Source-linked AI summary
Structurally Informed Connectivity Disruptions in Cocaine Use Disorder
Seyed Majid Razavi, Saeed Tajik Hesarkuchak, Triet M. Tran, Mehdi Zaeifi, Amirhossein Arezoumand, Farnaz Zamani Esfahlani, Jason A. Oliver, Sina Khanmohammadi
TL;DR
CUD is linked to widespread functional-network alterations whose mechanisms and clinical relevance remain poorly understood. The paper integrates individualized structural connectivity with dynamic functional connectivity and finds greater integration and recruitment but lower flexibility in CUD, with features related to weekly cocaine use.
Problem
The mechanisms underlying CUD-related functional-network alterations and their relationships to clinical and cognitive outcomes remain poorly understood.
Method
The study applies individualized structural priors and Laplacian spectral smoothing to dynamic functional-connectivity matrices, then analyzes multilayer community organization.
Results
CUD showed higher whole-brain integration and recruitment and lower flexibility than healthy controls, with network-specific effects across canonical systems.
Takeaways & Limitations
Structurally informed dynamic connectivity measures were informative about CUD network alterations and individual differences in weekly cocaine use.
Takeaways & Limitations
The cross-sectional group differences cannot establish whether the observed patterns reflect pre-existing vulnerability or consequences of prolonged cocaine exposure.
Abstract
from arXiv · showhide
Cocaine Use Disorder (CUD) is associated with widespread alterations in large-scale functional brain networks, yet the mechanisms contributing to these changes and their relationship to clinical and cognitive outcomes remain poorly understood. To address this gap, we introduce a framework to extract structurally informed dynamic functional connectivity patterns. We then leverage these connectivity patterns to characterize differences in functional brain network organization associated with CUD and to examine their relationship with clinical measures. Specifically, we applied Laplacian spectral smoothing to each participant's functional connectivity matrix using individualized structural priors derived from diffusion imaging. These structurally informed connectivity features were subsequently used to examine cross-network interactions and characterize dynamic community organization across functional brain states. Our findings indicate that individuals with cocaine use disorder exhibit increased integration and recruitment accompanied by reduced flexibility in the functional brain networks, with the most pronounced alterations in visual, attentional, and control systems. In addition, structurally informed functional connectivity features were predictive of weekly cocaine use within the CUD cohort. Overall, these results highlight the value of structurally informed dynamic connectivity measures for characterizing network-level alterations associated with cocaine addiction and for linking these alterations to clinically meaningful measures of cocaine use severity.
Introduction
CUD involves structural and functional brain-network disruptions linked to cognitive and emotional difficulties, but prior work often studied these domains separately. This study integrates individual structural connectivity with dynamic functional connectivity to characterize CUD-related network organization and clinical associations.
- CUD is associated with craving, compulsive drug seeking, impaired emotion regulation, attention deficits, and reduced response inhibition.
- Neuroimaging studies report CUD-related abnormalities in gray matter volume, white matter integrity, cortical thickness, and major white-matter tracts.
- Functional-connectivity disruptions have been reported within and between the default mode, salience, and central executive networks.
- Prior studies often treated structural brain alterations and functional connectivity as independent constructs.
- The study uses structural connectivity as an explicit prior for dynamic functional connectivity and evaluates network dynamics, CUD classification, and weekly cocaine use.
Materials and Methods
The study analyzed open-access multimodal MRI data from 76 participants with CUD or healthy control status. The framework combined structural, functional, demographic, and clinical data after standardized preprocessing and quality-control procedures.
- The dataset included 38 healthy controls and 38 individuals with cocaine use disorder.
- CUD status was established using the Spanish version 5.0.0 of the MINI-Plus interview.
- Participants contributed structural T1-weighted MRI, multishell diffusion-weighted imaging, and resting-state functional MRI.
- The framework comprised data preparation, network construction, multilayer community detection, and community-dynamics analysis.
- The imaging pipeline used fMRIPrep preprocessing for structural and functional MRI data and included motion-related quality control.
LIMB CON DMN
The framework represents cortical fMRI data as temporally segmented, structurally informed multilayer networks. It then applies multilayer community analysis to characterize organization across canonical functional systems.
- Network Construction: Functional connectivity was computed across non-overlapping temporal windows, while diffusion imaging generated subject-specific structural connectivity matrices.
- Network Construction: Structural smoothing produced SiFC layers that were stacked into a temporal multilayer network for community detection.
- Data Preparation: Resting-state fMRI was extracted from 200 cortical regions grouped into seven canonical resting-state networks.
- Data Preparation: ROI analyses were restricted to cortical parcels, so the LIMB label denotes cortical parcels assigned to the limbic network rather than subcortical limbic structures.
- Quality Control: Quality control excluded runs with excessive motion, defined as framewise displacement above 0.5 mm in more than 20% of volumes.
