1. Introduction

Biological fluorescence microscopy enables the selective observation of targets such as organelles, membranes, and proteins via the utilization of fluorescent dye in live specimens. Because many biological specimens are three-dimensional, confocal laser scanning microscopy or two-photon microscopy have been used for decades to acquire fluorescence optical sections, providing fluorescence tomography [1,2]. In these methods, the image acquisition time is determined by the scanning speed, width of the observed field, and pixel resolution to scan the focused laser light two or three dimensionally. These methods require a significant amount of time to acquire high-resolution and wide-field fluorescence images. Hence, light-sheet fluorescence microscopy (LSFM), also known as selective plane illumination microscopy (SPIM), was developed, wherein the irradiation laser forms a planar illumination resembling a sheet [3]. A thin laser beam (hereafter, laser sheet), which becomes elliptical as it passes through a cylindrical lens, selectively excites the fluorescence dye within the laser sheet, which enables optical sectioning to be obtained without scanning. Therefore, compared with confocal laser scanning microscopy and two-photon microscopy, the time required for image acquisition is shorter, and the generation of background fluorescence and photodamage to the observed sample is suppressed because only the observation cross-section is illuminated. Instead of shaping using a cylindrical lens, planar illumination can be created by scanning a focused laser beam, which is a method known as digital scanned laser light-sheet microscopy (DSLM) [4].

However, when observing the inside of biological specimens such as tissues, microstructures inside tissues that change the refractive index partially cause scattering, refraction, and diffraction, which results in stripe-shaped shadowing artifacts. Tissue clearing techniques wherein the refractive index in the specimen is matched with the solvent can be used to prevent refraction of illumination light at the sample surface or inside the sample; however, it is difficult to achieve a completely uniform refractive index; refraction and diffraction within a sample are inevitable. Further, there is a trade-off in that the high transparency obtained by tissue clearing allows light to pass through easily although the samples are killed [5–8]. Nevertheless, the observation of live samples is important to biologists, and therefore, multidirectional selective plane illumination microscopy (mSPIM) is applied. The laser sheet in mSPIM is formed by a cylindrical lens that illuminates the inside from multiple directions to suppress shadowing artifacts [9]. Furthermore, DSLM can be combined with mSPIM, i.e., a method known as multidirectional DSLM (mDSLM) [10].

Refractive index inhomogeneity described above causes random scattering. The scattering increases the background fluorescence by the undesirable irradiation of the off-sheet area. Consequently, the image contrast degrades in DSLM observations. Hence, the rolling shutter of a complementary metal oxide semiconductor (CMOS) camera, which is an image detector, can be used as a virtual slit to remove the signals from the off-sheet area. We can obtain only the line area irradiated with the laser light by synchronizing the scanning of both the line-formed laser and exposed area with the rolling shutter and by removing the fluorescence from off-sheet areas, which provides a pseudo confocal effect to improve the image contrast [11]. Both the mDSLM and pseudo confocal system enabled by a rolling shutter are effective in improving the image contrast. However, they are almost technically incompatible. Only one example that shows compatibility between them was reported [10]; however, there are still problems in that the illumination intensity decreases at both extremities of the line.

This study demonstrates that mDSLM and a pseudo confocal method can be combined and the combination is effective for observing three-dimensional bio-specimens. We developed a method to generate a pseudo confocal effect via image acquisition through global exposure by DSLM with a modified laser scanning sequence; subsequently, we investigated the effect of the proposed pseudo confocal method on the image contrast in artificial phantoms and living biological samples. The proposed method can be combined with mDSLM because the synchronization of the scanned laser line with the exposing shutter in a camera is not required. We combined mDSLM and the pseudo confocal method and investigated the effect of the pseudo confocal method on mDSLM imaging of mouse early embryos. We believe that the proposed combination of mDSLM and the pseudo confocal method can be extended to various areas to obtain high-contrast three-dimensional images of live tissues effectively, including those of organoids.

