1. INTRODUCTION

Mesoscopic fluorescence 3D imaging of model organisms in their entirety has emerged as an important imaging application in the life sciences. This encompasses monitoring systemic properties such as blood flow, every neuron in a behaving organism, or observing the metastatic cascade longitudinally in xenograft models. Moreover, advances in tissue clearing [1] and expansion [2] result in ever growing sample volumes that are rendered optically transparent. These applications require efficient 3D microscopes that possess rapid yet gentle volumetric acquisition capabilities and have sufficient spatial resolution, volumetric coverage, and optical sectioning capability.

Light-sheet fluorescence microscopy (LSFM) [3,4] has the right attributes for the aforementioned tasks, as it offers efficient 3D imaging combined with intrinsic optical sectioning, and in recent implementations also provides high volumetric acquisition rates compared to other imaging modalities such as confocal microscopy [5–7]. A large variety of LSFM instruments have been reported [4], some of which are cost effective and have been widely disseminated [8]. Normally, LSFM uses separate optics for illumination and fluorescence detection, which can complicate sample access for large samples or high-throughput applications, e.g., in multi-well formats [9]. As an alternative, open-top light-sheet geometries are gaining popularity [9–13], as they leave in principle “infinite” lateral accessibility for large tissue slices, series of organoids, or model organisms in multi-well plates. Among open-top architectures, oblique plane microscopy (OPM) is attractive as it employs only one primary objective for light-sheet illumination and fluorescence detection [14]. As such, the optical axis of the objective can be orthogonal to a coverslip or sample plate, a geometry for which objectives are usually designed and corrected for. In contrast, dual objective open-top LSFM architectures need aberration correction when interfacing with a coverslip at an off angle [13,15].

Another distinctive feature of OPM is its ability to rapidly image volumes by using galvo mirrors without mechanical motion of the sample or the objective. As such, 3D imaging at up to 300 volumes per second [5] has been demonstrated, as well as projection imaging at above 100 fps [16]. In contrast, conventional LSFM architectures use slower mechanical actuators to scan either the detection objective or the sample during a volumetric acquisition. With its speed potential, OPM has relevance for functional imaging in the neurosciences, especially with the advent of fast calcium and voltage sensors. Conceivably, mesoscopic OPM systems could monitor neurons across a sufficiently transparent brain or organism at 10–100 Hz, in volumetric or projection mode, respectively.

OPM can also be advantageously used in a continuous motion acquisition modality, where its galvo mirror stabilizes the image for stacking [17]. As such, multi-well plates containing organoids or model organisms can be screened volumetrically at high rates. In addition, a multi-angle projection mode [16] may provide different views of the 3D content at even higher rates. The review by Kim provides further detailed discussions of OPM, the involved technology, theory, and applications [18].

The concept of OPM, however, requires high numerical aperture objectives [14], as a large half opening angle is needed to give the light sheets sufficient tilt, and to be able to capture fluorescence light with the downstream optical train. The latter constraint is rooted in how the tilted light-sheet plane is imaged onto a camera in OPM: leveraging the principle of remote focusing [19], a distortion free and diffraction limited 3D image of the sample space is created by a secondary objective. When properly implemented, remote focusing maintains angles from the sample to the remote space; hence the fluorescence image of the light-sheet plane is tilted by the same angle as the light sheet emerges in sample space. A tertiary imaging system, whose focal plane overlaps with the remote image of the light-sheet plane, is then used to map the fluorescence onto a camera. Since the tertiary objective is tilted to the optical axis of the secondary objective, a light loss will occur, as the acceptance cones of the two objectives do not fully overlap. When using low NA objectives with a half opening angle below 30º, this light loss becomes total [20,21]. In principle, one can increase the numerical aperture, and hence the acceptance angle, of the tertiary objective to counter the light loss. However, there is an inverse tradeoff between field of view and numerical aperture [22]; hence this approach is not well suited for mesoscopic imaging. Ideally, all three objectives should possess similar fields of view to maximize volumetric coverage, and consequently all would have a similar numerical aperture.

Methods have been reported to “bend” light into the shallow acceptance cone of a tertiary objective in a mesoscopic OPM, such as reflective gratings [21], fiber face plates [20], or intentionally distorting the remote image space to make the image of the light sheet less inclined [23]. While they provide solutions for low NA OPM implementations of the detection arm, they all come with specific caveats. Gratings with fine linespacings compress the light cone in one spatial dimension, potentially lowering the resolution. In addition, higher diffraction orders may enter the tertiary objective in high numerical aperture scenarios. Fiber face plates impose micrometer-sized resolution limits based on their fiber spacing. Lastly, compressing the remote space introduces spherical aberrations, which will lower the system’s resolution.

