Adolescent substance use trajectories are associated with nucleus accumbens functional connectivity

Abstract

Background:

Adolescent substance use is associated with disrupted communication among brain regions underlying reward-driven behaviors (e.g., nucleus accumbens [NAcc]) and cognitive/emotional control (e.g., prefrontal cortex [PFC]; medial temporal lobe), which may be linked to poor future outcomes (e.g., substance use disorder). However, the relationship functional brain connectivity has with trajectories of adolescent alcohol, tobacco, and cannabis use has received limited attention.

Aims:

Investigate relationships between adolescent substance use trajectories and young adult whole-brain NAcc resting-state functional connectivity (rsFC).

Methods:

Substance use was assessed at ages 11, 13, 16, and 19. Subsequently (age 20), a subset of participants (N=299) completed a single neuroimaging session. Latent growth curve models estimated substance use trajectories that included the: intercept (age 14 use), linear slope (progression), and quadratic slope (acceleration), which served as predictors in neuroimaging analyses.

Hypotheses:

Substance use trajectories representing greater age 14 usage, faster progression of use, and acceleration of use across adolescence would show stronger NAcc rsFC with regions implicated in cognitive/emotional control.

Results:

Age 14 use (alcohol, tobacco, and cannabis) was associated with NAcc rsFC with dorsolateral PFC, dorsomedial PFC, parahippocampal gyrus (PHG), hippocampus, and amygdala. Progression of use was associated with NAcc rsFC with dorsolateral PFC, dorsomedial PFC, ventrolateral PFC, PHG, and amygdala. Finally, acceleration of use was linked with NAcc rsFC with dorsolateral PFC, ventromedial PFC, PHG, and hippocampus.

Conclusions:

NAcc rsFC with several brain regions (e.g., PFC subregions) varied with adolescent substance use, which may represent common neural mechanisms linking adolescent substance use with common psychological outcomes.

Introduction

Adolescence is a period of neurobehavioral development characterized by an increase in risk taking behaviors as the brain continues to mature (Casey et al., 2008; Casey and Jones, 2010; Crews et al., 2007; Squeglia and Gray, 2016). Substance use (most commonly alcohol, tobacco, and cannabis) is a risk taking behavior that often emerges during adolescence (Casey et al., 2008; Casey and Jones, 2010; Johnston et al., 2019; Kann et al., 2018; Miech et al., 2019). Substance use-related alterations in brain function may be potentiated during adolescence due to ongoing brain maturation during this developmental window. Adolescent substance use is highly prevalent in the United States. In fact, by 12^th grade, 60 to 70% of adolescents have used alcohol, 20 to 40% have used tobacco, and 30 to 45% have used cannabis (Johnston et al., 2019; Kann et al., 2018; Miech et al., 2019). Adolescent substance use has been linked with negative outcomes later in life, including increased risk of psychopathology and substance use disorder (SUD), as well as poor educational and occupational achievement (Braun et al., 2000; Grant et al., 2004; Han et al., 2017; Schuster et al., 2001; Volkow et al., 2014). Prior work suggests that substance use may alter brain function and connectivity (Khalili-Mahani et al., 2012; Lee and Telzer, 2016; Shen et al., 2016; Weissman et al., 2015). In particular, the nucleus accumbens (NAcc) has emerged as an important structure underlying many aspects of reward-driven and other motivated behaviors (Chang et al., 1994; Gregorios-Pippas et al., 2009; Hare et al., 2008). However, relatively little is known about the relationships that the pattern of substance use across adolescence (i.e., trajectories) has with the functional connectivity of the NAcc in young adulthood. Given that the functional connectivity of neural networks can predict emotional, cognitive, and clinical outcomes (Fox et al., 2007; Greicius et al., 2003; Khalili-Mahani et al., 2012; Seeley et al., 2007), identifying relationships between adolescent substance use trajectories and NAcc functional connectivity in young adulthood is an important area of investigation.

The functional connectivity of the NAcc has been implicated in the initiation and maintenance of substance use due, in part, to its central role in reward-related networks (Janes et al., 2012; Lichenstein et al., 2017; Motzkin et al., 2014). Specifically, the NAcc is the hub of the reward network and is connected to several brain regions that support executive function, including the ventromedial, ventrolateral, dorsomedial, and dorsolateral prefrontal cortex (PFC) (Janes et al., 2012; Lee and Telzer, 2016; Manning et al., 2015; Motzkin et al., 2014; Posner et al., 2016). These regions of the PFC operate in concert to subserve executive function via reciprocal connections that facilitate the flow of information (Goodman et al., 2018; Seeley et al., 2007). However, there is also functional specificity among these brain regions. For example, these subregions of the PFC support working memory (dorsolateral PFC), attention to salient stimuli (dorsomedial PFC), response suppression (ventrolateral PFC), and emotion regulation (ventromedial PFC) processes (Cho et al., 2010; Goodman et al., 2018; Seeley et al., 2007). These cognitive processes are also related to reward processing and substance use behaviors. In fact, as part of the executive control network, the PFC provides top-down modulation of other brain regions (e.g., NAcc, hippocampus, amygdala) that support reward processing (Koehler et al., 2013; Kravitz et al., 2015; Lee and Telzer, 2016; Martz et al., 2019; Watt et al., 2017). Indeed, prior work has found that risk taking behaviors (e.g., substance use), which are common during adolescence, may be linked with developmental changes in the connectivity between the NAcc (i.e., bottom-up reward driven) and PFC (i.e., top-down regulation) (Casey, 2015; Casey et al., 2016). Thus, investigating the resting state functional connectivity (rsFC) between these brain regions may further our understanding of the relationships between neurobiological processes and adolescent substance use. Ultimately, a better understanding of the neurobiology associated with substance use may support future efforts to develop novel intervention targets.

Prior neuroimaging work examining substance use and NAcc rsFC has produced mixed results, due, in part, to variability in participant characteristics and the use of cross-sectional substance use data. For example, many neuroimaging studies have been conducted with clinical samples (e.g., diagnosed substance use disorder vs. those who do not use), emphasized dichotomous participant groupings (e.g., heavy vs. no use), and relied on cross-sectional substance use data (Bracht et al., 2021; Gu et al., 2010; Ma et al., 2010, 2010; Upadhyay et al., 2010; Wilcox et al., 2011). Similarly, neuroimaging studies of adolescents have relied on either cross-sectional substance use data (e.g., Arienzo et al., 2020; Cservenka et al., 2014) or selected samples (e.g., exclusively low SES participants) (Lichenstein et al., 2017; Weissman et al., 2015). As a result, the relationship between adolescent substance use and NAcc rsFC remains unclear. General patterns among results in the extant literature are not apparent even between studies of the same substance and participants with similar characteristics (e.g., adults with heavy alcohol use: Bracht et al., 2021; Yang et al., 2021). Thus, although there is some consensus in the literature that the functional connectivity of the NAcc and PFC may be linked with substance use, the direction of these relationships remains unclear. Investigations with larger community samples of adolescents, such as the current study, are necessary to maximize generalizability and may help to contextualize prior work. Therefore, the present investigation utilized substance use data reported across adolescence as part of a longitudinal study of health risk behaviors.

