Open Access
Add to BibSonomy
Abstract
We present an elegant scheme for providing multi-directional illumination in selective plane illumination microscopy (SPIM). Light sheets can be delivered from one of two opposed directions at a time and pivoted about their center for efficient stripe artifact suppression using only a single galvanometric scanning mirror to perform both functions. The scheme results in a much smaller instrument footprint and allows multi-directional illumination with reduced expense compared with comparable schemes. Switching between the illumination paths is near instantaneous and the whole-plane illumination scheme of SPIM maintains the lowest rates of photodamage, which is often sacrificed by other recently reported destriping strategies. The ease of synchronization allows this scheme to be used at higher speeds than resonant mirrors typically used in this regard. We provide validation of this approach in the dynamic environment of the zebrafish beating heart, where imaging at up to 800 frames per second is demonstrated alongside efficient suppression of artifacts.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Selective plane illumination microscopy (SPIM) is an increasingly prevalent technique for fast and gentle multi-scale imaging of biological samples [1,2]. Typically, a laser beam is appropriately shaped to produce a thin sheet of light, illuminating a single optical section of a sample. The fluorescence emitted from the illuminated section is collected at 90° and imaged onto a camera. This widefield scheme delivers imaging speeds several orders of magnitude faster than equivalent point-scanning techniques (e.g., laser-scanning confocal microscopy) providing a route to uncover dynamics in living organisms and allowing large cleared tissues to be imaged in a reasonable time. Furthermore, SPIM features low rates of photodamage owing to two factors: i) plane-wise illumination confinement (all ballistic signal contributes positively to image formation) and ii) a high degree of illumination parallelization (signal is captured from the whole plane synchronously). This latter point is particularly crucial in highly optically sensitive or dim samples where the lowest peak intensities are required to ensure minimal impact on the system under study or to capture sufficient photons during a single camera exposure. Likewise, synchronous illumination renders the fastest acquisition modes of modern sCMOS cameras, easily accessible for high-speed imaging.
In mesoscopic and macroscopic samples (e.g., living embryos to intact optically-cleared tissues) the ability to employ multi-directional illumination via additional illumination paths is particularly beneficial to circumvent loss of contrast and attenuation of the light sheet towards the distal side of the sample [3]. Ideally, one achieves satisfactory penetration to at least the center of the sample, such that illumination from the two directions provides a pair of complementary images that can be combined for superior optical coverage of the sample.
One can envisage a variety of routes to provide and select double-sided illumination. These essentially belong to one of three categories: i) A laser beam is split into two, with one side optionally shuttered. This is wasteful with regard to laser power but can deliver illumination from both sides at the same time if desired. This may be beneficial, where line-confocal detection [4] or non-linearity [5] ensure that the signal decays with depth, but more typically, one wishes to deliver illumination from only one direction at a time such that the distal light sheet does not contribute to loss of contrast on the proximal side with respect to the other illumination arm. The shuttering can be achieved using mechanical, liquid crystal or acoustic devices as desired (in increasing order of expense/complexity). ii) A single laser beam is directed down one of two optical pathways in an alternating manner. This scheme is more optically efficient but precludes the ability to illuminate from both sides simultaneously. iii) Lasers that can be modulated are provided for each of the two illumination paths. This is optically efficient but a particularly costly option. Note, that the case of a dual fiber-coupled laser engine falls under ii) since the output is internally optically switched between the two fibers. Various schemes for double-sided light sheet generation that conform to this categorization are shown in Figure 1.
Fig. 1. Schemes for double-sided light sheet illumination. a) – b) Lossy schemes based on 50:50 beam-splitting. c) – d) Lossless schemes based on beam-sharing. The power transmission through the system is shown by the opacity of the colored arrow. The polarization state is shown by the black arrows where relevant. a) One beam is selectively shuttered. The shutter in this case may represent an optoelectronic device, including mechanical and liquid-crystal shutters, acousto optic-detectors/-modulators/tunable filters. Some of these shuttering methods may alter the polarization state, requiring up/downstream optics to reach a desired polarization output state. b) A chopper wheel can be used for fast shuttering of the beams for interleaved double-sided light sheet illumination. c) A flip mirror can be flipped into place to divert the beam to a second illumination path. d) A Pockels cell selectively rotates the polarization state of the input beam, which is then either reflected or transmitted by a polarization beam-splitter. To reach a desired polarization state a half-wave plate is required in one of the beam paths.
Given the often slow, wasteful or expensive schemes for double-sided illumination, further exploration is worthwhile. In general, scheme ii) whereby the beam is alternated between two pathways is preferable to maximize power throughput, but the majority of schemes are too complex/expensive (Pockels cell, AOD, dual-fiber-coupled laser engine) or too slow and prone to drift with multiple changes of position (flip mirrors). One may switch with much greater rapidity and repeatability using galvanometer-based scanning mirrors (galvos), however, performing the relatively simple function of binary switching leaves much of the capability of the galvo underutilized. Nevertheless, much finer control over the beam propagation could provide additional capabilities if the galvo is more completely utilized.
A further challenge for light sheet microscopy in general is the presence of striped artifacts. The typically low numerical aperture (small disparity of ray angles) of the light sheet and the orthogonal illumination-detection scheme highlights attenuation of the light sheet, manifesting as dark stripes in the image. These striped artifacts arise as a result of uneven illumination as absorbing structures directly attenuate the light sheet while scattering bodies result in areas of destructive and constructive interference beyond the occlusion.
