1. Introduction

Fluorescence microscopy in combination with specific labeling of biomolecules has been a standard tool for elucidating molecular mechanisms in biological systems for many decades. The advent of single-molecule detection and imaging [1] opened up the ability to observe the behavior of individual biomolecules without ensemble averaging as long as the emitters are at very low concentration [2,3]. This field was advanced further by the development of far-field super-resolution fluorescence techniques, which allow for imaging beyond the diffraction limit by using e.g. stimulated emission depletion microscopy (STED) [4,5], structured illumination microscopy (SIM) [6,7], or single-molecule active-control techniques, such as (fluorescence) photoactivated localization microscopy ((f)PALM) [8,9] and stochastic optical reconstruction microscopy (STORM) [10] (for reviews, please see e.g. Refs [11–17].). This review will center on the third group of methods which involve single-molecule localization as a key step.

When imaging thick samples, such as mammalian cells, tissue, and polymers and gels, background fluorescence arising from out-of-focus emitters reduces the ability to detect and localize individual molecules precisely. This is especially true when imaging samples in all three dimensions (3D). Even if individual molecules can be localized precisely, high labeling density is also required to extract high-resolution spatial information about the sample, as stipulated by the Nyquist criterion [18]. There is a finite number of fluorophores available in the sample and each fluorophore only emits a finite number of photons before photobleaching. A second potential issue with 3D imaging using conventional wide-field epi-illumination is premature photobleaching of fluorophores in parts of the sample that have not yet been imaged. This will reduce the resolution that can be achieved and in the worst case completely hinder imaging of the entire 3D structure. A third issue with excessive illumination is increased risk of collateral photodamage (e.g. due to formation of reactive oxygen species) when imaging sensitive live samples. These issues can all be mitigated by combining single-molecule super-resolution imaging with light sheet illumination [19], where the sample is optically sectioned by a sheet of light illuminating the image plane of the detection optics. However, light sheet illumination was originally designed for relatively low-magnification imaging of samples much larger than a single mammalian cell. To optimize the use of light sheet illumination together with single-molecule super-resolution imaging in individual mammalian cells, some adjustments of the optical design must be made: thinner light sheets for improved optical sectioning, high numerical aperture (NA) detection objectives for improved collection of the weak emission signal from single molecules, and a way of illuminating the cell at the coverslip without distorting the light sheet.

In this review, we will discuss the advantage of combining single-molecule super-resolution with various approaches for light sheet illumination. We begin by describing the principles of single-molecule localization microscopy in 2D and 3D, followed by a brief history, theory, and development of light sheet approaches. Finally, we will give examples of methods that successfully have married single-molecule super-resolution imaging with light sheet illumination for improved precision in mammalian cells.

2. Single-molecule super-resolution microscopy

The ability to detect single molecules optically has now been available for some decades [3,20]. This advance rests upon using optimal emitters (large absorption and high fluorescence quantum yield), as well as maximizing signal to noise for detection. Maximal signal arises from efficiently detecting emitted photons with high-NA optics (because the collection efficiency scales as NA²) and from the use of high-sensitivity detectors such as EMCCDs or sCMOS array cameras. Minimizing noise is done by using samples of the highest purity to reduce unwanted background and by using detectors with low intrinsic noise. Going further, single-molecule super-resolution microscopy has enabled imaging of sub-diffraction structures by combining two key additional ideas [11–17]. The first is the localization of an isolated fluorescent molecule with precision far better than the width of its corresponding image spot on a detector. When a point source of light such as a single molecule is imaged by a microscope, the diffraction of collected light through the apertures of the optical system results in a point spread function (PSF) with dimensions on the order of half the wavelength of the light (Fig. 1). However, if each photon in the measured PSF is treated as a sample of a probability distribution centered on the true position of the molecule, then an estimator of the center of the PSF can be used to localize the emitter in 2D with much finer precision than the width of the distribution. This is essentially an application of the central limit theorem, because each photon landing on the camera is an estimate of the emitter position.

 

Fig. 1 Fundamentals of localization-based super-resolution microscopy. The noisy image (a) of an isolated emitter can be fitted to a model function such as a Gaussian (b) to estimate the center position. The distribution of center position estimates (c) is much narrower than the point spread function. A diffraction-limited image with all molecules fluorescing simultaneously (d) can be super-resolved by sequentially localizing spatially-isolated molecules (e) to produce a reconstruction (f) which shows fine structural details. Localization precision σ_µx for various signal photons N and background photons per pixel β is plotted in (g), based on Eq. (1) for a 160 nm pixel size and 250 nm diffraction-limited spot size.

Download Full Size | PDF

The second idea that enables single-molecule super-resolution microscopy of static structures is the separation in time of fluorescence emission events from closely-spaced molecules. Since single-molecule localization requires PSFs to be isolated in space so that their peak positions can be precisely estimated, yet sufficient labeling density is needed to usefully sample a sub-diffraction structure, some chemical or physical on-off mechanism can be used to only activate a sparse subset of fluorophores (or to force most of them into dark states) at any given moment in time. If the concentration of emitting molecules is kept sufficiently low, each emitter can be localized individually, and over the course of many camera frames the different emitters in the sample can be activated, localized, and de-activated. The resulting data set is a movie in which single molecules throughout the field of view appear and disappear stochastically in time.

Following acquisition of a single-molecule “blinking” movie, an image analysis workflow must be implemented. In general, this consists of (i) PSF detection, such as via intensity thresholding or pattern matching; (ii) background subtraction using various schemes such as wavelet or temporal quantile filtering; (iii) localization of spots to estimate underlying emitter position, discussed in more detail below; (iv) subsequent filtering of localizations, such as discarding bad fits and combining estimates likely to come from the same molecule. The end result is a list of precise position estimates for molecules separated by sub-diffraction distances, effectively enabling the resolution of sub-diffraction structures by visualization of these localizations in a super-resolution reconstruction. To choose from among the available algorithms for each part of this process, the experimenter must balance considerations of data acquisition time, image analysis time, and the quality of the final reconstruction.

The quality of a super-resolution image is closely tied to the precision with which the single molecules can be localized. The localization precision is governed by various parameters such as the number of collected signal and background photons, the effective pixel size of the images, as well as the choice of position estimator. The simplest position estimator is the centroid, or average photon position, but superior estimators are preferable. Other non-fitting estimators have also been demonstrated such as the virtual window center-of-mass [21], radial symmetry [22], and Fourier domain algorithms [23]. Such methods are computationally inexpensive and do not require models of the PSF or the measurement noise. Alternatively, a simple 2D function such as a Gaussian or Airy function may be used as a model of the PSF, in combination with a fitting criterion such as a least squares or maximum likelihood estimator. The merits of the different approaches to position estimation are reviewed extensively elsewhere [24–27]. For the widely popular case of a 2D Gaussian fit with a constant background offset and an unweighted least squares estimator [28], Mortensen and associates derived an approximate closed-form solution for the localization precision one can expect, given by

In addition to localization precision, the resolution of super-resolution images is governed by other important factors such as the labeling density, blinking dynamics, sample drift, and the shape of the underlying structure being imaged. Various methods have been used to quantify experimental resolution such as imaging well-characterized biological structures or DNA origami-based calibration standards [31] and the more general Fourier ring correlation [32].

