Trajectories of Alcohol Initiation and Use During Adolescence: The Role of Stress and Amygdala Reactivity

Abstract

Objective:

Early alcohol use initiation predicts onset of alcohol use disorders in adulthood. However, little is known about developmental trajectories of alcohol use initiation and their putative biological and environmental correlates.

Method:

Adolescents (N = 330) with high or low familial loading for depression were assessed annually for up to 6 years. Data were collected assessing affective symptoms, alcohol use, and stress at each assessment. Adolescents also participated in a functional magnetic resonance imaging protocol that included measurement of threat-related amygdala and reward-related ventral striatum activity.

Results:

Latent class analyses identified 2 trajectories of alcohol use initiation. Early initiators (n = 32) reported greater baseline alcohol use and rate of change of use compared with late initiators and/or current abstainers (n = 298). Early initiators reported higher baseline levels of stressful life events (p = .001) and exhibited higher amygdala (p = .001) but not ventral striatum activity compared with late initiators. Early initiators were 15.3 times more likely to have a full drink (p < .0001), 9.1 times more likely to experience intoxication (p < .0001), and 6.7 times more likely to develop an alcohol use disorder by 19 years of age compared with late initiators (p = .003).

Conclusion:

Adolescents on a trajectory of early alcohol use initiation have higher levels of stress, have increased threat-related amygdala activity, are more likely to consume a full standard alcoholic drink, are more likely to experience early intoxication, and are at a heightened risk for the onset of an alcohol use disorder.

Alcohol use disorder (AUD) is a significant public health problem in the United States, with approximately one-third of the population experiencing a life-time episode of AUD.¹ Recent data from the Monitoring the Future Study indicated that as many as 23% of eighth graders have used alcohol at some point in their lifetime and nearly 9% reported being drunk at least once, with these estimates increasing to 61% and 23% by 12th grade, respectively.² Although experimentation with alcohol during adolescence is common, it comes with significant risk; younger age at alcohol use initiation is one of the strongest predictors of lifetime AUD diagnosis and predicts a more severe chronic prognosis, even after controlling for family and demographic factors.^3–8 Given the public health significance of AUD and the critical developmental period of adolescence during which alcohol use is typically initiated, it is important to identify the risk factors and subsequent outcomes associated with early alcohol use initiation and those associated with progression to AUD.^7,9,10

Historically, externalizing symptomology has been cited as the primary developmental pathway linked to the onset of alcoholism early in life.^11,12 More recently, however, there has been a shift toward elucidating the internalizing symptomology pathway for the onset of AUD during adolescence. The importance of the internalizing pathway is underscored by the following observations: substance use to alleviate negative affect is the primary self-reported reason for addiction among treatment samples^13,14; for vulnerable subgroups of the population (ie, children of alcoholic parents), internalizing symptoms are more likely to lead to a substance use disorder (which includes AUD) compared with participants without maladaptive beliefs about alcohol’s ability to aid in alleviation of negative affect and/or compared with participants without familial history of AUD^13,15–17; and the comorbidity rates of substance use disorders and affective disorders are higher than for any other sets of psychiatric disorders.^13,18 A recent review by Hussong et al¹⁹ highlighted that clinical depression and depressive symptoms predict earlier onset of alcohol use^20–22 and are indicators of problem drinking²³ and overall use.²⁴ Interestingly, the offspring of depressed parents have been shown to be at increased risk to develop not only depression but also substance use disorders.^25,26

Within the context of considering the affective pathway to AUD, functional neuroimaging studies have targeted the corticolimbic circuit, supporting recognition and reaction to danger, with a focus on the amygdala; and a corticostriatal circuit, supporting motivation and action, with a focus on the ventral striatum (VS). Historically, reward-related hypoactivity of the VS has been observed in individuals with AUD.²⁷ Similarly, studies have found that threat-related amygdala hypoactivity might be associated with familial risk for AUD and problem drinking, possibly because of diminished recognition and reaction to the consequences of excess alcohol consumption.^28,29 Intriguingly, 1 study reported that amygdala hyperactivity is protective against stress-related problem drinking, but only if the VS also is relatively hyperactive.²⁸ Building on this work, a recent study by this group examined reward-related VS activity and threat-related amygdala activity in the expression of problem drinking and AUD in a sample of university students.²⁹ The investigators reported that relatively increased amygdala activity coupled with decreased VS activity was characteristic of an affective pathway for AUD.²⁹ Conversely, relatively increased VS activity coupled with decreased amygdala activity characterized a more impulsive pathway.²⁹ Taken together, the results of these neuroimaging studies suggest that there could be neural signatures associated with an internalizing, or affective, pathway for AUD that are distinct from an externalizing or impulsive pathway.

