Liver molecular networks associated with drinking behavior in nonhuman primates

Abstract

Background

Consumed ethanol is primarily metabolized by the liver, with resulting products of ethanol metabolism, acetaldehyde and salsolinol that influence brain activity and alcohol drinking behavior. Alcohol consumption in humans is highly heritable with numerous associated genetic variants. Functional variants in the ADH and ALDH genes influence liver alcohol metabolism but only account for a small percentage of variance in consumption. We hypothesized that variation in hepatic molecular networks during the induction phase, where animals consume identical amounts of alcohol, predicted variation in drinking behavior during subsequent ad libitum access in nonhuman primates (NHPs).

Methods

We studied male rhesus macaques at baseline and during the uniform consumption phase that became discordant at the later ad libitum phase. The study design increased the likelihood of identifying functional molecular differences between light drinkers (LD) and very heavy drinkers (VHD) before animals exhibited differences in drinking behavior. We analyzed liver biopsies, provided by the Monkey Alcohol and Tissue Research Resource (MATRR), collected at baseline and after 3 months of uniform consumption, using multiomic and histologic methods.

Results

We found hepatic molecular pathways and networks differed between LD and VHD at baseline and in response to identical consumption. Notably, Sirtuin Signaling and a MYC‐regulated network were significantly enriched for differentially abundant molecules in both LD and VHD response to uniform alcohol consumption. Potential epigenomic mechanisms regulating response to alcohol consumption were significantly different with LD response primarily through microRNAs, and VHD primarily through DNA methylation. Histological analysis of liver biopsies showed no liver pathologies in either group.

Conclusions

Our findings of differences in molecular networks prior to alcohol consumption suggest genetic variation contributes to drinking phenotypes, and differences in molecular response to uniform alcohol consumption suggest epigenetic mechanisms regulating liver networks also contribute to the development and progression of drinking phenotypes in NHP.

We identified hepatic molecular networks that differed between light drinkers (LD) and very heavy drinkers (VHD) before they were exposed to alcohol, as well as epigenetic differences between LD and VHD in response to the same amount of alcohol. Identification of genetic and epigenetic variation that predicts drinking behavior in alcohol naïve individuals provides the foundation for early identification and individualized prevention of alcohol use disorder, potentially reducing the prevalence and impact of alcohol use disorder.

INTRODUCTION

It is well‐established that ethanol is primarily metabolized by the liver (Cederbaum, 2012). The preference for alcohol consumption in humans is highly heritable, and numerous genetic variants have been identified that are associated with alcohol consumption (Verhulst et al., 2015). Functional variants in the alcohol dehydrogenase (ADH) and aldehyde dehydrogenase (ALDH) genes influence alcohol metabolism in the liver, yet variants in these genes only account for a small percentage of variation observed in alcohol consumption (Edenberg et al., 2019). The major metabolites from ethanol metabolism, acetaldehyde and salsolinol, have been shown to influence brain activity and alcohol drinking behavior (Peana et al., 2016); however, additional, as of yet, unknown products of hepatic alcohol metabolism are also thought to influence alcohol consumption. Studies in a nonhuman primate (NHP) model of alcohol consumption show early drinking typographies are highly predictive of subsequent heavy drinking phenotypes (Grant et al., 2008). Drinking behavior at the final month of induction, when all animals consume identical g/kg amounts (1.5 g/kg) of alcohol, has been shown to predict future ad libitum drinking behavior (Grant et al., 2008). Analysis of alcohol‐naïve whole brain networks in NHP was also found to predict future heavy drinking (r = 0.88) (Rowland et al., 2021). Additionally, low cognitive flexibility in alcohol‐naïve NHP subjects assessed with a set‐shifting procedure was predictive of future heavy consumption (Shnitko et al., 2019).

A limitation of identifying functional genetic and epigenetic variants that influence alcohol metabolism, and possibly predict alcohol drinking behavior, is the inability to analyze liver molecular changes in humans prior to and during early controlled alcohol exposure. Studies in rats have shown tissue specificity of genetic variants with modules of genes in the brain and nonoverlapping modules of genes in the liver associated with alcohol consumption (Hoffman et al., 2018). Studies to determine whether variants in rats are relevant to humans are not feasible where longitudinal biopsies of healthy tissues are required. Of the 490 genes identified in rats that show tissue‐specific variation in gene expression associated with alcohol consumption (Hoffman et al., 2018), 58 have not been detected in human liver and 59 have not been detected in human brain. These transcript profiles highlight the tissue and species specificity of genetic variation that influence drinking behavior and support the need to investigate a model of alcohol consumption where liver metabolism is similar to humans. These studies also highlight the need for a study design that begins with alcohol‐naïve individuals and then transitions to alcohol consumption to identify genetic and epigenetic variants that contribute to variation in drinking behavior by potentially modulating alcohol metabolism and the generation of brain‐relevant intermediate metabolites.

In this study, we selected animals from a large cohort (n = 71) with extreme phenotypes for alcohol consumption. Samples were collected prior to alcohol exposure, and after 3 months of escalating doses of uniform ethanol consumption, in which all animals consumed the same amounts of alcohol (Figure 1A). Samples were used for an integrated omics approach (transcriptomic, epigenomic, and proteomic) to address the hypothesis that variation in hepatic molecular networks either prior to or during early uniform alcohol exposure predicts later ad libitum variation in drinking behavior in NHPs. Animals selected for analysis were discordant for daily drinking behavior at 6 months (light drinker (LD), 1.3 g/kg/day; very heavy drinker (VHD), 3.9 g/kg/day) and 12 months (LD, 1.5 g/kg/day; VHD, 4.0 g/kg/day), significantly increasing the likelihood of identifying molecular networks containing functional variants that differ between LD and VHD.

FIGURE 1.

(A) Overview of study design with each phase of the study and time points for liver biopsy collections; (B) Total ethanol consumption for each animal at 12 months. Red dots indicate animals selected for analysis of liver biopsies.

METHODS

Ethics approval

Tissues were acquired from the Monkey Alcohol Tissue Research Resource (MATRR; gleek.ecs.baylor.edu). All animal work was performed in strict accordance with the recommendations detailed in the Guide for the Care and Use of Laboratory Animals of the National Institute of Health, the Office of Animal Welfare, and the United States Department of Agriculture, and approved by the Oregon National Primate Research Center (ONPRC) Institutional Animal Care and Use Committee (IACUC). Surgical procedures were performed by a veterinarian, including liver biopsy as previously described (. Animals were provided nonsteroidal anti‐inflammatory and opioid analgesics during recovery as needed. Animal housing, handling, diet composition, and details are as described elsewhere).

