Large-scale brain network activation during emotional inhibitory control: Associations with alcohol misuse in college freshmen

Abstract

Background:

The transition to college is associated with increased alcohol misuse and a consequent increase in negative, alcohol-related social and health impacts. Traits associated with ongoing brain maturation during this period, including impulsivity in emotional contexts, may contribute to risky alcohol use.

Methods:

This functional magnetic resonance imaging (fMRI) study examined brain network activation strength during an emotional inhibitory control task (Go-NoGo), which required participants to ignore background images with negative or neutral emotional valence during performance. Participants were 60 college freshmen (aged 18-20 years, 33 women). Survey measures, completed at baseline and one-year follow-up (follow-up n=52, 29 women), assessed alcohol misuse (AUDIT), alcohol/substance use (C-CAPS), and negative consequences of alcohol use (YAACQ). Measures were examined relative to network activation strength, on the Negative NoGo > Neutral NoGo contrast, of four large-scale brain networks implicated in top-down regulation of cognition and attention: right and left lateral frontoparietal networks (rL-FPN; lL-FPN), dorsal attention network (DAN), and salience network (SN).

Results:

Activation strength of DAN was associated negatively with AUDIT (p=.013) and YAACQ (p=.004) at baseline, and with C-CAPS at baseline and follow-up (p=.002; p=.005), and also positively with accuracy on NoGo trials with negative backgrounds (p=.014). Activation strength of rL-FPN was positively associated with C-CAPS at follow-up (p=.003). SN activation strength was associated negatively with accuracy on NoGo trials with negative (p<.001) and neutral (p=.002) backgrounds, as well as with the accuracy difference between negative versus neutral NoGo trials (p=.003).

Conclusion:

These findings suggest that less engagement of large-scale brain circuitry that supports top-down attentional control, specifically during negative emotions, is associated with more problematic drinking in emerging adults who attend college. This pattern of network activation may serve as a risk marker for ongoing self-regulation deficits during negative emotion that could increase problematic alcohol use and negative impacts of drinking.

The transition to college is associated with substantial increases in alcohol consumption (Schulenberg and Maggs, 2002), resulting in a consequent increase in negative alcohol-related outcomes during this period. Negative impacts of alcohol use among youth include alcohol poisoning, blackouts, injuries, automobile accidents, physical and sexual assaults, sexually transmitted infections, interpersonal problems, poor academic performance (Hingson et al., 2017, Perkins, 2002), and risk for developing an alcohol use disorder (Bravo et al., 2018). Alcohol misuse, i.e. alcohol use that increases risk of such negative social and health impacts, reaches peak prevalence in emerging adulthood, between 18-25 years of age (Hingson et al., 2017). Common forms of alcohol misuse in this age group include, but are not limited to, binge and extreme binge drinking, heavy drinking, and drinking in risky contexts (Hingson et al., 2017). While the uptick in drinking during the college transition is undoubtedly impacted by a variety of social and environmental factors, such as reduced parental supervision, peer influence, and college culture, concurrent ongoing brain maturation may also play an important role (Silveri, 2012).

Continued refinement of brain circuitry implicated in executive control of cognition, attention and behavior is known to occur throughout emerging adulthood (Gogtay et al., 2004). This ongoing development can result in impulsive and risky behavior, including alcohol misuse (Casey et al., 2000) and may render neural circuitry particularly vulnerable to negative long-term effects of excessive alcohol use (Silveri, 2012). Acute alcohol exposure also increases impulsivity (Dougherty et al., 2008), decreases behavioral and attentional inhibition (Mulvihill et al., 1997), and negatively impacts decision-making (George et al., 2005), though these effects do not apply uniformly across all forms of impulsivity or decision-making (Karlsson et al., 2021). Nevertheless, drinking, together with ongoing brain maturation, can make young adults highly prone to risky behaviors, including continued drinking to excessive or binge levels (Spear, 2000), which may then further undermine self-control (Silveri, 2012, Weissenborn and Duka, 2003), synergistically increasing the likelihood of negative outcomes.

Powerful emotions can also erode efforts at self-regulation and increase impulsivity (Yong et al., 2010). The tendency to act rashly during negative emotions, or negative urgency, has been moderately to strongly linked to problematic alcohol use (Berg et al., 2015, Dick et al., 2010) and to alcohol use disorders (Fischer et al., 2004, Verdejo-Garcia et al., 2007), as well as specifically to binge drinking in emerging adults (Bo et al., 2016). Amplification of impulsivity by alcohol may be increased in emotion-laden circumstances, an effect that may be particularly pronounced in young women (Weafer, 2020). Furthermore, impulsive actions potentiated by drinking while experiencing negative emotions can include particularly self-destructive behaviors, such as risky sex, drug use, intoxicated driving, aggression and even deliberate self-injury or suicidal behaviors.