Additionally, a low NA primary objective launches the light sheet at a steep angle in sample space. This lowers the optical sectioning ability and axial resolution, as the light sheet is far from being orthogonal to the optical axis of the detection objective as in a classical LSFM architecture. Augmenting the light-sheet tilt angle has been achieved using a secondary illumination objective [9,24], or by placing a grating in front of the sample plane [23]. The former increases the complexity and alignment constraints, as two objectives now need to interface with the sample, negating the geometrical advantages of OPM over traditional LSFM. The latter comes with potential ghost images from higher diffraction orders, light losses in the excitation and detection path, and a color dependent tilt angle of the light sheet. As such, we found that there is still room for improvement for mesoscopic OPM designs.

Here, we introduce a reflection-based method to augment the light-sheet tilt angle. This is inspired by single objective selective plane illumination microscopy (soSPIM) [25], where a micromirror reflects the light sheet by 90°. We transfer this concept to mesoscopic OPM, where we increase the light-sheet tilt angle well beyond the half opening angle of the primary objective using a custom microprism. On the detection side, we leverage the concept of diffractive OPM (dOPM) to direct fluorescence light via a transmission grating to the tertiary objective. This resulted in a nearly isotropic lateral point spread function (PSF), as the grating overcomes “light-cone clipping” of traditional OPM systems and a low linespacing of the transmission grating leads to less light-cone compression as in prior work in dOPM [21].

To further simplify the design and ease the alignment burden, we introduce a lensless scanning system, which dispenses with two scan lenses of a traditional OPM optical train. We discuss the geometrical effects of this scanning mechanism in analytical detail. Nevertheless, a traditional LSFM architecture is in comparison more straightforward to build. We compare our OPM implementation to a conventional light-sheet architecture, and highlight the differences in the complexity of the system, their relative cost, and performance.

Lastly we explore how we can increase the volumetric coverage in mesoscopic OPM through optical tiling while keeping the axial resolution constant [26]. To our knowledge, this is the first time this has been implemented in an OPM system, and our architecture facilitates light-sheet focusing over a depth range of 1 mm. The system’s performance is evaluated by imaging fluorescent nanospheres in agarose gels. We demonstrate its application potential by imaging the vasculature and blood flow in zebrafish larvae, calcium signaling in Drosophila larvae, and zebrafish xenograft imaging in a 96-multi-well plate. The spatial resolution of the system allowed us to resolve fine details such as endothelial nuclei, shape changes of red blood cells, and metastatic niches. Our quantitative blood flow measurements across a zebrafish larva indicated the need for acquisition rates above 10 Hz. This further demonstrates the need for rapid mesoscopic imaging techniques.

2. METHODS

A. Light-Sheet Reflection by a Microprism

Reflection of a light sheet has been previously accomplished with a mirror near the sample in techniques such as soSPIM [Fig. 1(a)]. To translate this concept to an open-top, mesoscopic OPM system, the reflection of the light sheet must occur below the sample. This can be accomplished with a knife edge mirror [Fig. 1(b)] or via reflection in a higher refractive media [Fig. 1(c)]. When using the knife edge mirror with an air objective, the following constraints result due to Snell’s law: to achieve a desirable tilt angle of 45° of the light sheet in water, a 20° incidence beam at the air–glass interface is required. This means that the light travels at a shallow angle to the coverslip and the presence of a small gap between the mirror and coverslip would lead to a considerable distance traveled laterally before entering the sample. This in turn limits the field of view over which the beam can be scanned. In addition, a 45° tilt angle would also be close to the maximum tilt angle (48.6°) that could be achieved in such a configuration, as a result of refraction.

 

Fig. 1. Light-sheet reflections in different implementations. (a) In soSPIM, a mirror reflects a light sheet by 90°. (b) Using a knife edge mirror to tilt a light sheet in the sample space. (c) Using total internal reflection in a glass prism to tilt a light sheet. (b) and (c) assume air objectives.

Download Full Size | PDF

In contrast, in an optically denser medium such as a glass prism, the laser light travels at a steeper angle before entering the watery medium [Fig. 1(c)], resulting in a larger field of view that can be accessed. Further, the light-sheet tilt angle can in principle reach up to 90° in water when the total internal reflection condition is reached. Realistically, tilt angles of 60°–75° (to the coverslip normal) are possible, which might be of interest for mesoscopic OPM applied to shallow samples.

