Compounding Vulnerability in the Neurocircuitry of Addiction: Longitudinal Functional Connectivity Changes in Alcohol Use Disorder

Abstract

Aims

The addiction neurocircuitry model describes the role of several brain circuits (drug reward, negative emotionality and craving/executive control) in alcohol use and subsequent development of alcohol use disorder (AUD). Human studies examining longitudinal change using resting-state functional magnetic resonance imaging (rs-fMRI) are needed to understand how functional changes to these circuits are caused by or contribute to continued AUD.

Methods

In order to characterize how intrinsic functional connectivity changes with sustained AUD, we analyzed rs-fMRI data from individuals with (n = 18; treatment seeking and non-treatment seeking) and without (n = 21) AUD collected on multiple visits as part of various research studies at the NIAAA intramural program from 2012 to 2020.

Results

Results of the seed correlation analysis showed that individuals with AUD had an increase in functional connectivity over time between emotionality and craving neurocircuits, and a decrease between executive control and reward networks. Post hoc investigations of AUD severity and alcohol consumption between scans revealed an additive effect of these AUD features in many of the circuits, such that more alcohol consumption or more severe AUD was associated with more pronounced changes to synchronicity.

Conclusions

These findings suggest an increased concordance of networks underlying emotionality and compulsions toward drinking while also a reduction in control network connectivity, consistent with the addiction neurocircuitry model. Further, they suggest a compounding effect of continued heavy drinking on these vulnerabilities in neurocircuitry. More longitudinal research is necessary to understand the trajectories of individuals with AUD not adequately represented in this study, as well as whether this can inform effective harm reduction strategies.

Short Summary: Individuals with AUD had an increase in functional connectivity over time between emotionality and craving neurocircuits, and a decrease between executive control and reward networks. These findings suggest an increased concordance of emotionality and compulsions toward drinking with a reduction in control over urges, consistent with the addiction neurocircuitry model.

INTRODUCTION

Alcohol use disorder (AUD), a substance use disorder characterized by compulsive alcohol consumption, affects nearly 15 million Americans per year (SAHMSA, 2019). The addiction neurocircuitry model (Koob and Volkow, 2016) suggests a neurobiological approach to understanding AUD. Three stages of substance use behaviors are posited to correspond to specific brain regions and functions: binge/intoxication (reward and incentive salience processes in basal ganglia), withdrawal/negative affect (negative emotionality and stress processes in amygdala and limbic regions) and preoccupation/anticipation (craving, impulsivity and executive function in prefrontal cortex and insula; see also Supplementary Table 1 for a summary of these circuits). The brain networks involved in these stages provide healthy reinforcement learning and stress response, but repeated voluntary drug use requires changes in brain structure and function to maintain neurotransmitter homeostasis. Following these changes, the absence of alcohol induces a stress response and less responsiveness to non-drug rewards, which impacts behavioral control and alcohol salience leading to compulsive drinking. This cycle is self-reinforcing, resulting in treatment-resistant substance use disorders.

Significant amount of the work supporting this model is preclinical or behavioral (Kwako et al., 2019), leaving much still to be understood about how brain networks contribute to the acquisition and maintenance of AUD in clinical populations. Evaluations of intrinsic functional connectivity using resting-state functional magnetic resonance imaging (rs-fMRI) have revealed atypical connectivity related to AUD in the reward, emotionality, salience and executive control networks of individuals with AUD (Camchong et al., 2013a; Peters et al., 2015; Zhu et al., 2017; Dean et al., 2020), which predicts poorer cognitive performance (Galandra et al., 2019). Rs-fMRI measures may be particularly important to understand AUD severity or risk of worsening disease (Fede et al., 2019). However, to our knowledge, no work has been done to investigate changes in functional connectivity of adult individuals with AUD using a longitudinal design.

There is some evidence of alcohol-related changes over time in the neurocircuitry thought to underlie AUD. Adolescent studies have demonstrated that some neural differences may precede the development of a substance use disorder through altered inhibitory, reward and working memory processes, but also that these features then worsen with the continued heavy use of substances (Squeglia and Gray, 2016). Further, studies focusing on brain structure indicate widespread changes (primarily decline) in gray matter volume and white matter integrity associated with AUD, often sensitive to volume of alcohol consumed (Kroenke et al., 2014; Pfefferbaum et al., 2014; Topiwala et al., 2017).

