2.1.

Laser System

The blue shaded region of Fig. 1 is the laser system, where a 100 fs pulse at 1045 nm from the pump laser (Y-Fi, KMLabs, Boulder, Colorado, United States) is converted to $\sim 1300\mathrm{nm}$. A telescope magnifies the beam before propagation onto a spatial light modulator, which encodes a transverse spatial phase onto the incident Gaussian beam to ensure excitation of a single pure mode in the multimode fiber.²¹ Here, a linearly polarized (LP) ${(\mathrm{LP})}_{0,21}$ mode is excited in the custom 54-cm higher-order-mode fiber (core diameter of $97\mu \mathrm{m}$ and numerical aperture of 0.34). Initially in the fiber, a soliton is formed at the pump wavelength in the ${\mathrm{LP}}_{0,21}$ mode and subsequently the center wavelength of the soliton continuously red-shifts via soliton self-frequency shifting (SSFS). The soliton eventually reaches a wavelength at which it has the same group velocity as the ${\mathrm{LP}}_{\mathrm{0,20}}$ mode at a longer wavelength that is separated by a frequency detuning equal to the Raman gain peak of silica (13 THz). At this point, the energy of the soliton is transferred via the recently discovered nonlinear effect of SSMC to a new pulse in the ${\mathrm{LP}}_{0,20}$ mode. From this point, until the end of the fiber, the new pulse then frequency shifts via SSFS to 1317 nm where the pulse energy is measured to be 80 nJ.^11,22 Figure 2(a) shows the output spectra from the fiber before and after spectral filtering with a 1250-nm longpass filter. Figure 2(b) shows the spatial mode of the pulse centered at 1317 nm. It is seen to be in the ${\mathrm{LP}}_{0,20}$ mode, which resembles a truncated Bessel beam comprising a central peak surrounded by 19 concentric rings. This exact frequency conversion process requires that light initially possessed certain linear properties, which is ensured by launching the light in a specific higher order mode in the multimode fiber, which in this case is the ${\mathrm{LP}}_{\mathrm{0,21}}$ mode. Using an autocorrelator, the pulse duration was measured at 74 fs, corresponding to a peak power of $\sim 1.1\mathrm{MW}$. Based on the measured spectral bandwidth, it was found that the pulses are nearly transform-limited, an expected result considering the pulses are solitons.

Fig. 2

(a) Output spectrum before and after spectral filtering. (b) Experimental image of the ${\mathrm{LP}}_{0,20}$ output mode after spectral filtering. (c) Gaussian beam after spatial re-conversion.

2.2.

Spatial Mode Conversion

In the yellow shaded region of the experimental setup shown in Fig. 1, a longpass filter initially removes the residual pump light. Next, the ${\mathrm{LP}}_{0,20}$ fiber output mode is converted to a Gaussian-like beam using an axicon and lens, with a measured power conversion efficiency of 81% (theoretically 86%).¹¹ Figure 2(c) shows the image of the Gaussian beam after the spatial mode conversion. The following two lenses magnify and collimate the Gaussian beam to a beam diameter of 5 mm [full width at half maximum (FWHM)]. At this point, the pulse duration was measured at 100 fs using an autocorrelator. Though Bessel beams may be used to extend the depth of focus and for fast volumetric imaging of sparse samples,¹⁹ a Gaussian beam was utilized as it is more energy-efficient for imaging histological sections of excised nerves.

2.3.

