Attentional function and inhibitory control in different substance use disorders

Abstract

Attentional function in substance use disorder (SUD) is not well understood. To probe attentional function in SUD as a function of primary substance of abuse, we administered the attentional network task (ANT) to 44 individuals with Cocaine Use Disorder (CoUD), 49 individuals with Cannabis Use Disorder (CaUD), 86 individuals with Opioid Use Disorder (OUD), and 107 controls with no SUD, along with the stop-signal task (SST). The ANT quantifies the effects of (temporal) alerting cues and (spatial) orienting cues to reduce reaction time (RT) to targets, as well as probing how conflicting (target-incongruent) stimuli slow RT. The SST quantifies individuals’ ability to inhibit already-initiated motor responses. After controlling for sex representation and age, OUD and CaUD participants showed blunted alerting effects compared to controls, whereas CaUD and CoUD participants showed greater stimulus conflict (flanker) effects. Finally, CoUD participants showed a trend toward increased orienting ability. In SST performance, no SUD group showed a prolonged stop-signal reaction compared to controls. However, the OUD group (and CoUD group at trend level) showed prolonged “go” RT to targets and reduced hit rates. These data indicate differences in attentional function in persons with SUD as a function of the primary substance use.

1. Introduction

Substance use disorder (SUD) incurs tremendous medicolegal and psychosocial costs (Degenhardt and Hall, 2012), including harms associated with substance-induced cognitive impairment (e.g., drugged driving). For example, one study detected polysubstance use in biosamples of over 20% of fatally injured drivers (Brady and Li, 2013). Specific impairment of attention or reaction times (RT) during active intoxication by alcohol (Irwin et al., 2017), cannabis (Bondallaz et al., 2016; Ortiz-Peregrina et al., 2020), cocaine (Lundqvist, 2005), opioids (Cameron-Burr et al., 2021) or other narcotics (Saarialho-Kere et al., 1989) is well-documented. Less known are the chronic effects of these substances on attentional function (while not acutely intoxicated). Cross-sectional comparisons suggest that users of these substances show chronic impairments in attention as inferred by prolonged or variable RT compared to controls (Lundqvist, 2005).

A prevailing model of the brain’s attention system (Petersen and Posner, 2012) posits three separate components of attentional function thought to have separate anatomical and neurotransmitter substrates (Fan et al., 2005). A system of alerting in response to temporally-informative cues utilizes the norepinephrine (NE) system and the locus coeruleus to trigger a transient state of heightened alertness in preparation for an imminent, forthcoming target stimulus. Robust functioning of the alerting system facilitates motor responses to targets that are preceded by cues relative to responses to targets that are not forewarned. A system of orienting utilizes acetylcholine (ACh) to invoke parietal cortex and frontal eye fields (dorsal attention network) as well as the temporoparietal junction (ventral attention network) (Corbetta and Shulman, 2002) in response to cues of positional information about a forthcoming target, to prioritize regions of the sensory field for efficient target detection. Robust functioning of this system enables faster responses to spatially-cued targets. A third and more integrated component of attention is the executive system centered on the medial frontal cortex and the adjacent anterior cingulate cortex (ACC), as well as the insula/operculum. This system is invoked in the process of evaluating each stimulus for its contextual relevance, or “targetness”, in light of ongoing rule sets and potentially competing response tendencies. A robust executive attentional system would be indicated by a relative resistance to the “flanker” effect: a slowing of RT to targets presented along with distractors that are incongruous with the target or that prompt an opposing response tendency.

Attentional function is additionally relevant for the understanding of SUD-related cognition because sustained attention has been postulated as a required element for the successful performance of higher-order executive functions (EF) (Sturm and Willmes, 2001). Thus, poorer attention in persons with SUD may confound interpretation of response inhibition or other EF. For example, in a signal-detection task, sluggish attention to stimuli would compress the effective temporal window for referencing online rule sets to determine appropriate responses to each stimulus in time. This would result in inflated attribution of commission errors to impulsivity.

Attentional capacity has frequently been inferred from rates of successful responses (hits), RT variability, and metrics of signal detection accuracy such as d’ or A’ during response inhibition tasks such as go-nogo tasks or the stop-signal task (SST). However, all three key attentional networks described above can be directly probed with the Attentional Network Task (ANT), which was specifically designed to isolate alerting, orienting, and executive functioning (Fan et al., 2002). Function in the ANT has been tested in several populations with mental disorder. For example, participants vulnerable to social anxiety exhibited deficits in its spatial orienting component (Heeren et al., 2015; Wang et al., 2020), whereas those vulnerable to depression (Wang et al., 2020) and those with post-traumatic stress disorder (PTSD) (Leskin and White, 2007) showed impairment in the ANT executive/conflict component.

In drug users, acute withdrawal has typically been characterized by malaise and inability to sustain attention, where for example persons in extended cocaine abstinence have shown attentional decrements (Almeida et al., 2017). This has also been found in rodent models, wherein both acute withdrawal from either cocaine and heroin (Dalley et al., 2005) and extended withdrawal from cocaine self-administration (Vázquez et al., 2020) have also been shown to blunt attentional processing. To our knowledge, there have been only two studies of ANT performance in SUD. Salo and colleagues (Salo et al., 2011) applied audible alerting cues, and participants with methamphetamine use disorder showed a similar degree of RT benefit of alerting and orienting cues as in controls. However, they showed slower overall RT in all trials and more sensitivity to the flanker effect (executive deficit) (Salo et al., 2011). Conversely, persons with alcohol use disorder showed generally slower RT compared to controls, but only a decrement in the executive (flanker) component of the ANT (Maurage et al., 2014).

