Fluorescence microscopy [1] is typically evaluated based on four key parameters: temporal resolution, spatial resolution, signal-to-noise ratio (SNR), and imaging speed [2]. Light-sheet fluorescence microscopy (LSFM) selectively illuminates a plane in the biological sample [3], achieving the balance of the four aspects of fluorescence microscopy [4–7]. As a result, LSFM has become an ideal tool for three-dimensional (3D) imaging of live cells [8–13]. In LSFM, the thickness of the light sheet directly affects the slicing capability [14]. The primary challenge in LSFM is generating an ultra-thin light sheet for optical sectioning in order to improve both SNR and axial resolution [15–18].

The thickness of the light sheet is usually confined by the numerical aperture (NA) of the excitation objective. Traditionally, the perpendicular layout between LSFM excitation and detection puts many limits on the placement of the objectives and sample holders. For example, the dual-view inverted selective plane illumination microscope (diSPIM) [19] was designed to be user-friendly in terms of sample placement, but tilting the optical paths relative to the horizontal sample holder could introduce significant off-axis aberrations. This was especially true for higher NA beams, which limited the use of high NA objectives for generating thinner light sheets [20]. One solution to this problem was to use a single-objective lens for both excitation and detection, with the sample holder oriented orthogonally to the objective [21]. However, this approach has its own limitations: since the illumination and collection beams share the NA of the single objective, the NA was underutilized for exciting the light sheet. The oblique plane microscopy fully utilizes the NA to excite light sheet at an angle with respect to the optical axis; the fluorescence information is then detected and captured through remote-focusing with the same angle of inclination. This approach requires the use of three objectives and involves a complex optical setup. It also necessitates precise matching of the tilt angles and careful consideration of refractive index matching between the sample region and the detection objective lens focal region [22–24]. Another approach involved converting axial information into lateral information using a micromirror array [25]. Extracting axial information required a long working distance objective due to the large size of the micromirror array, which limited axial resolution to 2.4 μm. Alternative approaches included combining LSFM with a micro-reflector or using an atomic force microscopy (AFM) cantilever to reflect the light sheet by 90° into the sample [26–28]. It should be noticed that introducing reflective elements in the excitation beam path leads to the use of a long working distance objective lens. Objectives with long working distances tend to have lower NA, making it challenging to generate thin light-sheet excitation. Furthermore, with the reflective element in the excitation beam path, there is significant spherical aberration due to refractive index mismatch among the objectives, the reflector, and the sample, particularly in the cases of tight focusing with high NA objectives. The presence of spherical aberration further degrades the thickness of the light sheet. It therefore led to poor axial resolution. Additionally, these methods can be costly due to the need for a new AFM cantilever each time, and system tuning can be quite complicated.

In this paper, we present a new method for reflective type light-sheet microscopy (RTLIS) that is capable of achieving ultra-thin light-sheet excitation with a high signal-to-noise ratio (SNR) and axial resolution. RTLIS accomplishes this through several advantages. Different from previous efforts, by using a reflective mini-prism in the detection beam path, isotropic resolution can be achieved when the excitation and detection objectives are identical. An ultra-high NA objective lens (NA 1.4) can further be used for thinner light-sheet excitation, resulting in a thickness of 190 nm when using a 561 nm laser. Secondly, the use of a mini-prism enables the proposed scheme to be compatible with traditional transmission type fluorescence microscopes. It therefore spares the sophisticated sample mounting in conventional LSFM. We demonstrate these capabilities through fast 3D imaging of Drosophila Rh6 cells with dimensions of 7.23μm×26.8μm×31.4μm. The lateral resolution of the system was 309 nm, and the axial resolution was also 312 nm without deconvolution.

