Self responses along cingulate cortex reveal quantitative neural phenotype for high-functioning autism

Abstract

Attributing behavioral outcomes correctly to oneself or to other agents is essential for all productive social exchange. We approach this issue in high-functioning males with autism spectrum disorder (ASD) using two separate fMRI paradigms. First, using a visual imagery task, we extract a basis set for responses along the cingulate cortex of control subjects that reveals an agent-specific eigenvector (self eigenmode) associated with imagining oneself executing a specific motor act. Second, we show that the same self eigenmode arises during one's own decision (the self phase) in an interpersonal exchange game (iterated trust game). Third, using this exchange game, we show that ASD males exhibit a severely diminished cingulate self response when playing the game with a human partner. This diminishment covaries parametrically with their behaviorally assessed symptom severity, suggesting its value as an objective endophenotype. These findings may provide a quantitative assessment tool for high-functioning ASD.

Figure 1. Schematic representation of the multi-round trust game and agent-specific BOLD responses along cingulate cortex

(A) Schematic representation of the multi-round trust game. In each round, Player 1, the “Investor,” is endowed with 20 monetary units. The Investor chooses to send some portion I of this amount to Player 2, the “Trustee.” This amount is tripled to 3I and sent to the Trustee who returns some fraction f of the tripled amount. The interaction continues for 10 rounds, and the players maintain their respective roles throughout the game. For events labeled “ASD,” the Investor is a non-psychiatric control participant, and the Trustee is a male with autism spectrum disorder or an age- and IQ- matched male control. (B) Trust task events from which cingulate responses are taken. The base of the gray triangles identify the two 8-second epochs where the “other” and “self” responses are defined, in both cases for the Trustee's brain. The “other” response is taken as the peak BOLD response in the 8 seconds following the revelation of the Investor's decision averaged with its two flanking points, and the “self” response is taken as the peak response in the 8 seconds following submission of the Trustee decision averaged with its two flanking points. The middle panel depicts the 11 regions-of-interest along the medial bank of cingulate cortex. The left panel shows the cingulate “other” response for each of the 11 domains from a database of 100 Trustees in this game. The right panel shows the cingulate “self” response from the same database. The lower panels show that these agent-specific response patterns disappear outside the context of social exchange when the interactive partner is removed yet motor, monetary earnings, and visual aspects of the game remain constant (adapted from Tomlin et al., 2006). Maximum and minimum activations in the “other” response are 0.25% and -0.10% change in MR signal, and 0.30% and -0.20% in the cingulate “self” response.

Figure 2. Cingulate self-response and self-eigenmode identified in visual imagery task

(A) Schematic representation of the visual imagery task. In each trial, subjects were presented with a visual cue (2 s) indicating a target person to be presented in an upcoming video. The subsequent video clip (4 s) depicted a specialized athletic act (e.g., kicking, throwing, dancing). Subjects were instructed to close their eyes upon video offset, then presented with an auditory cue to either “watch it” or “do it.” “Watch it” indicated that subjects were to imagine watching the actions in the video again, keeping the perspective of a spectator (third-person perspective). The “do it” cue indicated that subjects were to imagine the actions in the video from the perspective of the target athlete (first-person perspective). A final auditory cue (“stop”) indicated the end of the trial, upon which subjects were to open their eyes in preparation for the next trial. (B) Imagery task identifies cingulate self basis. Spatial principal components analysis was performed on cingulate responses comprising participants peak % MR signal change extracted from 10 equally-sized spatial domains along the anterior- posterior axis of the medial cingulate cortex. The most posterior cingulate domain is marked with a “p,” and the most anterior domain is denoted “a.” Note that the spatial pattern of principal component 2 (dotted red circle) exactly resembles that of the “self” response seen directly in the time series of the “self” phase of both the visual imagery task and the multi-round trust game (Figures 1B and 2A). (C) Cingulate “self eigenmode” identified in visual imagery task. BOLD responses from the video-watching (“other”) and first-person (“self”) imagery phases of the visual imagery task are projected respectively onto the cingulate self-basis. The projection coefficients of the BOLD responses on principal component 2 (Figure 2B, dotted red circle) significantly differentiates the two experimental conditions (p = 1.87×10^-15) and warrants the term “self eigenmode” to describe this discriminating eigenvector.

Figure 3. Trust task reveals complementary agent-specific cingulate basis sets and self eigenmode

Spatial principal components analysis (sPCA) was performed on our database of normative cingulate “other” responses consisting of 100 adult Trustees' average peak % MR signal changes extracted from 11 distinct and equally sized spatial domains along the anterior-posterior axis of medial cingulate cortex. The results of the sPCA on cingulate responses in the “other” and “self” phases of the trust game are depicted here. Segment “p” refers to the most posterior cingulate domain; segment “a” refers to the most anterior cingulate domain. (A) The “other” phase of the Trust game is depicted here and references the time points surrounding the revelation of the Investor's decision to the Trustee brain. The spatial patterns of the first 3 principal components, accounting for 92.9% of the variance in cingulate activation, are illustrated here. Note that the spatial pattern of principal component 3 (dotted red circle) resembles the normative “other” response observed in the time series data, and is an inverted complement of the self eigenmode. (B) The “self” phase of the Trust game is depicted here and references the time points surrounding the submission of the Trustee's own decision in the game. The spatial patterns of the first 3 principal components, accounting for 92.0% of the variance in cingulate “self” activation, are illustrated here. The spatial pattern of principal component 3 (dotted red circle) closely resembles that of the self eigenmode and the normative “self” response reported directly from the time series in the visual imagery task and multi-round trust game (see Figure 1A and Tomlin et al., 2006).