2. Experimental procedures

2.1 Materials

Optical components such as lenses, mirrors, posts, and holders were purchased from Thorlabs, Inc. (Newton, New Jersey, USA) or Edmund optics, Inc. (Barrington, New Jersey, USA). Optical filters were purchased from Semrock, Inc. (Rochester, New York, USA) via a Japanese import agency, Opto-line, Inc. (Tokyo, Japan). The mouse embryonic stem cell line (AES0135: E14tg2a) used was obtained from the RIKEN Cell Bank (Tsukuba, Japan).

2.2 Optical setup

A custom-developed LSFM system was used in all experiments (Fig. 1(a)). The employed light sources were free-space-type continuous wave lasers (Juno-Compact 532 nm for fluorescent microspheres, Juno-Compact 561 nm for the mouse embryoid body (EB), Juno-Compact 488 nm for mouse early embryos; KYOCERA SOC, Co., Ltd. Kanagawa, Japan). The laser beam was focused on a galvanometric mirror scanner GM1 (VM500Plus, Cambridge Technology, Inc., Bedford, Massachusetts, USA) via a plano-convex lens L1 with a focal length (f) of 100 mm; the beam was then collimated via a plano-convex lens L2 (f = 100 mm) such that it was incident on the galvanometric mirror scanner GM2 (VM500Plus, Cambridge Technology). The beam deflected by GM2 was relayed by a 4-f relay system composed of plano-convex lenses L3 (f = 200 mm) and L4 (f = 200 mm), and it focused on a sample by an objective lens (OL) (M Plan-Apo 10x, Mitsutoyo, Kawasaki, Japan). The diameter of the focused beam at the beam waist was 9.4 µm, and the corresponding effective field of view along x axis (twice the Rayleigh length considering the approximate refractive index of the sample medium of 1.33) was 378 µm. The positions of galvanometric mirror scanners GM1 and GM2 were the conjugated positions of the focal plane and back-focal plane of the OL, respectively. The focused illumination laser beam passed through a borosilicate glass cuvette (10 mm (L) × 10 mm (W) × 12.5 mm (H)). Fluorescence light emitted by the sample was collected using an OL (Olympus 20x dry NA0.45, Olympus, Tokyo, Japan); subsequently, it was passed through a bandpass filter (BPF) (FF01-607/36-25 or FF01-525/45-25, Semrock) and imaged onto a scientific CMOS (sCMOS) camera (Zyla 5.5, Andor Technology, Belfast, Northern Ireland for fluorescence microspheres and mouse EB, and ORCA-Flash 4.0V3, Hamamatsu, Shizuoka, Japan for mouse early embryos) using a tube lens (TL, effective focal length of 180 mm). The readout mode of Zyla 5.5 was set to the global shutter while that of ORCA-Flash 4.0V3 was set to the rolling shutter with a global reset mode that effectively acts as the global shutter mode. Both the sample stage and collection objective lens are mounted on motorized translation stages driven by DC servo actuators (Z825B with a controller KDC101, Thorlabs) such that their positions can be scanned along the z axis.

 

Fig. 1. Optical setup and image acquisition procedure. (a) Two schematic views of present optical setup. OL, objective lens; BPF, band-pass filter; CAM, CMOS camera; L, plano-convex lens; TL, tube lens; GM, Galvanometric mirror. (b) Schematic illustration of image acquisition procedure. (c) Typical example of raw image set obtained using proposed method. Sample was 0.5 µm yellow fluorescent beads in 0.5% agarose hydrogel.

Download Full Size | PDF

2.3 Device control

A customized software was developed using Microsoft Visual C# to control the timing of laser scanning, imaging, and motion of the sample stage. A digital-to-analog converter (NI-USB6363, National Instruments Corp., Austin, Texas, USA) was used to control the laser power and the angle of the galvanometric scanners via variable analog voltages. The images were obtained using an external level trigger input to the sCMOS cameras. The exposure timing of the cameras were controlled by external TTL signals. The TTL signal was input to the camera during one stroke of laser scanning. Camera exposure is triggered at the same moment that laser scanning starts, and the exposure time was same as the laser scanning period.

