2.1.

Animal Preparation

All surgical procedures, performed according to the recommendations of the Canadian Council on Animal Care, were approved by the Animal Research Ethics Committee of the Montreal Heart Institute. Nine male C57BL/6 mice ($8.9\pm 0.2$ weeks old, $23.3\pm 2.3\mathrm{g}$ weight, Charles River, Wilmington, MA) were anesthetized via intraperitoneal injections of urethane ($2\mathrm{mg}/\mathrm{g}$ body weight) in a 10% ($\mathrm{wt}/\mathrm{vol}$) saline solution. Body temperature was maintained at 37°C with a feedback controlled heating blanket (MouseSTAT, Kent Scientific, Torrington, CT).

A tracheotomy was done in order to avoid respiratory distress.³⁴ Scalp was carefully removed and a stereotaxically guided burr-hole was made over the left somatosensory cortex [co-ordinates 1 mm lateral to midline, 1 mm caudal to bregma, Fig. 1(b)] for the injection of 4-AP and electrophysiological monitoring, using a 22-gauge needle.

Fig. 1

(a) Overview of the intrinsic signal optical imaging system. LEDs and laser diode are time-multiplexed and synchronized to the acquisition system. A tungsten electrode is used to record local field potential (LFP) on the left somatosensory cortex. (b) Functional regions on the mouse cortex and seed placement and size, manually constructed from the work of Bero et al.²⁸ Dotted circle shows the placement of the LFP electrode. (c) Electrophysiology of 4-AP induced seizure. Top: example showing some ictal discharges after an injection. 4-AP injection was finished at time 0. Middle: zoom on a single ictal discharge. Bottom: expanded view showing the onset of the discharge, the intermediate phase and the offset. (d) Filtered time traces of oxyhemoglobin (${\mathrm{HbO}}_2$) and deoxyhemoglobin (HbR) at the epileptic focus [dotted circle in (b)], during the seizure.

Mice were placed on a stereotaxic apparatus and subcutaneous electrocardiogram was recorded with a single-lead electrocardiography (ECG) amplifier (amp-b01, emka Technologies, Paris). The amplified ($1000\times \text{gain}$) and filtered signal (0.2 to 500 Hz) was sampled at 1 kHz. All vital signals were digitized by a data acquisition card (NI-USB 6353, National Instruments, Austin, TX) controlled by a custom-made graphical user interface developed in LabView (National Instruments), which also controlled the image acquisition. To prevent drying of the exposed skull, a custom chamber made of bone wax was adhered to the skull with ultrasound gel and filled with mineral oil, after positioning the microelectrode.

2.2.

Epilepsy Model and Electrophysiology

Epileptiform activity was achieved according to the procedures described by Zhao et al.,^31,32 briefly described below. Focal seizures were induced by the injection of 500 nL of the ${\mathrm{K}}^{+}$-channel blocking agent 4-AP (A78403, Sigma-Aldrich, St. Louis, MO) solution at 15 mM, through a glass micropipette with a micro-syringe pump controller (uMC4, World Precision Instruments, Sarasota, FL). The volume was injected at a rate of $50\mathrm{nL}/\min$. The borosilicate glass pipette (1 mm outer diameter, 0.7 mm inner diameter, 75 mm long, World Precision Instruments, New Haven, CT) was horizontally pulled on a programmable pipette puller (P-2000, Sutter Instruments, Novato, CA).

Following injection, the glass pipette was removed and then a tungsten microelectrode (0.5 to 2 MΩ) was placed at a depth of $\sim 500\mu \text{m}$ into the cortex, in order to record extracellular LFP at the injection site. The signal was filtered between 10 and 5000 Hz, amplified 1000 times with a microelectrode AC amplifier (model 1800, A-M systems, Sequim, WA), and then digitized at 10 kHz. LFP data was further filtered between 0.2 and 130 Hz using an order 4 Butterworth digital filter in postprocessing. An example of typical epileptiform activity is shown in Fig. 1(c).

2.3.