To better understand the long-term cognitive effects of substance abuse, we need to investigate how chronic exposure to certain drugs of abuse might selectively impact these attentional systems. Non-nicotine drugs of abuse may differ in the degree to which their chronic use perturbs the ACh system (which could affect orienting function), the NE system (which could affect alerting functions), or the dopamine (DA) system (which could affect multiple functions). For example, cannabinoids have been found to affect nicotinic ACh receptor function (Oz et al., 2014) in hippocampal cells (Nava et al., 2001), and opioids have been shown to blunt NE release from locus coeruleus cells (Ronken et al., 1993). Chronic cocaine use may affect multiple attentional systems, based on the critical role that dopamine plays in motivation (Salamone et al., 2005), cognitive effort (Westbrook et al., 2020), and related effects of acute stimulant administration to improve metrics of attention (Finke et al., 2010).

To determine how different components of attentional function may be awry in individuals with different SUDs, we utilized cases from the Phenotypic Assessment Battery (PhAB) feasibility study (Keyser-Marcus et al., 2021) wherein the feasibility of the ANT and other cognitive and mood assessments was evaluated as a potential standardized deep phenotyping battery for SUD, intended for use in clinical trials (Keyser-Marcus et al., 2021). We compared alerting, orienting, and executive function performance in the ANT between adults with no SUD and each of three SUD groups: cannabis use disorder (CaUD), opioid use disorder (OUD), and cocaine use disorder (CoUD).

To gauge the specificity of linkages between SUD to cognitive abnormalities, we also evaluated group differences in performance data on the SST, as a comparator task. The SST requires the participant to respond to “go” stimuli. In a fraction of trials, a stop signal occurs after a variable stop-signal delay (SSD) following the go stimulus, at which point the participant must curtail the response. Longer stop-signal reaction time (SSRT) is characteristic of persons with ADHD (Crosbie et al., 2013) and some SUD (Liao et al., 2014; Wang et al., 2018), as these individuals require more time to curtail an already-initiated motor response. Of the three attentional networks (Petersen and Posner, 2012), the executive network uniquely interacts with other integrative centers for goal maintenance and other higher functions. Thus, there is a possibility that executive deficits in the ANT found in a SUD group would also be found in SST performance. By including another EF task, we intended to obtain evidence as to whether EF decrements operationalized as flanker interference in the ANT generalized to other EF tasks.

Based on previous findings that substance users in various stages of withdrawal from cannabis (Broyd et al., 2016; Wallace et al., 2019) and cocaine (Almeida et al., 2017; Lundqvist, 2005) show attentional impairments compared to controls, we hypothesized that alerting (and possibly orienting) ability in the ANT would be worse in CaUD and CoUD groups compared to controls, with no difference between each other. Further, because previous cross-sectional experiments with the SST have shown that OUD (Liao et al., 2014) and CoUD subjects (Wang et al., 2018) showed increased SSRT relative to controls, but persons with CaUD did not (Smith et al., 2014), we hypothesized that executive (conflict) function in the ANT would be worse and SSRT would be longer in OUD and CoUD compared to controls.

2. Methods:

All recruitment and testing procedures were reviewed and approved by the Institutional Review Board of Virginia Commonwealth University (VCU).

2.1. Participants:

Participants were adults between 18 and 70 years old (inclusive), recruited from a pre-existing registry of individuals with and without SUD, community-based recruitment using electronic and print media and flyers, and recruitment of individuals with SUD from the VCU outpatient treatment facility. Prospective participants first underwent a brief pre-screening interview (in person or administered remotely), whereby likely-eligible individuals were invited to a laboratory screening visit. Criteria for inclusion in an SUD group required meeting DSM-5 criteria for CaUD, CoUD, or OUD, respectively. In cases of comorbid substance use or multiple SUD, the participant was grouped according to substance of preferred or greatest use. Exclusion criteria were current SUD to substances other than opioids, marijuana, cocaine, alcohol or nicotine, or meeting criteria for SUD-severe for a second (non-primary) comorbid) substance. Due to ubiquity of heavy alcohol use in non-alcohol SUD, diagnosis of mild to moderate AUD was not exclusionary. Candidates for any group were excluded for history of seizures, traumatic injury with loss of consciousness for more than 30 minutes, current or past psychosis, current mood disorder, or for suicidal/homicidal ideation. A history of depression or an anxiety disorder currently in remission was not exclusionary for any group. Demographic and other characteristics of the CaUD (n = 49), CoUD (n = 44), OUD (n = 86), and control (n = 108) groups are shown in Table 1.

Table 1.

Participant Characteristics.