The current study seeks to understand how intrinsic functional connectivity changes as AUD progresses through the lens of the addiction neurocircuitry model. To do so, we examined changes in rs-fMRI measures of functional connectivity longitudinally between two timepoints as a function of AUD status. We hypothesized that observed changes would correspond to the addiction neurocircuitry model, in that AUD would be associated with worsening trajectories of change in synchronicity between reward (e.g. insula, basal ganglia), executive (e.g. prefrontal cortex, anterior cingulate (ACC)) and emotionality (e.g. extended amygdala) regions. These networks of regions were defined based on Koob and Volkow’s (2016) neurocircuitry of addiction. Moreover, we anticipated that more severe manifestations of AUD (i.e. more drinking) between the timepoints would exacerbate the observed effects along these pathways.

METHODS

Participants

Subjects (n = 39) completed functional neuroimaging procedures on two or more occasions as part of a variety of protocols at the National Institutes of Health Clinical Center through the National Institute on Alcohol Abuse and Alcoholism. Only individuals who had scans at least 21 days apart were included. For individuals with more than two scans, the scans with the longest interval in between were selected. Subjects were recruited from the community and included healthy control volunteers without AUD (HCVs; n = 21) and individuals with an AUD diagnosis (iAUDs; n = 18), most of whom were engaged in treatment at the NIAAA’s 30-day inpatient program. Patients were not actively withdrawing from alcohol at the time of scan. See Table 1 for demographic information, by AUD group. All research activities were conducted as approved by the NIH intramural Institutional Review Board consistent with the Declaration of Helsinki, including appropriate data sharing provisions between protocols. Protocol details are available in Supplementary Materials 1.

Table 1.

Sample description

(A)	HCV (n = 21)		iAUD (n = 18)
	mean	sd	mean	sd	pdiff
Age	37.86	9.2	44.17	10.47	0.0555
Years of education	17.29	4.55	14.61	2.7	0.0301
Interval days between scans	352.86	380.29	553.17	613.45	0.2398
Estimated interval drinks	98.81	233.46	3869.31	4992.09	0.0052
Lifetime drinks	1060.48	985.17	56704.48	59679.01	0.001
AUDIT	2.26	1.66	27.61	7.71	4.18E-11
(B)
Sex (% male)	66.67%		55.56%		0.7033
Treatment status (% inpatient)	N/A		94.44%*		N/A
Smoking status (% smokers)	0.00%		61.11%		1.08E-04
Current substance use disorders (%)
Cannabis	0.00%		16.67%		0.1780
Cocaine	0.00%		16.67%		0.1780
Hallucinogens	0.00%		5.56%		0.9377
Current other diagnoses (%)
Depressive disorders	9.52%		33.33%		0.1504
Anxiety/PTSD	4.76%		44.44%		0.0107
Eating disorders	4.76%		0.00%		1
Obsessive/compulsive disorders	4.76%		0.00%		1
ADHD	0.00%		5.56%		0.9377
Race (%)					0.1628
White	40.91%		61.11%		-
Black/African American	52.38%		27.78%		-
Asian/Pacific Islander	4.76%		0.00%		-
Multiracial	0.00%		11.11%		-
Unknown/not reported	4.76%		0.00%		-
Ethnicity (% hispanic or latino)	0.00%		5.56%		0.9377

Notes: (A) Mean and standard deviations (sd) reported for continuous variables within each group. p_diff refers to the significance of the t-test for group differences in those variables. (B) Percentage of sample for each category within each group for variables of sex, comorbid substance use and diagnoses, race/ethnicity and treatment/smoking status. p_diff refers to the significance of the chi-sq test for group differences in those variables; the chi-sq test for Race was tested as a 5 × 2 table. Abbreviations as follows: iAUD: individuals with Alcohol Use Disorder; PTSD: Post Traumatic Stress Disorder; ADHD: Attention Deficit/Hyperactivity Disorder. *One individual was an inpatient at the first visit and an outpatient at the second visit

