Mapping brain networks in awake mice using combined optical neural control and fMRI

Abstract

Behaviors and brain disorders involve neural circuits that are widely distributed in the brain. The ability to map the functional connectivity of distributed circuits, and to assess how this connectivity evolves over time, will be facilitated by methods for characterizing the network impact of activating a specific subcircuit, cell type, or projection pathway. We describe here an approach using high-resolution blood oxygenation level-dependent (BOLD) functional MRI (fMRI) of the awake mouse brain-to measure the distributed BOLD response evoked by optical activation of a local, defined cell class expressing the light-gated ion channel channelrhodopsin-2 (ChR2). The utility of this opto-fMRI approach was explored by identifying known cortical and subcortical targets of pyramidal cells of the primary somatosensory cortex (SI) and by analyzing how the set of regions recruited by optogenetically driven SI activity differs between the awake and anesthetized states. Results showed positive BOLD responses in a distributed network that included secondary somatosensory cortex (SII), primary motor cortex (MI), caudoputamen (CP), and contralateral SI (c-SI). Measures in awake compared with anesthetized mice (0.7% isoflurane) showed significantly increased BOLD response in the local region (SI) and indirectly stimulated regions (SII, MI, CP, and c-SI), as well as increased BOLD signal temporal correlations between pairs of regions. These collective results suggest opto-fMRI can provide a controlled means for characterizing the distributed network downstream of a defined cell class in the awake brain. Opto-fMRI may find use in examining causal links between defined circuit elements in diverse behaviors and pathologies.

Fig. 1.

Opto-functional MRI (fMRI): mapping of distributed neural circuits causally downstream of a defined cell class in the awake mouse brain. A: left: schematic of the opto-fMRI setup for measuring distributed neural activation patterns in the awake mouse, downstream of optical perturbation of an opsin-expressing cell type. Shown is a mouse bearing an MRI-compatible head post, attached to a head post holder, and surrounded by a body restraint tube, which is in turn attached to a cradle with radio frequency (RF) coil attached (surrounding the head post; not shown for clarity; see Supplemental Fig. S1A for photograph of the RF coil). An optical fiber holder orients an optical fiber (200 μm wide) into the craniotomy, aimed at the cortical surface. The cradle is inserted into a 9.4-T MRI scanner. Top Right: boxcar protocol for delivery of a series of 15 s periods (indicated by blue bars, in this and subsequent figures) of 40 Hz pulse trains of 8 ms blue laser pulses (473 nm, 5 mW unless otherwise indicated), driven by a PC running MATLAB. Bottom right: images [echo planar images (EPIs)] are motion-corrected, slice-timing corrected, detrended, converted into percent signal change [expressed as a percent difference from the blood oxygenation level–dependent (BOLD) signal during “laser off” periods], and averaged across scans. This percent signal change is correlated, voxelwise, to the boxcar pattern of the protocol, shifted forward by 2.5 s to compensate for the BOLD signal delay to determine the voxels that are significantly changing. Significance was judged with respect to a Monte Carlo simulation at a multiple comparisons corrected P value of 0.05 for the boxcar correlation (uncorrected P < 0.005; r > 0.83). Bi: channelrhodopsin-2 (ChR2)-GFP expression, overlaid on a mouse atlas (Paxinos and Franklin 2001) in a representative mouse injected with FCK-ChR2-GFP lentivirus in left hemisphere somatosensory cortex (SI) (bregma AP −1.0 mm, ML 3 mm, DV −0.7 mm). Bii: voxels with significant increases in BOLD signal are indicated for the representative mouse (same as Bi) undergoing opto-fMRI. The image in Bii shows the voxels with significant increases in signal; voxel color indicates the boxcar correlation of the voxel; the data are shown for each of 4 0.5 mm thick EPI slices (150 × 150 × 500 μm voxel resolution), overlaid over corresponding 0.5 mm thick single slice T1 anatomical images (these slices are shown from posterior to anterior, shown from left to right). Scale bar in the figure and subsequent figures is 1 mm. Blue triangle indicates illumination site. The time course of SI activation for the representative mouse in Bii is plotted in Biii (red trace), averaged across all 4 scans of an opto-fMRI session. The black trace in Biii is the time course of SI activation (same set of contiguous voxels as for the red trace), when the laser was not on.

Fig. 2.

Positive and negative signals shown by opto-fMRI in the presence and absence of ChR2. A: voxels with significant increase in signal, colored according to boxcar correlation, for a wild-type mouse (non-ChR2 bearing) undergoing opto-fMRI under 0.5% isoflurane and a laser power of 10 mW, superimposed on the corresponding single-slice T1-anatomical image. Images were collected at 200 × 200 × 500 μm voxel resolution. B: the proportion of voxels throughout the imaged brain volume (y-axis) that possess uncorrected boxcar correlation P values below a given value (x-axis) are plotted; x- and y-axes are plotted on a log scale. Vertical dashed line indicates the threshold value of the multiple comparisons corrected P value of 0.05 (that is, uncorrected P = 0.005). C and D: voxels with significant decreases in signal, colored according to boxcar correlation, for a wild-type mouse (non-ChR2 bearing; the same mouse as in A) and for an FCK-ChR2-GFP lentivirus-injected mouse undergoing opto-fMRI (150 × 150 × 500 μm; 5 mW laser power), respectively (see Supplemental Fig. S2 for further supporting data). E: as in B and C, but for an MRI phantom and a laser power of 15 mW, in this case a 2-cm-diam plastic centrifuge tube (50 ml) filled with agarose and a LEGO brick. Shown here is a single coronal slice collected at a voxel resolution of 200 × 200 × 500 μm. Scale bar in A and C–E is 1 mm. Blue triangle in A and C–E indicates illumination site.

