Alcohol Attention Bias Modulates Neural Engagement during Moral Processing

Abstract

The neurobiology of typical moral cognition involves the interaction of frontal, limbic, and temporoparietal networks. There is still much to be understood mechanistically about how moral processing is disrupted; such understanding is key to combating antisociality. Neuroscientific models suggest a key role for attention mechanisms in atypical moral processing. We hypothesized that attention-bias towards alcohol cues in alcohol use disorder (AUD) leads to a failure to properly engage with morally relevant stimuli, reducing moral processing. We recruited patients with AUD (n = 30) and controls (n = 30). During functional magnetic resonance imaging, participants viewed pairs of images consisting of a moral or neutral cue and an alcohol or neutral distractor. When viewing moral cues paired with alcohol distractors, individuals with AUD had lower medial prefrontal cortex engagement; this pattern was also seen for left amygdala in younger iAUDs. Across groups, individuals had less engagement of middle/superior temporal gyri. These findings provide initial support for AUD related attention bias interference in sociomoral processing. If supported in future longitudinal and causal study designs, this finding carries potential societal and clinical benefits by suggesting a novel, leverageable mechanism and in providing a cognitive explanation that may help combat persistent stigma.

INTRODUCTION

The ability to make moral judgements likely evolved as part of a package of social cognitive processes supporting well-functioning social group behavior (Cosmides et al., 2018); thus, understanding moral cognition is a key goal for social neuroscience. Investigation into this topic focuses on the neural and cognitive dynamics underlying moral cognition, including its development, dysfunction, and translation to behavior. Societally, actions that are perceived or identified as harmful are defined as “wrong” or “immoral” (i.e., antisocial behaviors), while those that promote the wellbeing of the community are “right” or “moral” (i.e., prosocial behaviors; Schein & Gray, 2018); individual decision making processes are more complex.

Neuroscience has attempted to describe the mechanisms underlying intact moral processing, first with the dual process model (Greene et al., 2008). When the “correct” answer is less clear (i.e., harm is the same across both conditions in utilitarian moral processing dilemmas like the Trolley problem), Greene and colleagues found that both intuitive emotional processes within limbic circuits and top-down prefrontal cognitive processes were engaged (Greene et al., 2001). The more strongly evoked process outcompetes the weaker one, leading to judgements based either on the objective utility or the aversive response. However, the consensus in the field is increasingly that a more dynamic model more accurately fits naturalistic decision making (Van Bavel et al., 2015). For example, Moll and colleagues emphasize the importance of social context, rules, and convention processing in the temporal and parietal lobes (Moll et al., 2005). Meta-analysis has implicated areas consistent with these models as being involved in moral processing; specifically, these are frontal (ventromedial/dorsomedial prefrontal cortex [vm/dmPFC], frontal pole, orbitofrontal cortex [OFC], anterior cingulate cortex [ACC]), limbic (amygdala, posterior cingulate [PCC]), and temporal regions (temporoparietal junction [TPJ], temporal pole, superior/middle temporal gyrus [STG/MTG], superior temporal sulcus [STS], insula) (Bzdok et al., 2012; Fede & Kiehl, 2019). Finally, it is likely that the interaction of these brain regions, driven by vmPFC-amygdala connectivity, is involved in moral processing (e.g., Jung et al., 2016), either through development of moral values via reinforcement learning in early childhood (Blair, 2007) or through regulation and integration of emotion into decision making (Kim et al., 2011). This neuroscientific evidence, taken together, suggests that moral cognition is a balance of top-down decision making weighed against internal emotional senses of harms/rewards and societally defined system of values, standards, or contexts.

The study of abnormal moral cognition includes a large focus on psychopathy and vmPFC lesion patients who often develop psychopathic-like traits (e.g., Phineas Gage; Saver & Damasio, 1991). Psychopathy is a personality disorder characterized by moral violations, shallow affect, lack of remorse, parasitic lifestyle, criminal behavior, and sensation seeking (Hare, 2003). Lesion studies have illustrated that patients with damage to their vmPFC, including frontotemporal dementia patients, make more utilitarian moral decisions than typical individuals (Baez et al., 2014; Koenigs et al., 2007). Psychopathy has been associated with a number of neurobiological differences, including reduced gray matter volume in paralimbic regions (Ermer et al., 2012), disrupted frontal and temporal function during emotion-cognition interaction (Müller et al., 2008), and altered resting functional connectivity within and between salience, visual, executive control, and attentional networks (Espinoza et al., 2018). Importantly, although individuals high in psychopathy have different neural engagement during moral processing (Fede, Borg, et al., 2016; Glenn et al., 2009), they do not appear to have any difference in knowledge of moral rules (Aharoni et al., 2012); as stated succinctly in the title of one report, “Psychopaths know right from wrong but don’t care” (Cima et al., 2010). Transcranial magnetic stimulation (TMS) studies indicate that disruption of other sites (i.e., TPJ, dorsolateral prefrontal cortex) produce psychopath-like moral judgements (Jeurissen et al., 2014; Young et al., 2010).

Although psychopaths definitionally engage in criminal behaviors, most individuals with criminal histories are not high in psychopathic traits (Hare & Neumann, 2010). There are some studies finding atypical brain structures and activity in individuals who commit homicide (Cope, Ermer, et al., 2014; Raine et al., 1998) and violent offenders (Leutgeb et al., 2016), but other work has found no notable differences in activation associated with moral processing between lower security inmates and community volunteers (Fede, Borg, et al., 2016). In fact, though reoffending is predictable using neurobiology, the models rely on response inhibition-related ACC activity (Aharoni et al., 2013; Meldrum et al., 2018) or intrinsic measures like cerebral blood flow in parietal and temporal lobes at rest (Delfin et al., 2019) and structurally derived brain-age definitions (Kiehl et al., 2018) to predict criminal activity, rather than regions classically associated with moral processing Thus, the relationship between antisocial behavior and moral cognition is not as clear as definitions of morality may imply.

One explanation for the lack of a clear association between antisociality and moral cognition is the imperfect overlap between criminal and immoral behaviors. Not all immoral behaviors are illegal (e.g., cheating on a partner), and not all criminal behaviors are intrinsically morally relevant (e.g., substance use). Substance and alcohol use are associated with crime commission (BJS, 2004; Browne et al., 2022; Popovici et al., 2012) and increased odds of violence (Zhong et al., 2020), with recovery accompanied by reduced crime commission and more work productivity (Martinelli et al., 2020). However, and contrary to harmful moralization of substance the mechanistic link between moral cognition and antisociality in individuals with substance use disorders (SUDs) including alcohol use disorder (AUD) has not yet been clearly described. Individuals with SUDs are not characteristically high in psychopathic traits. In fact, higher levels of psychopathic traits are associated with less drug-cue reactivity (Cope, Vincent, et al., 2014). That being said, previous studies do find SUDs are associated with atypical processing during moral cognition, including increased utilitarian responses in polysubstance users (Carmona-Perera et al., 2012) and individuals with AUD (Khemiri et al., 2012), blunted heart rate in AUD (Carmona-Perera et al., 2013), reduced frontolimbic connectivity in cocaine use disorder (Verdejo-Garcia et al., 2014), and reduced hemodynamic response in the ACC and amygdala regions associated with moral processing in individuals with stimulant use disorders (Fede, Harenski, et al., 2016). Social perception related to moral cognition is notably understudied in alcohol use disorder (Pabst et al., 2022).

