Our genome-wide DNAm analysis of the alcohol-naïve dlPFC-A46 is the first study to be conducted using a biopsy of the brain in living individuals prior to their enrollment in a chronic alcohol self-administration protocol. This unique study resulted in two major findings. First, it provides a molecular snapshot of the alcohol-naïve dlPFC, with the DNAm profile, mapped genes, and associated signaling pathways that distinguish individuals with different future drinking behavior, and that may contribute to risky drinking. Second, it identified a subset of 10 DMRs that strongly predicted the amount of ethanol that animals would go on to consume during 6 months of open access. Importantly, the same set of DMRs showed moderate predictive ability in an independent cohort of macaques. However, we acknowledge several limitations of this prediction analysis. The nearly perfect prediction we see in the discovery set is most likely the result of overfitting given the sample size. In addition, the discovery set contained only male samples, while the test set included males and females. Thus, if the predictive DMRs (or a subset of them) are male specific, their performance will be limited when predicting the drinking levels of females. Even with these limitations, the predictive set proved able to predict drinking risk in the independent dataset. Our next goals are 1) to increase the sample size of the discovery set and test set to confirm, refine or add predictive DMRs and 2) expand the study to include an even mixture of males and females in the discovery and test sets to find DMRs that predict drinking levels regardless of sex and simultaneously determine if the predictive DMRs are sex specific.
Alcohol-naïve individuals display a vulnerable neural signaling pathway network in the dlPFC.
Chronic alcohol use gradually reduces cognitive control, deteriorating the individual’s ability to inhibit perseverative responses and adapt to changes in environmental contingencies65. The dlPFC is a component of the central executive network66 and alcohol use compromises its proper function67,68. Studies have shown that alcohol impairs neuroplasticity in the dlPFC69. Furthermore, modulation of neuroplasticity in the dlPFC through repetitive transcranial magnetic stimulation shows promise as a treatment for AUD70,71. In addition, in alcohol-naïve rhesus macaques, low cognitive flexibility assessed with a set-shifting procedure was predictive of future classification as a heavy alcohol drinker1. The results of the present study used samples from the same macaques included in Shnitko et al.1, and support the hypothesis that individual differences in dlPFC function in an alcohol-naïve state and risk for future chronic heavy drinking could be due, at least in part, to underlying unique epigenetic signatures establishing a vulnerable brain state. In all, these results suggest that pre-existing individual variability in dlPFC epigenetic footprint and function, impacting neural-specific signaling pathways, reduces cognitive flexibility and establishes a predisposition towards habitual behaviors, contributing to future risk for heavy drinking. Similar findings have been reported in the few available human longitudinal studies, where decreased academic performance was associated with future alcohol use72–74.
Among the neural-specific signaling pathways, three developmental functions, such as neurogenesis, CNS neuron differentiation, and forebrain development, indicate that our study could be capturing individual differences in cortical maturation processes, which is active during late adolescence in the primate brain75. Accordingly, and within individuals of similar chronological age (~ 5 years old, approximately the equivalent to 20 human years), such differences could be engrained in an inherited epigenetic code that preconditions individuals to a life-long battle with AUD upon a first encounter with alcohol intoxication. We cannot exclude the possibility that we are capturing different stages of the cortical maturation process, as it is expected individual variability in reaching a matured brain. Even in this scenario, our results indicate that exposure to ethanol while the cortex is immature poses serious risk for future development of AUD, as suggested (see review76). In addition to exclusively developmental signaling pathways, we identified enrichment in modulation of synaptic transmission, regulation of membrane potential, neurotransmitter and vesicle-transport, and glutamate receptor signaling as relevant molecular signatures in the dlPFC of vulnerable individuals. It is not surprising that additional enrichment is detected for signaling pathways with global functions that are required for the neural processes described above. These include regulation of the cytoskeleton and its components, metabolic modulation and transcriptional and translational regulation.
The rhesus macaques included in this study were all born and raised at ONPRC under the same environmental conditions (i.e. diet, enrichment in corrals, same experimental conditions), and never enrolled in another experimental procedure before their recruitment to this study. Nonetheless, in this controlled environment, they lived in corrals with large social groups, and exposed to diverse social interactions through their life. Thus, it is possible that, in addition to epigenetic inheritance77, early-life social stressors could have contributed to shaping their epigenome prior to their enrollment in this study.
