Alcohol consumption and MRI markers of brain structure and function: Cohort study of 25,378 UK Biobank participants

1. Introduction

Moderate alcohol consumption is very common yet conflicting associations with late-life brain and cognitive outcomes have been reported (Visontay et al., 2021). Without disease-modifying treatments for neurodegenerative disease, there is a necessary focus on modifiable risk factors such as alcohol. Even small adverse effects of moderate drinking on the brain may have substantial public health relevance given the widespread exposure to alcohol. Here we assess relationships between alcohol consumption and neuroimaging indices. Structural MRI measures can serve as biomarkers before cognitive decline in people later developing dementia (Mcconathy and Sheline, 2015). Until the establishment of UK Biobank, costs have prohibited collection of imaging sample sizes necessary to examine the impact of drinking at such low levels.

Whilst chronic heavy alcohol intake is known to associate with reduced brain volume (Pfefferbaum et al., 1992, Mackey et al., 2019), the impact of ‘moderate’ consumption (defined variably from < 14 (Kim et al., 2020) to < 25 units (Paul et al., 2008) weekly) has been contentious (Anstey et al., 2009, Sabia et al., 2018). This is reflected in alcohol guidelines which currently neglect the brain, relying solely on cardiovascular and cancer research (Care DOHaS, 2016). Previous work has reported associations with hippocampal atrophy in males drinking just 14–21 units weekly (Topiwala et al., 2017). Two UKB studies of alcohol consumption and structural neuroimaging have reported associations between alcohol consumption, at even lower levels between 7 and 14 units weekly, and grey and white matter measures (Evangelou et al., 2021, Daviet et al., 2022). The UK Biobank has high statistical power but also higher sensitivity to confound effects (Alfaro-Almagro et al., 2021). Thus residual confounding, which could cause spurious correlations, is a concern. Earlier analyses have controlled for only a limited number of (particularly image-related) confounders. Studies to date have also failed to distinguish between never and ex-drinkers, raising the possibility that ‘sick quitters’ could be influencing results (Marmot and Brunner, 1991). Whether low levels of alcohol consumption relate to the brain’s functional architecture has not been well studied. Functional connectivity of the brain, meaning synchrony of activity between regions, changes in dementia and has been observed in small studies of alcohol dependent individuals but not examined in large samples of non-dependent drinkers (Zhu et al., 2017). What is also unclear is whether certain clinical or demographic factors increase the risk of alcohol-related brain harm. This has public health relevance in terms of focusing interventions on those at greatest need (Paper, 2020). Understanding how medical comorbidities, such as hypertension and obesity, interact with alcohol use is unknown (Qizilbash et al., 2015). ApoE4 genotype is a well-established risk factor for Alzheimer’s disease (Belloy et al., 2019). It has been implicated in brain changes in later life, including in the hippocampus (Honea et al., 2009), a structure previously associated with alcohol consumption (Topiwala et al., 2017). Whether ApoE4 modifies effects of alcohol on the brain is unclear. There are a few substantiated claims that red wine has beneficial effects (Pignatelli et al., 2006). Conversely it is thought that certain drinking patterns, such as binging, may worsen the brain impact (Hunt, 1993).

To address these unanswered issues, we investigated alcohol consumption and brain measures in UKB. We had four main research questions:

1) Are previous associations between alcohol consumption and structural neuroimaging indices reproducible in voxel-wise analyses, and robust to observed confounding?

2) Does alcohol consumption associate with functional connectivity in the brain?

3) Do the following factors increase risk of alcohol-related brain effects: ApoE4, hypertension, high body mass index, beer and spirit consumption, and binge drinking?

4) How robust are associations between alcohol and MRI markers to unobserved confounding?

An exploratory objective was to investigate the functional significance, in terms of cognitive performance, of any alcohol-brain associations.