Ethanol self‐administration protocol

Seventy‐one rhesus macaques in seven cohorts completed the alcohol self‐administration protocol described below. Six males were selected for omics and histological analyses from Cohort 10 (2 light and 1 heavy) and Cohort 14 (1 light and 2 heavy). Each cohort of animals was housed in the same room with single cages for dispensing and monitoring of alcohol consumption. All animals had visual and audio contact with all other animals in the cohort and were provided daily enrichment. The detailed protocol and enrichment were previously published (Grant et al., 2008). Groups were age‐matched with animals ranging from 6 to 9 years of age (p = 0.52) and were weight‐matched at the beginning of the study ranging from 6.9 to 8.4 kg (p = 0.084). Additional information on these cohorts can be found at MATRR (gleek.ecs.baylor.edu).

One month acclimation and training

Monkeys were acclimated to laboratory personnel, then trained to provide their leg through an opening in the cage front for awake blood collection via saphenous or femoral venipuncture for measures of blood ethanol concentrations. All monkeys were trained to operate a drinking panel. When active, the panel signaled fluid and food availability through illumination of stimulus lights.

Four month induction phase (one month of flavored water, three months escalating doses)

Monkeys were induced to drink increasing volumes of an alcohol solution (4% w/v ethanol mixed in deionized water) in a stepwise fashion over four consecutive 30‐day periods for a total of 120 days. To induce drinking, a 1 g flavored food pellet was delivered every 5 min, with water the only available fluid. After water consumption became associated with pellet delivery, the monkeys underwent 1 month with water as the only available drinking fluid. During the second month, animals drank a predetermined volume of alcohol solution corresponding to 0.5 g/kg/day alcohol, followed by volumes of alcohol corresponding to 1.0 and 1.5 g/kg/day during the third and fourth months, respectively. Monkeys were allowed to drink only 4% w/v alcohol until the required dose of alcohol was reached (e.g., 0.5 g/kg/day), at which point animals were allowed to drink only water.

Voluntary drinking phase

Following the 4‐month induction phase, monkeys were given simultaneous access to water and alcohol (4% w/v) and allowed to voluntarily self‐administer alcohol and/or water for 22 h/day (1100–0900 h each day), 7 days/week for 12 months. Consumption was recorded daily by using weighing scales (Ohaus Corp., Parsippay, NJ) to measure change in the mass of dispensing the solution containers. Blood samples were collected every 5 days from the saphenous vein just before the lights were turned off (between 1800 and 1900 h) corresponding to 7 h into the 22 h sessions.

Food consisted of 1 g banana‐flavored pellets (carbohydrate, 63%; fat, 4%; protein, 22%; PJ Noyes, Lancaster, NH). Over the 12‐month duration of the experiment, the monkeys were required to eat their daily allotment of food in no fewer than 3 “meals,” with at least 2 h between each meal. A meal was defined by the proportion of daily food allotted to each monkey and the pace of the animal to obtain the food. The meal ended if one‐third of the daily food allotment was obtained at a time, or if the monkey took longer than 2 min to obtain a pellet.

Selection of animals for analysis of liver biopsy samples

From the 71 animals that completed the ethanol self‐administration protocol, age‐matched males were selected that were greater than one standard deviation from the mean for ethanol consumption. Kinship analysis was based on whole exome sequence data variants using KING (Manichaikul et al., 2010) available from the Macaque Genotype and Phenotype Resource (mGAP).

Blood sample collections

Blood samples (20 μL) for blood ethanol concentration measures were collected every fifth day at 30, 60, and 90 min after the start of each induction session, respectively. These times correspond to peak blood ethanol concentrations (BEC) when 0.5, 1.0, and 1.5 g/kg are gavaged into the stomach. Blood samples are sealed in air‐tight vials containing 0.5 mL of distilled water and 0.02 mL of isopropanol (10%; internal standard) and stored at −4°C until assayed (Hewlett‐Packard 5890 Series II, Avondale, PA, equipped with a headspace auto sampler, flame ionization detector, and a Hewlett Packard 3392A integrator or Agilent 6890 N).

Liver biopsy collections

Liver biopsies were collected prior to alcohol consumption (baseline) and at the end of the 4‐month induction phase described above. Animals were anesthetized with ketamine, intramuscular (i.m.), at a dose of 20 mg/kg body weight and prepared for biopsy using an aseptic technique. The region of the liver to be biopsied was located by generating an ultrasound image with an ATL Ultramark 4 plus ultrasound imager (Bothell, WA). A 2‐mm skin incision was made below the ribcage on the right side of the upper abdomen, and a liver sample was obtained using a 14‐gauge biopsy needle (Temno, Clearwater, FL) inserted into the midaxial line to a depth of 1.5 to 3 cm. The biopsy sample was flash frozen in liquid nitrogen and stored at −80°C until analysis (Ivester et al., 2007).

Clinical measures

Clinical measures blood chemistries including measures of liver function: alanine transaminase (ALT), aspartate aminotransferase (AST), gamma‐glutamyl transferase (GGT); and metabolic measures including glucose, total cholesterol, and triglycerides were performed as described previously (Ivester et al., 2007).

Liver histology

Histologic analyses were performed on liver biopsies collected at the same time as those used for omic analyses, that is, baseline and end of the uniform induction phase. Hematoxylin and eosin‐stained slides were evaluated from each liver biopsy; lipid content and other histologic features associated with steatohepatitis were evaluated by an experienced gastrointestinal and liver pathologist. In addition, picrosirius red staining was performed to examine the level of fibrosis in the liver parenchyma.

Transcriptomics: RNA seq

RNA extractions

Total RNA was extracted from rhesus monkey livers using the Direct‐zol RNA Miniprep Plus kit (Zymo Research). RNA samples were quantified by Qubit assays (Thermo Fisher Scientific), and RNA integrity was determined using an RNA ScreenTape kit on the TapeStation 4200 system (Agilent).

Sequence data generation and processing

Using a NEXTFLEX Combo‐Seq kit (PerkinElmer) and 500 ng RNA from each sample (RIN > 8.5), we followed the manufacturer's instructions to generate mRNA and miRNA libraries that were combined into one library for sequencing. Libraries were quantified using a KAPA Library Quantification kit (Roche Sequencing and Life Science). Pooled libraries were sequenced (1 × 101 bp) using the NovaSeq 6000 system (Illumina) to produce an average of 25 M reads per sample. Demultiplexed raw sequence reads were analyzed using a Partek Flow (Partek Inc.) pipeline. Four random bases at the 5′ end and the poly (A) tract plus downstream sequence at the 3′ end of reads were trimmed. Reads with <15 bases were excluded for alignment. First, trimmed reads were aligned using Bowtie and the rhesus‐MMUL8 mature miRNA (miRBase v22) reference, and aligned reads with at least one read count were quantified in Flow using an expectation/maximization (E/M) algorithm (implemented by Partek). Unaligned reads from the first round of alignment were iteratively aligned to the rhesus‐MMUL8 precursor miRNA (miRbase v22) reference. Afterward, unaligned reads from the second round of alignment were aligned using BWA and the rhesus‐MMUL8 reference, and aligned reads with at least 10 read counts were quantified by the Partek E/M method. Read counts for mature miRNAs and gene transcripts were normalized by the TMM method prior to differential gene expression analysis.

