Proteomics reveals profound metabolic changes in the alcohol use disorder brain

Changes in brain metabolism are a hallmark of Alcohol Use Disorder (AUD). Determining how AUD changes the brain proteome is critical for understanding the effects of alcohol consumption on biochemical processes in the brain. We used data-independent acquisition mass spectrometry proteomics to study differences in the abundance of proteins associated with AUD in pre-frontal lobe and motor cortex from autopsy brain. AUD had a substantial effect on the overall brain proteome exceeding the inherent differences between brain regions. Proteins associated with glycolysis, trafficking, the cytoskeleton, and excitotoxicity were altered in abundance in AUD. We observed extensive changes in the abundance of key metabolic enzymes, consistent with a switch from glucose to acetate utilization in the AUD brain. We propose that metabolic adaptations allowing efficient acetate utilization contribute to ethanol dependence in AUD.


Introduction
The effects of alcoholism are an ongoing risk to both mental and physical health, and to the wellbeing of society. The World Health Organisation estimates that 76.3 million people worldwide are diagnosable with alcohol use disorder (AUD) 1 , and it is estimated that the dangerous consumption of alcohol results in 2.5 million deaths worldwide every year and is the third leading risk factor for poor health globally. A more detailed understanding of the causes and consequences of AUD is relevant to mitigating its impact on society.
The detrimental cognitive effects caused by chronic alcohol consumption have been well documented 2,3 . However, the underlying molecular mechanisms and neurochemistry of alcoholism remain elusive. In this context, systems biology has been a powerful tool to investigate the molecular changes in the brain associated with AUD. Changes in brain gene expression have been studied with DNA microarrays 4 and RNAseq 5,6 , which found a multitude of genes and pathways that were associated with tolerance, dependence, and neurotoxicity in chronic alcohol abuse. In particular, there was a marked and coordinated reduction in mRNAs coding for myelin proteins in the frontal lobe in AUD patients, suggesting that chronic alcohol abuse either directly or indirectly affects myelin-related genes, or that a loss of expression may be due to oligodendrocyte susceptibility to ethanol 4 . This is consistent with alcohol-related neuropathology, where demyelination is observed in the pre-frontal lobe 7,8 .
Protein abundance does not necessarily correlate with transcript expression 9 , especially at a tissue level in a complex and heterogeneous organ such as the brain, which has a multitude of distinct cell types [10][11][12] . It is recognized that the brain and circulating proteome may therefore also help identify risk factors, biomarkers, and underlying causes of AUD 13 . Previous proteomic studies of the brain of human alcoholics have used two-dimensional gelelectrophoresis and mass-spectrometry [14][15][16] . One study of the dorsolateral pre-frontal lobe found a lower abundance of thiamine-dependent proteins, as well as enzymes involved in glycolysis and the TCA cycle in alcoholic patients 16 . A more recent study identified reduced abundance of creatinine kinase chain B, NADH 2 ubiquinone, fructose-bisphosphate aldolase C, and glyceraldehyde-3-phosphate dehydrogenase, proteins central to cellular metabolism 14 . While these studies suggest that AUD is associated with diverse changes to the brain proteome, this has not been tested with modern mass spectrometry proteomics. Here, we used sequential window acquisition of all theoretical mass spectra (SWATH-MS) proteomics to mine the deep brain proteome and identify proteomic correlates of AUD.

Materials and Methods
Tissue collection and storage All brain tissue samples were provided by the Queensland Brain Bank (QBB), School of Chemistry and Molecular Biosciences, The University of Queensland, with informed consent from the next of kin. QBB is the Brisbane node of the Australian Brain Bank Network, which is supported by the National Health and Medical Research Council. Post-mortem histopathological examination was conducted by qualified pathologists, and tissue samples were treated as previously described 17 . The Medical Research Ethics Committee of The University of Queensland approved the study (Clearance N° 0000105).