		Controls	CoUD	CaUD	OUD	F/Chi-sq	p
Sex:		44 M, 63 F	33 M, 11 F	28 M, 21 F	43 M, 43 F	15.043	.002
Age:		36.4a (15.0)	52.6b (8.6)	34.4a (13.0)	42.3c (118)	20.571	<.0001
	Range	18–70	33–66	18–66	20–69
Shipley AQ (verbal) Standard Score		95.3a (16.6)	82.5b (15.6)	86.7b (15.7)	87.1b (14.15)	7.694	<.0001
	Range	34–125	45–106	47–116	50–120
Shipley BQ (nonverbal) Standard Score		98.0 (16.0)	95.6 (8.7)	95.8 (12.1)	96.0 (11.2)	0.534	.659
	Range	65–131	81–119	75–120	69–133
PROMIS Depression Score		8.4 (3.3)	8.7 (3.3)	9.2 (4.2)	9.0 (3.5)	0.631	.596
	Range	4–18	4–16	4–20	4–16
PROMIS Anxiety Score		7.5 (3.3)	8.3 (3.6)	8.0 (3.7)	8.2 (3.5)	0.9 12	.437
	Range	4–16	4–15	4–16	4–16
Pittsburgh Sleep Quality Index global score		5.4a (3.0)	6.2a,b (3.7)	7.4b,c (3.2)	7.8c (3.2)	8.584	<.0001
	Rang	0–12	1–15	1–15	1–15
PTSD Checklist-5 (PCL-5) total score		13.5a (13.4)	21.4b (17.6)	21.7b (17.9)	20.7b (15.9)	5.229	.002
	Range	0–53	0–62	0–57	0–78
Incidence of regular tobacco smoking		5 (5%)	32 (73%)	22 (46%)	58 (70%)	105.552	<.0001
Fagerstrom Test for Nicotine Dependence		0.2a (0.9)	2.6b (2.8)	2.0b (2.7)	3.1 (2.7)	30.316	<.0001
	Range	0–7	0–8	0–7	0–9
History of DSM-5 mood disorder		16 (15%)	3 (7%)	9 (18%)	8 (9%)	4.129	0.248
Incidence of alcohol use disorder-mild*		0 (0%)	4 (9%)	14 (29%)	8 (15%)	9.924	0.007*
Incidence of alcohol use disorder-moderate*		0 (0%)	7 (16%)	2 (4%)	3 (4%)	6.997	0.03*
Incidence of cocaine use disorder-mild*		0 (0%)	3 (7%)	5 (10%)	15 (18%)	124.377	<.0001
Incidence of cocaine use disorder-moderate*		0 (0%)	17 (39%)	4 (8%)	11 (13%)	--	--
Incidence of cocaine use disorder-severe*		0 (0%)	24 (55%)	0 (0%)	0 (0%)	--	--
Incidence of cannabis use disorder-mild*		0 (0%)	11 (25%)	24 (4 8%)	8 (9%)	54.193	<.0001
Incidence of cannabis use disorder-moderate*		0 (0%)	5 (11%)	15 (3l%)	8 (9%)	--	--
Incidence of cannabis use disorder-severe*		0 (0%)	0 (0%)	10 (20%)	0 (0%)	--	--
Incidence of opioid use disorder-mild*		0 (0%)	2 (5%)	1 (2%)	11 (13%)	165.936	<.0001
Incidence of opioid use disorder-moderate*		0 (0%)	0 (0%)	1 (2%)	14 (16%)	--	--
Incidence of opioid use disorder-severe*		0 (0%)	0 (0%)	0 (0%)	60 (71%)

^a,b,c-levels not connected by the same letter are significantly different

^*Chi-Square analysis spanned severity level within each substance and excluded controls

2.2. Assessments

2.2.1. Phenotyping assessments:

During each laboratory visit, study candidates completed urine drug screen toxicology and alcohol breathalyzer (as objective measures of recent substance use). Due to frequency of typical use coupled with enduring (24–48 h) urine detection of even non-cannabis drugs, on practical grounds, participants revealing positive urine results were only rescheduled if staff observed signs of active intoxication. The PhAB battery included structured interviews, including the Mini International Neuropsychiatric Interview Version 7.0. (MINI) (Sheehan et al., 1998), to determine DSM-5 diagnoses related to SUD and other mental disorders, the Columbia Suicide Severity Rating Scale (Posner et al, 2008), to assess current suicidality, and a time-line follow-back interview (Sobell and Sobell, 2008) to clarify the primary substance of abuse. The computerized Shipley Institute of Living Scale-2 (Shipley et al., 2009) was administered to obtain an approximation of crystallized and fluid IQ. Individuals who were selected for study participation were then invited to return to complete the NIDA Phenotyping Assessment Battery (NIDA PhAB), a feasibility study of administration of behavioral performance and self-report measures assembled to assess individuals along six neurofunctional domains germane to SUD (Keyser-Marcus et al., 2021). The phenotyping self-report assessments included the PROMIS Depression 4a and Anxiety 4a scales (Broderick et al., 2013), the PTSD Checklist-5 (Weathers et al., 2013) and the Fagerstrom Test for Nicotine Dependence (Heatherton et al., 1991). We also obtained scores from the Pittsburgh Sleep Quality Index (PSQI) (Buysse et al., 1989) to control for poor sleep in SUD (Angarita et al., 2016). Phenotyping battery performance measures included the ANT and SST on a Windows PC workstation, administered using Inquisit 5 Lab software (Millisecond Software LLC, Seattle WA).

2.2.2. Attentional Network Task (ANT):

The NIDA PhAB contained an abbreviated version (approximately 9 minutes) of the original ANT task (Fan et al., 2002). The ANT presents “flanker” arrows as targets in one of two locations near a central fixation point, the direction of which must be indicated by the participant with a key press. In alerting trials, asterisks appear just before the target at the central fixation point (replacing the crosshair). In orienting trials, an asterisk in the actual location of the forthcoming target is presented. In other trials, the target is not preceded by any asterisk cues. Within each of these three trial types, some trials are “congruent”, wherein arrows of identical direction to the central target arrow appear to the right and left. In “incongruent” trials, the flanking arrows are in opposite direction as the target arrow. Each trial began with 900 ms of central fixation with a “+” crosshair, followed by the presentation of an arrow target pointing left or right that appeared either above or below central fixation until participant response within a 1700 ms window (Figure 1). The participant was instructed to respond on either the left or right mouse button to indicate the direction of the central target arrow. The target window was then followed by a variable (0–2800 ms) interval of “+” central fixation that resulted in a mean trial length of 4 s each. In “incongruent” trials, the target arrow was flanked by two arrows at right and at left that pointed in the opposite direction to the target arrow. In “congruent” trials, the four flanking arrows pointed in the same direction as the central target arrow.

