A hybrid open-top light-sheet microscope for multi-scale imaging of cleared tissues

Light-sheet microscopy has emerged as the preferred means for high-throughput volumetric imaging of cleared tissues. However, there is a need for a user-friendly system that can address imaging applications with varied requirements in terms of resolution (mesoscopic to sub-micrometer), sample geometry (size, shape, and number), and compatibility with tissue-clearing protocols and sample holders of various refractive indices. We present a ‘hybrid’ system that combines a novel non-orthogonal dual-objective and conventional (orthogonal) open-top light-sheet architecture for versatile multi-scale volumetric imaging.

sample holder [22][23][24][25][26][27]. This enables a wide range of modular sample holders to be used (e.g. well plates), as well as potential accessory technologies such as microfluidics, electrophysiology, and microdissection/aspiration. With OTLS systems, the angled orientation of the objectives reduces the usable imaging depth of the systems to well below each objective's native working distance. In addition, in OTLS microscopy, angling the optical paths with respect to a horizontal sample holder can introduce significant off-axis aberrations (especially for higher-numerical-aperture beams) unless the refractive index of the specimen and the sample holder are exquisitely well-matched (Supplementary Figure 1).
This can reduce the ease of use and versatility of high-numerical-aperture (high-NA) OTLS systems.
To overcome these issues, we considered the use of a single-objective architecture that has gained popularity in recent years, in which the illumination and collection beams share a single objective [28][29][30][31][32][33][34][35][36][37][38][39][40] (Figure 1c). For an open-top version of single-objective light-sheet microscopy, orienting the objective in the vertical direction with respect to a horizontal specimen holder makes use of the objective's full working distance and dramatically increases the system's tolerance to refractiveindex mismatch (Supplementary Figure 2). However, single-objective OTLS microscopy is incompatible with multi-scale imaging because, to the best of our knowledge, no currently available objective provides both sub-micrometer resolution and a large mesoscopic field of view (FOV). In addition, the use of a single objective both constrains and couples the illumination and collection beams such that there is less flexibility to tailor the axial and lateral resolutions (Supplementary Figure 3) [41]. Recently, we have explored the concept of a non-orthogonal dual-objective (NODO) OTLS configuration that makes use of a high-NA collection objective oriented in the vertical direction (similar to single-objective OTLS microscopy) in conjunction with a separate angled objective to provide lower-NA light-sheet illumination (Figure 1d). Our simulations indicate that a NODO configuration maintains the high index-mismatch tolerance of single-objective OTLS microscopy (Supplementary Figure 4) and offers increased flexibility to optimize lateral and axial resolutions [41].
To address the varied requirements of cleared-tissue light-sheet microscopists, we designed a 'hybrid' OTLS microscope that is the first system to use a non-orthogonal dual-objective (NODO)

Figure 1 | Hybrid open-top light-sheet (OTLS) microscopy. (a)
Orthogonal dual-objective (ODO) light-sheet microscopes typically place one or more objectives in the horizontal plane of the specimen, which constrains the specimen size and prevents the use of standard modular sample holders. (b) ODO open-top light-sheet (OTLS) microscopy systems overcome this constraint by tilting the objectives and positioning them beneath a horizontal sample holder. However, a consequence of this is a limited imaging depth and low tolerance to refractive-index mismatch between the specimen and the sample-holder material (which can reduce ease-of-use). (c) Singleobjective OTLS microscopy systems overcome both limitations. However, the use of a single objective for illumination and collection precludes broad multi-scale imaging and limits/couples the axial and lateral resolutions. (d) Non-orthogonal dual-objective (NODO) OTLS microscopy allows for laterally unconstrained imaging over the full working distance of a vertically oriented collection objective with high tolerance to index mismatch. The use of a separate objective for light-sheet illumination also enables more design flexibility to optimize axial and lateral resolutions. (e) The hybrid microscope architecture consists of three objectives, positioned below the specimen, mounted into a monolithic imaging chamber that is filled with an interchangeable immersion medium. One objective is used for light-sheet illumination, and the other two objectives are used for orthogonal dual-objective (ODO) and non-orthogonal dual-objective (NODO) collection. The ODO path provides fast meso-scale screening capabilities, and the NODO path enables targeted sub-micrometer imaging. By using a motorized stage, tiled imaging is possible with both paths over a large 12 x 7.5 x 1 cm (XYZ) imaging volume. This large imaging volume accommodates multiple intact cleared organs and large tissue slabs mounted in an array of specimen holders. The