Simultaneous three-photon and optical coherence microscopy deep within an intact mouse brain

Abstract

Multimodal microscopy combining various imaging approaches can provide complementary information about tissue in a single imaging session. Here, we demonstrate a multimodal approach combining three-photon microscopy (3PM) and spectral-domain optical coherence microscopy (SD-OCM). We show that an optical parametric amplifier (OPA) laser source, which is the standard source for three-photon (3P) fluorescence excitation and third harmonic generation (THG), can be used for simultaneous optical coherence microscopy (OCM), 3P fluorescence and THG imaging. We demonstrate the system performance in deep mouse brains in vivo with an OPA source operating at 1620 nm center wavelength. We visualized small structures such as myelinated axons, neurons, and large fiber tracts in white matter with high spatial resolution noninvasively using linear and nonlinear contrast mechanisms at depth exceeding 1 mm in the intact adult mouse brain. Our results show that simultaneous OCM and 3PM at the long wavelength window can be conveniently combined for deep tissue imaging in vivo.

Introduction

Optical microscopy is an essential tool for noninvasive monitoring of biological structures and functions in vivo. Due to light scattering and absorption by biological tissue, it becomes challenging to obtain images with high contrast and high resolution in deep tissue, and the imaging depth in the mammalian brain is ultimately limited by the scattering of light despite the progress over the last three decades^1–3. Because of the three-dimensionally confined excitation volume and relatively long excitation wavelength, two-photon (2P) microscopy achieved deep brain imaging at a depth of >500 μm in the adult mouse brain. Using 1280 nm excitation, a penetration depth of up to 1.6 mm can be achieved in vivo for imaging vasculature in the mouse neocortex with 2P excitation⁴. By carefully investigating the combined effects of both scattering and absorption, the spectral windows of 1.3 μm and 1.7 μm^5,6 have been shown to be optimum for deep brain imaging⁷, and long wavelength 3PM of neuronal structure and activity in the subcortical areas at >1-mm depth in vivo was demonstrated within these two wavelength windows. With specific fluorescence labels in different cell types, the structure and function of the cells, as well as cellular and sub-cellular dynamics, can be imaged quantitatively via multiphoton fluorescence microscopy. In parallel to the development of multiphoton fluorescence microscopy, a variety of approaches based on label-free contrast mechanisms, including reflectance confocal⁸, spectral reflectance confocal^9,10, OCM^11–14, multi-harmonic generation^15,16, Raman scattering^17,18, and photoacoustic microscopy¹⁹ have been developed in the last two to three decades. These modalities provide complementary information to multiphoton fluorescence microscopy.

Optical coherence tomography (OCT) combines coherence gating and spatial confocal gating, providing fast, non-invasive, and label-free imaging in deep within scattering tissues_. OCM performs higher numerical aperture (NA) imaging than OCT, and can provide cellular-resolution, label-free imaging of deep-brain structures^12,20. OCM can achieve a larger imaging depth than confocal microscopy by better rejecting multiply-scattered and out-of-focus light²¹. By using a laser source in the 1.7 μm window, an imaging depth >1.2 mm in mouse brains^11,12,14 with OCT and OCM have been achieved.

Multimodal optical imaging has shown promise to integrate complementary structural and functional information from different biological constituents and tissue types. Several groups have reported the combination of laser scanning optical coherence and 2P fluorescence microscopy^22–24 for tissue imaging including skin²⁵, eye^26,27, brain²⁸ as well as human carotid atheroma²⁹. The integration of multiphoton and OCM has applications for in vivo diagnosis, such as in various pathological skin conditions and the related pharmaceutical as well as cosmetic research. The integration of combines 2P fluorescence microscopy, second and THG microscopy, and optical resolution photoacoustic microscopy can provide access to complementary information to support investigations of the constituents of and their mutual interactions within tissue²⁹. Since carotid artery atherosclerosis is a main cause of stroke, understanding atherosclerosis biology is significant to develop targeted prevention and related treatment strategies, and this is driving the demand for multimodal imaging tools to investigate atheroma pathology with different contrasts.

