Intra‐striatal dopaminergic inter‐subject covariance in social drinkers and non‐treatment‐seeking alcohol use disorder participants Authors: Evgeny J Chumin, Mario Dzemidzic, Karmen K Yoder Published: 2024 ABSTRACT AbstractOne of the neurobiological correlates of alcohol use disorder (AUD) is the disruption of striatal dopaminergic function. Although regional differences in dopamine (DA) tone/function have been well studied, interregional relationships (represented as inter‐subject covariance) have not been investigated and may offer a novel avenue for understanding DA tone. Positron emission tomography (PET) data with [11C]raclopride in 22 social drinking controls and 17 AUD participants were used to generate group‐level striatal covariance (partial Pearson correlation) networks, which were compared edgewise as well as on global network metrics and community structure. An exploratory analysis examined the impact of tobacco cigarette use status. Striatal covariance was validated in an independent publicly available [18F]fallypride PET sample of healthy volunteers. Striatal covariance of control participants from both data sets showed a clear bipartition of the network into two distinct communities, one in the anterior and another in the posterior striatum. This organization was disrupted in the AUD participants' network, which showed significantly lower network metrics compared with the control participants' network. Stratification by cigarette use suggests differential consequences on group covariance networks. This work demonstrates that network neuroscience can quantify group differences in striatal DA and that its interregional interactions offer new insight into the consequences of AUD.Keywords: covariance networks, dopamine, network neuroscience, PET, striatum INTRODUCTION Disruption of striatal function and motivational circuitry is thought to underlie substance abuse and addiction, and dopamine (DA)‐related alterations are believed to play a primary role. 1 Despite being the most consumed substance of abuse, neurobiological mechanisms underpinning the transition to maladaptive alcohol use are not yet fully understood. 2 , 3 , 4 Human neuroimaging studies of dopaminergic (DAergic) tone (i.e. baseline measurement) and function (changes in DA due to task or pharmacological manipulation) employing positron emission tomography (PET) have reported alterations in the DAergic system in various alcohol use disorder (AUD) populations. 5 , 6 , 7 , 8 It has been proposed that AUD is characterized by a reduction in DAergic function (see, however, Yoder et al. 8 ). Yet, we know little about whether regional interrelationships within the striatal DA system contribute to AUD. The striatum is structurally interconnected with all major cortical structures, forming a topographically organized circuit loop that involves regulation of motor, sensory and cognitive processing (e.g. motivation, reward, action planning). 9 , 10 It is commonly subdivided into the caudate, putamen and ventral striatum (composed of the nucleus accumbens and ventral portions of anterior caudate and putamen). 11 DAergic inputs to the striatum originate from the substantia nigra pars compacta and ventral tegmental area in the midbrain. Previous work has shown that the striatum, although composed of cytoarchitecturally overlapping areas of structural input, can be parcellated into spatially distinct regions of interest based on structural 12 as well as functional connectivity. 13 In a prior study, Tziortzi et al. 12 showed that the homogeneity of DA D2 receptor availability (binding potential [BP]) was greater within the connectivity‐based parcellation relative to commonly used anatomical divisions. Therefore, if regional connectivity is influencing the observed PET DA BP, regional DAergic function within the striatum is likely interrelated and could be investigated from a network perspective. Unlike previous approaches that mostly focused on isolated brain areas, network neuroscience aims to study the brain as a system of interacting elements where each brain region is referred to as a node and edges describe the connectivity between them. Edges can be quantified as absence/presence (e.g. binary structural connectivity) or by assessing their strength (i.e. weight), which could be streamline density for structural connectivity or a similarity measure such as correlation for functional connectivity. 14 Additionally, given a homogeneous sample/group of participants, it is also informative to compute the similarity of inter‐subject covariance. 15 Group differences in inter‐subject covariance have been assessed for grey matter, 16 , 17 metabolic activity ([18F]fluorodeoxyglucose PET) 18 , 19 and amyloid and tau PET levels in Alzheimer's disease. 20 , 21 To our knowledge, there has only been one study on DAergic covariance, which investigated the extra‐striatal BP of [11C]FLB‐457 PET in Parkinson's disease. 22 In this retrospective analysis, [11C]raclopride PET data were utilized to generate group‐level striatal covariance networks in social drinking controls (CON) and community‐recruited, currently drinking individuals with AUD. Networks were then compared using graph theory metrics of density, degree, strength, clustering and modularity. 14 We hypothesized that the covariance structure of the AUD group network would be altered as a function of chronic alcohol misuse. Additional exploratory comparisons were carried out to ascertain the potential impact of comorbid cigarette use on DA covariance in the striatum, as it has been shown that cigarette use alters DA function. 