optical layout of the hybrid OTLS system is shown in (f). The ODO path uses a single tube lens to directly image the light sheet onto Camera #1. The NODO path uses a remote focus system consisting of three objective-tube lens pairs to image the non-orthogonal light sheet onto Camera #2. (g) In combination, the ODO and NODO paths enable imaging over a tunable lateral resolution range of 0.5 -10.7 μm at imaging speeds of ~5 mm 3 to 10 cm 3 per hour. The current set points of the system are highlighted. (h) At these set points, the full-width half-maximum (FWHM) resolutions (xyz) for the ODO path are 4.09 ± 1.07, 4.41 ± 0.83, and 5.48 ± 1.08 μm, and for the NODO path are 0.45 ± 0.07, 0.46 ± 0.06, and 2.91 ± 0.31 μm (mean ± standard deviation). These values are measured from reflectance imaging of 150-nm gold beads in an ECi-cleared agarose phantom (638-nm illumination). (i) Representative ODO imaging results of an entire intact CUBIC-cleared mouse brain with arterial (αSMA) and nuclear (SYTOX-G) staining. The ODO imaging path, with near-isotropic resolution, is able to clearly resolve vasculature in both the xy and xz planes (insets). (j-k) Targeted imaging of a sub-region centered on a branching arteriole using the NODO imaging path resolves individual smooth muscle cells and sub-nuclear features that are not resolved by ODO imaging. Scale-bar lengths are as follows: (i) 1 mm (insets, 500 μm), (j) 100 μm, and (k) 10 μm. All images are displayed without deconvolution.
configuration (for high-resolution imaging). For low-resolution imaging, a conventional ODO open-top system is integrated with the NODO system. This new hybrid system leverages the strengths and overcomes the limitations of previous systems (Supplementary Note 1), addressing the five requirements listed earlier: (1) simple open-top mounting of multiple specimens in standard holders such as well plates, (2) the ability to pair any tissue-clearing reagent with nearly any sample-holder material with negligible degradation in imaging performance (i.e., high tolerance to refractive-index mismatch, Supplementary Figure 5), (3) no lateral constraints on the specimen size (limited only by the travel range of the microscope stage), (4) a 1-cm imaging depth for comprehensive interrogation of intact mouse organs and thick tissue slabs, and (5) multi-scale imaging over an unprecedented range roughly corresponding to what is achieved with 2X to 40X objectives. These unique capabilities open the door for new light-sheet microscopy applications, including efficient multi-scale imaging workflows in which one or more large specimens must be rapidly screened at low resolution to identify localized regions of interest for quantitative interrogation at the sub-micrometer scale.
The layout of our new system is shown in Figure 1e. The system architecture features three main objectives that are selected to avoid geometric interference. All three objectives are positioned below the specimen, which provides an unobstructed open top that enables volumetric imaging over a large 12 x 7.5 x 1 cm (xyz) imaging volume. The imaging volume is limited in z by the objectives and in xy by the mechanical limits of the motorized stage. All three objectives are sealed into a monolithic imaging chamber through direct immersion or the use of a solid-immersion meniscus lens (SIMlens), which provides multi-immersion capabilities spanning the refractive index range of all current clearing protocols (Supplementary Figure 6) [26,27].
The optical layout of the system is shown in Figure 1f. To achieve an optimal combination of sub-micrometer resolution, large imaging depth, and compatibility with standard sample holders, we developed a new NODO light-sheet configuration [41]. By using the full numerical aperture (NA) of a vertically oriented objective for fluorescence collection, and a separate objective for non-orthogonal illumination, our new NODO architecture provides superior resolution to that of a single-objective lightsheet system (if based on the same primary objective) and relaxes the NA requirements for the remotefocus module that is necessary for these non-orthogonal light-sheet systems (Supplementary Figure   4). This allows for the use of a wider range of moderate-NA primary (collection) objectives for clearedtissue imaging, as well as simple air objectives at the remote-focus module rather than bespoke objective assemblies (Supplementary Figure 7). We carefully selected our NODO collection objective and its optical and mechanical specifications to enable the placement of a separate low-NA collection objective oriented orthogonally to the illumination objective (Supplementary Figure 8). This forms an additional ODO imaging path for rapid mesoscopic-resolution imaging. Similar to single-objective lightsheet systems, the NODO path uses a remote focus to re-image the non-orthogonal light sheet onto a camera, whereas the ODO collection path directly images the orthogonal light sheet onto a camera ( Supplementary Figures 9 -16 and Supplementary Video 1 contain optical models, a photograph of the hybrid OTLS microscope, and additional point spread functions for both imaging paths). The same illumination path is used to generate a light sheet with a variable width for both the NODO and ODO imaging paths.
