WorksPhysics and Astronomy
Two-photon light-sheet live imaging at kilohertz frame rate using birefringence-based pulse splitting
Abstract
Multiphoton microscopy is widely used for live imaging. However, its acquisition speed remains limited by fluorophore emission rates and photodamage. To increase the pixel rate of a two-photon microscope beyond a few megahertz (MHz), multi-point parallelized schemes have been proposed. Two-photon (2P) light-sheet microscopy emerges as an effective approach for high-speed multiphoton imaging of live specimens, as it enables parallelized excitation while minimizing the required laser power. However, optimizing the signal-to-photodamage ratio in 2P light-sheet microscopy requires to precisely control the illumination parameters, including both wavelength and pulse frequency. Since conventional femtosecond laser sources generally do not allow independent modulation of these parameters, the development of low-cost, efficient and robust strategies to modulate the temporal excitation profile is essential to fully exploit the advantages of 2P light-sheet microscopy. Here, we introduce a compact pulse splitting scheme that meets these criteria. We used cascaded birefringent crystals to convert each excitation laser pulse into an adjustable sequence of collinear sub-pulses. We demonstrate its effectiveness in optimizing 2P light-sheet imaging of live zebrafish embryos. We analyze the impact of pulse splitting on photobleaching, nonlinear photodamage, and imaging performance. Additionally, we demonstrate high-speed 2P imaging of the beating heart and brain calcium dynamics using red fluorophores in live embryos. We achieve kilohertz imaging frame rate, reaching more than 150 MHz pixel rates with fluorescent signal levels above 10 using a laser mean power and a peak intensity in the range of 100 mW and 0.1 at the sample, respectively. This adjustable pulse-splitting scheme allows full advantage to be taken of light-sheet illumination for fast in vivo 2P imaging.
Introduction
Scaling up the imaging speed in multiphoton microscopy is an active field of research [1, 2, 3, 4]. However, developing fast two-photon (2P) imaging for live applications requires overcoming challenging trade-offs between image quality, signal level, sample heating [3, 5, 6, 7], photobleaching [8], nonlinear photodamage [9, 10], and optical setup complexity [8, 11]. Among strategies for increasing 2P imaging speed [1, 2, 3], multipoint light-sheet microscopy enables excitation parallelization with minimal laser power and peak intensity, which is critical for live imaging [12, 13]. However, optimizing the signal-to-photodamage ratio in 2P light-sheet imaging requires laser tunability in both excitation wavelength and pulse frequency (or pulse repetition rate) [14, 15]. Wavelength adjustment is required to maximize fluorophore excitation while minimizing water absorption-mediated heating. Laser peak intensity modulation through pulse frequency adjustment is required to reach an optimal trade-off between signal, heating, and nonlinear photodamage for a given application [14]. More generally, it is recognized that tuning the laser pulse frequency is beneficial for optimizing multiphoton microscopy of live biological samples [1]. However, most femtosecond laser sources used in multiphoton microscopy are not tunable in both parameters. Consequently, there is significant interest in developing methods to modulate the temporal illumination profile of wavelength-tunable pulsed laser sources.
Here, we report on a straightforward and robust approach to adjust the temporal illumination profile in a 2P light-sheet microscope. Specifically, we adapted the principle of pulse stacking with birefringent crystals [16], which has been employed for ultrafast pulse shaping [17] and amplification [18, 19]. By propagating a short pulse through cascaded birefringent crystals with appropriate thickness and orientation, a burst of pulses with controlled delay and alternating polarization is generated. Here, two birefringent crystals are used to obtain a burst of 1, 2, or 4 collinear sub-pulses spaced by picosecond delays. This birefringence-based passive pulse splitting scheme offers several advantages for adjusting the average pulse frequency compared to other methods used in multiphoton microscopy, such as pulse picking [20, 21], pulse splitting [8], or time gating [22]. Indeed, it is straightforward to implement, it requires no beam recombination or alignment since the output beams remain inherently collinear, and the splitting rate can be adjusted by simply rotating the crystals. The energy throughput of the pulse splitting setup is nearly lossless, unless a control of the linearly polarized pulse train is required.
In this work, we demonstrated the potential of birefringence-based pulse splitting in 2P light-sheet microscopy by optimizing the laser pulse frequency when exciting red fluorophores. We first selected the optimal wavelength based on the water absorption and 2P absorption spectra of mCherry fluorescent protein, to achieve enhanced two-photon excited fluorescence (2PEF) while reducing thermal effects. In addition, by using the scaling laws from [14], which describe the dependence of 2PEF signal, nonlinear photodamage and heating on laser power and pulse frequency, we predicted the optimal pulse frequency to enhance the 2PEF signal, while minimizing nonlinear photodamage. Together, as detailed in the supplemental data (see Methods section VIII A for more information), the optimal imaging parameters were determined to be a wavelength of 1070 and a pulse frequency in the range of 16 . To obtain such illumination parameters, we employed a commercially available laser source at 1070 and 4 and demonstrated that birefringence-based pulse splitting could effectively increase the average pulse frequency to 16 . Hence, we achieved optimal live imaging of red fluorophores in zebrafish embryos at kHz frame rate, reaching more than 150 pixel rate.
Results
Efficient pulse splitting using birefringent crystals
The two-stage pulse splitting setup is illustrated in Figure 1a. It comprises two birefringent crystals (BC1-2) with thicknesses and , a half-wave plate (HWP), and a linear polarizer. Each birefringent crystal introduce a time delay between the extraordinary and ordinary axes, while their relative orientation enable to adjust the number of pulses per burst between 1, 2 or 4 (see Methods section VIII B for more information). The output pulses displayed alternating linear polarizations. The optional polarizer aligns them to a defined polarization, at the cost of a 50% energy loss. If polarization control is not required, it can be removed to avoid this loss. In this work, the polarization is controlled to improve fluorescence signal detection [10, 23].