The present study focused on alcohol, tobacco, and cannabis use as they are the most commonly used substances during adolescence (Johnston et al., 2019; Kann et al., 2018; Miech et al., 2019). Further, the current work focused on trajectories of adolescent substance use in a community sample, which is an approach that has received very little attention to date. Therefore, hypotheses for this investigation were broadly formulated to encompass each of the three growth parameters investigated given the variability of findings in prior work and the paucity of neuroimaging research investigating substance use trajectories in adolescents. We hypothesized that trajectories of alcohol, tobacco, and cannabis use that represent greater use at age 14 (i.e., intercept), faster progression of use (i.e., linear slope), and acceleration of use across adolescence (i.e., quadratic slope) would be linked with stronger rsFC between the NAcc and regions implicated in emotion regulation and cognitive control (e.g., dorsolateral PFC, dorsomedial PFC, ventromedial PFC, parahippocampal gyrus, hippocampus, and amygdala). Understanding the relationships that adolescent alcohol, tobacco, and cannabis use trajectories have with NAcc rsFC may elucidate the neural mechanisms by which adolescent substance use results in negative outcomes later in life.

Method

Participants

Participants were recruited from the Birmingham site of the Healthy Passages study, a large, multi-site longitudinal study of adolescent health-risk behaviors (Schuster et al., 2012; Windle et al., 2004). Participants were recruited from the 5^th-grade classrooms of local public-school districts and were interviewed at three time points (Wave 1 M_age = 11.2, SD = 0.5; Wave 2 M_age = 13.0, SD = 0.5; Wave 3 M_age = 16.2, SD = 0.5). An additional Wave 4 interview was conducted with the Birmingham cohort only, as part of the current study (Wave 4 M_age = 19.2, SD = 1.2; see Table 1 for additional sample information). Following completion of the Wave 4 interview, participants were screened for eligibility and invited to return for a single MRI session (N=350; M_age = 20.1, SD = 1.6). MRI exclusion criteria included: left-handedness, history of blood or circulation disorders (e.g., anemia or sickle-cell), diabetes, brain or spinal abnormalities, pregnancy, previous or current head injury (e.g., traumatic brain injury), and history of psychosis. Fifty-one participants were excluded from the MRI analyses due to poor data quality (e.g., excessive motion, image artifacts) and incomplete data (e.g., participant dropout, data loss). The final MRI sample consisted of 299 participants (M_age = 20.1, SD = 1.6; 103 White participants, 196 Black participants; 142 female, 157 male).

Table 1.

Summary substance use information for the present sample

	Wave 1 (N = 1593)
	Alcohol		Tobacco		Cannabis
Mean (SD)	0.10 (0.461)		0.10 (0.397)		0.04 (0.294)
Median	0		0		0
Mode	0		0		0
	Wave 2 (N = 1535)
	Alcohol		Tobacco		Cannabis
Mean (SD)	0.16 (0.580)		0.15 (0.70)		0.07 (0.470)
Median	0		0		0
Mode	0		0		0
	Wave 3 (N = 1374)
	Alcohol		Tobacco		Cannabis
Mean (SD)	0.80 (1.26)		0.71 (1.65)		0.64 (1.51)
Median	0		0		0
Mode	0		0		0
	Wave 4 (N = 1264)
	Alcohol		Tobacco		Cannabis
Mean (SD)	1.91 (1.71)		1.38 (2.27)		1.17 (1.95)
Median	2		0		0
Mode	0		0		0

Note: Substance use data above reflect the mean (standard deviation), median, and mode of the 8-category substance use variable for the entire sample of participants. These 8-category variables were computed for each substance at each wave of data collection and served as inputs to latent growth curve models, which were run separately for each substance.

Twenty-two percent of the initial Birmingham cohort of the Healthy Passages sample participated in the neuroimaging session. Therefore, differences in key variables were assessed (i.e., race, sex, and substance use) between those who completed neuroimaging and those who did not to determine whether the characteristics of the current sample were consistent with those of the Birmingham cohort of the Healthy Passages study. No difference was observed in the proportion of males and females [χ²(1) = 0.131, p > 0.05]. Further, alcohol [χ²(1) = 2.26, p > 0.05], tobacco [χ²(1) = 0.646, p > 0.05], cannabis [χ²(1) = 0.081, p > 0.05], and other drug use [χ²(1) = 0.230, p > 0.05] did not differ between participants who completed neuroimaging and those who did not. However, a larger proportion of Black participants completed neuroimaging than did not [χ²(1) = 12.20, p < 0.001]. Additional comparisons (e.g., comparisons across scanner) can be found in the Supplementary Methods. The original Healthy Passages study (Waves 1–3) was approved by the Centers for Disease Control and Prevention and the institutional review boards of all original site institutions. Informed consent or assent was obtained from all participants at each wave of data collection, and parental consent was obtained for all participants who were younger than 18 years old. All participants in the present study (Wave 4) provided written informed consent and all study procedures were approved by the University of Alabama at Birmingham Institutional Review Board.

Alcohol, Tobacco, Cannabis, and Other Drug Use

Participants reported their use of alcohol, tobacco, and cannabis at each of the four waves of data collection using a computer assisted self-interview (CASI). During this portion of the interview, participants were alone in the room and followed prompts on the computer screen, rather than being interviewed in person. The CASI was used to maximize participants’ comfort and reduce hesitancy to report potentially sensitive information, such as substance use. Participants reported whether they used each substance during the last 12-month period (yes/no), as well as how frequently they used each substance during the last 30 days (ranging from 1 = used on zero days to 7 = used on all 30 days). These two items (i.e., past 12-month use and past 30-day use) were recoded into a single variable that encompassed both dichotomous past-year use and frequency of past-month use. This new variable had 8 levels: 0 = no last year use, 1 = reported last-year use, but no last-month use, 2 = used 1 or 2 days in the last month, 3 = used 3 to 5 days in the last month, 4 = used 6 to 9 days in the last month, 5 = used 10 to 19 days in the last month, 6 = used 20 to 29 days in the last month, and 7 = used all 30 days in the last month. This approach was used to create separate variables for alcohol, tobacco, and cannabis use at each time point. These recoded substance use variables were included in latent growth curve models (LGCMs; described below) to characterize each participant’s individual trajectory of use across adolescence for each substance. Rates of substance use in the present sample approximated rates observed in large, nationally-representative community samples of adolescents (Johnston et al., 2019; Kann et al., 2018; Miech et al., 2019). Specifically, approximately 70% of adolescents had used alcohol, 35% had used tobacco, and 40% had used cannabis by age 19 (i.e., Wave 4 of data collection). The average use of each substance at each wave of data collection in this community sample can be found in Table 1. A detailed breakdown of the reported use of each substance at each wave of data collection can be found in the supplementary files of our previously published work (Purcell et al., 2024).

Given that many adolescents use more than one substance (Tomczyk et al., 2016), additional variables were created to account for potential co-use of alcohol, tobacco, and cannabis, consistent with our prior work (Purcell et al., 2024). As described above, participants reported whether they had used alcohol, tobacco, and cannabis in the last year (yes/no) during each wave of data collection. A sum score was created by adding the number of waves in which participants reported they had used the substance. Sum scores were computed separately for each substance (i.e., alcohol, tobacco, and cannabis), resulting in three variables ranging from 0 (no reported use of the substance at any wave) to 4 (reported use of the substance at all four waves).

