Nature Neuroscience: doi:10.1038/nn · In addition, the positive correlation between DLPFC and healthy participant choices also suggested that DLPFC patients, like that of OFC and

Supplementary Figure 1

DLPFC lesion reconstruction separated by hemispheres.

(a) Structural MRI slices illustrating the lesion overlap for left DLPFC patients. (b) Reconstruction for the right DLPFC patient. (c)Combined DLPFC lesion reconstruction overlaid to the left hemisphere for comparison.

Nature Neuroscience: doi:10.1038/nn.3798


Individual lesion reconstruction.

(a) DLPFC mean lesion volume was 125.76 cm3 with maximal cortical lesion overlap (>50%) in the Brodmann areas 6, 8, 9 and 46 and encompassed portions of the middle and superior frontal gyri in all patients. (b) OFC mean lesion volume was 72.29 cm3 with maximal cortical lesion overlap (>50%) in Brodmann areas 10, 11, and 47, centered in the OFC.



Choice behavior of patient cohorts in comparison with age-matched controls.

(a) Identical to Fig. 2B, except healthy participants have been partitioned such that they age-match patient cohorts on both mean and standard deviation. Consistent with results from the combined healthy comparison cohort, DLPFC cohort differed significantly from their age-matched healthy controls in terms of the paired difference in amount given between the Message and Choice conditions (Kruskal–Wallis test, p<.001, two-tailed), whereas this paired difference was similar between OFC and their age-matched controls (Kruskal–Wallis test, p<.50, two-tailed). (b) Identical to Fig. 2C, except healthy comparison participants were partitioned to age-match patient cohorts on both mean and standard deviation. Consistent with results from the combined healthy comparison cohort, DLPFC cohort differed significantly from their age-matched healthy controls in terms of the proportion of honest decision during conflict trials (Fisher’s exact test, p<.01, two-tailed) but not during no conflict trials (Fisher’s exact test, p>.50, two-tailed).



Distribution of honesty effect at the individual level.

To illustrate individual differences in the honesty effect, we calculated the empirical cumulative distributions (ECDF) for paired difference in amount given for each patient cohort, as well as their age-matched comparison groups. This highlights distributionaldifferences at all quantiles, and shows that the DLPFC cohort is associated with an overall shift in terms of the distribution of honesty effects.



Comparison of relative influences of honesty on altruistic giving between patient and healthy comparison cohorts.

To compare the influence of honesty on altruistic giving in lesion cohorts relative to healthy participants at a trial-by-trial level, we regressed the relative influence of honesty in DLPFC and OFC cohorts against the relative influence for the same question in healthyparticipants, respectively. The relative influence of honesty is defined as the change of altruistic giving from the Choice condition to theMessage game with identical monetary consequences, normalized by the highest possible payoff to the message sender. The x-axis represents the level of relative influence of honesty in healthy participants, and the y-axis is the relative influence in lesion cohorts. The diagonal represents an identical impact of honesty concerns in lesion and healthy comparison cohorts. If the regression slope is larger than 1, the corresponding lesion cohort exhibits increased sensitivity to honesty relative to healthy participants. On the other hand, if theslope is significantly smaller than 1, the lesion cohort shows a diminished sensitivity to honesty concerns. We found that the regression line for OFC is located around the 45-degree line (β=0.85), and cannot be rejected compared to the null hypothesis of β=1 (p>0.2), indicating OFC and healthy comparison cohorts demonstrated similar sensitivity to honesty concerns. In contrast, the regression line for DLPFC can be rejected compared to the null hypothesis of β=1 (β=0.37, p<0.001), which supports the idea of a reduced sensitivity to honesty in the DLPFC cohort. In addition, the positive correlation between DLPFC and healthy participant choices also suggested that DLPFC patients, like that of OFC and comparison cohorts, were sensitive to the degree to which their choices ‘harmed’ the otherplayer, making more self-interested choices when such choices were least damaging to the recipient. Thus it helps to argue against thepossibility that DLPFC lesion patients have difficulty in representing the recipient’s feelings which increased the overall tendency towardself-interest.