The overall magnification of each imaging path is tunable, limited at the extremes by either the NA or FOV of the collection objectives (see Supplementary Note 2 for further discussion of these two imaging modes), corresponding to a combined lateral-resolution range of ~0.4 -10 μm, and volumetric imaging speeds of ~5 mm 3 to 10 cm 3 per hour (Figure 1g). In the current microscope configuration, the NODO path provides an XYZ resolution of 0.45 ± 0.07, 0.46 ± 0.05, and 2.91 ± 0.31 μm (N = 437 beads), and the ODO path provides an XYZ resolution of 4.41 ± 0.83, 4.09 ± 1.07, and 5.48 ± 1.08 μm (N = 109 beads) (mean ± standard deviation, Figure 1h). Representative imaging results of a CUBICcleared mouse brain labeled with brain-wide arterial (αSMA) and nuclear (SYTOX-G) stains are shown in Figures 1i-k (Supplementary Video 2). The entire intact specimen is rapidly imaged with nearisotropic resolution using the mesoscopic ODO path of the hybrid system, which clearly resolves vasculature in all three dimensions. Detailed interrogation of a sub-region with the high-resolution NODO path resolves individual smooth muscle cells and sub-nuclear features, which are not resolved in ODO images of the same sub-region. We further spotlight the unique utility of our hybrid system in two example applications where multi-scale imaging enables time-and data-efficient experimental workflows.
First, we imaged axons in an intact mouse brain cleared in ethyl cinnamate (ECi)

(Supplementary Video 3). Tracking the axons of individual neurons is a challenging problemaxons
can be very thin (100 nm) and span very large distances (cm). To do this effectively, one typically relies on sparse and bright labeling of a few neurons along with high-resolution, high-contrast imaging of the entire brain (~0.5 cm 3 ) [42,43]. Sub-micrometer imaging of such large volumes generates data sets that are tens of terabytes in size, necessitating computationally intensive downstream pipelines for data handling, processing, and storage. The multi-scale imaging capability of our hybrid system greatly accelerates and simplifies this process by screening an entire brain at low resolution to identify target regions with imageable neurons, followed by high-resolution imaging of the identified regions of interest.
Using the ODO imaging path, a mouse brain was rapidly screened in ~1 hour with isotropic ~2μm voxels (~4-μm resolution), revealing brain-wide axonal projections (Figures 2a-b). Inspection of dense axonal projections in xy and xz planes through the midbrain confirms the ability of the system to provide near-isotropic 4-to 5-μm resolution throughout the entire intact brain (Figure 2c). A targeted region of interest around a cortical pyramidal neuron was then imaged at <0.5-um xy resolution using the NODO imaging path (Figure 2d-e). This imaging resolution is sufficient to discern spines and varicosities on individual dendrites and axons (Figures 2f-h). Importantly, imaging a 1-mm 3 subregion containing the targeted neuron at ~0.5 x 0.5 x 2.7 μm (xyz) resolution required only 24 minutes, generating a 180-GB data set. By comparison, imaging the entire 0.5-cm 3 brain volume at this resolution would require ~2 weeks of imaging and would produce 200 TB of imaging data.
In a second example of the unique capabilities of our hybrid OTLS system, an imaging experiment was performed that required broad multi-scale imaging across multiple specimens. We used our hybrid OTLS system to study metastatic colonies from two cancer cell lines (MDA-231 and OS-RC-2) throughout intact mouse brains (Supplementary Video 4). Due to the sparse and unpredictable spatial distribution of brain metastases, identifying those sites in whole brains is challenging without a rapid low-resolution screening method. Once metastatic sites are identified, highresolution quantitative analysis of these regions is also desired. While past studies have relied on laborious and time-consuming experiments with manual transfer of individual specimens between different microscope systems [44] (for low-resolution localization followed by high-resolution analysis), we analyzed six intact mouse brains in a single imaging session without manually removing or remounting the specimens (Figure 3a).
