Task design and graphical summaries of analytic strategies. (Study 1)

Two task-fMRI studies were conducted to identify the connectivity patterns upregulated by high or low PE. (A) In each trial, participants saw a scene-object pair and were asked to report how likely it would be to find the object in the scene using a 4-point scale that was active while the object was on screen. (B) Separate participants completed a gambling task wherein each trial they made a binary choice (“?” screen) and were then told whether the choice caused them to earn money (upward arrow) or lose money (downward arrow). The “+” panel represents a 15 s fixation period between blocks. Both earning or losing money could elicit high or low PE, depending on whether it deviates from prior trials’ outcomes. (Study 2) (A) Using rs-fMRI data collected from Study 1A participants, analyses examined changes in connectivity over time to assess whether participants spontaneously fluctuate between the high and low PE signatures found in Study 1. (B) Further analyses examined whether slight differences in Study 1B participants’ task-fMRI responses to PE mirrored differences in their rs-fMRI connectivity states. (Study 3) With rs-EEG-fMRI data from another pool of participants, analyses modeled the amplitudes of fMRI connectivity fluctuations over time and correlated this with EEG oscillatory power at different frequencies to assess which frequency best tracked the fMRI patterns.

Modularity behind the enhancing and weakening effects of prediction error.

(A) Each neocortical connection was submitted to a regression predicting connectivity strength based on PE. For clearer visualization, the shown 46 x 46 matrix simplifies the original 210 x 210 matrix by averaging ROIs per anatomical labels. Supplemental Figure S1 shows the exact ROIs used on a glass brain, illustrating their precise anatomy. (B) Modularity was assessed in terms of the 5% of edges showing the most positive coefficients and 5% showing the most negative coefficients. (C) Two high-PE modules and two low-PE modules are shown. (D & E) The four modules can be described anatomically as connections among four anatomical quadrants. (F) Analyzing these anatomically defined connections shows a significant Connectivity-Direction x PE interaction (β [standardized] = .20, p < .001). Boxes indicate quantiles.

Reproducing the posterior-anterior/ventral-dorsal PE effects using a gambling task.

(A) This matrix represents high/low PE paired t-tests for each connection. Like Figure 2, a 46 x 46 regional matrix is shown rather than a 210 x 210 ROI matrix for clarity. (B) Low PE enhances posterior-anterior connectivity while high PE enhances ventral-dorsal connectivity but note the High > Low PE ventral-dorsal effect is driven by ATL-LPFC connectivity, and the OC-IPL effect is minuscule. For the bars, the connectivity data was normalized by subtracting means across conditions. Boxes indicate quantiles.

Dynamic connectivity modules.

(A) Time-varying connectivity was computed between adjacent portions of the four bilateral quadrants from earlier. Given that each quadrant is composed of many ROIs, this involved computing time-varying connectivity for individual ROI-ROI pairs and averaging them to produce overall estimates. (B) Based on the time-varying connectivity of each edge, an edge-edge correlation matrix was produced. (C) Computing modularity of the edge-edge correlation matrix yielded the two illustrated modules.

Replication of the connectivity modules and subject-specific analyses.

Performing the Study 2A modularity analysis using the Study 1B participants yields a similar pair of modules as in Figure 4 but with slight changes to the interhemispheric connections.

Correlating connectivity fluctuations and oscillatory power.

(A) If posterior-anterior (PA) and ventral-dorsal (VD) time-varying connectivity measurements at each TR reflect samples from an oscillator, then their absolute difference (|PA – VD|) will correlate with the oscillator’s true amplitude (A) even if the oscillator’s frequency outpaces fMRI’s temporal resolution. (B) Correlations between |PA-VD| and amplitude/power at different frequencies allow assessing the frequency of the oscillator generating the PA/VD fluctuations; amplitude (A) and power (A2) are equivalent for the analyses, which used Spearman correlations. The fMRI-EEG analytic strategy is motivated by simulations: (C) One sine wave is used as a simulation input. (D) An autoregressive time series is also used as an input. (E) The product of the sine wave and autoregressive time series defined the simulated oscillator, which represents how fluctuations between states may vary in intensity over time. (F) Ventral-dorsal (VD) connectivity is defined as the sine-amplitude product, and posterior-anterior (PA) connectivity is defined as the inverse of this product. (G) Representing brain data recorded using fMRI, the VD and PA dots here are the average over the TR windows. No hemodynamic response function convolution was applied when making this figure for the sake of clarity, but convolution was performed for the actual simulation results in Supplemental Materials 5. (H) Using the PA and VD dots, the |PA – VD| differences were computed. (I-K) Green lines represent the time-frequency decomposition of the sine-amplitude product. The dots represent the power averaged for each TR window. The first TR window for 0.5 Hz power was removed as it was susceptible to edge artifacts.

Correlations between fluctuation amplitude and oscillatory power.

For all participants, Spearman correlations were measured between connectivity-fluctuation amplitude and power at every 0.5 Hz interval from 0-0.5 Hz. Visualized here, each dot represents the Cohen’s d effect size from a one-sample t-test assessing whether the mean correlation surpassed zero. This was done using either (A) power averaged across the shown frontal electrodes or (B) power averaged across the shown frontal/central/ parietal pool of electrodes.

Mean correlations between connectivity fluctuation and frontal oscillatory power.

The table lists the Spearman correlations between fluctuation amplitude and the frontal power, averaged across an established frequency band (e.g., Delta is the average of 1 Hz power, 1.5 Hz power, … 3.5 Hz power). *, p < .05; **, p < .01; *** p < .0001.