Multiphoton Microscope

The multiphoton microscope is shown in the red shaded region of Fig. 1. The system was designed de novo around the aforementioned 1300-nm fiber laser, for cost efficiency and performance optimization. Galvo mirrors (GVS002, Thorlabs, Newton, New Jersey, United States) are used to scan the beam. The scan and tube lenses comprise of several air-spaced commercial lenses from Thorlabs. The optical design was optimized in Zemax to minimize aberrations, resulting in diffraction-limited performance for optical ray angles below 11.3 deg. The scan lens consists of two LA1541-C lenses and a LA1031-C lens and the tube lens consists of two AC508-300-C. The signals are epi-collected by a $20\times$ ($\mathrm{NA}=1$) objective lens (XLUMPLFLN20XW, Olympus, Tokyo, Japan), separated by a 775-nm longpass dichroic mirror (FF775-Di01, Semrock Inc., Rochester, New York, United States), and then detected sequentially using emission filters by a photomultiplier tubes (PMT) (H7422-40, Hamamatsu Photonics, Hamamatsu, Japan). The collection path was designed in Zemax from commercial lenses (Thorlabs) to ensure maximum collection efficiency. The first lens (Thorlabs LA1256-A) is followed by an air-spaced lens consisting of three plano-convex lenses (Thorlabs LA1417-A). This optical design ensures a full collection field-of-view (FOV) of $\sim 1\mathrm{mm}$ in diameter. The sample stage enabled coarse lateral movement and axial scanning using a three-dimensional (3D) stepper motor (TRA25CC, Newport Corp., Irvine, California, United States). A moveable silver mirror downstream of the dichroic mirror enables optional widefield imaging. The pre-amp is a FEMTO DHPCA-100 (FEMTO Messtechnik GmbH, Berlin, Germany), and the data acquisition card is an NI PCIe-6353 (National Instruments Corp., Austin, Texas, United States). Samples were excited using the same wavelength and pulse duration, but images were sequentially collected beginning with the signal requiring the least pulse energy (i.e., SHG) to minimize the risk of photodamage. The power was adjusted using a combination of neutral density filters, and polarization control was employed to maximize the signal strength. Images were collected with specific bandpass-filters to isolate SHG, THG, and 3PE fluorescence signals. Through the appropriate choice of filters crosstalk between images was avoided, as evident from the signals being delocalized between the images. The power dependency of the three signals was verified by collecting emission as a function of illumination beam power. The microscope system was controlled using ScanImage and the images were processed using ImageJ software (Fiji Distribution, Version 1.52e).^23,24 Frame averaging was used to further minimize noise. Colors were arbitrarily assigned to the grayscale images. The pulse duration was measured using an autocorrelator after the objective lens at 114 fs. The point spread function (PSF) of the system was measured using 175-nm blue fluorescent beads; the results for the lateral ($0.65\mu \mathrm{m}$) and axial ($2.9\mu \mathrm{m}$) directions are shown in Fig. 3. For calibration and PSF measurements, the fiber laser was operated at a repetition rate of 1 MHz. For biological sample imaging, the repetition rate of the laser was increased to 10 MHz to improve scanning efficiency. In both instances, the maximum pulse energy at the sample plane was $\sim 10\mathrm{nJ}$.

Fig. 3

PSF measurements. (a) The lateral FWHM is $0.65\mu \mathrm{m}$. (b) The axial FWHM is $2.9\mu \mathrm{m}$.

A summary of the optical performance of the MPM system is given in Table 1.

Table 1

Performance of MPM system.

2.4.

Biological Samples

Experimental protocols were approved by the Mass Eye and Ear Internal Review Boards, and all methods were carried out in accordance with relevant guidelines and regulations. Fresh frozen sections of healthy human motor branch of obturator nerve and transgenic mice sciatic nerve were employed. Adult human nerve samples (patient age >18 years) were obtained fresh at the time of the free-tissue transfer for smile reanimation after informed consent among patients undergoing facial reanimation procedures. Sciatic nerves from thy1-YFP mice were harvested in accordance with Massachusetts Eye and Ear Institutional Animal Care and Use Committee approval. Specifically, animals were placed in an induction chamber and placed under general anesthesia by inhalation of 2% isoflurane in $0.6\mathrm{l}/\min$ ${\mathrm{O}}_2$. Once unconscious, isoflurane was increased to 5% and continued until cessation of breathing was noted for one minute, followed by immediate cervical dislocation. Nerves were immediately harvested from the carcasses. Whole mounts were kept at 4°C in saline-soaked gauze and imaged fresh within 2 h. Nerves for sectioning were processed using a previously described protocol.²⁵ Briefly, nerves were fixed by immersion in 2% phosphate-buffered paraformaldehyde, followed by cryoprotection in sucrose solution, cryosectioning at 1 to $2\mu \mathrm{m}$, and staining using a non-toxic myelin-specific dye followed by mounting on glass slides prior to multiphoton imaging.

3.1.