A biological, mechanistic understanding of antisociality and moral processing in SUD/AUD specifically is vital for two reasons. First, such an understanding may help to combat persistent stigma. There is a history of labeling people with mental illness (particularly those with SUDs) and those involved with the justice system as “immoral”- that is, that these individuals have some sort of irredeemable moral failing that explains their behavior (Frank & Nagel, 2017). Stigma associated with mental illness and/or justice system involvement has been associated with the perpetuation of several individual and systemic harms by discouraging treatment seeking, perpetuating health disparities, and marginalizing people of color and certain socioeconomic groups who are incarcerated at disproportionate rates (Corrigan & Watson, 2007; Feingold, 2021; Parcesepe & Cabassa, 2013). By illustrating that an “immoral action” is not reflective of a failing of character but instead, atypicalities in neural processes that are not specific to “morality” (e.g., attention), and for which there are potential mitigating therapies, we can continue to combat these existing stigmas. Second, a mechanistic understanding may lead to the identification of cognitive or neuroscientific intervention targets to reduce rates of antisocial behaviors and associated harms in individuals with AUDs/SUDs. This has potential harm reduction implications, in that individuals who continue to use substances could meanwhile be given tools to make more prosocial decisions, improving quality of social relationships and reducing any negative impact of their use on society (e.g., increasing attention to the potential harm to others of driving after drinking may lead individuals to choose not to drive in those circumstances and reduce DUI-related fatalities).

A potential mechanism that can explain moral processing differences transdiagnostically is attention modulation. Newman & Lorenz define “response modulation” to typically involve switching from a dominant response set to respond to peripheral stimuli, particularly those that are novel or aversive (2003). We use the more intuitive term “attention modulation” going forward, as the switching mechanism involved does not necessarily require any response, particularly in the cognitive domain. Appropriate attention modulation is key to moral cognition; harm caused by moral violations are often peripheral (e.g., the goal of a mugging is gaining money not injuring the victim) and moral behavior may require bystander intervention (e.g., noticing an assault and calling the police.) This has also been referred to as moral recognition (Reynolds & Miller, 2015). In his seminal study, Greene (2003) suggested that the level of salience of the peripheral moral stimuli determined for individuals the moral permissibility of taking action when societal benefit was equal.

Deficits in attention modulation are well-established in psychopathy. Lorenz and Newman examined peripheral emotional processing during lexical decision making. Inmates with and without psychopathy (n = 100) viewed emotional, neutral, or non-words (Lorenz & Newman, 2002). Non-psychopaths identified emotional words most quickly, evidence of emotional cueing facilitation. Psychopaths did not have this emotional facilitation, despite no difficulty identifying the emotional content of words. The authors conclude that in psychopathy, peripheral stimuli are not salient, regardless of emotional content, consistent with an attention modulation deficit. This body of work (Baskin-Sommers & Brazil, 2022; Baskin-Sommers et al., 2011; Larson et al., 2013), which the described study exemplifies, leads researchers to conclude that failure to integrate negative reactions to antisocial behavior is the key mechanism leading psychopaths to behave immorally (Blair, 2007).

We propose that an attention modulation mechanism may also lead to altered moral cognition in individuals with SUDs, including AUD. Aside from the modulation deficit already described, Newman and Lorenz (2003) suggested that hyper arousal towards specific sets can interfere with proper attention modulation, consistent with findings that alcohol cues lead to narrowing attentional scope in heavy drinkers (Hicks et al., 2015). Individuals with AUD (iAUDs) have increased neural reactivity (Zeng et al., 2021) and attention bias towards alcohol cues (Field & Cox, 2008), which interferes in goal related neural processing (Murray et al., 2022). Further, Gilman and Hommer have demonstrated that in iAUDs, the presence of alcohol distractors detracted from emotional processing (2008).

We report findings from an empirical study examining the proposed attention bias interference mechanism in AUD by designing a novel fMRI task to assess alcohol attention bias in moral processing. AUD was chosen as a pilot SUD for this phenomenon; SUDs, regardless of substance, are not expected to differ fundamentally in terms of neurobiological mechanisms (Koob & Volkow, 2016) and further, the legal status of alcohol removes any potential confounds with criminality related to the criminalization of other substances of abuse. We hypothesized that iAUDs (but not healthy control volunteers; HCVs) would have reduced neural engagement in regions previously implicated in moral processing (from Fede & Kiehl, 2019 meta-analysis; OFC, ACC, d/vmPFC, PCC, insula, STS, M/STG, TPJ) when moral stimuli were presented alongside alcohol cues.

MATERIALS AND METHODS

The procedures described are part of the first stage of a two-stage registered trial (NCT03535129), including objectives, study population, design, and outcomes. Only procedures and results relevant to the Stage 1 aim to investigate the attention modulation interference hypothesis are described here; the overall study investigates the effect of fMRI neurofeedback on craving reduction and associated symptomology in AUD. During the same imaging session as the procedure reported below, participants also completed a neurofeedback training optimization procedure, results from which are reported in another publication (Fede et al., 2023). Stage 1 of the trial (n = 60) is completed; Stage 2 of the trial is ongoing. All procedures were conducted as approved by the NIH Institutional Review Board.

Participants.

Sixty adult individuals were recruited from the Washington DC metro area to participate in this study. This was evenly split into groups of individuals with AUD (iAUDs; n = 30) and individuals without (HCVs; n = 30). This recruitment number was chosen after power analysis (effect size d = 1.235, α = .001, power = 0.8) indicated a required n of 25 per group. This effect size was estimated by averaging reported across four studies on the neural correlates of emotion or moral cognition in substance use disorder populations (Caldwell et al., 2015; Fede, Harenski, et al., 2016; Gilman & Hommer, 2008; Padula et al., 2011). Participants met criteria for the iAUD group if they were diagnosed with current moderate to severe AUD based on DSM-5 criteria; participants in the HCV group had no lifetime AUD diagnosis. See Table 1 for demographic information. Groups were recruited to match on age and sex. The majority of the iAUDs (n = 22) were participating in NIAAA’s alcohol 28-day inpatient treatment program at the time of their involvement in this study. This treatment program involved group and individual therapy as well as pharmacological intervention as indicated. These inpatient iAUDs were abstinent and in weeks 2–4 of their inpatient treatment while participating in this study, prior to routine administration of any medication expected to suppress craving (e.g., naltrexone). Therapeutic interventions were not specific to the mechanism of interest (i.e., did not explicitly involve cognitive behavioral therapy, mindfulness, or attention modification.) Other iAUDs were classified as “outpatient” and were not treatment seeking.

