Binge drinking is associated with higher cortisol and lower hippocampal and prefrontal gray matter volume: Prospective association with future alcohol intake

2.1. Participants

Participants were healthy community adults recruited from the greater New Haven, Connecticut area via flyers, radio and social media advertisements calling for individuals who ‘liked beer’. Participants were healthy, non-substance using (except for alcohol), beer drinking men and women (N = 113, 56 males/57 females) aged 18–57 years and categorized into the binge drinking (BD) and moderate non-binge drinking (MD) groups using NIAAA specified criteria (NIAAA. Drinking, 2021) and with evidence of binge drinking episodes (five or more drinks in men; four or more drinks in women; per drinking episode) for the BD and no episodes of binge drinking for the MD group. Young to middle aged individuals aged 18–57 years old were included as recent estimates show significant growth in prevalence of binge drinking in this age range (Binge, 2019). Six of the BD and 3 of the MD group reported using cigarettes recreationally, but none were daily users. In addition, 15 out of 57 women were taking oral contraceptives. Participants were excluded if they had a history of psychotic disorders, current pregnancy, any nonremovable metal in their bodies, a history of loss of consciousness for longer than 30 min and any current other medical, psychiatric, or substance use disorder including alcohol use disorder (AUD) and taking medications for such illnesses. The BD and MD groups were not significantly different on sex, race, IQ, education, nicotine use and handedness, however, they differed in age, with MD older than BD (Table 1) and thus age was a covariate in all analyses. The follow-up real-world drinking assessment phase included 43 BD and 26 MD from the original sample (61%; N = 69, 34 males/35 females) (see Table 1). All participants provided written informed consent to participate in the study, and all procedures approved by the Human Investigation Committee of the Yale University School of Medicine.

2.2. Screening and intake procedures

All participants completed interviews and questionnaires assessing medical, demographic, and substance use characteristics as well as to determine psychiatric diagnoses (using the Structured Clinical Interview for the Diagnostic and Statistical Manual IV-TR (First et al., 1995)). Participant's drinking history was assessed using the Cahalan Quantity Frequency Variability Index (QFVI) (Cahalan et al., 1969; Babor et al., 2011) which assesses alcohol intake patterns, including typical and maximum frequency and amounts of alcohol intake, types of alcohol usually consumed and time period of specific recent pattern of intake. At all appointments including the MRI scan day, alcohol and carbon monoxide breathalyzers and urine toxicology screens confirmed that participants were in an alcohol, nicotine and drug-free state. Participants were also asked to refrain from alcohol use prior to the MRI scan day for 24–48 h. MRI scan days for women was not scheduled during ovulation days, assessed via self-report and based on 28–30 day cycles.

2.3. Cortisol measurement

All participants arrived at the Yale Magnetic Resonance Research Center at 7:00 a.m. after overnight fasting to control for diurnal cortisol dynamics (Hansen et al., 2008). An intravenous line was inserted in the nondominant forearm for blood draws to obtain repeated plasma sampling at four timepoints, at pre-scan after IV insertion (7:15 a.m.), during the scan at 8:30 and 9:30 a.m. and at end of scan at 10:00 a.m. during the 2-h MRI scan period. Blood (4 ml per timepoint) was collected in EDTA-coated tubes (BD Biosciences, Franklin Lakes, N.J.). Blood samples were immediately stored on ice, and plasma was separated by centrifugation at 4 °C for 10 min at 1000×g. Aliquots of plasma were stored in polypropylene tubes at −80 °C until assayed. Cortisol levels were assessed via radioimmunoassay at the Yale Center for Clinical Investigation Core Laboratory (Blaine et al., 2020).

2.4. MRI data acquisition and preprocessing

Structural magnetic resonance imaging (sMRI) data were acquired using a high-resolution T1-weighted magnetization-prepared rapid gradient echo (MPRAGE) sequence (voxel size = 1 mm³, field of view = 256 × 256, TR = 300 msec, TE = 2.46 msec, flip angle = 7°) using two 3-T Siemens Prisma scanners (Siemens, Erlangen, Germany; BD: 53% Prisma A, 47% Prisma B; MD: 56% Prisma A, 44% Prisma B). There were no significant differences in the proportion of subjects from each group for whom data was acquired on different scanners.

Image segmentation (into gray matter, white matter, and cerebrospinal fluid) and registration were conducted using segmentation and the DARTEL registration algorithms (SPM12). By using the segmentation procedure, improvements were given to the registration model as well as including additional tissue probability maps to better model cerebrospinal fluid (CSF) and other nonbrain voxels. DARTEL (Ashburner, 2007) is a high-dimensional and diffeomorphic registration algorithm which has shown excellent performance among registration algorithms (Klein et al., 2009). To provide better initial estimates for segmentation, each image was manually set for space origin to the anterior commissure and aligned with the plane of the anterior and posterior commissures (SPM12 Display). Default settings were used for segmentation and for bias correction, spatial normalization, and segmentation of the original structural images in the same model. A modulation step was used in normalization to ensure the overall tissue amount was not altered. Resulting segmentations of GM were validated visually (Cerasa et al., 2009; Davies et al., 2009) as manual segmentation is considered the gold standard for evaluation of the quality of automated tissue classification (Anbeek et al., 2004; Bouix et al., 2007). In particular, cortical thickness relative to each subject's native space image and to a published report of cortical thickness (Fischl and Dale, 2000) was evaluated. Appropriate face validity was demonstrated by gray matter (GM) segmentation in all images. The first author (SF) performed all the preprocessing steps including image registration and segmentations of gray matter volume described above and manual visualization for quality control while blind to study group. After completing these steps, Dr. Fan was unblinded in order to perform second-level analyses. Additionally, no statistical outliers were identified in the extracted values representing segmented GM, white matter (WM) and CSF tissues.

