Organization of Propagated Intrinsic Brain Activity in Individual Humans

Abstract

Spontaneous infra-slow (<0.1 Hz) fluctuations in functional magnetic resonance imaging (fMRI) signals are temporally correlated within large-scale functional brain networks, motivating their use for mapping systems-level brain organization. However, recent electrophysiological and hemodynamic evidence suggest state-dependent propagation of infra-slow fluctuations, implying a functional role for ongoing infra-slow activity. Crucially, the study of infra-slow temporal lag structure has thus far been limited to large groups, as analyzing propagation delays requires extensive data averaging to overcome sampling variability. Here, we use resting-state fMRI data from 11 extensively-sampled individuals to characterize lag structure at the individual level. In addition to stable individual-specific features, we find spatiotemporal topographies in each subject similar to the group average. Notably, we find a set of early regions that are common to all individuals, are preferentially positioned proximal to multiple functional networks, and overlap with brain regions known to respond to diverse behavioral tasks-altogether consistent with a hypothesized ability to broadly influence cortical excitability. Our findings suggest that, like correlation structure, temporal lag structure is a fundamental organizational property of resting-state infra-slow activity.

Keywords: functional connectivity; hubs; infra-slow; networks; resting-state fMRI.

Figure 1

Individual and group patterns of latency structure. (A) Weighted lag projection maps from data averaged over individuals (MSCavg; left) and from one separate individual, “RP” (right). Weighted lag projections are one-dimensional representations of latency structure computed as a column-wise weighted mean of the time delay matrix. More strongly correlated regions are given greater time delay weights (see Methods). (B) Weighted lag projection maps for each of the MSC subjects. Note larger lag projection values in individual as compared to MSC averaged results, likely attributable to inter-subject variability in the latter.

Figure 2

Spatiotemporal structure of RSNs shows similarities and differences between the group and individual. Seed-based lag maps reflect each vertex’s lag with respect to the seed region. Seeds were placed in the dorsolateral prefrontal cortex (A) and precuneus (B) (white circles outlined in black). Maps are thresholded at zero-lag correlation r > 0.2. (A) The group average (left) and an individual, RP (right) show similar lagged relationships across distributed regions of the frontoparietal control network. (B) The group (left) and RP (right) also exhibit similar spatiotemporal features within the default mode network; however, the medial prefrontal cortex (pink arrow) is a notable exception.

Figure 3

Specificity and reliability of BOLD spatiotemporal structure in individuals. (A) Split-half similarity matrices computed for TD matrices from each of the MSC subjects and RP. Each element in these matrices is a Pearson correlation coefficient between half of one subject’s data and half of another (or the other half of the same) subject’s data. The final row/column reflects correlations with the MSCavg. The left panel includes all time delays (i.e., the full TD matrix). In the middle panel, time delays corresponding to region pairs with zero-lag correlation magnitude <0.2 (as determined by the MSCavg) are excluded from the similarity computation. Excluding unstable time delays in this way improves within- (on-diagonal) and inter-individual (off-diagonal) correspondence and makes individual-specificity more apparent (comparison of on-diagonal blocks versus the final row/column). Reliability and individual-specificity are further augmented by increasing the correlation threshold to 0.4 (right). (B) Split-half similarity matrices for lag projections—both unweighted (left) and weighted (right) column-wise means of time delay matrices, where weighting is with respect to zero-lag correlation magnitude. Lag projections, particularly when weighted, are relatively stable measures of latency structure and also exhibit individual-specificity. (C) Split-half similarity matrix for the full FC matrix from each subject, which is more stable than measures of latency structure. (D) Summary and comparison of intra-individual (red), inter-individual (blue) and individual-to-group (black) similarity for different spatiotemporal measures, represented as Fisher z-transformed correlation coefficients. Intra-individual, inter-individual, and individual-to-group similarity were separately compared for both TD matrices computed across the different FC thresholds and for unweighted and weighted lag projections. Statistical significance was assessed by two-tailed paired t-tests (N = 11; ^***P < 0.001; n.s., not significant).

Figure 4

Data requirements for different measures of BOLD spatiotemporal structure. (A) Example TD matrices and weighted lag projections computed from two randomly split halves of RP’s data. Rows and columns of the first half TD matrix have been sorted from early-to-late, and by network affiliation following (Laumann et al. 2015). Half 1 sorting was applied to the TD matrix from Half 2. Note largely orthogonal relationship of latency structure to RSNs manifesting as a wide range of TD values within matrix blocks (Mitra et al. 2014). Also note that early (late) regions tend to be early (late) in both within- (on-diagonal) and between-RSN (off-diagonal) relationships (Mitra and Raichle 2018). (B) Correlation of spatiotemporal structure from one half of RP’s data with the spatiotemporal structure computed from increasing amounts of data drawn from the other half. Correlation is computed for several measures of spatiotemporal structure: full TD matrix (pink), TD matrix excluding |FC| < 0.2 (orange), TD matrix excluding |FC| < 0.4 (red), unweighted lag projection (blue), weighted lag projection (black), and the full FC matrix (green). For FC-based thresholding, FC is determined by the FC matrix computed over all RP data. Correlation curves are represented as mean (solid line) and SD (shaded surrounding area) of correlation from 100 random samplings of the 88 ten-min sessions.

Figure 5

Regional patterns of variability in latency structure. (A) Upper: Map for RP reflecting, for each vertex, TD standard deviation (TD_std) across sessions and subsequently averaged across all of the vertex’s TD relationships (column-wise mean of TD_std). Lower: Each vertex’s mean zero-lag correlation magnitude. The strong inverse correlation between these maps (r = −0.91) implies that regions that exhibit the most session-to-session TD variability tend to have weaker correlations in general, making their TD relationships more prone to sampling variability. (B) Upper: Map reflecting column-wise mean of TD_std, as in (A), but where TD_std is computed across all MSC subjects rather than RP sessions. Lower: Each vertex’s mean zero-lag correlation magnitude, computed from the MSC average. Despite 300 min of data per subject, there is still contribution of sampling variability (r = −0.27 between upper and lower maps). However other sources of variability, such as anatomical, are now apparent.

Figure 6

Consistent features of latency structure across individuals. Spatial overlap of the 15% earliest (upper) and 15% latest (lower) regions, based on lag projections shown in Figure 1. The earliest regions show the greatest topographic consistency across subjects. Note high consistency in regions concerned with organizing action, including several premotor areas and anterior insula.

Figure 7

Early regions preferentially localize to areas proximal to multiple functional systems. (A) RSN parcellation computed from the MSC average correlation matrix, as previously published (Gordon et al. 2017c), provided for reference. (B) Upper: Community density map for the MSC average, averaged across correlation thresholds or “edge densities” (see Methods). Higher values indicate proximity to a greater number of RSNs. Lower: Participation coefficient map for the MSC average, averaged across correlation thresholds. (C) Upper: MSC average community density (upper) and participation coefficient (lower) maps in (B), thresholded by the earliest 5% of vertices in the MSC average weighted lag projection (shown in Fig. 1). (D–F) Same as (A–C), but for the individual subject RP. RP RSN parcellation, provided for reference, is from Laumann et al. (Laumann et al. 2015). (G) Standardized community density, for each individual, as a function of increasing latency. Black opaque curve represents average curve across individuals. Community density was averaged into 5% bins spanning the range of lag projection values, ordered from early-to-late. One-sample t-tests determined, for a given latency bin, mean community density across individuals that differed from the expected value. (^*P < 0.05 following Bonferroni correction for 20 bins). (H) Similar analysis as in (G), but for participation coefficient instead of community density.