Non-invasive long-term three-dimensional (3D) imaging of biological tissues is crucial when exploring physiological and pathophysiological processes that traverse multiple spatial and temporal scales and involve large cohorts of cells. However, the high photon-scattering properties of biological samples fundamentally limit imaging depth while deteriorating imaging quality and restricting imaging duration. In particular, for dense and highly scattering specimens such as tissue slices, organoids, developing embryos, and smaller organisms, it has been difficult to obtain a full view observation throughout the entire sample with high spatiotemporal resolution.

A number of volumetric microscopes have been proposed to address these challenges [1]. Among them, multi-photon microscopy has been proved to be a competitive tool for the imaging of thick turbid samples with subcellular resolution [2–4], leveraging the longer excitation wavelengths and nonlinear excitation property. However, achieving volumetric imaging inevitably requires trade-offs among the spatiotemporal resolution, imaging volume, and imaging duration, as well as the signal-to-noise ratio (SNR) [5,6]. Various improved multi-photon microscopes have attempted to reduce the phototoxicity and increase the imaging speed by employing strategies such as high-speed scanning [7,8], point spread function (PSF) sculpting [9–13], temporal multiplexing [14–16], and multi-angle line scanning [17,18]. Three-photon microscopy, with its high-order nonlinear excitation, presents a promising avenue for extending the imaging depth in scattering samples [19–22]. However, the requisite high laser pulse energy (microjoule level) and low pulse repetition rate (hundreds of kilohertz level) impose limitations on the imaging speed essential for functional volumetric imaging. In all of these approaches, it is necessary to increase the excitation intensity exponentially with the imaging depth in order to penetrate the tissues, raising concerns as to the tolerability of high laser powers and poor signal-to-background ratios (SBRs) [23]. In addition, multi-photon excitation exhibits higher-order dependencies with the excitation intensity in photobleaching and phototoxicity [24,25]. Consequently, long-term high-spatiotemporal resolution 3D imaging of thick scattering samples remains a persistent challenge.

Here, we propose a novel dual-objective two-photon microscope configuration, Duo-2P. In contrast to traditional two-photon microscopes, our bidirectional imaging system excites each half side of the sample, extending the maximum imaging depth to more than twice the original limit determined by the SBR. As compared to a single-sided two-photon microscope, Duo-2P achieves a remarkable one-order-of-magnitude reduction in total excitation energy without compromising the SNR at a sample thickness fivefold the scattering length. The additional fluorescence collected by the contralateral system further improves the image quality, especially in deeper layers of the sample, enhancing the SNR by a maximum of 1.4 times without the need to increase the excitation intensity or pixel dwell time. Thus, it is particularly well-suited for long-duration 3D imaging of thick scattering samples without spatiotemporal sparsity requirements. With these capabilities, we performed volumetric calcium imaging for thousands of neurons in the highly scattering suprachiasmatic nucleus (SCN) and depicted their neuronal behaviors in space and time.

The Duo-2P was designed to extend the depth of 3D imaging and enhance the SNR through bidirectional excitation and collection by dual objective [Fig. 1(a)]. We built two sets of resonant scanning two-photon microscopes, one in the upright and the other in the inverted configuration, and both aligned along the same optical axis. The objective on each side is responsible for imaging half of the sample’s volume, driven by a piezo objective scanner for axial focusing. The two scanners operate in coordination to image the upper and lower focal planes alternately, and the two $\mathrm{Z}$-stacks of images are synthesized to obtain a complete 3D dataset [Fig. 1(b)]. In our system, we used two-dimensional (2D) resonant scanners, which acquire a typical $512\times 512$-pixel image in 33 ms. A Pockels cell is utilized to gate the laser pulses from a Ti:sapphire femtosecond laser to control the bilateral excitation of the upright and inverted two-photon microscopes, frame by frame [Fig. 1(c)]. The scanner of the objective delivering laser excitation remains stationary while the contralateral objective scanner executes a vertical stepping motion to drive its high-NA, 350-g objective to the next positions, with 5–10 μm inter-layer steps. Meanwhile, the emitted fluorescence photons are collected simultaneously by the dual objectives and detected by two independent photomultiplier tubes (PMTs).

