1. Introduction

The 3D imaging ability of selective plane illumination microscopy (SPIM), i.e. light sheet microscopy, relies on the intensity profile of the excitation light sheet used for 3D imaging. The thickness, light confinement ability, and size of the light sheet determine the axial resolution, optical sectioning ability, and field of view (FOV) of SPIM respectively [1,2]. In order to improve the 3D imaging ability of SPIM on large specimens, tremendous efforts have been spent on the optimization of the light sheet intensity profile, so that the excitation light can be confined near the detection focal plane over a long distance as much as possible [3–7]. Unfortunately, the diffraction of light makes it impossible to optimize these properties at the same time. The excitation light is less confined as the length of a light sheet increases, which makes it challenging to image large specimens at high spatial resolutions using SPIM. Methods other than optimizing the light sheet intensity profile were developed to solve the problem [8–14]. One of the most effective approaches is to quickly move the light sheet axially within the imaging plane along the excitation light propagation direction, so that a high spatial resolution and good optical sectioning ability can be maintained in a FOV much larger than the size of the light sheet [15–23]. Tiling light sheet selective plane illumination microscopy (TLS-SPIM) is a method using this strategy to improve the 3D imaging ability of SPIM on large specimens [19,20].

In TLS-SPIM, a large field of view (FOV) is imaged by tiling a short but thin light sheet at multiple positions within the imaging plane and taking an image at each light sheet tiling position. The final image is reconstructed using the raw images collected at all tiles. It has been demonstrated that TLS-SPIM is capable of imaging large multicellular specimens of different sizes, ranging from live embryonic specimens to physically expanded cells and optically cleared biological tissues at spatial resolutions higher than that of conventional light sheet microscopes [20–23]. In addition, TLS-SPIM could adjust the intensity profile and the tiling position of the tiling light sheet in less than a millisecond, so that its 3D imaging ability can be optimized in real-time based on the biological specimen and the biological process being imaged.

Despite the improved 3D imaging ability of TLS-SPIM, the additional camera exposures required by the light sheet tiling process cause a problem. The imaging speed decreases, and the raw image size increases proportionally to the number of tiles, i.e. the number of camera exposures, required per image plane. Although these problems are less of an issue when the number of tiles and the sample size are small, it could be troubling when large specimens are imaged using TLS-SPIM at high spatial resolutions, which requires a large number of tiles per image plane. For example, when TLS-SPIM is used to image optically cleared biological tissues at micron level spatial resolutions, the total imaging time could be extended by a few hours or more compared to a non-tiling condition despite the improved spatial resolution. Meanwhile, hundreds of gigabytes or even terabytes additional raw image data are generated by additional camera exposures, which must be collected and processed later. It creates a heavy burden on the limited data collection and analysis bandwidth of most imaging systems.

In this research, we developed a novel method to address the problem. Discontinuous light sheets, created by scanning coaxial beam arrays synchronized with the rolling shutter of an sCMOS detection camera (Fig. 1), are used in TLS-SPIM to image large specimens, by which the imaging speed of TLS-SPIM is increased, and the raw image data size is deceased proportionally to the number of discontinuous light sheet waists, i.e. the number of coaxial beams contained in the scanning coaxial beam array. We investigate the method via numerical simulations and demonstrate its imaging ability by imaging optically cleared mouse brain.

 

Fig. 1. The concept and operation modes of using discontinuous light sheets in TLS-SPIM for 3D imaging. (a) A discontinuous light sheet with multiple waists images a larger effective area at each tile than that of using a continuous light sheet with a single waist. It improves the imaging speed and decreases the raw data size at the same time. (b) The imaging plane is imaged by tiling the same discontinuous light sheet at multiple positions. (c) The imaging plane is imaged by using multiple different discontinuous light sheets compensating each other.

Download Full Size | PDF

2. Methods

In order to increase the imaging speed and produce less raw images in TLS-SPIM, the number of tiles required per image plane must be decreased. In other words, a larger effective area must be imaged at each tile during each camera exposure. Thus, the tiling light sheet must be enlarged along the light propagation direction, while the light sheet thickness and light confinement ability should remain the same to ensure the same spatial resolution and optical sectioning ability. Excitation light sheets obtained by scanning “non-diffracting” beams or beam arrays, such as Bessel light sheets and Lattice light sheets, have been developed for the purpose. However, the excitation light confinement ability of “non-diffracting” light sheets drops quickly when the light sheet size increases, which reduces the spatial resolution and optical sectioning ability significantly [1–6]. Therefore, “non-diffracting” light sheets can’t satisfy the requirements.

