2.1.

Animal Surgery

All experimental procedures have been performed in accordance with guidelines from the Canadian Council on Animal Care. C57BL/6 adult mice (25 to 30 g of body weight) either wild type or transgenic, CX3CR1-GFP (Jackson Laboratory, B6.129P-$\mathrm{Cx}3\mathrm{cr}{1}^{\mathrm{tm}1\mathrm{Litt}}/\mathrm{J}$) or Thy1-YFP (Jackson Laboratory, B6.Cg-Tg[Thy1-YFP]16Jrs/J), were anaesthetized with a mixture of ketamine-xylasine ($100\mathrm{mg}/\mathrm{kg}$ and $10\mathrm{mg}/\mathrm{kg}$, respectively, by intraperitoneal injection) supplemented hourly. Body temperature was kept at $37.5\pm 2°\mathrm{C}$ with a feedback-controlled heating blanket.

To image the spinal cord in vivo, either the lumbar segment L5 was exposed by laminectomy and the meninges gently removed or the intervertebral space between lumbar segments L4 and L5²⁴ was cleared of muscular tissue and enlarged to a 1.5-mm diameter hole with a surgical drill. The latter intervention gave access to the spinal tissue with the meninges left in place, minimizing surgery invasiveness. A portable stereotaxic frame with two adjustable clamps for vertebral fixation, rostral, and caudal to the exposed section, was mounted on a micromanipulator (Sutter Instrument, MPC-200). An agar pool was formed around the exposed spinal segment and filled with artificial cerebrospinal fluid.

2.2.

Video-Rate Multimodal Nonlinear Microscope

The laser sources⁶ and the custom-made video-rate laser scanning microscope²⁵ are illustrated in Fig. 1. The laser system consists of an optical parametric oscillator (OPO) (APE, Levante Emerald) pumped by a frequency-doubled Nd:${\mathrm{YVO}}_4$ mode-locked laser (High Q Laser, picoTRAIN). The pump laser generates a 7-ps, 80-MHz pulse train of 532- and 1064-nm laser light. The OPO is pumped with 5 W of 532-nm laser light.

Fig. 1

Laser and imaging system for simultaneous CARS and TPEF imaging with a microendoscope. $T_p$: pump beam telescope, $T_s$: Stokes beam telescope, ${\mathrm{DF}}_1$: recombining dichroic filter, ${\mathrm{DF}}_2$ and ${\mathrm{DF}}_3$: detection dichroic filters, SPF: short-pass filter and BPF: band-pass filter.

CARS is a nonlinear process in which two laser beams at two different frequencies drive the emission of an anti-Stokes signal when their frequency difference matches a Raman active vibration of the sample.^26–28 In order to probe the ${\mathrm{CH}}_2$ symmetric stretch vibration of lipids at $2845{\mathrm{cm}}^{-1}$, the OPO is tuned to 816.8 nm and is mixed with a fraction of the 1064-nm laser beam. Both pump and Stokes beams are used as excitation for TPEF, but most of the signal originates from the use of the pump beam. The excitation beams are overlapped in space using a dichroic long-pass filter (Semrock, LP02-1064RU-25) and in time using a delay line before they are sent to the laser scanning microscope. Very critical to the perfect spatial overlap of the two beams at the focal spot, their sizes and divergences are independently adjusted with $2:1$ telescopes to maximize CARS signal generation. Small adjustments of the distance between the lenses making the telescopes compensate for chromatic aberration and/or wavefront mismatch that could arise in the optical path to the focal point. The microscope comprises a raster beam scanner formed by a gold-coated polygonal mirror for the fast axis (Lincoln Laser, DT-36-290-025) and a galvanometer mirror for the slow axis (Cambridge Technology, 6240H). The microscope field of view (FOV) is 187.5 μm by 187.5 μm with a $40\times$ objective lens (Olympus, LUMPlan Fl/IR, 0.8 NA, w).

