Development a flexible light‐sheet fluorescence microscope for high‐speed 3D imaging of calcium dynamics and 3D imaging of cellular microstructure

We report a flexible light‐sheet fluorescence microscope (LSFM) designed for studying dynamic events in cardiac tissue at high speed in 3D and the correlation of these events to cell microstructure. The system employs two illumination‐detection modes: the first uses angle‐dithering of a Gaussian light sheet combined with remote refocusing of the detection plane for video‐rate volumetric imaging; the second combines digitally‐scanned light‐sheet illumination with an axially‐swept light‐sheet waist and stage‐scanned acquisition for improved axial resolution compared to the first mode. We present a characterisation of the spatial resolution of the system in both modes. The first illumination‐detection mode achieves dual spectral‐channel imaging at 25 volumes per second with 1024 × 200 × 50 voxel volumes and is demonstrated by time‐lapse imaging of calcium dynamics in a live cardiomyocyte. The second illumination‐detection mode is demonstrated through the acquisition of a higher spatial resolution structural map of the t‐tubule network in a fixed cardiomyocyte cell.

been demonstrated using an electrically-tunable lens (ETL) to sweep the detection focal plane in synchrony with the axial position of the illumination sheet at 30 volumes per second (vps) [3]. However, ETLs only allow low-order defocus wavefront aberration to be applied and are therefore not so well-suited to refocusing high numerical aperture microscope objectives [4]. A tunable acoustic gradient lens has been used to achieve LSFM at 100 vps [5], but also with a relatively low NA detection lens. Oblique plane microscopy (OPM) uses a single microscope objective for both illumination and detection and has been used to achieve 3D LSFM imaging with sub-cellular resolution at 21 vps with an EMCCD camera [6]. This approach was combined subsequently with an additional optical relay allowing a rotating polygonal mirror to be used for sample-scanning at 20 vps with a sCMOS camera over a larger field of view but with lower spatial resolution [7], and more recently with a galvo mirror, allowing imaging at up to 321 vps with an intensified CMOS camera [8].
Other approaches to high-speed volumetric imaging include extended depth-of-field imaging in the detection path coupled with galvo-scanning of the position of the illumination plane, which has been demonstrated at up to 73 vps [9]. This approach uses a deformable mirror in a plane conjugate to the pupil of the collection objective to impart for example a cubic phase profile that, when coupled with deconvolution, enables an extended depth of field to be obtained. This approach only requires mechanical scanning of the axial position of the excitation sheet, but the performance of the deconvolution step depends on signal-to-noise ratio (SNR) and/or how sparsely the specimen is labelled. The light-field microscopy approach has also been used to demonstrate volumetric imaging at 200 vps [10], but again the SNR and/or the sparsity of fluorescence labelling affects the performance of the reconstruction algorithm.
We previously used OPM with a sCMOS camera to image dynamic calcium events in isolated cardiomyocytes [11]. In this previous work, the image acquisition speed was limited by the frame rate of the camera employed and by phototoxicity considerations. We therefore set out to explore other approaches that could increase the spatial resolution and image acquisition speed achievable, with the aim of also using this instrument to study dynamic calcium and voltage events in larger cardiac tissue slices in the future.
We report an inverted LSFM system with separate illumination and detection objectives. The design builds on previous inverted LSFM configurations such as objectivecoupled planar illumination (OCPI) microscopy [12] and inverted selective plane illumination microscopy (iSPIM) [13], where excitation and emission objectives lie in a vertical plane and use water immersion dipping objectives that sit in the sample mounting medium with no coverslip between objective and sample. For the high temporal resolution mode, we employed the multidirectional selective plane illumination microscopy (mSPIM) light sheet illumination mode [14], where an angle-dithered Gaussian beam focused in one direction is used for light-sheet illumination. High-speed 3D imaging is achieved by scanning the position of the detection focal plane along the detection objective's optical axis by using the folded remote refocusing approach [15] in synchrony with a galvo-scanning mirror in the excitation path that sweeps the axial position of the illumination beam. The remote refocusing configuration allows for rapid scanning of the image plane while avoiding mechanical interference with the sample compared to moving the primary LSFM objective [12]. The quadratic relationship between Gaussian beam waist size and confocal parameter leads to a design trade-off between optical sectioning and field of view. At the expense of optical sectioning, we under-filled the back focal plane of the illumination objective so the confocal parameter matched the diameter of a single cardiomyocyte cell.
For slower, higher resolution imaging, we designed a digitally scanned light-sheet microscopy (DLSM) light sheet mode, where a virtual light-sheet is generated by scanning a Gaussian beam focused in both directions across the sample [16]. For improved axial resolution, we expanded the size of the beam at the back focal plane of the illumination objective. In addition, to ensure a spatially invariant point spread function (PSF) across the field of view, we also implemented axially swept light-sheet microscopy (ASLM) [17][18][19].
For optical alignment, optical modifications and to image different types of samples we mounted the entire optical assembly on a single rotatable breadboard so illumination and detection objectives could be oriented with their optical axes lying in either a vertical or horizontal plane.
We present a characterisation of the system's spatial resolution by imaging thin layers of submicron fluorescent beads on microscope slides. To demonstrate the high temporal resolution mode, we show 3D time-lapse imaging of calcium dynamics in isolated cardiomyocytes. To demonstrate the higher-resolution imaging mode, we imaged a fixed cardiomyocyte cell with a fluorescently labelled membrane to generate a structural map of the t-tubule network.