Case selection
Clinic information including age, post-mortem interval, brain weight, and cause of death were retrieved from the QBB. Tissue from pathologically affected pre-frontal lobe and relatively spared motor cortex were sectioned in 0.5 g pieces. AUD was defined by National Health and Medical Research Council/WHO criteria as individuals who consumed greater than 80 g of ethanol per day for most of their adult lives. Controls either did not consume alcohol or were social drinkers who consumed less than 20 g of ethanol per day on average. Five males were selected for control cases and six males for the AUD cases, all from North-European descent. Age, post-mortem interval (PMI), and brain weight were matched as closely as possible in accordance with brain and case availabilities (Table 1). Alcoholic subjects were confirmed to not have Wernicke-Korsakoff syndrome 3 .

Whole tissue protein extraction
Whole tissue soluble protein extractions were prepared from the dorsolateral prefrontal gyrus of the pre-frontal lobe and the primary motor cortex of controls and alcoholic cases. The sectioned tissue pieces (0.5 g) were rapidly thawed in 0.32 M sucrose at 37 °C. Tissue was then immediately cooled to 4 °C on ice, weighed and homogenised in 10 volumes of ice-cold distilled water with a motor-driven Teflon-glass homogeniser. The homogenate was centrifuged at 1500 rcf for 5 min at 4 °C. Protein concentration was estimated via Lowry assay and adjusted to 3 mg/mL.

Mass spectrometry proteomics
Peptides were identified by LC-MS/MS using a Prominence nanoLC system (Shimadzu) and a TripleTof 5600 mass spectrometer with a Nanospray III interface (SCIEX) essentially as previously described 21 . Approximately 2 µg of peptides were desalted on an Agilent C18 trap (300 Å pore size, 5 µm particle size, 0.3 mm i.d. × 5 mm) at a flow rate of 30 µL/min for 3 min and then separated on a Vydac EVEREST reversed-phase C18 HPLC column (300 Å pore size,

Data analysis
Proteins were identified from DDA data using ProteinPilot 4.1 (SCIEX transformed normalized abundances of proteins with significant differences in abundance between groups with Cluster 3.0 using the average linkage method and Pearson's correlation method for similarity. Trees were generated using JavaTreeview 1.1.6.

Functional Enrichment analysis
Differentially abundant proteins (p < 10 -5 ) were subjected to gene ontology (GO) term enrichment analysis using DAVID Bioinformatics resources 26,27 . Significance was determined with a Benjamini-Hochberg adjusted p-value of < 0.05. The list of differentially abundant proteins that were identified in the AUD pre-frontal lobe was analysed using the protein database STRINGv10 database (http://string-db.org) to generate protein-protein interaction scores using default parameters 28,29 . The protein-protein interaction scores were downloaded and transferred to Cytoscape for visualisation 30 .

Results
Pathophysiological sex-related differences in AUD patients have been previously reported 31,32 , and drinking behaviour varies between ethnic groups 33 . We therefore chose to exclusively study the effects of AUD in males of Northern European descent to limit confounding effects.
We chose to study whole-tissue extracts as a means to glean generalised insights into the proteomes of diverse cell types and tissues.

Protein Identification
We identified proteins from DDA LC-MS/MS of high pH fractionated pooled samples and unfractionated individual samples of trypsin-digested whole tissue protein extractions from pre-frontal lobe and motor cortex from human brain of AUD patients and controls. We then used SWATH-MS to measure the abundance of proteins in each brain region in each individual.
A total of 614 unique quantifiable proteins were identified and measured with a critical local false discovery rate of < 0.01 (Supplementary Table 1).

