transparent samples able to minimize possible perturbations coming from refractive index variations, too (Gustafsson et al., 1999). In terms of linear and space invariant systems (Castleman, 1996), the image produced by the microscope is the convolution of the object with its own point spread function. This allows one to consider that a further image enhancement can be achieved by solving the deconvolution problem, which consists of the restoration of the characteristics of an object from a given image and a given point spread function (Bertero and Boccacci, 1998; Schrader et al., 1996; van der Voort and Contract grant sponsor: INFM. *Correspondence to: A. Diaspro, INFM, Biophysical Section, Genoa Research Unit and Department of Physics, University of Genoa, Via Dodecaneso 33, 16146 Genova, Italy. E-mail: diaspro@fisica.unige.it Received 2 February 1999; accepted in revised form 9 May 2000 2000 WILEY-LISS, INC.
CONFOCAL IMAGING OF SUB-RESOLUTION SPARSE OBJECTS 465 Strasters, 1995). This problem is called image restoration. The question then becomes whether it is possible to improve imaging properties substantially, even in such a very particular case as the imaging of sparse beads having a subresolution diameter. It has to be clearly stated and understood that this is different from discriminating densely spaced objects below the resolution. However, image restoration offers a strategy for image enhancement without the need to modify the optical apparatus significantly or to utilise near field detection schemes that suffer from the drawback of being surface analysis techniques (Betzig and Trautman, 1992). In the present work, a robust maximum likelihood image restoration algorithm (Bertero and Boccacci, 1998; van Kempen et al., 1996) has been used in conjunction with a measure of the experimental point spread function (Diaspro et al., 1999) to improve the discriminating capabilities in three-dimensional imaging of sparse subresolution objects acquired with a single pinhole confocal laser scanning microscope. We carried out our experiments with three intentions. First, we applied mathematical techniques to explore the extent to which the three-dimensional point spread function can be reduced and to demonstrate the achievement of a significant gain in the localisation of sparse subresolution fluorescent particles. Second, we aimed to utilise a procedure widely applicable to the majority of operating confocal microscopes. In order to fulfill this second intention, we used a measured point spread function acting as the fingerprint of the microscope and a robust and widely disseminated scheme for image restoration. The measure of the point spread function does not need to be repeated for any acquisition session if the operating conditions are quite similar. Third, we wanted to show that a significant enhancement is possible in fluorescence imaging using a single-pinhole scanning-head. Because of further limitations imposed by the photochemistry of the fluorescent molecules, the results are less spectacular than in the case of high reflective gold pointlike scatterers (Schrader et al., 1996). MATERIALS AND METHODS We used an advanced and ultracompact laser scanning confocal microscope system (PCM2000, Nikon SpA, Florence, Italy) based on a galvanometer pointscanning mechanism, a single pinhole optical path, and a very efficient all-fiber optical system for light delivery, both in excitation and in collection. The all-fiber solution avoids a dangerous vibration decoupling between the laser source and the scanning head even if they are placed on different tables. Moreover, by using three mechanically fixed possible pinhole diameters (20 m, 50 m, open), it is possible to repeat the same effect with excellent reproducibility and system stability. Photomultiplier electronic noise is greatly reduced by their positioning within the control unit, saving and avoiding background reflection signal collection within the scanning head. The scanning head is mounted on an inverted optical microscope Nikon Eclipse TE 300. We used a Plan Fluor oil immersion objective 100X/NA 1.3, the 488-nm line of an Argon-ion laser and standard side-window Hamamatsu R928 photomultiplier tubes. The 20- m diameter pinhole resulted in a back-projected radius two times smaller than the back-projected Airy disk. Our subresolution pointlike scatterers were 64 9-nm diameter fluorescent (excitation at 488 nm, emission at 515 nm) latex beads (Cat no. 17149, Polyscience, Warrington, PA). A drop of dilute sample of beads suspension was put on a coverslip of nominal thickness 0.17 mm and air-dried in a dust clean chamber. The beads were then covered with a drop of glycerol. Three-dimensional images