Alcohol use and gray matter structure: Disentangling predispositional and causal contributions in human studies

Abstract

Alcohol use is a growing global health concern and economic burden. Cross-sectional observations that alcohol involvement (i.e., initiation, use, problematic use, alcohol use disorder) is reliably associated with broad spectrum grey matter differences have been largely interpreted to reflect alcohol-induced atrophy that contributes to negative health outcomes. However, emerging data suggest that brain structure differences also represent pre-existing vulnerability factors for alcohol involvement. Here, we review evidence from human studies with designs (i.e., family-based, genomic, longitudinal) that allow them to assess the plausibility that these correlates reflect predispositional risk factors and/or causal consequences of alcohol involvement. These studies provide convergent evidence that grey matter correlates of alcohol involvement largely reflect predisposing risk factors, with some evidence for potential alcohol-induced atrophy. These conclusions highlight the importance of study designs that can provide causal clues to cross-sectional observations. An integrative model may best account for these data, in which predisposition to alcohol use affects brain development, effects which may then be compounded by the neurotoxic consequences of heavy alcohol use.

1. INTRODUCTION

Alcohol use is an international public health concern. It is the leading risk factor for mortality among those aged 15–49 (Griswold et al., 2018), and each alcoholic drink consumed in 2010 in the United States (U.S.) resulted in a $2.05 economic burden (Sacks et al., 2015). Alcohol is also associated with ongoing health crises; between 15% of opioid overdoses in the U.S. involved alcohol between 2008–2017 (Tori et al., 2020) and alcohol use disorder (AUD) is associated with increased COVID-19 risk and severity (Wang et al., 2021). The widespread prevalence and devastating consequences of alcohol have stimulated extensive research to understand its etiology and the mechanisms through which it contributes to disease.

Alcohol involvement (e.g., initiation, non-problem use, problematic use, AUD) has been reliably associated with smaller indices of brain structure (Mackey et al., 2019; Yang et al., 2016). Indeed, well-powered meta- and mega-analyses of AUD (Ns= 930 – 1,190) (Mackey et al., 2019; Yang et al., 2016) as well as large single studies of alcohol consumption (e.g., UK Biobank; N’s=550 – 36,678) (Baranger et al., 2020; Daviet et al., 2022; Evangelou et al., 2021; Holmes et al., 2016; Lange et al., 2017; Topiwala et al., 2021, 2017) have found that alcohol involvement is associated with smaller gray matter volumes and thinner cortex broadly across the brain. These cross-sectional observations are largely interpreted to reflect neurotoxic consequences of alcohol exposure that broadly contribute to alcohol-related negative health outcomes (Evangelou et al., 2021; Lange et al., 2017; Topiwala et al., 2021, 2017; Yang et al., 2016). However, emerging data challenging this narrative suggest that these brain structure correlates may, at least partially, reflect pre-existing risk factors for alcohol involvement (Baranger et al., 2020; Hatoum et al., 2021b, 2021a).

Here, we provide an overview of evidence from studies in humans that alcohol-related differences in brain structure may reflect neurotoxic consequences of alcohol exposure and/or predisposing risk. We contextualize alcohol-brain relationships by briefly reviewing two neural models of addiction: the neurobiological stage-based model, which focuses on substance-induced consequences, and the neurodevelopmental model, which focuses on typical patterns of brain maturation reflecting predispositional vulnerability to substance involvement. We then discuss human study designs that can be used to inform whether observed associations reflect causal consequences of alcohol involvement and/or predispositional risk factors (Figure 1) and what they have revealed about brain structure correlates of alcohol involvement (Figure 2). We first touch on longitudinal designs, which have been recently reviewed (Boer et al., 2022; Lees et al., 2021a, 2021b; Parvaz et al., 2022), and then discuss how these results can be integrated with the nascent literature leveraging family-based and genomic designs. Ultimately, by relying on convergent evidence from divergent study designs, we propose that an integrative predispositional/neurotoxic model best explains existing data, in which variability in brain structure predisposes individuals to alcohol involvement, which in turn has neurotoxic consequences, particularly when use is heavy and chronic (Baranger et al., 2020; Baranger and Bogdan, 2019; Harper et al., 2021a, 2021b) (Figure 3).

Figure 1. Schematic of experimental designs used to interrogate the association between alcohol use and brain metrics.

