Developmental Trajectories of the Orbital Frontal Cortex and Anhedonia in Middle Childhood and Risk for Substance Use in Adolescence in a Longitudinal Sample of Depressed and Healthy Preschoolers

Abstract

Objective

Deficits in reward processing are established in mood and substance use and are known risk factors for these disorders. Volume reductions of the orbital frontal cortex and the striatum, regions that subserve neural response to reward, have been identified related to anhedonia in depressive disorders and in substance use disorders. However, how structural maturation of these regions in childhood varies with anhedonia and predicts later substance use remains unknown.

Method

We investigated this using data from a sample of depressed and healthy preschoolers studied longitudinally that included 3 waves of neuroimaging from school age to adolescence. Three years after scan 3, subjects aged 13–18 participated in a comprehensive behavioral and substance use assessment. While not the primary aim of the study, the availability of 3 waves of brain data prior to adolescence made the data set of interest for an investigation of risk for substance use. Multi-level modeling was used to investigate the relationship between anhedonia and the growth trajectories of the striatum and orbital frontal cortex. Zero-inflated Poisson models were then conducted to determine if the intercepts and slopes of these trajectories predicted later alcohol and marijuana use frequency in adolescence.

Results

The anhedonia-by-age interaction was significant in the multi-level modeling of orbital frontal cortex but not striatal volume. Higher anhedonia was associated with steeper decline in orbital frontal cortex volume with age (t=−3.06, p=0.0025). Orbital frontal cortex volume at age 12 (t=−2.90, p=0.0038) and orbital frontal cortex thickness at age 12 (t=−4.11, p<0.0001) and trajectory over time (t=−4.91, p<0.0001) significantly and negatively predicted subsequent alcohol and marijuana use frequency but not depression during adolescence.

Conclusions

Findings suggest that the development of the orbital frontal cortex during childhood is strongly linked to experiences of anhedonia and that these growth trajectories predict substance use during a developmentally critical period.

Introduction

The ability to experience pleasure during the anticipation and receipt of reward is a fundamental human experience central to the drive to survive and thrive. Deficits in this domain have been identified in both mood and substance use disorders (1, 2). The symptom of anhedonia or the inability to enjoy previously pleasurable or rewarding activities is a core feature of depressive disorders. Abnormalities in reward processing have been identified from observational, behavioral and neurobiological studies as potent risk factors for the emergence of depression as well as substance use disorders (3).

Both human neurobiological studies and experimental animal studies have localized key brain regions involved in experiences of reward (4). This combined body of data validates the central role of the orbital frontal cortex and the striatum in neural responses to reward. The orbital frontal cortex plays a role in processing sensory input and in estimating the value of experiences of reward, including both natural (e.g., gustatory), monetary, and drug –related incentives (5, 6). The striatum drives appetitive behavior by facilitating incentive salience (e.g. wanting) and reward-related learning (7). While the OFC and striatal volume shows normative increases and then declines beginning in early adolescence (8), reductions in the volume of the orbital frontal cortex and the striatum have been identified in depressive disorders related to the symptom of anhedonia in adolescence (9–11). Variations in striatal and orbital frontal cortex volume have also been linked to substance use and later substance use disorders. Numerous studies have documented decreased gray matter volume in the orbital frontal cortex in substance-using individuals, including those using alcohol (12, 13) and marijuana (14–17). While the role of striatum in “wanting” is evident in reward-related behaviors such as substance use and substance use disorders, the role of the orbital frontal cortex may be related to the pursuit of goal-directed behaviors, or drive, which might evolve into habitual actions (18, 19), as well as learning associations between stimulus and reward (2), both highly relevant to risk for substance use disorders.

Of particular interest, associations between the orbital frontal cortex and striatum with substance use have been found during the reward-salient period of adolescence. In the United States, alcohol and marijuana are among the first substances to be used by Caucasian and African-American youth respectively, with onsets peaking between 15–17 years of age (20, 21). In addition, epidemiological studies suggest that childhood internalizing problems might modify risk, positively and negatively, for onset of substance use (22). Therefore, the extant literature supports the hypothesis that anhedonia-related changes in brain maturation might impact the onset and use of alcohol and marijuana. Both the orbital frontal cortex and ventral striatum have been investigated in the context of adolescent substance use. For marijuana, while most prior studies have conceptualized orbital frontal cortex volumetric alterations as consequences of heavy use (23), one longitudinal study found that smaller orbital frontal cortex volume at age 12 was related to onset of marijuana use by age 16 (15). A few studies also link orbital frontal cortex structure and adolescent alcohol use (e.g., (24)). Similar to marijuana findings, one adolescent co-twin-control study reported lower lateral orbital frontal cortex volume related to alcohol use. However as no volumetric differences were found within pairs of twins with varying levels of alcohol use, these findings suggested that changes in orbital frontal cortex volume might be a preexisting marker of risk and not a consequence of prolonged alcohol exposure (25). A link between structural differences in the striatum, and marijuana and alcohol use is less clearly established, with some studies suggesting ventral striatal volume increases in users (e.g.,(26)). These finding suggest that the orbital frontal cortex and ventral striatum are key areas involved in adolescent substance use and that while they may be vulnerable to the psychoactive effects of drugs, variations in their maturation might also precede and contribute to the onset of substance use.