Other multimodal imaging approaches, for example combining bioluminescence imaging and 2P fluorescence, can delineate the boundary and quantitatively analyze glioblastoma multiforme (GBM), and guide laser ablation surgery and theranostics³⁰ in animal in vivo. Multimodal imaging combining OCT and fluorescence³¹ can enable precise diagnosis of tumors in neurosurgery by providing tissue functional and structural information for surgeons. In particular, sodium fluorescein-guided resection of high-grade glioma (HGG) under “yellow fluorescence” mode has shown great clinical potential to help doctors distinguish tumors from peripheral brain tissues, to facilitate complete resection of gliomas with high precision^32,33. A multiple-contrast deep tissue imaging approach that can provide fast and accurate intraoperative information has potential applications to guide the resection of brain tumors, such as high-grade glioma³⁴, GBM³⁵ and cerebral metastases³³.

The depth penetration capability of conventional 2P microscopy (2PM) is limited^36–38. More importantly, the imaging depth of 2PM is typically well short of that achievable by long-wavelength OCM. Consequently, the full potential of multimodal imaging based on fluorescence microscopy and OCM has not been realized. Since 3PM permits deeper imaging than 2PM and naturally requires a long excitation wavelength, long-wavelength 3PM is a better choice than 2PM to combine with OCM for multimodal imaging in the deeper regions. An additional advantage of long-wavelength 3PM is the accompanying label-free signal of THG. Indeed, THG and 3PM have been combined routinely in previous works^5,6,39 for vasculature and neuronal imaging in the deep brain regions. Furthermore, OCM can provide complimentary information to multiphoton fluorescence imaging, including depth-resolved imaging of tissue structure, vascular networks⁴⁰, and refractive index⁴¹.

Here, we present a novel multimodal imaging system that integrates 3PM, OCM, and THG within a single multimodal laser scanning microscope. We use a single optical parametric amplifier (OPA) laser source, which is typically used only for 3PM, for both 3PM and spectral-domain OCM (SD-OCM). Use of a single source allows for simultaneous multimodal acquisition, which simplifies the system design as well as the co-registration of label-free structural (OCM and THG) and molecularly specific (3P fluorescence) information. Furthermore, by integrating an interferometer for the detection of the backscattered light for 3P excitation, we obtain high-resolution, depth-sectioned OCM images without additional light exposure and acquisition time. Extending beyond previous works that combined second harmonic generation (SHG) and 2P microscopy for shallow-layer imaging with complementary information, our approach provides valuable insights into deeper tissue layers. Our results facilitate a comparative analysis of each imaging modality, which is important for understanding their respective strengths and limitations for depth-brain microscopy.

Results

Multimodal microscopy setup combining OCM, THG, and 3P imaging

A key goal for multimodal microscopy is to enable the acquisition of high-quality co-registered multimodal images for each modality. The multimodal SD-OCM, 3PM and THG setup presented here is based on a typical multiphoton microscope, where we combined these modalities by sharing the same laser source, scanning unit, and imaging optics.