While the most commonly used substances during adolescence are alcohol, tobacco, and cannabis, they are not the only substances used (Johnston et al., 2019; Kann et al., 2018). Therefore, an additional variable was created to account for lifetime “other” drug use. During each wave, participants reported (yes/no) whether they had used illicit drugs or medications without a prescription “to get high or feel good” (e.g., inhalants, Ritalin, painkiller medicines, Xanax, Ecstasy, cocaine, heroin). The number of waves in which participants reported that they had used any “other” substances was summed, resulting in a single variable of “other” drug use. This variable ranged from 0 (no reported use of other drugs at any wave) to 4 (reported use of other drugs at all four waves).

Pubertal Status

During the CASI portion of the interview of the second wave of data collection (age 13), participants reported their perceived pubertal status (i.e., Tanner stage) (Dudovitz et al., 2015; Taylor et al., 2001). Participants were shown illustrations depicting Tanner stages for each of two physical features: pubic hair and breast development for girls and pubic hair and genital development for boys. Participants were then asked to select a single picture of each physical feature that most closely resembled their current stage of physical development. Possible responses ranged from 1 (pre-pubescent) to 5 (fully mature) and were then averaged across the two questions (i.e., pubic hair and breast/genital development) to produce an overall perceived pubertal status index. The average perceived pubertal status at age 13 for the present sample was 3.34 (SD = 0.83; range: 1.5 – 5).

Household Income

Participants’ caregivers reported their household income at Waves 1, 2, and 3. These values were transformed into a percentage of the federal poverty line, accounting for the size of the household. This transformed value was then averaged across all three waves to obtain an overall measure of household income across adolescence. Average household income for participants in this study was 274.47% (SD = 247.84; range = 0 – 1245.67%) above the federal poverty line.

Anatomical and Functional MRI

Data were acquired on 3T Siemens Allegra (n = 245) and Prisma (n = 54) scanner systems. High-resolution anatomical images (MPRAGE) were acquired to serve as an anatomical reference (T1-weighted, TR = 2300 ms, TE = 3.9 ms [Allegra], TE = 2.98 ms [Prisma], FOV = 25.6 cm, matrix = 256 × 256, 160 slices, slice thickness = 1 mm, 0.5mm gap). Blood oxygen level dependent (BOLD) fMRI data were acquired using a gradient echo echo-planar pulse sequence (TR = 2000 ms, TE = 30 ms, flip angle = 70°, FOV = 24 cm, matrix = 64 × 64, slice thickness = 4 mm, no gap). All MRI data were processed using the AFNI software package (Cox, 1996). Preprocessing steps included slice timing correction, extraction of tissue-based regressors (including cerebrospinal fluid [CSF] and white matter [WM]), and spatial smoothing using a 4mm FWHM kernel. First-level analyses accounted for variables of no interest, including average CSF time course, average WM time course, six motion parameters, six motion derivatives, and 111 bandpass time courses (bandpass filter: 0.01 < f > 0.1 Hz). These variables were regressed out of the gray matter time course for each individual. High motion TRs were defined as volumes in which greater than three percent of the voxels deviated by more than five times the median absolute deviation of the time series. These high motion TRs were excluded from individual participant data, consistent with prior work (Dark et al., 2020; Wood et al., 2015). Excluded volumes were ignored in subsequent statistical analyses. Participants with less than 80% usable TRs were excluded from data analysis. The functional dataset was normalized to the Montreal Neurological Institute (MNI) 152 template. Anatomically based seed regions for the left and right NAcc were created for each participant using to the MNI 152 atlas included in the mask creation tools within AFNI, resulting in average time courses for both the left and right NAcc. Correlation analyses were completed that compared the time courses for left and right NAcc, separately, to all other voxels in the brain. These Pearson correlation values were then converted into Fisher’s Z values to normalize the distribution for each participant. Finally, Fisher’s Z maps were resampled to 1mm isotropic voxels.

Data Analysis

Latent growth curve models.

Trajectories of alcohol, tobacco, and cannabis use across adolescence (i.e., between the ages of 11 and 19) were characterized using latent growth curve models (LGCMs), consistent with our prior work (Purcell et al., 2024). Briefly, LGCMs were run separately using Mplus version 8 (Muthén and Muthén, 2017) and included the full Birmingham cohort of the Healthy Passages sample (N=1,594) because larger LGCM sample sizes generally result in more accurate model fit and parameter estimation. Alcohol, Tobacco, Cannabis, and Other Drug Use across the 4 waves of data collection (described above) were input into LGCMs to estimate trajectories of substance use across adolescence. No other covariates were included in LGCMs. Robust maximum likelihood estimation was used due to non-normal distributions of the substance use variables. Full Information Maximum Likelihood (FIML) was used to handle missing data. FIML uses all available observations to generate model-specific parameter estimates, thus preserving the full sample size and reducing bias compared to other approaches to missing data (Enders, 2001, 2010).

The best fitting functional form of the growth model was determined separately for each substance (i.e., alcohol, tobacco, and cannabis) by comparing LGCMs that included only linear vs. both linear and quadratic growth components using Satorra-Bentler chi-square difference tests (Satorra and Bentler, 2010). Most participants reported no substance use at the first two waves of data collection (i.e., ages 11 and 13), which resulted in very little variability in substance use at the first two waves. Therefore, time was centered at age 14, rather than at age 11, to ensure variability in intercept estimates. Individual growth parameters (i.e., intercept, linear slope, and quadratic slope) of the best fitting functional form were output from each growth model (see LGCM results below and Purcell et al., 2024). These individual growth parameters were used in neuroimaging analyses in AFNI.

In the present study, the three growth parameters (i.e., age 14 use, progression of use, and acceleration of use) were conceptualized as representations of separate aspects of an individual’s overall pattern of substance use across adolescence. For example, age 14 use (i.e., intercept) represents “initial” substance use because time was centered at age 14 in LGCMs (described above). The progression of substance use across adolescence (i.e., linear slope) represents the linear change in substance use (i.e., increasing or decreasing) across adolescence. The acceleration of substance use across adolescence (i.e., quadratic slope) represents the rate of change in the progression of substance use across adolescence (i.e., acceleration or deceleration). A positive quadratic slope represents a U-shape growth curve, which reflects acceleration of substance use across adolescence. In contrast, a negative quadratic slope corresponds to an inverted U-shaped growth curve, which reflects deceleration of substance use across adolescence.

Neuroimaging.