As shown in Fig. 2(a), both the excitation and collection share the same focal plane on the excitation side. Meanwhile, on the opposite side, the objective’s focal plane does not align with the fluorescence position, leading to the unfocused contralateral collection. We quantified the efficiency of unfocused contralateral collection using Monte Carlo analysis with Zemax [Figs. 2(b) and 2(c)]. The simulation was based on the Henyey–Greenstein model (mean free $\mathrm{path}=46.7\mathrm{\mu m}$ at $\lambda =555\mathrm{nm}$, $g=0.93$, corresponding to a mean free path of 100 μm at the excitation wavelength of 920 nm). The results indicate that the fluorescence collection efficiency in two-photon microscopy primarily depends on the imaging depth, and the difference in collection efficiency between both sides becomes smaller as the imaging depth increases. The collection efficiency of Duo-2P is also shown in Fig. 2(d), alongside the collection efficiency of the epi-two-photon microscope (epi-2P) for comparison. Bidirectional fluorescence collection improves the fluorescence collection efficiency, consequently enhancing the SNR, as quantified in Fig. 2(e). Image detectors may encounter two primary sources of noise—additive noise, typically following a Gaussian distribution, and multiplicative noise following a Poisson distribution. In our calculations, we estimated the Gaussian noise distribution from real PMT data, assuming an identical and independent noise distribution on both sides. The results show that the bidirectional emission collection effectively improves the image SNR, achieving a maximum 1.4-time enhancement when the collection efficiency is equal on both sides. This enhancement is equivalent to that gained by a 1.8-fold increase in fluorescence intensity. Evidently, the simultaneous collection of contralateral fluorescence and epi-fluorescence eliminates the need for image registration during image fusion.

In volumetric imaging, the attenuation caused by light scattering and absorption within tissues needs to be compensated by increasing the excitation intensity exponentially with the increasing depth. Moreover, the situation becomes worse when the collection efficiency decreases deep in the tissues. By exciting each half of the sample, Duo-2P dramatically reduces the total excitation energy imposed on the tissues [Fig. 3(a)]. Furthermore, bidirectional photon collection allows for a reduction in excitation intensity without compromising the SNR. The efficacy of Duo-2P in decreasing the total excitation energy in 3D imaging is quantified through theoretical calculations, as depicted in Fig. 3(b). The ratio of the total excitation energy in epi-2P to that of Duo-2P exhibits an approximately exponential rise with an increasing sample thickness. Notably, at a sample thickness five times the scattering length, Duo-2P only requires about one-twelfth of the excitation energy compared to epi-2P, achieving a remarkable reduction of an order of magnitude.

To validate and quantify the Duo-2P method, tissue-like phantoms containing 5 μm diameter fluorescent microspheres were used for both Duo-2P [Fig. 4(a)] and epi-2P [Fig. 4(b)] imaging. The scattering length of the phantom was estimated to be approximately 95 μm by measuring the decay of the fluorescence signal [Fig. 4(c)]. A comparison of the two volumetric images indicates that the image quality of the shallow layers in Duo-2P is comparable to that of epi-2P. However, as the imaging depth increases, differences in SNR and SBR become apparent between the two methods [Fig. 4(d)]. Duo-2P attains optimal image qualities on both sides of the sample’s surface. The calculation of the SBR with increasing depth in volumetric imaging was conducted, indicating that SBR decreases with imaging depth. For imaging half of the sample’s thickness from each side, Duo-2P effectively expands the depth limited by the acceptable SBR to twice the original without introducing additional photodamage [Figs. 4(e) and 4(f)]. Moreover, the ratios of fluorescence collection efficiency in Duo-2P to that of epi-2P imaging, calculated by measuring the intensities of fluorescent microspheres, illustrate the improvement of fluorescence collection efficiency [Figs. 4(g) and 4(h)]. This improvement is evident even on the sample surface and increases with imaging depth. Thus, from the superficial to the intermediate layers of the sample, contralateral fluorescence collection efficiency incrementally increases to match that of epi-collection. This outcome is particularly beneficial for improving the image quality of intermediate layers, which are most affected by scattering in Duo-2P. The statistical analysis of SNR with different imaging depth was also performed, as shown in Fig. 4(i). Furthermore, the SNR enhancement from improved collection efficiency allows for a reduction in excitation intensity while maintaining the same SNR as in epi-2P [Fig. 4(j)].

