Prefrontal tDCS attenuates counterfactual thinking in female individuals prone to self-critical rumination

The tendency to ruminate (i.e., repetitive negative self-referential thoughts that perpetuate depressive mood) is associated with (a) an elevated propensity to maladaptively experience counterfactual thinking (CFT) and regret, and (b) hypo-activity of the left dorsolateral prefrontal cortex (DLPFC). The goal of this study was to investigate whether anodal transcranial direct current stimulation (tDCS) over the left DLPFC, in function of self-critical rumination tendencies, momentarily reduces counterfactual thinking and regret (assessed via self-report and psychophysiological indices). Eighty healthy participants with different levels of self-critical rumination received either anodal or sham tDCS while performing a decision making task in which they were repeatedly confronted with optimal, suboptimal, and non-optimal choice outcomes. The results showed that among rumination-prone individuals, anodal (versus sham) tDCS was associated with decreased CFT and attenuated psychophysiological reactivity to the differential choice outcomes. Conversely, among low rumination-prone individuals, anodal (versus sham) tDCS was associated with increased CFT and regret, but in absence of any effects on psychophysiological reactivity. Potential working mechanisms for these differential tDCS effects are discussed. Taken together, these results provide initial converging evidence for the adaptive effects of left prefrontal tDCS on CFT and regret to personal choice outcomes among individuals prone to engage in self-critical rumination.

www.nature.com/scientificreports/ by setting the point threshold to completely fill the progress bar to an amount (2600) that was slightly above the points that would be obtained when using the optimal strategy of consistently opening half of the boxes (2500). The paradigm was programmed in MATLAB 2018b (The MathWorks Inc,), using the psychtoolbox 60 .
Transcranial direct current stimulation (tDCS). TDCS was applied with a pair of saline-soaked sponge electrodes (5 × 7 cm = 35 cm 2 ), and delivered with a battery-driven stimulator (1 × 1 tDCS mini-CT, Soterix Medical Inc.) The anodal electrode was vertically positioned over F3 (corresponding to the left DLPFC) according to the 10-20 international EEG system, whereas the cathode was placed over the contralateral supra-orbital area (Fp2). This electrode positioning is in accordance with previous tDCS studies on emotional processing and the level B recommendation for treating major depressive disorders 23,61,62 . A current of 2 mA (current density = 0.06), with 30 s of ramp up was applied for 20 min with a 30 s ramp down at the end. 50% of the participants received active anodal tDCS, whereas the others received sham tDCS (between-subject design). For sham tDCS (i.e., placebo) the current was directly ramped down after the initial ramp up phase 62 . The tDCS procedure followed a singleblind design. Figure 2 shows a visualization of the electric field simulation of the utilized tDCS montage, using Soterix HD-Explore software.
Self-report measures. Online survey. To ensure comparable anodal and sham tDCS groups, an online survey assessing potential confounders (e.g., trait regret proneness, behavioral inhibition and reward sensitivity, symptoms of mood and anxiety disorders, and the habitual use of adaptive and maladaptive emotion regulation strategies) was carried out prior to the experiment. Independent t-tests (Supplementary Table S1) showed no group differences on any of these variables (all ps > 0.17).
Self-critical rumination. The habitual tendency to engage in self-critical rumination (e.g., "I often worry about all the mistakes I have made", "Sometimes it is hard for me to shut off critical thoughts about myself ") was assessed using the Self-Critical Rumination Scale (SCRS 54 ). The scale consists of ten items, rated on a 4-point (1 = not at all, 2 = a little, 3 = moderately, 4 = very much) Likert scale. Participants were asked to indicate how well each item described them. The scale displayed excellent internal consistency (Cronbach's α = 0.92).
Counterfactual thinking and regret. Counterfactual thinking and regret were assessed at the end of every four trials (i.e., during the task), and at the end of the task, when participants knew they did not perform well enough for the monetary reward (post-task; see "Supplementary Materials"). Participants were displayed "To what extent do you think about what others choices would have led to?" and "To what extent do you regret the made choices?", and responded using a visual analog scale (VAS), ranging from 0 = not at all, to 100 = a lot.