Study Type denotes experimental designs and exemplar papers. Diagrams provide illustrations of the study design and associations tested. Implications highlight conclusions that can be drawn from the study design. Pros emphasize strengths of the study type, while Cons summarize the limitations.

Figure 2: Regions with convergent evidence for causal and/or predispositional effects:

Regions where 2 or more different experimental designs have found evidence for either a causal effect or a predispositional effect. Arrows reflect the direction of association. Modalities are combined unless otherwise specified, as there is as yet limited evidence for modality-specific effects. Note that the inferior frontal gyrus is not highlighted, as there is no convergent evidence for any single area in this region, even though associations in overlapping areas have been identified. Similarly, regions with inconsistent directional associations (i.e., the supramarginal gyrus, the lateral occipital cortex, and the precentral and postcentral gyri) are also not shown.

Figure 3: An integrative model.

Predispositional risk (genetic and/or environmental) is associated with different trajectories of brain development and aging, leading to different brain metrics. The onset of chronic alcohol use occurs in the context of these developmental differences. Heavy and chronic alcohol use leads to brain metric reductions. Note that the age of 21 was chosen for illustrative purposes, as this represents the legal drinking age in most states within the United States and celebrating the 21st birthday is frequently associated with excessive drinking in the United States (Rutledge et al., 2008).

1.1. Brain-based Models of Addiction

1.1.1. Neurobiological Model of Addiction.

The stage-based neurobiological model of addiction postulates that substance/experience-dependent changes in corticostriatal and corticolimbic circuits promote 3 recurring and non-mutually exclusive addiction stages: binge/intoxication, withdrawal/negative affect, and preoccupation/anticipation (Koob, 2021; Koob and Volkow, 2010; Volkow et al., 2016). Briefly, in the binge-intoxication stage, stimulation of neural circuitry subserving reward, most notably the striatum, provides positive reinforcement that motivates substance use. With continued use and progression, the withdrawal/negative affect stage emerges and is characterized by substance reinforcement shifting from positive to negative. This stage is characterized by anhedonia, following substance-induced changes in neural reward circuitry, as well as distress and physiological and psychological reactivity to the absence of drug-present homeostasis, predominantly from the amygdala and insula. Finally, with repeated positive and negative reinforcement pairings, cognitive preoccupation with/anticipation of drug reward and/or relief results in craving and the diminished influence of top-down control associated with the prefrontal cortex on behavior, even when there are strong subjective desires to not use substances.

1.1.2. Neurodevelopmental Model of Addiction.

The neurodevelopmental model of addiction proposes that typical patterns of brain maturation leave adolescents and young adults vulnerable to substance involvement by prioritizing emotional (reward, negative affect) and social processing while cognitive control and regulation continue to develop (Casey and Jones, 2010; Somerville et al., 2010; Steinberg, 2008). The rapid development of striatal and other limbic regions (e.g., amygdala, hippocampus) in combination with relatively delayed development of the prefrontal cortex is theorized to contribute to broad risk-taking behavior as well as impulsive attempts to cope with negative emotion. These place adolescents and young adults at predispositional risk for the positive (e.g., binge/intoxication) and negative (i.e., withdrawal/negative affect) reinforcing aspects of substance involvement vulnerability, particularly in the context of an underdeveloped physiological tolerance and regulatory capacity (i.e., preoccupation/anticipation).

The stage-based neurobiological and neurodevelopmental models of addiction are not mutually exclusive. While they emphasize substance-induced changes and predispositional vulnerability, respectively, they each acknowledge the importance of the other. The neurobiological model has arisen predominantly from functional differences in brain activity and receptor density, while the neurodevelopmental model originated primarily from developmental structural differences. Nonetheless, they are united in emphasizing variability in brain regions associated with positive and negative reinforcement, affect, and prefrontal regulation in the context of substance involvement and addiction vulnerability.

2. Alcohol Involvement and Brain Structure: Assessing the Plausibility of Alcohol-Induced Change and/or Predispositional Risk

Cross-sectional research is limited in its ability to differentiate whether phenotypic correlates reflect risk factors, causes, consequences, and/or epiphenomena. While randomized controlled trials, pseudo-experiments, and experiments on non-human animals can provide insight into causal mechanisms, they do not typically study predisposing risk factors, and suffer from several limitations (e.g., ethical concerns of assigning heavy alcohol consumption in humans; ecological validity; conservation across species). There now exist several human study designs that can be leveraged to evaluate the plausibility that phenotypic correlations with alcohol involvement reflect alcohol-induced changes in brain structure and/or brain structure biomarkers of vulnerability to alcohol involvement. These designs broadly fall into three categories: longitudinal designs, family-based designs, and genomic designs (Figure 1).

