Effects of Gabapentin on Dorsal Anterior Cingulate Cortex GABA and Glutamate Levels and Their Associations With Abstinence in Alcohol Use Disorder: A Randomized Clinical Trial

¹H-MRS data for the present analysis were collected as part of a larger (N=96) 16-week, randomized, double-blind, placebo-controlled trial of gabapentin (titrated over 5 days to 1,200 mg/day) for individuals with AUD and a history of alcohol withdrawal syndrome. A history of posttraumatic stress disorder with stable symptoms was allowed given its comorbidity with AUD and potential response to gabapentin (19). The study protocol was approved by the Medical University of South Carolina Institutional Review Board. Briefly, after providing written informed consent and undergoing assessment, participants who met DSM-5 (20) criteria for AUD and a history of alcohol withdrawal syndrome, with no current major psychiatric illnesses, no other substance dependence, and no clinically significant alcohol withdrawal syndrome at baseline (defined as a score ≥10 on the Clinical Institute Withdrawal Assessment for Alcohol Scale–Revised [21]), were randomly assigned to receive gabapentin or placebo for 16 weeks, with drinking assessed during this period using a daily calendar method (the timeline followback) (22). The primary report found that significantly more individuals in the gabapentin group had no heavy drinking days and total abstinence compared with those in the placebo group (2). Consented and evaluable participants of the ¹H-MRS substudy (N=68) (Figure 1) completed their first scan immediately prior to treatment randomization (following ≥3 days of abstinence from alcohol; mean=4.24 days, SD=2.03) and their second scan approximately 14 days after randomization (median=14 days, range=11–21). The initial 3-T Trio (N=47) MRI scanner was upgraded to a Prisma-Fit 3-T scanner (N=21) (Siemens, Germany) partway through the study, with the same 32-channel phased-array head coil used across scanners. Time from randomization and scanner were examined as covariates across statistical models, but were eliminated, as they were universally nonsignificant (p values >0.10).

A structural scan was taken for ¹H-MRS voxel placement and tissue segmentation (256 sagittal slices; 1 mm thick, 50% gap). A 2.5×2.5×3 cm³ dACC voxel was placed on midsagittal T₁-weighted images, posterior to the genu of the corpus callosum with the ventral edge aligned to the dorsal edge of the callosum (23). The dACC was chosen because of its demonstrated role in brain response to alcohol cues, relapse to heavy drinking, and core neurobehavioral deficits associated with AUD, and because most previous ¹H-MRS AUD studies have measured this region (24–27). See Figure 2 for visualization of the voxel and sample spectrum. After placement of saturation bands 1 cm away from voxel faces and shimming via FASTESTMAP (28), single-voxel water-suppressed ¹H-MRS spectra were acquired using a MEGA-PRESS sequence (TR=2,000 ms, TE=68 ms, number of averages=256) with symmetric editing pulse frequencies for macromolecule suppression (1.9 ppm, 1.5 ppm) (29) and a PRESS sequence sensitive to glutamate (TR=2,000 ms, TE=40 ms, number of averages=128) (30). Each sequence was followed by a matched water-unsuppressed acquisition for phase and eddy current correction and concentration referencing. Common sequence parameters included TR=2,000 ms, 16-step phase cycling, spectral width=2.5 kHz, and 2,048 complex data points. MEGA-PRESS data were processed using the Gannet MATLAB toolbox (version 3.1), with frequency/phase correction applied prior to fitting (31). PRESS data were processed using LCModel, version 6.3, with a vendor-supplied GAMMA-simulated basis set and an analysis window of 0.2–4.0 ppm (32). Within-voxel tissue fractions of gray and white matter and CSF were calculated based on automated segmentation in SPM12 (Wellcome Department of Cognitive Neurology), using a volume mask generated in Gannet (33). Metabolite concentrations were normalized to unsuppressed water and corrected for within-voxel CSF fraction.

First, to evaluate the generalizability of our previously reported findings to the present subsample, the number of evaluable individuals with no drinking days (abstinent) over the entire 16-week trial was analyzed using a two-sample z test to produce a chi-square p value; individuals with missing drinking data were assumed to have been drinking (2). ¹H-MRS quality control was conducted via visual inspection of spectra (34), by investigators blind to condition (J.J.P. and T.R.B.), along with evaluation of spectral fit errors relative to their observed distribution. Spectra with fit errors >3*interquartile range above the third quartile or below the first quartile were excluded from analysis. Change variables were created for both glutamate and GABA by subtracting initial scan values from those at the second scan. The ratio of gray matter to brain matter was calculated as the amount of within-voxel gray matter over the sum of within-voxel gray and white matter. The relationship between change in metabolite values (the dependent variable) and medication group, drinking levels between scans (percent days abstinent), and their interaction was estimated with linear models. Each model included covariates measuring baseline percent days abstinent, the metabolite value at the first scan, and ratio of gray matter to brain matter. To further examine the interaction between medication group and between-scan drinking, PROCESS model 1 was used to calculate Johnson-Neyman significance regions (35). For the second part of the analysis, linear mixed-effect models were estimated using REML in the R package nlme (36). These models predicted weekly percent days abstinent over the remainder of the study (seven 2-week periods) following each participant’s second scan and used all available longitudinal data from each participant, providing unbiased parameter estimates and standard errors when missing values were at least missing at random (37). Medication group, change in metabolite values, and their interaction were included as predictors. Baseline percent days abstinent was incorporated as a covariate. These models were reestimated with baseline metabolite values entered in place of change in metabolite values to evaluate whether baseline metabolite levels could serve a treatment response prediction function, similarly to change in metabolite levels.