23 , 24 , 25 Finally, we used an open data set of healthy controls with [18F]fallypride (FAL) DA PET imaging 26 to replicate the observed striatal covariance structure and ensure that it was not specific to a PET tracer. METHODS This study performed a retrospective analysis of National Institutes of Health (NIH)‐funded data (R01AA018354) collected to study alterations in DAergic function in AUD. Magnetic resonance imaging (MRI) data from these participants have been previously published. 27 , 28 , 29 All participants provided written informed consent for all study procedures that were approved by the Indiana University Institutional Review Board in accordance with the Belmont Report. Participants were recruited via advertisement in local media, were 21–55 years old and were able to read, understand and complete all study procedures in English. Exclusion criteria were history or presence of any psychiatric, neurological or other medical disorder; current use of any psychotropic medication; positive urine pregnancy test; positive urine toxicology screen for illicit substances at interview; and contraindications for the MRI. The Semi‐Structured Assessment for the Genetics of Alcoholism was administered to confirm the presence or absence of AUD. 30 Participants also completed the following questionnaires: a medical history and demographics questionnaire, the 90‐day Timeline Follow‐Back for Alcohol Use (TLFB), 31 Alcohol Dependence Scale (ADS), Fagerstrom Test for Nicotine Dependence, 32 Edinburgh Handedness Inventory, 33 AUD Identification Test (AUDIT) 34 and an internally developed substance use questionnaire. Demographic data for 39 participants are summarized in Table 1. Education, biological sex and smoking variables differed between AUD and CON groups, and as expected, AUD scored significantly higher on the AUDIT. Six CON participants tested positive for illicit substances on scan day (five marijuana, one benzodiazepine) as well as one AUD participant (marijuana) and were included in the study due to rarity of endorsed use. TABLE 1Demographic data.Group (n)Age (years)Education* (years)Sex* (M/F)Handedness (R/L/RL)Race (W/AA/other)Smoking* FagerstromAUDIT* Social drinking controls (CON)2239.9 ± 10.815.8 ± 2.010/1217/4/111/9/233 ± 06.7 ± 3.6Non‐treatment‐seeking alcohol use disorder (AUD)1743.0 ± 10.113.5 ± 2.214/316/1/010/7/0105 ± 1.715.2 ± 5.2 Note: Data are shown as mean ± standard deviation or number of participants.Abbreviations: AUDIT, Alcohol Use Disorder Identification Test; F, female. M, male. R, right. L, left. RL, ambidextrous. W, White. AA, African American.*Significant group differences at p < 0.05, using a two‐sample t‐test or χ 2 test as appropriate. Demographic data. Note: Data are shown as mean ± standard deviation or number of participants. Abbreviations: AUDIT, Alcohol Use Disorder Identification Test; F, female. M, male. R, right. L, left. RL, ambidextrous. W, White. AA, African American. Significant group differences at p < 0.05, using a two‐sample t‐test or χ 2 test as appropriate. Imaging Structural MRI was performed using a Siemens 3T SKYRA (10 controls and 8 AUD) and a Siemens 3T PRISMA (12 controls and 9 AUD), with a high‐resolution T1‐weighted anatomical image acquired using a whole‐brain magnetization‐prepared rapid gradient echo (MPRAGE) sequence (1.05 × 1.05 × 1.2 mm3 voxels; 176 sagittal slices). PET data were acquired with a Siemens Biograph mCT scanner (Siemens, Erlangen, Germany). PET scans were initiated with a single bolus injection of 14.35 ± 0.92 mCi of the DA D2/D3 antagonist [11C]raclopride (RAC). A computed tomography (CT) scan was acquired for attenuation correction. Dynamic acquisitions were 50 min in length, where the first 10 frames were 30 s long, whereas the remaining 45 frames were 60 s long. Data were reconstructed with Siemens software using the Filtered Backprojection algorithm (5‐mm Hanning filter) into a 2‐mm3 isotropic resolution, with corrections for random coincidences, attenuation and scattering. Structural image processing Structural T1‐weighted MRI data for each participant were first denoised, 35 bias field corrected and tissue type segmented (fsl_anat, FSL v6.0 36 ) and then skull stripped with Advanced Normalization Tools (ANTs). 37 A combination of linear (flirt 6 then 12 degrees of freedom) and non‐linear (fnirt) transformations in FSL was used to register T1 volumes to a standard space (Montreal Neurological Institute template), with the inverse transformations applied to align the Melbourne subcortical atlas Scale II parcellation 13 to each participant's T1 image. 36 PET preprocessing RAC PET data preprocessing was done in the Statistical Parametric Mapping version 12 (SPM12) software (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/). For each participant, the first 14 frames (~10 min) were averaged to create an early mean PET image that was used as a target for realignment of individual frames to help correct for head motion. The mean image was then coregistered to the structural MRI with the transformation applied to dynamic data. Subcortical regions of interest from the Tian et al. 13 parcellation (left and right anterior and posterior caudate and putamen and nucleus accumbens shell and core) were used as nodes in the inter‐subject covariance network generation. An additional cerebellar grey matter (with the vermis excluded) reference region (with little or no D2/D3 binding) was defined in the native T1 space of each participant. Regional time activity curves for the striatal subregions and cerebellum were generated in Matlab (R2022b). DA tone was defined as RAC D2/D3 receptor availability (binding potential [BPND]; relative to nondisplaceable binding). 