As shown in Fig. 1(a), a mini-prism mechanically mounted together with the sample chamber acts as the beam reflector, which essentially transforms a traditional transmission microscope into reflective LSFM. The mechanical design for the mini-prism-based sample compartment is discussed in Appendix A. These design considerations take into account the cover glass thickness adapted for objectives, matching with standard sample holders to minimize optical aberrations and ensure the stability of the glass plate. An ultra-thin light sheet then is generated by a high NA objective (NA 1.0), which selectively illuminates a single plane within the sample. The sample chamber is specially designed as a square slot (50mm×5mm×3mm) made of glass, with both the bottom and sides having a thickness of 0.17 mm. The fluorescence signals excited by the light sheet are reflected by a prism adjoining the glass sample chamber and then detected by the detection objective (NA 1.0). Using the identical objectives for both light sheet excitation and detection allows for isotropic resolution in all three dimensions. Additionally, the use of non-specifically designed objectives lowers experimental costs. The design of the glass chamber allows for a wide variety of samples to be used in experiments.

In our system, a 561 nm Gaussian beam (Coherent, OBIS CORE LS 561) is expanded and collimated by a lens (Olympus/4×, NA 0.13) before passing through an axicon lens (Thorlabs, AX252-A) to be transformed into a Bessel beam. The beam is then scanned along the X-axis to generate light-sheet illumination in the nominal object plane by galvo X (Thorlabs, GVS012, 2D). The excitation objective (Olympus, 60×, water immersion, NA 1.0) focuses the light sheet onto the sample, projecting a square area of illumination that emits fluorescence signals. These signals are reflected by a mini prism and collected by the detection objective (Olympus, 60×, water immersion, NA 1.0). After passing through a band-pass filter (Chroma, D580/20 mm), 3D imaging of the sample is acquired on a sensitive sCMOS camera (Hamamatsu, Flash4.0 V3) by synchronously moving the galvo Z along the Z-direction and a motorized translation stage (Newport, LTA-HL) along the Y-direction; the detection path [indicated by the orange beam in Fig. 1(b)] is placed on the motorized translation stage. The moving speed of the translation stage is 1 mm/s; when volumetric imaging was performed, the excitation light sheet was scanned at the step of 300 nm. Accordingly, the detection system was vertically translated at the same step, taking the time of 30 μs for each step. The camera exposure time was at the ms scale. Therefore, the primary limiting factor for the volumetric imaging speed is the camera exposure time. All motions, camera exposure, galvanometer deflection, and optical switching of the laser are synchronously timed and controlled by a DAQ (data acquisition system) (NI, 6259) (see Appendix B). The system control and data processing use a computer (Dell, T3630).

In the coordinate system as shown in Fig. 1(a), the Y-direction mentioned herein is the optical axis direction for excitation optics. In conventional fluorescence microscopy, the fast dynamics in the excitation optical axis direction is usually difficult to capture. In the proposed RTLIS, to obtain an axial plane image such as XY, only one frame imaging is needed, which is very important for samples with a dynamic axial distribution.

In the experiments, we first measured the optical performance of the setup. The point spread function (PSF) of this system was characterized by imaging 80 nm fluorescence beads. The deconvolution imaging results are shown in Figs. 2(a) and 2(b). The data are fitted by Gaussian function. The full-width at half-maximum (FWHM) values of the PSF along X are 290 nm and along Z are 310 nm. Figures 2(c) and 2(d) demonstrate the 3D isotropic resolution around 300 nm, while the raw imaging results are shown in Appendix D, in which the FWHM values of the PSF along X are 309 nm and along Z are 312 nm.

Our proposed system allows for the direct use of a high NA objective lens for light-sheet excitation. However, whether a higher NA excitation objective necessarily results in a thinner light sheet remains an important topic of investigation. Light sheets can be divided into two categories based on the excitation beam: single-lobe light sheets and multi-lobe light sheets. In single-lobe light sheets, such as those generated by a Gaussian beam, a higher NA results in a smaller diameter of the focused spot but also an overly short length of the light sheet (see Appendix C). In contrast, multi-lobe light sheets, such as those generated by a Bessel beam, maintain an appropriate length with higher NA. A higher NA may introduce additional side lobes within the same field of view (FOV), sacrificing axial resolution and SNR and negating the benefits of using a higher NA excitation objective. We address this issue by simulating the beam used for light-sheet excitation and validating our simulation results experimentally by scanning gold nanoparticles.

