Focus-tunable two-photon fiberscope enabling in vivo imaging at selected depths

Abstract

Miniaturized two-photon (2P) imaging devices enable real-time in vivo and in situ imaging at subcellular resolution, highly valuable for clinical applications and basic research (such as neuroscience). However, achieving high-quality volumetric imaging at varying depths remains challenging. In this study, we demonstrated a 2P fiberscope capable of three-dimensional (3D) imaging over a cylindrical volume of a 350 μm diameter and a 400 μm depth. Depth scanning was achieved by incorporating a miniature electrowetting-based varioptic lens (VL) into a two-dimensional (2D) scanning 2P fiberscope, whose focus was tuned by modulating the VL drive voltage. The performance of the fiberscope was demonstrated by ex vivo imaging of fluorescently stained convallaria and GFP mouse brain sections, as well as in vivo dynamic GCaMP-based calcium imaging of cortical neurons in an awake mouse.

Two-photon (2P) microscopy has served as a powerful tool to image biological structures, thanks to its high resolution, depth-sectioning capability, resistance to tissue scattering, and the ability to achieve deeper penetration [1,2,3]. Depth scan in 2P imaging has also been achieved conveniently via mechanically shifting the optics in benchtop 2P systems [4] or by using a tunable lens [5,6].

Over the past decade, miniaturized 2P imaging devices have witnessed fast development to enable in vivo imaging and applications for both clinical [7] and neuroscience research [8,9,10] based on several approaches, including the use of GRIN lens to design micro-objectives, MEMS scanners, fiber-scanning technology, and/or fiber bundles. Among these technologies, the use of a double-clad fiber (DCF) and tubular piezoelectric actuator for two-dimensional (2D) scanning has enabled 2P fiberscopes to fully integrate excitation and collection, achieving ultra-compact size and ultralight weight [11,12,13,14] and making it suitable for endoscopic imaging. In these systems, high transverse resolution was achieved by using large-NA objectives [15] and/or a compound fiber cantilever [16].

With the goal of three-dimensional (3D) in vivo imaging, depth scanning in miniaturized 2P imaging devices has been actively explored [17]. Generally, two different kinds of depth scanning have been applied in miniaturized imaging systems. One method is mechanical scanning: adjusting the separation between optical elements [18] and scanning the microlens [19] or the entire miniature imaging device [20,21], which has been applied to fiberscopes using shape memory alloy (SMA) wires. However, these approaches usually result in heavy weight due to complicated mechanical structural design and suffer from unsatisfactory accuracy or limited scanning range determined by the mechanical actuator. The other method is to optically tune the focal length of the imaging optics, where the lens shape can be controlled (such as fluid-based lenses and deformable elastomeric lenses) [22,23], or the refractive index of the material can be changed (such as a liquid crystal lens) to achieve adjustable focus [24].

In this study, we opted for the second approach and realized depth scanning by incorporating an electrowetting-based variable lens (VL) into a 2P scanning fiberscope to enable focus tuning. The fiberscope weighs 1.9 g, with an outer diameter of 2.4 mm at the distal end, and is capable of focus tuning over a cylindrical volume field of view (FOV) with 350 μm diameter × 400 μm depth. We first evaluated its performances using fluorescent slides and microspheres and then demonstrated the focus-tuning capability using ex vivo samples (convallaria and mouse brain sections) as well as in vivo dynamic calcium imaging of cortical neurons in an awake mouse. The low electrical power consumption, fast response, high repeatability, motionless focus scanning, and ease of operation offer tremendous benefits for in vivo imaging. We expect this compact, flexible 2P fiberscope with depth scanning capability will open up promising opportunities in various clinical applications and neuroscience research.

Our imaging system is based on the resonant fiber-scanner technology for performing 2D spiral scanning by the use of a tubular PZT actuator, a compound DCF cantilever, and distal-end micro-optics, as detailed in prior studies [11,13]. The imaging system used a Ti:sapphire laser (Coherent, Chameleon Ultra II, repetition rate: 80 MHz) as the light source and a photomultiplier tube (Hamamatsu, H10770PA-40) for fluorescence detection. A GRISM pair enabled delivery of ~76 fs pulses at 920 nm to a sample via a fiberscope, with angle and spacing optimized [14]. The 2D spiral scanning operates at 2.3 kHz, providing a radial pixel size of 0.22 μm and a circumferential pixel size of 0.51 μm at the outermost circle. The depth scanning is controlled by tuning the VL, which determines the axial pixel size. The pixel dwell time is 0.2 μs, with a sampling rate of 5 MHz.

