Protocol to study neurobiological pathways to adverse patterns of substance use in adolescents with anxiety

Abstract

Anxiety is among the most common clinical challenges in adolescence and serves as both a potential precursor and exacerbator of substance use disorder (SUD) during a developmental period when substance initiation is prevalent. Yet outcomes are markedly heterogenous: some studies suggest multifinality in adverse patterns of substance use (APSU) — defined as developmentally atypical trajectories of substance involvement, distinct from but predictive of later SUD—among individuals with anxiety. The sources of variability remain unclear; thus, symptom detection alone provides limited insight into the underlying pathophysiology, predictability, progression, and preventability of anxiety-linked trajectories to APSU. This paper describes an ongoing 5-year study (National Institute on Drug Abuse-R01) that longitudinally examines maturational change in neural connectivity and stress physiology, in combination with risk and resiliency factors, to characterize pathways through which anxiety symptoms may lead to APSU in adolescents. Such longitudinal studies that track the evolution of predictors, mechanisms, and outcomes hold promise for informing the development of precision-based preventive and treatment interventions, with potential to enhance effectiveness and reach a broader population of youth.

1. Introduction

Elevated levels of anxiety symptoms in adolescents (with >30% of adolescents meeting diagnostic criteria for anxiety disorders across this developmental period) are associated with risk for adverse patterns of substance use (APSU) in a substantial subset of youth. APSU refers to atypical trajectories of substance involvement characterized by early initiation, rapid escalation in frequency or quantity, experimentation with multiple substances, and use in high-risk contexts which, together, increase the likelihood of transitioning to substance use disorders (SUD) in late adolescence or adulthood. Elucidating the pathophysiology of anxiety-related APSU is, thus, critical for accurate prediction and prevention [1,2]. Emerging evidence indicates that neurobiological and stress-response systems linked to anxiety may differentiate adolescents who initiate substance use early and progress rapidly to APSU from those who follow more normative trajectories or remain abstinent [3,4]. Moreover, previous findings suggest that variations in the maturation of neural and stress-regulatory systems contribute to distinct etiologies and anxiety symptom trajectories among adolescents with other risk factors for APSU [5,6]. Precise characterization of adolescent risk trajectories is critical for designing interventions that surpass the small to moderate effect sizes observed in even the most effective current programs. The prospective, longitudinal study described herein aims to elucidate the dynamic mechanistic processes that give rise to these distinct developmental trajectories and to provide a framework for translating findings from longitudinal studies into next-generation, precision-based intervention models.

1.1. Conceptual framework

A key to understanding how anxiety symptoms increase risk for APSU may lie in disruptions to neural pathways that support executive cognitive modulation of information processing and response to threat and reward. Preliminary evidence suggests that local alterations in threat-processing circuitry underlie symptoms of anxiety. The alterations are found particularly within the medial temporal lobe and the central extended amygdala (EAc) – including the central nucleus of the amygdala (Ce) and the bed nucleus of the stria terminalis (BNST) – and are accompanied by dysregulation of stress physiology, such as heart rate variability (HRV) and salivary cortisol (CORT), during anticipation or unpredictability of threat across preclinical animal models and human subjects research [7,8]. Parallel findings from human neuroimaging studies indicate that, when confronted with threat stimuli, individuals with higher anxiety levels show reduced activation in the prefrontal cortex (PFC) and other regions within the salience network – including the anterior insula and dorsal anterior cingulate cortex – which is critical for detecting and filtering relevant internal and external information, directing attention, and regulating switches between brain states [9]. This pattern reflects diminished top-down cognitive control over emotion.

Importantly, converging addiction neuroscience research in humans suggests that similar disruptions in top-down control over subcortical systems such as the amygdala and striatal circuits are implicated in substance use and related risk-taking behaviors [10]. These findings point to a shared neurobiological mechanism across anxiety and addiction in which reduced prefrontal regulation may contribute to heightened threat sensitivity and maladaptive reward-related sensation seeking in adolescence [11,12]. However, the developmental trajectories through which internalizing symptoms (anxiety) progress towards or diverge from APSU have yet to be fully explained.

Interactions between circuits involved in anxiety with those underlying cognitive and behavioral regulation may have explanatory value for this variability. Based on existing evidence to date, our model, which is reflected in Fig. 1, posits that a breakdown in coordination between cognitive control circuity (frontoparietal, ventrolateral PFC, cingulo-opercular, inferior frontal gyrus) and threat-processing systems manifests as behavioral disinhibition, reduced cognitive control, weakened neural connectivity, and dysregulated stress physiological responses – all of which are expected to predict APSU in adolescents with elevated anxiety symptoms [11–14]. Conversely, adolescents with anxiety who do not develop APSU are expected to exhibit stronger PFC-limbic connectivity.

Fig. 1.

Conceptual framework linking anxiety symptoms to adverse patterns of substance use (APSU). The model depicts hypothesized neurobiological and physiological pathways through which weak integration between cognitive control and threat-processing circuits contribute to impulsivity, stress dysregulation, anxiety, and APSU risk.

We further hypothesize that these neural and physiological features will increasingly distinguish adolescents with anxiety symptoms who do and who do not develop APSU over time, and that more prominent sex differences will emerge in both biomarkers and behavioral outcomes. This study will delineate maturational changes in neural connectivity within and between fronto-limbic subsystems, as well as physiological responses to acute stress, in adolescents with anxiety symptoms who do and do not progress to APSU. Altered threat-processing is a hallmark of anxiety and can impair functioning and reduce well-being in adolescent sufferers. When combined with deficits in cognitive control, behavioral disinhibition, and stress dysregulation, the result can be particularly detrimental amid the biological and social transitions of adolescence, which include heightened autonomy, peer influences, and weakened parental controls. To clarify the developmental progression of these biological and behavioral risk factors, we employ dynamic circuit analytic techniques to identify distinct trajectories from anxiety to APSU relative to no APSU, with direct implications for more precisely targeted interventions.

1.2. Developmental progression of risk for APSU in the context of anxiety symptom severity

Prospective, longitudinal studies have yet to delineate the role of emergent neural and stress physiological processes in the relationship between anxiety symptoms and APSU. Prior research has examined the neural and physiological correlates of anxiety [9,15] and, to a lesser extent, their association with SUDs [13]. However, little is known about the maturation and functional organization during adolescence – the developmental period when anxiety symptoms most commonly emerge. Moreover, developmental factors may shape these circuits differently in vulnerable individuals [16]. This gap in knowledge stands in contrast to extensive literature on associations between conduct disorder (CD) and SUDs [17–19], despite the markedly higher prevalence of anxiety disorders (31.9%) compared to CD (5–9%) during adolescence [1,20]. Gaps in knowledge regarding the neurodevelopmental trajectory of these associations, however, also exist for CD [21,22].

The longitudinal study described herein is designed to identify the dynamic and differential contributions of neural circuits supporting stress-regulatory, inhibitory, and attention systems in processing emotionally salient information that predicts clinical outcomes and APSU. Our focus is on how neurobiological and neurocognitive development – specifically the maturation of cognitive control and inhibition – interacts with alterations in stress physiological processes, including anticipatory and adaptive HRV and CORT responses to acute stress, to shape anxiety trajectories that lead to APSU. Clarifying the unique developmental patterns of these circuits and their interrelationships will advance understanding of how internalizing symptoms can transition into APSU in adolescence.

Through longitudinal follow-ups, the study will also illuminate the maturational stability and predictive value of atypical stress responsivity and impaired threat processing through adolescence. By focusing on an at-risk population – adolescents with anxiety symptoms who are oversampled for high scores on a validated predictive SUD screening tool – this research will identify biological processes that differentiate anxiety trajectories leading to APSU from those that do not. Findings will inform new etiological models and help pinpoint windows of opportunity during early and mid-adolescence for designing preventive interventions more precisely targeted to these underlying neural and physiological mechanisms.

