FMRI and intra-cranial electrocorticography recordings in the same human subjects reveals negative BOLD signal coupled with silenced neuronal activity

Two patients underwent implantation of subdural electrode grids (Adtech, Racine, USA, interelectrode spacing: 1 cm, electrode diameter: 2.3 mm; Data recording system for participant 1 (P1): 128 channels, 512 Hz sampling rate, 0.15–134.4 Hz bandpass filter, Micromed, Italy; Data recording system for participant 2 (P2): 128 channels, 2048 Hz sampling rate, 0.15–500 Hz bandpass filter, Micromed, Italy) to determine the site of epileptic foci for the purpose of possible surgical removal of the epileptogenic tissue. Implanted grids extended to healthy tissue in the occipital pole of the left hemisphere for P1 and the right hemisphere for P2. The patients gave written informed consent to participate in the study in compliance with the Declaration of Helsinki 2013. The experimental protocol was approved by the medical research ethics committee of the UMC Utrecht.

The two participants underwent pre-operative and post-operative fMRI scanning, respectively. Data were acquired at a Philips Achieva 3T scanner (Philips Healthcare, Best, Netherlands), using a 32-channel head coil and 3D PRESTO (Neggers et al. 2008) with the following parameters: FOV 99X256X256mm, resolution of 3 mm isotropic, 33 slices, flip angle: 10°, TE/TR: 38.7/27 ms. Functional volumes were acquired every 810 ms (Neggers et al. 2008), and run duration was 3 min. Anatomical T1-weighted (T1-w) data were acquired using an 3D MPRAGE sequence (number of excitations per inversion 180; TR/TE 10 ms/4.6 ms; flip-angle 8°; FOV 240 × 240 × 160 mm; 200 slices, 0.8 mm isotropic voxel size; total scan duration 602 s).

Avoiding a central circular region (0.4° of visual angle) and displaced by 20° of polar angle from the vertical meridian (Fig. 1A). The stimulus radius was 10° of visual angle. This size (10° of visual angle) was computed based on the distance between the participant’s point of view and the screen, as well as the screen size. 10° were determined by the fMRI environment. We could go higher when measuring ECoG but we were constrained by the fMRI environment. The same stimulus properties were adopted for ECoG and fMRI measurements. The dart-board consisted of a rectangular grid with a spatial frequency of 0.5° of visual angle. Each spoke of the dart-board pattern moved coherently in opposite radial directions with a temporal frequency of 10 Hz. Each stimulus lasted for 810 ms and left and right visual field stimulation was alternated with an interleaving 15 s baseline (gray screen). The visual stimulus in the ECoG setting was identical to the fMRI paradigm, but stimulus duration was shorter. For the ECoG recordings, each stimulus lasted 500 ms with an inter-stimulus interval of 2 s. The participants fixated on the center of the screen and were instructed to maintain stable fixation. Different stimuli and baseline durations were adopted for the fMRI and ECoG measurements. This is due to the different duration of the vascular and neuronal responses measured by two techniques, respectively. fMRI is characterized by a relatively slow hemodynamic response function that can take up to 20 s to return to baseline after a short impulse stimulation (Glover 1999). On the other hand, direct neuronal recordings, such as ECoG measurements, are characterized by a rapid response to short visual stimulation (few ms) and a rapid return to baseline (Gaglianese et al. 2017a, b). 12 stimuli were presented during each fMRI run, 6 on the left hemifield and 6 on the right hemifield. 30 stimuli were presented during one ECoG run, 15 on the left hemifield and 15 on the right hemifield. In P1 we acquired the fMRI run before the ECoG run, vice versa for P2.

Electrodes with epileptic artifacts, as determined by trained neurologists, were removed from further analyses, and signals were re-referenced to the common average of all remaining electrodes. For each participant, power spectral density (1–134 Hz) was estimated for each trial every 1 Hz by Welch’s periodogram, averaging with a 1 s window. Active trials were defined as the 500 ms of visual stimulation during contralateral or ipsilateral conditions. Rest epochs started 500 ms after the stimulus offset and lasted 1 s. Mean responses in the high frequency band (HFB; 65–95 Hz), alpha band (9–13 Hz) and beta band (14–30 Hz) (Hermes et al. 2012; Gaglianese et al. 2017a, b) were extracted for both conditions for further statistical analysis.