Optical Recording System

OIS were acquired with a 12-bit charge-coupled device (CCD) camera (Pantera 1M60, DS-21-01M60-12E, Teledyne Dalsa, Waterloo, ON) with a physical pixel size on chip of 12 μm and full resolution of $1024\times 1024\text{pixels}$. The setup used in this work is depicted in Fig. 1(a).

A custom-made interface controls the camera and records images and vital signs, while synchronizing acquisition and illumination. A macro lens ($105\text{m}\text{m }\text{f}/2.8\max$, Sigma Corp., Ronkonkoma, NY) with small focal depth (350 μm) was used. Reflectance images of the brain cortex were recorded with time-multiplexed illumination (525, 590, 625 nm) produced by 10 W LEDs (LZ4-00MA00, Led Engin, San Diego, CA). Illumination for speckle imaging was provided by 90 mW, 785 nm laser diode (L785P090, Thorlabs, Newton, NJ) and aperture was adjusted to f/8, so that the pixel size and speckle size were matched. The four temporally multiplexed wavelengths led to a full-frame rate of 5 Hz. A $2\times 2$ binning on camera was performed to allow continuous data streaming to hard drive. Illumination was further adjusted so that no part of the brain was under- or oversaturated by any of the wavelengths. The exposure time of the camera was set to 10 ms. The setup was mounted on an optic table with tuned damping (RS 2000, Newport, Irvine, CA) to avoid spurious signals from vibrations.

2.4.

Optical Imaging of Intrinsic Signals

The analysis of spectroscopic images was based on previously published work.^35,36 In short, reflectance images from each LED wavelength were recorded with the CCD camera and interpreted as changes in attenuation (optical density) $\mathrm{\Delta }\mathrm{OD}=\log (I_0/I)$, where $I$ is the reflected light intensity and $I_0$ is the incident light intensity. Relative changes in ${\mathrm{HbO}}_2$ and HbR were found using the modified Beer–Lambert law³⁷ and a Moore–Penrose pseudoinverse:

The differential path length factor, $D(\lambda )$, was taken from Ref. 38 and values out of the 560- to 610-nm range were extrapolated from Ref. 39. Total hemoglobin baseline concentration of 100 μM with 60% oxygen saturation was assumed³⁸ for the spectroscopic analysis. The hemoglobin extinction coefficients were obtained from Ref. 40, and the reflectance values were corrected for the spectral response of the CCD camera and convolved with the LEDs’ spectral power distribution.⁴¹

Images of each chromophore (${\mathrm{HbO}}_2$, HbR) were spatially smoothed with a Gaussian kernel of $11\times 11\text{pixels}$ ($0.6\times 0.6\mathrm{mm}$) with a 5-pixel standard deviation (0.3 mm). An anatomical image was recorded with illumination at 525 nm, and then the region corresponding to brain was manually selected using a closed spline curve to create a brain mask. All further processing was performed only on those pixels belonging to the brain mask.

2.5.

Speckle Contrast Imaging

In addition to imaging changes in hemoglobin concentrations, changes in CBF were computed by laser speckle contrast imaging. Speckle originates from the random interference of multiple backscattered coherent light. In the presence of moving scatterers the interference field fluctuates, creating intensity variations, known as speckle patterns.^38,42,43 When integrated over the exposure time of the camera, the speckle pattern becomes blurred in areas of increased blood flow. Images of blood flow were obtained by measuring the spatial contrast of the speckle $C$, defined as the ratio of the standard deviation to the average intensity $\sigma /\langle I\rangle$. The contrast is a function of the CCD camera exposure time $T$ and it is related to the correlation time ${\tau }_C$ of the speckle, which is assumed to be inversely proportional to the speed of the scattering particles.⁴⁴ A window of $5\times 5\text{pixels}$ ($0.3\times 0.3\mathrm{mm}$) was used to compute the speckle contrast images and the relative changes in blood flow were obtained with the following formula:

Since $\mathrm{\Delta }v/v_0$ laser speckle imaging underestimates CBF by $<5\%$,⁴⁵ in this work both quantities were assumed to be equal. Flow images then underwent the same spatial filtering as the hemoglobin images, and images from all three different contrasts were further resized to one quarter of the original size due to memory constraints.

