Reduced brain network segregation in alcohol use disorder: Associations with neurocognition

Abstract

The human brain consists of functionally segregated networks, characterized by strong connections among regions belonging to the same network and weak connections between those of different networks. Alcohol use disorder (AUD) is associated with premature brain aging and neurocognitive impairments. Given the link between decreased brain network segregation and age‐related cognitive decline, we hypothesized lower brain segregation in patients with AUD than healthy controls (HCs). Thirty AUD patients (9 females, 21 males) and 61 HCs (35 females, 26 males) underwent resting‐state functional MRI (rs‐fMRI), whose data were processed to assess segregation within the brain sensorimotor and association networks. We found that, compared to HCs, AUD patients had significantly lower segregation in both brain networks as well as poorer performance on a spatial working memory task. In the HC group, brain network segregation correlated negatively with age and positively with spatial working memory. Our findings suggest reduced brain network segregation in individuals with AUD that may contribute to cognitive impairment and is consistent with premature brain aging in this population.

• We compared functional brain network segregation between patients with alcohol use disorder (AUD) and health controls (HCs) using resting‐state functional MRI.

• Compared to HCs, AUD patients had poorer spatial working memory performance and lower segregation in the association and sensorimotor brain networks.

• Brain network segregation correlated positively with spatial working memory in HCs but not AUD patients, suggesting that low network segregation may contribute to AUD‐related cognitive impairment.

1. INTRODUCTION

Harmful use of alcohol has led to 3 million deaths globally in a single year and poses significant economic and social risks to both families and society. ¹ Alcohol use disorder (AUD) leads to impairments in multiple cognitive domains, including working memory (WM) and executive functioning, ² and research suggests that this might be a product of changes in brain structure and function. ³ , ⁴

Cognitive impairments endure in abstinent AUD individuals, as shown in a meta‐analysis of 62 studies examining the relationship between AUD and cognitive performance across 12 domains. ⁵ These cognitive impairments persist in AUD even after months or years of abstinence. ⁶ , ⁷ , ⁸ For example, AUD patients who have been abstinent for both 3 months and 3 years performed only at chance level in the Iowa Gambling Task, which measures decision‐making ability in a context mimicking real‐life reward and punishment contingencies. ⁹ Moreover, AUD can double the risk of verbal memory impairments in middle‐aged adults later in life. ¹⁰ This highlights the importance of characterizing the mechanisms underlying cognitive impairment in AUD.

Researchers have employed various brain imaging techniques to assess AUD's effects on the brain. ¹¹ Resting‐state functional MRI (rs‐fMRI) measures the spontaneous brain activity and connectivity when a participant is not performing explicit tasks and has provided valuable insight regarding the functional organization of the brain. Studies show that AUD is linked to reduced connectivity within the default mode network, stronger connectivity between reward and somatosensory regions, and stronger connectivity between somatosensory and frontal cortical regions. ¹² , ¹³ These findings may underpin the observed abnormalities in reward processing and sensory‐motor integration that persist among people with a history of heavy drinking. ¹⁴ , ¹⁵ , ¹⁶ Notably, a decrease in functional connectivity between the cingulum gyrus and the fusiform gyrus in the AUD group indicates potential deficits in visual attention, cognition, emotional control and reward systems. ¹³ Additionally, a reduction in functional connectivity between the left precentral region and the left cerebellum has been associated with relapse in individuals with AUD. ¹³ This suggests that reduced connectivity between the frontal lobe and the cerebellum may play a role in AUD relapse. These findings provide insight into functional distinctions in brain network configuration between individuals with AUD and control populations. However, it is noteworthy that no studies, to our knowledge, have investigated how network‐level connectivity changes might mediate cognitive impairments in AUD.