Given our goal of maintaining a compact, lightweight design while ensuring high image quality across depths, we adopted the optical tuning method and integrated an electrowetting-based VL (Corning, A-25H0-D0-33) into distal-end optics.

This approach not only eliminates the need for mechanical movement of optical components or the probe itself but also preserves collection efficiency due to electrowetting’s insensitivity to polarization, outperforming other VLs based on liquid crystal lenses. The optical power of the lens can be tuned from −35 to +35 diopters with a fine-tuning resolution of 37 mD by a driving voltage ranging from 30 to 65 V. As shown in Figs. 1(a) and 1(b), the curvature of the water–oil interface is adjusted by applying the voltage between the two rings of metal: one connected to the water layer and the other situated under the oil layer separated by a layer of insulator [22]. The VL has a 2.5 mm clear aperture and anti-reflective (AR) coatings optimized in the visible range, providing an optical transmittance of 95% at 520 nm. Although the transmittance at an excitation wavelength of 920 nm is 77%, we prioritized optimizing the emission light path to enhance fluorescence collection efficiency. To compensate for attenuation of excitation light across the tuning range, the laser power delivered through the fiberscope was appropriately increased to maintain a constant incident power at the sample. The VL is connected to a driver via a flexible printed circuit (FPC) cable for focus to a driver via a flexible printed circuit (FPC) cable for focus control (see Fig. 1(c)). The VL also has a fast response time of <10 ms, much shorter than the frame rate of the 2D scanning (typically 2–3 fps).

Fig. 1.

Photos and cross-sectional schematics of the electrowetting-based VL and the focus-tuning 2P fiberscope. (a) Schematic cross section of the A-25H0-D0-33 lens in its lowest optical power (divergent) with 30 V applied between the water and the metal beyond the oil layer. (b) Schematic cross section of the lens in its highest optical power (convergent) with 65 V applied between the water and electrode. (c) Photos of A-25H0-D0-33 lens. Upper, unpackaged lens; lower, packaged lens with the FPC cable. (d) Schematic cross section of the distal-end optics of the focus-tuning 2P fiberscope. (e) Photo of the fiberscope.

The schematic of the focus-tunable 2P fiberscope is shown in Fig. 1(d). The VL was positioned between two GRIN lenses (GRINTECH, LFRL-200-023-50), where the beam is nearly collimated. The performance of this configuration was confirmed by ZEMAX simulations, providing the widest range of working distances (WD). In our fiberscope, the 2P excitation and emission light share the same optical path. To mitigate the chromatic aberration between the excitation and emission wavelengths along the DCF (~1 m in length from the proximal end to the distal end) and distal-end optics, a diffractive optical element (DOE) was placed between the first GRIN lens and the VL, thereby enhancing collection efficiency [25,26]. Details of the DOE design are provided in our previous work [27]. As shown in Fig. 1(d), the distal-end optics were assembled with the composite fiber scanner within hypodermic metal tubing (2.4 mm outer diameter). Additionally, a 3D-printed plastic housing is used to protect the VL and secure its connection to the tubing assembly. The VL itself (including the ring electrode) has a diameter of 7.7 mm, while the plastic 3D-printed holder for the assembly of the VL measures 10 mm in diameter. The VL assembly (including the packaged VL and FPC cable) weighs 0.6 g, and the fiberscope (including the metal tubing and plastic housing) weighs 1.9 g in total.