AUD has a profound effect on the overall soluble brain proteome
It has been proposed that molecular and tissue changes caused by AUD occur primarily in the pre-frontal lobe, with the motor cortex remaining comparatively unaffected 34,35 . We therefore compared the proteome of the pre-frontal lobe in AUD patients and controls, and also compared the proteome of the motor cortex in AUD patients and controls. In addition, we compared the proteome of the pre-frontal lobe and the motor cortex in both controls and in AUD patients.
All comparisons showed large and significant differences in abundance for a subset of the quantifiable proteins (Fig. 1, Supplementary Tables 2 -5). The proteome of the pre-frontal lobe showed major differences in AUD, with 70 proteins having significantly lower, and six proteins significantly higher abundance than in the controls (Fig. 1A, Supplementary Table 2). The motor cortex also differed substantially, with 50 and seven proteins detected with significantly higher and lower abundances than in controls, respectively (Fig. 1B, Supplementary Table 3). Statistically significant differences in both of these comparisons appear to be skewed towards proteins with lower abundance in AUD patients, suggesting AUD causes a reduction in the abundance of selected proteins in both the pre-frontal lobe and motor cortex. However, we note that our proteomics workflow measures relative, not absolute, protein abundance. We also observed differences in the proteomes of the pre-frontal lobe and motor cortex in both controls and AUD patients (Fig. 2C, D, Supplementary Table 3, 4). We expected to detect differences in these proteomes, as these brain areas are responsible for different cognitive and physiological processes. Surprisingly, it appeared that AUD had a larger effect on the proteome than brain region ( Fig. 2A). To investigate this in more detail, we performed hierarchical clustering on the 126 proteins which showed statistically significant differences in abundance between AUD and controls or between brain regions. This analysis showed that the data clustered first on the basis of AUD patients versus controls, and next between pre-frontal lobe and motor cortex (Fig. 2). Together, this showed that AUD had a profound effect on the overall soluble brain proteome exceeding the inherent differences between brain regions.  Supplementary   Table 6.

Pathway analysis identifies differences in central metabolism in AUD patients
To discover higher-order pathways and processes altered in the brains of AUD patients, we performed GO term enrichment analysis on proteins that were significantly different in abundance. This showed that proteins in the pre-frontal lobe with lower abundance in AUD patients were strongly associated with glycolysis and gluconeogenesis (P < 0.05). No significant GO term enrichments were found in any other comparison. To identify additional molecular processes altered in the brains of AUD patients, we performed pathway analysis using STRING, and obtained an interaction map of proteins with significant differences in abundance in the AUD pre-frontal lobe (Fig. 3). This interaction map showed strong connectivity for proteins involved in glycolysis (PFKAP and G3P), with additional links to proteins of the Ras superfamily (RAP1B and RAB10), and proteins involved in oxidative phosphorylation (NDUS8 and AATC), the cytoskeleton (TBA8), and neurotoxicity (PRDX2 and AATM) (Fig. 4).
As our cohort of AUD patients was older than the controls, we performed correlation analysis of subject age and normalised protein abundance for each selected protein that was significantly different in abundance between AUD patients and controls. None of the proteins in the prefrontal lobe with significantly different abundance in AUD compared to controls were found to be correlated with age (Supplementary Table 7), confirming that the differences in protein abundance we observed in the pre-frontal lobe were dependent on AUD status. Some agedependent changes in protein abundance were observed in the motor cortex (Supplementary Table 7).

Discussion
We used SWATH-MS to compare the proteomes of pre-frontal lobe and motor cortex of AUD patients and controls. While we detected differences in the proteomes of pre-frontal lobe and motor cortex in both controls and AUD patients, the AUD-dependent changes in regionspecific proteomes were substantially larger ( Fig. 1 and 2). This emphasizes the profound effects on brain physiology associated with AUD.

