Exploratory multimodal fusion analysis of resting-state activity and mGlu5 receptor availability in alcohol use disorder

1056 Introduction: Metabotropic glutamate type 5 (mGlu5) receptors modulate the strength and plasticity of glutamatergic synapses across the brain, and so may be well-situated to influence network-level neural activity. Possible relationships between mGlu5 availability and resting-state activity have not yet been fully explored. Using positron emission tomography (PET) with 18F-FPEB, we recently found that cortical mGlu5 availability is higher in people with alcohol use disorder (AUD) within approximately 1 week of abstinence[1]. People recovering from AUD also show persistently altered spatial-temporal patterns of resting-state brain activity measured by functional magnetic resonance imaging (fMRI) [2]. The objective of this work was to assess the relationships between mGlu5 signaling and low-frequency oscillations in resting-state activity by exploring shared patterns of variance in healthy people and in AUD patients, where both systems are altered. Methods: PET scans with 18F-FPEB (159.4 ± 33.7 MBq) and 3T resting-state fMRI scans (2x6min multiband echoplanar images) were acquired in 17 subjects who met DSM-5 criteria for AUD (42.3 ± 8.65 years; 13M/4F) and 23 healthy control volunteers (HCs; 41.0 ± 12.9 years; 17M/6F). HCs were scanned at a single time point; AUD patients were scanned at two time points, approximately one week and four weeks after initiating medically supervised abstinence. Parametric images of PET volume of distribution (VT) were determined using equilibrium analysis, then entered into an independent component analysis (ICA) to achieve data-driven separation of spatial patterns of VT with distinct variation across subjects. ICA was also applied to fMRI images to identify resting-state networks. Low-frequency oscillations of activity within each component were assessed using fractional amplitude of low frequency fluctuation (fALFF). Correlations between loading of PET components from unimodal ICA and fALFF for each fMRI network component were assessed. To identify specific associations between mGlu5 and fALFF patterns, in a separate exploratory fusion analysis parallel group ICA + ICA [3] was performed. This approach performs ICA on imaging data from both modalities while optimizing correlation between expression of PET and fMRI component loading across subjects. Results: Unimodal ICA identified 3 PET components of mGlu5 (Fig A). Loading of PET component 2, with binding primarily in the parietal and superior frontal cortices, was positively correlated with fALFF in the posterior default mode network, left executive control network (ECN), salience and sensorimotor networks in HCs (p(FDR) 0.05). Multimodal analysis using parallel ICA extracted PET components spatially similar to those from unimodal ICA, characterized by binding in parieto-occipital regions as well as the caudate and thalamus (Fig B). Loading of these two components was correlated with coherence of the left ECN (r=0.35, p=0.02; r=0.37, p=0.01, respectively). Loading of MR-linked PET component 1 was lower in AUD patients at the later time point (p=0.038). Conclusions: Fronto-parietal mGlu5 receptor availability is correlated with low-frequency oscillations in resting-state activity in healthy people across a range of networks, suggesting that higher levels of mGlu5 may support stronger activity in resting-state networks. Multimodal parallel ICA provided consistent but more spatially specific results, identifying a relationship between mGlu5 binding in parietal cortex as well as striatum and thalamus with resting activity in the left ECN. AUD participants exhibited higher mGlu5 availability but weaker or absent mGlu5 relationships to fALFF compared with HCs, suggesting possible functional separation of these systems. Work is ongoing to further assess multimodal analyses exploring relationships between molecular imaging data and brain activity patterns.