We further hypothesize that sex will modulate maturation of the stress-response systems and neural processing of threat information, thereby influencing anxiety symptom trajectories leading to APSU. Sex differences in hypothalamic-pituitary-adrenal (HPA) and hypothalamic-pituitary-gonadal (HPG) function emerge during adolescence, potentially reflecting the regulatory influence of sex hormones that contribute to heightened stress-sensitivity [23–25]. Given that anxiety is less prevalent in boys than girls, we anticipate sex-specific pathways to APSU will be observed. Understanding these differences is critical to identifying the neurodevelopmental mechanisms that underly anxiety-related risk for APSU and critical for designing more precision-based, sex-informed interventions.

1.3. Study design

Our conceptual model proposes that developmental pathways linking anxiety symptoms to subsequent APSU are associated with distinct maturation patterns of neurobiological and stress-physiological responses measurable in adolescents who are substance-naïve at baseline. Adolescents with anxiety symptoms are oversampled for APSU risk using validated screeners that measure emotion dysregulation, behavioral impulsivity, susceptibility to peer pressure, friends’ conduct problems, and other risk-relevant characteristics to define a high-APSU-risk prodrome [26]. APSU risk status is treated as a continuous trait, with empirically derived thresholds distinguishing high- and low-risk groups. This sampling strategy ensures that at least 50% of participants are at elevated risk, thereby increasing the likelihood of capturing youth who initiate and escalate substance use during this study – an event that frequently occurs prior to age 15 in youths considered high risk for SUD [27–29]. Capturing this cohort will allow for the identification of early neurobiological and physiological precursors to APSU in adolescents with anxiety symptoms using feasible and adequately powered longitudinal methods.

To test our hypotheses, we are prospectively assessing adolescents with anxiety symptoms who are substance-naïve at baseline (age 12–14) and at two additional 12-month follow-up waves. Participants complete an acute stress task outside the scanner with physiological monitoring, as well as a resting state and task-based functional magnetic resonance imaging (fMRI) session targeting inhibition, negative valence processing, working memory for emotionally salient stimuli, and attention bias. All measures are collected at each wave, and an online exit survey in Year 5 will evaluate clinical symptoms and endpoint APSU for all participants simultaneously.

Our protocol focuses on neural connectivity at rest and during task performance using a 3-Tesla MRI scanner. Graph theory is applied to functional connectivity (FC) estimates from resting-state MRI and four task-based paradigms: (1) an Emotional N-Back (Emo N-Back) task assessing the impact of emotionally salient stimuli on working memory; (2) a Stop Signal (SST) task evaluating response impulsivity and inhibition; (3) a No-Predictable-Unpredictable Threat (NPU) task to measure cued anticipation of threat; and (4) an Emotional Faces Interference Task (EFIT) determining the impact of perceptual load on response to emotional distractors – a key indicator of neural maturity. Acute stress responses (CORT, HRV) are assessed using the Trier Social Stress Task for Children, TSST-C. In addition, participants complete a comprehensive battery of clinical, cognitive, psychological, and behavioral assessments to characterize behavioral phenotypes and environmental profiles (“environtypes”). All constructs are modeled using continuous measures to capture gradations and heterogeneity in the hypothesized associations. Primary statistical models will be developed and followed by secondary analyses examining sex differences in means levels, patterns, and parameter estimates. Although behavioral and anxiety-related changes are expected across all participants, only a subset will develop APSU. Identifying malleable biomarkers that distinguish adolescents at varying levels of risk for APSU is critical to improving predictive specificity and informing early, targeted interventions.

1.4. Hypotheses

At baseline, our first aim is to assess and characterize heterogeneity in neural connectivity, task performance, and acute stress reactivity (independently and interactively) in adolescents with anxiety symptoms at elevated risk for SUD. Our three primary hypotheses are that adolescents with higher anxiety levels will exhibit one or more of the following neural activation patterns: (1) local hyperactivation within threat-processing circuitry, (2) localized alterations in threat-processing regions, and/or (3) weak connectivity between limbic threat-processing and frontal cognitive control circuitry. We further expect task performance to parallel neuroimaging findings; for example, in the EFIT task, participants with greater anxiety and weaker prefrontal-limbic connectivity will show reduced ability to ignore emotional distractors when completing a cognitive task compared with those exhibiting stronger connectivity. We also hypothesize that adolescents with more severe anxiety symptoms will exhibit increased acute physiological stress dysregulation. Finally, we anticipate that dysregulation of physiological stress responses will moderate (i.e., amplify) the relationship between fronto-limbic connectivity deficits and anxiety symptom severity. Our second aim is to determine how baseline neural, behavioral, and physiological biomarkers moderate developmental pathways from anxiety symptoms to APSU. We expect that weak connectivity between threat and cognitive control circuitry, poorer task performance, and dysregulation of stress response at baseline will strengthen associations between anxiety and APSU risk. Our third aim is to determine whether longitudinal changes in brain activation and connectivity, task performance, and physiological stress regulation indices correspond to the trajectory of anxiety symptoms and APSU. We anticipate that delays in the developmental progression of cognitive control systems will be linked to impaired inhibition, a reduced ability to shift from emotional to cognitive strategies, heightened intra-limbic hyperactivation to threat, poorer task performance, and greater acute stress reactivity – each serving as a time-varying predictor of anxiety and emergent APSU trajectories.

2. Methods

2.1. Sample selection and recruitment

Population.

Our recruitment goal is 180 adolescents (ages 12–14; 47.5% male and 52.5% female) with anxiety symptoms who are substance-use naïve at baseline. Minimal prior substance use, defined as fewer than three lifetime occasions of alcohol, nicotine, or cannabis, will be permitted to ensure inclusion of high-risk youth while avoiding the effects of heavier use on brain function. The recruitment age range is ideal because: 1) it is a developmental period during which relevant traits (e.g., anxiety symptoms, conduct problems, cognitive and psychological status) can be reliably assessed using validated instruments; 2) most participants will be post-pubertal, reducing confounding by extreme variability in pubertal status; 3) it precedes the onset of regular substance use in most youths, allowing us to distinguish precursors from consequences of APSU based on psychological and neurophysiological markers [30,31]; and 4) it is well-suited to the recruitment, assessment, and scanning procedures required by this study.

Recruitment.

Participants are being recruited from multiple sources serving adolescents with anxiety symptoms and other risk factors for APSU, using strategies proven effective in our prior studies. In addition to outreach through schools, newspapers, social media, and community flyers, we are recruiting from mental health clinics treating adolescents with anxiety symptoms and related distress. For example, we have access to electronic health records from the University of North Carolina Child and Adolescent Psychiatry Clinic to identify potential participants who have endorsed psychological symptoms. Recruitment procedures are designed to maximize racial, ethnic, and socioeconomic diversity and achieve sex balance.

Screening.

Anxiety symptoms are screened using the State-Trait Anxiety Inventory for Children (STAI-CH) [32]. Adolescents endorsing at least one of the 20 items on the “Trait” scale and responding “often” to describe “how they usually feel” are eligible.

Retention.

Participant retention is supported through a structured sample maintenance plan employing multiple contact and engagement strategies. Study staff maintain contact every six months via email and annually via postal mail to minimize attrition. Appointment reminders and seasonal or holiday greetings are sent following the baseline visit. Consent forms include permission for contact via telephone and email. Staff maintain a warm and supportive rapport to ensure a positive participant experience. Several recruitment and statistical strategies are used to minimize attrition, account for selective attrition, and adjust for subgroups that may be over or underrepresented (e.g., weighting on demographic variables, MPlus, multiple imputation).

Participants.

The current enrolled sample to date (N = 90) includes 68% female and 32% male participants with a mean age of 13.36 years (SD = 0.89, range = 12–14). Across the current sample, 61% identified as White, 17% as Black or African American, 5% as Hispanic or Latino, 4% Asian, 1% as American Indian or Alaskan Native, and 12% as multiracial. From the current sample with baseline complete (N = 74), trait anxiety scores range from 22–58 (M = 36.1, SD = 8.94). From the current sample with baseline risk assessments scored (N = 66), 62.1% are at low risk, 21.2% are at moderate risk, and 16.7% are at high risk.

2.2. Study procedures

In this longitudinal study, youth complete a battery of assessments and tasks at baseline and 12- and 24-month follow-ups. This battery includes clinical assessments and surveys, a psychosocial stress protocol during physiological monitoring, a neurocognitive battery, and an MRI scanning session (Fig. 2).