Psychophysiological measures. Skin conductance responses (SCRs). Electrodermal activity was recorded at a 1000 Hz sample rate with the Biopac EDA100c amplifier, in conjunction with the Biopac MP150 (Biopac Systems Inc., Santa Barbara, CA). On the amplifier, the gain was set to 5 μS/V, the low pass filter set to 10 Hz, and both high pass filters set to DC mode (off). Gelled velcro finger electrodes were placed on the distal phalanges of the index and middle finger of the non-dominant (left) hand [63][64][65] . The data was collected in the Acqknowledge software on an external computer, together with event triggers that were sent by the MATLAB computer to the Biopac STP100c, via a USB interface. Using the LedaLab 3.4.9 MATLAB toolbox 66 , the data was downsampled to 50 Hz (to increase processing speed), artifact corrected using spline interpolation, smoothed using a 16 sample moving average, and filtered using a 1st order Butterworth 5 Hz lowpass filter. Continuous decomposition analysis then extracted the phasic information from the skin conductance signals, based on a SCR detection threshold of 0.01 µS 67 , and computed average SCR amplitude (expressed in µS) for every 8 s outcome reveal at the end of each trial.
Cardiovascular reactivity. Cardiovascular reactivity was recorded at a 1000 Hz sample rate with the Biopac ECG100c amplifier, in conjunction with the Biopac MP150 (Biopac Systems Inc., Santa Barbara, CA). On the amplifier, the gain was set to 5000, mode set to normal, low pass filter set to 35 Hz and the high pass filter set to  www.nature.com/scientificreports/ 0.05 Hz. Pre-gelled Ag/AgCL electrodes were placed according the measurement of a lead II ECG. The data was collected in the Acqknowledge software on an external computer, together with event triggers that were sent by the MATLAB computer to the Biopac STP100c, via a USB interface. Using the PhysioData Toolbox 0.4 68 , the signal was first filtered using a 1 Hz highpass filter and a 50 Hz lowpass filter, and R-peaks and interbeat intervals (IBIs) were subsequently automatically detected based on the following constraints: (a) minimum R-peak of 0.5 mV, (b) minimum IBI of 0.3 s, and (c) maximum IBI of 1.5 s. Based on visual signal inspection, artifacts in the R-peaks and IBIs were then removed and interpolated if possible. Next, a continuous heart rate signal was computed via shape-persevering piecewise cubic interpolation at 100 Hz of the valid IBI data. Based on this signal, HR was computed for every second during the 8 s outcome reveal, and 1 s prior to the outcome reveal (i.e., pre-outcome baseline). The change in HR (expressed in bpm) was then computed for every second during the 8 s outcome reveal, by subtracting HR during the pre-outcome baseline from the HR values at every second of the outcome reveal 1 .
Procedure. Before the lab session, on an online webpage participants read a description of the study, including the exclusion criteria and part of the cover story. To mask the true intent of the study, participants were made to believe they were participating in a study investigating the effects of tDCS on decision making behavior. In addition, they read an information form explaining (a) that tDCS induces a weak electrical current flowing from the anode to the cathode, through underlying brain regions, which thereby modulates their neural activity, and (b) the safety and potential short-term side effects (e.g., headache, dizziness, nausea) of tDCS. They were informed they would either receive active or sham tDCS during the lab session. Afterwards, they completed the online survey and based on this survey data, participants were pseudo-randomly (see "Supplementary Materials") assigned to either the anodal or sham tDCS group. At the start of the lab session, participants provided written informed consent. Participants were seated in front of a computer screen and were connected to the physiological recording equipment. Next, anodal or sham tDCS was administrated while participants were introduced to the task and subsequently performed it. At the end of the task, participants were compensated and verbally debriefed: the cover story and the nature of the regret-inducing paradigm was explained. The duration of the protocol was on average 40 min.