The system acquires images of $500\times 500$ pixels at a rate of 15 frames per second. The backscattered anti-Stokes signal at 662.8 nm or the fluorescence signal are collected in epi-detection configuration. They are separated from the excitation beams by a dichroic long-pass filter (Semrock, FF735-Di01-25×36) and spectrally filtered from unwanted residual light using two laser block filters (Semrock, FF01-750/SP-25). A dichroic long-pass filter (Semrock, FF605-Di02-25×36) sends the signal to the TPEF detector behind a band-pass filter (Chroma, D545/90m), and the remaining light goes to the CARS detector after being filtered by a band-pass filter (Semrock, FF01-655/40-25). A red-sensitive photomultiplier tube (PMT) (Hamamatsu, R3896) is used for both channels. When transgenic mice are imaged, TPEF and CARS imaging are performed simultaneously.

2.3.

Microendoscope

The microendoscope is 7.4 mm long, has a diameter of 1.4 mm, a magnification of $4.8\times$ and a working distance of 200 μm (GRINTech, GT-MO-080-018-810, 0.8 NA, w). It is held under the microscope by a clamp (Thorlabs, Micro V-Clamps) attached to a 3 axis stage (Newport, ULTRAlign). A $10\times$ injection lens (Olympus, PLN, 0.25 NA) is used to inject light into the proximal end of the microendoscope with approximately matching numerical aperture (NA, 0.25 versus 0.18). The total magnification of the microendoscopic system is $48\times$, and the FOV is 162.5 μm by 162.5 μm.

2.4.

ISG Procedure

A post-processing algorithm that alleviates respiration-induced image movement artifacts called ISG has been developed. The method relies on movies of distortion-free frames (e.g., 300 to 450) followed a posteriori by a custom-built Matlab routine (Mathworks, R2010a) that computes a two-dimensional (2-D) cross-correlation matrix²⁹ between a user-selected key frame and every other frames. The maximum of the 2-D cross-correlation matrix is then used as a parameter to select a user-specified amount of similar frames and create a new gated data subset. This subset is then aligned and averaged to a single frame, resulting in an image that is free from motion-related imaging artifacts. All images [except Figs. 2(a) and 2(c)] have been processed using this procedure. The ISG routine is available online at www.dcclab.ca/isg/.

Fig. 2

ISG reduces motion-related artifacts from imaging in living animals. (a) User-selected key frame. (b) Artifact-free image obtained with ISG. (c) Averaging of all frames of the movie. All scale bars are 15 μm. (Video 1, MOV, 9.4 MB). [URL: http://dx.doi.org/10.1117/1.JBO.17.2.021107.1]

2.5.

Image Processing for Histomorphological Analysis

Myelin histomorphometry is assessed by computing the $g$-ratio and the myelin thickness.³⁰ Image processing to measure these parameters from CARS images is performed with Matlab and ImageJ (NIH, 1.42q), following a previously described method.⁷ Briefly, three regions of interest (ROIs), each parallel to a myelinated axon (when present), are selected on all tiles of a $2\times 2$ grid overlaid on every image. The axon and myelinated fiber diameters are determined respectively by the inner and outer edges of the myelin structure. These values are extracted from a cumulative integral of the average line profile. The $g$-ratio is calculated by dividing the diameter of the axon by that of the fiber. Myelin thickness is half the difference between the axon and the fiber diameters. Because axons are randomly sampled at different positions along their diameter, the $g$-ratio distribution is skewed toward smaller values.⁷

2.6.

Signal-to-Noise/Background Ratio

In the case of myelin imaging with the CARS modality, the signal-to-noise ratio (SNR) is defined by the ratio of the mean intensity of a myelin sheath area (signal) to the standard deviation of the corresponding adjacent axonal region (background). Analogously, the signal-to-background ratio (SBR) is obtained by the quotient of the mean intensity of a myelin sheath area to the mean of the corresponding adjacent axonal region.