Several schemes have been developed to overcome these challenges. Huisken and Stainier developed mSPIM (multi-directional SPIM) combining double-sided illumination with effective destriping by resonantly pivoting the light sheet about the detection lens optical axis [3,6,7]. The larger disparity of ray angles and lack of coherence between the angled light sheets sampled during a single camera exposure resulted in effective suppression of striped artifacts, effectively allowing the illumination source to ‘see-around’ occlusions. Others have reported similar utilizing more costly acousto-optic modulators rather than resonant mirrors for single color imaging only. In principle, this allows higher frame rates, however, the angular beam displacement produced is small, which may limit the light sheet angular disparity and the destriping capability accordingly [8].
Keller et al. first reported the use of virtual light sheets produced by scanning a Gaussian beam (digitally-scanned light sheet microscopy, DSLM [9]), which also has the effect of breaking the coherence across the height of the light sheet (responsible for scattering induced artifacts). However, the sacrifice of parallelization is associated with an increase in the peak intensity to maintain a given signal rate and so increases the rate of non-linear photodamage [10]. Moreover, the ray angle disparity is unchanged and so scanning does not suppress artifacts resulting from absorption (e.g., by pigmented structures). Glaser et al. combined the mSPIM and DSLM approaches (mDSLM) whereby an elliptical beam is focused and scanned to produce a virtual light sheet [11]. While suffering from the same drawbacks of DSLM with regard to parallelization, the disparity between the light sheet NA within and outside the light sheet plane provided more effective suppression of striped artifacts. This concept was extended by Ricci et al., who additionally pivoted the elliptical beam at high-speed using acousto-optics, while achieving compatibility with line confocal detection under pivoting conditions owing to the broad ray fan of the elliptical beam in the light sheet plane [12]. Others still have employed single-axis diffusers or gratings which, while effective, have not obviated the need for moving components or have otherwise required additional optical components to be located in the object space between illumination objective and sample [13,14].
The original scheme of Huisken et al. remains particularly attractive owing to its low cost and compatibility with whole-scene synchronous illumination but has been limited regarding imaging speed. This is a result of the asynchronous nature of the camera exposure and the pivoting action. Under this condition, one heavily oversamples the pivot during the camera exposure to achieve effectively uniform sampling of all light sheet angles. For a resonant mirror operating at 1 kHz, this speed limits one to effective destriping at 100 - 200 frames per second. While faster resonant mirrors are available (into the 10 kHz range), suitable options are costly and too loud to discount their impact on living biological systems. Galvos provide ease of synchronization, allowing one to sample all angles equally (an integer number of times ≥1) during a single camera exposure. Galvo mirrors also provide scope for larger beam deflections than required for light sheet pivoting. Herein we report how this additional capacity allows a single galvo mirror to be used not only to perform synchronized light sheet pivoting but also to rapidly select between one of two illumination paths. For brevity, we refer to this concept as single-element pivoting/switching (SEPS). Moreover, we show how this extends the capabilities of mSPIM to imaging at up to 800 frames per second in the dynamic environment of the beating zebrafish heart.
Before discussing, specifically how this scheme allows high-speed destriped light sheet imaging, consider that a knife-edge mirror (KEM) can be used to separate an incident beam into two opposed reflected pathways. KEMs have been employed previously for illumination/emission control in light sheet microscopy. Notably Vladimirov et al. used a KEM to switch the illumination and emission direction between two objectives oriented at 90 degrees to each other [15]. Landry et al. used translating KEMs to control the periodicity of patterns for the generation of structured light sheets [16], while Dean et al. and Greer and Holy have used KEMs to distribute images onto different regions of single and multiple cameras respectively [17,18]. Here, a KEM is used as part of a scan system as shown in Figure 2. A galvo mirror telecentrically scans an input beam across the face of a KEM. Light incident on either side of the KEM will go on to form a light sheet from one of two directions. The galvo mirror is conjugated to the shared focal plane of two opposed illumination objectives by a series of relay lenses. A cylindrical lens with its non-zero power axis oriented with the galvo scan produces a light sheet at the shared focus. Varying the galvo angle within some range results in a pivoting effect of the light sheet for the respective side of the KEM.
Fig. 2. The SEPS scheme. Top: the galvo mirror, two scan relay lenses and knife-edge mirror form the basis of the pivoting and light sheet switching strategy. Middle: the collimated pivoting beam output from the second scan relay lens is focused by a cylindrical lens and the scanned line focus is relayed to the back focal plane of the illumination objectives via scan and tube lenses. Bottom: the illumination objectives collimate the light sheet in the xy (pivoting) plane, while focusing in xz plane. Note the changes in coordinate geometry in the panels. The system x = 0 axis is common to all three panels and defines the center of symmetry of the system.
Now consider the scan angles required for SEPS. The original mSPIM reported a pivot range of ± 5° [3]. The pivot range for a given galvo scan range will be determined partially by the relay optics used. However, as an exercise, consider a system with unit magnification (M = 1) from the galvo to the common illumination objective focus. A pivot range of ± 5° on either side requires a maximum total deflection of 20°. This is well within the capabilities of common galvo mirrors. Since the off-axis transmission of the beam through the system is only in the axis defining the light sheet height, for which the effect of aberrations is negligible, one maintains a clean light sheet focus without highly specialized multi-element scan optics. For the highest speed, translating a small pivot at the galvo to a large pivot at the sample (M < 1) is desirable and there is no practical limitation from the SEPS scheme for much larger pivot angles than ultimately desirable. In any case, the effect of an ever-larger pivot angle on destriping can be seen to diminish for angles of ± 3° in Supplement 1.