We seek a different solution to solve the problem. Instead of using light sheets with a continuous intensity profile, multiple light sheets separated far enough in the light propagation direction are used and tiled simultaneously for 3D imaging. It can also be considered as using a discontinuous light sheet with multiple waists instead of a single continuous waist (Fig. 1(a)) for 3D imaging. Obviously, the imaging speed increases, and the raw data size decreases proportionally to the number of light sheet waists. Discontinuous light sheets could be used in two ways in TLS-SPIM for 3D imaging. First, the imaging plane can be imaged by tiling the same discontinuous light sheet at multiple positions (Fig. 1(b)), which is the same as the operation in regular TLS-SPIM. Second, the imaging plane can be imaged by using multiple different discontinuous light sheets with light sheet waists compensating each other (Fig. 1(c)).

Discontinuous light sheets can be obtained by scanning coaxial beam arrays, which can usually be obtained by using diffraction optical elements (DOE). However, a DOE element only generates a coaxial beam array with a fixed intensity profile, which reduces the flexibility of TLS-SPIM. In addition, DOE elements usually work differently at different excitation wavelengths, which makes it difficult to maintain the same imaging quality for different excitation wavelengths. Therefore, we decided to generate coaxial beam arrays using the binary spatial light modulator (SLM) equipped in typical TLS microscopes, so that the microscope has the ability to use coaxial beam arrays with different intensity profiles, beam numbers and periods. It is also necessary to be able to tile the excitation beam array and switch between different beam arrays quickly to optimize the 3D imaging ability of TLS-SPIM using discontinuous light sheets in different applications.

The generation of coaxial beam arrays using binary optical devices has been studied extensively. We investigated two approaches to generate coaxial beam arrays using a binary SLM. We first investigated methods modified from the designing of Dammann gratings [24,25]. Although it is possible to generate coaxial beam arrays with certain beam numbers and periods following this approach, we found it difficult to control the intensity profile and the position of each beam in the beam array due to the limited resolution of the binary SLM. We therefore developed a more effective method based on pupil segmentation to generate coaxial beam arrays using the binary SLM in TLS-SPIM [26].

Briefly, the pupil of the excitation objective is divided into multiple groups of radial segments. Different phase maps corresponding to the generation of different excitation beams in a coaxial beam array are applied to different groups of radial segments. The desired coaxial beam array that consists of all beams is obtained by combining different phase segments together. Both the intensity profile and the position of each individual beam within the coaxial beam array can be controlled individually by adjusting the size of different radial pupil segments and the corresponding phase maps applied on them. As shown in Fig. 2, a four-beam coaxial beam array (Fig. 2(c)) was generated by dividing the excitation objective pupil into two groups of radial segments distributed alternatively and applying two different phase maps on them separately (Figs. 2(a) and 2(b)), by which each group of segments generate a pair of excitation beams at different tiling positions when the SLM is illuminated uniformly. The obtained coaxial beam array can be tiled along the light propagation direction by superimposing an additional spherical phase map to the phase map used to generate the coaxial beam array (Fig. 2(d)).

 

Fig. 2. The pupil segmentation method used to create coaxial beam arrays in TLS-SPIM. (a) One group of pupil segments and a binary phase map used to generate a pair of beams within a four-beam coaxial beam array. (b) The other group of pupil segments and the binary phase map used to generate the second pair of coaxial beams within a four-beam coaxial beam array. (c) The combined phase map used to generate the desired four-beam coaxial beam array. (d) The binary phase map used to tile the obtained four-beam coaxial beam array at a different position. Excitation numerical aperture (NA): NA_od=0.08, NA_id=0.03. (e, f) The intensity profile of the coaxial beam array in Fig. 2(c) at the indicated positions.

Download Full Size | PDF

Nevertheless, the obtained excitation coaxial beam array must be scanned to create a virtual discontinuous light sheet. The crosstalk between different beams of the coaxial beam array causes a problem when the beam array is scanned to create a discontinuous light sheet. The off-focus light of the beam array accumulates during the beam scanning process and results in strong off-focus excitation at the waist positions of the obtained virtual discontinuous light sheet (Fig. 3(a)). The problem is nearly identical to that of “non-diffracting” light sheets created by scanning “non-diffracting” beams (Fig. 3(b)). Undoubtedly, the off-focus excitation reduces the imaging ability of discontinuous light sheets.