All mouse brains were first placed in a standard 6-well plate (Figure 3b) and screened sequentially using fast mesoscopic ODO imaging (Figure 3c). After imaging, 34 total regions of interest (ROI) centered on metastatic colonies were manually identified across all mouse brains and sequentially imaged without having to move the specimens or specimen holder. An example ROI from each brain is shown in Figure 3d, with all 34 ROIs shown in Supplementary Figure 17. Visual inspection of whole-brain images reveals that metastatic growth from MDA-231 cancer cells exhibit vessel co-option (Figures 3e-g) whereas OS-RC-2 cancer-cell metastases do not. Morphological analysis of volumetrically segmented brain metastases, performed across the high-resolution ROIs, reveal statistically significant differences between the two metastatic cell lines. For example, the "volume per surface area" shape metric, which correlates with the roundness of a metastatic colony, was found to be higher for the OS-RC-2 cell line (Figure 3i). In addition, for the MDA-231 cancer cell line, the density of cancer cells in the vessel co-option ROIs was found to be lower than the adjacent metastatic colonies (Figure 3j). Both of these findings are consistent with the previous report from Kubota et al [44]. However, while this previous study required two separate microscope systems and several weeks of tedious imaging, including manual specimen transfer and co-registration of The ODO imaging path was used to rapidly screen multiple intact mouse brains containing metastatic colonies. (d) N = 34 total metastatic regions of interest (ROI) across all brains were identified and subsequently imaged at sub-micrometer resolution using the NODO imaging path (only one ROI per brain is shown for illustrative purposes). Visual inspection of a single brain with MDA-231 metastases in (e) revealed multiple colonies distributed throughout the brain, with signs of vessel co-option (f-g) that were not observed for OS-RC-2 metastases. (h) To quantify these phenotypic differences between the MDA-231 and OS-RC-2 metastases, the metastatic lumens (cyan) and cancer cells (red) were computationally segmented in all N = 34 ROIs. Quantification of the resulting segmentation masks revealed statistically (two-sample t-test) different 3D growth patterns between the two cancer cell lines, consistent with a previous report (i-j) [44]. Error bars in (i-j) denote standard deviation. Scale-bar lengths are as follows: (b-c) 1 cm, (d) 100 μm, (e) 1 mm, (f) 500 μm, and (g-h) 100 μm. All images are displayed without deconvolution. coordinates (personal correspondence with S.I.K.), our hybrid OTLS system allowed the entire imaging experiment to be completed in ~1 day (~2 hours of low-resolution imaging time per whole brain and ~20 minutes of high-resolution imaging time per ROI).
Compared to existing academic and commercial light-sheet microscopy systems, our hybrid OTLS system provides a unique combination of versatility and performance necessary to satisfy the diverse requirements of a growing number of cleared-tissue imaging applications. In addition to the imaging examples in Figures 1-3, the system has potential utility for multi-scale non-destructive 3D pathology of prostate cancer with ECi clearing (Supplementary Figure 18)  In the future, both the axial and lateral resolution of the hybrid OTLS system can be improved.
For example, the axial resolution of the system can be enhanced by incorporating a Bessel beam or by using a higher-NA Gaussian beam for light-sheet illumination [17,18,45,46] (Supplementary Figure   26). This is facilitated by the fact that unlike single-objective light-sheet systems, our new NODO architecture allows the illumination NA to be substantially increased without necessitating a concomitant decrease in the collection NA (i.e. decoupled optical paths as shown in Figure 1d). With the 45-deg orientation of the two low-NA objectives in our system (one on either side of the NODO collection objective), it may also be possible to perform dual-sided illumination with fusion deconvolution to achieve improved image quality (Supplementary Figure 27) [20,37,47]. The lateral resolution of the system can also be improved by increasing the collection NA of the multi-immersion objective. However, the optical cone angle cannot be increased beyond 40 -45 deg for a hybrid NODO / ODO system, as this would prevent the ODO objectives from being positioned at 45 deg with respect to the vertical axis.