2. Methods

3. Results

4. Discussion

In this large population-based neuroimaging study, alcohol was linearly and negatively associated with total grey matter volume. Higher alcohol intake was negatively associated with FA and mode, and positively associated with MD, L1-3. Increased functional connectivity within several networks associated with alcohol intake. Differences in total grey matter volume were observed in those drinking as little as 7–14 units (56–112 g) compared to <7 units weekly. Relationships were robust against unobserved confounding. Frequent binging, higher blood pressure and BMI were associated with steeper associations between alcohol and total grey matter volume. In contrast, alcohol beverage type consumed appeared to have little significance.

Non-drinkers comprised a higher proportion of females, higher mean material deprivation and lower educational qualifications compared with drinkers. Similar patterns have been observed in the wider UK population (Statistics OFN, 2017). The latest Office of National Statistics (ONS) data found higher proportions of females, those with lower income and no formal educational qualifications amongst the teetotaler group. The one discrepancy is that in the ONS data a lower proportion of teetotalers had never smoked, whereas in this UKB sample lower smoking was reported amongst non-drinkers.

Negative associations between alcohol and grey matter volume, as assessed with VBM, were spatially extensive. Older studies of non-dependent alcohol consumption and brain MRI have been somewhat conflicting (Topiwala and Ebmeier, 2018). Whilst some studies have reported lower white matter lesions and infarcts with light to moderate intake (den Heijer et al., 2004), others have reported higher atrophy (Mukamal et al., 2001) or lower grey matter volume in frontal, temporal and parietal lobes (Ewing et al., 2014). Our findings corroborate widespread associations with alcohol previously reported in the UKB in frontal, parietal and insular cortices, temporal and cingulate regions, putamen, amygdala and the brain stem (Evangelou et al., 2021, Daviet et al., 2022). In this UKB sample the strongest subcortical associations with alcohol we and others (Evangelou et al., 2021, Daviet et al., 2022) observed were negative with thalamus volume. Whilst we previously observed inverse associations between alcohol localized to hippocampal size in a separate smaller cohort sample (Topiwala et al., 2017), associations with hippocampal volume in UKB appear to be weaker. There is a great disparity in sample sizes, and therefore statistical power to detect small effects, between our previous analyses in Whitehall II (Topiwala et al., 2017) and studies in UKB. Daviet et al. found associations with alcohol in >90% of grey matter regions examined (Daviet et al., 2022), whereas Evangelou et al. reported a much narrower spatial distribution than we found, limited to cingulate, orbitofrontal regions, insula and thalami (Evangelou et al., 2021). Methodological differences may also play a role in explaining differing results. For example, our analyses have used FSL software in contrast to Evangelou et al. (Evangelou et al., 2021) who used SPM. Modulation (full vs. non-linear only), smoothing kernel (4.6–8 mm) and multiple comparison correction methods also differ. Daviet et al. (Daviet et al., 2022) examined IDPs derived from parcellations from atlases rather than voxel-based analyses used in this study and previously (Evangelou et al., 2021). There are potential ramifications of choices we made in our analysis pipelines. For example, we employed a small amount of spatial smoothing which is more sensitive to smaller anatomical differences and potentially less sensitive to differences of larger spatial extent. As cited previously, we found no evidence against associations with alcohol being monotonic (Evangelou et al., 2021). Furthermore, the observed widespread grey matter volume-alcohol associations in the current study persisted after additional sensitivity analyses not performed in previous studies, including adjustment for additional image-related confounders, and dividing abstainers into previous and never drinkers. This gives more confidence to the interpretation that alcohol is responsible for the brain structure associations, rather than other confounds known to impact brain measures (Alfaro-Almagro et al., 2021). Findings were unchanged when excluding previous drinkers. This lessens the possibility that individuals who have stopped drinking due to ill health, so called ‘sick quitters’, could be elevating the risk of non-drinkers (Marmot and Brunner, 1991), therefore underestimating the risks of drinking. A small number of voxels had positive associations with alcohol intake. The post-hoc analyses of lingual volume IDPs did not have convergent results, in fact negative associations with alcohol were observed as has been found by others (Daviet et al., 2022). The positive associations observed in the VBM analyses may have resulted from misregistrations, not practical to manually check at this scale. This is indirectly supported by the opposite findings using the FreeSurfer IDPs which are thought to be less prone to such problems. Additionally, many of the voxels with positive associations with alcohol in the VBM analyses were bordering CSF spaces and sinuses or supratentorial meninges. This raises the possibility of partial volume effects, although on visualization they appeared cortical. Smaller total grey matter volume was observed in participants drinking>7 units of alcohol weekly. This level of drinking is lower than currently defined as ‘low risk’ by the UK drinking guidelines (<14 units weekly) (Care DOHaS, 2016). Within UKB, others have observed associations with structural MRI at 7–14 units weekly and above (Daviet et al., 2022). Outside UKB, to our knowledge the lowest alcohol intake associated with MRI markers has been 14 units weekly (Topiwala et al., 2017). We hypothesise that the greater sample size of UKB enables detection of smaller effects. Whilst the effect size of the alcohol-total grey matter association was small in comparison to age, it was largest of the modifiable risk factors examined in this sample, making it relevant to public health.

