The ability to flexibly update and adapt our behavior plays a role in anxiety and fear-based disorders. For example, individuals suffering from anxiety and fear-based disorders repeatedly (and often erroneously) respond to feared stimuli as “dangerous.” This anxiety is frequently associated with avoidance of the anxiety-inducing situation and/or performance of safety behaviors (e.g., hand-washing compulsions in case of obsessive-compulsive disorder) to prevent or reduce anxiety. However, such avoidance and safety behaviors are ultimately counterproductive because they sustain the anxiety and prevent the individual from responding flexibly and learning that the feared consequences do not occur (Foa and Kozak, 1986; Salkovskis et al., 1996). Exposure-based psychotherapies aim to tackle this behavioral inflexibility by systematically guiding patients to approach anxiety-provoking situations they would rather avoid. In doing so, exposure therapy provides a corrective learning experience as patients learn that these situations are not as dangerous or threatening as originally thought (Mowrer, 1960; Foa and Kozak, 1986; Beck et al., 2005). This parallels the extinction of fear, and results in the acquisition of new, more adaptive associations and behavioral responses that inhibit excessive or irrational fear (Foa and Kozak, 1986; Harris et al., 2000; Bouton, 2004; Bouton et al., 2006; McConnell and Miller, 2014).

Despite strong experimental support, exposure-based psychotherapy often results in less-than-optimal clinical outcomes; drop-out is high, and many patients experience residual symptoms and a return of symptoms in the months or years following treatment (Heimberg et al., 1998; Craske, 1999; Otto et al., 2000; Dudai, 2006; van Minnen and Foa, 2006; Kindt et al., 2009; Taylor et al., 2012; Rothbaum et al., 2014). Whereas the persistence of symptoms suggests difficulties of the generalization of newly learned associations beyond the treatment setting, the return of fear indicates a failure of extinction-based inhibition over time. One reason why generalization may be limited and inhibition may fail is that unlike initial fear acquisition, which generalizes easily across contexts, subsequent extinction is contextually bound and does not readily generalize to other contexts (Bouton, 1993, 2002, 2004; Bouton et al., 2011; Todd et al., 2012). In other words, a stimulus-response relationship that has been extinguished is considered the “exception to the rule.” This is highlighted by observations in rodents that extinction occurs faster when the extinction context during extinction differs from the context in which initial fear learning took place. Here the change in context signals that something has changed. Yet, a return to the initial (fear) context results in a return of fear, indicating renewal (Bouton and King, 1983; Vansteenwegen et al., 2005), and if a subsequently encountered context is more akin to the initial acquisition context, renewal is enhanced (Todd et al., 2012).

A similar phenomenon has been observed in reversal learning. Like extinction, reversal learning relies on updating contingencies through associative learning. Namely, in standard reversal learning tasks, one of two stimuli is initially associated with a reward after which learned associations are reversed (Cools et al., 2002; Fellows and Farah, 2003; Remijnse et al., 2005). When this reversal takes place in a context that differs from that in which initial learning occurred, reversal learning happens faster (compared to when initial and reversal learning take place in the same context) (Thomas et al., 1985; McDonald et al., 2001). Facilitating the relevant cognitive mechanisms underlying generalization of learning as well as mechanisms to prevent a return of fear might ultimately improve real-world efficacy of exposure therapy and reduce relapse (Boschen et al., 2009).

The prefrontal cortex, and in particular the ventrolateral, dorsolateral, (ventro)medial prefrontal and orbitofrontal cortex, are implicated in stimulus-outcome associative learning and play a critical role in behavioral updating and responding to surprising events when contingencies change (Fletcher et al., 2001; Cools et al., 2002; Clark et al., 2004; Boettiger and Esposito, 2005; Ghahremani et al., 2009; Mitchell et al., 2009; Greening et al., 2011; Levy-Gigi et al., 2011; Izquierdo et al., 2017). For this reason, neuromodulation of the prefrontal cortex might be a promising target to impact the updating of behavior when contingencies change. For example, transcranial direct current stimulation (tDCS) – a type of non-invasive neuromodulation that alters intrinsic neuronal activity using a relatively weak but constant electrical current (Nitsche et al., 2008) – has been found to facilitate implicit learning using a probabilistic classification learning task (Kincses et al., 2004). Additionally, tDCS aimed to inhibit ventrolateral prefrontal cortex functioning was associated with more errors during reversal learning as expected (Albein-Urios et al., 2019) whereas tDCS aimed to excite the medial prefrontal cortex led to increased choice shifting following punishments during reversal learning (Panitz et al., 2022). However, in these prior experiments, context was not manipulated and, as we highlighted above, context affects flexible updating and renewal of previously learned associations, which is relevant when the goal of such research is to contribute to the development of novel treatments for fear-based disorders. To the best of our knowledge, there are no studies examining how tDCS affects updating of contingency-based associations while context is manipulated. Moreover, it remains unknown whether effects of tDCS on updating contingency-based associations also impacts subsequent generalization of learned associations to a novel context.