2.4 Preparation of artificial phantoms

Yellow-Orange (YO) fluorescent beads (0.50-µm Polyscience Fluoresbrite YO carboxylate microspheres #18720) embedded in an agarose hydrogel were prepared for the experiments to investigate the results obtained from the proposed method. Agarose powder (01163-05, Nacalai, Kyoto, Japan) and YO beads were dissolved in ultrapure water (Milli-Q, Merck Millipore, Burlington, Massachusetts, USA) (0.5% w/v) at 100 °C in a microwave. The solution was stirred and cooled to room temperature. Subsequently, the solution was poured into a glass cuvette and incubated for several tens of minutes at room temperature for gelation.

2.5 Establishment of mouse embryonic cell line and cell culture

We established an embryonic stem (ES) cell line comprising a fluorescently labeled membrane and nucleus. The green fluorescent protein in the pAcGFP1-Mem vector (Clontech, 632453, Takara Bio Inc., Japan) was replaced by the red fluorescent protein mCherry (Mem-mCherry). The gene fragment encoding Mem-mCherry was introduced into a mouse ES cell line (E14tg2a) with the PiggyBac Transposon Vector (System Biosciences, LLC, Palo Alto, CA, USA) based on the manufacturer’s instructions. The ES cells were cultured in high glucose Dulbecco's modified Eagle medium (DMEM) (D6046, Sigma–Aldrich Co. LLC, St. Louis, Missouri, USA) containing 10% FBS (16141-075, Gibco, USA), 1% penicillin–streptomycin (PS; Sigma–Aldrich, P4333), 1% GlutaMAX (35050–001, Gibco), 1% non-essential amino acids (NEAA; 11140–050, Gibco), 1% nucleosides (ES-008-D, Merck Millipore), 1% sodium pyruvate (S8636, Sigma–Aldrich), 0.1 mM 2-mercaptoethanol (2-ME; Sigma-Aldrich), and 0.1% leukemia inhibitory factor (LIF; NU0013–1, Nacalai, Japan) at 37 °C, 5% CO₂, and 95% humidity.

2.6 Preparation of embryoid bodies

An EB is a three-dimensional spherical aggregate of ES cells. For daily culture, ES cells were dissociated into single cells with 0.05% Trypsin–ethylenediaminetetraacetic acid (Gibco, 25300054) and seeded at 1 × 10⁵ cells on 10 cm plastic dishes (BD Biosciences, 353003) coated with 0.1% gelatin (EmbryoMaxR 0.1% gelatin ES-006-B, Merck Millipore). The ES cells were passaged every 2 days during long-term culturing. For a suspension culture, 1 × 10⁵ cells were seeded on 35 mm glass bottom dishes (D11130H, Matsunami-Glass, Osaka, Japan) as a low attachment culture condition with a DMEM solution comprising 10% KSR (10820–028, Gibco) instead of FBS, 1% PS, 1% GlutaMAX, 1% sodium pyruvate, 1% NEAA, 0.1 mM 2-ME, and 0.1% LIF, supplemented with an MAPK inhibitor, PD0325901(Stemgent, USA, Stemolecule 04–0006) and a GSK3 inhibitor, CHIR 99021 (Stemgent, Stemolecule 04–0004) at 1 and 3 µM, respectively. The ES cells formed EBs of various sizes in 2 days. We extracted the EBs of approximately 300 µm in diameter and embedded them in a glass cuvette with 0.5% (w/v) agarose hydrogel.

2.7 Preparation of mouse early embryo

A mouse embryo expressing histone H2B fused EGFP in cell nuclei (R26-H2B-EGFP mice, Accession No. CDB0238K: http://www2.clst.riken.jp/arg/reporter_mice.html) [12] was used. A living mouse embryo at E5.5 was harvested and embedded in a cube (NK system, CIDH-29) with 0.5% agarose hydrogel (01163-0, Nacalai) in phosphate buffered saline (PBS (-)) (#T900, Takara). The cube was set to the bottom of a customized glass cuvette filled with PBS(-). The animals were housed in environmentally controlled rooms. All experimental procedures using animals were reviewed and approved by the Institutional Animal Care and Use Committee of RIKEN Kobe Branch (A2001-03).

