Light Sheet Theta Microscopy for High-resolution Quantitative Imaging of Large Biological Systems

Advances in tissue clearing and molecular labelling methods are enabling unprecedented optical access to large intact biological systems. These advances fuel the need for high-speed microscopy approaches to image large samples quantitatively and at high resolution. While Light Sheet Microscopy (LSM), with its high planar imaging speed and low photo-bleaching, can be effective, scaling up to larger imaging volumes has been hindered by the use of orthogonal light-sheet illumination. To address this fundamental limitation, we have developed Light Sheet Theta Microscopy (LSTM), which uniformly illuminates samples from same side as the detection objective, thereby eliminating limits on lateral dimensions without sacrificing imaging resolution, depth and speed. We present detailed characterization of LSTM, and show that this approach achieves rapid high-resolution imaging of large intact samples with superior uniform high-resolution than LSM. LSTM is a significant step in high-resolution quantitative mapping of structure and function of large intact biological systems.


Introduction
The emergence of various tissue clearing and molecular labelling methods over the last decade is enabling unprecedented optical access to the structure and function of intact biological systems Chung et al., 2013;Dodt et al., 2007;Erturk et al., 2012;Hama et al., 2011;Ke et al., 2013;Kuwajima et al., 2013;Murray et al., 2015;Pan et al., 2016;Renier et al., 2014;Romanov et al., 2017;Susaki et al., 2014;Tomer et al., 2014;Yang et al., 2014). Most of these methods employ a cocktail of chemicals for membrane lipid dissolution and/or refractive index smoothening to render the tissue transparent (Richardson and Lichtman, 2015). Together with parallel advances in high-speed microscopy methods, these approaches have already proven to be highly effective in mapping of organs as large as the intact adult mouse brain (Lerner et al., 2015;Migliori et al., 2016;Romanov et al., 2017;Tomer et al., 2014). By providing a highly-detailed 3D view of the architecture of normal and abnormal intact tissues, these methods can accelerate our understanding of the structure and function of brains, a key goal of high profile BRAIN initiative, as well as provide mechanistic insights into the pathophysiology of the microarchitecture of diseased tissues. However, scaling up these approaches, while maintaining uniform high imaging quality, faces the challenges of clearing and labelling of large samples combined with high-resolution quantitative 3D imaging.
Here we address some of these challenges by developing a conceptually distinct microscopy framework: Light Sheet Theta Microscopy (LSTM). Building upon the principles of Light sheet microscopy (LSM) (Siedentopf and Zsigmondy, 1903;Stelzer, 2015), LSTM allows high-speed quantitative imaging of large intact tissues at uniform high resolution. The basic concept of LSM was first introduced more than a century ago (Siedentopf and Zsigmondy, 1903). LSM uses a thin sheet of light for planar illumination of a sample and an orthogonally arranged wide-field detection arm for simultaneously capturing the emitted signal with a high speed CCD or sCMOS camera (Migliori et al., 2016). Compared to other commonly used 3D imaging modalities, confocal and 2-photon microscopy, LSM places the minimum possible energy load on the sample and provides orders of magnitude faster imaging. LSM has been highly successful for experimentations in developmental biology (Huisken et al., 2004;Keller et al., 2008;Preibisch et al., 2010;Reynaud et al., 2015;Wu et al., 2013), cell biology Gao et al., 2012;Planchon et al., 2011), high-resolution whole brain neuroanatomy (Lerner et al., 2015;Tomer et al., 2015;Tomer et al., 2014) and neural activity mapping experiments (Ahrens et al., 2013;Bouchard et al., 2015;Chhetri et al., 2015;Holekamp et al., 2008;Tomer et al., 2015;Vladimirov et al., 2014). The samples larger than the field-of-view (FOV) of a microscope are imaged by sequential acquisition of overlapping image stacks (Tomer et al., 2014), which are then computationally stitched to result in the final image volumes.
