Associations between nucleus accumbens structural connectivity, brain function, and initiation of binge drinking

Abstract

Adolescent alcohol use is associated with increased risk for alcohol use disorders later in life; therefore, identifying biomarkers for initiation of heavy alcohol use, such as individual differences in the development of white-matter microstructure, may inform prevention strategies that improve public health. This prospective cohort study included 40 adolescents, ages 14 and 15, without substantial history of alcohol or drug use at baseline. Fractional anisotropy (FA), an index of white-matter microstructure, was assessed in pathways connecting the nucleus accumbens (NAcc) to the rest of the brain using diffusion tensor imaging. Path analyses were conducted voxel-wise within these pathways to examine direct effects of premorbid FA on number of months between baseline assessment and the onset of binge drinking and indirect effects mediated by NAcc activation during decision making assessed using functional magnetic resonance imaging. Adolescents with lower premorbid accumbofrontal FA began binge drinking sooner, an effect which was mediated by greater NAcc activation during decision making involving greater levels of risk and reward (P < .05 corrected). An additional direct effect of FA on duration to onset of binge drinking was observed in white matter near the ventral pallidum, as adolescents with lower premorbid FA in this region began binge drinking sooner (P < .05 corrected). Findings suggest that delayed maturation of prefrontal white matter is associated with less top-down control over striatal sensitivity to reward. These factors, along with individual differences in white matter proximal to ventral pallidum, may represent premorbid risk factors for earlier initiation of heavy alcohol use.

1 |. INTRODUCTION

Individual differences in neurodevelopment during adolescence may confer risk for a variety of neuropsychiatric conditions that emerge later in life,¹ such as substance use disorders. The development of white-matter pathways can be assessed by measuring fractional anisotropy (FA), a metric which is thought to reflect axon myelination, integrity, or coherence.² Longitudinal studies show that on average, FA increases during adolescence, but not all adolescents develop at the same rate.³ Understanding the association between individual differences in FA and the early initiation of alcohol use may aid in the identification of adolescents at increased risk for development of alcohol use disorders later in life,⁴ as the odds of developing an alcohol use disorder decrease by 14% with each year that initiation of alcohol use is delayed during adolescence.⁵

Early initiation of alcohol use may be related to the structure and function of the nucleus accumbens (NAcc), a region implicated in the reinforcing effects of alcohol,⁶ which promotes behavior congruent with motivationally relevant goals.⁷ College-aged binge drinkers have greater NAcc volume than age-matched control participants,⁸ and greater NAcc volume is associated with more frequent alcohol use in adolescence.⁹ Furthermore, in adolescents without a significant history of alcohol or substance use, greater NAcc activation during decision making involving higher levels of risk and reward predicted earlier initiation of binge drinking.¹⁰ The NAcc receives input from cortical (eg, medial prefrontal cortex), limbic (eg, amygdala), and dopaminergic brain regions (ie, ventral tegmental area) and relays information to the ventral pallidum and ventral tegmental area¹¹; however, it remains unknown whether individual differences in connectivity between these addiction-relevant regions is associated with the initiation of binge drinking.

The goal of this study was to determine if premorbid individual differences in FA in white-matter pathways connecting the NAcc to the rest of the brain were associated with duration to onset of binge drinking in adolescents. Cross-sectional studies have shown that adolescents who binge drink have lower FA than control participants.¹² Considering these effects may be partially attributed to premorbid differences, we anticipated that lower FA in 14 to 15 year olds would be associated with earlier onset of binge drinking. NAcc function during decision making has been linked to duration to onset of binge drinking in adolescents¹⁰ and to FA in adults¹³; therefore, we hypothesized that FA would have an indirect effect on duration to onset binge drinking mediated by NAcc activation during decision making involving risk and reward.