2.6.

Cerebral Metabolic Rate of Oxygen

Changes in the cerebral metabolic rate of oxygen consumption (${\mathrm{CMRO}}_2$) were calculated from the images of changes in CBF, total hemoglobin (HbT) and HbR using the steady-state relationship^46,47:

A central hypothesis of this relation is the absence of transients with decoupled hemodynamic components. Here, since resting-state networks were identified from data filtered between 0.009 and 0.08 Hz (see below), we hypothesized that for these networks, this relationship was maintained. The constants ${\gamma }_R$ and ${\gamma }_T$ were both assumed to be 1, which is within a physiologically plausible range (0.75 to 1.25).⁴⁶

2.7.

Functional Connectivity

In this paper, we followed the procedure outlined by White et al.²⁹ to produce seed-based correlation maps as a method for studying one aspect of functional connectivity; the steps are briefly described below.

Multimodal recording sessions (optical and electrical) of 15 min were carried out in resting-state conditions: Two sessions were performed before the injection of 4-AP and two more consecutive sessions were carried out $\sim 5\min$ after the release of the epileptogenic compound in the somatosensory cortex, giving a total of four sessions per mouse. Time courses of every pixel were temporally band-pass filtered (zero phase-shift fourth-order Butterworth filter) at 0.009 to 0.08 Hz, according to previous functional connectivity studies.^28,29 After temporal filtering, each pixel time trace was downsampled from 5 to 1 Hz.

A global brain signal was created from the average of all the pixel time traces. In order to account for coherent variability common to all pixels, this global brain signal was regressed from each pixel time course, using a general linear model from the package statistical parametrical mapping⁴⁸ (SPM8, www.fil.ion.ucl.ac.uk/spm) running on a Matlab (The MathWorks, Natick, MA) platform.

All seeds were manually placed a priori using the coordinates corresponding to left and right frontal, cingulate, motor, somatosensory, cingulated, and visual cortices,²⁹ as shown in Fig. 1(b). Seed time-courses were computed as the mean time course of the pixels within a 3.5-pixel (0.2 mm) radius from the seed locus. An example of seed time-trace during seizure is illustrated in Fig. 1(d). This seed is placed at the epileptic focus and its location is indicated with a dotted circle in Fig. 1(b). An increase in ${\mathrm{HbO}}_2$ and a correspondent decrease in HbR is observed at the seizure onset.

With the unilateral application of 4-AP, we expect bilateral differences between brain regions so the metric used to evaluate functional connectivity was a regional bilateral functional correlation, defined as the correlation between each seed time course and its contralateral homologue, yielding six values for each mouse. The Pearson’s coefficient $r$-values were converted to Fisher $Z$ measures using $Z(r)-\frac{1}{2}\ln [(1+r)/(1-r)]$ before performing the random effect rank sum tests. Since the assumptions for normal distributions might not hold in this study, significance was determined by the Wilcoxon rank sum test. $P$ values were corrected for multiple comparisons using false discovery rate (FDR) adjustment. Adjusted values were considered significant at $p<0.05$.

3.1.

Identification of Functional Networks

Functional networks were identified prior to 4-AP injection and in agreement with previous literature²⁹ across mice for ${\mathrm{HbO}}_2$ (see Fig. 2 for representative maps of four sample mice) albeit with a stronger lateralization of some networks.

Fig. 2

Seed-based ${\mathrm{HbO}}_2$ correlation maps for four mice. One control session and one post–4-AP injection session are displayed for each mouse. (F: frontal cortex, M: motor cortex, C: cingulate cortex, S: somatosensory cortex R: retrosplenial cortex, V: visual cortex; subscripts L and R refer to left or right hemisphere, respectively.) The scale for all correlation maps is from $r=-1$ to 1. Maps are shown overlaid on the anatomical image of the brain, acquired with green light. Seeds placement and sizes are indicated with black circles.