| Preparation of thin layer of fluorescent beads
To generate a thin layer of fluorescence beads for characterising the system point spread function (PSF), a stock solution of 200 nm fluorescent microspheres (T7280, TetraSpeck, Thermofisher Scientific) was diluted in Milli-Q (MQ) by 500:1. To avoid aggregated microspheres, stock solution was agitated for 10 seconds on an Eppendorf vortex prior to dilution. Clean microscope slides were first coated in poly-L-lysine by pipetting on 1 mL of 1% poly-L-lysine dissolved in MQ and leaving for 10 minutes. After aspirating remaining liquid with pipette, excess 1% poly-L-lysine was washed away by pipetting MQ against the slide surface. A volume of 250 μL of bead solution was then spread across slide and left for 10 minutes. Excess solution was then removed by gently pouring MQ against the face of the slide.

| Isolation of left ventricle rat heart cardiomyocytes
All studies were carried out with the approval of the local Imperial College ethical review board and the Home Office, UK and in accordance with the United Kingdom Home Office Guide on the Operation of the Animals (Scientific Procedures) Act 1986, which conforms to the Guide for the Care and Use of Laboratory Animals published by the U.S. National Institutes of Health under assurance number A5634-01.

| Preparation of fixed WGA-stained isolated cardiomyocytes
The suspended cells were attached to 25 mm × 25 mm glass slides using a thin layer of mouse laminin (Gibco) diluted with phosphate buffered saline (PBS) and incubated at 37 C, 3.0% CO 2 for 40 minutes. The cells were then fixed with 4% paraformaldehyde (PFA) (Honeywell) in PBS and treated with 5% Triton (Sigma Aldrich) to enhance the cell permeability to the stain. The fixed cells were stained with 5% WGA, Alexa Fluor 594 conjugate (Life Technology) for 30 minutes and contained in PBSfilled wells of a 6-well plate in the dark prior to imaging.

| Preparation of live dual-labelled cardiomyocytes
To simultaneously monitor the intracellular calcium dynamics and intracellular t-tubule structure, we implemented dual labelling of the isolated cells with the cell permeant calcium indicator Fluo-4AM (ThermoFisher Scientific), and CellMask Orange (CMO) plasma membrane stain (ThermoFisher Scientific). The cells were first incubated with 0.16% pluronic acid and 10 μM Fluo-4AM in DMSO for 25 minutes at room temperature, followed by the addition of 5 μg ml −1 CMO at 37 C for 5 minutes. During the incubation period, the cells were moved gently on a rocking shaker and protected from light. The cells were then collected by low-speed centrifugation, resuspended in the 1:1 mixture of DMEM and low Ca 2+ solution and kept in the dark prior to imaging.
For imaging, the cells were attached to a microscope slide using mouse laminin (Gibco) and bathed in Normal Tyrode (NT) containing (mM): NaCl (140), KCl (6), Glucose (10), HEPES (10), MgCl 2 (1), CaCl 2 (1), pH adjusted to 7.4 with NaOH. A 2 mm-high silicone strip around the perimeter of the slide retained the NT, and two 0.8 mm diameter platinum electrodes (PT005149, Goodfellows) separated by 5 mm were used to deliver field stimulation. Prior to imaging, the cells were field-stimulated with 2 ms pulses at 0.5 Hz for 30 s at ×1.5 the threshold voltage for contraction.