While our current design uses three separate objectives, it is theoretically possible to achieve improved performance with an optimized NODO objective (Supplementary Figure 28), or to achieve similar multi-scale OTLS performance using a single high-NA objective with a large FOV and working distance (see Supplementary Note 4 and Supplementary Figure 29). For example, the pupil of a single objective could be split into three separate regions, using the edges for NODO illumination and for mesoscopic ODO imaging, and only the center for high-resolution NODO collection. However, this would require an objective with extraordinary specifications that could be prohibitively expensive to design and manufacture. Therefore, in summary, our hybrid OTLS design represents a practical means, with commercially available optical components, of achieving an impressive balance of performance and versatility for a growing number of cleared-tissue imaging experiments in which rapid low-resolution screening of large volumes is desired in addition to high-resolution characterization of localized subregions.

Hybrid open-top light-sheet microscope
Light-sheet-based imaging is achieved using three optical arms. The first obliquely illuminates the specimen with a light sheet at 45 deg relative to the vertical axis. The second is oriented vertically, enabling high-resolution collection of the light-sheet-generated fluorescence in a NODO configuration.

Illumination optical path
The illumination optical arm is shown in Supplementary Figure 9. The illumination optics are designed to allow for the light-sheet properties (i.e., width, thickness, and depth of focus) to be adjusted. In addition, the optical path is designed to be compatible with the refractive index, n, of all current clearing protocols (i.e., it has multi-immersion capabilities) and to minimize chromatic aberrations and defocusing (i.e., variations in the illumination focal length as a function of wavelength).
Illumination light is fiber coupled into the system with a Gaussian numerical aperture (NA) of ~0.12 and collimated using an objective (RMS20X, Olympus). The beam diameter is then adjusted using a 4X variable beam expander (BE052-A, Thorlabs). This serves to adjust the overall NA of the light sheet. The variably expanded beam is then passed through an electronically tunable lens (ETL) that enables axial adjustment/alignment of the light sheet (EL-16-40-TC-VIS-5D-C, Optotune). The axially adjusted beam is relayed 1:1 using a pair of lenses (AC254-75A, Thorlabs) so that it can be scanned using a pair of large-beam-diameter galvanometric scanning mirrors (GVS012, Thorlabs). One mirror is scanned to create a digitally scanned light sheet [48]. The other scanning mirror is used to align the light sheet with the focal plane of the NODO and ODO imaging paths. A pair of achromatic doublet lenses (AC508-75A and AC508-200A, Thorlabs) are then used to relay the scanned beam to the back focal plane of a 2X illumination objective with NA = 0.10 (TL2X-SAP, Thorlabs). Finally, the illumination light travels through the SIMlens (Supplementary Figure 6) [26]. The SIMlens provides multi-immersion performance and prevents aberrations (spherical, off-axis, and chromatic) of the light sheet by minimizing refraction of the illumination rays as they transition from air into the immersion medium. In addition, since ray angles are preserved as they transition between air and the immersion medium, the SIMlens increases the NA of the illumination light sheet by a factor of n. When combined, this optical design yields a light sheet with tunable NA (0.025 -0.10 x n) and tunable width (0 -11 mm / n) limited by the field of view (FOV) of the illumination objective.
A digitally scanned light sheet was chosen over a cylindrical-lens approach (static light sheet), as this facilitates achieving a high level of tunability for multi-scale imaging. Moreover, large scanning mirrors were selected to fill the back focal plane of the final illumination objective and to avoid having to significantly magnify the beam after the scanning mirrors, which would reduce the lateral scanning range of the mirrors (i.e., constrain the maximum light-sheet width). In the current design, rotating the scanning mirror results in lateral scanning at a ratio of ~0.60 mm per deg / n. The maximum desired light sheet width (i.e., lateral scanning range) is 11 mm, corresponding to a scanning angle of ~18 deg, which is within the maximum scan range of the galvo scanner (20 deg).

NODO optical path
The physical layout, ZEMAX model, and objective options for this NA-maximized NODO imaging configuration are shown in Supplementary Figure 10. The NODO optical path of our system uses a multi-immersion objective (#54-12-8, Special Optics) with a long 1-cm working distance. This objective is compatible with all clearing protocols (n = 1.33 -1.56) and provides a NA of 0.483 (in air) that scales with the index of the immersion medium (e.g., NA ~ 0.75 at n = 1.56). The lens is oriented in the normal (vertical) direction with respect to the specimen holder/interface and is therefore non-orthogonal to the light sheet. To image this non-orthogonal light sheet, we use a remote-focus imaging strategy analogous to what is used for single-objective light-sheet systems, with the multi-immersion objective serving as the primary objective (O1) [28,29].