Assessments

Sociodemographic and drinking behavior data were collected through a larger NIAAA screening protocol at baseline; per screening protocol procedures, these measures were repeated if significant time (e.g. 1 year) had passed since original baseline. Exact inclusion criteria for scanning differed by study but included age 21–60, no MRI contraindications and negative urine drug screen for drugs such as tetrahydrocannabinol, methamphetamine, cocaine and opiates. The Structured Clinical Interview for the Diagnostic Statistical Manual (DSM)-IV or DSM-5 (SCID) was used to identify AUD status (First et al., 1997; First et al., 2015); individuals with SCID-IV diagnoses of alcohol dependence or abuse were defined as having an AUD for the purpose of this analysis (Compton et al., 2013). Severity of AUD at baseline was quantified using the Alcohol Use Disorders Identification Test (AUDIT; Saunders et al., 1993). Alcohol consumption between scans was estimated using data from a combination of sources including lifetime drinking history (Skinner, 1979), Alcohol Timeline Followback (TLFB; Sobell and Sobell, 2000) and clinical intake interviews (see Supplementary Materials 2 for detailed estimation procedures). Participants were required to have a BrAC of 0.00 g/dl before being scanned.

Image acquisition and processing

Participants underwent scanning on a Siemens 3 T MRI machine. While in the scanner, they completed a 5 or 10 min long rs-fMRI scan and were advised to stay awake with eyes open during the duration of the scan. An echoplanar-imaging pulse sequence was used (TR: 2000 ms, TE: 30 ms, flip angle: 90°, FOV: 24 × 24 cm, 38 mm slice thickness, 36 slices, multislice mode: interleaved). A structural scan was also acquired at each timepoint for use in co-registration.

The rs-fMRI data were processed using CONN (version 17e), a Matlab-based toolbox for functional connectivity analysis (Whitfield-Gabrieli and Nieto-Castanon, 2012). After truncating (i.e. using AFNI sub-brik selection to select the first 5 min) longer rs-fMRI scans to match the 5 min scans available for some subjects, single-subject data were processed using CONN’s default preprocessing pipeline (see Supplementary Materials 3 for a detailed description). Five minutes is considered adequate to generate stable estimates of resting state connectivity (Van Dijk et al., 2010).

Statistical analysis

A whole brain seed-to-voxel analysis was conducted. Seeds were selected based on indication of their possible implication in AUD in previous literature and defined using the anatomical FSL Harvard-Oxford atlas used by default for segmentation during the CONN processing procedure: hippocampus, parahippocampal gyrus, amygdala, nucleus accumbens (NAcc), putamen, caudate, globus pallidus, thalamus, insula, superior frontal gyrus (SFG), frontal orbital gyrus and ACC. For each seed, the left and right structure was modeled separately (except for the ACC, which was modeled as a single, bilateral cluster). See Supplementary Table 1 for more detail on the correspondence of selected seeds with the neurocircuitry of addiction framework (Koob and Volkow, 2016).

Seed-to-voxel connectivity was calculated as Fisher-transformed bivariate correlation coefficients between the BOLD timeseries in the seed and in each of the other voxels in the whole-brain volume. This was calculated for each subject at each timepoint separately. CSF/gray/white matter signals and motion were included as covariates at the single subject level. At the group level, the model included timepoint, AUD status, age, years of education, sex and number of interim days between timepoints. Specifically, we examined the interaction between AUD status (AUD – HCV) and timepoint (Time 2 – Time 1), holding the other variables constant (i.e. covarying). Timepoint was modeled as a within-subject factor. Whole brain results were thresholded at a voxel level P < 0.001 and a cluster level P < 0.05 FDR corrected in order to control the false positive rate. All reported seeds survived an additional Benjamini–Hochberg correction for false discover rate inflation due to multiple seed testing (See Supplementary Materials 4 for a detailed description of the procedure used.) All tests were two-sided, and coordinates are reported in MNI space. We use the term ‘connectivity’ and ‘pathways’ to refer to the BOLD signal correlations identified through the seed-to-voxel analysis conducted in CONN. These should not be confused with anatomical connections, such as neurotransmitter pathways identified primarily through preclinical models.

In order to interpret the significant interactions, correlations between the seeds and significant clusters were extracted then plotted using ggplot2 in R (Wickham, 2009). Further, measures of AUD severity at baseline (AUDIT score), estimated interval drinks and number of interval days between scans were examined in association with the connectivity values by plotting and conducting Pearson’s correlations in an exploratory manner. This was done to aid interpretation of the primary analysis effects and to check our assumptions in defining this model (i.e. that interval time within group, rather than differences in group distributions of interval time, contributed to the group x timepoint effects; that sustained AUD symptomology, rather than diagnostic group membership, contributed to observed group x timepoint effects). Given the exploratory nature of these analyses, formal statistical testing of these correlations was not appropriate and interpretation of these exploratory results should be done with these caveats in mind. The strongest correlations (based on effect size) are highlighted as notable in the results section; all correlations are reported in Table 2.

