A versatile oblique plane microscope for large-scale and high-resolution imaging of subcellular dynamics
Abstract
We present an oblique plane microscope (OPM) that uses a bespoke glass-tipped tertiary objective to improve the resolution, field of view, and usability over previous variants. Owing to its high numerical aperture optics, this microscope achieves lateral and axial resolutions that are comparable to the square illumination mode of lattice light-sheet microscopy, but in a user friendly and versatile format. Given this performance, we demonstrate high-resolution imaging of clathrin-mediated endocytosis, vimentin, the endoplasmic reticulum, membrane dynamics, and Natural Killer-mediated cytotoxicity. Furthermore, we image biological phenomena that would be otherwise challenging or impossible to perform in a traditional light-sheet microscope geometry, including cell migration through confined spaces within a microfluidic device, subcellular photoactivation of Rac1, diffusion of cytoplasmic rheological tracers at a volumetric rate of 14 Hz, and large field of view imaging of neurons, developing embryos, and centimeter-scale tissue sections.
Introduction
Light-sheet fluorescence microscopy (LSFM) first generated significant interest in the biological community as a result of its ability to image developing embryos with single-cell resolution, inherent optical sectioning, low phototoxicity and high temporal resolution (Huisken et al., 2004). Since then, LSFM has undergone a revolution, and depending on the optical configuration, can now routinely image biological systems that range from reconstituted macromolecular complexes (Keller et al., 2007) to intact organs and organisms (Chakraborty et al., 2019; Voigt et al., 2019). Unlike other microscope modalities that are used to image three-dimensional specimens (e.g. confocal), LSFM delivers light to only the in-focus portion of the specimen, and therefore substantially decreases the illumination burden on the sample. Further, light-sheet excitation combines powerfully with the million-fold parallelization afforded by modern scientific cameras, permitting massive reductions in illumination intensities without compromising the signal-to-noise ratio, which significantly reduces the rate of photobleaching and phototoxicity. Consequently, LSFM enables imaging of biological specimens for 1000’s of volumes (Dean et al., 2017).
Despite its advantages over other imaging modalities, its widespread adoption remains limited. In part, this is due to the slow adoption of cutting-edge LSFM systems by commercial entities, and consequently, the requirement that each lab assemble, align, maintain, operate their own LSFM instruments. Sample preparation is an additional problem, as the orthogonal geometry of LSFM systems often sterically occludes standard imaging dishes such as multi-well plates. Furthermore, the highest resolution LSFM systems, the reliance on high-NA water-dipping objectives places the sample in direct contact with non-sterile optical surfaces, which compromises long-term imaging. And these matters are made worse because modern LSFM systems often lack modalities that render microscopy routinely useful for non-experts, including sample environmental control (e.g. CO2, temperature, and humidity), oculars, and hardware-based autofocusing schemes.
Given these concerns, several labs have worked to identify single-objective imaging systems that combine light-sheet illumination, more-traditional sample mounting, and high numerical aperture (NA) fluorescence detection. For example, using the same objective for illumination and detection, Gebhardt et al created a bespoke atomic force microscope cantilever that reflected a light-sheet into the ordinary viewing geometry of an inverted microscope (Gebhardt et al., 2013). Similarly, several labs developed specialized microfluidics with micromirrors positioned at 45 degrees (Galland et al., 2015; Meddens et al., 2016). Nonetheless, these approaches are only compatible with low-NA illumination (which reduces resolution and optical sectioning), have a limited volumetric imaging capacity, and require that the reflective surface be placed in immediate proximity to the specimen, which drastically limits the field of view. More recently, a high-NA oil immersion lens (1.49) was combined with lateral interference tilted excitation (Fadero et al., 2018). However, this system required specialized imaging chambers, used light-sheets that were several microns thick and tilted relative to the imaged focal plane, and owing to the large refractive index mismatch between the objective immersion media and the specimen, suffered from spherical aberrations, which ultimately impact the resolution and sensitivity of the microscope in a depth-dependent manner (Fadero et al., 2018). Thus, each of these systems are incompatible with many experimental designs, including biological workhorses like the 96-well plate and large-scale tissue sections.
There is however one form of LSFM, referred to as oblique plane microscopy (OPM), that avoids these complications (Dunsby, 2008). Unlike most LSFMs that require the light-sheet to be launched from the side or reflected off of a cantilever or microfluidic device, an OPM illuminates the sample obliquely and captures the fluorescence with the same objective (Figure 1A). As such, an OPM can be assembled using a standard inverted or upright microscope geometry and is entirely compatible with traditional forms of sample mounting (including multi-well plates), environment control, and laser-based auto-focusing. In OPM, the fluorescence captured by the primary objective is relayed in an aberration-free remote focusing (Botcherby et al., 2012) format to a secondary objective, which creates a 3D replica of the fluorescent signal at its focus that is then imaged by a tilted tertiary objective onto a camera (Figure 1B). To date, most oblique plane microscopes have had low optical resolution because the tertiary objective fails to capture all of the high-angle rays that are launched from the secondary objective (Figure 1B). Nevertheless, it was recently demonstrated that an OPM can achieve ~300 nm scale resolution by combining air and water objectives with a coverslip-walled immersion chamber (Yang et al., 2019). Here, as light rays travel from a low to a high refractive index medium (e.g. air and water, respectively), they are refracted at the coverslip interface toward the tertiary objective in a manner that compresses the optical cone of light, and thereby permits capture of the higher angle rays (Yang et al., 2019; Figure 1C and Figure 1–figure supplement 1). Nevertheless, aligning this chamber, which requires that the coverslip be placed at the proper distance and angle relative to both the secondary and tertiary objectives, while maintaining a proper water immersion for the tertiary objective, although feasible, is technically challenging. Further, this arrangement was only compatible for secondary objectives with a maximum NA of 0.9, as higher NA objectives would collide with the water chamber.

Optical principle of oblique plane microscopy.
(A) The light-sheet, shown in light blue, is launched from the primary objective at an oblique illumination angle, and the resulting fluorescence cones of light, shown in green, are collected by the same objective. (B) In a traditional oblique plane microscope, a replica of the fluorescence collected from the primary objective is formed by the secondary objective at the focus of the tertiary objective. However, owing to the off-axis imaging geometry, high-angle rays cannot be captured by the tertiary objective. (C) If instead the light travels from a region of low refractive index (η1, air) to high refractive index (η2, water or glass), then the optical cone of light is compressed from an angle of α to β and refracted toward the tertiary imaging system, thus permitting its capture, and maximizing the resolving power of the microscope.
Here, to mitigate these challenges, we built a high-NA OPM equipped with a recently developed glass-tipped tertiary objective (Millett-Sikking and York, 2019) that eliminates the need for an immersion chamber and further improves instrument performance (Appendix 1). Compared to other single-objective LSFMs and OPMs, we demonstrate that this OPM provides a unique and impressive combination of field of view, resolution, volumetric imaging capacity, and speed. As proof of principle, we image a number of biological processes that would otherwise be challenging to observe without this unique combination of microscope geometry, speed, resolution, and field of view, including nuclear rupture in melanoma cells as they migrate through tightly confined spaces in a microfluidic device, immunological synapse formation, cleavage furrow ingression in developmental systems, rheological cytosolic flows, optogenetic activation of Rac1, and intact imaging of an complete coronal murine brain slice.
Results
Microscope design
We designed an OPM capable of leveraging the maximum resolving power and field of view of a bespoke glass-tipped tertiary objective (Millett-Sikking and York, 2019). In this design, a high-NA primary objective (100X, NA 1.35) with an angular aperture of ~74 degrees is matched to a secondary air objective (40X, NA 0.95) with a similarly large angular aperture, which relays the fluorescence to a tertiary imaging system that is oriented 30 degrees off-axis (See Materials and methods). Other angles are possible (between 0 and 45 degrees), so long as the angle is matched to that of the light-sheet in sample space (Appendix 2). We chose a silicone immersion primary objective because the refractive index of living cells (~1.40) is closer to silicone (1.40) than water (1.333) or oil (1.52), thus reducing spherical aberrations and improving the overall imaging performance (Phillips et al., 2012). In cases where imaging through an aqueous boundary is unavoidable, the refractive index of the solution can be adjusted in a non-toxic manner using readily available reagents (Boothe et al., 2017). To achieve aberration- and distortion-free imaging, the optical train was carefully designed to properly map the pupils of the primary and secondary objectives and lenses were selected that maximized the near-diffraction-limited field of view (Figure 1—figure supplement 2). Inspired by tilt-invariant imaging systems (Kumar et al., 2018; Voleti et al., 2019; Yang et al., 2019), the microscope was equipped with a high-speed galvanometer mirror conjugate to both the primary and secondary objective pupils, allowing for rapid light-sheet scanning in sample space, and rapid emission descanning prior to detection with the camera. Because the galvanometer mirror is the only moving part, sources of optical drift are minimized, and only a portion of the camera is necessary to detect the descanned fluorescence (~256×2048), which permits very high imaging rates (~800 planes per second). Furthermore, for illumination, we developed a versatile laser launch that can be reconfigured in a fully automatic fashion to illuminate the cells with either an oblique light-sheet that is equipped with resonant multi-directional shadow suppression (Huisken and Stainier, 2007), widefield, or a laser-scanned and near-diffraction-limited beam for localized optogenetic stimulation, fluorescence recovery after photobleaching, or photoactivation (See Materials and methods, Figure 1—figure supplements 3–7). Owing to the large number of optics, absorption and spurious reflections resulted in a 59 and 47% decrease in fluorescence transmission for the laser-scanning and stage-scanning variants of the microscope, respectively, at 30-degrees. Transmission improved only slightly (3%) when the optical system was arranged at 0-degrees, indicating that the tertiary objective was indeed capable of capturing most of the transmitted light under a 30-degree tilt. The entire microscope was built in a standard inverted format with a motorized sample stage, objective and sample heating, and a temperature and CO2-regulated environmental chamber.
Instrument characterization
An obliquely oriented light-sheet is launched from a single objective into the specimen at an angle of 30 degrees, simultaneously illuminating a two-dimensional plane along the X and S axes. By scanning the laser in the Y-direction and collecting images at each intermediate XS plane, a three-dimensional volume is acquired (Figure 2A,B and C). Computationally shearing places these data into its proper Euclidian context and results in a parallelepiped-shaped image volume, which is readily visualized by imaging 100 nm green fluorescent beads embedded in a 1% agarose gel (Figure 2D). Here, at the highest illumination NA (0.34), a narrow strip of beads coincident with the illumination beam waist appear sharp. By fitting each bead to a three-dimensional Gaussian function, the raw (e.g. non-deconvolved) axial resolution for these data are 587 ± 18 nm (mean and standard deviation of the Full-Width Half Maximum, FWHM, Figure 2E). As the NA of the illumination decreases, the length and thickness of the light-sheet grows. While this improves the uniformity of the resolution and contrast throughout the field of view spanned by the XS plane, it ultimately reduces the raw axial resolution to 736 ± 10 nm and 918 ± 12 nm, for NA 0.16 and NA 0.06 illumination beams, respectively (Figure 2D and E). For most biological experiments reported here, we used an illumination NA of 0.16, yielding a Gaussian beam that has a thickness and propagation length of ~1.2 and~37 microns, respectively. Importantly, in cases where the illumination beam is thicker than the depth of focus of the detection objective, optical sectioning (e.g. the ability to reject out-of-focus fluorescence) will be slightly reduced. Nonetheless, as is evident from the PSF and optical transfer function, the OPM presented here delivers adequate optical sectioning over the full range of excitation NAs used in this manuscript (Figure 2—figure supplement 1). Theoretically, the maximum imaging depth of our remote focusing system is ~60 microns, beyond which tiling in the Z-dimension can be performed until one reaches the working distance of the primary objective (300 microns). Of note, the choice of illumination angle is accompanied by tradeoffs in light-sheet thickness, imaging depth, detection efficiency, and resolution (Appendix 2). Indeed, we observed a gradual loss in NA and thus resolution as our tertiary imaging system was adjusted from a 0 to a 30-degree tilt. Unlike single-objective light-sheet systems that use oil immersion objectives (Fadero et al., 2018), spherical aberrations are not immediately evident (Figure 2D and E, and Figure 2—figure supplement 2). For an oblique illumination angle of 30 degrees and an excitation NA of 0.16, the raw axial resolution of our system is similar to the raw 666 nm axial resolution reported for the most commonly used square illumination mode of lattice light-sheet microscopy (Chang et al., 2020; Valm et al., 2017) or a spinning-disk microscope using the same NA 1.35/100X objective used here (Figure 2—figure supplement 3).

Microscope illumination geometry, light-sheet properties, resolution, and field of view.
(A) The light-sheet is launched from a single primary objective at a (B) 30-degree oblique illumination angle, and rapidly scanned in the Y-direction to acquire a 3D volume. (C) At each intermediate position, a two-dimensional plane along the X and S axes is acquired with a scientific CMOS camera. (D) 100 nm fluorescent beads embedded in agarose. Sharp strip of beads reveals the position of the illumination light-sheet beam waist and scan trajectory along the Y-axis. Parallelepiped data geometry results from oblique illumination and computational shearing of the data. (E) FWHM of beads for different illumination light-sheet NAs, as a function of distance from the coverslip. Raw data, no deconvolution. The confocal parameter of the light-sheets projected onto the Z-axis is 3.0, 7.5, and >12 microns, for NAs of 0.34, 0.16, and 0.06, respectively. (F) Lateral and axial resolutions of surface immobilized fluorescent nanospheres and using an excitation NA of 0.16. Raw data, no deconvolution (G) Dimensions of the field of view for the OPM presented here when operated in a laser-scanning format compared to lattice light-sheet microscopy, eSPIM, a microfluidic-based micromirror, and an atomic force microscopy (AFM)-based cantilever. Lateral dimensions of the figure represent the X and Y axes. (H) Representative point-spread functions for surface immobilized fluorescent nanospheres, before and after deconvolution. Scale Bar: 500 nm. (I) Maximum intensity projection of a MV3 cell expressing genetically encoded multimeric nanoparticles. Deconvolved data is shown. Scale Bar: 10 microns. (J) Fourier Ring Correlation analysis of intracellular resolution. Solid lines show mean value, shaded area is the 95% confidence interval.
In an effort to more-systematically evaluate microscope performance, we also measured the resolution for 100 nm coverslip-immobilized fluorescent beads. Here, we used an illumination light-sheet with an NA of 0.16. For raw data, we measured a resolution of 299 ± 21, 336 ± 16, and 731 ± 21 nm in X, Y, and Z, respectively, throughout a lateral field of view of ~180 x~180 microns (Figure 2F). In the central ~60 x~60 micron portion of the field of view, a slightly improved lateral resolution was observed (284 ± 12 and 328 ± 14 for X and Y, respectively). Nonetheless, the resolution was relatively uniform throughout the full ~180 x~180 micron field of view (e.g. the footprint of the imaging volume in X and Y), which importantly is 2.6x, 3.8x, 37.9x, and 268x larger than reported for lattice light-sheet microscopy (Chen et al., 2014), eSPIM (Yang et al., 2019), micromirror (Galland et al., 2015; Meddens et al., 2016), and atomic force microscopy (AFM) cantilever-based methods (Gebhardt et al., 2013), respectively (Figure 2G). Of note, the field of view in lattice light-sheet microscopy, eSPIM and the OPM presented here is bounded in the X-dimension by the optics of the microscope and the finite size of the illumination beam and the camera chip size. In contrast, the Y-dimension is essentially unlimited in a sample scanning format. Nonetheless, through careful optical design, the lateral extent of our imaging field (e.g. the X- dimension) in a laser-scanning format is ~2.6 x larger than those reported for eSPIM (Yang et al., 2019). This is an important design parameter for live-cell microscopy, since laser-scanning based imaging is much more rapid than sample scanning and tiling approaches.
As an alternative resolution estimation metric, we also applied image decorrelation analysis, which evaluates the cross-correlation in frequency space between an image and its frequency-filtered equivalent (Descloux et al., 2019). Decorrelation analysis for the 180 × 180 micron field of view resulted in an aggregate raw lateral resolution (e.g. the average of both the X and Y dimensions) of 325 ± 25 nm (mean and standard deviation, Figure 2—figure supplement 4), which is in good agreement with our FWHM measurements. Deconvolution is commonly used in LSFM to improve image contrast and resolution. Here, 20 iterations of Richardson-Lucy deconvolution yielded resolutions of 203 ± 24, 209 ± 33, and 523 ± 60 nm in X, Y, and Z, respectively, and representative point spread functions for both raw and deconvolved data are shown in Figure 2H. Furthermore, microscope performance remained robust even when imaging beyond the nominal focal plane of the primary objective (Figure 2—figure supplement 5). Importantly, a point-spread function obtained by imaging a sub-diffraction bead is sufficient to describe the resolution and optical sectioning capacity of a fluorescence microscope. Nonetheless, sub-diffraction beads do not necessarily capture the optical complexities of the intracellular environment. Thus, we also sought to estimate instrument performance using fixed cells expressing genetically encoded multimeric nanoparticles (GEMs), which are self-forming cytosolic 40 nm diameter icosahedral assemblies of fluorescent proteins (Figure 2I). Using Fourier Ring Correlation analysis, which evaluates the spatial frequency-dependent signal-to-noise for a pair of images, we measured an aggregate lateral resolution of 343 ± 12 and 220 ± 23 nm, for raw and deconvolved GEMs, respectively (Figure 2J). These resolution values were also in close agreement with those obtained with decorrelation analysis (322 ± 20 and 251 ± 3 nm, raw and deconvolved mean and 95% confidence intervals, respectively). By comparison, the raw lateral resolution for lattice light-sheet microscopy for a GFP-like fluorophore is 312 nm (Valm et al., 2017). Thus, when compared to other single-objective LSFMs, OPMs, and even lattice light-sheet microscopy, the OPM presented here achieves a unique combination of resolution, field of view, and rapid imaging.
Biological imaging of clathrin-mediated endocytosis, vimentin, and membrane dynamics
To evaluate microscope performance on biological specimens, we first imaged the endoplasmic reticulum in U2OS osteosarcoma cells (Figure 3A, and Video 1). Endoplasmic reticulum tubules were highly dynamic, and unlike methods that rely on imaging slightly out of a total internal reflection geometry (Li et al., 2015), could be imaged with high-resolution throughout the entire cell volume. For comparison, an image of the endoplasmic reticulum without deconvolution is provided in Figure 3—figure supplement 1. We also imaged vimentin in retinal pigment epithelial cells, which is an intermediate filament that is often associated with the epithelial to mesenchymal transition and hypothesized to reinforce polarity cues through crosstalk in the microtubule, actin, and integrin-mediated adhesion cellular systems. Here, vimentin appeared as moderately dynamic filamentous structures that extended from the perinuclear region to the cell periphery (Figure 3B, and Video 2). Vimentin filaments occupied both the apical and basal sides of the cell, as is visible in an axial cross-section through the cell (Figure 3C). To evaluate our ability to image more dynamic processes, we imaged clathrin -mediated endocytosis in retinal pigment epithelial cells that were labeled with the clathrin adapter protein AP2 fused to GFP (Figure 3D and E, Videos 3 and 4). Endocytosis could be observed through time, with individual endocytic pits appearing as point-like structures that initialized on the plasma membrane, locally diffused, and then disappeared upon scission and release into the cytosol. Lastly, we imaged MV3 melanoma cells tagged with a membrane marker. These cells displayed numerous dynamic cellular protrusions, including blebs and filopodia, which extended away from the coverslip and otherwise could not have been observed without high-spatiotemporal volumetric imaging (Figure 3F–H, Video 5). In particular, we could observe short lived filopodial buckling events (Figure 3—figure supplement 2).

High-resolution biological imaging.
(A) Endoplasmic reticulum in U2OS cells. Inset shows fine details in the dense, tubulated network. The lookup table was selected as it allows visualization of both bright and dim structures. (B) Vimentin in RPE hTERT cells. (C) Single slice through vimentin network. (D) Lateral and (E) axial view of clathrin-mediated endocytosis in ARPE cells. (F) Cortical blebs in MV3 melanoma cells. (G) Cross-section through MV3 cells at the 6th and (H) 12th time point. All data shown in this figure was deconvolved. All Scale Bars are 10 microns.
Endoplasmic reticulum dynamics in osteosarcoma U2OS cells expressing Sec61-GFP.
Time Interval: 0.84 s. Scale Bar: 10 microns.
3D stack of RPE hTERT cells expressing GFP-vimentin.
Data has been deconvolved and sheared into its proper Euclidian position. Scale Bar: 20 microns.
3D stack of ARPE cells tagged with AP2-GFP, a marker for clathrin-mediated endocytosis.
Data has been deconvolved and sheared into its proper Euclidian position. Scale Bar: 10 microns.
Maximum intensity projection of ARPE cells tagged with AP2-GFP, a marker for clathrin-mediated endocytosis.
Time Interval: 1.34 s. Scale Bar: 20 microns.
MV3 cells expressing the biorthogonal membrane marker, CAAX-Halo-Tag, labeled with Oregon Green.
Time Interval:1.09 seconcs, Scale Bar: 10 microns.
Natural killer cell mediated cytotoxicity
Natural Killer (NK) cells are one of the main components of the innate immune system and are able to directly recognize and destroy virally infected or oncogenically transformed cells via the formation of a multi-purpose interface called the immunological synapse with their target (Mace et al., 2014). An early landmark of this structure is the establishment of a complex actin scaffold (on the effector side) which formation (Mace and Orange, 2014) and dynamism (Carisey et al., 2018) are known to be critical to ensure the productive outcome of the cytotoxic process. By contrast, the contribution of the target cell in the establishment and maintenance of this unique structure, and the interplay between the plasma membranes of the two cells remains poorly understood, in part because most work has been performed on ligand-coated surfaces (Brown et al., 2011; Carisey et al., 2018). Much could be gained by imaging heterotypic cell-cell interactions directly, but this is particularly challenging as it requires a combination of resolution, speed, and field of view as these interactions are random and often short lived. Here, we imaged the formation of the cytolytic immunological synapse between a population of NK cells expressing a reporter for filamentous actin and myelogenous leukemia cancer cells expressing a fluorescent marker highlighting their plasma membrane (Figure 4A). Upon engaging with the cancer cell, the NK cell formed an immunological synapse with rapid actin accumulation in the plane of the immune synapse followed by the establishment of an actin retrograde flow along the cell axis in a similar fashion to what has been observed in T cells (Chen et al., 2014; Ritter et al., 2015). Interestingly, long tethers of membrane were pulled from the target cell synchronously with the actin retrograde flow (Figure 4B,C and D, Video 6). This could be a direct visualization of the first stage of trogocytosis, an important consequence of mechanotransduction at the immunological synapse that contributes to hypo-responsiveness in cytotoxic cells (Miner et al., 2015). Indeed, recent work performed on cytolytic T cells emphasized the crucial role of tension transmitted through the ligand-receptor axis on the organization of the underlying actin cytoskeleton (Kumari et al., 2020). Such observations highlight the importance of moving away from artificial ligand-coated surfaces, and to evaluate biological processes in more relevant contexts as enabled by high-resolution, high-speed, volumetric imaging.

Formation of an immunological synapse between a natural killer cell and a target cell.
(A). Subset of imaging field of view, showing a population of NK-92 natural killer cells expressing Life-Act-mScarlet and K562 leukemic cells expressing Lck-mVenus, which is myristolyated and localizes to the plasma membrane. Scale Bar: 20 microns. (B). K562 leukemic cell. (C) NK-92 natural killer cell. (D) Overlay of NK-92 and K562 cells during the formation and maturation stages of the immunological synapse. Scale Bar: 5 microns. (E) Upon formation of the synapse, centrifugal flows driven by the NK-92 cell results in displacement of the K562 cellular membrane highlighted by the presence of membrane tethers, visible in the left panel and emphasized in the color merged frame sequence (bottom). Data is shown as maximum intensity projection and was deconvolved. Time Interval: 11.33 s.
An NK-92 natural killer cell forming an immunological synapse with a target cell.