 

Fig. 3. The intensity profile comparison of discontinuous light sheets created by scanning coaxial beam arrays and Bessel light sheets created by scanning Bessel beams. (a) YZ max intensity projection of a discontinuous light sheet created by scanning the four-beam coaxial beam array in Fig. 2(c), and its intensity profile at the indicated positions. (b) YZ max intensity projection of a Bessel light sheet that has a comparable thickness and effective length with the discontinuous light sheet in Fig. 3(a), and its intensity profile at the indicated position. Excitation numerical NA: NA_od=0.08, NA_id=0.07. (c) YZ max intensity projection of the equivalent light sheet when the scanning beam array is synchronized with a 7.5 µm wide virtual confocal slit, and its intensity profiles at the indicated position using confocal slits of different widths. (d) YZ max intensity projection of the equivalent light sheet when the scanning Bessel beam is synchronized with a 7.5 µm wide confocal slit, and its intensity profile at the indicated position using confocal slits of different widths.

Download Full Size | PDF

Indeed, the diffraction of light dominates the tradeoff between the thickness, light confinement ability and size of light sheets regardless of the intensity profile. The light confinement ability of a light sheet always decreases as its usable portion increases. Fortunately, the off-focus fluorescence background of virtual lights sheets created by scanning beams can be suppressed by using an sCMOS camera and operating it in the light sheet readout mod [27,28]. In this mode, the exposure and readout of each pixel row are synchronized with the scanning beam, results in a detection effect equivalent to slit confocal detection, so that the majority of the fluorescence background created by the off-focus excitation light is rejected by the camera.

Despite the similarity between discontinuous light sheets and “non-diffracting” light sheets, that both of them extend the effective length of a light sheet by sacrificing the light confinement ability, there is a significant difference, which is the distribution of the unconfined off-focus excitation light. As shown in the comparison of a discontinuous light sheet and a Bessel light sheet with similar effective light sheet lengths (Figs. 3(a) and 3(b)), the discontinuous light sheet distributes the off-focus excitation light much further from the detection focal plane than the Bessel light sheet. Obviously, the off-focus fluorescence light created by discontinuous light sheets can be rejected by the virtual confocal slit much more effectively compared to that of “non-diffracting” light sheets because the fluorescence background is spread further away from the detection focal plane (Figs. 3(c) and 3(d)), which is a key advantage of discontinuous light sheets over “non-diffracting” light sheets, such as Bessel light sheets. We further studied the ability of virtual confocal slits of different widths in rejecting the off-focus excitation of discontinuous light sheets via numerical simulations. As expected, thinner confocal slits with widths comparable to the scanning beam thickness reject off-focus background much more effectively (Figs. 3(c) and 3(d)). However, thinner confocal slits cost more photons for imaging, and it is more challenging to maintain the synchronization between the scanning coaxial beam array and the virtual confocal slit, i.e. the camera rolling shutter, in practice.

Altogether, the simulation results suggest that TLS-SPIM could work more efficiently using discontinuous light sheets synchronized with the virtual confocal slit of an sCMOS camera. A larger effective area can be imaged using discontinuous light sheets at each tile, which improves the imaging speed and reduces raw data size at the same time.

3. Experimental setup

We verified the 3D imaging ability of TLS-SPIM using discontinuous light sheets on a tiling light sheet microscope designed to image cleared biological tissues. The schematic diagram of the microscope is shown in Fig. 4. The excitation laser beam with the wavelength of 488 nm is expanded to a beam diameter of ∼8 mm (L1 = 30 mm, L2 = 250 mm) and sent to a binary SLM assembly for phase modulation. The binary SLM assembly consists of a polarizing beam splitter cube, a half-wave plate and a 1280×1024 binary SLM (Forth Dimension Displays, SXGA-3DM). The modulated light is focused on an optical slit to block the undesired diffraction orders generated by the SLM, and the SLM is conjugated to a galvanometer mirror (Cambridge Technologies, 6215HP-1HB) through relay lenses (L3 = 300 mm, L4 = 175 mm). The Galvo mirror directs the illumination light to one of the two symmetrical illumination paths by offsetting the initial angle and creates a virtual excitation light sheet for sample illumination by scanning the laser beam. The modulated laser beam is further conjugated to the rear pupils of two excitation objectives through two pairs of relay lenses (L5 = L7 = 150 mm, L8 = L9 = 250 mm) to illuminate the specimen from two opposite directions.