RNA seq data analysis

Transcripts with ≤10 read counts across all samples were filtered out. Filtered sequence reads were normalized, annotated, and abundance determined using Partek Flow (Cox et al., 2021). Differentially expressed transcripts were identified by a two‐sided t‐test assuming equal variance (unadjusted p < 0.05). Gene expression data were deposited in the National Center for Biotechnology Information's Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) ‐ GEO Series accession number GSE285286.

Small RNA seq data analysis

Transcripts with ≤1 read counts across all samples were filtered out. Filtered sequence reads were normalized, annotated, and abundance determined using mirDeep2. Differentially expressed transcripts were identified by two‐sided t‐test assuming equal variance (unadjusted p < 0.05).

DNA methylation

DNA (1 μg) was bisulfite treated using the EZ DNA Methylation kit (Zymo, Irvine, CA). DNA methylation was performed using the Infinium MethylationEPIC BeadChip (Illumina, Inc.), which assays 865,918 human CpG sites. To account for genomic sequences in rhesus DNA that differ from the human sequence‐derived Illumina probes, we limited examination to CpG sites in which probe sequence locations were conserved with human and with fluorescent signals that were significantly above the negative controls, with a detection p‐value of 0.01. This resulted in a total of 445,728 potential CpG sites for differential methylation. All quality control and differential methylation analysis were performed with the R package, ChAMP (Morris et al., 2014). The data were filtered to include only CpG sites within 1500 bp of gene transcription start sites based on GeneCode annotations, resulting in 102,418 CpG sites for statistical analyses. Methylation data were deposited in the National Center for Biotechnology Information's Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) – GEO Series accession number GSE289402.

Proteomics

Sample processing

Details of preparation of liver proteomic samples were as described (Hamid et al., 2022). Approximately 5 mg of each liver tissue sample was homogenized in Tris buffer, precipitated overnight in acetone at −20°C, and centrifuged at 12,000g for 10 min. The protein pellet was dried, reconstituted in 100 mM of ammonium bicarbonate, and quantified. One hundred milligrams of protein was reduced for 30 min in the dark, and digested overnight with trypsin. Samples were cleaned and desalted using Thermo Scientific Pierce C18 Tips, dried, and reconstituted in 0.1% formic acid.

LC/MS data acquisition and analysis

LC/MS data were acquired as described (Hamid et al., 2022). One mg of each sample was loaded on a PepMap RSLC C18 easy‐spray column using Easy‐nLC 1200 coupled to an Orbitrap Lumos Tribrid Mass Spectrometer (Thermo Scientific). Peptides were separated using a 3‐h gradient. Mass spectral data were acquired using Thermo Scientific Xcalibur software. MS raw data were analyzed using MetaMorpheus (Hamid et al., 2022; Miller et al., 2019) using the Macaca mulatta (Rhesus macaque) reference proteome database from Uniprot with 44,389 entries (UP000006718). Peptide and protein quantification were done using the FlashLFQ approach. Protein intensities were normalized using global intensity normalization. In the final normalized data, missing values were imputed using random forest imputation workflow (Hamid et al., 2022; Stekhoven & Buhlmann, 2012).

Comparison of gene and protein abundance

Gene lists (Table S1) and protein lists (Table S2) were uploaded into Venny and Venn diagrams generated showing commonly expressed and differentially expressed genes and proteins.

Pathway and network analyses

For individual omic datasets, all quality molecules for the dataset were uploaded to Ingenuity Pathway Analysis (IPA; QIAGEN) with pathway and network enrichment analyses using differentially abundant molecules and the IPA Knowledge Base, and requiring direct connections based on experimental evidence among differentially abundant molecules. Right‐tailed Fisher's exact test was used to calculate enrichment of differentially expressed genes in pathways, p < 0.01 (Spradling et al., 2013). Regulatory network prediction required previous experimental validation of direct connections in liver or liver cells.

Integrated omic analyses

Multi‐omic data analysis combined the total gene, protein, and/or metabolite lists for all molecules that passed quality filters as appropriate for the data type. Pathway and network enrichment used the same parameters and statistical tests as for individual omic datasets, requiring experimentally validated direct connections.

miRNA–gene/protein pairing

Current pathway and network enrichment tools in IPA do not provide the means to filter direct connections based on inverse abundance between a miRNA and its target. In order to explore the biological relationship of our miRNA data, we performed miRNA–gene pairing in IPA for our miRNA, gene, and protein datasets, requiring opposite expression for experimentally validated or highly predicted interactions (e.g., miRNA upregulated and target gene downregulated). We merged the list of miRNA–gene/protein pairs with the list of genes and proteins in all significantly enriched pathways and networks. This analysis does not provide the means to statistically evaluate the significance of miRNA addition to a given pathway or network; however, this approach provides support for an epigenetic component of the liver response to alcohol consumption.

Identification of pathway and network genes previously associated with alcohol drinking behavior traits

Whole exome single nucleotide variant data were acquired from the mGAP database (Bimber et al., 2019). The following search terms, with all variations of names in the GWAS catalog, were used to query the current GWAS catalog (alcohol consumption, alcohol use disorder, alcohol dependence, alcoholism [heaviness of drinking], and alcohol dependence symptom count). Genes with associations to any of these traits were compared to the list of all differentially expressed miRNAs, genes, and proteins from our transcriptomic and proteomic datasets, and compared with the genes in proteins in multi‐omic significant networks and pathways using Galaxy Genome (Afgan et al., 2018). In addition, variants within or proximal to human GWAS genes were compared with variants in the rhesus in this study (Bimber et al., 2019).

RESULTS

Animal selection, clinical data, and alcohol consumption

Seventy‐one rhesus macaques in seven cohorts have been included in studies on alcohol consumption in which the first 16 months for each cohort followed the same study design (Monkey Alcohol Tissue Research Resource (MATRR, www.matrr.com)), including exposure to the same amount of alcohol during the induction phase (Baker et al., 2014, 2017; Grant et al., 2008).

We used a strategy of selecting extreme phenotypes from the 71 animals of three animals that became LD and three animals that became VHD during the self‐administration phase, allowing us to treat the complex trait of alcohol consumption as a Mendelian trait, that is, selecting animals at opposite ends of the population distribution which improves power to detect genetic variation associated with phenotypic variation (Emond et al., 2012). Animals were age and weight matched at the beginning of the study. The LD group weighed more than the VHD group at the end of the induction phase (LD: 8.4 ± 0.30 kg; VHD: 7.4 ± 0.08 kg, p‐value 0.006). Evaluation of standard blood chemistries, including measures of liver function alanine transaminase (ALT), aspartate aminotransferase (AST), gamma‐glutamyl transferase (GGT), and metabolic measures including glucose, total cholesterol, and triglycerides, showed no differences between LD and VHD groups at both biopsy time points—these results indicate there were no differences in liver function at baseline and the end of the induction phase when all animals consumed the same amount of alcohol (g/kg). Genetic analyses based on whole exome single nucleotide variants indicated that all animals were second degree relatives with each of the five other animals—kinship values ranged from 0.13 to 0.16, indicating a similar degree of relatedness among all study animals. An overview of the study design is shown in Figure 1A.

