The neural computation of human goal-directed behaviors in complex motivational states

Because the motives behind goal-directed behaviors are often complex, most behaviors result from the interplay between different motives. However, it is unclear how this interplay between multiple motives affects the neural computation of goal-directed behaviors. Using a combination of drift-diffusion modeling and fMRI, we show that the interplay between different social motives changes initial preferences for prosocial behavior before a person makes a behavioral choice. This increase in preferences for the prosocial choice option was tracked by neural responses in the bilateral dorsal striatum, which in turn lowered the amount of information necessary for choosing prosocial behavior. We obtained these results using a paradigm in which each participant performed the same behavior based on different, simultaneously activated motives, or based on each of the motives separately. Thus, our findings provide a model of behavioral choice computation in complex motivational states, i.e., the motivational setting that drives most goal-directed human behaviors.


3 5
During the study, participants were paired with four partners (confederates of the 1 3 6 experimenter). In the empathy condition, the participants repeatedly observed one of the confederates 1 3 7 (the empathy partner) receiving painful shocks in a number of trials, a situation known to elicit an 1 3 8 empathic response (Batson et al., 1995;Hein et al., 2016;Lamm et al., 2011) (see Methods for 1 3 9 details). The reciprocity motive is defined as the desire to reciprocate perceived kindness with kind 1 4 0 behavior (Gouldner, 1960;Hein et al., 2016;McCabe et al., 2003). Therefore, in the reciprocity

7 8 8
Note that the increase in prosocial choices in the multi-motive condition compared to the 1 7 9 reciprocity condition was observed although the motives were induced with equal strength (no 1 8 0 difference in induction ratings and frequency of prosocial choices between the motive conditions), and 1 8 1 had a comparable effect on prosocial choices.

8 2
To clarify this effect, we calculated the percent change in prosocial choices in the multi-motive 1 8 3 condition relative to each single motive condition where prosoc multi-motive equals the frequency of the prosocial choices in the multi-motive condition, 1 8 5 prosoc reciprocity equals the frequency of prosocial choices in the reciprocity condition, and prosoc empathy 1 8 6 equals the frequency of prosocial choices in the empathy condition.

8 7
The percent change of the multi-motive condition relative to reciprocity was significantly SEM)), indicating that the simultaneous activation of the reciprocity motive did not enhance the 1 9 3 empathy motive.

9 4
In the next step, we used hierarchical drift-diffusion modeling (HDDM) (Vandekerckhove et al., multi-motive condition that we observed relative to the reciprocity condition. We estimated the three

1 5
The percent change in the z-parameter of the multi-motive condition relative to the reciprocity

1 9
In addition, we had hypothesized that the combination of the two motives may increase the 2 2 0 amount of information that was processed to reach a decision (captured by the a-parameter; Fig. 1C, 2 2 1 right panel). Supporting this hypothesis, the a-parameter was higher in the multi-motive condition 2 2 2 compared to the reciprocity condition (lmm χ 2 = 4.76, P = .03), but not compared to the empathy 2 2 3 condition (lmm χ 2 = 2.21, P = .14). Correspondingly, there was a significantly positive percent change 2 2 4 in a-parameters in the multi-motive condition relative to the reciprocity condition condition, but not with the relative difference between the empathy and reciprocity motives (see Table   2 4 2 S2 for the whole-brain analysis).

4 3 4 4
Taken together, the DDM results showed that the combination of the two motives enhanced 2 4 5 participants' initial preferences for choosing the prosocial option, relative to the initial prosocial 2 4 6 preferences induced by the reciprocity motive (captured by the percent change in the z-parameter).

4 7
The combination of empathy and reciprocity also increased the amount of information that people 2 4 8 used to make a choice relative to the reciprocity motive, and, with a similar trend also relative to 2 4 9 empathy (captured by the percent change in the a-parameter). In contrast, the speed of information 2 5 0 accumulation, i.e., the efficiency of the choice process itself, was comparable between multi-motive 2 5 1 and single-motive conditions (no change in v-parameter).