Lower fractional anisotropy and mode, and higher mean and radial diffusivity were associated with higher alcohol consumption in this study. These findings are suggestive of alterations in white matter microstructure such as loss of myelin or axonal membranes (Friedrich et al., 20202020, Winklewski et al., 2018). The widespread associations we observed corroborate findings by another UKB study that examined DTI IDPs and also found associations with alcohol across many IDPs (Daviet et al., 2022). Furthermore, both studies highlighted strongest effects in the fornix. Previous studies reported more localized associations with alcohol in the corpus callosum (Topiwala et al., 2017) and corticospinal tracts (Evangelou et al., 2021). Somewhat surprisingly, a third UKB study did not find widespread associations between alcohol and diffusion IDPs (Evangelou et al., 2021). In fact, they observed positive associations between alcohol and corticospinal tract FA, which they suggest relate to crossing fibres. Again, methodological differences may explain these discrepancies. Evangelou used a smaller UKB sample (∼9000 participants vs. > 25,000 here and in Daviet et al.) thus had lower statistical power to detect small effects. Whilst our previous study (Topiwala et al., 2017) used, as in the current study, tract-based spatial statistics which allows examination of white matter microstructure in a finer grained manner, Daviet et al. (Daviet et al., 2022) and Evangelou et al. (Evangelou et al., 2021) examined multiple fibre tract IDPs. Correcting for global mean FA, associations between alcohol and FA appear less widespread. Adjustment for cardiovascular risk factors did not make a material difference to the strength of the associations, suggesting they were neither confounders nor on the causal pathway in our sample. Exploration of these factors was important, given that alcohol has known effects on cardiovascular risk (Roerecke et al., 2017, De Oliveira et al., 2000) and cardiometabolic disease has been highlighted as a mediator of alcohol-dementia relationships previously (Sabia et al., 2018).

This is the largest study, to our knowledge, of non-dependent alcohol consumption and functional connectivity. Higher alcohol intake was associated with increased functional connectivity within the default mode, attention, central executive and salience networks, and lower connectivity within the visual network. Of the few studies in alcohol dependent individuals, reduced functional connectivity in visual, executive, salience and default mode networks has been reported (Chanraud et al., 2011). Many factors could underlie the observed associations between alcohol intake and functional connectivity. Associations between connectivity and baseline alcohol intake could be the result of a participants’ chronic exposure to alcohol and subsequent neural changes (Fairbairn et al., 2021). Alternatively, higher weekly consumption could theoretically associate with intermittent mild symptoms or subclinical (minimally recognizable clinical findings) alcohol withdrawal (Jung and Metzger, 2010). These symptoms could include nausea, tremor, anxiety, higher blood pressure and pulse, amongst others.