Figure 1.

Diagram of Attentional Network Task. In each trial of the Attentional Network Task (ANT), the participant must report the direction of the center arrow of an array of arrows with the corresponding mouse button (L or R) within a 1700 ms window. Trials began on average 4 s apart, and were separated by a variable intertrial interval (ITI). The top screen sequence illustrates a “no-cue” trial, in where there is no spatial or temporal warning cue prior to the arrow target presentation. In this example, the flanking arrows are congruent with the central, left-pointing target arrow. The middle screen sequence illustrates an “alerting” cue trial in which a centrally-located row of asterisks replaces the fixation cross for 100 ms, 500 ms prior to the target. In this example, the flanking arrows are incongruent with the target arrow, and point in the opposite direction. The bottom screen sequence illustrates an “orienting” cue trial, wherein a row of asterisks appears 500 ms prior to the target in the precise location of the forthcoming target. In this example, the arrow target itself is congruous, because the flanking arrows point the same direction as the central arrow target.

In a factorial trial design, within each of the incongruent and congruent trial classes, there were three further trial subtypes. In center-cue trials, 400 ms into the anticipatory fixation period, the central “+” was replaced by an asterisk for 100 ms before reverting back to the “+” for the remaining 400 ms. Therefore, the participant was given a fixed 500 ms notice of the forthcoming target, but with no information about whether the target would appear above or below fixation. In spatial-cue trials, 400 ms into the anticipatory fixation period, an asterisk appeared for 100 ms either above or below the central fixation crosshair, in the exact screen location programmed for the forthcoming target, beginning at 500 ms before the target. This spatial cue thus provided both timing (alerting) and location information. By comparison, the center cue could be considered “alerting-only.” In the no-cue control condition, there was only the 900 ms of central +” before the target, offering no warning. Participants first completed 12 practice trials with feedback, then completed the main task of 120 trials, composed of n = 5 repetitions of each combination of: 3 (cue conditions: alerting, orienting, none) × 2 (flanker conditions: congruent, incongruent) × 2 (target positions: above, below central fixation) × 2 (target directions: left, right).

In the Stop Signal Task (SST) (Verbruggen et al., 2008), each 2-s trial began with a presentation of a fixation circle for 500 ms, participants were then presented with an arrow within this circle that either points right or left. Participants were instructed to press the “D” key if the arrow pointed to the left and press the “K” if the arrow pointed to the right. However, the participant was to withhold his or her response if an audible stop signal beep was presented shortly after the arrow. This stop signal was presented pseudorandomly in one-third of trials. The initial stop-signal delay (SSD) between the onset of the target and the stop signal was set to 250 ms. Using a staircase algorithm, the SSD was prolonged by 50 ms (up to 1150 ms) after each successful inhibition, but reduced by 50 ms after each failure to stop. This titration was intended to arrive at an SSD value that stabilizes near the point where the participant could successfully inhibit responding in 50% of stop trials. Participants were instructed to respond as quickly and accurately as possible and to not “wait out” a trial to see if a stop signal beep occurred.

Participants completed one practice block of 32 trials (8 stop trials: 24 no-stop trials) followed by three test blocks of 64 trials each (16 stop trials: 48 no-stop trials). The key metric of the SST is the stop-signal reaction time (SSRT), where longer SSRT values are indicative of a more “ballistic” uninhibited response style that is difficult to interrupt once initiated, such as in persons with ADHD traits (Crosbie et al., 2013; Janssen et al., 2015). Per a “race model” between response-initiating and response-curtailing processes (Verbruggen and Logan, 2009), the software calculated SSRT as the difference in RT between the mean RT in “go” trials (irrespective of the correct report of arrow direction) and the mean programmed SSD across all three runs of the task. Other metrics include mean rates of responding in stop-signal trials (percent commission errors) as well as mean RT to respond in stop trials. Finally, the software generated probability that a participant’s rate of successful stops in stop trials significantly deviated from 50%, which violates the race model and thus invalidates SSRT and SSD values.

2.3. Data analysis:

Data analysis was performed with JMP-SAS (SAS Institute, Inc). To characterize the within-subject effects of ANT cueing and target type, repeated-measures ANOVA was used to detect effects of ANT trial type on accuracy rates and RT. Each analysis featured two levels of target arrow congruency (congruent, incongruent) X three levels of cueing (no cue, alerting-(only) cue, spatial cue) as within-subject factors. For the calculation of the three primary attention metrics, only response data from correct-response trials were used. For each of the six ANT trial conditions, we applied the “Absolute Deviation around the Median”-(ADM) trimming method (Leys et al., 2013) to correct for one-tailed RT outlier distributions. ADM-trimmed mean RT values were then used per the calculations of Fan et al (Fan et al., 2005) to calculate the alerting effect as RTno-cue – RTcenter-cue, the orienting effect as RTcenter-cue – RTspatial-cue, and the conflict (executive) effect as RT_incongruent – RT _congruent (all cue type trials collapsed).

Next, the key attention metrics from the ANT were compared between groups in four simultaneous multiple regression analyses. The CoUD and OUD groups had an older mean age than control and CaUD groups, with greater representation of males in the CoUD group (Table 1). Therefore, in light of the normative decrease in attentional and other EF performance with adult aging (Cohen et al., 2019), in each regression analysis, group, age and sex were entered into the model (along with age and sex interaction effects with group) to account for age and sex differences between groups. Each of alerting, orienting, or conflict from the ANT along with SSRT from the SST was the dependent variable. For each pairwise detection of a performance difference in a SUD group relative to controls, follow-on analyses ware performed wherein: 1) PSQI global scores were entered into the model to ascertain whether accounting for poorer sleep in SUD eliminated the group difference and 2) SUD participants with comorbid AUD were excluded to determine if group differences were potentially driven by chronic alcohol effects in some SUD participants.