Data analytic plan. All data was analyzed in R 3.6.1 69 in conjunction with Rstudio 1.2.1335. Analyses were conducted by either fitting linear mixed models (LMMs) or generalized linear mixed models (GLMMs), depending on the distribution (i.e., normal or gamma) of the outcome variable (see "Supplementary Materials"). These models were fitted via the 'lmer' and 'glmer' functions of the 'lme4' and 'lmertest' R packages 70,71 . The sum of squares for the models were estimated using the type III approach, and the statistical significance level was set to p < 0.05. Continuous predictors were standardized prior to model fitting.  73 . Where applicable, p-values from follow-up tests were corrected for multiple comparisons using the false discovery rate correction 74,75 . First, to investigate the role of self-critical rumination on effects of tDCS on reported CFT and regret during the task, 2 LMMs were fitted, featuring CFT and regret every four trials during the task as dependent variables, and group (sham tDCS, anodal tDCS) as fixed factor, recent lost opportunities (i.e., lost opportunities of the last four trials) and self-critical rumination as continuous predictors, and subject as random intercept.
Second, to investigate the role of self-critical rumination on effects of tDCS on SCRs and cardiovascular reactivity to trial-based choice outcomes, a GLMM for a gamma distribution with a log-link function was fitted with the amplitude of the skin conductance response as dependent variable, whereas a LMM was fitted with the change in heart rate as dependent variable. Both these models featured group (sham tDCS, anodal tDCS), choice outcome (non-optimal, suboptimal, loss) as fixed factors, self-critical rumination as continuous predictor, and subject as random intercept. The LMM for the cardiovascular reactivity additionally featured time (i.e., elapsed seconds after choice outcome; 1, 2, 3, 4, 5, 6, 7, 8) as fixed factor. The effects where group or choice outcome (for psychophysiological analyses) were not implied, are reported in "Supplementary Materials" for brevity, as these fall outside the scope of our research aims.

Results
Participants were not able to correctly gauge their stimulation group (sham, anodal tDCS), as the proportion of incorrect guesses (63%) was higher than change level (50%), p = 0.03. Furthermore, there were no significant differences in the time of day during the experimental sessions between the the anodal and sham tDCS group, t = 0.93, p = 0.36.   www.nature.com/scientificreports/ p < 0.001. All remaining tDCS effects (i.e., group, group × self-critical rumination × recent lost opportunities) were non-significant (all ps > 0.26). This model (excluding the random subject intercept) accounted for 21% of the observed variance in counterfactual thinking during the task.

Effects of tDCS on self-reported counterfactual thinking and regret.
Regret. This LMM showed that self-critical rumination was positively associated with regret, β = 9.01, SE   Table S5) showed that heart rate changes were modulated by choice outcome (loss vs optimal and loss vs suboptimal) across all levels of self-critical rumination and in both tDCS groups (all ps < 0.001). However, at high self-critical rumination levels, in the sham group, heart rate acceleration was significantly larger during 'suboptimal' versus 'optimal' choice outcomes, b = 0.42, SE = 0.17, t = 2.46, p = 0.02, and this was not the case in the anodal tDCS group at this rumination level, b = 0.19, SE = 0.18, t = 1.05, p = 0.44. Furthermore, a group × time × self-critical rumination interaction was present, which is presented in "Supplementary Materials". All remaining tDCS effects (i.e., group, group × self-critical rumination, group × time, group × choice outcome × time, group × choice outcome × self-critical rumination × time) were non-significant (all ps > 0.34). This model (excluding the random subject intercept) accounted for 9% of the observed variance in SCRs.  www.nature.com/scientificreports/

Discussion
The goal of the present study was to investigate whether, as a function of habitual tendencies towards self-critical rumination, anodal tDCS over the left DLPFC momentarily attenuates CFT and regret, and to explore psychophysiological correlates of this hypothesized tDCS effect in response to varying choice outcomes related to goal progress (i.e., non-optimal, suboptimal, optimal). First, the finding that individuals who are prone to ruminate are also more prone to experience CFT and regret 3,4 , was replicated in the current study as evidenced by a positive association between self-critical rumination and reported CFT and regret among participants receiving sham tDCS. Furthermore, analyses showed that the physiological responses (SCR, cardiovascular reactivity) were correlated with self-reported CFT and regret, thereby establishing construct validity of the employed psychophysiological measures to capture processes related to CFT and regret. Based on the self-report assessments, among individuals inclined to engage in self-critical rumination, anodal (versus sham) left DLPFC tDCS was associated with decreased CFT, but not with decreased regret. In contrast, among individuals with low rumination inclinations, anodal (versus sham) tDCS was associated with both increased regret and CFT. Among individuals with moderate rumination inclinations, anodal (versus sham) tDCS was not associated with differences in CFT or regret. Taken together, these results provide evidence that the effect of anodal tDCS over the left DLPFC on CFT and regret depends on the habitual inclination towards self-critical rumination. Furthermore, the level of self-critical rumination also plays a crucial role when investigating the effects of tDCS on the psychophysiological measurements.