Longitudinal studies have identified several trajectories of alcohol use during adolescence; these have been commonly identified as a nonuser or stable low-use course, a chronic or high-use course, a course characterized by maturing out of drinking, and a later onset or increasing course.^30–33 Longitudinal studies have observed that chronic high use of alcohol experimentation is related to AUD; however, the relation between the trajectories of alcohol use during adolescence and other indices of use such as time of first full drink and/or time of first intoxication are poorly understood. Toward this end, it is unclear how experimentation with alcohol is related to consumption of a full alcohol beverage and how these behaviors in turn are related to intoxication and subsequent AUD onset.^32–34

In the present study, we sought to identify patterns of alcohol initiation and use in a cohort of adolescents at high or low familial risk for depression using trajectory analyses. Through the exclusion of a middle group, we enriched the risk for the onset of depressive disorders and, relatedly, AUDs. Then we sought to identify how trajectories of use differed with respect to putative behavioral risk factors, such as time to first full drink and time to first intoxication, and distal and proximal stressors. Further, we examined how the trajectories differed with respect to 2 neural markers of risk for problem drinking: increased threat-related amygdala activity and decreased reward- related VS activity.^28,29 Based on prior research, we hypothesized that increased exposure to stress, familial history for depression, increased threat-related amygdala activity, and decreased reward-related VS activity would be associated with greater alcohol use and possibly an early initiation of alcohol use.

METHOD

Participants

Participants in this study were recruited as part of the Teen Alcohol Outcomes Study (TAOS); sampling procedures for TAOS have been previously described.^35–39 Briefly, 1,089 youth 12 years 0 month to 14 years 11 months of age living within a 30-mile radius of the University of Texas Health Center at San Antonio were contacted and screened for being at high and low familial risk for depression. Adolescents were determined to be at high familial risk for depression if they had at least 1 first-degree and 1 second-degree relative with a lifetime history of major depression. Adolescents were identified as being at low risk for depression if they had no first-degree and minimal second-degree relatives (<20%) with a lifetime history of depression.²⁶ Once identified as meeting the age and familial risk requirements, adolescents were excluded if they met criteria for any psychiatric diagnoses at the baseline assessment (with the exception of anxiety in the high-risk group), including externalizing disorders (eg, conduct disorder, attention-deficit/hyperactivity disorder), or had already binge drank according to National Institute on Alcohol Abuse and Alcoholism criteria.⁴⁰

A total of 330 participants were recruited into the study; 164 to the low-risk group and 166 to the high-risk group. Participants were reassessed annually with diagnostic interviews and self-report measures of behavior including mood, anxiety, stress, and substance use. In this article we focus on the prospective risk to AUD onset and thus all results are reported relative to baseline measures of affective symptoms, stress, and neural reactivity. All 330 adolescents were followed at least once after baseline, with an overall mean number of 3.86 ± 1.41 assessments (Table 1 and Table S1, available online).

TABLE 1.

Participant Demographic Characteristics

	Early Initiators	Late Initiators and/or Current Abstainers	Degrees of Freedom	Test Statistic	p
Age at baseline (y)	13.32 (0.87)	13.42 (0.97)	1, 329	F = 0.32	.57
Number of interviews	4.13 (1.29)	3.84 (1.43)	1, 328	F = 1.19	.27
Gender			1	χ2 = 0.45	.50
Boys	14	149
Girls	18	149
Race			2	χ2 = 2.30	.32
White	19	167
Other	13	131
Familial risk for affective			1	χ2 = 0.01	.91
disorders
High	16	146
Low	16	152

Measures

Childhood Trauma Questionnaire.

The Childhood Trauma Questionnaire (CTQ) is a 28-item self-report measure and was used to assess exposure to 5 different types of childhood trauma: emotional abuse, physical abuse, sexual abuse, emotional neglect, and physical neglect.⁴¹ CTQ total score reflects different forms of maltreatment.

Stressful Life Events Schedule.

The Stressful Life Events Schedule (SLES) was used to assessed the occurrence of stressful life events during the past year.⁴² Use of the SLES and scoring in this study have been described previously.^35–37 Briefly, the SLES assesses the presence and occurrence of age-appropriate stressors in children and adolescents across several domains (eg, family, friends, school). Each stressor is given a subjective stress rating and an objective stress rating by a consensus panel.