of the stack were recorded as a set of 21 images (1,024 1,024 pixels) taken at 100-nm intervals along the z axis, with an integration period for each pixel of 38 s. The pixel size was 26 nm in the optical plane. RESULTS AND DISCUSSION As the size of the bead is less then one-seventh the wavelength in the medium and the beads are sufficiently sparse, we can also assume that each of the beads being imaged represents the experimental confocal point spread function. Figure 1 shows the three-dimensional rendering of a portion of the field of the imaged subresolution objects before and after image restoration. The experimental lateral and axial Full Width at Half Maximum (FWHM) are determined as 178 21 nm and 509 49 nm, respectively. These data, obtained by averaging over more than 30 objects within the acquired frames, are in good agreement with the theoretical values of 180 nm and 480 nm for the lateral and axial FWHM calculated using Huygens2 (Scientific Volume Imaging, The Netherlands) by means of scalar theory (van der Voort and Strasters, 1995) (Fig. 2). Moreover, these data are in agreement with an earlier evaluation of the point spread function of the system used (Diaspro et al., 1999). In optical units, the experimental lateral and axial FWHM is 2.80 and 6.93, respectively. We used one of the three-dimensional point spread functions we measured for our system and corrected it for a slight broadening using Huygens2. We used this as the point spread function for deconvolving the stack of subresolved pointlike fluorescent objects. It is worthy of note that this three-dimensional point spread function was acquired in an independent session and represents the fingerprint of our scanning head under the acquisition conditions described above. We used the iterative maximum likelihood estimation algorithm implemented within the Huygens2 software because it is optimally suited to restore low signal images (Bertero and Boccacci, 1998). It essentially uses a positivity constraint and is a non-linear image restoration approach. We checked the processed data in the Fourier domain or in terms of optical transfer function in order to identify possible computational artefacts. In this case, we had no computational artefacts. Moreover, we considered single particles. The experimental lateral and axial FWHM is improved by a factor of 1.95 in the lateral direction and 3.75 in the axial direction, resulting in 91 11 nm and 161 27 nm resolution, respectively. This amounts to a lateral and axial extent of the intensity profile of 1.43 and 2.15 in optical units and about 5 and 3 times smaller than the wavelength. We have shown that one is able to perform imaging of subresolution sparse objects with a confocal micro-
466 A. DIASPRO ET AL. Fig. 1. Three-dimensional rendering of a portion of the field of the imaged sparse subresolution objects before (top) and after (bottom) image restoration. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]
CONFOCAL IMAGING OF SUB-RESOLUTION SPARSE OBJECTS 467 Fig. 2. Lateral (top) and axial (bottom) intensity profiles in the focal plane of one of the pointlike objects of Figure 1 from experimental (dotted) and deconvolved experimental (solid line) data. Theoretical (circles) expectations calculated by Huygens2 using scalar diffraction theory have been reported for comparison. [Color figure can be viewed in the online issue, which is available at www. interscience.wiley.com.] scope operating in the fluorescence mode. From the results obtained, one can infer that one is able to achieve object detection of less than one third of a wavelength by fluorescence imaging with far-field microscopy. This applies to a matrix of sparse subresolution objects. Alternative optical designs allow a highresolution level (Pawley, 1995), and once realized at sufficient signal-to-noise ratio their behaviour can be further enhanced by deconvolution, as for the 4Pi-confocal scheme (Schrader et al., 1998). Unfortunately, the previously mentioned schemes suffer from the following drawback: within a reasonable waiting time of 10 years they cannot be considered as practicable techniques for all those users who nowadays make extensive use of the three-dimensional high-resolution capabilities of the confocal microscope. This high level of spatial discrimination is necessary for visualisation of local intracellular dynamics, especially transient phenomena or chemical dynamics within living cells. We think that our results, coupled to a physical spatial confinement of the excitation volume achievable with two-photon excitation modalities (Denk et al., 1990; Diaspro and Robello, 1999), can be of interest in the field of three-dimensional microscopy applications.