2 |. MATERIALS AND METHODS

2.1 |. Study design and participants

These data are from a larger, ongoing parent study of adolescent neurodevelopment, approved by the Oregon Health & Science University Institutional Review Board. Exclusionary criteria for both this study and the larger parent study at baseline included left handedness (Edinburgh Handedness Inventory¹⁴), DSM-IV psychiatric diagnoses,¹⁵ serious medical problems, significant head trauma, intellectual or learning disabilities, psychotic illness in a biological parent, prenatal exposure to drugs or alcohol, MRI contraindications, or pregnancy. Participants were also excluded at baseline for significant exposure to drugs or alcohol, including > 10 lifetime alcoholic drinks, > 2 alcoholic drinks on any one occasion, > 10 lifetime uses of marijuana, and > 4 lifetime uses of cigarettes, or any other drug use (Customary Drinking and Drug Use Record¹⁶). Based on prior work indicating the developmental epoch associated with the greatest functional imbalance between reward and cognitive control circuitry is between the ages of 14 and 15 (for a review, see Bjork and Pardini¹⁷), participants from the parent study were included in this study if they were 14 to 15 years old at baseline and if they emerged into binge-level alcohol use by the time of data analyses. The forty participants (18 male) included in this study were part of a previous study examining the association between brain function and duration to onset of binge drinking in 47 adolescents (six did not undergo diffusion weighted imaging and one did not pass quality assessment¹⁰). Portions of this sample have been included in other published reports including somewhat overlapping samples, but distinct research questions.^18–20

After baseline demographic and brain imaging assessments, follow-up phone interviews were conducted every 3 months to capture emerging alcohol and substance use until participants turned 21. If participants missed a follow-up assessment, a longer version of the 90-day Timeline Followback²¹ was administered to capture any missing data. In the follow-up assessments leading up to the first instance of binge drinking, on average, participants missed 1.00 (2.59 SD) follow-up interviews, and 72% of participants completed all follow-up interviews on time.

2.2 |. Measures

Baseline assessments included the two-subtest form of the Wechsler Abbreviated Scale of Intelligence²² to assess intellectual functioning, the Hollingshead Index of Social Position²³ to measure socioeconomic status via parent self-report, and the Children’s Depression Inventory to measure self-reported symptoms of depression.²⁴ The Hollingshead Index of Social Position assesses socioeconomic status based on the education and occupation of the parent who earns the higher income. Education and occupation scores range from 1 to 7, with 1 specifying attainment of a professional degree or professional occupation and 7 specifying less than 7 years of education or unskilled work. To calculate a final score, occupation scores are multiplied by 7, and education scores are multiplied by 4, and then combined, resulting in a score ranging from 11 (higher socioeconomic status) to 77 (lower socioeconomic status). Raw scores on the Children’s Depression Inventory were converted to age-normed T-scores (no participants exceeded a cutoff score of 65, which differentiates youth with and without clinical levels of depressive symptoms). During follow-up interviews, the 90-day Timeline Followback²¹ was used to calculate number of months to the initiation of binge drinking. Binge drinking was defined as ≥ 5 drinks per occasion for males and ≥ 4 drinks per occasion for females, based on criteria established by the NIAAA.²⁵

2.3 |. Risky decision making

Participants completed the Wheel of Fortune (WOF) Task^26,27 during functional magnetic resonance imaging (MRI). Trials were 10.5 seconds long and consisted of selection (3 seconds), anticipation (3.5 seconds), and feedback (4 seconds) phases with jittered intertrial intervals (1–11 seconds). During the selection phase, participants had two choices: 10% probability of winning $7 versus 90% probability of winning $1 (high risk/reward), 30% probability of $2 versus 70% probability of winning $1 (moderate risk/reward), and 50/50 chance of winning $2 (equal risk/reward). After making a decision, participants indicated how sure they were of winning (anticipation phase), and they received feedback about whether they won or not (feedback phase). Participants received a portion of their total earnings. The scanning session consisted of two 10-minute runs, with a total of 24 high risk/reward trials, 28 moderate risk/reward trials, and 20 equal risk/reward trials.

2.4 |. Image acquisition

During baseline visits, diffusion weighted images (DWI) were acquired on a 3T Siemens Magnetom Tim Trio with a 12-channel head coil (TR = 9,100 ms, TE = 88 ms, slices = 72, slice thickness = 2 mm). Diffusion gradients were applied along 30 directions (b-value = 1000 s/mm²), and six images were collected with no diffusion weighting (b-value = 0 s/mm²). Participants received either three (n = 28) or two (n = 12) DWI runs. A diffusion field map was also acquired (TR = 790 ms, TE1 = 5.19 ms, TE2 = 7.65 ms, flip angle = 60°, slices = 72, slice thickness = 2 mm) to correct for eddy current-induced field distortions.