One approach to assessing impulsivity and its role in problem drinking makes use of tasks, such as a Go-NoGo task, that taps into inhibitory control by requiring participants to hold back a primed or prepotent response. Increased impulsive commission errors on Go-NoGo tasks – e.g., button presses to NoGo cues – have been directly linked to binge drinking among college students (Henges and Marczinski, 2012, Nigg et al., 2006). Poorer Go-NoGo performance is also associated with higher self-reported levels of negative urgency (Dick et al., 2010). When implemented in conjunction with functional magnetic resonance imaging, Go-NoGo tasks have been used to identify brain regions implicated in inhibitory control and alterations in task-related brain activation associated with both current and future alcohol misuse. For example, in a longitudinal study of adolescents, individuals who went on to transition to heavy drinking showed reduced baseline activation in frontal, parietal, subcortical and cerebellar brain regions during NoGo versus Go trials, relative to adolescents who remained non-drinkers at follow-up (Wetherill et al., 2013). Integration of emotional stimuli or distractors into a Go-NoGo protocol, in which impulsive errors must be avoided in the context of negatively valanced stimuli, provides insight into the impact of negative emotions on inhibitory control and related brain activation. A study in healthy young adults (age 18-28 years) reported associations between risk-taking and activation of orbital and ventromedial prefrontal cortex during performance of a Go-NoGo task with neutral or aversive image distractors (Brown et al., 2015). Our previous work has demonstrated that higher recent incidence of binge drinking is significantly associated with decreased brain activation on inhibitory trials with a negative versus neutral background in college freshmen (Cohen-Gilbert et al., 2017). In this study, we found decreased activation of brain regions that subserve inhibitory control: dorsolateral prefrontal cortex (DLPFC), dorsomedial prefrontal cortex (DMPFC) and anterior cingulate cortex (ACC) (Cohen-Gilbert et al., 2017). Notably, the regions identified in this study comprise nodes of large-scale brain networks implicated in executive attention and control.

Close correspondence of brain architecture and functional connectivity of large-scale brain networks during rest and during performance of multiple cognitive tasks (Cole et al., 2014, Laird et al., 2011, Nickerson, 2018, Smith et al., 2009) suggests an intrinsic architecture of functional brain organization, making large-scale networks an important unit of analysis in the study of brain function supporting cognition and behavior. Many networks have now been implicated in the top-down control of attention and cognition. Among those that have been examined most extensively are the lateral frontoparietal network (L-FPN), the salience network (SN) and the dorsal attention network (DAN). The L-FPN, comprised primarily of lateral prefrontal cortex, intraparietal sulcus, posterior parietal lobe and ventral inferior temporal lobe, is sometimes referred to as the central executive network. This network provides cognitive flexibility and is involved in initiating and adjusting control and switching attention from external to internally-focused processes (Dosenbach et al., 2008, Uddin et al., 2019). The L-FPN reproducibly separates into two lateralized sub-networks with right L-FPN (rL-FPN) associated more reliably with executive functions, such as inhibitory control and working memory while left L-FPN (lL-FPN) is also associated with language-related processes. The SN, a prefrontal midcingulo-insular network, comprised primarily of ACC and the anterior insula/operculum, together with limbic and paralimbic subcortical structures, assists in the detection of novel, surprising and behaviorally relevant stimuli (Menon, 2015, Uddin et al., 2019). The DAN, a dorsal frontoparietal network comprised primarily of DLPFC, frontal eye fields, middle temporal motion complex, and superior parietal lobe, modulates externally-oriented attention and orienting, providing top-down attentional control (Corbetta and Shulman, 2002, Corbetta et al., 2008, Uddin et al., 2019). Though few existing studies examine associations between task-related network activation and drinking in young adults, a preliminary prospective examination of young adult binge drinkers found maximum drinks per occasion was associated with reduced average activation of a fronto-parietal network (Worhunsky et al., 2016). This study further reported that an escalation in the maximum number of drinks during a 12-month follow-up was associated with a larger difference in engagement of this network during successful inhibition versus error trials in a Go-NoGo task, and with higher self-reported impulsivity (Worhunsky et al., 2016). New data on the relevance of network functioning in alcohol and substance use is rapidly emerging; however, most of the work to date remains focused on assessment of network connectivity during rest (Wilcox et al., 2019).

The current study utilizes multivariate methods to examine activation of key networks of interest during the emotional Go-NoGo task. These multivariate methods (Nickerson, 2018) enable task-related network loadings reflecting the activation strength of the network to be examined relative to variables of interest, such as drinking measures, while parsing out effects of any overlapping networks and other artifactual processes. In this study, large-scale network activation was examined during performance of an emotional Go-NoGo task in a sample of healthy college freshmen in order to elucidate the impact of negative emotions on the activity of control networks during inhibitory control. These measures of network activation were then examined relative to measures of alcohol misuse and negative consequences of drinking reported at the time of scanning and at a one-year follow-up. Inhibitory control efforts during emotional versus neutral background conditions were predicted to require additional executive and attentional control. As such, it was hypothesized that increased activation strength in the networks of interest would be associated with improved task performance, particularly on negative relative to neutral trials. It also was predicted that a higher reported alcohol misuse, substance use, and negative consequences of drinking, both at baseline and one year later, would be associated with a failure to recruit L-FPN, DAN and SN on emotionally negative versus neutral inhibitory control trials. Marked sex differences exist in the prevalence and manifestation of alcohol misuse (McHugh et al., 2018), as well as in the neurobiological impacts of alcohol (Flores-Bonilla and Richardson, 2020), thus sex is an important variable to examine, and accordingly, was included in analyses.