The NK-92 cell (Natural Killer cell line) was labeled with Life-Act-mScarlet, and is shown in orange. The target cell (K562 leukemia cell line) was labeled with Lck-mVenus and is shown in cyan. Time Interval:11.33 s. Scale Bar: 10 microns.
Imaging in biological microchannels – microtubules and nuclear shielding
Cells migrating through 3D microenvironments such as dense stromal tissues must navigate through tight pores between matrix fibers and are thus rate-limited by the cross-sectional diameter of their nucleus (Wolf et al., 2013). Understanding how cells adapt and effectively navigate these complex microenvironments is fundamental to multiple biological processes such as development, tissue homeostasis, wound healing and dysregulated in cancer cell invasion and metastasis. Our understanding of the mechanisms governing nuclear movement and protection in 3D environments is constantly emerging. For example, we now know that during 3D migration cells can tolerate and repair nuclear constriction events that cause nuclear herniation, rupture and DNA damage (Denais et al., 2016; Raab et al., 2016). Unfortunately, observing these events in vivo remains low throughput. In contrast, microfluidic devices provide an accurate and reproducible model of this phenomenon whereby cells can be subjected to precisely-defined mechanical constrictions with unique sizes and shapes that serve to recapitulate the biological microenvironment (Garcia-Arcos et al., 2019; Raab et al., 2016). Nonetheless, imaging this biological process three-dimensionally requires a large field of view, and both high spatial and temporal resolution. Unfortunately, confocal microscopes are accompanied by excessive phototoxicity to permit longitudinal observation of migration through confined microchannels. While light-sheet microscopy is an attractive alternative, the glass-polydimethylsiloxane sandwich geometry of microfluidic devices is not compatible with traditional LSFMs that use water-dipping objectives and an orthogonal illumination and detection geometry, let alone cantilever- or micromirror-based methods. Here, using our OPM, we were able to circumvent this and volumetrically image nuclear positioning and microtubule dynamics as cells navigated mechanical constrictions (Figure 5A, Video 7). Here, the microfluidic device consists of ~4 microns tall large and small circular posts with constriction junctions between large and small posts of 2.5 microns and two microns, respectively (Figure 5B). When cells were allowed to migrate within these microchannels, cells generated long, microtubule-rich protrusions, and migrated in a polarized manner. The nucleus was visibly compressed when viewed in the axial direction (Figure 5C) and appeared to be surrounded by microtubules on both the apical and basal surfaces of the nucleus (Figure 5D and E). Importantly, as these cells squeezed through the microfluidic device, they adopted particularly elongated morphologies (~80 microns) that otherwise would be challenging to observe if it were not for the large field of view of our OPM.

Cell migration through PDMS micro-confinement channels.
(A) 1205Lu metastatic melanoma cells endogenously expressing eGFP-α-tubulin tagged microtubules with CRISPR (green) and a nuclear-localizing red fluorescent protein (3XNLS-mScarlet-I; magenta), migrating through a PDMS microchannel device. (B) Schematic drawing and fluorescence image of microfluidic device filled with TRITC-Dextran, where cells migrate in the horizontal dimension and squeeze between the pillars (large pillars are separated by 2.5 microns, and small pillars are separated by two microns). The white arrow in the fluorescence image marks the migration direction of the cells and also the scan direction (Y) of the light-sheet. (C) Axial cross-section of cells in a microchannel device shows top-down nuclear confinement as cells migrate through 4-micron tall channels. (D) Microtubule protofilaments wrap around both the basal and apical surfaces of the cell when migrating through confined spaces. (E) Zoom of the region shown in C. All data with the exception of (B) was deconvolved. All scale Bars: 10 microns.
1250Lu metastatic melanoma cells expressing GFP-alpha-tubulin (green) and a 3XNLS-mScarlet-I nuclear marker (magenta).
Nuclei often undergo compression and rupture as cell migrate and squeeze through pillars. Time Interval: 29.88 s. Scale Bar: 10 microns.
High-speed imaging of calcium transduction and cytoplasmic flows
Three-dimensional imaging typically requires scanning heavy optical components (e.g. the sample or the objective), which limits the volumetric image acquisition rate. Nevertheless, because the OPM described here adopts a galvanometer-based scan-descan optical geometry, camera framerate-limited imaging is possible (Kumar et al., 2018; Voleti et al., 2019; Yang et al., 2019). Thus, we sought to image fast biological processes, including calcium wave propagation and the rapid diffusion of cytoplasmic tracers. For the former, we used the small-molecule calcium sensor Fluo-3, and imaged rat primary cardiomyocytes at a volumetric image acquisition rate of 10.4 Hz (Figure 6A, Video 8). Here, imaging was sufficiently fast to observe calcium translocation during spontaneous cardiomyocyte contraction. Such imaging can improve the understanding of single-cell calcium waves which is important for cardiac physiology and disease (Gilbert et al., 2020). Likewise, we also evaluated the rheological properties of the cytoplasm by imaging in live cells genetically encoded multimeric nanoparticles. These nanoparticles appeared as near-diffraction-limited puncta that rapidly diffused and thus served as inert cytoplasmic tracers, and which we were able to image and track (Jaqaman et al., 2008) over 100 time points at a volumetric image acquisition rate of 13.7 Hz (Figure 6B, Video 9). Such tracking can help us understand how diffusion varies for different morphological domains, as well as how phenotypic changes alter cellular mechanics (Delarue et al., 2018; Hannezo and Heisenberg, 2019). Nonetheless, these questions can only be answered if equipped with the volumetric image acquisition speeds demonstrated here.

High-speed volumetric imaging of calcium waves and genetically encoded multimeric nanoparticles.
(A) Primary rat cardiomyocytes were labeled with the small-molecule sensor, Fluo-3, and imaged at 10.4 Hz. Deconvolved data is shown. Scale Bar: 10 microns. (B) Imaging rheological tracers in the mammalian cytosol at 13.7 Hz. Deconvolved data is shown. Scale Bar: 10 microns.
3D rendering of primary cardiomyocyte stained with Fluo-3, a small-molecule sensor for calcium (II).
Green indicates fluctuations in intracellular calcium levels, and gray represents the cell boundary (calculated as an average of all imaging frames). Time Interval: 0.096 s. Scale Bar: 10 microns.
Orthogonal maximum intensity projections of MV3 cells expressing cytosolic GEMs as rheological tracers.
Particles were detected and tracked with the uTrack-3D software package. Time Interval: 0.073 s.
Simultaneous volumetric imaging and optogenetic stimulation
In addition to its effects on proliferation and survival (Mohan et al., 2019), Rac1 also drives changes in cell shape and migration. Here, leveraging our ability to perform simultaneous volumetric imaging and optical stimulation (Figure 7A and B, Video 10), we deployed a photoactivatable variant of Rac1 (PA-Rac1) in mouse embryonic fibroblasts (Wu et al., 2009). Subcellular optical stimulation of cells expressing PA-Rac1 resulted in large-scale dorsal ruffles that propagated from the cell edge near the activation region toward the cell nucleus (Figure 7C and D, Videos 11 and 12). Importantly, such dorsal ruffles would not be visible unless imaged volumetrically with high spatiotemproral resolution. By analyzing the protrusion dynamics in multiple cells (See Materials and methods), those expressing PA-Rac1 (N = 7) showed statistically significant increases in protrusion speed (p=0.04) and protrusion duration (p=0.02), but not the frequency of protrusion-retraction dynamics, upon photoactivation. Importantly, optically stimulated control cells (N = 6) did not show any significant change in protrusion speed, duration, or frequency in response to photoactivation (Figure 7E).

Simultaneous subcellular optogenetic stimulation of PA-Rac1 and volumetric imaging of morphodynamic changes in MEF cells.
(A) Cell before optogenetic stimulation. (B) Localized optical stimulation of PA-Rac1 (within the blue box) was performed with a 488 nm laser operating in a laser-scanned illumination geometry synchronously with volumetric imaging using a 561 nm laser. Scale Bar: 10 microns. (C) Lateral maximum intensity projection of the cell during optical stimulation shows the dorsal ruffles moving from the cell periphery to the juxtanuclear cellular region. Scale Bar: 20 microns. (D) Orthogonal maximum intensity projection along the dotted line in (C) of dorsal ruffles. Scale Bar: 10 microns. (E) Hidden Markov model analysis gives the log-ratio difference between pre activation and activation response showing control cells (N = 6) with no difference in protrusion duration, speed, or frequency while cells expressing PA-Rac1 (N = 7) show statistically significant increases in protrusion speed (p=0.04) and duration (p=0.02) with no significant changes in frequency. All image data shown are raw.
Animation of light-sheet scanning and optogenetic activation.
Blue rays represent the light-sheet illumination, and green rays indicate the near-diffraction-limited epi-illumination. Both beams are scanned with a mirror galvanometer (bottom) that is conjugate to the back focal plane of the primary objective (top).
Mouse embryonic fibroblasts expressing PA-Rac1 and mCherry.
Movie shows 15 min in the absence of optical stimulation followed by 15 min of optical stimulation. The blue rectangle shows region undergoing optical stimulation with 488 nm light. An axial cross-section of the region in green is shown at the bottom. Scale Bar: 10 microns. Time Interval: 10 s.
MEF PA-Rac1-mCherry cell movie same as Video 11 show cell dynamics with Imaris 3D rendering.
Time Interval: 10 s. Scale Bar: 10 microns.
Large field of view imaging of cortical neurons and ventral furrow formation
Neurons rapidly transduce action potentials across large spatial distances via their axonal or dendritic arbors, respectively. However, many neuronal features, including synaptic boutons, are sub-micron in scale. Thus, imaging neuronal processes requires a combination of field of view, resolution, and speed (Figure 8A). Here, we imaged cortical neurons expressing GCaMP6f at a volumetric imaging speed of 7 Hz, and readily visualized both small scale morphological features as well as rapid action potentials (Video 13). Another field that benefits from fast imaging of large volumes is developmental biology, which aims to longitudinally track cell fate throughout each stage of embryological development. Indeed, many developmental programs, including ventral furrow ingression in Drosophila, are inherently three-dimensional as cells are rapidly internalized on the timescale of a few minutes along the anteroposterior axis of the embryo which spans ~230 microns. Here, we imaged ventral furrow ingression in a stage 6 Drosophila embryo (Figure 8B, Figure 8—figure supplement 1, Video 14). The formation of the ventral furrow, which is seen as a prominent groove running along the anteroposterior axis of the embryo is clearly visible. As ventral furrow formation completes, germ band extension is initiated, and cells move toward the ventral midline and intercalate between one another (Blankenship et al., 2006). This global cell movement causes the tissue to elongate, which is immediately followed by rapid cellular rounding and mitotic events.

Large-scale imaging of cultured neurons and drosophila embryo gastrulation.
(A) Cultured cortical neurons expressing the Ca2+ biosensor GCaMP6f. Maximum intensity projection of deconvolved data. (B) Stage 6 Drosophila embryo expressing gap43-mCherry. Scale Bar: 20 microns. Maximum intensity projection of deconvolved data is shown in both (A) and (B).
Dissociated cortical neurons expressing the Ca2+ biosensor GCaMP6f, imaged volumetrically at 7 Hz.
Scale Bar: 20 microns.
Stage 6 Drosophila embryo undergoing gastrulation.
Ventral furrow ingression occurs along the anteroposterior axis of the embryo and is immediately followed by rapid epithelial mitotic events. Time Interval: 23 s.
Tissue-scale imaging
In addition to the rapid laser scan/descan illumination geometry, OPM is also compatible with a sample scanning acquisition format that is essentially field of view unlimited. Indeed, by combining scan optimized equipment with fully automated fluidic handling, it is possible to image ~1 cm2 of a thin tissue in less than 45 min per color and perform biochemistry, such as sequential multiplexed labeling. To demonstrate this, we imaged an entire 30-micron thick slice of coronal mouse brain tissue (Figure 9A, Video 15) labeled with the nuclear marker DAPI. Within these data, even small features like nucleoli are clearly resolved from both lateral and axial viewing perspectives throughout the entire ~6×8 mm tissue slice (Figure 9B and C). Likewise, we also imaged a ~ 4×14 mm slice of 12-micron thick human lung tissue labeled for nuclei, angiotension-converting enzyme 2 (ACE2) mRNA, and surfactant protein C (SFTPC) protein (Figure 9D). Here, characteristic histological features, including bronchiole, alveoli and vasculature, are readily visible, albeit with molecular contrast and sub-cellular resolution (Figure 9E,F and G, and Video 16). Quantification of molecular expression within this tissue section provides spatial information on ~20,000 cells, and verifies our previous limited quantification of ACE2 expression in alveolar epithelial type II cells using confocal microscopy (Muus and Luecken, 2020). Indeed, because we were not sterically restricted by the orthogonal illumination and detection geometry (Figure 9—figure supplement 1), the lateral dimensions of this human lung specimen were 8- and 1.5-fold larger than those of the biggest sample imaged with lattice light-sheet microscopy (Gao et al., 2019). However, in the third dimension, lattice light-sheet microscopy has in principle a 6.7x larger reach (2 mm working distance of the typically employed NA 1.1/25X detection objective compared to 300 microns working distance of our primary objective). In practice, optical aberrations limit high-resolution light-sheet microscopy to depths of a few hundreds of microns, even for highly transparent samples. Furthermore, our approach is fully compatible with automated fluid exchange, which is increasingly important for projects like the Human Cell Atlas that necessitate iterative imaging approaches for spatial -omics of RNAs and proteins at the single-cell level throughout entire tissues (Chen et al., 2015).

Tissue-scale imaging.
(A) Maximum intensity projection of fused raw image for 30 um thick mouse brain tissue labeled for nuclei Scale Bar: 2 mm. (B) Individual raw XY slice of individual nuclei. Scale Bar: 250 microns. (C) Individual raw YZ slice of data in (B). Scale Bar: 250 microns. (D) Maximum projection of fused raw image for 15 um thick human lung tissue labeled for nuclei (magenta), SFTPC protein (cyan), ACE2 mRNA (orange). Scale Bar: 2 mm. (E) Maximum projection of raw data for yellow box in (A). Scale Bar: 0.5 mm. (F) Maximum projection of raw data for blue box in (A). Scale Bar: 75 microns. (G) 3D rendering of raw data for green box in (A). All data was deconvolved. Scale Bar: 25 microns.
Z-stack of raw data for 30-micron thick slice of coronal mouse brain tissue.
Sub-nuclear features such as nucleoli, are readily evident owing to the high-resolution and optical sectioning of the OPM.
Volumetric rendering of deconvolved data for human lung tissue as imaged with the stage-scanning variant of the OPM, which permits imaging of cm-scale objects.
Nuclei (magenta) and SFTPC protein (cyan). SFTPC is a commonly used cytosolic marker of alveolar type 2 cells, which is resolved with subcellular resolution in 3D. Scale Bar: 10 microns.
Discussion
High-resolution light-sheet microscopy has yet to be widely adopted in biological laboratories and core facilities owing to routine problems with sample drift, contamination, and lack of user friendliness. Here, we show that an OPM with customized optics overcomes these challenges, and combines the ease of traditional sample mounting, environment maintenance, and multi-position stage control with the gentle, subcellular imaging afforded by selective plane illumination. For example, cells were easily identified in a traditional epi-fluorescence format prior to volumetric imaging in the light-sheet mode, the focus was maintained with readily available hardware solutions, and the environment remained sterile with CO2, humidity, and temperature control. In addition to its ease of use, the OPM described here delivers spatial resolution that is on par with lattice light-sheet microscopy in its most commonly used square lattice illumination mode (Chen et al., 2014), albeit with a larger field of view and a volumetric imaging speed that is only limited by the maximum camera framerate and the emitted fluorescence photon flux. And unlike state-of-the-art multiview LSFM techniques that achieve a better axial resolution after image fusion and deconvolution, only a single imaging perspective is needed (Guo et al., 2020; Wu et al., 2013). Higher axial resolution can be achieved with Axially Swept Light-Sheet Microscopy (Dean et al., 2015) or the less commonly used hexagonal and structured illumination modes of lattice light-sheet microscopy. However, this requires a separate illumination objective, which introduces steric limitations, and is accompanied by additional drawbacks that include a shorter effective exposure time in Axially Swept Light-Sheet Microscopy, increased amounts of out-of-focus blur for hexagonal lattices, or the acquisition of five images per plane for structured illumination.
This is in stark contrast to previous generations of OPM (Bouchard et al., 2015; Dunsby, 2008; Kumar et al., 2011), which provided only moderate spatial resolution or a limited field of view. This revolution in OPM performance was triggered by the insight that one could combine a high-NA (~0.9) air secondary objective with a high-NA (~1.0) water-immersion tertiary objective (Yang et al., 2019). In such a system, the refractive index interface between the two objectives compresses and refracts the optical cone of light toward the tertiary objective and improves both sensitivity and resolution of the entire imaging system. In this work, we take this concept to its extreme with an optimized optical train, and we replace the tertiary objective (and its water chamber and coverslip) with a solid immersion objective that eases alignment and more efficiently compresses and refracts the optical cone of light (Millett-Sikking and York, 2019). We characterize the performance of this system and demonstrate that it has a lateral and axial resolving power that is similar to or better than many LSFM systems. We note that the theoretical NA of 1.28 would in principle allow even higher spatial resolution. Indeed, when using a zero-tilt angle for the tertiary objective, the system routinely delivered 270 nm scale raw lateral resolution across its field of view. This indicates that with reduced tilt angles, even higher resolution as demonstrated here should be possible. Why the resolution dropped notably in the tilt direction in this work is still under ongoing investigation.
As the volumetric image acquisition rate is not limited by piezoelectric scanning of either the sample or the objective, but rather a high-speed galvanometric mirror, very high temporal resolution is possible. Here, we demonstrated volumetric subcellular imaging at rates of 10 Hz, which permitted tracking of intracellular flows and calcium propagation. To demonstrate its utility, we performed a variety of imaging tasks that would be hard or impossible to perform on a traditional LSFM, including imaging in a microfluidic channel and the reproducible subcellular optogenetic activation of Rac1. Indeed, as shown by imaging cm-scale tissue slices, the stage-scanning variant of this OPM also permits applications that would otherwise not be feasible with comparable high-resolution light-sheet platforms.
As such, we believe that this OPM could displace laser-scanning and spinning-disk confocal microscopes as the workhorse of cell biology in both individual labs and user facilities. Importantly, its design allows integration into existing epi-fluorescence frameworks, which decreases the cost of building such an instrument. Furthermore, it combines spatial resolution of a spinning-disk microscope (Figure 2—figure supplement 3) with high volumetric acquisition speed and low phototoxicity afforded by LSFM. It provides comparable spatial resolution to leading light-sheet technologies, including the square illumination mode of lattice light-sheet microscopy, but with simpler sample handling, maintenance of a sterile environment, the ability to perform simultaneous multicolor imaging (Chang et al., 2019), and an essentially unlimited field of view in a sample scanning format. The principle drawback that is inherent to OPM systems is the reduced collection efficiency, which necessarily results from the large number of optics necessary to reorient the fluorescence emission (Kim et al., 2019). While these losses are non-negligible, they are in part offset by the higher overall NA of our OPM system relative to other light-sheet microscopes (Appendix 3), For example, in the absence of aberrations, and assuming a lower bound for the overall NA of 1.2 to collect at best ~1.35 and~2.56 times more photons than a NA 1.1 or NA 0.8 objective, respectively. Consequently, for a given laser power, we achieve image contrast and rates of photobleaching comparable to lattice light-sheet microscopy (Figure 9—figure supplement 2). And lastly, we reported only on the highest resolution OPM variant that incorporates the bespoke glass-tipped tertiary objective. Indeed, many different variants of OPMs that operate across a range of NAs (1.0–1.35) and magnifications (20 - 100X) have already been designed and numerically evaluated. Thus, we consider this the opening prelude to what may be the next generation of user-friendly, and broadly accessible LSFMs.
Materials and methods
Laser-scanning microscope setup
Request a detailed protocolThe entire microscope was built in a basic inverted geometry (RAMM-FULL, Applied Scientific Instrumentation) with a three-axis motorized stage controller. Two solid-state continuous wave lasers (Sapphire-488–200 and Sapphire-568–100, Coherent) and a continuous wave fiber laser (VFL-P-300–642-OEM1, MPB Communications) were independently attenuated with logarithmically spaced neutral density filter wheels, optically shuttered (VMM-D3, and LSS6T2, Uniblitz), and combined with dichroic mirrors (LM01-552-25 and LM01-613-25, Semrock). Much of this equipment was collected from a decommissioned OMX system, and future variants of the OPM microscope could include faster laser switching devices (e.g. solid-state lasers or acousto-optic devices). After the laser combining dichroic mirrors, the beams were focused through a 30-micron pinhole (P30D, ThorLabs) with a 50 mm achromatic doublet (AC254-050-A, ThorLabs) and recollimated thereafter with a 100 mm achromatic doublet (AC254-100-A, ThorLabs). Laser polarization was controlled with a half waveplate (AHWP3, Bolder Vision Optik) that was secured in a rotation mount (RSP1 × 15, ThorLabs). The beam was then either reflected with a motorized flipper mirror (8892 K, Newport) toward the epi-illumination (for alignment) or laser spot (for optogenetics) path or transmitted toward the light-sheet illumination path.
For the light-sheet path (Figure 1—figure supplement 4), the light was first expanded in one dimension with a pair of cylindrical lenses (F = 25 mm, #68–160 Edmund Optics and ACY254-100-A, ThorLabs), and then focused into a 1D Gaussian profile using a cylindrical lens (ACY254-50-A, ThorLabs) onto a resonant galvo (CRS 12 kHz, Cambridge Technology) to reduce stripe artifacts by rapidly pivoting the light-sheet in sample space (Huisken and Stainier, 2007; Figure 1—figure supplement 3). Such a 1D light-sheet offers 100% spatial duty cycle, which reduces phototoxicity and photobleaching compared to light-sheets that are obtained by laterally scanning a 2D laser focus. An adjustable slit was placed at one focal distance in front of the cylindrical lens. This slit was used to adjust the effective NA of the light-sheet, which determines the light-sheet thickness and propagation length. For most experiments, the NA of the light-sheet was set to ~0.2, which creates a light-sheet with a Rayleigh length of about 21 microns. Opening the slit allows increasing the NA to 0.34, which is the practical limit for the chosen objective and inclination angle. The light was then relayed with a 100 mm achromatic doublet (AC254-100-A, ThorLabs) over a polarizing beam splitter (PBS251, ThorLabs), through a quarter wave-plate (AQWP3, Bolder Vision Optik) onto mirror galvanometer (6215H, Cambridge Technology), which backreflects the light through the same quarter wave-plate and polarizing beam splitter toward a multi-edge dichroic (Di03-R405/4888/561/635-t3−25 × 36, Semrock) that reflected the light toward the primary objective. The galvanometer mirror allows control of the lateral positioning of the light-sheet. After the dichroic, the light was focused with a 200 mm tube lens (TTL200, ThorLabs), recollimated with a 39 mm scan lens (LSM03-VIS, ThorLabs), reflected off of a 1D galvanometer mirror (6215H, Cambridge Technology), focused by a 70 mm scan lens (CSL-SL, ThorLabs), recollimated with a 200 mm tube lens (TTL200, ThorLabs), and imaged into the specimen with the primary objective (100X/1.35 MRD73950 Silicone Immersion Objective, Nikon Instruments). For detection (Figure 1—figure supplement 7), the fluorescence was descanned with the galvanometer mirror, transmitted through the multi-edge dichroic, and an image is formed by the secondary objective (CFI Plan Apo Lambda 40XC, Nikon Instruments), which is detected by the tertiary objective (AMS-AGY v1.0, Special Optics) and focused with a tube lens (ITL200, ThorLabs) onto a sCMOS camera (Flash 4.0 v3, Hamamatsu). Each imaging channel was collected sequentially (e.g. after a complete Z-stack), and the fluorescence was spectrally isolated with emission filters placed in a motorized filter wheel (FG-LB10-B and FG-LB10-NWE, Sutter Instruments). Detailed imaging parameters are listed in Supplementary file 1 The scan galvanometer and resonant galvanometer were driven by a 28V (A28H1100M, Acopian) and a 12V power supply, respectively.