Electrodes exhibiting significant responses for the contralateral condition were selected by statistically comparing the mean responses in the HFB frequency range to the mean power of the rest epochs (paired t test, p < 0.05 and average spectral power difference > 0). Overall, following the described criteria, n = 33 and n = 27 electrodes were selected in in P1 and P2, respectively. In addition, for each selected electrode, we computed mean responses in alpha and beta power frequency band for both contralateral and ipsilateral conditions. The responses of each frequency band were converted to z-scores by subtracting the mean response across rest epochs from the mean responses (averaged across trials) for each condition and dividing by the standard deviation of the mean of the rest epochs. Significance across all selected electrodes with respect to the null hypothesis of no response was tested using t tests for each frequency band (p < 0.05, Bonferroni corrected).

In addition, to assess the mean HFB response over time during visual stimulation we filtered the data between 65 and 95 Hz using a 3rd order Butterworth filter in two directions (Hermes et al. 2012) and we calculated the smoothed log power of the analytic amplitude using the Hilbert transform.

To assess the robustness of the results, we varied the frequency range of the measured HFB power changes in the contralateral and ipsilateral visual stimulation during ECoG measurements. We opted for three different frequency ranges: low-gamma (31:64 Hz) high-gamma (65:130 Hz), and all-gamma (31:130 Hz). We computed the number of electrodes exhibiting significant responses for the contralateral condition within each frequency range (paired t test, p < 0.05 and average spectral power difference > 0). Moreover, within each frequency range, we assessed the response HFB response in the ipsilateral condition. Significance across all selected electrodes with respect to the null hypothesis of no response was tested using t tests for each frequency range (p < 0.05, Bonferroni corrected).

Functional data was motion corrected using the function 3dvolreg in AFNI (https://​afni.​nimh.​nih.​gov, Cox., 1996). The average motion corrected functional volume was computed and used as a reference for co-registration with the T1-w anatomical image. Co-registration was performed using the function 3dAllineate in AFNI, with mutual information as cost function.

Functional data was analyzed in native space with a standard GLM using a canonical hemodynamic response function (HRF). To extract the shape of single voxel HRFs a set of basis functions was fit to each voxel via the 3dDeconvolve function in AFNI. Voxels were selected for further analysis based on the contrast of contra-lateral minus ipsi-lateral condition (T values > 2, which corresponds to p < 0.05, uncorrected, see Fig. 2C–F, this value was only used for voxel selection to extract the global HRFs reported in Fig. 2D, F). We performed a bootstrapping analysis of all the selected voxels to estimate the shape of the contra-lateral and ipsi-lateral HRF (Fig. 2D, F; Efron and Tibshirani 1994). It is important to note that these HRFs are representative of all the selected voxels based on the contrast between contra-lateral and ipsi-lateral responses and do not reflect the response of a single voxel or below a single electrode, for this reason we refer to global HRF (see Fig. 2, caption). Selected voxels were bootstrapped 5000 times with replacement. For each iteration, the average HRF was computed for the contra-lateral and ipsi-lateral stimulation conditions. The HRFs obtained were fit using a double-gamma hemodynamic response function with three free parameters: amplitude, delay of the response relative to onset and delay of the undershoot relative to onset. Fitting was performed using non-linear regression in R (R Development Core Team 2010) and we computed the 95% bootstrapped confidence intervals of the fit (Fig. 2D, F).

The location of each implanted electrode was determined on a post-operative CT scan using an automatic clustering detection algorithm as described in Branco et al. (2018). The T1-w anatomical images were segmented automatically using FreeSurfer (https://​surfer.​nmr.​mgh.​harvard.​edu). White matter and pial surfaces were generated in Freesurfer and then imported in SUMA (https://​afni.​nimh.​nih.​gov). After co-registering the CT scan to the anatomical scan, thereby bringing ECoG and fMRI data in the same space, electrode coordinates were projected on the closest vertex on the cortical surface in the anatomical T1-w scans.

For each participant, electrodes responding to the task were selected based on ECoG HFB responses to contralateral stimulation. Power increase in the HFB has been directly associated to spiking activity (Fries et al. 2007; Miller et al. 2009; Hermes et al. 2015), and it provides a precise measure to localize active brain regions during visual stimulation.