Figure 4. Expression of the self eigenmode is (A) entirely absent in the “other” phase of the multi-round trust game and (B) reaches maximal amplitude subsequent to a player's decision during the “self” phase of the game

We projected the Trustees' cingulate hemodynamic response patterns at each TR (every 2 s) in both the “other” and “self” phases of the multi-round trust game into the three dimensional space whose axes comprise the first three eigenvectors of the neural self basis. We thus tracked the self eigenmode (PC 3 of the self basis) across the course of the Trust game; the development of the self eigenmode is seen distinctly on the self eigenmode axis in panel B. As shown here, the self eigenmode is (A) absent or inverted during the entire “other” phase of the game, (B) absent during the beginning of the “self” phase, appears subsequent to the “self” decision, and disappears again at the ending time points of the “self” phase.

Figure 5. Behavioral characterization of ASD, age- & IQ- matched control, and behavior-matched control pairs on the iterated Trust game

(A) Dyadic behavioral trajectories do not differ among ASD, normative, and age- and IQ matched controls. To characterize the behavioral exchange of each experimental pair, we formed vectors composed of the series of Investment ratios (I = Investment / 20) and fractions of Trustee Repayments (R = repayment /3· I) over the ten rounds of a dyad's exchange. The behavioral vector for each pair p is denoted: U^p =(I₁, R₁, I₂, R₂, …, I₉, R₉, I₁₀, R₁₀), where I_n represents the fraction Invested in round n and R_n represents the fraction repaid in round n. All Investors were adult controls; the indicated experimental groups are identified by the participants in the Trustee role. The behavioral vectors for each pair of each experimental group (ASD, age- & IQ control, behavioral control, database adult control) are projected onto the plane defined by the first two principal components of the behavioral data from 100 iterated trust exchanges in our database of adult controls. The projections of the 20-dimensional behavioral vectors onto this plane are identified as follows: blue circles = database adults, red squares = ASD, green triangles = age & IQ-matched controls, blue circles with overlaid purple stars = behavioral controls. Open shapes represent the projections of the behavioral vectors of each pair; solid black shapes represent the projections of the average behavioral vector of the indicated group. Behavioral controls are those 15 pairs from our normative database whose pattern of play most closely resembled that of the ASD pairs (see Experimental Procedures for detailed description of control selection). (B) Round-by-round trustee repayment ratio for ASD and age- & IQ- matched control groups. Average trustee repayment ratio across the 10-rounds of the trust game do not differ between ASD and controls, supporting intact cognitive understanding of the basic elements of the task. (C) Average Trustee earnings for ASD and age- & IQ- matched controls. Average total earnings on the multi-round trust game do not differ between the ASD and control group.

Figure 6

(A) Participants with ASD lack cingulate “self” response pattern. Cingulate responses to the revelation of a partner's decision (“other” response pattern) and following the submission of one's own decision (“own” response pattern) are shown for the ASD group (n = 12), age & IQ-matched controls (n = 18), and controls whose pattern of play was behavior-matched to the ASD-dyads (n = 15; see Experimental procedures for details of group selection). The revelation of a social partner's decision elicits an unperturbed cingulate response pattern in the ASD and control groups (left column). In stark contrast, the activation following one's own decision is missing in the ASD group (white asterisk; p = 0.001 two-tailed for ASD vs age & IQ-matched controls' three middle cingulate segments; p = 0.0004 for ASD vs behavior-matched controls). Maximum and minimum activations in the “other” response are 0.25% and -0.10% change in MR signal, and 0.30% and -0.20% in the cingulate “own” response. (B) Lack of cingulate “self” response pattern relates parametrically to ASD symptom severity. The reduction in the “self” response pattern in ASD participants correlates with symptom severity on the Autism Diagnostic Interview-Revised. The ASD participants' average cingulate response across the three middle segments and the fraction of maximum ADI subscale score are depicted: ADI total score (dark circles; r = -.73, p = .007), ADI social subscale (light circles; r = -.70, p = .011, and ADI communication subscale (open circles; r = -.69, p = .012). No significant correlation was observed between the ADI repetitive behavior subscale and cingulate “own” response (r= -.34, p = .28). Moreover, no significant correlations were observed between scores on any ADI symptom domain and any cingulate region's “other” response. (C) Cingulate response basis set reduces diminished ASD “self” response to a single projection coefficient. To test for between-group differences in the “own” cingulate response pattern, two-tailed t-tests were performed on the ASD and the two control groups' expansion coefficients onto each principal component (PC) of the neural “self” basis set. This analysis identified an attenuated response in the ASD group specifically along the self-eigenmode (PC3 in Figure 3B) compared with both the age- and IQ- matched controls (p = 0.0036) and the behavior-matched controls (p = 0.00043; see Experimental Procedures for control selection procedures). For all other comparisons between the ASD and control groups on the remaining PCs, p > 0.1. Projections of cingulate BOLD responses onto the self eigenmode of the trust game illustrate the diminished self response pattern in ASD (right panel).