Comparison of average ethanol consumption for the three LD and three VHD animals during the later ad libitum phases at 6 months and 12 months is shown in Table 1. LD were <1 SD and VHD more than 1 SD from the mean for all animals included in this protocol (n = 71, Table 1, Figure 1B). Throughout the 12 months of the ethanol‐free access phase, including the first 6‐month average (p‐value = 0.0004), second 6‐month average (p‐value = 0.0014), and 12‐month average (p‐value = 0.0005), VHD consumed >2.5 times more than LD.

TABLE 1.

Ethanol consumption and blood ethanol concentrations during the ad libitum phases.

	Average EtOH consumption (gm/kg)			Average blood EtOH conc (mg pct)
	First 6 months	Second 6 months	Twelve months	First 6 months	Second 6 months	Twelve months
Light mean (n = 3)	1.30	1.71	1.51	18.84	22.88	20.96
Very heavy mean (n = 3)	3.87	4.21	4.05	103.81	152.32	129.51
p‐value	0.0004	0.0014	0.0005	0.0190	0.0077	0.0101
Fold difference	2.98	2.46	2.68	5.51	6.66	6.18
Population mean (n = 71)	2.46	2.72	2.57	54.97	70.43	61.47
Population SD (n = 71)	1.01	1.06	0.99	18.52	24.31	22.73

Bold values indicate statistically significant differences between groups.

Histological analyses of liver biopsies

We performed histological analyses of the same baseline and 3‐month liver biopsies as were used for omic analyses. H&E staining was used to assess the quality of tissue sections, and Oil Red O staining was used to quantify lipid content. No differences were seen between LD and VHD biopsies for liver phenotype of lipid content at either time point, indicating that molecular differences at baseline and 3 months of uniform alcohol consumption are not due to liver pathologies (Figure 2).

FIGURE 2.

Representative micrographs from liver biopsies in (A) LD at baseline, (B) LD at the end of the induction phase, (C) VHD at baseline, and (D) VHD at the end of the induction phase. All images are 20× magnification. All images show occasional lipid droplets, ranging from 3% to 5% of the volume. No hepatocytic ballooning, lobular inflammation, or fibrosis was observed.

VHD baseline vs. LD baseline

We identified 21,438 genes that passed quality filters. Of these, 462 were differentially expressed between liver samples from VHD at baseline versus LD at baseline (p ≤ 0.05, Table 2, Table S1). We also identified 1858 proteins that passed quality filters, of which 259 were differentially abundant between LD and VHD at baseline (p‐value ≤ 0.05, Table 2, Table S2). Similar to previous work (Cox et al., 2021), we found that the integration of transcriptomic and proteomic data for pathway and network analyses improved statistical significance for some pathways and networks, while other pathways and networks decreased in significance compared with each individual omic datatype. Results for integrated transcriptomic and proteomic data for pathways (Table S3) and networks (Table S4) are provided. Integration of differentially abundant genes and proteins with DNA methylation (DNAm) data showed 821 differential DNAm sites in promoters of 303, and integration of inverse, differentially expressed miRNAs showed 95 targeted genes/proteins (Table 2).

TABLE 2.

Summary of omic results.

Omic type	Heavy base v light base	Light P Ind v light base	Heavy P Ind v heavy base	Heavy P Ind v light P Ind	Total quality
Differentially expressed miRNA	21	34	10	27	903
Differentially expressed mRNA	462	1032	811	535	21,438
Differentially abundant protein	259	151	275	23	1858
Differential DNAm	33,127	19,532	32,560	39,094	445,728
Gene/Protein pathways	148	171	189	20
Gene/Protein networks*	1	38	37	0
Gene‐Protein/miRNA inverse pairs	114	371	95	124
Unique genes‐proteins in miRNA pairs	95	288	90	105
Unique miRNAs in miRNA pairs	14	26	8	22
Diff DNAm site with gene‐protein ID	32,864	19,343	32,312	38,769
Diff DNAm site with diff gene‐protein ID	821	753	551	926
Diff DNAm with unique diff gene‐protein ID	303	419	551	332

Abbreviations: Base, baseline; P Ind, postInduction.

^* p value <0.05 and predicted activation state.

Pathway enrichment analysis revealed 143 pathways that differed between VHD and LD at baseline. Only Sirtuin Signaling and Inhibition of ARE‐Mediated mRNA Degradation increased in significance with transcriptomic/proteomic integration (p‐value = 4.9 × 10⁻⁷ and 7.6 × 10⁻⁶ respectively). Sirtuin Signaling included epigenetic regulators SIRT1 (upregulated in VHD vs. LD) and SIRT5 (downregulated in VHD vs. LD), transcription factor FOXO1 (downregulated), and stress responder SOD2 (upregulated). Inhibition of ARE‐Mediated mRNA Degradation included down regulation of EXOSC5, a component of the RNA exosome complex, MAPK1, integral to cellular processes such as proliferation, differentiation, transcription regulation, and development, and six subunits of the proteosome multicatalytic proteinase complex. Also of potential interest is the Mitochondrial Dysfunction pathway, which showed decreased significance with gene/protein integration, but still had the most significant p‐value of the integrated pathways; molecules in this pathway overlapped with Sirtuin Signaling (Table S3). Integration of molecules in the Sirtuin Signaling pathway with DNAm and miRNA data showed 62% with differential DNAm sites, including transcription factor FOXO1 and SIRT1; and 12% targeted by inversely expressed miRNAs (Table S5).

SIRT1, which regulates hepatic lipid metabolism and inflammatory processes, has been shown to influence the development of alcoholic fatty liver disease. In Drosophila, acute alcohol exposure is associated with downregulation of Sirt1 (Ramirez‐Roman et al., 2018), and in humans, chronic alcohol consumption has been shown to inhibit SIRT1, leading to deregulation of lipid metabolism and inflammation in the liver (You et al., 2015). SIRT1 has been shown to reduce chronic inflammation and rebalance bioenergetics toward homeostasis (Haigis & Sinclair, 2010). However, these studies did not include investigation of the epigenetic regulation of SIRT1. In our study, SIRT1 was upregulated and had a differential DNAm site in the promoter of VHD versus LD prior to alcohol consumption, suggesting differences in epigenetic regulation of SIRT2 in VHD versus LD NHPs. Additional investigations are required to assess the role of SIRT1 during this early phase of ethanol exposure and how early exposure may trigger this initial liver protective response.