Alterations in glutamate receptor signaling contribute to vulnerability for future risky drinking.
Analysis of the hub genes for the alcohol-naïve network pointed to a critical and central role of the glutamate receptor signaling pathway. These include multiple members of glutamate receptors (metabotropic and ionotropic; GRIN1, GRIN2A, GRIN3A, GRM2, GRIK3), the dopamine receptor (DRD1), multiple members of the voltage gated calcium channels that control neurotransmitter release (CACNA2D2, CACNA2D3, CACNG6, CACNB2, CACNB4, CACNA1C, CACNA1H and CACNA1I), important modulators of intracellular calcium dynamics in neurons (PVALB, RYR1), and numerous synaptic proteins (i.e. SHANK2, NRXN2, NRXN3, HOMER, DLGAP2). As expected, a combination of hypo and hypermethylated DMRs were associated with increasing amounts of alcohol. We currently do not know the specific functional implications of these DMRs. However, and based on their genomic location mapping to promoters and intragenic introns/exons, they likely contribute to modulating alternative transcript expression, as we have previously shown45,46. Follow up studies to clarify their role in modulating transcript and protein function are underway.
Further strengthening the relevance of the glutamate receptor pathway in establishing a vulnerable brain, one of the central genes in this pathway, SHANK2, is also one of the predictive genes identified in this study. Importantly, SHANK2-DMR DNAm levels were strongly correlated with numerous members of this signaling pathway (i.e. CACNA1C, GRIN1). Underscoring the broader significance of the SHANK2-DMR, its DNAm levels were also correlated with genes implicated in neuron differentiation and development, axonal transport, regulation of action potential, ECM organization and synaptic transmission.
SHANK2 (also known as ProSAP1) is an abundant postsynaptic scaffolding protein involved in establishing postsynaptic density of glutamate synapses through a dense network of molecular interactions78–85. SHANK2 loss of function and missense mutations have been extensively implicated in neurodevelopmental and neuropsychiatric disorders, including autism spectrum disorder (ASD), intellectual disability, developmental delay, schizophrenia and alcohol addiction82,85–98. Different mutations in SHANK2 can lead to increased or decreased expression of SHANK2, dysfunctional SHANK2 domains or truncated proteins lacking interaction domains. This can lead to alterations in protein-protein interactions and the organization of the postsynaptic protein network, and ultimately to neuropsychiatric disorders87.
In addition to the role of the SHANK2 variants, epigenetics may have a strong influence on the expression of the SHANK2 variant-mediated phenotypes99–101 by affecting the penetrance of SHANK2 variants, as well as the expression of the different SHANK2 isoforms101. This has been observed for Shank3, where tissue-specific promoter and intragenic DNAm regions with differential DNAm patterns were associated with alternative transcript expression in a brain region- and cell-type specific manner102–104. Furthermore, treatment with the DNAm inhibitor 5-azacytidine in cultured cells resulted in demethylation of SHANK3 CpG islands, and altered isoform-specific expression of SHANK3. Although it is unknown if a similar regulatory mechanism exists for SHANK2, this will be particularly important to discern, because the different SHANK2 isoforms determine the organization of the post-synaptic density zone at different developmental stages and/or different brain regions92,105, have different pre and postsynaptic functions, and the endogenous balanced expression of SHANK2 isoforms is essential for establishing and maintaining higher brain functions such as social and cognitive behaviors106. The different isoforms lack domains involved in protein-protein interactions that bridge glutamate receptors, scaffolding proteins and intracellular effectors to the actin cytoskeleton79,85. An analysis of human tissues showed that expression of SHANK2A (equivalent to the macaque SHANK2-204) is only detected in the brain107. In addition, SHANK2E (equivalent to the macaque longest transcripts, 201–203) was detected in the brain107, contrary to previous findings stating SHANK2E was not expressed in the brain86. Three SHANK2 exons (19, 20, and 23) that code for a region between the PDZ and the proline rich domains, were only detected in brain samples107. Another important finding from this study, was that there was interindividual variability in the proportions of these three transcripts expressed in the brain107; a feature that has been observed for other synaptic proteins108–110. The macaque predictive DMR is located within SHANK2, and ongoing studies are underway to determine if the differential DNAm in this region contributes to the regulation of alternative transcripts in the dlPFC of macaques.