| Microscope optomechanical design
As shown in Figure 1A, the system can be divided into excitation optics (highlighted by beam paths with blue lines) and detection optics (highlighted by beam paths with green and orange lines) and these optics meet at the primary LSFM excitation and detection objectives (O3 and O4). To aid sample navigation, a low-magnification widefield microscope is positioned beneath the LSFM. A full list of parts is given in Table S1.
Referring to Figure 1A and starting with excitation optics, the laser engine delivers five laser lines (425, 488, 515, 561, and 642 nm) to the optical breadboard via a polarisation maintaining single mode fibre. Objective lens (O1) collimates the Gaussian beam from the delivery fibre and after mirror M1, a pair of motorised flip mirrors (FM1 and FM2) are used to remotely switch between an mSPIM light-sheet mode and a DSLM light-sheet mode. For the mSPIM mode, the Gaussian illumination beam is focused in only one direction and is less tightly focused than for the DSLM mode.
In the mSPIM beam-path, the galvo scanning mirror (G1 θ-scan) is positioned in a plane conjugate to image planes of O2 and O3 for angle dithering of the light sheet to remove sample-induced shadow artefacts. Cylindrical lens (C1) is placed at its focal length from G1 and produces a focused waist that is conjugate to the back focal panes of O2 and O3. Imaging of the waist to the pupil of O2 is achieved using a 4-f pair of Plössl lenses (P1 and P2) with unit magnification.
For the DSLM mode the Gaussian beam diameter after O1 is scaled up to approximately fill the back apertures of O2 and O3 by using a ×5 beam expander consisting of a pair of achromatic doublets (D1 and D2). Linearly polarised excitation light from either the mSPIM or DSLM beam path is then coupled into a folded remote refocusing unit for X 0 positioning of the illumination beam waist in the sample volume. This comprises polarising beam splitter PBS1, quarter waveplate QWP1, microscope objective O2 and remote scanning mirror RM1. In order to achieve imaging of RM1 into the sample with equal lateral and axial magnification, the magnification from RM1 to the sample is chosen to be equal to the ratio of the refractive indices of the two spaces n sample / n remote = n water /n air = 1.33 [15]. O2 images the light sheet onto RM1 (7 mm diameter mirror), which is mounted on a linear voice-coil stage. For Y 0 and Z 0 positioning of the excitation beam, a pair of galvo scanning mirrors (G2 Z 0 scan and G3 Y 0 -scan) are used. Plössl lenses P3 and P4 form a 4-f relay, imaging the pupil of O2 onto G2 with ×0.5 magnification. P5 and P6 provide a ×1 magnification, 4-f image relay between G2 and G3 and 4-f relay P7&8 images G3 onto the back focal plane of O3 with a magnification of 2.67.
Microscope objectives O3 (×10 water dipping objective, 0.3 NA, N10XW-PF, Thorlabs) and O4 (×20 water dipping objective, 1 NA, 421452-9900-000, Zeiss) form the primary excitation and detection LSFM objectives, respectively. The optical axes of O3 and O4 are fixed to be parallel to the optical breadboard and orthogonal to each other by mounting them both on a custom-built monolithic block of aluminium. For precise control of the axial positions of O3 and O4 they are attached to the block via separate micrometre stages (M-SDS40, Newport).
The sample is mounted between the LSFM primary objectives on a three-axis stage that is itself mounted on a motorised linear translation stage (motorised linear translation stage MFA-CC and control box SMC100CC, Newport), which is used for stage-scanned volumetric image acquisition. Below the sample a low-magnification wide-field microscope (O6, D3, M14 and CCD) is used to guide coarse positioning of the sample before high magnification LSFM imaging.
For the detection path, a second folded remote refocusing system is used consisting of the microscope formed by O4 and T1 and the microscope formed by O5 and P9. Together, these two microscopes achieve imaging with equal lateral and axial magnification of the sample onto RM2. RM2 is mounted in a piezoelectric actuator that allows rapid axial sweeping of the detection beam-path focus. PBS2 and QWP2 are used to couple fluorescence out of the remote refocusing unit into a dual-spectral channel imaging unit.
The excitation light is linearly polarised in the Y 0 direction and therefore the fluorescence emission will be preferentially polarised in the Y 0 direction. This effect will be greatest for fluorophores with a high fluorescence anisotropy. In addition, the angled dielectric mirror coating on M11 may act as a retarder. Therefore, the half-wave plate (HWP) in the detection path was rotated to maximise the fluorescence signal transmitted by PBS2 and therefore maximising the fluorescence transmitted by the detection beam path.
Following the remote refocusing unit, a twodimensional adjustable slit (2DS) is used to limit the size of the image formed by T2. To image two spectral windows on a single camera simultaneously, mirrors M12 and M13, dichroic beam splitters DM1 and 2 (both 560 nm long pass, FF560-FDi01 −25 × 36, Semrock) and emission filters F1 and 2 (ET525/50 and ET630/75, Chroma, respectively) and T4 are used to spatially separate the two images on the camera (sCMOS).