Blood-oxygen-level dependent (BOLD) signal was measured using T2* weighted gradient echo-planar images which were collected axially, parallel to the anterior-posterior commissure (TR = 2000 ms, TE = 30 ms, flip angle = 90°, voxel size = 3.75 × 3.75 × 3.8 mm). Anatomical, high-resolution T1-weighted MPRAGE structural scans were collected in the sagittal plane (TR = 2300 ms, TE = 3.58 ms, inversion time = 900 ms, flip angle = 10°, voxel size = 1 × 1 × 1.1 mm).

2.5 |. DWI processing

All DWI data were visually inspected for motion and scanner-related artifacts.²⁸ Volumes (ie, diffusion directions) containing artifacts were censored; however, participants were excluded if the same volume had to be removed from all of the imaging runs to avoid overestimation of diffusion metrics.²⁹ Preprocessing consisted of concatenation of all DWI runs within a scan; registration of the diffusion field map to the first volume³⁰; correction for eddy current distortion, intensity inhomogeneities, and head motion; and subsequent adjustment of the gradient table.³¹ FSL’s dtifit³² was used to calculate FA at each voxel. FA maps were registered to a study specific template and subsequently to Montreal Neurological Institute (MNI) space with a one-step nonlinear transformation process using Advanced Normalization Tools (ANTS).³³ Finally, a Gaussian blur (sigma = 1 mm) was applied to FA images.

BEDPOSTX was used to estimate the angular distribution of local tract direction with a two-fiber model using Markov Chain Monte Carlo sampling.³⁴ The NAcc, hippocampus, and amygdala were defined using the Harvard‐Oxford Atlas. Because the ventral tegmental area and ventral pallidum are not delineated in many commonly used brain atlases, they were defined using probabilistic midbrain³⁵ and reinforcement learning atlases,³⁶ respectively. Finally, the NAcc is structurally connected to various regions within the prefrontal cortex (ie, anterior, posterior and medial orbital gyrus, gyrus rectus, subcollosal gyrus, and ventral anterior cingulate).³⁷ We defined these regions using a mask created by Tziortzi and colleagues, which combined these regions into a single region of interest (ROI), referred to as the limbic prefrontal cortex. All ROIs were transformed from MNI space to native DWI space using the inverse of the one‐step registration estimated with ANTS. ProbtrackX was used to determine connectivity between the NAcc and all other ROIs. Each tract was delineated twice, once with the NAcc as the seed and one other brain region as the termination/waypoint and vice versa. For every voxel in the seed region, ProbtrackX was used to sample from the posterior distribution of fiber orientations, and 5000 streamlines were traced through local fiber samples. Streamlines entering the termination/waypoint mask were retained, and those entering the contralateral hemisphere or brain regions without connectivity to the NAcc (eg, occipital cortex) were excluded. All maps were thresholded at 0.02% of total number of streamlines³⁸ to remove unlikely paths, added together, binarized, transformed to MNI space using ANTS, and averaged across all subjects. Only white‐matter voxels where at least 50% of the participants had connectivity between the NAcc and the rest of the brain were included in the group‐level analyses (Figure 1).

FIGURE 1.

White-matter pathways connecting the nucleus accumbens to the rest of the brain. Maps show voxels where at least 50% of participants had connectivity between the nucleus accumbens and the prefrontal cortex, ventral pallidum, amygdala, hippocampus, or ventral tegmental area

2.6 |. Functional MRI processing

As described previously,¹⁰ BOLD images underwent visual inspection for artifacts, slice time correction, motion correction, nonlinear normalization to standard space via the MPRAGE using a one-step registration to minimize interpolation error, spatial smoothing using a 6-mm Gaussian kernel, and calculation of percent signal change using Analysis of Functional NeuroImages (AFNI³⁹). The onset time of selection, anticipation and feedback events, and event duration were convolved with a gamma-variate hemodynamic response function to create repressors for the first‐level statistical model. Additional regressors modeled linear drift, motion, and censored volumes with motion (framewise displacement > 0.7 mm, uncensored segments of data with < 5 contiguous frames). Finally, contrast images of average percent signal change between high and moderate levels of risk/reward during the selection phase were created, and signal was extracted from the NAcc using an ROI defined using the Harvard-Oxford atlas. This contrast was selected because brain activation during decision making under varying levels of risk/reward may be useful for understanding real-world engagement in risky behavior.

