Moderate alcohol use is associated with decreased brain volume in early middle age in both sexes

The aim was to examine cross-sectional association between moderate alcohol consumption and total brain volume in a cohort of participants in early middle-age, unconfounded by age-related neuronal change. 353 participants aged 39 to 45 years reported on their alcohol consumption using the AUDIT-C measure. Participants with alcohol abuse were excluded. Brain MRI was analyzed using a fully automated method. Brain volumes were adjusted by intracranial volume expressed as adjusted total brain volume (aTBV). AUDIT-C mean of 3.92 (SD 2.04) indicated moderate consumption. In a linear regression model, alcohol consumption was associated with smaller aTBV (B = − 0.258, p < .001). When sex and current smoking status were added to the model, the association remained significant. Stratified by sex, the association was seen in both males (B = − 0.258, p = 0.003) and females (B = − 0.214, p = 0.011). Adjusted for current smoking, the association remained in males (B = − 0.268, p = 0.003), but not in females. When alcohol consumption increased, total brain volume decreased by 0.2% per one AUDIT-C unit already at 39–45 years of age. Moderate alcohol use is associated with neuronal changes in both males and females suggesting health risks that should not be overlooked.

www.nature.com/scientificreports/ an optimal age to study how alcohol intake associates to brain volume. We studied the association of moderate alcohol consumption and total brain volume in a birth risk cohort aged 39 to 45 years in males and females.