Figure 1 illustrates the multimodal microscopy setup based on a conventional laser scanning microscope design, where SD-OCM, THG, and 3P signals are collected in the epi-direction simultaneously. The major optical components of the multimodal microscopy setup are shown as two modules: the SD-OCM module and the 3P/THG module. The common source for the system is an optical parametric amplifier (OPA, Opera-F, Coherent) pumped by a 1035 nm femtosecond laser (Monaco, Coherent Inc.) and generates tunable wavelengths via its idler beam, ranging from 1200 nm to 2500 nm. In our imaging experiments, we used the idler beam’s output wavelength, which was fixed at 1620 nm. A silicon wafer was employed to compensate for the dispersion of the optics in the laser source and the microscope, including the objective⁴² at the long excitation wavelength 1700 nm optical window. The laser repetition rate was maintained at 1 MHz for imaging. We set the center wavelength at 1620 nm, and the spectrum of the laser output is shown in the inset of Fig. 1 (measured by a spectrometer, OSA207C, Thorlabs). The images were taken with a high-NA objective (Olympus XLPLN25XWMP2, 25×, NA 1.05), as shown in Fig. 1. The diameter of the illumination beam was measured to be 18.4 mm (1/e² full width), which overfilled the back-aperture of the objective (~15 mm). For the SD-OCM setup, the backscattered light was de-scanned, refocused by an aspherical lens (A110TM-C, Thorlabs) into the single-mode fiber (SMF-28, Thorlabs), and detected by an InGaAs linescan spectrometer (Wasatch Photonics, Cobra). In the OCM path, the reference arm and single-mode fiber-connected spectrometer were set up before the Si wafer (Silicon wafer). The Si wafer was used for dispersion compensation at 1620 nm to maintain a transform-limited pulse after the objective (Fig. 1). All the images were simultaneously acquired by a multifunctional data acquisition card (NI PCI 6251, National Instruments) and a Camera Link image acquisition board (NI PCI-1424, National Instruments). The sensitivity of the OCM system was measured to be 96 dB using the characterization methods previously described⁴³, with 2.1 mW average power incident on the sample and 1 MHz laser repetition rate. For the 3 PM setup, two photomultiplier tubes (PMTs) with GaAsP photocathodes (H7422-40, Hamamatsu Photonics) were employed for the detection of the 3P fluorescence and THG signal. Transimpedance amplifiers (C7319, Hamamatsu) were used to convert the PMT current to voltage with the gain 10⁶ V/A for the fluorescence channel and 10⁵ V/A for the THG channel. All images from THG, 3PM and OCM were acquired by a custom LabVIEW software (National Instruments).

Fig. 1. Setup of the multimodal microscopy system.

The multimodal imaging system includes two paths, the OCM path and the THG/3P fluorescence path. The three modalities, OCM/ THG/3P, share the same laser source, the scanning unit and the imaging optics. The THG/3P path employed two PMTs (PMT1 photomultiplier tube to measure 3P fluorescence, PMT2 photomultiplier tube to measure THG), two dichroic mirrors (DM2 dichroic mirror to separate 3P excitation beam from emission beam, DM1 dichroic mirror to separate THG and 3P signal), and bandpass filters on each path to further select the signals. The scanning unit and signal collection system (two PMTs for THG and 3P, InGaAs line CCD-based high-speed spectrometer for OCM) are synchronized for signal collection under computer control. CPA: Chirped Pulse Amplification (Coherent Inc.). OPA optical parametric amplifier (Opera-F, Coherent Inc.). BS beam splitter. PBS polarizing beam splitter. PMT photomultiplier tubes. Lens focal length: 150 mm for L1–L3, 750 mm for L4, and 6.24 mm for L5. The input port of single-mode fiber (SMF) of the fiber-coupled spectrometer acts as the confocal “pinhole”, which is optically conjugate to the objective’s focal plane.

For deep brain imaging, we designed the multimodal microscopy setup to maximize the delivery of illumination photons to the sample and the throughput of the backscattered photons for OCM. Although higher excitation power is needed for imaging deeper with 3PM, the OCM reference arm (due to its high reflectivity) requires a low power from the laser to interfere with the backscattered photons and generate an adequate spectrometer signal. Consequently, a 1:99 beam splitter (thick glass slide) was employed to split the illumination power to the sample arm and the reference arm so that nearly all of the laser power was available for multiphoton excitation. Another 92:8 pellicle beam splitter (BP108, Thorlabs) preserves the signal strength of the backscattered photons for SD-OCM when re-combining the reference arm and the signal arm⁴⁴. A λ/2 wave-plate and a PBS controlled the sample arm power, and a combination of λ/4 wave-plate as the mouse cranial window and λ/2 wave-plate in the sample arm allowed the optimization for achieving high throughput for both the illumination and the backscattered photons at the PBS⁸ (see Methods for details in mouse cranial window preparation).