Neuroimaging analyses were completed using multivariate modeling (3dMVM) in AFNI. Analyses were run separately for each substance (alcohol, tobacco, and cannabis) and hemisphere of the NAcc (i.e., left and right) for a total of 6 models. Individual estimates of the growth parameters from the LGCMs (i.e., intercept, linear slope, and quadratic slope) were included as variables of interest in all models to determine whether rsFC varied uniquely with any of these parameters. In addition, the following variables were included as covariates in each analysis: race, sex, pubertal status, household income, scanner (Allegra or Prisma), sum of waves of other drug use, and an index of substance co-use. Analyses using 3dMVM in AFNI are not able to accommodate missing data. Therefore, missing data (e.g., household income [n = 50] and pubertal status [n = 10]) were imputed in SAS (version 9.4, PROC MI). Imputed values were then output and used in neuroimaging analyses. The analyses were restricted to a whole-brain gray matter mask to reduce the number of voxel-wise comparisons. Consistent with prior work (Dark et al., 2020; Purcell et al., 2024), the gray matter mask was created from the average of the anatomical images of the participants in the present study (3dMean in AFNI). The averaged image was segmented into gray matter, white matter, and cerebrospinal fluid (3dSeg in AFNI). The segmented gray matter was used to mask data in subsequent analyses and included a total of 989,403 1mm³ voxels. A family-wise error (FWE) corrected, voxel-wise threshold of p < 0.05 was used to reduce Type I error. Monte Carlo simulations (AFNI: 3dClustSim with -acf option; uncorrected significance threshold p < 0.005) were used to determine the corrected significance threshold. Smoothness was averaged across subjects based on spherical autocorrelation function (ACF) parameters (3dFWHMx) derived from residual volumes from the first-level analysis. Results of this simulation yielded a critical cluster extent volume threshold of 517mm³. Additional Monte Carlo simulations were completed for the hippocampus/parahippocampal gyrus (PHG) and the amygdala using a small volume correction, based on a priori hypotheses for these regions (185mm³ and 84mm³, respectively). Results of the Monte Carlo simulations corresponding to AFNI clusterize options for nearest neighbor 1 and two-sided criteria were used to determine cluster volume thresholds for each simulation.

Results

Latent Growth Curve Models

Results pertaining to LGCMs have been reported previously (Purcell et al., 2024). Briefly, chi-square difference tests indicated that models that included terms for both linear and quadratic slope fit the data better than models that only included a term for the linear slope (alcohol χ²(4) = 12.41, p = 0.006; tobacco χ²(4) = 50.71, p < 0.001; cannabis χ²(4) = 37.16, p < 0.001). Therefore, intercepts, linear slopes, and quadratic slopes from the LGCM were included in the neuroimaging analyses for alcohol, tobacco, and cannabis. The model fit indices and chi-square difference tests have been reported previously (Purcell et al., 2024). Correlations among growth parameters can be found in Supplementary Table S1.

Neuroimaging

The neuroimaging models yielded a large number of results, which can be found in the supplementary material (Supplementary Tables S2-S13). Results presented in the following sections are focused on the primary aims and hypotheses of the present study (i.e., investigating relationships among alcohol, tobacco, and cannabis use trajectories and rsFC in regions of interest) and are included in the Discussion. A summary of these selected results can be found in Tables 2–4.

Table 2.

Summary of selected alcohol trajectory results

			Age 14 Alcohol Use
			Left NAcc						Right NAcc
	Region		Hemisphere		Direction		Region		Hemisphere		Direction
	dlPFC		Right		Negative		dlPFC		Right		Negative
							PHG		Left		Positive
			Progression of Alcohol Use
			Left NAcc						Right NAcc
	Region		Hemisphere		Direction		Region		Hemisphere		Direction
	dlPFC		Right		Negative		dlPFC		Right		Positive
	vlPFC		Right		Positive
	vlPFC		Left		Positive
	dmPFC		Left		Positive
	dmPFC		Left		Positive
			Acceleration of Alcohol Use
			Left NAcc						Right NAcc
	Region		Hemisphere		Direction		Region		Hemisphere		Direction
	dlPFC		Right		Positive		dlPFC		Right		Positive
							PHG		Right		Negative
							PHG		Left		Negative

Note: Summary of selected results of models examining the relationship between alcohol trajectories and nucleus accumbens (NAcc) resting state functional connectivity. Full results can be found in Supplementary Tables S2-S3. The left and right NAcc were used as seed regions in whole-brain functional connectivity analyses. Region and Hemisphere refer to the brain region to which the seed was connected, and Direction refers to the directionality of the relationship between the NAcc and the region to which it connected (i.e., either positive or negative). dmPFC = dorsomedial prefrontal cortex; dlPFC = dorsolateral prefrontal cortex; vlPFC = ventrolateral prefrontal cortex; NAcc = nucleus accumbens; PHG = parahippocampal gyrus.

Table 4.

Summary of selected cannabis trajectory results

			Age 14 Cannabis Use
			Left NAcc						Right NAcc
	Region		Hemisphere		Direction		Region		Hemisphere		Direction
	dlPFC		Left		Negative		dlPFC		Right		Negative
	dmPFC		Right		Positive		dlPFC		Left		Negative
	dmPFC		Left		Positive		dlPFC		Left		Negative
							dmPFC		Right		Positive
							dmPFC		Left		Negative
							Hippocampus		Right		Positive
							Hippocampus		Left		Positive
							Hippocampus		Left		Positive
			Progression of Cannabis Use
			Left NAcc						Right NAcc
	Region		Hemisphere		Direction		Region		Hemisphere		Direction
							dlPFC		Left		Negative
			Acceleration of Cannabis Use
			Left NAcc						Right NAcc
	Region		Hemisphere		Direction		Region		Hemisphere		Direction
	dlPFC		Left		Negative		dlPFC		Left		Negative
	vmPFC		Left		Positive		dmPFC		Right		Positive
	PHG		Left		Positive		PHG		Left		Positive
							Hippocampus		Right		Positive

Note: Summary of selected results from models examining the relationship between cannabis trajectories and nucleus accumbens (NAcc) resting state functional connectivity. Full results can be found in Supplementary Tables S6-S7. The left and right NAcc were used as seed regions in whole-brain functional connectivity analyses. Region and Hemisphere refer to the brain region to which the seed was connected, and Direction refers to the directionality of the relationship between the NAcc and the region to which it connected (i.e., either positive or negative). dmPFC = dorsomedial prefrontal cortex; dlPFC = dorsolateral prefrontal cortex; vmPFC = ventromedial prefrontal cortex; NAcc = nucleus accumbens; PHG = parahippocampal gyrus.

Alcohol – Age 14 Use (Intercept)

Age 14 alcohol use (i.e., intercept) was associated with rsFC of the NAcc and the dorsolateral PFC and the parahippocampal gyrus (Table 2; Supplementary Tables S2-S3). Specifically, age 14 alcohol use was negatively linked with rsFC of the bilateral NAcc and the right dorsolateral PFC such that greater age 14 alcohol use was linked with weaker NAcc-dorsolateral PFC rsFC. In contrast, greater alcohol use at age 14 was associated with stronger rsFC of the right NAcc and the left parahippocampal gyrus.

Alcohol – Progression of Use (Linear Slope)

Progression of alcohol use across adolescence (i.e., linear slope) was associated with rsFC of the NAcc and the dorsolateral PFC, ventrolateral PFC, and dorsomedial PFC (Table 2; Supplementary Tables S2-S3). Specifically, progression of alcohol had a negative relationship with rsFC of the left NAcc and the right dorsolateral PFC. In contrast, rsFC of the right NAcc and a more superior region of the dorsolateral PFC showed a positive relationship with progression of alcohol use. Additionally, progression of alcohol use across adolescence had a positive relationship with left NAcc rsFC with the bilateral ventrolateral PFC and two regions of the left dorsomedial PFC.