Mood and Feelings Questionnaire.

Depressive symptoms from the past 2 weeks were assessed with the 32-item childreport and parent-report versions of the Mood and Feelings Questionnaire (MFQ).⁴³ Higher total scores on the MFQ reflect a greater likelihood for the diagnosis of a depressive disorder.^44,45

Youth Self-Report and Child Behavior Checklist.

Depressive symptoms were further assessed using the Affective Problems, Anxious/Depressed, Withdrawn/Depressed, and Anxious Problems subscales of the 120-item Child Behavior Checklist (CBCL) and the Youth Self-Report (YSR).^46,47 The CBCL and YSR are valid screens for affective and anxiety disorders.⁴⁸

Screen for Child Anxiety Related Disorders.

Total scores on the parent and child reports of the Screen for Child Anxiety Related Disorders (SCARED-P and SCARED-C, respectively) were used to assess anxiety symptoms. The SCARED is a 41-item inventory rated on a 3-point Likert-type scale. Higher scores on the SCARED-C and SCARED-P indicate the presence of more anxious behaviors and is a valid instrument for the screening of anxiety disorders among youth.⁴⁹

Schedule for Affective Disorders and Schizophrenia—Present and Lifetime Version.

Lifetime psychiatric disorders according to the DSM-IV-TR were assessed using the Schedule for Affective Disorders and Schizophrenia for School-Age Children—Present and Lifetime Present Episode Version (K-SADS-PL).⁵⁰ The child and parent/ guardian served as informants and summary symptom assessments based on each informant were made by the clinical interviewer. At follow-up, the K-SADS-PL was used to assess the onset of an AUD including alcohol abuse or dependence.

Substance Use Questionnaire.

Quantity, frequency, and experiences of alcohol use were examined using the alcohol section of the Substance Use Questionnaire (SUQ), which was administered at baseline and all follow-up assess-ments.⁵¹ For this study, we focused on quantity and frequency of use, age when an adolescent consumed the first full standard drink of alcohol (eg, 5 ounces of wine, 12 ounces of beer, or 1 ounce shot of liquor), and age at first intoxication. To quantify the frequency and quantity of alcohol use, we used 2 questions from the SUQ. The first item asked, “In the past 12 months, how often did you drink beer, wine, wine coolers or liquor?” The 11-point scale contained options from 0 (not at all) to 11 (several times a day). The second item asked, “Think of all the times you have had a drink in the past 12 months. How much did you usually drink each time?” Responses ranged from 0 (I didn’t drink in the past 12 months) to 13 (more than 25 drinks). The dimensional scales of quantity of use and frequency of use (over the past year) were multiplied to calculate a “use metric” consistent with our prior work.³⁶ A score of 1 on this metric is indicative of consuming approximately less than 1 can or glass of alcohol 1 to 3 times a year; a score of 10 on this metric is indicative of consuming approximately 1 drink 2 to 3 times a month or having 4 drinks 4 to 7 times a year; a score of 20 on this metric is indicative of consuming approximately 4 drinks once a month or having 3 drinks 2 to 3 times a month.³⁶ To quantify age at first drink, an item on the SUQ which asks youth, “How old were you the first time you had a drink, not just a sip or taste?,” was recorded. Age at first intoxication was extracted from an item asking, “How old were you when you first got drunk, or very high on alcohol?”

Alcohol Expectancy Questionnaire.

The Alcohol Expectancy Questionnaire (AEQ) was used to measure alcohol expectancies.⁵² The AEQ included 68 statements regarding the effects of alcohol. The Cognitive and Motor Impairment, Global Positive Changes, Increased Arousal, Relaxation and Tension Reduction, Sexual Enhancement, Changes in Social Behavior, and Improved Cognitive and Motor Abilities subscales were scored; higher scores indicate more positive expectancies.

Functional Magnetic Resonance Imaging Tasks.

At baseline, participants performed an emotional face matching task associated with robust threat-related amygdala activity. We examined amygdala activity to general threat-related signals as represented by angry and fearful facial expressions and amygdala activity to each expression independently. The latter allows for modeling of amygdala activity to an explicit interpersonal threat in the form of angry facial expressions and an implicit environmental threat in the form of fearful expressions.⁵³ Participants also performed a card-guessing task with monetary incentive to elicit reward-related VS activity. Further details are presented in Supplement 1, available online. These 2 tasks have been used extensively, including prior studies reported on by our group.^{35,37–39,54}

Data Analysis

Growth Mixture Modeling.