2.1. Longitudinal designs.

Longitudinal designs can inform the plausibility of causality and/or predisposition by testing temporality (Figure 1). If a change in drinking precedes reductions in brain structure, this temporal precedence would suggest that alcohol involvement may influence brain structure. Alternatively, if brain structure differences preceded alcohol involvement changes, this would highlight the potential predisposing vulnerability of variability in brain structure. Here we briefly touch on the longitudinal literature, which has been recently reviewed elsewhere (Boer et al., 2022; Lees et al., 2021a, 2021b; Parvaz et al., 2022).

2.1.1. Evidence for Changes in Alcohol Involvement Preceding Changes in Brain Structure.

In one of the first well-powered studies tracking changes in alcohol involvement and brain structure, the National Consortium on Alcohol and Neurodevelopment in Adolescence (NCANDA) found that youth who initiated heavy drinking had accelerated volumetric reductions across the whole brain, with the strongest effects in the frontal cortex (Infante et al., 2022; Pfefferbaum et al., 2018). More recent work in this sample has found that even moderate drinking is associated with accelerated whole brain development (i.e., cortical thinning), particularly in younger participants (Sun et al., 2023). Additional studies including those incorporating multiple cohorts and larger samples have found that moderate and heavy alcohol use is associated with potentiated whole-brain cortical gray matter decline during adolescence, particularly within the middle and superior frontal gyri (El Marroun et al., 2021; Squeglia et al., 2015). Associations with many other regions have been reported in individual studies (e.g., precentral and postcentral cortex, superior and inferior parietal, anterior cingulate, orbitofrontal, frontopolar, insular, and superior temporal cortices; amygdala and hippocampus), though the robustness of these associations remains unclear (El Marroun et al., 2021; Infante et al., 2022; Pfefferbaum et al., 2018; Squeglia et al., 2015). It should also be noted that while differences in age-related change are frequently interpreted as accelerated development, it is unclear whether they reflect changes to the aging process. Higher predicted brain age is cross-sectionally associated with alcohol use (Ning et al., 2020), but longitudinal and genetically-informed studies are needed to determine whether it reflects a prepositional risk factor and/or a consequence of alcohol use.

Further temporal plausibility for alcohol-related changes in brain structure come from studies showing that abstinence following treatment is associated with increases in cortical and subcortical volumes including within regions implicated in prospective studies of alcohol involvement noted above (e.g., middle frontal cortex, orbitofrontal cortex, insula, cingulate, and hippocampus) (Charlet et al., 2018; Durazzo and Meyerhoff, 2020; Parvaz et al., 2022; Zou et al., 2018). Despite these increases, cases typically still have smaller volumes than controls and it remains unclear to what extent these differences reflect abstinence-induced neural recovery and/or epiphenomena of recovery/treatment (e.g., improved nutrition, hydration).

2.1.2. Evidence of Brain Structure Differences Preceding Alcohol Involvement.

Several studies have observed that brain structure temporally predicts future drinking behavior (Boer et al., 2022; Lees et al., 2021a). We and others have found that smaller indices of brain structure, particularly volume and thickness of the prefrontal cortex (e.g., superior and middle frontal gyri), as well as thickness of several other regions (e.g., VLPFC, frontal pole, precuneus, superior parietal cortex, supramarginal gyrus, temporal pole, and transverse temporal gyrus) prospectively predict alcohol use initiation among adolescents who were substance naïve at the scan (Baranger et al., 2020; Brumback et al., 2016; Squeglia et al., 2017). A partially overlapping set of regions has also been shown to be associated with later alcohol use in participants who were alcohol-exposed at the time of scanning, including reduced volume of superior and middle frontal gyri and insula, VLPFC, and anterior cingulate (Baranger et al., 2020; Cheetham et al., 2014; Squeglia et al., 2014). The IMAGEN study is a large longitudinal study of adolescent development (ages 14–19) which has greatly contributed to our understanding of the emergence of alcohol-use behaviors during this developmental stage (Mascarell Maričić et al., 2020). Some machine learning analyses using this sample, where approximately half the participants were alcohol-exposed at the first scan, have found that structure is useful for predicting future alcohol involvement and binge drinking, and that important regions are widely distributed across the brain (Rane et al., 2022; Whelan et al., 2014), but see also some null results with a priori ROIs (Seo et al., 2019). Work in this sample that has tested the significance of associations in individual regions has found that increased volume of the caudate and cerebellum was predictive of increased future alcohol use (Kühn et al., 2019) and further, using a causal Bayesian model, that accelerated gray matter development (i.e., shrinkage) of the left prefrontal cortex and bilateral temporal cortex causally precedes increased drunkenness frequency (Robert et al., 2020).

