Multimodal Indicators of Risk for and Consequences of Substance Use Disorders: Executive Functions and Trait Disconstraint Assessed From Preadolescence Into Early Adulthood

Abstract

Risk for substance use disorders (SUDs) is hypothesized to include behavioral disinhibition, a genetically mediated inability to inhibit or regulate behavior given task demands or motivational drives. In the present study, we examined developmental trajectories of multiple indicators of behavioral disinhibition assessed from preadolescence into early adulthood among individuals with versus without alcohol, tobacco, and cannabis use disorders. Participants were a population-based sample of 1,512 male and female twins from the Minnesota Twin Family Study, prospectively assessed at ages 11, 14, 17, 20, and 24. Multimodal indicators of behavioral disinhibition included measures of executive function (visuospatial working memory accuracy, antisaccade task performance) and mother- and self-reported trait disconstraint. Multilevel modeling analyses that accounted for the repeated measures and nested nature of the twin family data were used to examine premorbid (age 11) indicators of executive function and trait disconstraint prior to the onset of any SUD symptoms, as well as changes from preadolescence into early adulthood (ages 11 to 24). Premorbid deviations evident at age 11 among individuals who subsequently developed SUDs included poorer performance on the visuospatial working memory test and higher levels of trait disconstraint. In addition, individuals with SUDs did not demonstrate developmentally normative improvements in inhibitory control (i.e., antisaccade performance did not improve) or in their levels of trait disconstraint. We conclude that these deviations in both neurocognitive and dispositional correlates of behavioral disinhibition precede onset of SUDs and may confer risk for their development, and in addition, problematic substance use may exacerbate preexisting deviations and interfere with normative developmental trajectories of executive function and trait disconstraint, with deleterious consequences for functioning.

Substance use disorders (SUDs) are defined by the physical and psychosocial consequences of the problematic use of addictive substances, including alcohol, tobacco, cannabis, and other drugs (American Psychiatric Association, 2013). SUDs are alarmingly common, with lifetime prevalence estimates as high as ~40% (Hamdi and Iacono, 2014; Kessler et al., 2005; Moffitt et al., 2010). Substance misuse and SUDs have tremendous public health and individual implications. SUDs cost billions of dollars each year in lost productivity, medical complications, treatment and prevention services, and crime and incarceration (Bouchery et al., 2011; Cartwright, 2008). They are associated with comorbid physical and mental health problems, academic and occupational impairment, problematic interpersonal relationships, and violent victimization (e.g., Degenhardt et al., 2007; el-Guebaly et al., 2007; Hasin et al., 2007; Vaughn et al., 2010). Substance use typically onsets during adolescence. The 2016 National Survey on Drug Use and Health, a survey of adolescent and adult substance use in the United States, found that rates of substance use were low in preadolescence, increased during adolescence into early adulthood, and decreased in later adulthood (Substance Abuse and Mental Health Services, 2017a). Notably, although substance initiation during adolescence and early adulthood is normative in terms of prevalence, it is not necessarily harmless. A sizable number of adolescents aged 12 to 17 years (4%) and young adults aged 18 to 25 years (15%) meet diagnostic criteria for a SUD in the past year (Substance Abuse and Mental Health Services, 2017a), and earlier substance initiation and use predicts increased risk of developing a SUD in later adolescence and early adulthood (Substance Abuse and Mental Health Services, 2017b). Given the substantial negative implications of SUDs, there has been considerable interest in understanding factors that contribute risk for their development, as well as the consequences of SUDs for functioning in important domains. In the present paper, we considered the role of behavioral disinhibition for SUDs in adolescence and early adulthood.

Behavioral disinhibition has been defined broadly by our group and others as an inability to inhibit or regulate prepotent responses in the service of a goal (Iacono et al., 2008; Krueger et al., 2002; Young et al., 2000). This construct is conceptualized as reflecting genetic liability toward disinhibitory traits and related clinical problems (externalizing psychopathology) and is observable in brain-based and behavioral deviations measurable using multiple modalities, including neurocognitive performance measures, psychophysiological techniques, and self- and informant reports (see Iacono et al., 2008; Venables et al., 2018; Wilson et al., 2015b). Behavioral disinhibition is posited to reflect interactions between bottom-up (reward-based) mechanisms and the failure of top-down (control) mechanisms (e.g., Beauchaine & McNulty, 2013; Iacono et al., 2008). Control mechanisms include executive functions, which encompass a number of related processes, including inhibitory control, working memory, attention, decision making, error monitoring, behavioral flexibility, and emotional regulation (see Jurado and Rosselli, 2007; Miyake and Friedman, 2012). Executive functions show normative changes across development; although basic executive functioning processes are evident even in infancy and childhood, substantial refinement occurs during adolescence, ultimately yielding improved performance and increased efficiency in adulthood (see Jurado and Rosselli, 2007; López-Caneda et al., 2013).

There is ample evidence that SUDs are associated with deviations in the executive functions implicated in behavioral disinhibition (see Goldstein and Volkow, 2011; Jacobus and Tapert, 2013; López-Caneda et al., 2013; Volkow et al., 2012). Prominent contemporary models of addiction implicitly or explicitly attribute such deviations to a neurotoxic effect of substances on the brain and related functioning (cf. Goldstein and Volkow, 2011; Jacobus and Tapert, 2013; Volkow et al., 2012). However, the vast majority of research in this area has relied on crosssectional studies conducted among samples of heavy, chronic substance abusing adults, and has thus been unable to determine whether deviations reflect the consequences of substance exposure on functioning or whether deviations were actually evident prior to substance initiation or the development of problematic use.