Table 2.

Significant AUD Group * timepoint interaction in the seed-to-voxel analysis

Seed		Connectivity cluster							Mean (SD) change in connectivity		Correlation (r): Connectivity change by:
											interval days		estimated interval drinks	AUDIT	estimated drinks per day
Label	Lat.	Label	Lat.	x	y	z	k	p (size, FDR)	iAUD	HCV	iAUD	HCV	iAUD only
Parahippocampal Gyrus	R	vOFC	R	26	32	−10	118	0.014308	0.09 (0.2)	−0.16 (0.21)	−0.25	0.14	-0.20	-0.04	0.04
	L	PCC	R	12	−34	46	353	0.000003	0.12 (0.24)	−0.02 (0.27)	−0.08	−0.52	0.14	0.30	0.46
Nucleus Accumbens	R	Postcentral Gyrus	R	68	−14	12	359	0.000001	−0.2 (0.17)	0.15 (0.19)	0.12	0.12	-0.03	-0.35	-0.08
	L	ITG / Parahppocampal Gyrus	R	38	−16	−38	182	0.00466	0.13 (0.16)	−0.15 (0.16)	−0.09	−0.30	0.12	-0.07	0.58
Putamen	R	Postcentral Gyrus	R	66	−8	12	119	0.006406	−0.11 (0.17)	0.15 (0.23)	0.22	0.04	-0.06	-0.10	-0.05
Insula	R	Fusiform Gyrus / ITG	R	34	−14	−46	287	0.00002	0.15 (0.25)	−0.13 (0.25)	0.23	−0.33	0.27	-0.01	0.36
	L	Parahippocampal Gyrus / Fusiform Gyrus	R	32	−20	−28	744	<0.000001	0.12 (0.22)	0.04 (0.27)	0.41	−0.24	0.28	-0.25	0.25
Superior frontal gyrus	R	Fusiform Gyrus / Temporal Pole	L	−30	−8	−50	156	0.002686	0.15 (0.16)	−0.17 (0.23)	−0.08	−0.20	-0.04	0.18	0.06
Frontal orbital gyrus	L	Brainstem	-	0	−18	−42	117	0.018097	0.16 (0.2)	−0.09 (0.23)	−0.34	−0.40	-0.16	-0.06	0.00
ACC	-	Premotor Gyrus	L	−50	−2	24	187	0.001243	−0.14 (0.19)	0.13 (0.26)	−0.10	0.39	-0.25	0.04	-0.17
		Middle/Superior Frontal Gyrus	R	32	26	54	151	0.002664	−0.1 (0.19)	0.13 (0.29)	0.10	0.32	0.27	0.11	0.12
		Postcentral Gyrus	R	54	−2	38	131	0.00419	−0.15 (0.13)	0.21 (0.24)	−0.30	0.32	-0.46	-0.12	-0.36
		PCC / Visual Association	L	−26	−64	12	74	0.048457	−0.17 (0.21)	0.06 (0.2)	−0.39	0.11	-0.32	-0.08	-0.23

Notes: Statistics and descrptives for seed-to-voxel connectivity analyses; only pairs with significant group by timepoint interactions at the whole brain level are included. Abbreviations as follows: k: number of voxels in the cluster; Lat. (R/L): Laterality of brain region (Right/Left); FDR: false discovery rate correction; SD: standard deviation; iAUD: individuals with Alcohol Use Disorder.

RESULTS

Interaction between AUD status and timepoint on functional connectivity

There was a significant interaction effect (AUD group x timepoint) on functional connectivity with bilateral NAcc, parahippocampal gyrus and insula seeds; the left orbital gyrus seed; the right SFG seed and the ACC seed. See Table 2 for details about the clusters with which these seeds are connected and Fig. 1 for visual depictions of both the extent of the significant clusters and the interaction.

Fig. 1.