With multispectral acquisition, the same procedure was also performed for HbR and CBF estimated from speckle imaging; no significant change in spatial network organization was observed (data not shown). In one experiment, we performed the network analysis using a higher frame rate and compared with a similar analysis performed by dropping camera frames to investigate the effect of acquisition frame rate. No significant difference was observed confirming that the chosen 5-Hz acquisition rate was adequate (data not shown).

3.2.

Changes in Functional Connectivity After Injection of 4-AP

Out of nine mice, five showed clear seizure-like activity (see Table 1), while four showed high-intensity spiking activity. Seizure-like activity was characterized by fast rhythmic spiking activity of increasing amplitude and decreasing frequency, evolving into rhythmic spike and slow wave activity prior to gradual offset [see an example in Fig. 1(c)], while spiking activity presents fewer than 10 spikes per burst (data not shown). All mice were combined in the analysis of this section. To assess the impact of acute seizures and epileptiform activity on resting-state networks, resting-state sessions acquired following the injection of 4-AP were analyzed following the same methodology as above.

Table 1

Seizure data.a

The correlations between homologous seeds were investigated and compared before and after the injection of neurotoxin. Data extracted from the ${\mathrm{HbO}}_2$ time course showed a significant difference in retrosplenial seeds [Fig. 3(a)]. For HbR contrast, there was no significant difference in any paired seeds’ time trace [Fig. 3(b)], although a decrease in functional connectivity was seen on the somatosensory region, a trend also seen in flow and ${\mathrm{CMRO}}_2$ data. Seeds of the cingulate, somatosensory, and retrosplenial cortex saw significant changes when analyzed with CBF contrast [Fig. 3(c)]: a decreased bilateral functional connectivity was observed in the somatosensory seeds while increased connectivity was seen in the cingulate and retrosplenial regions.

Fig. 3

Regional bilateral functional correlation before and after the 4-AP injection, analysis done for every seed time-trace and its contralateral part. Contrasts shown: (a) ${\mathrm{HbO}}_2$, (b) HbR, (c) cerebral blood flow (CBF), and (d) cerebral metabolic rate of oxygen consumption (${\mathrm{CMRO}}_2$). $*p<0.05$, FDR corrected. Standard error bars shown ($\sigma /\surd N$), with $N=9$.

To assess whether these hemodynamic changes were associated with metabolic consumption patterns, we computed ${\mathrm{CMRO}}_2$ using steady-state formulas given the low-frequency fluctuations frequently documented in functional connectivity studies.^10,49–51 The data shows no significant differences [Fig. 3(D)]. However there was a decrease in functional connectivity in the somatosensory and retrosplenial cortex, albeit not significant after FDR adjustment. All results were evaluated with a Wilcoxon rank sum test.

Functional networks were derived from low-frequency filtered imaging data. We were also interested to measure correlation between raw signals, before and after seizures, to observe the degree of disruption in the whole signal prior to regression by the common signal (data not shown). The data shows that all seeds are positively correlated, indicating the presence of a common signal to all the pixels’ time course during epileptic events, hence the need to regress a global brain signal, obtained from the average time trace of all the pixels marked as belonging to the brain.

3.3.

Correlation Between Seizure Duration and Functional Connectivity

Restricting the analysis to the mice showing clear seizure activity, we aimed to investigate whether seizure duration impacted network changes. To calculate the changes of bilateral correlation, we subtracted the bilateral correlation during control sessions, $z_0(r)$, from the bilateral correlation during the post 4-AP injection sessions, $z_{4\mathrm{AP}}(r)$, and plotted this change versus the average seizure duration of the 4-AP sessions, as illustrated in Fig. 4. Overall correlations between seizure duration and changes in bilateral correlations were moderate. For the somatosensory cortex, where the toxin was injected, there was a positive correlation between the duration of the seizure and the changes in bilateral correlation. This same positive correlation was observed from the data extracted from the frontal cortex seeds. For other cortical regions, i.e., motor, cingulate, retrosplenial, and visual, a reverse effect was observed: the changes were negatively correlated with the seizure duration. For CBF contrast, the correlation showed to be significant in the somatosensory ($r^2=0.52$) and retrosplenial ($r^2=0.75$) regions. HbR contrast had a significant correlation for retrosplenial seeds ($r^2=0.50$), while ${\mathrm{HbO}}_2$ displayed a moderate (not significant) correlation for the motor seeds ($r^2=0.41$). The average length of seizures per subject was $68.7\pm 51\mathrm{s}$. All recording sessions were 863 s long. Table 1 provides the duration of seizure-like activity for every recording session, the percentage of the recording session that belongs to seizure-like activity, and the average duration of seizures per subject.