Method

Participants

Participants included in the baseline analyses were 60 healthy college freshmen (aged 18-20 years, 33 women) who engaged in a range of alcohol drinking behavior. All participants were enrolled in 4-year college programs. Demographic, IQ and alcohol use data are summarized in Table 1. Participants were screened and excluded for neurological disorders, head trauma with loss of consciousness, current use of psychotropic medication, and MRI contraindications such as claustrophobia or metal in the body. Participants were also excluded at screening for more than ten lifetime uses of marijuana, more than 25 lifetime uses of tobacco products or any period of regular tobacco use (weekly or more frequent), and/or more than one use of illicit drugs other than marijuana. Participants were alcohol abstinent a minimum of 24 hours prior to the scanning visit. Of the 65 total participants who completed the fMRI task, five were excluded from analysis for the following reasons: 1) one fell asleep during the task, 2) one provided false or unreliable demographic and survey data, 3) two could not adequately see the letter stimuli due to issues with the MRI-safe glasses, 4) one experienced high levels of anxiety in the scanner and aborted the session immediately after completing the task. Of the 60 participants included at baseline, 52 participants (29 women) also completed remotely administered one-year follow-up surveys and are included in follow-up analyses.

Table 1.

Demographic data and alcohol use at baseline

Measures	Range	Mean(stdev)
Age	18.2 - 20.0	18.9(0.4)
IQ	93 - 142	118.7(10.0)
Number of drinks (past 3 mo)	0 - 115	37.8(35.6)
Number of binges (past 3 mo)	0 - 20	4.2(4.8)
Maximum drinks/occasion (past 3 mo)	0 - 20	6.1(4.1)
*Days since last drink	1 - 66	12.7(13.3)

^*Days since last drink was calculated for participants who reported at least one drink in the prior 90 days (n = 55). All other data reflect the full sample (n = 60).

IQ was assessed via the Weschler Abbreviated Scale of Intelligence (WASI, 2 subscale). The Structured Clinical Interview for DSM-IV (SCID) was used to assess presence or absence of psychiatric disorders. While no participants reported a current psychological diagnosis during phone screening, five participants met criteria for past major depressive disorder (>6 months prior to participation). Two participants met criteria for past social phobia. One participant met criteria for past anorexia nervosa (6 years prior, fully remitted). One participant met criteria for current social phobia, generalized anxiety disorder, and bulimia nervosa (partially remitted). Eleven participants met criteria for an alcohol use disorder (n=9 mild, n=2 moderate, based on DSM 5 criteria), which is unsurprising given efforts to recruit heavy drinkers. All participants provided written informed consent prior to participation. This study was approved by the Mass General Brigham Institutional Review Board of McLean Hospital.

Alcohol Use and Clinical Measures

Survey measures were administered using RedCap software (Harris et al., 2009) during the baseline visit and remotely at one-year follow-up. Alcohol misuse was measured via the Alcohol Use Disorder Identification Test (AUDIT), a 10-item questionnaire that assesses quantity and frequency of alcohol use, binge drinking, dependence symptoms, and alcohol-related problems (Saunders et al., 1993). Negative consequences of alcohol drinking, including blackouts, risky behaviors, social consequences, and academic or occupational harms over the past year were assessed via the Young Adult Alcohol Consequences Questionnaire (YAACQ) (Read et al., 2007), a 48-item self-report retrospective assessment with possible scores ranging from 0 to 48. Two baseline YAACQ surveys were unusable due to a technical error. General functioning, including depression and anxiety levels, were assessed via the Counseling Center Assessment of Psychological Symptoms (C-CAPS), a 62-item multi-dimensional mental health assessment tool designed for use in college populations (Locke et al., 2011). The C-CAPS includes the following subscales: Depression, Generalized Anxiety, Social Anxiety, Academic Distress, Eating Concerns, Family Distress, Hostility, and Alcohol and Substance Use. The alcohol and substance use subscale includes six questions, five of which query problematic alcohol use and one of which asks about drug use. The primary clinical outcome variables of interest in the current study included AUDIT total score, YAACQ total score and C-CAPS Alcohol and Substance Use.

Alcohol use 90 days prior to the baseline scanning visit was further queried via a calendar-based Time-Line Follow-Back interview, data derived from this interview - number of drinks, number of binges, maximum drinks per occasion and days since last drink – are reported in Table 1. Survey measures of alcohol use, depression and anxiety at baseline and follow-up are reported in Table 2.

Table 2.

