Brain network segregation is associated with drug use severity in individuals with opioid use disorder

Abstract

Background:

Opioid use disorder (OUD) is associated with altered brain network connectivity, particularly in the fronto-parietal, default mode, and salience networks. Brain efficiency is maximized when networks are distinct (‘segregated’) yet maintain partial connectivity with other networks (‘integrated’). ‘Brain network segregation’ quantifies this balance by comparing the functional connectivity of nodes within and between networks. Previous research found lower brain network segregation in people with cognitive impairment, alcohol use disorder, and older age. We hypothesized that recent drug use severity in people with OUD would relate to reduced brain network segregation.

Method:

Forty treatment-seeking adults with OUD (72.5 % male) completed resting-state functional magnetic resonance imaging. We grouped 264 brain regions into 10 networks, categorized as “association” (higher-order cognition) or “sensorimotor” (sensory and motor) networks. Regression analysis tested the relation between drug use severity and brain network segregation of association and sensorimotor categories and specific networks. Partial correlations explored the effects of cognition (IQ and working memory), mood, and affect.

Results:

Drug use severity predicted lower brain network segregation of the association networks, particularly the fronto-parietal and salience networks, but not the default mode network. The relation between drug use severity and lower segregation of the sensorimotor networks depended on age. In exploratory analyses, positive affect related to greater salience network segregation.

Conclusions:

An altered balance of connectivity within and between brain networks may correspond with drug use severity, particularly in cognitive and salience-detection networks. Lower brain network segregation may indicate accelerated brain aging and be a target for OUD treatment.

1. Introduction

Approximately six million people in the United States experience opioid use disorder (OUD) each year (Substance Abuse and Mental Health Services Administration, 2023, 2024). Risks of OUD include opioid overdoses, medical complications, social dysfunction, cognitive deficits, and poor mental health (Hser et al., 2017; Mahoney et al., 2023; Rhee and Rosenheck, 2019; Zibbell, 2019). Underlying these poor outcomes, chronic substance use affects neural functioning in a variety of domains—including reward, emotion, stress, sensorimotor, learning, and executive function (Koob et al., 2023).

Understanding the widespread brain effects of OUD is critical to expanding therapeutic targets (Volkow and Boyle, 2018). Previous work has shown that resting-state functional connectivity (rsFC) assessed by functional magnetic resonance imaging (fMRI) can shed light on the impact of substance use disorders (SUDs) on distinct brain networks (i.e., widespread brain areas defined by shared functional domains) (van den Heuvel and Hulshoff Pol, 2010). The most common networks implicated in OUD are the same as those most implicated in SUDs in general: fronto-parietal network (FPN), default mode network (DMN), and salience network (SN) (Ieong and Yuan, 2017; Moningka et al., 2019; Zeng et al., 2024). Analysis of rsFC networks may help parse subgroups of SUD (Drossel et al., 2023), identify which networks protect from or contribute to SUD (Ersche et al., 2020), and improve targeted intervention (Wilcox et al., 2019).

Researchers have applied rsFC to compute graph theory-based metrics of global brain network organization, including modularity and clustering coefficient. These measures capture the overall topological architecture and efficiency of brain networks (Farahani et al., 2019). One such metric, brain network segregation (also known as system segregation), quantifies the balance between having segregated and integrated brain networks by representing the relative strength of a network’s within-network correlations (i.e., segregation) compared to its between-network correlations (i.e. integration) (Chan et al., 2014). The property of distinct yet partially integrated neural networks is a fundamental aspect of the human brain that balances network specialization with global integration (Tononi, 1998). Wig (2017) proposed that this balance of network segregation preserves functional heterogeneity/specialization in the brain, contributes to energetically efficient neural adaptation, and promotes resilience to neural damage. Chan and colleagues (2014) showed that brain network segregation predicts aging better than other graph theory metrics. To gain some specificity of where network segregation differences appear, previous studies have found meaningful differences when separately examining the “association” networks and “sensorimotor” networks (Chan et al., 2014; Manza et al., 2020). Association networks are involved in higher-order cognition and information integration while sensorimotor networks process incoming sensory and outgoing motor information.

