Neural representations of vicarious rewards are linked to interoception and prosocial behaviour

Every day we constantly observe other people receiving rewards. Theoretical accounts posit that vicarious reward processing might be linked to people’s sensitivity to internal body states (interoception) and facilitates a tendency to act prosocially. However, the neural processes underlying the links between vicarious reward processing, interoception and prosocial behaviour are poorly understood. Previous research has linked vicarious reward processing to the anterior cingulate gyrus (ACCg) and the anterior insula (AI). Can we predict someone’s propensity to be prosocial or to be aware of interoceptive signals from variability in how the ACCg and AI process rewards? Here, participants monitored rewards being delivered to themselves or a stranger during functional magnetic resonance imaging. Later, they performed a task measuring their willingness to exert effort to obtain rewards for others, and a task measuring their propensity to be aware and use interoceptive signals. Using multivariate similarity analysis, we show that people’s willingness to be prosocial is predicted by greater similarity between self and other representations in the ACCg. Moreover, greater dissimilarity in self-other representations in the AI is linked to interoceptive propensity. These findings highlight that vicarious reward is linked to bodily signals in AI, and foster prosocial tendencies through the ACCg.


Introduction
Figure 1. fMRI and Behavioural tasks. Participants completed three tasks across two sessions separated by at least one week A. fMRI session: The value representation task. This task measured neural similarity between representation of rewards and losses for the self and others. Participants passively witnessed financial gains (6 levels) and losses (6 levels) for themselves and for another unknown person, the Receiver. Catch trials asked either who was the recipient of the last outcome, or the magnitude of the reward on the previous trial, to encourage attention to the task. Similarity was calculated as the Pearson correlation between the spatial representation of self-and other-related neural activity. B Prosocial Effort task. Participants made choices about whether to exert different amounts of effort (30-70% of their own maximum grip strength) on a handheld dynamometer for variable amounts of reward (2-10 credits). Participants worked either to benefit themselves or an anonymous other. C. Interoceptive Respiratory Task. On each trial participants blew into a peak-flow meter twice, the first blow setting a standard for that trial. In the second blow, participants were required to achieve a percentage of the first, with participants estimating their actual percentage performance at the end of the trial. This was conducted in an internal condition with white noise played through headphones to prevent the use of external auditory cues, and an external condition where external auditory cues were available. The difference in estimation accuracy between internal and external conditions was taken as a measure of how much participants rely on internal vs external signals. Note that the sequence of screens is an illustration of the test, as participants were blindfolded.
conditions, an 'internal' condition where no external auditory cue could guide their performance, and 166 an 'external' condition where external auditory cues caused by their exhalation were available to be 167 used. Using this technique, a metric of reliance on interoceptive cues was calculated, by comparing

175
Behavioural results: reliance on interoceptive signals is associated with prosocial motivation.

176
Previous research has suggested that people's propensity to rely on interoceptive signals is 177 correlated with their levels of altruism (16, 55). However the tasks used to measure prosocial or 178 interoceptive tendencies may confound different processes together and there have been questions 179 of their validity (51,56,57). Here, we address these issues using tasks to measure prosocial 180 behaviour and interoceptive processes without such confounds. Firstly, studies have used economic 181 games in which the rewards delivered to the other person directly impact on the magnitude of rewards 182 obtained oneself. Although interesting for quantifying variability in one's desire to benefit others, they 183 cannot distinguish between two different motives -sensitivity to one's own rewards, or an increasing people were not incentivised to choose to work more often as the rewards that would be received by Figure 2. Reliance on interoceptive signals is associated with prosocial choices. A. Interoception score did not influence motivation for rewards in self trials. Participants did not show different patterns in their proportion of choices to work over rest (y-axis) depending on their level of reliance on internal vs external signals (x-axis) across different reward magnitudes. B. Interoception influences motivation for others' rewards. As participants relied more on internal than external cues, they chose to work more as the reward to be received by another person increased. Participants who relied more on external cues were more reluctant to work to benefit others regardless the reward on offer. Shaded areas show the 70% confidence interval around the slopes. Individual points show the score of each participant for each condition. C. People who rely more on internal vs external signals (x-axis), weighted rewards more similarly for self and other when making decisions to work to obtain rewards. Y-axis depicts the difference between self and other betas from two mixed models predicting choices separately for self and other. Higher values indicate self rewards are valued more than others'. D. Participants who relied more on internal vs external cues (x-axis), were more sensitive to others' rewards (beta estimates, y-axis). Shaded areas show the 95% confidence interval around the slopes. Individual points show the score of each participant. with significant effects of effort, reward and beneficiary, as well as interactions between reward and 232 beneficiary, and effort and beneficiary (see Supplementary Fig. S3A and S3B). In sum, we show 233 that how much someone relies on internal signals is linked to how prosocial they are and not to their 234 motivation to work for themselves. Specifically, more internally-focused individuals are more 235 incentivised by the rewards that can be obtained for another person, and work more when they can 236 obtain a bigger benefit for them.