Co-registration of images from different modalities is required to obtain high-resolution complementary information from multimodal imaging. In our system, OCM, THG, and 3PM shared the same laser source, scanning unit, and imaging optics in the illumination path, and the multimodal images were acquired simultaneously, which ensured co-registration of the lateral positions of all three modalities. For the axial co-registration, the THG focus was first determined by maximizing the signal at the interface of the cover glass and the water immersion layer. The confocal pinhole (i.e., the fiber input port) position was then adjusted to maximize the OCM signal. This process ensured that the THG focal plane was conjugated to the OCM collection pinhole. In Supplementary Fig. 1, the cross-section of the cover glass is shown, which confirms the co-registration of OCM and THG.

Typical OCM uses telecentric scanning (Supplementary Fig. 2a) to achieve a flat coherence gate and uniform magnification across the transverse-scanned field-of-view (FOV), at the cost of underfilling the objective lens. In our multimodal system, since our objective aperture is fully filled, a non-telecentric scanning scheme (see Supplementary Fig. 2b) was used to implement transverse beam scanning. This maximizes illumination intensity for 3P fluorescence and THG and optimizes transverse resolution for all three modalities. However, the non-telecentric scanning, which is routinely used in THG and 3P imaging, results in a coherence gate curvature in OCM because the focal plane of the objective does not correspond to the surface defined by the same optical path length⁴⁵. Additionally, residual spherical aberration in the optical system of the sample arm (in particular the telescope lenses used for beam expansion) can lead to further variation of focal depth with transverse scan position. These effects can be particularly accounted for in the OCM image by estimating the focal plane curvature and correcting it by fitting the curvature with parabolic function to estimate the applied correction for a flat FOV in post-processing (see Supplementary Fig. 2c–e).

In order to get simultaneous, high-resolution, co-registered information from each image voxel, closely matched spatial resolution for each modality is needed. Here, we summarize the theoretical expectations and present the experimental characterization of our system resolution by measuring small features in the mouse brain in vivo.

In our system, the OPA is tuned to a center wavelength of 1620 nm with a full-width-half maximum (FWHM) bandwidth of ~70 nm, resulting in a theoretical coherence length of ~11 μm. Since this is much larger than the theoretical confocal axial resolution (FWHM) (~1.90 μm) associated with our high-NA (NA = 1.05) objective, the SD-OCM axial resolution is determined by the objective lens NA and corresponding confocal detection. The theoretical FWHM axial resolution of our 3P imaging system is about 1.6 μm. The close match of theoretical voxel dimensions across the imaging modalities helps to facilitate our procedure to achieve co-registration of 3PM and OCM images in the axial direction (see Image Processing & Co-registration for further details), and to ensure that the information in the voxels of the three imaging modalities were collected from approximately the same sample volume.

For all the imaging modalities, we used the smallest features (mainly the small capillaries and the axons) we could identify at each depth to estimate the resolution of the system (Fig. 2). Figure 2h–k show that a high spatial resolution was achieved with all the three modalities with lateral resolutions ~1 μm and axial resolutions ~2.7 μm at the depth of ~780 μm. We also measured the lateral FWHM of the small features at various depths and observed a gradual degradation of the lateral resolution as the imaging depth increases (Fig. 2i). A comparative summary of SD-OCM, THG, and 3PM, including lateral and axial resolution, imaging frame rate, maximum illumination power, and maximum imaging depth, is provided in Supplementary Table 1 (Modality Comparison).

Fig. 2. Resolution characterization of the multimodal imaging system in deep adult mouse brain in vivo.

a–c OCM, THG, and 3P fluorescence imaging of mouse brain between 778 μm and 790 μm. The depth is coded by color. e–g Enlarged images of the boxes in (a–c) in three different projections. h-j The line profiles along the green and red arrows in (e–g). k Boxplot of the lateral and axial resolution (N = 5). l Boxplot of the lateral resolutions of SD-OCM, THG and 3-photon imaging measured at various imaging depths. Scale bars = 50 μm.