Regulatory network analysis resulted in 33 significant networks, but only one, regulated by Estrogen Receptor 1 (ESR1), had a coordinated directionality (p‐value = 1.32 × 10⁻⁶). The network included 70 targets of the network hub ESR1 (Table S4), including several molecules in the Sirtuin Signaling Pathway (SIRT1, MAPK1, and FOXO1) indicating overlap between the ESR1 regulatory network and Sirtuin Signaling (Table S5). Integration of genes and proteins in the ESR1 regulatory network with promoter DNAm data showed 44% with differential DNAm sites, and integration with inverse, differentially expressed miRNAs showed one miRNA targeting the network (Table S6).

It is interesting to note that although only males were included in this study, ESR1 and downstream targets differ between VHD and LD. Estrogens are traditionally considered female sex steroid hormones; however, females and males both express high levels of ESR1 in the liver, which mediates the hepatic response to estrogens (Qiu et al., 2017). In Esr1 knockout mice, carbohydrate and lipid metabolism in the liver are altered compared with controls, indicating an important metabolic role for ESR1 in the liver (Nevzorova et al., 2016). Although ESR1 expression in the liver is thought to be protective against alcohol‐induced liver injury (Colantoni et al., 2002), ESR1 alternative splicing in humans is associated with variability in drug metabolism and susceptibility to liver‐related diseases (Sun et al., 2019). In our results, we did not see differences in ESR1 expression between VHD and LD; however, our RNA Seq approach did not allow for in‐depth analysis of splice variants. ESR1 network downstream targets FOXO1, as well as SIRT1, have been associated with alcoholic hepatitis. FOXO1 is important for lipid metabolism and oxidative stress response (Kousteni, 2012) and is downregulated in patients with alcoholic hepatitis compared with healthy controls (Yao et al., 2019). In our study, FOXO1 was downregulated with increased methylation of a promoter DNAm site in VHD compared with LD, suggesting FOXO1 downregulation precedes alcoholic hepatitis, potentially through epigenetic mechanisms.

Our findings of a relatively small number of differentially abundant genes and proteins between alcohol naïve VHD and LD animals at baseline, with a large number of significantly different pathways, suggest biologically coordinated differences between alcohol naïve animals prior to becoming VHD or LD with ad libitum alcohol exposure. In addition, the highly significant Sirtuin Signaling pathway and ESR1 network, which included large numbers of epigenetic differences in promoter DNAm sites and targeting miRNAs, further highlight hepatic molecular differences between VHD and LD prior to alcohol consumption.

LD response to alcohol consumption: Postinduction vs. LD baseline

Comparing postinduction to baseline in LD, we identified 1032 differentially expressed genes and 151 differentially abundant proteins (p < 0.05, Table 2). Pathway enrichment analysis revealed 170 pathways that differed between LD postinduction and baseline, including top‐ranked Sirtuin Signaling (p‐value = 1.66 × 10⁻⁶, Table S7). Integration of differential DNAm sites showed differential DNAm sites in 419 differentially abundant genes promoters. Integration of inverse differential miRNAs revealed 288 pairs with 26 unique differentially abundant genes/proteins. In addition, some differentially abundant genes/proteins are potentially regulated by both DNAm and miRNAs (Table 2).

The Sirtuin Signaling pathway had 32 differentially abundant genes/proteins, including downregulated SIRT2 and upregulated SIRT5. Differential DNAm integration showed 12% of the differential DNAm sites associated with AKT1, a serine–threonine protein kinase that regulates many cellular processes; JUN, a transcription factor; CRTC2, a CREB‐Regulated Transcription Coactivator that regulates gluconeogenesis; and SP1, a transcription factor that regulates many cellular processes. In addition, 15% of the miRNAs targeting Sirtuin Signaling genes/proteins included mitochondrial subunits NDUFA11 and NDUFS2, mitochondrial membrane translocases TIMM8A and TOMM40, and transcription factor FOXO1 (Table S5).

Regulatory network analysis resulted in 38 networks with predicted activation states. Of these, five networks were predicted to be activated and 33 inhibited. The most significant network with the greatest number of molecules (n = 92) is regulated by MYC (p‐value = 1.72⁻⁰⁶, Table S8, Figure 3A). Among the molecules in the MYC network is SIRT2, which is downregulated in postinduction versus baseline (Table S8). Integration of DNAm and miRNA data showed 20% of the genes with differential DNAm sites and 33% targeted by miRNAs. In addition, five genes/proteins contained differential DNAm sites and were targeted by inverse, differentially expressed miRNAs. These results support epigenetic mechanisms influencing liver response to alcohol consumption through Sirtuin Signaling, MYC Signaling, as well as influencing liver mitochondrial structural proteins in LD NHPs (Table S9).

FIGURE 3.

Comparison of putative epigenetically regulated genes and proteins in the MYC network for (A) light postinduction versus baseline and (B) heavy postinduction versus baseline. Proteins are denoted by (P) after the name, green fill is down regulated, red fill is up regulated, orange outlines indicate miRNA targets, purple outlines indicate differential DNAm sites, green outlines denote both miRNA target and differential DNAm, and gray indicates no evidence of epigenetic regulation. Arrows indicate activation and T lines indicate inhibition.

VHD response to alcohol consumption: Postinduction vs. VHD baseline

We found 811 genes, 275 proteins, and 189 pathways that differed between VHD postinduction and baseline (p < 0.05, Table 2). Integration of differential DNAm sites showed 551 sites in differentially abundant gene promoters, and 95 genes/proteins targeted by inversely expressed miRNAs (Table 2). Sirtuin Signaling, the top‐ranked pathway (p‐value = 7.24 × 10⁻⁷, Table S10), included upregulation of SIRT3 and SIRT5, which were among the genes containing differential DNAm sites, and SIRT3, and were also targeted by inverse differentially expressed miRNAs (Table S5).

Regulatory network analysis resulted in 37 networks with predicted activation states. Of these, 10 networks were predicted to be activated and 27 inhibited. The top‐ranking network, which is inhibited by MYC, has 104 targets (p‐value = 4.65 × 10⁻⁰⁷, Table S11, Figure 3B). Integration of differential DNAm sites showed 41% of gene promoters with differential DNAm sites. Integration of differentially expressed miRNAs showed 6 miRNAs targeting 11% of the genes/proteins in the network. Seven genes/proteins contained both differential DNAm sites and were targeted by inversely expressed miRNAs (Table S9), suggesting regulation by both epigenetic mechanisms. It should be noted that among the statistically significant regulatory networks is an ATF4‐inhibited network that includes FGF21, which was decreased more than 5‐fold in VHD postinduction compared with baseline (Table S11). A recent study showed that mice lacking Fgf21 require longer recovery time for reflex and balance following ethanol exposure compared to wild‐type littermates (Choi et al., 2023). The large decrease in FGF21 in the VHD animals suggests a maladaptive response by the liver to alcohol exposure.