Alcohol – Acceleration of Use (Quadratic Slope)

Acceleration of alcohol use across adolescence (i.e., quadratic slope) was linked with rsFC of the NAcc and the dorsolateral PFC and the parahippocampal gyrus (Table 2; Supplementary Tables S2-S3). More specifically, acceleration of alcohol use had a positive relationship with rsFC of the bilateral NAcc and the right dorsolateral PFC, such that participants with greater acceleration of alcohol use had stronger bilateral NAcc-right dorsolateral PFC rsFC. Additionally, greater acceleration of alcohol use was associated with weaker right NAcc-bilateral parahippocampal gyrus rsFC.

Tobacco – Age 14 Use (Intercept)

Age 14 tobacco use was associated with rsFC of the NAcc and the dorsolateral PFC, dorsomedial PFC, parahippocampal gyrus, hippocampus, and amygdala (Table 2; Supplementary Tables S4-S5). Specifically, greater age 14 tobacco use was linked with weaker rsFC of the bilateral NAcc with several regions in the bilateral dorsolateral PFC. Further, rsFC of the left NAcc showed some specific relationships with age 14 tobacco use. The rsFC of the left NAcc had a positive relationship with a midline dorsomedial PFC cluster and a second cluster in the left dorsomedial PFC. Additionally, the rsFC of the left NAcc had a negative relationship with a third, more dorsal region of the dorsomedial PFC. Finally, there was a positive relationship between age 14 tobacco use and rsFC of the left NAcc with the bilateral parahippocampal gyrus, left hippocampus, and right amygdala, such that greater age 14 tobacco use was linked with stronger rsFC between the left NAcc and the parahippocampal gyrus, hippocampus, and amygdala.

Tobacco – Progression of Use (Linear Slope)

Progression of tobacco use across adolescence was linked with rsFC of the bilateral NAcc with the dorsolateral PFC, dorsomedial PFC, parahippocampal gyrus, and amygdala (Table 3; Supplementary Tables S4-S5). Specifically, progression of tobacco use had a positive relationship with rsFC of the bilateral NAcc and several regions within the bilateral dorsolateral PFC, such that faster progression of tobacco use across adolescence was linked with stronger rsFC between the NAcc and the dorsolateral PFC. There were also some specific relationships observed with the left NAcc that were not observed with the right NAcc. For example, faster progression of tobacco use was associated with stronger left NAcc rsFC with one cluster in the left dorsomedial PFC and weaker left NAcc rsFC with a second, more dorsal region of the left dorsomedial PFC. Additionally, progression of tobacco use across adolescence had a positive relationship with left NAcc-left parahippocampal gyrus rsFC. Finally, rsFC of the left NAcc with the left amygdala had a negative relationship with progression of tobacco use, such that faster progression was associated with weaker left NAcc-left amygdala rsFC.

Table 3.

Summary of selected tobacco trajectory results

			Age 14 Tobacco Use
			Left NAcc						Right NAcc
	Region		Hemisphere		Direction		Region		Hemisphere		Direction
	dlPFC		Right		Negative		dlPFC		Right		Negative
	dlPFC		Right		Negative		dlPFC		Right		Negative
	dlPFC		Left		Negative		dlPFC		Left		Negative
	dlPFC		Left		Negative		dlPFC		Left		Negative
	dmPFC		Midline		Positive
	dmPFC		Left		Negative
	dmPFC		Left		Positive
	PHG		Right		Positive
	PHG		Left		Positive
	Hippocampus		Left		Positive
	Amygdala		Right		Positive
			Progression of Tobacco Use
			Left NAcc						Right NAcc
	Region		Hemisphere		Direction		Region		Hemisphere		Direction
	dlPFC		Right		Positive		dlPFC		Right		Positive
	dlPFC		Left		Positive		dlPFC		Left		Positive
	dlPFC		Left		Positive		dlPFC		Left		Positive
	dmPFC		Left		Positive
	dmPFC		Left		Negative
	PHG		Left		Positive
	Amygdala		Left		Negative
			Acceleration of Tobacco Use
			Left NAcc						Right NAcc
	Region		Hemisphere		Direction		Region		Hemisphere		Direction
	dlPFC		Right		Positive

Note: Summary of selected results from models examining the relationship between tobacco trajectories and nucleus accumbens (NAcc) resting state functional connectivity. Full results can be found in Supplementary Tables S4-S5. The left and right NAcc were used as seed regions in whole-brain functional connectivity analyses. Region and Hemisphere refer to the brain region to which the seed was connected, and Direction refers to the directionality of the relationship between the NAcc and the region to which it connected (i.e., either positive or negative). dmPFC = dorsomedial prefrontal cortex; dlPFC = dorsolateral prefrontal cortex; NAcc = nucleus accumbens; PHG = parahippocampal gyrus.

Tobacco – Acceleration of Use (Quadratic Slope)

Acceleration of tobacco use across adolescence was positively associated with rsFC of the left NAcc with the right dorsolateral PFC, such that greater acceleration of tobacco use across adolescence was linked with stronger left NAcc-right dorsolateral PFC rsFC (Table 3; Supplementary Tables S4-S5).

Cannabis – Age 14 Use (Intercept)

Cannabis use at age 14 was linked with rsFC of the NAcc and the dorsolateral PFC, dorsomedial PFC, and hippocampus (Table 4; Supplementary Tables S6-S7). Specifically, rsFC of the bilateral NAcc and several regions in the bilateral dorsolateral PFC had a negative association with age 14 cannabis use. A positive association between rsFC of the left NAcc and the bilateral dorsomedial PFC was observed. The rsFC of the right NAcc was associated with two clusters in the dorsomedial PFC; there was a positive association between age 14 cannabis use and right NAcc-right dorsomedial PFC and a negative association between age 14 cannabis use and a separate, more dorsal region of the dorsomedial PFC. Finally age 14 cannabis use had a positive relationship with rsFC of the right NAcc and the bilateral hippocampus.

Cannabis – Progression of Use (Linear Slope)

Progression of cannabis use across adolescence had a negative relationship with right NAcc-left dorsolateral PFC rsFC, such that faster progression of cannabis use was linked with weaker connectivity between the right NAcc and left dorsolateral PFC (Table 4; Supplementary Tables S6-S7).

Cannabis – Acceleration of Use (Quadratic Slope)

Acceleration of cannabis use across adolescence was linked with rsFC of the NAcc and the dorsolateral PFC, dorsomedial PFC, ventromedial PFC, parahippocampal gyrus, and hippocampus (Table 4; Supplementary Tables S6-S7). Specifically, acceleration of cannabis use had a negative relationship with bilateral NAcc-left dorsolateral PFC rsFC. Further, acceleration of cannabis use was also linked with rsFC of the bilateral NAcc and the left parahippocampal gyrus, such that faster acceleration of use across adolescence was linked with stronger NAcc-parahippocampal gyrus functional connectivity. Additionally, acceleration of cannabis use had a positive relationship with rsFC of the left NAcc and the left ventromedial PFC. Finally, rsFC of the right NAcc with the right dorsomedial PFC, the left parahippocampal gyrus, and the right hippocampus all had positive relationships with acceleration of cannabis use.