Simultaneous in vivo multimodal imaging of mouse brain

In vivo simultaneous multimodal imaging of the mouse brain was performed to demonstrate the deep tissue imaging capability. Figure 4a shows the optical sections in mouse brain in vivo at various depths. The incident beam power corresponding to different imaging depths was adjusted to achieve similar signal strength for all depths less than 1 mm. The incident power after the objective was measured and recorded for each depth. At depths beyond 1 mm, the illumination beam power right after the objective was further increased and measured to be up to a maximum average power of 55 mW. The time for acquiring XY frames was kept to a constant 3.9 s (515 × 512 pixels) for all acquired images. The current imaging speed is primarily limited by the galvanometer scanning rate and the data acquisition bandwidth of the spectrometer.

Fig. 4. Multi-contrast imaging of neurons, myelinated axons and small blood vessels.

a Multimodal imaging at a depth of 784 µm. b–d OCM, THG, and 3PM images of the boxed region in (a). e, f Enlarged views of the green boxes in (b, c), respectively. g Line profiles in dashed black and magenta corresponding to the lines in (e, f), respectively. h THG signal intensity (a.u.) of horizontal and vertical myelin in (c). i, j Small blood vessels visualization and analysis. Left, enlarged views of the yellow boxes in (b, c), respectively; right, line profile corresponding to dashed lines in the left panels. Scale bars in (a, c): 50 µm. Scale bars in (e, f, i, j): 20 µm.

The three-photon excited fluorescence and THG signal within the focal volume is mostly generated by ballistic photons, while the collected signal includes both ballistic photons and multiply-scattered photons in THG and 3PM. From Fig. 3, it is possible to estimate the effective attenuation length (EAL), a parameter combining the scattering and absorption length of the tissue, for 1620 nm imaging in mouse brain tissue in vivo. For THG and 3PM, we take the average values of the blood vessels and neuron soma, respectively, as the THG and 3P signals, and we estimate the EAL as the depth where the signal attenuates by 1/e³. Note that the signals for EAL measurement are all normalized by the measured incident power after the objective. For 3-photon/THG signal, it is normalized by the cube of the measured illumination power after objective and for OCM signal, it is normalized by the measured illumination power after objective. Figure 4b shows the detected THG signal (in purple) as a function of imaging depth. The EAL is determined to be 389 μm between 60 μm and 800 μm depth in the neocortex of the mouse brain. The EAL is ~230 μm between 800 μm and 1000 μm depth in the external capsule. From the data of 3P fluorescence signal versus depth (in green, mostly from neuron somas), the EAL between 60 μm and 800 μm depth is approximately 392 μm.

Fig. 3. Simultaneous multimodal imaging of the mouse brain in vivo.

a 3D stack of simultaneous multimodal imaging of the mouse brain in vivo. Units of depth axis are in microns (µm). b THG and 3-photon signal attenuation curves of the neocortex (N.C.) and external capsule (E.C.) of mouse brain in vivo. The signal is normalized by the cube of the measured illumination power after objective and plotted on a log scale. c The signal attenuation curve (log scale) and noise floor in SD-OCM are shown in green and red, respectively. For OCM signal decay curve, the magnitude squared of the signal (|S|²) taken from the features of blood vessels is normalized by the incident sample power P_s (i.e., |S|² / P_s is shown). Noise (|S_noise|²) from OCM data is quantified from the image frame at position that is away from the focus, and is also normalized by the measured illumination power after objective at each depth (i.e., |S_noise|² / P_s is shown). Scale bars = 50 μm.

The signal generation and collection within the focal volume in SD-OCM are both dominated by ballistic photons. Therefore, ballistic photon attenuations in both the illumination and collection paths (from the double-pass detection geometry) need to be taken into count in the SD-OCM case when estimating the EAL in the mouse brain using OCM signal. For the SD-OCM data acquisition, the reference arm position was manually adjusted for each focus depth to position the focus at a fixed optical path delay leveraging live 2D cross-sectional OCT imaging, in order to avoid the sensitivity roll-off from the spectrometer^46,47 impacting the estimate of EAL. We measured the OCM signal as a function of depth, where the value at each depth is determined by averaging signals from multiple blood vessels at that same depth (Fig. 3c) and compared it with the en face frame-averaged signal decay in Supplementary Fig. 3. We found that the signal from a specific feature (e.g., from multiple blood vessels) versus depth provides an estimate of the EAL closer to that quantified from THG signal than the frame-averaged signal, which includes all features within each en face image, but nevertheless is highly dependent on the brightest features at each depth. The OCM noise floor was estimated from the average intensity in the out-of-focus depths. Note that the OCM noise floor is flat because the power in the reference arm was kept at a constant level and was much higher than the signal power from the sample arm at all depths.