Importantly, the total internal reflection in the glass prism reverses the scan direction and modifies the scan amplitude of the light sheet (Supplement 1, Fig. 1). As such, the light sheet is scanned in the opposite direction than the fluorescent light is descanned. A second consequence of the light-sheet mirroring is that the beam waist will be axially shifted as it is laterally scanned, an effect that is also present in soSPIM [25] (Supplement 1, Fig. 2 and Note 1). Both effects are compensated for in our experimental setup as detailed below.

B. Optical Setup

A schematic of the optical layout of our mesoscopic OPM system is shown in Fig. 2. Laser light (blue) is shaped into a light sheet by the illumination engine [27]. Two galvo mirrors (QS20Y-AG, Thorlabs) in an image space between the first and second tube lenses (TL1 and TL2, Both ITL 200, Thorlabs) perform “lensless” scanning of the laser beam. Compared to traditional galvo scanning in OPM, this arrangement dispenses with two scan lenses, which results in less complexity and higher light-throughput (a detailed analytical description is given in Supplement 1, Note 2). The light sheet emerges from the primary objective (O1, Olympus XLFLUOR4X/340, ${4} \times$ magnification, Numerical aperture 0.28) along the optical axis and undergoes total internal reflection in a microprism [Perkins Precision, BK 7, 4.2 mm thickness; see also subpanel (i) in Fig. 2]. Placing a slab of glass in front of an objective would typically result in spherical aberrations. However, the used primary objective is designed to image into a 5 mm deep water column. As such, a glass plate of equivalent optical path length can be introduced without causing deleterious spherical aberrations. A Zemax simulation is detailed in Supplement 1, Note 3 and Supplement 1, Fig. 3. In fact, without such a tall water column or glass plate, spherical aberrations do occur [28]. For this reason, an additional optical flat (4 mm thickness, BK7 glass) is placed after O2, and the grating substrate compensates spherical aberrations for O3. The proper balancing of spherical aberrations was verified by imaging fluorescent beads with widefield based illumination; spherical aberrations are minimized when blur rings are symmetrical above and below the central body of the PSF and the axial full width half maximum (FWHM) is minimized (Supplement 1, Fig. 4).

 

Fig. 2. Schematic setup of the mesoscopic oblique plane microscope. O1–O3, primary, secondary, and tertiary objectives; TL1–TL3, primary, secondary, and tertiary tube lenses; OF, optical flat. Inset (i) shows detail of the microprism that reflects the light sheet into the sample. Inset (ii) shows the working principle of the image space scanning. Inset (iii) shows how the blazed diffraction grating diffracts the first order towards the primary objective. O1–O3, Olympus XLFLUOR4X/340; TL1 and TL2, $f = {200}$ tube lens; ITL 200, Thorlabs; TL3, $f = {250}\;{\rm mm}$ achromat; Galvo1–Galvo3, QS20Y-AG, Thorlabs; Galvo4, resonant galvo; CRS4K, Novanta; PL, Powell lens, 10° fan angle, Laser line Canada; L1 and L2, $f = {50}$ achromat; L3, $f = {60}$ achromat; L4, $f = {30}$ achromat. All achromats were purchased from Thorlabs.

Download Full Size | PDF

Fluorescent light captured by the primary objective is descanned by the galvo mirror pair and reimaged with the secondary objective (O2, Olympus XLFLUOR4X/340, ${4} \times$ magnification, numerical aperture 0.28). A transmission grating (1200 gr/mm, GT13-12, Thorlabs) diffracts the fluorescent light into the tertiary objective (O3, Olympus XLFLUOR4X/340, ${4} \times$ magnification, numerical aperture 0.28), which is then imaged via the tertiary tube lens (250 mm focal length achromat, Thorlabs) onto an ORCA Flash 4 (${2048} \times {2048}$ Hamamatsu) or Kinetix (${3200} \times {3200}$ pixels, Photometrics) sCMOS camera. The overall magnification of the system is ${5.65} \times$, which is close to the theoretical value given by the magnification of the tertiary imaging system, i.e., ${250}\;{\rm mm}/45\;{\rm mm} = {5.555} \times$. The slight deviation from the theoretical value could be explained by tolerances of the focal lengths of the used lenses. This magnification ensures Nyquist sampling.