To match the ECoG measurements with the BOLD fMRI data, the fMRI signal under each selected electrode was extracted from gray matter voxels within 3 mm (radius) around the electrode center.

HRFs corresponding to the contra-lateral and ipsi-lateral stimulation conditions were extracted from the voxels associated with each selected electrode (see “EcoG—analysis” section for electrodes selection and Fig. 3 for an example) via deconvolution using the hemodynamic response estimation toolbox (https://​stat.​columbia.​edu/​~martin/​HRF_​Est_​Toolbox.​zip) and using the finite impulse response model (Miezin et al. 2000). A total of 18 timepoints was estimated after each stimulus, corresponding to 15 s (equivalent to the inter-stimulus interval).

The association between BOLD (PBR/NBR) and ECoG (HFB, alpha and beta power) at the single participant level was tested using a multivariate general linear model against the null hypothesis that no association was present. For each electrode that was selected (n = 33 and n = 27 in P1 and P2, respectively), the contralateral and ipsilateral response was derived from the deconvolved HRF and entered in a multivariate general linear model associated with ECoG mean power responses (HFB, alpha and beta). The same association across all the selected electrodes in P1 and P2 was tested using a generalized linear model with participants as a random effect (participant was coded as a categorical variable), accounting to some degree for variation between the two participants that took part in the experiment. Statistical analysis was performed in R (R Development Core Team 2010).

We have obtained the location of the main ROIs along the visual hierarchy for each participant (P1 and P2) using the atlas provided in Wang et al. (2015) and the standardized surfaces in SUMA (Saad and Reynolds 2012).

For each participant, we have assigned each electrode to one surface-based ROI, based on the closest geodesic distance over the surface between the electrode and the surface-based ROI.

Wang et al. (2015) atlas provides within-ROI subdivisions between: ventral and dorsal V1/V2/V3, ventral/occipital areas 1 and 2 (Brewer et al. 2005), lateral occipital area 1 and 2 (Larsson and Heeger 2006), areas V3a and V3b (Press et al. 2001) and temporal occipital area 1 and 2 (Amano et al. 2009).

To facilitate visualization, we collapsed between ventral and dorsal V1/V2/V3, ventral/occipital areas 1&2, lateral occipital area 1&2 and temporal occipital area 1&2.

ECoG and fMRI data were acquired in separate sessions from two participants, while they were presented with a unilateral visual stimulation (Fig. 1). The participants underwent ECoG grid implantation for the purpose of epilepsy monitoring, and grids extended to healthy tissue in the occipital pole, the grids were placed to cover the calcarine fissure as well as the close surrounding cortex on the dorsal and lateral portion of the occipital pole (see Fig. 2A, B for a schematic of the electrode placement for each individual participant, P1 and P2).

Contralateral visual stimulation during ECoG measurements did elicit a significant increase in the HFB power range (65–95 Hz) in N = 33 and N = 27 electrodes in P1 and P2, respectively. No significant responses were detected for the ipsilateral stimulation in this frequency range (one sample t test, t(32) = 2.24, n.s. and t(26) = 1.96, n.s, for P1 and P2, respectively, Bonferroni corrected, see Figs. 3C, F and 4A, D).

In both participants we observed a power increase in the HFB for the contralateral stimulation and a stimulus contrast reversal peak synchronous with the visual stimuli temporal frequency rate of 10 Hz and its harmonics (Fig. 2G, H), as previously reported (Gaglianese et al. 2017a, b). No significant power changes in the HFB range were detected for the ipsilateral stimulation for both participants (average power spectra across selected electrodes shown in Fig. 2G, H). All the t tests that follow regarding alpha and beta values responses were computed on the distribution of z-scores across all the selected electrodes that exhibit positive HFB power for contralateral visual stimulation (see “ECoG—analysis”).

For the contralateral visual stimulation, we measured an alpha power decrease for P2 among the selected electrodes (t(26) = − 4.53, p < 0.001, Bonferroni corrected). The same comparison did not reach statistical significance for P1 (average z-score across the selected electrodes = − 0.57, one sample t test, t(32) = − 1.57, n.s, Bonferroni corrected).