For widefield illumination (Figure 1—figure supplement 5), after being reflected by the flipper mirror, the light was focused onto a galvanometer mirror (6215H, Cambridge Technology) with an achromatic doublet (AC254-100-A, ThorLabs) that itself was mounted on a flipper mirror (8892 K, Newport). For laser spot illumination (Figure 1—figure supplement 6), the achromatic doublet was removed from the optical path, resulting in a collimated beam on the galvanometer mirror. By controlling this galvanometric mirror synchronously with the Z-galvanometer, the cell can be illuminated with arbitrary 2D patterns of light, which enables optogenetic stimulation, photoactivation, or fluorescence recovery after photobleaching. Thereafter, the epi-illumination and laser spot illumination paths were reflected off the aforementioned polarizing beam splitter toward the multi-edge dichroic. The widefield illumination path proved useful for identifying interesting cells and for focusing the microscope, and the fluorescence was detected after the first tube lens with flipper-mounted dichroic mirror (Di02-R488-t3−25 × 36, Semrock) onto a sCMOS camera (Grasshopper 3, FLIR). The X, Y, and Z-resolution of the primary objective when imaged in a traditional widefield format was 243 ± 11 nm, 242 ± 15 nm, and 604 ± 31 nm, respectively.
The data acquisition computer was a Colfax International ProEdge SXT9800 Workstation equipped with two Intel Xeon Silver 4112 processors operating at 2.6 GHz with 8 cores and 16 threads, 96 GB of 2.667 GHz DDR4 RAM, a Intel DC P3100 1024 GB M.2 NVMe drive, and a Micron 5200 ECO 7680 GB hard-drive for file storage. All software was developed using a 64-bit version of LabView 2016 equipped with the LabView Run-Time Engine, Vision Development Module, Vision Run-Time Module and all appropriate device drivers, including NI-RIO Drivers (National Instruments). Software communicated with the camera (Flash 4.0, Hamamatsu) via the DCAM-API for the Active Silicon Firebird frame-grabber and delivered a series of deterministic TTL triggers with a field programmable gate array (PCIe 7852R, National Instruments). These triggers included control of the optical shutters, galvanometer mirror scanning, camera fire and external trigger. The control software can be requested from the corresponding authors and will be distributed under an MTA with the University of Texas Southwestern Medical Center.
Stage scanning microscope setup
Request a detailed protocolThe stage-scanning oblique plane microscope was built using an inverted geometry with a three-axis motorized stage with a constant scan speed optimized X stage (FTP-2000, Applied Scientific Instrumentation). Five solid-state continuous wave lasers (OBIS LX 405–100, OBIS LX 488–150, OBIS LS 561–150, OBIS LX 637–140, and OBIS LX 730–30, Coherent Inc) contained within a control box (Laser Box: OBIS, Coherent Inc) were combined with dichroic mirrors (zt405rdc-UF1, zt488rdc-UF1, zt561rdc-UF1, zt640rdc-UF1, Chroma Technology Corporation). After the laser combining dichroic mirrors, the beams were focused through a 30-micron pinhole (P30D, Thorlabs) with a 30 mm achromatic doublet (AC254-030-A, Thorlabs), recollimated with a 100 mm achromatic doublet (AC508-100-A, Thorlabs), and steered through an adjustable iris. The adjustable iris was used to control the diameter of the laser beam and the light-sheet NA at the sample. Light passed through an electrotunable lens (EL10-30-C, Optotune) placed horizontally and relayed by two 100 mm achromatic doublets (AC254-100-A, Thorlabs) onto a 1-axis galvanometer mirror (GVS201, Thorlabs). The galvanometer mirror was placed in the back focal plane of a 300 mm achromatic doublet (AC508-300-A, Thorlabs). The line focus formed by pivoting the galvanometer mirror was formed on a 2-inch mirror that was used to control the light-sheet tilt at the sample plane. This mirror was placed in the back focal plane of a 180 mm achromatic doublet (AC508-180-A). This lens was placed such that the galvanometer mirror rotation was relayed to the back focal plane of the primary objective (100X/1.35 MRD73950 Silicone Immersion Objective, Nikon Instruments). Excitation light was reflected off a pentaband dichroic mirror (zt405/488/561/640/730rpc-uf3, Chroma Technology Corporation) and imaged into the specimen with the primary objective.
For detection, fluorescence was transmitted through the pentaband dichroic mirror, then transmitted through a 200 mm tube lens (MXA22018, Nikon Instruments), passed through an empty kinematic mirror cube (DFM1B, Thorlabs), a 357 mm tube lens assembly (AC508-500-A and AC508-750-A, Thorlabs), and an image is formed by the secondary objective (CFI Plan Apo Lambda 40XC, Nikon Instruments), which is detected by the tertiary objective (AMS-AGY v1.0, Special Optics) and focused with a tube lens (MXA22018, Nikon Instruments) onto a sCMOS camera (Prime BSI Express, Teledyne Photometrics). Each imaging channel was collected sequentially (e.g. after one complete strip scan of the stage), and laser light was blocked by two identical pentaband barrier filters (zet405/488/561/640/730 m, Chroma Technology Corporation), with one placed in infinity space before secondary objective and one in infinity space after the tertiary objective. The kinematic mirror cube after the primary tube lens was used to redirect light to either an inexpensive CMOS camera placed at the primary image plane (BFS-U3-200S6M-C, FLIR) or a wavefront sensor (HASO-VIS, Imagine Optic) to characterize the wavefront after the primary objective.
Acquisition was performed on a Windows 10 64-bit computer (Intel i7-9700K, 64 Gb memory, 12 TB SSD raid 0 array, Nvidia RTX 2060 GPU card) connected via 10 Gbps optical fiber to a network attached storage (DS3018XS, Synology), Data was acquired by scanning the scan optimized stage axis at a constant speed with the camera set to ‘Trigger First’ mode and triggered to start by the stage controller (Tiger, Applied Scientific Instrumentation) when the stage passed the user defined start point. The scan speed was adjusted so that the displacement between exposures was either 100 or 200 nm. This slow scan speed ensured that minimal motion artifacts occurred during stage scanning. The ‘Exposure Out’ trigger from the camera triggered one sweep of the galvanometer mirror across the FOV using a homebuilt data acquisition system (Teensy 3.5, PJRC and Power DAC module, Visgence, Inc). The camera chip was cropped to the area of interest containing the sample and data was saved as one TIFF file per image. Multiple laser lines are acquired sequentially by allowing the stage to reset to the original position and repeat the scan. A custom script in Micromanager 2.0 gamma sets the stage parameters and controls the scan. The stage-scanning post-processing control codes are available via the Quantitative Imaging and Inference Laboratory GitHub repository (http://www.github.com/QI2lab/OPM).
Transmission measurements
Request a detailed protocolTo evaluate the transmission efficiency, we measured the light throughput of the optical system with a 543 nm HeNe laser in transmission at both 0- and 30-degree tilts. Here, the diameter of the alignment laser was set to the size of the primary objective back pupil, with all of the optics and filters present in the optical path with the exception of the primary objective and the camera. When oriented at 30 degrees, the laser and stage-scanning variants had a 41% and 53% transmission, respectively, and at 0 degrees, we observed a 3% increase in transmission. 12% of the losses could be attributed to the scan lens, mirror galvanometer, tube lens combination, which can be eliminated with lens-free scanning (Boden and Volpato, 2020). Nonetheless, the collection efficiency of both variants is greater than OPM designs that use beam splitters in their detection path (Kim et al., 2019).
General alignment
Request a detailed protocolIn an effort to assist in the adoption of both the laser-scanning and stage-scanning OPM technologies described here, we provide a detailed discussion on how to align such systems in Appendix 4. Furthermore, a complete parts list for these microscopes, as well as other variants based on OPM, can be found in the work by Millett-Sikking and colleagues (Millett-Sikking and York, 2019).
Data post-processing
Request a detailed protocolFor the laser-scanning system, analysis was performed on the local BioHPC high-performance computing cluster. Data was sheared with Python using a script originally developed by Dr. Bin Yang which applies the Fourier Shift Theorem. In instances where deconvolution was used, the raw data was deconvolved in a blind fashion where the experimentally measured PSF served as a prior. Of note, for the resolution measurements, we resampled the image data by a factor of 2 prior to deconvolution, reducing the lateral pixel size from 115 nm to 57.5 nm. As evidenced by the FRC measurements, this corresponds to zero-padding of the Fourier transform of the data, which allows the iterative deconvolution algorithm to reconstruct slightly out-of-band information (Heintzmann, 2007). For most of the biological data, this resampling was not performed, as the data size would have become limiting. Leaving the data sampled laterally at 115 nm pixel size results in a highest possible Nyquist limited resolution after deconvolution of 230 nm.
Both of these data post-processing functions are available via the AdvancedImagingUTSW GitHub repository (https://github.com/AdvancedImagingUTSW). As an example, shearing and deconvolution of a 2048 × 256×450 voxel image took ~5 and~120 s, respectively. Nonetheless, we believe that with GPU computing, real-time processing may be a possibility. For rotation into the traditional epi-fluorescence-like orientation, the freely available IMOD software package was used (Kremer et al., 1996). For the stage-scanning system, analysis was performed on an Ubuntu 20.04 LTS computer (2x Intel E5-2600, 128 Gb memory, 16 TB SSD RAID 0 array, Nvidia Titan RTX card) connected via 10 Gbps optical fiber to the same network attached storage as the acquisition computer. First, a retrospective flat-field was calculated per channel (Peng et al., 2017). Using Numpy and Numba libraries in Python (van der Walt et al., 2011), data was split into tiles that fit in local memory, flat-fielded, orthogonally interpolated to deskew the stage scan (Maioli, 2017), saved as a BigDataViewer H5 file (Pietzsch et al., 2015), and stitched using default settings in BigStitcher (Hörl et al., 2019). The stage-scanning post-processing codes are available via the Quantitative Imaging and Inference Laboratory GitHub repository (http://www.github.com/QI2lab/OPM).
Analysis of optogenetic responses
Request a detailed protocolTo evaluate the response of MEFs to blue light activation in PA-Rac1 and control cells we used an algorithm previously described to assess cell morphodynamics (Welf et al., 2019). Briefly, maximum intensity projection images of cells (mCherry) were pre-processed in Fiji via simple ratio bleach correction and a median filter with a 230 nm pixel size. The cell edge was then automatically detected, and small regions of interest (windows) of size 920 nm x 920 nm were placed around the periphery of the cell to assess cell edge velocity over time. The velocity profile was then analyzed with a Hidden Markov Model defining different protrusive states (Welf et al., 2019), which were then used to assess protrusion speed, frequency, and duration. The Hidden Markov Model analysis was done using R package ‘depmixS4’ (Visser and Speekenbrink, 2010). Here, we quantified the measures as a function of cell type (control vs. optogenetics cells). Log-ratio was used to quantify change via the difference in protrusion parameters before and during activation (e.g. ). The protrusion parameters were then assessed for difference between pre-activation and activation using a t-test for MEF control (N = 6) and PA-Rac1 (N = 7).
Cell lines, Plasmids and Transfection
Request a detailed protocolNK-92 cells were obtained from ATCC (CRL-2407) and maintained in alpha minimum modified Eagle medium, 0.2 mM myoinositol, 0.1 mM beta-mercaptoethanol, 0.02 mM folic acid, 12.5% heat-inactivated horse serum, 12.5% heat-inactivated FBS (Sigma–Aldrich), 2 mM L-glutamine and non-essential amino acids (ThermoFisher Scientific), supplemented with 100 U/mL Il-2 (Roche). K562 cells were obtained from ATCC (CCL-243) and cultivated in RPMI medium with high glucose, supplemented with 10% of heat-inactivated FBS (Sigma–Aldrich), 2 mM L-glutamine and non-essential amino acids (ThermoFisher Scientific). Both NK-92 and K562 cells were maintained in 37°C, 5% CO2 tissue culture incubators and routinely confirmed to be mycoplasma negative using LookOut mycoplasma PCR detection kit (Sigma–Aldrich). pLifeAct-mScarlet-N1 and pLck-mVenus-C1 were gifts from Dorus Gadella (Addgene plasmids #85054 and #84337) and transfected by nucleofection using Amaxa Kit R per manufacturer’s instructions (Lonza). Positive cells were amplified under antibiotic selection pressure and sorted for low or intermediate expression level of the fluorescently tagged protein on an Aria II Fluorescence Activated Cell Sorter (BD). Each sorted population was then used for pilot experiments to determine the lowest possible expression level required for optimal imaging conditions. NK-92 cells expressing Life-Act-mScarlet and K562 cells expressing Lck-mVenus were mixed 1:1 and briefly spun down and resuspended in full prewarmed medium. The cell conjugates were then seeded in #1.5 uncoated glass bottom dishes and imaged as soon as possible for up to 45 min post mixing. All imaging was performed at 37°C using pre-warmed media and a stage top insert enclosing the sample.
To stably express EGFP-Sec61b in U2OS cells, a TET-inducible EGFP-Sec61b fusion was knocked into the AAVS1 safe harbor locus (Qian et al., 2014). The homologous recombination donor (HRD) was generated by EGFP tag and Sec61b cDNA (a gift from Dr. Jennifer Lippincott-Schwartz, Addgene #90992) into AAVS1-TRE3G-EGFP (a gift from Su-Chun Zhang, Addgene #52343), which was linearized with MluI and SalI and used for Gibson assembly. U2OS cells were cultured in RPMI (Gibco, A4192301) media supplemented with 10% FBS and Pen/Strep at 37°C, 5% CO2. For AAVS1 locus knock-in, 1 µg of HRD plasmid and 1 µg of AAVS1 T2 CRISPR plasmid (a gift from Masato Kanemaki, Addgene plasmid #72833) were transfected into U2OS cells using FuGENE HD (Promega) according to the manufacturer’s instructions (Natsume et al., 2016). 48 hr after transfection selection was initiated with 1 µg/mL of puromycin. The FBS used was not tetracycline free, which resulted in sufficient EGFP-Sec61b expression without the addition of doxycycline.
The human melanoma cell line1205Lu (a gift from Dr. Meenhard Herlyn, Wistar Institute) was genotypically characterized as previously reported (Smalley et al., 2007a; Smalley et al., 2007b) and grown in high glucose DMEM (11965092, Gibco, Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (10100147, Gibco, Thermo Fisher Scientific), 1x MEM Non-Essential Amino Acids (11140050, Gibco, Thermo Fisher Scientific) and 50 U/mL penicillin/50 µg/mL streptomycin (15070063, Gibco, ThermoFisher Scientific). pLenti CMV Hygro 3NLS-mScarlet-I was generated by Gateway Gene Cloning (Invitrogen, Thermo Fisher Scientific). First, 3XNLS-mScarlet-I (a gift from Dr. Dorus Gadella. Addgene #98816) was amplified by PCR (M0515, Q5 Hot start, New England Biolabs) to introduce 5’ EcoRI and 3’ XbaI restriction enzyme sites flanking either ends of 3XNLS mScarlet-I sequence (Bindels et al., 2017; Chertkova et al., 2020). The entry vector pENTR1A-GFP-N2 (FR1) (a gift from Dr. Eric Campeau and Dr. Paul Kaufman, Addgene #19364) along with the purified PCR fragment was digested with EcoRI and XbaI then ligated together with T4 DNA ligase (M0202, New England Biolabs). pENTR1a-3XNLS-mScarlet-I was then recombined using Gateway LR Clonase II (11791, Invitrogen, Thermo Fisher Scientific) as per the manufacturer’s instructions into pLenti CMV Hygro DEST (W117-1, a gift from Dr. Eric Campeau and Dr. Paul Kaufman, Addgene #17454) to create the final vector which was sequence verified by Sanger sequencing (Australian Genome Research Facility). CRISPR-CAS9 Endogenous EGFP tagging of the TUB1AB genetic locus in 1205Lu cells was performed as previously described (Khan et al., 2017) with the following modifications. Cells were transfected with 3.5 µg of each CAS-9-guide and EGFP donor knock-in vector using Lipofectamine 2000 (11668019, Invitrogen, Thermo Fisher Scientific) for 6 hr before replacing media to allow cells to recover. To achieve a uniformly fluorescent population of cells, cells were sorted using a MoFlo Astrios EQ cell sorter (Beckman Coulter) to isolate cells with dual-positive expression profile of EGFP and mScarlet-I using a gating strategy isolating the top ~20% of GFP and an intermediate ~30–60% population of mScarlet-I. To label the nucleus, 1205Lu cells were lentivirally transduced to stably overexpress 3XNLS-mScarlet-I as previously published (Coleman et al., 2003). After successful transduction, cells were grown in media containing 0.1 mg/mL Hygromycin (H3274, Roche).
MV3 cells were obtained from Peter Friedl (MD Anderson Cancer Center, Houston TX), and cultured in DMEM (Gibco) supplemented with 10% FBS (ThermoFisher) in 37°C and 5% CO2. MV3 expressing genetically encoded multimeric nanoparticles (GEMs) were made by infection with lentiviral construct from Addgene (Plasmid #116934) (Delarue et al., 2018). Cells were FACS sorted to purify a population of cells expressing T-Sapphire GEMs. The membrane marker expressed in MV3 cells was created from the membrane targeting (i.e. CAAX) domain of KRas fused to the HALO tag and cloned into the pLVX vector. Cells were infected with virus containing this construct and selected for expression by G418 resistance.
Mouse embryonic fibroblast cells (MEFs) were obtained from ATCC (MEF CF-1, SCRC-1040) and cultured in DMEM (Gibco) supplemented with 10% fetal bovine serum (ThermoFisher) at 37°C and 5% CO2. MEFs we infected with retrovirus encoding PA-Rac1 (pBabe-TetCMV-puro-mCherry-PA-Rac1, Addgene #22035) and selected for 1 week with 10 ng/mL of puromycin (10 ng/mL). Likewise, control cells were infected with lentivirus encoding cytosolic mCherry in a pLVX-Puro vector backbone (Clontech) (Wu et al., 2009). Imaging and photoactivation of MEFs was performed with cells plated on 35 mm dish with cover glass bottom (P35G-1.5–14 C, Mattek). For both control and PA-Rac1 cells imaging of mCherry with a 561 nm laser, and PA-Rac1 was stimulated with a 488 nm in a point scanning rectangular geometry and a laser power or 30 μW before the primary objective.
ARPE and RPE hTERT cells were generated and cultured as previously described (Aguet et al., 2013; Gan et al., 2016). The gap43-mCherry fly line used for imaging ventral furrow ingression is as previously described. (Martin et al., 2010). After dechornination and washing, the embryos were placed on the surface of a glass bottom Mattek dish and submerged in a droplet of water to prevent the embryos from drying out. Cardiomyocytes were isolated from the left ventricle of 1- to 2-day-old Sprague-Dawley rats. The isolation process and initial culture were described previously (Morales et al., 2016). Cells seeded on a 35 mm plate were stained with Fluo-3 AM calcium indicator (Thermofisher) at 1 µM for 20 min incubation after which seeded cells were image on the microscope using 488 nm excitation.
Rat primary neurons were obtained in accordance with protocols approved by the University of Texas Southwestern Medical Center Institutional Animal Care and Use Committee (IACUC). Cell Culture and Labeling Male and female embryonic day 18 (E18) primary cortical neurons were prepared from timed pregnant Sprague-Dawley rats (Charles River Laboratories, Wilmington, MA) as previously described (Bock and Herz, 2003). Embryonic cortices were harvested, neurons dissociated, and plated on 35 mm glass bottom culture dishes (MatTek P35G-1.5–10 C), coated with poly-D-lysine, at a density of 0.8 million neurons per dish. Neurons were cultured in completed Neurobasal medium (Gibco 21103049) supplemented with 2% B27 (Gibco 17504044), 1 mM glutamine (Gibco 25030081), and penicillin streptomycin (Gibco 15140148) at 37°C in a 5% CO2 environment. At day in vitro (DIV) two neurons were treated with AraC to prevent overgrowth of glia cells. Half of the culture media was renewed twice a week. On DIV5 neurons were infected with lentivirus encoding GCaMP6f. The lentivirus was generated by co-transfecting HEK 293 T cells with psPAX2, pMD2G (kindly provided by Didier Trono, Addgene numbers 12260, 12259), and pLV-GCaMP6f. Neurons were utilized for live-imaging at DIV12-16.
Mouse brain tissue was procured, cleared with ice-cold, nuclease-free 1x PBS solution and fixed with ice-cold 4%, nuclease-free PFA solution via transcardial perfusion. Following fixation, brain tissue was dissected and post-fixed for overnight at 4°C. Tissue was then cryo-preserved in 15% then 30% nuclease-free sucrose solution before freezing in OCT and cut into 40 micron coronal sections via vibratome. Sections were stored in nuclease-free, 1x PBS solution at 4°C prior to mounting and staining. To improve the optical clarity of the tissue samples, we removed light scattering proteins and lipids. To achieve this, the samples were first embedded to a polyacrylamide gel matrix and then enzymatically digested to remove proteins and chemically dissolved to remove lipids as previously described (Moffitt et al., 2016). In short, samples were washed for two minutes with degassed polyacrylamide (PA) solution consisting of 4% (vol/vol) 19:1 acrylamide/bis-acrylamide (161010144, BioRad), 60 mM Tris-HCl pH 8 (AM9856, ThermoFisher), and 0.3 M NaCl (AM9759, Thermofisher) and then washed for two minutes with PA gel solution which consists of PA solution with the addition of polymerizing agents TEMED (Sigma, T9281) and ammonium persulfate (A3678, Sigma). The PA gel solution was then aspirated from the coverslip-mounted sample. To cast the gel film, 200 µL PA gel solution was applied to the surface of a Gel Slick (50640, Lonza) coated glass plate and the coverslip-mounted sample was inverted onto the PA Gel solution, creating a thin layer of gel solution between the two panes of glass. The PA gel was allowed to cast for 1.5 hr. Coverslip and affixed gel film were carefully removed from the glass plate. Following the gel embedding, samples were washed twice for 5 min with digestion buffer consisting of nuclease-free water with 0.8 M guanidine-HCl (G3272, Sigma), 50 mM Tris-HCl pH 8, 1 mM EDTA and 0.5% (vol/vol) Triton X-100 (T8787, Sigma). Once complete, samples were incubated in digestion enzyme buffer consisting of digestion buffer supplemented with (0.5%) proteinase K (P8107S, New England Biolabs) and 5% Pronase (11459643001, Sigma), at a concentration of 80 U/mL, at 37°C for 24 hr to clear the tissue. Once cleared, tissue was washed in 2x SSC buffer three times for 5 min. After clearing, tissue was stained with DAPI at a concentration of 50 μg/mL in 2x SSC buffer overnight at 37°C. Sample was then washed in 2x SSC buffer two times for 5 min. The sample was finally mounted in a flow chamber (no-heat FCS2, Bioptechs) and immersed in SlowFade Diamond Antifade Mountant (S36967, ThermoFisher).
Excised sub transplant-quality human lung tissues from donors without preexisting chronic lung diseases were obtained from the Marsico Lung Institute at the University of North Carolina at Chapel Hill under the University of North Carolina Biomedical Institutional Review Board-approved protocols (#03–1396). Human lungs were fixed with 4% PFA solution for overnight at 4°C. The fixed lungs were washed with nuclease-free 1x PBS, incubated with 30% Sucrose in PBS for overnight at 4°C. Sucrose was replaced by 1:1 solution of 30% Sucrose:OCT and tissue was incubated for 1 hr at 4°C before freezing in OCT and cut into 10 to 15 micron sections using cryostat. Proximity ligation in situ hybridization (PLISH) was performed as described previously (Nagendran et al., 2018). Briefly, fixed-frozen human lung sections were re-fixed with 4.0% formaldehyde for 20 min, treated with 20 μg/mL proteinase K (Thermo Scientific, EO0492) in 1x PBS for 9 mins at 37°C, and dehydrated with up-series of ethanol. The sections were incubated with gene-specific probes (Supplementary file 2) in hybridization buffer (1 M sodium trichloroacetate, 50 mM Tris (pH 7.4), 5 mM EDTA, 0.2 mg/mL heparin) for 2 hr at 37°C followed by 4 washes each 5 min with hybridization buffer. Common bridge and circle probes were added to the tissue section and incubated for 1 hr at 37°C followed by a wash with 1x PBS containing 0.05% Tween-20 (PBS-T) for 5 min. Sections were incubated with T4 DNA ligase (New England Biolab, M0202T) for 2 hr at 37°C followed by 2 washes each 5 min with hybridization buffer and a wash with 1X phi29 DNA polymerase buffer (Thermo Scientific, B62). Rolling circle amplification was performed by using phi29 polymerase (Lucigen, 30221) for 12 hr at 37°C followed by 2 washes each 5 min with label probe hybridization buffer (2x SSC, 20% Formamide). Tissues were incubated with fluorophore-conjugated detection probes in label probe hybridization buffer for 30 min at 37°C followed by three washes each 5 min with label probe hybridization buffer. Immunostaining was performed using standard protocols. Briefly, tissue sections were incubated with primary antibodies against pro-SFTPC (Millipore, ab3786, 1:200 dilution) in 1x PBS-T containing 1% BSA and 5 mM EDTA for 1 hr followed by three washes each 5 min with PBS-T. Sections were incubated with Donkey anti-rabbit IgG secondary antibody (Thermo Scientific, A31573, 1:400 dilution) for 45 min followed by three washes each 5 min with PBS-T. Sections were mounted in medium containing DAPI (Thermo Scientific, 00-4959-52).