Table 1.

Sample Demographic and Self-Report Descriptives

		Overall (N = 60)				iAUD (n = 30)		HCV (n =30)
A) Continuous variables		mean	sd	min	max	mean	sd	mean	sd	t	df	p diff
	Age	39.23	13.1	21	64	40.8	11.38	37.63	14.6	1.05	48.42	0.30
	Years of Education	14.95	2.9	3	22	14.2	2.32	15.73	3.23	−3.01	53.45	0.0040
		mean	sd	min	max	mean	sd	mean	sd	F (1,52)		p ancova
	Socioeconomic Status (SSS)	5.95	1.95	1	10	5.1	2.02	6.8	1.47	15.44		0.003
BRIEF (n = 46)		50.11	11	35	77	56	11.46	45.16	7.88	12.25		0.001
ACQ (n = 58)		2.241	1.26	1	7	2.82	1.45	1.62	0.59	13.55		0.0006
	AUDIT	13.63	12.5	0	38	24.9	7.24	2.33	1.3	16.82		<2.2e-16
	IRI- Perspective Taking	18.5	4.88	4	27	18.3	3.94	18.67	5.73	0.01		0.92
	IRI- Fantasy	14.6	5.46	0	24	14	5.57	15.17	5.39	0.36		0.55
	IRI- Empathetic Concern	20.28	4.41	6	28	20.4	4.07	20.2	4.8	0.10		0.75
	IRI- Personal Distress	9.433	5.09	0	24	10.2	5.25	8.67	4.89	0.63		0.43
	MFQ- Harm	23.18	3.97	15	30	24.2	3.77	22.13	3.95	3.06		0.09
	MFQ- Fairness	22.85	3.86	14	30	22.3	4.32	23.43	3.3	1.78		0.19
	MFQ- Ingroup	16.2	5.54	0	28	17.4	4.79	14.97	6.03	1.41		0.24
	MFQ- Authority	17.77	5.98	5	29	18.8	5.42	16.73	6.43	0.45		0.50
	MFQ- Purity	17.08	6.95	2	30	18.9	5.84	15.23	7.56	1.96		0.17
	SRP	47.62	11.6	29	75	48.7	11.45	46.5	11.87	0.30		0.59
	TAS	41.03	11.5	21	72	45.4	11.93	36.67	9.41	7.49		0.008
		Overall (n = 60)			iAUD (n = 30)			HCV (n =30)
B) Categorical variables		Number	Percentage		Number	Percentage		Number	Percentage		χ2	df	p diff
Sex	Female	25	41.7%		12	40.0%		13	43.3%		0.069	1	0.79
	Male	35	58.3%		18	60.0%		17	56.7%
Patient status	Inpatient	N/A			22	73.3%		N/A			N/A
	Outpatient				8	26.7%
Race	Asian	4	6.7%		0	0.0%		4	13.3%		7.94	4	0.094
	Black/African American	20	33.3%		8	26.7%		12	40.0%
	Multiple Race	2	3.3%		2	6.7%		0	0.0%
	Unknown/Not Reported	3	5.0%		2	6.7%		1	3.3%
	White	31	51.7%		18	60.0%		13	43.3%
Ethnicity	Hispanic Latino	9	15.0%		5	16.7%		4	13.3%		0.094	2	0.95
	Not Hispanic or Latino	49	81.7%		24	80.0%		25	83.3%
	Unknown/Not Reported	2	3.3%		1	3.3%		1	3.3%

Notes: Descriptive statistics for each self-report variables collected, across the sample and broken down by group. A) Reports of means, standard deviations and ranges. Groups compared using a two-tailed Welch two-sample t-test (Age, Years of Education) or ANCOVA covarying Age, Education, and Sex. Unless otherwise noted, these variables refer to the overall score on the measure. Nine individuals in the AUD group and five individuals in the HCV group had invalid responses on the BRIEF and were excluded from statistics (F(df) = 1,38). Two individuals in the HCV group did not complete the ACQ. B) Reports of numbers in each category as well as the proportion of the total that number represents. Groups compared using a χ2 test. No individuals reported sex other than male or female. Patient status was not reported for the HCVs, including as an overall number, since patient status is not applicable. Abbreviations as follows: SSS- McArthur Scale of Subjective Social Status; BRIEF- Behavior Rating Inventory of Executive Function; ACQ- Alcohol Craving Questionnaire; AUDIT- Alcohol Use Disorders Identification Test; IRI- Interpersonal Reactivity Index; MFQ- Moral Foundations Questionnaire; SRP- Self-Report Psychopathy scale; TAS- Toronto Alexithymia Scale; iAUD- individuals with alcohol use disorder group; HCV- healthy control volunteer group; df- degrees of freedom; sd- standard deviation; t/χ2- test statistic; pdiff - p-value for the group differences test; pancova - p-value for the group term in ANOVA.

Exclusion criteria included: significant history of head trauma or cranial surgery, history of neurological disease, MRI contraindications (including pregnancy, claustrophobia, and presence of ferromagnetic objects), and history of non-substance related psychosis. Participants were not scanned if they tested positive for opiates, cocaine, or methamphetamine, if they were currently experiencing withdrawal from alcohol, or if they had a positive breath alcohol measure. Two individuals in the HCV group consented but did not complete MRIs due to disclosing an MRI contraindications (i.e., metal in body; n = 1) and claustrophobia requiring the MRI session to be ended before functional tasks could be completed (n = 1).

Procedures and Measurements.

All participants first completed a separate screening protocol that included appropriate data sharing provisions. Clinical and sociodemographic data collected as part of that study were used for initial eligibility evaluation. This included the Structured Clinical Interview for the DSM-5 (SCID; First et al., 2015), which was used to evaluate AUD diagnosis. After providing written, informed consent for the current study, participants completed self-report measures of executive dysfunction (Behavior Rating Inventory of Executive Function [BRIEF]; Gioia et al., 2000), psychopathy (Self-Report Psychopathy Scale [SRP]; Paulhus et al., 2009), empathy (Interpersonal Reactivity Index [IRI]; Davis, 1983), foundations of moral reasoning (Moral Foundations Questionnaire [MFQ]; Davies et al., 2014), alexithymia (Toronto Alexithymia Scale [TAS]; Bagby et al., 1994) and alcohol craving (Alcohol Craving Questionnaire [ACQ]; Singleton et al., 1994). Group differences on these measures were examined using Welch two-sample t-tests and reported in Table 1A. Group differences in categorical demographic variables (i.e., sex and race) were examined using χ² tests and reported in Table 1B. Finally, they completed behavioral tests of cognitive interference and selective attention (Stroop; Golden & Freshwater, 1978), and sustained attention (continuous performance test [CPT]; adapted from Riccio et al., 2002).