Figure 1. (a) Principle of birefringence-based pulse splitting with linearly polarized output: (i) When the input pulse polarization is set to , and the OA of the first birefringent crystal (BC1), the second birefringent crystal (BC2), the half-wave plate (HWP), and the polarizer (P) both are aligned at , the output pulse frequency remains unchanged; (ii) Adjusting the OA of BC2 to and the OA of the HWP to while maintaining others parameters unchanged doubles the output pulse frequency; (iii) Setting the OA of BC1 to , the OA of BC2 to , and the OA of HWP to results in a fourfold increase in the output pulse frequency. All OA angles of optical elements are defined relative to the vertical direction in this work. (b) Schematic of the 2P light-sheet microscopy setup. A wavelength-tunable laser beam passes through a pulse-splitting (PS) unit, then an electro-optic modulator (EOM), and is subsequently directed to a galvanometric mirror (GM). The beam is relayed through two scanning lenses, which together conjugate the GM plane to the back focal plane of the objective (OB) with appropriate magnification. The laser then illuminates the sample through OB1, while OB2 provides white-light illumination. The emitted 2PEF signals pass through OB3 and tube lens (TL) before being recorded by a camera. (c) Laser parameters for configurations with pulse frequencies of 2 MHz (conventional), MHz, MHz, 4 MHz (conventional), MHz, MHz and 8 MHz (conventional), along with the corresponding pulse train temporal profiles. The laser frequency denotes the source pulse repetition rate, laser source specifies the employed laser system, splitting indicates the optical splitter configuration, average pulse frequency represents the average pulse frequency at the output of the splitter, and the notation refers to the corresponding labeling used in this work.
To bypass the pulse splitting, the input pulse is linearly polarized at , aligned with the optical axis (OA) of the first birefringent crystal (BC1). The OAs of both crystals, the HWP, and the polarizer are also set to , resulting in no pulse splitting (1× in Figure 1a(Figure 1i)). However, rotating the OA of the second crystal (BC2) to and the OA of the HWP to , while keeping other parameters unchanged, doubles the output pulse frequency (2× in Figure 1a(ii)). Further rotating the OA of the first birefringent crystal to and setting the OA of the HWP to leads to a four-fold increase in pulse frequency (2×2× in Figure 1a(iii)). All OA angles are defined relative to the vertical direction in this work.
We selected Yttrium vanadate () as the birefringent crystal due to its excellent transparency in the near-infrared range, limited group-velocity dispersion (GVD) for pulses longer than 200 in crystals thinner than 30 , and high birefringence with a typical polarization mode delay of 1 . We experimentally verified these properties by implementing 2× and 2×2× pulse splitting using one or two crystals of different thicknesses ( , ). Using anti-cut crystals, we generated collinear beams with distinct pulse polarizations and delays, achieving transmission in the 1030–1070 range. Experimental measurements showed minimal dispersion, with pulse broadening limited to 60 over 30 of crystal for initial 230 pulses (Opera laser at 1070 ) and negligible for 385 pulses (Satsuma laser at 1030 , see Methods section VIII C for details about laser sources). Autocorrelator measurements confirmed a pulse delay of 0.76 at 1030 , closely matching the theoretical value of 0.70 derived from optical constants [24] (see Methods section VIII B for more information). Thus, the 10 and 20 crystals introduced pulse delays of 7.6 and 15.2 , respectively.
Based on these properties, we implemented 2× and 2×2× pulse splitting using two crystals (10 and 20 ) placed in the illumination beam path of our 2P light-sheet imaging setup (Figure 1b). In the 2× configuration, only the second birefringent crystal (20 ) was used and the first one was removed. For experimental validation, illumination at 1030 was provided by a Satsuma laser (fixed wavelength, tunable pulse frequency) while an Opera laser generated 1070 illumination (tunable wavelength, fixed pulse frequency at 4 ). The microscope setup consists of the pulse splitter unit, an electro-optic modulator, two opposing illumination objectives, and a detection objective positioned perpendicular to the illumination (Figure 1b; see Methods section VIII C for details). We precisely adjusted the relative crystal orientations to achieve equal energy per pulse.
No degradation of image spatial resolution, 2PEF signal, and fluorophore photo-bleaching during in vivo imaging using birefringence-based pulse splitting
To validate the effectiveness of our birefringence-based pulse splitting technique for multiphoton light-sheet imaging, we used a fixed-wavelength laser source capable of being tuned to either 2, 4 or 8 . We could then characterize the impact of pulse splitting on image spatial resolution, 2PEF signal, and fluorophore photo-bleaching by comparing the same average pulse frequency with or without pulse splitting (Figure 1c). The effect of the pulse splitter on spatial resolution was assessed by estimating the Point Spread Function (PSF) from experimental 3D (x-y-z) images of fluorescent beads (see details in Methods section VIII D). Measurements were performed under varying conditions: with and without crystals, as well as with crystals of different thicknesses and orientations. The results showed that the proposed pulse splitters did not introduce significant changes to the PSF. The axial full-width at half-maximum (FWHM) remained consistent at , confirming that pulse splitting did not induce beam divergence or misalignment between the ordinary wave (o-wave) and extraordinary wave (e-wave), see additional details in Table I of the Methods section.
| Average pulse frequency () | Laser frequency () | Splitting | Crystal thickness () | Axial FWHM mean () | Axial FWHM standard deviation () | Axial FWHM mean uncertainty () |
| 4 | 4 | no | 0 | 3.55 | 0.06 | |
| 8 | 4 | 3.43 | 0.14 | |||
| 16 | 4 | 3.55 | 0.13 |
Table I. Axial PSF quantification. To estimate the axial PSF, we measured the axial FWHM of the signal z-profile from 20 beads per measurement. 4 measurements are used. Here, -splitting is obtained by rotating the second crystal to bypass the second splitting, which explains why we have 30 mm of crystal thickness at 8 MHz average pulse frequency. The laser source is the Opera at 1070 nm and 4 MHz
To evaluate the influence of the pulse splitter on 2PEF signal, we compared the signal from fluorescent beads under various conditions at . The results demonstrated that the 2PEF signal at (without the splitter) is comparable to the signal at (with one-stage splitter), and the signal at (without the splitter) matches the signal at (with the two-stage splitter) (Figure 2a). These findings indicate that our pulse splitter successfully achieves the intended pulse peak intensity and burst of pulses, with the short pulse delay (shorter than the fluorescence lifetime) not causing signal saturation.

Figure 2. Analysis of 2PEF signals and photon bleaching effect. (a) Violin plots comparing the 2PEF signals with and without the pulse splitter under varying pulse frequency conditions. (b) Normalized mean 2PEF signal across multiple cells over time under different pulse frequency conditions. (c) Violin plots depicting the decay of 2PEF signals from selected cells, illustrating the bleaching effect over time at each pulse frequency condition. indicates the mean value, represents the number of cells analyzed, and indicates the -value of t-tests, representing statistical significance between different groups of decay rates. The experiments (c-d) were conducted under laser illumination at a wavelength of 1030 .
