Healthy volunteers were recruited using advertisements on social media. Of 80 participants who were evaluated, 15 cancelled the session, 2 did not meet inclusion criteria and 3 were excluded prior to analysis because their age exceeded two STD from the mean. Participants were grouped into light social drinkers (LD, N = 30, 15 females) and non-treatment seeking heavy drinkers (HD, N = 30, 15 females) based on self-reported number of drinks per week. Heavy drinking was defined according to the Swedish public health agency, as 15 or more drinks per week for males and 10 or more drinks/week for women (Socialstyrelsen 2011).

Inclusion criteria included consumption of alcohol on one or more occasions in the past 3 months, age 20–65 years, minimum high school education and sufficient knowledge of Swedish to understand participant information and experimental instructions.

Subjects were excluded if they were currently receiving or seeking treatment for alcohol problems, had used illicit substances in the past month, had any current clinically significant psychiatric disorder, any past or present psychotic or bipolar disorder, had ongoing psychoactive medication, were pregnant or nursing, or had any history of clinically significant neurological disorders. To determine eligibility, the Modified MINI Screen (MMS) (Alexander et al. 2007) was used to screen for psychiatric disorders, and the Drug Use Disorder Identification Test (DUDIT) (Berman et al. 2007) and urine drug screening was used to identify illicit drug use. All women performed a urine pregnancy screening.

Baseline personality traits were obtained using the NEO Five Factor Inventory (NEO-FFI) (McCrae & Costa 1987). The Barratt Impulsiveness Scale (BIS) was used to assess the behavioral construct of impulsiveness (Barratt 1965). The Family Tree Questionnaire (FTQ) (Mann et al. 1985) was used to asses family history of AUD. AUDIT was used to assess alcohol use problems, and also served as proxy for the presence of AUD. We didn’t evaluate participants for a formal diagnosis of AUD, and a potential presence of AUD was not exclusionary. Given the AUDIT scores of participants in our heavy drinker group, and published data on the correlation between AUDIT scores and AUD diagnosis, all our participants in the heavy drinker group likely met criteria for moderate and severe AUD (Källmén et al. 2019; Moehring et al. 2019). However, participants were included only if they were not currently seeking or receiving treatment for alcohol problems; participants with clinically significant alcohol use problems were counselled after completion of the study, and offered referral to treatment.

The study was approved by the Swedish Ethical Review Authority (ref. 2020/07255) and all participants provided written informed consent.

The study consisted of one afternoon visit, which included a screening assessment followed by a laboratory session (Fig. 1). Participants were asked to abstain from alcohol (72 h) and food (3 h) prior to the session.

Study timeline. During screening MMS Modified Mini Screen, DUDIT Drug Use Disorders Identification Test, AUDIT Alcohol Use Disorder Identification Test were used and BrAC Breath Alcohol Concentration measured. After inclusion, participants filled out the NEO-FFI NEO Five-Factor Inventory, FTQ Family Tree Questionnaire and BIS-11 Barratt Impulsiveness Scale. Participants received instructions and did a trial version of the Choice Alcohol-Food (CCAF) task. The task was then performed in front of a computer. After task completion, a blood sample was taken and finally, participants received the reward depending on CCAF task results

During screening, prospective participants were evaluated for eligibility by a research nurse or a physician and, upon inclusion, filled out questionnaires. Participants were then invited to carry out two different tasks, assessing choice preference for alcohol and social processing (the latter not presented here).

Before carrying out the task, participants were presented with four alcoholic drinks (lager beer, vodka, red wine and white wine) and bowls containing four snacks (chocolate, chips, nuts, and candy), and asked to pick one alcoholic drink and one snack depending on their preference. At this stage, the subjects could hold and smell the glasses and bowls containing the rewards, but not consume them. Pictures of the chosen items were later used to tailor the stimuli to participants’ preferences. Task instructions were given, and the subjects then practiced the task in preparation for the experimental session.

After completing the experimental session, participants provided a blood sample for future genetic analyses. At the end of the session, all subjects received one standard drink of their preferred alcoholic drink or one serving of their favorite snack, depending on which of these reward categories they had earned most points for during the task. Pictures, task instructions and analysis scripts are available on GitHub: https://github.com/NeuroIP/CCAF.