Survey measures at baseline and one-year follow-up

		Baseline		Follow-up
Measures		Range	Mean(stdev)	Range	Mean(stdev)
AUDIT	Total	0 - 17	6.2(4.2)	0 - 20	6.4(4.2)
YAACQ	Total	0 - 23	6.6(5.6)	0 - 43	9.1(9.4)
C-CAPS	Substance/Alc Use	0 - 3.0	1.1(0.8)	0 - 3.2	1.1(0.8)
	Depression	0 - 2.7	0.7(0.6)	0 - 3.2	1.0(0.7)
	Social Anxiety	0 - 3.1	1.6(0.7)	0.3 - 3.4	1.8(0.8)
	Generalized Anxiety	0 - 2.2	0.7(0.5)	0 - 2.8	1.0(0.7)

At baseline, n=60 for AUDIT and C-CAPS, n = 58 for YAACQ. At follow-up, n = 52 for AUDIT, n = 51 for C-CAPS and YAACQ. AUDIT scores: 0 = abstainer, 1-7 = low risk, 8-14 = hazardous alcohol use, 15+ = high risk for AUD. C-CAPS elevated cut points: Substance/Alcohol Use = 1.4, Depression = 1.7, Social Anxiety = 2.5, Generalized Anxiety = 1.7. Scores above threshold indicate high likelihood of clinically significant issues in that subscale area.

Emotional Go-NoGo Task

FMRI data were acquired during an emotional Go-NoGo task that required participants to ignore background images with positive, negative, or neutral emotional valence (see Cohen-Gilbert et al., 2017, for a full description). As in the prior study, in the Go-NoGo task, letters were presented sequentially in a small box at the center of the background image. Background images were selected from the International Affective Picture System (IAPS) (Lang et al., 2008) based on valence ratings, to be positive, negative, or neutral. Scrambled versions of a subset of these images served as non-emotional backgrounds that had no discernable image content. The task consisted of 480 trials, evenly split between background types (120 each). Each background image was presented once during the task. Participants were instructed to respond (button press) as quickly as possible to every letter stimulus, except for the letter ‘X’. Participant responses were recorded using a fiber optic response pad (fORP). Xs appeared on 25% of the trials such that participants acquired a prepotent tendency to press and NoGo trials required active inhibition of a response. The task, presented using E-prime 2, was synched to the scanner via RF pulse. The paradigm used a rapid event-related design, with each 1500ms trial consisting of 500 ms of fixation, followed by 350 ms of the background image presented alone, and then 650 ms in which the letter cue and background image were both visible. This rapid stimulus presentation created high levels of inhibitory demand and prevented ceiling effects in NoGo trial performance in this high functioning young adult sample. To avoid the use of extended or irregular inter-stimulus intervals, target (NoGo) trials were distributed within the stream of Go trials to create jitter. Go trials were treated as an implicit baseline and not modeled (Garavan et al., 2002). The task was presented in three runs (160 trials/run). Task performance measures included accuracy on NoGo trials, accuracy on Go trials, and reaction times on correct Go trials (see Table 3 for task performance data).

Table 3.

Go-NoGo Task Performance

Measures	Positive	Negative	Neutral	Scrambled
NoGo Accuracy (%)	67.8(16.1)	65.3(19.3)	68.2(17.3)	71.9(16.2)
Go Accuracy (%)	98.9(2.0)	99.0(1.8)	99.0(2.1)	99.0(2.2)
Go Reaction Time (msec)	346.4(31.3)	354.6(33.3)	345.8(32.2)	343.9(30.6)

Data represent the means and (standard deviations) from the full sample n=60.

Magnetic Resonance Imaging

All MRI data were collected using a 3 Tesla Siemens TIM Trio scanner (Erlangen, Germany) with a 32-channel head coil. High-resolution structural images were acquired using a T1-weighted multiecho Multiplanar Rapidly Acquired Gradient-Echo (ME- MPRAGE) 3D sequence in 4 echoes (TE=1.64/3.5/5.36/7.22ms, TR=2.1s, TI=1.1s, FA=12°, 176 slices, 1×1×1.3mm voxel, acquisition time=5 min) and were used to register functional images to standard space. FMRI data were collected in three runs (5:13 min/run) using whole-brain multiband gradient echo echo-planar imaging (EPI) with BOLD contrast. Images were acquired in 54 oblique, interleaved slices (TR/TE/FA=750ms/30ms/52°, FOV=220, voxel size: 2.8mm×2.8mm×2.8mm, multiband=6, GRAPPA=2). At the same resolution and slice locations, a fieldmap (TR=1000, TE=10/12.46ms, FA=90°, 2:44min) was acquired to permit offline correction of field inhomogeneities.

FMRI Data Processing

Data Pre-Processing:

FMRIB Software Library (FSL) software v6.0.11 (Smith et al., 2004) was used to preprocess raw data and for subsequent statistical analyses. Preprocessing steps included: motion correction, slice-timing correction, non-brain removal, spatial smoothing (FWHM 6mm Gaussian kernel), and grand-mean intensity normalization of the 4D dataset by a single multiplicative factor. Runs began with a 30s rest block (40 volumes) before task onset, which was removed prior to this analysis, thus no additional volumes were removed to allow for signal equilibration. No participants were removed from the analysis due to excessive motion in the scanner as subject motion was minimal and did not exceed 1.5 times the voxels size (max movement was 3.5 mm). To further remediate any effects of motion, an automated method for fMRI data denoising, ICA AROMA, an independent component analysis-based denoising tool, was applied to all datasets (Pruim et al., 2015). Following ICA-based denoising, data underwent temporal filtering, using a Gaussian-weighted least-squares straight line fit with a highpass cutoff=100sec, and fieldmap-based distortion correction. FMRI data were registered to MNI152 standard space by first registering data to each individual’s high-resolution structural image, using boundary-based registration (BBR), and then transformation into MNI stereotaxic space using the registration information from registering the high-resolution structural image to MNI152 standard space, which was done using FNIRT.