For ipsilateral visual stimulation we measured an alpha power decrease with respect to baseline in both participants (one sample t test, t(32) = − 5.71, p < 0.001 and t test, t(26) = − 4.53, p < 0.001, for P1 and P2, respectively, Bonferroni corrected, Fig. 4B, E).

Results for the beta band appeared to vary between participants. For P1, we measured a small but significant beta power decrease for the ipsi-lateral condition among the selected electrodes (average z-score across the selected electrodes = − 0.3, one sample t test, t(32) = − 4.51, p < 0.01 Bonferroni corrected, Fig. 4C) but not for the contra-lateral condition (average z-score across the selected electrodes = 0.25, one sample t test, t(32) = − 1.54, n.s, Bonferroni corrected). For P2 no significant power changes were observed in the beta-band for either condition (Fig. 4F).

Robust positive and negative BOLD fMRI signals were elicited in early and extra-striate visual cortex for the contralateral and ipsilateral stimulation conditions, respectively (P1, see Fig. 2A, C, D; P2, see Fig. 2B, E, F). The average shape of the positive and the negative hemodynamic response functions (HRFs) for cortical locations (voxels) significantly responding to the fMRI stimulus are displayed in Fig. 2D, F (global HRF, locations selected based on the contrast of contra-lateral minus ipsi-lateral condition). Overall in fMRI, contra-lateral stimulation elicited PBR, whereas the ipsi-lateral condition elicited NBR, as previously reported (Tootell et al. 1998; Gouws et al. 2014; Fracasso et al. 2018).

We analyzed the amplitude of the HRF extracted in the cortical locations corresponding to each selected electrode of each participant (N = 33 and N = 27 electrodes in P1 and P2, respectively, see “ECoG—analysis”). HRF for both contralateral and ipsilateral condition for a representative electrode of each participant was shown on Fig. 3A, B–D, E. During contralateral visual stimulation we observed PBR (one sample t test, t(32) = 6.72, p < 0.001 and t(26) = 4.57, p < 0.001, for P1 and P2, respectively, Bonferroni corrected, Fig. 4A, D), and NBR during ipsilateral visual stimulation (t test, t(32) = − 15.6, p < 0.001 and t(26) = − 4.58, p < 0.001 for P1 and P2, respectively, Bonferroni corrected see Fig. 4A, D).

To investigate the correspondence between fMRI BOLD responses and the neurophysiological responses as measured by ECoG we compared the z-score responses in the HFB, alpha and beta band and the correspondent PBR and NBR amplitude peak responses derived from the deconvolved HRF in the fMRI measurements for each selected electrode (see “ECoG–fMRI co-localization and local HRF response”).

Contralateral stimulation elicited PBR in the voxels underneath the electrodes (Fig. 3B, E), which were associated with an HFB power increase for the duration of stimulation (500 ms), starting few milliseconds after stimulus onset (Fig. 3C, F). In the ipsi-lateral condition, NBR measured underneath the electrodes (Fig. 3B, E) were linked to HFB power just around zero for the duration of stimulation (Fig. 3C, F). A summary of the relationship between ECoG measurements in the HFB, alpha and beta band and the BOLD responses for each selected electrode is reported in Fig. 4.

To quantify the relationship between the neuronal responses and the BOLD responses we performed three separate multivariate analyses across all selected electrodes, for both the contralateral and ipsilateral stimulation. In the first analysis we tested individually for each participant whether PBR could be explained by HFB, alpha and beta power in the contra-lateral condition. In the second analysis we tested NBR against HFB, alpha and beta power in the ipsi-lateral condition (see Table 1, P1 and P2). In the third analysis we pooled all the recording sites in a single data set and accounted for individual participant variability using a generalized linear model with participants as a random factor (multilevel approach, see Table 1, P1 and P2).

Given the small sample in the data set (two participants), we opted for reporting results at the individual participant level as well as with the multilevel approach.

We observed a positive linear relation between PBR and HFB responses for contralateral stimulation as well as a negative linear relationship with alpha power for the same condition among the selected electrodes (see Table 1). NBR in the ipsilateral condition did not show a clear linear association with ECoG measurements among the selected electrodes (Table 1).