3. Results:

Due to technical error, failure to comprehend or comply with task instructions, or having been administered a (discontinued) longer ANT version at PhAB study launch, 60 participants (n = 13 CoUD; n = 4 CaUD; n = 17 OUD; n = 26 controls) were excluded from ANT analysis. Six participants were excluded from all SST analysis.

3.1. Effects of ANT trial type on ANT performance

3.1.1. ANT Accuracy

As shown in Figure 2, Part A, participants were generally near ceiling performance in correctly indicating the direction of the central target arrow (within the time window) in congruent trials. Repeated-measures analysis indicated main effects of target incongruence (F(1,225) = 118.919, P < .0001) to reduce response accuracy. There was also a main effect of target cue type on accuracy (F(2,224) = 9.916, p < .0001), where an interaction effect of cue type X congruency (F(2,224) = 5.747, p = .004) on accuracy indicated that the effect of cues to improve accuracy was limited to the incongruent trials.

Figure 2.

Main effects of ANT trial conditions. Shown here are the accuracy rates (reporting proper direction of target arrow within the time window) (Panel A) and outlier-trimmed mean reaction times (RT) to trial targets (Panel B) as a function of whether the trial featured an advance warning cue (alerting or orienting) and whether the target arrow was congruous or incongruous with the direction of the flanking arrows. For both accuracy and for mean RT values, there were significant main effects of cue type and target congruency, as well as cue type X congruency interaction effects.

3..1 2. ANT reaction times

Repeated-measures analysis indicated main effects of target incongruence (F(1,225) = 978.432, p < .0001) to increase RT (Figure 2 Part B). There was also a main effect of target cue type (F(2,224) = 346.935, p < .0001) on RT, where alerting(-only) and spatial cues sped RT relative to uncued targets. Finally, there was a significant congruency X cueing interaction effect on RT (F(2,224) = 18.637, p < .0001), wherein the alerting(-only) cue sped RT relative to the uncued condition only when target arrows were congruent.

3.2. Main and interactive effects of group on ANT performance

For illustrative purposes, raw group means and uncorrected group differences are enumerated in Table 2. The following primary (regression) analyses control for group differences in age and sex representation, with adjusted group means illustrated in Figure 3.

Table 2.

	Controls	CoUD	CaUD	OUD	F-statistic	P
Attentional Network Task (ANT)
Alerting (ms)x	28.2 (30.7)	11.4b(27.5)	11.4b(34.7)	11.1b(43.5)	3.956	.009
Orienting (ms)x	59.7 (35.6)	49.9b (56.1)	58.4 (38.6)	67.7 (45.6)	1.357	.257
Executive/conflict (ms)x	119.6 (65.0)	159.0a(54.4)	140.5a(68.3)	122.2 (54.4)	3.896	.001
Stop Signal Task
Accuracy rate in no-stop trials (in %)	94.7 (8.3)	92.6 (11.2)	93.2 (8.0)	90.0b(10.7)	3.915	.009
Reaction Time in no-stop trials (ms)	687.6 (191.9)	756.6a (163.6)	746.3 (175.2)	776.4b(175.0)	4.129	.007
Rate of responses-stop trials (in %)	43.5 (7.0)	44.1 (14.8)	47.1 (16.9)	43.2 (10.7)	1.308	.272
Stop-signal reaction time (SSRT) (ms)y	246.3 (59.5)	249.1 (62.7)	241.8 (54.8)	254.8 (58.5)	0.392	.759
Stop-signal delay (mean) (ms)y	403.2 (190.5)	479.9 (187.2)	502.6b(185.6)	494.9b(199.5)	3.710	.013

^adifferent from controls at p < .10 in pairwise analysis

^bdifferent from controls at p < .05 in pairwise analysis

^xOutlier-trimmed values used for mean RT

^yOnly participants with stop rates not significantly different from 50%

F-statistics and p values pertain to uncorrected main effect 006Ff group in one-way ANOVA (not controlling for sex or age)

Figure 3.

Shown here are adjusted mean values for each of the ANT alerting effect (Panel A), orienting effect (Panel B) and Executive/conflict effect (Panel C), after inclusion of age, sex, and their interactions with group in the regression model. Error bars represent 95% confidence intervals. Horizontal brackets denote significant difference at p < .05.

3.2.1. ANT Alerting

There was a main effect of group on alerting (F(3,214) = 2.884, p = .039), but no main or interactive effects of age or sex on alerting (Figure 3). Pairwise comparisons indicated that the main group effect was driven by lower alerting performance in each of CaUD (p = .006) and OUD (p = .019) participants relative to controls. Raw mean and variance values of task performance are shown in Table 2 along as well as group-difference statistics after inclusion of age and sex in the regression models. Post-hoc sensitivity analyses indicated that neither controlling for poorer sleep quality in SUD (by inclusion of PSQI global scores in the model) nor excluding participants with mild-moderate comorbid AUD appreciably affected statistic significance of group differences in alerting.

3.2.2. ANT Orienting

There was a main effect of advancing age to diminish orienting ability (F(1,214) = 15.499, p < .001) as well as an age X group interaction effect (F(1,214) = 3.727, p = .021) and a trend toward a group difference (F(3,214) = 2.434, p = .066) in orienting. Exploratory follow-on analyses indicated that these effects were driven by improved orienting in the CoUD group compared to controls (p = .028) that was unaffected by addition of PSQI sleep scores in the model but was only a trend (p = .06) when CoUD participants with comorbid AUD were excluded. Exploratory bivariate correlations indicated that the age X group interaction was driven by negative associations between age and orienting in the CoUD (Spearman r = − .440, p = .013) and CaUD groups (Spearman r = − .315, p = .035) that were not evident in control and OUD groups (all p > .4). No other main or interactive effects of group, age or sex on orienting were significant.

