BOLD fMRI Correlation Reflects Frequency-Specific Neuronal Correlation

The brain-wide correlation of hemodynamic signals as measured with BOLD fMRI is widely studied as a proxy for integrative brain processes. However, the relationship between hemodynamic correlation structure and neuronal correlation structure remains elusive. We investigated this relation using BOLD fMRI and spatially co-registered, source-localized MEG in resting humans. We found that across the entire cortex BOLD correlation reflected the co-variation of frequency-specific neuronal activity. Resolving the relation between electrophysiological and hemodynamic correlation structures locally in cortico-cortical connection space, we found that this relation was subject specific and even persisted on the centimeter scale. At first sight, this relation was strongest in the alpha to beta frequency range (8-32 Hz). However, correcting for differences in signal-to-noise ratios across electrophysiological frequencies, we found that the relation extended over a broad frequency range from 2 to 128 Hz. Moreover, we found that the frequency with the tightest link to BOLD correlation varied across cortico-cortical space. For every cortico-cortical connection, we show which specific correlated oscillations were most related to BOLD correlations. Our work provides direct evidence for the neuronal origin of BOLD correlation structure. Moreover, our work suggests that, across the brain, BOLD correlation reflects correlation of different types of neuronal network processes and that frequency-specific electrophysiological correlation provides information about large-scale neuronal interactions complementary to BOLD fMRI.