38 Regional BPND was estimated with the multilinear reference tissue model (MRTM). 39 Network construction Inter‐subject covariance networks for each group consisted of the 12 striatal nodes described above with edges quantified as partial Pearson correlation coefficients, adjusting for the age, biological sex and smoking status of participants within group as these factors are known to influence DAergic system function. 24 , 40 , 41 , 42 , 43 , 44 This approach generates a signed fully connected adjacency matrix. For exploratory analysis of cigarette use, networks were created with partial Pearson correlation (age and sex adjusted) from three groups: non‐smoking CON (nsCON; n = 19), non‐smoking AUD (nsAUD; n = 7) and smoking AUD (smAUD; n = 10). Additionally, sparse networks were computed by identifying the edges of significant covariance via permutation testing (10 000 permutations) where a null distribution of networks was generated by randomly and independently scrambling the regional BPND values of each participant and recomputing the covariance matrix. This procedure destroys any existing regional inter‐subject relationships. A permutation‐based p < 0.05 value was then used to identify significant edges. Stricter significance thresholds resulted in fragmented networks in the AUD group, in which one or multiple regions became disconnected from the rest. Network fragmentation prevents direct group comparisons and limits interpretability; therefore, only the p < 0.05 thresholded networks are reported. To assess whether the covariance structure was specific to BPND, the tissue‐to‐plasma rate constant k 2 and tracer delivery R1 parameters from MRTM as well as region size (in voxels) were also used to generate group covariance networks. Fallypride validation data set Data from 25 healthy young adults (12 male and 13 female; age 18–24) as described by Castrellon et al. 26 were downloaded from OpenNeuro. 45 This data set was identified from a list of openly available PET data provided by Sala et al. 46 The provided standard space parametric images of simplified reference‐tissue‐model‐derived BPND were used to extract regional averages for the striatal nodes described above. Networks were constructed with partial Pearson correlation, with age and education as covariates. Global and nodal metrics The following measures were computed from group covariance networks as described in the Brain Connectivity Toolbox 14 : density, number of connections (thresholded networks only); node degree, number of connections for each region (thresholded networks only); positive and negative strength, sum of edge weights (i.e. positive or negative correlation coefficients) for each region (summing over its connections) and for the whole network (summing all edges) (unthresholded and thresholded networks); and clustering coefficient, a measure of the tendency of nodes in a network to cluster (using the weighted and signed variant 47 ) (unthresholded and thresholded networks). Community structure was estimated from each network with the multiresolution consensus clustering (MRCC) method, 48 with 10 000 initial partitions sweeping the resolution parameter space. A co‐assignment matrix (probability that any region pair is assigned to the community) was then computed, followed by application of a null model for identification of a single consensus partition (here, the alpha threshold for the null model was set at 0.05). For within‐community (a subset of regions grouped together by the MRCC algorithm) averaged BPND comparisons, the CON group consensus partition was a reference partition and applied to all groups. Statistical analysis Statistical comparisons were performed in Matlab R2022b 49 . Demographic, regional BPND and community‐averaged BPND comparisons were assessed via independent‐samples t‐tests or χ 2 tests as appropriate. Distributions of nodal network values were compared via a two‐sample Kolmogorov–Smirnov (KS) test. The three‐group exploratory analysis applied a one‐way analysis of variance (ANOVA) with Tukey–Kramer post hoc tests. Code availability Image processing and covariance network analysis scripts have been made publicly available: structural T1‐weighted image preprocessing (https://github.com/IUSCA/IUSM-ConnPipe), RAC PET processing (https://github.com/echumin/PET_processing_Code) and covariance network analysis (https://github.com/echumin/CovNet). RESULTS We first assessed whether any regional differences in BPND existed that may influence network covariance. Among all striatal regions/nodes, only the left anterior caudate BPND was lower in AUD than in CON at uncorrected significance that did not survive false discovery rate correction (Table 2; independent‐samples t‐test, p = 0.009, uncorrected, q FDR > 0.05). TABLE 2Regional size and binding potentials.Binding potentialRegion size (voxels)CONAUDCONAUDLeftNAcShell2.07 ± 0.341.92 ± 0.5474.9 ± 13.5566.18 ± 13.31Core2.28 ± 0.441.99 ± 0.5464 ± 8.3361.70 ± 14.73PUTAnt2.63 ± 0.372.59 ± 0.29320.27 ± 32.27325.18 ± 64.42Pos2.54 ± 0.322.48 ± 0.24380.50 ± 37.40405.94 ± 63.74CAUAnt2.22 ± 0.311.93 ± 0.35* 201.77 ± 29.16223.71 ± 47.73Pos1.78 ± 0.361.62 ± 0.31142.23 ± 17.62146.12 ± 30.34RightNAcShell1.90 ± 0.342.07 ± 0.3693.50 ± 16.2284.41 ± 21.79Core2.06 ± 0.452.14 ± 0.3870 ± 7.2867 ± 18.90PUTAnt2.72 ± 0.362.64 ± 0.30290.09 ± 29.65306.12 ± 65.70Pos2.55 ± 0.292.43 ± 0.24304.41 ± 33.68316.06 ± 59.01CAUAnt2.06 ± 0.372.01 ± 0.31228.04 ± 28.58239.94 ± 41.41Pos1.69 ± 0.361.57 ± 0.32147.95 ± 22.48148.47 ± 29.17 Note: Mean ± standard deviation of binding potential and region size values (in voxels from 2‐mm isotropic resolution volumes) for the 12 striatal regions of interest from the Scale II Melbourne subcortical atlas. 