Growth mixture models were used to identify groups of adolescents with similar courses of alcohol use based on the quantity and frequency metric derived from the SUQ. A baseline growth model was used to establish that a linear trajectory is the best representation of change in the longitudinal data.⁵⁵ We proceeded to the selection of an optimal number of class trajectories by following an iterative sequence.^55–57

As part of this sequence, model fit was evaluated through comparisons of the Lo-Mendell-Rubin adjusted likelihood ratio test p value, Bayesian information crite-rion,^55,56,58,59 posterior probabilities, entropy, and total group size.^60,61 Model fit of a “k class” model (eg, a 4-class model) was compared with the model fit of a “k — 1 class” model (eg, a 3-class model). This process concluded and the optimal number of classes was determined when the like-lihood ratio tests were no longer significant, when overextraction was evident, or when a class model failed to converge. All models were run with 400 random starts.⁶¹

Once the groups were identified, each case was assigned to a class based on the case’s most likely class membership. Participants were grouped by modal assignment given the value of entropy greater than 0.90 for the final identified class structure.⁶² Between-group trajectory differences were examined using Welch F ratios to adjust for differences in the homogeneity of variance. Bonferroni-adjusted α levels were used to conservatively adjust for the number of differences examined.

All growth mixture model analyses were fit using Mplus 8.0.⁵⁸ Given the longitudinal nature of the data, some cases were missing owing to attrition and thus a full-information maximum-likelihood estimator was used because these models assume the data are missing at random and produce estimates based on all available data.⁶¹ Descriptive and behavioral analyses were completed using SPSS (IBM Corporation, Armonk, NY) and R 3.3.3 (R Foundation, Vienna, Austria; Supplement 1, available online).

Functional Magnetic Resonance Imaging Statistical Analyses.

Blood oxygen level-dependent functional magnetic resonance imaging data were analyzed using SPM8 (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/). Further details on quality control procedures and extraction of blood oxygen level-dependent parameter estimates from task-related functional clusters are provided in Supplement 1, available online. An analysis of variance model was used to examine the functional magnetic resonance imaging data for each task (amygdala and VS, respectively) predicted by class. Guided by our prior research,^37,38 in which associations with familial history of depression and the experience of a stressful life event were found for reactivity to fearful expressions, we chose to examine amygdala reactivity to fearful and angry facial expressions, greater than geometric shapes separately.³⁸ Reactivity to fearful expressions likely reflects an individual’s general sensitivity to an implicit environmental threat, whereas reactivity to angry expressions likely reflects sensitivity to an explicit interpersonal threat. The Welch t test was used for post hoc comparisons after the outcomes of the omnibus analysis of variance tests.

Survival Analyses

“Survival” is defined as time to an event; “death” is the onset of that event. The analyses examined the age at first full standard drink (eg, 5 ounces of wine, 12 ounces of beer, or 1 ounce shot of liquor), age at first intoxication, and age at onset for AUD. Age at last assessment was used to censor adolescents not experiencing onset of these events. The Kaplan-Meier estimator was used to estimate the unadjusted survival function across survey waves. Time-dependent Cox proportional hazards models were used to estimate unadjusted and adjusted hazard ratios for onset of alcohol-related outcome at the current wave in relation to latent class growth model-identified group membership (early initiator [EI] versus late initiator [LI] and/or current abstainer). Survival analyses were performed using SPSS.

Identification of Alcohol Use Initiation Trajectories

The models described in the Method section were fit to the longitudinal alcohol use quantity by frequency metric data presented in Figure S1, available online. For the 2-and 3-class models, we encountered convergence issues including variance estimates that were negative. In these cases, we fixed covariances to 0⁸ (Table S2, available online). The models began to fail to converge with the 3-class models, suggesting that a model with more than 3 classes was unlikely; thus, we ceased to run models with more than 3 classes (Table S2, available online).

The model that estimated differences in means between 2 classes and that restricted the intercept and slope variance for classes 2 to 0, but that allowed these variances to be freely estimated in class 1, was the best fitting model. This model had the lowest information criteria (ie, Akaike information criterion, Bayesian information criterion, Akaike Bayesian information criterion) value, had a value of 0.92 for entropy, and all 3 likelihood ratio tests supported the use of this 2-class model. After selecting this model, we evaluated the differences in parameters between the 2 groups (Table S3, available online).