Statistical Modeling:

A voxel-wise general linear model (GLM) with all NoGo trials within the four background conditions (positive, negative, neutral, scrambled) each modeled as separate regressors (30 trials per condition) convolved with a gamma hemodynamic response function was implemented for each run. Contrasts of parameter estimates (COPE) maps were calculated for the contrast of interest: Negative NoGo v. Neutral NoGo. COPEs from the three task runs were combined using a second-level fixed effects GLM to create an average COPE map for each participant. These average COPE maps were used for further statistical analyses. Whole brain GLM findings for the NoGo > Go contrast can be found in Supplement 1.

Extraction of Network Activation Weights:

Task-related activation of key brain networks was extracted via a multivariate spatial regression approach (see Nickerson, 2018, for detailed methods). To achieve this, participant-level average COPE maps were concatenated to generate a single 4D data file for the contrast of interest. Open access network template spatial maps, derived from Human Connectome Project data (Smith et al., 2015) using group independent component analysis (https://www.humanconnectome.org/study/hcp-young-adult, n =1003, dimentionality=50 components), were then spatially regressed against the concatenated participant-level brain activation maps for the Negative NoGo > Neutral NoGo contrast to estimate the impact of negative emotional stimuli during response inhibition on the strength of activation of each associated network. Task-related network loadings thus reflected the activation strength of individual networks for each participant, taking into account any spatial overlap among networks. The advantage of this approach is that the average activation strength of a particular network can be disentangled from other networks, which is not possible with the conventional voxel-wise GLM approach used in the standard task fMRI analysis. In the latter case, the activation map is an aggregate spatial map reflecting all nodes that are modulated by the task condition – by taking these maps and doing a multivariate spatial regression against network templates, we can disentangle node activity with respect to contributions to individual networks from each node (or voxel). Networks of interest were four major networks previously implicated in top-down regulation of attention, cognition and behavior: left and right L-FPN (lL-FPN; rL-FPN), DAN, and SN. Templates used for these four networks of interest are shown in Figure 1. The L-FPN is examined bilaterally because it reproducibly separates into two lateralized sub-networks (Laird et al., 2011, Nickerson, 2018, Smith et al., 2009). A single L-FPN network is only observed at very low model orders of ICA. At such low model orders, (dimensionality <~20) many networks are mixed, so current practice typically employs larger model orders for studying large-scale brain circuits (dimensionality used to generate the current templates = 50). The SN and DAN do not separate into lateralized networks at any model order as they seem to be truly bilateral in nature (Corbetta et al., 2008, Dosenbach et al., 2008). We selected HCP templates for our networks as they were derived from a very large sample of healthy young adults (N=1003; four runs of 15 minutes of high-quality data for each subject) and comprised network template spatial maps that have been shown to correspond to task networks and that are highly reproducible (Laird et al., 2011, Nickerson, 2018, Smith et al., 2009). Network activation strengths for these networks were examined relative to the primary clinical outcome variables (AUDIT total, YAACQ total, and C-CAPS alcohol and substance use) and task performance (accuracy on NoGo trials and reaction times on correct Go trials). Go trial accuracy was close to ceiling and therefore not examined relative to brain network activation.

Figure 1.

Template networks of interest, extracted from Human Connectome Project data (https://www.humanconnectome.org/study/hcp-young-adult, n =1003, dimentionality=50 components) for left lateral frontoparietal network (lL-FPN) right lateral frontoparietal control network (rL-FPN), salience network (SN), and dorsal attention network (DAN).

Statistical Analysis

Behavioral and survey measures:

Repeated-measures ANOVAs were used to examine the impact of sex and trial background on three task performance measures: accuracy on NoGo trials, accuracy on Go trials, and reaction time on correct Go trials. Each ANOVA included trial background (positive, negative, neutral or scrambled) as within subject factors and sex as a between subject factor. Sex differences in alcohol use (AUDIT total, YAACQ total, and C-CAPS alcohol and substance use) were likewise examined via repeated-measures ANOVAs. Each of these ANOVAs included time point (baseline vs. one-year follow-up) as within subject factors and sex as a between subject factor. Age was included as a covariate in each analysis. Significant effects revealed by the ANOVAs were further explored via post hoc paired-samples t-tests comparing performance between each background type (m=6). Post hoc tests were corrected for false discovery rate (FDR) using the Benjamini-Hochberg procedure and a critical p-value of p<.05.