Multimodal contrast for deep mouse brain imaging in vivo

Multimodal cellular imaging combining 3PM, THG, and SD-OCM can provide multiple unique and complementary contrasts in the mouse brain in vivo. Here, we investigate the similarities and differences of these contrasts.

First, we present multimodal imaging of neurons and myelinated axons. Figure 4 shows 3P, THG, and OCM images simultaneously obtained in the deep cortical layers at the depth of 784 µm, and illustrates the complementary information with both labeled and label-free contrast, such as the neurons in the 3P fluorescence channel and myelinated axons in the THG and OCM channels. 3PM with fluorescence labeling strategies such as genetic targeting provides high-specificity information that is not accessible with OCM and THG imaging. In Supplementary Fig. 4, neurons show up as bright signals in 3PM, while they were not visible in the OCM and THG images. For OCM, the main source of image contrast comes from spatial variations in the linear refractive index. In the mouse brain, myelinated axons have a high refractive index^9,48 due to the high lipid content and exhibit a high scattering contrast. For the THG signal, it is sensitive to the nonlinear refractive index variations and the geometry of the features¹⁶. In Fig. 4c, the interface of myelinated axons and their surroundings can be visualized by THG imaging but not OCM. As shown in Fig. 4g, the THG signal of the myelinated axons is the strongest from the sides of the myelin tube and exhibits two peaks, while the OCM signal is mainly generated from the top and bottom of the myelin tube and has a single peak. This observation is due to the strong index mismatch at the two radial boundaries of myelin, resulting in THG signals from the two sides of the myelinated axons yielding two peaks in cross-section (Fig. 4c, g)^49,50. On the other hand, OCM’s reflection-based contrast, enhanced at surfaces (with surface normal that is) approximately aligned with the optical beam axis, typically shows a single strong peak from top/bottom interfaces (Fig. 4b, g)⁵¹. In Fig. 4g, the distance between the two peaks in the THG signal line profile (in magenta) is 1.9 µm, which corresponds to the diameter of the myelin. The FWHM of the Gaussian fit for the right peak of the THG signal line profile is 1.1 µm, highlighting the capability of THG to directly identify fine features in myelin. The FWHM of the OCM signal line profile (dashed black line) is 1.29 µm. Both horizontal and vertical interfaces in the focal volume will elicit a strong THG signal, but as previously noted for myelin fibers^52,53, the vertical interfaces (green arrows in Fig. 4c) generate a higher THG signal than the horizontal interfaces (yellow arrows in Fig. 4c) (see Fig. 4h). The vertically oriented myelin fibers are generally not observed in the OCM images.

We found that OCM images of myelin show patchy, fragmented structures along the myelinated axons (see the myelin in the center of Fig. 4e). This can be attributed to self-interference between the top and bottom interfaces of myelin, similar to thin-film interference⁵⁴. This distinctive pattern may help with identifying myelin in the deep mouse brain in vivo.

In Fig. 3, at ~900 µm, THG and OCM images report similar structures, but by ~950 µm OCM and THG provide complementary information on white matter structures, while fluorescence contrast declines due to limited RFP brightness/expression and stronger scattering. THG also decreases at depths beyond white matter because of its mostly forward-directed generation combined with multiple scattering in white matter. Deep-tissue performance can be enhanced by dispersion pre-compensation and adaptive optics, and by using brighter, red-shifted probes for enhanced signal.