To minimize aberrations in the remote focus relay, the overall magnification from the specimen to the remote focus (air) should be equal to the refractive index of the specimen, n. Given this requirement, the relay lenses and first remote objective (O2) must be carefully selected. For O2, a 20X objective is optimal, as a 10X or 40X objective would clip either the NA or FOV of our O1. Of the several companies that produce microscope objectives, Zeiss and Leica were avoided because chromatic aberrations are partially corrected in the tube lenses produced by these companies, which would complicate selection of the two relay tube lenses. Therefore, only objectives from Olympus and Nikon were considered, where 20X objectives from Olympus have a focal length of 9 mm, and 20X objectives from Nikon have a focal length of 10 mm. Factoring in the effective focal length of O1 (12.19 mm / n), the required relay lens magnification is ~1.219X for Nikon and ~1.354X for Olympus. Note that the magnification of the multi-immersion objective in our system scales inversely as a function of n. This allows us to use a fixed set of relay lenses and always satisfy the remote focusing magnification requirement. However, this would be problematic with alternative multi-immersion objectives, where the effective focal length does not vary with n, and therefore the magnification of the relay lenses would need to be adapted for each immersion medium.
Well-corrected tube lenses are available with a limited selection of focal lengths (100, 165, 180, and 200 mm). Although custom tube lens assemblies are possible [33], we found that off-the-shelf 200-mm (Nikon Tube Lens #58-520, Edmund Optics) and 165-mm (TTL165-A, Thorlabs) tube lenses provide a magnification of ~1.212X, which matches the requirement for Nikon. Therefore, we decided to select a 20X Nikon objective for O2. To avoid the need for a cover glass, we narrowed our selection to Nikon objectives designed for use without a cover glass. This yielded one option, the LU (now TU) Plan Fluor EPI 20X (NA = 0.45). We chose this objective for the O2 in the current system. The goal of O3 is to maximize light collection when tilted at the angle required to orthogonally image a remote version of the oblique light sheet within the specimen. As mentioned previously, in a single-objective light-sheet design, the light sheet angle would be limited by our chosen O1 to a maximum of 28.9 deg. This would require O3 to be tilted by at least 61.1 deg. At this extreme tilt angle, one way to prevent light loss is through a custom solid-or liquid-immersion objective [32,33]. However, a benefit of our NODO design is that the crossing angle can be increased to 45 deg, which reduces the tilt of O3 to 45 deg. At this tilt angle, an objective with NA = 0.95 is able to capture light up to NA = 0.45 and provide NA-maximized imaging. Therefore, we opted to use a NA = 0.95 air objective for our O3.
This use of a tertiary air objective makes alignment more straightforward and stable than in the case of a solid-or liquid-immersion objective [32,33]. This is especially the case for liquid-immersion objectives, where there may be evaporation or leakage of a liquid medium over time.
In terms of selecting an optimal O3 for NA-maximized imaging, we only considered objectives from Olympus and Nikon for the same reasons as mentioned previously. Both companies offer two types of objectives with NA = 0.95. These include 40X life-science objectives with a correction collar for cover-glass thicknesses ranging from t = 0.11 -0.23 mm, and 50X metrology objectives for imaging with no cover glass. Although it would be possible to permanently align or adhere a cover glass to the 40X objectives, we decided to select the CF IC EPI Plan Apo 50X objective from Nikon for our O3 for simplicity and ease-of-use. When combined with a 100-mm tube lens (TTL100-A, Thorlabs), our NODO imaging path provides a total magnification of 25 x n. This yields a sampling rate of ~2.71 (slightly better than Nyquist) when using a sCMOS camera with pixels spaced by 6.5 μm (pco.edge 4.1, PCO Tech). The corresponding FOV is ~ 0.53 mm / n (FOV = 0.40 -0.34 mm when n = 1.33 -1.56), which is not clipped by the 0.40-mm FOV of the 50X objective. In this configuration, the back aperture of the illumination objective is filled, yielding an illumination NA of ~0.10 x n. This corresponds to a confocal parameter of ~40 -50 μm and a pixel height of ~256 pixels for each raw camera image. This also corresponds to an axial re-focusing range that is well within the range of operation for an idealized objective, as specified by Botcherby et al. [28]. The pixel width of each raw camera frame is the full 2048 pixels of the sCMOS camera, corresponding to a FOV of ~0.34 -0.40 mm.