| Microscope hardware control, data acquisition and processing
For hardware control, data acquisition and processing, a HP Z840 Workstation with dual Intel ® Xeon ® Processors (3.40 GHz, 6 cores each), 128 GB RAM and a pair of 2TB SSDs (SSD 960 PRO 2TB, Samsung) in software RAID0 was used.
The acquisition software was written in LabVIEW. For accurate synchronisation of hardware, a digital acquisition (DAQ) box (USB-6363, NI) was used to trigger the camera and other hardware, that is, modulation of lasers, angular positions of galvo scanning mirrors, positions of the motorised sample stage and positions of the illumination and detection remote refocusing linear actuators.
MATLAB and ImageJ were used for image processing and data analysis.

| MSPIM and remote refocusing imaging mode
A summary of the two excitation modes is given in Table S2. For video-rate 3D imaging of isolated cardiomyocytes, the system was configured to use mSPIM light-sheet illumination. The focal plane of the detection imaging system was controlled by piezoelectric actuation of RM1, which was driven with an asymmetric saw-tooth motion profile with a 40 ms period. For the first 30 ms of each period, the waveform was increased linearly and the remaining 10 ms was used for fly-back. To minimise high frequency noise due to resonance in the mechanical system, the saw tooth profile was digitally low-pass filtered. The resulting linear region of the profile used for imaging was 28 ms. The amplitude of the mirror's motion profile was~33 μm which after accounting for the remote refocus double-pass arrangement and axial magnification of 1/1.33 between RM1 and the sample, corresponds to a 50 μm remote refocusing range in sample space. During volumetric acquisition, actual positions of the mirror were recorded by a capacitive sensor (E-509.C1A, PI). To record a single volume, the camera was externally triggered to acquire 50 images equally spaced in time during the first 30 ms of the refocusing period with~1 μm axial spacing.
Prior to acquiring volumetric image data, the asymmetric saw-tooth motion profile was sent to the piezoelectric actuator of RM1 and the actual mirror position was recorded by the capacitive sensor. The measured mirror motion was then used to generate a matched voltage waveform that was subsequently used during image acquisition to drive the G2 Z 0 scan galvo-this ensured that the Z 0 position of the illumination sheet matched accurately the position of the detection focal plane.
Fluo 4 and CMO were excited using the laser lines at 488 and 561 nm with fluorescence emission detected using filters ET525/50 and ET630/75, respectively. The excitation power per laser line measured at the back aperture of the primary excitation objective was 400 μW. This power was distributed over the whole field of view of the illumination objective corresponding to a diameter of 2.2 mm. This is larger than the field of view of the sCMOS camera of 256 × 50 μm and larger than a single myocyte (~100 μm), so only a fraction of the measured excitation power was incident on the cell.
The spectral image-splitter was used to acquire two spectral channels simultaneously. The two channels were then co-registered. The registration parameters were determined by acquiring using correction parameters determined from a transmitted-light image of a USAF 1951 test chart (Edmund Optics, Barrington, USA). The resulting image was then split into two and the two channels overlaid, with one channel subject to an affine transformation. The transformation was then manually adjusted to achieve the best registration and the same parameters were used for all subsequent image processing.
A sCMOS camera (Hamamatsu ORCA-Lightning) was externally triggered (synchronous readout trigger mode) by TTL pulses from the DAQ with a 0.5 ms exposure time and a peak frame rate of 1754 frames per second. A total of 750 volumes of 50 planes per volume in two spectral channels were acquired using a 2048 × 200 (H × V) region of interest (ROI) equating to 1024 × 200 × 50 voxels per spectral channel per volume. The total time-averaged voxel rate across both spectral channels was 0.51 GHz. To balance SNR against spatial resolution, the final tube lens (T4 in Figure 1A) was chosen to give a magnification of ×22 to 2.s.f, equivalent to a sample plane pixel size of 0.25 μm.
The light sheet was angle dithered by G1 with a sinusoidal waveform synchronised with the camera acquisition so that each camera frame was recorded during half of the sinusoid's period with constant phase offset.