The illumination engine is shown in Fig. 2, and a CAD rendering is shown in Supplement 1, Fig. 5. A Powell lens (10° fan angle, Laserline Optics, Canada) and a $f = {30}\;{\rm mm}$ achromatic lens (Thorlabs) are used to create a light sheet. An electro tunable lens (ETL, EL-16-40-TC, Optotune) and a galvo mirror (Galvo3 in Fig. 2), both conjugated to the back focal plane of the primary objective, are used to refocus and laterally scan and position the light sheet, respectively. The galvo mirror reverses the scan direction of the light sheet and keeps it in lockstep with the fluorescence descan. An offset applied to the same galvo mirror is used to fine align the light-sheet position to the focal plane of the tertiary objective. A resonant galvo (Galvo4 in Fig. 2, CRS4K, Novanta), conjugated to an image plane, is used for shadow suppression [29]. The ETL dynamically compensates for the shift of the beam waist during lateral scanning. Further, the ETL can also axially shift the light sheet on demand, which we leverage for tiling light-sheet microscopy. With focus compensation, we estimate a lateral scan range of 1.51 mm using the 4.2 mm thick microprism (Supplement 1, Note 1 and Supplement 1, Fig. 2). Laser light is provided by a fiber coupled LightHUB Ultra light engine (Omicron Laserage Laserprodukte, Germany). Waveforms to drive the opto-electronic components (galvos, ETL, laser modulation, and camera triggering) were created with an FPGA card (PCIe-7852, National Instruments), and are detailed in Supplement 1, Figs. 6 and 7 for a volumetric acquisition and the projection mode, respectively.

C. Transmission Grating to Diffract Light into the Tertiary Objective

The use of diffraction gratings to guide light into the shallow acceptance cone of a tertiary objective was first shown by Hoffmann et al. [21]. The idea is to place a grating at the desired image plane angle (i.e., along the image of the light sheet) and choose the grating pitch such that (typically) the first order is normal to the grating surface. Using a reflection grating, a very steep light-sheet plane was picked up in the first demonstration of dOPM. In our case, the image plane is less inclined, owing to the tilt angle augmentation of the light sheet. This allowed us to use a transmission grating instead of a reflective grating and use a lower linespacing (1200 gr/mm compared to 1800 gr/mm). The grating compresses the light cone in the diffraction direction [21], which yields to an elliptical light distribution in the pupil of the tertiary objective. However, this effect is lessened for gratings of lower linespacing, and in our system leads to a fairly symmetric PSF, even compared to conventional OPM systems that incur some levels of beam clipping [30].

The grating surface, where the diffraction happens, faces forward (Fig. 2, inset iii). This was chosen because the grating is on a glass substrate, which has the potential to induce Coma aberrations when traversed at an angle. With the diffraction happening at the front surface of the substrate, the central ray travels close to orthogonally through the glass slide, thereby minimizing such aberrations. Experimentally, the diffraction efficiency into the first order remained $\sim 25\%$ if the grating faced forward or backwards for a wavelength of 514 nm. Of note, the remote image space is slightly compressed in the third dimension, as we chose a unity lateral magnification from sample space to remote space, i.e., the primary and secondary lens, and the corresponding tube lenses, are the same (Supplement 1, Note 4 and Supplement 1, Figs. 8–10).

3. RESULTS

A. Imaging of Fluorescent Nanospheres

We first imaged 500 nm fluorescent beads (Invitrogen, FluoSpheres F8813) in 2% low melting agarose (Sigma Aldrich, A9045) placed in a Matek dish with a No 0 coverslip (Matek P35G-0-10-C). The dish was contacted to the microprism with a thin oil layer ($n = {1.52}$). We used such bead samples to calibrate the ETL based beam waist compensation and to assess the imaging performance of the system [Figs. 3(a) and 3(b)].

 

Fig. 3. Beam waist position during stack acquisition, visualized by imaging 500 nm fluorescent nanospheres in Agarose. (a) Without correction, beam waist is defocused during stack acquisition. (b) By refocusing with the ETL, the beam waist is held at a constant distance to the coverslip during a scan. Red lines in (a) and (b) help visualize the usable range of the light sheet’s beam waist.