Alcohol use disorder perturbs central brain metabolism.
GO term enrichment analysis enabled us to identify nine enzymes involved in glycolysis that were less abundant in AUD patients in the pre-frontal lobe. Many of these proteins were also less abundant in AUD patients in the motor cortex. Phosphofructokinase (PFKAP), the first committed step in glycolysis 36 , was the protein with the largest decrease in abundance in the pre-frontal lobe in AUD compared to controls ( Fig. 2 and 4, Supplementary Table 2). Several other glycolytic enzymes were also less abundant in AUD patients, including the pyruvate dehydrogenase complex, which catalyses the production of acetyl-CoA from pyruvate required for the TCA cycle 37 . Together, our data suggest that there is a substantial reduction in glycolytic flux in AUD. This is consistent with previous findings in human imaging studies that indicated that low doses of alcohol significantly decrease baseline brain glucose metabolism 38,39 , and that alcohol misuse causes a reduction in glucose metabolism 40 .
Alcohol is primarily metabolized in the cytosol of hepatocytes by alcohol dehydrogenase to acetaldehyde 41 , which is further oxidized to acetate 42 . This excess acetate then enters the blood circulation. Previous studies have reported an increased uptake of acetate into the alcoholic brain, where it is metabolised in astrocytes to acetyl-CoA to enter the TCA cycle 43 . While our proteomic data identified all enzymes of the TCA cycle, none of these showed significant changes in abundance in AUD (Supplementary Table 2). Consumption of acetyl-CoA in the TCA cycle results in a net production of NADH 44 . Increased acetate metabolism in the AUD brain would therefore be expected to increase the local NADH:NAD + ratio. This affects many metabolic processes, and in particular glycolysis, which requires NAD + . Together, our data are consistent with AUD being associated with (i) a switch in energy utilisation in the brain from glucose towards acetate, (ii) enhanced metabolism of acetate through the TCA cycle, (iii) an increase in cellular NADH, with (iv) a concomitant severe down-regulation of glycolysis.
These changes in cellular metabolism may contribute to alcohol dependence through a shift to a metabolic preference for ethanol-derived acetate.

Microtubules, vesicular trafficking, and the cytoskeleton
GTP-binding proteins from the RAS superfamily are critical components of diverse intracellular signalling networks 45 . Rap-1b and Rab-10 were significantly less abundant in both the AUD pre-frontal lobe and motor cortex, while Rab-7a and Rab-1b were significantly less abundant in the AUD pre-frontal lobe (Fig. 4, Supplementary Table 2). Rap-1b is a GABAβ receptor binding partner, responsible for initiating the signalling cascade that mediates GABAβ receptor cell surface expression via receptor recycling 46 . Normal functioning of Rap-1b results in slow and long-lasting inhibitory action via the GABA neurotransmitter 46 . The observation that Rap-1b was less abundant in AUD in both brain areas suggests that the AUD brain has a reduced capacity to modulate inhibitory outputs, in particular in the mesocorticolimbic dopamine system responsible for addiction 47 . GABAβ is a potential treatment target for AUD, as demonstrated by the use of the receptor agonist baclofen in reducing craving and withdrawal symptoms 48 . Rab-10 is proposed to be involved in neurite outgrowth during the processes of axonal growth, pathfinding, regeneration, and dendritic growth 49 . These processes require extensive membrane trafficking events for normal functioning [50][51][52] . Reduction in such intrinsic axon growth ability may suggest that the AUD brain is deficient in integration of synaptic inputs. The retromer is a protein complex associated with endosome to trans-Golgi network trafficking 53,54 . Rab-7a is a retromer effector responsible for cargo-recognition for the retrograde pathway for transmembrane receptor recycling 55 . Down-regulation of Rab-7a could therefore disrupt transmembrane receptor recycling to the pre-and post-synaptic membranes, adversely affecting the delicate neurochemistry of the pre-frontal lobe.
Microtubules are a fundamental cytoskeletal component of neurons that are important in a multitude of cellular and developmental processes. The core microtubule structure consists of heterodimers of a and b tubulin isoforms. Our data showed that tubulin alpha chains (TUBA1A, TUBA4A, and TUBA8) were less abundant in AUD in both the pre-frontal lobe and motor cortex, and that beta chain TUBB2A was less abundant in AUD in the pre-frontal lobe (Fig. 4, Supplementary Table 2). Loss of cytoskeletal protein is associated with selective neuronal tissue loss and tissue pathology observed in alcoholics 34 .