While the key roles of the orbital frontal cortex and the striatum in reward processing are relatively clear, how structural maturation of these regions during childhood relates to variation in hedonic tone, experiences of anhedonia and later reward-seeking and substance use and depression outcomes remains unknown. Early deficits in hedonic tone may co-occur with or influence the developmental trajectory of the orbital frontal cortex and striatum. In turn, these neurobiological alterations may influence later reward-related drive behavior and related disorders during adolescence. One hypothesis is that anhedonia through middle childhood leads to increased goal-directed, reward-seeking behavior, potentially compensatory, to obtain high hedonic valence experiences in the form of substance use in adolescence. Alternatively, the growth trajectory of the orbital frontal cortex has been shown to relate to risk for later substance use, independent of anhedonia (15), suggesting independent processes between orbital frontal cortex volume and adolescent substance use are also operative. Based on these established brain behavior relationships, an investigation of the developmental trajectory of striatum and orbital frontal cortex, as a function of anhedonia in relation to risk for later depression and/or substance abuse, is of interest. Building on the extant literature, this study aimed to test hypotheses about how behavior and brain in these domains change together or independently influence each other to predict the risk trajectory for later substance use and/or depression.

Utilizing data from a longitudinal neuroimaging study of children with early onset depression and healthy controls followed from preschool into adolescence, we examined the developmental trajectory of orbital frontal cortex and striatal volumes and how they varied as a function of anhedonia. While this study sample was not specifically designed to investigate risk for substance use and substance use disorder, it provided an ideal opportunity to investigate the trajectory of brain change in the context of varying anhedonia and depression across childhood to inform risk for first onset of substance use disorder in adolescence and risk for a later recurrence of depression. We hypothesized that orbital frontal cortex and striatal volumes would decrease over time as a function of increased anhedonia even when controlling for other depressive symptoms. Based on an extensive literature demonstrating the role of the orbital frontal cortex and ventral striatum in onset of alcohol and marijuana use, the first substances to be used by most U.S. youth, we then sought to explore whether these brain-related volumetric trajectories predicted later substance use and depression recurrence as the study sample entered adolescence, which is a key period of risk for onset of substance use and depression. To examine the role of anhedonia in the associations with substance use, we compared whether volumetric alterations related to, and independent of, anhedonia were equally predictive of alcohol and marijuana use during adolescence.

Methods

Three hundred and five children (54% White, 33% Black, 13% Other) aged 3–6.0 at baseline, oversampled for symptoms of depression, were recruited from the St. Louis metropolitan area for participation in a longitudinal study of preschool-onset depression. After a first behavioral-only study phase, healthy children and those with a history of depression (N=211) were invited to participate in further school age and adolescent follow-up to undergo neuroimaging along with behavioral and diagnostic assessments comprising 3 additional waves. N=193 children had useable relevant data at 1 or more scan waves (N=113 had data at all 3 scans). Proximal to the scan, children participated in behavioral assessments that included parent report (< age 8) on the Preschool Age Psychiatric Assessment (PAPA)(27), the Child and Adolescent Psychiatric Assessment (CAPA)(28) (parent report at age 8 and parent and child report > age 9) as well as the Child Depression Inventory (CDI-C)(29) which was obtained at the time of scan. Demographic, psychosocial and developmental characteristics were also assessed. Approximately 3 years after scan 3, subjects aged 13–18 returned for an assessment that included a diagnostic interview as well as the Composite International Diagnostic Interview (CIDI)(30) to assess substance use. All study methods were reviewed and approved by the Washington University School of Medicine Institutional Review Board. Written informed consent was obtained from all study participants.

Assessments

Anhedonia

Anhedonia was assessed by parent report on the PAPA at baseline and at each scan by child report on the anhedonia subscale of the CDI-C (29). The CDI-C anhedonia subscale is comprised of 8 items, and Cronbach’s alphas for the subscale were 0.608, 0.648, and 0.540 at Scans 1, 2, and 3, respectively. The CDI has been shown to make valid distinctions between clinical and non-clinical levels of anhedonia(31).

Alcohol and marijuana use

Use of alcohol and drugs was assessed approximately 3 years after scan 3 using the CIDI. Frequency of alcohol use (ALC) in the past 12 months (never, less than once/month, 1–3 days/month, 1–2 days/week, 3–4 days/week, nearly every day, every day) was assessed using 3 items: (1) frequency of ≥1 drink, (2) frequency of ≥5 drinks, and (3) frequency of intoxication. Frequency of marijuana (MJ) and other drug use in the past 12 months was assessed using similar response categories but only related to use ≥1 times. To cover the general liability to early substance use during adolescence, and capture the first instance of substance use in our major demographic groups (i.e., Blacks and Whites), a variable representing alcohol use/marijuana use frequency (ALC/MJ) was created by summing the categorical marijuana use frequency variable (0=never, 1=3 or fewer days/month, 2= > 3 days/month) with the categorical frequency of ≥5 drinks of alcohol (alcohol use) variable (0=never, 1=less than 1 day/month, 2=1 or more days/month). We chose a higher level of alcohol consumption to capture non-normative drinking in this population and exclude the possibility that respondents might refer to occasional drinking during family celebrations or holidays or subjective evaluations of sensations of intoxication. However, secondary analyses examined combined alcohol/marijuana use variables at other degrees of alcohol use (i.e., frequency of ≥1 drink and of intoxication).

Other time-varying covariates

MDD core score without anhedonia, prior to each scan (i.e., time-varying), was calculated as the number of the 8 MDD core symptoms, not including anhedonia, endorsed by the parent and/or child.