One strong candidate for functional abnormalities that may relate to cognitive deficits in AUD is the segregation of brain networks. Segregation is a fundamental property of brain organization, supporting specialized information processing within functionally related brain regions (e.g., the medial prefrontal and posterior cingulate cortex related to internal mental states). Highly connected regions are often grouped together as ‘communities’, such as the default mode network. ¹⁷ , ¹⁸ Segregated networks have dense connections within the same communities and sparse connections between different communities and can be quantified using modularity, clustering coefficient, local efficiency and other metrics originating from graph theory. ¹⁸ Stronger segregation plays a crucial role in brain network function ¹⁹ and is associated with better cognitive performance in crystallized intelligence, processing speed ²⁰ and long‐term episodic memory. ²¹ It has also been linked to improvements in cognition following cognitive training. ²² In the normal aging process, the brain network segregation of both the sensorimotor and association networks decreases with age and mediates the aging‐associated cognitive decline. ²³ Sensorimotor networks process basic perceptual inputs, while association networks integrate input from various sources and senses to support higher level functions. ²¹ The sensorimotor‐to‐association axis represents a primary ‘gradient’ that is a core organizing feature associated with healthy brain function. ²⁴ Disruptions in the segregation of sensorimotor and association cortices may therefore have profound consequences for the behaviour. ²⁵

Previous studies have shown that AUD is associated with premature cognitive and brain aging. ²⁶ Chronic alcohol use has been found to result in cognitive performance comparable to healthy individuals who are 10 years older. ²⁷ Additionally, it has been observed to accelerate the onset of brain volume decline by 5 years. ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² Notably, when compared with healthy controls (HCs), AUD patients experience a faster decline in grey matter volume in regions such as the precentral and superior frontal cortices with age. ³³ Since AUD exaggerates the effects of chronological aging, it is possible that they also share similar neural substrates for cognitive impairments. Therefore, we hypothesized that individuals with AUD will have lower brain segregation than HC individuals of a similar age.

In the present study, we characterized brain network segregation using rs‐fMRI and assessed participants' cognitive performance with the Cambridge Neuropsychological Test Automated Battery (CANTAB). We hypothesized that AUD patients would demonstrate lower brain network segregation in both the sensorimotor and association networks, which would correlate with age and impaired cognitive performance.

2. METHODS AND MATERIALS

2.1. Participants

The study consisted of 94 participants: 33 patients diagnosed with AUD and 61 HCs recruited at the National Institute on Alcohol Abuse and Alcoholism (NIAAA), National Institute of Health. The full screening procedures and exclusion criteria have been reported previously. ²³ , ³⁴ This is a secondary analysis of two existing datasets: One consists of AUD inpatients who participated in a ketogenic diet intervention study (protocol 17‐AA‐0152 ³⁴ ), and the other consists of healthy volunteers who participated in a brain fMRI study (protocol 14‐AA‐0144 ²³ ).

AUD patients were admitted for alcohol detoxification at the NIAAA and were randomized into a 3‐week ketogenic diet (KD; n = 19) or a standard American (SA; n = 14) diet intervention. Among them, 30 AUD patients (KD = 17, SA = 13) completed the rs‐fMRI scans on Weeks 1 and 3. The HC subjects participated in a different study with MRI acquisition parameters that were not exactly the same as those for the AUD protocol (see Section 2.4). All participants were medically screened to exclude ferromagnetic implants, major medical problems, chronic use of psychoactive medications, neurological problems, liver disease, kidney stones, head trauma, and current diagnosis of a severe substance use disorder (SUD; other than AUD in the AUD group or nicotine) or any other major psychiatric disorder that needed treatment for longer than a month in the past year as assessed by the Structured Clinical Interview for the Diagnostic and Statistical Manual of Mental Disorders (DSM‐IV) or DSM‐5. ³⁵ Women were neither pregnant nor breastfeeding. All participants were free of psychoactive medications within 24 h of study procedures (except benzodiazepines as needed for AUD detoxification) and had a negative urine drug screen on days of testing. The AUD and HC studies were approved by the Ethics Committee of the National Institutes of Health (Combined Neurosciences White Panel) and were in accordance with the Declaration of Helsinki. All participants gave informed written consent before participating in the studies.

Demographic information is provided in Table 1. The AUD and HC groups showed significant group differences in sex and education level. The SUD and nicotine use of both groups are provided in Table S3.

TABLE 1.

Demographic characteristics of alcohol use disorder (AUD) and healthy control (HC) groups. Continuous variables are reported as mean (standard deviation) and categorical variables as counts (percent).

	AUD (n = 30)	HC (n = 61)	p
Age (years)	41.70 (13.07)	43.16 (14.43)	0.64
Sex	9 females (30%) 21 males (70%)	35 females (57%) 26 males (43%)	0.01
BMI (kg/m2)	25.42 (3.69)	26.99 (4.81)	0.12
Race	12 Black/African‐American 15 White 3 Other	25 Black/African‐American 24 White 12 Other	0.43
Education (years)	13.37 (2.83)	16.02 (2.59)	<0.001

Note: Group differences in age, BMI and education were tested using two‐sample t‐tests, and group differences in sex and race were tested using Chi‐square tests.