Four participants (three in the gabapentin group, one in the placebo group) were excluded from the glutamate analyses and one (in the gabapentin group) from the GABA analyses because their ¹H-MRS scans failed quality control. Within-voxel gray matter–to–brain matter tissue fraction did not significantly differ between participants in gabapentin group (mean=0.65, SD=0.043) and the placebo group (mean=0.67, SD=0.039; p=0.081). (See Table S1 in the online supplement for ¹H-MRS quality control metrics and Table S2 for summary statistics of metabolite levels organized by scan and by group.)

Controlling for pretreatment glutamate level (β=−0.57, p<0.001) and percent days abstinent over the 90 days preceding screening (β=−0.14, p=0.745), as well as percent days abstinent between pretreatment and posttreatment scans (β=−0.81, p=0.069) and gray matter–to–brain matter tissue fraction (β=0.92, p=0.703), participants in the gabapentin group had greater decreases (and/or smaller increases) in glutamate between scans (β=−1.26, p=0.008). However, this main effect of treatment was qualified by a significant interaction with percent days abstinent between scans (β=1.68, p=0.005). Specifically, Johnson-Neyman analyses demonstrated that participants in the gabapentin group with <50% of days abstinent between scans had greater decreases (and/or smaller increases) in glutamate relative to those in the placebo group (p values <0.05), whereas participants in the gabapentin group with >95% of days abstinent between scans had greater increases (and/or smaller decreases) in glutamate relative to those in the placebo group (p=0.08) (Figure 3A). (See Figure S1 in the online supplement for spaghetti plots of glutamate levels by between-scan drinking by treatment group.)

Conversely, controlling for pretreatment GABA level (β=−0.77, p<0.001) and pretreatment percent days abstinent (β=0.09, p=0.445), as well as percent days abstinent between scans (β=0.37, p=0.002) and gray matter–to–brain matter fraction (β=−1.23, p=0.038), participants in the gabapentin group had greater increases (and/or smaller decreases) in GABA between scans (β=0.33, p=0.009). However, similar to the glutamate measures, this main effect of treatment was qualified by a significant interaction with percent days abstinent between scans (β=−0.48, p=0.002). Specifically, Johnson-Neyman analyses demonstrated that participants in the gabapentin group with <40% of days abstinent between scans had greater increases (and/or smaller decreases) in GABA relative to those in the placebo group (p values <0.05), whereas participants in the gabapentin group with >85% of days abstinent between scans had greater decreases (and/or smaller increases) in GABA relative to those in the placebo group (p values <0.05) (Figure 3B). (See Figure S2 in the online supplement for spaghetti plots of GABA levels by between-scan drinking by treatment group.)

After controlling for pretreatment percent days abstinent (β=0.57, p=0.003) and time in study (averaged over 2-week blocks) (β=−0.01, p=0.268), there was a significant interaction between medication group (main effect, β<−0.01, p=0.909) and change in between-scan glutamate levels (main effect, β=−0.14, p=0.051; interaction, β=0.28, p=0.007) on percent days abstinent during the remaining study treatment period. Specifically, Johnson-Neyman analyses demonstrated that participants in the gabapentin group who had decreased glutamate levels between scans (i.e., <−0.88 IU; Z<−1.24 or 7.8% of the sample) had fewer percent days abstinent for the remainder of the study relative to those in the placebo group (p values <0.05), whereas participants in the gabapentin group who had increased glutamate levels between scans (i.e., >0.79 IU; Z>0.86 or 18.8% of the sample) had more percent days abstinent relative to those in the placebo group (p values <0.05) (Figure 4A). Substituting between-scan changes in glutamate levels for baseline glutamate levels, we found a significant interaction between medication group (main effect, β=3.24, p=0.040) and baseline glutamate levels on percent days abstinent during the remaining study treatment period (main effect, β=0.22, p=0.028; interaction, β=−0.29, p=0.040) (see Figure S2 in the online supplement).

In contrast, when pretreatment percent days abstinent (β=0.54, p=0.017) and time in study (β=−0.01, p=0.299) were controlled for, there were no significant main effects of medication group (β=0.004, p=0.959) or change in between-scan GABA levels (β=0.26, p=0.287), nor of their interaction (β=−0.32, p=0.327), on percent days abstinent during the remaining study treatment period (Figure 4B). Substituting between-scan changes in GABA levels for baseline GABA levels provided similarly null findings (p values >0.50)