Discussion

The functional connectivity of the brain has been linked with clinical, emotional, and cognitive outcomes (Dark et al., 2020; Fox et al., 2007; Greicius et al., 2003; Khalili-Mahani et al., 2012; Seeley et al., 2007). For example, the functional connectivity of the NAcc may be particularly relevant in substance use, given this structure’s central role in the reward network (Chang et al., 1994; Gregorios-Pippas et al., 2009; Hare et al., 2008). Therefore, the present study used resting-state fMRI to investigate the relationships that adolescent substance use trajectories have with NAcc rsFC. Understanding these relationships may advance our knowledge of the neural mechanisms by which adolescent substance use leads to the poor future outcomes observed in prior work (Braun et al., 2000; Grant et al., 2004; Han et al., 2017; Schuster et al., 2001; Volkow et al., 2014). The present study found that trajectories of adolescent alcohol, tobacco, and cannabis use were linked with the rsFC of the NAcc with brain regions that support cognitive control (i.e., PFC) and emotion processes (i.e., hippocampus and amygdala). Consistent with our hypotheses, the trajectories of all three substances were linked with the rsFC of the NAcc and several brain regions including the dorsolateral PFC, dorsomedial PFC, ventrolateral PFC, ventromedial PFC, PHG, hippocampus, and amygdala. These findings suggest that the trajectory of adolescent substance use varies with neural communication between the NAcc and brain regions that support cognitive control and emotion regulation processes. Variations in NAcc rsFC could represent a neural mechanism linking adolescent substance use with the poor future outcomes observed in prior work (Braun et al., 2000; Grant et al., 2004; Han et al., 2017; Schuster et al., 2001; Volkow et al., 2014).

Relationships Across Substances and Trajectories

NAcc rsFC with the dorsolateral and dorsomedial PFC was most consistently associated with adolescent trajectories of substance use (Figure 1). Activity within the dorsolateral PFC has been linked with reward-related decision making and valuation of outcomes, with greater activity associated with active, effortful regulation of reward-related expectations (Delgado et al., 2008; Dixon and Christoff, 2012; Park et al., 2010). The disruption of dorsolateral PFC function may interfere with the regulation of reward-related structures (e.g., NAcc) and has been implicated in the loss of control over substance use that has been linked to SUD (George and Koob, 2010; Motzkin et al., 2014; Park et al., 2010). In fact, substance use appears to vary with rsFC between the dorsolateral PFC and NAcc. Thus, this disruption of dorsolateral PFC-NAcc connectivity may underlie the loss of control over substance use observed in prior research (George and Koob, 2010; Motzkin et al., 2014; Park et al., 2010). For example, NAcc-dorsolateral PFC rsFC is weaker in individuals with SUD than healthy control participants (Motzkin et al., 2014). Further, weaker rsFC between the NAcc and dorsolateral PFC has been linked with greater alcohol use in a community sample (Veer et al., 2019).

Figure 1.

Strength of NAcc-dorsolateral PFC rsFC varied with cannabis use trajectories. NAcc-dorsolateral PFC rsFC showed a negative relationship across substance use trajectories (i.e., age 14 use (left), progression of cannabis use (center), and acceleration of cannabis use (right). Graphs depict standardized residuals of NAcc-dorsolateral PFC rsFC and cannabis use trajectories. NAcc = nucleus accumbens; dlPFC = dorsolateral prefrontal cortex; L = left; R = right; rsFC = resting state functional connectivity.

In addition to the dorsolateral PFC, the present study also found that NAcc-dorsomedial PFC rsFC was consistently associated with substance use, across substances and trajectories. The dorsomedial PFC supports the direction of attention toward salient stimuli (Seeley et al., 2007; Shen et al., 2016). In fact, prior work suggests that attention is biased toward substance-related stimuli during the early stages of addiction (Al-Khalil et al., 2021; Litten et al., 2015). In addition, NAcc-dorsomedial PFC rsFC may be associated with substance use. More specifically, prior work has found weaker NAcc-dorsomedial PFC rsFC in participants with than without SUD (Motzkin et al., 2014; Shen et al., 2016). NAcc-dorsomedial PFC rsFC has been linked with the severity of addiction (e.g., nicotine dependence), such that weaker NAcc-dorsomedial PFC rsFC was associated with greater addiction severity (Hong et al., 2009). In summary, the results of the present study suggest that adolescent substance use varies with the rsFC of the NAcc and regions that process stimulus salience (i.e., dorsomedial PFC).

In addition to the dorsolateral and dorsomedial PFC, rsFC of the NAcc with the hippocampus and PHG was associated with substance use, across substances and trajectories. The hippocampus and PHG are integral to learning and memory and have been strongly implicated in reward learning (Adcock et al., 2006; Haber and Knutson, 2010; Wittmann et al., 2005). The hippocampus is part of the reward network and shares reciprocal connections with the NAcc (Everitt and Robbins, 2005). The hippocampus also appears to provide contextual information regarding substance cues (Everitt and Robbins, 2005). Relatively few investigations have examined the rsFC of the NAcc with the hippocampus and PHG. Further, the limited work that has examined NAcc-hippocampus rsFC was conducted with adults with chronic cocaine use (Gu et al., 2010). This prior work did not find a relationship between NAcc-hippocampus rsFC and cocaine use, which is not consistent with results of the current study. However, it is possible that these inconsistencies are due to differences in the severity of substance use of the participants included in each study. The present study included a community sample of young adults and examined recreational levels of substance use, in contrast to the prior work which examined associations between chronic cocaine use and brain functional connectivity (Gu et al., 2010). Prior work suggests that the progression from substance use to addiction is mediated, in part, by a switch from ventral striatal signaling (i.e., NAcc) to dorsal striatal signaling (i.e., caudate, putamen) (Everitt and Robbins, 2005; Gu et al., 2010; Porrino et al., 2007). Therefore, prior work focused on chronic substance use may not have observed potential relationships with NAcc-hippocampal rsFC that may have been present prior to the progression to addiction. The results of the present study suggest that the dorsolateral and dorsomedial PFC, hippocampus, and PHG are associated with NAcc rsFC across substances and trajectories. Thus, variations of NAcc rsFC with these brain regions may represent a vulnerability factor that is common across substances.

Several additional brain regions were also consistently linked with NAcc rsFC across substances and trajectories, however the pattern of the relationships differed. For example, across all three substances, age 14 use had a negative relationship with NAcc-dorsolateral PFC rsFC. In contrast, both positive (tobacco use) and negative (alcohol and cannabis use) relationships were observed between the progression of use and NAcc-dorsolateral PFC rsFC (Figure 2). The full breadth of the results can be found in Supplementary Tables S2-S7. Very few studies have examined associations between adolescent substance use trajectories and neural outcomes, thus there is currently no consensus in the literature regarding the reason relationships may be less consistent for the progression and acceleration of use. However, it is possible that individual differences in brain maturation may contribute to the observed variability across trajectories. For example, puberty is a time of significant hormonal changes, some of which have been linked with brain development (De Bellis et al., 2001; Giedd et al., 1997; Herting et al., 2015). Therefore, individual differences in pubertal status and brain development, which may coincide with increases in substance use for some adolescents, may vary with NAcc rsFC. Thus, the precise relationship substance use has with neural development may depend, in part, upon individual variability in pubertal status and brain development. The present study included a measure of perceived pubertal status to address this possibility. However, this measure was not available at all waves and represents a relatively rough estimate of pubertal status. Thus, additional research is necessary to better characterize the relationships that progression and acceleration of substance use have with individual variability in pubertal status and brain development.