13 Abbreviations: Ant, anterior; AUD, alcohol use disorder; CAU, caudate; CON, controls; NAc, nucleus accumbens; PUT, putamen; Pos, posterior.*Group differences at p < 0.05, uncorrected independent‐samples t‐test. Regional size and binding potentials. Note: Mean ± standard deviation of binding potential and region size values (in voxels from 2‐mm isotropic resolution volumes) for the 12 striatal regions of interest from the Scale II Melbourne subcortical atlas. 13 Abbreviations: Ant, anterior; AUD, alcohol use disorder; CAU, caudate; CON, controls; NAc, nucleus accumbens; PUT, putamen; Pos, posterior. Group differences at p < 0.05, uncorrected independent‐samples t‐test. Within‐participant Z‐scored RAC BPND values for each node are shown in Figure 1A,D for CON and AUD, respectively. The covariance network of the CON group showed a clear pattern of similarity among bilateral nucleus accumbens (shell and core) and anterior putamen that was anticorrelated with a separate set of regions composed of bilateral caudate (anterior and posterior) and posterior putamen (Figure 1B). This pattern was less evident in the AUD group network (Figure 1E). Edge weight distributions between the two group networks showed no differences (KS = 0.18, p = 0.2). Permutation significance‐based thresholding of correlations (edge weights) revealed a greater number of edges in CON relative to AUD (Figure 1C,F; network densities 0.73 and 0.42 for CON and AUD groups, respectively). To assess whether the observed covariance structure was specific to BPND, we also computed networks of MRTM flow constants k 2 and R1 as well as region size. These MRTM flow constant networks did not show the organization seen with BPND (Figure S1; no group differences were found for k 2 or R1 for any region), whereas volumetric covariance showed a pattern distinct from BPND covariance, displaying positive covariance among bilateral nucleus accumbens and putamen and a separate group composed of bilateral caudate regions (Figure S2). Inter‐subject group covariance of striatal [11C]raclopride binding potential. Subject series of RAC BPND for striatal regions of interest are shown for (A) control and (D) alcohol use disorder (AUD) participants. Covariance matrices (B, E) were computed as pairwise partial Pearson correlation and adjusted for age, biological sex and cigarette‐smoking status. Permutation testing identified edges of significant covariance (C, F) within each group (p < 0.05). Density values show fraction of non‐zero edges in the network. Ant, anterior; Pos, posterior; NAc, nucleus accumbens; PUT, putamen; CAU, caudate; CON, controls. Elementwise (edge‐level) group comparisons of the two networks identified three edges only at uncorrected permutation p < 0.05 significance (q FDR > 0.05): (1) left nucleus accumbens shell and anterior caudate, (2) left nucleus accumbens core and right nucleus accumbens shell and (3) left and right nucleus accumbens core covariances (Figure 2A). In addition, total positive and negative strengths were lower in the AUD for both unthresholded and thresholded networks (Figure 2B,C). AUD also had lower nodal measures of degree, clustering coefficient and positive and negative nodal strengths (Figure 2D–G). Group comparisons of control (CON) and alcohol use disorder (AUD) covariance networks. (A) Covariance differences of weights for CON minus AUD matrix (lower triangle) and permutation test p‐values (uncorrected) of edge‐level significant group differences (upper triangle). (B) Positive and (C) negative total strength values of covariance networks (unthresholded and p < 0.05 correlation significance thresholded). (D–G) Distributions of nodal measures of covariance networks: (D) node degree, (E) clustering coefficient and (F) positive and (G) negative nodal strengths. Differences in distributions from the two networks were assessed with a two‐tailed Kolmogorov–Smirnov test. *p < 0.05, **p < 0.005, ***p < 0.001, ****p < 0.00001. Ant, anterior; Pos, posterior; NAc, nucleus accumbens; PUT, putamen; CAU, caudate. Community detection identified a two‐community partition for each group. For the CON covariance network, the partition grouped bilateral nucleus accumbens and anterior putamen into one community and bilateral caudate and posterior putamen into another (Figure 3A). These communities were clearly visible in the correlation matrix, even without employing community detection. The covariance structure in the AUD included one large community and a second smaller community that was composed of bilateral posterior putamen and left posterior caudate (Figure 3B, lower triangle). From the initial set of 10 000 partitions, the probability of any region pair being assigned to the same community can be expressed in a co‐assignment matrix (Figure 3C). From the AUD co‐assignment matrix, it is evident that there is no clear community partition and that the homotopic similarity observed in CON is absent. The thresholded matrices for both groups (Figure 3D) showed no cross‐hemisphere covariance of nucleus accumbens and anterior putamen in AUD that was observed in the CON network. Additionally, using the CON partition as a reference, we quantified differences in communities between groups