Neuroimaging

Participants underwent neuroimaging using a 3.0-T Tim Trio; Siemens Healthcare GmbH scanner. The same scanner was used at all 3 scans. The 2 results of magnetization-prepared rapid acquisition gradient echo scans were assessed visually, and the best one was selected for further processing by blind raters. The selected magnetization-prepared rapid acquisition gradient echo image for each wave was processed using the longitudinal stream in FreeSurfer software package, version 5.3 (surfer.nmr.mgh.harvard.edu)(32). Several processing steps, such as skull stripping, Talairach transformations, and atlas registration, as well as spherical surface maps and parcellations, were initialized with common information from an unbiased within-patient template. This longitudinal stream reduces the bias that would otherwise be present in selecting a single scan result as baseline, and significantly increases reliability and statistical power. For striatal volume, we focused on the sum of bilateral caudate, putamen and nucleus accumbens volumes. For OFC volume, we aggregated bilateral medial and lateral orbital frontal parcels from the Desikan atlas, one of the atlases available and validated in the FreeSurfer structural processing program (33). To account for whole brain volume, OFC and striatal volumes were regressed on whole brain volume minus orbital frontal cortex/striatal volumes, and the residuals were then utilized as the dependent variables in analyses.

Analysis

Anhedonia and trajectories of OFC and striatal volume

We fit the data to a longitudinal multilevel linear model in SAS v9.3 to determine whether child-reported anhedonia at the three scan waves (centered at the mean 43.5) was significantly associated with the trajectory of residualized orbital frontal cortex or striatal volumes across time points. We chose to use multilevel linear models for the following reasons: (1) the ability to model non-independence across observations; (2) the ability to include categorical and continuous predictors at any level; and (3) robustness to missing data. Given our longitudinal sample, multilevel linear models allowed us to model both between- and within-person variation simultaneously, whereas, generalized linear models account for fixed, but not random, effects (34). The multilevel linear models included random intercept and slope components with an unstructured covariance structure. Influence statistics for the multilevel linear models were obtained, and one subject with an extreme restricted likelihood distance value for the multilevel linear model of residualized orbital frontal cortex volume was excluded from orbital frontal cortex analyses. Time was defined as age at scan (centered at median age 12), and an age squared variable was included in the models to account for quadratic slopes. The influence of anhedonia was modeled as an interaction with age. Other covariates were sex and MDD core score (excluding anhedonia) at the scan wave (centered mean 1.85). An interaction between MDD core score and age was included in the models to account for differences in orbital frontal cortex or striatal values over time due to depression severity. An online calculator (35) was used to identify the age bands in which age × anhedonia significantly predicted volumetric change. As a follow-up to the significant volume results for orbital frontal cortex, we used the same models to examine thickness and surface area to determine which component contributed to the volume effects. A Bonferroni corrected p-value of 0.025 (for orbital frontal cortex and striatum) was used to assess significance.

Trajectories of brain volume and substance use at follow-up

In brain regions for which a significant anhedonia by age interaction was identified, individual subject intercepts and slopes (i.e., residualized for covariates and age × anhedonia effects) were extracted and investigated as predictors of alcohol and marijuana use frequency 3 years after scan 3, as well as a variable representing use of either substance to represent the first substance used by youth in our sample. As more than 60% of the N=135 subjects reported no alcohol or marijuana use in the last 12 months, zero-inflated Poisson regression models were utilized to simultaneously estimate a logistic regression of the zero-inflated component of the outcome (no ALC or MJ use, or abstinence) and a regression of the continuous component of the outcome (frequency of ALC or MJ use) on the predictors. Zero-inflated Poisson regression is ideal for constructs, such as substance use containing excess zero-count data (e.g., 0 = no alcohol/marijuana use), as using standard ordinary least squares regression on this type of data is likely to bias results by confounding onset of use (0 vs. non-zero value) and progression (range of use within users). Zero-inflated Poisson regressions account for any overdispersion due to an excessive amount of zeros without biasing the parameters or standard errors (36–38). There are two components to the model; the first employs a binary distribution to generate structural zeroes, and the second component generates counts, some of which may be zero. Race, sex, and age at last assessment were included as covariates. To examine whether anhedonia played a critical role in the relationship between change in brain volume and substance use, analyses were redone using intercepts and slopes extracted from an orbital frontal cortex model that did not include anhedonia as a main effect or interaction (age × anhedonia) term.

In supplemental analyses, we examined whether age at first substance use was associated with orbital frontal cortex volume or thickness. Age at onset was coded as self-reported age when first used alcohol or marijuana, or age at the first assessment when a non-zero value was reported for use. To account for the possibility that some never-users may be censored (i.e., not had the opportunity to use the substance due to their current age), age at final assessment was used in place of age of onset for never-users. Additional supplemental analyses also examined alternate definitions of alcohol use (i.e., marijuana use with alcohol use ≥ 1drink frequency and marijuana with alcohol intoxication frequency).

Trajectories of brain volume and depression at follow-up

MDD severity at the later follow-up was also investigated as an outcome. Individual subject intercepts and slopes were included as independent variables in linear regression models of MDD severity (number of MDD core symptoms), covarying for race, sex, and age.

Results

Characteristics of the sample are shown in Table 1.

Table 1.