Studies consistently show that resting-state brain networks become less segregated over the course of normal adult aging and that less segregation is associated with poorer cognitive performance and dementia (Chan et al., 2014; Manza et al., 2020; Nashiro et al., 2017; Yue et al., 2017; Zhang et al., 2023). Only one study has examined brain network segregation in SUD. As expected, individuals with alcohol use disorder had lower brain network segregation in the association and sensorimotor networks compared to healthy control participants (Wang et al., 2024). Studies using similar connectivity metrics (e.g., small-worldness, clustering coefficient) have shown similar findings, such that SUDs may be characterized by less efficient and hyper-connected networks (Sjoerds et al., 2017; Wang et al., 2015). No previous research has examined network segregation in OUD, but one study found that people with OUD have decreased small-worldness and clustering coefficient (Jiang et al., 2013). These findings indicate that OUD is characterized by more random brain organization and possibly lower network segregation. Examining network segregation in people with OUD may provide insight into the effects of chronic opioid use on widespread brain network organization, efficiency, and adaptability.

We present analyses of brain network segregation in relation to drug use severity in individuals with OUD. These are secondary analyses of data from a clinical trial of injectable, once-monthly extended-release naltrexone (XR-NTX) for OUD (Shi et al., 2018). By testing network segregation against a continuous measure of drug use severity, we sought to understand the sensitivity of network segregation to OUD symptomatology. We hypothesized that drug use severity and age would independently predict lower brain network segregation in OUD, particularly in the FPN, DMN, and SN. We extend previous research by examining network segregation in OUD and testing the effects both broadly and within specific networks. In exploratory analyses, we expected that network segregation would increase across XR-NTX treatment and drug use severity would decrease. We further explored the associations of brain network segregation with emotion (depression, anxiety, negative affect, and positive affect) and cognition (IQ and working memory).

2. Method and materials

2.1. Participants

Treatment-seeking individuals with OUD from the greater Philadelphia area volunteered to receive up to three monthly XR-NTX injections. Data were collected between 2012 and 2014. Of the initial 46 participants, we obtained the final sample of N = 40 after excluding for incomplete scans (n = 4), excessive head motion (n = 1), and incomplete Addiction Severity Index (n = 1). See Table 1 for demographics and baseline characteristics.

Table 1.

Participant demographics and mean scores.

	n	Range	Mean (SD)
Age (years)	40	19 – 56	29.33 (9.21)
Education (years)	40	7 – 19	14.00 (2.18)
Drug severity	40	0.08 – 0.51	0.29 (0.11)
Segregation of Sensorimotor networks	40	0.30 – 0.60	0.46 (0.08)
Segregation of Association networks	40	0.18 – 0.45	0.34 (0.06)
Depression	39	0 – 19	6.97 (5.01)
Anxiety	38	0 – 20	6.76 (5.32)
Positive affect	24	12 – 49	29.29 (9.01)
Negative affect	24	12 – 45	25.75 (9.02)
IQ	40	80 – 138	102.43 (12.20)
Working memory	26	75 – 117	100.12 (11.45)
		%
Sex
Female	11	27.5 %
Male	29	72.5 %
Race
Asian	1	2.5 %
Black	3	7.5 %
Hispanic	3	7.5 %
White	33	82.5 %

2.2. Study procedures

This study was approved by the university’s Institutional Review Board, registered as NCT01587196, and conducted in accordance with the Declaration of Helsinki. All participants signed voluntary informed consent. Prior to their baseline MRI, participants underwent outpatient opioid detoxification. Following the baseline MRI, the on-treatment MRI and post-treatment MRI were completed a mean of 13.0 (SD = 7.7) and 112.2 (SD = 21.4) days later, respectively. The first XR-NXT injection occurred after the baseline scan (see Supplemental Materials).