| Stage-scanned DSLM + ASLM image acquisition mode
A stage-scanning image acquisition mode was used together with DSLM and ASLM in order to achieve a better axial resolution that is more uniform across the field of view compared to the mSPIM mode. The motorised stage scans the sample along the X direction (see Figure 1A). In this configuration, the final tube lens (T4 in Figure 1A) was chosen to give a magnification of ×44 to 2.s.f and an effective pixel size in the sample plane of 0.125 μm.
The camera was triggered at 20 Hz via the DAQ box to acquire a frame for every 0.1 μm of stage travel during continuous stage motion over a distance of 100 μm at a velocity of 5 μm s −1 , resulting in 1000 acquired images. For analysis, the images volumes were de-sheared in MATLAB into the light-sheet coordinate frame (see Figure 1A).
The DSLM illumination was produced by driving G3 (Y 0 -scan) with an asymmetric saw-tooth waveform profile at 20 Hz to match the camera frame rate. A 10 ms laser illumination pulse and camera exposure time of 10 ms were triggered simultaneously.
For ASLM acquisitions, a stage-scanned volume was acquired for several positions of RM1 in order to span the sample's extent in the X 0 direction. The confocal parameter of the DSLM illumination was~9 μm and therefore four contiguous RM1 positions spaced by 9 μm in sample space, to cover a total extent of~36 μm in the X 0 direction, were used for the data presented.
The camera was operated at 20 Hz and 1000 image frames were acquired for each of the four ASLM positions. When including the time taken for the stage to return to the start position after each scan, this resulted in a total image acquisition time of 210 s.
Following acquisition of the multiple image volumes, the regions containing the illumination beam waist for each ASLM position were extracted and combined in MATLAB to generate a volume with uniform optical sectioning across a field of view that is greater than the illumination beam confocal parameter. We note that ASLM is usually implemented with the axially-swept beam waist synchronised with the rolling-shutter of the sCMOS camera, see Section 4.
To image fluorescence from WGA conjugated to Alexa-594, the 561 nm laser line was used for excitation and a spectral window corresponding to emission filter ET630/75M was used in the detection path.