by computing within‐ and between‐module strengths as well as the mean BPND in each community. For unthresholded networks, within‐community positive strengths (Figure 3E) and between‐community negative strengths (Figure 3F) were lower in AUD. When examining the average within‐module BPND for the two communities (Figure 3G), there were no significant group differences for either module. Community structure of inter‐striatal covariance. (A) Social drinking controls (CON) network community detection revealed a bipartition of the network into two communities. (B) The non‐treatment‐seeking alcohol use disorder (AUD) covariance network was also a bipartition (lower triangle); however, these communities were distinct from CON, as seen in the upper triangle where regions are arranged by community affiliations form the CON network. (C) Co‐assignment matrices for CON (lower triangle) and AUD (upper triangle) show the probability of any two nodes being assigned to the same community from the initial ensemble of 10 000 partitions, illustrating the disrupted community structure in AUD relative to CON. (D) CON community structure ordered networks for CON (lower triangle) and AUD (upper triangle). (E, F) Total strengths within each community and the between‐community blocks of the matrices: (E) positive and (F) negative strengths. (G) Community average binding potentials were computed per participant and displayed by group. Ant, anterior; Pos, posterior; NAc, nucleus accumbens; PUT, putamen; CAU, caudate. Exploratory cross‐sectional analysis of cigarette use We stratified the sample on cigarette use and computed covariance networks for the nsCON (n = 19), nsAUD (n = 7) and smAUD (n = 10) subgroups. The three smoking controls were not sufficient to perform analysis in this group. Overall, we observed a similar pattern, where the AUD groups had altered covariance of homotopic regions, notably the nucleus accumbens, as well as reduced thresholded network densities (Figure 4A–C). Community structure analysis revealed that the highest variance in community assignment was in smAUD (Figure 4D–F, left triangles). The loss of cross‐hemisphere covariance is also observed in the community ordered matrices for each group (Figure 4D–F, right triangles), such that homotopic regions (e.g. the nucleus accumbens) no longer group into the same community within the AUD subgroups. Using nsCON partitions as reference, average BPND values were compared for each module across subgroups. Module 2 (composed of bilateral caudate and posterior putamen) had a significantly lower BPND in smAUD relative to the nsCON group (one‐way ANOVA F(2,33) = 4.75, p = 0.015, with post hoc Tukey–Kramer tests, p = 0.011). Covariance and community structure of non‐smoking control (CON) (A, D), non‐smoking alcohol use disorder (AUD) (B, E) and smoking AUD (C, F) subgroups. (A–C) Lower triangle matrices show the covariance networks (partial Pearson correlation, adjusted for age and sex) for (A) non‐smoking CON, (B) non‐smoking AUD and (C) smoking AUD. Upper triangle matrices show significant edges (p < 0.05, permutation testing) for each covariance network with network densities (number of non‐zero edges) reported above each matrix. (D–F) Multiresolution consensus clustering (MRCC) co‐assignment matrices (left triangle), with their respective community ordered covariance matrices in the right triangles (same data as (A–C) [lower triangles] but reordered based on identified community structure for each group network). Ant, anterior; Pos, posterior; NAc, nucleus accumbens; PUT, putamen; CAU, caudate. Validation in the independent FAL data set We leveraged a publicly available FAL PET data set 26 , 45 to assess whether the striatal DAergic covariance observed in our RAC PET data set was capturing DA‐receptor‐distribution‐specific variance. Comparison of FAL covariance and RAC covariance showed strong similarity (Figure 5), with a significant correlation of edge weights (Pearson's r = 0.66, p < 5 × 10−9). Interestingly, region pairs whose weights showed an opposite relationship were predominantly interhemispheric (Figure 5C). Comparison of inter‐subject group covariance networks from [11C]raclopride‐ and [18F]fallypride‐quantified striatal dopamine tone. (A) Subject series of fallypride Z‐scored BPND for striatal region of interest. (B) Side‐by‐side comparison of raclopride‐ (lower triangle) and fallypride‐derived (upper triangle) networks (partial Pearson correlation, adjusted for age and sex). (C) Edgewise cross‐correlation of network weights between the raclopride‐ and fallypride‐derived networks presented in (B). Colour coding distinguishes edges within and between hemispheres. Ant, anterior; Pos, posterior; NAc, nucleus accumbens; PUT, putamen; CAU, caudate. DISCUSSION Here, we show that network neuroscience methodology can successfully quantify the interrelationships of DAergic striatal tone to study the neural correlates of AUD. The structured covariance of RAC BPND observed among the striatal regions of the social drinking CON group was disrupted in the active AUD group. This pattern of covariance was absent when networks were generated from the tracer kinetic model flow parameters k 2 and R1 or from region size for either the CON or AUD group (Supplemental Figures). This covariance structure was also replicated in an independent data set from a publicly available FAL study of healthy volunteers from Castrellon et al. 26 In sum, these results suggest that the observed striatal covariance structure is not limited to our sample or the raclopride PET tracer but rather captures biologically meaningful information on DAergic tone in the striatum. We used the Melbourne subcortical atlas to parcellate the striatum into 12 nodes, which can be anatomically aggregated into the caudate, putamen and ventral striatum (which includes the nucleus accumbens). In CON groups from both data sets (RAC and FAL), a two‐community partition subdivided the striatum into nucleus accumbens and anterior putamen in one community and caudate and posterior putamen into another. Striatal circuitry is typically split into three functional domains: limbic (reward and reinforcement) involving the ventral striatum, associative (cognitive and executive function) involving the central striatal areas and motor (sensorimotor control) involving the dorsolateral regions. 