Consistency of choices across cohorts.

To test for the possibility that deficits in cognitive processes unrelated to honesty may have contributed to the behavioral differencesbetween the DLPFC patients and other cohorts, we examined whether DLPFC damage resulted in more random choice behavior, thusexerting downward bias on the effect of honesty. We matched trials between the Choice and Message conditions with identicalmonetary consequences and calculated the proportion of choices on which subjects made inconsistent choices between the twoconditions. We rejected the hypothesis that DLPFC damage was associated with more inconsistent choices between the Choice andMessage conditions compared to both OFC and healthy comparison cohorts (Fisher’s exact test, p<.01, two-tailed), arguing against the possibility that observed differences were due to random behavior on the part of the DLPFC patients. DLPFC participants weresignificantly more consistent in choices between the Message and Choice conditions than either OFC or healthy participants (Fisher’s exact test, p<.01 for each, two-tailed). That is, OFC and healthy participants differed more frequently across Choice and Messageconditions (52%±6.5% and 61%±3.9%, respectively) than DLPFC (22%±5.8%). In contrast, OFC and healthy comparison cohorts didnot differ (Fisher’s exact test, p>.50, two-tailed). Thus these results suggest that the ability to reason through possible outcomesremained intact in DLPFC lesion cohort. In part, this may be because we reduced theory-of-mind and working memory demands by presenting both players’ payoffs explicitly.



Probabilistic nature of outcomes.

The 80% probability of implementation was introduced to the Choice condition to equalize the approximate payoff considerations across the two conditions, as message receivers typically follow received messages about 80% of the time, according to both our data as wellprevious research2. A possible concern arises in that it may have had the additional consequence of giving participants some plausibledeniability, or an "out", in the Choice condition that is not present in the Message condition, which could have contributed toparticipants' greater selfishness in the Choice condition. To investigate this, we conducted an additional study on Amazon’s Mechanical Turk in which we manipulated this factor directly using a 2 (condition: Choice, Message) x 2 (probability of implementation: 80%, 100%)design. In particular, we introduced plausible deniability into the Message condition by telling participants, "When you choose a message, there is an 80% chance that it will be delivered to the other person. There is a 20% chance that the other message will be delivered" (Message-80%) versus "When you choose a message, it will be delivered to the other person" (Message-100%). Similarly, we removed plausible deniability from the Choice condition by telling participants, "When you choose an option, it will be implemented"(Choice-100%) versus "When you choose an option, there is an 80% chance that it will be implemented. There is a 20% chance that the other choice will be implemented" (Choice-80%). We ran this study in a between-subjects design (n=163) in which each participant made a single decision in one of the four conditions. Consistent with our previous results, a two-way ANOVA revealed a significant reduction in selfish choices in the Message condition relative to the Choice condition (F(1,159)=32.98,p<.0001), which was present across both levels of probability (t-tests, p < 0.006 for each). Moreover, we found that decisions were not significantly affected by theprobability of implementation (F(1,159)=1.86,p>.17), and there was no significant interaction between condition and probability (F(1,159)=0.43,p>.83). We observed significantly fewer selfish choices in the Message condition relative to the Choice condition across both the 80% (t(74)=5.31,p<.0001) and 100% (t(89)=2.86,p<.005) implementation probabilities. Thus, although we cannot rule out that lesion participants held substantially different beliefs regarding the probability of implementation, our results suggest that it is difficult for differences in beliefs to completely explain observed behavioral differences between the Choice and Message conditions. Future studies directly probing beliefs on part of lesion patients will be important to address this question as well as larger question ofcontribution of theory of mind to honesty and deception.



Comparison of honesty effect under real versus hypothetical payoffs.