2.1.3. Conclusions and Limitations.

Longitudinal studies provide temporal evidence that alcohol-related variability in brain structure both precedes and follows changes in alcohol involvement. Notably, longitudinal studies, regardless of the direction of evidence, have predominantly identified regions that play prominent roles in top-down regulation critical for behavioral control emphasized in the preoccupation/anticipation stage of the neurobiological model of addiction (e.g., DLPFC and VLPFC, anterior cingulate cortex, insula, and caudate). Together this evidence highlights the potential importance of neural circuitry associated with behavioral control and top-down regulation as a predispositional risk factor for alcohol involvement that may be further impacted by alcohol involvement, particularly with heavy use.

Evidence of temporality is intuitively appealing in informing causality. However, longitudinal associations must be interpreted in the context of their limitations, including the inability to rule out important counterfactuals. For example, alcohol involvement often unfolds during adolescence and young adulthood, during which developmentally regulated normative synaptic pruning and gray matter loss also occur (Bethlehem et al., 2022; Selemon, 2013). As such, longitudinal correlations may reflect epiphenomena of development or could even reflect risk (e.g., genetically conferred risk for alcohol use may increase the rate at which synapses are pruned). Alternatively, longitudinal associations may provide illusions of directional causality when effects actually occur in reverse. For example, it remains plausible that brain changes preceded alcohol use, but were undetectable with the sparse data collection that is typical of many longitudinal studies (e.g., scans every 2 years). Most importantly, longitudinal studies cannot determine whether differences in brain trajectories reflect effects of liability or exposure.

2.2. Family-based designs.

Two family-based designs (i.e., offspring and twin), have been predominantly used to draw inferences about predispositional and causal effects (Figure 1). Offspring studies of alcohol involvement typically compare alcohol naïve participants whose parents have AUD with those whose parents do not have AUD. If differences in brain structure are associated with AUD relatedness, they may reflect predisposing vulnerability, though the source (i.e., genetic or environmental) of that variability cannot be distinguished. Twin- and family-based studies typically recruit individuals who are monozygotic (~100% genetically similar) or dizygotic (~50% similar) twins or non-twin full siblings (~50% similar). Through modeling of genetic similarity, these studies can identify the contribution of genetic and environmental sources to the correlation between alcohol use and brain metrics. If there are a sufficient number of pairs discordant for alcohol involvement (e.g., a high drinker and low drinker in a pair), these studies can use discordancy models to test whether alcohol use may have causal effects on brain structure and/or whether brain structure may represent a predisposing risk factor for alcohol involvement. The more recently developed genomic relatedness model is an extension of the family-based model (Yang et al., 2011), where genomic data are used to estimate the genetic similarity between unrelated participants (i.e, beyond siblings, parents, or cousins). Genetic distance can then be used like genetic similarity in a twin analysis to identify the contribution of genetic and environmental sources to the correlation between alcohol use and brain metrics.

2.2.1. Evidence for Causal Effects of Alcohol on Brain Structure.

The Minnesota Twin Family Study has provided evidence that smaller hippocampal volumes associated with alcohol involvement may be attributable to alcohol use (Harper et al., 2021b; Wilson et al., 2018). Specifically, two studies with overlapping co-twin samples of women (study 1 cotwin n = 98; study 2 cotwin n = 224) found that twins with greater alcohol use had smaller hippocampal volume. Further, in MZ and DZ twins discordant for alcohol, heavier drinking twins had thinner lateral PFC, frontal/parietal medial cortex and frontal operculum (Harper et al., 2021a) than their co-twins. By showing that alcohol use is associated with smaller indices of brain structure even after accounting for shared genetics and environments, these data increase the plausibility that alcohol may induce changes in brain structure.