Based on the assessments of SCRs, across all levels of self-critical rumination, sham tDCS was associated with (a) larger SCRs for non-optimal versus suboptimal outcomes, and (b) larger SCRs for optimal versus nonoptimal outcomes. These findings are consistent with the notions that (a) no goal progress (i.e., non-optimal) is more emotionally arousing than making some goal progress (i.e., suboptimal), and (b) optimal goal progress is more arousing than some goal progress (i.e., suboptimal). Yet, anodal (versus sham) tDCS over the left DLPFC attenuated this SCR reactivity to the various choice outcomes (e.g., non-optimal, suboptimal, optimal), exclusively among individuals prone to self-critical rumination. Specifically, among rumination-prone individuals who received anodal tDCS, emotional arousal did not differ between suboptimal and optimal outcomes, and between suboptimal and non-optimal outcomes. This may suggest that, among self-critical rumination-inclined individuals, tDCS may have induced blunted emotional arousal to choice outcomes, as if individuals were indifferent for either relative positive or negative utility of their specific choice outcomes. Such reasoning is consistent with the above described finding that self-reported counterfactual thinking was attenuated by tDCS in this group, suggesting reduced psychological responsiveness to information of choice outcome.
Based on the assessments of cardiovascular reactivity, HR accelerations were significantly smaller during non-optimal outcomes compared to suboptimal or optimal outcomes, regardless of tDCS group or self-critical rumination tendencies. These findings are in line with past research showing attenuated HR acceleration or HR slowing to non-optimal outcomes 1,76 , negative feedback regarding motor performance 77 , and negative social feedback 49,50,78 , and this is thought to reflect a defensive preparatory mechanism in the face of threat or other aversive situations 43,45,51,79 . Furthermore, among individuals with high self-critical rumination tendencies who received sham tDCS, HR accelerations were larger during suboptimal versus to optimal outcomes. This was not the case among individuals with high self-critical rumination tendencies who received anodal tDCS, or among low or moderately predisposed individuals, irrespective of their tDCS group. Possibly, this enlargement of HR acceleration during suboptimal versus optimal outcomes among self-critical ruminators may reflect an increased tendency to cognitively elaborate 44,45 on suboptimal versus optimal outcomes. Such an explanation would be consistent with the notion that ruminators are prone to counterfactual thinking and regret in situations where a decision turns out well, but discover that another decision would have led to even better outcomes 80 , as this is the case when faced with suboptimal outcomes in the task. Interestingly, enlarged HR accelerations to emotional stimuli have been observed in several vulnerable populations, such as rumination, social anxiety and post-traumatic stress disorders [81][82][83] . Following this reasoning, the observed HR acceleration enlargement during suboptimal versus optimal outcomes may reflect a cardiovascular marker of the proneness to experience counterfactual thinking and regret in this population. Given that this differential HR reactivity to suboptimal versus optimal outcomes was not observed during anodal tDCS among ruminators, suggests that tDCS attenuated this differential HR reactivity. Taken together, again consistent with the results of the self-reported and SCR data, these findings suggest that increased cardiovascular reactivity to suboptimal versus optimal outcomes observed exclusively in rumination-prone individuals may be a cardiovascular marker of CFT and regret-proneness in this population and that tDCS over the left DLPFC can attenuate this differential reactivity.