Fig. 2.

Longitudinal design to assess risk trajectories from anxiety symptoms to adverse patterns of substance use (APSU). This longitudinal study design models functional trajectories through baseline, 12-month, and 24-month follow-up, linking anxiety severity to APSU. Repeated fMRI tasks and physiology measures indicate maturation associated with adaptive versus maladaptive outcomes.

We are capturing key measures of resting-state fMRI (rsfMRI), task-based fMRI, cognitive assessment, stress physiology, and clinical/behavioral ratings to determine developmental pathways of brain function, physiology, cognition, behavior, and anxiety symptoms and their correspondence with pathways to APSU. With these analyses, we can examine the longitudinal predictive value of maturational profiles of stress physiology and neural connections of interest in adolescents with anxiety symptoms and the relationship of this maturation to APSU. All measures and their corresponding domains are presented in Table 1. Detailed instrument descriptions, including reliability, validity, and variables are available in Supplemental Table 1.

Table 1.

Summary of study measures and analytical role within the study model.

Measure	Domain (Role in Models)
Clinical and Behavior
ALEXSA-R©	Risk factors for APSU (Predictor/Moderator)
State-Trait Anxiety Inventory	State and Trait Anxiety (Latent construct of anxiety)
Multidimensional Anxiety Scale for Children 2	Dimensions of anxiety (Latent construct of anxiety)
Anxiety Sensitivity Index	Fear of anxiety (Latent construct of anxiety)
KSADS-COMP	Psychopathology diagnoses (Covariate)
Adverse Life Experiences Scale	Adverse Events (Covariate)
Perceived Stress Scale	Perceived stress (Covariate)
Behavioral Activation and Behavioral Inhibition Scales	Approach/avoidance motivation (Predictor/Moderator)
Impulsive Behavior Scale	Impulsivity (Predictor/Moderator)
UCLA Loneliness Scale	Social connectedness (Predictor/Moderator)
ABCD Youth Screen Usage Questionnaire	Screen Time (Predictor/Moderator)
Pubertal Development Scale	Physiological maturation (Covariate)
Pittsburgh Sleep Quality Index	Sleep quality (Predictor/Moderator)
Tobacco, Alcohol and Drug Use Survey	Substance use (Outcome)
Timeline Follow Back	Substance use (Outcome)
Drug Use Disorders Identification Test	Drug use severity (Outcome)
Rutgers Alcohol Problem Index	Alcohol-related problems (Outcome)
Drinking Motives Questionnaire-Revised	Alcohol motives (Outcome)
Marijuana Problems Inventory	Cannabis-related problems (Outcome)
Marijuana Motives Questionnaire	Cannabis motives (Outcome)
Penn State Electronic Cigarette Dependence Index	Nicotine dependence (Outcome)
Neurocognitive
NIH Cognitive Toolbox	Neurodevelopmental maturity & cognitive functioning (Covariates)
WASI-II Full Scale IQ-2	Estimated intelligence (Covariate)
Physiological Testing
Trier Social Stress Test for Children	Acute Stress Task (Stimulus)
Salivary CORT level (HPA activation)	Endocrine stress response (Predictor)
Heart Rate Variability	Parasympathetic autonomic nervous system responsivity (Predictor)
Neuroimaging
Stop Signal Task	Inhibition (Predictor/Moderator)
Emotional N-Back Task	Working memory (Predictor/Moderator)
Emotional Faces Interference Task	Cognitive control (Predictor/Moderator)
No-Predictable-Unpredictable Threat	Threat reactivity (Predictor/Moderator)
Resting State fMRI	Regional interactions at rest (Predictor/Moderator)

2.2.1. Surveys

Assessment of Liability and EXposure to Substance use and Antisocial behavior, Revised©. The ALEXSA-R© system is composed of evidence-based measures of many predictors and early problem behaviors. The ALEXSA-R was originally designed to psychologically assess individual high-risk youths’ profiles of risks and resiliencies, as well as to obtain clinical measures of externalizing and internalizing symptoms for tailored preventive intervention ahead of middle adolescence when risky health behaviors typically sharply increase. It has since been used for program evaluations, SU etiology research, prevention intervention, and school-wide risk factor surveys [33,34]. Psychometrics of ALEXSA-R subscales have been good to excellent for ages 7 through 19 and the average longitudinal trends of 21 ALEXSA-R subscales have been quantified for ages 8 to 16 in youth experiencing chronic stress [28,35]. ALEXSA-R measures of resiliencies and risk factors are organized into 10 domains, based on factor analyses (Disinhibition, Sensation Seeking, Self-Management, Risk Behavior Perceptions, Family Discord, Parental Fortification, Social Contagion, Social Support, School Protection, and Neighborhood Risks); clinical measures (Aggression, Anxiety, Conduct Disorder Behaviors, Delinquency, Depression, 3 Substance Use Risk Indexes); and stand-alone factors (e.g., boredom, pubertal status). The ALEXSA-R is an illustrated, audio computer-administered self-interview platform with carefully orchestrated presentation of stimuli allows it to be completed by youth with illiteracy or other reading difficulties. ALEXSA-R scoring uses standardized norms based a U.S. nationally representative sample; thus, standardized scores and percentiles permit our sample to be compared to developmental “norms” of same-aged peers.

Part of the ALEXSA-R, the Youth Risk Index© (YRI), is being used to oversample (50%) those at high risk for adolescent SUD at baseline. Use of the YRI also helps interpret results in terms of their clinical implications as this tool is used in screening and referral to intervention for at-risk youths during well-child check-ups and evaluation of school-based prevention programs [30,36]. The YRI gauges propensity for SU and other risky health behaviors broadly, including risks that may arise from internalizing symptoms or stressors. In program evaluation data collected across 19 school districts by Ridenour and colleagues, 41.4% of youth with anxiety had high-risk YRI scores (personal communication, 2025). YRI scores are largely stable over time with a 2-week test-retest reliability intraclass correlation = 0.85 and 12-month test-retest reliability intraclass correlation = 0.60 [30,36]. YRI scores prediction of processes amenable to targeted intervention is limited to programs targeting externalizing outcomes and is thus one focus of this study [36]. Oversampling based on the YRI characterizes the sample at baseline to examine the value of contextual and developmental information and maps scores to progression of biomarkers and behavior.

Anxiety.

The State-Trait Anxiety Inventory for Children (STAI-C; child report) [37] is being used in conjunction with other measures of anxiety to develop a composite indicative of anxiety severity. The Trait subscale of the measure is first used for determining eligibility while the State subscale of the measure is administered immediately prior to the fMRI scan session. Both the State and Trait subscales are administered concurrently within a battery of clinical surveys including the Multiple Dimension Anxiety Scale 2nd Edition (MASC2) (child report and parent report) [38] and the Anxiety Sensitivity Index-3 (child report) [39]. The web-based Kiddie Schedule for Affective Disorders and Schizophrenia (KSADS-COMP) [40] is also administered through semi-structured diagnostic interviews with both parents and children to assess whether participants meet DSM-5 diagnostic criteria for several disorders (Major Depressive Disorder, Persistent Depressive Disorder, Disruptive Mood Dysregulation Disorder, Bipolar I Disorder, Bipolar II Disorder, Generalized Anxiety Disorder, Panic Disorder, Social Anxiety Disorder, Obsessive-Compulsive Disorder, Conduct Disorder, Attention-Deficit/Hyperactivity Disorder, and Post-Traumatic Stress Disorder, as well as Substance Use Disorders in both follow-up waves). Clinical measures assess how an individual is feeling in the present (e.g., STAI-C), in recent times more generally (e.g., MASC2 and ASI-R), or across the lifespan (e.g., KSADS-COMP).

Stress.

The Adverse Life Experiences Scale (ALES) (child self-report and parent report) [41] is included given that the incidence of stress exposure has profound influence on stress adaptations and neutral maturation and functioning, and it will be included as a covariate if warranted. The Perceived Stress Scale (PSS) (child self-report) [42] is also used to determine the level of stress an individual has experienced over the last month.