2.3. Measures

2.3.1. Drug use severity

The Addiction Severity Index-5th Edition (ASI) (McLellan et al., 1992) and its Drug Composite score (ASI-drug) was used to measure drug use severity. The ASI-drug focuses on the past 30 days and all abused substances except alcohol. It consists of self-reported days of use, number and severity of negative drug-related consequences, and ratings for treatment need. Higher scores indicated higher recent drug use severity.

2.3.2. Depression and anxiety

Depression severity was assessed by the Hamilton Depression Rating Scale (Hamilton, 1960), which is a 17-item clinician-rated measure. The internal consistency (Cronbach’s alpha [α]) was acceptable (α =.73). We assessed anxiety with the 14-item, clinician-rated Hamilton Anxiety Rating Scale (Hamilton, 1959). Internal consistency was good (α =.80). Higher scores indicated greater symptom severity.

2.3.3. Positive and negative affect

Positive and negative affect were assessed on the 10-item subscales of the Positive Affect Negative Affect Schedule (PANAS) (Watson et al., 1988). Participants self-reported the frequency that they experienced various positive and negative emotions. Internal consistency was good for the positive (α =.89) and negative (α =.89) subscales.

2.3.4. Cognition

General cognitive function (IQ) was calculated from the Vocabulary and Matrix Reasoning subtests of the Wechsler Abbreviated Scale of Intelligence (WASI) (Wechsler, 1999). We measured working memory using the Working Memory Index (WMI) of the Wechsler Assessment of Intelligence Scale-III (Wechsler, 1997). IQ and WMI scores were age-adjusted and standardized to the norming sample.

2.4. MRI data

2.4.1. Acquisition and pre-processing

MRI data were collected on a Siemens Trio 3 T system (Siemens AG, Erlangen, Germany) and preprocessed using a pipeline adapted from Ciric and colleagues (2018). Preprocessing steps included slice time correction, realignment, skull stripping, despiking, detrending, interpolation of outlier volumes, bandpass filtering (0.01–0.1 Hz), regressing out motion parameters and nuisance signals (white matter, cerebrospinal fluid), spatial smoothing, and normalization to the Montreal Neurological Institute space. Mean framewise displacement (FD) was calculated to control for head motion in analyses. See the Supplemental Materials for details.

2.4.2. Region of interest definition

Following previous network segregation studies (Chan et al., 2014; Manza et al., 2020), we defined 264 regions of interest (ROIs) based on the Power-264 brain atlas (Power et al., 2011). ROIs were a 5-mm sphere centered around the atlas coordinates and were assigned to one of 13 distinct brain networks. These networks were grouped into three categories: “association” networks (default mode, fronto-parietal, ventral attention, dorsal attention, cingulo-opercular, and salience); “sensorimotor” networks (hand sensory-somatomotor, mouth sensory-somatomotor, visual, and auditory); and “other” networks (subcortical, cerebellar, and memory retrieval) (Chan et al., 2014).

2.4.3. Brain network segregation

For each subject, we generated correlation matrices by calculating the Pearson correlation coefficients between the time courses of all ROIs. Fisher’s Z-transformation converted these correlations into normally distributed Z-values, resulting in Z-transformed correlation matrices representing the functional connectivity between each ROI and all other ROI’s. Network segregation was calculated at the level of each predefined network. For each network, network segregation was defined as the difference between the averaged Z-transformed correlations of ROIs within a given network (${\overline{z}}_w$) and the averaged Z-transformed correlations of ROIs between that network and all other networks (${\overline{z}}_b$). This difference (${\overline{z}}_w-{\overline{z}}_b$) was then normalized by the within-network connectivity (${\overline{z}}_w$) (Chan et al., 2014):

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

As such, network segregation represents the balance between having a segregated network (contributed to by ${\overline{z}}_w$) and an integrated one (contributed to by ${\overline{z}}_b$). The theoretical maximum of network segregation is 1.0, and 0.0 indicates ROIs are equally connected within and between networks. Values > 0.0 indicate within-network correlations are stronger than the between-network correlations and values < 0.0 indicate between-network correlations are stronger. At the network category level, we calculated segregation of the sensorimotor and association networks by averaging the network segregation of their constituent networks. Consistent with previous studies, and to avoid potentially spurious negative correlations, we analyzed only positive correlations (Chan et al., 2014; Manza et al., 2020; Zhang et al., 2023).