237
Given that in the prosocial effort task the majority of people choose to work more at lower 238 reward levels for themselves than for other people, this leads to a prediction that people with more 239 positive interoception scores (i.e. more internally focused, relying more on interoceptive signals) 240 would show less of a difference in decisions to work between self and other, as a function of reward.

241
Furthermore, given the results of the mixed model above, greater interoception scores should be 242 specifically associated with sensitivity to others' rewards and not with sensitivity to self rewards nor 243 sensitivity to effort. Thus, as a confirmatory, post-hoc analysis, we extracted beta parameters for effort 244 and reward from two mixed effect models where decisions to work or rest were taken separately for 245 self and other trials (see Materials and Methods equations 5 and 6). Thus, these indexes were a 246 proxy for individual differences in sensitivities to effort levels and magnitudes of rewards for self-247 benefitting and prosocial decisions.

248
From the results described above, we expected interoception to be associated with sensitivity 249 to others' rewards and not to effort or self rewards. We tested this using Pearson correlations 250 between the interoception score, and reward and effort betas for self, other and their difference.

251
Results revealed a significant negative correlation between interoception and the difference in reward 252 betas, such that people who relied more on internal signals were more similarly motivated for self and 253 other rewards, while people who were more externally focused valued more their own rewards 254 compared to others' (r(59) = -0.39, p < 0.003, Fig. 2C). To test if this effect of interoception on prosocial 255 motivation was specific to sensitivity to rewards, we correlated difference in effort betas with 256 interoception. We found null correlation (r(59) = 0.15, p = 0.26), which was significantly different from 257 the correlation between interoception scores and sensitivity to reward (Fisher's Z transformation, z = -258 3, p < 0.003). As expected, the interoception score also correlated with sensitivity to others' rewards 259 (r(59) = 0.27, p < 0.04, Fig. 2D), but not with self rewards (r(59) = -0.14, p = 0.29; Supplementary Fig.   260 S2B), and these effects were significantly different from each other (z = 2.23, p < 0.03), suggesting 261 specificity of the social effects. Furthermore, interoception did not correlate with neither other (r(59) = -262 0.08, p = 0.54) nor self effort betas (r(59) = -0.13, p = 0.32), and these correlations showed a significant rewards in the effort task. Taken together, these results reveal that people who rely more in internal 266 versus external signals are specifically more driven by others' rewards when deciding whether to 267 expend energy to act prosocially. 268 269 fMRI results: Different roles of the ACCg and the right dorsal AI in vicarious rewards.

270
Next, we examined whether the degree of similarity between neural patterns evoked by self 271 and other rewards during the fMRI value representation task was predicted by prosocial motivation

281
Five regression models were used to test whether the degree of similarity between self and 282 other representations of reward was associated with interoception, one in each ROI, with the similarity 283 of self and other representations being the dependent variable and participants' interoception score 284 (reflecting their reliance on interoceptive signals) as a predictor. Only one area, the RdAI, showed a 285 significant effect (b = -1.01, F(1,52) = 11.12, p < 0.008 FDR corrected). No effects were found in any of 286 the other ROIs. Within the RdAI, less similar responses between self and other rewards were linked to 287 people relying more on their internal signals (Fig. 3A), suggesting that a greater specialisation of 288 neural responses to others' rewards in this region occurs in people who are more aware and use 289 interoceptive signals.