Additionally, to demonstrate that the burst of pulses does not lead to increased photobleaching, we measured the 2PEF signal decay while imaging mCherry in live zebrafish embryos under illumination (details of the photobleaching experiment in Methods section VIII E). As shown in Figure 2b-Figure 2c, there was no significant increase in the photobleaching rate when imaging at , (with the splitter), and (with the splitter). This observation confirms that photobleaching is not sensitive to the short delays between pulses ( or ), which are significantly shorter than the fluorescence lifetime of the proteins. This is consistent with previous observations using GFP [2].
Wavelength adjustment to minimize water-mediated heating and improve red fluorescence signal
To investigate the impact of wavelength and pulse splitting on photodamage during 2P light-sheet imaging, we monitored the heart beating rate (HBR) in live zebrafish embryos, as previously reported [14] (see details in Methods section VIII F). Water absorption induces a linear heating effect, and the resulting temperature increase accelerates the HBR, as documented in previous studies. Notably, a linear relationship between HBR and temperature has been observed in live embryos under physiological conditions [14].
To demonstrate wavelength-dependent heating, we recorded the variation in HBR at excitation wavelengths of and (Figure 3a-Figure 3b). The results show a significant reduction in HBR variation and linear heating when the excitation wavelength is increased from to . Specifically, the relative HBR differences, () at and () at , correspond to a reduction factor of by changing the excitation wavelength from to . This finding aligns with the decreased water absorption at compared to by a factor of (Figure 6a in Methods section VIII A). These results confirm that the wavelength, which corresponds to a local minimum in water absorption, should be preferred to enhance the absorption of the target protein while minimizing linear heating during 2P light-sheet imaging. In addition, we note that mCherry 2P absorption almost doubles between and , further justifying the wavelength adjustment.

Figure 3. Experimental investigation of wavelength-dependent linear heating based on water absorption and zebrafish embryonic heart beat rate (HBR) measurements. (a) Representative instantaneous HBR in zebrafish over time under femtosecond laser illumination at 1030 (orange) and 1070 (red), showing the changes in relative to the baseline when the laser is on (pink time window). (b) Violin plot of ratios under 1030 (orange) and 1070 (red) femtosecond laser illumination, where each data point represents an independent measurement from a different individual zebrafish embryo. The violin plots show the median and the first and third quartiles. stands for the excitation wavelength, indicates the number of zebrafish examined, and represents the mean value.

Figure 6. Prediction of optimal laser parameter for live imaging of red fluorophores. (a) Comparison of the 2P absorption spectrum of mCherry (sourced from FPBase) and the water absorption spectrum, both normalized to their respective values at . The difference between and is highlighted (b) Mean power at the nonlinear photodamage threshold (), at the thermal effect threshold () and in optimal imaging conditions () depending on the laser pulse frequency. The shaded region between dashed lines represents the optimal imaging conditions with respect to nonlinear photodamage and laser frequency. Black dots indicate optimal imaging condition at (typically and ) and at (typically and )
Nonlinear photodamage modulation using birefringence-based pulse splitting
To further investigate photodamage when using birefringence-based pulse splitting, we examined the nonlinear photodamage threshold (corresponding to the maximum average power before inducing visible nonlinear photodamage, ) under various illumination conditions. We note that the splitting nonlinear photodamage experiment was conducted with a single birefringent crystal in place.
Using the laser at , we demonstrate that at and without pulse splitting (Figure 4, orange stars) is equivalent to at and at , respectively, when using or crystals (Figure 4, orange circles). This finding shows that the nonlinear photodamage threshold does not depend on pulse time distribution: whether the pulses are evenly distributed or delivered in bursts with the same peak intensity, the resulting photodamage threshold remains unchanged. Additionally, the threshold is unaffected by the crystal’s thickness.

Figure 4. Quantification of nonlinear photodamage threshold using zebrafish HBR in vivo. The mean power at the nonlinear photodamage threshold () is measured in zebrafish under femtosecond laser exposure across different conditions: no pulse splitting, pulse splitting, and pulse splitting, at pulse frequency of , , and , with illumination wavelengths of and .
When we introduced a second splitting stage (i.e., or , blue upward-pointing triangle in Figure 4), followed the expected scaling law, though with a slight upward shift. Specifically, the threshold was approximately higher than anticipated (orange upward-pointing triangle compared to the black dash-dot line in Figure 4). This discrepancy is likely due to imperfections in the birefringent crystals, which may have introduced a minor spatial displacement between the o-wave and e-wave polarization components. We verified that this increase in threshold with the splitting did not result in any loss in 2PEF signal (Figure 2a) or spatial resolution (see Table I in the Methods section. This suggests that the effect is not caused by significant degradation of the focus shape or peak intensity at the focal point. Such effects may only become evident at higher orders and not in 2PEF, which is beneficial as it raises the threshold for photodamage without compromising imaging performance.
Similar results were observed at 1070 wavelength. Both at 4 (Figure 4, red leftward-pointing triangle) and at (Figure 4, red rightward-pointing triangle) followed the same scaling law, indicating that nonlinear photodamage is independent of wavelength within this range. Additional details of measurements can be found in Table II in the Methods section. We were unable to measure for the configuration at 1070 due to a mean power limited to about 130 at the sample.
| Average pulse frequency () | Laser source | Laser frequency () | Splitting | Crystal thickness () | N sample size | mean () | standard deviation () | mean uncertainty () |
| 4 | ** | 4 | no | 0 | 5 | 77.9 | 8.9 | |
| 4 | * | 4 | no | 0 | 3 | 99.1 | 6.7 | |
| 4 | * | 2 | 20 | 4 | 104.1 | 15.1 | ||
| 8 | ** | 4 | 20 | 4 | 150.6 | 7.5 | ||
| 8 | * | 8 | no | 0 | 7 | 158.5 | 18.1 | |
| 8 | * | 4 | 20 | 5 | 172.0 | 20.5 | ||
| 8 | * | 4 | 10 | 4 | 183.1 | 33.8 | ||
| 8 | * | 2 | 9 | 233.3 | 16.2 | |||
| 16 | * | 4 | 5 | 417.2 | 38.9 |
Table II. Table of nonlinear photodamage threshold () measurements. The laser sources are (*) the Satsuma at 1030 nm and 2, 4 or 8 MHz and (**) the Opera at 1070 nm and 4 MHz
Together, these results demonstrate that adjusting the pulse distribution and peak intensity using birefringence-based passive pulse splitting can effectively modulate nonlinear photodamage. In addition, the nonlinear photodamage threshold behaves similarly whether pulses are evenly distributed or delivered in bursts.