2.5. Statistical analyses

The primary analyses included two hierarchical regressions with segregation of association networks or sensorimotor networks as the dependent variable. Independent variables in the steps of the regressions were: model 1) mean FD [covariate model correcting for fMRI head motion]; model 2) adding age and ASI-drug; and model 3) adding the ASI-drug×age interaction. In follow-up analyses, the six association networks were used, in turn, as the dependent variable. In exploratory analyses in a subsample, repeated-measures ANCOVAs tested the effect of XR-NTX treatment on network segregation from baseline to on-treatment and post-treatment. A t-test examined change in ASI-drug from baseline to post-treatment. Exploratory analyses used partial correlations to examine the relation between baseline network segregation and baseline HAM-D, HAM-A, PANAS subscales, IQ, and WMI.

We conducted statistical tests in SPSS (version 28.0) and used percentile bootstrapping with 5000 samples. Interactions were probed with simple slopes and Johnson-Neyman regions-of-significance tests (SPSS PROCESS Macro version 4.2). We controlled for Family Wise Error Rate (FWER) or False Discovery Rate (FDR) (Verhoeven et al., 2005). Specifically, we adjusted for having two tests in our main analyses by controlling for FWER with Bonferroni correction (α <.025). In follow-up analyses, we adjusted for the larger number of tests by controlling for FDR using the Benjamini-Hochberg procedure (Verhoeven et al., 2005).

3. Results

3.1. Association networks

Model 2 significantly improved upon model 1, $R_{\mathrm{change}}^2=.19$, p = .006. ASI-drug was significantly related to lower segregation of the association networks, B = −0.26, p = .006, 97.5 % CI [−0.45, − 0.06] (Fig. 1). ASI-drug accounted for 17 % of the variance in segregation of the association networks and showed a medium-to-large effect, r_partial(36) = −.48, p = .002. The trend of age relating to lower segregation of the association networks did not survive correction for multiple tests, B = −0.002, p = .046, 97.5 % CI [−0.004, 0.0002] (Figure S1 in Supplemental Materials). In model 3, the ASI-drug×age interaction did not significantly improve model fit, $R_{\mathrm{change}}^2=.02$, p = .23. See Table 2.¹

Fig. 1.

Scatter plot of segregation of association networks by drug use severity. Note. The variables are adjusted by regressing out age and motion (mean framewise displacement). Drug use severity is derived from the Addiction Severity Index’s Drug Composite Score.

Table 2.

Hierarchical multiple regression predicting segregation of association networks.

Model	R2 (change)	Change Sig.		B	SE	β	Sig.	97.5 % CI Lower	97.5 % CI Upper
1	0.24	0.001a	Intercept	0.39	0.02		<.001a	0.35	0.43
			Head motion	−0.55	0.17	−0.49	.002a	−0.85	−0.08
2	.43 (.19)	0.006a	Intercept	0.40	0.02		< .001a	0.36	0.43
			Head motion	−0.58	0.14	−0.51	< .001a	−0.83	−0.21
	(.02)		Age	−0.002	0.001	−0.26	.046b	−0.004	0.0002
	(.17)		ASI-drug	−0.26	0.09	−0.43	.006a	−0.45	−0.06
3	.45 (.02)	0.23	Intercept	0.39	0.02		< .001a	0.35	0.43
			Head motion	−0.54	0.15	−0.48	.002a	−0.79	−0.14
			Age	−0.002	0.001	−0.33	.044b	−0.005	−0.0003
			ASI-drug	−0.27	0.09	−0.46	.004a	−0.47	−0.07
			Age x ASI-drug	−0.01	0.01	−0.17	.27	−0.04	0.02

Note. Head motion = mean framewise displacement; B = unstandardized beta coefficient; β = standardized beta coefficient; CI = percentile bootstrapped confidence interval.