Characteristics of the Sample (N=193)

	Scan 1 (N=175)		Scan 2 (N=160)		Scan 3 (N=139)		Later Wave (N=135)
Characteristic	%	N	%	N	%	N	%	N
Female sex	48.0	84	48.8	78	49.6	69	48.9	66
Race
Caucasian	54.9	96	50.0	80	46.8	65	51.9	70
African-American	33.1	58	40.0	64	41.7	58	35.6	48
Other	12.0	21	10.0	16	11.5	16	12.6	17
Age in years
7	2.9	5	0.0	0	0.0	0	0.0	0
8	13.1	23	0.0	0	0.0	0	0.0	0
9	27.4	48	8.1	13	0.0	0	0.0	0
10	22.9	40	18.1	29	2.9	4	0.0	0
11	23.4	41	29.4	47	19.4	27	0.0	0
12	10.3	18	27.5	44	29.5	41	0.0	0
13	0.0	0	13.8	22	30.9	43	3.7	5
14	0.0	0	3.1	5	13.7	19	8.2	11
15	0.0	0	0.0	0	3.6	5	28.2	38
16	0.0	0	0.0	0	0.0	0	32.6	44
17	0.0	0	0.0	0	0.0	0	24.4	33
18	0.0	0	0.0	0	0.0	0	3.0	4
MDD diagnosis	14.9	26	5.6	9	6.5	9	11.9	16
Externalizing diagnosis	20.4	35	17.3	27	12.8	16	12.6	17
Alcohol use (≥5 drinks) frequency
Never							68.2	92
<1 day/month							25.9	35
≥1 day/month							5.9	8
Marijuana use frequency
Never							79.3	107
≤3 days/month							11.1	15
>3 days/month							9.6	13
	Mean	SD	Mean	SD	Mean	SD	Mean	SD
Age in years	10.32	1.27	11.81	1.23	12.97	1.16	16.25	1.11
Income to needs ratio	1.72	0.95	1.72	0.92	1.67	0.92	1.97	0.77
Months since Scan 1	0.00	0.00	18.08	6.28	32.59	8.14	70.69	10.31
CDI-C anhedonia T-score	44.37	8.09	43.50	7.99	42.47	6.36
MDD core score without anhedonia	2.23	1.87	1.56	1.43	1.72	1.39

Anhedonia and trajectories of orbital frontal cortex and striatal volume

The anhedonia by age interaction was significant in the multilevel linear model of orbital frontal cortex volume (residualized for whole brain minus orbital frontal cortex; Table 2). Linear and quadratic slopes were significant. The interaction between time-varying anhedonia and age at scan was significantly associated with orbital frontal cortex volume, with higher levels of anhedonia indicating steeper decline in orbital frontal cortex volume with age. The age × anhedonia term was significantly related to orbital frontal cortex volume prior to age 11.25 and after age 14.88. Figure 1 illustrates estimated trajectories of orbital frontal cortex volume at three values of anhedonia (mean and +/− 1 SD). Those with higher anhedonia scores had higher orbital frontal cortex volumes prior to age 11.25 than those with lower anhedonia scores; even though volumetric decreases occurred across all levels of anhedonia. There was a crossover effect with those with higher anhedonia scores showing steeper decreases. Orbital frontal cortex thickness, but not surface area, showed similar significant effects to volume (Table 2), including an association with age × anhedonia, with higher levels of anhedonia associated with a steeper decline in orbital frontal cortex thickness (Figure 2). These associations were significant prior to age 11.37 and after age 15.85. Of note, the interaction between age and MDD core score without anhedonia was not a significant covariate in either the volume or thickness model.

Table 2.

Multilevel Models of Orbital Frontal Cortex (OFC) Volume Residuals, OFC Surface Area, and OFC Thickness by Time-Varying CDI-C Anhedonia T-Score (N=193)

DV: OFC Volume Residuals	Estimate	SE	t	p
Intercept	−0.1131	0.1868	−0.61	0.5457
Age at scan	−0.1228	0.0261	−4.71	<0.0001
Age at scan squared	−0.0337	0.0106	−3.19	0.0017
Male gender	0.1328	0.2575	0.52	0.6066
MDD core score without anhedonia	−0.0352	0.0256	−1.38	0.1702
CDI-C anhedonia T-score	0.0029	0.0060	0.48	0.6291
MDD core score without anhedonia X Age at scan	−0.0148	0.0141	−1.04	0.2972
CDI-C anhedonia T-score X Age at scan	−0.0102	0.0033	−3.06	0.0025
DV: OFC Surface Area (mm2)	Estimate	SE	t	p
Intercept	9691.18	3301.10	2.94	0.0035
Age at scan	−27.3667	141.31	−0.19	0.8465
Male gender	499.29	306.37	1.63	0.1054
MDD core score without anhedonia	−20.1196	3.7569	−5.36	<0.0001
CDI-C anhedonia T-score	0.6699	0.9033	0.74	0.4594
MDD core score without anhedonia X Age at scan	0.8103	2.3846	0.34	0.7344
CDI-C anhedonia T-score X Age at scan	−0.4704	0.6023	−0.78	0.4356
DV: OFC Thickness (mm)	Estimate	SE	t	p
Intercept	10.8609	0.0429	253.12	<0.0001
Age at scan	−0.1555	0.0082	−18.97	<0.0001
Age at scan squared	−0.0134	0.0034	−3.90	0.0001
Male gender	0.0824	0.0584	1.41	0.1600
MDD core score without anhedonia	−0.0070	0.0087	−0.80	0.4235
CDI-C anhedonia T-score	0.0016	0.0020	0.79	0.4328
MDD core score without anhedonia X Age at scan	−0.0064	0.0047	−1.37	0.1717
CDI-C anhedonia T-score X Age at scan	−0.0032	0.0011	−2.96	0.0036

Figure 1.