Longitudinal studies of adolescents and young adults are informative in this regard. Behavioral disinhibition in adolescence, and particularly the failure of top-down, control mechanisms related to executive functions, may lead to a reduced ability to prevent or stop behaviors related to substance use and thereby confer risk for the development of SUDs. In addition, because of the normative maturational changes in the neural structures underlying executive functioning processes that occur during adolescence, this is likely a period of particular sensitivity to the harmful effects of substances (Casey and Jones, 2010; Steinberg, 2010)--substance exposure and the experience of SUDs during adolescence may alter or interrupt the normative development of executive functioning processes. The existing longitudinal studies indicate that premorbid deviations in varied indicators of behavioral disinhibition predict the quantity and frequency of substance use, the number of substances used, and substance-related problems and SUDs in adolescence (Aytaclar et al., 1999; Fontes et al., 2011; Khurana et al., 2013; Nigg et al., 2004; Peeters et al., 2015; Squeglia et al., 2017, 2014a; Tapert et al., 2002a). In turn, substance use, and particularly heavy use, predicts decrements in executive functioning performance and related processes in adolescence and early adulthood (Fried and Watkinson, 2005; Hanson et al., 2010; Nguyen-Louie et al., 2015; Squeglia et al., 2012, 2009; Tapert et al., 2002b; Tapert and Brown, 1999). Thus, there is evidence that behavioral disinhibition is a risk factor evident prior to the development of problematic substance use and may play a causal role in the development of SUDs. At the same time, there is also suggestive evidence that the experience of SUDs has negative implications for the normative improvements in executive functioning processes typically seen during adolescence and early adulthood above and beyond premorbid liability to executive function deviations and SUDs.

We address two primary study questions in the present paper: (1) Are premorbid deviations in indicators of behavioral disinhibition evident prior to the onset of SUDs? and (2) Are SUDs associated with alterations in the normative developmental trajectories of indicators of behavioral disinhibition from preadolescence into early adulthood? We focused specifically on alcohol, tobacco, and cannabis use disorders because these are the most commonly used and abused substances in adolescence and early adulthood (Substance Abuse and Mental Health Services, 2017a). We undertook a multimodal approach, including neurocognitive, psychophysiological, and behavioral-trait indicators, to assessing key mechanisms associated with behavioral disinhibition in a large, population-based sample of twins prospectively assessed at multiple time points from ages 11 to 24. We selected three indicators that index related but potentially distinct aspects of behavioral disinhibition, including: (1) performance on a visuospatial working memory test as a measure of updating executive function; (2) antisaccade performance in an eye tracking paradigm as a measure of inhibitory control; and (3) mother/self-reports of levels of trait disconstraint, a dispositional factor related to unreliable and impulsive behaviors. These variables are hypothesized to reflect aspects of behavioral disinhibition that may differentially relate to the neurocognitive risk for and consequences of SUDs and comorbid aggressive-delinquent forms of externalizing psychopathology (Iacono et al., 2008).

In structural models of executive function (e.g., Miyake & Friedman, 2012), indicators across neurocognitive tasks load onto a common executive function factor. Tasks requiring inhibitory control (e.g., antisaccade) load directly onto this common factor. Tasks requiring working memory updating and task-switching form respective subfactors. This model suggests inhibitory control is required for all executive functions, antisaccade is a direct measure of inhibitory control, and that working memory also reflects specific updating processes. Both working memory and antisaccade performance tap into response inhibition processes common across executive functions (ability to actively maintain the goals of a task and relevant goal-related information and effectively use this information; see Miyake and Friedman, 2012; Munakata et al., 2011). Working memory performance also encompasses specific updating abilities (facilitating effective gating and/or retrieval from long-term memory; Miyake and Friedman, 2012). Both the visuospatial working memory test and the antisaccade task we used in the present study tap into inhibitory control via their contributions to a common executive function factor (Miyake & Friedman, 2012), and the visuospatial working memory test additionally taps into processes unique to updating executive function. Questionnaire-based indicators (self- and informant reports) of behavioral disinhibition measure a trait-like (dispositional) propensity to fail to restrain behavioral impulses and affective reactions (Patrick et al., in press; Venables et al., 2018). We used the higher-order constraint factor (reversed) from Tellegen’s Multidimensional Personality Questionnaire (MPQ), which assesses impulsivity, risk taking, and unconventionality (Tellegen & Waller, 2008).

We hypothesized that premorbid deviations in these three indicators of behavioral disinhibition would be evident in preadolescence, prior to the onset of SUDs in adolescence and early adulthood, so that individuals who developed SUDs would show impaired performance on the working memory test and antisaccade task, and higher levels of disconstraint at age 11. We further hypothesized that the presence of SUDs would interfere in the normative improvements in these indicators of behavioral disinhibition that occur over development, so that individuals who developed SUDs would fail to show improved performance on the working memory test and antisaccade task, and decreasing levels of disconstraint, from ages 11 to 24.

Material and Methods

Participants and Procedures

Participants were 1,512 male and female twins from the Minnesota Twin Family Study (MTFS) (50% female). The MTFS is an ongoing community-based, longitudinal study of reared-together twins and their parents, and is one of several ongoing studies that comprise the Minnesota Center for Twin and Family Research (MCTFR); the study design and sample have been described extensively elsewhere (see Iacono et al., 2006, 1999; Iacono and McGue, 2002) and are only briefly reviewed here. The present study included a cohort of twins first recruited for participation at age 11. The MTFS design includes assessments every 3 to 5 years; the present study includes data from assessments at target ages 11 (M = 11.72, SD = 0.43), 14 (M = 14.80, SD = 0.53), 17 (M = 17.83, SD = 0.69), 20 (M = 21.10, SD = 0.82), and 24 (M = 25.01, SD = 0.90) years old, during which indicators of behavioral disinhibition were assessed, in addition to symptoms of substance abuse and dependence. Rates of retention across follow-up waves were universally high (ranging from 87% to 93% across assessments). Consistent with the demographic makeup of Minnesota during the targeted birth years (1977 to 1984), twins were predominately Caucasian (98%). This study was approved by the research ethics committee at the University of Minnesota and was performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki. All participants gave their informed consent (18 years and older) or assent (younger than 18 years) for their own participation; parents gave consent for their children’s (younger than 18 years) participation.

Measures

Behavioral disinhibition.

Multiple indicators of behavioral disinhibition were assessed using a multimodal approach that included differing aspects of executive functions, including performance on a visuospatial working memory test and antisaccade performance in an eye tracking paradigm, and mother/self-reports of trait disconstraint.

Working memory performance.