Although the molecular and cellular basis of the biological circadian clock has been well-established [26,27], little is known about system-level mechanism of the SCN, the master clock of biological rhythms. Volumetric calcium imaging of Neuromedin S (Nms) neurons in adult SCN can provide a groundwork for investigating the underlying biological principles of time computing and encoding mechanism. However, due to the dense distribution with $\sim \mathrm{20,000}$ heterogeneous neurons in a mere volume of $0.2{\mathrm{mm}}^3$, high scattering exponentially increases the difficulty of imaging [23]. With Duo-2P, we were able to image continuous calcium activities in thousands of Nms neurons in SCN slices, at a rate of 1.67 s/volume for 4 h, and we characterized the state-switching behavior of these neurons, as shown in Fig. 5. The entire imaging volume was $550\mathrm{\mu m}\times 550\mathrm{\mu m}\times 380\mathrm{\mu m}$ (190 μm on each side), corresponding to $512\times 512\times 50$ voxels. A total number of 4377 neurons were detected in the demonstrated recording. Baseline corrected calcium traces of 30 example neurons from three layers are shown in Figs. 5(a)–5(f). The majority of neurons were non-stationary, displaying activities that wax and wane over different time scales. Interestingly, the continuous 3D imaging visualized that a small number of neurons [8 out of 4337, as shown in Figs. 5(g) and 5(h)] exhibited the state-switching behavior transiting between negative pulse oscillation and positive pulse oscillation, and the fire frequency changed over time [Fig. 5(i)]. The results demonstrated the feasibility of long-term calcium imaging in fresh adult SCN slices, affirming its capability for 3D imaging of highly scattering thick specimens. Compared to traditional epi-2P, Duo-2P doubles the imaging depth, while the excitation energy required for volumetric imaging is substantially reduced. This reduction in excitation energy minimizes photodamage, allowing the acquisition of reliable, long-term, large-volume imaging data to capture infrequent relevant physiological processes.

The evolution of light-sheet microscopy [28–31] and light-field microscopy [32,33] has facilitated large-volume, long-term 3D imaging with subcellular resolution in weakly scattering samples. However, the scattering of fluorescent signals renders wide-field acquisition nearly impractical in highly scattering samples. Given the unparalleled advantages of two-photon microscopy in highly scattering samples, Duo-2P employs layer-by-layer Gaussian focus resonant scanning on both sides of the sample, effectively doubling the imaging depth compared to traditional two-photon microscopy. Our system utilizes 8 kHz resonant scanners, achieving a frame rate of 30 frames per second at a typical frame size of $512\times 512\mathrm{pixels}$. Bidirectional collection enhances the SNR without compromising the temporal resolution or causing additional photodamage to samples, thus compensating for the SNR reduction associated with resonant scanning. Simultaneous bidirectional fluorescence collection and accurate alignment of Duo-2P ensured volume reconstruction for the comprehensive 3D dataset without the need for image registration. An overlap of imaging volumes from the upper and lower sides of Duo-2P could further increase the continuity and integrity of the synthesized sample volume with image fusion. Moreover, the introduction of bidirectional excitation and collection can reduce the required excitation energy in thick scattering samples by more than an order of magnitude, offering substantial benefits for long-duration volumetric imaging.

In contrast to a tomographic imaging method such as two-photon synthetic aperture microscopy [18], which reduces phototoxicity through fewer scans, Duo-2P minimizes excitation energy input based on a new microscope configuration. This approach avoids the complexities of image reconstruction and spatiotemporal sparsity priors, and shows excellent robustness in dense samples with the resolution unaffected by imaging depth or label density. In deep-imaging three-photon microscopy, greater imaging depths often require lowering the laser repetition rate to prevent tissue damage, but lower repetition rates confound the temporal resolution. Duo-2P, on the other hand, achieves large-volume resonant scanning imaging through bidirectional excitation, effectively doubling the imaging depth limited by SBR. The layer-by-layer scanning, facilitated by the high-speed mechanical axial movement of the objective in Duo-2P, enables a large imaging range, distinguishing it from techniques that utilize remote focusing [34], acousto-optic lenses [35–38], or liquid lenses [39]. At the same time, the effective working distance of the objective is also doubled, which would be helpful for the imaging depth beyond the objective’s working distance in low scattering samples. In Duo-2P imaging, images are obtained alternately frame by frame from both sides with the objectives stepping during the excitation intervals. This is especially crucial for the heavy high-end two-photon objectives, since it eliminates the delays from the objective scanner’s responses between frames, maximizing the volumetric imaging speed.