Fig. 3.

Unbiased algorithmic mapping of neural targets recruited by SI pyramidal cell activation in the awake mouse brain. A: population data for 5 mice undergoing awake opto-fMRI. The image in A shows the voxels with significant increases in signal in any animal. Voxel color indicates the median of the boxcar correlation, taken across all animals for which that voxel exhibits a statistically significant increase, for the voxel. The data are shown for each of 4 0.5 mm thick EPI slices (150 × 150 × 500 μm voxel resolution), overlaid over corresponding 0.5 mm thick single slice T1 anatomical images (these slices are shown from posterior to anterior, shown from left to right). Scale bar in the figure is 1 mm. Blue triangle indicates illumination site. For completeness, response maps for all 5 animals injected with FCK-ChR2-GFP lentivirus and undergoing opto-fMRI are shown in Supplemental Fig. S4, and statistical robustness is explored in Supplemental Fig. S5. Bi: plot of regions of interest that are consistently activated across mice during SI pyramidal cell activation, generated by an unsupervised k-means clustering of all sets of contiguous significantly activated voxels obtained during the opto-fMRI experiment performed in A. Each filled circle indicates the centroid of a k-means derived cluster and is connected to the sets of contiguous significantly activated voxels that make up this cluster; each of these sets of contiguous significantly activated voxels is marked by an open symbol, localized to the location of the peak correlation voxel for that set of contiguous significantly activated voxels. The shape of the open symbol indicates which animal the set of contiguous significantly activated voxels is from. K-means derived clusters were considered as regions of interest (ROIs), with key ROIs (annotated based on what they correspond to in the atlas) labeled as SI (SI barrel field), c-SI (contralateral SI), SII (secondary sensory cortex), MI (motor cortex), and CP (caudoputamen). Bii: percentage of mice that exhibit opto-fMRI activations for each of the 11 ROIs generated in the k-means clustering of Bi, rank ordered, from left to right, by reliability of observation across animals.

Fig. 4.

Opto-fMRI analysis of brain state modulation of causal network dynamics: application to characterization of anesthesia effects on SI network recruitment. A: overlay of the sets of contiguous significantly activated voxels, color-coded by ROI membership, onto T1 anatomical images, in the awake (top) and 0.7% isoflurane anesthetized (bottom) state (see Supplemental Fig. S3 for raw datasets). For each set of contiguous significantly activated voxels, a colored point is shown, localized to the location of the peak correlation voxel for that set of contiguous significantly activated voxels. Scale bar, 1 mm. B: percent change in the BOLD signal, plotted over time, for each key ROI indicated by a label in Fig. 3Bi. The BOLD signal time course was averaged across all pulse trains within a session (consisting of 4 scans of 16 pulse trains each) and across all the voxels within the contiguous significantly-activated voxels associated with each key ROI. Solid lines indicate the average of the percent signal change, taken across animals, for awake (red) and anesthetized (blue) states; shaded areas indicate the SE of percent signal change (n = 5 mice). The dashed curves in B represent scaled canonical hemodynamic response functions (HRFs) fitted to the BOLD signal time course for each of the ROIs, in either the awake (red) or anesthetized (blue) state. C: peak percent change in the BOLD signal for the experiments plotted in Fig. 4B. **P < 0.0001, factor of anesthesia level (2-way ANOVA of anesthesia level × region). D: normalized cross-correlations of the time series of percent change in the BOLD signal, taken between pairs of key ROIs in the awake (top) and anesthetized (bottom, 0.7% isoflurane) states; normalized correlations were computed for each animal, averaged across animals, and plotted on a continuous color scale. **P < 0.0001, factor of anesthesia level (2-way ANOVA of anesthesia level × pairs of ROI).

fig. 5.

Opto-fMRI applied to transgenic mice. A: voxels with significant increases in BOLD signal are indicated, color coded by boxcar correlation, for a representative Thy1-ChR2 mouse undergoing high-resolution (100 × 100 × 500 μm) opto-fMRI (here performed under 0.5% isoflurane anesthesia; blue laser pulse power, 10 mW). The data are shown for a single EPI slice, overlaid over a corresponding 0.5 mm thick single slice T1 anatomical image. Scale bar in the figure is 1 mm. Blue triangle indicates illumination site. B: voxels in SI with significant increase in BOLD signal, for 3 Thy1-ChR2 mice undergoing high-resolution opto-fMRI; the leftmost of the 3 images is a zoomed-in subpicture of Fig. 5A. C: peak percent change in the BOLD signal (x-axis) is plotted as a function of cortical depth (y-axis, measured relative to the pia). Solid line indicates the average of the percent signal change taken across the mice shown in B; shaded areas indicate the SE of percent signal change (across n = 3 Thy1-ChR2 mice). Di: epifluorescence image of a section (50 μm thick) of the brain of a Thy1-ChR2 mouse; the ChR2 is fused with YFP. Multiunit activity (MUA) recorded across neocortical layers was measured using a laminar silicon electrode array (contacts spaced, 100 μm). A representative episode of MUA recorded from an electrode contact in layer V of cortex during 1 s pulse train (40 Hz, 8 ms pulses) illumination is plotted in Dii. MUA firing rate (E) and local field potential (LFP) power (F) during the illumination period (x-axis; LFP power is normalized to the power in the 1 s preceding the illumination period) are plotted as a function of cortical depth (0–1,200 μm; y-axis). Solid line indicates the average of the MUA firing rate and the LFP power; shaded areas indicate SE (n = 2 Thy1-ChR2 mice).