5 2
It is possible that the observed percent changes in the multi-motive condition relative to the 2 5 3 reciprocity condition (in the z-and the a-parameters) originate from an interplay between the two 2 5 4 motives when simultaneously activated in the multi-motive condition. However, as we observed no 2 5 5 significant difference between the multi-motive condition and the empathy-condition, it is also 2 5 6 conceivable that the empathy motive replaced the reciprocity motive when the two motives were 2 5 7 activated simultaneously. In this case, the observed percent changes in the multi-motive condition 2 5 8 would reflect the dominance of empathy over reciprocity, instead of an interplay between the two 2 5 9 1 1 motives. If empathy replaced the co-activated reciprocity motive, the relative difference in the z-2 6 0 parameters and a-parameters between the empathy and the reciprocity conditions should predict the 2 6 1 individual extent of the percent changes in the multi-motive condition relative to the reciprocity 2 6 2 condition. To test this, we calculated the relative differences in the z-parameters and a-parameters 2 6 3 between empathy and entered them as predictors in a regression analysis, and tested their effects on the observed percent 2 6 6 changes in the multi-motive condition (Δz multi-motive/reciprocity ; Δ a multi-motive/reciprocity ). This analysis revealed 2 6 7 no significant effects (β = 0.11, P = .55; interaction with parameter type (z vs a): β = -0.02, P = .93).

6 8
These results demonstrated that the difference between the two motives cannot account for the 2 6 9 changes in choice parameters in the multi-motive condition relative to the reciprocity condition, 2 7 0 bolstering the claim that the multi-motive effects observed reflect an interplay between the two 2 7 1 motives.

7 2
Next, we investigated how the interplay between the two motives, and the resulting changes in prosocial choices between the multi-motive condition and the reciprocity condition, using second-level 2 7 7 regression. As a main result, the first analysis revealed activations in the bilateral dorsal striatum that 2 7 8 were related to the individual change in prosocial preferences (P(FWE cluster-corrected ) = 0.018; center co-2 7 9 ordinates: x = 30, y = 2, z = -2; x = -28, y = 7, z = 1; Fig. 2B, Table 1). The stronger the percent 2 8 0 increase in initial prosocial preferences in the multi-motive condition relative the reciprocity condition, 2 8 1 the stronger the neural response in bilateral dorsal striatum.  The respective second-level regression with the percent change in the a-parameter revealed 2 8 7 neural activity in the left anterior insula on a lower, uncorrected threshold (Table S1).

8 8
Again, we tested the alternative hypothesis that the increase in dorso-striatal activity may 2 8 9 reflect the dominance of empathy (captured by the relative difference in z-parameters between 2 9 0 empathy and reciprocity, Δ z empathy/ reciprocity ), instead of an interplay between the motives. We extracted 2 9 1 the individual beta estimates from the observed bilateral dorsal-striatum cluster ( Fig. 2B; using the 2 9 2 entire clusters in both hemispheres) for use as a dependent variable in a linear regression. The 2 9 3 predictors were the percent change in z-parameters (Δz multi-motive/reciprocity ) and the relative empathy vs 2 9 4 reciprocity difference (Δz empathy/ reciprocity ). The results showed a significant relationship between the 2 9 5 individual increase in dorso-striatal activity and the percent change in the multi-motive condition 2 9 6 relative to the reciprocity condition (Δz multi-motive/reciprocity β = .65, P = .00003), but not between neural 2 9 7 activity and the difference in the z-parameters between the empathy and reciprocity conditions 2 9 8 (Δz empathy/reciprocity β = -.15, P = .28) (Fig. 2C).

9 9
1 3 We also conducted a whole-brain analysis that compared the effect of Δ z multi-motive/reciprocity and 3 0 0 the effect of Δ z empathy/reciprocity on the neural multi-motive vs reciprocity contrast. Supporting the results 3 0 1 shown in Fig. 2C, we found stronger dorsal striatal activity for the whole-brain regression with Δ z multi-3 0 2 motive/reciprocity compared to Δ z empathy/reciprocity (Table S2). Together, these results showed that neural 3 0 3 responses in the bilateral dorsal striatum tracked the changes in initial prosocial preferences in the 3 0 4 multi-motive condition relative to the reciprocity condition, but not differences in initial prosocial 3 0 5 preferences between the single-motive conditions.

0 6
We conducted a final analysis to specify the mechanism through which the multi-motive

1 6
We conducted path analyses (Rosseel, 2012) to test these two alternative models. The 3 1 7 individual beta-estimates of initial prosocial preference-related activity in the bilateral dorsal striatum 3 1 8 were used as predictor variables. The individual percent changes in the amount of information 3 1 9 (Δa combined/reciprocity ) served as the mediator, and the percent change in prosocial choices in the multi-3 2 0 motive condition relative to the reciprocity condition (Δprosoc multi-motive/reciprocity ) was entered as the 3 2 1 dependent variable (Fig. 3). The results revealed significant indirect paths (standardized path a condition relative to the reciprocity condition in the dorsal striatum was used as the predictor variable.