Methods
Sample. The participants were part of a consecutive cohort that initially included 1196 infants born with various birth risks in a single hospital in Helsinki between 1971 and 1974 16 . A total of 202 children died or had severe disabilities, e.g. cerebral palsy, blindness or Down syndrome, and were excluded from the cohort, leaving 994 participants for prospective follow-up 16,17 . The inclusion criteria were hyperbilirubinemia (bilirubin level > 340 µmol/l or blood transfusion), birth weight below 2000 g, Apgar score < 7, respiratory distress requiring external ventilation, maternal diabetes, hypoglycemia (blood glucose ≤ 1.67 mmol/l), septicemia, or severe neurological symptoms such as rigidity, apnea, hyperexitability, convulsions or prolonged feeding difficulty in the absence of other risks 16,17 . In addition to the birth risk cohort, 164 singletons without birth risks have been followed from childhood as controls. The previous follow-up investigations were conducted at 5,9,16, and 30 years of age both among the birth risk cohort and control participants 17 . During 2014-2016, a total of 414 members of the birth risk cohort and 83 controls agreed to be re-examined. They were 39-45 years old community-dwelling adults with normal school history. The follow-up assessment consisted of a neurological and a neuropsychological examination, MRI, and a 516-item questionnaire (available from 18 ) which also included the Alcohol Use Disorders Identification Test (AUDIT) 19 . Table 1. Previous MRI studies on moderate drinking and brain structure using volumetry, voxel-based morphometry or brain age estimates. Only studies that examined global brain structures (e.g. total brain, total gray matter, total white matter or total cerebrospinal fluid) are included. TBV total brain volume, GM Gray Matter, WM White Matter a In studies which examined both local and global volumes, we listed results from global volume analyses only. www.nature.com/scientificreports/ A total of 393 participants had both brain MRI at a single imaging facility in Helsinki and a completed AUDIT test. We excluded 40 participants for following reasons: alcohol abuse (n = 5), poor quality MRI (n = 3), volBrain image analysis not available (n = 2), neurological conditions or MRI abnormalities e.g. multiple sclerosis, traumatic brain injury, ventriculomegaly, or severe white matter changes (n = 21), severe psychiatric problems (n = 9). See below for sources of information. The final sample consisted of 353 participants (163 males and 190 females).
Demographic and health data. Health and demographic data were acquired with the 516-item questionnaire filled out either online or as a mailed survey and the medical examination. Hospital and outpatient clinic discharge diagnoses (ICD-8, ICD-9 and ICD-10; 1975 onwards) were collected from the Care Register for Health Care (HILMO) of the Finnish Institute for Health and Welfare 20 . Purchases of reimbursed prescription drugs were obtained from the Social Insurance Institution of Finland 21 . A total of 92% of participants consented to share their registry information. Cerebrovascular events and risks, i.e. hypertension, diabetes, hyperlipidemia, high Body Mass Index, and heart disease, were identified by a neurologist (J.L.) using given history, a clinical examination, questionnaire, and registry data. Other neurological conditions, alcohol abuse and psychiatric problems were evaluated by J.L. using all the above information. Wechsler Adult Intelligence Scale-Fourth Edition 22 was used to measure the full-scale IQ. Education was classified into three levels: basic education, including the obligatory 9 years, secondary, including completed high school, vocational or comparable education, which lasts typically 12 years, and higher education, e.g. university-education.
Alcohol consumption. The questionnaire included a Finnish translation of the AUDIT 19 . AUDIT-C, a three-item short version derived from AUDIT, was used as the measure of alcohol consumption 23 . AUDIT-C questions assess how often and how much one typically drinks and how frequently binge drinking occurs (range 0-12). The second AUDIT-C question "How many drinks containing alcohol do you have on a typical day when you are drinking?" includes response options ranging from 1 to 10 or more drinks. Response option 0 drinks (scored as 0) was added to capture responses from nondrinkers. AUDIT-C scores of 3 or below in females and 4 or below in males have been used to indicate low-level drinking, scores of 5-8 to indicate possibly harmful but moderate alcohol use, and scores of 9 or higher to indicate potential alcohol abuse 24,25 . Five single-item responses were missing in AUDIT-C and were imputed with zero. other substance use. Current smoking, lifetime cannabis and illicit drug use were assessed on the questionnaire in yes/no format. Smoking status was classified as a current smoker or current non-smoker. Eight participants did not provide information on cannabis use and nine participants on lifetime illicit drug use. Thirty nine participants did not provide information on current smoking, 22 of whom had reported to be non-smokers in a questionnaire given ten years earlier. All were included as non-smokers.
MRi acquisition and volumetric analysis. Brain MRI scans were obtained using two 1.5 T MRI scanners (Signa; General Electric, Milwaukee, USA). Sequences included a T1-weighted three-dimensional structural sequence (Cube), used for volumetric analysis. Parameters used were: time of repetition (TR) 540 ms, time to echo (TE) 9.9 ms, flip angle 90 degrees, spacing between slices 0.59 mm, pixel spacing 0.48/0.48 mm, slice thickness 1.2 mm, and acquisition matrix 256 × 256. Additionally, T2-weighted fluid attenuation inversion recovery (FLAIR), susceptibility weighted imaging (SWAN), and either diffusion weighted imaging or 30-direction axial diffusion tensor imaging (half of all cases) sequences were performed. All scans were visually assessed by an experienced neuroradiologist (R.V.) who was blinded to clinical data and volumetric quantification. Imaging results with all sequences evaluated were, with regard to age, normal in 332 (94%) while 21 participants had clinically irrelevant minor abnormalities (e.g. developmental venous malformations, microbleeds, non-specific hyperintense white matter changes, or small cerebellar infarction).
Brain volume was analyzed using the fully automated MRI volumetry system volBrain 26 , which analyzes 3D T1-weighted scans to calculate volumes of total brain tissue, gray matter, white matter, cerebellum, hippocampus, lateral ventricles, thalamus, caudate nucleus, putamen, globus pallidus, nucleus accumbens, and amygdala. We used the total brain volume with cerebellum and brain stem included. To control for individual differences in overall cranial size, adjusted total brain volume (aTBV) was calculated by dividing brain volume by intracranial volume and expressed as a percentage.
Statistical methods. Based on power calculations the required sample size to detect a regression coefficient 0.2 with 80% power was 184 and with 95% power 304. Power calculations were based on simple regression analysis between alcohol consumption and aTBV assuming 5% alpha level and using observed standard deviations AUDIT-C (SD = 2.42) and aTBV (SD = 2.38) which were calculated from all available cases. Power calculations were done using the G*power 3.1.9.2 27 .
Linear regression was used to examine the association between alcohol consumption and aTBV, first unadjusted and then adjusting for covariates. Results are given as unstandardized coefficients (B) with 95% CI. Age, sex, birth risk status, current smoking, and the neuroradiological status (normal/abnormal) were potential confounders. Those found to be associated with aTBV using ANOVA, t-tests or Pearson's correlation were included as covariates in the adjusted model. Normal distribution of residuals and equality of variances were