Inhibition and Impulsivity.

The Behavioral Activation and Behavioral Inhibition Scales (BIS/BAS) (child-self-report) [43] are used to assess motivations underlying behaviors in terms of reward and punishment. The UPPS-P for Children Short Form (child self-report) [44] is used to identify personality traits that might be linked to impulsive action such as APSU including sensation seeking, negative urgency, lack of planning, and lack of perseverance.

Social Engagement.

The UCLA Loneliness Scale (child self-report) [45] measures subjective feelings of loneliness and social isolation. The Youth Screen Usage Questionnaire (child self-report, developed for the ABCD Study^®) (child self-report) [46] assesses weekday and weekend screen time including both individual screen time platform data and total usage.

Physiological Characteristics.

The Pubertal Development Scale (PDS) (child self-report) [47] indicates self-perceived physical maturation across growth in height, body hair, skin changes, and sex-specific indicators (voice changes and facial hair for males and breast growth and menstruation for females). The Pittsburgh Sleep Quality Index (PSQI) (child self-report) [48] is used to measure subjective sleep quality and disturbances over the last month.

Substance Use.

The Tobacco, Alcohol, and Drug (TAD) use interview (child self-report) is computer administered (to increase accuracy of responses) to assess APSU, including age of onset, quantity, and frequency of use for an extensive list of licit and illicit substances, cumulative exposure for each substance reportedly used, composite severity score for all substances combined, and the composite polysubstance score. The Timeline Follow Back (TLFB) (child self-report) [49] is a retrospective, calendar-based tool to measure daily substance consumption over the last 90 days which captures the quantity and frequency of substance use. The Drug Use Disorders Identification Test (DUDIT) (child self-report) [50] screens for non-alcoholic drug-related problems and drug use severity yielding a total score indicating drug dependence.

Alcohol Use.

The Rutgers Alcohol Problems Index (RAPI) (child-self report) [51] measures alcohol-related problems regarding the past year. The Drinking Motives Questionnaire-Revised (DMQ-R) (child self-report) [52] assesses motivation for alcohol consumption based on four subscales: social, coping, enhancement, and conformity.

Cannabis Use.

The Marijuana Problems Index (MPI) (child self-report) [53] is adapted from the RAPI and designed to measure problems related to cannabis use over the past year. The Marijuana Motives Questionnaire (MMQ) (child self-report) [54] is used to assess the recreational and psychological motives for using cannabis based on five subscales: social, coping, enhancement, conformity, and expansion.

Nicotine Use.

The Penn State Electronic Cigarette Dependence Index (PS-ECDI) (child self-report) [55] assesses the level of dependence on electronic cigarettes based on frequency, cravings, withdrawal symptoms, and difficulty refraining from use in restricted areas.

Calculation of APSU.

The TAD produces a continuous measure of APSU based on combinations of measures of quantity and frequency of use for each drug separately for illicit drugs, tobacco, and alcohol, and diagnostic indicators of SUDs. Following the work of Chassin et al. and Ellickson et al. [56,57] frequency is defined as the number of times respondents used each drug (a) since the prior assessment (~12 months) and (b) in the past 30 days. Then, the quantity used per occasion, measured by how much was “usually used in a single day”, is multiplied by the frequency of days used in a given time frame (e.g., past year or past month). Thus, the level of use variable is interpreted as a measure of cumulative exposure in a given month and a 12-month period. Given that additional constructs are being measured, we also plan to adapt the Caspi severity index for each drug used by creating an ordinal scale with 4 categories (0–3) each for recency of use, past frequency, amount of usual use, and age of first use [58,59]. Scores across categories are summed and then divided by 20 to produce a severity score with values ranging between 0 and 1. For a composite severity score, these values for all drugs are further combined and averaged. To produce a composite score for polydrug use, the cumulative exposure and severity scores can be multiplied by number of drugs used. Severity is modeled longitudinally to characterize APSU over time. Based on the probability that the initiation and escalation of various substances may be differentially associated with the predictors and mechanisms we are focusing on, we will further examine divergent pathways to the use of alcohol, cannabis, cocaine, opioids and other substances. This detailed examination of substance use will enable distinctions between recreational and pathologic substance use.

2.2.2. Cognitive assessments

Neurocognitive Performance.

To obtain estimates of neurocognitive functioning, we are using selected tasks from the NIH Cognitive Toolbox [60] and the Weschler Abbreviated Scale of Intelligence 2nd Edition (WASI-II) [61]. The NIH Toolbox provides estimates of both fluid and crystallized cognitive abilities, with a strong set of subtests that assess selective executive functions. This set includes a Dimensional Card Change Sort task, a Flanker Inhibitory task, a Pattern Comparison Processing Speed task, a List Sort Working Memory task, an Oral Reading Recognition task, a Picture Vocabulary task, and a Picture Sequence Memory task. Decision making, set-shifting, attention, and working memory will be assessed as potential covariates. The WASI-II provides an estimate of overall intellectual functioning. Of the four subtests, we administer two, the vocabulary and matrix reasoning tests, wherein both verbal and nonverbal abilities can be derived. Both sets of measures are highly reliable. IQ score will be considered as a potential covariate, as warranted by the data.

2.2.3. Physiological stress responses

Social Stress Task and Stress Physiology.

The Trier Social Stress Test for Children (TSST-C) is used to elicit physiological stress responses [62]. The standardized social stressor involves a brief preparation period, followed by a public speaking task (storytelling) and a cognitive challenge involving mental arithmetic-counting backward from 1023 by 13 s performed before two neutral judges. Together, these elements combine uncontrollability and social-evaluative threat to reliably activate the HPA axis stress [63]. As previously described, the TSST-C consists of four parts: a 5-minute anticipatory stress phase (speech preparation); a 5-minute speech task, a 5-minute math task, and a 45-minute recovery phase (45 min post-task). Prior to the TSST-C participants complete a 5 min rest period, after which baseline cortisol (CORT) is collected. CORT samples are then collected at four additional 15-min intervals: immediately post-task and three other times across the recovery phase. During the TSST-C, HRV event markers are placed at the following intervals: (1) before and after the 5-minute rest period, (2) at each 5-minute transition between task components of preparation, speech, and math, and (3) at three 15-minute intervals during the recovery phase.

Salivary Cortisol (CORT).

A key measure of HPA axis activation, CORT increases rapidly in response to acute stressors [64–66]. Saliva is collected with the SalivaBio collection aid and tubes, kept cold, and stored at −15 °C until assayed. Time of day is regressed out of each time point prior to computation of the area under the curve with respect to increase (AUCi) to minimize effects of diurnal variation on CORT. AUCi is calculated from residualized measures according to Pruessner et al. [67] with time between measures to calculate AUCi. Peak CORT response is identified as the highest adjusted post-task measure for an individual. Percent change baseline-to-peak (BtP) is defined as the percent change between the adjusted pre-task and peak post-task CORT collections. Nonresponders are excluded in primary models, however, more in-depth examination of nonresponders will be conducted if there are sufficient numbers to determine differential responsivity to the task based on trauma history and other variables that may influence anxiety symptoms.

Heart Rate Variability (HRV).

Defined as the variations in time between successive heartbeats, or inter beat intervals (IBIs), HRV is a reliable indicator of overall health status and a sensitive index of autonomic stress reactivity [68,69]. Statistical decomposition of HRV allows quantification of rhythmic components that reflect distinct pathways of neural regulation. The most salient components are the respiratory oscillations known as respiratory sinus arrhythmia (RSA; high frequency (HF-HRV)) and low frequency HRV (LF-HRV). RSA primarily reflects vagal regulation by myelinated vagal efferent fibers, while LF-HRV is associated with blood pressure regulation through baroreceptor activity and peripheral vasomotor control [70,71]. Also, the heart period (HP, average IBI over time) reflects the sum of neural, neurochemical, and intrinsic influences on the heart.

HRV is a responsive biomarker for the TSST.