The sizes of samples that can be imaged with LSM is restricted along two dimensions, the detection and the illumination axes (Figure 1). Sample illumination is therefore restricted to few millimeters deep and wide, without a limit on length. Although LSM has been used for rapid high-resolution imaging of samples as large as mouse brains (Renier et al., 2014;Susaki et al., 2014;Tomer et al., 2014), image quality progressively reduces towards the center because of illumination light scattering (even with two-sided illumination). The reduction of image quality is further more severe for larger rat brain tissues (Stefaniuk et al., 2016). These fundamental limitations have precluded the use of LSM for high-resolution quantitative imaging of large samples such as rodent brain tissues, and for imaging of a sample with laterally extended geometries such as thick slices of human brain or physically expanded tissues (e.g. using Expansion Microscopy (Chen et al., 2015)). Recently, alternative optical configurations of LSM have been developed that partially address these limitations, including the rotation of the illumination and the detection axes by 45° relative to the sample surface normal as done in OCPI, iSPIM and diSPIM implementations (Holekamp et al., 2008;Strnad et al., 2016;Wu et al., 2011;Wu et al., 2013) and the generation of illumination light sheets through the detection objectives itself as implemented in SCAPE/OPM (Bouchard et al., 2015;Dunsby, 2008). iSPIM/diSPIM approach removes limits on the lateral dimensions of imaging volumes, although at the cost of reduced effective working distance of the detection objective, and therefore is restricted to relatively low-resolution detection for thick samples (Figure 1a). SCAPE/OPM approaches employ rotation optics to image an oblique plane in a sample, illuminated through the detection objective itself. This configuration enables fast volumetric imaging speeds for small volumes, although at the cost of reduction in image quality due to the collection of signal from non-native focal planes (Figure 1a). While these implementations have been highly successful for rapid imaging of small samples (e.g. C. elegans embryos) or small volumes of mouse brain cortex, they suffer from other geometric and image quality constraints when applied to larger samples.
We have developed Light Sheet Theta Microscopy (LSTM) to address some of these limitations. LSTM achieves planar imaging by employing obliquely arranged illumination light sheets from the same side of the sample as the detection objective. This configuration alleviates limitations on the lateral dimensions of the sample, while providing similar imaging depth, uniform highresolution and low photo-bleaching (higher than LSM for smaller samples, but similar to LSM for larger samples). Note that LSTM bear some resemblance to Line Scan Confocal Microscopy (LSCM) (Mei et al., 2012;Wolleschensky et al., 2006) at an abstract level, as effectively an illumination line profile is scanned in the detection focal plane (Figure 1a). In LSCM, the illumination line profile is generated by focusing of a collimated LASER beam through the detection objective itself, which is then rapidly scanned over the focal plane to result in high imaging speed (Wolleschensky et al., 2006). However, this increase in imaging speed comes at the cost of reduction in imaging quality (both axial and lateral resolution) and very high photobleaching (similar to Confocal). Moreover, the usage of a single objective for illumination as well as for signal detection reduces the flexibility needed for using optimal optics for efficient illumination and high resolution imaging (which has been a critical factor in the success of LSM).
These and other limitations make LSCM less suitable for high-resolution imaging of large samples.
Indeed, LSCM has only be successfully used for relatively low resolution imaging of very small samples (Mei et al., 2012;Wolleschensky et al., 2006). The key advantages of LSTM approach over LSCM includes high imaging resolution and low photo-bleaching, both due to the use of thin oblique illumination light sheets, generated using two separate illumination objectives, making LSTM ideally suited for rapid high resolution imaging of very large samples. We present detailed characterization of the LSTM approach using examples that include mouse and rat brain, as well as human brain slices. Furthermore, we demonstrate the unique suitability of LSTM for rapid volumetric imaging of highly motile animals. Through high-speed quantitative imaging of larger samples, LSTM could facilitate mapping of an entire post-mortem human brain (slab-by-slab) in a practical time-frame.

Light Sheet Theta Microscopy (LSTM)
LSTM includes a standard wide-field detection arm and two symmetrically arranged nonorthogonal (q<90°, relative to the sample surface normal) illumination arms for the generation of thin sheets of light that intersect at the detection focal plane (Figure 1,2). This approach results in a thin line illumination profile which is then scanned, in synchrony with the line-by-line rolling shutter detection of an sCMOS camera (virtual slit effect, Tomer et al 2014), to achieve thin optical sectioning (Figure 1). In contrast to LSM, the non-orthogonal optical configuration of LSTM does not place any restrictions on the lateral dimensions of the imaging volume, while still allowing access to the complete working distance of the detection objective, provides high imaging speeds (20 milliseconds per image acquisition, same as COLM implementation (Tomer et al., 2014)) and similarly low photo-bleaching for large samples (Figure 3d (Dean et al., 2015;Fu et al., 2016;Gao, 2015;Planchon et al., 2011;Tomer et al., 2015;Vettenburg et al., 2014)).