2. Optical scheme
The SEPS scheme as implemented herein is briefly outlined: A 4f relay is placed between the galvo mirror and the cylindrical lens. We refer to this relay pair as the scan relay (scan relay lenses 1 and 2). A 90° knife-edge mirror is placed intermediate to the scan relay (i.e., at 2f from the galvo mirror), resulting in a 90° reflection of the beam into one of two optical paths. A scanned 1D focus is produced by a cylindrical lens and relayed to the back focal plane of the illumination objective via a 4f system and the illumination objective translates the scan offset at the back focal plane to a tilt of the light sheet produced at the front focal plane. We refer to the relay lens set as the light sheet relay (light sheet relay lenses 1 and 2). Since the conjugacy of the scanning element (galvo) and illumination objective focus are maintained in the scanning direction (3× 4f systems), dithering the galvo mirror across a single side of the knife-edge mirror reproduces the desired pivoting effect.
The requirement to provide double-sided illumination, with identically sized light sheets further dictates that the two optical pathways from the galvo mirror to the sample are identical and that the galvo scan axis lies along the system y-axis at x = 0 (see Figure 2 for a description of the system axes). This yields a scheme whereby the galvo mirror, the first of the two additional relay lenses and the knife-edge mirror are shared between the two optical pathways and all other components are mirrored to produce two identical illumination paths.
Since some portion of the knife-edge mirror clear aperture on either side is reserved for pivoting, the 0° non-pivoted light sheet (i.e., equivalent to the canonical SPIM) is produced when the beam is offset some distance from the knife-edge of the mirror. Consequently, as default, the first relay lens is utilized in an off-axis configuration.
The position and orientation of turning mirrors is also important to ensure that the pivot is oriented correctly and that conjugacy is maintained. Moreover, any rotation of the pivot orientation must be constrained to 90° turns such that the beam scan at the back focal plane of the illumination objective is oriented in y only. To provide ease of realignment, a steering mirror is placed conjugate to the back and front focal planes of the illumination objectives. To match the length of the two optical pathways it is necessary to place an additional turning mirror at an arbitrary plane (between the cylindrical lens and second scan relay lens is typical, with its precise location dependent on the front/back working distances and thicknesses of lenses throughout the illumination path). The galvo mirror to knife edge mirror subsystem can be rotated to correctly orient the scan at the back focal plane in y (incorrect orientation will orient the scan in z).
3. Experimental setup
The diverging output of a fiber-coupled laser engine (Toptica Photonics, MLE, 405, 488, 561 and 640 nm laser lines) is collimated (f = 40 mm), cropped using a pair of adjustable slits, and directed onto a galvo mirror (Scanlab Dynaxis 3S). The scan relay system is produced by two achromatic doublet lenses (2” diameter, SR1/2 f1,2 = 100 mm, Thorlabs) with a protected aluminum coated knife-edge mirror placed at the common focal plane (OptoSigma). A line focus is produced by a cylindrical lens (cyl., fcyl = 50 mm, Thorlabs) and conjugated to the back focal plane of the illumination objective (obj. Nikon 10×/0.3W, CFI60 10XW) via the light sheet relay (scan,tube f1,2 = 75 mm, Thorlabs) to produce a light sheet at the front focal plane of the objective. The optical path after the knife-edge mirror is duplicated in mirrored form on the opposite side. Fluorescence is detected via a water dipping objective lens (Nikon CFI75 16×/0.8, water-dipping) and imaged onto the entrance port of an image splitter (Hamamatsu Gemini) via a tube lens (ftube = 200 mm Nikon, system magnification = 16×). An alternative tube lens was used exchangeably to provide doubled magnification (ftube = 400 mm Thorlabs, system magnification = 32×). The image splitter images two color channels separated by a dichroic mirror (560 nm longpass, Chroma) and individually spectrally filtered (525/50 nm bandpass, 570 nm longpass, Chroma) onto an sCMOS camera (Andor Zyla 4.2). The sample is positioned via 3 translation stages (Physik Instrumente, M-111-DG, C-884), a rotation stage (Physik Instrumente, U-651, C-867) and a custom sample holder.
The galvo mirror is controlled and synchronized to the camera exposure via a DAQ card (National Instruments, NI-USB-6341). The camera was run at its full frame rate in overlap mode (approx. 1/texposure) providing the galvo waveform frequency. The waveform offset voltage allowed one or the other illumination side to be used during acquisition and to switch between illumination sides in <10 ms. The waveform phase was selected to initiate the pivot with the galvo at its largest angular displacement for the respective illumination side. For an acquisition comprising n images, n + 1 galvo oscillation periods were generated to ensure that the first and last frame (which only overlap with the subsequent or preceding image for half of their total exposure time) sampled all light sheet angles equally. The timing is illustrated in Supplement 1. It should be noted that other readout/exposure modes such as simulated global shutter are compatible with the scheme requiring only simple reconfiguration of the galvo waveform.
Since the beam path requires substantial folding to achieve a centering of the galvo scan system on the system yz plane at x = 0, while providing sufficient degrees of freedom for alignment, we include an additional flattened schematic of the illumination path(s) and a 3D CAD rendering of the microscope described as Supplement 1.