^aSignificant effect that survived Bonferroni correction (α =.025, 97.5 % CI)

^bNon-significant trend that did not survive Bonferroni correction (α =.025, 97.5 % CI)

3.2. Sensorimotor networks

Model 2 significantly improved upon model 1, $R_{\mathrm{change}}^2=.20$, p = .017. Only age was significantly related to segregation of sensorimotor networks, B = −0.004, p = .007, 97.5 % CI [−0.007, − 0.0005], while the effect of ASI-drug was not significant, B = −0.09, p = .44, 97.5 % CI [−0.35, 0.17]. Age accounted for 19 % of the variance in segregation of sensorimotor networks and showed a medium-to-large effect, r_partial(36) = −.45, p = .005. This effect was qualified by a significant improvement in model fit when adding the ASI-drug×age interaction in model 3, $R_{\mathrm{change}}^2=.12$, p = .01. This interaction was on the edge of the corrected significance level, B = 0.04, p = .026, 97.5 % CI [−0.003, 0.079], but we probed it due to the significant improvement in model fit. Simple slopes showed ASI-drug was related to lower segregation of sensorimotor networks at 1 SD below mean age, B = −0.41, p = .019, but not at 1 SD above mean age, B = 0.33, p = .10 (Figure S2). The Johnson-Neyman tests showed a significant negative effect of ASI-drug at age ≤ 23 (n = 14)². which could alternatively be interpreted as a significant negative effect of age when ASI-drug scores were low (i.e., below the mean [n = 18]). In the age ≤ 23 group (see Table S1), we found a large negative partial correlation (controlling for mean FD and age) between ASI-drug and segregation of sensorimotor networks, r_partial(10) = −.63, p = .029 (Fig. 2). Likewise, there was a large negative partial correlation between age and segregation of sensorimotor networks only in participants with low ASI-drug scores, r_partial(14) = −.78, p < .001 (Figure S3). See Table 3¹. Removing two potential outliers retained a significant ASI-drug×age interaction, while the effect within the age ≤ 23 group was reduced (see Supplementary Results).

Fig. 2.

Scatter plot of segregation of sensorimotor networks by drug use severity and age. Note. The variables are adjusted by regressing out age and motion (mean framewise displacement). Drug use severity is derived from the Addiction Severity Index’s Drug Composite Score.

Table 3.

Hierarchical multiple regression predicting segregation of sensorimotor networks.

Model	R2 (change)	Change Sig.		B	SE	β	Sig.	97.5 % CI Lower	97.5 % CI Upper
1	0.02	0.38	Intercept	0.48	0.03		< .001a	0.41	0.53
			Head motion	−0.20	0.22	−0.14	.34	−0.59	0.43
2	.22 (.20)	0.017a	Intercept	0.47	0.03		< .001a	0.40	0.52
			Head motion	−0.14	0.24	−0.10	.53	−0.54	0.47
	(.19)		Age	−0.004	0.001	−0.47	.007a	−0.007	−0.0005
	(.01)		ASI Drug	−0.09	0.12	−0.12	.44	−0.35	0.17
3	.34 (.12)	0.015a	Intercept	0.49	0.03		< .001a	0.42	0.54
			Head motion	−0.25	0.23	−0.18	.25	−0.63	0.43
			Age	−0.003	0.002	−0.30	.15	−0.006	0.001
			ASI Drug	−0.04	0.09	−0.06	.64	−0.26	0.18
			Age x ASI Drug	0.04	0.02	0.39	.026b	−0.003	0.08

Note. Head motion = mean framewise displacement; B = unstandardized beta coefficient; β = standardized beta coefficient; CI = percentile bootstrapped confidence interval.