290
Next, we hypothesised that variability in vicarious reward responses would be related to how 291 incentivised a participant was by others' rewards in the prosocial effort task. We used the same 292 regression approach in each ROI, to test whether similarity in the neural response between self and 293 other reward was related to a beta-weight reflecting someone's sensitivity to others' rewards in the 294 prosocial effort task. We found a significant effect in only one ROI, the ACCg (b = 0.07, F(1,53) = 7.36, 295 p < 0.05 FDR corrected, Fig. 3B). Within the ACCg, greater similarity in responses between self and 296 other reward was predictive of increased incentivisation by others' rewards in the prosocial effort task.

297
That is, people who showed a greater increase in choosing to help others as the rewards on offer 298 increased, showed more similar neural patterns between self and other rewards. Note that both the 299 results within the ACCg and RdAI were also present independently of the statistical method used to Analyses of the behavioural data revealed that interoception and sensitivity to others reward investigate whether the effects in the ACCg and RdAI remain when this shared variance is included in 308 the model, two regression models were conducted, one in each region, in which the beta weight 309 measuring incentivisation by other rewards from the effort task, and the interoception score, were  A. Interoception was associated with multivariate similarity between self and other rewards only in the RdAI, among five ROIs (p < 0.008 FDR corrected). People who relied more on internal relative to external signals showed more dissimilar neural responses to reward between self and other. Y-axis depicts the similarity values, with higher values meaning more similarity between self and other reward representation. B. Prosocial motivation -specifically how incentivised a participant was to obtain rewards for others -was associated with multivariate similarity between self and other rewards only in the ACCg (p < 0.05 FDR corrected). Participants who were more motivated to work for others' rewards showed more similar neural patterns between self and other rewards in the ACCg. X-axis corresponds to the reward betas obtained from a mixed model where decisions to work for others in the prosocial effort task was predicted by effort, reward and their interaction. Positive values indicate higher weights for others' rewards.
Interestingly, the association between interoception and AI was specifically located in its 382 dorsal part in the right hemisphere. This is consistent with previous work that proposes functional and 383 anatomical division in the AI (63, 64). Thus, its ventral portion is believed to be mainly involved in 384 affective reactivity toward salient outcomes impacting self and others, while its dorsal area (especially 385 in the right hemisphere) has been more specifically linked to interoception, providing a bodily map for 386 a wide range of mental processes (32,34,38,46,(63)(64)(65)(66). Posterior insula, involved in primary 387 interoceptive representation of the physiological state of the body through thalamocortical pathways, Notably, our results suggest that response to others' reward in the ACCg better reflects 397 people's motivation to perform prosocial acts than their interoceptive propensity. Classical accounts 398 have suggested that there is a functional dissociation between these regions, with the insula 399 implicated in affective and bodily processes and frontal areas including the ACC implicated in 400 motivating behaviour (34,(43)(44)(45). However, the presence of vicarious reward signals in both the AI 401 and ACCg did not seem to fit with such a division. Here we show that it is the degree to which the AI 402 represents others' rewards as distinct from self rewards that is associated with interoceptive signals, 403 and the same distinctiveness in ACCg that is associated with levels of motivation, specifically the 404 willingness to help others. As such, the hypothesised division between AI and ACC suggested by  there was evidence that greater similarity in patterns between self and other reward was linked to 435 increased prosociality in the effort task. Although this might argue in favour of a common-currency 436 account, it is notable that the multivariate technique examines spatial disparity in terms of patterns, 437 rather than overlap in voxels as examined with traditional methods. This work highlights that neither 438 account in isolation can explain how the brain processes social information. This is also reflected in 439 how the interaction between interoception and prosocial behaviour might be implemented in the brain.