Alcohol-naïve predictive DNAm signals identify several signaling pathways modulating synaptic transmission that define vulnerability for future risky drinking.
Based on the network analysis described above, it is not surprising that most of the predictive genes exert critical functions during cortex development and in the adult brain. These results strengthen the hypothesis that DNAm signals that are inherited or acquired early-in-life could predetermine an individual’s risk for alcohol misuse.
We described above the relevance of the glutamatergic system in vulnerability, with a central role for SHANK2. Interestingly, the DNAm levels of SHANK2 were highly correlated with that of two other predictive genes (brown module), SMAD3 and SLC9A3. Thus, indicating that synaptic transmission, ECM organization and sodium ion transmembrane transport (Fig. 7B) are connected biological functions and relevant in determining vulnerability.
Smad proteins play a broad role in multiple organ systems, including in the brain, regulating ECM organization, axonogenesis and axonal growth, synapse formation and synaptic plasticity. Soluble factors (TGFβ or BMP) induce phosphorylation of Smads by MAPK, GSK, CDKs, CamKII, which results in the activation of the receptor-associated Smad2 and 3 (R-Smads), and the Smad complex translocation to the nucleus where it regulates transcription of target genes111–113 involved in the above mentioned functions.
The development of the nervous system, as well as proper brain function in adults are all closely linked to Smad proteins. Our study of individuals in late adolescence and transitioning into adulthood might be capturing functions of SMAD3 in these two developmental stages. During brain development, neurogenesis involves migration, and settlement, stopping proliferation of neuronal progenitor cells (NPCs). These processes, as well as further differentiation, including dendritic and axonal growth and orientating, synapse formation and myelination, depend on Smad proteins. TGFβ activates Smad3 signaling, diminishing neurogenesis in the hippocampus114 and promoting neuronal differentiation in hippocampal and cortical progenitors115,116. An intact canonical TGFβ/Smad2,3 cascade modulates calcium influx117 in basal forebrain cholinergic neurons, which are essentially involved in the organization of cortical brain structures, learning and memory. In the adult brain, disturbed cellular or subcellular Smad localization and sequestration or deposition are commonly seen in neurodegenerative disorders118,119–121. Furthermore, alcohol treatment increased Smad3 expression in the cortex of mice122, while it was downregulated in the hippocampus after chronic alcohol use123. In addition, a gene co-expression network containing Smad3 was negatively correlated with CIE treatment and blood alcohol levels, suggesting that CIE exposure dysregulates Smad3-related pathways124.
Through tightly controlled alternative splicing events, SMADs encode different isoforms with distinct tissue-specific expression profiles and functions. For instance, R-SMADs with deletions in the linker region or the MH1 domain usually lose the transcriptional activities or have deficient DNA-binding and nuclear translocation ability125. This is critically important for Smad3, which through DNA interaction motif within its MH1 domain125 binds directly to the TGFβ responsive element, CAGA box. However, the mechanism by which alternative splicing is regulated is poorly understood. Our results suggest that the DMR might contribute to the regulation of alternative expression of transcripts lacking the MH1 domain, and thus, altering downstream transcriptional regulation mediated by TGFβ. Our data suggests that different levels of DNAm could be contributing to individual differences in SMAD diversity through alternative splicing regulation. Consequently, Smad-dependent neural processes, such as axonal growth, ECM organization, synapse formation and synaptic plasticity, are altered, rendering individuals more vulnerable for certain behaviors.