2.1 Using PSF engineering to extract axial information

The above discussion describes how a super-resolution image is formed of emitters in the focal plane of the microscope’s imaging objective. However, the standard PSF is not conducive to 3D imaging because it changes very slowly as a function of z (the optical axis), it is symmetric about the focal plane, and its intensity falls of rapidly within a few hundred nanometers of that plane. To perform 3D localization, a number of methods have been devised using multi-plane imaging [33], interferometry [34,35], and PSF engineering (Fig. 2). For a recent comprehensive review, see [13]. Here we will focus on the last of these approaches, which entails modification of the PSF of the microscope such that it efficiently encodes 3D position of the emitter over a chosen axial range.

 

Fig. 2 Experimental demonstrations of engineered point spread functions (PSFs) used for 3D localization microscopy. The arrows (right) represent the applicable axial ranges of the different PSFs, and the range over which the PSFs were imaged. (a) Astigmatic [37]. Scale bar ~0.5 µm. Reprinted from [37]. Reprinted with permission from AAAS. (b) Phase ramp [50]. Reprinted with kind permission from Springer. (c) Double-helix [41]. Scale bar is 2 µm. Reprinted with permission from Ref [41]. (d) Accelerating beam [52]. Scale bar is 1 µm. Reprinted by permission from Macmillan Publishers Ltd: Nature Photonics [52], copyright (2014). (e) Corkscrew [51]. Reprinted with permission from [51]. (f), (g) Tetrapods [54]. Scale bars are 2 µm and 5 µm in (f) and (g), respectively. Reprinted from [54] with permission from the American Chemical Society (http://pubs.acs.org/doi/abs/10.1021%2Facs.nanolett.5b01396). (h) Schematic of the optical design used for PSF engineering when implemented using a reflective element for phase modulation. Figure adapted from Ref [75]. with permission from the Royal Society of Chemistry.

Download Full Size | PDF

In principle, PSF engineering can include modulations of both the amplitude and phase of the PSF, but owing to the photon-limited nature of single-molecule imaging, the losses inherent in amplitude modulation are generally not desirable. Instead, a variety of phase aberration patterns have been devised which alter the PSF shape with minimal loss of signal. One simple implementation involves inserting a weak cylindrical lens between the objective and the tube lens to introduce a slight defocus of the PSF along one axis in the image plane [35–37]. The result is an elliptical astigmatic PSF with eccentricity which changes as a function of the axial position of the emitter, enabling 3D localization over a range of about 1 µm, depending on the power of the cylindrical lens. A more general configuration for PSF engineering involves placing a phase-modulating element in a plane conjugate to the pupil plane of the objective, i.e. the Fourier plane of the microscope [38]. The phase modulation can be imparted by a static, dielectric mask (e.g. lithographically-etched quartz) [39] or by a programmable, dynamic element such as a deformable mirror (DM) [40] or a liquid crystal spatial light modulator (SLM) [41]. These options have various advantages and disadvantages. Whereas dielectric masks offer minimal loss and can produce features with fineness limited only by the fabrication method, programmable masks offer the flexibility to change the phase pattern. DMs have high efficiency but relatively few actuators, which limits the resolution of the available phase patterns and precludes the generation of certain features such as phase vortices. SLMs have a large number of pixels, but are designed to work only for one linear polarization of incident light, effectively halving the collected photons. This drawback can be circumvented via a “pyramid geometry”, in which the polarizations are separated and rotated such that each is incident on the SLM in the correct orientation and imaged separately [42].

The axial position can be encoded into the PSF shape in various ways. One intuitive approach is to change the PSF so that a particular parameter of its image depends on the axial position. A calibration curve can then be produced, using an experimental measurement or theoretical model of the PSF, to relate this parameter to the axial position. For instance, the astigmatic PSF encodes the axial position in a function of the PSF width along the x and y axes, which can be estimated from a fit to a 2D elliptical Gaussian function [36,37,40,43]. The double-helix phase mask [41,44–49] produces a PSF with two lobes which rotate around a center point such that the angle reports on the axial position; the phase ramp [50], corkscrew [51], and self-bending [52] PSFs similarly encode axial position into relative motion of two lobes. These lobes can be fit with models such as 2D Gaussian functions to extract the lobe positions, and axial position can be calculated from the output parameters of these fits. Alternatively, the entire image can be used to encode the axial position, such as in the case of the very long axial range Tetrapod PSF family [53–55], which forms complicated images that vary rapidly with defocus and require a 3D model function of the PSF to be used to precisely extract the axial position from the shape. Such a model can be either numerically calculated [53,56] or interpolated from experimental images [57], and then used in a maximum likelihood or least squares estimation procedure. Another consideration in PSF engineering for 3D localization is the wavelength-dependence of the phase delay imparted by a phase mask, which results in different PSF shapes for different emitter colors. This feature has enabled precise estimation of emitter wavelength simultaneously with position [58–60], as well as the design of multi-color phase masks simultaneously producing a different PSF for 3D localization at each of two desired wavelengths [61].

The CRLB, introduced in Section 2 above, provides a useful means of characterizing engineered PSFs in terms of their information content. The CRLB has served both as a metric for comparing different PSFs [62–64] and as a tool for designing PSFs via optimization [53,65]. Some important conclusions can be drawn from CRLB analysis. In order to precisely encode the emitter position in all three dimensions, the available photons must be spread out on the detector into a useful shape, resulting in a larger PSF than the “clear aperture” PSF used for 2D imaging. This entails a slightly lower localization precision laterally but a substantial improvement axially. In general, the larger the axial range of a PSF, the greater the 3D CRLB. Importantly, the larger size of engineered PSFs on the detector reduces the number of molecules that can be visualized in a single frame before the PSFs overlap, so more frames need to be acquired (and analyzed) for the same number of localizations. To analyze dense fields of view in which PSFs overlap with one another on the detector, multi-emitter localization and compressive sensing algorithms have been employed and extended to 3D by incorporating the depth-dependence of the PSF shape [54,66,67]. The combination of PSF engineering with biplane and multifocus imaging has also been used to enable high-density imaging [68–71]. It is also important to note that the CRLB is a theoretical bound, and does not necessarily reflect the true localization precision that is obtained in experiment. Experimental localization precision will be governed by the choice of estimator [72], uniformity of the background, as well as the presence of optical aberrations that degrade the experimental PSF [56,73,74].

