Competition for the Confocal Microscope?

Transcription

1 Competition for the Confocal Microscope? M. Schropp 1,2, Ch. Seebacher 1,2, A. Deeg 2, A. Dovzhenko 3, Olaf Tietz 3, K. Palme 3, and R. Uhl 1,2 1 Bioimaging Zentrum der LMU München, Grosshadernerstrasse 2-4, D Martinsried, Germany 2 TILL I.D. GmbH, Am Klopferspitz 19, D Martinsried, Germany 3 Institute of Biology II, Molecular Plant Physiology, Schänzlestr. 1, D Freiburg, Germany Keywords: 3-D sectioning, live cell, confocality, speed, cell viability 1. Summary Cells are 3D-objects and their function is only fully understood if they are examined in their original 3D-context. The gold-standard for 3D-cellular-imaging is the confocal microscope, which employs optical means for removing unwanted out-of-focus information. However, such a confocal point scanner is slow due its sequential scanning-mode, and it is not very gentle to cells due to the high intensities needed for obtaining a decent signal-to-noise ratio during short pixel-dwell-times. Spinning-disk systems, on the other hand, are fast and gentle by virtue of their multipoint scanning, but their confocality suffers at the same time. We will demonstrate a third approach, which combines the virtues of the two and avoids their draw-backs. Our novel 3D-Structured Illumination Microscope (SIM) employs 2-D hexagonal illumination patterns and combines optical (linescanning) and mathematical means for suppressing and removing unwanted out-of-focus information. By comparing the three 3D-imaging techniques we can show that in all but the thickest samples SIM equals resolution and confocality of the confocal point-scanner, yet it is at least an order of magnitude faster and more gentle to live cells. Compared with a spinning disk-system, SIM is almost as fast under realistic conditions equally fast but it excels with respect to its resolution (particularly in z) and its sectioning capabilities (confocality, not to be confused with z-resolution). We feel that our XL-SIM-approach may well become the method of choice for the majority of applications which - so far - have been dealt using classical confocal microscopes. 2. Introduction Over the past decades the fluorescence microscope has developed into an indispensable tool for the study of cellular processes in live cells. This was not only due to great progress achieved on the side of instrumentation, but also due to the discovery of fluorescent proteins and the availability of genetic tools to express these proteins anywhere in live cells under observation [1]. In order to detect and distinguish tiny signals from nanometer-scale objects, microscope objectives must gather as much light as possible, and this, in turn, reduces their depth of sharpness, often to such an extent that the sought after features are completely buried in out-of-focus haze. The classical cure for this problem is the confocal microscope [2], which uses optical means for blocking unwanted out-of-focus light. While this gold-standard of 3D microscopy - used in thousands of central imaging facilities - provides images sufficient for most purposes, it suffers from two major drawbacks: (i) its sequential point-scanning takes time, hence image acquisition is slow and fast biological processes cannot be followed in real time, and (ii) the short pixel dwell-time of the process necessitates exceedingly high excitation intensities, which are invariably harmful for live cell material: they not only bleach the respective fluorophores, but by means of secondary photochemistry in the cells they damage the cells under observation to such an extent, that often one can only study the process of deceasing cells. To overcome this, multipoint-scanners have been employed, which illuminate the sample not with a single excitation focus, but with many hundred of them simultaneously [3]. Emission is then recorded with an equal number of conjugated confocal pinholes. This parallel-approach accelerates image acquisition considerably, while, at the same time, light can be collected for much longer time from a given fluorophore, hence the danger of photodamage is greatly reduced. Consequently multipoint-confocals (in most cases spinning- disk systems) have become the gold-standard for fast live-cell imaging, yet for every multipoint-confocal there are still ten point-scanning confocals, which are used for tasks they are not really suited for. So what is the drawback of the multipoint-confocal? It is cross-talk between neighboring illumination spots. In the following we shall describe an alternative approach, which uses a combination of mathematical and optical means to remove unwanted out-of-focus information. It unites the sectioning capabilities of a single-point scanner with the speed of a multipoint-scanner and thus qualifies to become a new gold-standard for 3Dlive cell imaging. In the following we will compare its performance to that of a point-scanning and a spinning-disk confocal and based on these results we shall arrive at the conclusion, that our novel approach of a structured confocal microscope, by combining the virtues of both and avoiding most of their respective drawbacks, constitutes the method of choice for most applications requiring 3D-sectioning for live cells. 531