Estimated Trajectories of Orbital Frontal Cortex (OFC) Volume Residuals Over Time by CDI-C Anhedonia T-Scores

Figure 2.

Estimated Trajectories of Orbital Frontal Cortex (OFC) Thickness Over Time by CDI-C Anhedonia T-Scores

As shown in Supplemental Table 1, even after accounting for baseline anhedonia, age × anhedonia remained significantly associated with orbital frontal cortex volume and orbital frontal cortex thickness. Further, findings also remained significant after inclusion of MDD core score (without anhedonia).

The anhedonia by age interaction in the multilevel linear model of striatal volume (residualized for whole brain minus striatum) was non-significant (p=0.69).

Trajectories of Orbital Frontal Cortex volume and substance use at follow-up

Alcohol use

As shown in Table 3, residualized (for race, sex, MDD core, anhedonia, age × anhedonia) intercepts from the multilevel linear models of orbital frontal cortex volume and orbital frontal cortex thickness significantly and negatively predicted the alcohol use frequency at age 12 (i.e., intercept). Change in alcohol frequency (i.e., slope) was also negatively correlated with orbital frontal cortex thickness (p=0.0057) but not volume (p=.47). There was no effect on the intercept or slope of the zero-inflated component (i.e., change in volume or thickness did not predict alcohol abstinence or initiation).

Table 3.

Zero-Inflated Poisson Regression Models of Alcohol Use (≥5 Drinks) Frequency Categories by Individual Subject Intercepts and Slopes from Multilevel Linear Models (MLM) of Orbital Frontal Cortex (OFC) Volume Residuals and OFC Thickness by Time-Varying CDI-C Anhedonia (N=135)

	Estimate	SE	t	p
Zero-inflated: alcohol abstinence
Intercept	8.0334	5.3138	1.51	0.1306
Intercept from MLM of OFC volume residuals	−0.2931	0.2261	−1.30	0.1949
Caucasian race	1.3022	0.8949	1.46	0.1456
Male gender	−0.4264	0.7879	−0.54	0.5884
Age at last assessment	−0.5672	0.3407	−1.66	0.0959
Continuous: frequency of use
Intercept	−3.1602	2.4822	−1.27	0.2030
Intercept from MLM of OFC volume residuals	−0.2063	0.0705	−2.93	0.0034
Caucasian race	−0.0643	0.3450	−0.19	0.8521
Male gender	0.2422	0.3298	0.73	0.4628
Age at last assessment	0.1944	0.1529	1.27	0.2037
Zero-inflated: alcohol abstinence
Intercept	8.5058	4.7860	1.78	0.0755
Slope from MLM of OFC volume residuals	7.5032	4.1092	1.83	0.0679
Caucasian race	0.2992	0.6811	0.44	0.6604
Male gender	0.0636	0.7470	0.09	0.9322
Age at last assessment	−0.5182	0.3036	−1.71	0.0879
Continuous: frequency of use
Intercept	−3.0709	2.3429	−1.31	0.1900
Slope from MLM of OFC volume residuals	1.0316	1.4191	0.73	0.4673
Caucasian race	−0.5545	0.3014	−1.84	0.0658
Male gender	0.5337	0.3106	1.72	0.0858
Age at last assessment	0.2057	0.1425	1.44	0.1487
Zero-inflated: alcohol abstinence
Intercept	22.8310	13.4559	1.70	0.0897
Intercept from MLM of OFC thickness	−1.5359	1.4005	−1.10	0.2728
Caucasian race	0.6745	0.9768	0.69	0.4899
Male gender	−0.9416	1.1661	−0.81	0.4194
Age at last assessment	−0.4434	0.4822	−0.92	0.3578
Continuous: frequency of use
Intercept	12.2404	4.1696	2.94	0.0033
Intercept from MLM of OFC thickness	−1.5851	0.3743	−4.23	<0.0001
Caucasian race	−0.4532	0.2956	−1.53	0.1252
Male gender	0.1746	0.3350	0.52	0.6021
Age at last assessment	0.3061	0.1617	1.89	0.0583
Zero-inflated: alcohol abstinence
Intercept	3.1761	14.7041	0.22	0.8290
Slope from MLM of OFC thickness	−36.9453	76.1419	−0.49	0.6275
Caucasian race	1.1750	1.5443	0.76	0.4467
Male gender	−0.3272	1.2195	−0.27	0.7884
Age at last assessment	−0.6618	0.4954	−1.34	0.1816
Continuous: frequency of use
Intercept	−9.3125	4.5857	−2.03	0.0423
Slope from MLM of OFC thickness	−33.4619	12.0919	−2.77	0.0057
Caucasian race	−0.2361	0.3057	−0.77	0.4400
Male gender	0.3532	0.3628	0.97	0.3302
Age at last assessment	0.2399	0.1890	1.27	0.2042

Marijuana use

Residualized orbital frontal cortex volume and orbital frontal cortex thickness were significantly associated with the slope for the zero-inflated component, i.e., onset of marijuana use, with positive associations for volume (p=0.0009) and particularly strong negative effects for thickness (p<0.0001; Table 4). There was also evidence that change in orbital frontal cortex thickness (but not volume) was negatively associated with change in frequency of marijuana use (i.e., slope). Intercepts for the zero-inflated component as well as for continuous frequency were unrelated to orbital frontal cortex change.

Table 4.