Following segmentation, single-level GM and WM segments were warped to the space of the mean image averaged from all the study participants and normalized to the Montreal Neurological Institute (MNI) template space using default settings (DARTEL), which include resampling to 1.5 mm³ voxels (reducing memory demands for the large number of parameters estimated by the registration algorithm). Final outputs were modulated GM segments (1.5 mm³ voxels) smoothed using an 8-mm Gaussian filter (Rando et al., 2011, 2013a, 2013b). Gaussian smoothing reduces the effects of residual misregistration on potential group differences and reduces departures from normality that may occur at some voxels (Ashburner and Friston, 2000).

2.5. Follow-up phase: post-scan daily diary assessment of alcohol intake over 30 days

A large subgroup of BD (N = 43) and MD (N = 26) (61% of the original sample) participated in a follow-up phase added after study initiation. In this phase, participants reported their daily alcohol intake up to 36 days (50 participants reported drinking data for 30 days while 19 provided daily data for up to 36 days) using a smartphone application (MetricWire, https://metricwire.com/). They were asked to complete the prompt at 5 pm each day. Typically, the participants were asked these questions: 1) “Did you drink alcohol since the last prompt?“, 2) “How much did you drink in total? (If they answered “yes” on question 1)”, 3) “Did you drink anything yesterday after your last survey; how much did you drink in total?“. The amount of alcohol intake per day was calculated as the sum of the amount since the last prompt the day before (Q3) plus total alcohol intake on that day (Q1-2). We used these questions to derive three metrics of real-world drinking behavior reported daily for the 30-day monitoring period. First, to calculate the average number of alcoholic drinks per day, which was the sum of alcohol drinks was divided by the number of monitoring days. Second, number of drinking days was the total days when alcohol was consumed during the assessment period, and third, number of heavy drinking days was the number of days on which binge drinking occurred (5 or more drinks/occasion in men and 4 or more drinks/occasion in women).

2.6. Data analysis

Between-group differences on demographic, clinical characteristics, baseline and prospective drinking measures as well as basal cortisol levels were examined using IBM SPSS Statistics version 26 (64-bit edition).

All structural MRI images were preprocessed in SPM 12 (see above) and second-level statistical between-group comparisons were conducted using AFNI (version 16.0.09). To test our hypothesis that GMV would differ between BD and MD, a linear mixed-effect (3dLME) model (Chen et al., 2013) was created in AFNI, with whole brain GMV analyzed as a function of Group (BD vs MD) with covariates for age, sex, educational level, and global brain size (i.e., estimated total intracranial volume (TICV), calculated by summing voxel-wise native space GM, WM and CSF segments per subject). To correct for multiple comparisons, we used family-wise error correction determined by Monte Carlo simulation using AFNI's 3dClustSim (bi-sided first-nearest neighbor). A threshold p value of 0.005 was considered statistically significant for whole-brain GMV analysis and an additional family-wise cluster correction of p = 0.05 was used.

For post-hoc analyses examining the directionality of associations between cortisol, GMV and future drinking, Bioimage Suite version 35 (Joshi et al., 2011) was utilized with a whole brain significance threshold of p < 0.005 and an additional cluster correction p = 0.05. Similarly, to test the third and final hypothesis exploring whether high cortisol levels were associated with changes in GMV in BD, and whether cortisol and GMV were associated with future alcohol intake in BD, Bioimage Suite version 35 was used to conduct whole-brain correlation analysis in the BD group at the whole brain significance threshold of p < 0.005 and an additional cluster correction of p = 0.05 was used.

3.1. Participants

Binge and moderate drinkers were not significantly different on sex, race, IQ, education, handedness, and smoking status, as well as lifetime incidence of mood and anxiety disorders. On average, binge drinkers were younger than moderate drinkers, (t(111) = -2.78, p = 0.003) (Table 1); accordingly, age and also sex and education were included as covariates in all group differences analyses. As expected, baseline assessment of drinking revealed participants in the BD group had significantly more alcohol drinking days (p < 0.001) and consumed more per drinking day (p < 0.001). However, the groups did not differ in the number of years of alcohol use (Table 1).