The beam simulation is based on the beam propagation method (BPM). The BPM represents an analytical solution to the Helmholtz equation [29], as shown in Eq. (1). The accuracy of the simulation was validated [30]: ∇2E(r→)+4π2λ02n2(r→)E(r→)=0.(1)

The basic idea of BPM is to divide the propagation path traveled by the beam in the Z-direction into N parts with a small distance dz between adjacent parts. The light field distribution of each cross-section can be calculated from the light field of the previous cross-section so that the light field of each section can be obtained by the method of split-step calculation.

E(r→) is the complex amplitude of the electric field at the coordinate vector r→=(x,y,z), λ0 is the wavelength of light in a vacuum, and n(r→) is the refractive index, composed of n0 and δn(r→). In the paraxial approximation, while the beam propagates forward along the Z-axis, the complex amplitude of the electric field can be expressed as Eq. (2): E(r→)=U(r→)exp(j2πλ0n0z),(2)where U(r→) is the amplitude at the entry of the dz slice. According to the idea of the split-step calculation of the beam propagation theory, the change in U(r→) at each step is linear, the incident surface is defined as Uk, and the exit surface is defined as Uk+1 of each slice. The iterative relationship for Uk+1 and Uk can be expressed as Eq. (3): Uk+1=exp[jλ0dz4πn0∇⊥+j2πλ0δn(r→)dz]Uk.(3)

During the propagation of light fields, the additional phase of angular spectrum diffraction in the frequency domain and the additional phase of non-uniform refractive index distribution in the spatial domain are mainly calculated. The angular spectrum diffraction in the frequency domain is solved by Fourier transformation (FT), and the iterative relationship shown in Eq. (4) is obtained: Uk+1=[j2πλ0δn(r→)dz]×IFT{FT{Uk}exp[−jπλ0dzn0(fx2+fy2)]}.(4)

Considering that the incident angle at high NA is much larger than the paraxial approximation, we apply the propagation equation under a large incident angle [31], as shown in Eq. (5): Uk+1=[j2πλ0δn(r→)dz]×IFT{FT{Uk}⁢exp[−jλ0dz2πn0+4π2n02−λ02(fx2+fy2)(fx2+fy2)]}.(5)

Equation (5) is used to investigate the optical characteristics of the excited Bessel beam and scanning light sheet using a high NA objective. In the simulation, a 60×, NA 1.0, water immersion objective is used. The entrance pupil of the objectives is set to be the cutoff frequency corresponding to the edge of the projection on its anterior focal plane. High frequency plane wave components beyond this edge do not pass through the objectives. Therefore, the image on the anterior focal plane of the objectives is set to be tangential to the projection edge in order to determine the NA of the objectives. Additionally, the spectral distribution of Bessel beams in the frequency domain is a ring. The spectral distribution of Uk at the back focal plane of the objectives is set to form a ring distribution with an outer diameter of 6 mm and a thickness of 0.2 mm, and the corresponding NAs are 1.0 and 0.97, consistent with the experiment. After FT, the beam continues to propagate split-step in the Z-direction to obtain the Bessel beam under high NA focusing. The beam is scanned along the X-direction to obtain the corresponding optical field distribution of the light sheet. The sample environment used is a water-based solution, the refractive index of the medium is set to 1.33, and the wavelength of light in the medium is 561 nm.

The results of the simulation in Figs. 3(a)–3(d) and the results of the experiments in Figs. 3(e)–3(h) agree with each other well. According to the YZ distribution of a Bessel light sheet both in simulation results and experiment results as shown in Figs. 3(b) and 3(f), respectively, the Bessel light sheet still maintains a proper length of 8μm when NA is 1.0. The side lobe of the Bessel light sheet is distributed beyond 800 nm from the center, which can be seen in Figs. 3(a), 3(b), 3(e), and 3(f), and the intensity of the side lobe is 30% of the main lobe, which is shown in the normalized intensity distribution [inset of Figs. 3(a) and 3(e)]. The axial resolution and the optical sectioning in LSFM should be decided by the overlap of excitation (green) PSF and detection (blue) PSF [Fig. 3(i)]. The PSF of the detection path [PSFd(x,y,z)] is characterized in Fig. 3(j). The PSF of the excitation path [PSFe(x,y,z)] is shown in Figs. 3(f)–3(h), so the overall PSF [PSFo(x,y,z)] can be written as PSFo(x,y,z)=PSFe(x,y,z)×PSFd(x,y,z).(6)