Zero-Inflated Poisson Regression Models of Marijuana Use Frequency Categories by Individual Subject Intercepts and Slopes from Multilevel Linear Models (MLM) of Orbital Frontal Cortex (OFC) Volume Residuals and OFC Thickness by Time-Varying CDI-C Anhedonia (N=137)

	Estimate	SE	t	p
Zero-inflated: marijuana abstinence
Intercept	−144.9453	2.6622	−54.45	<0.0001
Intercept from MLM of OFC volume residuals	−92.7248	174.4580	−0.53	0.5951
Caucasian race	358.0357	153.6707	2.33	0.0198
Male gender	−156.1959	422.8904	−0.37	0.7119
Age at last assessment	−17.2575	50.3923	−0.34	0.7320
Continuous: frequency of use
Intercept	−6.3711	2.2635	−2.81	0.0049
Intercept from MLM of OFC volume residuals	−0.1871	0.0982	−1.91	0.0568
Caucasian race	0.3343	0.3160	1.06	0.2901
Male gender	−0.1413	0.3121	−0.45	0.6507
Age at last assessment	0.3381	0.1394	2.43	0.0153
Zero-inflated: marijuana abstinence
Intercept	8631.4828	96.6595	89.30	<0.0001
Slope from MLM of OFC volume residuals	2021.8787	610.7295	3.31	0.0009
Caucasian race	984.1515	721.9290	1.36	0.1728
Male gender	941.9319	743.8689	1.27	0.2054
Age at last assessment	−677.0268	46.0206	−14.71	<0.0001
Continuous: frequency of use
Intercept	−5.5617	2.4598	−2.26	0.0238
Slope from MLM of OFC volume residuals	−0.8607	1.7057	−0.50	0.6138
Caucasian race	0.0247	0.2817	0.09	0.9301
Male gender	0.3570	0.2889	1.24	0.2166
Age at last assessment	0.2735	0.1519	1.80	0.0718
Zero-inflated: marijuana abstinence
Intercept	−977.6172	201.3198	−4.86	<0.0001
Intercept from MLM of OFC thickness	262.9481	581.2684	0.45	0.6510
Caucasian race	−385.6956	48.2515	−7.99	<0.0001
Male gender	−62.2419	364.4789	−0.17	0.8644
Age at last assessment	−119.4254	420.1010	−0.28	0.7762
Continuous: frequency of use
Intercept	2.1503	4.5573	0.47	0.6370
Intercept from MLM of OFC thickness	−0.6201	0.3927	−1.58	0.1143
Caucasian race	−0.3023	0.2952	−1.02	0.3058
Male gender	0.2429	0.2893	0.84	0.4011
Age at last assessment	0.2339	0.1513	1.55	0.1222
Zero-inflated: marijuana abstinence
Intercept	−1445.989	0.0404	−35772	<0.0001
Slope from MLM of OFC thickness	−12198	0.0069	0.00	<0.0001
Caucasian race	82.9762	0.0404	2052.86	<0.0001
Male gender	119.6468	0.0404	2960.08	<0.0001
Age at last assessment	−48.6716	0.6872	−70.83	<0.0001
Continuous: frequency of use
Intercept	−15.3208	3.5677	−4.29	<0.0001
Slope from MLM of OFC thickness	−54.2207	17.8396	−3.04	0.0024
Caucasian race	−0.0296	0.2821	−0.10	0.9165
Male gender	0.4101	0.2912	1.41	0.1590
Age at last assessment	0.3583	0.1444	2.48	0.0131

Alcohol or Marijuana use

Alcohol use and marijuana use were moderately correlated with each other (r = 0.68). Results of the zero-inflated Poisson models of alcohol /marijuana use are shown in Table 5. Residualized (for race, sex, MDD core, anhedonia, age × anhedonia) intercepts from the multilevel linear models of orbital frontal cortex volume and orbital frontal cortex thickness significantly and negatively predicted subsequent alcohol/marijuana use frequency (Supplemental Figure 1), even after accounting for zero-inflation (i.e., majority non-users). Thus, even after accounting for prior anhedonia during childhood, and its effects on orbital frontal cortex volume, subjects with smaller orbital frontal cortex volume at age 12 reported more frequent alcohol/marijuana use 3 years post scan (p=0.0038). The association between the intercept for orbital frontal cortex thickness and frequency of alcohol/marijuana use was highly significant as well (p<0.0001, Table 5), even after adjustment for sex, race and age at alcohol/marijuana use assessment. In addition, slopes from the multilevel linear model of orbital frontal cortex thickness were significantly (p<.0001) and negatively associated with alcohol/marijuana use abstinence and frequency (i.e. with sharper declines in thickness, the likelihood of onset of alcohol/marijuana use and frequency of use increased).

Table 5.

