Imaging localized fast optical signals of neural activation with optical coherence tomography in awake mice

Abstract

We report optical coherence tomography (OCT) imaging of localized fast optical signals (FOSs) arising from whisker stimulation in awake mice. The activated voxels were identified by fitting the OCT intensity signal time course with a response function over a time scale of a few hundred milliseconds after the whisker stimulation. The significantly activated voxels were shown to be localized to the expected brain region for whisker stimulation. The ability to detect functional stimulus-evoked, depth-resolved FOS with intrinsic contrast from the cortex provides a new tool for neural activity studies.

As traditional microelectrode recordings of neural cell signals can damage brain tissue, and fluorescent tags of neural voltage activity are still being optimized and do not meet all needs, there is a role for non-contact and label-free fast optical recordings of the neural signal. Since the original study in 1949 [1], many scientists have demonstrated in vitro fast optical signal (FOS) measurements of neural activity in isolated nerves [2] and cultured neurons [3]. FOS measurements have also been demonstrated in brain slices [4] and in vivo in rats during somatosensory stimulation [5] using a high-speed CCD camera and high-sensitivity photodiodes, but have remained elusive perhaps because of a low signal-to-noise ratio.

Optical coherence tomography (OCT) is having growing impact in the neurosciences because of its ability to quantify absolute cerebral blood flow and track capillary dynamics with high spatio-temporal resolution [6] and to detect slow hemodynamic changes in response to functional stimulation that generally occurs on the time scale of several seconds [7,8]. Exploiting the highly sensitive measures of optical phase changes provided by OCT, the potential exists for using OCT to measure the FOS induced by endogenous contrast mechanisms which happens on a time scale of tens to hundreds of milliseconds [5,9]. Recent studies with OCT have shown that FOS can be observed in the sea slug abdominal ganglion [10], squid giant axons [11,12], and Drosophila central nerves [13]. Mice are widely used as a model species for functional neuroscience research and show optical property change detected by OCT during seizure activity in vivo [14], but OCT detection of the FOS in awake living mice has not been demonstrated. Here we show that OCT can measure the localized FOS in response to whisker stimulation in the cerebral cortex of awake mice.

All animal experiments were conducted following the Guide for the Care and Use of Laboratory Animals, and the experimental protocol was approved by the Institutional Animal Care and Use Committee of Boston University. To minimize the number of animals used, we prepared two 12- to 16-week old C57BL/6 mice (22–28 g, Charles River Laboratories) for the study in this Letter and used them repeatedly for our experiments. Bilateral cranial windows were made by removing the skull overlaying both hemispheres while keeping the midline intact. A custom-made PEEK head bar was attached to the remaining skull using dental acrylic, as shown in the inset of Fig. 1(a). Two half-crystal skulls (LabMaker, Germany) were used to cover the craniotomies, secured to the skull edges using glue and dental acrylic [15]. The animals were allowed to recover for two weeks; then they were trained to be head fixed for another two weeks using sweetened condensed milk as a reward before starting the experimental imaging sessions.

Fig. 1.

(a) Schematic drawing of the experimental setup. (b) Map of surface vasculature across one hemisphere of the mouse brain obtained with LSCI and overlapped with the relative change to the baseline showing the activation region; scale bar: 1 mm. The green square in (b) outlines the imaging region for the OCT angiogram (c) used for image registration. The red line in (c) shows the XY position of repeated Bscans (d) for FOS detection; scale bar in (c) and (d): 100 μm. (e) Whisker stimulation protocol.