An additional predictive gene involved in synaptic transmission and regulation of cognition is the roundabout gene, ROBO3. The role of Robo3 has been extensively studied as a master regulator of axon pathfinding and its role in guiding the commissural axons in the spinal cord and hindbrain during development126 127,128. Interestingly, two Robo3 isoforms, Robo 3.1 and Robo 3.2, are tightly spatially regulated and have opposite functions129, ensuring proper brain wiring. Axon guidance molecules also play critical roles in cell migration130,131 during cortical development132. Radial progenitor cells (RGPCs), which give rise to all cortical neurons and glia, undergo mitosis to either self-renew, differentiate into cortical neurons or generate intermediate progenitor cells (IPCs). IPCs self-renew or differentiate into cortical neurons133–135, serving as the main determinant for cortex expansion in mammals135–140. Gain of function experiments showed that high Robo1/2 expression is sufficient to induce differentiation of progenitor cells into neurons in the mouse cortex141. Given the age of the monkeys in our study, it is possible these mechanisms of cortical expansion are still ongoing, and ROBO3 might be playing a role in cortical expansion. Furthermore, it is possible that innate individual differences in ROBO3 DNAm and function might determine differential cortical organization that could predispose individuals to risky drinking. Additional studies are needed to better understand the regulatory role of these DMRs in ROBO3 expression and its function in modulating synaptic neurotransmission.
BARX2 is a developmentally important homeodomain protein with activator and repressor domains that, upon binding to the promoter of target genes, regulates the expression of neural cell adhesion proteins N-CAM, CREB and ATF2142. N-CAM is important during (1) corticogenesis by orchestrating proliferation, migration and differentiation of NPCs143, (2) synapse development and (3) for mature synaptic plasticity144. Similarly, CREB is also known to be involved in very early stages of cortical neuronal differentiation145, and is a master transcriptional regulator of neuronal plasticity in the adult brain146. Given the position of the DMR (promoter), its hypermethylation could result in downregulation in BARX2 expression, which will need to be explored in the future.
The p53 inducible gene 3 (TP53I3), a downstream molecule of the tumor suppressor p53, is involved in apoptosis and oxidative stress response in cancer studies147,148. This predictive gene was enriched in the nucleotide excision repair pathway, that was part of a larger correlated network (yellow module; Fig. 6E) involving lipid and carbohydrate metabolism, RNA splicing and translation, ECM organization, GPCR, Ca2+ signaling, glutamate receptor, synapse organization and neuron differentiation pathways. At the later stages of cortical development, with active cell proliferation, migration and differentiation, synapse establishment and remodeling, it is not surprising that cell death mechanisms are actively present to coordinate and ensure proper cortical organization and function149.
Alcohol-naïve individual differences in GTPase binding activity as predictors of future risky alcohol intake.
The Ras-like superfamily of G proteins functions as molecular switches to regulate a wide range of primary and versatile cellular functions, including signal transduction, cellular growth, transformation, transport, motility, traffic, and adhesion150. The Ras-like proteins rely on their structural changes between guanosine triphosphate (GTP)-binding conformation and guanosine diphosphate (GDP)-binding conformation, acting as binary signaling switches biochemically, and transmitting upstream signals to downstream molecules151,152. Recent evidence has implicated the Ras family of GTPases, their upstream receptors (i.e. TrkB) and regulators (protein kinases, PKA and PKC), and their downstream mitogen-activated protein kinases (MAPK) and phosphoinositide 3-kinase (PI3K) signaling pathways as key biochemical cascades linking activation of NMDA receptors and calcium influx with phosphorylation and trafficking of AMPA receptors during synaptic plasticity153. Drugs of abuse stimulate signaling pathways that activate these molecules, leading to altered gene transcription and protein synthesis, which contribute to the long-term changes in synaptic function and neural networks that lead to addiction (see154).
