Intrinsic functional architecture predicts electrically evoked responses in the human brain

Abstract

Adaptive brain function is characterized by dynamic interactions within and between neuronal circuits, often occurring at the time scale of milliseconds. These complex interactions between adjacent and noncontiguous brain areas depend on a functional architecture that is maintained even in the absence of input. Functional MRI studies carried out during rest (R-fMRI) suggest that this architecture is represented in low-frequency (<0.1 Hz) spontaneous fluctuations in the blood oxygen level-dependent signal that are correlated within spatially distributed networks of brain areas. These networks, collectively referred to as the brain's intrinsic functional architecture, exhibit a remarkable correspondence with patterns of task-evoked coactivation as well as maps of anatomical connectivity. Despite this striking correspondence, there is no direct evidence that this intrinsic architecture forms the scaffold that gives rise to faster processes relevant to information processing and seizure spread. Here, we demonstrate that the spatial distribution and magnitude of temporally correlated low-frequency fluctuations observed with R-fMRI during rest predict the pattern and magnitude of corticocortical evoked potentials elicited within 500 ms after single-pulse electrical stimulation of the cerebral cortex with intracranial electrodes. Across individuals, this relationship was found to be independent of the specific regions and functional systems probed. Our findings bridge the immense divide between the temporal resolutions of these distinct measures of brain function and provide strong support for the idea that the low-frequency signal fluctuations observed with R-fMRI maintain and update the intrinsic architecture underlying the brain's repertoire of functional responses.

Fig. 1.

Analysis schematic. (A) CCEPs in response to single-pulse electrical stimulation of each electrode pair were recorded at all implanted electrodes. Responses were categorized as significant or nonsignificant if the average waveform exceeded a statistical threshold of 6 SDs during an interval of 50–500 ms poststimulation. For visualization purposes, the locations of significant responses were overlaid onto the individual's reconstructed pial surface. The magnitude of CCEPs at each electrode was computed as a z score relative to baseline of each stimulation site. (B) RSFC maps associated with spherical ROIs underlying the stimulated electrodes were computed for each stimulation site. For visualization, the RSFC maps were displayed on the reconstructed pial surface. To examine the correlation between RSFC and CCEP values, Fisher's z-transformed correlation coefficients were extracted from the voxels underlying each electrode and correlated with the CCEPs recorded from the same electrodes. The schematic shows data from Subject 1 as an illustration.

Fig. 2.

Relationships between CCEPs and RSFC. (A) Group and single-subject analysis. Across all stimulation stites, electrodes exhibiting significant CCEPs (+, dark gray bars) also exhibited significantly higher RSFC z values than electrodes exhibiting nonsignificant CCEPs (−, light gray bars), at the group level t(5) = 9.3, P < 0.001, two-tailed t test), and in all but one of the six individual subjects (S1–S6; *P < 0.05; **P < 0.01). (B) Exclusion of proximal electrodes. Significantly higher RSFC z values were still observed at electrodes exhibiting significant CCEPs after excluding electrodes within 2.5 cm of the stimulation site (P < 0.001, two-tailed t test). (C) Correlation between RSFC and CCEPs. We computed the correlation between CCEP amplitude and RSFC, across all electrodes, for each stimulation/seed site in each subject, regardless of whether a significant CCEP response was observed (gray lines). Two-tailed t tests indicated that the mean relationship across all sites (black line) was significant in each subject (all P < 0.001) as well as across all six subjects [t(5) = 21.4, P < 0.001]. By including all electrodes, this correlation is independent of a priori assumptions implied by amplitude thresholding. The mean r-correlation value of all trend lines is reported for each subject (n = number of trend lines = number of stimulation sites). CCEPs are reported as z scores, and RSFC is reported as a Fisher's z value.

Fig. 3.

Probing Broca's area (A) and Wernicke's area (B) in one patient (S1). Patterns of CCEPs and RSFC are overlaid onto subjects’ pial surfaces. Selected significant (black lines) and nonsignificant (gray lines) evoked potentials are shown. The dotted boxes designate the area of interest for determining the significance of the evoked potential (corresponding to the N2 phase). Significant CCEPs (green circles) were observed in close proximity to the stimulation site (e.g., Broca's area) but also in spatially distal areas (e.g., posterior middle and superior temporal cortex). The gray dots indicate electrodes with nonsignificant CCEPs. Areas exhibiting significant RSFC with the seed ROI underlying the stimulation site are shown in yellow/red (positive RSFC) and blue (negative RSFC). Note the striking overlap of electrodes exhibiting significant CCEPs and regions exhibiting significant RSFC. The scatter plots show the relationship between the RSFC Fisher's z values and CCEP z scores for all recorded electrodes (Left) and for electrodes that exhibited significant CCEPs (Right). The bar graphs show the mean RSFC z values for significant (black) and nonsignificant CCEP electrodes [P < 0.001]. (C) Probing sensorimotor areas. The remarkable overlap between patterns of RSFC and CCEPs was also observed using seeds and stimulation sites in sensorimotor areas. Examples from two subjects (S1 and S2) are shown. CCEPs are reported as z scores, and RSFC is reported as a z-correlation value.

Fig. 4.

Negative RSFC correlations. (A) CCEP responses were grouped according to whether they were observed at electrode sites exhibiting negative RSFC (400 most negative RSFC values), nonsignificant RSFC (400 RSFC values centered at 0), or positive RSFC (400 most positive RSFC values). Averaged full-wave rectified waveforms are plotted for each group. (B) Peak CCEPs for negative and positive RSFC in the N2 time range was calculated using different percentile groups. Although CCEP amplitude increased with increasing positive RSFC values (*P < 0.01), no change in CCEP amplitude was observed at sites exhibiting negative RSFC. (C) Mean CCEPs were calculated for each RSFC group and each subject (S1–S6). CCEP z scores were consistently higher for the positive RSFC group than for the nonsignificant and negative RSFC groups. There was no consistent difference between mean CCEP magnitude for the negative and nonsignificant RSFC groups across subjects.