3. Results and discussion

3.1 Image acquisition

We constructed an optical setup to perform mDSLM (Fig. 1(a)). The procedure for pseudo confocal image acquisition with the setup is as follows (Fig. 1(b)):

In this study, we chose three tilt angles (−3°, 0°, 3°) for mDSLM. Increasing the number of angles, causing elongation of the overall exposure time, would reduce shadow artifacts further in principle. An alternative method to reconstruct an image using masks has been reported [13]. Although this technique is useful for background rejection, the masking process requires an electronic slit (rolling shutter) or the estimation of laser line positions. In this study, we applied MIP because of its simplicity and because it can be applied easily in the case that the refraction at the sample surface (or inside the sample) is not negligible.

A single plane image can also be acquired by continuously scanning m × n laser lines at an interval of Δy / n (hereafter, conventional (m)DSLM). In the conventional (m)DSLM, all laser lines are scanned in one exposure of the camera, including the repetition due to multi-directional irradiation. In the present study, we simulated conventional (m)DSLM images by summation of frames acquired by the pseudo confocal method instead of actual scanning with the method described above to compare conventional and pseudo confocal images of the same area.

The number of lines per frame (m) was set to 16 lines per 1024 pixels (central pixels cropped from full pixels, corresponding to 332.8 µm) for fluorescence beads and EBs, and it was set to 16 lines per 2048 pixels (corresponding to 665.6 µm) for mouse early embryos. These values were selected as a value that provides sufficient interval to avoid overlap between adjacent lines (See Fig. S1 for comparison between illuminations with different number of lines). Number of frames to scan the entire area (n) was set to 16. The speed of displacement of a laser line between set positions, which depends on the performance of GM2, also determines m or the exposure time per frame since excitation during the laser moving causes background fluorescence.

For volumetric imaging with multiple optical sections, the sample was scanned along the z-axis to change the relative position of the light-sheet. The position of the collection objective lens was scanned simultaneously to maintain its focal plane at the position of the light-sheet.

Although the multiline laser was produced by the discrete scanning, we tested a microlens array as an alternative candidate. The small structure between the two microlenses caused undesired scattering, which degraded the line form. Moreover, the uniformity of the intensity and the shape of each line depended on the processing accuracy of the microlenses. Therefore, we opted for the scanning procedure.

3.2 Comparison of results in high- and low-density fluorophores in sample

First, we compared the image quality between the proposed pseudo confocal method and the conventional DSLM in a phantom sample fabricated using 0.5 µm yellow fluorescent beads and 0.5% agarose hydrogel. Exposure time per frame was 20 ms and overall exposure time to obtain single plane image was 640 ms (16 frames). The bead densities were 3.64 × 10⁷/mL and 7.28 × 10⁸/mL. An improvement in the image contrast was confirmed visually (Figs. 2(a) and (b)). The full width at half maximum (FWHM) was estimated on the x-, y-, and z-axes by fitting the volumetric image of a single bead with the Gaussian distribution for the x- and y-axes or the Lorentz distribution for the z-axis to confirm the confocal effect by the proposed method. The FWHMs on all axes were independent of the bead density (Figs. 2(c)–(e), green).

 

Fig. 2. Quality evaluations of image obtained using conventional DSLM and proposed pseudo confocal method in high- and low-density fluorophores in the sample. (a, b) Typical images obtained using conventional DSLM (a) and proposed method (b), in high (left) and low density (right). (c–e) Comparison of FWHMs of fluorescent bead images between conventional SPIM (green) and proposed method (cyan), on x- (c), y- (d), and z-axes (e). (f) Comparison of image contrast between conventional DSLM (green) and proposed method (cyan).