Figure 2.

Strength of NAcc-dorsolateral PFC rsFC varied with progression of use across substances. NAcc-dorsolateral PFC rsFC had a negative association with progression of alcohol (left) and cannabis use (right) and a positive relationship with progression of tobacco use (center). Graphs depict standardized residuals of NAcc-dorsolateral PFC and progression of substance use. NAcc = nucleus accumbens; dlPFC = dorsolateral prefrontal cortex; L = left; R = right; rsFC = resting state functional connectivity.

Substance Specific Relationships

In addition to patterns of rsFC that were common across all substances and trajectories, the present findings also identified some substance-specific patterns of NAcc rsFC. For example, tobacco, but not alcohol or cannabis, use trajectories were associated with NAcc rsFC with the amygdala. Specifically, age 14 tobacco use was positively associated with the NAcc-amygdala rsFC (Figure 3), while the progression of tobacco use was negatively associated with NAcc-amygdala rsFC. The amygdala is heavily involved in emotion expression and activity within this region may underlie craving and negative emotionality that is associated with withdrawal (Cheng et al., 2003; Knight et al., 2005; Koob and Volkow, 2016; Shaham et al., 2003; Um et al., 2020). Prior work has implicated NAcc-amygdala rsFC in substance use. Specifically, stronger NAcc-amygdala rsFC has been observed in adults with high nicotine dependence compared to low dependence (Shen et al., 2016). Thus, this prior work supports the findings of the present study, which demonstrated a positive association between tobacco use at age 14 and NAcc-amygdala rsFC. However, the results pertaining to progression and acceleration of substance use are more difficult to contextualize within the extant literature given that few studies have examined substance use trajectories. Regardless, the present study suggests that trajectories of tobacco use are associated with the rsFC of the NAcc and amygdala. It is possible that the patterns of NAcc rsFC observed in the present study could be related to factors associated with substance use, such as craving, relapse, and resilience. Future work exploring the relationships that rsFC patterns have with other features connected with substance use (e.g., craving, relapse, resilience) may help to clarify the variability of the present results.

Figure 3.

Strength of NAcc-amygdala rsFC varied with age 14 tobacco use (left), but not with age 14 alcohol (center) or cannabis (right) use. NAcc-amygdala rsFC showed a positive relationship with age 14 tobacco use such that greater age 14 tobacco use was associated with stronger NAcc-amygdala rsFC. Graphs depict standardized residuals of NAcc-amygdala rsFC and age 14 substance use. NAcc = nucleus accumbens; L = left; R = right; rsFC = resting state functional connectivity.

The ventrolateral PFC is another region that showed substance-specific patterns of NAcc rsFC. Specifically, the progression of alcohol use was positively associated with NAcc-ventrolateral PFC rsFC, a relationship which was not observed with other substances (Figure 4). The ventrolateral PFC is part of the executive control network, which underlies a broad range of processes that include cognitive flexibility, attentional control, and goal-directed decision making (Seeley et al., 2007; Sutherland et al., 2012; Zhang and Volkow, 2019). More specifically, ventrolateral PFC function has been linked with response inhibition and emotion regulation, tasks on which individuals with SUD tend to perform poorly (Al-Khalil et al., 2021; Nestor et al., 2011; Ochsner and Gross, 2008; Ridderinkhof et al., 2004). Additionally, NAcc-ventrolateral PFC rsFC has been previously implicated in substance use. For example, prior work found weaker NAcc-ventrolateral PFC rsFC in adults with opioid dependence compared to healthy control participants (Upadhyay et al., 2010). Similarly, substance-naïve adolescents with a positive family history of Alcohol Use Disorder (AUD) had weaker NAcc-ventrolateral PFC rsFC compared to youth without a family history of AUD (Cservenka et al., 2014). Thus, prior work suggests that weaker NAcc-ventrolateral PFC rsFC is linked with substance use and may represent a pre-existing vulnerability for individuals with a family history of AUD.

Figure 4.

Strength of NAcc- ventrolateral PFC rsFC varied with progression of alcohol, but not tobacco or cannabis use. Greater progression of alcohol use across adolescence was associated with stronger NAcc-ventrolateral PFC rsFC. No relationship was observed with NAcc-ventrolateral PFC rsFC and progression of tobacco or cannabis use. Graphs depict standardized residuals of NAcc-ventrolateral PFC rsFC and progression of alcohol use. NAcc = nucleus accumbens; vlPFC = ventrolateral prefrontal cortex; L = left; R = right; rsFC = resting state functional connectivity.

In contrast to prior work, results of the present study indicate that progression of alcohol use across adolescence has a positive relationship with NAcc-ventrolateral PFC rsFC. Emerging evidence suggests that stronger rsFC between the NAcc and regions that support executive control (e.g., dorsolateral and ventrolateral PFC) may be linked with abstinence from alcohol use (Camchong et al., 2013a, 2013b; Seeley et al., 2007). Specifically, prior work found that stronger coupling of the NAcc and executive control regions was linked with better performance on cognitive flexibility tasks in long-term abstinent participants (Camchong et al., 2013a, 2013b). Further, a dose-dependent effect was observed in which longer abstinence was linked with stronger coupling of the NAcc with executive control regions (Camchong et al., 2013b). Thus, taken together with prior work, the results of the present study suggest that the progression of alcohol use across adolescence is associated with NAcc-ventrolateral PFC rsFC in young adulthood. However, future work is necessary to more fully understand whether the pattern of NAcc-ventrolateral PFC rsFC observed in the present study may relate to executive control processes (response inhibition, attentional control) that affect risk and resilience to SUD.