While engineered PSFs are a powerful tool for performing precise 3D localization that enables super-resolution imaging of whole cells, an important challenge that such samples introduce is the substantial increase in background fluorescence from additional molecules at other z positions. In particular, since engineered PSFs require spreading out of photons, sensitivity to background is further increased relative to the 2D imaging case, so careful illumination and detection of single molecules is of paramount importance [72].

2.2 Improving localization precision by increasing the signal-to-background ratio

As indicated in Section 2, the localization precision in single-molecule super-resolution imaging can be improved by two different approaches: (1) increasing the signal by using brighter emitters or (2) reducing background by various imaging designs that suppresses out-of-focus fluorescence. This review will primarily be focused on approach (2), but a brief overview of approach (1) is given below.

Using fluorescent proteins (FPs) is a convenient and specific labeling strategy which has revolutionized live cell imaging [76]. However, FPs are generally dimmer than most synthetic dyes, which reduces the localization precision for single-molecule imaging. FPs also require transfection, and the expression of FPs may affect cellular function. In addition, there is a limited practical number of different FP-labeled proteins in a cell, and photoswitchable and photoconvertible FPs mainly come in only a few colors. Synthetic dyes are generally brighter than FPs, but they can, on the other hand, yield a higher background from unspecific binding of fluorophores. They also often require fixation and permeabilization for labeling of structures inside cells. The labeling efficiency of synthetic dyes can be limited, and this can negatively impact the resolution of the reconstructed image. Large efforts are therefore continuously being made to improve parameters of fluorophores such as brightness, photostability, excitation and emission wavelengths, blinking and activation properties, cell permeability, and labeling specificity [77–81]. The researcher should take care to select the optimal fluorophore and imaging conditions for the specific experiment of interest.

The second approach to improve the localization precision is to reduce background from out-of-focus fluorescence using optical sectioning in the excitation path. This is particularly important in samples such as whole mammalian cells (on the order of 10 µm thick) or in thicker samples such as embryos and tissues (meaning samples from several 10s of µm to several mm thick), where the large axial extent of the sample causes high background when using conventional wide-field epi-fluorescence illumination (Fig. 3).

 

Fig. 3 Comparison between epi-, confocal, total internal reflection fluorescence (TIRF), and light sheet illumination. Epi- and confocal illumination provides no axial confinement of the irradiation, which causes unnecessary background (epi), photobleaching, and photodamage. Confocal illumination improves contrast by blocking out-of-focus light but requires scanning to build up an image. TIRF provides excellent optical sectioning, but is limited to imaging within a few hundred nm from the coverslip. Light sheet provides an elegant means to illuminate only the sample plane that is imaged, which reduces both background, photobleaching, and photodamage.

Download Full Size | PDF

An obvious alternative to wide-field epi-illumination is confocal point-scanning microscopy, where a focused laser spot is scanned across the sample to build up an image pixel by pixel. Confocal imaging is a standard tool in many laboratories which provides improved contrast by rejecting out-of-focus light using a pinhole aperture. However, even though light is only detected from the image plane, the excitation light is not axially confined and illuminates the entire thickness of the sample. This causes premature photobleaching and photodamage in regions of the sample outside of the imaging volume. This method is also inherently slower than wide-field approaches due to the point-scanning implementation. Axial confinement of the excitation can be achieved using multi-photon microscopy, where the nonlinear multi-photon excitation only excites a very small volume around the focus. However, this technique is still limited by the inherently slow raster imaging. The speed of the acquisition can be improved using spinning disc confocal microscopy, where the confocal microscopy concept is parallelized by employing an array of pinholes on a rotating disc. This concept has recently been paired with single-molecule super-resolution imaging [82,83]. However, this technique is still limited by the risk of premature photobleaching of labeled molecules outside of the detection volume, just as in conventional confocal imaging, which ultimately will reduce the number of detectable molecules. One way to mitigate the issue with premature photobleaching was recently demonstrated [84] by combining spinning disc confocal microscopy with DNA points accumulation for imaging in nanoscale topography (DNA-PAINT) [85–87]. Since this method relies on transient binding of a large pool of dye-labeled DNA strands to target structures, it relaxes the requirement of low photobleaching outside of the measurement volume. However, this method is not compatible with conventional cell labeling using e.g. FPs or direct immunolabeling using fluorophore-conjugated antibodies, which might prevent implementation in some laboratories. Also, in spinning disc confocal methods, both the excitation intensity and the detection efficiency can be limited by the disc, which reduces the precision with which single molecules can be localized.

Total internal reflection fluorescence (TIRF) microscopy [88] is a widely adapted technique for single-molecule super-resolution imaging [89,90]. Here the sample is excited by the evanescent wave that penetrates into the sample chamber following total internal reflection of the excitation beam at the coverslip. This provides excellent background reduction due to its thin optical sectioning. However, this method is limited to a probing depth of a few hundred nm from the coverslip, and does not allow for imaging of structures deep within a cell or in 3D over an extended axial range.

A particularly useful option is to use light sheet microscopy, which is the focus of the rest of this review. Here, the sample is excited by a thin sheet of light illuminating the image plane of the detection optics. This allows for imaging of the entire transverse image plane simultaneously while providing intrinsic optical sectioning where only the current image plane is illuminated. This optical sectioning provides improved contrast by background reduction and reduced photobleaching of fluorophores before they are actually imaged. In addition, this sectioning reduces photodamage, which is a crucial consideration when imaging live cell samples. The development and implementation of light sheet illumination has paved the way for numerous studies within embryogenesis and developmental biology in a large variety of species [19,91–100].

3. Light sheet illumination for reduced background in thick samples

Light sheet microscopy was invented over a century ago, and numerous different designs have been developed since then to achieve 3D imaging and improved contrast by fast optical sectioning of thick samples (for reviews, please see e.g. Refs [101–103]). The first demonstration of light sheet microscopy was performed already in 1902 by Siedentopf and Zsigmondy, with a method called “ultramicroscopy” [104]. In this early design, orthogonal illumination was used for darkfield imaging of gold nanoparticles in glasses. Light sheet illumination was implemented for fluorescence imaging of biological samples in 1993 by Voie et al., who created a light sheet using a cylindrical lens [105]. This method, termed orthogonal-plane fluorescence optical sectioning (OPFOS), was used to observe the 3D structure of cochlear features. Later, a similar light sheet design termed thin light-sheet microscope (TLSM) was used to study microbes in sea water in a large field-of-view (FOV) [106]. In 2004, an updated design termed selective plane illumination microscopy (SPIM) was developed that allowed for non-invasive imaging of live embryos where the sample could be rotated for sequential acquisition of multiple views [19]. Since then, light sheet technologies have advanced rapidly and have been the gold standard for 3D and 4D imaging of developmental processes and live species behavior.