The source, authentication method, and mycoplasma state for all cell lines are provided in Supplementary file 3.
Microfluidic printing, fabrication, and casting
Request a detailed protocolMasks were designed using Tanner L-edit IC Layout (Mentor, Siemens Business) and printed using a Heidelberg µPG 101 mask writer (Heidelberg instruments) to create a 5-inch square chrome photomask on a soda lime glass substrate. After photomask printing, photomasks were developed in AZ 726 MIF developer (MicroChemicals), etched in chromium etchant (651826, Sigma–Aldrich) and cleaned in an ultrasonic acetone bath (V800023, Sigma–Aldrich). To create master molds, 4-inch round silicon wafers were surface etched with hydrofluoric acid (339261, Sigma-Aldrich) to remove the surface silicone oxide layer in order to ensure photoresist adhesion to the underlying silicon wafer. Master mold of approximately 5-micron height were created using manufacturers recommended spin coating and post-spin coating processing guidelines for SU-8 2005 negative photoresist (MicoChemicals). SU-8 2005 spin coated silicon wafers were contact printed using the developed photomasks on an EVG620 mask aligner (Ev Group) with 360 nm light exposure at a dose of 120 mJ/cm2 and post-baked according to manufacturer recommended guidelines (MicroChemicals). Prior to the first casting, each master mold was vapor deposition silanized with Trichloro (1H, 1H, 2H, 2H-perfluorooctyl) silane (448931, Sigma-Aldrich) in a vacuum desiccator (Scienceware, Sigma-Aldrich) to create an anti-stiction layer for PDMS demolding as previously described (Qin et al., 2010). To create micro channels, Polydimethylsiloxane elastomer (PDMS) castings of the master mold were made using SYLGARD 184 silicone elastomer (Dow Corning) and prepared and cured according to manufacturer’s recommended instructions. PDMS microchannels were cut to size using a razor blade before cell seeding ports were created using a 3.0 mm diameter biopsy punch (ProSciTech) at opposite ends of a 1500-micron length microchannel array. PDMS microchannel were bonded to glass bottom dishes (P35G-1.0–14 C, Mattek) using a corona tool (Electro-Technic Products Inc) as previously published (Haubert et al., 2006), sterilized with 100% ethanol before being incubated overnight with 20 µg/mL bovine telo-collagen I in PBS (5225, Advanced Biomatrix). On the day of assay, each dish was replaced with fully complemented media before seeding cells into each of the seeding ports.
Silanization of coverslips for tissue slices
Request a detailed protocolTo promote covalent adhesion of tissue and polyacrylamide gel to glass coverslips, coverslips were silanized according to previously published methods (Moffitt et al., 2016). In short, 40 mm, #1.5 coverslips (0420-0323-2, Bioptechs) were washed at room temperature in solution consisting of 1:1 (vol/vol) 37% HCl and Methanol for 30 min, rinsed in DI water three times and then dried at 60–70°C. Following this, coverslips were washed in solution consisting of chloroform with 0.1% (vol/vol) triethylamine (TX1200, Millipore) and 0.2% (vol/vol) allyltrichlorosilane (107778, Sigma) for 30 min at room temperature. Finally, coverslips were washed once in chloroform and once in ethanol and then baked for 1 hr at 60–70°C. Coverslips were then stored in a vacuum desiccation chamber before use.
Image based resolution metrics
Request a detailed protocolFourier ring correlation (FRC) was performed by first blind denoising individual raw images to remove correlated noise due to the sCMOS readout architecture (Banterle et al., 2013; Broaddus et al., 2020; Hörl et al., 2019; Krull et al., 2019; Van den Eynde et al., 2019; van Heel and Schatz, 2005). Raw images were then deconvolved, deskewed, and rotated into the coverslip reference frame as described above. The axial size of resulting raw and deconvolved imaging volumes was reduced to match the axial size of the cells within the region of interest. Because the transformed volumes were oversampled in the axial direction, even and odd images could be split into separate image stacks. FRC curves and resolution estimates were calculated using the 1/7 resolution criteria for all paired images. The mean and 95% confidence intervals were calculated across all FRC curves. Image decorrelation analysis was performed using provided FIJI plugin and default settings to the same images used for FRC (Descloux et al., 2019). The mean and 95% confidence intervals were calculated using all resolution estimates.
Bleaching comparison between OPM and lattice light-sheet microscopymicroscopy
Request a detailed protocolThe Lattice Light-Sheet Microscope used for photobleaching experiments has been described in detail elsewhere (Chang et al., 2019). In an effort to facilitate comparison with the original Lattice manuscript (Chen et al., 2014), we modified our system such that it used the 28.6X NA 0.66 (54-10-7, Special Optics) and 25X NA 1.1 (CFI75 Apo 25XC W, Nikon Instruments) objectives for illumination and detection, respectively. A 500 mm tube lens (AC508-250-A, Thorlabs) is used to achieve a final magnification of 62.5X. We chose an annular mask with an outer and inner diameter of 3.76 and 2.98 mm, which corresponds to an outer and inner NA of 0.269 and 0.213, respectively. The selected annular mask and the corresponding pattern on the spatial light modulator results in a square lattice light-sheet with an effective NA of 0.16 (Chang et al., 2020), which matches the illumination NA of the OPM. The resulting lattice was rapidly dithered to create a time-averaged sheet of light during the image acquisition. Photobleaching experiments were performed on ARPE cells expressing EB3-mNeonGreen seeded on #1.5 polymer coverslip imaging dishes (μ-Dish 35 mm, ibidi) or 5 mm diameter coverslips for the OPM and Lattice Light-Sheet Microscopes, respectively. In both setups we used similar imaging conditions that included an excitation wavelength of 488 nm, a laser power of 470 μW at the back pupil of the illumination objective, a 20 ms camera integration time, 201 image planes acquired with a 0.5 μm axial step size (before shearing), for 50 time points. EB3 comets were automatically detected using u-Track (Jaqaman et al., 2008), and their intensity plotted through time with MATLAB.
Appendix 1
Theoretical differences between the eSPIM (Yang et al., 2019) and the OPM presented here. A complete discussion is reported elsewhere (Millett-Sikking and York, 2019).
Our cumulative theoretical NA is >1.28, whereas eSPIM has an NA >1.06.
Our FOV is theoretically diffraction limited throughout a 110 × 110 micron region and expected to achieve 80% of the NA throughout a field of view of 220 microns, whereas eSPIM achieves a field of view of ~70×70 microns.
The remote refocus optics remain theoretically diffraction limited throughout a range of ± 30 microns, whereas eSPIM is ± 10 microns.
Working distance of our primary objective is 300 microns, whereas eSPIM has a 170-micron long working distance.
The glass-tipped tertiary objective can operate with tilt angles from 0 to 45 degrees, whereas eSPIM is limited by housing and water chamber to 0–30 degrees.
The choice of a silicone immersion objective reduces depth-dependent spherical aberrations when imaging within biological samples.
The tertiary objective is dedicated, plan, color corrected between 450–700 nm, and can extract images from any choice of secondary objective so long as they have ~100 microns of clearance. In contrast, eSPIM uses a water-dipping lens with a coverslip that is not plan or coverslip corrected and can only be paired with a limited selection of secondary objectives. Furthermore, the water chamber is quite complex.
The tertiary objective can be paired with a NA 0.95 secondary objective, whereas the eSPIM design had a maximum NA of 0.9 for the secondary objective. This is in part possible by the increased clearance engineered into the glass frustum that is attached to the tertiary objective.
The intermediate tube lenses and scan lenses deliver the full 200-micron diameter field of view from the primary objective, whereas eSPIM uses non-optimized achromatic doublets which limit the field of view to ~100 microns.
The secondary objective used here is a widely available fluorescence objective, whereas the secondary objective used in eSPIM is an industrial objective that has been discontinued.
Appendix 2
Choice of illumination angle, and tradeoffs associated with the 0 to 45-degree tilt
The tertiary objective is capable of accepting tilt angles between 0 and 45 degrees; a choice that is accompanied by several tradeoffs. A detailed discussion, and derivation, is provided by Millett-Sikking and colleagues (Millett-Sikking and York, 2019), and is freely available online (see https://github.com/amsikking/SOLS_optimum_tilt/blob/master/SOLS_optimum_tilt_quadratic_estimate.ipynb). These can be summarized as follows:
By definition, the excitation light-sheet and emission path must share the same primary objective NA.
As the tilt angle increases, thinner light-sheets can be produced, as more NA is available to shape the illumination beam.
As the tilt angle increases, the extent in the z-direction (normal to the coverslip) that can be covered with a given light-sheet confocal parameter increases (e.g. the useful length of the light-sheet). In contrast, a smaller tilt angle reduces the usable height of the imaging volume that can be covered with the light-sheet. Thus, from a light-sheet perspective, a large tilt angle is advantageous.
For low tilt angles, the detection efficiency and resolution improve owing to decreased reflective losses at the air-glass interface of the tertiary objective. Furthermore, if the tilt angle is 18 degrees or less, no light-clipping occurs (Figure 1—figure supplement 1B).
Pragmatically, one wants the light-sheet to be thin enough that its beam waist is comparable to the depth of focus of the detection objective. With the NA 1.35 silicone immersion objective used here, this results in a tilt angle that is close to 30 degrees.
Appendix 3
Estimation of theoretical NA and light-collection angle
To analyze the theoretical NA, it is helpful to first analyze the collection angle of the tertiary objective, which has a nominal NA of 1 for oil immersion (n = 1.51). The tertiary objective does not use oil, but has a glass frustum attached, which essentially mimics oil immersion and creates an optically flat refractive index interface at the objective’s focal plane. The half opening angle of the tertiary objective (41.47 degrees) reaches exactly the critical angle for total internal reflection at the glass-air interface. This means that the objective theoretically can collect light over a half opening angle of 90 degrees, or two pi steradian solid angle. This becomes obvious by the formulae for the critical angle and the definition of the NA:
The critical angle for a glass-air interface is given as:
The definition of NA, can be rearranged to:. For NA = 1 and n = 1.51, a half opening angle α of 41.47 results, which is obviously the same as above for the critical angle.
Another way to look at our tertiary objective is to compare its function to an ‘elusive’ NA one air objective. A NA of 1 would mean a half opening angle of 90 degrees in air, which would be highly beneficial for a tertiary objective in any OPM system to capture the most light emanating from the secondary objective. However, an NA one objective is not practical, as it would either have an infinitely large pupil for a finite working distance, or a working distance of zero for a finite pupil. Therefore, so far air objectives have always had an NA smaller than one.
In a sense, the tertiary objective presented here is the best approximation to an NA one air objective. This thought experiment also makes clear why the tertiary objective presented here has zero working distance: indeed, its frustum exactly ends at the focal plane of the objective.
To understand what we can do with such a tertiary objective, we next analyze which angles emanating from the secondary air objective (NA of 0.95, 71.8 degrees half angle) can be collected, depending on the tilt angle.
Let us first consider a zero-tilt angle, that is the secondary and tertiary objective are in-line and share the same optical axis (Figure 1—figure supplement 1A). Two marginal rays (red and magenta) coming out of the secondary objective at the highest allowed angles (±71.8 degree) are getting refracted toward the normal of the air-glass interface. This notably reduces the half opening angle from 71.8 degrees to 26.2 degrees inside the glass medium.
Next we consider a tilt angle of 18.2 degrees. In Figure 1—figure supplement 1B, the same marginal rays (red and magenta) are still both coupled into the tertiary objective. But the magenta ray enters the tertiary objective at the critical angle. This is the limiting case for which still the full light-cone coming out of the secondary objective can be transmitted into the tertiary objective.
If the tertiary objective is tilted more than 18.2 degrees, in our case by 30 degrees, one can see that it cuts into the cone of light of the secondary objective (light blue, Figure 1—figure supplement 1C). Thus, the light that can be coupled in at the critical angle occurs now at an opening angle of 60 degree at the secondary objective, which reduces the useful angular aperture of the secondary objective by 11.8 degrees. The red marginal ray is still coupled in at the glass interface, thus the full half angle of 71.8 degrees can be used on this side of the pupil.
Thus, an angular range of 60+71.8 = 131.8 degrees can be transmitted for a tilt angle of 30 degree. This corresponds to a useful NA of 1.4*sin(131.8/2)=1.28 of the primary objectives’ NA (nominally 1.35). Normal to the tilt direction, no clipping occurs by the tertiary objective. Thus in this direction, an theoretical NA of 1.4*sin(71.8)=1.32 can be used.
The exact solid angle used by the presented system cannot be derived from the simplified sketches shown here but must be computed analytically or numerically from the overlap of the Ewald’s spheres of each objective. Nevertheless, the value of 1.28 can serve as lower bound of the theoretical NA of our system. Since the upper bound is not much higher (1.32), we have decided to use the lower bound as an approximation here.
Appendix 4
Microscope alignment
For alignment purposes, a collimated green laser beam is injected into objective one from above in the diascopic direction. To align optical elements downstream, objective one is removed, and the size of the alignment beam is adjusted with an iris such that it either has a similar size as the back pupil of the primary objective, or that it becomes a narrow pencil-like beam (e.g. the beam divergence introduced by a lens is smaller than the divergence of the beam itself and can thus travel straight through a lens without changing much in size). Using the small pencil beam for alignment, back reflections are checked for every lens introduced into the optical train. Using the larger collimated beam, the correct distance and alignment between each pair of lenses is confirmed with a shear plate interferometer (e.g. the beam remains collimated). The alignment of the secondary objective is particularly critical. We place an iris downstream of where objective two will eventually go, as well as a frosted glass window with an aperture just before it (DG10-1500-H1-MD, ThorLabs). The hole is adjusted carefully in its height and lateral position such that the beam downstream is not steered or clipped in any way (checked by iris downstream). Now Objective two is introduced. It is carefully translated and rotated, while checking the back reflections on the frosted glass disk. If the correct alignment is found, a series of concentric rings should appear on the frosted glass disk. Simply holding the frosted glass disk by hand into the setup is not good enough, as minute displacements of the glass disk change the apparent back reflections from the objective. The hole of the glass disk needs to be well centered on the alignment beam.
Another critical alignment step is that the pupils of objective 1 and objective two are conjugated to each other. This can be verified by the ‘beam wobble method’, which is introduced below: Send a small pencil alignment laser beam over the OPM galvo mirror and put a card after objective 2. The beam emerging from objective two should be slightly diverging. Adjust the iris to obtain the smallest spot on the card. Now put a sinusoidal drive signal on the galvo mirror. If the objective is in the correct position, the beam exiting the objective does not move. If it is axially at the wrong position, the beam will wobble back and forth. If this is the case, carefully translate the objective axially and find the position where the beam appears stationary on the card. This is the position where objective two is conjugate to the galvo mirror. For Objective 1, inject a laser beam in the opposite direction (it can be the optogenetics laser beam, or the light-sheet beam without the cylindrical lens in place), which will pass over the OPM galvo and exit Objective 1. Use again a card after objective one and limit the beam size of the laser such that the smallest spot appears on the card. Use the same wobble trick to find the position of objective one where the beam is stationary. Now both objective’s pupil planes are conjugate to the galvo, and thus they are also conjugate to each other.
To align the tertiary imaging system, we recommend to first install it in straight transmission. If correctly aligned, the collimated green alignment laser (full beam width, objective one removed) has to appear on the center of the sCMOS camera and it has to form a sharp focus (FWHM slightly above on pixel). When imaging 100 nm fluorescent beads, the microscope system has to generate near-diffraction limited PSFs (~270 nm FWHM for green emission, no visible aberrations on the beads) over the full field of view (full chip of the sCMOS). It further has to deliver this performance over a ±10-micron focus range. To test this, translate beads on a coverslip up and down 10 microns, then refocus the tertiary objective to focus on them. No visible aberrations should appear at the axial positions outside the nominal focal plane. These measurements reveal that the aberration-free remote focusing works correctly. If aberrations occur outside of the nominal focal plane, check carefully if a slight lateral displacement of objective one relative to objective two can improve it. We found that such alignment errors show up strong in the non-nominal focal planes. Only if good 3D imaging performance is achieved in straight transmission we can proceed further. Now the tertiary imaging system can be tilted. Importantly, when it is correctly aligned, the alignment laser beam (objective one removed, full beam width) should appear centered and focused (the beam width should only marginally increase compared to the straight transmission case) on the sCMOS camera.
For routine alignment checks, objective one is removed every week, and the focusing of the alignment laser is verified on the sCMOS camera. Irises in the beam path are used to see if there is any obvious drift of any components. Typically, the stage that holds objective three drifts a bit with temperature, as well as the relative focal planes of objective 2 and 3. If all the rest of the beam path is correct (i.e. beam travelling centered through the irises), then usually a slight translation of objective three is needed to bring the beam back to focus and the center of the camera. Temperature drift of the focal plane of the primary objective one is corrected separately by forming a widefield image after the first tube lens on an inexpensive CMOS camera. If this widefield image is in focus, is well centered and looks aberration-free, the image on the sCMOS camera after the tertiary imaging system should also be centered and aberration-free. This procedure allowed us to maintain near-diffraction limited performance over long time periods of repeated use.
Data availability
Manuscript data is available on Zenodo, under the https://doi.org/10.5281/zenodo.4266823.
-
ZenodoA Versatile Oblique Plane Microscope for Large-Scale and High-Resolution Imaging of Subcellular Dynamics - Public Data.https://doi.org/10.5281/zenodo.4266823
References
-
Fourier ring correlation as a resolution criterion for super-resolution microscopyJournal of Structural Biology 183:363–367.https://doi.org/10.1016/j.jsb.2013.05.004
-
ConferenceRemoving structured noise with Self-Supervised Blind-Spot networks2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI). pp. 159–163.https://doi.org/10.1109/ISBI45749.2020.9098336
-
Universal light-sheet generation with field synthesisNature Methods 16:235–238.https://doi.org/10.1038/s41592-019-0327-9
-
Deconvolution-free subcellular imaging with axially swept light sheet microscopyBiophysical Journal 108:2807–2815.https://doi.org/10.1016/j.bpj.2015.05.013
-
Optically sectioned imaging by oblique plane microscopyOptics Express 16:20306.https://doi.org/10.1364/OE.16.020306
-
LITE microscopy: tilted light-sheet excitation of model organisms offers high resolution and low photobleachingThe Journal of Cell Biology 217:1869–1882.https://doi.org/10.1083/jcb.201710087
-
3d high- and super-resolution imaging using single-objective SPIMNature Methods 12:641–644.https://doi.org/10.1038/nmeth.3402
-
Reconstitution of cell migration at a glanceJournal of Cell Science 132:jcs225565.https://doi.org/10.1242/jcs.225565
-
Calcium signaling in cardiomyocyte functionCold Spring Harbor Perspectives in Biology 12:a035428.https://doi.org/10.1101/cshperspect.a035428
-
Rapid image deconvolution and multiview fusion for optical microscopyNature Biotechnology 38:1337–1346.https://doi.org/10.1038/s41587-020-0560-x
-
Computer visualization of three-dimensional image data using IMODJournal of Structural Biology 116:71–76.https://doi.org/10.1006/jsbi.1996.0013
-
ConferenceNoise2Void - Learning denoising from single noisy images2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR.pp). pp. 2124–2132.https://doi.org/10.1109/CVPR.2019.00223
-
Cell biological steps and checkpoints in accessing NK cell cytotoxicityImmunology & Cell Biology 92:245–255.https://doi.org/10.1038/icb.2013.96
-
Integration of contractile forces during tissue invaginationJournal of Cell Biology 188:735–749.https://doi.org/10.1083/jcb.200910099
-
Single objective light-sheet microscopy for high-speed whole-cell 3D super-resolutionBiomedical Optics Express 7:2219.https://doi.org/10.1364/BOE.7.002219
-
Acquisition of activation receptor ligand by trogocytosis renders NK cells hyporesponsiveThe Journal of Immunology 194:1945–1953.https://doi.org/10.4049/jimmunol.1402408
-
A BaSiC tool for background and shading correction of optical microscopy imagesNature Communications 8:ncomms14836.https://doi.org/10.1038/ncomms14836
-
Soft lithography for micro- and nanoscale patterningNature Protocols 5:491–502.https://doi.org/10.1038/nprot.2009.234
-
The NumPy array: a structure for efficient numerical computationComputing in Science & Engineering 13:22–30.https://doi.org/10.1109/MCSE.2011.37
-
Fourier shell correlation threshold criteriaJournal of Structural Biology 151:250–262.https://doi.org/10.1016/j.jsb.2005.05.009
-
depmixS4: anrpackage for hidden markov modelsJournal of Statistical Software 36:i07.https://doi.org/10.18637/jss.v036.i07
-
Spatially isotropic four-dimensional imaging with dual-view plane illumination microscopyNature Biotechnology 31:1032–1038.https://doi.org/10.1038/nbt.2713
Decision letter
-
Melike LakadamyaliReviewing Editor; University of Pennsylvania, United States
-
Anna AkhmanovaSenior Editor; Utrecht University, Netherlands
-
Rory PowerReviewer
In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.
Acceptance summary:
In this manuscript, authors describe an Oblique Plane Microscopy method that uses a bespoke glass-tipped tertiary objective with higher numerical aperture to achieve a combination of higher resolution and larger field of view imaging of biological specimens mounted in a range of imaging chambers including microfluidic devices. This advancement provides a combination of improved resolution, imaging throughput and ease of use over existing methods.
Decision letter after peer review:
[Editors’ note: the authors submitted for reconsideration following the decision after peer review. What follows is the decision letter after the first round of review.]
Thank you for submitting your work entitled "A Single-Objective Light-Sheet Microscope with 200 nm-Scale Resolution" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Rory Power (Reviewer #3).
Our decision has been reached after consultation between the reviewers. There was a very detailed, in-depth discussion of the reviewer comments during the consultation session. The overall conclusion that the reviewers and the reviewing editor reached during the consultation session is what formed the basis for the final decision. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife. The summary of the most pertinent points from the consultation was that the main claims of the manuscript, in particular the improved spatial resolution, are not substantiated and the data presented do not support the main conclusions of the manuscript. In addition, the conceptual innovation was deemed insufficient and technical improvements were deemed incremental. I hope that you will find the extensive comments of the reviewers useful for improving your manuscript.
Reviewer #1:
In this manuscript the authors describe an advancement of the single-objective, tilted-plane light-sheet concept. The general idea is to create a 2D light-sheet with as large of angle into the sample as the objective can support. The emission light field is relayed by the objective and other optics to a place in the emission pathway where the illumination/emission plane can be imaged by a high numerical aperture lens that is tilted to the optical axis. This allows the formation of an un-tilted plane on a detector. As is described in the Introduction of the paper, the work presented here is the evolution of this concept. Earlier designs used lower numerical aperture objectives to “un-tilt” the image and somewhat compromised the performance by not capturing all the light at the highest angles. In this design, no light is lost due to compromises in NA. This is accomplished by using a custom-designed objective that is essentially a glass immersion lens, improving the NA via higher index of refraction.