| Quantification of System PSF
The spatial resolution of the system was characterised by imaging thin layers of fluorescent beads (see above). Fluorescence was excited using the 488 nm laser line and detected through a ET525/50 emission filter. Images were recorded using the stage-scanning image acquisition mode with 0.1 μm between frames. For each remote refocus position, image volumes were acquired at three different sample heights in the Z direction, resulting in beads appearing in the raw acquired images at different positions along the light sheet (X 0 ) propagation direction. Two data sets were acquired, one with mSPIM illumination and one with DSLM + ASLM illumination. For mSPIM illumination, bead volumes were also acquired for three different remote refocus positions.
The acquired image stacks were de-sheared in MATLAB into the microscope frame and the volume was then axially resampled using bilinear interpolation to achieve a uniform voxel size in all three dimensions. A thresholding step was applied to create a binary mask of the volume and connected components within the mask were identified using MATLAB's bwconncomp function. The threshold was chosen by increasing the threshold value until the decreasing number of detected components reached a plateau. A cuboid ROI about the weighted centroid of each bead image was then selected for each bead. In order to select appropriate individual bead images for analysis, ROIs were excluded if the ROI touched the edge of the volume imaged, if the ROI overlapped with another ROI, and if the total volume or total intensity within the connected component fell outside the main peak in the distribution of that parameter over all connected components. This process was used to exclude small bright regions caused by hot sCMOS pixels and larger bright regions corresponding to clumps of multiple beads.
First, a three-dimensional Gaussian fit with offset is applied to the ROI in order to obtain an initial estimate of the position of the peak intensity in X 0 and Y 0 . Next, a 1D Gaussian fit with offset was applied axially (Z 0 ) through the laterally integrated intensity to determine the FWHM (Z 0 *) and axial position of the peak intensity. The X 0 Y 0 plane corresponding to the position of the axial peak intensity was then extracted and 1D Gaussian plus offset curves were fitted in the X 0 and Y 0 directions through the corresponding lateral centre position estimated by the initial 3D fit. Finally, a 1D axial fit through the X 0 and Y 0 centres found from the two 1D lateral fits is used to estimate the axial FWHM in the Z 0 -direction. with positive RF indicating increasing distance from O4. Gaussian fits to the data for an exemplar bead image is shown in Figure S1A and the corresponding histograms for all beads quantified for zero remote refocus and at the light-sheet beam waist are shown in Figure S2A. The corresponding results from the PSF characterisation method are presented in Table 1. For the mSPIM mode, the average measured bead image FWHM in the plane of the light sheet over all distances across the FOV and overall RF positions in the X 0 and Y 0 directions are 0.51 and 0.47 μm, respectively. The FWHM in the X 0 direction is slightly higher than for the Y 0 direction, which is due to the stage motion in the X direction during the sCMOS exposure time and due to mechanical vibrations of the translation stage used that can be seen in the X 0 Z 0 profiles shown in Figure 2. Accounting for the 200 nm bead diameter, the 0.125 μm pixel size and stage scan motion in the X direction during the camera exposure, and assuming these can be approximated by independent Gaussian distributions of equivalent FWHM gives an Note: Data was obtained from image stacks of 200 nm fluorescent beads and compared for positive and negative remote refocus (RF) and different positions across the FOV along the light-sheet propagation direction. The bead FWHM are estimated from the FWHM (mean ± SD) of unweighted 1D Gaussian fits to linear profiles through subvolumes containing the bead images. The total number of beads analysed in each measurement is indicated by n. estimated measured PSF FWHM in the X 0 and Y 0 directions of 0.46 and 0.40 μm, respectively. For comparison, the theoretical scalar detection PSF lateral FWHM for an emission wavelength of 525 nm and NA 1.0 is 0.27 μm. The average measured 1D axial (Z 0 ) bead FWHM over all RF positions for the centre of the FOV is 1.64 μm. Removing a Gaussian combination of bead size and stage motion in Z 0 direction over the sCMOS integration time gives an estimated measured PSF FWHM of 1.63 μm, which compares reasonably well with the theoretical lower limit estimated from the 1D axial detection scalar PSF FWHM of 1.24 μm. The corresponding average laterallyintegrated axial bead (Z 0 *) FWHM over all RF positions for the centre of the FOV is 1.79 μm and accounting for bead size and stage motion gives an estimated FWHM of 1.78 μm. For all RF positions, the axial resolution is reduced away from the centre of the FOV as expected due to the Gaussian illumination profile, see Figure 2A. The estimated beam FWHM-based on the measured beam waist after O1, accounting for the nominal magnifications of the train of relay optics and focusing by O3-is 2.2 μm.