Excitotoxicity and oxidative damage in the AUD brain
Excitotoxicity is a common pathological hallmark of neurodegenerative diseases. This process results in a calcium overload in the mitochondria, triggering compensatory respiration and the generation of reactive oxygen species (ROS) 56,57 . Long-term ROS irreversibly damage the electron transport chain and initiate apoptotic and necrotic events within the cell 58 . Two members of the antioxidant peroxiredoxin family effective against hydrogen peroxide and peroxynitrite, peroxiredoxin 2 and 3, were significantly decreased in abundance in both the pre-frontal lobe and the motor cortex in patients with AUD. Enzyme kinetic experiments have demonstrated that hyperoxidation of peroxiredoxin 1 causes rapid inactivation of the enzyme, a phenomenon which is conserved in peroxiredoxin 2 and 3 59,60 . Peroxiredoxin 2 is more susceptible to hyperoxidation due to slower disulphide formation 61 . This is consistent with our data, as peroxiredoxin 3 had a smaller AUD-dependent fold-change than peroxiredoxin 2 in both brain locations. Our data are therefore consistent with increased oxidative stress in AUD, causing chronic overoxidation and degradation of peroxiredoxins.
A key example of neuro-excitotoxicity is glutamate toxicity 62 . Chronic alcoholism has been associated with hyperglutamatergic activity in the mesocorticolimbic brain areas 63 . Alcohol non-competitively inhibits NMDA receptors, the binding targets of glutamate 64 . The enzymes cytoplasmic aspartate aminotransferase (AATC; also known as GOT1) and glutamine synthetase (GLUL) are important regulators of glutamate levels and have been implicated in brain neuroprotection 65 . Our data showed a decrease in AATC and GLUL abundance in the AUD pre-frontal lobe (Fig. 4), consistent with increased levels of glutamate and enhanced excitotoxity in the AUD brain.

Epigenetic and post-translational modifications
The differences we observed in the proteomes in the brains of subjects with AUD may be associated with epigenetic regulation and/or protein post-translational modifications (PTM) 66 .
Epigenetic changes have previously been implicated in alcohol abuse, where changes in NADH:NAD + ratios affect the activity of proteins responsible for gene expression and silencing 67,68 . Increased DNA methylation has been associated with a decrease in synaptic proteins in the pre-frontal lobe of rats with alcohol-induced behaviour 69 . In addition, acetyl-CoA is used in histone acetylation by histone acetyltransferases to regulate gene expression 70 , and also modifies diverse other proteins with acetylation to regulate their functions. The PTM of acetylation is dynamic and reversible, but changes in the NADH:NAD + ratio associated with AUD would be expected to perturb the balance of this modification. Alcohol-induced acetylation has been shown to affect several proteins 71 . GAPDH has emerged as an important regulator of metabolism, with acetylation at multiple lysine residues potentially affecting its activity 72 . It has also been reported that lysine acetylation of a tubulins is increased in alcoholic cases compared with controls 34 . It is therefore possible that an increased NADH:NAD + ratio and acetyl-CoA concentration in the AUD brain cause inappropriate increases in DNA methylation and protein acetylation that systemically perturb translation and directly disrupt the functions of diverse proteins.

Conclusion
Using SWATH-MS proteomic analyses we identified large and significant AUD-dependent differences in the abundance of proteins in pre-frontal lobe and motor cortex. We observed profound changes in metabolic enzymes consistent with a switch from glucose to acetate utilization in the AUD brain. The proteins with differences in abundance in AUD may be useful as biomarkers of disease. Further studies are necessary to clarify the mechanisms underlying these differences, whether they are causes or effects of AUD, and the potential role of DNA methylation and/or lysine acetylation in regulating these changes. We also propose that metabolic adaptations allowing efficient acetate utilization contribute to ethanol dependence in AUD.