440
We showed that ACCg and RdAI are more functionally connected when people observe others' 441 reward outcomes than self. This could suggest that these areas, and potentially the link between 442 interoception and prosocial behaviour, are mainly socially-specific. However, interpreting the links 443 between behaviour, multivariate representations and connectivity seems too speculative, and 444 conciliating these three approaches is beyond the scope of this study. Thus, the connectivity results 445 add complexity to our main findings, and encourage future research to shed light on the 'social brain' 446 debate. As such, common-currency and specialisation rooted accounts will need to be adapted to 447 account for multivariate similarity analyses, and individual differences in such patterns, to move 448 forward our understanding of the implementation of social information processing in the brain (47).

449
Finally, we found a behavioural link between prosocial motivation and interoception.

450
Theoretical accounts posit that interoceptive signals might drive people to behave more prosocially 451 (12,16,60,80). Here, we show that people who have higher propensity to be aware and use 452 interoceptive signals are more incentivised and motivated to help others, especially when the reward 453 to be received by others increases, than people who rely more on external signals. Thus, we expand 454 previous research reporting levels of altruism in economic games associated with levels of 455 interoception (16). We used measurements that allow us to identify that increased sensitivity to 456 rewards for others was specifically associated with reliance on interoceptive signal, rather than 457 changes in one's own experience of reward or effort. This might relate to the potential link between 458 empathetic processes and interoception (81). Our results therefore support the notion that prosocial 459 behaviour may be driven by how sensitive someone is to their internal bodily signals, mediated by 460 sensitivity to others' outcomes.

461
In conclusion, we show that vicarious reward processing, interoception and prosocial 462 behaviour are closely linked. People who show a greater propensity to be aware and use 463 interoceptive signals are more incentivised when they can obtain rewards for others and act more 464 prosocially. Neural representations of passively observed vicarious reward in the AI and ACCg were catch trials, participants were asked to indicate as quickly as possible if recipient of the previous trial 525 was the self or the receiver (i.e. "You or Receiver?") or to identify the amount that was awarded in the 526 previous trial compared to a randomly generated amount (e.g. "Won £15 or Lost £9") using a 527 response box. Accuracy at the catch trial task was computed by counting the number of correct 528 answers and dividing it by the total number of catch trials. Participants who had an accuracy below 529 50% of the catch trials were excluded from analysis (n = 3).

530
Behavioural session: Prosocial effort task 531 The prosocial effort task measures how motivated participants are to obtain rewards for self 532 and others (8,48,52). Participants in this task were instructed that, as Deciders, they would be paired 533 with one of the Receivers of the experiment, but not the same one they had in the fMRI session. The 534 task consisted of participants making decisions between options with different magnitudes of financial 535 rewards (represented by number of credits) in exchange for different levels of physical effort (grip 536 force). For each trial, participants chose between two options: a rest baseline option associated with 537 no effort and low reward (1 credit); and a work offer option, which results in higher monetary gain (2, 538 4, 6, 8 and 10 credits) for higher effort that varies across 30, 40, 50, 60 or 70% of each participant's the work option was chosen, participants needed to make the effort required to obtain the credits on 542 offer -they needed to squeeze a handle with the required force with their dominant hand for 1 s out of and reward levels in the work option varied independently over trials, and each effort-reward 546 combination was repeated three times per beneficiary condition, giving a total of 150 trials: 75 self 547 trials and 75 other trials. The rest option with one credit was used to make sure that there was a 548 conscious and motivated decision to choose the alternative work option. If a choice was not selected, 549 zero credits were given. All trials had the same duration, controlling for potential temporal discounting 550 effects. Participants who did not actively choose any of the options for more than 10% of the trials 551 were excluded from analysis (n = 1).

552
Before participants made decisions in the prosocial effort task, they were asked to grip a 553 handheld dynamometer with as much force as they could to determine their MVC. Thus, the task 554 measures similar effort levels across participants regardless of their variability in strength. After MVC 555 estimation and prior to the main decision task, participants completed 18 trials where they 556 experienced each effort level three times, and also learned to associate each level of effort with the 557 elements in the pie chart (e.g. one element of the pie chart corresponded to 0% force, i.e. the rest 558 option).

