Overview of the task structure.

During the offer phase, participants made choices between rest, a fixed low reward (1 credit) involving no effort, and work, a variable higher reward (2, 4, 6, 8, or 10 credits) involving more effort (30, 40, 50, 60, or 70% of participant’s individual MVC). Higher effort levels were indicated by a more filled in circle. Participants had 3.2s to make their choice and the chosen option was then highlighted with a box for 0.3s. If participants decided to work, then during the force phase, they had to squeeze the hand dynamometer at the required effort level (shown by the yellow line) for one second during a three second window. If they decided to rest, then the yellow line was displayed at the bottom of the bar and participants did not have to squeeze. During the outcome phase, the number of credits participants earned either for themselves or the next participant was displayed. Each phase of each trial was separated by a variable jitter shown at the bottom of the diagram.

Mean (+/-SEM) cortisol and stress ratings at each sample time point.

The grey box indicates the time participants spent in the scanner doing the tasks. Participants completed six runs of 25 trials for the prosocial effort task. Before each run participants experienced either an adapted version of the Montreal Imaging Stress Test (MIST; Dedovic et al., 2005) or the counting task from the Trier Social Stress Test (Kirschbaum et al., 1993).

Estimated marginal means (+/-SEM) are plotted on a logit scale and were extracted using the emmeans package in R (Lenth, 2022).

Follow-up tests showed that the three-way interaction between Group, Recipient and Effort was driven by a significant interaction between Group and Recipient at a low level of effort (i.e., Effort level 2).

The interaction between SVO angle and perceived stress on the proportion of prosocial choices.

Participants with a more individualistic SVO angle (-1SD; black dashed line) became more selfish at increasing levels of perceived stress, which was not seen in more prosocial participants (+1SD; orange solid line).

Results from the ROI analysis in the AI and dACC.

These show percentage signal change to SV during the offer phase associated with perceived stress on self and other trials.

Regions resulting from a whole-brain analysis in which activity scaled more with SVself compared to SVother that covaried with Perceived Stress (using a statistical threshold of p<0.05 FWE corrected at the cluster level having thresholded at p<0.001 across the whole brain).

The relative summed BIC values for each model.

A: We used maximum likelihood estimation and a model comparison approach to determine which model best characterised participants’ choices. Twelve models were compared which included either one (1K) or two discount parameters (2K: Kself, Kother) , either one (1B) or two (2B: βself, βother) temperature parameters, and either a parabolic (P), linear (L) or hyperbolic (H) discount function. Model 7 (2K 1B parabolic) had the lowest BIC across the whole sample. Model 10 (2K 2B parabolic) had the lowest BIC in the stress group. B: Across the whole sample Model 7 (77.66%) was a better fit compared to model 10 (22.34%) for the majority of participants so was chosen for all analysis.

Parameter recovery.

We stimulated data based on our trial structure and determined whether the parameters were recoverable for the best fitting model. There was excellent recovery as showed by the strong relationship between fitted and simulated (or ‘true’) parameters.

Type III Wald test on choice data from the GLMM.

The binary dependent variable was choice (0=rest,1=work). Group, Recipient, Effort, Reward, and their interactions were fixed effects. We included a subject-level random intercept and a random slope for Recipient. Significant results are shown in bold.

Post-hoc comparisons of choice data.

The interaction between Group and Effort, and the interaction between Group, Recipient and Effort, are shown. All P values are Bonferroni corrected and significant results are shown in bold. Means were extracted using the emmeans package in R (Lenth, 2021).

Type III Wald test on force data from the LMM.

Group, Recipient, Effort, Reward, and their interactions were fixed effects and force was the dependent variable - area under the curve during the force period relative to each participants’ MVC. We included a subject-level random intercept and a random slope for Recipient. Significant results are shown in bold.