Non-invasive vmPFC-tDCS modulates the rational decision-making
We applied a within-subjects design so that each participant received excitatory and inhibitory stimulation of the vmPFC on two different days with at least 48 hours in between (Methods Fig. 1). The tDCS was optimized for targeting the vmPFC and minimizing the impact on other brain regions (Methods Fig. 2). In both excitatory and inhibitory conditions participants were stimulated with a current strength of 1.5mA over 10 minutes and were unable to differentiate between stimulation conditions (see SM2.1). Immediately after stimulation, participants performed the gambling task, where gambling behavior and event-related magnetic fields were measured. At the beginning of each trial, participants received an initial amount (‘gambling stake’), which was varied (25ct, 50ct, 75ct, 100ct) to enhance the credibility of the gambling paradigm (see Fig. SM2). Participants were asked to either keep a smaller amount or to gamble for the full amount. The 'keep' option was either framed as a gain (Fig. 1A; gain-frame: receive a smaller but safe amount) or framed as a loss (Fig. 1A; loss-frame: subtraction of a smaller safe amount). Although the final monetary amounts were equivalent in both frames, participants chose the ‘gamble’ option in the loss-frame and accepted the safe smaller amount in the gain-frame more often, replicating the framing-effect10 (Fig. 1B; z = 2.64 , p = 0.008). Next, we specifically tested the hypothesis that noninvasive vmPFC stimulation modulates the framing-effect as index of rational decision-making. Indeed, a logistic regression with the predictors of stimulation (excitatory, inhibitory) and frame (gain-frame, loss-frame) revealed a significant interaction (z = 2.19, p = 0.029) modulating the decision (keep, gamble). Importantly and supporting our hypothesis, excitatory vmPFC stimulation resulted in a significantly reduced framing-effect compared to inhibitory stimulation (Fig. 1B; χ² = 14.74, p < 0.001). Furthermore, excitatory stimulation significantly reduced the proportion of gambling choices in the loss-frame (χ² = 12.24, p < 0.001) but not in the gain-frame (χ² = 0.086, p = 0.769) supporting the hypothesis that reduced loss-aversion after excitatory stimulation underlies the attenuated framing-effect.
To uncover underlying neural correlates of the framing-effect and its interaction with stimulation, we computed a 2x2x2 repeated-measures ANOVA on source-localized MEG data (the dependent variable; see SM1.4) with the factors stimulation (excitatory, inhibitory), frame (gain-frame, loss-frame) and decision (keep, gamble). The interaction of stimulation by frame was significant at the junction of right anterior temporal and right orbitofrontal regions in a mid-latency time interval between 220 and 270ms (p-cluster = 0.023; Fig. 1C). Post-hoc t-tests of this interaction revealed less activation in the gain-frame after excitatory compared to inhibitory stimulation (t = 1.77, p = 0.044), while activation in the loss-frame showed the opposite trend-significant effect (t = -1.65, p = 0.055). The relatively greater neural response to the ‘loss-frame’ after excitatory tDCS suggests an improved inhibition of loss-aversion, most probably leading to a reduced salience of losses and eventually resulting in an attenuated framing-effect as seen in the behavioral stimulation effects. The location of the cluster nicely converges with fMRI findings of Martino and coworkers8, who showed that, in addition to vmPFC areas, enhanced activity at right orbitofrontal cortex regions (OFC) was also associated with more rational decision-making as reflected by reduced framing-effects.
Next, we aimed to investigate whether the more rational decision-making following excitatory versus inhibitory vmPFC stimulation as shown above in the ‘keep’ option generalized to decision-making depending on the odds (i.e. when choosing the ‘gamble’ option). Here, the risk-to-lose or the chance-to-win respectively was varied in steps of 20%, 40%, 60% and 80%, indicated by the size relation of the inner blue and yellow circles in both frames (Fig. 2A). Of course, participants increasingly avoided gambling with increasing risk-to-lose (z = -9.60, p < 0.001). But importantly, the effect of risk-to-lose on gambling behavior was significantly modulated by stimulation (risk-to-lose by stimulation: z = 6.60, p < 0.001; Fig. 2B). Post-hoc tests revealed more risky-choices in the two low-risk conditions (20%: χ² = 35.81, p < 0.001 and 40%: χ² = 4.81, p = 0.028) but fewer risky-choices in the two high-risk conditions (60%: χ² = 37.32, p < 0.001 and 80%: χ² = 40.97, p < 0.001) after excitatory compared to inhibitory stimulation. A representation of the averaged winnings achieved in the game across all choices (‘keep’ or ‘gamble’; Fig. 2C) elucidates the consequences of the participant’s behavior. A constant choice of the ‘keep’ option would have resulted in averaged winnings of 25 cents across all risk conditions (Fig. 2C dotted line). A constant choice of the ‘gamble’ option would have resulted in averaged winnings of 50 cents at 20% risk, 37.5 cents at 40% risk, 25 cents at 60% risk and 12.5 cents at 80% risk (Fig. 2C dashed lines). Thus, participants behaved quite rational as they almost reached the maximal win for all risk conditions. However and importantly, a t-test across all trials (t(11519) = 2.23; p = 0.026) revealed that after excitatory vs. inhibitory stimulation averaged wins were significantly higher. Post-hoc t-tests showed significantly higher winnings in the lowest 20% (t = 1.74, p = 0.041) and predominately in the highest 80% (t = 2.95, p = 0.002) risk conditions after excitatory compared to inhibitory stimulation. Averaged winnings did not differ in the 40% condition (t = 0.62, p = 0.267) and could not differ in the 60% risk-to-lose conditions because both ‘keep’ and ‘gamble’ choices resulted in an identical amount of 25ct.