2.7 |. Group-level analyses

Path analyses were conducted voxel-wise in the mask of white-matter pathways connecting the NAcc to the rest of the brain using neuropointillist (http://github.com/IBIC/neuropointillist) in conjunction with the lavaan package⁴⁰ in R (v3.5.0). Because white-matter microstructure could theoretically impact BOLD response during the WOF task, we tested for direct effects of FA on duration to onset of binge drinking, as well as indirect effects through NAcc activation, in the high versus moderate risk/reward condition, with socioeconomic status as a covariate (Figure 2). The model was fit using maximum likelihood estimation with bootstrapped standard errors (k = 5000). To correct for multiple comparisons, a cluster-forming threshold of P < .05 was applied to direct and indirect effect maps, and an updated version of AFNI’s 3dClustsim, which simulates noise assuming that spatial auto-correlation is given by a mixed-model, was used to determine the number of voxels needed to achieve cluster-level significance at a two-sided threshold of P < .05. Mean FA was extracted from clusters where significant direct or indirect effects were detected. Using mean values, models were re-run in R, and standardized coefficients were generated (Table 1). Based on prior findings indicating that proportion of risky selections was associated with duration to onset of binge drinking, bivariate correlations were used to also examine the association between proportion of risky selections and white-matter microstructure. Furthermore, path analyses were conducted to examine whether task performance mediated the association between white-matter microstructure and duration to onset of binge drinking.

FIGURE 2.

Mediation model. The path diagram outlines the mediation model that was tested voxel-wise in white-matter pathways connecting the nucleus accumbens to the rest of the brain. Direct effects of fractional anisotropy (depicted by a map of mean fractional anisotropy in this sample) on duration to onset of binge drinking (histogram shows distribution of duration to onset of binge drinking by sex) were examined. Furthermore, the hypothesis that the association between fractional anisotropy and duration to onset of binge drinking was mediated by nucleus accumbens activation during decision making in the high versus moderate risk/reward condition was tested by examining the significance of the indirect effect

TABLE 1.

Location of significant direct and indirect effects

	MNI Coordinate			Standardized Coefficients (Standard Error)a
Location of White Matter Cluster	x	y	z	Path a	Path b	Direct c′	Indirect* (a × b)	Total* (a × b) + c
Right hemisphere proximal to the ventral pallidum	10	1	−14	−0.17 (0.16)	−0.47 (0.11)	0.43*(0.12)	0.08 (0.08)	0.51 (0.10)
Left hemisphere proximal to the ventral pallidum	−24	−1	−9	−0.10 (0.17)	−0.51 (0.11)	0.40* (0.12)	0.05 (0.09)	0.45 (0.11)
Left hemisphere proximal to the medial orbital gyrus	−14	−38	−19	−0.59 (0.10)	−0.56 (0.15)	0.02 (0.16)	0.33* (0.10)	0.31 (0.15)
Right hemisphere proximal to the medial orbital gyrus	15	−43	−7	−0.56 (0.10)	−0.53 (0.18)	0.03 (0.17)	0.30* (0.10)	0.33 (0.13)
Right hemisphere proximal medial temporal lobe	33	16	−7	−0.52 (0.10)	−0.53 (0.15)	0.04 (0.19)	0.28* (0.10)	0.31 (0.16)
Left hemisphere anterior corona radiata	−22	−23	10	−0.59 (0.12)	−0.56 (0.19)	−0.03 (0.21)	0.33*(0.12)	0.30 (0.16)

^aStatistical significance of direct and indirect effects was evaluated voxel-wise. Values presented in the table were calculated post hoc using mean FA from significant clusters and represent circular analyses. Therefore, significance of direct effect a, direct effect b, and the total effect are not presented, and coefficients are included only for descriptive purposes.

^*P < .05 corrected for multiple comparisons in voxel-wise analysis.

