Altered Patterns of Dynamic Functional Connectivity Underpin Reduced Expressions of Social-Emotional Reciprocity in Autistic Adults

Abstract

To identify the neurocognitive mechanisms underpinning the social difficulties that characterize autism, we performed functional magnetic resonance imaging on pairs of autistic and non-autistic adults simultaneously whilst they interacted with one another on the iterated Ultimatum Game (iUG)-an interactive task that emulates the reciprocal characteristic of naturalistic interpersonal exchanges. Two age-matched sets of male-male dyads were investigated: 16 comprised an autistic Responder and a non-autistic Proposer, and 19 comprised non-autistic pairs of Responder and Proposer. Players' round-by-round behavior on the iUG was modeled as reciprocal choices, and dynamic functional connectivity (dFC) was measured to identify the neural mechanisms underpinning reciprocal behaviors. Behavioral expressions of reciprocity were significantly reduced in autistic compared with non-autistic Responders, yet no such differences were observed between the non-autistic Proposers in either set of dyads. Furthermore, we identified latent dFC states with temporal properties associated with reciprocity. Autistic interactants spent less time in brain states characterized by dynamic inter-network integration and segregation among the Default Mode Network and cognitive control networks, suggesting that their reduced expressions of social-emotional reciprocity reflect less efficient reconfigurations among brain networks supporting flexible cognition and behavior. These findings advance our mechanistic understanding of the social difficulties characterizing autism.

Keywords: autism; dynamic functional connectivity; reciprocity; social interaction.

FIGURE 1

Experimental paradigm. On experimental (iUG) rounds, Proposers selected between two alternative monetary divisions to offer the Responder (Choice), after which their choice was presented, and Responders had to accept or reject the proposal (Offer). The Responder's decision was then presented subsequently (Decision). The rounds were separated by the jittered inter‐trial interval (ITI). The same sequence and timings were followed in Control (CTRL) rounds, but Proposers chose between two alternative divisions of color and the Responder decided whether to accept or reject the offer. In these examples, the unfair offer made by the Proposer on a Proposer–Responder (PR) round of the iUG is rejected by the Responder (left), while the offer made on the CTRL round is accepted (right).

FIGURE 2

Processing pipeline for dFC. After pre‐processing, the fMRI data were parcellated into 400 regions. The representative time‐series for every region across all participants were entered into the Bayesian Mixture of Factor Analyzers (BMFA) model. The model was estimated repeatedly with random initialization and 2 to 15 States. The optimal run of the BMFA estimation was selected according to metacriterion comprising several metrics. Each resulting state was subsequently described by a group‐specific FC matrix and its probability time‐series for each participant. The winner‐takes‐all (WTA) approach was employed to construct participants' state sequences, which were used to compute temporal characteristics for each state. These characteristics were then subjected to statistical comparisons. The BMFA model was estimated separately for groups of Proposers and Responders.

FIGURE 3

Comparisons of behavioral metrics between non‐autistic Proposers (left) comprising the AA/NA (green) or NA/NA dyads (gray), or between autistic and non‐autistic Responders (right) comprising AA/NA (green) or NA/NA dyads (gray). Top to bottom: Role‐specific reciprocity parameters calculated from trial‐by‐trial monetary exchanges of the iUG, and proportions of fair offers (those presenting the least advantageous inequity from the Proposer's perspective) and their acceptances across Proposer–Proposer (PP) and Proposer–Responder (PR) rounds. Boxplots illustrate medians (horizontal lines) within interquartile ranges, with means presented as crosses. Note the ceiling effects where non‐autistic Responders from NA/NA dyads accepted almost all the fair offers. Note: *p < 0.05, **p < 0.01.

FIGURE 4

Latent brain states. Matrices depict functional connectivity among all nodes of the seven brain networks characterizing each of the latent brain states identified from Responders (top) and Proposers (bottom) across both AA/NA and NA/NA dyads. Functional connectivity is expressed as pairwise Pearson correlation coefficients computed from co‐variances identified with BMFA. Matrices are organized by brain networks in each hemisphere. DAN = dorsal attentional network, DMN = default mode network, FPN = fronto‐parietal network, LN = limbic network, SMN = somato‐motor network, VAN = ventral attentional network, VN = visual network.

FIGURE 5

Differences between players of AA/NA (green) and NA/NA (gray) dyads in the temporal characteristics of latent brain states computed across the entire iUG: The bar charts present medians and interquartile ranges. S1–S5 = State 1–State 5; *p < 0.05; **p < 0.01.

FIGURE 6

Condition‐specific differences in the temporal characteristics of latent brain states between players of NA/NA (gray) and AA/NA dyads (green). Boxplots present medians and interquartile ranges. CTRL = control, PP=Proposer–Proposer, PR = Proposer–Responder, S1–S5 = State 1–State 5; *p < 0.05; **p < 0.01.

FIGURE 7

Brain‐behavior relationships. Associations between player‐specific reciprocity parameters estimated with the modeling procedure and the temporal characteristics of latent brain states exhibited by Responders (top) and Proposers (bottom). Note: NA/NA = dyads comprising a non‐autistic Responder and Proposer (gray), AA/NA = dyads comprising an autistic Responder and non‐autistic adult Proposer (green).