 

Fig. 4. The TLS microscope designed for cleared tissue imaging. (a) The schematic diagram of the TLS microscope. (b) The optomechanical design of the TLS microscope.

Download Full Size | PDF

Two Mitutoyo MY5X-802 objectives are used as the excitation objectives of the microscope. The emitted fluorescence is collected with an Olympus MVX10 Macro Zoom microscope equipped with a 1 × 0.25 NA long working distance detection objective (Olympus, MVPLAPO 1×) and imaged onto an sCMOS camera (Hamamatsu, Orca Flash 4.0 v3). The sample is mounted on a sample holder driven by a 3D translational stage (Physik Instrument, Q522.230) and immersed in the imaging buffer during 3D imaging.

4. Results and discussion

We evaluated the imaging performance of TLS-SPIM using discontinuous light sheets by imaging a fixed mouse brain expressing GFP in microglia cells and optically cleared using CUBIC2 [29]. We first examined the imaging ability of TLS-SPIM using discontinuous light sheets produced by scanning coaxial beam arrays with different beam thicknesses, beam numbers and beam array periods. A ∼6 µm wide virtual confocal slit was used in imaging with all discontinuous light sheets. Clearly, the 3D imaging ability of the microscope is improved by using a virtual confocal slit (Figs. 5(a)–5(d)), while both the spatial resolution and optical sectioning ability were maintained at the light sheet waist positions of all discontinuous light sheets (Figs. 5(e)–5(j)). In addition, the spatial resolution was further improved by using thinner discontinuous light sheets while the effective imaging area increases at the same time (Figs. 5(c)–5(d) and Figs. 5(i)–5(j)).

 

Fig. 5. The imaging performance comparison of TLS-SPIM using continuous and discontinuous light sheets. (a, b) XY and YZ max intensity projection of a sample volume imaged using a continuous light sheet created by scanning the excitation beam. Excitation NA: NA_od=0.045, NA_id=0.015. (c, d) XY and YZ max intensity projection of the same sample volume imaged using the same light sheet with the scanning beam synchronized with a 6 µm wide virtual confocal slit. (e-j) XY and YZ max intensity projections of the same sample volume imaged using discontinuous light sheets created by scanning a two-beam, three-beam and four-beam coaxial beam array synchronized with a 6 µm virtual confocal slit. Excitation NA: NA_od=0.08, NA_id=0.03 for all discontinuous light sheets. The inserts show zoomed in views of the indicated cells in each panel. Scale bars, 200 µm in Figs. 5(a) and 5(b), 10 µm in all inserts.

Download Full Size | PDF

Next, we compared the imaging ability of discontinuous light sheets with Bessel light sheets of similar effective lengths (Fig. 6). As predicted, the spatial resolution of images obtained using Bessel light sheets decreases significantly as the light sheet length increases despite the use of a 6 µm virtual confocal slit, because the off-focus excitation of Bessel light sheets stays too close to the detection focal plane to be blocked by the confocal slit effectively (Figs. 6(a)–6(f)). On the contrary, the same spatial resolution and optical sectioning ability were maintained within a larger area at light sheet waist positions by using a discontinuous light sheet (Figs. 6(g)–6(h)).

 

Fig. 6. The imaging performance comparison of TLS-SPIM using discontinuous light sheets and Bessel light sheets. (a, b) XY and YZ max intensity projections of a sample volume imaged using a continuous light sheet created by scanning the excitation beam synchronized with a 6 µm virtual confocal slit. Excitation NA: NA_od=0.08, NA_id=0.03. (c, d) XY and YZ max intensity projections of the same sample volume imaged using a twice long Bessel light sheet created by scanning the excitation beam synchronized a 6 µm virtual confocal slit. Excitation NA: NA_od=0.08, NA_id=0.06. (e, f) XY and YZ max intensity projections of the same sample volume imaged using a three times long Bessel light sheet created by scanning the excitation beam symphonized with a 6 µm virtual confocal slit. Excitation NA: NA_od=0.08, NA_id=0.067. (g, h) XY and YZ max intensity projections of the same sample volume imaged using a discontinuous light sheet created by scanning a three-beam coaxial beam array synchronized with a 6 µm virtual confocal slit. Excitation NA: NA_od=0.08, NA_id=0.03. (i-n) XY max intensity projections of the selected image volumes indicated in Figs. 6(a), 6(c), 6(e), 6(g) . (o-t) YZ max intensity projections of the selected image volumes indicated in Figs. 6(b), 6(d), 6(f), 6(h). Scale bars, 200 µm in Figs. 6(a) and 6(b), 20 µm in Figs. 6(i) and 6(o).