Updating executive function was assessed using a visuospatial working memory test. Participants completed a computerized version of a visuospatial working memory test (Zald and Iacono, 1998) at ages 11, 14, 17, 20, and 24. The test was conducted in a quiet, darkened room. Participants were seated comfortably and their heads were placed on an adjustable chin-forehead rest to maintain a standard 27-cm distance from a computer monitor Participants were asked to visually fixate on a small cross presented in the center of the computer screen; after 3 sec the fixation cross disappeared and a small target (an asterisk) was presented at one of 16 positions evenly distributed along the circumference of an imaginary circle 4.5 cm from the fixation point. The test included three conditions completed in three blocks that were presented in the same order to all participants, a No Delay, Delay, and Verbal Interference condition (the Verbal Interference condition was first introduced at age 14). In each condition, participants were asked to indicate the location of the target stimulus using a PXL lightpen; prior to beginning the test, participants were shown how to hold and respond with the lightpen and where to place their hand between responses. In the No Delay condition, participants received 16 targets (1 in each location on the imaginary circle) and were instructed to “Indicate the location of the dot as accurately as possible;” the target remained on the screen until the participant indicated its location using the lightpen. In the Delay condition, the target stimulus was only presented for 200 msec, after which the screen went dark for a variable delay of 0.5 and 8 sec; participants received 16 targets once for each delay time (32 targets total). Following the delay, a completely blank lightened screen appeared and participants were instructed to “Indicate the location where the dot had been as accurately as possible.” The 0.5 sec delay was selected to assess accuracy of representation prior to much degradation; the 8-sec delay was selected to assess accuracy of representation after some degradation. Target positions and delay times were intermixed in a pseudorandom order. In the Verbal Interference condition, the target stimulus was again presented for 200msec, after which the screen went dark for a delay of 8 sec. During the delay period, three- and four-letter words appeared at the center for the screen at a rate of one word every 2 sec and participants were asked to read the words aloud to ensure compliance with the task; participants received 16 targets total. The Verbal Interference condition was included to increase the attentional load and to limit the use during the delay period of the strategy of fixating on the location in which the target stimulus had appeared by requiring participants to focus on the center of the screen while reading words. Words were selected from a list of the most frequently used words in the English language (Kučera and Francis, 1967).

The pixel location of each response in each condition was recorded using automated software and the difference between the pixel location of the participant’s response and the pixel location of the actual location of the target stimulus was calculated in millimeters (see Figure 1). The mean displacement from the target in millimeters was calculated separately for each condition, adjusting for the total number of responses for that condition, at each age. Cases in which fewer than 9 targets (n = 4) or more than 16 targets (n = 2) were administered due to an equipment malfunction were excluded from further examination. Because the No Delay condition assessed hand-eye coordination and ability to locate an object in space, rather than working memory, this condition was not considered further. We examined coherence in performance across the 0.5- and 8-sec Delay and the 8-sec Verbal Interference conditions at each age by computing zero-order correlations and principal components analyses, which indicated shared variance across these conditions at each age (mean r = .48, mean 67% cumulative variance across assessments). We thus computed a psychometrically robust indicator of working memory aspects of executive functions (i.e., distance in millimeters from the target) by averaging across the three Delay (0.5, 8 sec) and Verbal Interference (8 sec) conditions separately at each age (two Delay conditions at age 11 because Verbal Interference was first introduced at age 14).

Figure 1.

Representative responses to the (A) No Delay, (B) 0.5-sec Delay, (C) 8-sec Delay, (D) 8-sec Verbal Interference conditions on the visuospatial working memory test.

Antisaccade performance.

Inhibitory control was assessed using performance on an antisaccade task. Participants completed the antisaccade task (Hallett, 1978; Malone and Iacono, 2002) at ages 11, 14, 17, 20, and 24 as part of an ocular-motor assessment battery. The ocular-motor battery included prosaccade and antisaccade tasks. The prosaccade task, in which participants were asked to make a saccadic eye movement in the direction of targets appearing at one of several eccentric locations, always immediately preceded the antisaccade task; only the antisaccade task is considered here. The ocular-motor battery was conducted in a quiet, darkened room. Participants were seated comfortably and their heads were placed on an adjustable chin-forehead rest to maintain a standard 82-cm distance from a computer monitor. Participants were asked to visually fixate on a small cross presented in the center of the computer screen with a circle subtending 0.4° of visual arc at the initial central fixation point; at unpredictable times between 1.5 and 2.5 sec, the fixation cross disappeared and a target stimulus identical to the fixation cross was presented approximately 6° to either the left or the right of the fixation point. Participants were asked to visually fixate on the fixation cross when it was presented in the center of the computer screen and to respond to the appearance of the target by moving their eyes approximately the same distance to the opposite side (i.e., an antisaccade) of the screen. The target remained on the screen for 1.5 sec, after which it disappeared and the central fixation cross reappeared, signaling the participant to return his/her gaze to the central fixation point. The task included 20 trials, with the target appearing an equal number of times to the left or right of the fixation point.

Eye movements for each response were measured using the corneal reflection technique, as well as electrooculographic (EOG) activity. Participants wore empty spectacle frames attached to an adjustable headband. An infrared light source mounted on the spectacle frames was reflected off the cornea and detected by a pair of sensors on either side of the infrared light source using an Eye Trac Model 210 monitoring system, which is linear to ± 10° with resolution of 0.25° of visual angle. EOG activity was recorded using two pairs of Ag/AGCl electrodes. The first pair of electrodes was placed superior and inferior to one eye and recorded eye blinks and other artifacts not readily observable in the infrared reflection; the second pair was placed at the outer canthi of each eye and recorded horizontal eye movements. Data from each pair of electrodes were input to a separate amplifier in a Grass Model 12A Neurodata acquisition system, amplified 5,000 times, and filtered with a bandpass of 0.10 to 100 Hz (6 dB attenuation, or half-amplitude frequency, with a roll-off of 6 dB per octave); an electrode on the shin served as a ground. All recordings were digitized at 256 Hz to 12 bits resolution.

Participant responses on each trial were coded by a trained rater as one of five mutually exclusive responses: a correct response (the participant made an initial saccade in the opposite direction of the target), a direction error (the participant fixated on the target or made an initial saccade toward the target followed by a return to midline), a self-corrected error (the participant made an initial saccade toward the target followed by a second saccade in the opposite direction of the target that went past the midline), an anticipatory response (the participant made a saccade within 80 ms of the appearance of the target, which is unlikely to have been visually guided), or no response (the participant remained fixated on the central fixation point) (see Figure 2). Participant responses were determined primarily from the infrared signal, with EOG activity used to assist with the accurate identification of saccades and coding of responses, as needed (Calkins et al., 2001). Trials coded as direction errors and self-corrected errors were considered response errors and the proportion of response errors (direction errors, self-corrected errors) to total valid responses (correct responses, direction errors, self-corrected errors) was calculated across trials (anticipatory responses and no responses were not included in calculations) to compute an indicator of antisaccade performance (i.e., percentage response errors) separately at each age. Cases in which fewer than 7 trials were coded as either correct responses or response errors (n = 67) were excluded from further examination.