optimizing in vivo 2P light-sheet imaging using birefringence-based pulse splitting
To demonstrate the potential of birefringence-based pulse splitting, we used the results presented above to optimize 2P light-sheet imaging using red fluorophores. 2P light-sheet microscopy is a powerful technique for capturing fast biological processes in vivo. However, 2PEF signal and imaging rate is limited by linear heating and nonlinear photodamage. Based on previous studies, we predicted live imaging of the beating heart in zebrafish embryos using mCherry labeling would be optimal at 1070 and with a pulse frequency in the range of 16 (see Methods section VIII A for a detailed rationale). We used birefringent-based pulse splitting to reach such frequency with a wavelength tunable laser source at 1070 and fixed 4 pulse frequency. Adjusting the wavelength is critical to enhance 2PEF signal and limit water absorption, while optimizing the laser pulse frequency limits high-order nonlinear photodamage.
We then conducted in vivo heart imaging experiments on zebrafish embryos labeled with mCherry protein, an average power at the sample of 110 at 1070 , and with a pulse frequency. Under these laser conditions, the corresponding laser peak intensity was in the order of 0.1 . Such illumination parameters enabled us to reach sufficient 2PEF signals to capture the fast dynamics of the atrio-ventricular canal at 1052 frame rate, corresponding to a pixel rate of 147 (700 (x) by 200 (y) pixels, bi-directional scanning), achieving volumetric acquisition of over 150 z-sections, with 2000 frames at each z-depth, as shown in Figure 5a and Supplementary Video 1. Using the same laser parameters, we also captured the beating heart in its entirety (700 (x) by 500 (y) pixels, forward scanning) at 465 frame rate, collecting volumetric data of over 150 z-sections, with 1000 frames at each z-depth, as shown in Figure 5b and Supplementary Video 1. tary Video 2. This corresponds to a pixel rate of . We note that it is well above the achievable speed of point-scanning microscopy, which requires at least one pulse per pixel and therefore a pixel rate below the laser pulse frequency. We also observe that photobleaching remained limited: after recording 2000 frames at frame rate and 1000 frames at frame rate, the 2PEF signal decreased by 34% and 18%, respectively. See Methods Section VIII G for the zebrafish preparation.

Figure 5. High-speed 2P light-sheet imaging of beating heart and neuronal activity in live zebrafish embryos. (a) 4D reconstruction of the atrio-ventricular canal in the beating heart of a 3-day post-fertilization zebrafish embryo labeled with Histone-mCherry, captured at 1 kHz frame rate ( average pulse frequency, mean power at sample, pixel rate). (b) 4D reconstruction of the entire beating heart of a 4-day post-fertilization zebrafish embryo labeled with Histone-mCherry, captured at 465 Hz frame rate ( average pulse frequency, mean power at sample, pixel rate). (c) in vivo calcium imaging in zebrafish embryo showing spontaneous activity using the jRGECO1b indicator ( average pulse frequency, mean power at sample), recorded at 465 Hz frame rate ( pixel rate). (d) Neuronal activity profiles () for selected regions of interest (ROIs) from (c), demonstrating detailed activity patterns at 465 Hz frame rate. The experiments (a-d) were launched under laser illumination at a wavelength of 1070 nm. Grid spacing 100 in (a,b). See Methods section VIII H for the image visualization details.
To quantify the typical signal level obtained during kHz frame rate imaging of the beating heart, we estimated the mean electron count per pixel within segmented cell nuclei which are in focus as a signal metric (see Methods Section VIII I for more details). Nuclear segmentation was performed independently in each frame using Cellpose [25], without spatial binning. Across the dataset, the mean signal was consistently in the range of 9 to 10 . Considering the quantum efficiency of the camera in the 600 nm wavelength range is about 80%, it corresponds to more than 10 . The background signal was measured to be less than 1 , resulting in an average signal-to-background ratio in the range of 10.
To demonstrate the potential of birefringence-based pulse splitting for high-speed 2P imaging in neurosciences, we performed in vivo calcium imaging using the jRGECO1b red indicator [26]. We captured images at frame rate ( pixel rate, 700 (x) by 500 (y) pixels, forward scanning) with an average power at the sample of 91 and a pulse frequency (Figure 5c). As shown in Figure 5d, spontaneous neuronal activity without extra stimulus was clearly identified at frame rate (see details of neuron activity data processing in Methods Section VIII J). There is a strong interest in neurosciences to capture neuronal activity at high-speed using red indicators, so that imaging and blue-shifted optogenetics strategies can be combined without cross-talks [27, 28].
Conclusion and discussion
In this work, we presented an efficient pulse splitting strategy using birefringent crystals to optimize the signal-to-photodamage ratio in 2P light-sheet microscopy, achieving pixel and kHz frame rates. Experimental evaluation of the 2PEF signal level, spatial resolution, photobleaching, and nonlinear photodamage with the pulse splitter validated its effectiveness in maintaining image quality while mitigating photodamage. Meanwhile, optimizing the excitation wavelength from to reduced the heating effect and enhanced the 2PEF signals from red fluorophores. It demonstrated the advantage of implementing such a pulse splitter with a wavelength-tunable laser source. Notably, photobleaching and nonlinear photodamage thresholds exhibited similar behavior whether the pulses were distributed evenly or delivered in bursts with 7-15 ps delays when using such a pulse splitter.