^aSignificant effect that survived Bonferroni correction (α =.025, 97.5 % CI)

^bNon-significant trend that did not survive Bonferroni correction (α =.025, 97.5 % CI)

3.3. Specific association networks

Fig. 3 shows spring-embedded layouts (Cytoscape version 3.10.1) visualizing edges and nodes for each network. Visual inspection showed increased spread among the FPN and SN nodes in people with higher ASI-drug. Networks with significant effects are described below, while other association networks results are in the Supplemental Materials.

Fig. 3.

Spring-embedded layouts of resting state functional connectivity by network. Note. Plots were created using 5 % edge threshold. Observe the greater spread between nodes among the Fronto-parietal and Salience networks in the high drug severity group, as was substantiated by the statistical results. Low drug severity: Addiction Severity Index Drug Composite ≤ −0.5 SD below the mean. High drug severity: Addiction Severity Index Drug Composite ≥ 0.5 SD above the mean.

3.3.1. Salience network

FDR-corrected α = .008. Model 2 significantly improved upon model 1, $R_{\mathrm{change}}^2=.27$, p = .002. ASI-drug was significantly related to lower SN segregation, B = −0.46, p = .004, 97.5 % CI [−0.82, − 0.16]. ASI-drug accounted for 24 % of the variance in SN segregation, r_partial(36) = −.52, p < .001 (Figure S4). Age showed a non-significant but trending relation to reduced SN segregation, B = −0.003, p = .040, 97.5 % CI [−0.007, 0.000]. The ASI-drug×age interaction did not significantly improve model fit, $R_{\mathrm{change}}^2=.01$, p = .62.

3.3.2. Fronto-parietal network

FDR-corrected α = .017. Model 2 showed a trend toward improving upon model 1 after correction for multiple tests, $R_{\mathrm{change}}^2=.18$, p = .022. There was a significant effect of ASI-drug being related to lower FPN segregation, B = −0.39, p = .011, 97.5 % CI [−0.73, - 0.10]. ASI-drug accounted for 18 % of the variance in FPN segregation, r_partial(36) = −.43, p = .005 (Figure S4). Age was not significantly related to FPN segregation, B = −0.001, p = .688, 97.5 % CI [−0.004, 0.003]. The ASI-drug×age interaction did not significantly improve model fit, $R_{\mathrm{change}}^2=.01$, p = .64.

3.4. Exploratory results

3.4.1. Network segregation and treatment outcome

Outcome analyses included participants who completed the post-treatment assessments (n = 16). ASI-drug scores significantly reduced from baseline to post-treatment, t(15) = 6.43, p < .001, 95 % CI [0.14, 0.26], Hedges’ g = −1.57. This is consistent with a benefit of naltrexone treatment, such that 100 % of the ASI-drug scores at post-treatment were below the baseline mean. In contrast, network segregation for both the association and sensorimotor networks was stable across all three time points (p’s > .23; see Table 4 and Supplemental Materials).

Table 4.

Comparing mean network segregation and drug use severity across time.

	Time 1	Time 2 Mean (SD)	Time 3	Time 1 vs. Time 2		Time 1 vs. Time 3		Time 2 vs. Time 3
				p	g	p	g	p	g
Days since Time 1	–	13.0 (7.7)	112.2 (21.4)	–		–		–
Segregation of Association networks	.355 (.066)	.334 (.086)	.332 (.072)	.24	− 0.30	.16	− 0.36	.93	− 0.02
Segregation of Sensorimotor networks	.436 (.090)	.428 (.111)	.448 (.071)	.70	− 0.10	.58	− 0.14	.43	0.20
Drug use severity	.262 (.103)	–	.062 (.072)	–		< .001	1.57	–

Note: Drug use severity was not collected at Time 2.

Time 1 = baseline/pre-treatment, Time 2 = on-treatment (week 2 or 3), Time 3 = post-treatment (>1 month post-injection); SD = standard deviation. g = Hedges’ g.

