Multi-omics signatures of alcohol use disorder in the dorsal and ventral striatum

Alcohol Use Disorder (AUD) is a major contributor to global mortality and morbidity. Postmortem human brain tissue enables the investigation of molecular mechanisms of AUD in the neurocircuitry of addiction. We aimed to identify differentially expressed (DE) genes in the ventral and dorsal striatum between individuals with AUD and controls, and to integrate the results with findings from genome- and epigenome-wide association studies (GWAS/EWAS) to identify functionally relevant molecular mechanisms of AUD. DNA-methylation and gene expression (RNA-seq) data was generated from postmortem brain samples of 48 individuals with AUD and 51 controls from the ventral striatum (VS) and the dorsal striatal regions caudate nucleus (CN) and putamen (PUT). We identified DE genes using DESeq2, performed gene-set enrichment analysis (GSEA), and tested enrichment of DE genes in results of GWASs using MAGMA. Weighted correlation network analysis (WGCNA) was performed for DNA-methylation and gene expression data and gene overlap was tested. Differential gene expression was observed in the dorsal (FDR < 0.05), but not the ventral striatum of AUD cases. In the VS, DE genes at FDR < 0.25 were overrepresented in a recent GWAS of problematic alcohol use. The ARHGEF15 gene was upregulated in all three brain regions. GSEA in CN and VS pointed towards cell-structure associated GO-terms and in PUT towards immune pathways. The WGCNA modules most strongly associated with AUD showed strong enrichment for immune response and inflammation pathways. Our integrated analysis of multi-omics data sets provides further evidence for the importance of immune- and inflammation-related processes in AUD.


Introduction
Alcohol Use Disorder (AUD) is a major contributor to the global disease burden, with a 23 prevalence of ~17% among 12-month alcohol users in the US 1, 2 and an estimated 24 heritability of 49% 3 . Knowledge about the molecular mechanisms can foster understanding 25 of causes and promote prevention. Recent genome-wide association studies (GWASs) have 26 identified 29 genetic loci associated with Problematic Alcohol Use (PAU), a proxy of AUD 4 . 27 While GWASs identify increasing numbers of disease-associated loci, the functional 28 interpretation of many of these findings remains inconclusive. Analyzing the transcriptome 29 can extend the understanding of the molecular mechanisms underlying AUD, by identifying 30 associated gene expression patterns. Findings can in turn be integrated with results from 31 GWASs and epigenome-wide association studies (EWASs) to identify the pathomechanisms 32 underlying disease. 33 Processes in the central nervous system are considered to play a major role in the etiology 34 of addiction, and the transition from chronic alcohol consumption to AUD 5 . Therefore, it is of 35 particular interest to examine molecular changes associated with addiction in brain tissue. 36 So far, only few studies have been conducted in postmortem human brain tissue to identify 37 transcriptional changes associated with AUD [6][7][8] . These studies mainly focused on the 38 prefrontal cortex (PFC) one important part of the neurocircuitry of addiction 9,10 . The first 39 transcriptome-wide study in the PFC found DE genes implicated in neuronal processes, 40 such as myelination, neurogenesis, and neural diseases, as well as cellular processes, such 41 as cell adhesion and apoptosis 11 . In Brodmann Area 9 downregulation of calcium signaling 42 pathways has been observed in individuals with AUD compared to controls 12 . In the same 43 study, a weighted gene co-expression analysis (WGCNA) pointed towards modules 44 associated with AUD case/control status, which were enriched for nicotine and opioid 45 signaling, as well as immune processes. Another study in the PFC (Brodmann Area 8) 46 showed that co-expression networks associated with lifetime alcohol consumption were 47 RNA extraction and -sequencing 98 RNA was extracted from frozen tissue according to the manufacturer´s protocol using the 99 Qiagen RNeasy microKit (Qiagen, Hilden, Germany). The RNA Integrity Number (RIN) of all 100 samples was determined using a TapeStation 4200 (Agilent, Santa Clara, CA). RIN 24 . Minimal pre-filtering 117 was applied, removing genes with normalized counts <10 for more than two samples. 118 Technical replicates were merged prior to differential expression analysis using the 119 collapseReplicates function as implemented in DESeq2. For the differential gene expression 120 analysis, we included age, sex, RIN, pH-value of the brain, and postmortem interval (PMI) as 121 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 7, 2021. ; covariates, because of their known influence on gene expression [25][26][27] . Results were filtered 122 for differentially expressed (DE) genes with an absolute log2 fold change larger than 0.02. P-123 values were adjusted for multiple testing using Benjamini-Hochberg correction (FDR<0.05) 124 28 . All downstream analyses were performed for DE genes with FDR<0. 25. 