In the light-sheet system, the PSFe(x,y,z) is the intensity distribution of the scanned light sheet [Ie(x,y,z)], so PSFo(x,y,z) can be written as PSFo(x,y,z)=Ie(x,y,z)×PSFd(x,y,z).(7)

The theoretical result and the experiment result of overall PSFo(x,y,z) of the light-sheet system are shown in Fig. 3(k) and Fig. 2, respectively. The theoretical resolution of RTLIS is 290 nm in XY, and 270 nm in Z, when only the main lobe is considered. The experimental resolution of RTLIS is 290 nm in XY, and 310 nm in Z, so there is little impact brought by the side lobe, but the system also shows around 300 nm isotropic distribution in 3D without deconvolution and deep learning algorithms. The system design allows for the use of objectives with higher NA for light-sheet excitation, achieving higher axial resolution without sacrificing lateral resolution. At the same time, the system also allows for light sheet imaging with a Gaussian beam, further improving contrast, but at the expense of the FOV to some extent. Detailed experiments and simulations of the Gaussian beam light sheet are discussed in Appendix C.

Fluorescent microspheres with diameters of 2, 5, and 10 μm are used as our imaging samples. The microspheres are mixed with agarose solution and injected into the glass chamber. Figure 4(a) shows the sectioned fluorescence images of one cluster of three microspheres. Figure 4(b) shows the optical sections of 10 μm microspheres at different depths. Figure 4(c) shows the fluorescence image of a 2 μm microsphere. Figures 4(d)–4(f) show the 3D views of two microspheres.

The Rh6 cell of the Drosophila eye is used for the bio-cellular imaging. The Drosophila eye provides a powerful system to study cell-fate determinations: it develops from a flat epithelium into a complex three-dimensional structure of multiple cell types in less than a week [32]. In the Drosophila compound eye, each of the 800 ommatidia is a single optical unit containing photoreceptor cells (PRs). The eight PRs form widely expanded membrane structures, rhabdomeres, which contain the photosensitive rhodopsin (Rh). The discovery of the three-dimensional structure of Rh6 cells in adult Drosophila eye has significant implications for the study of gene expression. RFP expression in Rh6 cells of living flies was assessed by neutralizing the cornea using water immersion. This feature is well-suited for fluorescence imaging with excitation wavelength at 561 nm. Using RTLIS, we achieved simultaneous 3D volume imaging of the Rh6 cell throughout a whole sample at a speed of 100 frames per second. Figure 5 shows the 3D views of the fluorescence images of the Rh6 cells. The volume size is 7.23μm×26.8μm×31.4μm.

In summary, we have demonstrated reflected ultrathin light-sheet microscopy with isotropic 3D resolution. With this configuration, a high NA objective lens can be fully utilized for both excitation and detection, improving spatial resolution, especially axial resolution, achieving 300 nm isotropic resolution in 3D. Furthermore, higher axial resolution can be achieved with a higher NA excitation objective (see Appendix C). Additionally, the ingeniously designed glass chamber allows for easily transforming a conventional inverted microscope into light-sheet microscopy, potentially making it the simplest LSFM setup. The proposed scheme can also be easily integrated with other imaging modalities, such as super-resolution microscopy for promising biomedical applications. It should be noticed that employing high NA excitation objectives also leads to a reduced axial extent of the beam, even when using Bessel beams as reported in current work. This trade-off necessitates sacrificing the field of view in order to achieve thinner light sheets. To address this limitation, one can integrate axially swept light-sheet microscopy [33,34], thereby enabling high resolution imaging over a larger field of view.