Therefore, we tested whether inhibition of the left prefrontal cortex using tDCS impacted updating of avoidance-based associative learning when contingencies reversed in both a different context and in the same context as initial learning. Additionally, we examined whether tDCS during contingency reversal affected subsequent generalization of learned associations to a novel context. Given the role of context and the prefrontal cortex discussed above, we predicted (a) that reversal of associations would be easier (i.e., less errors) if the context differed from initial learning than if it remained the same, and (b) that cathodal tDCS would impair the updating of associative learning during reversal. As to generalization of learning, we predicted that last learned rules (during reversal) would generalize to a novel context because of recency effects when reversal occurs in the same context as initial learning. However, when reversal occurs in a different context from initial learning, we hypothesized participants would learn with 50% change that a stimulus can be bad or not, and thus no clear preferences will generalize to a novel context. We remained agnostic as to whether tDCS will affect this generalization. We used tDCS as a temporary non-focal lesion technique, based on the idea that tDCS modulates Hebbian processes (Kronberg et al., 2020) relevant for associative learning-related plasticity.

This study aimed to test whether cathodal tDCS targeting the left prefrontal cortex, including the ventrolateral and dorsolateral prefrontal cortex, could affect the updating of previously learned avoidance-based associations when this updating occurred in the same or different context from initial learning. This was based on prior literature highlighting the relevance of the ventrolateral and dorsolateral prefrontal cortex for the updating of previous associations (Fletcher et al., 2001; Cools et al., 2002; Clark et al., 2004; Boettiger and Esposito, 2005; Ghahremani et al., 2009; Mitchell et al., 2009; Greening et al., 2011; Levy-Gigi et al., 2011; Izquierdo et al., 2017). Additionally, we tested whether tDCS would impact the subsequent generalization of learned rules to a novel context. This was tested in two separate experiments.

Results from Experiment 2 suggest that cathodal tDCS over EEG coordinate F3 impaired the updating of avoidance-based associative learning irrespective of context. The impeding effect of tDCS on reversal was also observed when the statistical model included the learning rate (i.e., Trial Number) as revealed by a significant three-way interaction with context. Specifically, reversal learning rate was lowest when participants received active tDCS and reversal occurred in a different context, whereas learning rate was highest when participants received sham during this condition (i.e., tDCS + reversal in different context < tDCS + reversal in same context < sham + reversal in same context < sham + reversal in different context). Although the main effect of tDCS was not significant in Experiment 1, perhaps due to low power, when entering learning rate to the model, we observed a similar pattern and three-way interaction as in Experiment 2. Again, indicating that participants who received active tDCS had the lowest learning rate when reversal took place in a different context from initial learning, and the highest learning rate was observed when participants received sham and reversal took place in a different context (with tDCS + reversal in same context and sham + reversal in same context falling in between). This replication of findings in two separate samples, one mostly including individuals from the general population (Experiment 1) and one sample consisting mostly of undergraduate university students (Experiment 2), reinforces the conclusion that tDCS targeting the prefrontal cortex can impair updating of previously learned avoidance-based associations. Our findings are broadly consistent with a prior study in which high definition, cathodal tDCS targeting the ventrolateral prefrontal cortex, but not sham or stimulation of the dorsomedial prefrontal cortex, resulted in higher perseverative errors during probabilistic reversal learning (Albein-Urios et al., 2019).