Download Full Size | PDF

To estimate the resolution, we analyzed the FWHM of 500 nm fluorescent nanospheres in a 2% low melting agarose gel over a depth of 330 µm (which corresponds to the extent of the beam waist in the $z$ direction). Analysis of 1720 beads resulted in ${2.33} \pm {0.37}\;{\unicode{x00B5}{\rm m}}$, ${2.77} \pm {0.32}\;{\unicode{x00B5}{\rm m}}$, and ${10.16} \pm {1.49}\;{\unicode{x00B5}{\rm m}}$ for the FWHM in the $x$, $y$, and $z$ directions, respectively (${\rm mean} \pm {\rm standard\, deviation}$). As shown in Figs. 4(a) and 4(b), the PSFs remained constant across the field of view.

 

Fig. 4. Imaging of 500 nm fluorescent nanospheres in Agarose. (a) Maximum intensity projection over a 330-µm depth. (b) Magnified views of the white boxes in (a). (c) and (d) Point spread function selected from one nanosphere image.

Download Full Size | PDF

As with any light-sheet microscope employing Gaussian beams [31], the axial resolution and length of the usable beam waist are coupled. In our system, we set the NA of the light sheet to 0.04. Since the light sheet propagates along a plane tilted by 45° to the coverslip, the “height” above the coverslip (i.e., the imaging depth) that is covered with the beam waist is further reduced by a factor of ${\sim}{0.707}$. Increasing the depth coverage could be achieved by lowering the excitation NA, which would reduce the axial resolution. Alternatively, one can refocus the light-sheet waist to different depths for tiling, which has been implemented in conventional LSFM configurations [26], but not in OPM to our knowledge. One aspect that complicates the implementation in OPM is that the light sheet is typically composed of marginal rays of a high NA objective, which would require complex wavefront shaping for proper refocusing. In our case, the light sheet leaves the objective along its optical axis, and as such makes refocusing analogous to refocusing a light sheet in a conventional LSFM architecture. This appeared to work well to dynamically stabilize the waist position during scanning using small refocus amounts, and hence we explored whether we could also use the same mechanism for tiling microscopy. We imaged six individual stacks with different axial waist positions [Fig. 5(a)] over a depth range of 1 mm, and computationally fused the volumes to a single stack [Fig. 5(b)]. The individual stacks were fused with a custom Python script using weighted average fusion based on a sigmoidal function (available at [32]). Of note, the imaging speed of our mesoscopic OPM system was faster than any notable sample motion; thus no initial computational registration was required. For more complex or faster processes, more advanced stitching and registration pipelines are available [33]. The axial resolution over a height of 1 mm was maintained at ${9.225} \pm {1.151}\;{\unicode{x00B5}{\rm m}}$ (${\rm mean} \pm {\rm standard}\;{\rm deviation}$, ${\rm n} = {417}$).

 

Fig. 5. Increasing the depth range while maintaining constant $z$ resolution via optical tiling. (a) Six consecutive volumes are shown where the $z$ position of the beam waist was varied, starting at the bottom (near the coverslip). Red lines indicate the beam waist location. (b) Fusion of the six tiles. Insets show magnified views of the boxes near the bottom, middle, and top of the fused volume.

Download Full Size | PDF

B. Imaging of Zebrafish and Drosophila Larvae

To demonstrate the intravital imaging capabilities of our mesoscopic OPM system, we performed fast live imaging of entire zebrafish larvae at three days post fertilization (dpf). Zebrafish expressed the vascular marker Tg(kdrl:EGFP) and the red blood cell marker Tg(gata1a:DsRed) in a casper background [34]. To mount the zebrafish larvae, they were anesthetized with 200 mg/l Tricaine (Sigma Aldrich, E10521) and placed in a Matek dish with a No. 0 coverslip (Matek P35G-0-10-C) at the bottom. A second glass coverslip on top provided further mechanical support during imaging. Zebrafish husbandry and experiments followed established protocols and have been approved and conducted under the oversight of the Institutional Animal Care and Use Committee (IACUC) at UT Southwestern under protocol number 101805 to Gaudenz Danuser.

The field of view of the mesoscopic OPM system covered the whole zebrafish larval vasculature [Figs. 6(a)–6(f)]. To obtain higher resolution and better depth coverage, we fused two tiled volumes. As with any LSFM technique, light scattering by the sample induced some blur further away from the detection objective [Fig. 6(d)]. Nevertheless, the head and tail vasculature of the zebrafish larva were well sectioned throughout the volume [Figs. 6(d)–6(f)]. Importantly, the mesoscopic OPM system resolved fine details [magnified views in Figs. 6(b), 6(c) and 6(e), 6(f)], including nuclei of endothelial cells (bright, localized spots) and the intricate branching patterns of zebrafish vasculature, including parachordal lymphiangioblasts.