Group 1 was composed of 32 adolescents whose trajectories were considered EI type with a progressive increase in alcohol use experimentation and consumption over time (Figure 1). Group 2 was composed of 298 participants who were classified as LI type, with relatively no alcohol use at baseline and a slower increase in initiation of alcohol use or continued abstinence across adolescence. All comparisons between groups were done using α 2-tailed testing to be most conservative; for analyses for which the Levene test of equality of variances suggested unequal variances between groups, we report results of the Welch t test, which pools variances across groups.

FIGURE 1.

Class-Specific Alcohol Use Trajectories Based on Estimated Means (±95%CI) for Early Initiators (Red) and Late Initiators and /or Current Abstainers (Blue)

Note: The scale for the alcohol use metric (frequency by severity) ranges from 0 to 143.

We performed a set of post hoc analyses with abstainers (n = 97) as a separate group from the LI group and found no difference in the main dependent variables. Thus, we retained the 2-group classifications as suggested by the latent class growth model.

RESULTS

The 2 groups were similar in age across all interview waves, ethnicity, gender, number of interviews, and familial risk for depression (Table 1 and Table S1, available online). Across the first 5 waves ofthe study, the EI group had greater alcohol use levels compared with the LI group. At the sixth study assessment the 2 groups had comparable levels of alcohol consumption (Figure 1 and Table S4, available online). After correction for multiple comparisons, the AEQ Increased Arousal subscale (p = .002), the AEQ Social Behavior subscale (p = .002; Table S5, available online), and overall SLES objective stress (p = .001) were significantly higher in the EI group (Table 2 and Table S6, available online, for full test statistics). The differences between the 2 groups remained through follow-up; however, they were not consistently different throughout the remaining 4 time points (Table 2 and Tables S5 and S6, available online).

TABLE 2.

Phenotypic Differences Between Early Initiators (EI), Late Intiators (LI), and/or Current Abstainers

		EI, Mean (SD)	LI and/or Current Abstainers, Mean (SD)	Welch t Test	Degrees of Freedom	p
AEQ	cognitive and motor impairment	2.72 (0.68)	2.69 (0.65)	0.06	1, 37.65	.808
AEQ	global positive changes	0.75 (0.76)	0.52 (0.86)	2.58	1, 40.33	.116
AEQ	increased arousal	1.88 (0.94)	1.29 (1.06)	10.75	1, 40.67	.002*
AEQ	relaxation and tension reduction	2.13 (1.01)	1.72 (1.20)	4.24	1, 49.70	.043
AEQ	sexual enhancement	1.31 (1.04)	1.22 (0.98)	0.20	1, 33.34	.656
AEQ	changes in social behavior	0.75 (0.80)	0.27 (0.53)	10.64	1, 34.09	.002*
AEQ	improved cognitive and motor abilities	0.16 (0.37)	0.11 (0.33)	0.40	1,36.77	.526
CBCL	anxiety problems	0.60 (1.33)	0.72 (1.08)	0.27	1, 33.33	.607
CBCL	anxious/depressed	1.65 (2.12)	1.78 (1.92)	0.12	1, 35.61	.73
CBCL	withdrawn/depressed	0.9 (1.40)	1.12 (1.82)	0.61	1, 42.12	.438
CBCL	affective problems	1.13 (2.26)	0.88 (1.62)	0.37	1, 33.46	.548
MFQ_C	summary score child report	11.34 (11.17)	9.41 (8.19)	0.90	1, 34.68	.349
MFQ_P	summary score parent report	4.53 (5.82)	3.1 (5.19)	1.77	1, 36.54	.192
SCARED_C	summary score child report	13.28 (7.97)	16.13 (10.53)	3.43	1, 43.67	.071
SCARED_P	summary score parent report	10.72 (19.25)	6.41 (6.20)	1.59	1, 31.70	.217
YSR	anxious/depressed	3.18 (2.33)	3.79 (3.26)	1.64	1, 38.56	.208
YSR	withdrawn/depressed	2 (1.98)	2.69 (2.20)	3.11	1, 35.66	.087
YSR	affective problems	2.83 (2.56)	2.81 (2.73)	0.001	1, 35.04	.974
YSR	anxiety problems	1.89 (1.42)	2.13 (1.93)	0.63	1, 37.78	.432
SLES	objective stress rating	12.42 (6.84)	7.94 (5.08)	12.55	1, 33.65	.001*
CTQ	total score	35.4 (10.52)	32.89 (6.93)	1.62	1, 31.70	.212

Note:AEQ = Alcohol Expectancy Questionnaire; CBCL = Child Behavior Checklist; CTQ = Childhood Trauma Questionnaire; MFQ_C = Mood and Feelings Questionnaire: Child Report; MFQ_P = Mood and Feelings Questionnaire: Parent Report; SCARED_C = Screen for Child Anxiety Related Disorders—Child Report; SCARED_P = Screen for Child Anxiety Related Disorders—Parent Report; SLES =Stressful Life Events Schedule; YSR = Youth Self Report.