As stated earlier, based on the self-report assessments, among individuals with low self-critical rumination tendencies anodal tDCS was associated with an increased propensity to experience counterfactual thinking and regret. However, in contrast to self-critical rumination-prone individuals, tDCS was not associated with changes in physiological reactivity to the differential choice outcomes among these individuals. Given this absence of tDCS effects on physiological reactivity in this population, it could be argued that this self-reported increased CFT and regret propensity does not necessarily reflect a maladaptive affective/cognitive outcome, as CFT and regret are generally, in healthy populations, thought to serve adaptive functions to improve future decision making 9 . Such a reasoning would also be consistent with correlational and experimental data showing that the DLPFC is involved in both counterfactual thinking 84,85 and the maintenance of goal-oriented intentions and behaviors 86 . Furthermore, the current results show that regardless of tendencies towards self-critical rumination, perceived recent lost opportunities was more strongly related to counterfactual thinking and regret in the anodal tDCS group. This suggests that anodal tDCS over the left DLPFC generally increased the sensitivity towards counterfactual information, consistent with the role of the DLPFC in the value encoding of hypothetical outcomes from specific actions 85,87 and maintenance and implementation of goal-oriented behavior 86  www.nature.com/scientificreports/ as attempting to make the best decisions based on prior actions. Altogether, anodal tDCS could have generally influenced prefrontal networks involved in the value encoding of (counter)factual outcomes 85,87,88 . At the same time, depending on altered fronto-limbic connectivity associated with ruminative tendencies, anodal tDCS could have increased connectivity within this fronto-limbic network 18 , as this network is involved in down-regulation of limbic reactivity 21,89,90 , which would be in line with the observed tDCS effects on physiological reactivity to choice outcomes among ruminative-prone participants. Such a reasoning is consistent with research suggesting that interindividual differences such as preexisting activation of specific neural networks, significantly influence the outcome and efficacy of tDCS applications in mental health outcomes 91 . On a neurocognitive process level, the rumination-associated hypo-activity is thought to involve impairments in attentional processes 17,92 , such as negative attentional biases and impaired disengagement from negative information. Furthermore, prefrontal tDCS has been shown to beneficially modulate these attentional processes, in turn attenuating emotional reactivity 26,93,94 and ruminative processes 25 . Interestingly, by using a similar task as the current paradigm, attentional deployment towards factual (i.e., points that have been achieved) versus counterfactual outcomes (i.e., points that could have been achieved) has been shown to influence CFT and regret, as measured by both self-report and functional neuro-imaging 95 . Based on this empirical data, it is possible that the effect of tDCS over the left DLPFC on CFT among high self-critical ruminators may be mediated by the modulation of these attentional processes. Given that these propositions fit well into existing theory and empirical data, the current data generates new testable hypotheses for future research potentially unraveling underlying neurocognitive mechanisms in these observed tDCS effects.