 

Fig. 6. Imaging of Zebrafish vasculature. Fluorescently labeled vasculature, Tg(kdrl:EGFP), in a three days post fertilization (dpf) zebrafish larva, as imaged with our mesoscopic OPM. (a) $x {\text -} y$ maximum intensity projection of the entire zebrafish with (b),(c) magnified views (head and tail vasculature) of the boxed regions in (a). Arrowheads indicate selected endothelial nuclei, and arrows point to parachordal lymphangioblasts. (d) $x {\text -} z$ maximum intensity with (e),(f) $x {\text -} z$ maximum intensity projected magnified views (head and tail vasculature) of the boxed regions in (d). Arrowheads indicate selected endothelial nuclei.

Download Full Size | PDF

To demonstrate the potential for rapid volumetric imaging, we imaged blood flow dynamics without optical tiling (one stack was acquired per timepoint). With ${4} \times {4}$ binning, we could achieve a volumetric imaging rate of 5 Hz (Visualization 1).

While it was possible to identify and track blood cells manually with anatomical knowledge of zebrafish vasculature, 5 Hz was too slow for automated analysis of blood flow dynamics with optical flow analysis (Supplement 1, Note 4). Therefore, we leveraged our recently introduced projection imaging technique [16], where the light sheet is rapidly swept through the 3D volume, and the fluorescence image is synchronously scanned (“sheared”) over the image sensor. To this end, we added a single galvanometric mirror (Thorlabs QS20Y-AG) in front of the camera for optical shearing. Together, this enabled rapid time-lapse projection imaging at 12 Hz imaging rate with coverage of the whole larva over 100 timepoints (Visualization 2).

The resulting imaging data enabled qualitative and quantitative analysis of blood flow across the entire larva (Fig. 7). As blood cells are sparse objects in the blood stream, we applied maximum intensity projections over the time-lapse to obtain a more continuous map of vessel perfusion [Fig. 7(a)]. Moreover, color-coding of subsequent timepoints [Fig. 7(b)] revealed the directionality of flow within the vessels. Thereby, the mesoscopic OPM system also captured fine details such as subtle differences of red blood cell shapes in different vessels [Figs. 7(c)–7(f)]. To quantitatively determine the blood flow direction and magnitude across a whole larva, we computed multiscale optical flow using Farnebäck’s method [35], implemented in OpenCV [36] to measure the frame-to-frame blood cell movement. As movement is only measured in the presence of cells, we sampled the velocity vector corresponding to the 95th percentile speed measured over the video duration to reconstruct the underlying blood flow field at each pixel position [Fig. 7(g), Supplement 1, Note 4 and Supplement 1, Fig. 11]. Our analysis clearly resolved the directionality of blood flow within the whole zebrafish vasculature present in the projection images.

 

Fig. 7. Imaging and quantification of blood flow in zebrafish larvae. Blood flow dynamics in a 3 days post fertilization (dpf) zebrafish larva, as imaged with our mesoscopic OPM in a projection format at 12 Hz over 100 timepoints. (a) The maximum intensity of the hundred frames (gray) provides a visual impression of the vasculature and a map of which vessels were perfused. In red, a single timepoint of the movie is shown from which the average signal over the time-lapse was subtracted to highlight individual, bright red blood cells. Insets depict the regions that were magnified in (c)–(f). Scale bar: 250 µm. (b) Ten subsequent timepoints of the movie (744 ms) were color-coded and overlaid on each other. (c),(d) Magnified views of the subintestinal vein (SIV) plexus. White arrows indicate blood flow direction. Scale bar: 100 µm. (e),(f) Magnified views of the dorsal longitudinal anastomotic vessel (DLAV), three intersegmental vessels (Se), the dorsal aorta (DA), and posterior cardinal vein (PCV). Scale bar: 100 µm. (e) White arrowheads highlight that red blood cells adopt different shapes in different vessels, including elongated shapes in intersegmental vessels and compact, spherical shapes in large vessels with fast flow. (f) Color coding further allowed to identify the vessel identity and distinguish intersegmental artery (SeA) and intersegmental vein (SeV). White arrows indicate blood flow direction. (g) Result of quantitative analysis of blood flow with optic flow using Farnebäck’s method. The color code and arrow orientation indicates the directionality of the flow, and arrow length and color saturation the magnitude of the flow.

