Perseverative behavior under uncertainty is enhanced by tryptophan depletion but diminished with subclinical obsessive–compulsive symptoms

Abstract

Serotonin is critically involved in flexible decision-making, yet the preponderance of evidence comes from non-human studies. Many insights about the serotonin's function have come from the technique of dietary acute tryptophan depletion (ATD). However, it is unclear whether ATD modulates choice behavior during probabilistic reversal learning (PRL), a widely used model of behavioral flexibility with significant translational and clinical value. To determine whether ATD affects latent choice tendencies, we applied computational models of reinforcement learning to PRL data from healthy human volunteers (n = 62; 29 females) who had received either ATD or placebo in a randomized, double-blind, placebo-controlled procedure. A secondary objective was to ascertain how model parameters related to clinically relevant self-report questionnaires. ATD did not affect the updating ("learning rates") or deployment ("reinforcement sensitivity") of value representations driving choice. However, ATD increased "stimulus stickiness", the tendency to choose a previously chosen visual stimulus again, regardless of reinforcement--a pattern seen in stimulant use disorder (SUD). Greater subclinical obsessive-compulsive symptoms were associated with lower stimulus stickiness, a pattern seen in clinically diagnosed obsessive-compulsive disorder (OCD). Low reinforcement sensitivity was instead related to intolerance of uncertainty and symptoms of depression and anxiety. Individuals with higher trait impulsivity showed lower reward and higher punishment learning rates, which parallels findings in SUD. Collectively, these results point to a role for serotonin in compulsive tendencies. They underscore the utility of computational modelling in illuminating the microstructure of behavior, which could point towards new markers of vulnerability to psychopathology.

Distinct computational mechanisms underlying cognitive flexibility deficits in impulsivity and compulsivity

“…In the exact model, the choice hysteresis parameter was significantly different from 0: 1.06±0.05, t(391)=21.8, p<0.0001 and was negatively associated with OCI-R total score in line with previous studies 7,14,15,17 : 𝛽𝛽 𝑂𝑂𝑂𝑂𝑂𝑂𝑅𝑅 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 = −0.17 (𝑆𝑆𝑆𝑆 = 0.05), 𝑝𝑝 = 0.0007 (controlling for the covariates:…”

“…A tendency to repeat previous choices irrespective of their value has been traditionally captured with an additional parameter during choice [47][48][49] . This "choice hysteresis" (tendency to repeat same action) parameter has previously been linked to compulsivity 7,14,15,17 . However, we found that choice hysteresis was no longer necessary when accounting for learning imprecisions, aligned with previous simulations 24 (Supplementary Information).…”

“…Finally, in line with previous study 24 , accounting for learning imprecisions resulted in recategorizing of more than 70% (74.6±1.5%) (mean±s.e.m) of the decisions labelled as 'nongreedy' (i.e., choices where an option with lower expected value has been chosen) under the exact model, as 'greedy' under the noisy model. This re-evaluation resulted in a dissociation between compulsive and impulsive traits previously linked to the non-greedy decisions 7,33,36 with only compulsivity positively contributing to the exploratory decisions (cf Supplementary Information).…”

“…Phenomenologically, compulsive behaviours are believed to arise from excessive rigidity and cognitive inflexibility, which leads to maladaptive compulsive perseveration and reduced goaldirected behaviour [6][7][8][9] . In contrast, premature and exaggerated switching is archetypical for impulsivity 1,10,11 .…”

Serotonin Reduces Belief Stickiness

“…Several studies have related serotonin, OCD, and obsessive-compulsive symptoms to stimulus stickiness [72][73][74][75] . Those studies used reversal-learning tasks, but analyzed behavior with S-R-learning models without state inference 73,75 . Our findings, like prior ones 69,70 , show that participants perform state inference in these tasks.…”