Zero-Inflated Poisson Regression Models of Summed Marijuana Use and Alcohol Use (≥5 Drinks) Frequency Categories by Individual Subject Intercepts and Slopes from Multilevel Linear Models (MLM) of Orbital Frontal Cortex (OFC) Volume Residuals and OFC Thickness by Time-Varying CDI-C Anhedonia (N=135)

	Estimate	SE	t	p
Zero-inflated: marijuana /alcohol abstinence
Intercept	8.1699	5.2383	1.56	0.1188
Intercept from MLM of OFC volume residuals	−0.2889	0.2199	−1.31	0.1888
Caucasian race	1.2379	0.8548	1.45	0.1476
Male gender	−0.3746	0.7803	−0.48	0.6312
Age at last assessment	−0.5740	0.3373	−1.70	0.0888
Continuous: frequency of use
Intercept	−3.0915	2.4523	−1.26	0.2074
Intercept from MLM of OFC volume residuals	−0.2017	0.0696	−2.90	0.0038
Caucasian race	−0.0951	0.3401	−0.28	0.7797
Male gender	0.2691	0.3288	0.82	0.4131
Age at last assessment	0.1906	0.1513	1.26	0.2076
Zero-inflated: marijuana /alcohol abstinence
Intercept	8.9732	4.7298	1.90	0.0578
Slope from MLM of OFC volume residuals	8.8264	4.7813	1.85	0.0649
Caucasian race	0.3490	0.6779	0.51	0.6067
Male gender	0.0255	0.7276	0.04	0.9720
Age at last assessment	−0.5352	0.2998	−1.79	0.0743
Continuous: frequency of use
Intercept	−2.9693	2.3376	−1.27	0.2040
Slope from MLM of OFC volume residuals	1.3176	1.5667	0.84	0.4003
Caucasian race	−0.5484	0.2978	−1.84	0.0655
Male gender	0.5180	0.3044	1.70	0.0888
Age at last assessment	0.2024	0.1418	1.43	0.1535
Zero-inflated: marijuana/alcohol abstinence
Intercept	21.3707	12.6235	1.69	0.0905
Intercept from MLM of OFC thickness	−1.3092	1.2100	−1.08	0.2793
Caucasian race	0.4801	0.9149	0.52	0.5997
Male gender	−0.7338	1.0460	−0.70	0.4830
Age at last assessment	−0.5013	0.4553	−1.10	0.2709
Continuous: frequency of use
Intercept	11.8839	4.1470	2.87	0.0042
Intercept from MLM of OFC thickness	−1.5213	0.3699	−4.11	<0.0001
Caucasian race	−0.5022	0.2995	−1.68	0.0936
Male gender	0.2242	0.3327	0.67	0.5004
Age at last assessment	0.2862	0.1570	1.82	0.0683
Zero-inflated: marijuana /alcohol abstinence
Intercept	−1525.124	21.6553	−70.043	<0.0001
Slope from MLM of OFC thickness	−13151	7.6639	−1716.0	<0.0001
Caucasian race	186.4047	364.0409	0.51	0.6086
Male gender	30.3424	76.6287	0.40	0.6921
Age at last assessment	−54.3685	24.3988	−2.23	0.0259
Continuous: frequency of use
Intercept	−16.5681	2.7172	−6.10	<0.0001
Slope from MLM of OFC thickness	−64.1517	13.0766	−4.91	<0.0001
Caucasian race	−0.4468	0.2225	−2.01	0.0446
Male gender	0.5044	0.2202	2.29	0.0220
Age at last assessment	0.3818	0.1090	3.50	0.0005

Supplemental analyses found no association between age at initiation of alcohol/marijuana use and intercepts or slopes for orbital frontal cortex volume residuals or thickness (Supplemental Table 2). Results for other definitions of alcohol/marijuana use (Supplemental Table 3: marijuana use with alcohol use ≥ 1drink frequency; Supplemental Table 4: Marijuana use with alcohol use intoxication frequency) were highly similar to those for the primary phenotype of marijuana use with alcohol use ≥ 5 drink frequency.

Even after controlling for prior marijuana or alcohol use (i.e., at scans during childhood that were concurrent with orbital frontal cortex assessment), residualized intercepts for orbital frontal cortex volume, and both intercepts and slopes for orbital frontal cortex thickness remained associated with alcohol/marijuana use frequency during adolescence (i.e., 3 years post scan 3).

In contrast, MDD core score at the later follow-up was not significantly associated with residualized (for race, sex, earlier MDD core score, anhedonia, age × anhedonia) intercepts or slopes from multilevel linear models of orbital frontal cortex volume (intercept p=0.8152, slope p=0.3588) or orbital frontal cortex thickness (intercept p=0.4112, slope p=0.8174) controlling for race, sex, and age.

Anhedonia at scan 1 and 2 were not correlated, while anhedonia at scan 3 was weakly correlated with alcohol/marijuana use frequency (r=0.19, p=0.0474). Consistent with this, intercepts and slopes for orbital frontal cortex volume and thickness extracted from models residualizing for race and sex, but not for anhedonia or the age × anhedonia term, were also significantly and similarly associated with alcohol/marijuana use (Supplemental Table 5). Similar models (without anhedonia) were run for striatum and were non-significant. These findings suggest that while prior anhedonia may play a role in the relationship between the orbital frontal cortex and later alcohol and marijuana use, some of the relationship between the orbital frontal cortex and substance use arises independent of anhedonia.

Discussion

Implications for Brain Development

The roles of the orbital frontal cortex and the striatum in processing of rewarding experiences have been relatively well established (4, 39). Deficits in reward processing are known to be associated with risk for SUD and MDD in adolescence (1, 3). The findings presented here from a longitudinal neuroimaging study of depressed and healthy preschoolers that continued into early adolescence, extend this literature by demonstrating that the developmental trajectory of orbital frontal cortex volume and anhedonia, co-vary together over school age and early adolescent development. Notably, this effect was driven by thickness of the orbital frontal cortex and not surface area. This finding suggests that the structure of this region and its reward processing function are linked across childhood development. Of interest, this relationship was not found in the striatum despite its established role in reward processing. It was also notable that this effect was specific for anhedonia, as this relationship remained even when MDD severity without anhedonia was accounted for in the model. Further it was also specific for the measures of anhedonia proximal to the measures of brain structure, as the findings remained even after controlling for anhedonia measured at baseline during the preschool period. This finding suggests a close dynamic interplay between anhedonia and orbital frontal cortex thickness across middle childhood development.