A second system optimized for a larger field of view (FOV) was also constructed and practically differs only in the absence of the dual color image splitter, and the system magnification of 11.1× (defined by the Olympus XLPLN10XSVMP 10×/0.6, multi-immersion objective (ftube = 200 mm). As above an alternative tube lens was used exchangeably (ftube = 400 mm Thorlabs) for doubled magnification of 22.2×. Furthermore, a removable 3× beam expansion in the y direction was included (using two cylindrical lenses, f1 = -50 mm f2 = 150 mm, Thorlabs) to allow for a uniformly illuminated FOV or for the Gaussian profile of the light sheet in y to be clearly apparent. The data comprising Figures 3, 4 and 5 were produced using this larger FOV version, whereas Figures 6 and 7 were produced with the prior version optimized for high imaging speed.
Fig. 3. Light sheet uniformity under synchronized and unsynchronized pivoting conditions. The pivoting frequency multiple and absolute pivoting frequency is shown in top left of each panel. Three sequential images are acquired at 100 frames per second (10 ms exposure time). The light sheet y-intensity profile is displayed for each image (averaged across all x rows). The dashed region demarcates the usable region of the desired light sheet (> 80% peak intensity). The green, blue and orange boxes show the region over which the > 80% condition is met for the three images.
Fig. 4. Sectioning and resolution resulting from the SEPS scheme assessed in a fluorescent bead sample. a) xy view: a maximum intensity projection (along the z-axis) of a z-stack acquired for a bead sample over a volume 600 × 600 × 250 µm. b) xz view: a maximum intensity projection (along the y-axis) of the resliced bead stack over the central sub-volume with dimensions 600 × 75 × 250 µm. Only a sub-volume is shown to more clearly show the shape of the point spread function when the volumetric data is projected.
Fig. 5. Assessment of SEPS for destriping under quasi-static conditions in a zebrafish larva (72 hpf) stained with a lipophilic dye. Top: The spectrum of the 0× (no pivoting) and 1× (pivoting at the camera frame rate) images and the difference between the two. The spectrum is shown as the absolute value of the fast fourier transform of the image. Additional frequencies along the y-axis are present in the 0× spectrum as a result of the stripes visible in the image. Bottom: The relative information ratio, Ix:y,relative for images acquired with pivoting frequencies 0–4× the camera frame rate are shown. The error is given as the standard deviation of Ix:y,relative across a full z-stack of images (200 panes captured with 2 µm plane spacing). The blue arrows indicate the direction of light sheet propagation.
Fig. 6. Assessment of SEPS for destriping under dynamic conditions. a) Identical phases in the heartbeat of a transgenic zebrafish larva (48hpf, (Tg(kdrl:Hsa.HRAS-mCherry, myl7:dsRed))) additionally counterstained with a lipophilic dye. Top: 0× (no pivoting), Bottom: 1× (pivoting at the camera frame rate, 400 frames per second, 2.5 ms exposure). The blue arrows indicate the direction of light sheet propagation. b) The green boxed region is shown across 9 heartbeat cycles. The red box shows the mean image across all heartbeats. The information ratio, Ix:y,relative is shown for the SEPS scheme, the standard deviation is calculated for the 9 heartbeats and reflects the variation in striping for both the 0× and 1× datasets resulting from the transient presence of blood cells, which varies from one heartbeat to the next. c) the destriping capability of the SEPS scheme for several phases of the heartbeat (note the 0 phase is as a) but arbitrary rather than referenced to any particular feature of the beating cycle). The destriping can be seen most clearly in the lipophilic stain (green) but can also be seen by careful inspection of the myocardium (red channel). The destriping is clearly seen in the zoomed images (0× left, 1× right)
Fig. 7. Assessment of SEPS for destriping under dynamic conditions with higher pivoting harmonics and at high frame rates. a) Sub-regions extracted from identical phases in the heartbeat of a transgenic zebrafish larva (48hpf, Tg(kdrl:Hsa.HRAS-mCherry, myl7:dsRed)) additionally counterstained with a lipophilic dye. The region of interest is extracted from behind the atrium with respect to the light sheet propagation for 0×, 1×, 2×, 3×, and 4× pivoting frequencies, corresponding to 0, 200, 400, 600, and 800 Hz for imaging at 200 frames per second. b) Images of the zebrafish beating heart at arbitrary phases in the beating cycle under doubled-magnification (32×) at 800 fps showing effective destriping for all 1× cases. The blue arrows indicate the direction of light sheet propagation.
In both systems the light sheet length and thickness, which are related quantities, were controlled by the use of an adjustable slit to crop the beam size and to provide the optimum sectioning over the field of view required for the sample. However, all direct comparisons between imaging modes including the datasets for Figs. 5–7 are made without adjusting the light sheet dimensions. In both cases the light sheet height (y) was controlled via appropriate beam expansion to achieve better than a ± 10% intensity variation over the height of the light sheet for the required number of pixel rows (512/2048 for the first and second microscopes respectively). The SEPS scheme is not restricted to any particular light sheet dimensions and could be redesigned using multi-element scan and tube lenses to allow for much higher illumination NA than used here (NA < 0.15).