2 3. Results and Discussion The confocal microscope (a) attributes all light passing through the pinhole to the position, where its focus lies at a given point in time. Light originating in other planes is mostly, but not completely rejected, it also contributes to the signal as long as its direction is such that it can pass the pinhole or appears to have passed it. That is why confocality is always not as good as the theoretical z-resolution of a system and why single point confocals show a residual signal at positions where no signal ought to be. The effect becomes increasingly evident with heavily stained samples, where more fluorophores in other planes contribute to this kind of false signal. In a multipoint confocal (b) the excitation volumes of neighboring foci overlap and form a continuum only a few micrometers above and below the focal plane. This bulk excitation diminishes with z-distance, but at the same time more pinholes are passed by false photons or virtually originate from them. Thus the false-photon-contribution to the real signal is much more prominent than in a single point-scanner. In addition, with multipoint confocals scattering becomes an issue, because it can create an additional class of false photons. As a consequence, z-stacks obtained with a spinning disk confocal shows photons originating from certain object planes long before the focal plane of the objective has reached these planes. In a structured illumination microscope (SIM, c) the sample is also excited with a structured illumination pattern, but no attempt is made to eliminate photons originating from outside the pattern by optical means. Instead these photons are also recorded and used for later mathematical removal. To allow such mathematical tricks, one needs to know not only intensity values when a given object-point is illuminated itself (left panel in fig. 1c), but also when it ought to be dark and only neighboring areas are illuminated (fig. 1c, right panel). Thus SIM requires shifting a pattern and recording several phase-images, from which suitable algorithms can extract the fraction of photons arising from planes outside the focus and can remove them completely. Sectioning achievable with a SIM microscope is not only (significantly) better than with a multipoint confocal, but also noticeable better than that of a point-scanning confocal as we will demonstrate below! a b c grid shi Fig. 1: Schematic drawing of the excitation and emission beam-paths in a single-point confocal system (a), a multipoint scanner (b) and a structured illumination microscope (SIM, c). 532

3 However, the saying that there is no free lunch also holds in microscopy: While optical removal of out-of-focus photons also removes their statistical photon-noise 1, their mathematical removal retains these noisy photons! Hence, the more unwanted photons are removed mathematically, the more the resulting desired signal is polluted with noise originating elsewhere. In essence this means: samples, which are thin, not highly scattering or not heavily stained, i.e. samples with little out-of-focus information, can be trimmed to yield optimal tomographic information. However, the more unwanted information needs to be removed, the more noise spills over to the sought after signal! Rescue comes from combining optical and mathematical means for removing unwanted out-of-focus information. A slit-shaped lightbar is modulated two-dimensionally and scanned uni-directionally over the sample. The resulting slit-image is recorded with a rolling shutter scmos camera, whose electronic slit-width is adjusted so as to reject (optically) all light originating from areas outside the slit-width, and whose position is strictly synchronized with the movement of the slitshaped light-bar. The combination of the slit-scanner with the rolling shutter forms a slit-confocal microscope, which differs from previous slit-confocals [4] in that the slit is modulated by a pattern as needed for SIM! In conventional SIM, a one-dimensional line-grid is used and (phase-) images are taken at 3 different grid positions. In contrast to this, our approach employs a spatially more uniform 2-dimensional, hexagonal grid, which needs to be phase-shifted 7 times before enough information is sampled for the calculation of well-sectioned images. We call this hexagonal SIM-version X-SIM, and the hexagonal Line-confocal version XL-SIM. The fillfactor of a hexagonal pattern is 1/3, 1.5x lower than that of conventional linegrid-sim. This leads to a further reduction of out-of-focus light and this improves the signal-to-noise ratio (S/N). Fig. 2 shows such a structured slit illumination, which, in conjunction with a current scmos camera, can be swept over a sample in 10 ms. Moving the grid between sweeps takes < 1 ms, therefore a complete set of phase-images can be recorded in 76 ms. At first sight this is considerably slower than a spinning disk confocal, which in theory acquires up to 2000 fps, yet in reality the low transfer-efficiency of microoptics-empowered spinning disk systems necessitates exposure times of ms, hence there is no real speed advantage of the spinning disk concept. Fig. 2 modulated slit-shaped excitation pattern With respect to sectioning, however, there is a clear performance advantage of the XL-SIM approach compared to a spinning disk confocal! Fig. 3 shows sections of the widely used FluoCells Prepared Slide # 3, taken with an Andromeda Spinning Disk Confocal from FEI Munich (left) and an XL-SIM system in our lab (right). It is quite obvious that the spinning disk images are by far not as crisp as the XL-SIM images and that the intensity-valleys between intensity peaks never reach zero in places where it should be dark! 1 which increases with the square-root of the number of detected photons in case of Poisson distributed photon noise 533