While the above O3 selection provides NA-maximized imaging, our O1 also provides a 1-mm FOV to enable FOV-maximized imaging (with slightly worse resolution). To achieve this, O3 can be

ODO optical path
The physical layouts and ZEMAX models of the ODO imaging configurations are shown in Supplementary Figure 13-14. While the NODO imaging path can provide sub-micrometer resolution, the FOV of the multi-immersion primary objective (O1) is restricted to 1 mm, which is insufficient for fast mesoscopic imaging. This tradeoff between NA and FOV is standard across all currently available microscope objectives (e.g., no current clearing-compatible objectives can simultaneously offer submicrometer resolution over a mesoscopic FOV) (Supplementary Figure 8) [49]. Therefore, in our hybrid system, we achieve low-resolution imaging with a second independent ODO imaging path.
The ODO collection path uses the same objective as the illumination optical path (TL2X-SAP, Thorlabs). The objective is similarly used in conjunction with a SIMlens (fabricated by BMV Optical), which provides multi-immersion performance and prevents axial chromatic aberrations in the ODO imaging path [26]. In addition, the SIMlens increases the NA of the collection path by a factor of n. This yields an effective NA = 0.10 x n.
The current set point of the ODO collection path lies between the NA-and FOV-maximized imaging extremes and uses a tube lens with a 200-mm focal length (TTL200-A, Thorlabs). This provides a magnification of 2 x n and an effective FOV of 6.6 mm / n, which undersamples the collection NA of the system when using a sCMOS camera with a pixel spacing of 6.5 μm (pco.edge 4.1, PCO Tech, 2048 x 2048 pixels). In this configuration, the back aperture of the illumination objective is underfilled to yield an illumination NA of ~0.04 x n. This results in a confocal parameter of ~250 -300 μm, which corresponds to a digital image height of ~128 pixels for each raw camera image. The width (in pixels) of each raw camera image is the full 2048 pixels of the sCMOS camera, corresponding to a distance of ~4.25 -5 mm in the sample. This combination of illumination and collection NA provides near-isotropic resolution (Supplementary Figure 16Error! Reference source not found.).
The magnification of the ODO imaging path is easily adjusted by changing the tube lens. NAmaximized imaging is achieved using a tube lens with a 400-mm focal length (AC508-400-A, Thorlabs).
This provides a magnification of 4 x n and an effective FOV of 3.3 mm / n, which corresponds to a near-Nyquist sampling rate of ~2.1 when using a sCMOS camera with a pixel spacing of 6.5 μm. FOVmaximized imaging is achieved using a tube lens with a 100-mm focal length (TTL100-A, Thorlabs).
This provides a magnification of 1 x n and an effective FOV of 13.33 mm / n. See Supplementary Note 2 for more discussion of NA-versus FOV-maximized imaging modes.

Image acquisition and post processing
Image strips are collected with a combination of stage-scanning and lateral/vertical tiling. The stagescanning firmware is used to send a TTL SYNC trigger signal from the XY stage to the sCMOS camera for reproducible initiation of each imaging strip. After triggering, the camera is set to free-running mode and acquires the desired number of frames for a given image-strip length (as the sample is scanned by the stage at a constant velocity), with a spacing between adjacent frames that is identical to the sampling pitch of the raw camera images. For each raw frame, the camera uses the standard rolling shutter, where the shutter rolls from the center of the camera chip in both directions, as opposed to the light-sheet readout mode, where the shutter rolls from the top to the bottom of the camera chip. Using the standard rolling shutter, the rolling directions are oriented along the light sheet propagation direction. This orientation allows the imaging speed to be increased when the pixel height of each raw frame is cropped to match the confocal parameter of the illumination light sheet.
At the start of each exposure, the camera sends a trigger to an analog output DAQ card (PCIe-6738, National Instruments). The DAQ card then sends output voltages to the lasers, galvos, and ETL (for alignment only, not axial sweeping). To reduce motion blur, the lasers and galvos are triggered with a delay to only illuminate once the shutter has rolled across the full pixel height of the camera frame, resulting in a strobing effect. For the 128-or 256-pixel height of the ODO and NODO paths respectively, the rolling time, troll, is 0.625 and 1.25 ms. The exposure time, texp, is set to 3 x troll, resulting in a total exposure time, ttot, of 4 x troll, or 2.5 and 5 ms for the ODO and NODO paths. This corresponds to a data rate that is 1/4 the maximum data rate of the sCMOS camera. The lateral scanning mirror is actuated with a sawtooth waveform and completes a single period within the total exposure time, ttot, of the raw camera frame, corresponding to a frequency of ~400 and ~200 Hz for the ODO and NODO paths. The ETL and second mirror are set to a pre-calibrated DC voltage for the entire image strip to yield an in-focus light sheet that is axially aligned to the center of the camera chip.