Regarding the implementation of pulse splitting, the approach based on cascaded birefringent crystals has several advantages. It is low-cost, loss-less, easy to align, compact, straightforward to incorporate into any pulsed illumination path and compatible with a large range of pulse frequencies. The nature of the crystal can be adapted to the wavelength used. Importantly, the splitting strategy is adjustable: a simple crystal rotation enables the user to bypass the splitting or adjust the number of splits. In case of pulse broadening, group-velocity dispersion can be pre-compensated. While this strategy can be applied to point-scanning 2P microscopy, similarly to other pulse splitters [8], the short delay between pulses, in the range of 7-15 ps in our case, can be a limitation if evenly distributed pulses is required [4]. Using the proposed pulse splitter with light-sheet microscopy, we successfully achieved 2P imaging at kHz frame rate in live zebrafish embryos to capture cell motion in the beating heart and calcium dynamics in the brain with subcellular resolution and using red fluorescent protein labels. We reached more than 150 pixel rate with, typically, 10 photons pixel, 100 mean power and 0.1 peak intensity at the sample. According to our results and previous report [14], it is significantly below the nonlinear photodamage threshold and corresponds to less than 1 °C sample heating. These performances compare very favorably with the characteristics reported for alternative fast multiphoton imaging schemes [1, 2, 3, 29]. Additionally, we note that the effective pixel rate in point-scanning multiphoton microscopy cannot exceed the laser pulse frequency, since at least one pulse is required per pixel [4, 21]. This limitation does not apply to 2P light-sheet illumination, in which a single pulse illuminates a line of pixels on the camera. In this study, we achieved a pixel rate one order of magnitude higher than the laser pulse frequency (up to 165 compared to 16 , respectively).
Due to fluorescent scattering, fast multiphoton methods with camera-based detection necessarily involve a compromise in imaging depth compared to single-point detector approaches [1, 3, 12, 13]. By contrast, multifocal (camera-based) detection, such as with micro-lens arrays [3], can achieve substantially higher imaging speeds when fully developed. However, this comes at the cost of increased mean illumination power. Among these methods, 2P light-sheet microscopy requires the least increase in mean power due to its orthogonal geometry, which enables excitation parallelization with a single beam [13]. In the present study, we demonstrate progress toward realizing this potential using pulse splitting. By investigating the limits imposed by the signal-to-photodamage trade-off, we reach optimal laser illumination in terms of wavelength, mean power, and peak intensity. The resulting strong, parallelized multiphoton excitation is compatible with high-speed live imaging.
Funding
This work was supported by Agence Nationale de la Recherche (ANR-EQPX-0029, ANR-18-EURE-0002, ANR-19-CE16-0019, ANR-19-CE11-0005, ANR-20-CE16-0026); Fondation Bettencourt Schueller (Biomedical Engineering Seed Grant Program); European Research Council (951330 HOPE).
Acknowledgments
We thank members of the Laboratory for Optics and Biosciences for discussions on microscopy, Jean-Marc Sintes for mechanical engineering, Emilie Menant and Isabelle Lamarre-Jouenne for zebrafish husbandry. We thank Volker Bormuth (Jean Perrin Laboratory, Paris) for providing the jRGECO1b line of Zebrafish.
Disclosures
The authors declare no conflict of interest.
Data availability
Data underlying the results presented in this paper are available from the corresponding author upon reasonable request.
Methods
Prediction of optimal laser parameters for in vivo 2P light-sheet imaging with a red fluorophore
Water-mediated sample heating has been previously reported as a limitation in 2P light-sheet microscopy [14, 15], and confirmed in this report (see section II C). Considering the absorption spectra of both water and mCherry [30] (Figure 6a), the optimal excitation wavelength for enhancing the 2PEF signal while minimizing thermal effects is . This wavelength coincides with a local minimum in water absorption, allowing for higher mean power without inducing excessive heating. Additionally, is near the peak of mCherry absorption, thereby maximizing the 2PEF signal. We note that compared to wavelength when using a pulse frequency tunable laser source as previously reported [14], increasing the wavelength by increases the 2PEF signal by a factor 2.2 and decreases water absorption by a factor 0.6. As a consequence, according to [14] and water absorption, the maximum mean power before reaching the thermal effect threshold (typically sample heating in our 2P light-sheet microscope) is typically at and at .
After selecting the excitation wavelength, the laser pulse frequency determines the mean power that can be used before reaching the nonlinear photodamage threshold. The optimal laser mean power at the sample must satisfy the conditions and , where is the mean power threshold at which nonlinear photodamage is observed, such as optical breakdown and cavitation bubble formation [14]. The scaling law established in [14], for a similar biological sample at shows that nonlinear photodamage has an order of and scales with the laser pulse frequency as follows:
We assume that does not vary between and . As previously reported [14, 15], decreasing the laser pulse frequency below the conventional value at constant mean power is a way to increase 2PEF signal by increasing the laser peak intensity while maintaining constant heating. However, this approach carries the risk of increased nonlinear photodamage and photobleaching. At a given the optimal laser pulse frequency is governed by Eq. (1) (see Figure 6b). Together, we predicted that at wavelength, a maximum mean power at the sample of can be used with an optimal pulse frequency of , as shown in Figure 6b.
YVO4 birefringent crystal: pulse delay and broadening calculation
In this work, we used and undoped Yttrium Orthovanadate () a-cut crystals with AR coating at (Casix Inc.). is a uniaxial birefringent crystal with a single optical axis (OA) governing its anisotropy. As long as the linear polarization of the laser pulse is not aligned with the crystal’s OA, the pulse is split into two temporally separated pulses: one pulse with a polarization perpendicular to the OA (ordinary ray) governed by the refractive index and the other, with a polarization parallel to the OA (extraordinary ray) governed by the refractive index . The delay between the two pulses travelling a distance through the crystal is then described as:
where refers to the speed of light. Here, we used a-cut crystals, with the OA parallel to the crystal surface and orthogonal to the illumination beam. In this case, the two co-aligned pulses travel the same distance , which is the thickness of the crystal. The refractive indices and are 2.1691 and 1.9592 at , respectively (2.1656 and 1.9575 at ) according to the Refractive Index database [24]. Thus, the pulse delay per crystal thickness is around at and at .
Pulse broadening between the input and output pulse duration ( and , respectively) after traveling a distance through the crystal and the output is estimated as:
where GVD refers to the Group Velocity Dispersion. For the ordinary ray, the GVD is 203.58 and 192.42 at and , respectively. For the extraordinary ray, the GVD is 288.7 and 272.62 at and 1070 [24], respectively. For instance, theoretically, an input pulse 230 will be broadened after 30 of crystal to 240 and 250 for ordinary and extraordinary pulses, respectively.