In young adults, significant differences in HRV indices have been observed in the arithmetic and memory task in comparison to the relaxation stage [72]. Induction of psychosocial stress increases HR and decreases HRV in patients with panic disorder and healthy controls, but activation of the sympathetic and/or inhibition of the parasympathetic autonomic nervous system is more pronounced in the first group [73]. Higher HRV indicates greater cardiac vagal tone; individuals with higher resting levels of cardiac vagal tone have greater flexibility responding to stressors and can regulate emotional responses without relying on sympathetic arousal response mechanisms. The Porges-Bohrer method will be used to extract HRV components (RSA) [74]. HRV will enable assessments of links between the stress response state, anxiety symptoms, and APSU.

During the Trier Social Stress Test (TSST), adolescents typically exhibit RSA withdrawal—a decrease in vagal tone—reflecting adaptive mobilization to cope with social-evaluative threat [75,76]. However, findings in anxious youth are mixed. Some studies report exaggerated RSA withdrawal in trait-anxious or PTSD samples, suggesting heightened autonomic reactivity [77], while others observe blunted RSA changes or even RSA augmentation, indicating vagal inflexibility or compensatory regulation strategies [78]. Longitudinal evidence suggests that adolescents with high resting RSA who demonstrate RSA augmentation during TSST show decreasing anxiety trajectories over time, whereas those with RSA withdrawal exhibit increasing anxiety symptoms, highlighting vagal flexibility as a potential resilience factor [78]. Additionally, Yim et al. [79] demonstrated that experimental manipulations of TSST can systematically vary stress arousal across developmental stages, emphasizing the importance of task design in interpreting RSA responses and anxiety outcomes. These findings challenge the assumption that RSA suppression is uniformly adaptive and underscore the importance of considering both baseline RSA and dynamic RSA responses when evaluating anxiety risk.

Beyond the magnitude of RSA change, temporal dynamics during stress and recovery phases provide additional insight into anxiety vulnerability. Dynamic modeling approaches reveal that slower RSA oscillations during TSST predict elevated state anxiety, while slower oscillations during recovery are associated with higher trait anxiety [80]. Similarly, adolescents with high resting RSA exhibit greater emotional reactivity to social-evaluative stress, suggesting that vagal tone may facilitate active engagement rather than passive coping [81]. Contextual factors such as childhood adversity further moderate these associations; high resting RSA combined with robust RSA reactivity predicts accelerated recovery and lower internalizing symptoms among adversity-exposed youth [82]. Collectively, these findings indicate that RSA reactivity during TSST interacts with baseline vagal tone and environmental context to shape anxiety trajectories, supporting polyvagal theory’s emphasis on flexible vagal modulation as central to adaptive stress regulation.

To obtain the physiological data, a BIOPAC MP160 system (BIOPAC Systems, Inc., Camino Goleta, CA, USA) is used. The BIOPAC acquires physiological data at a 2000 Hz sampling frequency. The signal acquired by the BIOPAC MP160 system is ECG from a 3-lead configuration on the chest.

Three programs are used to segment, edit, and analyze the ECG data for cardiac rhythms. First, CardioPeak and Segmenter software [83,84] is used to extract the IBI, expressed as the time in milliseconds between heart beats, by subtracting the time between consecutive R-peaks from the ECG data and segmenting the IBI data by tasks. Second, CardioEdit Plus software [85] is used to review and edit the resulting IBI files for artifacts. Editing consists of adjusting incorrect R-peak identifications and/or integer arithmetic (i.e., dividing intervals between heart beats when detections of R-peak from the ECG are missed or adding intervals when spuriously invalid detections occur). Third, CardioBatch Plus software [74,86–88] is used to analyze the edited IBI files. RSA is calculated using the Porges–Bohrer method [74], which quantifies the amplitude of RSA with age-specific parameters that are sensitive to the maturational shifts in the frequency of spontaneous breathing. The Porges–Bohrer method, as applied in CardioBatch Plus, includes the following steps: (a) IBI are timed to the nearest millisecond to produce a time series of sequential heart periods; (b) sequential heart periods are resampled to 2 Hz to produce time-based data; (c) The time-based series is detrended using a cubic moving polynomial filter (MPF), 21-point for RSA that is stepped through the data to create a smoothed template and the template is subtracted from the original time-based series to generate a detrended residual series; (d) the detrended time series is bandpassed to extract the variance in the heart period pattern associated with spontaneous breathing (0.12–0.40 Hz); (e) the resulting bandpassed time series are divided into epochs (30 sec, in this case), (f) the natural logarithm of the variance of the bandpassed time series epoch is calculated as the measure of the amplitude of RSA [89], (g) epochs across tasks are averaged to obtain the RSA in that particular task. Proper treatment of the beat-to-beat heart rate pattern enables reliable assessment of cardiac vagal tone (i.e., RSA) even under conditions of changing respiratory parameters that confound less robust measures of variance [90]. These procedures are statistically equivalent to frequency domain methods (i.e., spectral analysis) for calculation of the amplitude of RSA when heart period data are stationary [91].

2.2.4. Neuroimaging

MRI acquisition and processing.

Imaging data are acquired on a 3T Siemens MAGNETOM Prisma scanner using a 64-channel head coil. Acquisition parameters are harmonized with those of the Adolescent Brain Cognitive Development (ABCD) Study to facilitate data integration across adolescent neuroimaging cohorts. Accordingly, we implemented the following MRI parameters: 1) a 3D T1-weighted magnetization-prepared rapid acquisition gradient echo scan, for cortical and subcortical segmentation; 2) a 3D T2-weighted variable flip angle fast spin echo scan, for detection and quantification of white matter lesions and CSF; 3) a high angular resolution diffusion imaging (HARDI) scan, with multiple b-values, for segmentation of white matter tracts and measurement of diffusion parameters; 4) high spatial and temporal resolution resting state and task fMRI scans [92].

Specifically, in our study, a high resolution structural T1-weighted scan (6.25 min) is first acquired for co-registration of functional (fMRI) and diffusion MRI (dMRI) images. Participants then complete two 6.1 min resting state runs scans (TR=800 ms, 454 vol per run) followed by 4 functional tasks, with the first three presented in a randomized order: (1) Stop Signal Task (SST) (2 runs of 180 trials, TR=800 ms each, 440 vol with a duration of 5.9 min per run), (2) Emotional Faces Interference Task (EFIT) (2 runs with 12 blocks of 4 trials, TR=800 ms each, 454 vol with a duration of 6.1 min per run), (3) Emotional N-Back (Emo N-Back) (2 runs, TR=800 ms each, 335 vol with a duration of 4.5 min per run) and (4) No-Predictable-Unpredictable Threat (NPU) (2 runs with 3 blocks of 3 8-s cues, TR=800 ms each, 440 vol with a duration of 5.9 min per run).

Diffusion-weighted images are collected on a 3T Siemens MAGNE-TOM Prisma scanner using a 64-channel head coil. A single-shot spin-echo EPI sequence was acquired (TR = 4200 ms, TE = 90 ms, flip angle = 90°, multiband factor = 3, bandwidth = 1700 Hz/px), yielding 81 axial slices with 1.7 mm isotropic resolution (FOV = 240 mm, matrix = 140 × 140). Diffusion weighting is applied along 102 directions using a monopolar scheme with 2 b-values: b = 0 and b = 3000 s/mm². To correct susceptibility distortions, two additional spin-echo EPI images with opposite phase-encoding directions (AP and PA; TR = 12 400 ms, TE = 89–90 ms, 1.7 mm isotropic) are acquired for use with TOPUP or equivalent preprocessing tools. The diffusion sequence duration is 8 min.

The total scan length is about 90 min. Importantly, the Framewise Integrated Real-time MRI Monitoring (FIRMM; https://firmm.io/) software is used to assess motion in real time during the scan, providing feedback during scanning and enabling additional acquisitions when excessive movement is detected.

Stop Signal Task (SST).

During the SST task, participants are instructed to indicate the direction of an arrow as pointing left or right as quickly and accurately as possible. Participants must refrain from responding when the left or right arrow is followed by an up arrow. An algorithm changes the time between the presentation of the right or left arrow and the presentation of the Stop Signal (up arrow).

Emotional Faces Interference Task (EFIT).