3 |. RESULTS

3.1 |. Participant characteristics

On average, adolescents were 15.07 (SD = 0.59) years old, had IQ scores of 111.66 (SD = 9.50), scored 27.63 (SD = 13.25) on the Hollingshead Index of Social Position, had T-scores of 43.10 (SD = 6.01) on the Children’s Depression Inventory, and began binge drinking 29.63 (SD = 19.46) months after baseline assessments. Furthermore, 22 of the participants were female, and 18 were male. Duration to onset of binge drinking was not related to any demographic variables (P > .15), except the Hollingshead Index of Social Position (r = 0.33, P < .05). Consistent with the exclusion criteria, at baseline, participants used cigarettes 0.05 (SD = 0.32) times, used alcohol 0.30 (SD = 1.0) times, and used marijuana 0.63 (SD = 1.84) times. Analyses were also conducted in the 29 participants who were completely alcohol and drug naïve at baseline.

3.2 |. Associations between FA, NAcc function, and duration to onset of binge drinking

There was a direct effect of FA on duration to onset of binge drinking in bilateral white matter proximal to the ventral pallidum (P < .05 corrected, Figure 3). Adolescents with lower FA in white matter proximal the ventral pallidum began binge drinking sooner. An indirect effect of FA on duration to onset of binge drinking, mediated by NAcc activation in the high versus moderate risk/reward condition, was observed in white matter proximal to the bilateral medial orbital gyrus, left medial temporal lobe, and the corona radiata (P < .05 corrected, Figure 4). These indirect effects were driven by the negative association between FA and NAcc activation in the high versus moderate risk/reward condition. As previously reported,¹⁰ adolescents with greater NAcc activation in the high versus moderate risk/reward condition began binge drinking sooner. In adolescents that were completely drug and alcohol naïve at baseline, direct effects were significant in white matter near the right ventral pallidum and indirect effects were significant bilaterally in white matter near the medial orbital gyrus (P < .05 corrected).

FIGURE 3.

Direct effects of fractional anisotropy on duration to onset of binge drinking. Statistical maps display regions where there was a direct effect of fractional anisotropy on duration to onset of binge drinking (P < .05 corrected). The scatterplot illustrates that adolescents with lower fractional anisotropy in white matter near the right ventral pallidum began binge drinking sooner. A similar effect was observed in white matter near the left ventral pallidum

FIGURE 4.

Indirect effects of fractional anisotropy on duration to onset of binge drinking. Statistical maps display regions where there was an indirect effect of fractional anisotropy on duration to onset of binge drinking (A, P < .05 corrected). The scatterplot between fractional anisotropy in white matter proximal to the left medial orbital gyrus and nucleus accumbens activation during decision making in the high versus moderate risk/reward condition illustrates that adolescents with lower fractional anisotropy have greater nucleus accumbens activation during decision making involving high versus moderate levels of risk and reward (B). Scatterplots also demonstrate that individuals with lower fractional anisotropy and greater nucleus activation during decision making involving high versus moderate levels of risk and reward began binge drinking sooner (B). A similar effect was observed in other clusters with a significant indirect effects

3.3 |. Associations between FA and proportion of risky selections

Proportion of risky selections was negatively correlated with FA in clusters located proximal to the ventral pallidum in the right (r = −0.45, P = .003) and left (r = −0.42, P = .006) hemisphere (Figure 5). Adolescents who made more risky choices had lower FA in these clusters. In path analyses examining whether proportion of risky selections mediated the association between FA in these clusters and duration to onset of binge drinking, the indirect effect was not significant (P’s > .5). No other significant associations between risky selections and FA were detected (P’s > .5).

FIGURE 5.

Associations between fractional anisotropy and decision making behavior. Adolescents with lower fractional anisotropy near the right (r = −0.45, P = .003) and left(r = −0.42, P = .006) ventral pallidum made a greater proportion of risky selections

4 |. DISCUSSION

This work extends previous findings indicating that adolescent binge drinkers have abnormalities in white-matter microstructure,¹² by demonstrating that premorbid differences in FA are also associated with duration to onset of binge drinking. Findings suggest convergence between structural and functional MRI, as adolescents with lower medial orbital gyrus FA had greater NAcc activation during decision making involving risk and reward and began binge drinking sooner. However, the results also highlight the utility of a multimodal neuroimaging approach for identifying risk factors for future substance use, as a direct effect of FA on duration to onset of binge drinking was observed in white matter near the ventral pallidum, independent of the effect of NAcc function on duration to onset of binge drinking.