Figure 1(a) shows the experimental setup, including a spectral domain OCT system (Thorlabs, Inc., Telesto III) and a custom-made laser speckle contrast imaging (LSCI) system. Air puff stimuli to the whiskers were delivered through a glass capillary connected by a plastic tube to a picopump (Picospritzer III, Parker, Inc.), which was controlled by the host computer. The OCT system had a broadband near-infrared light source with a center wavelength of 1,310 nm and a bandwidth of 170 nm, giving an axial resolution of 3.5 μm. For FOS detection, we used a 2× objective (NA = 0.1) that provided a lateral resolution of 10 μm. The A-line sampling rate was 76,000 A-lines/s. The LSCI system [16] was integrated with the OCT system in order to identify the responding region through the blood flow response caused by neurovascular coupling; this response is known to occur on a slower time scale than that of neural activity. A laser diode with a wavelength of 785 nm (LP785, ThorLabs) was used for LSCI illumination. The laser speckle images were acquired with a CMOS camera (acA2040-90 μmNIR, Basler) with an exposure time of 5 ms through a polarizer and a camera lens (VZM 600i, Edmund Optics). Spatial speckle contrast images were calculated as the ratio of the standard deviation of intensity over the mean using a moving 7 × 7 pixel window. The blood flow index map was obtained by taking the inverse of the squared speckle contrast [17]. We mapped the responsive brain region based on the blood flow response detected by LSCI. Briefly, the mean LSCI time course for each pixel was obtained by averaging across stimulation trials, and the relative change of LSCI was calculated dividing by the mean of the baseline. An activation image was then obtained by taking the mean of the relative change during the time period of 2–4 s after the stimulation onset [17], as shown as the pseudo color overlap in Fig. 1(b). The position at which OCT Bscans were acquired was determined by considering this activation image.

An OCT angiogram [18] was acquired [Fig. 1(c)] to register the FOS detection region of repeated OCT Bscans with the LSCI image. Figure 1(d) shows a representative OCT X-Z Bscan image acquired at the transverse position indicated by the red line in Fig. 1(c). The OCT Bscan frame rate was 193.8 Hz, which is sufficient to detect the FOS induced by the neural response to whisker stimulation expected to occur over 10–150 ms following each individual air puff stimulation [5,9]. Figure 1(e) presents the whisker stimulation protocol. During the 5 s stimulation period, 10 individual 60 ms long air puffs were delivered at 2 Hz. Note that all 35 s of LSCI data was analyzed to obtain the LSCI activation map, while only the 5 s of OCT Bscan acquisition during stimulation was used for the OCT-based FOS data processing.

For FOS data processing, we analyzed the OCT Bscan intensity response to each one of the air puff-induced whisker stimulations. There were 969 Bscans acquired for each 5-s trial (10 trials in total). To minimize motion artifacts, we tested two imaging registration methods, including global phase compensation plus image shift correction [ 19] and intensity-based image registration (MATLAB function: imregdemons). We noted that the intensity-based image registration outperforms the former method and the former method also suffers from unstable phase variation after registration. Therefore, we chose to use the intensity-based image registration method which outputs the intensity signal. Next, the mean signal time series for each voxel was obtained by averaging across the 100 individual stimuli, resulting in a three dimensional dataset with dimensions [nz, nx, nt], where, nz and nx are the number of spatial voxels in the axial and transverse directions, respectively, and nt = 95 is the number of time points. Visualization 1 shows the time series averaged over the 100 stimuli showing negligible motion artifacts after image registration.

Figure 2 illustrates a FOS data processing procedure based on the OCT signal intensity for a dataset obtained from the activation brain region identified by LSCI. A spatial mask was employed to exclude voxels with a signal variation std/mean > 2% before the stimulation onset (i.e., nt_pre = 10). This spatial masking was employed to remove voxels containing moving red blood cells that cause large signal fluctuations. The remaining voxels are expected to be within the brain parenchyma. An example of this masking is shown in Fig. 2(a). For the remaining voxels, the signal time course was fit with the response function [20,21]

$$f\left(t\right)=a{\left(t-t_0\right)}^2\cdot e^{-b\left(t-t_0\right)}\cdot s\left(t-t_0\right),$$ (1)

where a, b, and t₀ are fitting coefficients, and s (t) is the step function. In Eq. (1), t₀ represents the latency of the response, and we constrained it within the range of 20–150 ms, as it takes time for air to move along the tube (~1 m) with an additional 10 ms for the brain to respond to the movement of a whisker. The fitting accuracy was estimated by the coefficient of determination, R².

Fig. 2.

FOS signal data processing. (a) Masking: voxels satisfying the fitting criteria that std/mean before stimulation ≤2% (blue) are overlapped on the original OCT Bscan image. (b) Fitting accuracy R². (c) R² > 0.6: maximum variation of voxels having fitting accuracy greater than 0.6. (d) Representative mean OCT signal intensity change (dots) and FOS response function fit result (black solid line) of voxels marked in (c).