In the EFIT task, participants are required to identify an ‘N’ or an ‘X’ on a string of letters presented on top of neutral, fearful, and mad faces. There are two conditions: 1) low load condition, where letter strings consist of all target letters (all Xs or all Ns); and 2) high load condition, where the letter strings consist of 5 non-target letters and 1 target letter. Participants are asked to respond as quickly and accurately as possible.

Emotional N-Back Task (Emo N-Back).

For the Emo N-Back task, participants are presented with pictures of faces (happy, fearful, or neutral) and places one at a time and are instructed to determine whether each picture presented is a “Match” or a “No match”. In the 0-back, low load condition, the “match” occurs when the picture presented is the same as a fixed target picture presented at the beginning. In the 2-back, high load condition, the “match” occurs when the picture presented is the same as the one shown two pictures ago.

No-Predictable-Unpredictable Threat Task (NPU).

The NPU task asks participants to associate symbols with faces, meaning an ‘X’ will be followed by a fearful face, an ‘O’ will be followed by a neutral face, and a ‘?’ will be followed by either. Participants are told to press a button when each symbol is shown to maintain attention. The symbols and corresponding faces are shown. fMRI data is preprocessed using fMRIPrep [93], a Nipype based tool [94]. The preprocessing is performed in 3 steps. First, a B0-nonuniformity map (or fieldmap) is estimated based on two (or more) echo-planar imaging (EPI) references with FSL topup [95]. Second, the anatomical data is preprocessed. Each T1w image is corrected for intensity non-uniformity (INU) with N4BiasFieldCorrection [96], distributed with ANTs 2.5.0 [97] (RRID:SCR_004757). The T1w-reference is then skull-stripped with a Nipype implementation of the antsBrainExtraction.sh workflow (from ANTs), using OASIS30ANTs as target template. Brain tissue segmentation of cerebrospinal fluid (CSF), white-matter (WM) and gray-matter (GM) are performed on the brain-extracted T1w using fast (FSL version 6.0.1, RRID:SCR_002823) [98]. An anatomical T1w-reference map is computed after registration of 2 T1w images (after INU-correction) using mri_robust_template (FreeSurfer 7.3.2) [99]. Brain surfaces are constructed using recon-all (FreeSurfer 7.3.2, RRID:SCR_001847) [100], and the brain mask estimated previously is refined with a custom variation of the method to reconcile ANTs-derived and FreeSurfer-derived segmentations of the cortical gray-matter of Mindboggle (RRID:SCR_002438) [101].

Brain surfaces are reconstructed using recon-all (FreeSurfer 7.3.2, RRID:SCR_001847) [100], and the brain mask estimated previously is refined with a custom variation of the method to reconcile ANTs-derived and FreeSurfer-derived segmentations of the cortical gray-matter of Mindboggle (RRID:SCR_002438) [101]. Volume-based spatial normalization to two standard spaces (MNI152NLin6Asym, MNI152NLin2009-cAsym) is performed through nonlinear registration with antsRegistration (ANTs 2.5.0), using brain-extracted versions of both T1w reference and the T1w template. The following templates are selected for spatial normalization and accessed with TemplateFlow (23.1.0) [102]: FSL’s MNI ICBM 152 non-linear 6th Generation Asymmetric Average Brain Stereotaxic Registration Model [103] RRID: SCR_002823; TemplateFlow ID: MNI152NLin6Asym, ICBM 152 Nonlinear Asymmetrical template version 2009c [104] RRID: SCR_008796; TemplateFlow ID: MNI152NLin2009cAsym. Grayordinate “dscalar” files containing 91k samples are resampled onto fsLR using the Connectome Workbench [105].

Lastly, the functional MRI data is preprocessed. For each functional run, the following procedure is followed. First, a reference volume is generated using a custom methodology of fMRIPrep, for use in head motion correction. Head-motion parameters with respect to the BOLD reference (transformation matrices, and six corresponding rotation and translation parameters) are estimated before any spatiotemporal filtering using mcflirt (FSL) [106]. The estimated fieldmap is then aligned with rigid-registration to the target EPI (echo-planar imaging) reference run. The field coefficients are mapped on to the reference EPI using the transform. The BOLD reference is then co-registered to the T1w reference using bbregister (FreeSurfer) which implements boundary-based registration [107]. Co-registration is configured with six degrees of freedom. Several confounding time-series are calculated based on the preprocessed BOLD: framewise displacement (FD), DVARS and three region-wise global signals. FD was computed using two formulations following Power (absolute sum of relative motions [108] and Jenkinson (relative root mean square displacement between affines [106]). FD and DVARS are calculated for each functional run, both using their implementations in Nipype (following the definitions by Power et al. [108]. The three global signals are extracted within the CSF, the WM, and the whole-brain masks. Additionally, a set of physiological regressors were extracted to allow for component-based noise correction [109]. Principal components are estimated after high-pass filtering the preprocessed BOLD time-series (using a discrete cosine filter with 128 s cut-off) for the two CompCor variants: temporal (tCompCor) and anatomical (aCompCor). tCompCor components are then calculated from the top 2% variable voxels within the brain mask. For aCompCor, three probabilistic masks (CSF, WM, and combined CSF+WM) are generated in anatomical space. The implementation differs from that of Behzadi et al. [109] in that instead of eroding the masks by 2 pixels on BOLD space, a mask of pixels that likely contain a volume fraction of GM is subtracted from the aCompCor masks. This mask is obtained by dilating a GM mask extracted from the FreeSurfer’s aseg segmentation, and it ensures components are not extracted from voxels containing a minimal fraction of GM.

Finally, these masks are resampled into BOLD space and binarized by thresholding at 0.99 (as in the original implementation). Components are also calculated separately within the WM and CSF masks. For each CompCor decomposition, the k components with the largest singular values are retained, such that the retained components’ time series are sufficient to explain 50 percent of variance across the nuisance mask (CSF, WM, combined, or temporal). The remaining components are dropped from consideration. The head-motion estimates calculated in the correction step are also placed within the corresponding confounds file. The confound time series derived from head motion estimates and global signals are expanded with the inclusion of temporal derivatives and quadratic terms for each [110]. Frames that exceed a threshold of 0.3 mm FD or 1.5 standardized DVARS are annotated as motion outliers. Additional nuisance timeseries are calculated by means of principal components analysis of the signal found within a thin band (crown) of voxels around the edge of the brain, as proposed by Patriat et al. [111]. The BOLD time-series are resampled onto the left/right-symmetric template “fsLR” using the Connectome Workbench [105]. Grayordinates files containing 91k samples are also generated with surface data transformed directly to fsLR space and subcortical data transformed to 2 mm resolution MNI152NLin6Asym space. All resamplings can be performed with a single interpolation step by composing all the pertinent transformations (i.e. head-motion transform matrices, susceptibility distortion correction when available, and co-registrations to anatomical and output spaces). Gridded (volumetric) resamplings are performed using nitransforms, configured with cubic B-spline interpolation.

After preprocessing, several QC metrics are calculated after Ciric et al. [112], mainly QC-FC correlations. Post-processing and FC analyses are carried out using the CONN toolbox [113]. Nuisance regressors from fMRIPrep are included in the denoising step, consisting of time series from white matter and cerebrospinal fluid (with temporal derivatives and quadratic terms), six head-motion parameters (translations and rotations, with derivatives and quadratic expansions), and motion outlier regressors to censor high-motion volumes (spike regression). After regression with Ordinary Least Squares (OLS), the residual BOLD signal is band-pass filtered (0.01–0.08 Hz) to reduce physiological noise (cardiac and respiratory artifacts). Finally, voxel time series are averaged within regions of interest (ROIs), and Pearson correlations between ROI time series are computed and Fisher-transformed to z-scores for group-level analyses.

Connectivity and network analysis.