Figure 2.

Representative coded responses to the (A) target on the antisaccade task, (B) anticipatory response, (C) self-corrected error, (D) direction error, (E) correct response.

Trait disconstraint.

Trait disconstraint was assessed using mother- and self reports on questionnaires. Mothers reported on participant personality using the Multidimensional Personality Ratings (MPR; Cukrowicz et al., 2006; Oliva et al., 2012; Tackett et al., 2008) at age 11 and participants reported on their own personality using a version of the Multidimensional Personality Questionnaire (MPQ; Tellegen and Waller, 2008) at ages 17 and 24. The MPR is a 34-item (1 = my child is definitely low on this trait, 4 = my child is definitely high on this trait) parent-report measure developed to approximate the adult MPQ; the version of the MPQ used is a 198-item (1 = definitely true, 4 = definitely false) self-report measure of normal-range personality traits. The MPR and MPQ yield scores on three higher-order dimensions of positive emotionality, negative emotionality, and (dis)constraint, each comprised of lower-order trait facets. The lower-order MPR/MPQ disconstraint substraits includes reflected control (reflected: lack of inhibition, impulsivity), harm avoidance (reflected: risk-taking, sensation-seeking), and traditionalism (reflected: impropriety, unconventionality). MPR disconstraint is computed as a simple sum across the lower-order subtraits and MPQ disconstraint is computed as a weighted sum across the lower-order subtraits. Representative MPR disconstraint items are “My child often does things on the spur of the moment without giving it much thought” and “My child would rather do something risky and dangerous than something boring or tedious.” Representative MPQ disconstraint items are “I often act without thinking” and “I am often not as cautious as I should be.” Internal consistency, indexed by Cronbach’s alphas, was adequate to good in the present sample (MPR: .73 at age 11; MPQ: .76 to .85 across lower-order traits at age 17, .81 to .87 across lower-order traits at age 24).

Substance misuse.

Participants reported on alcohol, tobacco, and cannabis abuse and dependence symptoms at age 17 using a version of the Substance Abuse Module (SAM; Robins et al., 1987) of the Composite International Diagnostic Interview (CIDI; Robins et al., 1988) and their mothers reported on participants’ symptoms using a version of the Diagnostic Interview for Children and Adolescents--Revised (DICA-R; Reich & Welner, 1988) updated to include DSM-IV criteria; a best-estimate procedure was used to assign symptoms if they were endorsed by either participants or their mothers. Participants also reported on abuse and dependence symptoms since the previous assessment using the SAM of the CIDI at ages 20 and 24. SUD diagnoses were assigned if participants met criteria for 2 or more DSM-IV symptoms of abuse or dependence, consistent with current DSM-5 criteria for SUDs¹. Diagnostic interviews were conducted by extensively trained interviewers with bachelor’s or master’s degrees in psychology or a related discipline. All diagnostic interviews were reviewed in case conferences with at least two advanced clinical psychology graduate students, and consensus was required prior to assigning each symptom. Computer algorithms were used to assign diagnoses at each time point. Interrater reliability for SUD diagnoses was assessed on a randomly selected subsample of 600 MTFS participants (kappas ≥ .94). The cumulative lifetime prevalence of SUDs through the age-24 assessment was n = 534 (39%) for alcohol use disorder, n = 575 (42%) for tobacco use disorder, n = 281 (21%) for cannabis use disorder, and n = 180 (12%) met criteria for all three SUDs; n = 606 (43%) of participants did not meet criteria for a SUD (for any licit or illicit substance) at any assessment point. We classified participants into SUDs groups for analyses (no SUD versus an alcohol, tobacco, or cannabis use disorder, in separate groups, and no SUD versus all three SUDs). Because SUDs groups were non-mutually exclusive, comorbid substance use was possible within each SUD group. The alcohol use disorder group included n = 346 (66%) and n = 213 (40%) participants with comorbid tobacco or cannabis use disorders, respectively; the tobacco use disorder group included n = 346 (64%) and n = 229 (42%) participants with comorbid alcohol or cannabis use disorders, respectively; and the cannabis use disorder group included n = 213 (79%) and n = 229 (83%) participants with comorbid alcohol or tobacco use disorders. Because we were interested in premorbid indicators of behavioral disinhibition (i.e., prior to the onset of problematic substance use), we excluded from further analyses n = 7 (0.5%) participants who met criteria for any substance abuse or dependence symptoms at age 11, assessed using mother reports on the DICA-R.

In addition to substance abuse and dependence symptoms, participants also reported on their use (ever used, quantity, frequency) of alcohol, tobacco, cannabis, and other substances at ages 11 and 14 using a Computerized Substance Use (CSU; McGue et al., 2014) survey and at ages 17, 20, and 24 using the SAM of the CIDI. Descriptive data for substance use at each age is presented in Table 1.

Table 1.

Descriptive Statistics for Substance Use at Each Age

Substance Use	Age 11	Age 14	Age 17	Age 20	Age 24
Ever used alcohol	2%	33%	74%	95%	98%
Used alcohol every day/nearly every day	0%	0%	1%	2%	3%
Usual number of drinks	3.38 (2.40)	3.65 (2.90)	5.51 (3.82)	4.81 (3.31)	3.92 (3.00)
Largest number of drinks	3.54 (2.33)	4.32 (3.14)	10.84 (9.54)	13.13 (9.39)	12.09 (8.82)
Ever used tobacco	4%	36%	67%	79%	83%
Used tobacco every day/nearly every day	8%	20%	43%	49%	47%
Ever used cannabis	1%	13%	38%	54%	59%
Used cannabis every day/nearly every day	0%	10%	20%	22%	21%

Note. Frequency (percentage) and means and standard deviations (SDs) for substance use at each assessment.