3.4.2. Network segregation and mental health

We explored associations between mental health variables and networks that related to drug use severity in the previous analyses. Pearson partial correlations (controlling for mean FD, age, ASI-drug) tested whether segregation of the association networks, sensorimotor networks, SN, and FPN were related to depression severity (n = 39), anxiety severity (n = 38), positive affect (n = 24), and negative affect (n = 24)³. Results showed mostly small and non-significant correlations (Table S1). However, SN segregation significantly and positively correlated with positive affect, r_partial(19) = .62, p = .0029 (16-correlation FDR-corrected α =.0031; Fig. 4).

Fig. 4.

Scatter plot of salience network segregation by positive affect. Note. The variables are adjusted by regressing out age and motion (mean framewise displacement). Positive affect = Positive affect subscale of the Positive Affect Negative Affect Schedule.

3.4.3. Network segregation and cognition

Pearson partial correlations (controlling for mean FD, age, ASI-drug) tested whether segregation of the association networks, sensorimotor networks, SN, and FPN were related to IQ (n = 40) and working memory (n = 26)³. Results showed no significant correlations between network segregation and these cognitive measures (Table S1).

4. Discussion

We examined brain network segregation in individuals with OUD seeking treatment with XR-NTX. Consistent with our hypothesis, drug use severity was related to lower network segregation in association networks, specifically the SN and FPN. Age predicted lower network segregation in association networks at a trending, non-significant level. In the sensorimotor networks, drug use severity was only related to lower network segregation in younger adults. Exploratory analyses revealed an association between positive affect and greater SN segregation. We found no relation between brain network segregation and cognition and no change in network segregation during XR-NTX treatment. Overall, our findings suggest that pre-treatment drug use severity and affect are related to a brain organizational structure that supports functional specialization and energetically efficient neural adaptation.

When examining specific association networks’ segregation from all other brain networks, drug use severity was only related to FPN segregation and SN segregation. In Fig. 3, the SN and FPN nodes are less densely packed (i.e., less segregated) in individuals with high drug use severity. Although we did not find an effect with DMN segregation, previous research has observed various alterations in DMN functional connectivity in addiction (Zhang and Volkow, 2019). Network segregation is only one aspect of brain architecture, such that the impact of substance use on DMN connectivity may be less robust in regard to segregation or limited to other metrics of network integrity In contrast to the task-negative DMN, the task-positive FPN activates during tasks to support cognitive functions (Schimmelpfennig et al., 2023). The SN helps switch between the FPN and DMN, which helps identify contextually relevant stimuli and integrate it to guide behavior (Menon, 2023; Menon and Uddin, 2010; Schimmelpfennig et al., 2023). The relation of drug use severity to lower FPN and SN segregation suggests consequences for cognitive performance and control over highly salient stimuli. Indeed, individuals with OUD have impaired executive functioning (Arias et al., 2016). As we did not find an association between FPN segregation and working memory, future studies should examine other executive functions. Alternatively, it is possible that OUD-related damage decoupled FPN segregation from executive function, which aligns with a study on alcohol use disorder (Wang et al., 2024). Future research should investigate whether changes in SN and FPN brain network segregation improves cognitive impairment in OUD.

The potential link between segregation of sensorimotor networks and drug use severity was more nuanced. We found that higher drug use severity was related to lower segregation of sensorimotor networks only in the younger participants (≤23 years old). This effect should be considered in context of the limitations of the smaller and mostly male subsample as well as susceptibility to potential outliers. Though preliminary, this result suggests that the brains of younger adults with OUD may be more vulnerable to drug-related changes in segregation of sensorimotor networks. We interpret this with caution, given the relatively mature development of sensorimotor networks connectivity by late adolescence (Zielinski et al., 2010). However, the relation between segregation of sensorimotor networks and drug use behavior may have been amplified in young adults because they have not yet reached full maturity in other brain areas that help control behavior, such as the prefrontal cortex (Arain et al., 2013). Previous studies have found that opioid use and craving are associated with sensorimotor network connectivity, regardless of age (Lichenstein et al., 2021; Yang et al., 2023). Sensorimotor networks integrate with top-down cognitive networks to prepare, implement, and regulate action (Gordon et al., 2023). As such, dysconnectivity in sensorimotor networks may contribute to addiction via increased cognitive dysfunction and impulsivity (Kebets et al., 2019) as well as enhanced automaticity of the motor and sensory aspects of drug use behaviors (Yalachkov et al., 2010). Considering our preliminary result, longitudinal samples are needed to explore the developmental trajectories of segregation of sensorimotor networks in people with OUD.