Results
Characteristics of the sample are given in Table 2. AUDIT-C mean indicated low-level to moderate use. Ninety participants had cerebrovascular risks, but none had cerebrovascular events. Males consumed more alcohol than females, M The unadjusted association of alcohol consumption and aTBV was significant with B = − 0.258 (Table 3). When sex and current smoking status were added to the model, the association of alcohol consumption with the aTBV remained statistically significant (see Table 3). When the analysis was stratified by sex (Fig. 1)

Discussion
A higher amount of alcohol consumption was associated with a smaller brain volume among moderate drinkers aged 39 to 45 years. The effect persisted after adjusting for sex and smoking, and it appeared in males and females independently. A one-unit increase in AUDIT-C corresponded to approximately a 0.2% reduction in total brain www.nature.com/scientificreports/ volume. Previously, the association of moderate alcohol consumption and brain volume in early middle-age has been unclear especially in females. Two previous studies found smaller total brain volume associated with moderate drinking in a combined group of males and females with a mean age of over 60 years 8,9 . In these studies, the magnitude of the total brain volume difference between non-drinkers and moderate drinkers was approximately 1%. These are the only findings in line with our results. Two other studies of total brain volumes have reported inconsistent findings 5,6 . The first study of moderately drinking adults mainly around their forties and fifties found no reduction in total brain volume 5 , but the sample size (n = 91) was considerably smaller than in our study. The second study found a larger total brain volume associated with moderate drinking 6 , but the participants were around 80 years of age. Small effects may be difficult to differentiate from age-related variability when the sample is small or includes elderly people. For comparison, normal age-related total brain volume decline is approximately 0.4% per year after the age of 60 years 14,15 , but very slow around 40 years, i.e., in the age of our group. Brain ageing has recently been found to accelerate with daily or almost daily consumption of alcohol in a sample of participants above the age of 45 13 .
Interpretation of most previous studies is challenging because of small sample sizes, selection bias due to questionable recruitment methods (e.g. newspaper advertisements), and incomplete data of concomitant conditions. Also, many of the studies use complex regression models which include a large number of covariates in the analysis, which causes concerns of over-fitting and insufficient statistical power 29 . We confirmed by power calculations that we had a sufficient number of participants for analyzing aTBV at a satisfactory statistical power. The effect of alcohol use on smaller brain structures (e.g. hippocampus) has been analyzed in several studies 1 . Table 3. Associations between alcohol consumption (AUDIT-C) and total brain volume adjusted for intracranial volume (aTBV). B unstandardized regression coefficient beta. www.nature.com/scientificreports/ We calculated that using the same power assumptions in our volBrain analysis, including, for instance, the hippocampus would have required 1009 participants (slope = 0.0017, SD = 0.046). The increasing use of open datasets and biobanks will help in achieving adequate sample sizes. Methods of estimating the amount of consumed alcohol vary between studies. Commonly used methods include the number of drinks consumed per week 4,5 , often categorized into drinking groups [6][7][8][9]12 , and lifetime intake in kg 5,10,11 (Table 1). The AUDIT-C score measures the frequency and quantity of drinking as well as the frequency of heavy drinking occasions 23 . Although they are not directly comparable, all methods give estimates of consumption which can be used in regression analysis. The current sample had a group mean indicating low-level use, comparable to the levels found in a Finnish population-based study 30 . Many of the alcohol related health risks are linear 31 . All cutpoints are therefore arbitrary but still often needed in clinical use. In the Finnish study, the highest specificity for heavy use (defined by 16 drinks per week in males and 10 in females) was reached with an AUDIT-C cutpoint of 9 in males and 7 in females, while cutpoints of 7 and 5 were suggested for better sensitivity 30 . Others have also found scores of 9 or higher to indicate potential alcohol abuse and related health risks 24,25 . In our sample only four individuals reported a high score. Those with confirmed abuse were excluded.