Optical setup
The described multiphoton light-sheet microscope incorporates two illumination paths and a single detection path. Illumination is provided by femtosecond lasers. We first used a Ytterbium fiber femtosecond laser (Satsuma, Amplitude Laser Group, France), which is tunable in laser frequency (or pulse repetition rate) at a fixed wavelength 1030 . We also used an optical parametric amplifier (Opera-HP pumped at 4 by a Monaco amplifier, Coherent) tuned at 1070 wavelength at a fixed 4 laser frequency. The laser output passes through a telescope system to adjust the beam diameter, producing a 1 collimated beam. The collimated beam then enters a pulse-splitting unit, which consists of two birefringent crystals, a half-wave plate, and a polarizer.
The collimated illumination light from the pulse-splitting unit is directed to an electro-optic modulator (EOM, Conoptics), which rapidly controls its transmission power. The laser beam is then reflected by a system of two galvanometric mirrors responsible for -axis and -axis scanning, mounted on a custom 2D translation stage. By adjusting this translation stage, the beam can be precisely translated at the back focal plane of the illumination objective to align the system. After reflection from the galvanometric mirrors, the beam passes through a 4f optical system consisting of lens A ( ) and lens B ( ), providing a 2X magnification. The beam then reaches the back focal plane of a 10X water-dipping illumination objective with a numerical aperture (NA) of 0.3 (N10XW-PF, Nikon). A second illumination objective provides white-light illumination, aiding in sample positioning and monitoring. These two opposing illumination objectives are mounted on independent linear stages, allowing precise axial adjustment of the illumination beam focus relative to the detection objective.
On the detection part, the 2P signal is collected by a 25X apochromatic water dipping objective with an NA 1.10 and a working distance of 2.0 (N25X-APO-MP, Nikon). The collected signal then passes through a 100 focal length tube lens (TTL100-A, Thorlabs) before being captured by a qCMOS camera (ORCA-Quest 2, Hamamatsu) with a pixel size of 4.6 . With this combination of detection objective, tube lens, and camera, we achieve an image pixel size of 0.37 . A low-pass filter (720/SP, BrightLine) is placed between the detection objective and the tube lens to block residual laser light. For 3D imaging, the detection objective is mounted on a piezo translation stage, which is synchronized with the -axis movement of the illumination beam controlled by the galvanometric mirror.
The imaging sample is embedded in agarose held by a glass capillary, which is positioned in a top-open chamber. A motorized stage (MP285, Sutter Instrument) controls the sample’s translation along the , , and axes, while a rotation stage enables rotation about the -axis. Notably, the sample remains stationary during image acquisition. All peripheral devices—including laser illumination power (regulated by the EOM), the shutter, galvanometric mirrors, piezo stage, camera, and motorized sample stage—are controlled and synchronized via custom LabVIEW (National Instruments) software. Image acquisition is performed at a maximum speed of 3.6 per pixel along the -axis. Imaging Modes: (i) Forward Scanning: 700 (x) by 500 (y) pixels at 465 frame rate. (ii) Bi-Directional Scanning: 700 (x) by 200 (y) pixels at 1052 frame rate. Notably, the bi-directional scanning mode significantly reduces the flyback time of the galvanometric mirror compared to forward scanning. Additionally, in both forward and bi-directional scanning modes, the camera detection line is synchronized with the illumination beam along the -axis.
The frame rate can be adjusted by modifying the line exposure time along the -axis. To acquire a sample plane, the light sheet must be precisely aligned with the camera’s focal plane. The alignment process involves first adjusting the illumination beam via a galvanometric mirror to bring it into the camera’s focal plane. Fine-tuning is then performed by adjusting the objective’s position using a piezo translation stage.
The light sheet is generated by scanning the Gaussian beam within the focal plane. During image acquisition, the light sheet is synchronized with the camera’s line exposure. For 3D data acquisition, the sample remains stationary while the imaging plane is adjusted by moving the objective and shifting the illumination plane using the galvanometric mirror.
PSF analysis
The point spread function measurement was performed in 1% low melting point agarose (Sigma Aldrich) containing fluorescent beads (Invitrogen TetraSpeck Microspheres, 0.2 T7280) diluted at a 1:1000. Stacks in different positions within the sample chamber were acquired using an isotropic pixel spacing (0.368 0.368 ). For illumination, a 1070 laser was used, and the emission was collected in the red spectrum. The FWHM in the plane is determined by fitting the intensity distribution in the corresponding planes with the Gaussian function as described above. The average axial FWHM is determined by averaging the quantifications of several beads in different depths within the mounting medium. The axial PSF quantification result is shown in Table I.
Photobleaching experiment
The photobleaching measurement is performed in zebrafish embryos expressing mCherry, which are imaged in the tail region to minimize artifacts caused by cell movement. Different areas were imaged under each condition. Photobleaching experiments were conducted at a constant fluorescent signal level by adjusting to maintain a constant , refers to the pulse frequency.
Because the illumination has a Gaussian profile in the x-direction, photobleaching is more pronounced at the centre of the field of view. Therefore, the central region of the image, 60 wide in the x-direction, is selected for analyzing the 2PEF signal decay. First, the cells are chosen by segmentation. Then, the experimental data are fitted by an exponential decay fit using a script written in Python. The photobleaching rate is then defined as the decay rate of an exponential fit:
Heart Beat Rate analysis
To investigate photodamage while imaging the zebrafish beating heart, we used the same experimental strategy as previously reported in [14, 15], using HBR as a metric to evaluate photodamage. In each HBR analysis experiment, the heart was imaged at 168 fps or 465 frame rate using white-light illumination, capturing a total of 22,000 images with a resolution of 700 500 pixels, with binning by a factor of 20. The initial 3,000 images (representing 18 seconds of acquisition) are used to establish the baseline heart beat rate (HBR). Following this, the femtosecond laser is activated for the next 7,600 images (from 18 to 63 seconds) at a specified mean power () and pulse frequency (), with the embryo illuminated in the same manner as during 2P light-sheet imaging. During this time, the change in HBR (HBR) is measured. For the final 11,400 frames (from 64 to 131 seconds), the laser was turned off, allowing the HBR to return to its baseline level, assuming the variation is reversible.