The WD of the fiberscope in immersion oil (Cargille, type A microscope immersion oil) was measured throughout the 30–65 V driving voltage range. We placed a green fluorescence reference slide on a precision motorized linear stage (Newport LTA-HS) and recorded the WD distance at each voltage. We then compared the experimentally measured WD with the simulated WD (in ZEMAX) across various voltages applied to the VL. As shown in Fig. 2(a), a depth scanning range of 400 μm (WD from 614 to 214 μm) was obtained. The discrepancy between the simulated and measured working distances was smallest (~3 μm) at the lowest voltage (30 V) and increased at higher voltages, reaching 24 μm at 65 V. At 49 V, where no optical power is provided by the VL, the simulated WD was 411 μm, and the measured WD was 415 μm. In addition, we measured the fluorescence collection efficiency of the fiberscope, and the normalized results are shown in Fig. 2(b). The highest efficiency occurred when the VL’s optical power was near zero, and an ~10% decrease in efficiency when the VL’s optical power was at its maximum. With this in mind, during volumetric imaging, the fluorescence intensity variations across imaging depths corresponding to different driving voltages can be digitally compensated. We also accounted for potential hysteresis of the VL by changing the voltage tuning direction. The VL exhibited high repeatability, and no hysteresis was observed over its tuning range.

Fig. 2.

Fiberscope’s performance. (a) Fiberscope’s WD tuning range versus driving voltage and the measured resolutions at three different WDs. (b) Normalized fluorescence collection efficiency of the fiberscope versus driving voltage. The black dashed line indicates 49 V, at which the VL lens provides no optical power or focusing.

The spatial resolution of the fiberscope was measured by imaging across 200 nm diameter fluorescent microspheres (Fluoresbrite 09834) and estimating the full width at half maximum (FWHM) of the point spread function (PSF) of an individual microsphere (Gaussian fitting for lateral resolution and Lorentzian fitting for axial resolution). Table 1 shows both the simulated and the measured resolutions at three imaging depths in immersion oil: 230 μm (64 V), 415 μm (49 V), and 605 μm (31 V). The simulated resolution remains consistent—or even slightly improves—with increased imaging depth, as the simulated image space NA, i.e., the effective NA at the sample plane, slightly increases at larger depths. Note that the ZEMAX model of the VL provides only an approximate estimation, and the simulation serves primarily as a guideline for tuning the WD. A slight reduction in the measured resolution was observed with increasing depth. This degradation is attributed to the reduced focusing power of the lens assembly and potential misalignments during fiberscope assembly. We then conducted 2P imaging with the focus-tunable fiberscope on two ex vivo samples. To test its ability of focus tuning, we used two thin slides: a convallaria rhizome slide (stained with acridine orange, around 60 μm thick) and a GFP mouse brain slide (around 40 μm thick). During imaging, the samples were positioned at a WD of 420 μm, corresponding to the midpoint of the depth-tuning range. An FOV of a 350 μm diameter over a depth range of ~150 μm (with a driving voltage from 42 to 55 V, in 1 V intervals for the convallaria slide and 0.5 V intervals for the brain slide) was imaged. The reconstructed volumetric images of the convallaria and mouse brain slides are presented in Figs. 3(a) and 3(b), respectively, with three adjacent layers (−20, 0, +20 μm) displayed on the side. For both slides, the layer with the strongest fluorescence intensity was defined as the 0 μm reference layer. When imaging from −20 to +20 μm, the images transitioned from blurred and dim to sharp and bright and then became defocused again—validating the fiberscope’s depth-tuning capability. With this confirmed, we proceeded to in vivo neural activity imaging to demonstrate the system’s potential for 3D volumetric data acquisition.

Table 1.

Spatial Resolution at Different Depths

Depths (μm)	Driving Voltage (V)	Image Space NA	Simulated Lateral (μm)	Measured Lateral (μm)	Simulated Axial (μm)	Measured Axial (μm)
230	64	0.34	1.21	1.24	18.01	16.78
415	49	0.35	1.20	1.31	17.11	19.62
615	31	0.35	1.19	1.44	16.42	26.89

Fig. 3.

3D ex vivo image of the convallaria slide and mouse brain GFP slide with 2D images at the depth of the slide (denoted as 0 μm) and 20 μm above/below. (a) Convallaria slide under 10 mW excitation at 920 nm. (b) Mouse brain GFP slide under 20 mW excitation at 920 nm.

