Computational and Neurobiological Substrates of Cost-Benefit Integration in Altruistic Helping Decision

Abstract

Although altruistic behaviors, e.g., sacrificing one's own interests to alleviate others' suffering, are widely observed in human society, altruism varies greatly across individuals. Such individual differences in altruistic preference have been hypothesized to arise from both individuals' dispositional empathic concern for others' welfare and context-specific cost-benefit integration processes. However, how cost-benefit integration is implemented in the brain and how it is linked to empathy remain unclear. Here, we combine a novel paradigm with the model-based functional magnetic resonance imaging (fMRI) approach to examine the neurocomputational basis of altruistic behaviors. Thirty-seven adults (16 females) were tested. Modeling analyses suggest that individuals are likely to integrate their own monetary costs with nonlinearly transformed recipients' benefits. Neuroimaging results demonstrate the involvement of an extended common currency system during decision-making by showing that selfish and other-regarding motives were processed in dorsal anterior cingulate cortex (ACC) and right inferior parietal lobe in a domain-general manner. Importantly, a functional dissociation of adjacent but different subregions within anterior insular cortex (aINS) was observed for different subprocesses underlying altruistic behaviors. While dorsal aINS (daINS) and inferior frontal gyrus (IFG) were involved in valuation of benefactors' costs, ventral aINS and middle INS (vaINS/mINS), as empathy-related regions, reflected individual variations in valuating recipients' benefits. Multivariate analyses further suggest that both vaINS/mINS and dorsolateral prefrontal cortex (DLPFC) reflect individual variations in general altruistic preferences which account for both dispositional empathy and context-specific other-regarding tendency. Together, these findings provide valuable insights into our understanding of psychological and neurobiological basis of altruistic behaviors.SIGNIFICANCE STATEMENT Altruistic behaviors play a crucial role in facilitating solidarity and development of human society, but the mechanisms of the cost-benefit integration underlying these behaviors are still unclear. Using model-based neuroimaging approaches, we clarify that people integrate personal costs and non-linearly transformed other's benefits during altruistic decision-making and the implementations of the integration processes are supported by an extended common currency neural network. Importantly, multivariate analyses reveal that both empathy-related and cognitive control-related brain regions are involved in modulating individual variations of altruistic preference, which implicate complex psychological and computational processes. Our results provide a neurocomputational account of how people weigh between different attributes to make altruistic decisions and why altruistic preference varies to a great extent across individuals.

Keywords: altruistic behavior; cost-benefit integration; empathy; model-based fMRI.

Figure 1.

Experimental design and behavioral results. A, The procedure of the experiment. Participants performed the tasks in two sessions on two separate days. In session 1, participants performed the noise rating task, noise and visual stimuli association task, and interpersonal helping task outside fMRI scanner. In session 2, participants performed the noise and visual stimuli association task again outside scanner and the interpersonal helping task in the scanner. B, The procedure of noise rating task. Participants rated unpleasantness for noise stimuli with different levels of intensity on a VAS. C, We estimated a power function (red curve) with individual's unpleasantness rating data, and then selected 10 levels of noise stimuli specific for each participant from mild to extremely unpleasant with equal interval of subjective unpleasantness difference between adjacent levels. D, The procedure of noise and visual stimuli association task. Each of the 10 selected noise stimuli was associated with each of 10 different visual cues (i.e., blue trumpets). Visual stimuli with more trumpets correspond to noise stimuli with higher unpleasantness score. E, The procedure of interpersonal helping task. In each trial, participants decided whether or not to forgo a certain amount of money to prevent the partner from receiving a clip of noise stimuli with a certain level of unpleasantness. Each trial began with a sentence “Pairing, please wait!” on the screen for 1–7 s. Then, the participant's own portrait and a faceless silhouette representing the partner, together with participants' cost amount and a visual cue representing the noise unpleasantness level the partner will receive were presented on the screen for 3 s. Then, the question “Whether to donate?,” together with “yes” and “no” options, was presented in the lower part of the screen. The participant had to make his/her choice within 3-s time limitation. F, Behavioral results in interpersonal helping task for both sessions. Left panel, WTP is depicted as a function of recipients' (partners') noise unpleasantness level in session 1. Error bars indicate SEM, CNY, Chinese Yuan. Right panel, Helping rate is depicted as a function of benefactors' (participants') cost amount level and recipients' (partners') noise unpleasantness level across all the trials over all participants in session 2. Each cell represents one specific cost amount level-noise unpleasantness level pairing condition.