Using a cylindrical lens to form the light sheet as in SPIM, the cross-section of the light sheet will have a Gaussian beam profile both axially and laterally. To improve the lateral uniformity of the light sheet intensity distribution, a method termed digitally scanned laser light-sheet fluorescence microscopy (DSLM) was developed where a virtual light sheet is generated by scanning of a focused Gaussian laser beam [107].

A common factor for the conventional SPIM and DSLM techniques is that they utilize a Gaussian beam to form the light sheet. The thickness of the light sheet, $\omega$, along the direction of beam propagation can then be described according to:

3.1 Improving the light sheet characteristics for bulk imaging

Many different variants of light sheet designs based on either SPIM or DSLM have been developed to address specific shortcomings for various biological applications where single-molecule sensitivity is not required (bulk). These developments have been driven by a need for e.g. improved image quality, improved axial resolution over a larger FOV, and faster acquisition rates. The user has to choose carefully to balance these features against the need for low photobleaching, low photodamage, and minimal complexity of the optical setup.

This section provides a historical overview of the improvements made in recent decades to the original SPIM and DSLM designs to enhance image quality, axial resolution, FOV size, and acquisition rates for cases where single-molecule sensitivity is not required in thick samples such as multi-cellular organisms, embryos, tissues, and single cells. Some of these improvements have been used when adapting designs for single-molecule imaging in single cells, discussed in Section 4; others could potentially be implemented in future single-molecule designs.

In the first versions of SPIM, large samples were illuminated from a single side. This caused scattering of the excitation light as it passed through the sample, leading to stripes and shadow artifacts that degraded the final image quality. This issue was mitigated by the extension from the basic SPIM setup to multi-directional SPIM (mSPIM) [108] and dual-sided ultramicroscopy [109] based on light sheets entering the sample from different directions. Image quality has also been improved using multi-view reconstructions [110] and image deconvolution [111].

The axial resolution was improved by the implementation of a large aperture lens to create a thinner light sheet, resulting in high-resolution (HR) OPFOS [112] and high-resolution SPIM [113]. To extend the range over which the sample could be imaged with a thin light sheet, the investigators scanned the sample laterally and only used the image of the sample when illuminated with the thinnest part of the light sheet. A similar approach termed thin-sheet laser imaging microscope (TSLIM) using bi-directional illumination was later reported [114] and improved with respect to scanning time and photobleaching rates [115]. The sectioning capability and image contrast of scanning methods have also been improved via a number of background reduction techniques such as confocal line acquisition that is synchronized with the beam scanning [116,117]. Instead of scanning the sample, Dean et al. implemented extended focusing with a tunable lens to sweep the beam waist of a scanned light sheet in the direction of beam propagation and combined this with confocal line acquisition to achieve good axial resolution over an extended FOV [118]. A similar approach was used by Zong et al. who instead used two-photon excitation to reduce background [119], and by Gao who used a spatial light modulator to defocus the beam to achieve tiling of the light sheet along the direction of beam propagation [120]. Scanning of the laser beam both axially and along the beam propagation direction reduces the dwell time and therefore requires higher peak intensities. To mitigate this issue, Dean et al. developed axially swept light sheet microscopy (ASLM) in which a thin light sheet is created by a cylindrical lens and then swept along the propagation direction using remote focusing [121]. This method achieves good axial resolution throughout a large FOV and background is reduced through confocal line acquisition. A similar tiling approach with remote focusing has also been demonstrated [122]. In many of these tiling approaches, regions outside of the detection volume are still illuminated even if the detected background is suppressed. This might cause additional photobleaching and photodamage of the sample even if good contrast is achieved. Another drawback with these scanning methods is the inherently slower acquisition rate, which can limit the applications of such approaches. A novel approach to parallelize the acquisition in DSLM and somewhat mitigate this problem was recently presented by Dean et al., who, by staggering two (or more in principle) beams laterally and simultaneously scanning them axially, imaged several image planes with minimum cross-talk [123].

Wu et al. developed a different design called inverted SPIM (iSPIM) to achieve fast imaging with high axial resolution, although in a smaller FOV [96]. This design tilts both the illumination and detection objectives by 45° with respect to the sample, and the complete light sheet module (commercially available, ASI, Inc., Eugene, OR) can be added to any conventional inverted microscope. This design has been used extensively, and it has also provided the foundation for more recent designs, as discussed in e.g. Section 3.2, as well as for some single-molecule approaches, as discussed in Section 4. This dual-objective design uses one objective for excitation and one for detection, and both roles can be reversed if a second imaging camera is available. While a clever approach, this method has a limited NA due to the close proximity of the two objectives, and the objectives may need to be dipped into the buffer solution above the sample.

To capture fast events in a live 3D sample, large efforts have been made to increase axial scanning and image acquisition rates. In a design termed objective coupled planar illumination (OCPI) microscopy, the alignment of the light sheet and the image plane was simplified by rigidly coupling a thin light sheet with the detection objective [124]. This allowed for faster axial scanning through the sample and was used to image neuronal activity in mice. To improve the acquisition rate, several different designs have also been developed based on the original multi-view SPIM design by Huisken et al. in 2007 [108]. Multi-view light sheets based on SPIM approaches include multi-view SPIM (MuVi-SPIM) [125], four-lens SPIM [126], and dual-view iSPIM [127,128]. Multi-view approaches based on DSLM have also been demonstrated, such as DSLM with dual illumination objectives [129], simultaneous multi-view (SiMView) imaging [130], isotropic multi-view (IsoView) light-sheet microscopy [131], multi-view light sheet with electronic confocal slit detection [132], and high-speed SiMView (hs-SiMView) [133]. In a recent approach, the spatiotemporal resolution and collection efficiency of (d)iSPIM was improved by using reflective coverslips [134]. In hs-SiMView, the sample is kept stationary while the light sheet is translated using galvanometer scanners and the detection objective is moved rapidly using piezo positioners to co-align the image plane with the light sheet position. In a similar approach called electrically tunable lens (ETL) SPIM, the translation of the detection objective is replaced by remote focusing by a tunable lens, eliminating the need for physical translation of the objective [135]. An approach complementary to translating the image plane is to extend the depth-of-field (DOF). This was demonstrated by Olarte et al., who introduced a cubic phase in the back focal plane of the detection objective using a deformable mirror [136]. This method accommodates imaging with the light sheet at any position throughout the axial range while keeping the sample stationary. A similar design was implemented using a transmissive phase mask instead of a deformable mirror for imaging of neural activity in larval zebrafish [137]. Another volumetric imaging technique which also eliminates the need for physical scanning is spherical-aberration-assisted extended depth-of-field (SPED) light sheet microscopy [138], which utilizes an extended DOF in combination with the optical sectioning of a light sheet. Another recent method combines acousto-optical scanning of the light sheet with an acoustic-optofluidic lens for detection over an extended DOF to achieve inertia-free imaging with high temporal resolution [139].