2.2. AUD withdrawal medications

AUD participants were assessed with the Clinical Institute Withdrawal Assessment—Alcohol revised (CIWA‐Ar) ³⁶ at admission and then approximately every 2 h until withdrawal ceased within the first week. If the CIWA‐Ar scores were ≥8 on admission, patients were provided with benzodiazepines (oxazepam or diazepam) to treat withdrawal symptoms. At admission, 16 out of 19 patients who would be subsequently randomized to KD received benzodiazepines (3 diazepam, 13 oxazepam, 3 none), and 10 out of 14 patients who would be randomized to SA received oxazepam (0 diazepam, 4 none). There were no significant group differences in benzodiazepine use on the MRI day (5 out of 19 in the KD group and 2 out of 14 in the SA group received benzodiazepines; ChiSq = 0.7, p = 0.68).

2.3. Neurocognitive assessment

AUD and HC participants completed the following assessments from the CANTAB suite ³⁷ : Intra‐Extra Dimensional Set Shift (IED), Pattern Recognition Memory (PRM), Reaction Time (RTI), Stockings of Cambridge (SOC), Spatial Span reverse‐mode (SSP) and Spatial Working Memory (SWM). See Supporting information for details on the tasks.

2.4. MRI acquisition

All participants underwent scanning on a 3T Magnetom Prisma scanner (Siemens Medical Solutions USA Inc.) with a 32‐channel head coil. AUD participants completed their first MRI an average of 4.2 days after admission and repeated the second and third scans in Weeks 2 and 3 of treatment. The rs‐fMRI data of AUD patients were acquired using a multiplexed gradient echo‐planar imaging (EPI) sequence with the following parameters: multiband factor = 8, anterior‐posterior phase encoding, Repetition Time (TR) = 891 ms, Echo Time (TE) = 37 ms, Flip Angle (FA) = 52°, 72 slices and isotropic voxel size = 2 mm. The same parameters were used for HC subjects except that TR = 720 ms. Participants were instructed to keep their eyes open during the 15‐min resting‐state scan while a fixation cross was presented on a black background. For the structural imaging, a T1‐weighted 3D magnetization‐prepared gradient‐echo image (MP‐RAGE) sequence was used for HC participants with the following parameters: TR = 2400 ms, TE = 2.24 ms, FA = 8° and isotropic voxel size = 0.8 mm. And the anatomical brain images of AUD participants were acquired with a MP‐RAGE sequence using the following parameters: TR = 2200 ms, TE = 4.25 ms, FA = 9° and isotropic voxel size = 1 mm.

2.5. MRI preprocessing

The rs‐fMRI data were preprocessed using Analysis of Functional NeuroImages (AFNI). ³⁸ , ³⁹ After removing the first six scans, the data were motion‐corrected, normalized to the standard MNI space, smoothed with a 5‐mm Full‐width at half‐maximum (FWHM) Gaussian kernel and band‐pass filtered (0.01–0.1 Hz). To further control the effect of head motion, scrubbing was also performed to cut time points with a frame‐wise displacement larger than 0.5 mm, and their previous one and subsequent two time points.

2.6. Region of interest definition

Using the Power‐264 brain atlas ⁴⁰ (Figure 1A), we defined 264 spherical brain regions of interest (ROIs) with a 5‐mm radius that belong to 13 large‐scale functional brain networks. Among these networks, the default mode, fronto‐parietal, ventral attention, dorsal attention, cingulo‐opercular and salience networks were grouped into the ‘association’ network; the sensory hand, sensory mouth, visual and auditory networks were grouped into the ‘sensorimotor’ network. ²³ Other networks include the subcortical, cerebellar and memory retrieval networks. For each ROI, we extracted the mean time series across voxels. We then calculated the Pearson correlation coefficients between the ROIs and converted them to Fisher‐z values for further analysis.

FIGURE 1.