This analysis involves ROIs within networks related to threat processing (medial temporal lobe and central extended amygdala [EAc], including the central nucleus of the amygdala [Ce] and the bed nucleus of the stria terminalis [BNST]), disinhibition (amygdala and striatal circuits), and cognitive control (frontoparietal, ventrolateral PFC, inferior frontal gyrus, cingulo-opercular, including the anterior cingulate cortex subregions: dorsal, subgenual, and pregenual), defined with two commonly-used functional brain atlases: the Schaefer atlas for cortical ROIs [114] and the CIT-168 for subcortical ROIs [115] (Fig. 3). We assessed FC by correlating the fMRI timeseries across each pair of ROIs, resulting in a FC matrix for each subject for each timepoint and each task (rest, NPU, EFIT, and Stop Signal). Weighted FC matrices were submitted to graph theory network analyses. Each connectivity matrix is being transformed into a brain graph with each ROI constituting a node and correlation values constituting edges. Each brain graph will be separated into the networks of interest—threat processing, inhibition, and cognitive control. Metrics summarizing topological properties of each graph will be calculated, producing continuous measures quantifying coherence within and integration across networks. Within-module degree (WD) and participation coefficient (PC) are calculated to quantify within-network connectivity strength and across-network integration, using Matlab and Brain Connectivity Toolbox functions [116].

Fig. 3.

A. Sagital view of cortical ROIs. B. Axial view of subcortical ROIs. C. Anterolateral view of subcortical ROIs. Abbreviations: ACC, anterior cingulate cortex; Amy, amygdala; DLPFC, dorsolateral prefrontal cortex; EXA, extended amygdala; Hip, hippocampus; Ins, insula; NAcc, nucleus accumbens; OFC, orbitofrontal cortex; Pu, putamen; VMPFC, ventromedial prefrontal cortex. Figure made by extracting regions of the Harvard-Oxford atlas using FSL functions and MRIcroGL and Inkscape for image edition.

3. Data analysis plan

Our general analytic strategy includes descriptive analysis to examine means, ranges, and standard deviations indicative of general trends and patterns in predictor and outcomes measures. Redundancy among measures of the same or similar characteristics (e.g., anxiety symptoms, task performance, stress response) will be reduced by constructing composite measures to represent their common latent construct, based on confirmatory factor analyses. Regression models (aligned with the distribution of variables in a particular analysis) will be used to examine the three hypotheses with hierarchical mixed models to test the remaining hypotheses, accounting for autocorrelations within each participant using the best fitting error covariance structure. Moreover, unconditional growth-curve models will identify and account for time-related change in APSU (e.g., linear, quadratic, or cubic “growth”) [16,117–127]. Medication and other treatments will be tested in sensitivity analyses to determine whether they ought to be statistically controlled or more carefully integrated into statistical models.

At baseline, our first aim is to assess and characterize heterogeneity in neural connectivity, task performance, and acute stress reactivity (independently and interactively) in adolescents with anxiety symptoms at elevated risk for SUD. To assess baseline relationships and characterize heterogeneity amongst our variables (Hypothesis 1), regressions analyses (based on the distribution of the variables in a particular analysis) will examine the linkages among functional brain connectivity (through threat processing, impulsivity, and cognitive control fMRI tasks), task performance, and anxiety symptoms.

Regression analyses will also be used to evaluate the association between physiological stress response and anxiety symptoms at the baseline (Hypothesis 2). First, latent profile analyses will be conducted to identify the three expected neural connectivity patterns: (i) local hyperactivation in threat processing circuitry, (ii) local alterations in threat processing circuitry, and (iii) weak connectivity between threat and cognitive control circuitry. After identifying latent profiles, their associations with baseline task performance and anxiety will be evaluated using regression analysis. Interaction terms between neural connectivity and stress physiological responses will be tested to examine how the relationship between neural connectivity and anxiety symptom severity vary by stress physiological responses (Hypothesis 3) for anxiety symptoms.

Our second aim is to determine how baseline neural, behavioral, and physiological biomarkers moderate pathways from anxiety symptoms to APSU. To determine the extent to which pathways from anxiety symptoms to APSU are moderated by baseline biomarkers and task performance, data from 3 waves and exit survey will be used to estimate agebased (6 year-age range from 12 to 18 for the trajectory) growth-curve mixed models on APSU (as a continuous measure).

A series of increasingly complex models will be run in 4 steps. First, to determine the overall shape of change over time, we will estimate unconditional growth-curve models with the intercept centered at the age of 12 at wave 1 so that the start (time = 0) of the growth curve represents 12 years of age. Different model specifications will be considered, comparing models with linear change, quadratic change, and accelerated change (including linear and quadradic). The best fitting model to represent the basic developmental course of anxiety symptoms will be selected. Standard fit indices will be used to compare models, including descriptive measures of overall model fit, such as the Akaike information criterion, the Bayesian information criterion, and the sample-size adjusted Bayesian information criterion.

Second, the latent classification of neurological measures (to reflect the combined variation of threat processing, impulsivity, cognitive control), stress response, and task performance at baseline and time-varying anxiety symptoms will be added to examine their relationship with the APSU growth curves.

Third, the interaction between anxiety symptoms and each of the 3 measures, including 2-group typology, stress response, and task performance, will be added to the growth-curve model. Two-level HLM [119] for Step 3 will be specified as follows (assuming linear and quadratic growth), accounting for within individual correlations.

$$\text{Level-1 Model (Repeated-Observation Model): Outcome}{\left(\text{APSU}\right)}_{\text{ti}}={\text{π}}_{\text{0i}}{\text{+π}}_{\text{1i}}{\text{Age}}_{\text{ti}}{\text{+π}}_{\text{2i}}{\text{Age}}_{\text{ti}}^{\text{2}}{\text{+π}}_{\text{3i}}\text{;}{\text{Anxity}}_{\text{ti}}{\text{+π}}_{\text{4i}}\left[{\left(\text{two-category neurological}\text{measure or stress response or task performance}\right)}_i\text{X}{\left(\text{Anxiety}\right)}_{\text{ti}}\right]+{\text{e}}_{\text{ti}}$$

$$\text{Level-2 Model}\left(\text{Person-Level Model for i}\right){\text{: π}}_{\text{0i}}{\text{= β}}_{\text{00}}{\text{+ β}}_{\text{01}}{\left(\text{two-category neurological measure or stress response or task performance}\right)}_{\text{i}}{\text{+ r}}_{\text{0i}}{\text{; π}}_{\text{1i}}{\text{= β}}_{\text{10}}{\text{+ r}}_{\text{1i}}{\text{; π}}_{\text{2i}}{\text{= β}}_{\text{20}}{\text{+ r}}_{\text{2i}}$$

The interaction between age variables and primary predictors will be examined as well to explore whether the coefficients for the APSU curve change by the biomarkers, task performance, and heir interaction with anxiety symptoms. The random effects for the intercept and slopes will be introduced sequentially to assess the necessity to include them.

Our third aim is to determine whether longitudinal changes in brain activation and connectivity, task performance, and physiological stress regulation indices correspond to differences in trajectories of anxiety symptoms and APSU. To examine changes in biomarkers and task performance in correspondence with the trajectory of anxiety symptomatology and APSU, two-level mixed models (ages 12 to 18 years) will be specified as follows where biomarker and task performance measures are age-varying variables for the growth curves of anxiety symptoms or APSU. As with Aim 2, unconditional growth curves of APSU will be explored to determine the best fit (shape) for its trajectory.

To test the hypotheses for Aim 3, the model will be specified (assuming linear and quadratic growth) as follows:

$$\text{Level-1 Model}\left(\text{Repeated-Observation Model}\right)\text{: Outcome}{\left(\text{Anxiety or APSU}\right)}_{\text{ti}}{\text{= π}}_{\text{0i}}{\text{+ π}}_{\text{1i}}{\text{Age}}_{\text{ti}}{\text{+ π}}_{\text{2i}}{\text{Age}}_{\text{ti}}^{\text{2}}{\text{+ π}}_{\text{2i}}{\left(\text{two-category neurological measure or stress response or task performance}\right)}_{\text{ti}}{\text{+e}}_{\text{ti}}$$

$$\text{Level-2 Model}\left(\text{Person-Level Model for i}\right){\text{: π}}_{\text{0i}}{\text{= β}}_{\text{00}}{\text{+r}}_{\text{0i}}{\text{; π}}_{\text{1i}}{\text{= β}}_{\text{10}}{\text{+r}}_{\text{1i}}{\text{; π}}_{\text{2i}}{\text{=β}}_{\text{20}}{\text{+r}}_{\text{2i}}$$

To advance our understanding of sex-specific pathways to anxiety symptoms and APSU, we will first estimate fully interactive models with sex. If it is necessary, we will stratify the sample by sex to empirically test how each model pathway differs by sex. Additionally, throughout our model building, we will control for related covariates, such as SES and stressor exposures in multivariate analysis. In sum, the multilevel modeling approach will enable strong inferences about timing, cumulative adversity exposure, and growth curves of anxiety symptoms and APSU. The results generated from these models will provide tests of whether patterns of neural activation and connectivity, neurocognitive functioning, and contextual measures are predictive of diverging growth-curves of anxiety symptoms and subsequent APSU.