3.2. Participants in the post-scan prospective daily alcohol intake follow-up phase

The subgroup of participants who completed the follow-up phase did not differ significantly from the main cohort in age, sex, years of education, IQ, race, alcohol use variables, proportion of smokers, or lifetime diagnoses of mood disorders. Consistent with baseline group data, BD relative to MD participants who completed the follow up phase also consumed significantly more drinks per day (t(74) = 4.45, p < 0.001) and had significantly more drinking days (t(88) = 3.64, p < 0.001) and heavy drinking days (t(88) = 5.32, p < 0.001) during the 30-day follow-up (Table 1).

3.3. Differences in whole brain GMV by drinking group

Consistent with our hypothesis, we found that individuals in the BD group had significantly lower GMV than MD in regions including bilateral hippocampus, right dlPFC, bilateral motor areas (pre-motor, pre-SMA, SMA), left primary sensory cortex, right angular gyrus, right dorsal parietal central cortex, and bilateral visual cortex (Table 2, and Fig. 1a). No GMV clusters were identified that were significantly higher in BD compared to MD. To ensure that these differences were not driven by other stress-related psychopathology, we ran post-hoc analyses excluding individuals with lifetime mood (N = 18) and anxiety disorders (N = 15) and found that the GMV differences between BD and MD groups remained significant. Sex was included as a factor and there were no sex main effects or interactions in the whole-brain gray matter volume when comparing binge drinkers to moderate drinkers after controlling for sex as one of the confounding variables in our brain imaging model. Furthermore, post-hoc secondary LME analyses assessed whether age differences between drinking groups moderated GMV differences between drinking groups, and we found no significant interactions between age and drinking groups on whole brain GMV (see the Supplemental Methods).

An additional post-hoc exploratory analysis was also conducted to specifically assess sex differences in the GMV differences between BD and MD groups. Drinking Group effects were different in males and females in the regions of bilateral amygdala and hippocampus and left SMA. The significant interactions were driven by lower GMV in the right hippocampus and left SMA in BD males relative to MD males, and lower GMV in the left hippocampus and right hippocampus extended to the amygdala area in BD females relative to MD females (see the Supplemental Methods, Results, Table S1 and Fig. S1).

3.4. Higher fasting morning cortisol in binge drinking group

Cortisol levels dropped significantly during the 7:00am-10:00am scan window (main effect Time: F(3) = 37.39, p < 0.001), as expected based on morning drop in cortisol known to occur in the 24-h diurnal variation of cortisol levels (Blaine et al., 2019) (Fig. 1b). Notably, cortisol levels differed significantly between BD and MD (main effect Drinking Group: F(1) = 5.79, p = 0.018; Drinking Group x Time: F(7) = 17.47, p < 0.001). Post-hoc independent samples t-tests revealed that BD had significantly elevated cortisol levels at 8:30 a.m. (t = 2.70, p = 0.011) and 9:30 a.m. (t = 2.43, p = 0.023), but not at the 7:15 a.m. and 10:00 a.m. timepoints (Fig. 1b). These higher cortisol levels in BD were not significantly associated with recent past alcohol use. Also, there were no sex differences in the cortisol elevations seen in the BD vs. MD at the 8:30 and 9:30 a.m. timepoints, however female BD were significantly higher than the male BD group at the first 7:15 a.m. timepoint (p < 0.05), and no significant sex differences were seen in the MD group.

3.5. Relationship between altered GMV and concurrent high cortisol levels in BD

We next tested whether the elevated cortisol levels in BD were associated with differences in whole-brain GMV within this cohort. Lower GMV in bilateral dlPFC, premotor and SMA were significantly associated with higher cortisol levels at the 8:30 a.m. (whole brain corrected, FWE p < 0.05) (Fig. 1c). Post-hoc secondary analyses were performed to examine the role of sex in the association between brain GMV and cortisol levels and no significant sex-specific associations were observed. Notably, the cortisol associated SMA GMV was at a different location to the SMA clusters observed in the sex X drinking groups interaction effect (see the Supplemental Methods, Table S1 and Fig. S1). Additionally, we observed no significant association between hippocampal GMV and cortisol in the BD group.

3.6. Associations between lower GMV and higher future alcohol intake in BD

Finally, we tested the predictive validity of GMV changes in BD by assessing its impact on levels of future alcohol intake. Using the prospective follow-up phase drinking data, we examined relationships between whole brain GMV and future daily drinking across a subsequent 30-day period. Lower GMV in the left orbito-frontal cortex (OFC)/amygdala, left ventral anterior cingulate cortex, bilateral dlPFC, left dorsal ACC, medial PFC, left insular, right premotor/supplementary motor area, bilateral angular gyrus, right primary visual cortex, and right visual motor cortex were significantly associated with greater number of future any drinking days in BD (Table 3, Fig. 2a). Also, lower GMV in bilateral OFC, left dorsal ACC, bilateral dlPFC, left anterior PFC, premotor and bilateral supplementary motor area, right primary motor area, bilateral insular, right temporal pole and bilateral visual cortex was also associated with higher number of future heavy drinking days in BD (Table 3, Fig. 2b-2c). There were no significant associations with amounts of alcohol intake that met the corrected whole brain threshold. Post-hoc secondary analyses also revealed no significant associations between GMV and future drinking days in the MD group.