125 Pathway Analysis 126 Gene-set enrichment analysis was performed using the R package fgsea (v.1.12.0) 29 16 were performed to identify 132 modules of co-expressed genes and co-methylated CpG-sites. We assessed the relationship 133 of these modules with AUD case/control status and tested the overlap between associated 134 modules. WGCNA clusters the input matrix according to a dynamic tree cutting algorithm, 135 using a soft power threshold that approximates the criterion of scale-free topology 136 (R signed 2 >0.80). Resulting soft power thresholds for expression networks were 6 for CN, 5 for 137 PUT, and 14 for VS; for methylation networks, all power thresholds were 2. 138 To identify methylation networks associated with gene expression, beta values from 139 normalized intensities of all samples from which gene expression data were available were 140 filtered for promoter-associated CpG-sites based on the manufacturer's manifest (Illumina,  141 San Diego, CA). The resulting 105 796 CpG-sites were used as input. 142 For the RNA-seq data, count matrices were normalized using the DESeq2 function 143 normalizeCounts and variance stability transformation was applied. 144 Networks were constructed using following settings: minimum module size=30, 145 mergeCutHeight=0.25, maxBlockSize=36 000. In WGCNA, modules are labeled using 146 colors. In the results section modules are labeled according to type of data, brain region, and 147 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 7, 2021. ; color assigned in the analysis, e.g. "e-VS-pink" for module "pink" from the WGCNA analysis 148 of gene expression data in the ventral striatum. For each module, its eigengene was 149 calculated and correlated with AUD status. For modules associated with AUD status, we 150 Thus, promoter-associated CpG-sites were used in the analysis. 158 At the module level, gene-set overlap tests were performed using the R package 159 33 . Here, Fisher's exact test is used to identify significant overlap. For 160 each brain region, the overlap of the AUD-associated co-expression and co-methylation 161 modules was tested. 162 GWAS Enrichment Analysis 163 We analyzed enrichment of DE genes with an FDR<0.25, and genes in AUD-associated 164 WGCNA modules in GWAS summary statistics using Multi-marker Analysis of GenoMic 165 Annotation (MAGMA, v.1.08b) 34 . We performed GWAS enrichment analysis for several 166 SUDs, such as alcohol use disorder and problematic alcohol use 4 , cannabis use disorder 35 , 167 and a recent GWAS comparing individuals with opioid use disorder with unexposed 168 controls 36 . Results with p<0.05 were considered statistically significant. 169 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  Table  189 2; DE genes significant at FDR<0.10 are listed in Supplementary Table S2 ( is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 7, 2021. ; https://doi.org/10.1101/2021.10.04.21264523 doi: medRxiv preprint were found to be related to cilia-and microtubule-associated GO-terms, while none of the 195 Hallmark gene-sets was significantly enriched. GO-term and Hallmark gene-set analysis in 196 PUT samples showed enrichment for immune processes, such as "acute inflammatory 197 response to antigenic stimuli" (padj=0.006) and "adaptive immune response" (padj=0.006). 198 In the VS the most significantly enriched GO-terms were also related to cilia and 199 microtubules, as well as antigen processing. All GO-terms and Hallmark gene-set with Module "e-CN-magenta", consisting of 328 genes, showed the strongest association with 205 AUD status (r=0.42, p=2.89*10 -4 ). In PUT module "e-PUT-black", of the 25 modules (median 206 size 249 genes, range: 33-5 381) identified, was most strongly correlated with AUD (r=0. 41,207 p=2.31*10 -4 ). For expression data from the ventral striatum, 16 modules with a median size 208 of 429 genes (range: 35-9 708) were identified; module "e-VS-pink" had the strongest 209 association with AUD (r=0.41, p=0.009). Interestingly, in a GO-term analysis the three AUD 210 associated modules were all enriched for immune processes, such as "defense response" 211 and "inflammation response". There was also a wide overlap of the genes in the three 212 modules: 174 (22.54%) were partially shared between all three modules corresponding to 213 the three brain regions, while another 21.76% were shared between at least two modules 214 ( Figure 2B). 215 In the GWAS-enrichment analyses, the WGCNA modules "e-CN-magenta" and "e-PUT-216 black" were not enriched for signals from SUD-GWASs. "e-VS-pink" was enriched for genes 217 associated with Cannabis Use Disorder (p=0.043). 218 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  In the present study, we identified DE genes, co-expression networks, and pathways 257 associated with AUD in the dorsal and ventral striatum. The results were integrated with 258 DNA-methylation data and results from GWASs of SUDs. 259 We discovered that one gene (ARHGEF15) was consistently upregulated in all investigated 260 brain regions of AUD cases compared to controls. ARHGEF15 encodes a specific guanine 261 nucleotide exchange factor for the activation of Ras homolog family member A (RhoA), a 262 GTPase, which has been linked to higher blood pressure and hypertension over the 263 Rho/ROCK signaling cascade 37 . It is postulated that the Rho Guanine Nucleotide Exchange 264 Factor 15 negatively regulates excitatory synapse development by suppressing the synapse-265 promoting activity of EPHB2. EPHB2 deficiency has been linked to depression-like behaviors 266 and memory impairments in animal studies 38 . In line with this, genetic variation within 267 ARHGEF15 has been associated with hematocrit, red blood cell count, and hemoglobin 268 concentration 39 , but also with psychiatric traits, such as neuroticism and worries 40 as well as 269 bipolar disorder 41 . 