The ability to use standard cell preparations, coverslips and optics makes the tilted-plane concept extremely practical and powerful. The advancement reported here is incremental but takes the concept from some NA compromise to no NA compromise, so it is not clear why a titled-plane setup would not use this approach. I therefore think there is enough potential impact to publish in eLife.
Given that the impact of this work is largely of its ability for more researchers to build and engage with light-sheet instruments, the authors should provide more practical information on setup and alignment. For example, a picture in the supplementary information of an actual implementation, a basic alignment guide and supporting optical/mechanical parts. Or at minimum, explain how they will get this in the hands of users/builders.
Reviewer #2:
In this manuscript, Fiolka and co-workers attempt to improve the spatial resolution of state-of-the-art implementations of oblique plane illumination microscopy (OPM), and apply their technique to image subcellular dynamics in single cell samples. The title of the manuscript made me enthusiastic that the authors had achieved a technical breakthrough in light-sheet fluorescence microscopy (LSFM), but to my dismay my enthusiasm has been severely dampened upon reading their manuscript carefully. Not only have the authors failed to convince me that they have improved significantly upon state-of-the-art OPM, their data and analysis do not support their central claim of “200 nm scale” resolution. There is little here that advances light-sheet microscopy and the authors appear to have either wilfully or accidentally ignored a large body of literature demonstrating LSFM implementations with better collection efficiency and resolution than what they claim here. Biological insight is scant at best, with experiments presented as “one-offs” without any statistical rigor or attempt at reproducibility. eLife asks us to evaluate manuscripts that are of the “highest scientific standards and importance in all areas of the life and biomedical sciences”, and this paper does not meet this bar.
1) LSFM imaging with a single high NA objective is nothing new, and published work already demonstrates higher NA imaging with better collection efficiency and resolution than what is reported here. There are multiple papers that achieve LSFM with a high NA lens with NA > 1.1. The earliest I could find in my literature review is Gebhardt et al., published in 2013. Here a reflective surface in combination with 1.35 NA and 1.40 NA objectives was used to study single molecule transcription factor binding in live mammalian cells. This concept was refined and made easier to use in 2015 and 2016 by Galland et al. and Meddens et al., who used reflective microfluidic chips to perform single-molecule imaging of a variety of cell- and embryonic samples at high NA detection. Meddens used a 1.2 NA lens and Galland a 1.3 NA lens, and Galland specifically is notable since by slightly raising the sample, imaging of the entire cellular sample is possible without clipping (unlike in Gebhardt). While neither Meddens nor Galland used a 1.35 NA oil objective, this is presumably a trivial exchange on their system, and even with their reported objective lenses I suspect the lateral spatial resolution is improved over this work (more on that below). In reading Galland in particular, I was struck by the conceptual similarity and goals outlined in that manuscript and this one: “The soSPIM configuration enables broad modularity. Phase, differential interference contrast, wide-field, high-resolution and super-resolution imaging can be performed on the same microscope. Switching optical magnification does not involve any realignment. The requirement of perfect mechanical alignment of the two objectives used in traditional SPIM is obviated by the use of a single objective combined with software-controlled alignment of the light sheet along the mirror axis. Long-term stability is enhanced by the use of a single objective and the ability to implement the perfect-focus system built into inverted microscopes.”
Galland seems to achieve many of the stated goals of this technique.
And then there is “LITE” microscopy, published in JCB in 2018. Here an angled light-sheet (2.4 deg with respect to the horizontal) is used combination with high NA optics, up to an NA of 1.49. This latter NA enables higher resolution imaging than that shown here, as assayed by images of beads with FWHM < 250 nm, pre-deconvolution.
Like the OPM technique described in this manuscript, all of these techniques have their drawbacks – the first three require the addition of mirrored surfaces to bend the light-sheet, and LITE presumably suffers from out-of-focus illumination over the field of view and spherical aberration when using the 1.49 NA lens. A second drawback is that none of these techniques can easily achieve the speed of OPM-based methods.
Nevertheless, a significant advantage that all these “true” single-objective LSFMs have over the authors' work is access to the full NA of the detection lens, without the inherent losses that come from using 3 detection objectives (a 71% light loss is a hefty price to pay for a system that does not achieve the full NA of the primary lens). A biologist wishing to eke out the maximum signal and resolution from her dim samples (e.g. as single molecules or endogenously tagged CRISPR constructs often are) might well choose to capitalize on the advantages of these published techniques, despite their drawbacks. Thus, why is none of this prior work discussed in the manuscript? The most charitable explanation is that the authors of this work are unaware of it, but I am worried that in their effort to “sell” their current technique the authors have purposefully ignored this previous body of work. An intellectually honest comparison of their method in comparison to these other, higher NA methods would improve this work – in particular the claim that they achieve the “highest lateral resolution in light sheet microscopy” would seem to be strictly false based on these prior LSFMs operating at higher NA.
2) Resolution claims are not backed up by the data, and do not show a significant advance over previous work.
a) The authors claim a “200 nm scale lateral resolution”. Their sole evidence for this claim appears to be the images of 100 nm beads that they deconvolve with the Richardson-Lucy algorithm. Using RL on beads as an exclusive resolution metric is at best naïve, and at worst misleading and wrong. RL is known to turn beads into points, i.e. regardless of the physical limits of the optical system it is possible to claim an arbitrarily good resolution if one over-deconvolves. How were the number of iterations decided and why are the convergence curves not shown? The numbers the authors produce raise other red flags -
b) The authors deconvolve their beads until they reach a value of 189 ± 6 nm. The cutoff frequency for the lens they use, assuming 500 nm light, is λ/(2 * NA) = 500 nm / (2 * 1.35) = 185 nm. At face value, the deconvolved numbers would then suggest, in the best-case scenario, that the authors achieve an NA very close to 1.35. Indeed, the number they report makes me wonder if they simply deconvolved their data until they got close to this theoretical value. Unfortunately, their own raw measurements suggest otherwise, as for the same beads in widefield mode (sans OPM detection) they report ~240 nm FWHM. Assuming that there are not additional aberrations in this primary lens, and the ~240 nm corresponds to the full 1.35 NA, in OPM mode the authors are relatively far from achieving diffraction-limited performance – they report 284 nm x 328 nm lateral FWHMs in the OPM configuration. Given that resolution scales with NA, these numbers would suggest (at best) an NA of 1.35 (240/284) = 1.14 in the direction normal to the scan, and an NA of 1.35 (240/328) = 0.99 in the direction along the scan. The combination seems to be no (or marginally) better than what is achieved in lattice light-sheet microscopy with a 1.1 NA lens.
c) The authors argue that OPM occupies a unique niche among LSFMs, in its ability to easily integrate into existing optical microscopy workflows. I am sympathetic to this argument, but am not convinced that the technique described here is conceptually new or even advantageous compared to the previous state-of-the-art in this field: Yang et al.5. As the authors correctly point out, the conceptual advance realized by Yang is that sticking a piece of high refractive index material between secondary and tertiary lenses enables better performance than previous OPM. Using a bespoke objective instead of the awkward coverslip/water construction used in Yang is definitely a step in the right direction towards making the method more practical and commercially viable but is not a conceptual advance. On top of this, Yang reports pre-deconvolution bead-based FWHM values of 316±8nm and 339±18 nm laterally and 596 ± 32 nm axially. The lateral values are ~10% worse than those reported here, which is sufficiently close that I have a very difficult time believing that the authors' work represents a notable advance over the previous work in Yang et al. i.e. the minor improvement they report in pre-deconvolution FWHM values is unlikely to make any qualitative difference to the kind of data/insight obtained against this previously published paper, nor does it justify a quantitative jump from “300 nm scale” to “200 nm scale”. Moreover, the axial FWHM value reported in Yang is ~25% better than the 823± 31 nm number reported here. Taken together these comparisons would suggest that this work represents a rotation in parameter space rather than an advance.
d) Perhaps most remarkably, the authors make no attempt to verify “200 nm scale” in any of their images, e.g. by empirically investigating if they can separate biological features at this distance or by running Fourier-based methods that provide pixel-based resolution maps. Doing so would go a long way to clarifying if indeed they have achieved “200 nm scale” resolution vs. something more like “250 nm-scale” or even “300 nm scale” resolution in biological samples. Ideally these calculations would be performed with and without deconvolution, so that the effect of deconvolution is cleanly separated from the “raw” imaging performance. Speaking of raw performance, it is unclear to me which of the presented data is deconvolved vs. raw. Do the authors present any raw data anywhere in this paper? This point needs to be clarified.
e) What is the pixel size resulting from the imaging system, and the data presented here? It is concerning that all data, with the exception of the beads, seems to have been at least displayed in 115 nm x 115 nm x 100 nm voxels. Nyquist alone would dictate that “200 nm scale” lateral resolution is not possible.
3) The axial resolution values the authors report are significantly worse than state-of-the-art multiview LSFM techniques8-11, which can achieve up to a near twofold axial resolution enhancement compared to their (likely) over-optimistic deconvolved values of 570 nm. These references need to be cited as they set the bar as far as axial resolution, not the lattice light-sheet system. Speaking of axial resolution, why is the axial FWHM not reported in addition to the lateral FWHM measurements in Fiugre 2—figure supplement 5? Please provide all XYZ measurements before deconvolution, so the “raw” FWHM values are evident for the interested reader. This will help avoid confusion about likely over-aggressive deconvolution.
Reviewer #3:
The authors report a single objective light sheet (SOLS) microscope that provides unprecedented numerical aperture for light sheet imaging. The microscope utilizes a recently available solid-immersion objective lens to re-image an appropriately relayed tilted image plane onto a camera while encoding high NA information at smaller ray angles, thus allowing their capture without steric interference between the remote objective pair. This represents a natural progression from the eSPIM (Yang et al., 2019), which first illustrated the use of ray compression in SOLS. The application of this technology to several applications that would otherwise be extremely challenging in a light sheet/live-imaging context follow and illustrate that the microscope can image traditional sample preparations, in microfluidic chips and with high volumetric image rates. These applications provide an appropriate showcase for the technology and promising routes for further investigation without providing substantive biological conclusions. Given the technological focus of the article, I believe the article nonetheless warrants publication in eLife when the issues below are addressed. More generally, I believe this work and the recent developments on which it is built, will be of the utmost significance to the cell biology community in the coming years.
The following major points should be addressed before publication:
1) Having said that this is a technologically focused article, at least in terms of the innovation in the approach, I actually felt that the article was lacking in this regard. I am aware that much of the development around the solid-immersion objective has been reported via non peer-reviewed sources (GitHub etc.). However, as the first peer-reviewed study to utilize this technology, a more thorough description of the system would nonetheless seem warranted, particularly with regard to optical simulations, aberrations/useable viewing field (e.g. comparing simulations with observations) and a discussion of the various trade-offs etc. This may seem outside the scope of a typical eLife publication and the manuscript as prepared but without it, the greatest achievement of this work (the microscope development, not the biology in this case) is sidelined. The authors are uniquely placed to be able to provide this information.
2) The authors include several comparisons with lattice light sheet microscopy that could be supported better with corresponding data and expanded discussion. This is noted with regard to the comments below. Similarly, I believe that the authors are uniquely able to provide the additional information via simulations and data that likely already exists e.g. fluorescent bead PSF measurements (e.g. Chang et al., 2019.) The authors would also benefit from comparing against some other reports in high-resolution light sheet microscopy e.g. Bessel beam (one/two-photon excitation, axially-swept light sheet microscopy).
Introduction : The authors note that previous LSFM implementations require mounting either on a small cover glass (e.g. lattice light sheet) or in agarose tubes (more classical SPIM-type systems). However, the authors fail to note that in the latter case, this is actually a superior mounting method e.g. for developing embryos as the soft surrounding environment promotes normal development (e.g. Kaufmann et al., Development, 2012), whereas a hard glass interface does not. I appreciate that these applications are outside the scope of what is presented here but I still think further clarification is warranted and that the desire to use traditional preparations is rather an issue for cell cultures and high-throughput imaging. The authors should nonetheless note that coverslip mounting may produce biological artifacts in the context of their earlier studies using collagen gels for mounting with a long working distance SPIM approach (Welf et al., Dev. Cell, 2016).
Results: Please comment on the choice of angle and the trade-offs associated with the 0 – 45 degree tilt.
Subsection “Instrument Characterization”: Can you comment on how this compares to eSPIM (Yang et al., 2019). The lateral resolution appears to be slightly superior but the axial resolution seems to be lower (assuming a Gaussian beam for eSPIM). Is this a result of a longer thicker light sheet or a difference in how tilted the light sheet is with respect to the detection axis? This is surprising given the additional ca. 0.2 NA provided by the reported system.
and an 823 nm axial resolution?
Subsection “Instrument Characterization” final paragraph: Can you comment on the treatment here. Taking the efficiency simply as NA2 gives ca. 1.5, 1.9x respectively. This is without taking into the account the small NA loss relative to the 1.35 provided by the Si-immersion objective. Nevertheless, this analysis is welcome and generally underreported in single-objective light sheet microscopy.
Figure 2 a) Is there a reason to choose a different color lookup table for this image?
Discussion second paragraph: Repetitive with regard to the introduction.
“…it has the highest lateral resolving power (~200 nm) in light-sheet
microscopy. Indeed, owing to the narrow depth of focus provided by the optical design, the axial resolving power” This is the deconvolved resolution. It would be helpful to include PSFs from light sheet microscopes and report both the raw and deconvolved resolutions in each case. The authors previous efforts regarding field synthesis should yield the required data (Chang et al., 2019).
Regarding contrast and axial resolution, which are tied together, the lattice light sheet supposedly maintains a thin profile over a larger distance than a purely Gaussian light sheet. The authors should report on the useable field of view when noting comparisons of contrast and resolution (and whether this is dominated by the requirements of a thin light sheet or rather the detection optics/remote focus scheme). The Gaussian sheet should produce better contrast at the focus whereas the lattice trades some contrast for field of view. However, the high detection NA in the SOLS case does mean that the light sheet will be thicker at waist than the depth of field and so contrast is also lost here. Again, the authors are uniquely placed to make these comparisons in a more rigorous manner. At the minimum, it would be helpful to report the length and thickness of the light sheet produced at the 0.2 – 0.3 NA (reported in the Materials and methods), to provide some measure of the effective field of view and contrast for the lattice and SOLS cases.
With regard to laser power requirements and synchronous multicolor imaging, this is largely a result of how lattice light sheets have typically been generated. The authors own previous research regarding field synthesis (which should be cited) has reduced these issues for anyone who may be choosing whether to construct a SOLS or lattice light sheet system. The primary issue remaining is the lack of isolation between the objectives and media and the steric issues.
Subsection “Data Post-Processing”: Please comment on the time taken and computational resources required for shearing and deconvolving the datasets. Since the SOLS system appears a good fit for imaging facilities, the associated throughput rate is a consideration.
[Editors’ note: further revisions were suggested prior to acceptance, as described below.]
Thank you for submitting your article "A Versatile Oblique Plane Microscope for Large-Scale and High-Resolution Imaging of Subcellular Dynamics" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Anna Akhmanova as the Senior Editor. Two of the original reviewers were consulted as well as a new expert for third opinion. The following individual involved in review of your submission has agreed to reveal their identity: Rory Power (Reviewer #3).
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, when editors judge that a submitted work as a whole belongs in eLife but that some conclusions require a modest amount of additional new data, as they do with your paper, we are asking that the manuscript be revised to either limit claims to those supported by data in hand, or to explicitly state that the relevant conclusions require additional supporting data.
As you will see from the comments, the reviewers agreed that the revisions improved the paper and that the presented technique is interesting and provides certain advantages over existing methods. As such, the reviewers found the paper appropriate for eLife. That being said, two of the reviewers had very strong remaining criticism on the way the claims are presented. They felt that the pros and cons of the new method were not fairly discussed and that the comparisons to other methods were not appropriately presented. In particular, the reviewers felt that the trade-off coming from the poor light efficiency of the system was not properly discussed, comparisons to existing methods tended to highlight the strengths of the current method while not properly discussing the weaknesses such as " listing one dimension of an acquired sample (the lung tissue explant here) that's larger (8X) than what was imaged in other publications and then claiming this instrument is an improvement while leaving out the other dimensions that are clearly worse (45X thinner). ". In the consultation session, there was also concern over how the field of view comparisons were made and whether a tiled acquisition was being compared to a non-tiled acquisition with other methods, which the reviewers did not find to be a fair comparison.
Overall, the reviewers and myself believe that the claims need to be tempered down and the presentation on pros/cons of the method must be improved. I append below detailed comments from the reviewers, which should help you make these improvements.
Reviewer #3:
The authors have made substantive changes to the manuscript following the first round of reviews. I remain unswayed in my opinion that this work is suitable for publication in eLife given required improvements. I believe that the changes made have improved the manuscript.
In particular, the assessment of the system PSF and resolving capability has been greatly improved by including the non-deconvolved PSFs, details of the deconvolution and Fourier ring correlation estimate of resolution. The discussion of true single objective light sheet systems and comparison with eSPIM places the reported technique in context and I believe helps to make its case far more than detracting from it. The appendices should go some way to assisting others in building such a system.
There are still some changes based on my suggestions, which under ideal circumstances would be adopted. However, the author's arguments for not having done so are well reasoned. My primary concern was the frequent comparison with lattice light sheet microscopy. Here I still find the manuscript somewhat lacking but do appreciate that the authors field synthesis lattice light sheet system does not use the typical configuration of objective lenses making comparison difficult. The benefits of the reported system run deeper than pure resolution claims in any case and given the author's recent publication (Cheng et al. 10.1364/OE400164) reporting on the Gaussian to lattice light sheet comparison the omission here is reasonably justified.
All things considered, I would recommend the article for publication.
Reviewer #4:
First, I wish to commend the authors for their thorough revision. They have addressed many of my concerns and I feel better about the overall manuscript. The emphasis on field of view over resolution, the Fourier-based methods of assessing resolution, the citing of previous work, the additional details put forth in the appendices, and more accurately referring to their method as an OPM have all significantly improved the work. Here are my remaining concerns/comments:
a) The authors now more fairly describe the advantages of their method over previous similar microscopes. They still do not clearly state what I see as weaknesses of their method, which seems important given the extensive comparisons to LLSM and other systems:
i) relatively low sensitivity. In their revised manuscript, the authors now seem to admit that the advantage in light efficiency offered by the 1.35 NA lens is less than optical losses through their emission path. In sum, these losses would seem to imply a strictly lower sensitivity than the more “conventional” LLSM with 1.1 NA detection. Please discuss this point clearly in the main text, rather than relegating it to Materials and methods.
ii) Low effective NA vs. theoretical NA. I appreciate the appendix describing the NA calculations, but I am still not convinced that the number presented is an “effective NA”, i.e the NA of an equivalent widefield with primary lens using this NA. In particular, I am still bothered by the apparent loss of resolution compared to their widefield performance, which I commented on previously and repeat here:
“Unfortunately, their own raw measurements suggest otherwise, as for the same beads in widefield mode (sans OPM detection) they report ~240 nm FWHM. Assuming that there are not additional aberrations in this primary lens, and the ~240 nm corresponds to the full 1.35 NA, in OPM mode the authors are relatively far from achieving diffraction-limited performance – they report 284 nm x 328 nm lateral FWHMs in the OPM configuration. Given that resolution scales with NA, these numbers would suggest (at best) an NA of 1.35 (240/284) = 1.14 in the direction normal to the scan, and an NA of 1.35 (240/328) = 0.99 in the direction along the scan. The combination seems to be no (or marginally) better than what is achieved in lattice light-sheet microscopy with a 1.1 NA lens.”
It is very reasonable to suppose that NA would scale with the apparent size of a subdiffractive object, regardless of suboptimal Strehl ratio, MTF, hardware etc. And if the NA of the primary lens is effectively less than 1.35, this would seem to further diminish the “effective” NA that is achieved in this manuscript. This point is important to address properly, because it has implications both for comparisons to other microscopes and it suggests that the deconvolved numbers the authors present are still overly optimistic (e.g. if the NA is not in fact 1.28, physically it does not seem possible to achieve a deconvolved resolution of ~203 nm as in Figure 1H). I would suggest replacing “effective NA” with “theoretical NA” everywhere the former is referred to, and being very clear in the manuscript that there is still residual resolution loss: i.e. the theoretical NA is not achieved at the tilt settings used in this paper. The authors could prove me wrong by providing a suitably compelling measurement of the “effective NA” rather than the theory – presenting a theoretical argument does nothing to convince me that in fact they achieve this value, which their own measurements seem to contradict.
b) Regarding deconvolution, I am still having difficulty understanding how ~203 nm resolution is possible with 115 nm voxels. Once more, Nyquist alone would seem to render this number incorrect. The authors' argument in using a factor 1.414 seems to boil down to: “other people have done it, so we should do it too”. This is not a good argument in my opinion, as it leads to a result that contradicts common sense. The authors digitally upsample the bead data before deconvolution, but I do not see how such digital upsampling could effectively beat Nyquist.
The authors in their rebuttal seem to suggest that presenting the raw values sans deconvolution is unusual in multiview LSFM (“I think this is by far more forthcoming than many other manuscripts, including Multiview LSFM techniques, which only report resolution after iterative deconvolution”). I am not sure what they are talking about: Wu 2013 (diSPIM), Keller 2015 (IsoView), classics in this area, state these values clearly in the main text.
c) The authors seem to acknowledge in their rebuttal that they do not achieve diffraction-limited performance (“we agree we do not achieve diffraction-limited performance”), but I still find multiple places in the manuscript where this is stated or implied. Please address this point.
d) Depth. What sets the effective depth limit of this technique, as presented here? 25-30 μm is relatively modest for a lens with a 300 μm working distance, so discussing theoretical and practical limits would be useful for a biologist who is searching for a practical solution for their sample.
e) The authors state numerous times that they achieve resolution on par with or better than LLSM. Please explicitly state the corresponding values reported in LLSM (or measure them with an LLSM with 1.1 NA detection), where relevant.
f) “And unlike state-of-the-art multiview LSFM techniques that achieve a slightly better axial resolution, only a single objective and imaging perspective is needed (Guo et al., 2020; Wu et al., 2013).” Wu 2013 achieves ~350 nm axial resolution, Wu 2017 achieves ~300 nm axial resolution, and Guo 2020 enhances the effective axial resolution of LLSM to ~380 nm resolution. These are not “slightly better”, they are significantly better (indeed, the degree of improvement is more convincing than what is shown here relative to LLSM). Also, as I think the authors acknowledge in their rebuttal, they do not use a single objective, rather they use 3. Please remove “slightly” for accuracy and/or rethink this sentence.
Reviewer #5:
In this manuscript, "A versatile oblique plane microscope for large-scale and high-resolution imaging of subcellular dynamics", Sapoznik et al. describe a variant of an oblique plane microscope (OPM) that makes use of a custom designed tertiary objective to capture a greater portion of the emitted rays through the oblique refocusing objectives. OPM combined with light sheet is attractive because it offers the potential for optically sectioned, low phototoxicity imaging using only a single primary imaging objective with 180 degree physical access. As the authors note, this concept itself is not new and many descriptions and implementations have been described before (originally with Dunsby in 2008 and more recently with Bouchard 2015 and Yang 2019). Thus, the primary technical innovation here is the implementation of the custom-designed solid immersion lens.
In general, I think this is a useful addition to the field and could be suitable for publication in eLife. An open top-light sheet microscope with increased sample accessibility is indeed useful. However, in its current form, the manuscript reads more like an aggressive sales pitch rather than a balanced discussion of the pros and cons of different microscope approaches. The presentation is over-dismissive of prior work, often mis-representing or cherry-picking specific comparisons to make this current instrument appear better, while neglecting to mention or discuss trade-offs wherein the instrument might perform worse. I noticed that the authors disclose financial relationships with companies that sell products that compete with commercial versions many of the instruments compared here. However, I feel that the readers of eLife would benefit if the authors focused on more thoroughly documenting their own scientific contribution, together with its trade-offs, rather than (often incompletely or inaccurately) characterizing prior work from others.