Duo-2P holds great potential for imaging applications involving overall observation of highly scattering samples such as organoids, pancreatic islets, and embryos, and in vivo imaging of small organisms. In this study, we have demonstrated the application of Duo-2P in SCN slices, which posed great challenges for previous imaging methods. In the future, the configuration of bidirectional imaging could be integrated with other imaging techniques such as temporal multiplexing or PSF sculpting to further increase the imaging speed and volume. The configuration of symmetrically arranged collection modules on both sides in Duo-2P is also suitable for imaging applications such as second harmonic generation [40] and coherent anti-Stokes Raman scattering [41]. Moreover, the imaging depth of Duo-2P could be further increased by introducing three-photon excitation [19–22,42], and the reduction of excitation energy by a dual-objective configuration, coupled with a high-repetition-rate laser, might break the trade-off between temporal resolution and cellular health in three-photon microscopy. However, Duo-2P has inherent physical size limitations between objectives, which would permit in vivo imaging only to organisms of several millimeter thickness.

In summary, we propose a bidirectional resonant scanning two-photon microscope and have imaged continuous calcium activities of thousands of Nms neurons in adult SCN slices and characterized their state-switching behavior. Compared to traditional two-photon microscopy, this method markedly reduces the total excitation energy used for 3D imaging in thick scattering samples, increases the image SNR, and doubles the imaging depth. This method exhibits versatility across various sample and labeling densities and can be readily extended to other nonlinear microscopy systems.

The Duo-2P system is shown in Fig. 6. The light source for two-photon excitation is a commercial femtosecond Ti:sapphire laser (Chameleon Vision S, Coherent) with a repetition rate of 80 MHz, and the central wavelength is 920 nm. The Pockels cell based electro-optic modulator (350-80, ConOptics) is placed in a reversed position to adjust the laser polarization state. The laser path is switched through the PBS (CCM1-PBS252, Thorlabs) between the upper and lower sides depending on the polarization state. Two independent acousto-optic modulators (MT110-A1.5-IR, A&A Optics) are used to adjust the laser power on the upper and lower sides. The maximum average power output of the objective is about 480 mW. The laser beam is directly expanded by variable-magnification beam expanders (ZBE1B, Thorlabs) on the upper and lower sides to match the size of the back pupil planes of different objectives. The excitation modules (OPX1100, Thorlabs) and fluorescence collection modules (BDM3214S, Thorlabs) are symmetrically arranged on the two sides of the sample plane. The Duo-2P system uses two independent two-dimensional resonant scanners (LSK-GR08, Thorlabs) for high-speed scanning. The fluorescence signal is collected by dichroic mirrors (FF705-Di01, Semrock), bandpass filters (FF03-525/50, Semrock), and PMT (H10770A-40, Hamamatsu). Two water immersion objectives (CFI75 Apochromat 25XC W, Nikon) are used in the Duo-2P system for high-NA fluorescence collection. The imaging sample is placed in the custom-made imaging chamber with a cover glass as the bottom. The lower objective images through the cover glass, and the upper objective is directly immersed in the culture medium. The spherical aberration correction rings of the objectives are adjusted accordingly. Bilateral volumetric imaging is performed by the mechanical axial scanning of the objectives using the piezo objective scanners (PFM450E, Thorlabs). The Duo-2P system uses an FPGA-based controller (TVS-MMC-01, Transend Vivoscope) for real-time control of galvanometer scanning, laser power modulation, and image acquisition. In practice, we have developed the control system and software to integrate two different imaging modes, the traditional epi-2P mode and the Duo-2P mode, to compare the performance of these two methods. In epi-2P mode, only one side of the resonant scanner is used and image acquisition is limited to the corresponding side. In Duo-2P mode, the resonant scanners on both sides are used and the laser alternated between the upper and lower sides each frame. Images on both sides are acquired at the same time. The Duo-2P system has been calibrated by a fluorescent target (Argo-HM, Argolight) for the same field-of-view (FOV) size of the two sides. With the adjustment of the relative position between the upper and lower sides microscopes by high-load translation stages, the centers of FOVs from the two sides have been aligned as well. The proportional-integral-derivative parameters of piezo objective scanners have also been well tuned for smooth and accurate movement with allowed longer response time.