270 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 7, 2021. mitochondrial chaperone, which has been associated with progressive brain atrophy 43 and 275 with the cellular response to alcohol-induced stress 44 . In a recent GWAS, CLPB was 276 associated with the amount of alcohol consumed on a typical day (p=9.67*10 -5 , N=116 277 163) 45 . 278 DE genes in the ventral striatum were enriched for GWAS signals of PAU, but not AUD. This 279 could be a result of the larger sample size of the PAU GWAS, but also point towards 280 differences in genetic variation as responsible for differential expression. 281 Our results from the pathway and network analyses further underline immune-related effects 282 of chronic alcohol exposure; the pathway and network modules most strongly associated 283 with AUD case-control status were also enriched for immune system and inflammation 284 processes. This was observed for all three brain regions, and both in expression and 285 methylation data, providing further evidence for the important role of immune processes in 286

AUD. 287
These results strongly reflect the well-described effect of chronic alcohol exposure on 288 different aspects of the innate and acquired immune systems 46 . Chronic alcohol exposure 289 accelerates the inflammatory response and reduces anti-inflammatory cytokines 46 . An 290 activated immune response in response to chronic alcohol exposure has been shown on the 291 cell level 47 , as well as on the transcription 47 , and protein levels 48,49 . In a previous EWAS, we 292 found strong enrichment of immune processes in differentially methylated CpG-sites 293 associated with alcohol withdrawal 50 . Neuroinflammation has been repeatedly associated 294 with AUD and both the glutamate excitotoxicity and the production of acetaldehyde, key 295 processes in AUD metabolism, have been suggested to produce an inflammatory response 296 in the brain 51 . On a phenotypic level, there is also widespread overlap between symptoms of 297 inflammation and of SUDs, such as anhedonia, depression, and decreased cognitive 298 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 7, 2021. ; https://doi.org/10.1101/2021.10.04.21264523 doi: medRxiv preprint functioning 52 . In addition, in postmortem human brain studies in the PFC, hippocampus, and 299 orbitofrontal cortex, increased mRNA levels of HMGB1 encoding a proinflammatory cytokine 300 and toll-like receptor genes have been associated with alcohol consumption in AUD cases, 301 providing evidence for chronic neuroinflammation in response to alcohol [53][54][55] . Notably, there 302 is an overlap of findings not only on the single-gene level but also on the level of pathways 303 and networks/modules. This overlap underlines that alcohol consumption has common 304 biological effects in different brain regions, i.e., most prominently, effects on immune and 305 inflammation processes. 306 Several limitations apply to our study. First, we cannot distinguish between effects being a 307 consequence of chronic alcohol consumption or addiction. Second, although we corrected 308 for PMI, which can influence tissue quality as a confounding factor, it cannot be ruled out 309 that other characteristics not easily accounted for, such as cause of death, or blood alcohol 310 level for which the majority of individuals have missing data, influenced gene expression. 311 In summary, the present study provides further evidence from multi-omics data sets for the 312 importance of immune-and inflammation-related processes in AUD. Notably, drugs that 313 reduce neuroinflammation to reduce drinking, such as phosphodiesterases, may be 314 promising approaches for novel treatment options for AUD. Recently published randomized 315 controlled trials suggest that a phosphodiesterase inhibitor reduces heavy drinking whereas 316 an antibiotic compound was not effective 56,57 . A deeper understanding of the underlying 317 mechanisms will enhance the discovery of drug targets and drive forward the development 318 of precision medicine within in this field. 319 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 7, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.    . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Data are presented as count (n/n; n (%)) or mean (±SD), PMI: post-mortem interval, pH: pH-value of the brain, p: p-value of t-Test comparing cases and controls.

*significant difference between cases and controls
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 7, 2021. ; https://doi.org/10.1101/2021.10.04.21264523 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 7, 2021. ; https://doi.org/10.1101/2021.10.04.21264523 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 7, 2021. ; https://doi.org/10.1101/2021.10.04.21264523 doi: medRxiv preprint