Effects of Transcranial Direct Current Stimulation on Reward Learning in Subclinical Depression.

The development of non-invasive brain stimulation techniques offers new approaches for the treatment of depression. TMS is an approved treatment for treatment-resistant depression in the UK and USA. tDCS is currently under investigation as a potentially cheaper, safer and more accessible alternative. A recent meta-analysis suggests that it has a moderate antidepressant effect (Razza et al., 2020).

One potential way of improving the antidepressant efficacy of tDCS might be to combine it with a reward learning task. There is evidence from rodent and human studies that tDCS enhances activity-dependent synaptic plasticity and behavioural learning and retention (Fritsch et al., 2010; O'Shea et al., 2017; Reis et al., 2009).

In depression, people prioritize the processing of negative information at the cost of positive information and this negative bias is theorized to play a major role in the maintenance of depressive symptoms (Kube, Schwarting, Rozenkrantz, Glombiewski, \& Rief, 2020). Therefore, by having depressed participants learn from reward, and increase their learning/retention via tDCS, this could potentially counteract negative bias and thus increase the antidepressant potential of tDCS.

In an earlier proof-of-concept study in young healthy volunteers (without low mood), we showed that tDCS applied during, but not before, an information-bias learning task increased reward learning rates. Here, we will test for the same effect in young healthy volunteers with subclinical depression. The study will involve online screening, a face-to-face screening session and two tDCS + learning sessions. Task performance will be fitted with a computational reinforcement learning model. Analyses will include both computational and non-computational measures of task behaviour.

Primary hypotheses:

1. Bifrontal compared to sham tDCS applied during the information-bias learning task will increase the learning rate from positive outcomes.
2. Bifrontal compared to sham tDCS applied during the task will increase the percentage of win-driven choices.
3. The effects of bifrontal tDCS will be cognitive state dependent, i.e. the predicted effects will be specific to stimulation applied during but not prior to learning.

Secondary hypothesis:

A growing body of evidence suggests that depression and anxiety are characterised by a reduced ability to adjust learning rates to volatility of action-outcome contingencies (Browning, Behrens, Jocham, O'Reilly, \& Bishop, 2015; Gagne, Zika, Dayan, \& Bishop, 2020; Pulcu \& Browning, 2017). Therefore, we will also investigate whether bifrontal tDCS might increase participants' ability to adjust their learning rates to the volatility context. This is exploratory, since there is no current evidence that tDCS can improve this ability.

Sample size: Our primary goal is to test hypothesis 1 that bifrontal compared to sham tDCS applied during the information-bias learning task will increase the learning rate from positive outcomes. In our previous study in healthy volunteers we observed this effect, which was most pronounced in the losses-volatile condition (paired t-test contrasting bifrontal and sham tDCS in the losses volatile blocks: t(19) = 2.88, p = .009), with an effect size of Cohen's d = 0.522. To detect an effect of this size with a power of 80% and an alpha level of .05 requires 31 participants. To account for a potential overestimation of the effect size we will recruit 40 participants, which should yield a power level of .89 for the effect size observed in our previous study. To keep sample sizes equal across conditions, we will recruit another 40 participants for the offline condition.

Screening: Participants will be asked to fill out a pre-screening online depression questionnaire (Beck Depression Inventory - II (BDI)) and a safety screening form to identify potential contraindications to tDCS. Participants with a BDI score of ≥10 will be invited to a 1-hour screening session. The structured clinical interview (SCID) will be administered and the researcher will interview the participant about their tDCS safety screening questionnaire. If there are no contraindications to tDCS and no indications of current or history of bipolar disorder, the researcher will schedule two tDCS testing sessions.

Experimental task: Participants will perform the Information Bias Learning Task developed by (Pulcu \& Browning, 2017). The same task protocol as in our previous study will be used (Overman, Sarrazin, Browning, \& O'Shea, 2021). In short, participants will be asked to press a button to choose between two shapes on each trial. After they choose, the word "win" and "loss" will appear on the screen, each associated with one of the two shapes. The wins and losses are independent of each other, i.e. both can also appear with the same shape. If the chosen shape is associated with a win this leads to a 10 pence financial gain; a loss is minus 10 pence. The cumulative total is shown continuously in the bottom centre of the screen throughout the task. Participants start with a total of £1.50. This task is designed to assess the extent to which participants choices are relatively influenced by win vs. loss outcomes. Participants will perform 6 task blocks of 80 trials. In the first and sixth block ("both volatile blocks"), the wins and losses each have a 75% probability of being associated with one of the shapes. The shape the wins or losses are associated with changes over time. In the second to fifth task blocks, one outcome is associated with one shape in 75% of the trials. The shape the wins or losses are associated with changes over time. The other outcome is associated with both shapes in 50% of the trials. Each participant performs two "wins-volatile" and two "losses-volatile" blocks in alternating order. Half of the participants perform a "win-volatile" block first, and the other half a "loss-volatile" block first.