Next, we present multimodal imaging of blood vessels. For small blood vessels, the dark streak or bright line patterns of the flowing red blood cells (RBCs) in vessels have previously been revealed by THG imaging at 1700 nm⁵⁵ and 1300 nm⁵⁶. Here, fine stripe patterns show up in both THG and OCM signal (RBCs, the dotted lines in Fig. 4i, j). Within a single-frame acquisition, compared with the patchy structures observed in OCM data from static myelin, the fine strip patterns generated by fast flowing RBCs tend to appear less structured. This is due to their dynamic nature, which can produce varying patterns over a short distance within a single blood vessel. The blood vessels are not fluorescently labeled, and so appear as dark shadows in 3P fluorescence images (Fig. 4d).

Figure 5a shows a three-dimensional reconstruction of large horizontal blood vessels in the shallow regions of the mouse brain by OCM and THG imaging. In both SD-OCM and THG images, we can visualize the intensity fluctuations across the blood vessel caused by the motion blur of moving RBCs (Fig. 5b, c). In the OCM image, the middle of the blood vessel generated a stronger signal than the two sides, whereas THG signal levels were similar for both center and sides. This leads to the FWHM of OCM signal having a smaller than the THG signal of the same blood vessel (Fig. 5e). We also observed differences in signal from blood vessels of different sizes (see Supplementary Figs. 5 and 6). Vertical blood vessels appear dark compared to the surroundings in OCM (Fig. 5f and Supplementary Fig. 7a); however, the corresponding THG signal is stronger due to the contributions from the RBCs. The walls of blood vessels, both in vertical and horizontal orientations, generate strong THG signals, while producing weak or no detectable OCM signals (Fig. 5b, f). The vessel wall thickness, as estimated from the THG image (the right panels in Fig. 5f and Supplementary Fig. 7), is ~2 µm. This is consistent with the vessel wall thickness of arterioles or venules in the mouse brain⁵⁷.

Fig. 5. Multi-contrast imaging of large blood vessels in mouse brain by multimodal imaging.

a Three-dimensional reconstruction of multimodal images from a mouse brain at depth up to 80 µm. b Normalized frames of OCM, THG, and merged signals at a depth of 34 µm. OCM and THG profiles of the green and magenta dotted lines across the vessels are displayed in (c). d Normalized frames of the OCM, THG, and merged signals of one large blood vessel at various depths (the center of the blood vessel was set at 0 µm and the depths of the other images are measured relative to the center). e Plot of OCM and THG signals FWHM values corresponding to (d). f Normalized frames of OCM, THG, and merged signals of a vertical blood vessel and a horizontal one (magenta dashed boxes) at a depth of 535 µm. The line intensity profiles crossing the blood vessel walls in the THG image are plotted on the right. Scale bars in (b, c), 50 µm; scale bar in (f), 20 µm. All frames were individually normalized only by the per-frame maximum pixel value to optimize display dynamic range for visualization purpose.

Discussion

In this paper, we demonstrated the use of a single OPA laser source, commonly used in three-photon microscopy (3PM), for simultaneous OCM, THG, and three-photon excited fluorescence imaging, and performed multimodal, high-spatial-resolution imaging of the intact adult mouse brain in vivo at depths >1 mm. 3PM is emerging as a valuable tool for studying neuronal structure and function using exogenous dyes or fluorescent proteins because of its capability for deep tissue penetration in the mouse brain. Meanwhile, OCM and THG could provide unique and complementary label-free contrast of brain features such as blood vessels, red blood cells, and myelinated axons. Notably, achieving this with a single modality is difficult because reliably identifying and registering the same vessels or fibers at depth is challenging. Therefore, the multiple contrast mechanisms of these three modalities could offer complementary structural and physiological information. For faster dynamic imaging, higher speeds could be achieved by reducing the scanning field of view or using a resonant scanner and faster spectrometer. This will allow our platform to further simultaneously record calcium activity with 3PM (using calcium indicators) while acquiring label-free THG and OCM, allowing direct correlation of neuronal/vascular function with fine structural context.