1) The resolution claims here are provided by FHWM measurements of isolated beads and of xy (lateral) measurements using correlation based approaches on the images. Importantly, optical sectioning is not discussed at all. In this instrument, the resolution is provided by the high NA primary objective, but due to the thick lightsheets (0.16 NA Gaussian beams) used in most of the measurements, the lightsheets are substantially thicker than the depth of field for the detection. This would be readily apparent as the "missing cone" in the optical transfer function and I suspect it would also affect correlation based axial resolution measurements if they were conducted on cells or embryos rather than isolated beads. In contrast, axial resolution in several other approaches (Gaussian, Bessel, Lattice etc) is obtained by using a lightsheet that is thin-compared to the detection depth of field. The authors' own prior work, Dean et al., 2015, have highlighted the advantages of this approach, but there is no mention of the relationship between optical sectioning and resolution here.
FWHM measurements in isolation are a limited picture of resolution. The authors should present optical transfer functions in the xy, xz, and yz planes to demonstrate true resolution and to what extent their system fills in the missing cone. They should also perform the raw image-based correlation measurements in the axial dimension (as they already do now for the lateral dimension) on a range of biological samples. This could be easily done with existing data.
2) The authors describe that, due to the optical path required for OPM, the instrument loses 71% of the photons from the sample before they hit the camera. I.E. it has a transmission efficiency of 29%. This is in stark contrast to the standard SPIM systems which due to the simple widefield detection optics should operate at very close to the ~80-90% transmission efficiency of commercial objectives. Further, it's not clear at what imaging angle this measurement was performed. This information is also important because the efficiency decreases further as the primary objective is utilized further away from its design standard angle of 0 degrees. In the appendix, the authors describe how the 1.2 NA of this system would somewhat compensate for this poor transmission over the 1.1NA lens used in other variants, but this analysis assumes that all rays have equal transmission efficiencies which generally isn't true. Thus even without any additional optics, an OPM system operating with an off-centered pupil will have a lower transmission efficiency than an objective with the same effective NA operating on axis.
Experimental characterization of how the transmission efficiency between a 1.2 NA objective operating with a centered pupil vs. a higher NA objective with an effective 1.2 NA due to pupil decentering with OPM and associated optics, and how this declines with the OPM angle would be extremely useful for the field. The authors are ideally poised to make these type of measurements.
Regardless, this is an important discussion, so I'm unclear why the measurements are not mentioned in the main text under "Instrument Characterization". They are instead provided in the fourth paragraph of Materials and methods section under "Laser Scanning Microscope Setup". Given that microscope end users are willing to pay thousands more for a back thinned camera with a 10% increase in quantum efficiency than a non-back thinned unit, the authors should mention that the OPM implementation here comes at the cost of reduced detection efficiency in the main text. The statement that the instrument is "sufficiently sensitive to detect single molecules (data not shown)" does not adequately address this concern. Especially when other approaches have clearly demonstrated their utility for live cell and super-resolution single molecule imaging with both dyes and fluorescent proteins alike. It's also not clear how the authors define "without obvious signs of phototoxicity". How was this measured and what is considered obvious?
3) I have concerns about how prior methods are discussed. For example, Lattice Lightsheet Microscopy is not a single instrument, but a general description of the use of excitation patterns based on optical lattices for microscopy. The specific choices for excitation objective and detection objective as well as the type of lattice light sheet and extent of optical confinement used will determine the resolution and optical sectioning and can be chosen/optimized for specific applications. Thus, statements like "Lattice Lightsheet microscopy has a resolution of xx" or "requires a 5mm coverslip" are no more accurate than saying "Gaussian beam microscopy has a resolution of xx and requires agarose sample embedding". It all depends on how one decides to configure an instrument and for what purpose they choose to balance the tradeoffs. It's fine to say that a specific publication reported certain values, but this should be accompanied by some context. In many cases, design considerations may have been chosen to make an instrument that is optimized for a different purpose or with additional features than the one presented here.
https://doi.org/10.7554/eLife.57681.sa1Author response
[Editors’ note: The authors appealed the original decision. What follows is the authors’ response to the first round of review.]
Reviewer #1:
In this manuscript the authors describe an advancement of the single-objective, tilted-plane light-sheet concept. The general idea is to create a 2D light-sheet with as large of angle into the sample as the objective can support. The emission light field is relayed by the objective and other optics to a place in the emission pathway where the illumination/emission plane can be imaged by a high numerical aperture lens that is tilted to the optical axis. This allows the formation of an un-tilted plane on a detector. As is described in the Introduction of the paper, the work presented here is the evolution of this concept. Earlier designs used lower numerical aperture objectives to “un-tilt” the image and somewhat compromised the performance by not capturing all the light at the highest angles. In this design, no light is lost due to compromises in NA. This is accomplished by using a custom-designed objective that is essentially a glass immersion lens, improving the NA via higher index of refraction.
The ability to use standard cell preparations, coverslips and optics makes the tilted-plane concept extremely practical and powerful. The advancement reported here is incremental but takes the concept from some NA compromise to no NA compromise, so it is not clear why a titled-plane setup would not use this approach. I therefore think there is enough potential impact to publish in eLife.
Given that the impact of this work is largely of its ability for more researchers to build and engage with light-sheet instruments, the authors should provide more practical information on setup and alignment. For example, a picture in the supplementary information of an actual implementation, a basic alignment guide and supporting optical/mechanical parts. Or at minimum, explain how they will get this in the hands of users/builders.
We would like to thank the reviewer for their kind words, and we are happy that they recognize the power of the approach and believe that our method has sufficient impact for publication in eLife. We also agree that providing practical information on the setup and alignment of the microscope would be useful. As such, we now provide a basic alignment guide in Appendix 4. Furthermore, we have tried to make the Materials and methods section as exhaustive as possible, while pointing potentially interested readers to online sources of information that include complete parts lists.
Alignment:
We now provide in Appendix 4 a section with tips and tricks for aligning the microscope.
We are also working pro bono with Applied Scientific Instrumentation so that they may provide turnkey optical components for assembling the microscope.
In the Materials and methods, Laser Scanning Microscope Setup, second paragraph, last sentence, we now point individuals to the complete parts list provided by Millett-Sikking et al.
We are working to place a CAD rendering of the microscope online, but this is a work in progress. Nonetheless, if the reviewer sees this as critical, we assure you that we will have this completed prior to publication of the manuscript.
Reviewer #2:
In this manuscript, Fiolka and co-workers attempt to improve the spatial resolution of state-of-the-art implementations of oblique plane illumination microscopy (OPM), and apply their technique to image subcellular dynamics in single cell samples. The title of the manuscript made me enthusiastic that the authors had achieved a technical breakthrough in light-sheet fluorescence microscopy (LSFM), but to my dismay my enthusiasm has been severely dampened upon reading their manuscript carefully. Not only have the authors failed to convince me that they have improved significantly upon state-of-the-art OPM, their data and analysis do not support their central claim of “200 nm scale” resolution. There is little here that advances light-sheet microscopy and the authors appear to have either wilfully or accidentally ignored a large body of literature demonstrating LSFM implementations with better collection efficiency and resolution than what they claim here. Biological insight is scant at best, with experiments presented as “one-offs” without any statistical rigor or attempt at reproducibility. eLife asks us to evaluate manuscripts that are of the “highest scientific standards and importance in all areas of the life and biomedical sciences”, and this paper does not meet this bar.
We appreciate the reviewer’s careful analysis of the manuscript, as well as their forthright criticism.
Resolution: We have substantially strengthened our resolution measurements and they now include Decorrelation Analysis and Fourier Ring Correlation Analysis on both beads and biological data, before and after deconvolution. We also present raw and non-deconvolved data for both beads and cells. These measurements again confirm that our microscope delivers resolution that is equal or better than Lattice Light-Sheet Microscopy for raw, and non-deconvolved data (see 10.1364/OE.400164, 10.1038/nature22369, 10.1364/BOE.11.000008). Further, we now more clearly show that this level of resolution can be maintained over a large field of view.
Literature: It was not our intention to “wilfully ignore” single objective light-sheet fluorescence microscopes based upon micromirrors, cantilevers, or “lateral interference tilted excitation”, and we have rectified this error by discussing these methods in the Introduction and comparing our performance to them throughout the manuscript. From our analysis, it is clear that many of these techniques cannot compete with the field of view, volumetric imaging capacity, resolution, imaging speed, and practical usability that we demonstrate with our method.
Biology: Admittingly, many of the biological experiments presented in our manuscript are indeed “one-offs”. However, we only ask that the reviewer holds us to the same standard as every other microscopy development paper, including those published in eLife (10.7554/eLife.14472, 10.7554/eLife.32671, 10.7554/eLife.40805, 10.7554/eLife.45919, 10.7554/eLife.46249). We consider it unfair to stipulate that we need to solve a biological conundrum while improving the field of view by a factor of 3.7 and matching the resolution of Lattice Light-Sheet Microscopy, a method which is onerously complex and incompatible with some of the samples imaged here. We have now more systematically analyzed PA-Rac1 stimulation and can show a significant increase in protrusion speed and duration compared to control cells.
Statistical rigor: This statement is demonstrably false, as we report quantitative metrics for the diffusion of cytosolic tracers, protrusion dynamics upon optical stimulation of Rac1, and evaluate the point-spread function of a large number of beads. Importantly, each measurement is accompanied by either standard deviations or 95% confidence intervals. With regard to the statistical robustness of the remaining biological observations, we are in agreement with Yu et al. (10.7554/eLife.46249) when they state that “the goal of the current paper was to demonstrate a technology, rather than test a hypothesis, [so] we did not pre-determine any sample sizes for this study.”
Scientific Standards: We have not only met the scientific standards commonly adopted in optical microscopy but exceeded them by meticulously providing details regarding the acquisition, processing, and analysis of our data, and by making publicly available all data and software. Historically, we have been penalized for not deconvolving our data, as well as for deconvolving our data. Unlike many other manuscripts, we clearly state the resolution of our raw data, and limit the number of iterations of Richardson-Lucy to avoid misleading results. Indeed, we limit the resolution enhancement to a factor of , which is commonly accepted by many in the super-resolution fields as theoretically appropriate. In contrast, publications using lattice light-sheet microscopy regularly deconvolve their data until they achieve up to 1.9-fold resolution enhancement in the axial resolution and with one notable exception (10.1038/nature22369) fail to disclose raw resolution values (10.1364/OE.400164., 10.1364/BOE.11.000008).
Importance in all areas of the life and biomedical sciences: We have presented a truly versatile microscope that uniquely delivers resolution, field of view, speed, subcellular optogenetics, and the ability to image an incredibly diverse collection of specimens that range from single cells to entire tissues. We politely disagree with your diagnosis that it is not of importance in all areas of the life and biomedical sciences.
1) LSFM imaging with a single high NA objective is nothing new, and published work already demonstrates higher NA imaging with better collection efficiency and resolution than what is reported here. There are multiple papers that achieve LSFM with a high NA lens with NA > 1.1. The earliest I could find in my literature review is Gebhardt et al., published in 2013. Here a reflective surface in combination with 1.35 NA and 1.40 NA objectives was used to study single molecule transcription factor binding in live mammalian cells.
This is a classic paper from Sunnie Xie’s lab which introduces a light-sheet from the diascopic direction with a water immersion objective and a reflective cantilever. Nonetheless, by placing the specimen in contact with the water immersion objective and cantilever, it will no longer be sterile. Furthermore, the specimen must be in immediate proximity to the cantilever tip, which is technically challenging and limits the field of view to roughly 1/200th of what we report for our OPM. Further, the light-sheet cannot reach the bottom of the cells, as it would get aberrated by the coverslip. Proper use of the AFM-based microscope also requires recalibration for every imaging error (see their methods). Perhaps most importantly, this method is incompatible with many of the specimens that we image with our OPM, including the microfluidic, tissue sections, neurons, etc. As such, while we agree with the author that it is a demonstration of a high NA single objective light-sheet, it lacks the general usability and performance (Field of view, volumetric imaging rate) that we demonstrate. But we include it now in our manuscript.
This concept was refined and made easier to use in 2015 and 2016 by Galland et al. and Meddens et al., who used reflective microfluidic chips to perform single-molecule imaging of a variety of cell- and embryonic samples at high NA detection. Meddens used a 1.2 NA lens and Galland a 1.3 NA lens, and Galland specifically is notable since by slightly raising the sample, imaging of the entire cellular sample is possible without clipping (unlike in Gebhardt). While neither Meddens nor Galland used a 1.35 NA oil objective, this is presumably a trivial exchange on their system, and even with their reported objective lenses I suspect the lateral spatial resolution is improved over this work (more on that below). In reading Galland in particular, I was struck by the conceptual similarity and goals outlined in that manuscript and this one: “The soSPIM configuration enables broad modularity. Phase, differential interference contrast, wide-field, high-resolution and super-resolution imaging can be performed on the same microscope. Switching optical magnification does not involve any realignment. The requirement of perfect mechanical alignment of the two objectives used in traditional SPIM is obviated by the use of a single objective combined with software-controlled alignment of the light sheet along the mirror axis. Long-term stability is enhanced by the use of a single objective and the ability to implement the perfect-focus system built into inverted microscopes.”
Galland seems to achieve many of the stated goals of this technique.
Galland et al. use a sophisticated microfluidic method with embedded micromirrors to illuminate the specimen with a light-sheet. While we agree that changing the objective on this system is trivial, the photolithography, sequential anisotropic and dry silicon etching, and sputtering of gold surfaces to create a micromirror, is not. Furthermore, such a design requires that the specimen be placed into a microcavity, which again limits the types of specimens that can be imaged. Even for single cell aggregates, Galland et al. had to collect multiple images per image plane with the illumination beam defocused by 20 microns with an electrotunable lens, and then stitch the images together. Similarly, also the bottom of the cell cannot readily be imaged, as the light-sheet would be distorted. Very little information is provided regarding how they performed or analyzed their resolution, and a single value is provided (565 nm) for the nuclear lamina. While this may be similar to our axial resolution upon increasing the excitation NA, insufficient information is provided for proper evaluation as statistics are not provided. For samples larger than 28 microns, they have to substantially decrease the illumination of the NA. Thus, our OPM is much more user friendly and capable of imaging a much more diverse array of samples.
Meddens et al. also use a sophisticated microfluidic method with an embedded micromirror, but their illumination train does not have a tunable lens. Also, rather than having microcavities, Meddens et al. flow the cells into a microfluidic chip, which necessitates that the sample advantageously settles in a region adjacent to the mirror. Also, no information is provided regarding their resolution in the absence of STORM imaging. Thus, this method suffers from many of the same drawbacks as Galland et al. with regard to resolution, ability to image diverse specimens, field of view, speed, and more. Indeed, it is clear from their images that our OPM method delivers superior imaging results.
And then there is “LITE” microscopy, published in JCB in 2018. Here an angled light-sheet (2.4 deg with respect to the horizontal) is used combination with high NA optics, up to an NA of 1.49. This latter NA enables higher resolution imaging than that shown here, as assayed by images of beads with FWHM < 250 nm, pre-deconvolution.
As the reviewers has mentioned later, the use of a high NA oil immersion lens results in significant spherical aberrations, which necessarily leads to a depth dependent lateral and axial resolution as well as a diminished sensitivity (owing to a decreased Strehl ratio). Fortunately, while changing the objective in such a system is trivial, one must ask why they didn’t use a water or silicone immersion objective in the first place. Unfortunately this means that they are only competitive with our technology at the glass interface. Once they image a few microns into the specimen, the performance noticeably deteriorates. Also, in an effort to cover a field of view 150 microns, they use a light-sheet that is 4.3 microns thick. Thus, their optical sectioning will also be quite poor owing to a large amount of out-of-focus illumination and image blur. It is essentially “widefield” from the side. For higher NA illumination beams, the quadruple-slit photomask used results in periodic intensity modulation along its propagation axis of the illumination beam, rendering it non-quantitative. While quite convenient, LITE requires custom imaging chambers and is thus not compatible with the diverse array of specimens that we imaged (e.g., microfluidics, 96-well plates). Ultimately, LITE is in principle capable of imaging a similar sized field of view as us, but at a complete loss of optical sectioning capability. Further it lacks the imaging speed and versatility that makes our system powerful and is really only competitive in terms of resolution in the region adjacent to the glass coverslip.
Like the OPM technique described in this manuscript, all of these techniques have their drawbacks – the first three require the addition of mirrored surfaces to bend the light-sheet, and LITE presumably suffers from out-of-focus illumination over the field of view and spherical aberration when using the 1.49 NA lens. A second drawback is that none of these techniques can easily achieve the speed of OPM-based methods.
Thank you, we agree completely. We find that the addition of a mirrored surface near the specimen drastically limits the versatility of the microscope and makes routine manufacture of the mirrored surface or operation of the microscope quite technical. We agree that these techniques need to be discussed and their strength and weaknesses need to be better detailed.
Nevertheless, a significant advantage that all these “true” single-objective LSFMs have over the authors' work is access to the full NA of the detection lens, without the inherent losses that come from using 3 detection objectives (a 71% light loss is a hefty price to pay for a system that does not achieve the full NA of the primary lens). A biologist wishing to eke out the maximum signal and resolution from her dim samples (e.g. as single molecules or endogenously tagged CRISPR constructs often are) might well choose to capitalize on the advantages of these published techniques, despite their drawbacks.
The collection efficiency of microscopes varies widely and is rarely reported. For example, the collection efficiency of a single molecule confocal microscope can be as low as ~3% (see 10.1529/biophysj.108.134346). Nonetheless, we do agree that every microscope genre occupies its own niche. If a biologist needs every photon, we would recommend a TIRF (and potentially HiLO or grazing incidence modes) microscope because it combines high NA optics, 100% illumination duty cycles, optical sectioning, and aberration free detection. Unfortunately, TIRF and HiLO are not very useful for volumetric imaging. Nonetheless, we would not recommend that someone image with a severely aberrated PSF either (e.g., like LITE), as this will necessarily degrade their sensitivity as well.
Perhaps most importantly, we have clearly demonstrated that despite our losses, we are able to image a diverse range of specimens with negligible photobleaching or phototoxicity. Although we do not present the data, we have also performed live-cell single molecule imaging of genetically encoded fluorescent proteins with this microscope. This single molecule sensitivity is consistent with the eSPIM paper, as well as Xiang Zhang’s paper (which uses a polarizing beam splitter in the detection path that automatically throws away 50% of the detected light without accounting for other losses, 10.1038/s41592-019-0510-z). Thus, we consider the losses manageable, and we are also working on reducing them (e.g. reducing the number of relay lenses, e.g. via lensless optical scanning approaches).
Thus, why is none of this prior work discussed in the manuscript? The most charitable explanation is that the authors of this work are unaware of it, but I am worried that in their effort to “sell” their current technique the authors have purposefully ignored this previous body of work. An intellectually honest comparison of their method in comparison to these other, higher NA methods would improve this work – in particular the claim that they achieve the “highest lateral resolution in light sheet microscopy” would seem to be strictly false based on these prior LSFMs operating at higher NA.
As previously mentioned, we now discuss this literature in the manuscript. We did not purposefully ignore these papers in an effort to “sell” our manuscript. To suggest that we would ignore a subset of the literature in an effort to advance our own science at the cost of our colleagues is offensive. Instead, it was an honest error of judgment. When we wrote this manuscript, we were focused on lattice light-sheet microscopy, which is by many perceived as the dominant and ultimate light-sheet microscope for high-resolution imaging. While the reviewer is correct that single objective microscopes with high NA lenses have the potential to beat the resolving power of a lattice light-sheet instrument, lattice is much more versatile in the range of samples that are being imaged. As such, we were too narrowly focused on lattice light-sheet microscopy as the main competition.
Also, we made the semantic error of referring to our oblique plane microscope as a single objective light-sheet microscope. Regardless, we do not consider these techniques competitive in terms of instrument performance (again, diversity of specimens, speed, field of view, optogenetics…) when compared to our oblique plane microscope. We agree that the term highest resolution is as of now not warranted, and we have reduced the resolution claims accordingly.
2) Resolution claims are not backed up by the data, and do not show a significant advance over previous work.
a) The authors claim a “200 nm scale lateral resolution”. Their sole evidence for this claim appears to be the images of 100 nm beads that they deconvolve with the Richardson-Lucy algorithm. Using RL on beads as an exclusive resolution metric is at best naïve, and at worst misleading and wrong. RL is known to turn beads into points, i.e. regardless of the physical limits of the optical system it is possible to claim an arbitrarily good resolution if one over-deconvolves. How were the number of iterations decided and why are the convergence curves not shown? The numbers the authors produce raise other red flags -
We now provide orthogonal measures of instrument resolution, including Decorrelation Analysis and Fourier Ring Correlation Analysis, and they are largely in agreement with the previous measurements. Importantly, we report these values for raw and deconvolved objects including beads and intracellular targets. Nonetheless, we would also like to state that a large number of microscopy papers report RL deconvolved resolution values (e.g., Lattice, iSIM, ISM, diSPIM, …).
RL: While we agree that RL deconvolution can result in unrealistic resolution values for point sources, we clearly state in the manuscript that we perform the minimum number of iterations necessary (always less than 20) to achieve a resolution enhancement, which is considered acceptable by in the super-resolution field as appropriate (i.e. iSIM, ISM, and other confocal photon-reassignment techniques claim two fold resolution improvement, where a factor improvement comes physically from the small pinholes, and another comes from iterative deconvolution. If their resolution gain by iterative deconvolution was illegitimate, it would essentially put them out of the super-resolution field). And the 1.41 gain is significantly less than others in the field, including the ~1.9 axial resolution enhancements in Lattice Light-Sheet Microscopy. Additionally, we did perform the convergence analysis, but did not present the data in this manuscript as it is not as clear cut as we hoped it would be (see Author response image 1, the axial resolution started to level off after 15 iterations, but the lateral resolution was still shrinking). Thus, we used the empirical square root of two criterion, which is not uncommon in the microscopy field. But importantly, we provide the raw data too, so the reader can clearly judge what comes from raw performance, and how much from deconvolution.
We are however sympathetic to the issues with deconvolution, and we have reduced our claims based on the deconvolved values, as estimated by FRC and image decorrelation analysis. We now emphasize the performance over the entire field of view, which we think is impressively uniform. This is a more valuable advantage of our microscope than resolution gains in the range of tens of nm.
b) The authors deconvolve their beads until they reach a value of 189 +/- 6 nm. The cutoff frequency for the lens they use, assuming 500 nm light, is λ/(2 * NA) = 500 nm / (2 * 1.35) = 185 nm. At face value, the deconvolved numbers would then suggest, in the best-case scenario, that the authors achieve an NA very close to 1.35. Indeed, the number they report makes me wonder if they simply deconvolved their data until they got close to this theoretical value. Unfortunately, their own raw measurements suggest otherwise, as for the same beads in widefield mode (sans OPM detection) they report ~240 nm FWHM. Assuming that there are not additional aberrations in this primary lens, and the ~240 nm corresponds to the full 1.35 NA, in OPM mode the authors are relatively far from achieving diffraction-limited performance – they report 284 nm x 328 nm lateral FWHMs in the OPM configuration. Given that resolution scales with NA, these numbers would suggest (at best) an NA of 1.35 (240/284) = 1.14 in the direction normal to the scan, and an NA of 1.35 (240/328) = 0.99 in the direction along the scan. The combination seems to be no (or marginally) better than what is achieved in lattice light-sheet microscopy with a 1.1 NA lens.
It is well known that the Abbe diffraction limit is an optimistic estimate for resolution and real-world high NA imaging systems achieve ~20% less resolution, due to many factors (Strehl ratio <1, attenuation of marginal rays, modulation transfer function, etc.). Thus, we do not agree with the estimates of the effective NA based solely on the achieved resolution. If we calculate analytically the opening half angle that we use, the effective NA for detection is 1.28 (Appendix 3). We agree that our measurements underperform in the oblique plane microscope format. We noticed that if we tilt the tertiary imaging path to zero degree, the lateral raw resolution gets uniformly ~270nm, which is more in line with the estimated NA. This is after going through a lot of lenses, which verifies that the basic optical configuration and alignment is sound. That said, it also shows that the tilting of the tertiary imaging system causes some residual resolution loss and stipulates that lower inclination angles could be more advantageous for achieving the highest resolution. We discuss these trade-offs now more clearly in Appendix 2.