5
The percent change of the amount of information in the multi-motive condition relative to the 3 3 6 reciprocity condition served as the mediator and the percent change in prosocial choices in the multi-3 3 7 motive condition relative to the reciprocity condition was entered as the dependent variable. The interacting motives affect behaviors, e.g., the computation of social choices, are poorly understood.

4 8
Our findings provide such a mechanism. We show that multiple social motives, per se, impede the Prior to the task, the individual thresholds for pain stimulation were determined for the participants and 4 1 9 all the confederates. Next, the participants and confederates were assigned to their different roles by a 4 2 0 manipulated lottery (drawing matches). In order to ensure that each participant was always assigned 4 2 1 her designated role as a participant (pain recipient during motive induction; decider during the decision 4 2 2 task), the drawing of the matches was organized in such a way that she always drew the last match.

2 3
The confederates were assigned to the roles of the empathy partner, the reciprocity partner, the multi-4 2 4 motive partner, or the baseline partner, counterbalanced across participants.

2 5
In accordance with these roles, two of the confederates first went to an ostensible other

3 7
Participants spent approximately 60 min in the scanner and the entire procedure lasted about 2.5 4 3 8 hours. To avoid possible reputation effects, which could influence participants' behavior, participants 4 3 9 were informed that they would not meet the confederates after the experiment.

1
Each empathy-induction trial started with a colored arrow shown for 1,000 ms, which indicated the 4 4 2 empathy partner. After this cue and a jittered (1,000-2,000 ms) fixation cross, the same colored flash German). The scale ranged from -4 (labeled "very bad") to +4 (labeled "very good") and was visually 4 4 8 1 8 displayed in steps of 1. Before analysis, the induction ratings were recoded such that high positive 4 4 9 values reflect strong responses to the induction procedure (strong empathy motive). Participants had 4 5 0 to respond within 6 s. The inter-trial interval was 1,500 ms. Empathy induction consisted of 12 trials: 4 5 1 nine that were ostensibly painful for their partner (i.e., the confederate).

5 2
Reciprocity induction 4 5 3 Each reciprocity-induction trial also started with an arrow colored in the reciprocity partner's color, 4 5 4 which pointed toward the seating position of the reciprocity partner (left or right) and was shown for 4 5 5 1,000 ms. Next, the participants were shown a flash displayed to the right and a crossed-out flash 4 5 6 displayed to the left of a centered fixation cross. Participants were told that this was the decision 4 5 7 screen, which the reciprocity partner also saw while making her decision to either save or not save the 4 5 8 participant from painful stimulation. After a jittered interval of 2,000 to 4,000 ms, a box appeared

0 3
Similarly, the reciprocity-induction procedure included trials that were identical to the empathy-5 0 4 induction trials, except that the reciprocity partner only received non-painful stimulation on these trials, 5 0 5 as visually represented by a light-colored flash. In total, the reciprocity-induction procedure consisted 5 0 6 of 12 of these control trials and 12 reciprocity trials (see above), i.e., 24 trials (identical to the other 5 0 7 conditions).

0 8
Baseline induction 5 0 9 2 0 The baseline procedure consisted of 24 trials in total, 12 trials in which the baseline partner only 5 1 0 received non-painful stimulation and 12 trials in which the computer decided whether the participant 5 1 1 would be saved from a painful stimulus or not. This computer's decision was visually represented by a 5 1 2 white-colored box either appearing around the crossed-out flash (saving the participant) or the normal 5 1 3 flash (not saving the participant). It was clearly explained to the participant that the white box did not 5 1 4 represent the decision of a person, but indicated the computer's choice.

1 5
Pain stimulator 5 1 6 For pain stimulation, we used a mechano-tactile stimulus generated by a small plastic cylinder (513 g).