After initial data collection in a neutral environment, and prior to the MRI scan, participants spent approximately 10 minutes in an alcohol cued environment (a room decorated to resemble a bar/pub; i.e. a “bar lab”). Participants then repeated the Stroop assessment in that alcohol cued environment and were taken to the MRI scanner. Participants underwent one scan session on a 3 Tesla Siemens Prisma MRI machine. Two individuals in the HCV group and one individual in the AUD group did not complete their MRI scan session; all analyses of behavioral and imaging results do not include those individuals. Participants were asked to complete the sociomoral alcohol attention task (SAAT), described below and depicted in Figure 1, during an echoplanar-imaging (EPI) pulse sequence (TR: 2000 msec, flip angle: 90 degrees, FOV: 24x24 cm, 38 mm slice thickness, 36 slices, multislice mode: interleaved.) A MPRAGE structural scan was also acquired for use in co-registration.

Figure 1:

Graphical depiction of sociomoral alcohol attention task (SAAT).

In each SAAT trial, two picture stimuli were presented to the subjects: a distractor and a cue. Distractors consisted of either alcohol or neutral objects; cues consisted of negative moral (e.g., a man pointing a gun at a woman, a group of kids bullying another kid), negative emotional (e.g., a crying baby, a person with a nail piercing their foot), or neutral (e.g., a person looking out a window, a person watching tv) pictures. Picture cue stimuli were initially selected from established stimuli sets (OASIS, IAPS), supplemented with related images found via internet image search (Kurdi et al., 2017; Lang et al., 1997), and refined using a pilot study (described in Supplemental Materials 1 and Supplemental Table 1) . The final set of cues was selected to include 40 cues belonging to each of the three categories, for a total of 120 trials. Each trial lasted 3 seconds, with an interstimulus interval of 1–3 seconds. This was an implicit moral processing task; participants were asked to indicate if the pictures depicted an indoor or an outdoor location and were not told that any content would be morally relevant. This behavioral response was meant to maintain attention to the task and was not of interest as an outcome measure. Code for this task is available at https://osf.io/zhuvj/.

Following their scan, the participants in the current study were asked to rate a random subset of the cues (m = 50) on the same scales (outside the scanner). Participants were also asked to rate a random subset of the distractors (m = 50) on amount of craving evoked. Post-scan ratings for each cue type, broken down by group, are presented in Supplemental Table 2. In post-scan ratings, moral and neutral cues differed significantly on valence, moral content, and social acceptability, but not arousal or social content (i.e., rate of images containing people, rate of clear faces visible; see Supplemental Table 1B and 2).

Behavioral Analysis.

The effect of group on each self-report and behavioral variable was analyzed in separate one-way ANCOVA models, covarying age, sex, and years of education, using the aov function in R. Group differences in age and years of education were compared using Welch two-sample t-tests. Behavior was quantified as percentage of trials with “correct” responses (i.e., indoor or outdoor); descriptives are reported in Supplemental Table 3. Individuals with poor task performance (<80% accuracy; n = 3) were excluded from imaging analyses

Image Processing and Analysis.

fMRI data collected for each participant was processed separately using AFNI (v21.0.20) using the following pipeline: first, we shifted slice time courses so that each voxel data was aligned to the same temporal origin while identifying outliers in the time series. Then volumes across the time series were aligned to the first EPI volume and to the skull-stripped structural image of the subject. Transformation to standard MNI space was conducted using a non-linear warping procedure. Data was then smoothed using a 4 mm full-width at half-maximum Gaussian kernel and scaled to a mean of 100 (min 0, max 200). Outlying TRs and those with motion derivatives of 0.3 mm or greater were censored from all analyses that followed. Individuals who had 30% of trials or greater censored (n =2) were excluded from the group analyses (final N = 52; n_HCV = 25, n_AUD = 27).

Conditions of interest were modeled time locked to stimuli presentation. Specifically, six conditions of interest were modeled (listed in cue-distractor format): moral-alcohol; moral-neutral; emotional-alcohol; emotional-neutral; neutral-alcohol; and neutral-neutral. We used the 3dmaskave (through AFNI) to extract mean betas associated with each condition in separate region of interest (ROI) masks. These are spherical masks with 5mm radii centered on peak coordinates reported in a meta-analysis of moral cognition (Fede & Kiehl, 2019). This meta-analysis reported 20 peak coordinates across several frontal, temporal, and limbic anatomical regions (ACC, OFC, dmPFC, vmPFC, amygdala, PCC, insula, STS, STG, MTG, and TPJ); given that two PCC peaks were within 5mm of one another, we excluded the second from our analysis such that the final ROI number was 19.

To test our hypothesis, we conducted a 2[Cue Type: Moral, Neutral] x 2[Distractor Type: Alcohol, Neutral] x 2[Group: HCV, iAUD] mixed ANCOVAs for each of these ROIs using the aov function in R. The interaction between Cue and Distractor was modeled as nested within subject. Age, sex, and years of education were modeled as covariates. The cue*distractor interaction term was considered the main effect of task; the group*cue*distractor interaction term was considered the group effect on task and was our primary effect of interest. Given that emotional stimuli without moral content were not the primary focus of this study, and low power of this small, initial study, we did not include them in the ANCOVAs. ROIs with significant model effects were explored post-hoc using plots and descriptive statistics. All plots were created using ggplot2 in R (Wickham, 2009).

Importantly, although we planned to include the demographic covariates in our model a priori, their interaction effects were not an original focus of the analysis. These terms were included according to our standard practices, consistent with general neuroimaging recommendations to include relevant covariates in fMRI statistical models (Nichols et al., 2017). These are relevant due to evidence of associations with brain structure (reviewed in Hyatt et al., 2020) and function (e.g., Filippi et al., 2013; Arenaza-Urquijo et al., 2021; Fede et al., 2019); thus, demographic features can potentially mask or drive results in conditions of interest. Including covariates that are expected to influence outcome variables has generally been shown to increase power (Kahan et al., 2014). We report the interaction of these covariates with our terms of interest for the sake of completion.

RESULTS

Behavioral Results

Demographics:

There were significant group differences in years of education (p_diff = 0.004) and socioeconomic status (p_aov = 0.003), but no significant differences in age, sex, race, or ethnicity. Specifically, iAUDs reported less education and lower social standing compared to HCVs. See Table 1 for descriptives, overall and broken down by group, and test statistics.