The LSTM 1-AS approach provides a simpler implementation, although at the cost of nonuniformity in the planar illumination and low axial resolution (because of the need to use relatively low numerical aperture objectives for illumination). The 2-AS approach allows for uniform planar illumination and detection to enable high-resolution quantitative imaging. To characterize the two LSTM modes and to compare them with LSM, we imaged micron-sized fluorescent beads and CLARITY-cleared (Tomer et al 2014) human brain tissue stained with nuclear marker DAPI  (Supplementary Figure 2). We also developed adapters to mount a prism mirror for the optical alignments. The entire sample chamber assembly can be translated in 3 dimensions to acquire the image volumes. This approach allows for full exploration of various parameters of the system (such as the angular separation between the illumination and detection arms) and acquiring data from large samples by providing rigid monolithic illumination and detection units with translational and rotational degrees of freedom.
The final overall LSTM illumination configuration includes a LASER source, collimators (~10 mm output beam diameter), ETL, cylindrical lens, galvo scanner, scan lens, tube lens and illumination objective (Figure 2a). In addition, we incorporated an iris, after the collimator, to remove the peripheral spread of Gaussian beams, a one dimensional slit, before cylindrical lens, to control the effective numerical aperture of illumination and a second iris at the conjugate plane, between scan lens and tube lens, to control the light sheet height. The detection arm is composed of a detection objective, emission filter, tube lens and an sCMOS camera.
Since LSTM involves scanning of a line illumination-detection profile generated by the intersection of the light sheet and the detection plane, we used static sheets (generated by the use of a cylindrical lens and associated optics), instead of a dynamic sheet (generated by rapid scanning of a pencil beam) to maximize the imaging speed. The cropping of peripheral parts of the large input diameter beam with an iris ensured a relatively uniform intensity distribution profile across the static light sheet. We used a galvo scanner to achieve rapid translation of light sheets perpendicular to their propagation direction. Finally, for the 2-AS mode, we also needed rapid translation of the thinnest part of the sheet along the propagation direction. Possible approaches here include fast piezo motors to translate the illumination objectives, using holographic spatial light modulators or an electrically tunable lens (ETL) able to drive the divergence and convergence of a collimated beam. The use of piezo motors for rapid scanning of objective often results in vibrations and require additional settling time (Ahrens et al., 2013;Tomer et al., 2015), and the spatial light modulators are limited in modulation speed because of slower refresh rates. ETLs, on the other hand, can achieve high frequency modulation of focal point position without the need for moving optics of significant mass (Fahrbach et al., 2013). We thus tested an ETL based approach and found it to be highly effective for achieving uniform simultaneous 2-axes scanning ( Figure   1d, Supplementary Videos 1).
The LSTM assembly was optically aligned by placing a prism mirror (with fine scratches in the center, see Supplementary Figure 2 for mounting arrangements) in the focal plane of detection optics, to visualize the location and cross-section of the light sheet relative to the detection focal plane. The light sheet positioning parameters were optimized such that the thinnest part was in alignment with the center of the field-of-view of the detection plane. Next, the mirror was replaced with a high concentration (>2%) agarose gel containing fluorescent beads (Note, the high concentration of agarose was used to ensure that only the surface plane of the agarose gel was visible during the alignment optimization.). Optimal galvo scanner and ETL parameters for achieving uniform planar illumination across the entire field-of-view were identified by examining the extent and quality of the illuminated beads located on the surface.