The collection efficiency in relation to the imaging depth and defocus distance was obtained using Zemax OpticsStudio. Except for the objective (patent: US6501603 B2), paraxial lenses were used for the simulation. The Henyey–Greenstein scattering model was applied to the object with a mean path of 46.7 μm at the wavelength of 550 nm. The object and image were conjugated initially for the source point at the focal plane of the objective and detector at the image plane. The distance between the object and the objective’s focal plane was introduced to simulate the contralateral collection. The impact of coverslips on the collection efficiency is ignored. The collection efficiencies were also acquired with the simulation of different source depths in the scattering sample.

For the simulation of SNR in Fig. 2(e), we mainly consider two sources of noise, additive noise following a Gaussian distribution with zero mean and the multiplicative noise following a Poisson distribution, which are generally independent. The SNR of epi-2P with additive and multiplicative noise was defined as $${\mathrm{SNR}}_{\mathrm{epi}}=\frac{E(S_{\mathrm{epi}})}{\sqrt{{\sigma }_{\mathrm{epi}}^2+E(S_{\mathrm{epi}})}},$$(1)where $E(S_{\mathrm{epi}})$ represents the expected signal $S_{\mathrm{epi}}$ on the detector, and ${\sigma }_{\mathrm{epi}}$ represents the standard deviation of Gaussian noise in epi-collection images. The SNR of Duo-2P was calculated by introducing the bidirectional signal collection defined as $${\mathrm{SNR}}_{\text{bidirectional}}=\frac{(1+{\alpha }^2)E(S_{\mathrm{epi}})}{\sqrt{{\sigma }_{\mathrm{epi}}^2+{\alpha }^2{\sigma }_{\text{contralateral}}^2+(1+{\alpha }^2)E(S_{\mathrm{epi}})}},$$(2)where $\alpha$ represents the weight for introducing the contralateral signals, which was obtained by calculating the ratio of collection efficiency on both sides: $\alpha =S_{\text{contralateral}}/S_{\mathrm{epi}}$. And ${\sigma }_{\text{contralateral}}$ represents the standard deviation of Gaussian noise in contralateral collection images.

As the light absorption is negligible compared to scattering in most biological samples at the near-infrared wavelength range, the laser intensity of the two-photon microscope at the sample surface increases exponentially with imaging depth to compensate for the photon scattering and maintain the same signal intensity [3], which is illustrated in Fig. 3(a). The required laser power at the imaging penetration depth $z$ is given by $$P_z=P_0{e}^{z/l_s},$$(3)where $P_0$ represents the laser power when imaging the sample surface and $l_s$ represents the scattering length. Thus, the total excitation laser energy for the volumetric imaging by epi-2P and Duo-2P is defined as $$E_{\mathrm{epi}\text{-}2\mathrm{P}}=\sum _{m=1}^{m=n}{P}_m{t}_{\mathrm{frame}},$$(4)$$E_{\mathrm{Duo}\text{-}2\mathrm{P}}=2\sum _{m=1}^{m=n/2}{P}_m{t}_{\mathrm{frame}},$$(5)where $P_m$ represents the laser power at the $m$th layer of a total of $n$ layers and $t_{\mathrm{frame}}$ represents the scanning time for one frame. Taking into account the simulation collection efficiency $\eta$ in Fig. 2(c) and the calculated fluorescence intensity $F$ with same SNR in Fig. 2(e), the laser power at the $m$th layer with the same SNR is calculated as $$P_m=P_0{e}^{z_m/l_s}/\sqrt{\eta F}.$$(6)