Figure 2(b) shows the fitting accuracy R² map overlapped on the OCT Bscan image. Voxels with good fitting accuracy (R² > 0.6) were designated as a FOS response voxel, and the maximum relative change (|ΔI/I|) map of those FOS response voxels is shown in Fig. 2(c). We noted diversity in the response time course, including both increasing and decreasing signals, different signal latencies, different peak times, and different relaxation times, as shown in Fig. 2(d). The short peak and relaxation time courses [top row of Fig. 2(d)] are similar to the FOS associated with action potentials as reported from the in vitro studies [11,12,22] and are similar to the modeling of ionic flux induced changes in neural volume inducing the optical scattering changes [21].

Figure 3 shows that OCT voxels with fitting accuracy R² > 0.6 were largely co-localized with the region of brain activation identified by LSCI. To test whether the OCT signal change potentially resulted from bulk motion, we scanned a large region with a 3 mm width covering both the center and periphery of the activated region, as shown in Fig. 3(a). Figure 3(b) shows the OCT angiogram used to register with the LSCI activation map [Fig. 3(a)]. The activation region is on the left side of the OCT Bscan in Fig. 3(b). Figure 3(c) shows the maximum relative change of “active” voxels with fitting accuracy R² > 0.6. There are more active voxels in the region corresponding to the blood flow response map in Fig. 3(a). Figure 3(d) shows the lateral variation of the total number of activated voxels along the B-scan direction (left axis) and its ratio over the total number of masking voxels at each transverse position (right axis). It is evident that there are significantly more activated voxels in the center of the activated region compared to the surrounding region. Figure 3(e) zooms in on the orange rectangle marked in Fig. 3(c) showing the maximum relative change measured at each voxel. Figures 3(f) and 3(g) present the average time courses (top panel) and the relative change resulting for all stimuli for two representative voxels marked with magenta and green circles in Fig. 3(e), respectively. We see a robust OCT intensity signal response across stimulations.

Fig. 3.

More activated voxels were within rather than outside the active region. (a) Blood flow response to whisker stimulation mapped with LSCI. The cyan rectangle in (a) shows the OCTA imaging region (b). The red lines in (a) and (b) show the repeated Bscan positions, which cover both the active region, as detected by LSCI, and the peripheral region. (c) Significantly activated voxels identified by FOS data processing. (d) Axially summed total number of activated voxels (left axis, blue dots) and the ratio of activated voxels over the total number of masked voxels (right axis, orange dots) at each X position. (e) Zoomed-in figure of the yellow rectangle in (c). (f) and (g); top panel, averaged OCT intensity time course (dots) and the fitted response function (black solid line) for the activated voxels marked in magenta and green circles in (e), respectively; bottom panel, OCT intensity change in response to all stimuli of the two voxels.

We further compared the total number of activated voxels within and outside the activated region. Smaller OCT Bscan regions (1 mm scanning width in X direction) were collected at different regions across multiple whisker stimulation experiments using the two trained mice. Figure 4(a) shows an example of the OCT repeated Bscan positions placed in the center and peripheral regions. Figure 4(b) shows the maximum relative change of activated voxels with fitting accuracy R² > 0.6 in the center region (green outline) and in the peripheral region (light blue outline), respectively. Clearly, we see that there are more activated voxels in the center region (n = 313) than in the peripheral region (n = 29). Figure 4(c) further shows the statistically significant results (p < 0.001, student t test) comparing the ratio of the total number of activated voxels normalized by the total number of considered voxels within the active region (R_resp, n = 5) and outside (R_control, n = 5) obtained from the two mice. We note that there is an approximately fourfold increase in the number of active voxels detected in the center (R_center = 10.2 ± 2.9%, mean ± std, n = 5), compared to the periphery region (R_periphery = 2.7 ± 1.2%, mean ± std, n = 5).

Fig. 4.