2.2.2. Evidence for Brain Structure as Predispositional Risk for Alcohol Involvement.

Using a familial and co-twin control design in the Human Connectome Project (n = 804), we (Baranger et al., 2020) found that the correlation between alcohol use and smaller volume of the middle and superior frontal cortex and insula is partially attributable to shared genetic effects and a predispositional effect of brain structure on alcohol use, with no evidence for non-shared environmental or causal effects. Studies in the Minnesota Twin Family Study (Harper et al., 2021a; Wilson et al., 2015) have also corroborated findings in the middle and superior frontal cortex, for both volume and thickness. These studies additionally identified predispositional effects of brain structure in several other regions, including lower amygdala volume, greater cerebellar volume, and lower thickness and volume in the pars triangularis and middle and inferior temporal gyri (Wilson et al., 2015), as well as thinner cortex in frontal operculum, frontal medial cortex, and parietal medial areas (Harper et al., 2021a). More recently, a novel genomic relatedness analytic technique in the UK Biobank (n=20,190) identified genetic correlations between alcohol use and greater volume of the lateral occipital cortex (de Vlaming et al., 2021).

The largest study to-date examining family history of substance use used data from the Adolescent Behavioral and Cognitive Development study (ABCD; N=11,875) (Lees et al., 2021b). While this study did not focus solely on family history of alcohol misuse or dependence, it provides the strongest evidence that a family history of substance use is associated with brain structural differences in substance-naive youth. This study found evidence for lower whole-brain cortical thickness, as well as local associations with lower thickness in the precentral and paracentral lobules, superior and inferior parietal lobules, precuneus, middle temporal gyrus, banks of the superior temporal sulcus, entorhinal cortex, and lateral occipital sulcus, as well as increased surface area of the precentral lobule and lateral occipital sulcus. This study, however, did not replicate prior findings that had been observed across multiple smaller samples, including smaller amygdala volume (Dager et al., 2015; Hill et al., 2013) and greater cerebellar volume (Benegal et al., 2007; Comstock et al., 2019; Hill et al., 2011, 2016, 2007). The young age of the ABCD study participants relative to these samples may contribute to differential findings.

2.2.3. Conclusions and Limitations.

Family-based designs have predominantly provided evidence that variability in brain structure represents a pre-existing vulnerability factor for alcohol involvement, though some evidence for causal effects exists. The most replicable finding has been the association of reduced volume/thickness of the middle and superior frontal gyrus with pre-existing liability to alcohol use. As with longitudinal findings, this highlights the potential importance of neural circuitry associated with behavioral control and top-down regulation as a predispositional risk factor for alcohol involvement. A single sample (the Minnesota Twin Family Study) has repeatedly observed potential causal effects of alcohol use on reduced hippocampal volume. The hippocampus is hypothesized to interact with the prefrontal cortex in driving the preoccupation/anticipation stage in the stage-based neurobiological model of addiction, by influencing the processing of contextual information (Koob and Volkow, 2010).

Twin-based designs are particularly powerful when testing for causal and predispositional effects, as they can adjust for confounding genetic and environmental influences. It is important to note that while twin-based designs can find support for a causal effect, they cannot rule out reverse causality (i.e., brain structure differences cause differences in alcohol use). Twin-based models also assume that environmental influences are shared to the same extent by monozygotic and dizygotic twins (Derks et al., 2006). Although it is difficult to validate this assumption, violations lead to inflated estimates of predispositional and causal effects (Dalmaijer, 2020).

2.3. Genomic Designs: Polygenic Risk Scores and Genetic Causal Models.

Following the proliferation of GWAS (genome-wide association studies) there has been rapid development of techniques using these generated data to inform the etiology of psychopathology (i.e., LD score correlation, polygenic scores, mendelian randomization, and latent causal variable analysis). A widely-used approach is LD-score correlation (LDSC) (Bulik-Sullivan et al., 2015), which assesses the concordance of effects between GWAS for two traits. For example, a significant correlation between alcohol involvement and brain structure indicates that genetic variants associated with differences in brain metrics are also associated with alcohol involvement. Polygenic scores (PGS) (Bogdan et al., 2018) can also be derived by applying GWAS results to an independent target sample, and reflect a fraction of an individual’s genetic risk (Figure 1). PGS can be applied to samples of alcohol-naïve individuals to estimate associations between brain structure and genetic liability to alcohol involvement. When applied to alcohol-naïve individuals, any observed correlations plausibly reflect predisposing vulnerability and/or epiphenomena, as opposed to consequences of exposure, though other exposures could also be considered (e.g., prenatal).