4 Fig. 3 Three-color image (40x, NA 0.95) of a mouse kidney section, stained with Alexa Fluor 488 WGA, Alexa Fluor 568 Phalloidin, and DAPI) (Molecular Probes slide #3). The left image (courtesy Rainer Daum, FEI Munich) was acquired with a FEI Munich Andromeda spinning disk system, powered by mw lasers and with exposure times of 80 ms each, whereas the right image was taken from the same slide, using 40 mw laser-power and also 80 ms total exposure time. The difference between the two approaches is even more striking when comparing x-z-sections of the same object obtained with the two 3D-imaging concepts (Fig. 4). Clearly the XL-SIM image is much more well defined and better resolved. Fig. 4 x-z-sections of the same object as in fig. 3, left spinning disk confocal, right XL-SIM. Fig. 5 displays a comparison of a sectioned image obtained with a conventional point-scanning confocal (Nikon C1 CLSM) and our XL-SIM approach. The better rejection of out-of-focus haze provided by XL-SIM is not as profound as in the comparison with a spinning disk confocal, yet it is clearly visible. Note that the point-scanning system needed more than 10x longer! 534

5 Fig. 5 COS7 cells, Actin GFP-labelled. Objective Nikon 40x, 1.2W. Left: LSM with 1024 x 1024 pixel scanned in 2 s. Right: Detail of an XL-SIM image of 2048 x 1024 pixel recorded in 210 ms. 4. Summary We have demonstrated that XL-SIM, a method combining optical and mathematical means for sectioning of 3D-objects under the light microscope, provides images as good or better than those of a conventional point-scanning confocal microscope, the current gold standard for 3D sectioning microscopy. With respect to imaging speed, XL-SIM outperforms a classical confocal by a factor of 10x or more, and while we have no quantitative measures as to how much photo-damage is reduced in XL-SIM, we are convinced that it is several orders of magnitude! This conviction stems from the fact that a total exposure-time of 75 ms in XL-SIM means that the average time a given sample-voxel is exposed to the excitation light amounts to 500 µs, whereas in a point-scanning confocal it can t be longer than 1 µs, otherwise image acquisition gets exceedingly slow! And as it is well known, that photo-damage increases with intensity more steeply than in a linear fashion [6], a 500x longer exposure yielding the same signal as a 1 µs exposure in pointscanning should be 500x or more less harmful to cells! When compared to a spinning disk confocal - the current gold standard for fast live cell imaging we conclude that XL-SIM can be as fast as the former, but it provides significantly better sectioning. In principle the spinning disk could be significantly faster, however this speed-advantage can rarely be utilized under realistic measurement conditions, since in the usual micro-lens- (Yokogawa CSU X1, Andor Dragonfly, GE Opera Phenix) or micro-mirror (FEI Andromeda) based designs only < 10% of the laser-power is available for excitation, hence a decent S/N ratio necessitates exposure times comparable to those of a XL-SIM system! Faster frame-rates can be achieved with XL-SIM simply by reducing the field over which the slit is scanned. The only spinning disk system that can reduce field-size and hence exposure-times without sacrificing the S/N-ratio is the Andor Dragonfly, which has a variable telescope in the excitation beam. We conclude that XL-SIM may challenge the two most popular methods for 3D-sectioning under the light microscope! Acknowledgements The support by BMBF (Microsystems FKZ ) is gratefully acknowledged! References [1] [2] [3] [4] R. Tsien Handbook of Biological Confocal Microscopy (Pawley, James, Ed.) ISBN Any Way You Slice It A Comparison of Confocal Microscopy Techniques. James Jonkman & Claire M. Brown, J Biomol. Tech. (2015) 26(2): A line scanning confocal fluorescent microscope using a CMOS rolling shutter as an adjustable aperture. E. MEI, et al., 247,

6 [5] Neil, M. A. A., Juskaitis, R. and Wilson, T. Method of obtaining optical sectioning by using structured light in a conventional microscope. Optics Letters (1997). 22: [6] Lattice Light Sheet Microscopy: Imaging Molecules to Embryos at High Spatiotemporal Resolution, Bi-Chang Chen et al., Science Oct 24; 346 (6208):