^*Significant at p < .05 after Bonferroni correction.

The EI group had an early age of alcohol use initiation (mean 13.68, SD 0.88) compared with the LI group (mean 15.05, SD 1.67; F_1,213 = 20.41, p < .001). Of note, 29 youth in the LI group reported a lifetime initiation with alcohol but did not report frequency or severity of use. In addition, for these 29 youth, their age at first initiation was not reported. However, at initiation the 2 groups did not differ in quantity of alcohol consumed (EI, mean 2.68, SD 3.98 versus LI, mean 1.97, SD 4.92; F₁, ₂₁₄ = .59, p = .44). However, consistent with the description of EI youth as having a more rapid progressive increase in alcohol use experimentation and consumption over time, we found that EI youth escalated to higher levels of consumption than LI youth across follow-up (Table S4, available online). A total of 127 (38.3%) of the 330 adolescents in the study had a standard full drink (eg, 5 ounces of wine, 12 ounces of beer, or 1 ounce shot of liquor) at some point during the study. All EI youth had a full drink compared with 32% of LI youth. In addition, 78 (23.6%) of the 330 adolescents in the study experienced self-reported intoxication. Of the EI youth 75% had experienced intoxication compared with 18% of LI youth. Kaplan-Meier survival estimates found significant differences in the cumulative probability of consuming a first full drink (χ² = 214.81, df = 1, p < .0001) and being intoxicated (χ² = 104.86, df = 1, p < .0001 by log-rank test) in EI compared with LI adolescents (Figure 2A). Similarly, Cox proportional hazards regression showed the EI group to be 15.3 times more likely to have a full drink (95% CI 9.5–24.6, p < .0001) and to be 9.1 times more likely to be intoxicated (95% CI 5.5–15.1, p < .0001) compared with the LI group (Figure 2B). A relatively small number (n = 9; 2.7%) of the 330 adolescents in the study met full criteria for an AUD based on DSM-IV-TR criteria. AUD occurred in 5 of 298 LI youth (1.7%) and 4 of 32 EI youth (12.5%). The cumulative probability of developing an AUD in EI versus LI adolescents was statistically significant (χ² = 11.91, df = 1, p = .001; Figure 2C). Similarly, Cox proportional hazards regression showed the EI group to be 6.7 times more likely to develop an AUD (95% CI 1.8–24.9, p = .005) compared with the LI group (Figure 2).

FIGURE 2.

Kalpan-Meier Survival Curves to Incidence of (A) First Full Drink,(B)First Intoxication, and (C) Onset of Alcohol Use Disorder (AUD)

Note: The cumulative probability of having a standard drink, experiencing intoxication, or developing substance use disorder for early initiator adolescents (red) was higher than for late initiator and/or abstainer adolescents (blue).

Neural Signatures of Alcohol Initiation

Amygdala Activity in EI Versus LI Group.

Overall, the task produced robust threat-related bilateral amygdala activity (Montreal Neurological Institute [MNI] −22, −2, −18, t₂₃₀ = 15.69, k = 172 voxels,p corrected for family-wise error [pFWE-corrected] < .05; MNI 22, −4 −18, t₂₃₀ =18.96, k = 231 voxels, pFWE-corrected < .05; Figure 3A). Robust bilateral amygdala activity was found for fearful (MNI −22, −2, −18, = 14.46, k = 172 voxels, pFWE-corrected < .05; MNI 22, –2, −18, t₂₃₀ = 18.06, k = 229 voxels, pFWE-corrected < .05) and angry (MNI −22, −2, −18, t₂₃₀ = 12.15, k = 172 voxels, pFWE-corrected < .05; MNI 20, −4, −16, t₂₃₀ = 14.81, k = 231 voxels,pFWE-corrected < .05) facial expressions. Group comparisons showed that although there were no significant differences in general threat-related activity, there was higher mean activity specifically to fearful facial expressions in the EI group compared with the LI group (n = 231; 22 EI, 209 LI; Welch t_29.21 = 3.72, p = .0008; Figure 3B). Results were specific to fearful expressions, because group differences for angry expressions were not significant.