Download Full Size | PDF

The resolution did not improve on the x–y plane; in fact, it degraded slightly (Figs. 2(c) and (d), cyan). Confocal light sheet microscopy has no significance on lateral resolution [14]. This same discussion applies because our method is a confocal detection method. The slight degradation was attributed to image reconstruction. The MIP adopts a maximum intensity as a pixel value not only onto the peak of a fluorescence spot, but also onto its tail, which affects the broadening of the peak width. Frequency domain analysis of the images in Figs. 2(a,b) confirmed that the lateral resolution was almost preserved between conventional and pseudo confocal methods (see Fig. S2 for the details). Meanwhile, a 1.2-fold improvement was observed in the z-axis at both densities (Fig. 2(e), cyan). The image contrast depended on the bead density caused by the crosstalk fluorescence from the off-sheet area (Fig. 2(f), green). Furthermore, the proposed method improved the image contrast, particularly in the denser condition (Fig. 2(f), cyan). Thus, the proposed method resulted in the improvement of the axial resolution and image contrast.

3.3 Comparison of results in high- and low-density micro scatterers in sample

The blurring and/or degradation of a fluorescent image in a DSLM observation was assumed to be caused by microstructures inside a sample. We evaluated the FWHMs of the x-, y-, and z-axes and image contrast in a phantom of 0.5% agarose hydrogel with 0.5% intralipid to investigate the pseudo confocal effect against the existence of microstructures (Figs. 3(a)–(c)). The exposure time per frame was 20 ms, and the overall exposure time to obtain a single plane image was 640 ms (16 frames). The PSF is the summation of the original and scattered PSFs in the presence of microstructures, and the original PSF is dominant at an observation depth of less than 200 µm [15,16]. Therefore, although the image contrast degraded in the presence of the intralipid, the FWHMs were not affected as much (Fig. 3(a)–(c), green). In the proposed pseudo confocal method, the presence of the intralipid did not affect the improvement in the image contrast (Figs. 3(a)–(c), cyan). In addition, the light sheet was not deformed by the presence of the intralipid, whereas the background fluorescence increased (Fig. S3), which decreased the degradation of the image contrast. The present method can recover the image contrast degraded by the intralipid to the same level as the no-intralipid condition (Fig. 3(d), cyan). This result suggests that the confocal effect reduced background fluorescence. The background fluorescence showed a uniform distribution instead of a Gaussian distribution (Fig. S3(c)). Therefore, the image contrast may be improved only by the confocal effect.

 

Fig. 3. Quality evaluations of image obtained using conventional DSLM and proposed pseudo confocal method in high- and low-density scatterers in the sample. (a–c) Comparison of FWHMs of fluorescent bead images between conventional DSLM (green) and proposed method (cyan), on x- (a), y- (b), and z-axes (c). (d) Comparison of image contrast between conventional DSLM (green) and proposed method (cyan).

Download Full Size | PDF

3.4 Observation of mouse EB

We successfully confirmed the improvement in image contrast under the fluorescence dense and microstructure presence conditions. The next experiment involved actual live tissue imaging. As a model sample, we prepared a mouse EB, which is a floating spherical cell aggregate. The membrane was labeled with a red fluorescent protein variant by fusing a membrane-targeting signal. We set an EB with a diameter of 300 µm into a 0.5% agarose hydrogel and acquired images with conventional DSLM and the proposed method (pseudo confocal with DSLM) at depths of 50, 100, and 200 µm apart from the surface of the EB. The exposure time per frame was 20 ms, and the overall exposure time to obtain single the plane image was 640 ms (16 frames). As the depth of the observation plane increased, the internal structure became blurred, and the contrast decreased in the conventional DSLM (Figs. 4(a)(c)(e), left). The image contrast, sharpness, and brightness at the 100 µm depth were similar to that at the 50 µm depth in the present methods; the background fluorescence degraded the image contrast in the conventional DSLM (Figs. 4(a)(c)(e), right). The line profiles highlighted the difference between the conventional DSLM and the proposed method (Figs. 4(b), (d), and (f)). As shown in the profiles, the spatial resolution was only slightly degraded by the depth of the observation plane; however, the background fluorescence increased. In general, the proposed method maintained the image quality at a deeper position, although the background fluorescence at 200 µm depth could not be completely removed. Because the illumination intensity decreased due to absorption and scattering inside the sample depending on the penetrating distance of the laser light (Fig. 4, white arrowhead), the brightness reduced along the laser illumination axis (Figs. 4(a)(c)(e), right, see Fig. S4 for profiles in the far region). This problem can be alleviated by illuminating the sample from both sides [9].