(A) AUD-related increases in connectivity over time. (B) AUD-related decreases in connectivity over time. For anterior cingulate connectivity: plots of seed connectivity not pictured are substantially similar to connectivity between the anterior cingulate and postcentral gyrus. See Supplementary Figure 1 for all plots. (A/B) For each seed region—(upper left panel) representation of significant seed-to-voxel connections. Axis labels provided for perspective reference as follows: S: superior, I: Inferior, A: Anterior, P: Posterior, R: Right, L: Left; (lower left panel) cluster of significant activation at the voxel-wise P < 0.001/cluster size P < 0.05 FDR corrected level. Cool colors represent negative connectivity, while warm colors represent positive connectivity. (Upper right panel) line graph of connectivity coefficients at each time point by each group. (Lower right panel) scatter plot of change in connectivity coefficients by interval days by each group. Linear line of best fit overlayed for interpretation. Multiple columns of line graphs represent multiple significant clusters, identified by number label (e.g. 1, 2, 3, 4).

Specifically, iAUDs had an increase in connectivity over time between the left NAcc and an inferior temporal gyrus (ITG) cluster, the right SFG and a fusiform gyrus cluster, the left frontal orbital gyrus and a brainstem cluster, the right parahippocampal gyrus and a ventral OFC cluster, the bilateral insula and clusters in the right fusiform/parahippocampal gyrus and the left parahippocampal gyrus and a posterior cingulate cortex (PCC) cluster. Except the left parahippocampal gyrus and left insula paths, HCVs had a decrease over time in connectivity between these regions.

On the other hand, iAUDs had a decrease in connectivity over time (where HCVs had an increase in connectivity over time) between the right putamen and NAcc seeds and a postcentral gyrus cluster as well as between the ACC and clusters in the middle/SFG, premotor gyrus, postcentral gyrus and PCC/visual association area.

Post hoc analysis: associations between connectivity change and interval characteristics

Interval length

For the significant connections identified above, interim time between scans was associated with more pronounced differences in connectivity change between iAUDs and HCVs most notably in the left parahippocampal-PCC path; the insula-fusiform/parahippocampal paths; and the ACC–PCC, premotor and postcentral gyrus paths. On the other hand, interval time attenuated the differences in connectivity change between groups in the right parahippocampal-OFC path. See Table 2 for a complete list of correlations between connectivity change and interim time. See Fig. 1 for scatter plots corresponding to the association between connectivity change and interim time.

Alcohol measures (iAUDs only)

For the significant connections identified above, estimated interval drinks was most notably associated with greater increases in connectivity in the left posterior-fusiform/parahippocampal gyrus path, and greater decreases in connectivity in the paths between the ACC seed and PCC/postcentral clusters. It was also associated with an attenuation of the observed AUD group effect on connectivity between the ACC seed and the middle/SFG cluster. This pattern remained consistent when estimated drinks per interval day were used instead of total estimated interval drinks. Also notable, estimated drinks per interval day were associated with increased connectivity between timepoints in the left parahippocampal-PCC and left NAcc-ITG paths.

Baseline AUDIT was most notably associated with increased connectivity over time in the left parahippocampal-PCC path along with decreased connectivity over time in the right NAcc-postcentral gyrus path. Baseline AUDIT was also associated with attenuation of the AUD group effects along the left insula—parahippocampal gyrus path.

See Table 2 for a complete list of correlations between connectivity change and drinking measures. See Supplementary Fig. 2 for scatter plots corresponding to the association between connectivity change and drinking measures for each pathway.

DISCUSSION

In order to characterize how intrinsic functional connectivity changes with sustained AUD, we analyzed rs-fMRI data from individuals with and without AUD collected on multiple visits as part of various research studies at the NIAAA intramural program from 2012 to 2020. Further we explored how these changes were sensitive to the amount of time between scans and the estimated amount of alcohol consumed during the interval. We found that over time, iAUDs had increased functional connectivity between negative affect and reward/incentive salience networks but decreased synchronicity between executive function and reward networks. These findings suggest iAUDs have an increased concordance of emotionality and compulsions toward drinking accompanied by a reduction in control over those urges. Moreover, in many of these circuits, more severe AUD and more drinking led to more atypical patterns of connectivity, suggesting additive harm with continuing heavy drinking.