Data Analyses

We first examined patterns of data missingness in indicators of behavioral disinhibition across study assessments. We then conducted preliminary analyses, including descriptive statistics and zero-order correlations, for each of the indicators of behavioral disinhibition, at each age, computing the correlation matrix using full information maximum likelihood estimation in Mplus 8 (Muthen and Muthen, 1998–2017) because data were largely missing by study design (see below). We conducted multilevel modeling analyses that are robust to missing data and that accounted for the repeated measures and interdependence of the twin family data to address the two primary study questions, whether SUDs are associated with (1) premorbid (age 11) deviations in behavioral disinhibition and (2) changes in behavioral disinhibition from preadolescence into early adulthood (ages 11 to 24). Specifically, we conducted three-level multilevel models comprising time-varying variables (behavioral disinhibition indicators assessed at ages 11, 14, 17, 20, and 24) at level 1, nested within individual participants at level 2, nested within twin families at level 3. First, we fit a series of baseline models to estimate effects of age at level 1 for each indicator of behavioral disinhibition; age in years was centered at age 11 so that the intercept of each model reflects the age-11 level of the behavioral disinhibition indicator and the slope represents change from ages 11 to 24. These models quantify developmental trajectories of behavioral disinhibition. Next, we fit a series of group comparison models that explicitly tested whether behavioral disinhibition trajectories differed as a function of SUDs by adding dummy coded variables representing alcohol, tobacco, and cannabis use disorders, as well as a group with all three SUDs (0 = no SUD, 1 = SUD, in separate models) to the intercept and slope parameters at level 2 (i.e., a cross-level interaction between SUD group and age). Effects at the intercept indicate whether indicators of premorbid (age 11) behavioral disinhibition differ as a function of SUD group and effects at the slope indicate whether rates of change (from ages 11 to 24) in behavioral disinhibition differ as a function of SUD group. Because disconstraint was assessed using different measures at age 11 (MPR) and ages 17 and 24 (MPQ), we included a dummy coded variable representing measure type at level 2 in disconstraint models. Participant sex was included as a covariate in all models by adding a dummy coded variable representing sex at level 2². In each model, the variance component for the level-1 intercept was allowed to vary randomly across participants. All multilevel modeling analyses were conducted using Scientific Software International’s HLM 7.02 (Raudenbush et al., 2016) using full maximum likelihood estimation.

Results

Missing Data

The number of participants with data for each indicator of behavioral disinhibition at each study assessment is presented in Table 2. Data were missing primarily due to the nature of the study design and occasional changes made during the course of longitudinal assessments (i.e., not all measures were administered to all participants at all assessments). The visuospatial working memory test was first administered at the end of the age-11 assessment and was thus completed by only a small number of participants; due to funding constraints, only about half of participants completed the working memory test or the antisaccade task at the age-14 assessment. Data were also missing to a lesser extent because not all participants completed all measures. Some participants who were unable to return to the laboratory for in-person study assessments instead completed assessments via telephone and thus did not complete the working memory test or the antisaccade task; not all participants participated in all assessments (though missing one or more assessments did not preclude participating in subsequent assessments). We examined whether participants who did versus did not complete each measure at each assessment differed in terms of SUDs status and found little evidence that participants who did not complete the working memory test or antisaccade task had different rates of SUDs, though participants with missing disconstraint data at each age had somewhat higher rates of SUDs. To account for data missingness, we selected a multilevel modeling approach to analyses that uses maximum likelihood estimation, a more powerful technique than traditional approaches (e.g., listwise or pairwise deletion). Maximum likelihood estimation uses all of the available information (complete and incomplete, as well as auxiliary variables) to identify the parameter values with the highest probability of producing the sample data, thus yielding unbiased parameter estimates when data are missing at random (Baraldi and Enders, 2010; Enders, 2010).

Table 2.

Descriptive Statistics for Indicators of Behavioral Disinhibition at Each Age

Age	Behavioral Disinhibition
	Working Memory Distance From Target (mm)		Antisaccade Response Errors (%)		Trait Disconstraint
	n	Mean (SD)	n	Mean (SD)	n	Mean (SD)
11	77 (5%)	12.26 (3.26)	1441 (95%)	.53 (.24)	1279 (87%)	120.29 (23.89)
14	632 (49%)	11.53 (3.03)	710 (55%)	.33 (.20)	-	-
17	894 (89%)	9.97 (2.41)	990 (99%)	.28 (.20)	1319 (95%)	113.75 (15.57)
20	1088 (95%)	9.40 (2.02)	776 (68%)	.24 (.19)
24	834 (77%)	9.12 (1.90)	1053 (98%)	.22 (.18)	1224 (96%)	109.12 (15.70)

Note. Sample sizes, means, and standard deviations (SDs) for each indicator of behavioral disinhibition at each assessment. Working memory distance from target is mean distance in millimeters from the target across conditions. Antisaccade response errors are proportion in percentages of response errors to total valid responses. Trait disconstraint is mother reported at age 11 and self-reported at ages 17 and 24; trait disconstraint was not assessed at ages 14 or 20 (indicated by a -). Total N = 1,512. Available n for each measure varies at each assessment due to study design (not all measures were administered at all participants at all assessments) and because not all participants participated at each assessment; percentage given indicates the proportion of participants at each assessment who completed the measure.

Preliminary Analyses

Descriptive statistics for indicators of behavioral disinhibition at each age are presented in Table 2. As expected, there were mean-level decreases in visuospatial working memory and antisaccade errors, as well as mother/self-reported levels of disconstraint, from preadolescence into early adulthood. Zero-order correlations among indicators of behavioral disinhibition at each age are presented in Table 3. Intra-indicator correlations across ages were generally moderate to large for visuospatial working memory and antisaccade performance, and for disconstraint; interindicator correlations were generally modest for visuospatial working memory and antisaccade performance, and nonsignificant for disconstraint. That is, the indicators of behavioral disinhibition that more directly reflect executive functions (updating: visuospatial working memory; inhibitory control: antisaccade performance) showed greater coherence with one another than they did with trait disconstraint.

Table 3.