Two of the predictive DMRs mapped to two genes with a role in regulating Ras and Rap1 proteins. TBC1D30, a RAB8 GAP protein, regulates membrane precursor-vesicle trafficking to the plasma membrane and localizes in neurites of cultured hippocampal neurons in early developmental stages; while in mature neurons localizes to dendrites, but not in axons155–157. In neuronal cells, downregulation of RAB8 inhibits anterograde formation and transport of vesicles and avoids neurite outgrowth155,158. In addition, RAB8 is also involved in GluA1-AMPA receptors trafficking from the endoplasmic reticulum to the Golgi complex, and their delivery at the post-synapse159,160. RAB8 works in coordination with RAB11 in mediating recycling of endosomes, and activity-dependent insertion or removal of AMPA receptors from the post-synaptic plasma membrane, which triggers synaptic plasticity, in the form of LTP and LTD157,161,162. While additional studies are needed to better understand the regulatory role of this DMR in regulating the expression of the RAB8 alternative transcripts, we hypothesize that individuals at highest risk may have lower levels of one or all transcripts, a unique ratio of the RAB8A and RAB8B isoform, or different levels of these isoforms as compared to those individuals at lower risk. Given the role of RAB8 in mediating AMPA receptor trafficking and availability at the synapse, differences in RAB8 may contribute, together with SHANK2 and other genes within the glutamate receptor signaling, to establishing an innate glutamatergic system that increases risk for addiction.
MEX3B encodes an RNA binding protein (RBP) and functions in the spatial regulation of the small GTPase Rap1163 which plays a role in the regulation of integrin-mediated cell adhesion, cadherin-based cell-cell junction integrity, and in regulating the polarity of migrating neurons in the developing mouse cortex through the control of N-cadherin164–166. MEX3B is also a co-receptor of Toll-like receptor 3 (TLR3)-mediated signaling, promoting its activation167. TLRs play important roles mediating chronic ethanol exposure-linked neuroinflammation168–170. For instance, chronic ethanol intake increases the expression of TLRs, including TLR3, which then induces neuroinflammation through the nuclear factor-κβ171,172. Activated NF-κβ translocates to the nucleus and binds to target gene promoters/enhancers, regulating the expression of various cytokines, adhesion molecules, chemotaxis factors and acute phase protein173. A study showed that ethanol increases the expression of TLR3 and NF-κβ and produces neuroinflammation and anxiety-like behavior in male C57BL/6 mice174. In addition to inducing neuroinflammation, HMGB1, NF-κβ, TLR, and others can contribute to ethanol dependence175. Assuming the hypermethylated DMR leads to reduced MEX3B expression, then individuals with higher risk for heavy alcohol use will have reduced MEX3B, and consequently less activated Rap1 and/or TLR3 activation. This could lead to deficiencies in proper neuronal migration during cortex development and potentially less neuroinflammation at baseline. Additional studies are needed to determine the role of MEX3B in establishing vulnerability for risky drinking.
Alcohol-naïve individual differences in MAPK signaling as predictors of future risky alcohol intake.
DUSP2, a dual-specificity phosphatase, acts as a negative regulator of MAPKs by dephosphorylating both phosphotyrosine and phosphoserine/threonine residues176. Most studies on DUSP2 function have been conducted in the field of immune response and cancer177. The few available studies on DUSP2’s role in the CNS showed Dusp2 mRNA increased expression in forebrain neurons resistant to ischemia, but not in vulnerable neurons, suggesting a protective role of DUSP2 against ischemic stress178–180. Other studies suggest that DUSP2 modulates p38MAPK activity to optimize its functions in synaptic plasticity and cytoskeletal stability 181. Clarifying the functional effects of the DUSP2 DMR are of particular interest given previous suggestions that impaired DUSP2 expression could exacerbate p38MAPK-mediated responses (i.e. neuronal death), contributing to the development of brain pathologies181, including addiction182.
Alcohol-naïve individual differences in superenhancer TMEM44-AS1 as predictor of future risky alcohol intake.
The gene TMEM44 encodes 8 different transcripts, which differ in use of alternative promoters, splicing and 3’ ends. The predictive DMR maps to the last intron of the longer transcripts, and could potentially be involved in regulating alternative 3’ ends. TMEM44 is expressed in brain, but its function is currently unknown. Interestingly in humans, the homologous DMR overlaps with the last exon of the TMEM44-AS1, which encodes a superenhancer. Superenhancers are clusters of epigenetic regulatory elements that are developmentally established, are highly sensitive to perturbations and drive expression of genes that define cell identify and control cell-type specific functions183,184. It has been shown that superenhancers regulate genes controlling neuronal function and identity185,186. The predictive DMR is hypomethylated in future heavy drinkers, and if it alters the expression of the superenhancer, this specific pattern of DNAm might establish diverse downstream effects determining future risk.