 

Fig. 4. Evaluation of image qualities of embryoid body (EB) obtained using conventional DSLM and proposed method (pseudo confocal). Images of EB at three depths of observation planes obtained using conventional (left) and proposed method (right) shown in (a), (c), and (e). Intensity distributions along highlighted dashed lines in (a), (c), and (e) shown in (b), (d), and (f), respectively.

Download Full Size | PDF

3.5 Observation of mouse early embryos

Finally, we implemented a two-side illumination and demonstrated an imaging of a mouse early embryo (E5.5) with conventional (m)DSLM and the pseudo confocal (m)DSLM (Fig. 5). The focus position of the illumination beams from the two sides (along x axis) were matched by adjusting the position of illumination OLs. The pseudo confocal images were acquired separately for each side of illumination and finally merged via the MIP. A comparison of images obtained using the conventional DSLM (Fig. 5(a)), conventional mDSLM (Fig. 5(b)), pseudo confocal method with DSLM (Fig. 5(c)), and pseudo confocal method with mDSLM (Fig. 5(d)) showed an improvement in image quality using the proposed method. The exposure time per frame was 10 ms, and the overall exposure time to obtain a single plane image was 960 ms for pseudo confocal DSLM (32 frames) and 2880 ms for pseudo confocal mDSLM (96 frames). Background fluorescence causing blurring and reduction of image contrast was significantly reduced in pseudo confocal (m)DSLM images. A stripe-shaped shadowing effect is obvious in the pseudo confocal DSLM image, whereas it is not noticeable in the conventional DSLM image. This is probably because the uneven illumination is blurred by background fluorescence in the conventional DSLM, whereas the original uneven illumination is directly observed in the pseudo confocal DSLM as the pseudo confocal method is not affected by it. As expected, multidirectional laser scanning eliminated the shadowing effect. Improved image contrasts were confirmed at all observation depths (50, 75, 100, and 125 µm from the surface) (Fig. S5). By plotting an intensity profile on a line in the images, it was observed that the effective suppression of the background signal was enhanced by the pseudo confocal method (Figs. 5(e–g)). Although image acquisition using the proposed method required a longer time compared with that using the actual conventional mDSLM (acquired in a single 100 ms exposure), only approximately 3 s were required for a single cross-section, i.e., it was sufficient for the imaging of a living embryo. The results confirmed that the combination of the pseudo confocal method and mDSLM is effective and applicable to three-dimensional live tissues.

 

Fig. 5. Comparison of images of living mouse early embryo (E5.5) via conventional DSLM (a), conventional mDSLM (b), pseudo confocal method with DSLM (c), pseudo confocal method with mDSLM (d). Images obtained with illuminations from both sides and mDSLM angle (θ₀) of −3°, 0°, and 3°. Images in (a and b) are constructed via the simple summation of all obtained images. Images in (c and d) are constructed via the MIP of images captured at θ = 0°. Pseudo confocal image with mDSLM (c) obtained by summation of all MIP images captured with all θ₀ values. Regions indicated by dashed squares in (b and d) are enlarged in (e) and (f). Intensity profiles along red solid lines in (e and f) are compared in (g).

Download Full Size | PDF

4. Conclusion

We demonstrated a high-contrast image acquisition method that enables the combination of the pseudo confocal effect and mDSLM using a general DSLM optical system. The proposed method for the pseudo confocal effect improved the resolution in the z-axis and the image contrast in the fluorescence dense and microstructure presence conditions. The observations of EB and embryo as biological sample models with dense phosphors showed that the image quality can be maintained without blurring even in the deeper observation area. In conclusion, the pseudo confocal method with a global shutter improved the optical sectioning ability in mDSLM without requiring any additional equipment. This study highlights the further potential of mDSLM for observing thick biological specimens by the pseudo confocal method.