Besides several strengths, such as the ecologically valid experimental paradigm, some limitations of the current study have to be mentioned. First, because female populations tend to be more rumination-prone 55 , the sample consisted exclusively of females, but at the cost of limiting the generalizability of the results. Furthermore, past research has suggested that the stage of the menstrual cycle can influence tDCS effects 96,97 , and this was not controlled for in the current design. Given the relatively large sample size, it could be argued that these hormonal effects could be balanced out between the two groups. Second, although it could be considered as a limitation that a between-subject design was employed without measuring baseline sensitivity to the experimental paradigm, this specific design was chosen to prevent habituation and desensitization to the paradigm due to repeated exposure. Moreover, to sufficiently accommodate this limitation, the tDCS groups were matched on a wide range of potential confounding variables, thereby ensuring group comparability. Third, the experimental design would have been methodologically stronger if a double-blinded tDCS procedure 98 would have been used. Fourth, although the current sample size was significantly larger than commonly employed sample sizes in tDCS studies 29 , power analyses indicated that there was only adequate power to detect moderate to large effect sizes, and additional post-hoc power analyses reported in the "Supplementary Materials" furthermore corroborate this notion. This warrants the need to employ even larger sample sizes in future tDCS research, as tDCS effects tend to be weak and between-subject designs also inherently involve more variance compared to within-subject designs 29,97 . Finally, although the employed tDCS protocol was motivated by both previous tDCS research on rumination 25,99 and a priori electric field simulations, some critical reflections have to be made regarding (a) the design of the tDCS groups, (b) the electrode size, and (c) the position of the cathode electrode. First, the absence of an additional anodal tDCS group in which a brain region is targeted that is not implicated in ruminative processes, hinders clear conclusions regarding the specificity of the tDCS montage on the observed rumination-dependent tDCS effects. For instance, based on the involvement of the orbitofrontal cortex (OFC) in decision outcome valuation, in a previous study it was shown that cathodal tDCS over the OFC was associated with less intense self-reported emotions to factual and counterfactual outcomes in a gambling task 100 . In this former study, the cathode was placed over the left OFC, while the anode was placed over the right lateral occipital cortex. Second, due to the relatively large electrode size in the current study, the focality of the tDCS is reduced and the contribution of brain regions neighboring the DLPFC in the observed effects is highly likely [101][102][103] . Nevertheless, in view of an ongoing paradigm shift towards brain network perspectives in neuroscience 104,105 and the importance of brain networks in affective disorders 106 , a (clinical) advantage of tDCS is its ability to modulate large-scale brain networks [107][108][109] , rather than brain regions in isolation, but at the cost of reduced focality. Third, in the current study the cathode was placed over the right supra-orbital area, and the current distribution of tDCS is suggested to be influenced by both the position of the anode and cathode electrode 103 . Consequently, the possibility that the observed effects were driven rather by the cathode than the anode cannot be excluded and this could potentially have been mitigated by extracephalic placement of the cathode. However, computational modelling studies suggest that extracephalic cathode montages result in a more widespread current distribution 110,111 , thereby reducing focality. Taken together, as a variety of parameters define the current distribution of tDCS 30,103,112 , in order to clearly elucidate the neural circuitry involved in tDCS effects, future tDCS studies would greatly benefit from the concurrent use of neuro-imaging methods such as functional magnetic resonance imaging or high density electroencephalogram, and/or the application of HD-tDCS, which features increased focality 101,113,114 .
Taken together, these findings show that in a context of self-relevant decision making, left prefrontal tDCS momentarily attenuates (1) self-reported counterfactual thinking (but not regret) and (2) differential psychophysiological reactivity to choice outcomes differing in their utility towards goal progress, specifically among individuals prone to self-critical rumination. These findings extend existing tDCS literature showing adaptive effects on emotional reactivity to various (non-self-referent) emotional stimuli 35,94,115 , by employing (self-referent) personal choice outcomes as the presented stimuli. Moreover, based on the notion that excessive counterfactual thinking may fuel regretful self-critical ruminative cognitions, thereby increasing the risk for the development of depression 3,9 , these initial results may suggest that tDCS could be of therapeutic use in attenuating counterfactual thinking, specifically among individuals who are prone to repetitively and self-critically engage with these thoughts. However, further research is warranted to systematically investigate the potential prospective effects of tDCS on the temporal dynamics between counterfactual thinking, self-critical rumination, and depressive www.nature.com/scientificreports/ symptomatology. Nevertheless, the current results are clinically relevant because feelings of contentment regarding one's personal life decisions are important in mental health 1,2 , and individuals increasingly experience distress surrounding personal decision making (outcomes) in modern western societies 116 , as these societies are characterized by a large number of behavioral possibilities and typically promote individualistic self-deterministic values, associated with a high pressure on individuals to achieve goals and success 15  www.nature.com/scientificreports/