Imaging of Human Obturator Nerve

The microscope was used for SHG and THG epi-detection imaging of human and murine peripheral nerve. Figures 4(a) and 4(b) show images of axially frozen-sectioned human obturator nerve obtained sequentially with different bandpass filters (BPF). For this label-free sample, the SHG signal (617/73 nm BPF) was generated from collagen in the three nerve sheath layers (endoneurium, perineurium, and epineurium). Collagen is the main tissue component responsible for an SHG signal in peripheral nerve tissue. For validation of label-free SHG imaging, picrosirius red was employed on separate tissue sections as an efficient alternative to immunohistochemistry for high-specificity labeling of collagen-dense structures. Picrosirus red-stained collagen was imaged using brightfield microscopy²⁶ as shown in Fig. 4(c). A THG signal (420/50-nm BPF) was predominately generated from the myelin sheaths of myelinated axons [Fig. 4(b)], demonstrating near-identical imaging results compared to widefield fluorescence imaging of myelin-specific dye stained sections, as shown in Fig. 4(d). Though the THG signal is predominantly emitted in the forward direction, there is sufficient backward-scattering in turbid media (such as peripheral nerve) to obtain a sufficient signal using epi-detection as demonstrated here and by other groups.^9,27 The combination of photobleaching-free SHG and THG images reveal complementary information allowing characterization of nerve morphology in unstained tissues, as shown in Fig. 4(e). Figure 4(f) shows myelinated axon quantification of unlabeled cryosectioned human peripheral nerve performed through segmentation of a THG image using commercial machine learning software.²⁸

Fig. 4

Label-free imaging of ex vivo human obturator nerve cross-sections using a novel 1300-nm fiber laser-powered multiphoton microscope. (a) Second-harmonic generation imaging of neural connective tissue collagen, post objective average power 7.5 mW. (b) Third-harmonic generation imaging of myelinated fibers, post objective average power 15 mW. (c) Picrosirius red staining of myelinated fibers (conventional light microscopy). (d) Myelin-specific dye (FluoroMyelin Green, Molecular Probes, Eugene, Oregon, United States) staining of cross-section (widefield fluorescence microscopy). (e) Merged image combining SHG (red) and THG (green) signals. (f) Automated histomorphometry of peripheral nerve human obturator nerve section imaged by THG. The FOV is $380\times 380\mu \mathrm{m}$, and the scale bar is $50\mu \mathrm{m}$.

3.2.

Imaging of Sciatic Murine Nerve

Figure 5 shows imaging of a sciatic nerve cross-section from a thy1-YFP mouse using the multiphoton microscope developed here. Three-photon-excitation fluorescence images were collected using a third filter ($535/60\mathrm{nm}$ BPF) to separate them from SHG and THG signals. Figures 5(a)–5(c) show a THG signal from myelin, SHG signal from collagen, and three-photon fluorescence excitation signal from YFP-labeled axons. In Fig. 5(d), the three images are merged to obtain a multicolor image.

Fig. 5

Multiphoton-microscopy image of ex vivo sciatic nerve of a thy1-YFP mouse. FOV is $210\times 240\mu \mathrm{m}$. Scale bars are $25\mu \mathrm{m}$. Post objective average power was 12.5 mW. (a) Third-harmonic generation (blue) shows the myelinated fibers. (b) Three-photon excited fluorescence of YFP (green) shows axons. (c) Second-harmonic generation (red) shows collagen fibers. (d) Merged image of THG, 3PE, and SHG signals.

To estimate the nonlinear order and required power for multiphoton imaging, the dependence of the signal on excitation power was measured. The plot of signal strength versus excitation power is shown in Fig. 6, along with fits. The black-dashed lines represent the power order as a free-fitting parameter, and the red-dashed lines represent fits with fixed power order. The fits show excellent agreement with the data, indicating capture of appropriate emission. Excitation power exhibits a second-order increase in the SHG signal ($\times 1.90$) and third-order increase in the 3PE ($\times 2.99$) and THG ($\times 3.24$) signals.

Fig. 6

Power dependence of the three signals measured in Fig. 5. (a) Three-photon signal. (b) THG signal. (c) SHG signal.

To assess whether this novel 1300-nm fiber-based multiphoton microscope could potentially be employed for in vivo imaging of mammalian nerve, freshly harvested whole mount murine sciatic nerve from a thy1-YFP mouse was imaged. Figure 7 shows the results for 3PE [Fig. 7(a), YFP] and THG imaging [Fig 7(b), myelin] using non-descanned epi-detection at a depth of $120\mu \mathrm{m}$ into the tissue. The two images are merged in Fig. 7(c).

Fig. 7

Multiphoton-microscopy images of wet mounted sciatic nerve of a thy1-YFP mouse at a depth of $120\mu \mathrm{m}$, post objective average power was 30 mW. (a) Three-photon excited fluorescence of YFP (green) shows axons. (b) Third-harmonic generation (blue) signal shows the myelinated fibers. (c) Merged image of 3PE and THG signals. About 50 frame averages were used to enhance image quality. FOV is $480\times 480\mu \mathrm{m}$. Scale bars is $50\mu \mathrm{m}$.