Download Full Size | PDF

Finally, we imaged a ∼4 mm³ sample volume using tiling discontinuous light sheets operated in two modes described above (Fig. 7). We first imaged the sample volume in ∼1 minute at a spatial resolution of ∼2 by 2 by 4 µm³ by tiling a discontinuous light sheet containing three waists at three positions (Figs. 7(a)–7(f)). The imaging speed was improved, and raw data size was decreased by 3 times compared to that of using a regular continuous light sheet, which would require 9 tiles. Next, we imaged the same sample volume using a discontinuous light sheet containing four waists and a compensating discontinuous light sheet containing three waists that are slightly longer and thicker than that of the four-waist discontinuous light sheet (Figs. 7(g)–7(j)). The same image volume was imaged at roughly the same spatial resolution in ∼40 seconds, as only two tiles were required to image the entire FOV, which would require 7 tiles to image using a continuous tiling light sheet. It represents a 3.5 times improvement of the imaging efficiency. The final result was reconstructed using either group of images by selecting and stitching the areas corresponding to the light sheet waist positions in all tiles following the method described in the previous publication (Figs. 7(k) and 7(l)) [18]. As shown, TLS-SPIM worked much more efficiently without sacrificing the 3D imaging ability by using discontinuous light sheets created by scanning coaxial beam arrays that are synchronized with the rolling shutter of the sCMOS detection camera.

 

Fig. 7. Different operation modes using discontinuous light sheets in TLS-SPIM. (a-f) XY and YZ max intensity projections of a sample volume imaged using a three-waists discontinuous light sheet tiled at three positions. Excitation NA: NA_od=0.08, NA_id=0.03. (g-j) XY and YZ max intensity projections of the same sample volume imaged using a four-waists discontinuous light sheet and a three-waists discontinuous light sheet compensating each other. Excitation NA: NA_od=0.08, NA_id=0.03 for the four-waists discontinuous light sheet and NA_od=0.06, NA_id=0.02 for the three-waists discontinuous light sheet. (k, l) XY and YZ max intensity projections of the reconstructed result, which could be obtained using the results shown in Figs. 7(a)–7(f) or Figs. 7(g)–7(j). Scale bars, 200 µm.

Download Full Size | PDF

5. Conclusions

In summary, we developed a novel method to improve the imaging efficiency of TLS-SPIM without affecting its 3D imaging ability. Discontinuous light sheets created by scanning coaxial beam arrays synchronized with the detection camera rolling shutter are used in TLS-SPIM for 3D imaging. We investigated the method and evaluated its imaging ability via numerical simulations and experiments. We show that discontinuous light sheets improve the imaging speed and reduce the raw data size by imaging larger effective areas at each tile without affecting the spatial resolution and optical sectioning ability, but at a cost of more critical requirements on the microscope alignment due to the use of the detection camera rolling shutter.

The exact improvement could be achieved is determined by the implemented discontinuous light sheets and the operation mode, which could be adjusted based on the requirements of the specific application. Such an improvement brings significant benefits when TLS-SPIM is used to image large specimens at high spatial resolutions. It could potentially reduce the total imaging time by hours even days and reduce the raw data size by terabytes or more.

Although we demonstrated the use of discontinuous light sheets on a TLS microscope designed to image optically cleared biological tissues at micron level spatial resolutions, the method is applicable to all TLS microscopes with similar optical configurations regardless of the optical components being used, the spatial resolution to be achieved, and the applications to be focused on. In addition, the described method used to produce coaxial beam arrays and discontinuous light sheets based on pupil segmentation provides more freedom to deliver and control the excitation light in TLS-SPIM. It enables more flexible optimizations of the 3D imaging ability of TLS-SPIM, that could introduce additional benefits in its applications.