11 , 50 , 51 These three subdivisions follow an anterior–posterior gradient with topographically organized parallel and integrative circuits with the cortex. 10 , 52 The community organization of RAC BP, which largely divides the anterior and posterior striatal covariance network nodes, is generally consistent with these functional domains of the striatum. Although the covariance structure of CON groups between the RAC and FAL data sets was similar (Figure 5), qualitative differences were seen in the interhemispheric covariance of these networks. With the FAL data set, 26 our motivation was to determine whether observed DA covariance is tracer specific. A factor that could be contributing to these differences in covariance between the RAC and FAL data sets is the distribution of DA D2/D3 receptors in the striatum. Both RAC and FAL are generally considered nonselective D2/D3 antagonists. Although FAL has a higher D3 selectivity relative to RAC, it is not sufficient to consider it D3 selective. 53 Autoradiography studies have shown that both D2 and D3 are present in the striatum, with a greater D2 density throughout. 54 Additionally, there is evidence that the D2/D3 density ratio differs in precommissural (anterior) and postcommissural (posterior) caudate and putamen. 55 Taking this into account and in combination with the comparison of RAC and FAL covariance networks in Figure 5, it is plausible that the regional receptor distribution contributes to the observed covariance variability between PET tracers. The difference in age (FAL data set participants were younger than RAC data set controls) could also be a contributing factor as age‐related changes in DA function are known to occur over the lifespan. 40 , 42 , 44 , 56 , 57 However, given the available data, we were unable to explicitly test whether the variability in interhemispheric covariance seen between RAC and FAL in Figure 5B is driven by age. After establishing that DAergic striatal covariance networks possessed organizational features, we investigated potential alterations in AUD, as substance use disorders are believed to disrupt the DAergic system. 58 Our analyses showed that the covariance structure of the AUD group network is indeed altered, showing less organization and lower network degree, density and strength (Figure 2), supporting previous literature that alcohol dependence disrupts DAergic tone in the striatum 59 while providing novel insight into the interregional relationships of DA in the striatum. Three edges showed uncorrected significance in comparison between AUD and CON: ipsilateral left hemisphere nucleus accumbens shell and anterior caudate, and contralateral left accumbens core and right hemisphere accumbens shell and core (Figure 2A). There is prior evidence of asymmetry in alcohol‐related DA release in the ventral striatum, 8 , 60 and therefore, it may be that the left–right synchrony of DA function is disrupted in chronic alcohol misuse. However, due to the limited sample size and uncorrected significance of our edgewise testing, these findings should be replicated in larger cohorts where corrections for multiple comparisons would be feasible. The interpretation of structured DA tone covariance is challenging, as it cannot be thought of as a form of direct or indirect connectivity. These networks are derived at the group level and therefore likely reflect anatomical properties shared by humans (mammals), which can become altered in disease. It is possible that this structure is a macroscale manifestation of neuronal dynamics in the striatum. In support of this, preclinical research in rodents with calcium imaging of striatal microcircuits has shown that manipulations of the striatal DAergic system alter the network firing properties of striatal neuronal ensembles. 61 , 62 DA tone, as investigated here, is a combination of receptor density and synaptic and extrasynaptic DA concentration. PET studies in AUD and more broadly addiction have shown alterations in both DA tone and DA release via pharmacological challenge paradigms. 63 Therefore, it is likely that structured covariance at baseline in healthy individuals indirectly reflects striatal neuronal activity as a function of DA tone. This is consistent with the cross‐sectional differences in covariance networks as observed here. The Melbourne subcortical parcellation used here was derived from functional MRI connectivity data with a gradient approach, which delineated boundaries based on sharp changes in subcortical to whole‐brain connectivity. 