Self-Report Measures:

There were significant group differences in executive function, alcohol craving, and alexithymia. Specifically, iAUDs had more indication of executive dysfunction (measured by the BRIEF; p_aov = 0.001) and alexithymia (measured by the TAS; p_aov = 0.008) compared to HCVs. iAUDs also self-reported more alcohol craving (measured by the ACQ; p_aov = 0.0006) compared to HCVs. There were no significant group differences in interpersonal reactivity (perspective taking, fantasy, empathy, or personal distress; measured by the IRI), moral foundations (measured by the MFQ), or psychopathy (measured by the SRP). See Table 1A for detailed descriptives, overall and broken down by group, and test statistics for all measures.

Behavioral Tasks:

There were no significant group differences in performance on the CPT or Stroop. See Table 2 for detailed descriptives, overall and broken down by group, and test statistics for all behavioral performance metrics.

Table 2.

Behavioral Task Results

	Overall (N = 55)*				iAUD (n = 26)		HCV (n =29)
CPT	mean	sd	min	max	mean	sd	mean	sd	F (1,47)	pancova
RT, Overall	0.42	0.06	0.31	0.57	0.42	0.07	0.42	0.06	0.36	0.55
RT, Commision Errors	0.35	0.08	0.14	0.58	0.35	0.08	0.36	0.07	2.48	0.12
RT, Correct	0.41	0.06	0.31	0.57	0.42	0.07	0.41	0.06	0.36	0.55
Accuracy	87.97%	8.04%	68.33%	100.00%	88.81%	7.61%	87.21%	8.47%	1.49	0.23
Commision Error Rate	9.42%	5.63%	0.00%	20.30%	8.75%	5.59%	10.03%	5.69%	0.69	0.41
Ommision Error Rate	2.61%	4.46%	0.00%	17.50%	2.44%	4.23%	2.76%	4.72%	1.35	0.25
Stroop	Overall (N = 60)				iAUD (n = 30)		HCV (n = 30)		F (1,51)	pancova
Interference Score, Neutral Environment	52.1	6.69	38	71	51.97	7.53	52.23	7.37	0.50	0.48
Interference Score, Alcohol Cued Environment	55.22	7.39	40	77	55.73	6.99	54.69	6.45	1.55	0.22
Interference Score, Alcohol - Neutral Enviornment	2.2	8.92	−46	19	3.77	6.93	0.63	10.44	0.64	0.43
Number of Errors, Alcohol - Neutral Environment	0.22	0.92	−2	3	0.2	0.96	0.23	0.9	0.21	0.65

Notes: Descriptive statistics (means, variances, and ranges) for behavioral tasks of sustained attention (CPT) and cognitive inference (Stroop), reported across the sample and broken down by group. Groups compared using a two-tailed Welch two-sample t-test. One HCV subject did not complete the Stroop within the alcohol cued environment. *statistics, descriptives, and number of subjects included reflect the removal of 5 outliers who had <60% accuracy on the CPT or average response time >700 ms. Abbreviations as follows: RT- response time; CPT- continuous performance task; AUD- individuals with alcohol use disorder group; HCV- healthy control volunteer group; sd- standard deviation; F()- test statistic (degrees of freedom); pancova - p-value for the group differences test

Imaging Results

There was a significant three-way group*cue*distractor interaction in the mPFC (BA9) ROI (p = 0.0068); iAUDs had reduced mPFC activity during moral-alcohol compared to moral-neutral cue-distractor pairs; this pattern was not present for neutral cues or in the HCV group. (See Figure 2 for plots of betas in the mPFC ROI by group and condition and Table 3A for group effects on task test statistics for all ROIs). In ROIs without group effects, there were significant cue*distractor interactions in ROIs located in the left amygdala (p = 0.017), left STS (left: p = 0.0221), STG (p = 0.0321), MTG (p = 0.0145), and PCC (dorsal portion of BA 23 subdivision; p = 0.0393). Specifically, in the STS and PCC ROIs, there was increased BOLD response in neutral-alcohol cue pairs (compared to neutral-neutral) but similar response during moral cues regardless of distractor types. In the amygdala, MTG and STG ROIs, there was an attenuated response to moral cues when paired with alcohol distractors compared to pairings with neutral distractors; the opposite pattern was observed for neutral cues. (See Figure 3 for plots of betas in the ROIs by condition and Table 4 for task main effect test statistics for all ROIs.)

Figure 2:

Brain activity (betas) plotted by cue type, distractor type, and group for the medial prefrontal cortex (mPFC) region with a significant group*cue*distractor type interaction. (Left) Red circle plotted on MNI_152_T1_2009 anatomical image represents ROI mask (5mm radii sphere around peak coordinates listed in Table 3) from which average beta-values were extracted. (Bar plots) Colored bars represent means; error bars show standard error. Overall plots correspond to primary analysis; split plots to significant covariate*group*cue*distractor interactions. (Scatter/line plots) Dots represent individual values. Betas (Difference) corresponds to the value of the alcohol distractor condition minus the neural distractor condition for each cue type. Lines are linear best fit lines for the association between beta difference and age within each cue type. Abbreviations: AUD- alcohol use disorder; HCV- healthy control volunteer; ROI- region of interest.

Table 3.

Mixed ANCOVA statistics for Group*Cue*Distractor model terms (Group Effects on Task)

			A) Group*Cue*Distractor		B) Age* Group*Cue*Distractor		C) Sex* Group*Cue*Distractor
ROI	Location Notes	Peak Coordinate	F	p	F	p	F	p
OFC		−38,26,−14	0.417	0.5215	0.038	0.8472	0.067	0.7966
ACC		0,50,8	2.165	0.1480	0.172	0.6810	2.268	0.1390
dmPFC		0,52,28	2.135	0.1509	1.522	0.2237	6.208	0.0165
mPFC	more dorsal, BA 9	0,52,22	8.050	0.0068	4.102	0.0488	5.008	0.0302
	more ventral, BA 10	−2,54,14	3.212	0.0798	0.001	0.9819	0.001	0.9720
vmPFC		−2,48,−14	0.138	0.7125	0.260	0.6129	3.077	0.0862
Amygdala	Left	−22,−4,−18	0.010	0.9189	4.974	0.0308	1.262	0.2672
	Right	22,−2,−14	0.745	0.3930	0.085	0.7720	0.259	0.6130
PCC	dorsal, BA 23	−4,−48,30	3.023	0.0889	0.251	0.6192	0.574	0.4525
	ventral, BA 31	−2,−56,24	0.609	0.4394	0.402	0.5293	0.577	0.4515
	dorsal, BA 31	−4,−22,40	3.773	0.0584	0.088	0.7677	1.110	0.2977
Insula	Left	−34,14,−10	0.112	0.7399	0.008	0.9313	0.055	0.8151
	Right	40,2,−2	0.067	0.7973	0.286	0.5952	1.140	0.2912
STS	Left, towards MTG	−58,−14,−12	2.924	0.0941	0.636	0.4293	1.011	0.3201
	Right, towards STG	54,−12,−12	0.346	0.5591	1.549	0.2197	0.001	0.9723
STG		46,4,−20	0.061	0.8066	0.068	0.7948	0.414	0.5234
MTG		−60,−24,−10	0.794	0.3776	0.466	0.4985	1.709	0.1977
TPJ	Left	−48,−60,24	0.621	0.4348	0.132	0.7182	3.302	0.0759
	Right	52,−56,22	1.444	0.2360	0.216	0.6450	0.427	0.5170