To calculate instantaneous HBR, custom Python scripts are employed. A fast Fourier transform (FFT) was applied to the temporal signal from each pixel over a 10-second time window. The 30 pixels with the highest signal-to-noise ratio in the FFT are then selected, based on the ratio of the FFT peak within the time window to the average FFT value, to determine the instantaneous HBR. The relative variation in HBR, denoted as HBR, is defined as the difference between the HBR after the laser is switched on and the baseline HBR (HBR = HBR - HBR), as shown in Figure 3a.
Zebrafish imaging
We used two zebrafish lines. To obtain an ubiquitous mCherry labeling of cell nuclei in embryos without iridophores and melanophores, we used casper crossed with Tg(ubi:H2B-mCherry) provided by the AMAGEN zebrafish facility (ama000049Tg on zfin.org). We also used the Tg(elavl3:jRGECO1b) zebrafish line originating from the Ahrens laboratory [26], which express jRGECO1b (red calcium reporter) in all neurons. Eggs from the Ecole Polytechnique zebrafish facility were raised in the dark at 28°C for three to five days. For imaging, embryos were anaesthetized with 0.001% (100 mg/L) Tricaine (MS-222, Sigma Aldrich) and embedded in 1% (10 g/L) low melting point agarose (Sigma Aldrich) as described [12]. Imaging was performed at room temperature (20-23°C) in the imaging chamber filled with 0.01% Tricaine solution. All experiments were performed with zebrafish embryos at 3-5 days post-fertilization before independent feeding larval forms, which complied with the European directive 2010/63/UE.
Image visualization
In Figs. 5a,b, 4D reconstruction was achieved using post-acquisition time synchronization especially for periodic motion, as previously described [31]. 3D rendering and manual heart segmentation (red cells) using Imaris. For the display of images in Fig. 5a (Fig. 5b), the pixel histogram was linearly stretched and gamma adjusted using the value of 2.28 for red cells and 1.1 for white cells (2.0 for red cells and 1.0 for white cells). For the display of Supplementary Video 1 (Supplementary Video 2), the pixel histograms were linearly stretched and gamma adjusted using values of 2.28 for red cells and 1.1 for white cells in 10-time binned images, and 1.2 for both red and white cells in the raw data (2.28 for red cells and 1.1 for white cells in 5-time binned images, and 1.74 for red cells and 1.72 for white cells in the raw data). All datasets were background-subtracted prior to processing.
Signal level quantification
To quantify the signal level, we measured the average electron count per pixel within individual in-focus cells [25]. To convert the recorded signal to electron counts, we first calibrated the gain of each pixel and subtracted the dark current [32]. The signal in each pixel was then divided by its gain to determine the electron count, which is explicitly calculated as
In this experiment, 2000 frames were captured in a single imaging plane at 1052 frame rate as shown in Figure 7a. We randomly selected several frames for cell nucleus segmentation as shown in Figure 7b. After segmenting the nuclei, we tracked the temporal variation in signal intensity for each cell throughout the frame sequence. The average electron count per pixel was calculated for each segmented cell at each time point. From one sample, we selected 3 planes at depth to quantify the signal level. Due to the periodic movement of the beating heart, the electron count per pixel exhibited a cyclic behavior, with signal intensity gradually decreasing from peak to peak due to photobleaching, as visible in Figure 7c. To accurately quantify signal intensity, we used the maximum value of the average electron count per pixel as a representative measure for each segmented cell, as shown in Figure 7c. The signal level quantification results of multiple representative cells are presented in Figure 7d. The electron count can be converted to a photon count considering the quantum efficiency, as described in Eq. (5),

Figure 7. Quantification of signal detection level in cells based on electron counts. (a) Captured images from a 2000-frame acquisition of the beating heart at a single plane in a 3-day post-fertilization zebrafish embryo labeled with histone mCherry, recorded at 1 kHz. (b) Example of cell nucleus segmentation from randomly selected frames. (c) Illustration of cyclic behavior in electron counts per pixel, attributed to the beating heart, derived from the segmented cells in (b). (d) Quantification results of electron counts from different zebrafish embryos, showing the median and the first and third quartiles. represents the number of selected cells and indicates the mean value.
where refers to the typical quantum efficiency at wavelength .
Neuronal activity data processing
The temporal profile of neuronal activity is computed by spatially averaging the jRGECO1b signal within a selection of ROIs (see in Figure 5c). The normalized profile is then obtained using:
where baseline is defined as the first decile value of , computed along the temporal dimension using a rolling filter with a 17.2 -time window, and is the nearest lower integer of the lowest value [3].
References
References
- [1]J. Wu, N. Ji, and K. K. Tsia, Speed scaling in multiphoton fluorescence microscopy, Nature Photonics 15, 800 (2021).DOI
- [2]J. Wu, Y. Liang, S. Chen, C.-L. Hsu, M. Chavarha, S. W. Evans, D. Shi, M. Z. Lin, K. K. Tsia, and N. Ji, Kilohertz two-photon fluorescence microscopy imaging of neural activity in vivo, Nature Methods 17, 287 (2020).DOI
- [3]T. Zhang, O. Hernandez, R. Chrapkiewicz, A. Shai, M. J. Wagner, Y. Zhang, C.-H. Wu, J. Z. Li, M. Inoue, Y. Gong, B. Ahanonu, H. Zeng, H. Bito, and M. J. Schnitzer, Kilohertz two-photon brain imaging in awake mice, Nature Methods 16, 1119 (2019).DOI
- [4]S. Xiao, J. T. Giblin, D. A. Boas, and J. Mertz, High-throughput deep tissue two-photon microscopy at kilohertz frame rates, Optica 10, 763 (2023).