Comparison of response to ethanol in LD and VHD

Our integrated omic analysis of liver biopsies collected before and after 3 months of uniform alcohol consumption (weight adjusted), comparing animals that subsequently became LD or VHD during free access conditions, showed 54 shared upregulated and 180 shared downregulated molecules. However, the results also revealed that 292 upregulated and 422 downregulated molecules were unique to LD, and 279 upregulated and 251 downregulated molecules were unique to VHD. Comparison of differentially abundant genes and proteins with differential DNAm sites showed 310 increased and 363 decreased DNAm in LD, and 201 with increased and 635 with decreased DNAm in VHD. Comparison of miRNAs targeting differentially abundant genes and proteins with inverse expression revealed 19 miRNAs upregulated and 15 downregulated in LD, and 5 upregulated and 5 downregulated in VHD, with no common miRNAs targeting differentially expressed genes in LD and VHD.

Pathway enrichment analysis revealed that the Sirtuin Signaling pathway is central to the alcohol consumption response for both LD and VHD. Although some molecules overlapped between LD and VHD with upregulation of SIRT5 and ACSS2, and downregulation of AKT1, MYC, SP1, and TRIM28, there were significant differences in the majority of pathway molecules. Differences included 32 differentially abundant pathway molecules in LD, including downregulation of SIRT2, with only 13 in VHD, including downregulation of SIRT3. Putative epigenetic regulation of Sirtuin Signaling also differs significantly, with LD revealing four genes (12.5%) containing differential DNAm sites and five (16%) targeted by inverse miRNAs, whereas VHD showed six genes (46%) with differential DNAm sites and two (15%) targeted by inverse miRNAs. More specifically, FOXO1, SIRT2, and seven molecules in the NADH:Ubiquinone Oxidoreductase Core Subunit of the mitochondrial membrane respiratory chain NADH dehydrogenase (Complex I) are downregulated in LD but not in VHD. FOXO1 is a transcription factor that promotes gluconeogenesis in hepatocytes (Yuan et al., 2008, 2009) and regulates autophagy with oxidative stress (Zhao et al., 2010). Little is known about SIRT2, but evidence indicates it plays roles in cell differentiation, apoptosis, and is associated with multiple types of cancer. The NADH Core subunit catalyzes electron transfer from NADH through the respiratory chain (Lebon et al., 2007). In addition, overlapping genes MYC and SIRT5 were associated with differential DNAm sites in VHD but not LD, again emphasizing both genetic and epigenetic differences between LD and VHD. There is very little information in the literature on the function of SIRT2, SIRT3, and SIRT5, making it difficult to interpret the impact of these differences on Sirtuin Signaling. However, the differences in directionality and epigenetic alterations in Sirtuin Signaling suggest LD NHP, but not VHD NHP, hepatic energy management is responsive to ethanol consumption.

Integrated omic analysis also revealed networks regulated by MYC as central to the response to alcohol consumption in both LD and VHD. Comparison of the LD and VHD MYC networks showed more striking differences than Sirtuin Signaling. Although there were two common upregulated and 31 common downregulated genes and proteins, there was no overlap with putative epigenetic regulation for these common genes and proteins. In LD, 15% of the genes contained differential DNAm promoter sites, with 41% of the genes in the VHD MYC network. In addition, 32% of the genes in the LD MYC network were targeted by differential inverse miRNAs, with only 11.5% in the VHD MYC network.

Additional differences between VHD and LD response to ethanol consumption included inhibition of FGF21 within the ATF4 network in VHD NHP, but not LD. FGF21 is a hormone and antioxidant that mediates its effects by activating noradrenergic neurons in the locus coeruleus region of the brain, which regulates arousal and alertness. An alcohol intoxication study in mice found that Fgf21 was strongly induced by ethanol in the liver, similar to humans, and stimulates arousal from intoxication, which is inhibited in Fgf21 liver‐specific knockout mice. These results suggest that FGF21 acts on the liver–brain axis and protects against ethanol‐induced intoxication (Choi et al., 2023). Our findings that FGF21 does not change in LD, but decreases in VHD with initial alcohol consumption, again highlight the significant molecular differences between LD and VHD.

Canonical pathway annotation of genes and proteins in the LD MYC network showed enrichment of Acute and Chronic Inflammatory Processes, Sirtuin, mTOR, Hormone, Insulin Receptor, Telomerase, Hepatic Fibrosis, and Autophagy Signaling. Annotation of the VHD MYC network showed multiple processes related to cancer, hypertrophy, and senescence. The VHD MYC network, unlike the LD MYC network, is not enriched for molecules in the Sirtuin Signaling pathway, indicating additional molecular differences. Comparison of LD and VHD MYC networks indicates that although MYC is a key regulator for alcohol response in both groups, the molecular changes in LD appear to be related to nutrient sensing and hormonal regulation, whereas molecular changes in VHD appear to be primarily related to dysregulation of cell proliferation and differentiation, indicating that downstream consequences of MYC signaling differ according to genetic and epigenetic background. Studies in humans and mice show upregulation of MYC with chronic alcohol consumption and associated liver fibrosis (Nevzorova et al., 2016) highlighting differences between acute exposure to alcohol in our study compared with chronic exposure to alcohol in previous studies.

VHD postinduction vs. LD postinduction

Comparison of LD with VHD at the end of the induction phase (after 3 months of uniform alcohol exposure), we found 535 (2.5%) differentially expressed genes and 23 (1.2%) differentially abundant proteins (p < 0.05, Table 2). Pathway enrichment analysis revealed 20 pathways that differed between VHD and LD, but Sirtuin Signaling was not among the pathways (Table S12). Regulatory network analysis resulted in 5 networks that were predicted to be activated. The top‐ranking network regulated by XBP1 contained only 12 targets. In addition, an ESR1 network was significantly enriched but did not have a predicted activation state (Table S13).

Based on our findings of an ESR1 network in VHD versus LD at baseline, and given that the network had the second most molecules of those that were significant postinduction, we further explored network composition and putative epigenetic regulation. The network contained epigenetic regulator SIRT1 (upregulated), and transcription factor FOXO1 (downregulated). Fourteen of the molecules in the network showed abundance values opposite of what is predicted based on the published literature, potentially explaining why such a large regulatory network did not have a predicted activation state. Epigenomic analysis showed 67 % of the molecules in the network contained differential DNAm sites, and 6 (21%) were targeted by inversely, differentially expressed miRNAs (Table S6). These results indicate that there are significant epigenetic differences between VHD and LD at the end of 3 months during which animals drank the same amounts of ethanol.