Notes: Statistics from mixed ANCOVA models run for each region of interest (ROI) for model terms containing the full Group*Cue*Distractor interaction reported here. Within each ROI, reported statistcs are for terms within the same model and were not run seperately. For all term F tests, degrees of freedom were (1,45). ROIs with significant terms bolded, trend level results italicized. Location notes offered for clarity when ROIs have similar labels. Peak Coordinates are reported in MNI space as (x,y,z). Although Education*Group*Cue*Distractor effects were modeled, there were no significant effects of that term. Abbreviations - ROI: region of interest; OFC: orbitofrontal cortex; (A/P)CC: (anterior/posterior) cingulate cortex; (d/v/m)PFC: (dorso/ventro/medial) prefrontal cortex; STS: superior temporal sulcus; (S/M)TG: (superior/medial) temporal gyrus; TPJ: temporoparietal junction; BA: Brodmann Area; F: ANCOVA test statistic corresponding to model term; p: significance level corresponding to reported test statistic.

Figure 3:

Brain activity (betas) plotted by cue type and distractor type for each of the regions with significant cue*distractor type interactions. (Left) Red circle plotted on MNI_152_T1_2009 anatomical image represents ROI mask (5mm radii sphere around peak coordinates listed in Table 4) from which average beta-values were extracted. Colored bars represent means; error bars show standard error. Abbreviations: ROI - region of interest; STS- Superior Temporal Sulcus; PCC- Posterior Cingulate; MTG- Middle Temporal Gyrus; STG- Superior Temporal Gyrus.

Table 4.

Mixed ANOVA statistics for Cue*Distractor model terms (Task Main Effects)

			Significant Higher Order Term?	Cue*Distractor
ROI	Location Notes	Peak Coordinate		F	p
OFC		−38,26,−14		0.368	0.5470
ACC		0,50,8		2.495	0.1210
dmPFC		0,52,28	Sex*Group*Cue*Distractor	0.696	0.4085
mPFC	more dorsal, BA 9	0,52,22	Age/Sex*Group*Cue*Distractor	0.385	0.5379
	more ventral, BA 10	−2,54,14		2.383	0.1297
vmPFC		−2,48,−14		0.443	0.5092
Amygdala	Left	−22,−4,−18	Age*Group*Cue*Distractor	6.150	0.0170
	Right	22,−2,−14		2.231	0.1420
PCC	dorsal, BA 23	−4,−48,30		4.507	0.0393
	ventral, BA 31	−2,−56,24		3.705	0.0606
	dorsal, BA 31	−4,−22,40		2.246	0.1400
Insula	Left	−34,14,−10		0.513	0.4775
	Right	40,2,−2		3.193	0.0807
STS	Left, towards MTG	−58,−14,−12		5.623	0.0221
	Right, towards STG	54,−12,−12		3.528	0.0668
STG		46,4,−20		4.893	0.0321
MTG		−60,−24,−10		6.468	0.0145
TPJ	Left	−48,−60,24		3.881	0.0550
	Right	52,−56,22		2.717	0.1060

Notes: Statistics from mixed ANCOVA models run for each region of interest (ROI) for model terms containing the main task effect Cue*Distractor interaction. "Significant Higher Order Term" refers to the model effects reported in Table 3. Within each ROI, reported statistcs are for terms within the same model and were not run seperately. For all term F tests, degrees of freedom were (1,45). ROIs with significant terms bolded, trend level results italicized, if not superceeded by a higher order term. Location notes offered for clarity when ROIs have similar labels. Peak Coordinates are reported in MNI space as (x,y,z). Although covariate*Cue*Distractor effects were modeled, there were no significant effects of those terms. Abbreviations - ROI: region of interest; OFC: orbitofrontal cortex; (A/P)CC: (anterior/posterior) cingulate cortex; (d/v/m)PFC: (dorso/ventro/medial) prefrontal cortex; STS: superior temporal sulcus; (S/M)TG: (superior/medial) temporal gyrus; TPJ: temporoparietal junction; BA: Brodmann Area; F: ANCOVA test statistic corresponding to model term; p: significance level corresponding to reported test statistic.

In addition to the primary three-way interaction of group*cue*distractor, there were several ROIs with significant higher-order interactions with age and sex. These ROIs were the mPFC (BA9; age*group*cue*distractor p = 0.0488; sex*group*cue*distractor p = 0.0302), the left amygdala (age*group*cue*distractor p = 0.0308), and the dmPFC ROI (sex*group*cue*distractor p = 0.0165). The hypothesized group*cue*distractor effect (i.e., reduced activity in moral-alcohol compared to moral-neutral cue-distractor pairs but not present for neutral cues) was present in younger (mPFC, left amygdala) and female (mPFC, dmPFC) iAUDs, but was attenuated in older/male iAUDs, respectively. See Table 3 for statistics for each ROI for the age* (B) and sex* (C) group*cue*distractor model terms. There were no significant education*group*cue*distractor*education interaction effects. Demographic interaction effects are plotted in Figures 2 & 4.

Figure 4:

Brain activity (betas) plotted by cue type, distractor type, and group for additional regions with significant covariate*group*cue*distractor type interaction. (Left) Red circle plotted on MNI_152_T1_2009 anatomical image represents ROI mask (5mm radii sphere around peak coordinates listed in Table 3) from which average beta-values were extracted. (Bar plots) Colored bars represent means; error bars show standard error. Overall plots correspond to primary analysis; split plots to significant covariate*group*cue*distractor interactions. (Scatter/line plots) Dots represent individual values. Betas (Difference) corresponds to the value of the alcohol distractor condition minus the neural distractor condition for each cue type. Lines are linear best fit lines for the association between beta difference and age within each cue type. Abbreviations: AUD- alcohol use disorder; HCV- healthy control volunteer; ROI- region of interest

In many ROIs without these higher order interactions, lower order model terms were significant; these were not effects of interest and are reported (for all ROIs and remaining model terms) in Supplemental Tables 4 and 5.