- [5]K. Podgorski and G. Ranganathan, Brain heating induced by near-infrared lasers during multiphoton microscopy, Journal of Neurophysiology 116, 1012 (2016).DOI
- [6]K. Charan, B. Li, M. Wang, C. P. Lin, and C. Xu, Fiber-based tunable repetition rate source for deep tissue two-photon fluorescence microscopy, Biomedical Optics Express 9, 2304 (2018).DOI
- [7]T. Wang and C. Xu, Three-photon neuronal imaging in deep mouse brain, Optica 7, 947 (2020).DOI
- [8]N. Ji, J. C. Magee, and E. Betzig, High-speed, low-photodamage nonlinear imaging using passive pulse splitters, Nature Methods 5, 197 (2008).DOI
- [9]A. Hopt and E. Neher, Highly Nonlinear Photodamage in Two-Photon Fluorescence Microscopy, Biophysical Journal 80, 2029 (2001).pubmed.ncbi.nlm.nih.gov/11259316
- [10]P. Luu, S. E. Fraser, and F. Schneider, More than double the fun with two-photon excitation microscopy, Communications biology 7, 364 (2024).DOI
- [11]J. Demas, J. Manley, F. Tejera, K. Barber, H. Kim, F. M. Traub, B. Chen, and A. Vaziri, High-speed, cortex-wide volumetric recording of neuroactivity at cellular resolution using light beads microscopy, Nature Methods 18, 1103 (2021).DOI
- [12]T. V. Truong, W. Supatto, D. S. Koos, J. M. Choi, and S. E. Fraser, Deep and fast live imaging with two-photon scanned light-sheet microscopy, Nature Methods 8, 757 (2011).DOI
- [13]W. Supatto, T. V. Truong, D. Débarre, and E. Beaurepaire, Advances in multiphoton microscopy for imaging embryos, Current Opinion in Genetics & Development Developmental Mechanisms, patterning and evolution, 21, 538 (2011).DOI
- [14]V. Maioli, A. Boniface, P. Mahou, J. F. Ortas, L. Abdeladim, E. Beaurepaire, and W. Supatto, Fast in vivo multiphoton light-sheet microscopy with optimal pulse frequency, Biomedical Optics Express 11, 6012 (2020).
- [15]F. M. Gasparoli, A. Escobet-Montalbán, J. Early, G. D. Bruce, and K. Dholakia, Is laser repetition rate important for two-photon light sheet microscopy?, OSA Continuum 3, 2935 (2020).DOI
- [16]H. E. Bates, R. R. Alfano, and N. Schiller, Picosecond pulse stacking in calcite, Applied Optics 18, 947 (1979).DOI
- [17]S. Zhou, D. Ouzounov, H. Li, I. Bazarov, B. Dunham, C. Sinclair, and F. W. Wise, Efficient temporal shaping of ultrashort pulses with birefringent crystals, Applied Optics 46, 8488 (2007).
- [18]S. Zhou, F. W. Wise, and D. G. Ouzounov, Derived-pulse amplification of ultrashort pulses, Optics Letters 32, 871 (2007).
- [19]L. Daniault, M. Hanna, D. N. Papadopoulos, Y. Zaouter, E. Mottay, F. Druon, and P. Georges, High peak-power stretcher-free femtosecond fiber amplifier using passive spatio-temporal coherent combining, Optics Express 20, 21627 (2012).DOI
- [20]D. Stachowiak, J. Bogusławski, A. Gluszek, Z. Laszczych, M. Wojtkowski, and G. Soboń, Frequency-doubled femtosecond Er-doped fiber laser for two-photon excited fluorescence imaging, Biomedical Optics Express 11, 4431 (2020).DOI
- [21]J. Song, J. Kang, U. Kang, H. S. Nam, H. J. Kim, R. H. Kim, J. W. Kim, and H. Yoo, SNR enhanced high-speed two-photon microscopy using a pulse picker and time gating detection, Scientific Reports 13, 14244 (2023).DOI
- [22]S. A. Engelmann, A. Tomar, A. L. Woods, and A. K. Dunn, Pulse train gating to improve signal generation for in vivo two-photon fluorescence microscopy, Neurophotonics 10, 045006 (2023).DOI
- [23]G. d. Vito, P. Ricci, L. Turrini, V. Gavryusev, C. Müllenbroich, N. Tiso, F. Vanzi, L. Silvestri, and F. S. Pavone, Effects of excitation light polarization on fluorescence emission in two-photon light-sheet microscopy, Biomedical Optics Express 11, 4651 (2020).DOI
- [24]M. N. Polyanskiy, Refractiveindex.info database of optical constants, Scientific Data 11, 94 (2024).DOI
- [25]M. Pachitariu and C. Stringer, Cellpose 2.0: how to train your own model, Nature Methods 19, 1634 (2022).DOI
- [26]Y. Mu, D. V. Bennett, M. Rubinov, S. Narayan, C.-T. Yang, M. Tanimoto, B. D. Mensh, L. L. Looger, and M. B. Ahrens, Glia Accumulate Evidence that Actions Are Futile and Suppress Unsuccessful Behavior, Cell 178, 27 (2019).DOI
- [27]H. Dana, B. Mohar, Y. Sun, S. Narayan, A. Gordus, J. P. Hasseman, G. Tsegae, G. T. Holt, A. Hu, D. Walpita, R. Patel, J. J. Macklin, C. I. Bargmann, M. B. Ahrens, E. R. Schreiter, V. Jayaraman, L. L. Looger, K. Svoboda, and D. S. Kim, Sensitive red protein calcium indicators for imaging neural activity, eLife 5, e12727 (2016).DOI
- [28]M. Kannan, G. Vasan, C. Huang, S. Haziza, J. Z. Li, H. Inan, M. J. Schnitzer, and V. A. Pieribone, Fast, in vivo voltage imaging using a red fluorescent indicator, Nature Methods 15, 1108 (2018).DOI
- [29]A. Kazemipour, O. Novak, D. Flickinger, J. S. Marvin, A. S. Abdelfattah, J. King, P. M. Borden, J. J. Kim, S. H. Al-Abdullatif, P. E. Deal, E. W. Miller, E. R. Schreiter, S. Druckmann, K. Svoboda, L. L. Looger, and K. Podgorski, Kilohertz frame-rate two-photon tomography, Nature Methods 16, 778 (2019).
- [30]N. C. Shaner, R. E. Campbell, P. A. Steinbach, B. N. G. Giepmans, A. E. Palmer, and R. Y. Tsien, Improved monomeric red, orange and yellow fluorescent proteins derived from discosoma sp. red fluorescent protein, Nature Biotechnology 22, 1567 (2004).DOI
- [31]M. Liebling, A. S. Forouhar, M. Gharib, S. E. Fraser, and M. E. Dickinson, Four-dimensional cardiac imaging in living embryos via postacquisition synchronization of nongated slice sequences, Journal of Biomedical Optics 10, 054001 (2005).pubmed.ncbi.nlm.nih.gov/16292961
- [32]B. Mandracchia, X. Hua, C. Guo, J. Son, T. Urner, and S. Jia, Fast and accurate sCMOS noise correction for fluorescence microscopy, Nature Communications 11, 94 (2020).DOI