To provide a descriptive analysis of electrode location and activity along the visual hierarchy we have obtained the location of the main visually responsive ROIs for each participant (P1 and P2, see Fig. 5) using the atlas provided in Wang et al. (2015), collapsing any within-ROI subdivision and the standardized surfaces in SUMA (Saad and Reynolds 2012).

PBR is associated with the contra-lateral condition and NBR is associated with the ipsi-lateral condition, throughout the visual hierarchy. We observe limited NBR responses in contralateral visual stimulation that tend to be spread along the cortical hierarchy without a specific pattern. We interpret these responses as reflecting a mix of activity between surround suppression effects due to visual stimulation at the edge of population receptive field centers (pRF, Dumoulin and Wandell 2008; Fracasso et al. 2016a, b; Zuiderbaan et al. 2012), and ‘blood stealing’ from neighboring portions of stimulated cortex (Shmuel et al. 2006).

Alpha increases, when present, were observed in electrodes located in early visual cortex (V1–V3). Alpha decreases were detected across all ROIs for both stimulus conditions, although significant alpha decreases between the two participants (P1 and P2) were observed only in the ipsi-lateral condition.

To assess the robustness of the HFB results presented, we varied the frequency range of the measured HFB power changes in the contralateral and ipsilateral visual stimulation during the ECoG measurements. We opted for three different frequency ranges: low-gamma (30:64 Hz) high-gamma (65:130 Hz), and the all-gamma band (all-gamma: 30:130 Hz), see Fig. 6.

For each frequency range and participant, results were very similar to those observed with the frequency range 65–95 Hz reported in the section “ECoG responses to contralateral and ipsilateral stimulation”. Contralateral visual stimulation elicited a significant increase in the HFB power, while no significant responses were detected for the ipsilateral stimulation in this frequency range.

For P1, contralateral visual stimulation elicited a significant response in 33, 32 and 33 individual electrodes, for the all-gamma, high-gamma and low-gamma frequency ranges, respectively. No significant HFB responses were detected for the ipsilateral stimulation (one sample t test, t(32) = 1.61, n.s., t(31) = 2.18 n.s., and t(32) = 0.45 n.s., for the all-gamma, high-gamma and low-gamma, respectively.

For P2 results were virtually indistinguishable from P1, contralateral visual stimulation elicited a significant response in 27, 27 and 27 individual electrodes, for the all-gamma, high-gamma and low-gamma frequency ranges, respectively. No significant HFB responses were detected for the ipsilateral stimulation (one sample t test, t(26) = 1.61, n.s., t(26) = 2.05, n.s., and t(26) = − 0.12 n.s., for the all-gamma, high-gamma and low-gamma, respectively.

Several processes have been proposed to account for NBR and it is likely that different mechanisms are responsible, depending on the specific experimental conditions (Goense et al. 2012a, b). Initially, NBR was believed to be the consequence of a decrease in cerebral blood volume (CBV) due to neighboring positive BOLD activity, referred to as the blood-stealing effect (Harel et al. 2002). This mechanism can account for negative BOLD signal located close to positive BOLD signal.

Further experiments have shown that NBR could be observed following a reduction of neuronal activity and a decrease in cerebral blood flow (CBF) (Shmuel et al. 2002, 2006; Devor et al. 2008; Boorman et al. 2010; Goense et al. 2012a, b), suggestive of a completely different neurovascular coupling mechanism compared to the blood-stealing effect. In our experiment, PBR and NBR were extracted from the same location in the same hemisphere, which agrees with the latter.

In the literature, NBR are also observed in the sensory-motor domain. During unilateral hand movement, the contralateral cortex is activated, showing increases in HFB power as well as PBR. However, in the ipsilateral side NBR is routinely observed (Devor et al. 2008; Kastrup et al. 2008; Schafer et al. 2012). Interestingly, in this case, ipsilateral NBR co-occurs with neuronal excitation as well as trans-callosal inhibition (Nass 1985; Allison et al. 2000). This pattern of results has been attributed to active inhibition by inhibitory GABA interneurons in the ipsilateral cortex (Devor et al. 2008; Kastrup et al. 2008; Schafer et al. 2012). The combination of excitation and inhibition in the ipsilateral side could vary and was dependent on the task demands (i.e., the movement rate). However, recent evidence suggests that activation of inhibitory neurons can increase local cerebral blood flow independently of net ongoing neuronal activity (Anenberg et al. 2015).