Liver alcohol response genes compared to genetic variants associated with alcohol behavior‐related traits

To determine whether differentially expressed genes and proteins in LD and VHD NHP livers had previously been associated with alcohol‐related traits in humans, we compared our data to the human Genome Wide Association Study (GWAS) catalog (Buniello et al., 2019). To do so, we generated a list of genes associated with alcohol consumption from the human GWAS catalog using search terms alcohol consumption, alcohol dependence, alcohol use disorder, alcohol co‐dependence, alcohol dependence symptom count, and alcoholism, and identified 1023 single nucleotide polymorphisms in, or proximal to, genes. Merging the GWAS list with differentially abundant genes and proteins from our datasets (for all comparisons described above) we found 35 SNPs in or near 27 genes/proteins. Among these were alcohol dehydrogenase 5 (ADH5, intronic, rs29001570), aldehyde dehydrogenase 1 family member B1 (ALDH1B1, 3’ UTR, rs3043), and aldehyde dehydrogenase 2 family member (ALDH2, missense, rs671). For genes in the Sirtuin Signaling pathway and the MYC network, one SNP (rs72716801‐G) located ~100,000 bp upstream of the MYC gene has previously been associated with alcohol use disorder in humans (Table 3). Although these common genes and proteins are potentially of interest, this is a very small percentage of the differentially abundant genes and proteins observed in this study, further highlighting the power of comparing response to ethanol in naïve animals and selection of animals extremely divergent for their preponderance to later ethanol consumption.

TABLE 3.

Differentially abundant genes and proteins with human GWAS loci associated with alcohol consumption‐related traits.

Gene ID	Protein ID	p‐val	FC	Risk allele	Comparison	Chr	Nucleotide	SNV name	pubMedID
ABI3BP		0.043	−1.56	rs12632235‐T	Base VH v Base L	3	100,817,439	rs12632235	26,365,420
ACSS3	F6T734	0.050	−1.20	rs11114787‐T	Base VH v Base L	12	81,201,920	rs11114787	31,358,974
ADH5	H9Z5A5	0.029	−1.79	rs29001570‐C	Base VH v Post Ind VH	4	99,073,253	rs29001570	28,937,693
ALDH1B1	F7H5N9	0.044	−1.49	rs3043‐?	Base VH v Post Ind VH	9	38,397,357	rs3043	31,959,922
ALDH2	A0A5K1TZ33	0.023	−1.45	rs671‐?	Base L v Post Ind L	12	111,803,961	rs671	31,959,922
BRAP		0.031	−1.81	rs11066001‐?	Post Ind L v Base L	12	111,681,366	rs11066001	31,591,379
BRAP		0.031	−1.81	rs3782886‐C	Post Ind L v Base L	12	111,672,684	rs3782886	30,940,813
DCC		0.027	5.55	rs768048‐C	Post Ind VH v Post Ind L	18	52,759,027	rs768048	21,529,783
ELAVL4		0.018	5.08	rs7517344‐A	Post Ind VH v Post Ind L	1	50,246,288	rs7517344	31,358,974
GCKR	F7BYJ8	0.025	1.68	rs11127048‐G	Base L v Post Ind L	2	27,529,595	rs11127048	28,937,693
GCKR	F7BYJ8	0.025	1.68	rs1260326‐?	Base L v Post Ind L	2	27,508,072	rs1260326	31,959,922
GCKR	F7BYJ8	0.025	1.68	rs1260326‐C	Base L v Post Ind L	2	27,508,072	rs1260326	30,679,032
GCKR	F7BYJ8	0.025	1.68	rs4665985‐C	Base L v Post Ind L	2	27,531,010	rs4665985	28,485,404
GCKR	F7BYJ8	0.025	1.68	rs780094‐T	Base L v Post Ind L	2	27,518,369	rs780094	28,485,404
MAPT		0.019	−6.40	rs62055546‐A	Base VH v Base L	17	45,887,200	rs62055546	30,643,258
MAPT		0.019	−6.40	rs62062288‐A	Base VH v Base L	17	46,019,186	rs62062288	30,336,701
MYC		0.024	−26.22	rs72716801‐G	Post Ind VH v Base VH	8	127,636,480	rs72716801	30,940,813
NEIL2		0.009	1.82	rs804292‐G	Post Ind VH v Base VH	8	11,786,405	rs804292	23,942,779
NQO1	F7CGF2	0.012	1.63	rs1800566‐A	Base VH v Base L	16	69,711,241	rs1800566	30,679,032
PDE4B		0.013	2.72	rs2310752‐A	Post Ind VH v Base VH	1	65,926,721	rs2310752	31,358,974
PPP2R2B	F6VS43	0.022	1.89	rs1864982‐A	Base L v Post Ind L	5	146,941,259	rs1864982	19,581,569
RORB	A0A1D5QER8	0.041	−1.35	rs77123275‐T	Base L v Post Ind L	9	74,459,180	rs77123275	30,643,258
SECISBP2L		0.035	2.58	rs4775792‐T	Post Ind L v Base L	15	48,995,880	rs4775792	30,643,258
SORL1		0.021	−1.85	rs10790449‐C	Base VH v Base L	11	121,630,696	rs10790449	31,998,841
SORL1		0.021	−1.85	rs485425‐C	Base VH v Base L	11	121,674,275	rs485425	31,358,974
TCF12		0.043	1.94	rs11071294‐A	Post Ind L v Base L	15	56,785,359	rs11071294	30,643,258
TCF4		0.020	−2.03	rs9320010‐A	Post Ind VH v Post Ind L	18	55,386,665	rs9320010	31,358,974
TCF4		0.020	−2.03	rs1788030‐T	Post Ind VH v Post Ind L	18	55,377,966	rs1788030	30,643,258
TENM2		0.044	29.28	rs10078588‐A	Post Ind L v Base L	5	167,389,170	rs10078588	31,358,974
TNS3		0.042	−1.55	rs334524‐A	Post Ind VH v Base VH	7	47,526,558	rs334524	29,460,428
TRMT10C		0.015	−1.66	rs142338804‐A	Post Ind VH v Base VH	3	101,487,318	rs142338804	30,643,258
TUFM	H9FZ92	0.019	1.45	rs113079736‐A	Base L v Post Ind L	16	28,848,861	rs113079736	30,643,258
USP28		0.032	−2.64	rs1713675‐A	Post Ind VH v Base VH	11	113,790,626	rs1713675	30,643,258
XPNPEP1	F7DGY1	0.018	0.79	rs142488468‐C	Base VH v Post Ind VH	10	108,996,100	rs142488468	30,643,258
ZNF124		0.025	−1.57	rs3738443‐A	Post Ind VH v Base VH	1	247,184,886	rs3738443	21,314,694

Abbreviations: Base, baseline; Chr, chromosome; FC, fold change; L, light drinkers; Post Ind, postinduction; p‐val, p‐value; SNV, single nucleotide variant; VH, very heavy drinkers.