468 A. DIASPRO ET AL. ACKNOWLEDGMENTS We acknowledge Cesare Fucilli, Massimo Fazio, and Marco Raimondo for their technical support. REFERENCES Bertero M, Boccacci P. 1998. Introduction to inverse problems in imaging. Philadelphia: IOP Publishing. Betzig E, Trautman JK. 1992. Near-field optics: microscopy, spectroscopy and surface beyond the diffraction limit. Science 257:189 195. Bianco B, Diaspro A. 1989. Analysis of the three dimensional cell imaging obtained with optical microscopy techniques based on defocusing. Cell Biophysics 15:189 200. Brand U, Hester G, Grochmalicki J, Pike R. 1999. Super-resolution in optical storage. J Opt A: Pure Appl Opt 1:1 7. Brankenhoff GJ, van der Voort HTM, van Spronsen EA, Nanninga N. 1989. Three-dimensional imaging in fluorescence by confocal scanning microscopy. J Microsc (OXF) 153:151 159. Castleman KR. 1996. Digital image processing. Englewood Cliffs, NJ: Prentice Hall. Denk W, Strickler JH, Webb WW. 1990. Two-photon laser scanning fluorescence microscopy. Science 248:522 526. Diaspro A, Robello M. 1999. Two-photon excitation of fluorescence in three-dimensional microscopy. Eur J Histochem 43:70 79. Diaspro A, Annunziata, S, Raimondo M, Robello M. 1999. Threedimensional optical behaviour of a confocal microscope with single illumination and detection pinhole through imaging of subresolution beads. Microsc Res Tech 45:130 131. Gustafsson MGL, Agard DA, Sedat JW. 1999. I 5 M: 3D widefield light microscopy with better than 100 nm axial resolution. J Microsc (OXF) 195:10 16. Hänninen PE, Hell SW, Salo J, Soini E, Cremer C. 1995. 2-photon excitation 4pi confocal microscope-enhanced axial resolution microscope for biological research. Appl Phys Lett 66:1698 1700. Heintzmann R, Cremer C. 1998. Laterally modulated excitation microscopy: improvement of resolution by using a diffraction grating. Proc SPIE 3568:185 195. Hell, SW and Stelzer EHK. 1992. Properties of a 4Pi-confocal fluorescence microscope. J Opt Soc Am A 9:2159 2166. Hell SW, Lindek S, Cremer C, Stelzer EHK. 1994. Measurement of the 4Pi-confocal point spread function proves 75 nm resolution. Appl Phys Lett 64:1335 1338. Klar TA, Hell SW. 1999. Sub-diffraction resolution in far-field fluorescence microscopy. Opt Lett 24:954 956. Pawley J. 1995. Handbook of biological confocal microscopy. New York: Plenum Press. Schrader M, Hell SW, van der Voort HTM. 1996. Potential of confocal microscope to resolve in the 50 100 nm range. Appl Phys Lett 69:3644 3646. Schrader M, Hell SW, van der Voort HTM. 1998. Three-dimensional superresolution with a 4Pi-confocal microscope using image restoration. J Appl Phys 84:4033 4042. Sheppard CJR. 1989. General considerations of diffraction theory of 3-D imaging. Eur J Cell Biol 48(Suppl. 25):29 32. Shotton DM. 1989. Confocal scanning optical microscopy and its applications for biological specimens. J Cell Sci 94:175 206 van Kempen GMP, van der Voort HTM, Baumann JGJ. 1996. Comparing maximum likelihood estimation and constrained Thikonov- Miller restoration. IEEE Eng Med Biol 15:76 83. van der Voort HTM, Strasters KC. 1995. Restoration of confocal images for quantitative image analysis. J Microsc (OXF) 178:165 181. Wilson T. 1990. Confocal microscopy. London: Academic Press.