13 The observed structured covariance in the present data may be related to the functional connectivity of these regions. Indeed, Tziortzi et al. 12 derived a structural connectivity‐based parcellation of the striatum in which regional DA release was more homogeneous within connectivity‐based subdivisions rather than within anatomical structure‐based subdivisions. In AUD, investigations into the connectivity of the striatum have focused on cortico‐striatal connectivity, with seed‐based approaches indicating alterations in fronto‐striatal and fronto‐insular connectivity. 28 , 64 It is possible that prior methodological challenges associated with functional imaging of subcortical structures have limited the analysis of intra‐striatal connectivity. As MRI scanners and sequences improve in their ability to image deep nuclei and better quantify intra‐striatal connectivity, the combination of PET‐derived covariance and functional MRI‐derived connectivity networks with methods from network science could provide unique insights into circuits critical in AUD and substance use disorders. Finally, we carried out an exploratory analysis where the sample was stratified into cigarette‐smoking and non‐smoking AUD and non‐smoking CON participants. With only three smoking CON participants, the sample was insufficient to generate a covariance network. As expected, the nsCON group had the highest network density, whereas the density was greater in smAUD (dual‐substance use) relative to nsAUD (single‐substance use) (Figure 4A–C). Interestingly, co‐assignment, a property of the community structure of the networks, showed less organization in the smAUD network relative to the non‐smoking group networks (Figure 4D–F). Although these exploratory findings are intriguing, further work is needed in larger samples and in otherwise healthy cigarette‐smoking participants to help determine the degree to which network disruption in co‐use of tobacco and alcohol is distinct from alcohol or cigarette use alone. Parcellation selection is an important factor in network‐based analyses and requires an informed choice based on the scientific hypotheses and data type (e.g. imaging modality, tracer) and resolution. Here, we elected to apply the multiscale Melbourne subcortical atlas generated from a large sample of functional MRI data. 13 Scale II of this parcellation generally recapitulates the defined anatomical divisions of striatal anatomy commonly employed in DAergic PET imaging of the striatum (i.e. anterior/posterior caudate and putamen and ventral striatum), with one distinction that the nucleus accumbens (ventral striatum) is partitioned into its ‘shell’ and ‘core’ subdivisions. (Note: Although the naming convention was taken from the rodent literature, the human subdivisions do not resemble rodent anatomy; they are contiguous ventral posteromedial and dorsal anterolateral segments). In humans, this division is observed in both immunohistochemical 65 and functional imaging studies. 13 , 66 However, it is not typically reported in DAergic PET studies due to their small size. 66 The Melbourne parcellation subdivides the nucleus accumbens into subregions that are modestly larger than the spatial resolution limits of both the Siemens mCT 67 and the GE Discovery ST. 68 Thus, the measured radioactivity in the subregions will include both a ‘spill‐in’ contribution from the ventral aspects of the caudate and putamen and a ‘spill‐out’ (dilution) of signals to adjacent white matter and cerebrospinal fluid compartments. Therefore, the BP values of the ‘shell’ and ‘core’ are likely affected by both partial volume effect and proximity to larger nuclei with greater apparent D2 availability. As such, the values from these subregions may not represent accurate estimates of BPND. However, we chose to remain consistent with the Scale II Melbourne subcortical parcellation in order to facilitate comparisons across studies and imaging modalities. These findings demonstrate the utility of a network‐based approach in studying alcohol and substance use disorders such as AUD. The covariance strategy has been long applied to fluorodeoxyglucose, 18 amyloid and tau 21 , 69 PET data. Recently, the emphasis on what has been referred to as ‘molecular connectivity’ has begun to grow. 46 , 70 It is important to consider that covariance network applications require a homogeneous group/sample and enough participants to accurately estimate the networks. 15 Our analyses utilized a retrospective sample of CON and AUD participants who were selected based on strict criteria for AUD and alcohol consumption behaviours, thus making a covariance analysis appropriate. As for group size, the adage is always ‘more is better’; however, PET imaging is costly and time/resource intensive, making large sample sizes less feasible. The number of participants used to generate covariance networks in our primary analysis ranged from 17 to 25 participants/group, which we believe to be sufficient to derive a representative group network. Additionally, inference is made at the group level, that is, with the assumption that any observed differences in two networks are driven by the variables of interest, which in the present work are AUD and cigarette use. Lastly, these networks help us to understand the potential significance of interregional relationships. However, care must be taken when interpreting the results, as the biological meaning is not as apparent as in conventional PET neuroimaging analyses. Additional work is needed to fully appreciate the functional relevance of interregional relationships in the striatal DA system. CONCLUSION In summary, we showed that striatal DAergic tone can be successfully investigated using inter‐subject covariance. Additionally, we demonstrated the presence of