4. System performance
First, we demonstrate that the scheme can produce even light sheet illumination across several camera frames. Deviations from telecentricity of the scan that occur due to imperfect placement of optics, clipping of the beam, and any misalignment between the illumination and detection objective foci result in an imperfect pivoting of the light sheet under normal conditions. This can be observed as an apparent shift of the light sheet center in y. For conventional mSPIM, this does not cause issues since each camera exposure sees an average of the light sheet positions. Small shifts in fact act to homogenize the light sheet y-intensity profile at a small cost to spatial duty cycle (i.e., only ca. 80–90% of the intended field of view is illuminated at any particular instance). However, when the pivoting rate is similar to the camera frame rate (but not synchronized therewith), one observes this variation across a series of camera frames as some angles are ‘seen’ more than others. Under conditions of synchronization (i.e., the light sheet pivot rate is an integer multiple of the camera frame rate), all angles are equally sampled during an exposure, again leading to a uniformly illuminated field. To demonstrate that this is apparent, a uniformly fluorescent phantom (2% agarose with a small quantity of fluorescein) was imaged under different conditions for three consecutive frames (Fig. 3). Under unsynchronized conditions with a pivot rate slower or similar to the camera frame rate (0.33×, 1.66×), the variation in the effective light sheet position was clearly visible for the three frames shown. When the pivot rate was much faster (7.66×, comparable with classical mSPIM) the light sheet was close to homogeneous owing to oversampling of the pivot during a camera exposure. Under synchronized conditions, the light sheet was fully uniform from one frame to the next, since every pivot angle was sampled exactly once during the exposure in this case. Note, that the light sheet height was reduced by removing the 3× cylindrical expander to better observe the Gaussian profile within the field of view.
Having shown that the SEPS scheme produces a homogeneous light sheet, we explore the pivoting behavior and its ability to deliver well resolved images. If the beam scanning, cylindrical lens focusing axis and the detection objective focal plane do not lie in the same plane, the effect of the beam-scanning will be to effectively broaden the light sheet, thus resulting in a high background from out of focus emitters and poor axial resolution. Figure 4 shows images acquired from stacks of fluorescent beads (TetraSpeck, 0.5 µm, ThermoFisher) in 1% agarose gel using the second of the two microscopes described across a volume of 600 × 600 × 250 µm. The lateral PSF dimensions, as determined using PSFj [19] were calculated as 755 ± 22 and 861 ± 48 nm (full-width at half-maximum). The discrepancy between the two axes is typical and owes to mild astigmatism resulting from imaging through an agarose gel cylinder, which exhibits a weak focusing effect in one axis. The axial PSF dimension was 5.25 ± 0.268 µm. The absence of notable background and well confined axial PSF are indicative of the excellent planewise confinement of the light sheet across the FOV under pivoting conditions.
Next, we demonstrate the effective destriping of the SEPS scheme under quasi-static conditions (i.e., the changes in the sample were slow on the timescale of the imaging carried out). For this demonstration we noted that stripes were most noticeable for densely labeled structures. Zebrafish larvae (72 hours post fertilization, hpf), see Supplement 1 for a statement regarding zebrafish handling) were stained with a live imaging compatible lipophilic dye, Bodipy TR methyl ester (Thermo Fisher), which rapidly and near uniformly stains the zebrafish, thus providing an extremely challenging candidate for destriping. A z-stack of images of the zebrafish were acquired in the absence of pivoting and with pivoting at 1, 2, 4 and 8× the frame rate (100 fps, 100–800 Hz) to assess whether the larger number of pivots per exposure of conventional mSPIM had an effect on destriping under these conditions. To assess the destriping, one requires a quality metric. Since the stripes constitute information, albeit artifactual, simple statistical measures such as the image variance or even more complex entropy-based metrics [20,21] will often determine that a destriped image has lower quality than a striped image. For this reason, we compared the relative ratio of information in the image x and y axes. Since the light sheet propagates along the system x-axis, the stripes will primarily appear as dark and light bands aligned to this axis. As such, the additional striping information will appear along the y-axis of the image. To view the relative effect of destriping on the two axes, the metric calculates the Fourier transform of the image and takes the ratio of the standard deviation along x and y (sum across all x and y pixel rows) as a measure of the information content:
The SEPS scheme inherently includes double-sided illumination. However, the images from Figure 5 are shown only for single-sided illumination. The corresponding images for illumination from either side and the fusion of the two illumination views (which are acquired as separate z-stacks following the scheme of Supplement 1, Supplementary Figure 2 c) are shown in Supplement 1, highlighting the well-characterized benefits to optical coverage afforded by double-sided illumination. Moreover, Visualization 1 demonstrates that the SEPS concept is capable of imaging with interleaved double-sided illumination (following the scheme of Supplement 1, Supplementary Figure 2 a). In this case, a transgenic zebrafish larva was imaged (Tg(kdrl:GFP)) at 11.6 frames per second.
Having demonstrated that the SEPS scheme operates effectively under quasi-static conditions, we move to a more challenging, dynamic environment. The embryonic zebrafish heart beats at 2 - 3 Hz and blood flow velocities may be as large as 1.5 m/s [22] presenting a challenge for destriping. To objectively compare the 0× and 1× cases, one must synchronize datasets with respect to the heart beat phase and analyze across multiple beats to average out variations owing to the variable number of blood cells present visible in a given beat. The details of this procedure are given in Supplement 1. To provide a counterstain to the specific contrast of an available transgenic zebrafish line (Tg(kdrl:Hsa.HRAS-mCherry, myl7:dsRed)) [23] where the vasculature and myocardium are labelled with the red fluorescent proteins mCherry and dsRed, respectively, zebrafish were stained with Bodipy 493/503 (ThermoFisher). The heart itself did not provide a good structure for comparison, again owing to the variable presence of blood cells. However, the region downstream (with respect to the propagation of the light sheet from left to right) of the atrium was ideally suited as a uniformly stained region of interest, over which the destriping could be assessed. This region is shown for a single heartbeat phase in 48 hpf zebrafish larvae with and without light sheet pivoting in Figure 6(a). The same region is shown for 9 consecutive heartbeats and an image formed as their average in Figure 6(b). One clearly sees a strong destriping effect with the light sheet pivoting even under these dynamic conditions.
