GABAergic modulation of conflict adaptation and response inhibition

Adaptive behavior is only possible by stopping stereotypical actions to generate new plans according to internal goals. It is response inhibition —the ability to stop actions automatically triggered by exogenous cues— that allows for the flexible interplay between bottom-up, stimulus driven behaviors, and top-down strategies. In addition to response inhibition, cognitive control draws on conflict adaptation, the facilitation of top-down actions following high conflict situations. It is currently unclear whether and how response inhibition and conflict adaptation depend on GABAergic signaling, the main inhibitory neurotransmitter in the human brain. Here, we applied a recently developed computational model (SERIA) to data from two studies (N=150 & 50) of healthy volunteers performing Simon and antisaccade tasks. One of these datasets was acquired under placebo-controlled pharmacological enhancement of GABAergic transmission (lorazepam, an allosteric modulator of the GABA-A receptor). Our model-based results suggest that enhanced GABA-A signaling boosts conflict adaptation but impairs response inhibition. More generally, our computational approach establishes a unified account of response inhibition and conflict adaptation in the Simon and antisaccade tasks and provides a novel tool for quantifying specific aspects of cognitive control and their modulation by pharmacology or disease. Author Summary Our capacity to prepare for situations that afford conflicting responses (conflict adaptation) and to stop our immediate impulses in these scenarios (response inhibition) are the hallmark of cognitive control. As these abilities require both the stopping or slowing of response tendencies, a natural question is whether they are mediated by inhibitory neurotransmission in the brain. Here, we combined computational modeling with two experiments to investigate how conflict adaptation and response inhibition interact with each other (experiment 1) and how these are modulated by lorazepam (experiment 2), a positive modulator of the GABA-A receptor, one of the main inhibitory receptors in the human brain. Using our computational model to disentangle conflict adaptation and response inhibition, our results indicate that while lorazepam impaired response inhibition, it improved conflict adaptation. Thus, our results suggests that conflict adaptation is mediated by GABA-A neurotransmission.

: Antisaccade task. A central fixation cue was presented for 1000 to 2000ms. Its color (blue or yellow) indicated subjects to saccade to the peripheral stimulus (congruent trial) or to saccade in the opposite direction (incongruent trial). The peripheral stimulus was presented for 1000ms. Simon task. Subjects were instructed to press a left ('x') or right (',') key depending on the color (blue or green) of a peripheral cue (display duration 1500ms) following a fixation period of 500ms. On congruent trials, the right-left location of the cue and the correct button matched each other; on incongruent trials, they were in opposite locations. SERIA model. Reaction times and actions are assumed to be the outcome of a race to threshold between independent linear accumulators (processes), whose slopes take different, random values on each trial. Initially, the automatic process starts after a short delay from the cue presentation. This process can be stopped by the inhibitory process, if the latter is the first to hit threshold. When this occurs, the outcome of the race between two controlled processes that represent congruent and incongruent responses decides the action. The reaction time on a trial is assumed to be the threshold-hit-time of the corresponding process.

141
The number of excluded subjects increased t o 12 in the antisaccade task, as 142 11 subjects had 50% or more trials excluded and one subject's ER was higher 143 than 80%.

144
For clarity, in the following, "conflict level" refers to the N-1 trial. Congruent 145 trials are considered low conflict trials, and incongruent trials high conflict 146 trials.

152
Regarding ER, we again found no evidence of conflict adaptation as the 153 congruency effect was significantly higher after high conflict trials compared 154 to low conflict trials (Δ = 9%; = 0.004).

164
; < 10 -5 ). ER followed the same pattern (see Fig. 2D  Histograms show empirical data and solid black lines display the model fits. B) Accuracy function in the antisaccade task. Accuracy plots were generated by sorting trials in RT percentiles (20, 40, 60, 80 and 100%) and plotting the mean accuracy in each percentile against the corresponding mean RT. C) Delta plot in the antisaccade task. As with the accuracy plots, in delta plots trials are binned in RT percentiles and the congruency effect (the difference between mean RT on incongruent and congruent trials) is plotted against the pooled mean RT. D-F) RT distribution, accuracy and delta plot in the Simon task, similar to A-C.

183
Rather than merely predicting the RT histograms, we aimed to reproduce and 184 explain the time course of the congruency effect revealed by delta plots (21).

185
In this analysis, trials are binned in RT quantiles and either the accuracy or the 186 RT congruency effect (i.e., the difference in RT between incongruent and 187 congruent trials) are plotted against the quantile-specific mean RT.

188
As shown in Fig. 3B&E   responses, which is separable from the negative slope of the delta plots. This is absent in the antisaccade task, where no significant conflict adaption was 268 detected.

269
The success of SERIA in fitting the data from these two tasks begs the question

365
In the antisaccade task (Fig. 8) (Fig. 3A&D), mean RT and ER (Supp. Fig. 1), but it also captured 408 the time course of the congruency effect as visualized in the delta plot analysis 409 (Fig. 3B

877
In general, SERIA fitted the data from both tasks more accurately and had 878 higher WAIC than single process models with a comparable number of 879 parameters. Thus, the structural flexibility inherent to dual process models 880 explained RT distributions and ER better than single process models.