Exploratory analyses showed that higher network segregation of the SN was associated with greater self-reported positive affect. Given the role of the SN in detecting and integrating rewarding, engaging, and emotional stimuli (Menon, 2011), a relatively segregated SN may increase functional and energy efficiency SN to directly support positive affect through processing of positive environmental stimuli and internal states. Further, greater SN segregation may support emotion regulation skills (Fresco et al., 2017; Pinto et al., 2023), which would increase positive affect (Tugade and Fredrickson, 2007). One study of healthy young adults found a link between positive affect and lower SN within-network rsFC (Qi et al., 2021), inconsistent with our finding. However, network segregation is the balance of within versus between network connectivity, and people with OUD may have different neural underpinnings of positive affect. SN’s relation to positive affect but not negative affect is consistent with other work showing differential networks involved in positive and negative affect (Qi et al., 2021; Rohr et al., 2013). This also aligns with our expected non-significant correlation between positive and negative affect (Serafini et al., 2016; Watson et al., 1988).

In this first study to test network segregation response to treatment, we did not observe a significant change across three months of XR-NTX injections, despite a significant decrease in drug use severity. Network segregation may not be as malleable to XR-NTX as self-reported drug severity, which encompasses behavior and subjective ratings. Although XR-NTX alters functional brain response (e.g., reduced cue-reactivity) (Shi et al., 2018), interventions with profound effects on network connectivity and neural adaptability/flexibility may be more suitable for altering connectivity within and between brain networks. Non-invasive neurostimulation is one promising tool. For example, transcranial direct stimulation in people with methamphetamine use disorder modulated rsFC of the DMN, FPN, and SN, which was associated with reduced methamphetamine craving (Shahbabaie et al., 2018). Research suggests that stimulation deeper in the brain through low intensity focused ultrasound may alter network rsFC of subcortical regions involved in SUD (Chou et al., 2024; Kuhn et al., 2023). Further, the psychedelic substance psilocybin may open a window for flexibility in network connectivity such as in decreased network modularity associated with reduced depression (Daws et al., 2022). Future work could examine whether psilocybin-induced flexibility enables a complementary normalization of network segregation/integration in people with SUDs.

This study has limitations that constrain the generalizability of the results. The study sample was relatively small and from predominantly White and male individuals with heroin and prescription opioid use, reflecting the demographics and drug use prior to the broad introduction of synthetic opioids to the illicit drug supply in the United States (Casillas et al., 2024; Spencer et al., 2023). We lacked a healthy control sample and could not directly establish whether OUD generally relates to brain network segregation deficits and direct comparison to studies with different rs-fMRI scan times is inconclusive (Han et al., 2024). Nevertheless, mean brain network segregation in our OUD sample was significantly lower than the healthy control and alcohol use disorder groups in our previous study (Wang et al., 2024) (ps < .001; see Supplemental Results). The associations between brain network segregation and drug use were cross-sectional, and causality and direction are unknown. Datasets that measure rsFC prior to and during substance use will better describe changes in network segregation.

In conclusion, this first study of brain network segregation in OUD showed that participants with greater recent drug use severity had lower network segregation in cognitive networks, especially the FPN and SN. Inefficient executive functioning and salience processing may contribute to or be impacted by ongoing substance use and its consequences in people with OUD. Our preliminary evidence that segregation of sensorimotor networks was related to drug use severity only in young adults indicates the need for developmental studies. Our findings provide better understanding of reduced efficiency and premature aging of brain networks in OUD and encourage testing interventions to address brain network architecture in OUD.

Appendix A. Supporting information

Supplementary data associated with this article can be found in the online version at doi:10.1016/j.drugalcdep.2025.112863.