2.7.

Statistical Analysis

Values are expressed in terms of $\text{mean}\pm \text{standard error}$ of the mean unless otherwise stated. Statistical analysis of the data was performed using Matlab. Student’s $t$-test and Mann-Whitney $U$-test were used to evaluate differences between the mean and the median of two distributions respectively.

3.1.

Reduction of Motion-Related Imaging Artifacts with ISG

Figures 2(a) to 2(c) illustrates a typical implementation of the ISG procedure. The original movie has been acquired from the dorsal root at the lumbar enlargement level of a mouse spinal cord and contains 310 frames (Video 1). The user-selected key frame is shown in Fig. 2(a). The graininess of the key frame arises from the low photon count (the shot noise of the PMT), intrinsic to a high-frame-rate acquisition system. Figure 2(b) shows the movement artifact-free image that has been obtained with the ISG procedure while Fig. 2(c) displays the result from averaging every frame composing the movie (without the use of ISG). Comparison among Fig. 2(a), Fig. 2(b), and Fig. 2(c) illustrates the benefits of the ISG procedure in terms of SNR, SBR, and image blurring over unprocessed images. The SNR for the ISG procedure ($16.3\pm 2.0$) compared to the single frame ($4.1\pm 0.2$) is significantly higher (five ROIs, one image, student’s $t$-test, $p\le 0.001$). Furthermore, the SBR for the ISG procedure ($2.7\pm 0.2$) versus the averaging ($1.7\pm 0.1$) is also significantly higher (five ROIs, one image, student’s $t$-test, $p\le 0.001$). More importantly, myelin sheaths are much sharper when the ISG procedure is used, giving a more reliable morphological representation of the imaged region.

3.2.

Comparison of CARS Images Obtained with the Objective Lens and the Microendoscope

The objective lens, the injection lens, and the microendoscope are depicted in Fig. 3(a) for size comparison. Figures 3(b) and 3(c) show in vivo CARS imaging of myelin sheaths from the dorsal column of a mouse spinal cord at the lumbar enlargement level, acquired with the $40\times$ objective lens and the microendoscope respectively. Laser power at the sample was set to 75 mW (pump) and 25 mW (Stokes) for both conditions. The near infrared wavelengths (816.8 nm and 1064 nm) and the short pixel dwell time ($\simeq 267\mathrm{ns}$) ensure that no damage is done to the tissue. The PMT gain and all other acquisition parameters were kept constant throughout the imaging session. Each image is an average of 30 frames. Signal strength (integrated intensity) is higher in the case of the objective lens by 27% (two images), but optical resolutions (both lateral and axial) are similar due to their identical NA.¹² The SNR ($\simeq 15$) and the SBR ($\simeq 3$) are also comparable for both lenses (10 ROIs, two images, student’s $t$-test, $p\ge 0.05$). The graph in Fig. 3(d) represents the intensity profiles highlighted by the ROIs in Figs. 3(b) (dashed) and 3(c) (full).

Fig. 3

The use of the microendoscope does not deteriorate in vivo CARS image quality. (a) Photo of the $40\times$ objective lens, the $10\times$ injection lens, and the microendoscope. The same spinal cord region was imaged in a live mouse with (b) the objective lens and (c) the microendoscope. (d) Intensity profiles corresponding to the ROIs in (b) and (c). All scale bars are 15 μm.

3.3.

Video-Rate Multimodal Nonlinear Microendoscopy

To show the compatibility of the microendoscope with other nonlinear imaging modalities, we imaged the static morphology of axons with TPEF [Fig. 4(a)] simultaneously with myelin sheaths with CARS [Fig. 4(b)] on Thy1-YFP transgenic mice. Figure 4(c) shows the overlay of both modalities. Each image is an average of 30 frames. The graph in Fig. 4(d) represents the intensity profiles of the red and green channels highlighted by the ROI in Fig. 4(c). Their mutually exclusive signal location confirms the intrinsic chemical specificity of the CARS signal for myelin structures.

