Investigating the Mechanisms of Decision-Making Using Non-Invasive Brain Stimulation

"Weight" references a parameter that is extracted from recorded choice behavior by means of computational modeling. Three basic models (one map-based, two stimulus-outcome association-based) are fit to each subject's choice data from each session. Two hybrid models - combining the map-based with either of the stimulus-outcome association-based components - are also fit. The weight parameter will be extracted from the hybrid model that fits most subjects' behavior best under the control condition (high-frequency stimulation). It describes the extent to which the subject relies on the map-based relative to the association-based component/strategy for making their choices.

Given the presumed roles of hippocampus in map-based and striatum in association-based decision-making, the investigators expect the weight parameter to increase under hippocampal tTIS and to decrease under striatal tTIS relative to high-frequency control stimulation.

Whether, out of two simultaneously presented stimuli, the one associated with the higher reward has been chosen on a given trial.

Early after a shift in reward contingencies, using a map-based strategy should increase the probability of choosing correctly. This is because the new reward contingencies may not have been experienced yet (disabling association-based strategies) and so can only be inferred (requiring map-based strategies). Both strategies should be equally successful later after a shift once all new contingencies have been experienced. Averaged across all trials, the investigators therefore expect to observe higher accuracy under hippocampal and lower accuracy under striatal tTIS as compared to control stimulation. The investigators also expect subjects' individual choice accuracies to positively correlate with their map-based weight parameter estimates.

Time between presentation of the two stimuli for the present trial and the subject's choice of one of the two stimuli.

The investigators will analyze the effects of hippocampal and striatal tTIS as compared to control stimulation on response time in an exploratory manner.

Different candidate models are fit to the recorded choice data, two of which are on the "stimulus-outcome association-based" and one on the "map-based generalization" side. Hybrid models (= weighted sum of one stimulus-outcome association-based element and the map-based generalization element) are also fit. WAIC (Watanabe, 2010) will be used to determine which model fits which subject best under any given stimulation condition. The investigators expect the proportion of subjects best described by the map-based (association-based) model(s) to increase (decrease) under hippocampal tTIS as compared to control stimulation, with the opposite pattern showing under striatal tTIS.

Note: Should ≥ 50 % of subjects perform at chance level in the proximity judgments in the first session (last eight blocks), the investigators will use subjects' *subjective* representations (as collected at the end of the last choice task session) instead of the objective ones as a basis for the map-based model.

The investigators will use (a) the difference in the most recently observed values of the stimuli presented to the subject's left vs. right and (b) the difference between the actual, current values of these stimuli (which may not have been observed yet) to predict whether the subject will have chosen the left vs. right stimulus. Subjects pursuing a stimulus-outcome association-based strategy will exclusively rely on directly experienced reward contingencies (strong regression weight of (a) but not (b)). Subjects generalizing across related stimuli will be able to infer a stimulus' current value even if that one has not been observed recently (strong regression weight of (b) but not (a)).

The investigators expect the regression weights of (a) to increase and (b) to decrease under striatal tTIS (opposite under hippocampal tTIS) as compared to control stimulation.

The investigators will use (a) the difference in the directly experienced past values of left vs. right stimuli and (b) the difference in the actual, current reward values of left vs. right stimuli (which may not have been observed yet) as predictors of response time. Subjects should respond more quickly if the value difference is larger (less conflict). Under this assumption, a comparison of regression weights (a) and (b) will reveal whether subjects' behavior is more strongly governed by (differences in) experienced values or by (differences in) inferred true values.

The investigators will analyze the effects of striatal and hippocampal tTIS as compared to control stimulation on the regression weights in an exploratory manner.

Subjects will learn the latent relations between stimuli on the first day by means of a navigation task and a proximity judgment task (among two stimuli, identifying the one that is closest to a third). Both tasks are presented in an alternating manner. Proximity judgment task accuracy is the proportion of trials on which the closer stimulus was correctly identified. Proximity judgment task accuracy in the last 8 blocks is expected to be above chance and to increase over blocks as subjects learn the relations between stimuli. Proximity judgments are additionally collected at the beginning of the second, third, and fourth sessions *before* stimulation onset to determine subjects' current knowledge of the latent relations. Stable performances are expected across sessions 2 through 4 and between stimulation conditions. Supplementary analyses of choice behavior will be constrained to subjects with above-chance proximity judgment accuracy or include this accuracy as a covariate.

Time between presentation of the three stimuli for the present trial and the subject's response as to which of the two options they believe to be closest to the third stimulus. This measure will be analyzed in an exploratory manner.

• Which sensations do subjects report during familiarization with the stimulation? In the familiarization, the investigators will gradually expose the subject to a current intensity of 1, 2, 3, 4 mA (summed across both channels) both for tTIS and for high-frequency control stimulation (order counterbalanced between subjects and subjects blind to type of stimulation). Subjects are asked to report the strength and type of any sensations they may experience.
• Which side effects do subjects report relating to each of the three stimulation sessions?

The investigators expect tTIS and high-frequency control stimulation to be tolerable (Vassiliadis et al., 03/2024).

Are subjects' classifications of real vs. placebo stimulation for each session (as collected after the final stimulation session) correct above chance?

The investigators expect tTIS and high-frequency control stimulation to be indistinguishable from subjects' point of view, ensuring successful blinding (Vassiliadis et al., 03/2024).

Not an outcome variable but a variable that will be collected to characterize the study sample.

Not an outcome variable but a variable that will be collected to characterize the study sample.

German version (Hautzinger & Bailer, 2003). This is not an outcome measure but a variable that will be collected once at the beginning to characterize the study sample. It will not be analyzed as a function of stimulation condition. Sum scores range from 0 to 60, where higher scores indicate higher depressive symptom load.

Oldfield (1971). German translation. This is not an outcome measure but a variable that will be collected once at the beginning to characterize the study sample. It will not be analyzed as a function of stimulation condition but will be explored as a potential moderator of stimulation effects. Per side (left and right), between 0 and 20 "+"s can be marked, where a higher number of "+"s indicates stronger use of the respective hand.

Subjects will learn the latent relations between stimuli on the first day by means of a navigation task (deciding whether moving forward vs. backward will yield the shortest path from one stimulus to another) and a proximity judgment task. Both tasks are presented in an alternating manner. Navigation task accuracy is the proportion of correct forward/backward choices. The investigators expect navigation task accuracy in the last 8 blocks to be above chance and to increase over blocks as subjects learn the relations between stimuli. Navigation task accuracy is further expected to correlate positively with proximity judgment task accuracy. Because the navigation task is only completed in the first study session, it cannot be analyzed as a function of stimulation condition.

Time between presentation of the two stimuli for the present trial and the subject's response as to whether moving forward vs. backward will yield the shortest path from one to the other. This measure will be analyzed in an exploratory manner.

Subjects will be asked at the end of the final stimulation session to visualize their mental representation of stimuli by dragging and dropping those into a virtual arena on the screen. Importantly, there is a "correct" arrangement (iterative sequence of stimuli), which subjects are never fully shown but are supposed to learn through the tasks presented on Day 1. The subjective arrangement is considered fully correct if it corresponds to the true underlying arrangement.

Supplementary analyses of choice behavior will be constrained to subjects with correct arrangements. The investigators expect the weight parameter increase under hippocampal tTIS vs. control stimulation to be more pronounced in this subsample than in the subsample with ambiguous or incorrect arrangements. The same is expected for the hippocampal tTIS-contingent increase in the influence (regression coefficient) of the true value difference on choice.