| DSLM + ASLM imaging mode
Exemplar bead images obtained from the DSLM + ASLM imaging mode are shown in Figure 2B with quantitative values reported in Table 1. Gaussian fits to the data for an exemplar bead image is shown in Figure S1B and the corresponding histograms for all beads quantified at the light-sheet beam waist are shown in Figure S2B. The average X 0 and Y 0 bead image FWHM over all positions in the FOV for this mode are 0.52 and 0.45 μm, respectively. Accounting for bead size, pixel size and stage motion gives 0.46 and 0.38 μm, respectively. Again, the resolution in the X 0 direction is slightly poorer due to stage motion and vibrations along the stage-scan (X) axis. The average 1-D and spatially averaged bead FWHM for 3.2 | 3D imaging of calcium dynamics in isolated cardiomyocytes at 25 volumes per second Figure 3 shows two-colour, 3D time-lapse imaging of a spontaneous calcium wave in a live cardiomyocyte cell acquired at 25 volumes per second with 50 planes per volume by remote refocusing. The red channel shows cell membrane labelled with CMO and the green channel shows fluorescence from intracellular calcium concentration indicator Fluo-4. The cell was electrically paced prior to the acquisition and then imaged without stimulation for 30 seconds (see Section 2 and also Video S1). Midway through the time-lapse, a spontaneous calcium wave and associated cellular contraction was observed. Figure 3A shows a montage of central orthogonal planes from a temporally contiguous subset of volumes lasting 2.8 seconds around this event. As can be seen from Figure 3A,B, the calcium wave propagates from the centre of the cell outwards. Figure 3C,D show the same time points but using orthogonal maximum intensity projections (MIP). To further show the spatio-temporal resolution of this mode, Figure S3 plots temporal profiles of Fluo-4 intensity for individual voxels (0.25 × 0.25 × 0.5 μm 3 ) within the cell shown in Figure 3.
Across the time-lapse the periodic structure of the membranes can be seen. Larger uniformly fluorescent regions that are particularly visible in the MIP images are attributed to membrane fragments from other dead cells in the suspension prior to mounting on the microscope. Across the time lapse we see calcium dynamics in the context of the t-tubule/membrane structure.

| Higher spatial resolution imaging of T-tubule membrane network in isolated cardiomyocytes
To illustrate the relative improvement in spatial resolution of DSLM + ASLM light sheet mode compared to the F I G U R E 4 3D imaging with stage scanning image acquisition and DSLM+ASLM of a fixed cardiomyocyte fluorescently labelled with wheatgerm-agglutinin (WGA) conjugated to Alexa-594. A shows central orthogonal planes when using mSPIM light sheet excitation and (B) shows central orthogonal planes when using DSLM light sheet excitation combined with ASLM. To compare the spatial resolution between the mSPIM and DSLM +ASLM modes, intensity line profiles corresponding to the yellow lines in the central orthogonal planes shown (A) and (B) are plotted in C to E. The primed Cartesian coordinate labels Y 0 -X 0 , Y 0 -Z 0 and Z 0 -X 0 refer to the plane orientation w.r.t the light sheet coordinate frame of reference mSPIM light sheet mode on a biologically relevant sample, both modes were used to image the t-tubule membrane network of a cardiomyocyte. The cell was fixed and fluorescently labelled with WGA conjugated to Alexa-594 and both illumination modes were used to acquire volumes with stage scanning. Figure 4A shows central orthogonal planes when using the mSPIM illumination mode and Figure 4B shows central orthogonal planes when using the DSLM + ASLM illumination mode. The improved optical sectioning provided by the DSLM + ASLM mode is especially noticeable at the edge of the cell membrane, and is most apparent in the Z 0 -Y 0 planes shown in Figure 4. Line profiles across the orthogonal central planes shown in Figure 4C-E show quantitatively that the DSLM + ASLM mode improves axial sectioning while lateral resolution is comparable to the mSPIM mode as expected. Figure S4 shows the constituent raw ASLM images for different illumination beam axial focus positions that were fused computationally to generate the data shown in Figure 4.