LSTM characterization
A series of calculations were used to assess and compare various properties of LSTM (summarized in Figure 3). First, we devised a method to calculate the physical geometric constraints of arranging a given set of detection and illumination objectives in a non-orthogonal configuration (Supplementary Figure 3). The main physical parameters used in the calculations are the working distances and the diameters of both the illumination and detection objectives. We calculated the range of physically-allowable, relative angular arrangements that enable light sheets to intersect the detection focal plane at their thinnest parts, while also ensuring that illumination objectives remain above the physical extent of the detection objective (Supplementary Figure 3). For instance, only angular configuration of 43-62 degrees for Macro 4x/0.28NA/29.5mmWD (Olympus) as the illumination objective and 10x/0.6NA/8mmWD (Olympus) as the detection objective are possible (Figure 3, Supplementary Figure 3). Note that the working distance of this illumination objective is given for use in air, and therefore we calculated the approximate effective working distance as shown in Supplementary Figure 3. Next, determined the influence of angular separation of illumination and detection arms on the resulting image volumes. We first calculated and compared the illumination path length in LSTM to LSM: shorter the illumination path length the better the image quality. In LSM, the illumination light sheet needs to penetrate the entire width of the sample for complete coverage, whereas in LSTM the effective illumination path length depends on the angular arrangement and the tissue thickness(t): t/cos(q). As shown in It is evident that LSTM imparts significantly higher energy load, compared to LSM, for imaging of smaller samples, while similar for larger samples. The LSTM energy load also decreases when using higher angular configuration (between illumination and detection axes) and higher magnification objectives. Nevertheless, we provide strong empirical evidence of no consequences in terms of photo-bleaching (summarized in Figure 3f), by imaging of fixed samples of various sizes and shapes, and also by high-speed long-term live functional imaging of a freely moving animal.
In LSTM, from an illumination path length stand point, minimizing the angular separation will increase the imaging quality. However, when the effect of q on the effective light sheet thickness (approximated as b/sin(q), Figure 3c), which determines the axial resolution, is measured, an inverse relationship is found: the more the q the better the axial resolution. Because illumination is provided via a relatively low NA objective (0.28) for which the light scattering has much smaller effect on the illumination side, we decided to maximize the angular separation (~60°) to achieve higher axial resolution. All the experiments were performed using this configuration.

LSTM enables rapid quantitative imaging of large samples with uniform high-resolution.
We first tested the use of lower NA illumination (hence larger field-of-view and thicker sheet) in LSTM 1-AS configuration in a large sample, a CLARITY-cleared thick coronal section of Thy1-eYFP transgenic mouse brain (Figure 4). While, the LSTM 1-AS mode allowed for high-quality imaging of the section, image quality was reduced (marked with dotted-rectangles) for peripheral most portions of the field-of-view, even for the low NA illumination configuration. This result was similar to the imaging performance of a LSM system employing Gaussian beams for illumination.
By allowing use of high NA objectives for illumination (hence thinner sheet), LSTM 2-AS mode enables high-resolution imaging with uniform quality. To assess the quantitative imaging performance of LSTM, we performed imaging of cleared intact mouse central nervous system of Thy1-eYFP transgenic mouse. As demonstrated in Figure 5a and SupplementaryVideo 3, LSTM enables rapid high-resolution quantitative imaging of these large samples without any reduction in the image quality across the sample dimensions. We further imaged a large (~9.6mm x 13.5 mm x 5.34 mm) coronal slice of CLARITY-cleared Thy1-eYFP transgenic mouse brain, with 10x/0.6NA/8mmWD (Figure 5b, Supplementary Video 4) and 25x/1.0NA/8mm (Figure 5c) objectives, and larger input beam diameter (to employ the full available NA of 0.28) of the illumination objective. Note that this sample was expanded ~1.5-2 fold by incubation in glycerol solution  to result in ~1-5-2 folds expansion. As demonstrated by zoom-in views of various locations of the samples, LSTM provides high uniform quality across the samples.