Figure 2.

Computational modeling results. A, Model comparison results. Model F1.2 (interdependent-and-nonlinear weight of recipients' benefits model) outperforms than all the other models in the RFX-BMS analysis. Model F1.2 has the highest exceedance probability (xp = 0.71), suggesting that the probability that model F1.2 is more likely implemented than all the other models is 71%. B, Correlation between BEES score (dispositional empathy) and log-transformed κ in model F1.2 (other-regarding preferences). C, Bar plots show that cross-validation prediction accuracies are significantly higher than chance level (i.e., 0.5) for all the 15 models of interest. Error bars indicate SEM. D, Scatter plots for correlations between estimated parameters with model F1.2 in the two sessions. E, Model parameters recovered from simulated response data for each participant; 100 sets of response data were simulated with model F1.2, each participant's specific cost amount–noise unpleasantness level pairing choice set, and his/her own best-fitting parameters. Then, model parameters in model F1.2 were estimated with these 100 sets of response data for each participant's cost amount–noise unpleasantness level pairing choice set, and averaged across the 100 sets of simulated parameters. Scatter plots show the association between the averaged simulated parameters (y-axis) and the estimated parameters fitted by observed behavioral data (x-axis) across all the participants. Dashed blue lines are the diagonal lines. Each dot represents one participant; ***p < 0.001.

Figure 3.

Parametric analysis results in GLM 1. dACC (A) showed positive associations with ${\mathrm{SV}}_{\mathrm{cost}}$ (blue region), and rIPL (B) showed positive associations with ${\mathrm{SV}}_{\mathrm{benefit}}$ (orange region). ROI conjunction analysis revealed that part of dACC and rIPL are associated with both ${\mathrm{SV}}_{\mathrm{cost}}$ and ${\mathrm{SV}}_{\mathrm{benefit}}$. Parametric estimate values corresponding to each of the two modulators (${\mathrm{SV}}_{\mathrm{cost}}$ and ${\mathrm{SV}}_{\mathrm{benefit}}$) were extracted from dACC (A, right panel) and rIPL (B, right panel) identified in GLM 1 conjunction analysis. The parametric estimate values were the averaged values across the voxels in a within 3-mm edge cube and centered at the peak coordinate of each region (ACC: 9, 38, 19; rIPL: 54, −58, 49). C, Bilateral daINS/IFG showed positive associations with ${\mathrm{SV}}_{\mathrm{cost}}$. Neural results were thresholded at voxel- wise p < 0.001 uncorrected and cluster-wise FWE corrected p < 0.05. Error bars indicate SEM.

Figure 4.

Parametric analysis results in GLM 2. A, MPFC, including VMPFC, showed positive associations with utility difference between chosen and unchosen choice. B, MCC/SMA, left IFG and right DLPFC showed negative associations with utility difference between chosen and unchosen choice. Neural results were thresholded at voxel-wise p < 0.001 uncorrected and cluster-wise FWE corrected p < 0.05.

Figure 5.

Mediation analysis results. A, Whole-brain analyses showed the association between neural representation of recipients' benefits in vaINS/mINS with log-transformed κ in model F1.2 (other-regarding preferences). B, Scatter plot for the correlation between parametric estimates of recipients' benefits in vaINS/mINS and log-transformed κ. Parametric estimates in vaINS/mINS was averaged values across the voxels in a region within 3-mm edge cube and centered at the peak coordinates (45, 8, −5). C, Path diagram shows the mediation pathway. The predictor variable (BEES scores) shown on the left predicts neural representation of recipients' benefits in vaINS/mINS (path a for the mediator variable). The mediator variable (vaINS/mINS) predicts individuals' other-regarding preferences (log-transformed κ) after controlling for individuals' empathy disposition (path b). The effect of dispositional empathy on other-regarding preferences after controlling for the mediator variable (path c') is not significant any more, suggesting that neural estimates of vaINS/mINS fully mediate the effect of dispositional empathy on other-regarding preferences. Path coefficients are labeled on the lines. Bootstrapping analysis suggested that this indirect effect is significant after 20,000 bootstraps.

Figure 6.