3.2.3. ANT Executive (conflict)

There was a main effect of group on conflict effects (of incongruous flankers to increase RT) (F(3,314) = 3.403, p = .019) and trends toward main effects of advancing age (F(1,214) = 2.901, p = .090) and female sex (F(1,214) = 3.068, p = .081) to associate with increased conflict. Pairwise comparisons indicated that the trend was driven by more severe conflict effects compared to controls in both CaUD participants (p = .026) and CoUD (p = .028) participants. These group decrements were unaffected by inclusion of PSQI scores in the models or by the exclusion of participants with comorbid AUD. The interaction effects of group with either age and sex on conflict effects were not significant.

3.3. Stop-signal task performance

Hit rates were generally high overall, where on average each group correctly reported the direction of the arrow in no-stop trials at least 90% of the time. There was a main effect of group on hit rates in no-stop trials (F(3,268) = 3.585, p = .014) that was driven by lower hit rates in OUD (p = .002) and CoUD (p = .077) participants compared to controls. There was also a trend toward a group X age interaction (F(3,268) = 2.129, p = .097) on hit rates, driven by OUD and CoUD groups. With regard to RT in the no-stop trials, there was a main effect of group (F(3,268 = 3.862, p = .01) that was driven by slower responses in the OUD group (p < .001) and the CoUD group (.066). A significant age X group interaction effect (F(3,296) = 4.758, p = .003) was driven by a positive association between age and go RT in the control group (Spearman r = .238, p = .014) and a negative association between age and go RT in the OUD group (Spearman r = −.251, p = .022). Finally, all groups successfully stopped in 53–57% of stop-signal trials with no main effect of group. However, there was a significant group X age interaction (F(3,268) = 3.003, p = .031) and a trend toward a main effect of age (F(1,268) = 3.303, p = .07) on rates of successful stops. These effects were driven by decreasing rates of successful stops with age in the OUD group (Spearman r = −.307, p = .005).

Analysis of SSRT and mean SSD was restricted to those participants whose stopping rate did not significantly differ (p >.05) from the ideal 50% rate. This reduced subset of participants was composed of n = 29 CoUD, n = 39 CaUD, n = 58 OUD, and n = 88 controls. The omnibus regression indicated only a trend toward a main effect of sex on SSRT (F(1,202) = 3.692, p = .055) with higher adjusted mean SSRT in women (264.8 ms) compared to men (238.9 ms).

Analysis of SSD values indicated solely a significant age X group interaction effect (F(1,202) = 2.851, p = .039). This was driven by trends toward a positive correlation between age and SSD in controls (Spearman r = .186, p = .084) but a trend toward a negative correlation in the OUD group (Spearman r = −.240, p = .069). There were no main or interactive effects of group, age, or sex on SSD.

3.4. Correlations among ANT and SST performance metrics and premorbid IQ estimates

As shown in Table 3, there were some within-subject correlations between ANT, SST, and Shipley scores. Within the ANT, conflict effects did not correlate within-subject with either alerting or with orienting. Conversely, alerting ability correlated negatively with orienting (Spearman r = −.289, p < .0001). Among the participants with valid SSRT values in the SST, SSRT correlated positively with the ANT conflict effect (Spearman r =.173, p = .028). In other words, subjects who were prone to the incongruous flanker arrow effect to slow responses in the ANT also required more time to stop a prepotent response in the SST. SSRT did not correlate with either alerting or with orienting. Shipley AQ and BQ scaled scores, derived from vocabulary plus abstraction (AQ) and block-design (BQ), respectively, also correlated negatively with executive/conflict effects in the ANT.

Table 3.

	ANT Alerting	ANT Orienting	ANT Executive	SST SSRT	Shipley AQ	Shipley BQ
ANT Alerting ANT	1.00
Orienting ANT	−.289 c	1.0
Executive	−.065	.036	1.0
SST SSRTx	−.060	.054	.173 a	1.0
Shipley AQ	.192b	.091	−.263 c	−.076	1.0
Shipley BQ	.096	.041	−.200 b	−.020	.522 c	1.0

^adenotes P < .05

^bP < .01

^cP < .001

^xOnly participant subset with valid SSRT values included

4. Discussion:

We administered the ANT to a relatively broad and diverse sample of individuals with several different primary SUD, such as OUD, CoUD, CaUD as well as control participants to determine if substance of choice was associated with performance deficits in three key areas of attentional function. Compared with controls, we found that subjects with CaUD or OUD showed reduced alerting (response readiness) attention on the subtest of ANT. We also found that both CaUD and CoUD participants demonstrated an exaggerated effect of incongruent (flanker) stimuli to delay responding to target arrows compared to controls in the conflict/executive trials of the ANT. These decrements remained significant after removing participants with comorbid AUD in follow-on analyses. Interestingly, after controlling for their older age and male predominance, CoUD participants showed a trend toward improved spatial orienting function of attention, which reflects the ability to leverage advance information about the location of the forthcoming target. Finally, the groups did not differ in impulsivity as indexed by SSRT in the stop-signal task.