Network Construction
Network construction combined subject-specific diffusion-derived structural connectivity with windowed functional connectivity. A tunable graph filter smoothed each functional layer before producing structurally informed multilayer data for community analysis.
- Subject-specific structural connectivity was derived from the Schaefer-200 parcellation using probabilistic tractography and SIFT2-weighted streamline counts.
- Functional time series were divided into 19 non-overlapping 30-second windows to compute 200 × 200 Pearson correlation matrices.
- The normalized structural Laplacian was eigendecomposed to construct a low-pass graph filter with spectral response h(λ) = 1/(1+τλ).
- Bilateral graph smoothing incorporated anatomical structure without forcing functional connectivity to exactly match the structural network.
- The resulting SiFC layers were post-processed for zero diagonals and nonnegativity before forming a subject-specific multilayer tensor.
Multilayer Community Detection
The study detects communities in structurally informed functional-connectivity layers using generalized multilayer modularity, comparing within-layer structure with a strength-preserving null model while coupling adjacent temporal layers.
- Multilayer Community Detection: Generalized multilayer modularity identifies communities by comparing observed structurally informed connectivity with a Newman–Girvan null model.The null model preserves node-strength sequences while randomizing connections.
- Multilayer Community Detection: The modularity framework uses structurally informed functional connectivity matrices that are symmetric, nonnegative, and have zero diagonal.Each matrix corresponds to one temporal layer of the multilayer network.
- Multilayer Community Detection: The modularity model represents each temporal layer as a diagonal block and connects adjacent layers through ordinal interlayer couplings.Community assignments are optimized jointly across the multilayer supra-graph.
- Multilayer Community Detection: One hundred independent Louvain repetitions per subject address stochastic optimization, while joint optimization keeps community labels comparable across temporal windows.Analysis parameters were held constant across participants.
- Multilayer Community Detection: The resolution parameter γ controls detected community size, with lower values producing fewer larger communities and higher values producing more smaller communities.The interlayer coupling parameter ω controls the strength of identity links connecting each node to itself across layers.
Analysis of Community Dynamics
The analysis summarizes dynamic community organization through module allegiance, recruitment, integration, and flexibility, then tests group differences and classification using these network-level measures.
- Analysis of Community Dynamics: Module allegiance quantifies how often pairs of regions share a community across temporal layers and independent Louvain runs.Its values range from 0 to 1 and support network-level summaries.
- Analysis of Community Dynamics: Recruitment measures within-network community co-assignment, whereas integration measures co-assignment between a canonical network and other networks.Recruitment reflects internal cohesion and integration summarizes cross-system coupling.
- Analysis of Community Dynamics: Flexibility measures how frequently each brain region changes community assignment between consecutive temporal layers.The study reports network-averaged flexibility across seven canonical networks and node-wise flexibility across the cortex.
- Analysis of Community Dynamics: Group differences in flexibility, recruitment, and integration are assessed with two-sided nonparametric permutation tests using 10,000 permutations.The tests compare observed differences in group medians with null distributions generated by permuting group labels.
- Analysis of Community Dynamics: Logistic regression compares conventional functional-connectivity and structurally informed models using 21 RSN-level predictors to differentiate CUD from healthy controls.Predictors include flexibility, recruitment, and integration across seven networks.
- Analysis of Community Dynamics: PLS-DA evaluates whether the 21 dynamic network variables distinguish low weekly cocaine dosage of 1–3 g/week from high dosage of 4–6 g/week within CUD.The dosage groups were selected for clinical interpretability and sufficient cross-validation sample sizes.
Results
CUD was associated with higher whole-brain integration and recruitment but lower flexibility, with network-specific alterations and improved prediction when structural priors informed connectivity features.
- Whole-brain dynamics: CUD showed higher integration (p = 0.0001, q = 0.0001) and recruitment (p = 0.0001, q = 0.0002), but lower flexibility (p = 0.0005, q = 0.0009).These whole-brain differences were measured relative to healthy controls.
- Network-specific dynamics: Integration was elevated across all seven networks, with FDR-significant effects in SM, DAN, SVAN, LIMB, CON, and DMN, but not VIS.VIS showed a nominal integration increase that did not survive correction.
- Network-specific dynamics: Recruitment increases survived FDR correction in VIS (p = 0.0013, q = 0.0065) and DAN (p = 0.0188, q = 0.0494), while other effects were nominal or absent.CON and LIMB showed nominal effects; SVAN and DMN showed no evidence of group differences.
- Network-specific dynamics: Flexibility was reduced after FDR correction in VIS (p = 0.0043, q = 0.0120), DAN (p = 0.0004, q = 0.0028), and CON (p = 0.0017, q = 0.0068).No reliable flexibility differences were observed in the other reported networks.