Fig. 4

Video-rate multimodal nonlinear microendoscopy. Static morphology of axons with (a) TPEF of YFP-labeled axons simultaneously with (b) CARS of myelin sheaths. (c) Overlay of both modalities. (d) Intensity profile corresponding to the ROI in (c). Time-lapse imaging of a microglial cell surveying state. (e) TPEF of GFP-labeled microglia (red) and CARS of myelin sheaths (green). (f) and (g) TPEF microendoscopy at subsequent time points. Arrowheads highlight some extensions and retractions of processes. All scale bars are 15 μm. (Video 2, MOV, 0.5 MB). [URL: http://dx.doi.org/10.1117/1.JBO.17.2.021107.2]

Next, we performed time-lapse imaging of the basal dynamical behavior of microglial cells with TPEF microendoscopy [Figs. 4(e) red, 4(f), and 4(g)] along with CARS [Fig. 4(e) green] on CX3CR1-GFP transgenic mice. Its highly ramified morphology with no preferential orientation confirms that the cell illustrated in Figs. 4(e) to 4(g) was imaged in an unperturbed environment.³¹ Arrowheads highlight regions where the cell extends and retracts its processes as a function of time. The original movie, covering 33 minutes, shows this highly dynamical behavior called surveying state (Video 2). To quantify microglial motility, we measured the length changes and the velocity of individual processes (one cell, nine processes) with MTrackJ. On average, processes extended and retracted $32.27\pm 1.94\mathrm{\mu m}$ at a velocity of $1.12\pm 0.09\mathrm{\mu m}/\min$, in agreement with literature values.^32,33 Note that microglia remained in their surveillance mode, indicating that imaging through the meninges via the intervertebral space with the microendoscope did not induce noticeable inflammation.

3.4.

Live Animal Myelin Histomorphometry

In order to validate that the microendoscope is suitable for live animal myelin histomorphometry, a study on five mice was conducted. Images of myelin sheaths in the coronal plane were obtained in the dorsal column and dorsal root. A semi-automated custom-built routine retrieved all relevant morphological parameters,⁷ after being processed using the ISG procedure.

Morphological data are shown in Fig. 5. The median of the $g$-ratio histogram [Fig. 5(a)] was $0.546\pm 0.003$ (red, five animals, 56 images, 591 measurements) for axons of the dorsal column while in the dorsal root it was $0.616\pm 0.005$ (blue, five animals, 38 images, 357 measurements). It is found that those two axonal populations have a significant difference in the median of the $g$-ratio distributions (Mann-Whitney $U$-test, $p\le 0.001$). The scatter diagram of the $g$-ratio versus the axon caliber [Fig. 5(b)] reveals variations in $g$-ratio between 0.10 and 0.70 for the dorsal column (circles) and from 0.20 to 0.80 for the dorsal root (squares); their medians were separated by 0.07. Axons from the dorsal column were generally smaller in size (1 to 5 μm diameter) than those from the dorsal root (2 to 10 μm diameter). Finally, Fig. 5(c) shows the myelin thickness histogram for both axonal populations, and it is found that their medians (dorsal column: $0.88\pm 0.01\mathrm{\mu m}$ and dorsal root: $1.32\pm 0.02\mathrm{\mu m}$) are significantly different (Mann-Whitney $U$-test, $p\le 0.001$). Axons from the dorsal column had a myelin thickness on the order of 0.50 to 1.25 μm while those of the dorsal root ranged from 0.75 to 2.00 μm.

Fig. 5

Live animal myelin histomorphometry differentiates two axonal populations with small differences in $g$-ratio and myelin thickness. (a) $g$-ratio histogram. (b) Scatter plot of the $g$-ratio versus the axon diameter. (c) Myelin thickness histogram.