(A) Two hundred sixty‐four regions of interest (ROIs) belonging to 13 networks are used for the brain segregation calculation. Different colours represent different networks (cyan: hand, orange: mouth, purple: cingulo‐opercular, pink: auditory, red: default mode, grey: memory retrieval, blue: visual, yellow: fronto‐parietal, black: salience, brown: subcortical, teal: ventral attention, green: dorsal attention, pale blue: cerebellar). (B) Lower segregation in the sensorimotor and association networks was found in alcohol use disorder (AUD) patients, compared to healthy controls (HCs). Error bars represent standard error of mean. **: p < 0.01. (C, D) Correlation between age and segregation of sensorimotor (C) and association (D) networks in the AUD and HC group. (E, F) Correlation between Spatial Span reverse‐mode (SSP) span length and segregation of sensorimotor (E) and association (F) networks in the AUD and HC group.

2.7. Brain segregation

As in previous studies, ²¹ , ²³ the segregation of brain networks was defined as the relative strength of within‐network connectivity when compared with between‐network connectivity. The formula is:

$$\text{Network segregation}=\frac{{\overline{Z}}_w-{\overline{Z}}_b}{{\overline{Z}}_w}$$

where ${\overline{Z}}_w$ represents the mean Fisher‐z value of all connections within the current network, and ${\overline{Z}}_b$ represents the mean Fisher‐z value of connections between the current and other networks. Following a previous study, ²³ only positive connectivity values were included in the analysis, and the segregation of each network was averaged to obtain the final measure of ‘sensorimotor’ and ‘association’ segregation. Other networks beyond the sensorimotor and association networks were not examined in the current study.

2.8. Statistical analyses

For all comparisons between the AUD and HC groups, sex, BMI, race and education were corrected in the statistical analysis. Analyses of covariance (ANCOVA) were used to investigate differences in brain segregation and cognitive performance between the AUD (Week 1) and HC groups. Two one‐way ANCOVAs were used for the sensorimotor‐ and association‐network segregation, respectively.

Pearson correlations were used to correlate brain network segregation with age and with cognitive task performance showing difference between the two groups. In the correlation analyses, outliers (>3 SD) in segregation were excluded. Correlation slopes between age and segregation, and between cognitive performance and segregation were compared between groups using Fisher's z statistics. We also tested the potential mediating role of brain segregation in the relationship between group and cognitive performance using the PROCESS macro ⁴¹ in SPSS version 28.

Furthermore, patients with AUD demonstrate low brain glucose metabolism, ⁴² which, in healthy volunteers, covaried with brain network segregation. ²³ As heavy alcohol users shift brain energy utilization from glucose to acetate, which might jeopardize brain function during withdrawal, ⁴³ we hypothesized that a ketogenic diet intervention would boost brain energy in AUD patients, hence improve brain network segregation and cognition. ³⁴ , ⁴⁴ Therefore, we performed the following exploratory analyses in the AUD group: (1) associations between CIWA‐Ar withdrawal and brain network segregation, (2) effects of 3‐week abstinence on brain network segregation and (3) effects of diet intervention on brain network segregation and behaviour using repeated measures ANOVAs (see Supporting information). Both benzodiazepine usage (yes/no) and age were added as covariates to the ANOVAs.

3. RESULTS

3.1. Group difference (AUD vs. HC) in brain segregation

Compared with the HCs, the AUD participants showed lower brain segregation in both the sensorimotor (F(1,83) = 11.703, p = 0.001, η ² = 0.124; Figure 1B) and association (F(1,83) = 11.188, p = 0.001, η ² = 0.119; Figure 1B) networks. The group differences remained significant after including age as a covariate (sensorimotor: F(1,82) = 10.886, p = 0.001, η ² = 0.117; association: F(1,82) = 10.444, p = 0.002, η ² = 0.113). To further explore the contribution of each individual network, we compared their segregation between groups and found that comparing with AUD patients, the HC group showed higher brain segregation in sensory hand, sensory mouth, auditory, memory retrieval, visual, fronto‐parietal, and ventral and dorsal attention networks (Table S2).

As reported in Manza et al., ²³ we found similar negative correlations between age and segregation in both the sensorimotor and association networks in the HC group (sensorimotor: r = −0.306, p = 0.025, Figure 1C; association: r = −0.299, p = 0.028, Figure 1D); however, these correlations were not significant in the AUD group (sensorimotor: r = −0.013, p = 0.953, Figure 1C; association: r = −0.228, p = 0.272, Figure 1D). The correlation coefficients did not differ between groups (sensorimotor: z = −1.277, p = 0.201; association: z = −0.326, p = 0.745). CIWA‐Ar withdrawal did not correlate with brain network segregation in the AUD group (sensorimotor: r = −0.116, p = 0.582; association: r = −0.009, p = 0.966).