similar organization in two separate D2/D3 PET tracers. We then showed that this organization is disrupted in AUD, with preliminary evidence for greater disruption in comorbid alcohol and cigarette use. Network approaches to neurotransmitter PET data, recently termed ‘molecular connectivity’, offer a novel avenue for increasing our understanding of alcohol and substance use disorders by studying interactions of regional neurotransmitter patterns. AUTHOR CONTRIBUTIONS Karmen K. Yoder conceptualized the study design, provided funding for the work, secured all regulatory approvals, supervised all aspects of the work and provided intellectual input for the generation and revision of this manuscript. Evgeny J. Chumin performed the image processing and data analyses; developed code that facilitated processing and analyses; drafted, edited and revised the manuscript; and provided intellectual input. Mario Dzemidzic designed the MRI acquisition sequence, contributed to code for processing and analyses and provided intellectual input for the generation and revision this manuscript. All authors critically reviewed the manuscript and approved the final version for publication. CONFLICT OF INTEREST STATEMENT The authors declare no conflicts of interest. AUTHOR STATEMENT All authors have read and approved the manuscript for submission and made a substantial contribution to the conception, design, gathering, analysis and/or interpretation of data and a contribution to the writing and intellectual content of the article. All authors acknowledge that they have exercised due care in ensuring the integrity of the work. None of the original material contained in the manuscript has been submitted for consideration nor will any of it be published elsewhere. All participants provided written informed consent for all study procedures that were approved by the Indiana University Institutional Review Board in accordance with the Belmont Report. Supporting information Figure S1. Z‐scored values of the tissue‐to‐plasma rate constant k 2 (A, B) and tracer delivery R1 (C, D) for social drinking controls (CON) and active alcohol use disorder (AUD) participants. Inter‐subject covariance networks were computed as partial Pearson correlation, adjusting for age, biological sex, and smoking status. Ant – anterior, Pos – posterior, NAc – nucleus accumbens, PUT – putamen, CAU – caudate, CON – controls, AUD – alcohol use disorder. Figure S2. Z‐scored values and covariance networks of region size (in voxels) for social drinking controls (CON) and active alcohol use disorder (AUD) participants. Inter‐subject covariance networks were computed as partial Pearson correlation, adjusting for age, biological sex, and smoking status. Ant – anterior, Pos – posterior, NAc – nucleus accumbens, PUT – putamen, CAU – caudate, CON – controls, AUD – alcohol use disorder. ACKNOWLEDGEMENTS The authors would like to thank Dr. Olaf Sporns for feedback on method development for this manuscript. We are also thankful to the research imaging technologists of the Indiana Institute of Biomedical Imaging Sciences. In addition, we would like to acknowledge the assistance of Dr. Meredith Halcomb and study coordinators Karen Hile and James Walters. This work was supported by the National Institute of Alcohol Abuse and Alcoholism grants R01 AA01835402 (KKY), R21 AA02265002 (KKY), T32 AA07462 (EJC) and F31 AA025518 (EJC). This research was supported, in part, by the Lilly Endowment, through its support for the Indiana University Pervasive Technology Institute and, in part, by the Indiana METACyt Initiative. Chumin EJ, Dzemidzic M, Yoder KK. Intra‐striatal dopaminergic inter‐subject covariance in social drinkers and non‐treatment‐seeking alcohol use disorder participants. Addiction Biology. 2025;29(11):e70008. doi: 10.1111/adb.70008 DATA AVAILABILITY STATEMENT Image processing and covariance network analysis scripts are publicly available: structural T1‐weighted image preprocessing (https://github.com/IUSCA/IUSM-ConnPipe), RAC PET processing (https://github.com/echumin/PET_processing_Code) and covariance network analysis (https://github.com/echumin/CovNet). Data can be made available upon reasonable request to the corresponding author. REFERENCES Associated Data Supplementary Materials Figure S1. Z‐scored values of the tissue‐to‐plasma rate constant k 2 (A, B) and tracer delivery R1 (C, D) for social drinking controls (CON) and active alcohol use disorder (AUD) participants. Inter‐subject covariance networks were computed as partial Pearson correlation, adjusting for age, biological sex, and smoking status. Ant – anterior, Pos – posterior, NAc – nucleus accumbens, PUT – putamen, CAU – caudate, CON – controls, AUD – alcohol use disorder. Figure S2. Z‐scored values and covariance networks of region size (in voxels) for social drinking controls (CON) and active alcohol use disorder (AUD) participants. Inter‐subject covariance networks were computed as partial Pearson correlation, adjusting for age, biological sex, and smoking status. Ant – anterior, Pos – posterior, NAc – nucleus accumbens, PUT – putamen, CAU – caudate, CON – controls, AUD – alcohol use disorder. Data Availability Statement Image processing and covariance network analysis scripts are publicly available: structural T1‐weighted image preprocessing (https://github.com/IUSCA/IUSM-ConnPipe), RAC PET processing (https://github.com/echumin/PET_processing_Code) and covariance network analysis (https://github.com/echumin/CovNet). Data can be made available upon reasonable request to the corresponding author.