Mendelian randomization and Latent Causal Variable (LCV) analysis can be used to test for plausible causal effects between brain structure and alcohol involvement. Mendelian randomization treats genetic variants as instrumental variables, which includes assumptions that the genetic variant has a causal effect on the measured trait and that any effect on the outcome is only through its association with the trait of interest. Latent causal variable (LCV) analysis was developed to address the shortcomings of MR, including bias by pleiotropy and sample overlap, though it cannot test for bi-directional effects. Several papers provide greater explanation of mendelian randomization and LCV as well as critiques of these approaches (Davies et al., 2018; O’Connor and Price, 2018).

2.3.1. Genomic Designs: Alcohol and Brain Structure.

Leveraging genomic designs to understand the putative directionality of alcohol-brain relationships is in its infancy. Two studies have applied PGS to the question of alcohol use and brain structure using data from substance naïve children of European ancestry who completed the first wave of the ABCD study. These found that PGS for alcohol-involvement (i.e., problematic alcohol use; drinks/week [DPW]) are associated with brain structure (i.e., problematic use: lower volume in the frontal pole, greater supramarginal gyrus thickness; DPW: greater postcentral gyrus cortical thickness, greater cortical surface area; mixed directionality for cortical thickness) (Hatoum et al., 2021b; Rabinowitz et al., 2022). PGS were also recently applied in alcohol-exposed older adults in the UK Biobank (N=36,799), with evidence of associations between DPW PGS and greater total surface area, lower thickness in the superior frontal, caudal middle frontal, and inferior parietal cortices, and greater thickness in the insula and lingual gyri (Liu et al., 2022). LDSC has been applied in two studies to date, with evidence for genetic correlations between DPW and lower cortical thickness of the right superior frontal cortex (Liu et al., 2022), as well as lower global cortical thickness, greater total surface area, and greater surface area of the postcentral gyrus (Rabinowitz et al., 2022).

The few mendelian randomization studies examining alcohol use and brain structure have yielded conflicting results, including some evidence for bi-directional effects between subcortical volumes and alcohol involvement (Logtenberg et al., 2021). Notably, none of these associations remained significant after multiple testing when accounting for non-significant pleiotropy in models. Another study found weak evidence for potential causal influences of alcohol use on accelerated brain aging (Bøstrand et al., 2022). We used LCV analyses to more fully model the polygenicity of problematic alcohol use and brain structure, as well as its pleiotropy (Hatoum et al., 2021a). This analysis found evidence for a causal effect of brain structure on problematic alcohol use, including greater volume of the pars opercularis, greater thickness of the cuneus, and lower volume of the basal forebrain, but no evidence for a causal effect of alcohol use on the brain when phenotypes were reversed. Consistent with these findings, a recent study using mendelian randomization found evidence for a negative effect of global cortical thickness on DPW and binge drinking, with no evidence for regional effects or effect of alcohol use on the brain (Mavromatis et al., 2022).

2.3.2. Conclusions and Limitations.

Echoing results from twin studies, genomic designs have provided initial evidence that the genetic architecture of brain structure and alcohol involvement are shared. After accounting for this pleiotropy, there is evidence that brain structure may represent a pre-existing liability to alcohol involvement. However, regions identified by genomic models largely do not overlap with regions found in longitudinal and family-based studies (with the exception of whole-brain effects); this is likely due in part to the widespread use of region of interest analyses in longitudinal and family-based studies, where analyses are restricted to a small number of regions with a priori interest. GWAS for alcohol use phenotypes and brain metrics have only just reached a sufficient sample size for genomic analyses (Grasby et al., 2020; Zhou et al., 2020). It is quite likely that some of the current findings will not be reproduced, and that new insights will emerge, as samples further increase.