559
Behavioural Session: Interoception respiratory task 560 The goal of this task (51) was to assess how much participants rely on their internal signals 561 relative to exteroceptive cues when assessing the force of their exhalation. To assess participants' 562 force of exhalation, a standard peak flow meter was used. Participants who had a history of breathing 563 difficulties did not undertake this task. On each trial, participants were first required to make a large 564 exhalation into the peak flow meter. This first exhalation was taken as the 'standard' for that trial, i.e.,

565
100% performance. Participants were then required to perform a second exhalation. In this second, 566 'target' exhalation, participants were told to aim to perform a percentage of the standard exhalation 567 (e.g., 30%) for that trial. There were four possible targets for each trial -30, 50 70 or 90% of the 568 standard. Once the participant had performed the target, they were asked by the experimenter to

591
During the task, and across conditions, participants were blindfolded, to prevent them using 592 visual information to aid performance. For both internal and external conditions, participants 593 completed six repetitions of the four targets presented in a random order, with a total of 24 trials per 594 condition. The order in which the internal and the external conditions were presented was 595 counterbalanced across participants. To ensure that the 30% target trial could be measured, 596 participants were required to surpass a threshold of 200 L/min in their standard exhalation. If they did 597 not accomplish this, the standard blow was repeated until the threshold was surpassed. No feedback 598 about participants' performance was provided across the experiment.

599
The peak-flow meter into which participants performed their exhalations was gently secured in 600 a horizontal position using a vice clamp and elevated in line with each participant's mouth using a 601 stand. Participants were instructed to keep their hands resting at the bottom of the stand during 602 exhalations, using their hands only to locate the gauge prior to performing exhalations, and to be still 603 in the chair and sat upright, without pushing the mouthpiece forward while exhaling. Trials in which 604 these conditions were not accomplished were repeated or removed from analysis. Participants with 605 more than 10% of missed trials in either of the two conditions were excluded (n = 3).

608
Interoception score: reliance on internal vs external signals.

609
To calculate participants' reliance on their interoceptive signals, the absolute error scores 610 were computed for each trial of the interoception respiratory task, such that: between the actual performance of the participant in target T as a percentage of the standard, and 616 their estimation E, divided by their actual performance T. The mean of the AEs is computed 617 separately for the internal and the external conditions for each participant. The interoception score 618 was then the difference in performance between the internal and the external condition, such that: Where the interoception score for the participant j is the difference between their mean AEs in 621 the external condition and in the internal condition. Scores below zero indicate that a participant made 622 more mistakes in the internal than the external conditions, suggesting that they are more externally-

623
focused. The interoception score therefore reflects how much participants rely on internal versus 624 external signals.

625
Association between interoception and motivation 626 For the prosocial effort task, choices to work relative to rest were taken as an index of 627 motivation to obtain self and other rewards. Mixed effects models were used to predict trial-by-trial

643
As a post-hoc analysis, we tested for Pearson correlations between the interoception scores (http://www.rbmars.dds.nl/CBPatlases.htm). We created the ACCg mask by modifying the original 725 parcel using the imcalc function in SPM12. Thus, the left hemisphere parcel for the area 24ab was 726 duplicated onto the right hemisphere to create a bilateral mask. Furthermore, those voxels located in

740
Functional connectivity analyses

741
We used beta series regressions (59) to assess individual differences in the functional 742 connectivity between the ROIs in in each task condition. Specifically, for each participant, we 743 calculated the mean value in ACCg and RdAI at each trial and calculated the regression coefficient 744 between their combination across all trials for self and other. This allowed us to obtain a beta estimate 745 parameter indicating the functional connectivity between ACCg and RdAI for each participant and 746 condition. Paired t-test was used to test for differences in connectivity between self and other trials.
above three standard deviations were excluded (n = 3