FIGURE 3.

(A) Mean Reativity to Feaful Expressions Within Anatomically Defined Amygdala Regions of Interest (p < .05,corrected for family-wise error; *p<.01). (B) Amygdala Activity for Contrast of Fearful Face Greter Than Shapes Between Early Initiators (Red) and Late Initiators and/or sCurrent Abstainers (Blue) Controlling for Multiple Comparisons (p < .01 corrected).

VS Activity in EI Versus LI Group.

The task elicited robust reward-related activity in the left VS (MNI −14, 14, −12, t₂₁₂ = 4.39, k = 51 ,voxels, ,pFWE-corrected < .05). Group comparisons showed no significant differences in VS activity between the EI and LI groups (n = 213; 20 EI, 193 LI, Welch t_27.82 =, —1.48, p = .15; Figures S2a and S2b, available online).

DISCUSSION

Here we report on trajectories and correlates of alcohol use and AUD in a sample of adolescents. We identified 2 trajectories of alcohol use initiation: EI and LI. Examination of alcohol use patterns between these groups showed that EI youth have a younger age of consumption of a standard drink and younger age at first intoxication, which contribute to a greater likelihood of an adolescent-onset AUD. Youth in the EI group reported consuming larger quantities of alcohol, a more rapid acceleration in use, and were at significantly increased risk for developing an AUD. EI youth also exhibited higher amygdala activity to implicit environmental signals of threat in the form of fearful facial expressions, had more positive expectancies regarding alcohol consumption, and had higher recent stress at baseline. Our findings suggest the importance of higher levels of stress and greater amygdala activity among adolescents who will initiate alcohol use earlier and add to our prior observations in a subset of these adolescents showing that increased regional cerebral blood flow in mesolimbic regions is associated with current and future alcohol use initiation.³⁶

Our findings are further consistent with those of Nikolova et al.²⁹ who reported that higher amygdala activity to an implicit environmental threat specifically predicted future problem drinking in response to stressful life events and AUD in university students.⁴⁰ However, the effect of higher amygdala activity observed in their study was expressed as problem drinking and AUD only in participants with relatively low reward-related VS activity. We did not observe VS activity differences between the EI and LI groups and did not see a moderating effect of VS activity (Supplement 1, available online). The absence of VS effects in our analyses could reflect developmental phenomenon because Nikolova et al. analyzed data from 18- to 22-year-old university students. Further, the difference in VS contributions to alcohol use could reflect an affective pathway to alcohol use that is independent of VS activity. It is important to consider these findings in the context of our study design. The sample was enriched for affective disorders by recruitment of adolescents with heightened familial risk for depression. Furthermore, the study design excluded adolescents with externalizing disorders and this could explain the null VS reactivity results, because adolescents with externalizing disorders or with externalizing sympto-mology have been shown to have indiscriminate and heightened VS reactivity to reward stimuli.^63,64 Future work with more diagnostically inclusive samples is needed to simultaneously evaluate the unique and shared contributions of internalizing and externalizing pathways to AUD.

Our findings are contrary to several prior reports showing that amygdala hypoactivity is associated with problem drinking and heightened familial risk for affective disorders.^53,54 There are several possible reasons for this divergence. First, prior studies did not isolate amygdala activity to different forms of threat; thus, risk-related hypoactivity might not apply to an implicit environmental threat. Second, the extant literature captures a broad and not necessarily overlapping age range of participants. Thus, risk could be associated with relative hyper-or hypoactivity across different developmental windows, as has been observed in studies of mood and anxiety.^63,64 Third, the size of our EI group was relatively small, and this could have contributed to lack of convergence with studies finding that problem drinking might be related to heightened familial risk for affective disorders. Fourth, the contribution of amygdala hypoactivity to risk for AUD might be conditional on the dynamics of other neural circuits. Interestingly, Nikolova et al.²⁸ found evidence for increased stress-related problem drinking not only in participants with relatively high amygdala and low VS activity but also in participants with the opposite pattern of relatively low amygdala and high VS activity. They demonstrated that the former risk pattern is associated with higher negative affect and possible problem drinking as a form of coping, whereas the latter pattern is associated with higher impulsivity and possibly problem drinking as a form of disinhibition and poor decision making. Thus, prior studies that reported amygdala hypoactivity among individuals at familial risk for AUD might have been in individuals with contemporaneously high VS activity that was not measured.³⁸