Next, we demonstrate that LSTM indeed outperforms LSM for uniform high-resolution imaging of large samples (Figure 6). For this we cleared a large slice of rat brain and stained it with a relatively uniform label to visualize all the blood vessels. Previous attempts of using LSM to image rat brain resulted in poor image quality apart from the periphery of the tissue (Stefaniuk et al., 2016), as also expected because of illumination scattering. We challenged the LSTM approach with this very large sample (~2 centimeters wide and >5 mm deep) for a direct comparison with LSM imaging performance of the same sample. Note that we ensured to choose a highly transparent tissue (Figure 6a inset) for a fair comparison. As shown in Figure 6 and Supplementary Video 5, indeed LSTM allowed rapid uniform high-resolution imaging of the entire tissue, whereas LSM resulted in progressively poor image quality towards the middle of the sample, similar to the previous report (Stefaniuk et al., 2016). To complement these observations, we performed rapid high-resolution imaging of a large (3.32 cm x 1.93 cm x 1 mm) uniformly expanded (using Expansion Microscopy approach (Chen et al., 2015)) brain slice (Figure 6c-d and Supplementary Video 6). Expansion Microscopy (ExM) method is enabling unprecedented high super-resolution access to biological tissues, although also presenting unique challenges on imaging and data handling front. For example, imaging of this sample with the state-of-art Confocal or 2-photon microscopy will take several weeks of continuous imaging. We demonstrate unique suitability of LSTM for ExM samples by imaging (using 10x/0.6NA detection objective) the entire expanded tissue in ~22 hours, yielding 723,200 images (2048x2048 pixels). The resulting dataset reveals the finest details of brain neuronal architecture (e.g. dendritic spines), while providing complete coverage. In addition, we also demonstrate imaging of a large piece of human brain tissue labelled with a uniform nuclear label DAPI. Human brain tissue, being dense, scatters the illumination light heavily and thus has been proven to challenging to be imaged by LSM approach. As shown in Supplementary Figure 1, LSTM indeed enabled uniform highresolution imaging of large piece of human brain tissue.
Finally, we demonstrate the unique compatibility of LSTM in capturing the nervous system dynamics of a highly motile animal. Live samples often undergo substantial rearrangements in their body shape and cellular density, which significantly alter their local optical properties.
Although LSM based imaging methods have been effective in capturing the cellular dynamics of developing embryos and neuronal activity of immobilized Zebrafish larvae, LSM remain susceptible to large changes in shape and density of motile sample, mainly because of the use of orthogonal illumination. This limitation has been partly addressed recently by utilizing a sophisticated array of hardware and software components that facilitate real-time adaptation of light-sheet parameters (Royer et al., 2016). LSTM, with its non-orthogonal illumination, may provide a simpler and highly effective solution. We tested this hypothesis by performing rapid volumetric calcium imaging of highly motile Hydra, which has been recently established as an effective model for exploring the role of neuronal network activities on apparent behaviors (Dupre and Yuste, 2017). We found that, indeed, LSTM enables aberration free calcium imaging of freely behaving Hydra which is undergoing drastic changes in its body shape and cellular density in the recording duration (Figure 7a and Supplementary Video 7). In a way, the non-isomorphic body deformation of Hydra represents the worst-case scenario for tracking the activity of every neuron during behavior. We validated LSTM datasets by extracting and comparing neuronal traces with previous observations, finding excellent agreement (Dupre and Yuste, 2017). Note that, for this first demonstration, we used a relatively slow process of step-wise motion of the sample stage to acquire the image stacks. LSTM mechanism is straightforward to combine with piezo motor-based synchronous rapid scanning of the detection objective and also with extended detection depth-offield .
Thus, LSTM allows high-resolution quantitative imaging of large intact biological systems with no limitations on the lateral dimensions and high uniform quality. The sample thickness that can be imaged remains limited by the working distance of the detection objective and also by the level of tissue transparency and the penetration of labelling reagents.