One can define two types of stripes in the images, those arising from quasi-static features (present across all heart beats) and those arising from dynamic features (unique to each heartbeat). Across several heart beats, dynamic features should average out while quasi-static features persist. The topmost part of each ROI image corresponds to a region not obscured by the beating heart and so contains only quasi-static stripes. The lower part of the image contains primarily dynamic stripes. This is evident from the averaged images. For the 0× case, one clearly sees the static stripes at the top, while the lower part appears more uniform. Averaged over a larger number of heartbeats still, the dynamic stripes should fully disappear and any remaining stripes would be quasi-static in nature. In the 1× case, one sees suppression of both the quasi-static and dynamic stripes suggesting that even for a single pivot per image, the destriping effect was substantial. Moreover, the small feature in the top right of the ROI, which is not visible in the 0× images, became apparent in all images. Calculating an information ratio as previous suggests a 14% increase in the relative x to y content from the 0× to 1× case with a standard deviation of 3%. As such, all heart beats are expected to show a decrease in stripes for the 1× case. The heartbeat phase chosen is convenient for analysis purposes but from visual inspection one sees a robust destriping across heartbeat phases (Figure 6(c)). While the stripes were clearly present in the bodipy channel, ignoring any spectral differences in absorption/scattering, they were equally present in the dsRed/mCherry signal as shown in Figure 6(d).
Clearly, the SEPS destriping scheme has a measurable and positive effect on image quality even in the dynamic environment. However, it is possible that higher pivoting rates may more effectively destripe images where the timescale for changes in the stripe pattern is short with respect to the exposure time of the camera as may be the case in the heart. This effect was explored by following the procedure for imaging the zebrafish beating heart, but varying the frequency multiplier from 0 - 4× (0–800 Hz for imaging at 200 fps). Analysis of the destriping capability was performed across a similar ROI as previous, with the assessed region downstream of the atrium/ventricle with respect to the light sheet propagation. All non-zero pivoting frequencies resulted in a clear reduction of stripes compared with the 0× case. The most prominent stripes were present across all images, highlighting that multiple pivots per image do not visibly offer any major improvement to destriping. This assessment was supported by the Ix:y,relative values, which showed an increase of 35% from 0× to 1×, but only an additional 4% from 1× to 4× (Figure 7(a)). The zebrafish heart being a dynamic environment, benefited from faster imaging still to minimize motion artifacts. The required frame rate is thus proportional to the feature size to be resolved. Established methods for zebrafish heart imaging employ moderate-high NA objectives with low magnification to achieve high camera frame rates (400 fps) and so undersample with respect to the theoretically achievable optical resolution. Doubling the magnification may thus achieve higher resolution in principle, however the allowable motion blur should scale accordingly, thus requiring a doubled frame rate. Investigation of the resolution was outside the scope of this work, nevertheless, we demonstrate that at doubled magnification we were able to use SEPS to perform destriping in the beating zebrafish heart at 800 fps. The field of view height was limited to a quarter of that presented in Figure 6 (half due to the requirement for doubled frame rate, half due to the doubled magnification). The destriping effect of all 1× frames is clear, with some dynamic striping resulting from transient passage of blood cells remaining.
5. Discussion
High-speed light sheet imaging is made possible by modern sCMOS cameras operating at hundreds of frames per second. For imaging with double-sided illumination, ideally the volume would be imaged in an interleaved mode, whereby successive planes are illuminated from left and right sequentially. Doing so, requires use of the simulated global shutter mode of the camera, since any overlap between image i and i + 1 is not desirable owing to the alternate illumination directions. Typically, this requires that the switching time is less than or equal to the exposure time, thus allowing switching between successive laser exposures (Supplement 1). Switching on a timescale that could feasibly be as little as 1–2 ms is out of reach of most devices shown in Figure 1. Indeed, galvo settling times are typically on the order of 10 ms for the large changes in scan angle required. However, since the pivoting action for both light sheet directions is at one of its maxima where the KEM faces meet, one may continuously scan across the knife-edge to effectively transition between the two sheets with a near zero galvo step. Even allowing for some unused range at the knife-edge to avoid diffraction effects, the galvo settling time for small (>>1 degree) steps, is typically >> 1 ms. So, for interleaved double-sided imaging, there is effectively no limitation to the frame rate set by the galvo mirror. Clearly for the alternate case of sequential imaging of volumes, whereby the illumination side is switched only between volumetric acquisitions, the requirements for switching are more relaxed (Supplement 1). High-speed imaging capable of capturing e.g., the motion of the embryonic zebrafish heart is typically performed using only one illumination side [24,25]. In this case, the rolling shutter overlap mode of the camera may be used, thus effectively doubling the camera frame rate. For full sampling of all light sheet angles during a single camera exposure by every pixel line, one must perform the pivoting at an equal rate to the camera frame rate (Supplement 1).