By excluding adolescents with externalizing sympto-mology and enriching the sample for familial risk for depression, we aimed to examine an affective pathway for alcohol initiation and AUD during adolescence. This was motivated by our interest in linking prior research establishing a relation between familial risk for depression and lifetime AUD with that showing early alcohol use initiation increases risk for lifetime AUD.^{5,9,10,24,65–67} However, we did not find independent or additive effects of familial risk for depression or internalizing symptomology on early alcohol use or AUD onset. This could be attributed to the relatively small number of EI participants in our sample and the classification of enriched risk for depression requiring only 1 parent with depression.^68,69 Secondarily, it is plausible that the risk conferred through family is masked by the age of this cohort. It is worth noting that the average age at our final follow-up was 18 years, whereas the reported average age of onset of AUD is 22 to 24 years.⁷⁰ Thus, although the EI and LI groups might not have differed in the risk for depression, over time negative affect could differentially affect the behaviors of high risk and low risk youth in each initiation group. The mediating relation between depression risk and AUD might emerge as participants enter young adulthood, at which time AUD onset and major depressive disorder onset are more likely. This likelihood is underscored by prior research showing that alcohol primarily serves as a negative reinforcement¹³ and that negative affect precedes the onset of AUD.

The finding that EI youth escalate their drinking more rapidly and consume more alcohol is consistent in part with prior research showing that escalation during adolescence leads to more negative alcohol dependence outcomes such as increased risk of alcohol-related problems and alcohol dependence.^33,71 We found that EI youth escalated alcohol use and were more likely to consume a full standard drink of alcohol and more likely to have drank alcohol to intoxication. This suggests that a trajectory of steady low use of alcohol is likely to escalate to consumption of alcohol in larger quantities and with greater frequency during adolescence. This point is further highlighted by our finding that, in addition to being 15 times more likely to consume a full drink, EI youth are almost 5 years younger than LI youth when they do so (Figure 2). In addition, EI youth are approximately 4 years younger than LI youth when they first experience intoxication, with the average age of first intoxication among EI youth who report experiencing intoxication being 15.1 years compared with 19.0 years in LI youth. Collectively, these patterns suggest that adolescents who initially engage in early initiation of alcohol use rapidly progress to consuming full drinks of alcohol to the point of intoxication. This trajectory of rapid increase of alcohol consumption across adolescence leads to an increased likelihood of AUD as can be seen in our preliminary observation that EI youth are 6 times more likely to have an AUD by 19 years of age.

Our findings that EI youth are more likely to develop an AUD has similarly been observed across numerous prior studies^5,65,72,73; our findings also corroborate prior literature, which has shown that there is more than 1 pattern of alcohol use initiation in adolescence. However, our findings diverge in that we identify 2 unique trajectories of alcohol use initiation, whereas other literature has reported on as many as 5 trajectories of alcohol use initiation.^21,30,32,33 Our identification of 2 trajectories of alcohol initiation could reflect the sampling strategy used that excluded adolescents with internalizing and externalizing symptomol-ogy disorders and was enriched for familial risk for depression. In addition, we derived an alcohol use metric based on the quantity and frequency of alcohol use; other research has suggested that there is variability in the number of latent groups identified because there are differences in metrics of use reported.³¹

Our study is not without limitations. First, a limited number of questions capturing quantity and frequency of alcohol use were used. A fuller spectrum of alcohol use questions could have identified a different pattern of alcohol initiation in this cohort. Second, a large number of scans were removed due to issues of imaging quality control. However, participants who had useable scans did not differ from participants without usable scans on demographic and clinical characteristics (Tables S7 and S8, available online), suggesting this did not bias our results in any one direction. Third, our overall sample of 330 adolescents, although large for a neuroimaging study, is relatively small for growth mixture model analyses, which depend on large samples. However, our model fit statistics suggest that although our sample is small, our overall model fit to these data is excellent. Fourth, only a small number of adolescents were followed long enough to document the initial onset of an AUD and we have a large percentage of attrition in follow-up. As with many longitudinal designs, attrition throughout the study duration is an issue in our sample and could influence our ability to detect differences between our EI and LI youth levels of alcohol consumption at the final study assessment point. We did not find that EI and LI youth had different numbers of interviews (Table 1), and levels of attrition did not differ (EI = 9.0%, LI = 10.5%; χ² = 0.188, df= 1, p = .664); however, attrition could have influenced our ability to detect differences between groups on variables of interest. Therefore, our analyses examining the contribution of early alcohol initiation and amygdala reactivity to initial AUD onset must be viewed with caution until more of the sample has been followed into young adulthood.