Discussion
Understanding the function of complex biological systems -especially highly complex mammalian brains -requires access to the intricate details of underlying structure and molecular composition, along with the functional dynamics. Over the last decades, a number of methods for clearing tissues to facilitate the interrogation of the structure and molecular architecture of large intact tissues have been developed. Methods such as CLARITY (Chung et al 2013, Tomer et al 2014 can render opaque tissues transparent while at the same time preserving structural and molecular content. Together with advances in high-speed imaging methods, these approaches provide new possibilities for understanding the functioning of tissues in health, and in diseased states such as malignant tumors and neurological damage. Methods that allow acquisition of high-resolution information on a practical time-frame such as LSM (Stelzer 2015, Migliori et al 2016 are effective for imaging these transparent tissues because of their inherent low photo-bleaching and the high imaging speeds. The LSM approach of illuminating a sample with a thin sheet of light, and detecting the emitted signal with an orthogonally arranged detection arm provides two main advantages: minimal energy load and high imaging speed. We developed a highly-optimized implementation of LSM, called COLM Tomer et al., 2014) which allowed high-resolution imaging of entire intact mouse brains in a matter of hours. However, LSM as a general approach has been restricted in the lateral dimension of image volumes because the illumination light sheet enters from the side of the sample, and needs to penetrate the entire sample and the working distance of the illumination objective places a hard-physical limit of the imaging volume. In addition, physical tissue expansion approaches (Chen et al., 2015;Ku et al., 2016) for achieving high imaging resolution are producing even larger samples. To address these challenges, we developed a conceptually distinct imaging framework, called Light Sheet Theta Microscopy (LSTM). Similar to LSM, LSTM is based on planar illumination, but achieves this goal by using non-orthogonally arranged illumination objectives to produce light sheets that intersect the detection plane in a line profile, which is then synchronously (along with line-by-line detection of sCMOS camera) scanned along the detection plane. An immediate advantage of such a configuration is that it alleviates the restrictions on lateral sample dimensions, while providing uniform image quality for achieving true quantitative imaging. Note that LSTM imparts higher energy load (compared to LSM) for smaller samples, but similarly low photo-bleaching for larger samples. For all the imaging volumes reported in this study, no photo-bleaching is observed, further demonstrating the unique suitability of LSTM for imaging of samples of varied sizes and shapes. We found that a strategy of simultaneous 2-axes illumination (i.e. along and perpendicular to light sheet propagation) provided the best imaging performance. In comparison to LSM, LSTM allows imaging of larger samples such as a ~2 cm wide and ~5mm thick rat brain slice -with high uniform resolution across the entire sample.
Moreover, LSTM is uniquely compatible with live imaging of highly motile samples, which undergo non-isomorphic changes in body shape and cellular densities.
LSTM is expected to allow uniform high-resolution imaging of large samples including thick slabs of cleared and labelled post-mortem healthy and diseased human brains as well as imaging of large animal intact brains, including rat and primate brains. Moreover, LSTM may facilitate in situ detection of thousands of transcripts in expanded tissue samples. Future work will include integration of super-resolution approaches (such as structured illumination) and simultaneous multi-view imaging.

Illumination depth calculations.
We used geometric calculations (Figure 3b) to estimate the maximum illumination path lengths of LSTM as / cos q , where t is the sample thickness to be imaged and q is the angle between the illumination propagation direction and the detection axis.
The maximum illumination path length in LSM would be the same as the sample width (w). We calculated the ratio of these illumination path lengths which was converted into a binary representation by thresholding at 1, and plotted as a heat map, shown in Figure 3b.
LSTM effective light sheet thickness calculations. Due to the non-orthogonal incidence of the light sheet on the detection plane, the effective light sheet thickness can be approximated as the projection of the original thickness on to the detection direction, resulting in /sin(q), where b is the original light sheet thickness at the most focused position, and q is the angle of incidence relative the detection axis. The relationship was plotted in the graph shown in Figure 3c.
For whole brain clearing, transcardiac perfusion was performed with 20 ml HM solution, followed by overnight incubation at 4°C. The rat brain was perfused with 4% paraformaldehyde (PFA), post-fixed for 16 hours and then frozen in isopentane for storage. The frozen brain was thawed at room temperature in PBS buffer, sliced and incubated in HM solution overnight at 4°C.
The human brain tissue was incubated in 4% PFA for ~2 days, followed by incubation in HM solution up to days at 4°C. All the perfused tissues were de-gassed and transferred to 37°C for 3-4 hours for hydrogel polymerization. The tissues were cleared by incubating (with shaking) in clearing buffer (4% (wt/vol) SDS, 0.2 M boric acid, pH 8.5) at 37°C until clear (2-3 weeks). The rat brain and human brain slices were incubated in the HM solution overnight at 4°C followed by degassing to replace the oxygen with nitrogen, and incubation at 37℃ for 3-4 hours for polymerization. After washing off any remaining HM solution, the tissues were incubated in buffered clearing solution (4% (wt/vol) SDS, 0.2 M boric acid, pH 8.5) at 37℃ with shaking until the tissue became clear (1-4 weeks). Afterwards, the tissue was washed with 0.2 M boric acid buffer (pH 8.5) with 0.1% Triton X-100 for up to 24 hours. The cleared tissue was labelled with DAPI (1 µg/mL final concentration) and/or blood vessels marker tomato lectin (Vector Labs, FL-1171), by incubating in the labelling solution for 3-4 days. After washing with the buffered solution (0.2 M boric acid buffer, pH 7.5, 0.1% Triton X-100), the tissue was transferred into 85-87% glycerol solution in graded fashion (i.e. 25%, 50%, 65% and finally 87%) for final clearing and imaging. All the image volumes were acquired with a 2 or 5 microns step-size.