To compare honesty effect under hypothetical and real payment, the latter being more common in previous studies on signaling game experiments, we collected additional data of 14 participants on 3 sets of outcomes identical to Gneezy2 using real payoffs. First, we adopted the analysis in Gneezy2, comparing proportions of selfish choices based on the same sets of outcomes across Gneezy2, study with real payoffs and the study with hypothetical payoffs. The proportions of selfish decisions are quite similar across 3 datasets (Choice condition: 66%±4% (Gneezy 2005) vs. 74%±12% (real payoffs) vs. 70%±8.82% (hypothetical payoffs), Message condition:35%±4.27% vs. 34%±12.66% vs. 21%±7.84% (Chi-square test, p>.30 for both comparisons, two-tailed). In addition, following the analysis in the present study, we compared paired difference of amount given between the Message and Choice condition with and without real payment. We found a small reduction in the honesty effect under real payoffs (2.76±.37 vs. 3.38±.62, with and without real payoffs respectively), but these were not significant (two-sample t-test, p>0.10, two-tailed). Note that we did not include Gneezy2 in this comparison, as paired differences are not possible due to its between-subject design.



Affective involvement of message and choice conditions.

The fact that we did not observe an effect of OFC damage on behavior in the present investigation is perhaps surprising given existing data on the role of OFC in prosocial behavior. One possible explanation for the OFC results in the present study is that our gameslacked a strong affective component, including the need to regulate emotional responses to self-generated or external stimuli that isoften attributed to OFC functioning. In particular, Koenigs et al.16 found that OFC patient decisions deviated from those of healthyparticipants only in moral dilemmas involving a high degree of conflict between emotional responses and utilitarian goals. To investigate this possibility, we ran an additional study in which healthy participants (n=51) recruited online from Amazon’s Mechanical Turk made a series of comparisons between the degree of affective involvement in our task versus the moral dilemmas borrowed from Koenigs et al.16. Subjects were randomly assigned to rate the emotional intensity of a trial from our Choice condition or Message condition in comparison with one of each from the following types of moral dilemmas as classified according to Koenigs et al.: (i) high-conflict between emotion and utilitarian concerns, (ii) low-conflict between emotion and utilitarian concerns, and (iii) impersonal cases, whichare associated with lower emotional involvement. To the extent that low affective involvement helps to explain the lack of difference between the OFC and healthy comparison cohorts observed in our study, it should be the case that individuals experience substantially less emotional involvement in our task than in both High- and Low- Conflict dilemmas, and that this involvement should be comparable to, or lower than, the Impersonal dilemmas. Consistent with this possibility, we found that participants rated both the Choice and Message conditions of our task as being significantly lower in emotional intensity than all three types of moral dilemmas used in Koenigs et al. (one-sample t-test p < .0001 for all comparisons, two-tailed). Similarly, past explanations of OFC patients’ lower givingrates in the standard Dictator Game have centered on the role of OFC in guilt aversion. The current design minimizes guilt byinstructing the participant that the recipient will never know the original payoff amounts (Online Methods), which perhaps accounts for the lack of effect of OFC damage. This view is consistent with previous observations that DLPFC is involved generally in cognitivecontrol of impulses, be they emotionally-derived or not. However, our hypothesis is necessarily speculative given the lack of directmanipulation of emotional intensity in the current task. Indeed, other studies have suggested that OFC patients exhibit increased, ratherthan decreased, emotional reactivity to direct personal frustration or provocation16. Future studies combining economic games with emotional manipulations are thus needed to clarify these questions. An additional possibility is that behavioral differences insteadreflect differences in lesion cohort characteristics, such as age or etiology (Online Methods Table 1). For example, it is possible that, due to the TBI nature of damage in our OFC cohort, OFC damage actually had the effect of shifting patient behavior from a premorbidabnormal state into the normal range. Future experiments with larger cohorts varying in in etiology will be needed to test these hypotheses.



Task interfaces.

(A) Message condition. (B) Choice condition.