However, we do agree with the statement that the practical resolution we achieve is not dramatic: Fourier based analyses put our lateral resolution at 220+-23nm (FRC) and 251+-3nm (Image Decorrelation) for deconvolved biological data, lattice light-sheet microscopy reports a lateral resolution value of 250nm. Nevertheless, the fact that one can get similar or even better performance with a single primary lens, enabling much more flexibility on sample mounting, is an important and remarkable finding that allows us to image a much more diverse set of specimens. Lastly, we notice that despite our efforts to unlock even higher resolution in OPM, it is equally advantageous to increase the field of view and the resolution uniformity, as this permits imaging throughput. We can cover 180x180 micron with quite uniform resolution, which is after reconsideration, more valuable than arguing about a +/- 10 nm resolution gain. As such, we de-emphasized some of our resolution claims and shifted some attention towards the field of view. Notably, lattice light-sheet microscopy uses a detection lens that is not corrected for the visible, and does not use a proper tube lens, which limits its useful field of view.
c) The authors argue that OPM occupies a unique niche among LSFMs, in its ability to easily integrate into existing optical microscopy workflows. I am sympathetic to this argument, but am not convinced that the technique described here is conceptually new or even advantageous compared to the previous state-of-the-art in this field: Yang et al.5. As the authors correctly point out, the conceptual advance realized by Yang is that sticking a piece of high refractive index material between secondary and tertiary lenses enables better performance than previous OPM. Using a bespoke objective instead of the awkward coverslip/water construction used in Yang is definitely a step in the right direction towards making the method more practical and commercially viable but is not a conceptual advance. On top of this, Yang reports pre-deconvolution bead-based FWHM values of 316± 8nm and 339± 18 nm laterally and 596 ± 32 nm axially. The lateral values are ~10% worse than those reported here, which is sufficiently close that I have a very difficult time believing that the authors' work represents a notable advance over the previous work in Yang et al. i.e. the minor improvement they report in pre-deconvolution FWHM values is unlikely to make any qualitative difference to the kind of data/insight obtained against this previously published paper, nor does it justify a quantitative jump from “300 nm scale” to “200 nm scale”. Moreover, the axial FWHM value reported in Yang is ~25% better than the 823± 31 nm number reported here. Taken together these comparisons would suggest that this work represents a rotation in parameter space rather than an advance.
We are intimately familiar with the eSPIM (Bin Yang is also a co-author on this manuscript), and we consider our manuscript to be a major advance in making the eSPIM design routinely useful for a much larger regimen of specimens. Indeed, our field of view is 3.7-fold larger than that of eSPIM, which takes the system from a dedicated single cell imaging system to one that can handle neurons, embryos, and even tissues. We also introduce the first optogenetic module that provides the level of control necessary to stimulate cells with arbitrary 2D patterns synchronously with volumetric imaging. While our resolution is similar, our detection path is much more apochromatic, and our alignment much more robust. We can also image deeper into a specimen owing to the larger working distance of the primary objective. We now detail all of the advantages between our system and eSPIM in Supporting Note 1.
We have not focused in the initial manuscript on axial resolution and used a rather low NA setting, which would fit a wide variety of samples. However, for the sake of exploring what axial resolutions are possible, we systematically varied the effective excitation NA. We can achieve the same axial raw resolution as in Bin Yang’s work using a Gaussian beam, albeit we note that this mode is best for shallow, adherent cells (applies both to eSPIM and our work).
d) Perhaps most remarkably, the authors make no attempt to verify “200 nm scale” in any of their images, e.g. by empirically investigating if they can separate biological features at this distance or by running Fourier-based methods that provide pixel-based resolution maps6,7. Doing so would go a long way to clarifying if indeed they have achieved “200 nm scale” resolution vs. something more like “250 nm-scale” or even “300 nm scale” resolution in biological samples. Ideally these calculations would be performed with and without deconvolution, so that the effect of deconvolution is cleanly separated from the “raw” imaging performance. Speaking of raw performance, it is unclear to me which of the presented data is deconvolved vs. raw. Do the authors present any raw data anywhere in this paper? This point needs to be clarified.
It is commonly accepted in the field that careful analysis of sub-diffraction beads is the best method to evaluate instrument performance. Evaluating biological targets for which no ground truth exists is potentially problematic. Nonetheless, we now provide Fourier-based analyses (FRC and image decorrelation analysis) of instrument resolution with and without deconvolution on a biological specimen. We also now provide images of beads before and after deconvolution (Figure 2), as well as a non-deconvolved image of the ER and a Drosophila Embryo (Figure 3—figure supplement 1 and Figure 8—figure supplement 1.). We also show the optogenetic data in its raw form, including top and cross-sectional views. We also clearly state which data is in raw and which in deconvolved form in a table in Supplementary file 1.
e) What is the pixel size resulting from the imaging system, and the data presented here? It is concerning that all data, with the exception of the beads, seems to have been at least displayed in 115 nm x 115 nm x 100 nm voxels. Nyquist alone would dictate that “200 nm scale” lateral resolution is not possible.
The first two dimensions in the voxel (115 nm) are physically set by the magnification of the optical train. The third dimension (100 nm) is set by the step size of the laser scan. In this manuscript, we Nyquist sampled according to our raw resolution. For deconvolution, the bead data, and biological data where resolution was quantified, was interpolated onto a finer grid prior to deconvolution. Without resampling, the maximum resolution (i.e. where critical Nyquist sampling is achieved) is 230 nm. We did not resample most of the biological data before deconvolution, because its massive size would become limiting. The term “200nm scale resolution” has been removed from the manuscript.
3) The axial resolution values the authors report are significantly worse than state-of-the-art multiview LSFM techniques8-11, which can achieve up to a near twofold axial resolution enhancement compared to their (likely) over-optimistic deconvolved values of 570 nm. These references need to be cited as they set the bar as far as axial resolution, not the lattice light-sheet system. Speaking of axial resolution, why is the axial FWHM not reported in addition to the lateral FWHM measurements in Figure 2—figure supplement 5? Please provide all XYZ measurements before deconvolution, so the “raw” FWHM values are evident for the interested reader. This will help avoid confusion about likely over-aggressive deconvolution.
We have not optimized axial resolution in the first version of the manuscript, but rather used a low NA light-sheet to cover most samples. If we increase the light-sheet NA, we achieve an axial raw resolution of 587± 18nm, which is in line with eSPIM and a better raw resolution value than ever reported for the widely used lattice light-sheet microscope (in the most commonly used square lattice mode). We now report the axial raw resolution for different settings of the light-sheet. We feel that these values are competitive with the main other technique used to image such samples as presented in the manuscript, the lattice light-sheet microscope.
We have provided in Figure 1 raw XYZ measurements on FWHM, as well as the deconvolved values. In the new Figure 2, we now also provide raw XYZ resolution values across the field of view, and supplement them with an Image decorrelation measurement. Further, we report FRC and image decorrelation analysis for the lateral resolution before and after deconvolution as well. I think this is by far more forthcoming than many other manuscripts, including Multiview LSFM techniques, which only report resolution after iterative deconvolution. We believe that the reader can get a clear picture how much is gained in raw resolution with our setup, and what is gained by deconvolution. We feel we have been forthcoming and transparent about our resolution values, much more so than any other publication in the light-sheet field that we are aware of.
While multiview techniques do achieve excellent axial resolution, at least two objectives have to interface with the sample. It is in this manuscript the explicit goal to have only one objective interfacing with the sample to enable broader applicability of various samples. You simply cannot fit a multi-well plate or other bulky sample holders in the mm-sized sample volume of a multiview microscope.
We now explicitly mention the multiview methods in the Discussion: “In addition to its ease of use, the OPM describe here delivers spatial resolution that is on par with Lattice Light-Sheet Microscopy, albeit with a larger field of view and a faster volumetric imaging capacity. And unlike state-of-the-art multiview LSFM techniques that achieve a slightly better axial resolution, only a single imaging perspective is needed.”
It is worth mentioning that these multiview LSFM techniques necessitate the same RL algorithm for image fusion that is the source of much consternation here. For example:
Wu et al. (10.1038/nbt.2713) – Converges after 30 iterations.
Guo et al. (10.1038/s41587-020-0560-x) – New back projector that converges in ~10 iterations.
Wu et al. (10.1038/s41467-017-01250-8) – Reflective imaging geometry, report between 20 and 100 iterations of RL in Supplementary file 3.
Reviewer #3:
The authors report a single objective light sheet (SOLS) microscope that provides unprecedented numerical aperture for light sheet imaging. The microscope utilizes a recently available solid-immersion objective lens to re-image an appropriately relayed tilted image plane onto a camera while encoding high NA information at smaller ray angles, thus allowing their capture without steric interference between the remote objective pair. This represents a natural progression from the eSPIM (Yang et al., 2019), which first illustrated the use of ray compression in SOLS. The application of this technology to several applications that would otherwise be extremely challenging in a light sheet/live-imaging context follow and illustrate that the microscope can image traditional sample preparations, in microfluidic chips and with high volumetric image rates. These applications provide an appropriate showcase for the technology and promising routes for further investigation without providing substantive biological conclusions. Given the technological focus of the article, I believe the article nonetheless warrants publication in eLife when the issues below are addressed. More generally, I believe this work and the recent developments on which it is built, will be of the utmost significance to the cell biology community in the coming years.
The following major points should be addressed before publication:
1) Having said that this is a technologically focused article, at least in terms of the innovation in the approach, I actually felt that the article was lacking in this regard. I am aware that much of the development around the solid-immersion objective has been reported via non peer-reviewed sources (GitHub etc.). However, as the first peer-reviewed study to utilize this technology, a more thorough description of the system would nonetheless seem warranted, particularly with regard to optical simulations, aberrations/useable viewing field (e.g. comparing simulations with observations) and a discussion of the various tradeoffs etc. This may seem outside the scope of a typical eLife publication and the manuscript as prepared but without it, the greatest achievement of this work (the microscope development, not the biology in this case) is sidelined. The authors are uniquely placed to be able to provide this information.
We agree that a more thorough description of the solid-immersion objective would be advantageous. We have detailed some design choices and their consequences in Appendices 1-3. However, in the spirit of supporting the open scientific ideals of Andrew York and Alfred Millett-Sikking (which we believe are well-aligned with the mission of eLife), we have agreed to let them keep many of these details on their GitHub page and to cite it accordingly as a Zenodo submission. Additional complications arise from non-disclosed proprietary information (that even the corresponding authors are not privy too) regarding the tertiary objective per an agreement between Calico, ASI, and Special Optics. Nonetheless, we now provide a bullet list of these differences in Appendix 1.
2) The authors include several comparisons with lattice light sheet microscopy that could be supported better with corresponding data and expanded discussion. This is noted with regard to the comments below. Similarly, I believe that the authors are uniquely able to provide the additional information via simulations and data that likely already exists e.g. fluorescent bead PSF measurements (e.g. Chang et al., 2019.) The authors would also benefit from comparing against some other reports in high-resolution light sheet microscopy e.g. Bessel beam (one/two-photon excitation, axially-swept light sheet microscopy).
The reviewer is correct that we are uniquely suited to be able to provide this data, but wanted to note that our Field Synthesis work was performed with a pair of NA 0.8/40X objectives, and not the much more expensive objectives used in the original lattice manuscript. As such, a comparison between our NA 1.35 and a NA 0.8 objective would not be fair. Also performing such a careful analysis is no small undertaking, especially during a pandemic. Nevertheless, we feel strongly that by providing the raw and deconvolved data for each microscope technique we have published, we have left the readers a more transparent way to compare optical performance between the different microscopes. Raw resolution information has been completely absent in the seminal lattice light-sheet publications, and it is our hope that a recent publication from our lab (Chang et al., 10.1364/OE.400164) is sufficient to shed some light on this topic.
Introduction : The authors note that previous LSFM implementations require mounting either on a small cover glass (e.g. lattice light sheet) or in agarose tubes (more classical SPIM-type systems). However, the authors fail to note that in the latter case, this is actually a superior mounting method e.g. for developing embryos as the soft surrounding environment promotes normal development (e.g. Kaufmann et al., Development, 2012), whereas a hard glass interface does not. I appreciate that these applications are outside the scope of what is presented here but I still think further clarification is warranted and that the desire to use traditional preparations is rather an issue for cell cultures and high-throughput imaging. The authors should nonetheless note that coverslip mounting may produce biological artifacts in the context of their earlier studies using collagen gels for mounting with a long working distance SPIM approach (Welf et al., Dev. Cell, 2016).
We have adopted this recommendation.
Results: Please comment on the choice of angle and the trade-offs associated with the 0 – 45 degree tilt.
Appendix 2 now discusses this. In short, you want to tilt as little as needed for best resolution and light-collection. However, the shallower the light-sheet, the longer it needs to be and the thicker it gets. Further, there will be less numerical aperture for the light-sheet generation remaining. A reasonable criterion we have come up with is to require that you find a tilt angle that leaves enough numerical aperture for the light-sheet left such that one can create a beam waist that is as large as the axial width of the detection PSF. For our system, this tilt angle turns out to be pretty close to 30 deg.
Subsection “Instrument Characterization”: Can you comment on how this compares to eSPIM (Yang et al., 2019). The lateral resolution appears to be slightly superior but the axial resolution seems to be lower (assuming a Gaussian beam for eSPIM). Is this a result of a longer thicker light sheet or a difference in how tilted the light sheet is with respect to the detection axis? This is surprising given the additional ca. 0.2 NA provided by the reported system.
We have estimated in the original submission the numerical aperture of the light-sheet illumination. We have now added a camera conjugate to the pupil plane for better estimation of the numerical NA of the light-sheet. We have systematically varied the NA of the light-sheet and we can match the axial resolution reported by Bin Yang. However, this comes at the expense of volumetric coverage, i.e. a smaller range in the z-direction (normal to the coverslip) can be covered with thinner light-sheets. This is however not different to Yang’s work.
and an 823 nm axial resolution?
Thank you for pointing this out.
Subsection “Instrument Characterization” final paragraph: Can you comment on the treatment here. Taking the efficiency simply as NA2 gives ca. 1.5, 1.9x respectively. This is without taking into the account the small NA loss relative to the 1.35 provided by the Si-immersion objective. Nevertheless, this analysis is welcome and generally underreported in single-objective light sheet microscopy.
We have addressed this now more carefully in the Appendix 3. The effective NA is conservatively estimate as 1.28. As such, the efficiency is only 35% higher than the NA 1.1. lens. This is a lower bound that neglects the slightly higher NA that is available in the direction normal to the tilt.
Figure 2 a) Is there a reason to choose a different color lookup table for this image?
This lookup table was selected as it allows visualization of both bright and dim structures. We now clearly state this in the figure legend.
Discussion second paragraph: Repetitive with regard to the Introduction.
We agree, but Introductions and Discussions often have some redundancy.
“…it has the highest lateral resolving power (~200 nm) in light-sheet
microscopy. Indeed, owing to the narrow depth of focus provided by the optical design, the axial resolving power” This is the deconvolved resolution. It would be helpful to include PSFs from light sheet microscopes and report both the raw and deconvolved resolutions in each case. The authors previous efforts regarding field synthesis should yield the required data (Chang et al., 2019).
We now provide PSFs in the raw and deconvolved state. While we agree that it would be helpful to include PSFs from other microscopes, as previously stated, that requires substantial work, for which we currently lack the time in the lab due to the COVID pandemic.
Regarding contrast and axial resolution, which are tied together, the lattice light sheet supposedly maintains a thin profile over a larger distance than a purely Gaussian light sheet. The authors should report on the useable field of view when noting comparisons of contrast and resolution (and whether this is dominated by the requirements of a thin light sheet or rather the detection optics/remote focus scheme). The Gaussian sheet should produce better contrast at the focus whereas the lattice trades some contrast for field of view. However, the high detection NA in the SOLS case does mean that the light sheet will be thicker at waist than the depth of field and so contrast is also lost here. Again, the authors are uniquely placed to make these comparisons in a more rigorous manner. At the minimum, it would be helpful to report the length and thickness of the light sheet produced at the 0.2 – 0.3 NA (reported in the Materials and methods), to provide some measure of the effective field of view and contrast for the lattice and SOLS cases.
At the moderate NAs used for illumination with our OPM, Gaussian beams and the square lattice from Lattice Light-Sheet Microscopy (the most commonly used variant) have indistinguishable propagation lengths (in terms of FWHM and confocal parameter) and beam waists (see Chang et al., 10.1364/OE.400164). Thus, we would expect that a Gaussian beam actually provides superior contrast for a given field of view since it does not have side lobes structures that result in out-of-focus illumination.
We now provide measurements for the propagation length and NA of the illumination beams used in Figure 2. As can be seen at the highest illumination NA (0.34), the beam waist is evident in the axial resolution for beads suspended in an agarose gel. Here, the axial resolution is dominated by the light-sheet thickness. As the illumination NA decreases and the beam waist grows larger than the depth of focus, the axial resolution eventually approaches ~800 nm.
With regard to laser power requirements and synchronous multicolor imaging, this is largely a result of how lattice light sheets have typically been generated. The authors own previous research regarding field synthesis (which should be cited) has reduced these issues for anyone who may be choosing whether to construct a SOLS or lattice light sheet system. The primary issue remaining is the lack of isolation between the objectives and media and the steric issues.
Owing to objections from reviewer #2, we have removed the comment regarding laser power. However, we have adopted your recommendation that Field Synthesis should be cited, as well as mentioning the sterics and isolation of the media.
Subsection “Data Post-Processing”: Please comment on the time taken and computational resources required for shearing and deconvolving the datasets. Since the SOLS system appears a good fit for imaging facilities, the associated throughput rate is a consideration.
The research code we are using can clearly be optimized, and we believe that it may be possible to shear and deconvolve the data in real-time using GPU. Nonetheless, we now provide an approximate time for shearing and deconvolution in the data post-processing section.
[Editors’ note: what follows is the authors’ response to the second round of review.]
Reviewer #3:
The authors have made substantive changes to the manuscript following the first round of reviews. I remain unswayed in my opinion that this work is suitable for publication in eLife given required improvements. I believe that the changes made have improved the manuscript.
In particular, the assessment of the system PSF and resolving capability has been greatly improved by including the non-deconvolved PSFs, details of the deconvolution and Fourier ring correlation estimate of resolution. The discussion of true single objective light sheet systems and comparison with eSPIM places the reported technique in context and I believe helps to make its case far more than detracting from it. The appendices should go some way to assisting others in building such a system.
There are still some changes based on my suggestions, which under ideal circumstances would be adopted. However, the author's arguments for not having done so are well reasoned. My primary concern was the frequent comparison with lattice light sheet microscopy. Here I still find the manuscript somewhat lacking but do appreciate that the authors field synthesis lattice light sheet system does not use the typical configuration of objective lenses making comparison difficult. The benefits of the reported system run deeper than pure resolution claims in any case and given the author's recent publication (Cheng et al. 10.1364/OE400164) reporting on the Gaussian to lattice light sheet comparison the omission here is reasonably justified.
All things considered, I would recommend the article for publication.
We would like to thank the reviewer for once again taking the time to carefully evaluate our manuscript. In an effort to accommodate your last request, we now provide PSFs obtained by Dr. Talley Lambert at Harvard Medical School on a Betzig-derived Lattice Light-Sheet Microscope. As can be seen in Figure 2—figure supplement 3, and the quantitative measurements in the corresponding figure caption, the two PSFs are comparable.
Reviewer #4:
First, I wish to commend the authors for their thorough revision. They have addressed many of my concerns and I feel better about the overall manuscript. The emphasis on field of view over resolution, the Fourier-based methods of assessing resolution, the citing of previous work, the additional details put forth in the appendices, and more accurately referring to their method as an OPM have all significantly improved the work. Here are my remaining concerns/comments:
a) The authors now more fairly describe the advantages of their method over previous similar microscopes. They still do not clearly state what I see as weaknesses of their method, which seems important given the extensive comparisons to LLSM and other systems:
i) relatively low sensitivity. In their revised manuscript, the authors now seem to admit that the advantage in light efficiency offered by the 1.35 NA lens is less than optical losses through their emission path. In sum, these losses would seem to imply a strictly lower sensitivity than the more “conventional” LLSM with 1.1 NA detection. Please discuss this point clearly in the main text, rather than relegating it to Materials and methods.
We have moved this back into the Microscope Design portion of the main text. The text now reads “Owing to the large number of optics, spurious reflections resulted in a 59% and 44% decrease in fluorescence transmission for laser scanning and stage scanning variants of the microscope, respectively, at 30-degrees. Transmission improved slightly (3%) when the optical train was arranged at 0-degrees.”
ii) Low effective NA vs. theoretical NA. I appreciate the appendix describing the NA calculations, but I am still not convinced that the number presented is an “effective NA”, i.e the NA of an equivalent widefield with primary lens using this NA. In particular, I am still bothered by the apparent loss of resolution compared to their widefield performance, which I commented on previously and repeat here:
“Unfortunately, their own raw measurements suggest otherwise, as for the same beads in widefield mode (sans OPM detection) they report ~240 nm FWHM. Assuming that there are not additional aberrations in this primary lens, and the ~240 nm corresponds to the full 1.35 NA, in OPM mode the authors are relatively far from achieving diffraction-limited performance – they report 284 nm x 328 nm lateral FWHMs in the OPM configuration. Given that resolution scales with NA, these numbers would suggest (at best) an NA of 1.35 (240/284) = 1.14 in the direction normal to the scan, and an NA of 1.35 (240/328) = 0.99 in the direction along the scan. The combination seems to be no (or marginally) better than what is achieved in lattice light-sheet microscopy with a 1.1 NA lens.”
Ideally, the resolution of our microscope would be identical to the resolution of the microscope objective when operating in a widefield imaging mode. However, this expectation seems unrealistic since we are using the objective under such non-ideal conditions. Indeed, we are imaging above and below the nominal focal plane of the primary objective and using physically imperfect optics to create a replica of this fluorescence above and below the nominal focal plane of secondary objective. As previously stated, real lenses rarely reach the full resolution predicted by ideal models. Thus, some degradation should be expected.
It is very reasonable to suppose that NA would scale with the apparent size of a subdiffractive object, regardless of suboptimal Strehl ratio, MTF, hardware etc. And if the NA of the primary lens is effectively less than 1.35, this would seem to further diminish the “effective” NA that is achieved in this manuscript. This point is important to address properly, because it has implications both for comparisons to other microscopes and it suggests that the deconvolved numbers the authors present are still overly optimistic (e.g. if the NA is not in fact 1.28, physically it does not seem possible to achieve a deconvolved resolution of ~203 nm as in Figure 1H). I would suggest replacing “effective NA” with “theoretical NA” everywhere the former is referred to, and being very clear in the manuscript that there is still residual resolution loss: i.e. the theoretical NA is not achieved at the tilt settings used in this paper. The authors could prove me wrong by providing a suitably compelling measurement of the “effective NA” rather than the theory – presenting a theoretical argument does nothing to convince me that in fact they achieve this value, which their own measurements seem to contradict.
We understand the reviewer’s concern and have adopted the term theoretical NA throughout the manuscript. We also state in the main text that the system performs closer to the expected NA when reducing the tilt angle of the tertiary imaging system, and that there is some residual resolution loss when significant tilt is introduced in the tertiary imaging system. In the instrument characterization section, the text now reads “Of note, the choice of illumination angle is accompanied by trade-offs in light-sheet thickness, imaging depth, detection efficiency, and resolution (Appendix 2). Indeed, we observed a gradual loss in NA and thus resolution as our tertiary imaging system was adjusted from a 0 to a 30-degree tilt.”
b) Regarding deconvolution, I am still having difficulty understanding how ~203 nm resolution is possible with 115 nm voxels. Once more, Nyquist alone would seem to render this number incorrect. The authors' argument in using a factor 1.414 seems to boil down to: “other people have done it, so we should do it too”. This is not a good argument in my opinion, as it leads to a result that contradicts common sense. The authors digitally upsample the bead data before deconvolution, but I do not see how such digital upsampling could effectively beat Nyquist.