Increased connectivity in affective pathways

Across seed regions, iAUDs had increased connectivity (as opposed to the decrease observed in HCVs) along paths between each seed (frontal, left accumbens and insula) and clusters in a temporal area encompassing portions of the ITG, fusiform gyrus and parahippocampal gyrus. Lesion studies suggest a critical role of this temporal region in empathy and emotion regulation (Rankin et al., 2006; Kumfor and Piguet, 2012) and related processes such as sharing negative emotion (Sung et al., 2018), though also in associative memory recall (Sperling et al., 2003) and when considering self-confidence (Morales et al., 2018).

Although this cluster is not specifically hypothesized within the addiction neurocircuitry model (Koob and Volkow, 2016), that framework may be a useful lens for interpretation of these results. As outlined in the introduction, the three circuits posited by that model reinforce substance use behaviors, contributing to a cycle of addiction. The frontal and insula seeds we defined here are part of the preoccupation/anticipation circuit underlying impulsivity and craving associated with addictions, while the NAcc plays a central role in the binge/intoxication circuit leading to use through reward and incentive salience processes. Functionally, the extended ITG/parahippocampal region could be considered part of the withdrawal/negative affect circuit based both on its function, described in the preceding paragraph, and on its previous implication in functional networks associated with emotional experience (Raz et al., 2016) and social connection (Bickart et al., 2012). Thus, these results indicate that with continued progression of AUD, there is an increased coupling of negative affect with reward seeking and impulsive behaviors.

The association between negative emotionality and drinking has been observed in college-aged social drinkers (Armeli et al., 2010), and may play an important role in the development of AUD. In individuals who drink to cope, negative emotionality actually induces alcohol motivation (Ostafin and Brooks, 2011) and alcohol consumption (Dvorak et al., 2014); drinking to cope is a pathway by which young adults develop AUD (Kenney et al., 2018). Evidence that the connection between negative emotionality, reward processes and cognition increases as problematic alcohol consumption continues suggests a persistent and worsening vulnerability that might contribute to more severe alcohol-related problems.

Decreased connectivity in behavioral control pathways

ACC and right basal ganglia connectivity with a cluster in the right postcentral gyrus was decreased in individuals with AUD. This postcentral gyrus, extending into premotor areas, has been implicated in a variety of motor functions including speech production and control (Barrett et al., 2004; Price et al., 2006; Tourville et al., 2008), response action (Liddle et al., 2001), effortful motor coordination (Goble et al., 2010) and intentional eye blinking (Kato and Miyauchi, 2003). The ACC is classically involved in conflict detection and cognitive control (Botvinick et al., 2001), as well as in emotion control (Bush et al., 2000) and reward-based decision making (Bush et al., 2002).

It is possible that our ACC results are driven by salience rather than executive network involvement, given that our ACC seed contained both ventral and dorsal regions. Evidence suggests that the dorsal ACC can be better characterized as part of a salience network (Seeley et al., 2007). However, both salience and cognitive aspects of ACC functionality are part of the preoccupation/anticipation circuit of the addiction neurocircuitry model. Thus, either way these decreases in connectivity over time observed in iAUDs are best characterized as desynchronization of the preoccupation/anticipation and binge/intoxication neurocircuits.

Taken together, these results represent a pattern of decreasing synchronization between top-down control processes and right lateralized salience/reward regions, hypothesized by the addiction neurocircuitry model as highly influential in the development and maintenance of addiction (Koob and Volkow, 2016). Our results then can be interpreted as a greater reduction in control over alcohol seeking behaviors; these reductions have been shown to be associated with more craving and relapse (Volkow et al., 1999; Kalivas, 2009; Wang et al., 2012). A more nuanced interpretation may indicate a shift from voluntary, initial drug use, in which NAcc-prefrontal dopaminergic and prefrontal-NAcc glutamatergic projections underly reinforcement learning (Everitt et al., 2008), relying on intact executive-reward structures, to compulsive, drug seeking characterized by drug-cued hyperactivity and drug-absent hypoactivation in these glutamatergic pathways.