3. Brain Structure and Alcohol Involvement: Evidence for Predisposition and Neurotoxic Consequences

Longitudinal, family-based, and genomic models provide evidence that alcohol-related variability in brain structure reflects predispositional risk for alcohol involvement as well as neurotoxic consequences of moderate and heavy use. As each of these study designs is marked by unique assumptions and limitations in their ability to inform causality, convergent findings across methodologies are particularly informative (Munafò and Davey Smith, 2018). Convergent evidence predominantly supports predispositional effects of brain structure on alcohol involvement (Figure 2). Most notably, evidence for predispositional effects have been observed across experimental designs for reduced whole brain cortical thickness, increased cortical surface area, with evidence for regional effects in the DLPFC, insula, frontal pole, precuneus, superior parietal cortex, middle and inferior temporal cortex, cerebellum, and amygdala. That genomic risk for alcohol use is enriched for genes expressed in the DLPFC as well as other brain regions further supports these notions (Baranger et al., 2020). Many of these regions are theorized to contribute both to the negative affect and preoccupation/anticipation addiction stages in the neurobiological model of addiction, as well as developmental vulnerability during adolescence/young adulthood in the neurodevelopmental model of addiction (e.g., DLPFC, insula, and amygdala).

There is less convergent evidence for potential neurotoxic consequences of alcohol involvement on brain structure. There was support for putative neurotoxic consequences of alcohol involvement in the DLPFC and hippocampus. However, the longitudinal evidence for DLPFC associations have not been demonstrated to be independent of preexisting liability (e.g., they are no longer significant after correcting for family history (Pfefferbaum et al., 2018)). While research in model organisms (e.g., mice, rats, and macaques) has found evidence that supports potential neurotoxic effects of alcohol (for a review see (Hiller-Sturmhöfel and Spear, 2018)), it is important to consider that dependent brain measures in non-human animal studies often differ from those used in human studies (e.g., number of neurons vs gray matter volume). Experiments in model organisms that are more methodologically consistent with human studies have yielded substantial variability in regions implicated and direction of associations, suggesting that MRI-derived indices of brain structure may not recapitulate the same sources of alcohol-related variance as cellular assessments (Coleman Jr et al., 2011; Coleman et al., 2014; Pfefferbaum et al., 2006; Shnitko et al., 2019). The relatively limited support for alcohol-induced neurotoxicity contrasts with broad narratives often offered by cross sectional studies (Evangelou et al., 2021; Lange et al., 2017; Topiwala et al., 2021, 2017; Yang et al., 2016). Nevertheless, there is agreement that high levels of alcohol use cause brain and behavioral change in Wernicke-Korsakoff syndrome and fetal alcohol syndrome (Arts et al., 2017; Hoyme et al., 2016; Wozniak et al., 2019). However, the majority of the associations seen in the neuroimaging literature are not attributable to these rare syndromes (Wozniak et al., 2019) and we find that longitudinal, family-based, and genomic models provide more robust support for predispositional models.

Rather than view predispositional and causal perspectives as opposing, we note that both may underlie alcohol-brain structure correlations. Genetic and environmental risk for alcohol use influence brain development, effects that may then be compounded by the neurotoxic consequences of alcohol use (Figure 3). Altered brain development may additionally mediate the effects of risk on later alcohol use initiation and escalation. Acknowledging the evidence supporting both perspectives highlights areas where future research is needed. Here we briefly summarize four key directions:

Refining estimates of the causal effect of alcohol use:

The integration of these literatures demonstrates that correlational studies will invariably over-estimate the causal effects of alcohol use on brain structure, as correlational effects are clearly confounded by genetic and environmental risk. On the other hand, studies that have identified causal effects of alcohol use necessarily have a relatively smaller sample size, owing to the reliance on longitudinal or twin-based samples. As such, these studies may be less representative of the population and risk over-estimating effect sizes. Further, the causal effect of alcohol use likely depends on participant age, as well as the quantity and frequency of use. For example, while alcohol consumed is frequently modeled as a linear term, some evidence points to non-linear effects (Pfefferbaum et al., 2018). Additionally, given the rapid development of the brain from childhood to young adulthood (Bethlehem et al., 2022), one might expect that the causal effects of alcohol use may be moderated by age. Understanding the influence of these factors will be critical for identifying whether there are periods and behaviors leading to greater risk for detrimental effects. Estimating the causal effect of alcohol use on the brain, independent of predispositional confounding, will be crucial for refining messaging on the public health burden of alcohol use.