1 7
The projectile was shot against the cuticle of the left index finger using air pressure (Impact Stimulator,

2 6
Choice task 5 2 7 The choice task was identical in all the conditions. Participants were asked to repeatedly Regression analyses were conducted using the R-packages "lme4 and "car" (R Core-Team, 5 4 2 2018). The frequency of prosocial choices was calculated for each participant for each condition 5 4 3 (empathy, reciprocity, multi-motive, and baseline) and entered as a dependent variable into a linear 5 4 4 mixed model (lmm) with conditions as fixed effects (empathy, reciprocity, multi-motive, and baseline) 5 4 5 and participants as random effects. To investigate the differences between the social motives more 5 4 6 closely, additional lmm analyses were conducted that only included the multi-motive condition and the 5 4 7 reciprocity or the empathy condition as fixed effects and participants as random effects. Parallel 5 4 8 analyses were conducted for reaction times and the DDM parameters v, z, and a.

4 9
To test whether the relative difference between empathy and reciprocity on the z-parameter 5 5 0 and a-parameter could explain the percent changes of these parameters in the multi-motive condition 5 5 1 compared to the reciprocity condition, the percent change values (Δz multi-motive/reciprocity and 5 5 2 Δ a multi-motive/reciprocity ) were entered as dependent variables in a linear regression model. The respective 5 5 3 relative differences (Δz empathy/reciprocity and Δ a empathy/reciprocity ) and one regressor modeling the parameter 5 5 4 type (z-parameter, a-parameter) were included as predictors.

5
To test whether the relative difference between empathy and reciprocity on the z-parameter 5 5 6 could explain the effect in dorsal striatum, beta estimates from the neural contrasts between the multi-5 5 7 motive reciprocity conditions in the bilateral dorsal striatum were entered as the dependent variable, 5 5 8 and Δ z multi-motive/reciprocity and Δ z empathy/reciprocity were entered as predictor variables. showed that models that allowed for trial-by-trial variation in the drift rate, v, the non-decision time, t, 5 6 6 the initial decision preference, z, and the amount of processed information, a, yielded the best model values < 1.01) (Gelman & Rubin, 1992). Parameters of interest from the best-fitting model were 5 7 0 2 2 extracted for further analysis. Specifically, for each participant, the condition-specific v-parameters, z-5 7 1 parameters, and a-parameters were extracted (resulting in 12 parameters per participant).

7 2
In the next step, the parameters were entered as dependent variables in lmms, with conditions 5 7 3 as fixed effects and participants as random effects (one model per parameter). For closer investigation 5 7 4 of the effects between the social motives, additional lmm analyses were conducted that only included 5 7 5 the multi-motive condition and the reciprocity condition or the empathy conditions as fixed effects and 5 7 6 participants as random effects.

9 3
Spatial preprocessing included realignment to the first scan, and unwarping and coregistration to the 5 9 4 T1 anatomical volume images. Unwarping of geometrically distorted EPIs was performed using the 5 9 5 FieldMap Toolbox. T1-weighted images were segmented to localize grey and white matter, and 5 9 6 cerebro-spinal fluid. This segmentation was the basis for the creation of a DARTEL Template and 5 9 7 spatial normalization to Montreal Neurological Institute (MNI) space, including smoothing with a 6 mm 5 9 8 (full width at half maximum) Gaussian Kernel filter to improve the signal-to-noise-ratio. To correct for 5 9 9 low-frequency components, a high-pass filter with a cut-off of 128 s was used. additional regressor of no interest was included, which modeled the potential effects of session.

1 4
For the second-level analyses, contrast images for comparisons of interest (empathy > reciprocity, 6 1 5 multi-motive > empathy, reciprocity > empathy, and multi-motive > reciprocity) were initially computed 6 1 6 on a single-subject level. In the next step, the individual images of the main contrast of interest (multi-6 1 7 motive > reciprocity) were regressed against the percent change in the z-parameter (Δz multi-6 1 8 motive/reciprocity ) and a-parameter (Δa multi-motive/reciprocity ) in the multi-motive condition, relative to the 6 1 9 reciprocity condition, using second-level regressions. To test if the neural response in the dorsal 6 2 0 striatum was related to the relative difference in z between empathy and reciprocity (Δz empathy/reciprocity ), 6 2 1 the (multi-motive > reciprocity) contrast was regressed against the empathy vs reciprocity z-6 2 2 differences (Δz empathy/reciprocity ) and the multi-motive z-enhancement (Δz multi-motive/reciprocity ) in the same 6 2 3 model. The individual beta-estimates of the neural multi-motive condition > reciprocity contrast were 6 2 4 extracted from the bilateral clusters in the dorsal striatum resulting from the second-level regression 5 4