Testing sessions: Participants will be asked to fill out mood and anxiety questionnaires. At the beginning of the first session the Trait subscale of the State Trait Anxiety Inventory (STAI)(Spielberger, 1983) will be completed. In both sessions, before and after performing the task participants will complete the State subscale of the STAI and the Positive and Negative Affect Scales (PANAS)(Watson, Clark, \& Tellegen, 1988)). The researcher will explain the computerised task, and the participant will get the opportunity to practise the task. The researcher will then set up the tDCS equipment with the anode over the left dorsolateral prefrontal cortex (DLPFC) and the cathode over the right DLPFC, in the F3 and F4 EEG electrode positions. The researcher will briefly test the tDCS to ensure that the participant is comfortable with the stimulation. The participant will perform one ("both volatile") baseline block of the task without tDCS. In the 'online' condition, tDCS will be applied during the task - for a duration of 20 minutes at 2 mA during the second and third task blocks. After the stimulation ends, participants will perform the remaining task blocks 4 and 5 (to test for persistence of any effects post-tDCS), followed by a repeat of the "both volatile" block performed at baseline. In the 'offline' condition, all procedures will be the same except that stimulation will be applied after the baseline ("both volatile") block and before the remaining blocks, while participants sit at rest.

Statistical analysis:

To test hypothesis 1, that bifrontal compared to sham tDCS applied during the task will increase the learning rate from win outcomes, we will run a repeated measures analysis of variance (ANOVA) on the learning rates in the "wins-volatile" and "losses-volatile" blocks, including the within-subject factors tDCS Condition (bifrontal vs. sham), Valence (win vs. loss outcomes), Volatility (wins-volatile vs. losses-volatile blocks) and Time (first vs. second half of the four task blocks). Baseline learning rates for wins and losses in the first task block of the first session will be included as covariates.

First, we will test for an interaction effect between tDCS Condition and Valence. To test our key prediction, we will run a priori planned contrasts of real vs. sham tDCS on win learning rates. Since in our previous study, the tDCS-induced increase in win learning rates was most pronounced in the losses-volatile condition, we will contrast the real-sham win learning rate separately for the losses-volatile and wins-volatile conditions.

To test hypothesis 2 that bifrontal compared to sham tDCS will increase the proportion of win-driven choices, we will run an ANOVA on the proportion of win-driven choices including the factors tDCS Condition, Volatility and Time. We will test for a main effect of tDCS Condition, as well as for an interaction effect between tDCS Condition and Volatility. The interaction effect will be followed up by planned contrasts of bifrontal versus sham tDCS, separately for the wins-volatile condition and the losses-volatile conditions.

To compare the effect of online versus offline bifrontal tDCS on the win learning rate and the proportion of win-driven choices (hypothesis 3), we will run ANOVAs on the combined dataset (online and offline condition), adding the factor "Stimulation Time" (before vs. during task performance). For learning rates, we will test for a main effect of Stimulation Time and/or an interaction between Stimulation Time and tDCS Condition and/or Valence. Follow-up ANOVAs will separate the data by Stimulation Time and analyse each dataset as described for Hypothesis 1 above, to test for an effect of tDCS on win learning rates. The effect in each condition (whether significant or not) will be quantified and contrasted across the two Stimulation Time conditions, to test the hypothesis that online but not offline tDCS will increase win learning rates. The same analysis approach will be used for the proportion of win-driven choices.

Regarding our secondary hypothesis that online bifrontal compared to sham tDCS might improve participants' ability to adjust their learning rate to the volatility context, we will conduct exploratory analyses, using ANOVA to test the effect of tDCS on the difference in win and loss learning rates between volatile and stable blocks (i.e. dependent variables will be: Win delta LR (win learning rate in wins-volatile blocks minus win learning rate in losses-volatile blocks) and Loss delta LR (loss learning rate in losses volatile blocks minus loss learning rate in wins volatile blocks). We will test for an interaction between Stimulation Time and tDCS Condition on Win and Loss delta LRs. This will be followed up by planned contrasts within and between the online and offline tDCS Conditions, as described above for hypothesis 3.