Alerting capacity represents the ability for a temporally-informative cue to mobilize proactive attentional control just prior to appearance of a target, in order to respond more readily to that target. Whereas control subjects responded on average 27 ms faster to cued vs uncued targets (adjusted means), this difference was only about 8–12 ms in CaUD and OUD, respectively. The observation that CaUD and OUD participants on average cannot leverage a temporal warning compared to controls is in line with findings that persons with chronic cannabis use (Bondallaz et al., 2016; Dahlgren et al., 2020) and chronic illicit opiate use (Cameron-Burr et al., 2021) show attentional deficits in tasks that require constant vigilance for environmental events, such as simulated driving. The alerting cue in the ANT shows conceptual similarities with an activated brake light of the car one is following, as such a situation signals an impending speed reduction that necessitates preparing to brake to avoid collision.

We also detected an exaggerated executive/conflict effect in the ANT task performance in CaUD and CoUD participants. Compared to controls, CaUD and CoUD participants were more prone to the “flanker effect” of incongruous stimuli flanking the target arrow to retard RT. Whereas controls were on 123 ms slower (adjusted mean) to report the direction of target arrows with incongruent vs congruent flanking arrows, this slowing was over 140 ms in participants with CaUD and 172 ms in participants with CoUD. There was also a trend toward a main effect of age to increase vulnerability to the flanker effect. Such age effects on the executive ANT component was also found in the Sperduti study (Sperduti et al., 2016).

Regarding spatial orienting, after controlling for age and sex, CoUD participants showed greater orienting in the ANT compared to controls. Controls responded roughly 60 ms faster (adjusted mean) to targets preceded by cues that were informative about the location of the upcoming target compared to targets that were preceded by centrally-appearing cues that provided a temporal alert alone. CoUD responded 93 ms faster when targets were spatially cued. However, unadjusted means (Table 2) indicate slightly reduced orienting in CoUD compared to controls. Because there was a main effect of age, wherein older participants showed blunted spatial orienting, and because CoUD were significantly older than controls, adding age and the age X group interaction terms to the model likely inverted the group difference. We note that worse spatial orienting with advancing age may extend to neurotypical adults in that advancing age magnified the RT difference in responding to targets preceded by valid vs invalid (misdirecting) spatial cues in the Posner attention task in several reports of neurotypical adults (e.g. (Arif et al., 2020)). Future studies with more precisely age-matched groups could illuminate these findings.

Several laboratory studies have shown that sleep deprivation degrades attentional function and hinders RT, including in the ANT (Martella et al., 2011). We therefore performed post hoc sensitivity analyses to indirectly assess whether attentional decrements in SUD groups could simply be attributed to poorer sleep common in SUD (Angarita et al., 2016). To index this, PSQI scores were added to the multiple regression analysis as a simultaneous covariate. However, the group-wise attentional decrements described above still survived. It may be that because PSQI total scores were actually on average in the problematic (5+) range in our controls, the effects of PSQI scores could have been minimized in the models derived from our sample. Alternatively, the PSQI might be an imperfect metric of actual sleep deprivation in that one recent study found no relationship between PSQI scores and other subjective metrics of sleep quality in neurotypical young adults (Zavecz et al., 2020).

Of all three ANT components, we found that the executive component was the most consistently and strongly linked to the IQ approximations of Shipley-2 AQ and BQ scores. Moreover, the executive component of the ANT was the only component that correlated with SSRT (inhibitory capacity) of the SST. Of note, SSRT itself is thought to index a general executive functioning factor (Friedman and Miyake, 2017). This specificity of the executive network association with IQ reflects the Peterson-Posner conceptual model of components of attention (Petersen and Posner, 2012), wherein the executive component indexed by the incongruent flanker effect is thought to be attentional network most heavily integrated with other brain networks, such as by virtue of its requirement for resistance to interference and inhibitory control.

We also found that there were no group differences in SSRT itself as the primary metric of the SST. This stands in contrast to several other reports wherein SUD, especially disordered use of stimulants (but not cannabis) (Lee et al., 2019) has been linked to longer SSRT. We posit two primary possible reasons for a negative finding. First, despite administering standard task instructions for the SST, roughly 31% of cases (86/280) had to be dropped from analysis of SSRT for having stop probabilities significantly different from the ideal of 50% despite the programmed titration of SSD across trials to attain this. All groups generally inhibited 54–57% of the time, suggesting that despite standard experimenter instruction to press quickly, participants tended to try to wait out stop signals. Of note, both “Go RT” values in no-stop trials and mean SSD values tended to be over 200 ms higher than those reported in previous SST experiments in SUD populations (Filbey and Yezhuvath, 2013; Li et al., 2006). Using the typical calculation of SSRT as mean ‘go’ trial RT minus mean SSD, SSRT values were artificially compressed to those akin to the previous reports.

Second, our controls were not supranormal. Rather than university campus controls of convenience, our community-recruited controls were likely less cognitively adept compared to controls in other cross-sectional studies of cognition in SUD. In our sample, the mean scaled Shipley-2 AQ and BQ scores of controls were only 95.5 and 98.2 respectively, such that BQ (non-verbal) IQ estimates were no higher in controls compared to SUD groups. Inasmuch as SSRT has been found to be solely a component of a general g-like cognitive factor (Miyake and Friedman, 2012), it stands to reason that there would also be a minimal decrement in SSRT in our SUD participants compared to our controls. Collectively, our negative finding with SSRT should be interpreted with caution and may not generalize to other participant samples or to SST experiments with incentive structures or other features that better deter waiting out trials.