Download Full Size | PDF

To further demonstrate the fast-imaging capabilities of the mesoscopic OPM system, we imaged calcium signaling in muscle cells of Drosophila first instar larva using volumetric imaging at 2 Hz over 50 timepoints. The resulting imaging revealed the intricate firing within a larva (Supplement 1, Fig. 12, Visualization 3).

C. Multi-Well Imaging of Zebrafish Larvae

To highlight the potential of our mesoscopic OPM for high-throughput imaging in a multi-well format, we imaged fixed zebrafish larvae at 5 dpf that have been xenografted with A375 melanoma cancer cells with a Rac1P29S mutation [37] and labeled with mRuby2-tractin. The zebrafish were injected with cancer cells into the common cardinal vein at 2 dpf and were fixed at 5 dpf with paraformaldehyde. For imaging, the fish were pipetted into a Hashimoto ZF plate (model number HDK-ZFA101-02a), which required no further manual alignment of the fish due to the funnels in each well. The well plate was then mounted on a home-made holder, which was aligned and translated over the microprism with a manual 3D stage.

The projection mode allowed us to quickly assess the spread of cancer cells in xenografted zebrafish expressing the vascular marker Tg(kdrl:EGFP) [38]. Some xenografted fish showed metastatic spreading to the tail [Figs. 8(a), 8(c), 8(e), and 8(h)], while in other fish, cancer cells were only found near the injection site [Figs. 8(b), 8(d), and 8(g)]. In one fish, no cancer cells survived 3 days after injection [Fig. 8(f)]. To study the cells that spread to the tail, we imaged one fish in 3D [Figs. 8(i)–8(l)]. Zooming into the tail (maximum intensity projection Figs. 8(i) and 8(j)] revealed that several cancer cells extravasated out of the vasculature and formed micrometastases in the caudal hematopoietic tissue [Figs. 8(k) and 8(l)].

 

Fig. 8. Imaging of zebrafish xenografts in a multi-well plate. (a)–(h) Eight fish, labeled with the vasculature marker Tg(kdrl:EGFP)(gray), and injected with A375 melanoma cells labeled with mRuby-tractin (magenta), as imaged in the projection mode. A gamma correction of 0.75 was applied to the vasculature channel. (i) Magnified view of the boxed region in (a), imaged in 3D. (j) Magnified view of the boxed region in (i). (k) Magnified view of the boxed area in (j); only the red channel is shown. (l) same area as in (k), but only the vasculature channel is shown. Red arrows point to the locations of the cancer cells. (i),(j) Maximum intensity projection. (k),(l) Single slice. Scale bar: (a)–(i) 500 µm; (j),(k) 100 µm.

Download Full Size | PDF

4. DISCUSSION

We have introduced a mesoscopic OPM system that combines concepts of soSPIM, dOPM, and tiling light-sheet microscopy, as well as a new lensless galvo scanning mechanism. Taken together, this resulted in a compact system that improves 3D imaging performance over previous mesoscopic OPM systems. To demonstrate the power of our mesoscopic OPM system, we imaged larval zebrafish vasculature and blood flow dynamics in vivo, resulting in a quantitative map of blood flow dynamics across an entire larva.

A key element in our mesoscopic OPM design is the microprism, which augments the light-sheet tilt angle via total internal reflection. Thereby, we can create light-sheet tilt angles that exceed the half opening angle of the primary objective. We chose a light-sheet tilt angle of 45°, as it forms a good compromise between optical sectioning and volumetric coverage (i.e., the height above the coverslip that can be covered with a given beam waist). Nevertheless, our method can also allow higher tilt angles. As an example, a prism angle of 61° instead of 71° would result in a light sheet that is tilted by $\sim 30^\circ$ to the coverslip.

Importantly, our approach via total internal reflection is wavelength independent. This is a considerable advantage compared to using a diffraction grating close to the sample plane, as introduced by Shao et al. [23], which induces a wavelength dependent tilt angle. Consequently, we can perform multicolor imaging without modifying the setup, in contrast to Shao et al., who needed to change the tilt of the tertiary imaging system for different colors. The grating will also create unwanted higher diffraction orders with the fluorescent light. While they do not appear prominently in the images presented by Shao et al., they still subtract valuable signal from the zero order. Lastly, a zero-diffraction order of the excitation light needed to be blocked with a moving mask, which will limit the ultimate speed of the grating method. In contrast, total internal reflection occurs with such a high efficiency that we did not need to block any unwanted excitation light.