(a) LSCI relative change shows the responding region and regions of interest (ROIs) for FOS detection Bscan positions. (b) FOS activated voxels (R² > 0.6) within the active region [dark solid green line marked in (a)] and outside the active region [periphery, right, solid blue line marked in (a)]. (c) Statistical results of the ratio of the total number of activated voxels normalized by the total number of masked voxels between, within, and outside the active region; student t test, p < 0.001, mean ± std, n = 5. (d) Definition of t_Latency and t_Peak. (e) Histograms show the distribution of t_Latency (up) and t_Peak (bottom) of all activated voxels from all ROIs within the active region (center ROI, left axis) and outside ROIs (periphery ROI, right axis). (f) Histogram shows the distribution of maximum signal change relative to the baseline (Change_Peak) both within the active region ROIs and outside ROIs.

We analyzed the latency of the response onset (t_Latency) and the peak time from the onset (t_Peak) of the activated voxels both within and outside the active region. The top and bottom panels of Fig. 4(e) present the histograms of t_Latency and t_Peak of all activated voxels (R² > 0.6) from the five acquisitions in the active region (left axis, n_voxel = 4294) and the five acquisitions outside the active region (right axis, n_voxel = 1239). We see that the latency has a peak in the distribution at t_Latency = 41 ms. The peak time [bottom panel of Fig. 4(e)] exhibits a Gamma distribution peaked at 105 ms. From the histogram distributions, we note that the t_Latency of the peripheral region tends to shift to longer latency, while the t_Peak distributions are quite similar between the regions. We further compared the maximum change relative to the baseline, and we see a larger response within the center region compared to periphery, as shown in Fig. 4(f).

We also see in Fig. 4(f) that both positive and negative intensity variations are observed in the FOS measurements. This is consistent with the notion that the observed signal changes arise from phase shifts in the light reflected from the activated neurons caused by neural cell deformation that can randomly obtain either sign [21,23]. We estimate the phase change required to produce this observed intensity change using two phasors corresponding to the static E_s and the dynamic E_d components of the detected electric field. The phase change Δϕ of the dynamic component that will induce a 5% intensity change, $\left|\frac{{\left|E_s+E_d{e}^{i\Delta \varphi }\right|}^2-{\left|E_s+E_d\right|}^2}{{\left|E_s+E_d\right|}^2}\right|=0.05$, is obtained for the initial amplitude range |E_d|/|E_s| from 0.1 to 2, and the initial relative phase range from 0 to 2π. The average phase change required is about 0.16 rad, which corresponds with a speed of deformation of $\upsilon =0.16\text{ rad}\ast \frac{\lambda }{2\pi }/\left(t_{\mathrm{peak}}-t_{\mathrm{latency}}\right)\sim 0.5\text{ nm}/\text{ms}$. This is consistent with the speed observed in ex-vivo recordings [23].

A potential confounding factor for the FOS detection using OCT may relate to the living animal’s motion artifacts induced by bulk motion or cardiac and respiratory cycles which produce signal fluctuations that can significantly exceed the intrinsic changes produced by neural activity [13]. To eliminate this confounding factor, we acquired data from a large Bscan area which covers regions both within and outside the center of neural activation, as shown in Fig. 3, and applied rigid image registration for the whole imaging region. The result clearly shows significant correlation between the total number of activated voxels detected with OCT and the closeness to the activation center identified by LSCI, indicating that the induced OCT signal change is not due to motion artifacts that are expected to be global. The functional hyperemia response of the blood supply may also confound the FOS detection. To rule out this possibility, we used the masking process to exclude voxels with baseline fluctuation greater than 2% that would likely indicate the presence of a blood vessel. Additionally, note that the peaking time of activated voxels (80–180 ms with a peak at ~110 ms) detected in this Letter is much shorter than the typical peak time of the hemodynamic response, i.e., a few seconds [7,24], suggesting the detected FOS is not from the hemodynamic response. As discussed in the literature [2,10,25–28], we also believe that the OCT detected FOS signal change arises from cellular optical scattering change, but we believe that the OCT signal is also arising from phase changes in the reflected light producing a signal intensity change. We do not yet know if the observed changes are coming from the neural cell body, axons, synapses, or dendrites. Further work is needed to resolve this question. Ideally, future work would integrate these OCT measurements with two-photon microscopy (TPM) to compare the OCT results directly against calcium signals measured by TPM and resolved in individual cellular compartments. Our results nonetheless demonstrate the ability of OCT to robustly detect the FOS in response to whisker stimulation in awake mice.