Taken together, our findings could be seen as supporting abnormalities in different aspects of attentional processing that are somewhat specific to certain SUD. Participants with CaUD and OUD did not benefit much from alerting signals in the ANT, where OUD also showed longer reaction times on go trials in the SST. These performance patterns could reflect deficiencies in ability to proactively recruit cognitive resources within contexts that provide early attentional cues meant to facilitate rapid response. Indeed, executive control has been characterized as involving both proactive mechanisms, which bias attention to optimally respond to task demands, and reactive mechanisms, which are activated during performance monitoring or to resolve conflicting information (Braver, 2012). Proactive attention toward alerting cues in the ANT is an optimal strategy for enhancing reaction time, and an inability to enhance attention when presented with temporal cues could represent a failure to recruit proactive cognitive task engagement.

However, this lack of proactive attention that minimizes the alerting effect in CaUD and OUD did not appear to translate to the orienting effect. It may be that spatial cues that can elicit saccades could activate reactive attention toward the target area, where such added visual salience could also explain improved (age-corrected) orienting in CoUD. This possibility is supported by findings in older adults, who typically value reactive accuracy over proactive speed in performance tasks, and demonstrate a diminished alerting effect but intact orienting effect (Williams et al., 2016). Proactive attention is also necessary for fast responding on the SST, as careful attention for the presentation of target arrows allows one to respond quickly on go trials. Instead, a more reactive cognitive strategy may favor reduced vigilance for target appearance and slower reactions, which additionally facilitates later stopping (increased SSD). To further investigate this possibility of a shift in cognitive control functions in CaUD and OUD, future studies ought to critically examine specific disruptions in the neural salience network, which is trans-diagnostically linked to cognitive perturbations in psychopathology (McTeague et al., 2017, 2016), as a function of exposure to different substances of abuse. To isolate SUD decrements in proactive vs reactive control, such studies could use versions of the SST that present cues that indicate probability of needing to stop (Zandbelt and Vink, 2010).

In addition to the atypical SST behavior in our sample, these findings should be considered in light of other limitations. Most importantly, there were significant differences in sex and age between groups. The pragmatic approach of the core PhAB study was to include individuals with SUD “as they are.” This not only resulted in CoUD participants being older and predominantly male compared to other groups, but also required allowing for SUD comorbidities. Notably, polysubstance use is ubiquitous in the United States (Crummy et al., 2020), and polysubstance use disorder is characteristic of most SUD treatment seekers in some programs (Lin et al., 2021). Therefore, attentional differences associated herein with certain substances of abuse cannot be attributed to neurotoxicity from the preferred substance with certainty. We caution too that observed group differences could even be manifestations of premorbid neurocognitive features that lent themselves to drug use (Crews and Boettiger, 2009) as well as governed preference for regular use of one drug class over another (Badiani et al., 2011). Second, as with any cross-sectional comparison of cognition, the degree to which impaired motivation in a participant group contributed to its cognitive task performance deficit cannot confidently be determined. However, it could be argued that sustained motivation when performing a dull or strenuous task is itself a (dopaminergic) component of executive function (Cools et al., 2019; Westbrook et al., 2020). We also note that orderly within-subject effects of trial conditions on RT in the ANT suggest that our participants attended reasonably well to the task.

Moreover, due to concerns over participant testing burden, we discontinued the full, original version of the ANT that also presented “double-cue” trials composed of asterisk rows both above and below fixation (Fan et al., 2002) and switched to a shortened version. This mid-project change in testing, together with other instances of task noncompliance resulted in several cases omitted from ANT analysis. Also to minimize overall burden (while focusing on cognitive phenotyping), questioning on substance use histories in the source study was limited to identification of DSM-5 diagnoses and estimation of preferred substance. Finally, to accommodate participant availability, attentional tasks were not administered at the same time of day in all participants, and testing was scheduled irrespective of participant chronotype (which was not probed as part of the PhAB battery), thus introducing chronobiological variance.

In conclusion, this study explored attentional function aspects in several different SUDs all probed in the same deep phenotyping feasibility study that included the ANT as a dedicated task of attention. Data analysis revealed that persons with CaUD and OUD showed blunted capacity for alerting attentional function and persons with CaUD and CoUD showed increased conflict interference effects (executive function) compared to non-SUD control participants, even when ostensibly not acutely intoxicated. Moreover, these attentional decrements were detected in the general absence of overall cognitive decrements vs. control, as indexed by SSRT and the Shipley-2 BQ scaled scores. Our findings of blunted alerting in CaUD and OUD participants comport with, and could perhaps partially explain the aberrant performance of cannabis or opioid users in driving simulations (Cameron-Burr et al., 2021) (even while sober (Brown et al., 2019)), as well as their greater rates of motor vehicle violations (Macdonald et al., 2004) and accidents (Windle et al., 2021). Our findings of increased conflict/executive interference effects in persons with CaUD and CoUD reflect their poorer performance on executive function tasks in some reports (Crean et al., 2011; Spronk et al., 2013) and perhaps their blunted engagement of frontocortical circuitry by inhibitory tasks (Zilverstand et al., 2018). Future experiments, such as in non-polysubstance-using populations, could recruit monosubstance users or users with less comorbidity, or could add pupillometry or other physiological metrics of arousal to the ANT, as well as additional phenotypic assessments of real-world driving and other attentional/motor function. Finally, larger sample sizes might allow for an exploration of potential sex differences in attention in SUD.

HIGHLIGHTS.

• We administered the Attentional Network Task to controls and to persons with Cocaine Use Disorder (CoUD), Cannabis Use Disorder (CaUD), Opioid Use Disorder (OUD)

• After controlling for group differences in sex representation and age, both OUD and CaUD participants showed blunted alerting effects compared to controls

• After controlling for sex representation and age, both CaUD and CoUD participants showed greater conflict effects of target-incongruent stimuli than controls

• No substance use disorder (SUD) group showed prolonged stop-signal reaction time compared to controls

• These data indicate differences in components of attentional function between persons with SUD as a function of the substance of choice