Potential associations between task performance (NoGo accuracy, Go accuracy, and reaction times on correct Go trials) on negative and neutral trials, and alcohol measures (AUDIT total, YAACQ total, and C-CAPS alcohol and substance use) at baseline and one-year follow-up were assessed via multiple linear regressions. In each separate regression, one performance measure, age, and sex were included as independent variables (IVs) and one alcohol measure was included as a dependent variable (DV). FDR for this set of regressions was controlled using the Benjamini-Hochberg procedure and a critical p-value of p<.05 (m=36, 3 performance measures x 2 background types x 3 alcohol measures x 2 timepoints).

Associations between network activation, alcohol misuse, and task performance:

To examine associations between network activation strengths and baseline drinking, three multiple linear regressions were performed. Separate regressions examined baseline scores on each alcohol use measure: AUDIT total, YAACQ total, and C-CAPS alcohol and substance use (DVs). IVs were activation strength in the four networks of interest (lL-FPN, rL-FPN, DAN, SN), as well as age and sex, included as covariates. FDR for these regressions was controlled using the Benjamini-Hochberg procedure and a critical p-value of p<.05 (m=12, 4 predictors of interest x 3 regressions).

Given that drinking behavior can change considerably during college, assessment of alcohol misuse across multiple timepoints (freshman and sophomore years) enabled the assessment of whether network activation related to future as well as current problem drinking. To examine the association between network activation strengths at baseline and drinking one year later, three additional multiple linear regressions were performed. Separate regressions examined 1-year follow-up scores on each alcohol use measure: AUDIT total, YAACQ total, and C-CAPS alcohol and substance use (DVs). Each of these regressions’ IVs included activation strength in the four networks of interest (lL-FPN, rL-FPN, DAN and SN), as well as baseline scores on the same alcohol measure, age, and sex, included as covariates. Baseline scores were included to examine whether associations could be detected between network activation and follow-up alcohol use scores could be detected above and beyond baseline associations. FDR for these regressions was controlled using the Benjamini-Hochberg procedure and a critical p-value of p<.05 (m=12).

To examine associations between network activation strength and inhibitory control during negative and neutral background conditions, three multiple linear regressions were performed with activation strength of the four networks of interest as IVs, as well as age and sex, included as covariates. NoGo trial accuracy (DV) was examined separately for negative trials and neutral trials. To parallel the Negative NoGo > Neutral NoGo contrast used to extract network activation strength, the relative impact of the two background types on performance (negative – neutral) was also examined as a DV. FDR for these regressions was controlled using the Benjamini-Hochberg procedure and a critical p-value of p<.05 (m=12).

To examine associations between network activation strength and processing speed during negative and neutral background conditions, three additional multiple linear regressions were performed with activation strength of the four networks of interest as IVs, as well as age and sex included as covariates. Reaction time on correct Go trials (DV) was examined separately for negative trials and neutral trials as well as the difference in reaction times between the two conditions (negative – neutral). FDR for these regressions was controlled using the Benjamini-Hochberg procedure and a critical p-value of p<.05 (m=12).

Sex differences in network activation strength were examined via a single repeated measures ANOVA which included network (lL-FPN, rL-FPN, DAN, SN) as a within subject factor and sex as a between subject factor. Age was included as a covariate.

Results

Behavioral and Survey Measures

Task performance for the full sample is summarized in Table 3. A repeated measures ANOVA revealed a significant effect of background on NoGo accuracy, F(3,174)=7.05, p<0.001. Post hoc paired samples t-tests showed significantly lower NoGo accuracy on negative (t(59)=4.42, p<.001), positive (t(59)=2.95, p=.005), and neutral (t(59)=2.72, p=.009) trials, relative to scrambled trials. Results also showed lower accuracy on negative relative to neutral trials (t(59)=2.02, p=.048), though this effect did not survive correction for multiple comparisons. No significant main effect of sex or interaction effects were observed.

A repeated measures ANOVA also revealed a significant effect of background on Go trial reaction time, F(3,174)=14.41, p<0.001. Post hoc paired samples t-tests showed significantly slower reaction times on negative relative to positive (t(59)=4.31, p<.001), neutral (t(59)=5.59, p<.001) and scrambled (t(59)=5.20, p<.001) trials. No significant main effect of sex or interaction effects were observed.

No significant main or interaction effects of background and sex on Go trial accuracy were observed.

Repeated measures ANOVAs revealed no main or interaction effects of sex and time point (baseline vs. follow-up) on the three alcohol-related survey measures (AUDIT total, YAACQ total, CCAPS alcohol/substance use).

Linear regressions revealed no significant associations between task performance and alcohol-related measures (AUDIT total, YAACQ total, CCAPS alcohol/substance use) at either baseline or follow-up.

Associations Between Network Activation and Alcohol Misuse

Associations between network activation strength for the Negative NoGo > Neutral NoGo contrast and alcohol-related measures are reported in Table 4 and Table 5. Activation strength of DAN was negatively associated with AUDIT score at baseline (p=.013), though this effect was marginal (corrected alpha value = .013 for this rank), but not at follow-up. DAN was also significantly negatively associated with the C-CAPS alcohol/substance use at baseline and follow-up (baseline p=.002, follow-up p=.005) and with alcohol-related consequences on the YAACQ at baseline (p=.004) but not follow-up. RL-FPN activation strength was significantly positively associated with C-CAPS alcohol/substance use at follow-up only (p=.003). LL-FPN and SN activation strengths were not significantly associated with alcohol or substance use measures at either time point.