Fig. 4

Changes in bilateral functional correlation plotted versus seizure duration (expressed as a percentage of the recording session) for different cortical regions and different contrasts.

4.1.

Acute Seizure Activity Leads to Both Decrease and Increase in Functional Connectivity

In the case of ${\mathrm{HbO}}_2$, a significant decrease of functional connectivity was observed in retrosplenial cortex. However, the difference between control and post-4-AP injection were not significant in HbR networks. In blood flow contrast, a significant increase of functional connectivity was observed in both cingulate and retrosplenial paired seeds, and there was a significant decrease for somatosensory cortex.

Previous literature using fMRI^11,20,21,24 mostly documented reduced functional connectivity in several functional networks in epileptic patients measured at rest. While a clear distinction is required with current experiments which focused on resting-state networks during acute epileptiform events, the picture emerging from this work indicates a complex interplay between the different components of the hemodynamic response leading to both increases and decreases in connectivity in our mouse model. While HbR, which closely correlates to BOLD-fMRI, saw a nonsignificant decrease in connectivity in most seeds ($4/5$), ${\mathrm{HbO}}_2$ and CBF saw their bilateral correlations move in opposite directions in seeds closest to the injection site (somatosensory and retrosplenial, for which changes were significant). Hemodynamic component dependent correlation changes indicate a potential uncoupling of flow and metabolism, even in low-frequency networks.

These observations suggest that using a single modality, e.g., fMRI BOLD, to investigate resting-state network in epilepsy may remain limited without controlling for flow or oxygenation. Whether the observed changes are due to changes in physiology (however, no significant change was observed on our ECG), high metabolic demand at the seizure location, or neural activity remains to be investigated. Further work with a chronic model investigating networks outside of epileptic periods may help reveal more coherent changes across hemodynamic components.

4.2.

Correlation Between Electrophysiological Activity and Resting-State Correlations

Since electrophysiology was measured simultaneously, we could investigate the effect of seizure duration on pairwise network correlations. Our results showed a dependence of the changes in bilateral functional correlations on seizure duration but remained mitigated due the low values of $r^2$ obtained in this analysis. Whether these changes are also present in subacute conditions, during simple spiking events, and the cause for network changes observed in patients remain to be studied. Our results are further limited by the fact that electrophysiology was measured in the somatosensory cortex; it may be that similar activity measured at other seeds may lead to a different outcome. Keeping the skull intact apart from the injection site allowed studying an almost intact cortex but the absence of multiple LFP recording sites implied that the spatial dependence of the LFP response remained a potentially confounding variable that was not investigated.

4.3.

Metabolic Consumption Reflects Hemodynamic Changes

${\mathrm{CMRO}}_2$ is potentially the closest marker to neuronal activity. Here, we hypothesized that the steady-state relation between hemodynamic signals to estimate ${\mathrm{CMRO}}_2$ held for the resting-state networks due to their low-frequency support. Assuming this hypothesis to be valid and comparing network correlation changes obtained by ${\mathrm{CMRO}}_2$, our analysis displayed changes that followed similar trends as changes in HbR, except for the visual and frontal seeds where small differences were observed. This observation may in turn support HbR-weighted fMRI-BOLD as a proxy for resting-state network in epilepsy despite the known physiological confounds and hemodynamic network uncoupling displayed here.