Next, we asked whether inter-subject correlation of functional activity and functional connectivity distinguish different phases of value learning. We expected to find an interaction between ISC and ISFC during learning, such that activity would be highly aligned across subjects early during learning, indicated by high values of ISC, and that functional connectivity would be highly aligned later during learning, indicated by high values of ISFC. Our hypothesis is driven by the notion that initial phases of learning would evoke common patterns of activity across individuals, which might in turn reinforce common patterns of functional connections across subjects near the end of learning. We first examined the dynamics of ISC during value learning (Fig.2C) and found that the average ISC of all regions increased from day 1 to day 2 and, subsequently, continued decreasing through day 4 (One-way ANOVA, \(F(3,448)=38.3121,p=2.7e-21\)). Overall, the average ISC was significantly greater during all four days of value learning compare to at rest (\(t=11.49,p=1.28\times 10^{-20}\); Fig. S2A). The resulting trends in ISC dynamics during value learning suggest that ISC might support two phases of learning: (i) increased ISC between day 1 and day 2 may be associated with increasing constraints on activity, perhaps as a result of common neurophysiologic mechanisms across subjects that facilitate early stage learning of the task mechanics, and (ii) decreased ISC from day 2 through day 4 may be associated with less constrained dynamics, perhaps as subjects explore cognitive strategies to increase their performance on the task. Interestingly, the average ISC of subjects at rest was near zero (Figure S2A-B, see supplementary results), suggesting that activity is minimally constrained in the absence of a common stimulus. Across the four days, the average ISC of the region during value learning across the region were different (One-way ANOVA, \(F(3,448)=38.31,p=2.7*10^{-21}\)) and there was no significant difference between day 1 and day 3 ( \(F(1,112)=2.73,p=0.1010\)). We next examined the dynamics of ISFC during value learning (Fig. 2D) and found a significant difference in average ISFC between brain regions across the four days (\(F(3,448)=22.77,p=9.95*10^{-14}\)). Specifically, ISFC significantly increased from day 1 to day 2 and peaked during day 3. The increasing trend in ISFC during value learning suggests that functional connectivity becomes more constrained to a common topological pattern across subjects, perhaps reinforcing the functional network associated with performing the value learning task.
We then explicitly tested for potential interactions between ISC and ISFC over time. Overall, we found that ISC and ISFC were significantly positively correlated on each of the four days (day 1: \(r=0.2684,p=0.0430\), day 2: \(r=0.3185,p=0.0000\), day3:\(r=0.1970,p=0.0375\), day 4: \(r=0.2982,p=0.0015\)), suggesting that increased group-level constraints on activity are related to increased group-level constraints on functional connectivity. However, despite the correlation between ISC and ISFC on any given day, we observed that their rate of change through the duration of the task was indeed significantly different (Fig. 2E and Table 2 for statistical differences). During early stage learning on days 1 and 2, ISC significantly exceeded ISFC – suggesting that activity is more rigidity constrained to common organizational rules across the cohort than functional connectivity. During later stage learning on day 3 and 4, ISFC significantly exceeded ISC – suggesting that activity becomes more autonomous than functional connectivity and functional activity becomes more constrained than functional activity. These nuanced dynamics point to a potential driver-follower mechanism of constrained functional activity preceding constrained functional connectivity over four days of value learning.
Regional variability of functional constraints during learning
Next, we asked: if the dynamics of whole-brain inter-subject correlations map unto different phases of value learning, which brain systems might be most complicit in these phases? To investigate this question, we first partitioned brain regions into objectively defined functional modules using the GenLouvain community detection algorithm \cite{de2011generalized} (see Materials and Methods and SI). Briefly, community detection is applied to the functional network constructed from data of each task session and parses brain regions into functional modules such that brain regions within the same module exhibit strong functional connections and brain regions between different modules have weak functional connections. In order to obtain a single representative partitioning of brain regions across subjects and scans, we computed the module allegiance matrix \cite{bassett2011dynamic} – capturing the probability that two regions belong to the same functional module (Fig. 3A). By applying a final round of community detection to the module allegiance matrix, we identified seven modules that were associated with different putative brain systems, including fronto-temporal (FT) which covered most of the limbic lobe, sensorimotor network (SM), auditory network (AUD) including hippocampus and amygdala, the common default mode network (DMN), Language (LAN) network, Visual (VIS) network and three subcortical regions - putamen, caudate and thalamus (PCT) (Fig. 3B; see Table 1 for list of regions in each community).