Zero-Order Correlations Among Indicators of Behavioral Disinhibition at Each Age

	1	2	3	4	5	6	7	8	9	10	11	12	13
Working memory
1. Age 11	-
2. Age 14	.37***	-
3. Age 17	.56***	.52***	-
4. Age 20	.60***	.42***	.52***	-
5. Age 24	.44***	.47***	.44***	.54***	-
Antisaccade
6. Age 11	.34***	.17	.11	.15	.11	-
7. Age 14	.22*	.27**	.16	.24**	.15	.49***	-
8. Age 17	.26**	.28**	.17	.23*	.15	.48***	.65***	-
9. Age 20	.21*	.17	.17	.23*	.22*	.34***	.44***	.50***	-
10. Age 24	.31***	.18	.11	.16	.20*	.33***	.53***	.50***	.54***	-
Trait disconstraint
11. Age 11	.05	.15	.07	.09	−.01	.07	.03	.09	−.01	.03	-
12. Age 17	−.05	.09	.04	.03	−.05	.07	.04	.07	.07	.04	.35***	-
13. Age 24	−.12	.08	.04	.02	−.03	.04	−.04	.02	.07	.03	.33***	.66***	-

Note. Total N = 1,512. Available ns for each measure at each age varied from 77 to 1441 due to study design (not all measures were administered to all participants at all assessments) and because not all participants completed each measure at each assessment. The correlation matrix was computed using full information maximum likelihood estimation, which accounts for missing data.

Developmental Trajectories of Behavioral Disinhibition

We conducted a series of multilevel models to quantify developmental trajectories of indicators of behavioral disinhibition from ages 11 to 24. A summary of results for each behavioral disinhibition indicator is presented in Table 4. The baseline models quantify effects of age; significant positive coefficients for the intercepts of these models indicate that each behavioral disinhibition indicator differed significantly from zero at age 11, and the negative coefficients for the slopes indicate that each behavioral disinhibition indicator evidenced normative changes from ages 11 to 24 in the complete sample (i.e., decreases in working memory distance from the target, antisaccade response errors, and disconstraint).

Table 4.

Summary of Results of Multilevel Modeling Analyses for Indicators of Behavioral Disinhibition

Model	Behavioral Disinhibition
	Working Memory		Antisaccade		Trait Disconstraint
	Intercept (age 11)	Slope (change from age 11 to 24)	Intercept (age 11)	Slope (change from age 11 to 24)	Intercept (age 11)	Slope (change from age 11 to 24)
	Coef (SE)	Coef (SE)	Coef (SE)	Coef (SE)	Coef (SE)	Coef (SE)
Baseline	11.27 (0.17)***	−0.156 (0.014)***	0.44 (0.01)***	−0.018 (0.001)***	115.85 (1.12)***	−0.810 (0.083)***
SUDs groups
Alcohol	0.46 (0.26)+	−0.025 (0.022)	−0.02 (0.01)	0.003 (0.001)*	8.98 (1.51)***	0.356 (0.124)**
Tobacco	0.67 (0.28)*	−0.020 (0.024)	−0.01 (0.01)	0.003 (0.001)**	10.61 (1.67)***	0.433 (0.132)***
Cannabis	0.79 (0.35)*	−0.031 (0.029)	−0.01 (0.02)	0.005 (0.002)**	11.32 (2.05)***	0.396 (0.188)*
Alc/tob/can	0.90 (0.44)*	−0.048 (0.037)	−0.03 (0.02)	0.005 (0.002)**	11.20 (2.29)***	0.400 (0.203)*

Note. Multilevel models quantifying developmental trajectories of indicators of behavioral disinhibition from age 11 to age 24. Baseline models quantify effects of age. In baseline models, the intercept represents the age-11 behavioral disinhibition indicator and the slope represents change in the behavioral disinhibition indicator from ages 11 to 24. SUDs models quantify effects as a function of SUDs by including dummy coded variables representing alcohol, tobacco, and cannabis use disorders, and all three (alc/tob/can) SUDs (0 = no SUD, 1 = SUD, in separate models). In SUDs models, positive values for the coefficients at the intercept indicate higher scores on the behavioral disinhibition indicator at age 11 for the SUD group relative to the no SUD group; when the coefficients for the slope for the baseline models are negative, positive values for the coefficients at the slope for SUDs models indicate a slower rate of change from ages 11 to 24 and negative values indicate a faster rate of change for the SUD group relative to the no SUD group. All models include participant sex as a covariate.

^*p < .050.

^**p < .010.

^***p < .001.

⁺p < .100.

Developmental Trajectories of Behavioral Disinhibition as a Function of SUDs

We next conducted a series of multilevel models to quantify developmental trajectories of indicators of behavioral disinhibition from age 11 to age 24 as a function of SUDs. A summary of results for each behavioral disinhibition indicator for alcohol, tobacco, and cannabis use disorders, as well as a group with all three SUDs, is presented in Table 4; results map onto the trajectories for each SUD group depicted in Figure 3 (i.e., the coefficients for the intercepts for each SUD model in Table 4 quantify the deviation from the no SUDs model at age 11, depicted as mean values for each group at age 11 in Figure 3, and the coefficients for the slopes for each SUD model in Table 4 quantify the deviation from the no SUDs model in change from age 11 to age 24, depicted as mean values for each group from ages 11 to 24 in Figure 3).

Figure 3.

Developmental trajectories of indicators of behavioral disinhibition from age 11 to age 24 as a function of SUDs: (A) working memory (mean distance from target in millimeters), (B) antisaccade (proportion in percentages of response errors to total valid responses), (C) trait disconstraint. Values are means for each indicator at each age for each group, adjusted for participant sex.

Visuospatial working memory performance.

Positive coefficients for the intercepts of alcohol, tobacco, and cannabis models, as well as all three SUDs, indicate that individuals who subsequently developed a SUD evidenced poorer visuospatial working memory performance at age 11 compared to individuals who did not develop a SUD; results were significant for tobacco and cannabis use disorders, and the all three SUDs group, and marginal for alcohol use disorder (p = .075). That is, there was evidence that visuospatial working memory deficits evident at age 11, before the onset of any SUD symptoms, predicted the development of SUDs. By contrast, the negative but nonsignificant coefficients for the slopes for each SUD model indicate that even individuals who subsequently developed a SUD showed developmentally normative improvements in visuospatial working memory performance from age 11 to age 24 that were comparable to those shown by individuals who did not develop a SUD; that is, all groups (no SUDs and SUDs) showed improvements on the visuospatial working memory test from preadolescence into early adulthood.

Antisaccade performance.