In vivo neuroactivity imaging was performed by measuring the GCaMP6 m fluorescence from firing neurons through a cranial window over the somatosensory cortex on a head-fixed mouse (Jax, #005359). The overall setup is shown in Fig. 4(a). During imaging, the mouse’s head was secured by using a head-restraining bar, and the 2P fiberscope was gently positioned above the cranial window with a gap of 50 μm from the #1 cover glass. All animal housing and experimental procedures were conducted according to protocols approved by the Institutional Animal Care and Use Committee (IACUC) of Johns Hopkins University.

Fig. 4.

(a) Photo of the in vivo mouse neuroimaging setup. (b) 3D in vivo calcium imaging of mouse’s cortical neurons under 35 mW excitation power at 920 nm, with 2D images at three different depths below the surface of the cortex. (c) ΔF/F curves of three representative neurons labeled at each depth.

3D imaging was performed over a 350 μm diameter FOV and a 400 μm depth (with a driving voltage from 30 to 65 V, in 1 V intervals). At each depth, 100 frames of images were collected with a frame rate of 2 fps. For 3D visualization, we averaged every five frames to reduce the noise and performed maximum intensity projections to visualize all firing neurons. Figure 4(b) shows the reconstructed volumetric calcium imaging of cortical neurons in an awake mouse, with three layers at depths of 228 μm, 274 μm, and 319 μm below the cortex surface displayed on the right-hand side. Individual neurons within the 3D volume are clearly resolved. The temporal calcium dynamics (ΔF/F curves) for three representative neurons at each layer were extracted using the CaImAn data processing pipeline [28] and illustrated in Fig. 4(c).

To sum up, we demonstrated a focus-tunable 2P fiberscope by integrating an electrowetting-based VL onto a 2D resonant fiber-optic scanner, achieving a depth scanning range of 400 μm. The performance of the fiberscope was evaluated by imaging fluorescent microspheres, ex vivo samples, and in vivo neurons of a mouse’s somatosensory cortex. Compared with the method with physically moving parts for depth scanning, this motionless focus scanning design is more attractive for in vivo and in situ imaging. Several miniaturized 2P imaging devices have been reported, capable of depth scanning using commercially available tunable optics. The 2P-FCM (~2.5 g) employed an electrowetting tunable lens (from −16 to +36 diopters, OD 7.8 mm) and achieved a depth-tuning range of 180 μm [29]. The FHIRM-TPM 2.0 (4.2 g) integrated an electrically tunable lens (from −30 to +30 diopters, 10×10×15 mm³) and also enabled a 180 μm tuning range [30]. More recently, by incorporating a micro-tunable quartet (from −24 to +51 diopters, 4.5×4.5×2.2 mm³), the MINI2P (2.4 g) achieved a 240 μm scanning range [31]. In comparison, our work presents the first demonstration of depth scanning in a 2P fiberscope. By employing a VL (from −35 to +35 diopters, OD 7.7 mm), we achieved an even larger depth-tuning range of 400 μm. Additionally, the DCF-based design enables both 2P excitation and fluorescence collection through a single fiber, resulting in an ultralight probe with a total weight of 1.9 g. Based on our current design, several potential upgrades can expand its future applications. First, increasing the length of the FPC cable beyond the current commercially available 6 cm would allow the driver board to be relocated to the proximal end, facilitating easier attachment of the fiberscope to the mouse head for two-photon neural imaging in freely behaving mice. Second, leveraging the VL’s rapid response time (<10 ms), the scanning mode can be shifted from multi-layer 2D scans to rapid sequential 1D depth scans over an area, enabling time-resolved volumetric neuroimaging. Lastly, since the distal-end GRIN lens has a diameter of only 2 mm, customizing the VL packaging and holder could further reduce the overall size (and weight) of the 3D fiberscope. With advances in fabrication techniques and improved integration of the VL into scanning algorithms, we anticipate the next-generation focus-tunable 2P fiberscope to be more compact and capable of flexible volumetric imaging.

In addition to neuroscience research, this technology has the potential to ease its operation for clinical use (e.g., by eliminating the need for mechanical adjustment and shortening the detection time) and to provide substantial benefits in ensuring safe and effective surgical procedures. We hope that this focus-tunable 2P fiberscope will help unlock promising opportunities across a wide range of applications.

Data availability.

Data underlying the results presented in this paper can be obtained from the authors upon reasonable request.