| DISCUSSION
Compared to the previous demonstrations of time-lapse volumetric imaging at video rate or faster cited in the Introduction, our light-sheet approach provides the best spatial resolution demonstrated to-date. The work of Fahrbach et al. [3] did not characterise the spatial resolution, but the theoretical limit can be inferred from the 0.3 NA illumination and 0.3 or 0.5 NA detection lenses employed. In SCAPE [7], the spatial resolution was reported as (2.5, 2, 3.6 μm) in Figure S8 of Reference [7]. In SCAPE 2.0 [8], the spatial resolution was reported as (1.21, 0.60, 1.55 μm) in Supplementary Note 1 Figure 1C of Reference [8].
In the mSPIM mode, the image acquisition rate was limited by the frame rate of the sCMOS camera employed. For a ROI of 2048 × 200 pixels the maximum frame rate achievable was 1754 fps. The rate of refocusing was limited by the power of the piezo-actuator used to axially translate O5. In this work, we acquired images as O5 is swept at 1.1 μm ms −1 . During the return sweep, O5 was swept at achieved an axial refocus velocity of 3.3 μm ms −1 and therefore this actuator is able to achieve an axial scan range of 50 μm in the sample at a frequency of 75 Hz. Faster sweep rates could be achieved by using a more powerful electronic amplifier and/or by reducing the mass of the mount used to hold the remote scan mirror RM2 (currently~13 g).
In the DSLM + ASLM mode, in the future, it will be possible to use rolling shutter detection on the sCMOS camera synchronised with the DSLM illumination mode to achieve 1D confocal detection to reduce scattered light when imaging in more scattering samples [21,22]. For the work presented here, the sCMOS camera was operated at 20 fps for a ROI of 1152 × 1152 pixels 2 in order to avoid any potential stage motion-induced mechanical vibrations of the sample. In principle, the speed of this acquisition mode could be increased to 178 fps for the Hamamatsu Fusion sCMOS camera or to 540 fps for the Lighting Hamamatsu sCMOS camera. If the sCMOS camera is rotated by 90 , it will instead be possible to synchronise the rolling shutter with the ASLM imaging mode, which would enable the ASLM image acquisition speed to be significantly increased [18] by avoiding the need to acquire multiple images for each stage position. However, this would not be compatible with 1D confocal detection.
The spatial resolution of the optical system is not diffraction limited in the X 0 and Y 0 directions. The theoretical diffraction-limited lateral resolution, that is excluding any motion of the translation stage during image acquisition and detector pixel size, should be 0.27 μm for a detection wavelength of 525 nm. Replacing PBS2 with a mirror so as to bypass O5 yielded no improvement in spatial resolution (data not shown) and so these components are not thought to be the cause. We also simulated the performance of the train of lenses T1, P9 and T3-5 in OpticStudio (v. 18.1, Zemax) and found it to be diffraction limited. We therefore suspect that the sub-diffraction-limited performance may be due to sub-optimal spacing of the Zeiss components O4, T1 and GB-the spacing between O4 and T1 was larger than the manufacturer's recommended value to provide sufficient space further down the optical system for placing PBS2 between P9 and O5. We have designed a new optical system that should overcome this limitation but have not implemented it yet. The spatial resolution of the illumination optical system in the DSLM + ASLM mode was also not diffraction limited and we attribute this to the λ RMS surface flatness (λ = 633 nm) of each of the three galvo mirrors employed. We hope to improve the illumination optical system in the future using mirrors with improved surface flatness.

| CONCLUSION
We have presented an LSFM system capable of operating in video-rate volumetric and lower-speed DSLM + ASLM LSFM imaging modes that is intended to study dynamic events in isolated cardiomyocytes and cardiac tissue slices.
In the higher-speed mode, we have demonstrated the first video-rate LSFM imaging employing the remote refocusing approach of Botcherby et al. [15] to scan the detection plane in synchrony with axial translation of the illumination light sheet. This has enabled a 1024 × 200 × 50 voxel volume to be acquired at 50 Hz (voxel acquisition rate of 0.51 GHz), and this was demonstrated by imaging single cardiomyocytes with sub-cellular spatial resolution. For a remote refocus range of 90 μm and across a field of view of 50 μm, the average spatial resolution for this imaging mode was determined to be 0.46 μm (X 0 ) and 0.40 μm (Y 0 ) in the plane of the light sheet. In the centre of the field of view, the laterally integrated 1D profile in the Z 0 direction, that is, the optical sectioning strength (Z 0 *), was 1.78 μm when averaged over a remote refocus range of 90 μm.
The use of ASLM in the lower-speed imaging mode enables an improved spatial resolution in the direction perpendicular to the light sheet and also enables this resolution to be maintained over a larger field of view. The spatially averaged PSF (Z 0 *) FWHM was found to be 1.19 μm in this mode across a field of view of 116 μm. We imaged a fixed cardiomyocyte and demonstrated the improved axial resolution obtained in this mode compared to the mSPIM illumination mode.