Nonsignificant coefficients for the intercepts of alcohol, tobacco, and cannabis models, as well as all three SUDs, indicate that individuals who subsequently developed a SUD evidenced antisaccade performance at age 11 that was comparable to that evidenced by individuals who did not develop a SUD. However, the positive, significant coefficients for the slopes for each SUD model indicate that individuals who subsequently developed a SUD failed to show comparable developmentally normative improvements in antisaccade performance from age 11 to age 24 relative to those shown by individuals who did not develop a SUD. That is, the SUDs groups made more antisaccade errors from adolescence into early adulthood than expected, given normative improvement in task performance.

Trait disconstraint.

Significant, positive coefficients for the intercepts of alcohol, tobacco, and cannabis models, as well as all three SUDs, indicate that individuals who subsequently developed a SUD evidenced higher levels of disconstraint at age 11 compared to individuals who did not develop a SUD. In addition, the positive, significant coefficients for the slopes for each SUD model indicate that individuals who subsequently developed a SUD failed to show comparable developmentally normative decreases in disconstraint from age 11 to age 24 relative to those shown by individuals who did not develop a SUD; that is, the SUDs groups evidenced greater disconstraint from preadolescence into early adulthood than expected.

Discussion

We examined whether premorbid deviations across multimodal indicators of executive functions and trait disconstraint were evident prior to the onset of SUDs and whether presence of SUDs was associated with alterations in the normative developmental trajectories from preadolescence into early adulthood for these indicators. We found that individuals who subsequently develop alcohol, tobacco, and cannabis use disorders in adolescence or early adulthood did in fact show deviations in preadolescence, evidenced by poorer premorbid performance on a visuospatial working memory test and higher premorbid levels of trait disconstraint. We further found that individuals with alcohol, tobacco, and cannabis use disorders failed to show developmentally normative changes from adolescence into early adulthood, as evidenced by persistently poorer performance on an antisaccade task and higher levels of trait disconstraint over time. Taken together, these results suggest that aspects of behavioral disinhibition may confer risk for the development of SUDs and that SUDs may have negative consequences for normative developmental trajectories of executive functions and trait disconstraint over time.

The key role of executive dysfunction implicated for SUDs has long been emphasized in models of addiction, which has largely been based on research documenting executive functioning deficits among heavy, chronic substance using adults (see Dom et al., 2005; Goldstein and Volkow, 2011; Jacobus and Tapert, 2013; López-Caneda et al., 2013; Oscar-Berman and Marinković, 2007; Volkow et al., 2012). More recent research using causally informative longitudinal study designs in samples of adolescents and young adults has begun to differentiate premorbid deviations that predict the subsequent development of SUDs versus those reflecting the consequences of SUDs on executive functions and other neural systems (Aytaclar et al., 1999; Fontes et al., 2011; Fried and Watkinson, 2005; Hanson et al., 2010; Khurana et al., 2013; Nguyen-Louie et al., 2015; Nigg et al., 2004; Peeters et al., 2015; Squeglia et al., 2017, 2014b, 2012, 2009; Tapert et al., 2002a, 2002b; Tapert and Brown, 1999). The existing literature and the present results highlight the importance of individual differences in indicators of behavioral disinhibition for conferring risk for the development of SUDs. The relative immaturity of the adolescent brain and consequent deficits in the top-down control mechanisms underlying executive functions may make it generally more difficult for adolescents to inhibit behaviors related to substance use. However, we found that it was individuals with relative deviations in premorbid indicators of updating executive processes (performance on the visuospatial working memory test) and trait disconstraint who developed a SUD. Because we excluded any participants who met criteria for any substance abuse or dependence symptoms at age 11, these results reflect deviations evident prior to the onset of problematic substance use. At the same time, our finding that having a SUD was associated with relatively worse trajectories of indicators of inhibitory control (performance on the antisaccade task) and trait disconstraint from adolescence into early adulthood is consistent with the supposition that the adolescent brain and related executive functioning and behavioral processes may be particularly vulnerable to the harmful effects of substances.

Although the present paper speaks to the role of indicators of behavioral disinhibition for the development of SUDs, it is important to note that there are likely multiple interrelated processes at play. Dual-systems models of adolescent risk-taking and substance use emphasize an imbalance in the respective maturation during adolescence and early adulthood of the brain regions underlying bottom-up reward-related processes and top-down control processes (Casey and Jones, 2010; Steinberg, 2010). Heightened reward sensitivity and incentive motivation may bias adolescents to seek out rewarding, novel, and sensation-enhancing stimuli (including substances) while insufficiently developed and inefficient executive functions and related abilities may make resisting such reward-seeking behaviors difficult or impossible. Our group has previously found evidence that morphological deviations in brain regions linked to control processes (e.g., gray matter thickness in frontal regions) reflect a pre-existing propensity toward problematic substance use in adolescence, while more risky decision making, indexed by poor performance on the Iowa Gambling Task, reflects a substance exposure-related effect on reward processing (Malone and Iacono, 2002; Wilson et al., 2015a). Taken together, these lines of research speak to the critical role of the neural systems underlying indicators of behavioral disinhibition for understanding the development of SUDs in adolescence and early adulthood.

It is notable that the pattern of results was generally consistent across the different substances, highlighting the implications of behavioral disinhibition for substance use in general, rather than for a particular substance of abuse. This likely reflects the high rates of comorbidity across SUDs, and is consistent with a model of externalizing psychopathology in which SUDs share a common etiology with each other, as well as with other forms of disinhibited, aggressive, and delinquent behaviors (e.g., antisocial personality disorder; Iacono et al., 2008; Krueger et al., 2002). At the same time, that results did not differ for alcohol and tobacco versus cannabis is an interesting finding. Because cannabis was illegal in all 50 United States at the time of data collection, one might expect greater levels of behavioral disinhibition would be necessary to promote its initiation and abuse, relative to alcohol and nicotine, both of which are legal among adults. Future research that considers the role of contextual factors (e.g., legality, acceptability, availability) for associations between behavioral disinhibition and SUDs will be informative.