Raw camera frames are streamed from the camera to RAM and subsequently saved directly as a single HDF5 file and XML file with the associated metadata, which is suitable for immediate processing with BigStitcher via ImageJ [50,51]. This involves on-the-fly saving of down-sampled copies of the image strip (2x, 4x, 8x, and 16x) in the hierarchical HDF5 file format. Additionally, the GPU-based B3D compression filter can be optionally added to the HDF5 writing process to yield 5-10X compression with negligible loss of usable information content [52]. However, both processing steps slow down the net data rate from RAM to disk. On-the-fly saving of down-sampled copies of the image strips slows the data rate by ~2x, and the B3D algorithm slows down the process by another ~2x. This net speed reduction of ~4x motivated our selection of the camera frame rate mentioned previously. For our experiments, this reduction in post-processing times and data-storage requirements are worth the reduction in imaging speed. However, it is important to note that the imaging speeds could be increased by 2x if the aforementioned processing steps are omitted from the acquisition procedure, in which case the imaging speed would become limited by the duty cycle of the illumination strobing, where texp = troll. The imaging speeds quoted in Figure 1g assume the 4x-reduced data rate.
A ~15% overlap is used for both vertical and lateral tiling. Different wavelength channels are acquired sequentially. For each image strip "tile" that is acquired by laterally scanning the specimen in the y-direction, all channels are acquired by cycling through various laser/filter combinations and rescanning that image strip for each laser/filter setting, before moving to the next tile position. When tiling vertically, the laser power is increased with depth per a user-defined exponential relationship, P = P0 × exp(z/μ), to account for the attenuation of the illumination light sheet as it penetrates deeper into the specimen (typically μ = 5 -20 mm -1 ). Finally, if desired, all of the imaging tiles can be aligned and fused into one contiguous 3D image as an HDF5 or TIFF file output using BigStitcher [50]. The entire image acquisition is controlled using a custom-written Python program that is available from the authors upon request.

Computer hardware
During acquisition, the images are collected by a dedicated custom workstation (Puget Systems) equipped with a high-specification motherboard (Asus WS C422 SAGE/10G), processor (Intel Xeon W-2145 3.7GHz 8 Core 11MB 140W), and 256 GB of RAM. The motherboard houses several PCIe cards, including 2 CameraLink frame grabbers (mEIV AD4/VD4, Silicon Software) for streaming images from the camera, a DAQ card (PCIe-6738, National Instruments) for generating analog output voltages, a 10G SFP+ network card (StarTech), and a GPU (TitanXP, NVIDIA). Datasets are streamed to a local 8 TB U.2 drive (Micron) that is capable of outpacing the data rates of the microscope system. Data is then transferred to a mapped network drive located on an in-lab server (X11-DPG-QT, SuperMicro) running 64-bit Windows Server, equipped with 768 GB RAM and TitanXP (NVIDIA) and Quadro P6000 (NVIDIA) GPUs. The mapped network drive is a direct-attached RAID6 storage array with 15 × 8.0 TB HDDs. The RAID array is hardware based and controlled by an external 8-port controller (LSI MegaRaid 9380-8e 1 GB cache). Both the server and acquisition workstation are set with jumbo frames (Ethernet frame), and parallel send/receive processes matched to the number of computing cores on the workstation (8 physical cores) and server (16 physical cores), which reliably enables ~ 1.0 GB sec −1 network-transfer speeds.

Preparation of ECi-cleared mouse brain
Labeling and clearing was carried out as previously described [43]. For sparse labeling, a Slc17a7-Cre mouse (8 weeks old, female) received a systemic injection, via the retro-orbital sinus, of a mixture of Cre-dependent Tet transactivator (PHP-eB-Syn-Flex-TRE-2x-tTA) and a reporter virus (PHP-eB-CAG-