4. Discussion and future directions

This prospective, longitudinal study was designed to evaluate the complex relationship between anxiety symptoms and APSU, hypothesized to be mediated by neurobiological and physiological developmental processes. Variables of interest are well-defined in our sample and tracked over time in relation to APSU in a relatively large sample of 180. This design enables both within- and between-subject analyses to: (1) characterize baseline association between neurophysiological and behavioral factors in adolescents with anxiety symptoms; (2) predict divergent trajectories toward or away from APSU based on patterns of neurobiological, behavioral, and physiological response systems, (3) track developmental change in stress response systems in relation to trajectories from anxiety to APSU, and (4) differentiate potentially distinctive neural pathways that influence progression to APSU.

4.1. Implications for more precisely targeted preventive intervention

Anxiety symptoms can undermine intervention efforts aimed at preventing APSU when program components do not address the anxiety. As such, this study is designed not only to expand our knowledge about biomarkers associated with the relationship between anxiety and APSU, but also to identify developmental points at which trajectories diverge, where some youth with anxiety symptoms move towards APSU, while others do not. Identifying biomarkers that signal risk for adverse outcomes will help reveal novel targets for the next generation of precision-based interventions [128]. Although this study primarily focuses on biological “risks” that may propel youth toward APSU, it also seeks to identify compensatory neurobiological and physiological resources (i.e., strong fronto-limbic connectivity and adaptive stress regulation) that may attenuate risk and promote resilience in the context of anxiety. Given the experience-dependence and malleable nature of neurodevelopment and stress physiology, identifying characteristics that exert protective effects (e.g., effective cognitive control) will reveal opportunities to strengthen these mechanisms and avert maladaptive trajectories.

Despite substantial progress, the heterogeneity in prevention outcomes remains largely unexplained, and prevention programs yield only small to modest effect sizes on average. One reason may be prevention science’s reliance on traditional assessment methods (e.g., interviews, behavioral assessments, and direct observations) that capture observable behavior but not the underlying neural and physiological processes driving it. As a result, even refinements to programs based on identified mediators and moderators have not substantially improved intervention efficacy. Information gleaned from this study will be critical to identifying neural, physiological, and cognitive mechanisms that can be directly targeted to enhance prevention outcomes.

Elucidating individual and subgroup differences in biomarkers that map onto distinct anxiety pathways to APSU will help guide the development of intervention components tailored to the mechanisms underlying anxiety symptoms. For example, youth whose anxiety is linked to heightened connectivity in medial temporal regions may benefit most from stress-reduction or mindfulness-based approaches, whereas those whose anxiety reflects diminished PFC control may respond better to cognitive training aimed at strengthening executive regulation. Identifying such proximal biomarkers in emerging disorders is crucial for pinpointing malleable targets for early intervention, identifying subgroups differentially responsive to specific intervention components, predicting treatment outcomes, and ultimately improving outcomes. Important advances in research including neuroimaging and other biomarkers have revealed activity within relevant neural circuits that is associated with behavioral change reflecting protection from high-risk trajectories [129]. Without the integration of biomarker research, further refinement of prevention programs will overlook a primary driver of behavior – the brain. Improving effect sizes translates to the ability to design more effective interventions for a greater number of recipients, ultimately improving the health and well-being of these youths, easing the significant burdens they experience. In sum, preventative approaches informed by biological variation among subtypes of youth with anxiety symptoms promise to be vastly more effective than non-specific interventions directed toward a heterogeneous population [117].

4.2. Clinical applications for this research

In prevention science, identifying adverse patterns of adolescent substance use is critical because they function as early warning indicators—allowing intervention before neurobiological adaptation and behavioral reinforcement consolidate into addiction. As such, this hypothesis-driven study will further establish linkages between neurobiological patterns of brain activation and connectivity, and stress physiology, to gauge and use clinically as a proxy for risk status, and to inform types of interventions, and outcomes in response to interventions. The information provided can further guide replication studies using the NIH-funded Adolescent Brain Cognitive Development (ABCD) study about specific targets for intervention for a particular high-risk population. These study attributes can provide more meaningful interpretations of neural activation patterns found in the ABCD study that will significantly boost its implications for prevention.

The assessment and screening tools used in this study provide a comprehensive system for measuring risk factors for SUD that are used clinically and for program evaluation. Also, given evidence for associations between the measures these instruments assess and resting state network connectivity of the frontoparietal network [26] there is potential to construct a rare bridge between practice and neurophysiological risk. And because these instruments are standardized to the U.S. national population of 8- to 13-year-olds, the generalizability of our results to other adolescents with anxiety symptoms will increase their clinical significance. As such, our findings may help to distinguish typologies of anxious youth at differential risk for externalizing behaviors such as APSU and that require different approaches to intervention. Galán [36] recently reported that well-child patients aged 10–13 who were screened and scored in the “at risk range” and then referred to intervention were 11% less likely to initiate new substance, had a 29% reduction in alcohol and/or tobacco use, and experienced reduced anxiety (Cohen’s d = 2.62). Thus, once specific measures produced by the current battery have been isolated and reliably associated with APSU, those measures may serve as a proxy for neurobiological vulnerability – capturing cognitive, physiological, and symptom-based indicators of risk – across externalizing and internalizing pathways in a variety of settings, including clinics and schools. Thus, mapping neural substrates to other markers lends itself to a more precision-based approach to intervention. Our data will in turn permit exploration of how neurobiological vulnerability combines with other risk factors to impact youths’ trajectories toward or away from SU/SUD.

4.3. Lessons learned

Preliminary experience implementing the study has illuminated several practical and conceptual lessons. Recruitment efforts have proven unexpectedly difficult, perhaps because features of our sample, such as heightened anticipatory anxiety and avoidance of novel situations, may act as barriers to participation, particularly in the MRI environment which is unfamiliar for most youth. Moreover, a marked (albeit expected) sex difference in ease of recruitment has emerged, with female participants enrolling at far higher rates than male participants. This pattern likely reflects both the greater prevalence of anxiety symptoms among adolescent girls and the potential sociocultural factors that discourage boys from identifying or labeling symptoms of anxiety. Finally, the second wave of data collection thus far has exhibited lower reported rates of substance use than initially projected. This may signal that the current sample is at lower risk than the goal sample that is oversampled for high-risk adolescents. It may also represent generational shifts in adolescent social behavior among individuals who entered this developmental stage in the wake of the COVID-19 pandemic, which may have altered peer social contexts and parental proximity [130].

4.4. Successes and next steps

The study has achieved several noteworthy successes. To date, response rates for the second wave have been high, with 90% of families readily scheduling the follow-up visit once contacted. To reduce participant burden, study visits are offered seven days a week, including in the evenings, and include dedicated breaks for rest and a meal built into each session. Participant engagement is encouraged through creative incentives such as individualized refrigerator magnets of each child’s brain image, as well as 3D printed keychain models of each child’s brain. Targeted recruitment will be implemented moving forward through prioritizing electronic health record searches that identify adolescents with clinically significant anxiety symptoms and adolescents with high-risk characteristics. This will improve representation across the risk spectrum and increase the likelihood of capturing the divergent trajectories towards APSU that was intended in the study.

Supplementary Material

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.addicn.2026.100258.

Data availability

This is not a data paper so only one mention of partial and preliminary findings is included. Once we are further along in data collection, they will become public.