Additional Analyses

In order to assist with interpretation of unexpected individual difference results, we conducted supplemental analyses examining group differences in demographics and AUD severity measures based on treatment status within the iAUD group. Sub-groups differed qualitatively but non-significantly on age (mean_out = 35.14; mean_in = 43.36; p_diff = 0.175) and sex (female_out:57%; female_in: 32%; p_chi-sq = 0.229). Outpatients had less severe AUD as measured by the AUDIT (mean_out = 19.13; mean_in = 27.05; p_diff = 0.0024), but qualitatively higher alcohol craving (measured by the ACQ; mean_out = 3.83; mean_in = 2.45; p_diff = 0.11). Finally, outpatients reported higher acute craving in response to alcohol pictures compared to neutral pictures (mean_alc = 4.55; mean_neut = 3.43; p = 0.02) while inpatients did not (mean_alc = 2.44; mean_neut = 3.25; see Supplemental Table 2B).

To rule out the possibility that socioeconomic status (SES), which differed significantly between the iAUD group and the HCV group, drove our group effects, we ran a post-hoc supplementary model for each ROI with a SES*group*cue*distractor term. Results were substantively the same (see Supplemental Table 6A) except for significant SES*group*cue*distractor interactions in the bilateral TPJ ROIs. Consistent with other ROIs, engagement was lower during moral-alcohol compared to moral-neutral cue-distractor pairs (see Supplemental Figure 2). SES was lower in the outpatient compared to inpatient group.

DISCUSSION

The reported study in part supports the hypothesis that attention bias towards alcohol cues would interfere with moral processing, as measured by brain activity in regions previously identified as engaged during sociomoral cognition. When viewing image stimuli with moral content in the presence of alcohol distractors (but not neutral distractors), iAUDs had reduced neural engagement in the mPFC ROI; this pattern was not observed for neutral cues or in the HCV group. This pattern was also seen for the left amygdala and dmPFC ROIs in sub-groups of iAUDs (younger and female iAUDs, respectively). Across groups, there was a main effect of task in that individuals had reduced left amygdala, MTG and STG engagement when moral stimuli were paired with alcohol distractors.

In addition to being implicated through meta-analyses in moral cognition (Bzdok et al., 2012; Fede & Kiehl, 2019), the mPFC is a well-established major hub of the default mode network (Fox et al., 2005), which is typically characterized as reflective of internally focused tasks or envisioning the internal states of others (Buckner et al., 2008). The meta-analytical association test map generated by Neurosynth (on June 17^th, 2024; Yarkoni et al., 2011) for the central coordinates of our mPFC cluster support this characterization (top association terms include: default mode, beliefs, theory of mind, and mental states). The mPFC, particularly extending in the dorsal aspect, may be important in moral cognition through its connectivity with other default mode regions relevant to theory of mind (including the PCC/precuneus and TPJ) and empathy (amygdala); as a hub, it is thought to be vital in integrating information from these social cognitive processes into coherent judgements (Bzdok et al., 2015). Notably, the integrative role of the mPFC is not specific to moral cognition, but is considered to be somewhat specialized to affective/motive and reflexive dimensions of processing (Miller & Cohen, 2001). Lesions to this region produce alterations in a variety of social cognitive, self-processing, and emotion processes (see Liebermann et al., 2019 for a thorough review of the evidence for those processes). Individuals high in psychopathy have reduced gray matter volume and blunted activity during social cognition tasks in this extended mPFC region (Anderson & Kiehl, 2012).

There were no group differences in self-report measures of empathy, moral foundations, or psychopathy. The lack of group differences in these trait-level characteristics suggests individual differences in morality are not driving the group differences in effects reported here, consistent with our hypothesis that attentional processes, rather than emotional or moral traits, drive differences in brain activity corresponding to moral processing. It is also consistent with previous work finding affective biases in criminal offenders were not predicted by trait empathy or moral reasoning (Sadek et al., 2021), though this same study found IQ predicted this affective bias. This latter reported effect may correspond to our finding that iAUDs had higher general executive dysfunction, though importantly not specific deficits in sustained attention or inhibition of cognitive interference. Alexithymia, which was elevated in our iAUD group compared to controls, has previously been shown to impair moral reasoning (Patil & Silani, 2014a), though the proposed mechanism for this impairment is via reduced empathy (Patil & Silani, 2014b), which we did not observe. An alternative mechanism, more consistent with our observed trait assessments, is that alexithymia influences moral reasoning through a cognitive anchoring bias (Li et al., 2022). There is a high prevalence rate of alexithymia in iAUDs (45–67%; Thorberg et al., 2009). Therefore, even if group differences in alexithymia drive our main group results, the findings reported here still have important implications for the role of cognitive/attentional biases in sociomoral processing within iAUDs.

Unexpectedly, demographic variables (age and sex) interacted importantly with AUD status in the influence of alcohol distractors on neural engagement while viewing moral cues; as age increased, and in males, the hypothesized interference effect was attenuated. Although we included age, sex, and education in the model as nuisance variables, we did not expect them to interact in a significant way a priori. The literature suggests the potential for age to impact attention processes, as aging is associated with reduced visual attention capacity (Kunstler et al., 2018) and reduction in engagement of visual-attention brain networks (Roski et al., 2013). Males also have been shown to have slower performance and more extensive engagement during visual attention and switching tasks (Kuptsova et al., 2015). Taking this body of work in total, we would expect older males to have the largest neural, alcohol-interference effect, contradicting our observed effects.

Also unexpectedly, the observed pattern was not specific to the more complex model terms; we also saw interactions between alcohol cue reactivity and demographic variables, where younger and female (but not older male) iAUDs had elevated neural response to alcohol cues (see Supplemental Table 4B). Although alcohol cue reactivity in iAUDs is well established in the literature (Schacht et al., 2013; Zeng et al., 2021), previous work has suggested a potential impact of age and sex. Alcohol craving seems to decrease with age (Hintzen et al., 2011), while males have higher neural activation in response to alcohol pictures (Kaag et al., 2019; Petit et al., 2013). This is only partially consistent with our findings, as we saw neural patterns suggesting elevated alcohol cue reactivity in younger or female iAUDs.