Differentiation between right vaINS/mINS and daINS/IFG. A, Neural representations of recipients' benefits in vaINS/mINS (yellow) mediate the effect of dispositional empathy on other-regarding preferences, and activity in daINS/IFG (cyan) is associated with ${\mathrm{S}\mathrm{V}}_{\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{t}}$. B, Insular subregions template (k = 3 solutions) from Kelly et al. (2012): dorsal anterior insular (green), ventral anterior and middle insular (red), and posterior insular (blue). C, Mapping vaINS/mINS (yellow) and daINS/IFG (cyan) onto Kelly's insular subregions template suggests that vaINS/mINS is mainly located in the vaINS and mINS, and daINS/IFG is mainly located in the dorsal anterior part of insular and IFG. D, Activity in daINS/IFG showed significant associations with ${\mathrm{S}\mathrm{V}}_{\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{t}}$; and, none of activity of ${\mathrm{S}\mathrm{V}}_{\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{t}}$ in vaINS/mINS, activity of ${\mathrm{S}\mathrm{V}}_{\mathrm{b}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{f}\mathrm{i}\mathrm{t}}$ in daINS/IFG or in vaINS/mINS was significant from 0, ***p < 0.001. E, Scatter plots for the correlations between log-transformed κ and parametric estimates of ${\mathrm{S}\mathrm{V}}_{\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{t}}$ in vaINS/mINS and daINS/IFG (left panel), and scatter plots for the correlations between log-transformed κ and parametric estimates of ${\mathrm{S}\mathrm{V}}_{\mathrm{b}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{f}\mathrm{i}\mathrm{t}}$ in vaINS/mINS and daINS/IFG (right panel). The correlation coefficient of ${\mathrm{S}\mathrm{V}}_{\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{t}}$ was not significant for both vaINS/mINS and daINS/IFG (left panel), and the correlation coefficient of ${\mathrm{S}\mathrm{V}}_{\mathrm{b}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{f}\mathrm{i}\mathrm{t}}$ was significant for vaINS/mINS but not for daINS/IFG (right panel). The parametric estimate values for vaINS/mINS were the averaged values across the mINS template in Kelly et al. (2012); and the parametric estimate values for daINS/IFG were the averaged values across a cluster combining the anterior dorsal INS in Kelly et al. (2012) and the IFG in AAL templates. If parametric estimates of ${\mathrm{S}\mathrm{V}}_{\mathrm{b}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{f}\mathrm{i}\mathrm{t}}$ were extracted from the peak coordinates of right daINS/IFG (MNI coordinates: 48, 17, 1) and right vaINS/mINS (MNI coordinates: 45, 8, −5) identified in previous analyses, the correlation coefficient between other-regarding preferences (i.e., log-transformed $\mathrm{\kappa }$) and ${\mathrm{S}\mathrm{V}}_{\mathrm{b}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{f}\mathrm{i}\mathrm{t}}$ signal was significantly stronger in vaINS/mINS (r = 0.64, p < 0.001) than in daINS/IFG (r = 0.41, p = 0.02; Z = 2.27, p = 0.023). Whole-brain neural results were thresholded at voxel-wise p < 0.001 uncorrected and cluster-wise FWE corrected p < 0.05. Error bars indicate SEM.

Figure 7.

Illustration of the IS-RSA. A, Procedure of performing IS-RSA. First, we created a parameter RDM, which measured the dissimilarity across participants in general altruistic preference that was calculated by the Euclidean distance between each pair of participants in z-scored BEES (a measure of dispositional empathy) and z-scored log-transformed κ (a measure of task-specific altruistic preference) driven from the winning model (also see the scatter plot showing the relationship between the two measures; each dot represents the data of a single participant). Next, we built a neural RDM for each of the hypothesized ROIs (here we used bilateral vaINS extending to mINS as an example), which was measured by the correlation distance between the multivoxel patterns in each ROI of SV_cost (or SV_benefit) of each pair of participants. Last, we calculated the Spearman rank-order correlation between these two RDMs and implemented a permutation test with Bonferroni correction to confirm the statistical significance. Notably, neural RDMs shown here were based on parametric contrasts map of SV_cost. Multivoxel patterns (heatmaps in gray scale) shown here were only for illustration. B, ROIs used in IS-RSA. ROI masks were defined based on a whole-brain parcellation given a meta-analytic functional coactivation map of the Neurosynth database (http://neurovault.org/collections/2099/). RDM, representational dissimilarity matrix; SV, subjective value; sbj, subject; BEES, balanced emotional empathy scale; daINS, dorsal anterior insular; mINS, middle insular; vaINS, ventral anterior insular; TPJ, temporoparietal junction; DLPFC, dorsolateral prefrontal cortex.