Our selection of three different multimodal indicators of behavioral disinhibition that reflect related but distinct aspects of the construct allowed us to consider potentially different associations with SUDs. Both working memory and antisaccade performance tap into general executive functioning processes (i.e., inhibitory control), with the former also relating to updating executive functioning processes. Whereas trait disconstraint is conceptualized as a dispositional tendency toward a lack of behavioral and affective restraint. Our finding that premorbid deficits in performance on a visuospatial working memory test predicted the development of SUDs may speak to the importance of impaired updating executive function for conferring risk for SUDs and points to potential mechanisms by which this particular aspect of behavioral disinhibition might lead to substance initiation and eventual misuse. In contrast, our finding that SUDs were associated with less developmentally normative improvement in antisaccade performance may speak to the negative implications of substance misuse for a broader indicator of inhibitory control--the ability to appropriately inhibit strong prepotent responses in order to correctly execute appropriate behavioral responses. Our finding that premorbid trait disconstraint predicted the development of SUDs is consistent with previous research (Aytaclar et al., 1999; Fontes et al., 2011; Khurana et al., 2013; Nigg et al., 2004; Peeters et al., 2015; Squeglia et al., 2017, 2014a; Tapert et al., 2002a) in speaking to the key role of this trait in indexing risk for SUDs and related problems, but the finding that individuals with SUDs also failed to show developmentally normative decreases in trait disconstraint comparable to individuals without SUDs is relatively novel in highlighting the potential implications of substance misuse for this dispositional construct. It is worth noting the relatively modest intercorrelations among the working memory and antisaccade performance indicators at each age, as well as their nonsignificant intercorrelations with disconstraint. As we discuss in greater detail elsewhere, there are notable challenges when relying on single measures of a construct assessed using different methods (Patrick et al., in press; Venables et al., 2018). Rather than suggesting that certain indicators better reflect the behavioral disinhibition construct, we believe the present findings speak to the importance of including multiple indicators assessed using multivariate methods when modeling constructs.

Although our prospective, longitudinal design allowed us to track indicators of behavioral disinhibition and SUDs over time, and our exclusion of participants with any SUD symptoms at the initial assessment allowed us to examine premorbid indicators of behavioral disinhibition, we are nonetheless limited in determining the causal role of behavioral disinhibition for SUDs. That is, although our finding that premorbid deviations in behavioral disinhibition indicators precede the onset of SUDs is consistent with causal risk, it may be the case that behavioral disinhibition in fact reflects a larger underlying liability toward SUDs and related psychopathology (e.g., Iacono et al., 2008). Future research using causally informative study designs, and in particular research that takes a more fine-grained approach to investigate the dynamic nature of the interrelationships among indicators of behavioral disinhibition and substance initiation, regular use, and abuse will help to illuminate causal relationships, identify the mechanisms by which substance use and abuse develops over time, and inform our understanding of associations at different stages of the addiction process.

The present study had a number of strengths, including the large, population-based sample that had been prospectively assessed from preadolescence into early adulthood with high participation rates. However, it is important to also note several limitations. Our inclusion of multiple indicators of behavioral disinhibition assessed at multiple time points using a multimodal approach that included neurocognitive, psychophysiological, and behavioral measures allowed us to examine developmental trajectories of related but distinct indicators of behavioral disinhibition during a critical developmental period, the transition from preadolescence into early adulthood. Methodologically, however, the selection of measures that are developmentally appropriate for repeated use over such a large age span can prove difficult, and it is possible that restriction of range across groups may have affected our ability to detect differences in antisaccade performance at age 11 (i.e., a floor effect) or improvements in visuospatial working memory performance from age 11 to 24 (i.e., a ceiling effect). Although the sample had minimal participant attrition due to missed assessments over multiple time points spanning approximately 13 years, as is the case with many longitudinal studies, our introduction of additional measures at different assessments and funding constraints that limited administration of some measures at some assessments resulted in substantial rates of missing data at some ages. We selected a multilevel modeling approach that uses maximum likelihood estimation that incorporates all available information and yields unbiased parameter estimates (Baraldi and Enders, 2010; Enders, 2010), but this is nonetheless an important consideration when interpreting the results. Our examination of the most common substances of abuse, alcohol, tobacco, and cannabis, allowed us to conduct analyses in relatively large subsamples of participants with SUDs, but the present paper cannot speak to less commonly abused but nonetheless critically important substances (e.g., opioids). We allowed comorbid substance use within each SUD group because co-occurring use of alcohol, tobacco, and cannabis during adolescence and early adulthood is the norm, rather than the exception (Substance Abuse and Mental Health Services, 2017a), but results are as such less informative as to specific effects for individual substances. In addition, although the majority of SUDs onset by early adulthood (Kessler et al., 2005), it is nonetheless possible that some participants in the no SUD group, who did not meet criteria for any licit or illicit substance use disorder at any assessment point, will eventually develop a SUD. Finally, although representative of the demographic makeup of Minnesota state during the targeted birth years, the lack of racial and ethnic diversity in the sample limits generalizability. Notably, because this is a population-based sample (of twins born in Minnesota between 1977 and 1984), the present sample has broader distribution of key socioeconomic indicators (e.g., income, education level, rural/urban residence) than is typically found in the usual sample of self-selected community volunteers. Because these factors may be highly confounded with problematic substance use and SUDs (Casswell, Pledger, & Hooper, 2003; Galea & Vlahov, 2002; Hiscock, Bauld, Amos, Fidler, & Munafo, 2012), the present study is informative as to implications of behavioral disinhibition for SUDs. Nonetheless, future research is needed that better delineates relationships between critical risk factors often associated with socioeconomic disadvantage (e.g., less parental monitoring, poorer-quality schools, lead exposure, less safe neighborhoods) and the development of behavioral disinhibition and associated substance misuse.

Conclusions

SUDs are associated with behavioral disinhibition. The present findings for multimodal indicators of behavioral disinhibition suggest this association reflects a failure in the top-down control mechanisms related to executive functions. Because of the significant normative maturational changes in executive functioning and related processes that occur during adolescence, as well as the disproportionate onset of SUDs in adolescence and early adulthood, this is key developmental period for investigations of behavioral disinhibition and the development of SUDs. Premorbid deviations in updating executive function and trait disconstraint evident in preadolescence precede the onset of SUDs and may confer risk for their development. In turn, SUDs may exacerbate preexisting deficits in inhibitory control and trait disconstraint and may have deleterious consequences for functioning during the critical transition from adolescence into early adulthood.

Highlights.

• Executive functioning impairment and trait disconstraint are implicated in substance use disorders

• Executive functioning indicators and trait disconstraint predicted substance use disorders

• Substance use disorders disrupted normative developmental trajectories of executive functioning indicators and trait disconstraint