In addition to substance-specific effects for age 14 tobacco use and the progression of alcohol use, substance-specific effects were also observed for the acceleration of cannabis use. Specifically, greater acceleration of cannabis use across adolescence was linked with stronger NAcc-ventromedial PFC rsFC (Figure 5). The ventromedial PFC is part of the default mode network (DMN), a network of brain regions that are often less active during cognitive tasks, but which have been linked with self-referential processing (Andrews-Hanna, Reidler, Huang, et al., 2010; Zhang and Volkow, 2019). Prior work suggests that ventromedial PFC, in particular, underlies processing information of personal relevance (Andrews-Hanna, Reidler, Sepulcre, et al., 2010; Benoit et al., 2010; Ochsner et al., 2005). Further, ventromedial PFC activity has been linked with poor decision making in individuals who use substances (Bechara et al., 2001; Konova et al., 2019; Vaidya et al., 2012). Prior work also suggests the function of both the NAcc and ventromedial PFC can be impacted by substance use (e.g., Khalili-Mahani et al., 2012). Further, previous research has found NAcc-ventromedial PFC rsFC is stronger in those with SUD than control participants (Arienzo et al., 2020; Wilcox et al., 2011). Further, the strength of NAcc-ventromedial PFC rsFC was positively associated with several metrics of alcohol use (e.g., number of binge drinking episodes, number of drinks in 24 hours, alcohol craving) (Arienzo et al., 2020; Bracht et al., 2021). Thus, these findings appear to suggest that stronger NAcc-ventromedial PFC rsFC represents a risk factor for future substance use (Arienzo et al., 2020; Bracht et al., 2021). For example, others have suggested that stronger NAcc-ventromedial PFC rsFC is linked with alcohol craving, which could result in increased alcohol consumption (Bracht et al., 2021). In contrast, however, recent work has found that stronger NAcc-ventromedial PFC rsFC is associated with lower risk of substance use relapse (Wang et al., 2021; Yang et al., 2021). Thus, it is possible that stronger NAcc-ventromedial PFC rsFC could be protective in some circumstances. Taken together, the results of the present study indicate that the acceleration of cannabis use across adolescence is associated with NAcc-ventromedial PFC rsFC. Additional work is necessary to clarify whether, and under what circumstances, this pattern of rsFC represents a risk or resilience factor.

Figure 5.

Strength of NAcc-left ventromedial PFC rsFC varied with acceleration of cannabis, but not alcohol or tobacco use. NAcc-ventromedial PFC rsFC showed a positive relationship with acceleration of cannabis use. Greater acceleration of cannabis use across adolescence was associated with stronger NAcc-ventromedial PFC rsFC. Graphs depict standardized residuals of NAcc-left ventromedial PFC rsFC and acceleration of substance use. NAcc = nucleus accumbens; vmPFC = ventromedial prefrontal cortex; L = left; R = right; rsFC = resting state functional connectivity.

Differences in NAcc rsFC Across Trajectories

There were several brain regions where the rsFC of the NAcc was linked with substance use across substances and trajectories, though the directions of relationships varied. For example, rsFC of the NAcc and PHG was negatively associated with the acceleration of alcohol use, positively associated with acceleration of cannabis use, and not significantly associated with the acceleration of tobacco use. Further, there were several regions where rsFC of the NAcc was linked with one substance and not another. For example, NAcc-amygdala rsFC was linked only with tobacco trajectories. One possible explanation for the variation in results across substances could be differences in the mechanisms of action of alcohol, tobacco, and cannabis. Although we did not directly test this hypothesis, prior work has established that many psychoactive substances lead to the release of dopamine (DA) in the NAcc and ventral tegmental area, although the precise mechanisms of action differ (Di Chiara and Imperato, 1988; Koob and Bloom, 1988; Volkow et al., 2019; Wise, 2008). Prior work suggests that cannabis and nicotine exert their effects via specific receptors, while alcohol has effects on a wide range of receptor types (Bertrand and Terry, 2018; Draycott et al., 2014; Fowler et al., 2008; Silveira et al., 2017; Volkow et al., 2019; Wiers et al., 2016). Moreover, recent work indicates that specific receptors may play a role in behaviors associated with substance use. For example, a group of participants with AUD with about 1 week of abstinence had greater availability of the metabotropic glutamate 5 (mGlu5) receptor compared to control participants, particularly within the dorsolateral and ventromedial PFC (Hillmer et al., 2021). Further, emerging evidence suggests that receptor availability may contribute, in part, to rsFC signals observed in fMRI studies (Kojima et al., 2022; Smart et al., 2022). Thus, it is possible that different distributions of receptors may contribute to the variability in findings across substances in the present study.

In addition to differences in mechanism of action, the variability observed in the present study could be due to differences in the effects of substances on glial cells. Although we did not directly examine glial cells in the current investigation, emerging evidence suggests that some glial cells express receptors that can be activated by some psychoactive substances (Aryal et al., 2021; Carrera et al., 2020). Glial cells support neuronal signaling and synaptic development (Fields et al., 2015; Pannasch et al., 2011; Stogsdill and Eroglu, 2017; Ullian et al., 2001). In fact, prior work suggests that glial cells may contribute to rsFC signals observed with fMRI (Kahali et al., 2021). Thus, different psychoactive substances may have distinct relationships with glial function, which could contribute to the variability in the findings across substances observed in the present study. Taken together, the results of the present study suggest that there may be substance-specific associations between adolescent alcohol, tobacco, and cannabis use trajectories and NAcc rsFC in young adulthood.

Limitations and Future Directions

The findings of the present investigation should be interpreted within the context of their limitations. The present study utilized longitudinally reported substance use data, however, there was only a single MRI scanning session. Thus, causality cannot be established in the present investigation. Further, because the neuroimaging session was conducted after substance use was initiated, we are unable to determine whether relationships observed between NAcc rsFC and substance use trajectories represent a cause or a consequence of use (Miller et al., 2024). Additionally, the present study relied upon data collected as part of a larger study of health behaviors (Schuster et al., 2012; Windle et al., 2004). Thus, we were limited in some respects by measures included in the larger study. For example, measures related to substance use (e.g., craving, diagnostic assessments) were not included in the parent study. Future work will be necessary to better understand whether the relationships observed in the present study are linked with other outcomes associated with substance use, such as the development of SUD. Similarly, vaping was not evaluated in the initial study because data collection began in 2004, prior to the rise in popularity of vaping among adolescents (Gilbert et al., 2020). Thus, it is possible that tobacco use trajectories in the present investigation may not accurately reflect all participants’ actual nicotine use. Finally, some aspects of our analysis approach deserve mention. First, all three growth parameters were included in neuroimaging models to evaluate the unique contributions of each growth parameter. Second, we did not examine trajectories of substance co-use or other drug use in the present investigation. Third, we have not systematically explored the impact of covariates (e.g., substance co-use) on NAcc rsFC. While these analysis choices are not necessarily limitations, they are important to bear in mind when interpreting the present results and they may serve as important avenues for future investigation.

Conclusions

The present study investigated the relationships that alcohol, tobacco, and cannabis trajectories have with the rsFC of the NAcc. The current findings suggest that adolescent substance use trajectories are associated with rsFC of the NAcc in young adulthood. More specifically, alcohol, tobacco, and cannabis use at age 14 was linked with the rsFC of the NAcc and dorsolateral PFC, dorsomedial PFC, PHG, hippocampus, and amygdala. Additionally, the progression of alcohol, tobacco, and cannabis use was linked with the rsFC between the NAcc and dorsolateral PFC, dorsomedial PFC, ventrolateral PFC, PHG, and amygdala. Finally, the acceleration of alcohol, tobacco, and cannabis use was linked with rsFC between the NAcc and dorsolateral PFC, ventromedial PFC, PHG, and hippocampus. Several brain regions were associated with NAcc rsFC across substances (i.e., dorsolateral PFC, dorsomedial PFC, PHG, hippocampus). Thus, these brain regions may represent a general (i.e., not specific to a particular substance) neural correlate associated with adolescent substance use. Identification of common neural correlates associated with adolescent substance use may inform prevention and early intervention strategies. Future work can build upon the current findings by exploring relationships that NAcc rsFC has with behavioral outcomes associated with substance use, such as progression to substance use disorder, craving, relapse, and recovery.