Table 4.

Network activation on Negative NoGo > Neutral NoGo contrast: associations with baseline alcohol use measures.

	AUDIT score		CCAPS substance use		YAACQ total
Network	Standard β	p	Standard β	p	Standard β	p
left L-FPN	0.253	0.068	0.174	0.194	0.148	0.283
right L-FPN	−0.198	0.170	−0.012	0.930	−0.034	0.817
DAN	−0.335	0.013 *	−0.411	0.002 *	−0.404	0.004 *
SN	−0.087	0.534	0.020	0.833	−0.003	0.981

^*significant following B-H correction

Table 5.

Network activation on Negative NoGo > Neutral NoGo contrast: associations with 1 year follow-up alcohol use measures, controlling for baseline levels.

	AUDIT score		CCAPS substance use		YAACQ total
Network	Standard β	p	Standard β	p	Standard β	p
left L-FPN	0.025	0.856	0.234	0.037	0.117	0.334
right L-FPN	0.160	0.275	0.362	0.003 *	0.060	0.650
DAN	−0.142	0.335	−0.370	0.005 *	−0.289	0.040
SN	−0.060	0.672	−0.127	0.265	0.026	0.842

^*significant following B-H correction

Associations Between Network Activation and Task Performance

Associations between network activation strength for the Negative NoGo > Neutral NoGo contrast and task performance measures are reported in Table 6 and Table 7. These findings show that activation strength of SN was negatively associated with NoGo trial accuracy on negative trials (p<.001), neutral trials (p=.002) and with the difference in NoGo accuracy between negative and neutral trials (negative – neutral; p=.003). DAN activation strength was significantly associated with NoGo accuracy on negative trials only (p=.014). lL-FPN, rL-FPN activation were not significantly associated with performance.

Table 6.

Network activation on Negative NoGo > Neutral NoGo contrast: associations with NoGo trial accuracy.

	Negative		Neutral		Negative - Neutral
Network	Standard β	p	Standard β	p	Standard β	p
left L-FPN	−0.035	0.766	0.009	0.949	−0.073	0.590
right L-FPN	0.185	0.134	0.229	0.133	−0.034	0.808
DAN	0.285	0.014 *	0.161	0.223	0.241	0.068
SN	−0.651	<0.001 *	−0.444	0.002 *	−0.434	0.003 *

^*significant following B-H correction

Table 7.

Network activation on Negative NoGo > Neutral NoGo contrast: associations with Go trial reaction times.

	Negative		Neutral		Negative - Neutral
Network	Standard β	p	Standard β	p	Standard β	p
left L-FPN	0.078	0.583	−0.013	0.924	0.246	0.083
right L-FPN	0.032	0.831	0.164	0.276	−0.341	0.023
DAN	0.094	0.491	0.032	0.815	0.171	0.209
SN	−0.388	0.010	−0.399	0.008	−0.007	0.963

A repeated measures ANOVA showed no significant main or interaction effects of sex or network on network activation strength.

Discussion

The current study examined activation of large-scale regulatory brain networks during inhibitory control efforts in the context of emotionally negative versus neutral distractors in a sample of college freshmen. Drinking often potentiates deleterious impulsive actions, an effect that is heightened in the context of negative emotions. Such behaviors, including continued or escalated alcohol misuse, have significant and lasting negative effects on brain regions critical to cognitive and emotional regulation, particularly in youth (Brown et al., 2008). The failure to recruit regulatory circuitry in the context of negative emotion may also represent an important risk factor for problematic or excessive drinking. Findings revealed that less recruitment of the DAN, a network implicated in executive attention – i.e., goal-directed visuospatial attention for selecting and linking stimuli and responses, was associated with increased alcohol misuse and negative consequences of drinking, measured at baseline. DAN activation strength was further associated with the C-CAPS alcohol and substance use measure at follow-up, even when controlling for the baseline association. These findings suggest that in emerging adults with more problematic drinking, negative emotional information interferes more with engagement of large-scale brain circuitry supporting top-down attentional control than in emerging adults with less alcohol misuse. This alteration of DAN activation during inhibitory control in the context of negative emotional stimuli may also serve as an early marker of risk for future problems related to alcohol or substance use.