Adolescents with lower FA in medial orbital gyrus white matter displayed greater activation in NAcc during decision making involving risk and reward, suggesting that top-down projections from the medial PFC are important in regulating reward responses in the striatum. Support for this hypothesis comes from a study in rodents which demonstrated that optogenetic stimulation of dopamine neurons increased BOLD signal in the NAcc and that this effect was suppressed by activation of neurons in the medial PFC.⁴¹ Similarly, a study measuring functional connectivity in individuals with varying levels of psychopathy demonstrated that those with stronger NAcc activation in response to increasing subjective reward value had weaker functional connectivity between medial PFC and NAcc.⁴² Lower FA in white-matter pathways connecting the medial PFC to the NAcc has also been observed in young adult smokers,⁴³ suggesting the strength of cortico-striatal connectivity and its impact on reward processing may present a risk factor for various forms of psychopathology and substance use.

We also found a direct effect of FA on duration to onset of binge drinking in the white matter proximal to ventral pallidum, a region involved in both “liking” and “wanting” of rewarding stimuli.⁴⁴ Consistent with the hypothesis that the ventral pallidum plays a role in substance use disorders, microinjections of GABA_A receptor agonists or inhibition of GABA_A α1 receptor subunit expression in the ventral pallidum decrease ethanol intake.^45,46 To better understand how FA in these regions impacts propensity for binge drinking, future research could examine how FA in these regions relates to subjective response to alcohol and the motivation to obtain alcohol.

Our prior work in this sample demonstrated that although proportion of risky selections and NAcc activation during decision making involving high versus moderate levels of risk and reward were not correlated, both were significantly associated with duration to onset of binge drinking.¹⁰ Consistent with prior studies showing that structural connectivity is associated with decision making,^13,47 we extend our prior work by demonstrating that adolescents with lower FA in white matter near the ventral pallidum are more likely to choose options associated with higher potential for risk and reward. More research is needed to determine how FA in the white-matter clusters near the ventral pallidum is linked to task-evoked and resting-state brain function, as we did not detect significant associations between FA and decision making in white-matter regions correlated with NAcc activation.

Although this prospective cohort study has several strengths, including the ability to examine the adolescent brain prior to substantive alcohol and drug use, some limitations warrant mentioning. The voxel-wise implementation of the mediation analysis allowed for more precise spatial localization of the direct and indirect effects of FA on duration to onset of binge drinking than an ROI approach; however, there are strengths associated with an ROI approach, such as the ability to conduct analyses in native space thereby minimizing registration error. Mediation models do not prove causation, and the interpretation of associations between white-matter microstructure, brain function, and duration to onset of binge drinking is limited by the correlational nature of the analysis. Limitations inherent to DWI, such as the inability to differentiate afferent and efferent pathways and the heterogeneity in underlying biological features that might impact FA (eg, axon density, myelination⁴⁸), limit interpretation of results. Finally, longitudinal studies examining differences in baseline FA between adolescent drinkers and control participants have produced mixed results.^20,49 Because earlier alcohol use is associated with greater risk for negative consequences later in life and because experimenting with binge drinking during adolescence and young adulthood is common, the use of a continuous outcome measure in this study may have provided greater sensitivity for identifying associations between FA and future alcohol use. However, our sample is relatively small, and replication in independent samples would support the veracity of the effect, provide a more reliable estimate of the effect size, and allow for examination of sex differences.

This study provides novel information indicating that premorbid individual differences in white-matter microstructure during adolescence are associated with decision making, NAcc sensitivity to risk and reward, and future alcohol use. Developing a better understanding of both the genetic and environmental factors that influence white-matter integrity may be useful for identifying adolescents at greatest risk for heavy alcohol use. Furthermore, interventions that strengthen implicated neural circuitry may be more effective at promoting adaptive decision making and curbing heavy alcohol use.