To compare the performance of traditional epi-2P and Duo-2P (shown in Fig. 4), we imaged the fluoresce-labeled microspheres in tissue phantoms. The phantom consists of low-melting-point agarose (1%, Macklin) containing nonfluorescent microspheres (1 μm, Invitrogen) at a concentration of $9.0\times {10}^9\text{beads}/\mathrm{mL}$. The scattering anisotropy and scattering length are 0.88 and 100 μm, which were calculated with Mie theory. The concentration of fluorescent polystyrene microspheres (5 μm, Thermofisher) is $1.5\times {10}^7\text{beads}/\mathrm{mL}$. The scattering properties are not affected significantly by the fluorescent microspheres at this concentration. The agarose gel containing fluorescent microspheres was well mixed and cooled. The imaging experiments were performed in both Duo-2P mode and traditional epi-2P mode. The original image size shown in Figs. 4(a) and 4(b) is $1000\times 1000\times 480\mathrm{pixels}$, corresponding to a volume of $220\mathrm{\mu m}\times 220\mathrm{\mu m}\times 700\mathrm{\mu m}$. We performed repeated imaging for three times for each $Z$-position in both the Duo-2P mode and traditional epi-2P mode. The range of average laser power on the sample surface was 2–51 mW under Duo-2P mode and 2–480 mW under epi-2P mode.

The decay lengths were quantified by analyzing the fluorescence intensities of fluorescent microspheres at different depths, given the same excitation laser power. Three different FOVs from the same depth range were selected. For each FOV to be measured, volumetric stacks were taken with steps of 1.5 μm from the shallowest to the deepest. The excitation laser power was adjusted to avoid saturation at the top layer. The bias voltage of the PMTs was measured by acquiring the images with the PMTs’ shutters closed and calculating the mean grayscale value of the images. The pixels that were three standard deviations above the mean intensity were segmented as the fluorescent signal of the current depth, after averaging and bias subtraction. The logarithms of the mean intensities of the segmented pixels from the three FOVs were taken, and linear decay curves were fit to the logarithms. The fitting coefficient can indicate the length in micrometers through which the fluorescent signal will decay by a factor of constant $e$. The fluorescence intensity of two-photon excitation is proportional to the square of the excitation intensity. Thus, the scattering length is two times the reciprocal of the fitting coefficient from the fluorescent signals.

Volumetric images of the fluorescent microspheres in the tissue phantoms were taken with epi-2P mode and Duo-2P mode for the calculation of SBR in Fig. 4(f). The laser power was increased exponentially with imaging depth to maintain the fluorescence signals. We first removed the bias of the images. The signal pixels were segmented by binarizing the images, and the artifacts were removed by the area threshold. Circle fitting was used to define the signal and background regions, as shown in Fig. 4(e), where the pixels within the circle were segmented as the signal, and the pixels around the circle and within the ring as the background. The mean intensity of the signal pixels was calculated as $S$, and the background pixels as $B$. SBR could be obtained as follows: $\mathrm{SBR}=S/B$. We did the calculations for all the microspheres within the volume layer by layer.

For the comparison of collection efficiency in Fig. 4(h), volumetric images of fluorescent microspheres in the tissue phantoms were acquired from the $Z$ position for every 50 μm of depth in both epi-2P mode and Duo-2P mode. The laser power was increased exponentially with imaging depth to maintain the fluorescence signals. In epi-2P mode, only epi-collection images were acquired, while both epi-collection and contralateral collection images were acquired in Duo-2P mode. The signal pixels of fluorescent microspheres were segmented by subtracting the background and binarizing the images, and the artifacts were removed by the area threshold. The mean intensities of the signal pixels from epi-collection and contralateral collection were calculated as $I_{\mathrm{epi}}$ and $I_{\text{contra}}$, respectively. The comparison of collection efficiency could be measured by the ratio of total fluorescent microspheres’ intensities from Duo-2P mode and epi-2P mode to $I_{\mathrm{epi}}$ as follows: ${\mathrm{Ratio}}_{\text{collection}}=(I_{\mathrm{epi}}+I_{\text{contra}})/I_{\mathrm{epi}}$.

For the SNR, which varies with the intensity of the excitation power, the images were acquired with a gradient of excitation power from the $Z$ position for every 50 μm of depth in both epi-2P mode and Duo-2P mode. The bias of the images was subtracted first, and the signal pixels were segmented by binarizing the images. For Duo-2P, we estimated the difference of contralateral collection and epi-collection by calculating the ratio of averaged images at the same depth and the excitation power from epi-collection frames and contralateral collection frames pixel by pixel. Using the weights of contralateral/epi collection ratio $\alpha$, we fused the images collected in both modes. The mean intensities of the signal pixels of each microsphere were calculated as $S$, and the standard deviation as $N$. The SNR could be calculated as follows: $\mathrm{SNR}=S/N$. For easy comparison in Fig. 4(i), the images within the same range of SNR as epi-2P were screened across different excitation powers at the same imaging depth.