3.2. Group difference (AUD vs. HC) in cognitive function

When examining the group differences in cognitive performance, we found that the AUD group scored significantly lower than HC on the SSP length (F(1,77) = 4.153, p = 0.045, η ² = 0.051), which measures WM capacity. After controlling for age, there was a non‐significant trend for group difference in the same direction (F(1,76) = 3.337, p = 0.072, η ² = 0.042). There were no group differences on other tasks (Table S1). Therefore, only the SSP performance was considered in the following analyses.

Although it is difficult to rule out a benzodiazepine effect given that only the AUD group received them, we found no significant effects of benzodiazepine use (i.e., use on the MRI day and cumulative use) on brain network segregation and cognitive performance within the AUD group (network segregation: all p's > 0.323; cognitive performance: all p's > 0.479).

3.3. Correlation between segregation and cognition in AUD and HC

Further brain‐behavioural association analysis revealed that for both sensorimotor (Figure 1E) and association (Figure 1F) networks, higher segregation was correlated with better SSP performance in the HC group (sensorimotor: r = 0.406, p = 0.004; association: r = 0.483, p = 0.001), but not the AUD group (sensorimotor: r = −0.067, p = 0.754; association: r = −0.058, p = 0.783). Comparisons of correlation coefficients revealed a significant group difference in both sensorimotor (z = 2.059, p = 0.039) and association segregation (z = 2.449, p = 0.014). These findings still hold when adding age as an additional covariate (sensorimotor: AUD: r = −0.081, p = 0.712, HC: r = 0.328, p = 0.024; association: AUD: r = −0.180, p = 0.401, HC: r = 0.380, p = 0.008).

3.4. Mediation analysis

When adding age as a covariate, brain segregation in both sensorimotor and association networks did not significantly mediate the relationship between group (AUD vs. HC) and SSP performance (sensorimotor: indirect effect: unstandardized B = −0.210, 95% CI [−0.567, 0.058], direct effect: B = −0.447, 95% CI [−1.215, 0.322]; association: indirect effect: B = −0.303, 95% CI [−0.721, 0.024], direct effect: B = −0.340, 95% CI [−1.090, 0.410]).

3.5. Effect of alcohol abstinence and ketogenic diet intervention

Brain network segregation measures did not improve over the 3‐week abstinence (sensorimotor: F(2,52) = 0.696, p = 0.503; association: F(2,52) = 0.747, p = 0.479). Moreover, while the KD group showed elevated associative brain network segregation compared to the SA group in the first week of scanning (sensorimotor: t(28) = −1.790, p = 0.084; association: t(28) = −2.521, p = 0.018; when blood BHB level was 1.694 for KD and 0.139 for SA), there was no significant main effect of diet (sensorimotor: F(1,26) = 0.490, p = 0.490; association: F(1,26) = 2.095, p = 0.160) and no significant interaction effect of intervention‐by‐time (sensorimotor: F(2,52) = 1.109, p = 0.338; association: F(2,52) = 0.794, p = 0.458) (see Supporting information).

The SSP task performance did not improve with 3‐week abstinence (F(1,26) = 0.078, p = 0.783), and there was no main effect of diet intervention (F(1,26) = 0.495, p = 0.488) and no interaction effect of intervention‐by‐time on SSP performance (F(1,26) = 0.873, p = 0.359).

4. DISCUSSION

In line with our hypothesis, we observed lower brain segregation in both the sensorimotor and association networks in AUD compared to HCs. However, it is noteworthy that brain network segregation was significantly negatively correlated with age exclusively in the HC group. This observation deviates from our hypothesis, which proposed a similar link between reduced segregation and age in the AUD group. Additionally, brain network segregation displayed a positive correlation with SSP task performance only within the HC group, but not the AUD group.