One interpretation of iterative deconvolution is that it makes an image sharper by extrapolating low frequency information into a higher frequency space where previously there was little or no information. By zero-padding the data, we are essentially introducing empty frequency space that the iterative deconvolution routine can extrapolate into. Importantly, the FRC data supports this view. Here, the frequency support where meaningful information is contained cuts off sharply, and the iterative deconvolution extends this domain to higher frequency values. Thus, it does not matter if that space into which sample information is extrapolated is created by zero-padding, or by finer spatial sampling (which also creates void frequency space, albeit with some white noise). This is supported by work by Dr. Rainer Heintzmann which shows that one can get ~1.5x resolution “out of band” with Richardson Lucy owing to non-negativity assumptions and iterations in real space (DOI: 10.1016/j.micron_2006.07.009), depending on the sample and imaging conditions. Please also note that diSPIM by necessity makes use of the extrapolation capability of RL deconvolution. The physical coverage in reciprocal space from two views leaves two large missing cones in diagonal directions. Instead of filling these voids physically with information by acquiring views from other orientations, they are filled by RL deconvolution. However, while diSPIM never showed that this out-of-band information is valid, we show here in our work evidence using the Fourier ring correlation.
Overall, we think that this discussion about what RL deconvolution can and cannot do is reserved to specialized journals and publications such as the one by Rainer Heintzmann. We mention this now in the Materials and methods section.
The authors in their rebuttal seem to suggest that presenting the raw values sans deconvolution is unusual in multiview LSFM (“I think this is by far more forthcoming than many other manuscripts, including Multiview LSFM techniques, which only report resolution after iterative deconvolution”). I am not sure what they are talking about: Wu 2013 (diSPIM), Keller 2015 (IsoView), classics in this area, state these values clearly in the main text.
We apologize for the miscommunication. The raw resolution for each view is often reported in multiview LSFM manuscripts, but the fused resolution is often only reported after Richardson-Lucy deconvolution. This likely results from the fact that without deconvolution, 6-8 views are necessary to sufficiently sample Fourier space (see DOI: 10.1364/OE.15.008029). Nonetheless, these manuscripts are insightful:
Wu et al., 2013:
Axial resolution (Supplementary Table 1) – 1.47 microns raw, 0.8 microns deconvolved (a factor of 1.83).
Arithmetic fusion (Supplementary Table 1) – 0.98 microns raw, 0.62 microns deconvolved (a factor of 1.58).
Joint fusion and deconvolution (Supplementary Table 1) – 0.33 microns axial resolution.
Acquisition (see Online Methods) – “In each 3D stack, 50 or 100 xy planes separated by az step of 1 micorn of 0.5 microns were imaged”
Upsampling (see Online Methods) – “ViewA is upsampled (i.e., coarsely sampled axial pixels were linearly interpolated to obtain an isotropic voxel size of 0.1625 x 0.1625 x 0.1625 microns3”.
Conclusion – Wu et al. upsample their data from 0.5 microns or 1 micron to 0.1625 microns and report a resolution of 0.33 microns. Their deconvolution is much more aggressive than ours.
Chhetri et al. (not Keller), 2015:
Axial resolution (Supplementary Figure 6) – 3.01, 2.95, 2.63, or 3.34 microns, depending upon the imaging axis.
Arithmetic fusion – Not provided.
Multiview Deconvolution (Supplementary Figure 6) – Report a resolution of 410, 420, and 450 nm.
Acquisition – Lateral voxel size of 0.4 microns reported. Axial dimension not reported.
Upsampling (Methods, IsoView multiview image registration) – “In the first step of our registration procedure, we perform a coarse image alignment. After cubic interpolation for generating voxels with isotropic size…”
Conclusion – Chhetri et al. upsample their data and report a resolution of ~400 nm. Also appears that their deconvolution is much more aggressive than ours.
As such, we believe that your concerns about extending beyond Nyquist sampling should also be applied for their deconvolved results. But as stated before, this is not contested by theoretical work on iterative deconvolution.
c) The authors seem to acknowledge in their rebuttal that they do not achieve diffraction-limited performance (“we agree we do not achieve diffraction-limited performance”), but I still find multiple places in the manuscript where this is stated or implied. Please address this point.
We no longer state or imply that we achieve diffraction-limited performance throughout the text with a few exceptions. In Appendix 1, bullet point 2, we clearly state that the FOV is theoretically diffraction limited (not experimentally diffraction limited), which is accurate. Same with bullet point 3. The remainder of the text now states “near-diffraction limited”, which is accurate, and backed up by our wavefront measurements (Strehl = 0.97 at 0-degrees, and 0.91 at 30-degrees, see Figure 1—figure supplement 2).
d) Depth. What sets the effective depth limit of this technique, as presented here? 25-30 μm is relatively modest for a lens with a 300 μm working distance, so discussing theoretical and practical limits would be useful for a biologist who is searching for a practical solution for their sample.
The effective imaging depth depends on two separate factors. The first and most intuitive factor is the working distance of the primary objective, which as you state, is 300 microns. The second factor is the distance over which the remote focusing system properly functions, which influences extent of the W dimension in Figure 1. In both the linear and tilted configurations, we achieved ~60-80 microns of high-quality remote focusing, and this allowed us to recently image through the entire tail of a zebrafish. Importantly, one can still tile the data in the Z direction within the working distance of the objective, provided the sample has not a significant refractive index mismatch. In an effort to communicate this topic, we now clearly state these practical limits in the main text of the manuscript. Specifically, it reads in the instrument characterization section that “Theoretically, the maximum imaging depth of our remote focusing system is 60 microns, beyond which tiling in the Z-dimension can be performed until one reaches the working distance of the primary objective (300 microns)”.
e) The authors state numerous times that they achieve resolution on par with or better than LLSM. Please explicitly state the corresponding values reported in LLSM (or measure them with an LLSM with 1.1 NA detection), where relevant.
The seminal lattice manuscript, as well as many that followed, provide insufficient evidence to back up their resolution claims. There is however one exception (DOI:10.1038/nature22369) that clearly states that their lateral and axial resolution varies between 294-370 nm and 649-947 nm, respectively, depending on the laser illumination wavelength. Likewise, we recently published a manuscript (DOI:10.1364/OE.400164) that exhaustively evaluated light-sheet properties (thickness, confocal parameter, resolution, OTF, etc.) for both Lattice and Gaussian-based light-sheets. Cumulatively, these two manuscripts clearly show that we achieve resolution that is on par or better than LLSM. Nonetheless, we now provide further evidence of this in the form of PSFs acquired on a LLSM by Talley Lambert at Harvard Medical School (a system built by Betzig’s group). Again, these data show that we are indeed on par, or better than, a LLSM.
The text now states: “And for an oblique illumination angle of 30 degrees, these raw axial resolutions are similar or better than the 666 nm axial resolution reported for the square illumination mode of Lattice Light-Sheet Microscopy”.
“By comparison, the raw lateral resolution for Lattice Light-Sheet Microcopy for a GFPlike fluorophore is 312 nm…”.
f) “And unlike state-of-the-art multiview LSFM techniques that achieve a slightly better axial resolution, only a single objective and imaging perspective is needed (Guo et al., 2020; Wu et al., 2013).” Wu 2013 achieves ~350 nm axial resolution, Wu 2017 achieves ~300 nm axial resolution, and Guo 2020 enhances the effective axial resolution of LLSM to ~380 nm resolution. These are not “slightly better”, they are significantly better (indeed, the degree of improvement is more convincing than what is shown here relative to LLSM). Also, as I think the authors acknowledge in their rebuttal, they do not use a single objective, rather they use 3. Please remove “slightly” for accuracy and/or rethink this sentence.
As stated previously, multiview light-sheet microscopes only achieve axial resolutions better than ours after image fusion and deconvolution. As we clearly show, our best raw axial resolution of 580 nm is significantly better than the 1.47micron raw axial resolution reported by Wu et al., 2013. And, these manuscripts are much more aggressive with their deconvolution than we are.
But in an effort to be constructive and not endlessly debate what resolution gains come from iterative deconvolution, maybe let us discuss what is physically possible before any deconvolution. The highest axial raw resolution for light-sheet before deconvolution is about 380nm (Dean et al., 2015), our best resolution is 200nm worse. Dean et al. used an NA of ~0.8 for light-sheet generation, ours is less than half of that. We can agree that when a single objective is used for both illumination and detection, you have less angular range available than in a dual or multi-objective geometry.
The reviewer concern that we use three objectives is again semantics in our opinion. In our setup, there is only one objective that interfaces with the sample. As mentioned before, it is clear that when one uses multiple objectives to illuminate and detect the sample, then one can get more access from a greater angular range, at the expense of a complicated sample interface.
In an effort to communicate this more effectively, we now state that “And unlike state-of-the-art multiview LSFM techniques that achieve a better axial resolution after image fusion and deconvolution, only a single imaging perspective is needed.”
Reviewer #5:
In this manuscript, "A versatile oblique plane microscope for large-scale and high-resolution imaging of subcellular dynamics", Sapoznik et al. describe a variant of an oblique plane microscope (OPM) that makes use of a custom designed tertiary objective to capture a greater portion of the emitted rays through the oblique refocusing objectives. OPM combined with light sheet is attractive because it offers the potential for optically sectioned, low phototoxicity imaging using only a single primary imaging objective with 180 degree physical access. As the authors note, this concept itself is not new and many descriptions and implementations have been described before (originally with Dunsby in 2008 and more recently with Bouchard 2015 and Yang 2019). Thus, the primary technical innovation here is the implementation of the custom-designed solid immersion lens.
In general, I think this is a useful addition to the field and could be suitable for publication in eLife. An open top-light sheet microscope with increased sample accessibility is indeed useful. However, in its current form, the manuscript reads more like an aggressive sales pitch rather than a balanced discussion of the pros and cons of different microscope approaches. The presentation is over-dismissive of prior work, often mis-representing or cherry-picking specific comparisons to make this current instrument appear better, while neglecting to mention or discuss trade-offs wherein the instrument might perform worse. I noticed that the authors disclose financial relationships with companies that sell products that compete with commercial versions many of the instruments compared here. However, I feel that the readers of eLife would benefit if the authors focused on more thoroughly documenting their own scientific contribution, together with its trade-offs, rather than (often incompletely or inaccurately) characterizing prior work from others.
We appreciate that you believe that this is a useful addition to the field and agree with your sentiment that open top light-sheet microscopes are indeed useful. It was not our intention to be over-dismissive of prior work, but rather to clearly delineate the differences between our technique and that of others, which was necessary to address many concerns raised by the reviewers in the first revision of this manuscript. Importantly, with regard to the OPM technology described here, we do not have any competing financial interests:
1) We have worked with Applied Scientific Instrumentation to make many of the optical element and made it easier for other labs to get started building their own OPM systems. Nonetheless, this arrangement has been purely pro bono. We do not hold patents on OPM or SCAPE technology.
2) Discovery Imaging Systems, LLC was formed in an effort to sell a cleared tissue light-sheet microscope based upon our Axially Swept Light-Sheet Microscopy (ASLM) technology. However, this venture is pending until the patent for ASLM is finalized.
3) The remaining of the competing financial interests are pharmaceutical in nature and can be wholly attributed to Drs. Carlos Arteaga and Ariella Hanker.
1) The resolution claims here are provided by FHWM measurements of isolated beads and of xy (lateral) measurements using correlation based approaches on the images. Importantly, optical sectioning is not discussed at all. In this instrument, the resolution is provided by the high NA primary objective, but due to the thick lightsheets (0.16 NA Gaussian beams) used in most of the measurements, the lightsheets are substantially thicker than the depth of field for the detection. This would be readily apparent as the "missing cone" in the optical transfer function and I suspect it would also affect correlation based axial resolution measurements if they were conducted on cells or embryos rather than isolated beads. In contrast, axial resolution in several other approaches (Gaussian, Bessel, Lattice etc) is obtained by using a lightsheet that is thin-compared to the detection depth of field. The authors' own prior work, Dean et al. 2015, have highlighted the advantages of this approach, but there is no mention of the relationship between optical sectioning and resolution here.
As this reviewer is aware, all light-sheet microscopes are accompanied by complex trade-offs that include illumination confinement, field of view, lateral resolution, axial resolution, sensitivity, speed, optical sectioning, etc… However, recent work has shown that the most widely used square lattices does not improve optical sectioning or axial resolution relative to Gaussian beams (DOI: 10.1364/OE.400164 and 10.1364/BOE.11.000008), which is why we chose to proceed with the latter.
In our view, the most promising methods to improve optical sectioning, axial resolution and field of view is either light-sheet generation with two photon Bessel beams, structured illumination, or using the ASLM principle, all of which come with intrinsic drawbacks. In principle, both methods are compatible with our microscope, but we have not realized these options in our microscope yet. Other light-sheet types (hexagonal lattice, 1PE Bessel, Airy beams) come with a significant loss of optical confinement.
For most biological imaging performed, we used a Gaussian beam with an effective NA of 0.16 and an anticipated thickness and confocal parameter of 1.2 and 36.9 microns, respectively. While this thickness is larger than the depth of focus, we would like to note that it is only ~1.33-fold larger. We now explicitly mention this in the instrument characterization as follows: “For most biological experiments reported here, we used an illumination NA of 0.16, yielding a Gaussian beam that has a thickness and propagation length of 1.2 and 37 microns, respectively. Importantly, because the illumination beam is thicker than the depth of focus of the detection objective, optical sectioning (e.g., the ability to reject out-of-focus fluorescence) is slightly reduced.”
FWHM measurements in isolation are a limited picture of resolution. The authors should present optical transfer functions in the xy, xz, and yz planes to demonstrate true resolution and to what extent their system fills in the missing cone. They should also perform the raw image-based correlation measurements in the axial dimension (as they already do now for the lateral dimension) on a range of biological samples. This could be easily done with existing data.
We now provide OTFs in Figure 2—figure supplement 1 and demonstrate that our system does indeed partially fill the missing cone. For comparison, we also provide a widefield OTF, which clearly shows the singularity at the origin and the missing cone of information. There is in our opinion a vast difference between a widefield OTF and any of the light-sheet OTFs that we show. The widefield system cannot reject any background, and hence has a singularity at its origin, which makes any deconvolution attempts difficult.
Owing to complexities that arose from the correlated noise distribution of modern CMOS cameras, computational shearing of the data and the parallelepiped shape of the imaging volume, Fourier Shell correlations turned out to be complex and problematically variable (we spent ~2 months working on this during the first revision cycle). Consequently, we provided Fourier Ring Correlation and Decorrelation Analyses of the lateral resolutions, and these results were largely in agreement with the FWHM and resolution metrics based in frequency space. Given this agreement between the to approaches of resolution measurements in the lateral dimensions, we do not see any reason why the axial resolution measured in real space should be inaccurate.
2) The authors describe that, due to the optical path required for OPM, the instrument loses 71% of the photons from the sample before they hit the camera. I.E. it has a transmission efficiency of 29%. This is in stark contrast to the standard SPIM systems which due to the simple widefield detection optics should operate at very close to the ~80-90% transmission efficiency of commercial objectives. Further, it's not clear at what imaging angle this measurement was performed. This information is also important because the efficiency decreases further as the primary objective is utilized further away from its design standard angle of 0 degrees. In the appendix, the authors describe how the 1.2 NA of this system would somewhat compensate for this poor transmission over the 1.1NA lens used in other variants, but this analysis assumes that all rays have equal transmission efficiencies which generally isn't true. Thus even without any additional optics, an OPM system operating with an off-centered pupil will have a lower transmission efficiency than an objective with the same effective NA operating on axis.
For the previous submission, we performed the measurement at a 30-degree tilt. We have more systematically repeated this measurement by carefully matching the diameter of the alignment laser to the size of the primary objective back pupil, with all of the optics and filters present in the optical path with the exception of the primary objective and the camera. The laser and stage scanning variants had a 41% and 53% transmission when oriented at 30 degrees. Accordingly, the scan lens, galvo, tube lens combo was responsible for the additional 12% decrease in transmission between the two systems. Placing the stage scanning variant at a 0-degree tilt only improved the transmission by 3% to 56%. This verifies the design predictions for our tertiary detection objective (i.e. capturing most of the light-cone emanating from the secondary objective under a moderate tilt angle). We now detail these measurements in the Materials and methods section.
Experimental characterization of how the transmission efficiency between a 1.2 NA objective operating with a centered pupil vs. a higher NA objective with an effective 1.2 NA due to pupil decentering with OPM and associated optics, and how this declines with the OPM angle would be extremely useful for the field. The authors are ideally poised to make these type of measurements.
Regardless, this is an important discussion, so I'm unclear why the measurements are not mentioned in the main text under "Instrument Characterization". They are instead provided in the fourth paragraph of Materials and methods section under "Laser Scanning Microscope Setup". Given that microscope end users are willing to pay thousands more for a back thinned camera with a 10% increase in quantum efficiency than a non-back thinned unit, the authors should mention that the OPM implementation here comes at the cost of reduced detection efficiency in the main text. The statement that the instrument is "sufficiently sensitive to detect single molecules (data not shown)" does not adequately address this concern. Especially when other approaches have clearly demonstrated their utility for live cell and super-resolution single molecule imaging with both dyes and fluorescent proteins alike. It's also not clear how the authors define "without obvious signs of phototoxicity". How was this measured and what is considered obvious?
In the final paragraph of the Discussion, we now mention the reduced collection efficiency. Also, we have removed the statements regarding the detection of single molecules and phototoxicity.
3) I have concerns about how prior methods are discussed. For example, Lattice Lightsheet Microscopy is not a single instrument, but a general description of the use of excitation patterns based on optical lattices for microscopy. The specific choices for excitation objective and detection objective as well as the type of lattice light sheet and extent of optical confinement used will determine the resolution and optical sectioning and can be chosen/optimized for specific applications. Thus, statements like "Lattice Lightsheet microscopy has a resolution of xx" or "requires a 5mm coverslip" are no more accurate than saying "Gaussian beam microscopy has a resolution of xx and requires agarose sample embedding". It all depends on how one decides to configure an instrument and for what purpose they choose to balance the trade-offs. It's fine to say that a specific publication reported certain values, but this should be accompanied by some context. In many cases, design considerations may have been chosen to make an instrument that is optimized for a different purpose or with additional features than the one presented here.
In theory, lattice light-sheet is a generic imaging approach that is compatible with a range of objective and thus sample preparations. However, in practice, it is built and sold in a manner that does require 5 mm coverslips. This is how it was implemented by Betzig, 3i, and over 120 different labs. Also, the way that Lattice light-sheet is described in the literature makes comparisons challenging. It can be operated in a structured illumination, hexagonal, and square illumination modes, each with different inner and outer NAs, beam lengths, thicknesses, etc. For the comparisons here, we are referring the square illumination mode that accounts for >95% of the use cases (see Table S1, DOI: 10.1038/s41592-0190327-9), and which we have performed careful measurements (DOI: 10.1364/OE.400164). We now try to specify that we are referencing the axial resolution of the square illumination mode (the lateral resolution remains unaffected). Also, the text now reads “Sample preparation is an additional problem, as the orthogonal geometry of LSFM systems often sterically occludes standard imaging dishes such as multi-well plates.”
https://doi.org/10.7554/eLife.57681.sa2Article and author information
Author details
Funding
Cancer Prevention and Research Institute of Texas (RR160057)
- Reto P Fiolka
National Institutes of Health (R00 GM120386)
- Jens C Schmidt
National Institutes of Health (R01HL068702)
- Douglas P Shepherd
- Rory Kruithoff
National Institutes of Health (R33CA235254)
- Reto P Fiolka
National Institutes of Health (R35GM133522)
- Reto P Fiolka
National Institutes of Health (K25 CA204526)
- Erik S Welf
National Institutes of Health (P30 CA142543)
- Carlos L Arteaga
National Institutes of Health (1R01MH120131-01A1)
- Kevin M Dean
National Institutes of Health (1R34NS121873)
- Kevin M Dean
National Institutes of Health (5P30CA142543)
- Kevin M Dean
Damon Runyon Cancer Research Foundation (DFS-24-17)
- Jens C Schmidt
Chan Zuckerberg Initiative (HCA3-0000000196)
- Yoshihiko Kobayashi
- Purushothama Rao Tata
- Douglas P Shepherd
Australian Research Council (FT190100516)
- Samantha J Stehbens
Rebecca L. Cooper Medical Research Foundation (PG2018168)
- Samantha J Stehbens
University of Queensland (RM2018002613)
- Samantha J Stehbens
Company of Biologists (JCSTF1903138)
- Robert J Ju
Robert A. Welch Foundation (I-1950-20180324)
- Konstantin Dubrovinski
National Institutes of Health (R01GM110066)
- Konstantin Dubrovinski
Human Frontier Science Program (LT000911/2018C)
- Jaewon Huh
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Acknowledgements
The authors would like to thank Dr. Dana Reed for her generous support, as well as the lab of Professor Joseph Hill for providing primary cardiomyocytes for calcium imaging, and Dr. Rosa E Mino, Dr. Madhura Bhave, and Dr. Marcel Mettlen for providing the ARPE Cell line tagged with AP2-GFP. We thank Tamara Terrones for her technical assistance in preparing primary neurons. Microfluidics were prepared in part at the Queensland node of the Australian National Fabrication Facility, a company established under the National Collaborative Research Infrastructure Strategy to provide nano- and microfabrication facilities for Australia's researchers.
Senior Editor
- Anna Akhmanova, Utrecht University, Netherlands
Reviewing Editor
- Melike Lakadamyali, University of Pennsylvania, United States
Reviewer
- Rory Power
Publication history
- Received: April 8, 2020
- Accepted: November 9, 2020
- Accepted Manuscript published: November 12, 2020 (version 1)
- Accepted Manuscript updated: November 16, 2020 (version 2)
- Version of Record published: December 1, 2020 (version 3)
- Version of Record updated: December 7, 2020 (version 4)
- Version of Record updated: February 1, 2021 (version 5)
Copyright
© 2020, Sapoznik et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Metrics
-
- 27,490
- Page views
-
- 1,198
- Downloads
-
- 63
- Citations
Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.
Download links
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Further reading
-
- Cell Biology
- Neuroscience
Prolonged exposure to loud noise has been shown to affect inner ear sensory hair cells in a variety of deleterious manners, including damaging the stereocilia core. The damaged sites can be visualized as ‘gaps’ in phalloidin staining of F-actin, and the enrichment of monomeric actin at these sites, along with an actin nucleator and crosslinker, suggests that localized remodeling occurs to repair the broken filaments. Herein, we show that gaps in mouse auditory hair cells are largely repaired within 1 week of traumatic noise exposure through the incorporation of newly synthesized actin. We provide evidence that Xin actin binding repeat containing 2 (XIRP2) is required for the repair process and facilitates the enrichment of monomeric γ-actin at gaps. Recruitment of XIRP2 to stereocilia gaps and stress fiber strain sites in fibroblasts is force-dependent, mediated by a novel mechanosensor domain located in the C-terminus of XIRP2. Our study describes a novel process by which hair cells can recover from sublethal hair bundle damage and which may contribute to recovery from temporary hearing threshold shifts and the prevention of age-related hearing loss.
-
- Biochemistry and Chemical Biology
- Cell Biology
Autophagy is an essential catabolic pathway which sequesters and engulfs cytosolic substrates via autophagosomes, unique double-membraned structures. ATG8 proteins are ubiquitin-like proteins recruited to autophagosome membranes by lipidation at the C-terminus. ATG8s recruit substrates, such as p62, and play an important role in mediating autophagosome membrane expansion. However, the precise function of lipidated ATG8 in expansion remains obscure. Using a real-time in vitro lipidation assay, we revealed that the N-termini of lipidated human ATG8s (LC3B and GABARAP) are highly dynamic and interact with the membrane. Moreover, atomistic MD simulation and FRET assays indicate that N-termini of LC3B and GABARAP associate in cis on the membrane. By using non-tagged GABARAPs, we show that GABARAP N-terminus and its cis-membrane insertion are crucial to regulate the size of autophagosomes in cells irrespectively of p62 degradation. Our study provides fundamental molecular insights into autophagosome membrane expansion, revealing the critical and unique function of lipidated ATG8.