Genetic variants in the cohort

To determine whether genetic variants in human alcohol consumption associated genes (GWAS above) segregated in our cohort, we identified genetic variants in our cohort within the human GWAS alcohol gene list. To do so, we merged gene IDs from the human GWAS list with whole exome sequence data in the mGaP database (Bimber et al., 2019). No risk variants were homozygous for any SNPs in either the LD or VHD groups. Four SNPs, two in ACACB, one in TOR1AIP1, and one in VARS2 were homozygous among all animals in one group and heterozygous among all animals in the other group (Table 4). None of these genes were found in Sirtuin Signaling or the MYC and ESR1 regulatory networks. These results are consistent with our findings of differences in alcohol responsive pathways and networks that are very likely influenced by epigenetic regulatory variants.

TABLE 4.

Differentially abundant genes and proteins with coding region single nucleotide variants that differ between very heavy and light drinkers.

Gene ID	Compare	Ch	Position	Light genotype	Heavy genotype	Ref allele	Minor allele Freq	Type	Impact	Change
ACACB	Post Ind L v Baseline L	11	108,913,454	C/A	C/C	C	0.23	Intergenic	Modifier	C>A
ACACB	Post Ind VH v Base VH Post Ind L v Basel L	11	108,913,469	G/C	G/G	G	0.22	Intergenic	Modifier	G>C
TOR1AIP1	Post Ind L v Base L	1	37,109,093	G/G	G/A	G	0.21	Missense	Moderate	Asp378Asn
VARS2	Post Ind VH v Base VH Post Ind VH v Post Ind L	4	138,999,359	A/A	A/C	A	0.26	Missense	Moderate	Leu737Val

Abbreviation: Ch, chromosome.

CONCLUSIONS

Alcohol consumption is a complex trait influenced by genetics and the environment. Identifying functional variants that influence drinking behavior is challenging in humans due to the inability to control environmental factors, including the unlikelihood of studying alcohol‐naïve subjects followed by controlled consumption of alcohol; whereas, the use of NHPs allows control of environmental factors, which increases our power to detect genetically influenced molecular mechanisms underlying variation in drinking behavior. In addition, the liver is the primary organ metabolizing ethanol, and genetic and epigenetic variation significantly influence variation in ethanol metabolism. The use of NHPs provides the means to better understand the molecular and epigenetic mechanisms and pathways influencing alcohol metabolism in an animal model similar to humans and allows liver biopsy collection prior to ethanol consumption at defined periods when animals are still healthy. Biopsied liver for the current study was obtained from male rhesus macaques that underwent a well‐characterized ethanol self‐administration paradigm (Emond et al., 2012) that has proven validity with known individual differences in alcohol consumption, as well as a normal distribution of drinking phenotypes representative of both extremes of the alcohol drinking spectrum paralleling what is observed in humans (Grant et al., 2008).

A major challenge for identifying variation in molecular mechanisms underlying complex traits is the large sample size typically needed to identify contributing variation. To increase the likelihood of detecting genetic and epigenetic variation influencing the complex trait of ethanol consumption, we used an extreme phenotype approach, in which we selected animals at both ends of the phenotypic distribution (extreme phenotypes), that is, the frequency of alleles that contribute to the trait is enriched in the phenotype extremes, allowing for use of a small sample size to identify novel genes, proteins, and variants, and nominate mechanisms that underlie these differences (Emond et al., 2012).

Analyses of liver biopsies at baseline, before alcohol exposure, showed that hepatic molecular pathways and networks already differed between LD and VHD. In addition, the response to uniform doses of alcohol for 3 months differed between LD and VHD, including putative epigenetic regulatory mechanisms. Notably, Sirtuin Signaling and a MYC regulated network were significantly enriched for differentially abundant genes and proteins in both LD and VHD response to alcohol consumption. MYC has previously been associated with alcohol use disorder (Kranzler et al., 2019). However, the composition and putative epigenetic mechanisms regulating the networks differed with a greater response of miRNAs than DNAm in LD, and a greater response of DNAm than miRNAs in VHD.

Histological and blood analyses showed no evidence of liver damage in either group at baseline and at the end of the induction phase, indicating that the observed differences were not due to liver pathology. Our findings suggest that genetic variation and epigenetic mechanisms regulating alcohol consumption response in naïve animals differ between animals that later become LD and VHD, influencing liver metabolism of alcohol. Hepatic molecular differences prior to alcohol exposure, as well as differences in initial response to alcohol exposure, may predict eventual alcohol consumption between LD and VHD NHPs. The observed molecular differences between VHD compared with LD suggest a dysregulated response similar to the metabolism of alcohol by individuals with aldehyde dehydrogenase 2 deficiency (Edenberg, 2007). Additional studies in these animals at later time points are required to determine whether the VHD hepatic response is maladaptive or adaptive.

Limitations

In this exploratory study, we included three LD and three VHD rhesus macaques. Although these animals were selected from the extremes of the 71 animals that have been included in the study to maximize the likelihood of identifying molecular differences, the small sample size limits statistical power. In addition, the current study included only males; future studies must include females to determine the extent to which our findings are sex‐specific.

SUMMARY

Our findings from the comparison of VHD and LD at baseline (alcohol naïve), and the comparison of initial uniform exposure to alcohol with baseline, show molecular differences that are not apparent from human genetic studies to date. In addition, we show marked differences in epigenetic responses with miRNAs appearing to play a major role for LD response, and DNAm for VHD response. Histological analyses of liver biopsies showed no differences between LD and VHD after 3 months of uniform alcohol consumption, supporting the hypothesis that the observed molecular differences are due to variation in genetic and epigenetic regulation, not liver pathology. The extensive metabolic and genetic similarities of this Old World Primate with humans suggest that findings from these investigations will be translatable to humans. Identification of genetic and epigenetic variation that predicts drinking behavior in alcohol naive individuals would provide a foundation for early identification and individualized prevention of alcohol use disorder, potentially reducing the prevalence and impact of alcohol use disorder. Further investigations are required to validate the differing genetic and epigenetic mechanisms regulating metabolic differences in the livers of VHD versus LD NHP, determine whether the VHD response is adaptive or maladaptive, and identify blood signatures associated with liver molecular variation as a first step for translation to prediction of alcohol use in humans.

CONFLICT OF INTEREST STATEMENT

The authors have no conflict of interest to declare.

Supporting information

Table S1

Table S2

Table S3

Table S4

Table S5

Table S6

Table S7

Table S8

Table S9

Table S10

Table S11

Table S12

Table S13

Cox, L.A. , Daunais, J.B. , Howard, T.D. , Li, G. , Puppala, S. , Chan, J. et al. (2025) Liver molecular networks associated with drinking behavior in nonhuman primates. Alcohol: Clinical and Experimental Research, 49, 2470–2484. Available from: 10.1111/acer.70162

DATA AVAILABILITY STATEMENT

The data that supports the findings of this study are available in the supplementary material of this article.