In the future, our multimodal approach could also potentially utilize the phase information from OCM for sensing of tissue-induced aberrations^58,59 as well as refractive index measurement^41,60. The tissue-induced aberration information provided by computational adaptive optics^61,62 can then be used to perform hardware adaptive optics for aberration correction⁶³. Such closed-loop aberration sensing and correction can be especially beneficial for 3PM because the higher-order nonlinear excitation is more sensitive to wavefront imperfections⁶⁴. Therefore, the combination of 3PM and SD-OCM offers not only the linear and the nonlinear contrast from the sample simultaneously, but also the intriguing possibility of using the information obtained in one modality to enhance the imaging performance of the other.

Methods

Mouse preparation

All animal procedures were reviewed and approved by the Cornell Institutional Animal Care and Use Committee. We used male B6.Cg-Tg(Thy1-Brainbow1.0)HLich/J mice (8 weeks old, The Jackson Laboratory) for red fluorescence protein (RFP)-labeled neuron imaging. Animals were prepared using the methods described in the previous work^8,65. We used a 5 × 5 mm² square-shaped λ/4 waveplate (WPQ501, Thorlabs) as the cranial window instead of the normal cover glass. Craniotomies were centered at 2.5 mm posterior and 2 mm lateral to the Bregma point. As the mice were not subjected to Cre recombinase, RFP tdimer2(12) was the only fluorescent protein expressed. During the surgery, the mice were anesthetized with 1.0 to 1.5% isoflurane in oxygen.

In vivo imaging of mouse brain

We imaged RFP-labeled neurons in a B6.Cg-Tg(Thy1-Brainbow1.0)HLich/J mouse two weeks after surgery. The mouse, imaged under anesthesia with 1.0% isoflurane in oxygen, was mounted on a motorized stage (M-285, Sutter Instrument Company) for axial scanning (although animals were used in this study, no euthanasia procedures were performed, as all experiments were non-terminal and did not require sacrifice). Deuterium oxide (i.e., heavy water, D₂O) was used as the objective immersion medium to reduce the absorption caused by water (H₂O) when imaging at 1620 nm. Because the index of refraction of brain tissue is larger than that of the immersion medium (heavy water), the actual imaging depth within the tissue was ~5–10% more than indicated by the depth measurement (i.e., the raw axial movement of the objective). For collecting fluorescence emission from RFP, we used a 593-nm long-pass filter (Semrock, FF01-593/LP-25) in combination with a band-pass filter (FF01-630/92-25). For collecting third-harmonic signal, a 558 ± 10 nm bandpass filter (Semrock) was used. For the detection of the third-harmonic/3P fluorescence signal, a GaAsP photomultiplier tube (H7421-40, Hamamatsu Photonics) was used. For OCM signal collection, an InGaAs linescan spectrometer (Wasatch Photonics, Cobra) was used. At each imaging depth, multiple frames were acquired and then averaged during imaging processing stage. More details on OCM data processing is described in OCM image reconstruction section.

OCM image reconstruction

SD-OCM images were reconstructed from spectral data through the following steps: wavelength to wavenumber domain resampling, reference background subtraction, dispersion compensation (resulting from different amounts of glass in the sample vs. reference arm), and finally an inverse Fourier transform along the wavenumber axis to convert data from the spectral-domain to the space-domain. In addition, sample tilt and coherence gate curvature were compensated by adding a linear phase ramp in the spectral domain for each A-scan (Fig. S2). The slope of the phase ramp at each transverse location was estimated by fitting a second-order polynomial to the cover glass surface. For OCM image processing, because OCM data are inherently three-dimensional, slight axial motion can be corrected across multiple acquired frames, helping to mitigate issues caused by in vivo biological movements such as heartbeats. Specifically, a bulk shift based on cross-correlation to a reference image was applied to eliminate minor motion artifacts throughout the entire acquisition at the same focus location. In the lateral direction, we averaged frames acquired at the same depth after taking the magnitude of the reconstructed complex OCT data.

Author contributions

X.Y., S.L., and F.X. built the imaging system, acquired the data, and analyzed the data. X.Y., S.L., F.X., M.W., C.X., and S.A. discussed the results. X.Y., S.L., and F.X. prepared the manuscript with input from all authors. C.X. and S.A. supervised the project and edited the manuscript.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to large file sizes but are available from the corresponding author on reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s44303-026-00148-9.