The interaction between tDCS and context is noteworthy. Characteristically, both in humans and rodents, changes in contextual cues from initial to reversal learning increases the reversal learning rate (Bouton, 1993; McDonald et al., 2001). This is because context functions to retrieve the initially learned stimulus-outcome contingencies hindering updating of the subsequently reversed stimulus-outcome contingencies. Thus, when the context during reversal differs from initial learning, this interference is reduced, facilitating reversal learning. This led to our prediction that reversal would be more difficult when the context did not change, and which might be more prone to interference from tDCS. Although this is not entirely what we observed, i.e., we did not see a significant main effect of context during reversal, the most plausible explanation for the observed three-way tDCS, context, and learning rate interactions is grounded in the observation that the prefrontal cortex, including the dorsolateral and ventromedial prefrontal cortex and their connectivity, are important for context-dependent valuation to guide choices (Rudorf and Hare, 2014). Namely, disruption of the prefrontal cortex due to cathodal tDCS might have specifically impaired the processing of context during reversal updating.

An alternative possible explanation for the effect of tDCS is that tDCS generally disrupted executive functioning, including working memory and/or attention. This is consistent with multiple prior reports of tDCS targeting the left dorsolateral prefrontal cortex yielding altered working memory performance (Fregni et al., 2005; Ohn et al., 2008; Hoy et al., 2013). However, if tDCS impaired executive functioning, we would expect tDCS to mainly impair reversal when it occurred in the same context as initial learning given the greater interference of prior learned associations and thus greater burden on executive functioning, which we did not observe. Likewise, if the negative effect of tDCS was due to distraction (i.e., the physical sensation of tDCS), we would expect to observe an overall effect of tDCS even when learning rate was taken into account.

A second focus of this study was to test whether the manipulation of context and application of tDCS during reversal subsequently influenced generalization of stimuli preferences in a novel (never-before encountered) context. We predicted that last learned associations, i.e., associations learned during reversal, would generalize to a novel context due to recency, and that this would be more pronounced after reversal occurred in the same context as initial learning, compared to when reversal occurred in a different context. We reasoned that the change in context allowed participants to learn, with equal likelihood, that either image would result in a loss (initial learning) in one context and no loss (reversal) in another context. On the other hand, reversal in the same context may “overwrite” learned associations during reversal promoting generalization of last learned rules. Our data appears to partially support this prediction. That is, last learned associations where the stimulus that resulted in a loss during initial learning, but then reversed to no longer result in a loss, was preferred in the novel context. Yet, this was only true after reversal had occurred in a different context. This was true irrespective of tDCS in Experiment 2, but only true for participants who received active tDCS in Experiment 1. These results suggest that, despite tDCS impairing the updating of context-dependent stimulus-outcome associations, the effect of tDCS on the generalization of rules as detected by preferences in a novel context is minimal.

Limitations of the current work include the relatively small sample sizes, especially in Experiment 1. In Experiment 1, our data indicated that participants made significantly fewer errors in Context 2 compared to Context 1 across initial learning trials, and which may reflect differences in perceptual quality between these two contexts. However, this difference between the contexts during initial learning was not replicated in Experiment 2, which included a larger sample. Other limitations are that our task only involved one reversal as opposed to reversing back and forth between primary and secondary learned associations which might have allowed examining the effects of tDCS on renewal, and the administration of a fixed number of trials for each learning phase. Although this was done to allow the duration of the tDCS to match the duration of the task administration, individual variability in reaching the learning criterion might have led to over practice in some individuals and could have influenced subsequent task phases. Finally, we applied a constant electrical current (2 mA) which likely resulted in variable electrical field values in participants (Evans et al., 2020). It is possible that individual electrical dosimetry might yield different outcomes and future studies are needed to address this general issue concerning tDCS.

Nevertheless, our results support a causal role for the left prefrontal cortex in the updating of avoidance-based associations, a neurocognitive process that lies at the heart of exposure-based psychotherapy for anxiety and fear-based disorders. These findings encourage further exploring the effects of non-invasive brain stimulation on flexible updating of learned associations to ultimately augment exposure-based treatments for anxiety and fear-based disorders.

Data will be made available upon request following VA regulations.

The studies involving human participants were reviewed and approved by the VAPHS IRB (Experiment 1) and Brown University IRB (Experiment 2). The patients/participants provided their written informed consent to participate in this study.

MW-F designed the experiment, collected and analyzed the data, and wrote the manuscript. SG designed the experiment and wrote the manuscript. CF collected and processed the data and wrote the manuscript. AA and NP interpreted the data and wrote the manuscript. JC processed the data and wrote the manuscript. RB designed the experiment, interpreted the data, and wrote the manuscript. All authors contributed to the article and approved the submitted version.