For isotropic tissue expansion, a Thy1-YFP mouse brain slice (250 um, perused and fixed with 4% paraformaldehyde and sliced with vibratome) was gelled and digested following the protein retention expansion microscopy (proExM) protocol (Tillberg et al., 2016). The sample was stored in 1x PBS before changing the buffer to 65% Glycerol (with 2.5 mg/mL DABCO) for the LSTM imaging.
For live imaging of Hydra, we followed the procedures described previously (Dupre and Yuste, 2017). Transgenic Hydra expressing GCaMP6s in neurons (Dupre and Yuste, 2017) were maintained in the dark at 18°C and were fed freshly hatched Artemia nauplii once a week, or more frequently when the colony needed to grow. Animals were mounted between two coverslips (VWR cat# 89015-724 and VWR cat#16004-094) using a 100um spacer (Grace Bio-Labs cat#654006) and were imaged with LSTM using 10x/0.6NA objective (Olympus).

Image analyses.
A TeraStitcher (Bria and Iannello, 2012) based pipeline (Tomer et al., 2014) was used for stitching of acquired image stack tiles of all the datasets. Maximum intensity projections, and other linear image contrast adjustments were performed using Fiji (Schindelin et al., 2012;Schneider et al., 2012) and MATLAB. All volume renderings were performed using Amira (FEI).
All the fluorescent beads image analysis was performed using Fiji. To calculate the axial Full Width at Half Maximum (FWHM), x-z projections of beads image stacks were used. For individual beads a line intensity profile was calculated along the central position, followed by manual calculations of full width at half maximum intensity values. For Hydra data analyses, manual tracking of selected neurons was performed using TrackMate (Tinevez et al., 2017), and the generation of graphs and movies was performed using MATLAB.
Illumination energy load calculations. The procedure is summarized in Supplementary Figure   5. To calculate the total illumination energy load in LSTM, we performed a simulation of the stepwise scanning of sample through the illuminating light-sheet. A horizontal plane across an entire sample can be imaged with approximately non-overlapping thin sheets of light, therefore, the total energy is calculated by step-wise scanning of the sample through the illumination volume. All voxels receiving illumination are incremented by 1. The final energy load is calculated as the total sum of accumulated illuminations in all voxels. The procedure was implemented of a range of parameters and two detection objectives (10x/0.6NA/8mmWD and 25/1.0NA/8mmWD). Each voxel in LSM imaging is illuminated w/f times, where w is the width of the sample, and f is the field-of-view size of the detection arm. Therefor the total energy load is ~(w/f)*number of voxels.
Note that the energy load in LSM as well as LSTM scales up by the same constant factor in LSTM and LSM, which cancels out in ratio, Data availability statement. All the datasets reported in this paper, ranging in tens of terabytes will be made available on request. Complete CAD model of LSM and other related resources will be made available with the manuscript and on a dedicated resources webpage. Complete parts list is included as a supplementary table.
Materials & Correspondence requests to Raju Tomer (raju.tomer@columbia.edu).    Table   1 for complete parts list.    A CLARITY-cleared intact Thy1-eYFP transgenic mouse brain, with attached spinal cord, was imaged with LSTM microscopy (10x/0.6NA objective). High-resolution 3D rendering, using 2x2 fold down-sampled data, is shown for the entire tissue and for a sub-volume marked by magenta rectangles. Two orthogonal views, and zoomed-in images are shown as marked by corresponding colors. The bounding boxes are 11.8mm x 27.6 mm x 5.2 mm for the whole sample, and 5.1mm  The bounding box size is 1.2mm x 1.2mm x 1mm. Note that, even with 10x/0.6NA objective, dendritic spines can be unambiguously identified. Imaging with 25x/1.0NA objective will result in further increased resolution, although at the cost of substantial increase in data size. See also Supplementary Video 6 for detailed volume rendering.