Brain network segregation is a composite metric of network organization that summarizes the relative strength of within‐community connections in comparison to between‐community connections. ¹⁷ , ¹⁸ Therefore, the lower segregation of both the sensorimotor and association networks in AUD patients compared to HC might reflect a diminished level of specificity in information processing within the sensory, motor and more integrated brain regions in the AUD group. Importantly, it is worth mentioning that this neural deficit is not unique to AUD, as many psychiatric disorders exhibit reduced brain segregation. Examples include Alzheimer's disease, ²⁵ schizophrenia, ⁴⁵ , ⁴⁶ autism spectrum disorder and attention‐deficit/hyperactivity disorder (see Cao et al. ⁴⁷ for review).

While network segregation is a macroscopic brain metric, recent literature has linked it to relatively microscopic processes including the cholinergic basal forebrain and noradrenergic locus coeruleus function. ⁴⁸ Shine argues that the cholinergic projections from the basal forebrain to the cortex promote brain segregation, and the noradrenergic projections from the locus coeruleus to the cortex promote integration. The combined activity of these two systems modulates the balance between segregation and integration on a whole‐brain level. ⁴⁸ Therefore, it is conceivable that the altered network segregation in AUD reflects abnormal activity and/or connectivity within and between the cholinergic basal forebrain and cortex. This is particularly relevant given that alcohol modulates acetylcholine receptors, ⁴⁹ , ⁵⁰ with both nicotinic and muscarinic receptors implicated in alcohol‐related behaviours such as dependence, frequency of use, self‐administration and cue‐induced reinstatement. ⁴⁹ , ⁵¹ , ⁵² A frequent confound in brain imaging studies with AUD is the high co‐morbidity that exists with tobacco smoking, which has been shown to affect brain functional connectivity ²⁶ and decrease network segregation. ⁵³ For our study, 53% of AUD patients and none of the HC participants were smokers. Nevertheless, adding smoking status as a covariate to the ANCOVA did not change the significant effect of AUD on segregation (sensorimotor: F(1, 81) = 7.634, p = 0.007, η ² = 0.086; association: F(1, 81) = 7.386, p = 0.008, η ² = 0.084).

Notably, we were able to replicate the findings of a previous study revealing a negative correlation between age and network segregation in healthy adults. ²¹ , ²³ However, we did not observe a significant age‐related segregation decline in the AUD group. It is possible that segregation within the younger AUD patients has already decreased to a level comparable to that of elderly HCs, resulting in a more uniform level of segregation across all AUD patients, irrespective of age. In a recent study, greater age was associated with lower alcohol risk scores on the Alcohol Use Disorders Identification Test (AUDIT), positive alcohol expectancy and craving. ⁵⁴ Another study examined drinking levels in an AUD cohort that did not maintain abstinence and found that individuals with low drinking levels after relapse had significantly more frontal grey matter volume and thalamic volumes than those with high drinking levels, with no differences in brain volume measures between individuals with low drinking levels and those who maintained abstinence. ⁵⁵ This suggests that even a reduction in drinking levels might be beneficial for brain volume recovery. Therefore, we also considered the possibility that the absence of age‐related segregation decline in the AUD group might be due to a decrease in problem drinking with age, which could contribute to the recovery of brain measures. However, when measuring AUD severity with AUDIT and Alcohol Dependence Scale, we did not find any significant correlations between either age and severity scores (|r|'s < 0.070, p's > 0.05), or severity scores and network segregation (|r|'s < 0.003, p's > 0.05) in the present study. It is also possible that we did not have the requisite sample size to see a significant association between brain segregation and age in the AUD group, since such correlation for the association network in AUD was about the same size as that of the HC group but was not significant.

As indicated in Figure 1C,D, the segregation of sensorimotor and association networks in AUD patients in their twenties is similar to those of HCs in their sixties and seventies. This finding provides additional support for the premature aging hypothesis in AUD, which has been evidenced by many previous structural MRI studies. For example, the grey matter volume of the precentral and superior frontal cortices ³³ and hippocampus ⁵⁶ declines more rapidly with aging in AUD compared to controls. Our study further contributes functional evidence to the premature aging hypothesis in AUD.

In addition to abnormally low network segregation, patients with AUD also show impairments in SWM performance. Such impairment is consistent with previous findings ⁵⁷ and has been associated with poorer clinical outcomes, including heavier substance use ⁵⁸ and higher relapse risk. ⁵⁹ However, the group difference would not survive multiple comparison (p = 0.045) and should be interpreted with caution.