Study findings suggest that there is a close link between experiences of anhedonia and orbital frontal cortex thickness as they change across school age and early adolescent development. It is notable that this is occurring during the developmentally normative process of pruning and myelination and associated gray matter thinning, known to begin in early adolescence (40). The finding of effects driven by thickness and not surface area is not surprising given that development and variation in cortical thickness and surface area are driven by dissociable genetic (41–44), evolutionary (45) and neurobiological mechanisms (46–49) and show different trajectories across neurodevelopment (50–56). For example, very early in life, surface area expansion reflects at least in part the generation of cortical columns (46–48), while thickness is thought to reflect the creation of neurons within these columns (46). In healthy children, there is evidence that thickness changes may be the primary contributor to developmental changes in volume during adolescence (50), hypothesized to reflect at least in part synaptic pruning (57–59). Thus, though speculative, it is possible that our findings for orbital frontal cortex thickness could reflect processes related to variation in synaptic modification. However, animal research or human developmental data that focus on Diffusion Tensor Imaging (DTI) are needed to more directly test this hypothesis. Interestingly, the finding of both steeper rates of growth and subsequent thinning have been reported in a number of risk conditions and has been dubbed an “acceleration/deceleration” pattern of brain maturation. Such a pattern has been observed in high risk states (e.g., early life institutionalization and other forms of extreme stress) described in some mental disorders (60, 61).

Implications for Substance use and Substance Use Disorder Risk

Of clinical significance, changes in orbital frontal cortex volume and thickness were related to frequency of alcohol and marijuana use (both individually and as a composite) and they were not related to later risk for depression. This finding is consistent with the role of the orbital frontal cortex in substance use previously reported (14, 15) and suggests it is specific to SUD rather than depressive outcomes. Due to its role in reinforcement learning and decision-making related to valuation of reward, the orbital frontal cortex has been more notably implicated in the development of cue-reactivity and drug – seeking in the latter stages of addiction (14). The orbital frontal cortex is known to be enriched for CB1 receptors and is shown to modulate goal-directed behavior (62). This makes it a specific region of interest for marijuana and alcohol (63, 64)-related outcomes. Accordingly, atypical orbital frontal cortex functional connectivity has been noted in heavy marijuana and alcohol (13, 65, 66) users (16, 35).

While most studies have emphasized changes in orbital frontal cortex as a consequence of prolonged substance use, at least three studies suggest that variability in orbital frontal cortex volume and thickness may precede substance use. Cheetham et al. (2012) found that orbital frontal cortex volume at age 12 was associated with onset of marijuana use by age 16. Further, Malone and colleagues found that while orbital frontal cortex volume was negatively related to alcohol use in a dose-dependent manner, these differences were not significant within pairs of twins who differed in their extent of alcohol use suggesting that genetic and environmental predisposing factors for substance use that twins were matched for also played an important role in the risk relationship. In addition, one study suggested that prenatal exposure to tobacco smoking might moderate the relationship between orbital frontal cortex thinning and drug experimentation, again underscoring the importance of considering links between orbital frontal cortex structure and drug use in addition to causal effects of substance use on the orbital frontal cortex (67).

Our finding is also important because it provides a compelling test of an alternate structural brain-based hypothesis for the relationship between substance use in youth and anhedonia or depression (68). While most studies have proposed self-medication as the primary pathway linking depression/anhedonia and substance use, our longitudinal analysis suggests that anhedonia-related developmental changes in orbital frontal cortex structure may contribute to future escalations in substance use. Further, the fact that these trajectories did not predict depressive outcomes lends further credence to this developmental neurobiological rather than self-medicating model. However, associations between volumetric change in orbital frontal cortex and alcohol/marijuana use were also evident in models that did not include anhedonia, indicating that there is also an independent relationship between orbital frontal cortex volumes and substance use in youth. Regardless of whether childhood anhedonia mediates the relationship, our study provides compelling support for the role of dynamic developmental changes in orbital frontal cortex structure and substance use during adolescence.

Limitations of the study include the availability of only 3 scan waves, as ≥ 5 waves would allow for a more powerful platform to disentangle the direction of effects. However, the availability of 3 scan waves across adolescence is more than has been commonly available in the literature to date. The use of a sample enriched for depressed symptoms in the preschool period may limit generalizability of the findings suggesting that similar investigations should be conducted in community samples. However, the fact that this sample would be at uniquely high risk for depression in adolescence and that these trajectories failed to predict adolescent depression is striking. While key variables were controlled in this analysis, medication exposure and co-morbid disorders were not accounted for. Future studies should investigate neural measures of reward processing during fMRI in the orbital frontal cortex and related areas simultaneous to measures of structure across time to further elucidate these relationships. In addition, given the centrality of alterations in reward processing to mental disorders including depression and substance use disorder, how these change trajectories predict later measures of response to reward in early adulthood would be of critical interest to elucidate the developmental neurobiology of these disorders.

Study findings harness longitudinal developmental data to provide novel mechanistic pathways linking the developmental trajectory of the orbital frontal cortex and experiences of anhedonia to later risk for alcohol and marijuana use and abuse. These findings extend the available literature on the role of alterations in reward response in these risk trajectories and suggest specific and novel developmental targets for prevention.