Learning Mechanisms to Predispose Risky Alcohol Drinking Behaviors During Young Adulthood

Alcohol use disorder (AUD) is a mental disorder that negatively affects personal health and burdens the global health system. Alcohol-attributed harms can also extend beyond the drinkers to other people in the society through increased road traffic accidents and more interpersonal violent behaviors. The effects of this disorder make it crucial to investigate predisposing mechanisms in order to identify at-risk individuals and further develop novel interventions. Although aberrant learning and dysfunctions in decision-making have been observed in individuals with AUD, it is not yet clear whether they predispose the development of risky drinking behaviors or result from repetitive alcohol use. To disentangle this, we studied the drinking behaviors of a community sample comprising participants who were 18–24, which is when the prevalence of alcohol use typically peaks. This thesis investigates whether two types of learning mechanisms—the balance between goal-directed and habitual control and the susceptibility to interference between Pavlovian cues and instrumental behaviors—are associated with the development of risky alcohol drinking behaviors. For Study 1, we assessed how goal-directed and habitual controls at 18 predispose alcohol use development over the course of 3 years. Goal-directed and habitual control, which are informed by model-based (MB) and model-free (MF) learning, were assessed with a two-step sequential decision-making task during functional magnetic resonance imaging. Three-year drinking trajectories were constructed based on the Alcohol Use Disorders Identification Test (AUDIT-C; assessed every 6 months) and a gram/drinking occasion measure (binge drinking score; assessed yearly). Latent growth curve models were applied to examine how the MB and MF controls were associated with the drinking trajectories. We found that MB control was negatively associated with the development of the binge drinking score trajectory. In contrast, MF reward prediction signals in the ventromedial prefrontal cortex and the ventral striatum (VS) were associated with a higher starting point and a steeper increase/less decrease in AUDIT-C, respectively. For Study 2, we investigated the cross-sectional association between the susceptibility to interference between Pavlovian cues and instrumental behaviors and risky (binge) drinking behaviors at age 18. During a Pavlovian-to-instrumental transfer (PIT) task, the participants were instructed to “collect good shells” and “leave bad shells” while the appetitive (monetary gain) or aversive (monetary loss) Pavlovian cues were presented in the background. The behavioral interference PIT effect was characterized by an increased error rate (ER) during incongruent trials (“collecting good shells” in the presence of an aversive Pavlovian cue or “leaving bad shells” during the presentation of an appetitive Pavlovian cue) in comparison to congruent ones. Overall, the individuals demonstrated a substantial behavioral PIT effect. Neural PIT correlates were found in the VS, dorsomedial, and lateral prefrontal cortices (dmPFC and lPFC, respectively). High-risk drinkers, in comparison to low-risk drinkers, exhibited a stronger behavioral PIT effect, decreased lPFC responses, and increased trend-level VS responses. Moreover, the effective connectivity from the VS to the lPFC during the incongruent trials was weaker for the high-risk drinkers, which indicates that the altered interplay between bottom-up and top-down neural responses may contribute to the poor interference control performance of this group. During Study 3, we further examined whether the susceptibility to Pavlovian cues during conflict trials was associated with the development of drinking behaviors over 6 years from ages 18 to 24. The drinking behaviors were again constructed based on the AUDIT-C and the binge drinking score. The PIT task was assessed at ages 18 and 21. Following Study 2, the increased ER in the incongruent condition compared with the congruent condition (along with the neural responses in the VS, lPFC, and dmPFC during the incongruent trials) were included in the latent growth curve models as predictors. A stronger VS response during a conflict at age 18 was associated with a higher starting point in both drinking trajectories but was negatively associated with the development of the binge drinking score trajectory. At age 21, high ER and enhanced neural responses in the dmPFC were associated with a risky AUDIT-C trajectory that started to emerge and develop until age 24. Through exploratory cluster analyses of the drinking trajectories, we identified two subgroups: the drinking behavior in the 'late riser' group escalated after age 21, whereas the drinking of 'early peakers' culminated at this age and then declined. The late risers displayed enhanced dmPFC responses and higher ER during conflict at age 21. Interestingly, this group also exhibited an increased ER from ages 18 to 21. Taken altogether, the unbalanced goal-directed to habitual control, informed by less MB and more MF control, appears to be a strong predisposing candidate mechanism that underlies the development of risky drinking behaviors during young adulthood. At age 18, the susceptibility to interference between Pavlovian cues and instrumental behaviors was associated with risky drinking behavior. The development of risky drinking behaviors over the 6 years was associated with the behavioral interference PIT effect at age 21 and its change from ages 18 to 21. Researchers could further explore the dynamics in PIT to predict risky drinking behaviors in the future