Dichotomous versus additive impact of problematic alcohol consumption

Much of the model discussed above posits a dichotomous (casual drinking vs. AUD) ‘shift’ (sometimes characterized as a ‘hijacking’) of these three circuits in addiction (Koob and Volkow, 2010; Hall et al., 2015). In the healthy state, voluntary drinking engages intact circuits of reward, reinforcement and relief. In the compulsive drinking/addicted state, neurotoxic and homeostatic changes in these networks lead to higher thresholds for reward signaling, automatized alcohol seeking and withdrawal symptoms. To explore whether our results better reflect this dichotomous ‘shift’ in neurocircuitry or the alternative hypothesis, a gradual and additive change in the functioning of these systems with increased AUD severity, we qualitatively examined the association between connectivity changes observed in our main analysis and continuous measures of drinking volume and AUD severity.

More severe alcohol use at baseline was associated with attenuation of observed AUD related change in left insula connectivity between the timepoints. Although preliminary, this suggests that individuals who already had severe AUD at the first timepoint had few further changes, while individuals closer to the threshold between so-called ‘healthy’ and ‘addicted’ neurocircuitry states had that ‘shift’ during the interim time, as their progression of AUD continued. Mechanistically, the insula’s role in addiction is the representation of interoceptive effects of drug use, where afferent pathways carry body state information associated with substance use from the brainstem to the posterior insula, then subjectively re-represents it in the anterior insula in communication within the limbic system (Craig, 2002), at which point it can be recalled to trigger what has been called ‘conscious craving’ (Naqvi and Bechara, 2009). Preclinical work demonstrates a dichotomous and causal link between the introspective insula and drug seeking behavior; temporary disruption of the insula led amphetamine-place preference in conditioned animals to turn ‘off’, and when disruption ceased, back ‘on’ (Contreras et al., 2007); a similar effect has been seen in lesion studies of smokers (Naqvi et al., 2007).

Though our findings are consistent with this binary (presence or absence) of an internal representation of the subjective effects of alcohol, and thus supportive of the binary group model we used in our primary analysis, we should note that other studies provide contradictory evidence. For example, Kosuke et al. (2020) found that hyper-connectivity between the insula and default mode network associated with gambling disorder was positively correlated with duration of the disease, rather than simply presence or absence. Notably, the referenced study was cross-sectional; it is possible that with continued follow-up, the researchers would have observed the lack of progressive neural change seen in individuals in our sample.

Other neurocircuitry impacted by AUD over time may reflect sensitivity to degree of alcohol use severity, inconsistent with a dichotomous shift between healthy and disordered states. In particular, we saw patterns possibly indicating that more severe baseline AUDIT and more daily alcohol consumption in the interim was associated with increased connectivity between the parahippocampal gyrus and the PCC; decreased connectivity in ACC paths was also related to alcohol severity measures. Other research has suggested that changes in brain function are sensitive to severity of disease. For example, in a previous single-timepoint study that partially overlapped with the sample described here, we demonstrated that functional connectivity features are predictive of alcohol use severity within an AUD sample (Fede et al., 2019); other groups have also used resting state connectivity to predict early relapse within an AUD sample (Camchong et al., 2013b; Blaine et al., 2020; Muller and Meyerhoff, 2020) and across substances (Moeller and Paulus, 2018; Yip et al., 2020). Though this pattern reflects vulnerability to continued drinking, it also identifies potentially valuable targets for harm reduction-based treatment approaches; recent work demonstrates that tDCS treatment alters global resting connectivity corresponding to reductions in relapse (Holla et al., 2020).

Some of our post hoc analysis results were unexpected considering the previous literature, suggesting the need for alternative hypotheses in future studies on this topic. We saw patterns where increases in left NAcc-ITG connectivity were more pronounced in individuals with higher interim drinking rates and that right NAcc-brainstem connectivity was more desynchronized in individuals with more severe AUD at baseline. Previous work demonstrates a ventral to dorsal striatum shift in cue reactivity in compulsive drinkers (Vollstädt-Klein et al., 2010). Though there was an effect of AUD group on right putamen connectivity, there was no observed association with magnitude of AUD progression. That being said, our study was conducted in the absence of drug-cues, and although rs-fMRI measures of intrinsic functional connectivity can suggest functional differences, they can also correspond to structure changes (e.g. alcohol-related changes to white matter integrity, which have been demonstrated longitudinally in AUD; 13).