The thickness of the light sheet normally determines the axial resolution that can be achieved, and for SM imaging the thickness is a crucial parameter to consider for improving the localization precision through background reduction. Tiling of the light sheet can extend the FOV that can be imaged with a thin light sheet, but here the sample is still excessively illuminated, which might cause premature photobleaching and photodamage. Several strategies have therefore been developed to make the light sheet thinner and to break the tradeoff between thickness and useful FOV inherent in Gaussian beam light sheets. These strategies are discussed in Section 3.2.

3.2 Strategies for improving the optical sectioning through beam shaping

While Gaussian beams focused along one axis are straightforward to implement for light sheet illumination simply using a cylindrical lens, alternative beam shaping strategies have been used to gain certain advantages over this basic scheme at the cost of some complexity in the optical design (Fig. 4). In place of the cylindrical lens, a scanned Gaussian beam [140] or a Powell lens [141,142] can be used to improve the uniformity of the light sheet. Similar results have also been achieved using a binary-pupil phase mask [143,144]. Bessel beams have the useful property of staying focused over a long distance along the propagation direction, so they have been utilized to create a 1D “pencil” of illumination which can be scanned to create a thin plane of excitation light. Bessel beams also have a self-healing property [145–147] which is valuable when propagating through scattering media [148], but the beams have prominent side lobes which excite fluorophores above and below the intended light sheet position. To reduce the effect of the side lobes of the Bessel beam, multiphoton excitation [140,149,150] has been demonstrated [151,152], as well as lattice light sheet illumination [153–155], in which a lattice of Bessel beams is formed coherently so that the side lobes will be suppressed in the image plane and the lattice can be dithered to create a uniform sheet. Another alternative beam shaping scheme to mitigate the effect of side lobes is illumination with an Airy beam, combined with deconvolution, to provide an extended depth-of-focus and similar self-healing properties [156–158]. However, this technique has not yet been used in the single-molecule imaging regime.

 

Fig. 4 Simulated beam profiles using the sample-oriented axis convention with x the propagation direction of the beam and z the optical axis of the collection objective. Gaussian yz (a) and xz (b), Bessel yz (c) and xz (d), and Airy yz (e) and xz (f) are shown. The yz profiles of four beams are compared: Gaussian beam formed by cylindrical lens (g), scanned Gaussian beam (h), scanned single-photon Bessel beam (i), and scanned two-photon Bessel beam (j). All scale bars are 10 µm. Each image is scaled to span the color map on the right.

Download Full Size | PDF

Structured illumination with both Gaussian [94,118,159–164] and Bessel beams [152,152,154,165–167] has been employed to improve resolution, and the combination of light sheet illumination with stimulated emission depletion [168–170] and RESOLFT microscopy [171] has been applied to improve targeted readout-based super-resolution imaging techniques.

4. Light sheet illumination for single-molecule super-resolution imaging

The sensitivity and spatiotemporal resolution of light sheet microscopy have improved drastically in recent years with the developments discussed above. However, when aiming to image single molecules with high precision in single mammalian cells, there are a few additional aspects to consider compared to conventional light sheet imaging with single-cell resolution in large samples.

Since the localization precision of single molecules depends on both the detected signal photons and on the background, one would optimally want (i) high collection efficiency of the weak emission signal from the single molecules, and (ii) a thin light sheet for reduction of out-of-focus background emission that produces haze. Single-molecule experiments generally use high-NA collection optics to maximize the detected signal. Reducing the background by using a thinner light sheet will improve the ability to localize a single molecule as described in Eq. (1). One issue with satisfying both (i) and (ii) is that high-NA lenses have a limited working distance and a large physical profile, which creates a steric constraint for concurrent use in conventional light sheet designs, such as SPIM and DSLM. Imaging with a high-NA detection objective also requires the sample to be positioned very close to the objective, preferably at the coverslip. An additional aspect to consider is therefore (iii) introduction of the light sheet into the sample chamber in a way that allows for illumination of entire adherent cells at the coverslip without inducing distortions of the light sheet at the chamber-coverslip interface. Several different designs have been developed to address these challenges and optimize imaging of single molecules with high precision in mammalian cells.

An early demonstration of single-molecule imaging using light sheet illumination was provided by Ritter et al., who tracked single molecules in the nucleus of the large salivary gland cells of C. tentans larvae [172]. Their light sheet was formed using a cylindrical lens and a long working distance illumination objective, and directed through the glass wall of the sample chamber before exciting the sample positioned on the bottom coverslip of the chamber. However, since they used this traditional SPIM design, where the light sheet had a Gaussian beam profile which diverged away from focus, they could not image all the way down to the coverslip without distorting the light sheet at the interface at the bottom surface when introducing the beam into the glass chamber. This prevented imaging within tens of µm from the coverslip and they were limited to imaging 100-200 µm from the coverslip using a 40x NA 1.2 detection objective. Their implementation thus prevents imaging of most adherent mammalian cells, which are on the order of 10 µm thick, using higher-NA detection objectives with shorter working distances.

More recently, Strnad et al. developed a design that allowed for imaging of pre-implantation embryos all the way from zygote to blastocyst with high axial resolution [173]. Their design consisted of a custom-made inverted DSLM microscope with a lower-NA illumination objective and a 100x NA 1.1 detection objective mounted in close proximity at 90° relative to each other. The sample was mounted in a custom-made sample chamber with a rounded bottom that allowed it to be positioned in between the two objectives. Although this study did not demonstrate any single-molecule imaging, their design could in principle be used for single-molecule super-resolution imaging in adherent mammalian cells. However, steric hindrance between the objectives limits the NA that can be used.

In the following sections, we will describe and discuss designs that can combine light sheet illumination with single-molecule super-resolution imaging with high precision in mammalian cells (Fig. 5).

 

Fig. 5 Schematics of different designs that can be used for light sheet single-molecule super-resolution imaging with improved contrast. HILO: highly inclined and laminated optical sheet [174]; IML-SPIM: individual molecule localization with selective-plane illumination microscopy [175]; RLSM: reflected light sheet microscopy [176]; LSBM: light-sheet Bayesian microscopy [177]; LLS: lattice light-sheet [154]; soSPIM: single-objective SPIM [178]; SO-LSM: single-objective light-sheet microscopy [141]; TILT3D: tilted light sheet microscopy with 3D PSFs [49]. Schematics are not to scale.