Cohort Sample size Age Gender

(F) Years of

Education Estimated

WAIS1 Etiology Hemisphere

DLPFC 6 57 (8.37) 4 16.17

(2.86) 99

(8.50) Stroke (6) Left (5) Right (1)

OFC2 7 46.71 (16.86) 3 15.14

(2.85) 109.83 (9.26)

Traumatic brain injury3 (6)

Tumor resection (1)

Bilateral (6) Left (1)

DLPFC age-matched comparison 11 54.40

(9.94) 5 15.77 (1.05)

107.8 (15.98) - -

OFC age-matched comparison 16 44.13

(15.65) 7 15.81 (1.07)

105.5 (13.26) - -

Parentheses contain standard deviations. WAIS: Wechsler Adult Intelligence Scale. 1 WAIS scores were estimated from Shipley Institute of Living Scale. 2 WAIS in OFC lesion cohort is taken as average over 6 patients, as one did not complete the IQ test. Supplementary Table 1: Demographic information of age-matched healthy comparison cohorts for DLPFC and OFC, respectively, on mean and standard deviation.


Option A Option B Own Other Own Other 15 5 5 15 6 5 5 15 6 5 5 6 5 20 20 5 5 10 6 5

10 12 12 10 10 5 5 20 5 10 10 5 6 5 10 4.99

10 4.99 4 5 10 6 10 5 8 10 10 12

Supplementary Table 2: Full table of trial options.


Cohorts Choice1 Message1 Choice – Message2

L DLPFC (N=5) vs.

healthy comparison

! > .10

Kruskal-Wallis test

! < .005

Kruskal-Wallis test

! < .005

Wilcoxon rank sum test

L DLPFC (N=5)

vs. OFC

! > .10

Kruskal-Wallis test!! < .005

Kruskal-Wallis test!! < .005

Wilcoxon rank sum test

L DLPFC (N=5) vs.

R DLPFC (N=1)

!!"!#$%#!Wilcoxon rank sum test



1 Null hypothesis: Amount given is the same across cohorts in Choice and Message conditions. 2 Null hypothesis: Paired difference between Choice and Message conditions is the same across cohorts. All tests are two tailed.

Supplementary Table 3: DLPFC laterality effects. Previous rTMS studies have suggested that hemispheric differences in DLPFC contributions to behavior, and specifically that the right DLPFC might be particularly important for social decisions. Accordingly, we assessed both the robustness of our findings in left DLPFC patients, as well as possible hemispheric differences. We found that our results were robust to exclusion of the patient with right DLPFC damage (row 1). Additionally we did not observe a significant association between DLPFC laterality and amount given under either the Choice condition, the Message condition, or the difference between Message and Choice conditions (row 2). However, we note that given our lesion cohort composition, we lack adequate power to detect hemispheric differences. Future studies using larger cohorts will be needed to address this important issue.


Cohort Choice Condition

Message Condition

Honesty Effect

DLPFC .82 (.05)

.75 (.09)

-.07 (.08)

OFC .79 (.07)

.43 (.06)

-.36 (.10)

Healthy Comparison

.73 (.05)

.29 (.04)

-.44 (.06)

Supplementary Table 4: Effect of messages on weight placed on own payoff. In the Choice condition, all participants placed similar weight on own payoff, placing greater weight on one’s own payoff than that of the receiver. In the Message condition, OFC patients and healthy participants placed significantly greater weight on the receiver’s payoff, whereas DLFPC patients did not exhibit a significant shift. All

cohorts exhibited similar elasticity of substitution, captured by ! ! !!!! (DLPFC: !!!" ! !!!, OFC:

!!!" ! !!", and healthy comparison: !!!" ! !!"), and were not significantly different (chi-square test, ! ! !!", two-sided). Honesty effect is calculated as the paired difference between the weights in the Message and Choice conditions. Parentheses contain bootstrap standard errors.


Documents

Nature Neuroscience: doi:10.1038/nn · In addition, the positive correlation between DLPFC and healthy participant choices also suggested that DLPFC patients, like that of OFC and