All animal experiments were performed in accordance with the Animal Care and Use Committee of Peking University accredited by AAALAC International, and the procedures were approved by the Animal Care Committee of PKU-Nanjing Institute of Translational Medicine (Approval ID: IACUC-2021-023). All mice were group housed, at 20°C–22°C in a 12-h light/dark cycle, with ad libitum access to water and food. The Nms-Cre::GCaMP6s mice used in this study were generated by crossing Nms-Cre mice (JAX #027205) [43] with Rosa26-lsl-GCaMP6s mice [44] for at least two generations.

For SCN slice preparation, Nms-Cre::GCaMP6s male mice aged 6–8 weeks were used. Mice were anesthetized with isoflurane and subsequently decapitated. The brain was quickly removed and immersed in ice-cold oxygenated sectioning solution containing (in mmol/L) the following: 110 choline chloride, $25{\text{NaHCO}}_3$, 25 D-glucose, $7{\mathrm{MgCl}}_2·{6\mathrm{H}}_2\mathrm{O}$, 2.5 KCl, $1.3{\mathrm{NaH}}_2{\mathrm{PO}}_4$, $0.5{\mathrm{CaCl}}_2$, 1.3 L-sodium ascorbate, and 0.6 sodium pyruvate [45]. An approximately 350-μm coronal SCN slice was prepared using a vibratome (VT 1200, Leica) and then incubated in artificial cerebrospinal fluid (aCSF) containing (in mmol/L): 124 NaCl, $24{\text{NaHCO}}_3$, 10 D-glucose, 3 KCl, $1.25{\mathrm{NaH}}_2{\mathrm{PO}}_4$, $2{\mathrm{CaCl}}_2$, and $1{\mathrm{MgSO}}_4$ at 34°C for 30 min [46].

For calcium imaging of Nms neurons shown Fig. 5, an acute brain slice containing the SCN was mounted in a specially modified imaging chamber (RC-27LD, Warner) equipped with a coverslip at the base. The brain slice was supported by Lycra threads, ensuring a gap of 500 μm between the slice’s bottom surface and the coverslip. Filtered aCSF was driven by a peristaltic pump (BT100-2J, Longer) to flow through both the upper and lower sides of the brain slice at a flow rate of 2–3 mL per minute and suctioned into a waste bottle subsequently. The imaging chamber base (PM-7D, Warner) was heated by a customized temperature controller to consistently maintain a temperature of 35°C. The aCSF was pre-warmed and maintained at a steady temperature of 32°C–34°C within the imaging chamber.

Volumetric image stacks were acquired by Duo-2P with a size of $512\times 512\mathrm{pixels}$ and 50 slices, corresponding to a volume of $550\mathrm{\mu m}\times 550\mathrm{\mu m}\times 380\mathrm{\mu m}$, as the voxel size was set to $1.07\mathrm{\mu m}\times 1.07\mathrm{\mu m}\times 7.6\mathrm{\mu m}$. Each side of the Duo-2P imaged one half of the entire thickness of the SCN slice, and the imaging speed was 0.6 volume/s. The SCN slice was imaged continuously for 6 h, and the imaging paused for several minutes every half an hour for perfusion status checks and initial focal plane calibrations.

After acquisition of calcium imaging data, the image stacks were split as time-lapse images for every layer. The captured data of 9000 volumes with stable focal plane within the last four and a half hours were used for analysis. The images from epi- and contralateral collections were fused frame by frame. The lateral movements were corrected, and the background was subtracted using MATLAB scripts. Neuron segmentation was accomplished manually using ImageJ. The intensity of neuron activities at every sampling point was calculated as the mean intensity of the segmented pixels. Traces of $\mathrm{\Delta }\mathit{F}/\mathit{F}$ were obtained by using the median of a 60-s-window around each sampling point as the baseline. We utilized continuous wavelet transformation with Gabor wavelet as the basis to obtain the spectrum of the given traces.

Acknowledgment. We thank Danlei Wu for useful discussion in Zemax simulation. We thank Jiazhi Zhang and Dr. Haiwen Li for helpful suggestions regarding the data analysis.