Limitations

The results reported here are not from studies designed to be longitudinal. Though we have attempted to control for the resulting confounds, there are still some potential concerns. A large portion of our AUD sample participated in our inpatient treatment program, relapsed at some point after completion, and were readmitted. This pattern of relapsing and return to treatment informed our interpretation of the results as a progression of AUD (i.e. continued neurobiological damage caused by heavy chronic drinking). Therefore, the observed changes likely reflect trajectories of additional impact of chronic AUD. Patients who remained abstinent or those who relapsed but did not return for treatment were not represented, so we cannot rule out the possibility that our results apply only to this subset of iAUDs. Indeed, our results should not be applied to individuals who have sustained recovery from AUD; this trajectory was not evaluated in our sample.

Moreover, our measure of drinking between timepoints is only an estimate. Portions of the interim may have been spent in abstinence or in treatment programs, though this was accounted for when estimating interval drinking. The iAUD and HCV groups did not significantly differ on the average number of interval days between scans, age or sex, but the groups did differ on years of education. All of these variables were included in the CONN analysis as covariates to control for these factors to the extent it is statistically possible. Many iAUDs were smokers or had comorbid substance use disorders that may have contributed to the findings; however, since these were absent from the HCV group, it was not appropriate to include these as covariates. Similarly, iAUDs had higher rates of anxiety. See Table 1B for comorbidity rates. In general, the small to moderate sample size limited our power to explore all the variations in patient trajectories (length of abstinence, number of previous detoxes), multiple timepoints, methodology and group differences. This is particularly true in that we did not have the ability to split our data into two sets to conduct such exploratory secondary analyses in an independent sample. Even the limited exploratory correlations we presented to interpret our main results may present a biased view of the true associations with brain connectivity (see Kriegeskorte et al., 2009 for a discussion of the impacts of double-dipping), and we reiterate again the importance that they be interpreted cautiously and solely as context for the main analysis.

As scans were collected across several protocols, scanner attributes varied. All subjects completed an eyes-open resting state fMRI EPI protocol with identical parameters on a Siemens 3 T scanner. However, some scans were collected on a Skyra using a 20-channel head coil, while some were collected on a Prisma using a 32-channel coil. There was no significant association between scanner types and connectivity or group membership. Marginal findings are reported in Supplementary Materials 5.

Amount of alcohol consumption between scans was conceptualized as a measure of AUD progression, based on findings that greater volume of alcohol consumption is estimated to result in more disease burden, deaths and early disability, particularly in marginalized communities (Rehm et al., 2009). The word ‘progression’ does not mean that harm was not reduced compared with hypothetic harm that could have occurred at pre-treatment levels of drinking. Additionally, although the addiction neurocircuitry model we reference uses clinical and function-based labels (e.g. withdrawal), our study was not designed to evaluate the clinical correspondence between the model and behaviors (see Kwako et al., 2016 for a description of what such a study would need to entail.) Instead, we are functional connectivity during resting state, thought to reflect the intrinsic nature of these network connects outside of task demands.

Finally, there are limitations associated with seed correlation analyses such as the one reported here. The seeds used here were from the FSL Harvard-Oxford atlas, which is anatomically rather than functionally defined (Jenkinson et al., 2012). Many of the seeds are quite large and may be functionally segmentable into several subregions belonging to dissociable resting state networks (Doucet et al., 2019). It is possible that a study using similarly labeled seeds based on other atlases would not replicate our results, or that different functional segmentation approaches would suggest alternative interpretations. In order to minimize results driven by between-atlas variability here, we have used only seeds from the FSL Harvard-Oxford atlas.

CONCLUSIONS

We report observations of distinct patterns of change in connectivity over time in individuals with AUD and HCVs. In particular, this work illustrates the way in which the cycle of addiction spirals into worsening disorder through altered network connectivity with the reward/incentive salience network, which becomes more tightly coupled with emotionality circuits but desynchronized with control functions. It may also provide initial evidence highlighting not only a dichotomous shift toward conscious craving, but an additive impact of alcohol use in much of the circuitry underlying AUD, where individuals with more severe AUD continue to be most vulnerable to increased alteration of functional network connectivity. However, these circuits may also be the most receptive to harm reduction strategies of reduced drinking. Formal longitudinal studies and experimental designs evaluating neural changes in connectivity following treatment are needed to better understand addiction neurocircuitry in clinical AUD.

FUNDING

Research funded by National Institute on Alcohol Abuse and Alcoholism (NIAAA; ZIAAA000126; Intramural, PI: Momenan).

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to declare.