Previously introduced methods for double-sided light sheet generation have required additional optics to deliver effective stripe artifact suppression [3]. Likewise, reported schemes for destriping have either been too slow/unsynchronized for imaging in the zebrafish beating heart [24,25], or make undesirable sacrifices with respect to parallelization [11,12]. While simple methods based on 1D diffusers could allow for destriping at the high frame required, current reports have only demonstrated this capability at up to 100 frames per second [12,13]. The SEPS scheme reported herein solves the dual challenges of efficiently delivering double-sided light sheet illumination while suppressing stripe artifacts even under the dynamic conditions present in the zebrafish beating heart. The scheme makes no sacrifices to the underlying SPIM architecture while offering these dual capabilities. Furthermore, the scheme is inexpensive to implement. Comparing with a typical mSPIM system utilizing a flip mirror for selection of one of the two illumination paths and a pair of resonant mirrors for light sheet pivoting, the SEPS scheme requiring only one movable element (as opposed to three) is typically cheaper. Correct implementation of the SEPS scheme does require greater care in alignment. In mSPIM it is generally necessary to very closely align the axes of the cylindrical lens, galvo scan and detection plane, particularly for large FOVs (where misalignment of the cylindrical lens axes is readily apparent) and at moderate-high NA (where the depth of field is small and so out of focus excitation is more likely to arise from relative misalignments between the three). However, the SEPS scheme must ensure the scan axis is perfectly aligned in two pathways, essentially precluding differential correction of the scan axis orientation between the two. For a given camera chip size, the alignment challenge scales as $\; C = N{A^2}/M$ (where M and NA are the system magnification and emission path numerical aperture respectively). Nevertheless, we have demonstrated that the scheme can be employed for M = 16, NA = 0.8 (C = 0.04), M = 10, NA = 0.6 (C = 0.036). The primary challenge for alignment concerns improper rotation of the scan axis through the several 90 ° periscopes required to guide light from the galvo mirror to the sample from two directions. With some care, the alignment can be achieved without improper rotation of this axis.
The SEPS scheme has been demonstrated to deliver effective destriping up to an imaging speed of 800 frames per second, twice that which is typically used for imaging at cellular resolution scale in the zebrafish beating heart. While an exposure time of 2.5 ms (giving a maximum camera frame rate of 400 fps) is considered sufficient to avoid motion blur on the scale of the microscope resolution, faster imaging still may allow for higher resolution, since the amount of blur permissible will scale accordingly. The galvo used presents a practical lower limit on the exposure time of ca. 1.25 ms, equivalent to 800 fps. This added capability should allow for stripe artifact suppression at shorter length scales or indeed for imaging other dynamic environments such as those associated with freely moving animals. For generation II sCMOS cameras, imaging at 800 fps, already necessitates a narrow field of view, typically insufficient to encompass the entire beating heart. Generation III sCMOS camera offer up to 5× higher imaging speeds. To be able to perform effective destriping at up to 2,000 fps will require alternative scan technologies capable of operating at 2 kHz synchronized or 20 kHz non-synchronized.
6. Summary
A single element pivoting switching (SEPS) scheme was introduced for cost and space-efficient high-speed double-sided light sheet delivery and stripe suppression. The SEPS scheme allows interleaved and sequential double-sided light sheet delivery with a speed limited only by the camera frame rate. Moreover, the destriping capability was demonstrated under pseudo-static and dynamic conditions. Surprisingly, pivoting at higher frequency harmonics of the camera frame rate was not associated with a significant improvement in image quality, demonstrating that pivoting is only necessary at the frame rate of the camera. We have demonstrated that the SEPS scheme is ideally suited to artifact-suppression in the widely studied zebrafish beating heart. This scheme opens the door for miniaturized multi-view light sheet microscopes free from performance sacrifices relative to their larger counterparts that use multi-element schemes for switching and pivoting.
Funding
Human Frontier Science Program (LT000321/2015-C); Morgridge Institute for Research.
Acknowledgments
We thank the members of the Huisken lab for useful discussions and critical comments on the manuscript.
Disclosures
The authors declare no conflicts of interest.
Data availability
The data presented in this paper are not publicly available but may be obtained from the authors upon reasonable request.
Supplemental document
See Supplement 1 for supporting content.