Males consumed more alcohol and they also had larger brain volumes, and therefore the effect of sex was analyzed in more detail. Based on the results, the rate of change was of the same magnitude among sexes. In earlier studies, association between gray matter volume and alcohol use has been demonstrated in males 10 . However, females with alcohol use disorder appear equally susceptible to alcohol related brain damage as males, e.g. in the Framingham cohort study, that included a wide range of ages and consumption levels, the association between alcohol consumption and brain volume was actually more pronounced in women than in men 8 . The sex difference in susceptibility at low consumption levels or at a younger age has not been demonstrated. In the age range of our study, the existing studies have not shown unequivocal association between alcohol use and global brain volume in women 5,10,11 . In our sample, the association was visible in both sexes, analyzed combined as well as separately. When stratified by sex, a difference was seen only in the effect of smoking status. Current smoking is associated with a decreased brain volume 32 and increased brain ageing 13 . Current smokers also in our study had a smaller brain volume, as compared to non-smokers. However, smokers also consumed more alcohol, which probably, at least in part, explains this result. Smoking status did not explain the association between brain volume and alcohol consumption when both sexes were included.
A key strength of our study is that our participants come from a socio-economically and medically very homogenous longitudinal cohort, and they have been followed from birth. Participants were nearly the same age, resulting in minimal age-related variability which enhances the sensitivity to detect small effects. We also had previous and current medical registry data, self-reports, and a clinical examination to identify concomitant factors and exclude major conditions. None of the included participants had marked abnormalities in the four MRI sequences that were used. The imaging quality was ensured by using scans from a single facility and excluding scans with suboptimal quality. A fully automatic volumetric segmentation, unbiased by subjective user input, was used 26 . Further, we refrained from analysis of the small brain structures, as sample size calculus indicated insufficient statistical power.
Limitations may be caused by the cohort's structure, the measurement of alcohol consumption, and the choice of volumetric technique. Eighty-two percent of participants had a history of a birth risk. Thus, it is not a customary population sample cohort. However, participants with severe consequences or disabilities were excluded already in childhood and aTBV did not differ between different birth risks or controls. The participants were living independently, mostly with jobs and, according to Statistic Finland 33 , an education level corresponding to the general population of Finland. We think that it is very unlikely that birth risk history would affect our results. The estimates of alcohol consumption were based on self-reported typical consumption using AUDIT 19 , which may be prone to under-reporting 34 . However, we were able to identify and exclude excessive drinkers using the comprehensive medical data. Using volBrain for volume measurements may complicate comparison with other studies conducted with e.g. the commonly used FreeSurfer. We chose volBrain because, in our study, accurately measuring the ratio of total brain volume to intracranial volume was important. The estimated intracranial volume from FreeSurfer may be biased by total brain volume 35 and also the Freesurfer manual suggests using another method for determining the intracranial volume 36 . Further, volBrain is light on computing resources and the stability of brain volume in the middle age has been compellingly demonstrated using volBrain 14 .

conclusions
We found a direct association between moderate alcohol consumption and decreased brain volume at early middle-age in both males and females. Understanding of the mechanisms of moderate drinking on the brain is incomplete, but even moderate alcohol consumption may have a harmful effect already in middle-age. Recent systematic analysis on alcohol use and global disease burden suggests that the level of consumption that minimizes health loss is zero 31 . The risk that even moderate drinking poses on the brain should not be overlooked.

Data availability
Restrictions apply to the data. Although the participant-level data do not include participant identification, the ethics review board decision demands confidentiality. Pseudonymized data are available to a qualified investigator from the corresponding author.