- CUD classification: SiFC classification achieved an AUC of 0.771 versus 0.620 for FC, yielding ∆AUC = 0.151 with permutation p = 0.007.The comparison used stratified 10-fold cross-validation; SiFC also showed a smaller standard deviation across folds.
- Weekly cocaine-use classification: For weekly cocaine-use classification, SiFC achieved higher AUC (0.831 vs. 0.781) and balanced accuracy (0.732 vs.0.659) than FC.SiFC also had greater total VIP: 19.22 ± 0.33 versus 18.16 ± 0.36 for FC.
Discussion
CUD showed a more integrated, recruited, and less flexible dynamic community organization, with network-specific effects and structurally informed features related to cocaine-use intensity. These findings are constrained by causal, connectivity-sign, and sample-size limitations.
- Whole-brain findings: CUD showed higher whole-brain integration and recruitment but lower flexibility than healthy controls.Reduced flexibility indicates less frequent changes in community assignments and greater temporal stability.
- Network-level findings: Integration was higher across all resting-state networks, with FDR-corrected effects in SM, DAN, SVAN, LIMB, CON, and DMN.VIS showed a nominal integration increase.
- Network-level findings: Recruitment increases were strongest in VIS and DAN, while flexibility reductions were most pronounced in visual, attentional, and control networks.LIMB and CON recruitment effects and SM flexibility effects were weaker and nominal.
- Predictive analyses: Structurally informed connectivity features improved discrimination between CUD and healthy controls across classification metrics.The model also showed stronger evidence for above-chance performance and higher aggregate VIP than independent FC.
- Clinical relevance: Within CUD, structurally informed features distinguished high from low weekly cocaine use using a distributed signature across integration, recruitment, and flexibility.Strong contributors included integration in VIS and DMN, recruitment across several networks, and flexibility in DMN and LIMB.
- Limitations: The findings cannot establish whether the network patterns reflect pre-existing vulnerability or consequences of prolonged cocaine exposure.The study also used positive functional connectivity networks, and its modest sample size was N = 76.
Conclusion
The study introduced a structurally informed dynamic connectivity framework for characterizing network reconfiguration in CUD. It identified increased cross-network integration and recruitment, reduced flexibility, and features related to weekly cocaine-use intensity, while emphasizing the need for further validation.
- Conclusion: The framework combined subject-specific structural connectivity priors with time-resolved functional connectivity and multilayer community detection.It quantified integration, recruitment, and flexibility across canonical brain networks.
- Conclusion: CUD showed increased cross-network integration and recruitment together with reduced flexibility, especially in visual, attentional, and control networks.The pattern was described as more rigid and less reconfigurable functional community organization.
- Conclusion: Structurally informed network features distinguished individuals with CUD from healthy controls and captured differences between high and low weekly cocaine use.Discriminative features included visual and default mode integration, distributed recruitment, and flexibility within default mode and limbic systems.
- Conclusion: Further studies should validate the findings in larger independent cohorts and examine their longitudinal stability and diagnostic specificity.The conclusion also calls for testing whether the signatures are specific to CUD or broader across substance-use disorders.
Author Contributions Statement
The author contributions covered data curation, analysis, investigation, methodology, validation, visualization, drafting, supervision, and review and editing.
- Author Contributions: SMR led data curation, formal analysis, investigation, methodology, validation, visualization, and original-draft writing.STH contributed data curation, investigation, and visualization.
- Author Contributions: TMT contributed methodology and investigation, while MZ and AA contributed visualization and investigation.FZE, JAO, and SK contributed writing review and editing; SK also contributed conceptualization, investigation, and supervision.
Supplementary Material
Supplementary analyses assessed sensitivity to structural-prior strength, temporal-window length, and network-level dynamic community measures. Across tested prior strengths, the principal whole-brain group-difference directions remained consistent.
- Whole-brain sensitivity: Whole-brain sensitivity analyses compared Integration, Recruitment, and Flexibility across τ = 0.1, 0.5, and 0.7.Boxplots displayed individual participants with medians and interquartile ranges.
- Whole-brain sensitivity: Across structural-prior strengths, CUD showed higher Integration and Recruitment and lower Flexibility than healthy controls.The reported direction of group differences was robust to τ, although flexibility effects weakened with stronger regularization.
- Network-level sensitivity: Network-specific analyses examined Integration, Recruitment, and Flexibility across VIS, SM, DAN, SVAN, LIMB, CON, and DMN.Separate figures used τ = 0.1, τ = 0.5, and τ = 0.7.
- Parameter sensitivity: Parameter heatmaps varied sliding-window length across 30s, 60s, and 120s and structural-prior strength τ across 0.1, 0.3, 0.5, and 0.7.They summarized CUD-minus-HC differences for Integration, Recruitment, and Flexibility by network.