References
1. J. Huisken, J. Swoger, F. Del Bene, J. Wittbrodt, and E. H. K. Stelzer, “Optical sectioning deep inside live embryos by selective plane illumination microscopy,” Science 305(5686), 1007–1009 (2004). [CrossRef]
2. R. M. Power and J. Huisken, “A guide to light-sheet fluorescence microscopy for multiscale imaging,” Nat. Methods 14(4), 360–373 (2017). [CrossRef]
3. J. Huisken and D. Y. R. Stainier, “Even fluorescence excitation by multidirectional selective plane illumination microscopy (mSPIM),” Opt. Lett. 32(17), 2608–2610 (2007). [CrossRef]
4. G. De Medeiros, N. Norlin, S. Gunther, M. Albert, L. Panavaite, U. M. Fiuza, F. Peri, T. Hiiragi, U. Krzic, and L. Hufnagel, “Confocal multiview light-sheet microscopy,” Nat. Commun. 6(1), 8881–8888 (2015). [CrossRef]
5. T. V. Truong, W. Supatto, D. S. Koos, J. M. Choi, and S. E. Fraser, “Deep and fast live imaging with two-photon scanned light-sheet microscopy,” Nat. Methods 8(9), 757–760 (2011). [CrossRef]
6. G. Shah, K. Thierbach, B. Schmid, J. Waschke, A. Reade, M. Hlawitschka, I. Roeder, N. Scherf, and J. Huisken, “Multi-scale imaging and analysis identify pan-embryo cell dynamics of germlayer formation in zebrafish,” Nat. Commun. 10(1), 5753 (2019). [CrossRef]
7. B. Schmid, G. Shah, N. Scherf, M. Weber, K. Thierbach, C. P. Campos, I. Roeder, P. Aanstad, and J. Huisken, “High-speed panoramic light-sheet microscopy reveals global endodermal cell dynamics,” Nat. Commun. 4(1), 2207 (2013). [CrossRef]
8. G. Sancataldo, V. Gavryusev, G. de Vito, L. Turrini, M. Locatelli, C. Fornetto, N. Tiso, F. Vanzi, L. Silvestri, and F. S. Pavone, “Flexible Multi-Beam Light-Sheet Fluorescence Microscope for Live Imaging Without Striping Artifacts,” Front. Neuroanat. 13, 1–8 (2019). [CrossRef]
9. P. J. Keller, A. D. Schmidt, J. Wittbrodt, and E. H. K. Stelzer, “Reconstruction of zebrafish early embryonic development by scanned light sheet microscopy,” Science 322(5904), 1065–1069 (2008). [CrossRef]
10. L. Gao, L. Shao, C. D. Higgins, J. S. Poulton, M. Peifer, M. W. Davidson, X. Wu, B. Goldstein, and E. Betzig, “Noninvasive imaging beyond the diffraction limit of 3D dynamics in thickly fluorescent specimens,” Cell 151(6), 1370–1385 (2012). [CrossRef]
11. A. K. Glaser, Y. Chen, C. Yin, L. Wei, L. A. Barner, N. P. Reder, and J. T. C. Liu, “Multidirectional digital scanned light-sheet microscopy enables uniform fluorescence excitation and contrast-enhanced imaging,” Sci. Rep. 8(1), 13878 (2018). [CrossRef]
12. P. Ricci, G. Sancataldo, V. Gavryusev, A. Franceschini, M. C. Müllenbroich, L. Silvestri, and F. S. Pavone, “Fast multi-directional DSLM for confocal detection without striping artifacts,” Biomed. Opt. Express 11(6), 3111 (2020). [CrossRef]
13. M. A. Taylor, G. C. Vanwalleghem, I. A. Favre-Bulle, and E. K. Scott, “Diffuse light-sheet microscopy for stripe-free calcium imaging of neural populations,” J. Biophotonics 11(12), 1–9 (2018). [CrossRef]
14. S. M. Salili, M. Harrington, and D. J. Durian, “Note: Eliminating stripe artifacts in light-sheet fluorescence imaging,” Rev. Sci. Instrum. 89(3), 036107 (2018). [CrossRef]
15. N. Vladimirov, F. Preusser, J. Wisniewski, Z. Yaniv, R. A. Desai, A. Woehler, and S. Preibisch, “Dual-view light-sheet imaging through a tilted glass interface using a deformable mirror,” Biomed. Opt. Express 12(4), 2186 (2021). [CrossRef]
16. J. R. Landry, R. Itoh, J. M. Li, S. S. Hamann, M. Mandella, C. H. Contag, and O. Solgaard, “Tunable structured illumination light sheet microscopy for background rejection and imaging depth in minimally processed tissues,” J. Biomed. Opt. 24(04), 1 (2019). [CrossRef]
17. K. M. Dean, P. Roudot, E. S. Welf, T. Pohlkamp, G. Garrelts, J. Herz, and R. Fiolka, “Imaging subcellular dynamics with fast and light-efficient volumetrically parallelized microscopy,” Optica 4(2), 263 (2017). [CrossRef]
18. P. Theer, C. Mongis, and M. Knop, “PSFj: know your fluorescence microscope,” Nat. Methods 11(10), 981–982 (2014). [CrossRef]
19. C. J. Greer and T. E. Holy, “Fast objective coupled planar illumination microscopy,” Nat. Commun. 10(1), 4483 (2019). [CrossRef]
20. J. He and J. Huisken, “Image quality guided smart rotation improves coverage in microscopy,” Nat. Commun. 11(1), 150 (2020). [CrossRef]
21. L. A. Royer, W. C. Lemon, R. K. Chhetri, Y. Wan, M. Coleman, E. W. Myers, and P. J. Keller, “Adaptive light-sheet microscopy for long-term, high-resolution imaging in living organisms,” Nat. Biotechnol. 34(12), 1267–1278 (2016). [CrossRef]
22. H. C. Yalcin, A. Amindari, J. T. Butcher, A. Althani, and M. Yacoub, “Heart function and hemodynamic analysis for zebrafish embryos,” Dev. Dyn. 246(11), 868–880 (2017). [CrossRef]
23. N. C. Chi, R. M. Shaw, S. De Val, G. Kang, L. Y. Jan, B. L. Black, and D. Y. R. Stainier, “Foxn4 directly regulates tbx2b expression and atrioventricular canal formation,” Genes Dev. 22(6), 734–739 (2008). [CrossRef]
24. M. Mickoleit, B. Schmid, M. Weber, F. O. Fahrbach, S. Hombach, S. Reischauer, and J. Huisken, “High-resolution reconstruction of the beating zebrafish heart,” Nat. Methods 11(9), 919–922 (2014). [CrossRef]
25. M. Weber, N. Scherf, A. M. Meyer, D. Panáková, P. Kohl, and J. Huisken, “Cell-accurate optical mapping across the entire developing heart,” Elife 6, 1–14 (2017). [CrossRef]