On the neural level, we performed a 2x4 repeated-measures ANOVA with the factors stimulation (excitatory, inhibitory) and risk-to-lose (20%, 40%, 60%, 80%) on source-localized MEG data (dependent variable). This analysis revealed an interaction of stimulation by risk-to-lose between 310 and 460ms (Fig. 2D) at left anterior temporal and left orbitofrontal regions (p-cluster = 0.025) and thus later but laterally symmetric to the above reported interaction of stimulation by frame (Fig. 1C). Post-hoc tests of neural activity within this cluster revealed that this interaction was driven by the lowest and highest risk conditions (20%: t = -1.56, p = 0.065; 80%: t = 1.85, p = 0.038), while the medium risks, did not show significant effects of stimulation (40%: t = -0.82, p = 0.209; 60%: t = 0.23, p = 0.409). Convergent with the behavioral effects, this neural pattern suggests that excitatory compared to inhibitory stimulation facilitates rational decision-making towards maximized winnings.
Taken together, the behavioral data consistently indicates a modulation of decision-making towards increased rationality after excitatory compared to inhibitory vmPFC-stimulation. The reduced framing-effect and greater ability to estimate risks is mirrored by the neural data providing relatively enhanced inhibition of loss-aversion and high-risk options after excitatory versus inhibitory stimulation.
Non-invasive vmPFC-tDCS modulates loss-aversion in feedback-processing
After the participant's choice, the feedback on win or loss was indicated by green and red circles with the amounts of the wins or losses in the middle, respectively (Fig. 3A). Finally, participants rated their subjective hedonic valence and emotional arousal in response to each outcome. Having established that neurostimulation significantly modulated the rationality of decision-making, here we aimed to determine the effects of tDCS on rational feedback-processing in particular on its modulation of loss-aversion. We addressed this question by computing a mixed effects linear regression with the predictors stimulation (excitatory, inhibitory), outcome (gain, loss) and decision (keep, gamble). A main effect of outcome (t = -25.08, p < 0.001; Fig. 3A) reflected the trivia that gains were rated more positive than losses. Importantly, feedback evaluations were overall (across keep and gamble decisions; Fig. 3A&B) rated more positive after excitatory than inhibitory stimulation (t = 2.99, p = 0.003). A main effect of decision (t = -5.16, p < 0.001; keep > gamble) was mainly driven by a less negative evaluation of losses in the ‘keep’ condition after excitatory stimulation which will be further discussed below. While stimulation did not affect the factors decision and outcome alone (stimulation by decision: t = -0.17, p = 0.867; stimulation by outcome: t = 1.65, p = 0.245), the three-way interaction was significant (stimulation by decision by outcome: t = 5.33, p < 0.001). Post-hoc repeated-measures ANOVAs that were conducted separately for the 'keep' and 'gamble' conditions revealed a significant interaction of stimulation by outcome in the ‘keep’ condition (F(1, 35) = 11.48, p = 0.001; Fig. 3A), while the respective interaction was insignificant in the 'gamble' condition (F(1, 35) = 1.79, p = 0.190; Fig. 3B) indicating that the three-way interaction was mainly driven by the 'keep' condition. To further elucidate the influence of outcome in the three-way interaction and the effect of stimulation on rational decision-making, we calculated the difference of gain-ratings minus loss-ratings after ‘keep’ and ‘gamble’ decisions and calculated t-tests comparing excitatory and inhibitory stimulation. With respect to modulations of the framing-effect, the difference of gain-ratings minus loss-ratings in the ‘keep’ condition (Fig. 3A) was of major interest since both options held the same monetary value while monetary outcomes were very different in the ‘gamble’ condition (Fig. 3B). As predicted, the ‘framing-difference’ was smaller following excitatory compared to inhibitory tDCS (t(35) = -3.14, p = 0.003), reflecting less loss-averse feedback-processing, while the corresponding t-test in the ‘gamble’ condition was insignificant (t(35) = 1.64, p = 0.111).
To investigate the neural responses to the feedback stimuli, we performed a 2x2x2 ANOVA employing the factors stimulation (excitatory, inhibitory), decision (keep, gamble) and outcome (gain, loss). In a repeated-measures ANOVA, this did not reveal a significant main effect of stimulation nor significant interactions with the factor stimulation. This was most likely due to the extremely strong main effect of outcome (loss >> gain) - reflecting the strong loss-aversion since losses are usually rated twice as salient as gains - that explained a large part of variance (see SM2.4.3). Nevertheless, we tested our specific hypothesis regarding the modulation of framing-effects via tDCS by subtracting gains from losses in the ‘keep’ option (i.e. the relevant comparison to evaluate the effect of stimulation on framing) and used the resulting difference for a t-test (excitatory versus. inhibitory). This revealed a significant cluster (p-cluster = 0.045) in the dorsomedial prefrontal cortex in a late time interval between 470 and 510ms (Fig. 3C). Post-hoc tests of neural activity within this cluster revealed greater activations in response to losses after excitatory compared to inhibitory stimulation (t = 2.79, p = 0.010) while stimulation did not modulate gain trials (t = -0.79, p = 0.437). The increased responses to losses in the ‘keep’ condition following excitatory stimulation suggest facilitated inhibition of loss-aversion leading to less negative evaluations of losses (Fig. 3A) and eventually more rational evaluation of feedback stimuli.
In summary, convergent to effects in the decision-making phase, the analysis of feedback-processing also revealed that excitatory compared to inhibitory stimulation reduced behavioral loss-aversion and enhanced inhibition of loss-aversion in neural measures.