Besides the aspects of efficiency and wavelength dependency, there are geometrical effects. Light-sheet scanning can occur anywhere across the grating. In contrast, in our method, light-sheet scanning must occur across the tilted face of the prism, so more care for adjusting the position must be taken as the tilted flank of the prism cannot be used for fluorescence imaging. Additionally, the total internal reflection on the prism causes a reversal of the scan direction, and a defocusing of the light sheet. In our setup, the reversal of the scan direction is countered with the light-sheet positioning mirror in our illuminator. This mirror has so far always been motorized in our setups, as it is used to align the light-sheet position to the focal plane. Thus, here, the mirror not only ensures the proper alignment of the two but is also used for compensating for the scan reversal. Of note, it is also conceivable to place the dichroic right after the primary objective and use this space to couple in the light sheet. Our solution minimized in comparison the total number of lenses needed.

Furthermore, the electro tunable lens (ETL) in our illuminator introduces optical tiling to OPM systems besides its function to compensate the defocusing of the light-sheet waist. As demonstrated here, this can extend the reach of our mesoscopic OPM in the third dimension up to 1 mm, about three times the range shown by Shao et al. We believe that this capability is of importance for large transparent organisms, or samples that are not attached to the coverslip surface, which may include freely moving larvae. Lastly, expansion microscopy and tissue clearing now produce very large transparent samples. As such, improvements of the axial reach are important for the next generation of mesoscopic OPM systems.

In comparison to a traditional LSFM system, an OPM imposes more stringent alignment criteria for its remote focusing system to work properly and possesses a more complex detection path featuring three objectives. Nevertheless, in our system the number of major components is not that different from a traditional LSFM setup: while our OPM system needs one additional objective and two more galvos, a conventional LSFM system may need a large stroke objective piezo to perform fast 3D imaging. While these components are similar in cost, the performance is different. A traditional LSFM system will have a more light-efficient detection path and is expected to achieve higher 3D resolution, in particular when using axially swept light sheets [39]. However, in terms of acquisition speed, the OPM is expected to outperform the traditional LSFM systems, as large stroke mechanical actuators, combined with a heavy objective, typically have a much lower bandwidth than galvo mirrors. Lastly, in terms of sample access, a traditional LSFM geometry is not compatible with multi-well plates, whereas the OPM, by virtue of its single primary objective, is fully capable of imaging in well plates. We compare in detail the components, alignment, and cost for our OPM system and a traditional LSFM in Supplement 1, Note 6 and Supplement 1, Fig. 13.

The lateral width of the field of view is currently limited by the camera to enable Nyquist sampling. For a ${3200} \times {3200}$ pixel camera, a width of ${\sim}{3.68}\;{\rm mm}$ results. With cameras with more pixels, a 5 mm width might be possible given the field of view of the objectives used. In the scan direction, the prism imposed a limit of 1.51 mm. The glass prism thickness could be doubled if the optical flat after the secondary objective were left away. This can be seen as follows: somewhere in the system, an aberration correction for a combined 15 mm deep water column must occur. This correction can occur at any stage, i.e., in front of the primary objective, in the remote space, or in both spaces. Assuming a field of view of 5 mm of the primary objective, this would enable ${\sim}{2.78}\;{\rm mm}$ scan range (2.21 mm would be used up by the inclined flange of the prism).

The high sampling requirements may slow down acquisition speed. Based on the manufacturer’s specifications, we estimate that volumetric acquisition rates of 8 Hz for Zebrafish embryos could be possible (Supplement 1, Note 8), or faster by dispensing with Nyquist sampling. Practically, we had trouble streaming and buffering the resulting data flow, which our computational hardware could only handle with binning for the 5 Hz volumetric acquisition shown here. An alternative could be an adaptive, smart sampling approach [40]. As an example, one camera channel could record the entire fish at reduced 3D spatio-temporal resolution, while another channel covers a sub volume (such as the brain), features that require higher resolution (such as cancer cells), or processes that can be interrogated by fast 2D projection imaging.

In conclusion, we have introduced a compact mesoscopic imaging system that features large volumetric coverage and comparatively high spatial resolution. This has been enabled by light-sheet mirroring to increase its tilt angle, the use of a transmission grating to diffract light into the acceptance cone of the tertiary objective, and optical tiling. An image-based scanning mechanism further simplifies the optical train of our system. As such, we hope that this system will find widespread applications in biological and biomedical research, as it is easy to build and offers high spatiotemporal resolution in the mesoscopic imaging realm.