Download Full Size | PDF

4.1 Single objective lens oblique light sheet illumination and detection

In conventional SPIM and DSLM techniques, steric constraints between the illumination and detection objectives prevent the use of very high NA detection objectives and imaging of single cells close to the coverslip. This issue was circumvented in 2008 by Tokunaga et al., who used a single high-NA objective both for oblique light sheet illumination and for fluorescence detection [174]. In this implementation of pseudo-TIRF illumination, called highly inclined and laminated optical sheet (HILO) microscopy, a laser beam with an incident angle at the coverslip just below the critical angle for TIR intersects the image plane at a shallow angle. This oblique illumination roughly optically sections the sample in the center of the FOV, independent of translation of the image plane. This property is advantageous when imaging 3D samples with a small FOV, especially since the angle of the light sheet limits the useful FOV for conventional imaging. The thickness of the light sheet,$dz$, can be tuned by adjusting the incident angle, $\theta$, roughly according to $dz=R/\tan (\theta )$, where $R$is the extent of the illuminated area in the image plane.

The intensity of the light sheet varies with the incident angle and increases with decreasing sheet thickness. As the incident angle converges towards the critical angle and TIRF illumination, the intensity of the excitation light is enhanced four-fold as compared to the incident light [179]. This is caused by constructive interference between the incident and reflected light, which doubles the electric field and quadruples the intensity. This effect can be useful for single-molecule super-resolution imaging in that higher intensities can be achieved with the same power from the laser. However, varying the intensity of the light can change the photophysics of the imaged fluorophores, which is a relevant property for some single-molecule applications. The user should thus be aware of and account for this effect, in addition to the decay of the evanescent field away from the coverslip, when imaging close to TIRF conditions.

HILO microscopy typically yields a light sheet thickness of several microns, and was shown to improve the signal-to-background ratio by a factor of 7.6 for imaging of GFP-tagged proteins using a light sheet thickness of 7 µm and a FOV of 20 µm. Here the divergence of the beam was reduced by inserting a field stop in a plane conjugate to the sample plane.

A very similar approach termed variable-angle epifluorescence microscopy (VAEM) using the same concept was independently demonstrated by Konopka et al. [180], but in this approach a field stop was not implemented. In this initial study, VAEM was used to study GFP-tagged proteins in the plant cell cortex.

HILO/VAEM is a convenient approach to achieve optical sectioning using a standard epi-fluorescence microscope with a high-NA objective and using conventional sample preparation and mounting. However, potential drawbacks are that the parameters of the light sheet are coupled, that the light sheet typically is several µm thick, and that the illumination is non-flat with regard to the image plane. This limits the FOV for conventional imaging.

To solve the limitation arising from the non-planarity between the illumination and image planes, Dunsby developed an optical method to rotate the image plane using an additional pair of objective lenses in the detection path [181]. This increases the useful FOV that can be imaged. However, this adds complexity to the optical setup and the additional lenses reduce the detected signal. The geometry also limits the detection NA to 0.98 in an optimal case and scanning of one of the remote lenses is required to jointly move the illumination and image planes. This method was recently adapted using a stage-scanning approach to allow for 3D imaging using commercially available 96-well plates [182].

A more recent single-objective oblique light sheet technique that alleviates the need for physical translation is swept confocally-aligned planar excitation (SCAPE) microscopy [183]. Here scanning of the light sheet and de-scanning of the detection plane is achieved using an oscillating polygonal scanning mirror. This design allows for high-speed imaging of 3D samples over a large FOV. However, it was demonstrated using a lower-NA detection objective, which limits its applications for single-molecule imaging.

By adding a second objective lens perpendicular to the light sheet plane of a HILO setup, Theer et al. solved the problem of angle mis-match between the oblique HILO light sheet and the detection plane with a method called πSPIM [184]. Light sheet generation and conventional epi-fluorescence imaging were performed using a high-NA objective and an NA 1.1 water-immersion objective was used for light detection. This allowed for imaging of adherent cells all the way down to the coverslip using conventional sample mounting. The performance of the setup was demonstrated by imaging of endocytosis in yeast. Single-molecule imaging should in principle be possible, although the detected signal will be limited by the NA 1.1 detection objective.

4.2 Individual molecule localization with selective plane illumination microscopy

The first design utilizing SPIM for single-molecule super-resolution imaging was developed by Zanacchi et al. in 2011 and termed individual molecule localization with selective plane illumination (IML-SPIM) [175]. The authors used a conventional SPIM setup to create a light sheet either 4 µm or 1.8 µm thick and detected fluorescence from up to 150 µm thick samples using either an NA 0.8 or an NA 1.1 detection objective. Using PALM with astigmatism for 3D single-molecule imaging, the authors reported lateral and axial localization precision of 35 nm and 65-140 nm, respectively, depending on sample and setup characteristics. To achieve activation of the fluorophores, a 405 nm light sheet was co-aligned with the excitation light sheet.

In contrast to HILO, here the light sheet plane can be positioned anywhere in the sample and the light sheet plane is parallel to the image plane, allowing a larger FOV for imaging with a short axial range. However, the sample was mounted in agarose gel in a glass capillary. This mounting method is not compatible with imaging of single, adherent cells. The relatively low NA of the detection objective also limits the achievable localization precision for single molecules. Regardless of these drawbacks, this design demonstrates that 3D single-molecule super-resolution imaging can be achieved using a relatively simple optical system. More recently, this design was coupled with two-photon photoactivation for improved performance and penetration depth [185].

4.3 Reflected light-sheet microscopy

To circumvent the limitations of standard SPIM imaging for single-molecule imaging in adherent cells [172,175], Gebhardt et al. developed a method called reflected light sheet microscopy (RLSM) [176]. In this design, both the illumination and detection objectives are mounted vertically. The excitation and activation light sheets are co-aligned, formed by a cylindrical lens, focused by a high-NA illumination objective, and reflected at 90° into the sample using a polished atomic force microscopy (AFM) cantilever as a mirror positioned directly adjacent to the sample. This allows for excellent optical sectioning with light sheets with tunable thicknesses down to 0.5 µm, although with the corresponding reduction of the FOV characteristic for Gaussian beams. This method enables single-molecule imaging of adherent cells using conventional sample preparation and mounting and uses an NA 1.4 detection objective. The de-coupling of the illumination and detection NAs facilitates concurrent use of a thin light sheet for background suppression and a high-NA detection objective for improved detection of signal photons, yielding superior localization precision for single molecules. This was demonstrated by tracking of single FPs in mammalian cells and the results showed improved performance as compared to HILO. An additional benefit is that the design can be implemented on most conventional epi-fluorescence microscopes.

A similar design was recently demonstrated where the AFM cantilever was replaced by micro-prisms mounted on the coverslip for reflection of the light sheet [186]. However, in either design, an axial slice of about 2 µm closest to the coverslip is inaccessible to the light sheet. This prevents imaging of the lower part of adherent cells. In addition, the use of an AFM cantilever or custom-mounted micro-prisms adds to the complexity of the setup. These factors might be limiting for some applications, but RLSM provides means for single-molecule imaging with improved precision. The extension to 3D single-molecule imaging should also be straightforward by combining RLSM with PSF engineering (see also Section 4.8).