The CCAF task was modified from (Hogarth & Hardy 2018). Participants were instructed to collect points toward either an alcohol drink or a snack that they redeemed at the end of the session. They were told that the more points they collected toward a reward, the better the chance to receive that reward in the end. The CCAF task trial is presented in Fig. 2. During each trial, participants were first presented with two concurrent pictures showing their preferred snack and their preferred alcoholic drink (3000 ms). During this interval, participants chose one of the two pictures by clicking mouse buttons with their index or middle finger. Following their choice, there was a jittered fixation interval (1500–3500 ms), after which participants were presented with a feedback image (2000 ms), showing the points earned during the trial for the respective reward category, and the running total in brackets. Each alcohol and snack picture was associated with either 1 or 3 points, shown on the side of the picture. This feature introduced three relative point levels. When both pictures were associated with either 1 or 3 points, the relative point levels were equal (0). When the relative point level differed, it could be in favor of alcohol (+ 2) or snacks (-2). Each relative point level was balanced across trials, with 32 trials per relative level and 96 trials in total. The position of alcohol and snack images (left or right) was counterbalanced across trials. At the end of the session, participants who accumulated more alcohol points than snack points in total received one standard drink of their preferred alcoholic drink, while those who accumulated more snack points received a portion of their favorite snack.

An illustrative Concurrent Choice Alcohol-Food (CCAF) task trial. Participants made forced choices between accumulating alcohol or food points. The accumulated points during the trial were then presented after a jittered fixation period. Total points were also displayed in brackets, and a trial countdown was presented at the bottom of the screen. Conditions were balanced based on relative point level (-2, + 0, + 2). The trial depicted here illustrates an increased relative point level for food compared to alcohol (-2 relative point level). Pictures associated with alcohol/snacks were based on the selections made by participants prior to the session

Sample size was based on a power analysis using G*Power, based on a within-subject factor with 3 levels (i.e. relative point level: -2, 0, + 2) and a between-subject factor with 2 levels (i.e. group: light drinkers, heavy drinkers). Assuming an effect size of Cohen’s f ≥ 0.31 (moderate effect size, as seen previously (Hogarth & Hardy 2018) and an α = 0.05, a total sample size of 58 subjects was required to detect a between-group behavioral effect with ≥ 80% power.

The individual percentage of alcohol choice, reflecting the number of trials when alcohol was chosen as a percentage of the total number of choice trials was calculated. Data were highly skewed, with near-ceiling and -floor effects for relative point levels + 2 and -2 respectively. Therefore, percentage of alcohol choice data were first logit transformed and then used for all subsequent analyses.

To identify predictors of percentage of alcohol choice, we entered relative point level (-2, 0, + 2) and group (light drinkers, heavy drinkers) as independent variables and sex (female, male) as covariate in a linear mixed-effects (lme) model.

As a metric of choice behavior, we assessed the relative point level at which alcohol and snacks were equally chosen in the two groups. First, percentage of alcohol choice data according to relative point level were separately fitted, using logistic regression, for heavy and light drinkers, respectively. Parametric bootstrapping was performed with 1000 samples to obtain 95% confidence intervals on model parameters. Points of subjective equality (PSEs) for the two groups were estimated using the fitted sigmoid curves resulting from the logistic regressions to interpolate the relative point levels at which alcohol and snacks were equally likely to be chosen. In addition, we further calculated individual PSEs and characterized each participant as having an overall preference for alcohol, snack or neither of the two (based on 95% bootstrapped CIs of the PSE overlapping 0). Frequencies of light and heavy drinkers in each category were then compared using Chi2 test.

Finally, we calculated the association between alcohol use severity and percentage of alcohol choice. First, we performed a lme model with AUDIT, relative point level, and group as independent variables and sex as covariate. Percentage of alcohol choice data were fitted using logistic regression across all participants relative to AUDIT scores, separately for each relative point level. Parametric bootstrapping was performed with 1000 samples to obtain 95% confidence intervals on model parameters. Slopes were obtained from the model fit.

Data were analyzed using R version 4.0.3 with the quickpsy package version 0.1.5.1 (Linares & López-Moliner 2016) and SPSS version 28.0.1. Graphs were created in R with the ggplot2 package version 3.3.6 (Wickham 2016).