Moving to the visual domain, ipsilateral NBR has been observed previously (Tootell et al. 1998; Gouws et al. 2014; Fracasso et al. 2018), and was also shown to be dependent on task demand, with attention-demanding tasks on the contralateral stimuli leading to stronger ipsilateral NBR (Gouws et al. 2014). Overall, a variety of different sources have been identified to be responsible for the observation of NBR and these sources have been shown to differ depending on the exact task demands.

As expected, our results showed a positive correlation between PBR and HFB responses elicited by contralateral visual stimulation among the selected electrodes. PBR is known to be coupled with high-frequency electrophysiological responses observed in intra-cortical recordings (Logothetis et al. 2001), ECoG (Lachaux et al. 2007; Siero et al. 2014; Gaglianese et al. 2017a, b), MEG (Brookes et al. 2005) and EEG (Ball et al. 2008; Mulert et al. 2010). Moreover, PBR were correlated with decreased alpha power in the contralateral condition, consistent with multiple studies combining fMRI and EEG (Goldman et al. 2002; Laufs et al. 2003; Feige et al. 2005; Scheeringa et al. 2009) and with the idea that alpha activity reflects functional inhibition (Jensen and Mazaheri 2010).

In the current investigation NBR measured during ipsilateral visual stimulation were accompanied with an absence of HFB power response measured by ECoG. In contrast, an earlier study showed an association between a decrease in BOLD signal and decreased neural firing rate in multi-unit activity below spontaneous activity (Shmuel et al. 2006; Boorman et al. 2010). The difference between our and earlier findings could be explained by the type of electrodes adopted in ECoG compared to animal neurophysiology. Indwelling electrodes are used in the latter case, which measure responses from one or several neurons, whereas in the former electrodes are placed over the brain surface, which sample from several hundreds of thousands of neurons. If a relatively small number of neurons exhibits a decrease in firing rate, this may not be observed when measuring with surface electrodes.

A significant decrease in alpha power in response to ipsilateral stimulation was measured across responding electrodes, in both participants. Although no linear association was found between NBR and alpha power responses, this alpha power decrease was concomitant with NBR (Fig. 4B, E).

We can speculate about the neuronal underpinnings of the alpha power decrease. The electrodes in the ipsilateral hemisphere are pooling from neuronal populations with a corresponding population receptive field. It could be argued that the suppressive surround of this population receptive field might extend over the participant’s midline, into the contralateral visual field. Thus, our contralateral stimuli might stimulate the suppressive portion of the corresponding population receptive field, yielding NBR (Zuiderbaan et al. 2012). However, in the scenario just described we would expect an increase in alpha power as reported in Harvey et al. (2013). This is not compatible with our data, as we observed a decrease in alpha power instead.

Alternatively, literature from the sensory-motor domain shows that unilateral hand movement induces NBR in the ipsilateral motor cortex, which could arise from the activity of inhibitory GABA interneurons that are actively suppressing neuronal activity via trans-callosal connections (Schafer et al. 2012).

Moreover, Mullinger and colleagues used median nerve stimulation and EEG recordings showing ipsilateral decreases in BOLD signal concomitant with decreases in the alpha power range (8–13 Hz; Mullinger et al. 2014). This may suggest that activity in inhibitory GABA interneurons is related to the ipsilateral alpha power decreases observed by Mullinger and co-workers.

Based on the results obtained in the sensory-motor domain (Schafer et al. 2012; Mullinger et al. 2014) we speculate that GABAergic inhibition might drive the alpha power decrease we observed in the ipsilateral condition. Note that this decrease was observed for the same electrodes that showed a response in HFB and an associated decrease in NBR. However, the data available does not allow us to draw conclusions about the underlying GABA mechanism at play.

ECoG–fMRI research can positively affect also non-invasive EEG findings. EEG can provide large-scale information about electrophysiological and hemodynamic measurements, while ECoG can provide evidence on a meso- and micro-scale, showcasing tuned and localized neuronal responses and their link to behavior (Schölvinck et al. 2010; Marino et al. 2019; Seeber et al. 2019).