4.4 Light-sheet Bayesian microscopy

In a design called light-sheet Bayesian microscopy (LSBM), Hu et al. formed a light sheet using a cylindrical lens, focused it using a low-NA illumination objective, and redirected it horizontally to coincide with the image plane in the sample using a Pellin-Broca prism [177]. The sample was mounted on top of the prism at an angle and fluorescence was detected using an NA 1.0 detection objective positioned vertically above the sample. The design resembles those of HILO [174] and πSPIM [184], but here the sample is angled instead of the illumination and image planes. The overlap of the light sheet and the image plane allows for imaging of a large FOV. In this study, the authors used a light sheet with a measured thickness of 1.8 µm FWHM, which provides good sectioning while still covering a FOV large enough to image ~10 µm-sized human embryonic stem cells (hESCs). A 405 nm light sheet was co-aligned with the excitation light sheet for (re-)activation of fluorophores. The performance of the setup was demonstrated for single-molecule super-resolution imaging throughout entire hESCs. Since the light sheet has a thickness on the order of microns, the large sample volume in mammalian cells, as compared with thinner samples such as bacterial cells and/or when using TIRF illumination, can lead to substantial overlapping of emitter PSFs. To address this, localization was performed both using a conventional PALM/STORM algorithm and using Bayesian bleach-and-blink (3B) image reconstruction [187,188] to better localize molecules in dense regions. This resulted in a reported lateral resolution of 50-60 nm. The setup has also been used for dSTORM imaging of the spatial organization of T-cell receptors on T-cells with a lateral localization precision of 14 nm [189].

Although the sample can be imaged directly in conventional glass-bottomed culture dishes, the addition of the prism pathway and the custom-made sample holder adds complexity and prevents direct addition to a conventional microscope. However, the use of a prism might be less prone to vibrations and might thus provide improved stability of the light sheet for long-term imaging, as compared to e.g. the AFM cantilever approach used in RLSM.

4.5 Lattice light-sheet microscopy

To improve the optical sectioning capability of existing light sheet implementations, Chen et al. developed lattice light-sheet (LLS) microscopy [154]. As mentioned in Section 3.2, scanned Bessel beam light sheets suffer from increased background arising from the side lobes of the beam [166]. The LLS addresses this by spacing a linear array of Bessel beams with the correct periodicity, such that the beams can be made to interfere constructively in the main lobes and destructively elsewhere. This greatly reduces the background contribution of the side lobes of conventional Bessel beams. This 2D optical lattice was implemented using a binary phase pattern on an SLM, an annular mask, and an illumination objective that focuses the lattice in the image plane. This design can be used in two configurations: either SR-SIM, or by dithering of the lattice to create a very thin, uniform light sheet that can be combined with single-molecule super-resolution imaging. In addition to reducing the effect of side lobes, dithering of multiple beams allowed for imaging with much lower peak intensity delivered to the sample relative to that of a single scanned Bessel beam.

The design has successfully been used with several different lattices and sectioning as thin as ~400 nm, which is thinner than the DOF of a high-NA detection objective. The fluorescence light was detected using an NA 1.1 detection objective, which was mounted orthogonal to a custom-designed NA 0.65 illumination objective in an iSPIM-like configuration. Steric hindrance between the objectives and between the detection objective and the coverslip prevents the use of commercially available higher-NA detection objectives.

Tracking and super-resolution imaging of single molecules was demonstrated using the dithered LLS configuration to create a ~1.0 µm thick (FWHM) light sheet over a 50-µm FOV. Using astigmatism to encode axial position, the entire nuclear lamina was reconstructed with reported lateral and axial localization precision of ~9 nm and 45 nm, respectively. The setup has also been used e.g. for 3D PAINT imaging of ER structures [190], and for multi-color PALM and PAINT imaging of DNA, the nuclear lamina, and intracellular membranes [155]. A theoretical study has demonstrated that isotropic resolution can be achieved by combining three identical objectives in iSPIM geometry with skewed versions of lattice light sheets. However, this approach is aimed at improving the SR-SIM configuration of LLS and it further adds complexity to the design [191].

LLS provides a powerful tool for single-molecule imaging in samples ranging from single, adherent cells to multicellular spheroids tens of µm across. However, the complexity and cost of the optics and electronics needed for the setup might prevent implementation in many laboratories. To provide a simpler alternative, a design was recently developed that offers multi-color imaging with improved light efficiency without the need for an SLM [192]. However, here the side lobes were not suppressed as in LLS, and the performance for single-molecule imaging has yet to be demonstrated with this technique.

4.6 Single-objective selective-plane illumination microscopy

A recent method that only requires a single objective lens for light sheet generation and fluorescence detection was developed by Galland et al. and termed single-objective selective-plane illumination microscopy (soSPIM) [178]. Here, the light sheet is formed either using a cylindrical lens or using beam scanning. The beam is angled at the back focal plane of the objective lens to create a vertical light sheet offset from the center of the FOV. The light sheet is then redirected by a micro-mirror mounted at 45° in a custom-made sample chamber to overlap with the horizontal image plane, similar to an inverted RLSM design. A beam-steering unit and an electrically tunable lens are used to translate the light sheet axially and in the image plane. Since the illumination and detection optics are coupled, axial scanning requires synchronous translation of the objective lens, translation of the light sheet beam at the mirror, and defocus of the light sheet beam to keep the beam waist in the center of the FOV.

The thickness of the light sheet can conveniently be tuned by the choice of objective lens and the size of the beam at the back aperture, with resulting light sheets as thin as 1.2 µm measured when using an NA 1.3 objective. Thinner light sheets, however, limit the useful FOV due to the Gaussian beam tradeoff.

This method was used for single-molecule super-resolution imaging in 2D and in 3D using astigmatism with resulting lateral and axial localization precision of 22 nm and 47 nm, respectively, at the bottom of the sample chamber. Improved contrast was achieved for samples ranging from single cells to embryos, although the smaller samples showed a larger improvement over conventional SPIM techniques. The optical/electrical design is relatively complex compared to some other light sheet designs and the custom-made sample chamber requires sophisticated fabrication approaches. A simpler fabrication approach was recently demonstrated, although with a resulting light sheet thickness of 2.4 µm [193]. The soSPIM design can be implemented on any conventional epi-fluorescence microscope. It allows for imaging of entire adherent cells using thin light sheets and high-NA objectives, which are beneficial for single-molecule imaging. The design can also easily be extended for 3D imaging with a larger axial range by combination with PSF engineering (see also Section 4.8).