An alternative, and in our opinion, more likely explanation is that demographic effects were incidental to substantive patient status effects within our iAUD sample. 73.33% of our iAUDs were in an inpatient treatment program for AUD at the time of their participation, while the others were non-treatment seeking AUD “outpatients”. The outpatients were qualitatively younger (mean age = 35.14) and more female than our inpatients, although these differences did not reach statistical significance. The inpatients reported more severe AUD but despite this elevated severity, inpatient iAUDs in our sample had qualitatively lower alcohol craving than outpatient iAUDS, though alcohol craving in both iAUD groups was still elevated compared to HCVs. This pattern was also reflected in self-reported cue induced craving on the post-scan questionnaireswhere outpatients reported higher craving in response to alcohol pictures compared to neutral pictures. Meta-analysis has previously shown that alcohol treatment attenuates alcohol-cue induced neural activity (Zeng et al., 2021). If this interpretation is true, it is likely that the pattern observed in our younger, female iAUDs, consistent with our hypothesis, is more generalizable to typical non-treatment seeking iAUDs in the community. Another way of interpreting this lack of alcohol cue-reactivity effect in our inpatient iAUDs is the alcohol distractors we used did not adequately or consistently induce craving. Given the observation of the expected effect in the non-treatment seeking iAUDs and the sourcing of these images primarily from previously published cue-reactivity studies (e.g., Naqvi et al., 2015), we think this is less likely than alcohol treatment attenuating alcohol craving.

Our results indicate that our task successfully elicited brain activity previously associated with moral processing. There was an effect of cue type (or higher order interaction; see Supplemental Table 5A) in the majority of the hypothesized ROIs, which were derived from published meta-analysis of moral cognition (Fede & Kiehl, 2019). However, there was not a significant moral versus neutral cue contrast in several notable regions (right amygdala, vmPFC, dorsal PCC, bilateral insula, right STS and right TPJ). A possible explanation for this incomplete replication is that this study used an implicit task of moral cognition, meaning that participants were not explicitly told to consider the moral content of the material or to make a moral judgement. Although we are confident the images had moral content the participants could identify (the images were rated as depicting “morally wrong” scenes by both a pilot sample and our current sample after the scan; see Supplemental Tables 1 & 2), meta-analysis indicates implicit compared to explicit moral tasks elicit less brain activity in the TPJ, PCC, and mPFC (Fede & Kiehl, 2019). The implicit nature of the task was by design and necessary to answer the research question, as a prompt to rate moral content would have masked any potential attention bias effects. For example, we did not observe group differences in Stroop performance in the alcohol-cued environment, possibly due to the explicit nature of the task demands.

This smaller effect of implicit moral stimuli on eliciting corresponding neural response may have generally limited the study power. For example, in several ROIs, model terms containing the three-way interaction of interest reached trend level (vmPFC, PCC, left STS and TPJ). Limited power might also explain the significant alcohol bias interference effects (i.e., cue*distractor interactions) that unexpectedly do not differ between iAUDs and HCVs. In examining these results qualitatively (Supplemental Figure 1) and considering trend level results, the hypothesized effect does appear qualitatively present in the vmPFC, PCC, MTG, and left STS, although not the STG or TPJ. An alternative explanation for this regional lack of group effect may also relate to differences in functional roles of temporal gyrus regions in moral processing. According to Moll’s theory (2005), temporal regions contribute to the recall of social rules and concepts; moreover, the STG has been specifically implicated in incongruence between simultaneous, emotional information (Watson et al., 2013). These processes may not be impacted by alcohol bias in the same way, instead responding to social contextual detail congruence that is not unique to iAUDs.

It is unclear if the interference effect illustrated by the observed cue*distractor interactions are specific to moral cues. There is reason to suspect it is not; Gilman & Hommer (2008) previously found a similar pattern in a study examining negative emotionality. This is not necessary for an attention interference mechanism to have utility in understanding atypical moral processing. Emotion is an important process within moral cognition (Greene et al., 2004; Moll et al., 2005), which relies on a balance of emotion, social, and fronto-cognitive processes. In fact, the previously reviewed attention modulation deficit of psychopathy explicitly applies not only to moral cognition but to reinforcement learning (Patterson & Newman, 1993), lexical decision making (Lorenz & Newman, 2002), and facial emotion processing (Shane & Groat, 2018), among others. Though it was not our primary focus, for the reasons above, we did collect neural response data during presentation of negative emotional, non-moral cues for the sake of completeness. Given the limited power available for additional analyses, we did not systematically analyze this condition. Qualitative examination of plots suggests ROI differences in how similar BOLD signal was during emotional and moral cues. We present these plots as Supplemental Figure 1 for the reader’s interpretation, but do not place any particular emphasis on them, as they do not represent statistical results. Our ROI approach also cannot rule out involvement of any other areas of the brain not specifically examined here.

In addition to power limitations, our sample is limited in representation, which may reduce the generalizability of the findings. Based on self-report, the sample was 50% White, 30% Black/African American, and predominantly not Hispanic or Latino (80%). This distribution is consistent with the diversity of the metropolitan area in which the study was conducted; however, the results should be replicated in a sample that includes more individuals coming from additional racial and ethnic backgrounds.

Finally, higher order interactions with demographic variables should be interpreted with caution, as this study is underpowered to tease apart the unique effects of these demographic features, and demographic differences may be incidental to other more meaningful AUD characteristics (e.g., patient status). To rule out the possibility that underpowered demographic terms were driving or masking effects of interest, we did run models post hoc for each of the ROIs without any covariates; results were substantively the same (see Supplemental Table 6B).

Conclusions and Future Directions

Our study provides initial evidence for an attention bias interference mechanism during moral processing in AUD. However, more work is needed to understand whether this generalizes to attention mechanisms within moral processing transdiagnostically. Further, in our unexpected findings, we have identified several research questions around individual differences that may impact the attention bias interference mechanism. This can be addressed by replicating these findings in samples powered a priori to address sex and age differences, as well as extension to other psychopathology with attention dysfunction. Moreover, the predictive utility and causal implications of these findings need to interrogated in future work. Building on this finding with longitudinal designs and experimental methods that allow researchers to manipulate attention is a necessary next step. We are currently conducting two such follow-up studies. The first examines the predictive utility of neural engagement during the task described here for post-treatment outcomes (including antisocial behavior and interpersonal aggression) in a new inpatient AUD sample (NCT03535129). The second examines causality by using TMS to excite and inhibit attention switching during peripheral moral processing in a new HCV and a new community AUD sample (NCT05611502).

Until the results of such follow-up studies are established, we cannot speak to the promise of this mechanism to inform application of interventions, such as attentional bias modification (Heitmann et al., 2018). This study is only the first step in realizing such potential promise. However, we believe continued exploration of potentially leverageable cognitive mechanisms in sociomoral contexts, such as the attention bias interference mechanism discussed here, is vital as it can serve to combat persistent stigma about AUD, particularly the incredibly harmful myth that substance use is an intractable moral failing.

DATA AVAILABILITY STATEMENT

The data that supports the findings of this study are available from the senior author (R.M.) upon reasonable request. Data will also be shared in the NIAAA Data Archive in compliance with ethics approvals and with NIH’s Data Management and Sharing Policy following completion of NCT03535129.