The negative associations of binging with total grey matter volume is in keeping with other health outcomes, including mortality (Kauhanen et al., 1997), breast cancer (White et al., 2017), and cardiovascular disease (Mckee and Britton, 1998). Peak ethanol levels are higher during a binge. Binging followed by abstinence can precipitate alcohol withdrawal (Day and Daly, 2022). Withdrawal increases glutamate release (Hermann et al., 2012), microglial activation, inflammatory cytokines (Marshall et al., 2013) and can lead to brain injury (Jung and Metzger, 2010). Repeated binging-withdrawal cycles could magnify effects (Brown et al., 1988). Whilst most of our subjects were not alcohol-dependent, many reported binges, potentially resulting in subclinical withdrawal. Given the marked consequences for the brain of withdrawal, this represents a potential explanation for our observed interaction between alcohol intake and binging frequency in predicting grey matter volume.

Some studies have postulated protective effects of moderate drinking are strongest in wine (Gronbaek et al., 1995), due to polyphenols levels in grape skins. Our findings support the hypothesis that it is ethanol itself that is on the causal pathway of brain effects. Associations of wine-drinking with higher educational level and socioeconomic status may explain apparent health benefits (Mortensen et al., 2001).

We observed steeper associations between alcohol consumption and total grey matter volume in those with higher blood pressure and BMI. Higher blood pressure and BMI have been previously implicated in reducing grey matter volumes (Qizilbash et al., 2015, Beauchet et al., 2013) but their interaction with alcohol has not been explored. One mechanistic hypothesis is that hypertension may facilitate diffusion of ethanol throughout brain tissue. Animal models have demonstrated dysfunction of the blood brain barrier (BBB) in hypertension (Mackenzie et al., 1976, Biancardi et al., 2014). ApoE4 can also break down the BBB (Montagne et al., 2020), and therefore could hypothetically facilitate ethanol diffusion. We found no significant interaction between alcohol consumption and ApoE4, perhaps because of limited power due to the small number of E4 homozygotes. Animal models have shown synergistic effects of obesity and alcohol on steatohepatitis (Xu et al., 2011). BMI may increase the adverse impact of alcohol via the generation of toxic ceramides through a liver-brain axis (De La Monte et al., 2009). Ceramides have been linked to hippocampal atrophy (Kim et al., 2017) and risk of Alzheimer’s disease (Mielke et al., 2012). Alternatively, obesity could impair the intestinal barrier, facilitating bacterial endotoxin entry and pro-inflammatory cascades (Yan and Schnabl, 2012, Frazier et al., 2011, Brown, 2019).

MRI measures were associated with cognitive test performance, but direct associations between alcohol and cognition were only observed amongst participants with lower education. This could reflect a protective effect of education in line with cognitive reserve theory, or ceiling effects of certain measures (tower rearranging) (Stern et al., 1999). MRI may be more sensitive to the effects of alcohol than behavioral measures, especially given UKB’s limited cognitive battery which does not cover certain domains known to be affected by alcohol, such as verbal fluency (Topiwala et al., 2017, Fawns-Ritchie et al., 2020, Nowakowska-Domagała et al., 2017). Additionally, an unmeasured confounder, positively correlated to ‘moderate’ drinking and cognition, such as premorbid IQ could mask associations (Topiwala et al., 2017).

Funding

AT is supported by a Wellcome Trust fellowship (216462/Z/19/Z) and KPE by the UK Medical Research Council (G1001354 & MR/K013351/1) and the European Commission (Horizon 2020 732592). This work was also supported by the Li Ka Shing Centre for Health Information and Discovery, NIH grant (TMS, TN: R01EB026859), and a Wellcome Trust award (TN: 100309/Z/12/Z).

CRediT authorship contribution statement

Anya Topiwala: Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing. Klaus P. Ebmeier: Writing – original draft, Writing – review & editing. Thomas Maullin-Sapey: Formal analysis, Methodology, Writing – review & editing. Thomas E. Nichols: Formal analysis, Methodology, Visualization, Supervision, Writing – original draft, Writing – review & editing.