4.7 Single-objective light sheet microscopy in a microfluidic channel

A method similar to soSPIM was developed by Meddens et al. and called single-objective light-sheet microscopy (SO-LSM) [141]. Here, the micro-mirror design was replaced by a microfluidic chip with reflective side walls. The side walls are angled at 45° to redirect the light sheet into the sample, which is positioned inside of the microfluidic channels. The microfluidic chip provides a sealed environment where the imaging conditions can be adjusted rapidly and in an automated fashion. In contrast to the soSPIM design, the reflective surface does not extend below the sample mount surface. For imaging of entire cells, the authors therefore mounted the samples on an agarose gel pad to suspend them above the coverslip. The reflective walls of the chip also obstruct imaging with wide-field epi-fluorescence using these chambers. Apart from these factors, the design has similar benefits and drawbacks to soSPIM. The authors demonstrated 3D single-molecule super-resolution imaging of mitochondria in whole cells using astigmatism to extract axial information. Since the design can be implemented on a conventional epi-fluorescence microscope using a high-NA objective, this technique can also conveniently be combined with PSF engineering with extended axial range (see also Section 4.8).

4.8 3D Tilted light sheet microscopy with 3D point spread functions

Very recently, our group developed a method that combines tilted light sheet microscopy with 3D PSFs (termed TILT3D) [49]. Here, the light sheet is formed by a cylindrical lens, focused by a long working distance, low-NA objective, and reflected by a prism into a glass-walled sample chamber at a tilt relative to the focal plane of the detection objective. This tilt allowed for imaging of adherent cells all the way down to the coverslip using very simple optics. The light sheet could be translated both laterally and axially using a motorized mirror. Importantly, the 3D PSFs work over an axial range exceeding the light sheet thickness, so that all molecules in the tilted illumination region produce useful images on the camera. The axial position of any illuminated molecule is determined by the shape of the PSF itself, not by the precise position of the light sheet. By combining light sheet illumination with detection using long axial range PSFs, there was no need to create an extremely thin light sheet using complex optics or electronics, since the single molecules were detected in the entire axial range illuminated by the tilted light sheet. This design allowed for 3D single-molecule super-resolution imaging of whole cells cultured on conventional coverslips using an NA 1.4 detection objective. The light sheet and PSF engineering modules can easily be added to any conventional epi-fluorescence microscope. A simple two-lens 4f optical configuration with a phase mask is added to the detection channel to produce the engineered PSF.

In the initial demonstration, a 2.1 µm thick light sheet was used together with the double-helix PSF for single-molecule imaging, resulting in lateral and axial localization precisions of 16 nm and 24 nm, respectively. By addition of a second detection channel or use of a deformable mirror to encode and alternate between PSF phase patterns, fiducial beads in the sample could be detected using a longer axial range PSF than what was used for single-molecule imaging. In this study, alternating fiducial detection was demonstrated using a Tetrapod PSF with a 6 µm range and epi-illumination. This range was chosen so that a fiducial bead could be detected within the entire axial thickness of the imaged cells, and there was no need for scanning e.g. down to the coverslip for fiducial imaging. This allowed for both convenient stitching of different z-slices and for live drift correction regardless of the position of the fiducial beads in the sample. By simple translation of the tilted light sheet axial position along with simultaneous translation of the detection objective, the approach was used for imaging of the entire nuclear lamina in mammalian cells.

The optical sectioning of TILT3D would be improved by the use of a thinner light sheet. Nevertheless, this approach improved the signal-to-background ratio in mammalian cells up to five-fold. By matching the axial range of the used PSF to the thickness and tilt of the light sheet over the used FOV, TILT3D provides a very simple and effective platform for 3D single-molecule super-resolution imaging with high precision in mammalian cells.

5. Discussion and future directions

Fluorescence light sheet microscopy and single-molecule super-resolution microscopy were both developed over a decade ago. In recent years, continuous development has enabled merging of the two for imaging of single molecules with improved precision and resolution throughout mammalian cells. The inherent optical sectioning capability of light sheet illumination, with reduced out-of-focus background, photobleaching, and photodamage as a result, is an excellent solution for any wide-field microscopy application, and it is especially beneficial for imaging of single molecules in mammalian cells.

In this review, recent methods that have successfully married single-molecule super-resolution imaging with light sheet illumination for improved precision in single cells have been described and compared. The presentation will hopefully aid users in choosing a design with their specific application in mind. The choice of method should be made considering both the needed optical sectioning ability and the experimental implementation requirements, where some designs allow for thinner light sheets, but at the price of increased complexity and cost.

All of the methods described in Section 4 can also be employed for single-particle and single-molecule tracking (demonstrations in e.g [46,154,172,174,176]). When using the light sheet position as readout of the axial position of one or a few emitters, active feedback can be used to keep the light sheet at the emitter position. If numerous emitters are tracked simultaneously, then the light sheet needs to be axially scanned to track the 3D dynamics of all emitters over time. If instead PSF engineering is used to encode the axial position of emitters, the 3D position of multiple emitters can be tracked throughout the entire light sheet volume without scanning. Axial scanning of the light sheet is then only required if samples thicker than the light sheet are to be studied. If very long axial range PSFs are used, 3D emitter position can be detected throughout the entire thickness of mammalian cells without requiring movement the sample stage or the detection objective. The light sheet then only functions to optically section the cell and select which emitters are excited in a given frame. However, the reader should keep in mind that there is a tradeoff between the thickness of the light sheet (and thus the axial range that can be tracked simultaneously) and the useful sectioning by the light sheet. This is especially important for very long range PSFs, which require good signal-to-background ratio and lower density of emitters.

As a next step to improve the performance of light sheet single-molecule imaging, adaptive optics (AO) can be implemented in both the illumination and the detection paths of the microscope [194–197]. In the illumination path, AO can be used to reduce the effect of optics- (static) and sample-induced (varying) aberrations to improve the light sheet characteristics. In the detection path, AO can be used to correct for optics- and sample-induced aberrations of the PSF, improving its shape and thereby improving the accuracy and precision of localization algorithms. Such AO can be implemented using e.g. DMs or SLMs.

Light sheet single-molecule tracking and super-resolution imaging will also benefit from the ongoing development of new fluorophores with improved brightness, photostability, excitation and/or emission wavelengths, blinking and activation properties, cell permeability, target labeling strategy, labeling specificity, and biocompatibility.

The user has to optimize the setup design choice with respect to the light sheet thickness, physical coverage, acquisition speed, and complexity required for the application of interest. New implementations are continuously being developed to improve not only the precision, the resolution, and the preservation of sensitive samples, but also to make the designs easier to implement to allow more users to take advantage of these powerful methods. This will lead to a better understanding of the nanoscale structures and mechanisms of the complex workings of biological systems.