All four large-scale brain networks examined in the current study – lL-FPN, rL-FPN, DAN and SN - play crucial and interconnected roles in regulating how sensory information is filtered, integrated with internal references and goals, and mapped onto a behavioral response (Laird et al., 2011). While aspects of network function for these networks have previously been linked to alcohol and substance use disorders (Wilcox et al., 2019, Zhu et al., 2017) and risky drinking (Sousa et al., 2019, Worhunsky et al., 2016), in the current study, DAN network activation strength showed the strongest and most consistent relationship with outcome measures reflecting alcohol misuse and related consequences. This may, in part, reflect the specific demands of the emotional Go-NoGo task, in which a stable visuospatial priority map, or “attentional spotlight,” such as is thought to be provided by the DAN, can provide a highly effective approach to mitigating the disruptive impact of the task-irrelevant background images by narrowing focus to the small, centered, task-relevant letter. The positive relationship observed between DAN activation and accuracy on negative NoGo trials supports this possibility. However, given the absence of effects of alcohol misuse on task performance, these findings suggest that while increasing alcohol misuse is associated with reduced implementation of a relatively stable narrowing of attentional focus subserved by the DAN, it is also associated with the utilization of alternate compensatory mechanisms or strategies that support equivalent task performance. Notably, the current findings did not support the notion that this alternate mechanism was related to altered activation of SN, suggesting that the relative salience of the negative versus neutral distractor images was not driving differences in alcohol misuse. Increased right L-FPN activation was related to the C-CAPS measure of alcohol and substance use problems at follow-up but not baseline, suggesting a possible compensatory role for this executive network, though no relationship was observed at baseline or for the solely alcohol-focused measures. The absence of an effect at baseline could result from the screening out of most substance use in the freshmen year baseline sample, which then increased by sophomore year. Interpretations, however, must remain speculative in the absence of a broader pattern of effects.

While alcohol misuse was most consistently associated with altered DAN activation, task performance was linked primarily to activation of the SN. Reduced SN activation (or increased deactivation) on negative versus neutral NoGo trials was associated with improved accuracy on inhibitory trials with negative and neutral backgrounds as well as a smaller drop in accuracy on negative relative to neutral NoGo trials, suggesting that individuals who did not increase SN activation in response to negative versus neutral images were generally less distracted by the emotionally negative images and thus more able to inhibit impulsive errors on negative trials. These findings are congruent with the characterization of the SN as a network which refocuses attention to novel, surprising, or emotional stimuli. In the case of the current task, the emotional backgrounds are salient despite not being relevant to the assigned task and switching attention to these images and away from the less emotionally salient but task-relevant letter cues is detrimental to successful task performance.

Sex was not found to be a significant predictor in any of the analyses. However, sex differences might be revealed by a larger sample size and the current analyses did not look for moderating effects of sex that may exist. Future studies should examine such effects as impulsivity and risk taking are greater in men (Cross et al., 2011). Alcohol misuse is also higher in men than women, though this gap has been narrowing over the past decade, particularly among youth (Dir et al., 2017). Furthermore, among individuals with substance use disorders, women show marked increases in impulsivity and risk-taking, particularly when intoxicated (Weafer, 2020). Women are also more likely to drink alcohol to mitigate stress or negative emotions, making emotion-related impulsivity an important factor in alcohol misuse in women (McHugh et al., 2018). Indeed, affective reactivity has been reported to be associated with heavy drinking in women but not men (Pedrelli et al., 2018). Considering these sex differences, brain network activation associated with impulsivity and alcohol misuse may differ between men and women and merits further investigation.

A limitation of the current study is the inclusion of both correct and incorrect NoGo trials within the regressors for each background condition, made necessary by the relative infrequency of NoGo trials, further subdivided by background condition, and the wide range of NoGo trial accuracy scores observed in this sample. Thus, correct and incorrect trials were combined to provide a sufficient number of events for all participants in each condition. This introduces a potential confound when contrasting background conditions with different levels of accuracy. However, accuracy differences for the primary contrast of interest (negative NoGo versus neutral NoGo) in the current sample was not statistically significant. Given this equivalence in performance, the contrast should, as intended, primarily reveal differences in brain activation related to processing different types of distracting emotional information while attempting to withhold a prepotent response, rather than activation differences related to successful versus unsuccessful inhibitory control efforts.

The targeted selection of college freshmen in this study sample provides insight into the neural mechanisms at play in problematic alcohol use during the college transition, a time often associated with an increase in alcohol consumption (Schulenberg and Maggs, 2002). However, as this study is targeted towards understanding college drinking behaviors, caution must be used when generalizing results to emerging adults who do not attend college and can demonstrate distinct trajectories of alcohol and substance use relative to college-attending cohorts, including lower rates of binge drinking and higher rates of substance use (Hingson et al., 2017). The current sample was also deliberately capped in its use of drugs and tobacco in order to focus on identifying the neural correlates of alcohol use. This does, however, exclude a large population of problem drinkers who also use illicit drugs or tobacco. Further inquiry into the convergent and dissociable effects of brain network activation during inhibitory control on alcohol misuse and other forms of substance use would help extend the generalizability of the current findings into the broader population of problem drinkers.

Overall, study findings support the utility of large-scale brain network activation as an approach to exploring functional task-related brain activity. While highly nuanced parcellation of altered within-network and nodal connectivity are possible, analyses at the full network level facilitate theory-driven hypothesis testing and are useful in establishing links between network function and complex aspects of human cognition and behavior, such as alcohol misuse. The finding of reduced DAN activation with increased alcohol misuse and negative consequences of drinking during the first years of college could have future potential clinical implications as it suggests a reduced ability to attentionally screen out negative emotional content in young problem drinkers.