We also observed a group difference in the association between segregation and SSP performance. Within the HC group, larger SWM capacity is correlated with higher segregation of sensorimotor and association networks, suggesting that their normal WM function benefits from localized information processing throughout the brain. In contrast, for AUD patients, there is no significant correlation between WM capacity and brain segregation. It should be noted that the current study employed a static measure of brain network segregation, while brain connectivity is dynamic by nature. ⁶⁰ , ⁶¹ When compared with static functional connectivity, dynamic functional connectivity has been shown to be more robust in predicting individual differences in cognition. ⁶² , ⁶³ , ⁶⁴ , ⁶⁵ Therefore, it is possible that dynamic functional connectivity, which provides a more precise time‐dependent profile of brain networks, may play a more crucial role in mediating WM task performance in AUD patients. While no study to date has directly examined the role of other brain network properties in SWM performance, researchers have identified other neural substrates underlying WM in individuals with AUD. These include connectivity between the posterior cingulate cortex and cerebellum, ⁶⁶ grey matter density in regions of the salience network ⁶⁷ and regional cerebral blood flow in the anterior/middle cingulate cortex, insula and thalamus ⁶⁸ in AUD.

The present study has several limitations. First, it consists of a secondary analysis of existing data collected in previous studies, and because the AUD and HC groups were enrolled in two separate studies, they were not proactively matched for age, sex or education levels. The sample size was also not determined a priori, as these were convenience samples. Second, the study was based on the analysis of cross‐sectional data, revealing correlational rather than causal relationships. It remains unclear if the compromised WM capacity and network segregation are negative consequences of AUD, or predisposing factors of AUD, which would require future longitudinal studies. Third, although we did not find significant effects of benzodiazepine use on brain network segregation and cognitive performance in AUD participants, it is difficult to rule out the benzodiazepine effect since only the AUD group received them. Fourth, the effect size of the group differences in segregation and cognition was medium to large, based on Cohen's benchmarks. ⁶⁹ , ⁷⁰ , ⁷¹ Our small sample size (i.e., 30 AUD patients) may have limited our ability to detect small effect sizes. Future studies with a larger sample may reveal more subtle brain and behavioural differences between AUD and HC individuals. Finally, while we ruled out SUDs in the HC group based on the DSM‐5 or IV, we only excluded SUDs that required treatment in the past year for the treatment‐seeking AUD patient group. As shown in Table S3, the AUD group contained more smokers (46.7% vs. 0%) and individuals with cannabis use disorder (36.7% vs 0%), stimulant use disorder (13.3% vs 0%) and opioid use disorder (3.3% vs 0%). Although the effect of AUD on brain network segregation remained after controlling for smoking status, we did not correct for other SUDs in between group analyses.

AUTHOR CONTRIBUTIONS

Gene‐Jack Wang, Peter Manza, Nora D. Volkow and Corinde E. Wiers designed the study and collected the data. Xinying Wang, Peter Manza, Zhenhao Shi and Corinde E. Wiers analysed the data. All authors interpreted the results. Xinying Wang, Peter Manza, Xinyi Li, Astrid Ramos‐Rolón, Nathan Hager, Nora D. Volkow, Zhenhao Shi and Corinde E. Wiers wrote the manuscript. All authors provided critical revision of the manuscript for intellectual content and approved the final version of the manuscript.

CONFLICT OF INTEREST STATEMENT

None of the authors have conflicts of interest to disclose.

ETHICS APPROVAL STATEMENT

The studies were approved by the Ethics Committee of the National Institutes of Health (Combined Neurosciences White Panel) and were in accordance with the Declaration of Helsinki.

Supporting information

Figure S1. (A, B) Repeated measures ANOVA on the segregation of sensorimotor (A) and association (B) networks revealed no main effect of group or time, and no interaction effect between them. Error bars represent standard error of mean.

Table S1. Difference in CANTAB performance between the AUD (week 1) and HC group. Scores on each measure are reported as mean (standard deviation).

Table S2. Difference in the segregation of individual networks between the AUD and HC group.

Table S3. SUD diagnoses and nicotine use in the AUD and HC groups.

Wang X, Manza P, Li X, et al. Reduced brain network segregation in alcohol use disorder: Associations with neurocognition. Addiction Biology. 2024;29(12):e13446. doi: 10.1111/adb.13446

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.