Identifying how predispositional risk influences brain development:

That predispositional risk for alcohol involvement influences brain metrics implies an effect on brain development, though we know little about the underlying mechanisms. Studies of environmental risk factors (e.g., low socioeconomic status) have found reduced cortical expansion in early childhood, as well as accelerated cortical thinning in adolescence (Hair et al., 2022; Mcdermott et al., 2019), consistent with theories of adaptive accelerated brain development following early life stress (McLaughlin et al., 2019). Indeed, emerging work suggests that developmental trajectories of the subcortex and orbitofrontal cortex mediate the effects of environmental risk factors on adolescent substance use and risky behaviors (Barch et al., 2022; Luby et al., 2018). However, no study to date has examined the effects of genetic risk for alcohol use on longitudinal brain development. Twin-based and genomic methods can be integrated with longitudinal analyses to address this question, and to identify plausible molecular pathways that may underlie these effects.

Determining whether brain development mediates the effects of predispositional risk:

There is little direct evidence supporting the hypothesis that brain development mediates the effects of genetic or environmental predispositional risk on later alcohol use initiation and escalation, despite speculation pointing to this mechanism (Barch et al., 2022; Hatoum et al., 2021a; Luby et al., 2018). This lack of evidence is largely attributable to the difficulties in testing causal hypotheses using observational data. Given the recent advent of large scale longitudinal neuroimaging studies (i.e., the ABCD and HBCD studies (Casey et al., 2018), https://hbcdstudy.org/) and the increasing sample size of GWAS of AUD, alcohol use, and brain metrics, further opportunities to test this hypothesis will soon arise.

Disentangling disease specific and general processes:

There is growing evidence that many genetic and environmental risk factors are shared across substance involvement (Hatoum et al., 2022). In tandem, substance use disorders are also associated with effects in overlapping brain regions (Mackey et al., 2019). This raises the possibility that reduced brain metrics may partially reflect general addiction liability (Hatoum et al., 2022), rather than alcohol specific effects. Indeed, the brain-based models of addiction referenced here are not specific to alcohol (Casey and Jones, 2010; Koob and Volkow, 2010). The extent to which alcohol-based findings generalize to other substances will be important for improving our understanding of the neurobiological and genetic underpinnings of these traits and potentially identifying causal influences of specific substances.

4. Limitations

Given the broad range of applicable methods, effect size estimates are largely not comparable across methods. As a result, there is no clear way to resolve discrepancies (e.g., differences in the direction of an association across methods). This is further compounded by variation in neuroimaging methods, including different brain metrics (i.e., volume, thickness, and surface area), reliance on region-of-interest analyses in much of the literature, variability in whether lateralized effects are considered (i.e., left vs. right or an average of left and right), and differences in the definition of region boundaries (e.g., which brain atlas is used). As a result, there is insufficient data for a robust meta-analysis of predispositional or causal effects. Here, we attempt to synthesize results by focusing on associations that are directionally consistent across different methodologies. However, we are unable to account for differences in effect size or sample size across methods, as any attempt to do so would be inherently subjective. Future work will be needed to more fully resolve these discrepancies. Further, the majority of the reviewed studies did not examine differences associated with sex or demographic factors such as socioeconomic status. While there are reports that the association of alcohol use with brain structure may differ between sexes (Grace et al., 2021; Seo et al., 2019), others have found no differences when formally testing the interaction (Pfefferbaum et al., 2018) or when repeating analyses in the same sample (Rossetti et al., 2021), and findings in general are mixed (see (Verplaetse et al., 2021) for a recent review). Even so, sex and other demographic variables are known to influence rates of substance use initiation (Amaro et al., 2021; Fonseca et al., 2021) as well as trajectories of adolescent brain development (Barch et al., 2022; Bethlehem et al., 2022; Lenroot et al., 2007). Further work considering the contribution of sex and other demographic variables is needed.

5. Conclusions

Correlations between alcohol use and reduced brain metrics are highly replicable and robust. There is growing evidence that these correlations are largely attributable to the influence of predispositional genomic and environmental risk factors on brain metrics. Evidence also supports causal effects of alcohol involvement on brain structure, though this evidence is less robust. Ultimately, models explicitly incorporating and testing the plausibility of predisposition and causation are needed to advance our understanding of the dynamic etiology underlying alcohol involvement, to identify pre-existing biomarkers of vulnerability, and to attenuate the negative outcomes associated with alcohol involvement.

Highlights:

1. Grey matter correlates of alcohol involvement largely reflect predisposing risk factors.

2. Alcohol-induced atrophy is an additional source of variation.

3. An integrative model may best account for these data.