OBESITY, DIABETES, COFFEE, TEA AND CANNABIS USE ALTER RISK FOR ALCOHOL-RELATED CIRRHOSIS IN TWO LARGE COHORTS OF HIGH-RISK DRINKERS

246 words, main text 4349 words, 49 references, 5 Tables, 2 Figures. OBESITY, DIABETES, COFFEE, TEA AND CANNABIS USE ALTER RISK FOR ALCOHOL-RELATED CIRRHOSIS IN TWO LARGE COHORTS OF HIGH-RISK DRINKERS John B. Whitfield PhD FRCPath*, Steven Masson FRCP, Suthat Liangpunsakul MD, Sebastian Mueller MD PhD, Guruprasad P. Aithal PhD, Florian Eyer MD, Dermot Gleeson MD FRCP, Andrew Thompson PhD, Felix Stickel MD PhD, Michael Soyka MD, Ann K. Daly PhD, Heather J. Cordell DPhil, Tatiana Foroud PhD, Lawrence Lumeng MD, Munir Pirmohamed PhD FRCP, Bertrand Nalpas MD PhD, Jean-Marc Jacquet MD, Romain Moirand MD PhD, Pierre Nahon MD PhD, Sylvie Naveau MD, Pascal Perney MD PhD, Paul S. Haber MD PhD, Helmut K. Seitz MD, Christopher P. Day MD, Philippe Mathurin MD PhD, Timothy R. Morgan MD, Devanshi Seth PhD*, for the GenomALC Consortium.


th April 2020 The Editors-in-Chief
The American Journal of Gastroenterology Dear Editors,

AJG-20-0154, Revision
Thank you for your initial decision on this paper, and your invitation to respond to the comments and submit a revision. The revised paper, with and without highlighting of changes, and our response letter will be uploaded through your website.
All authors have seen the comments and suggested responses, and the revised paper, and their suggestions have been incorporated.
We look forward to hearing from you again.
Sincerely John B Whitfield, PhD On behalf of the authors Email: John.Whitfield@qimrberghofer.edu.au Phone: (+61) 7 3362 0229 Sustained high alcohol intake, often associated with alcohol dependence, can lead to alcoholrelated liver diseases including cirrhosis. The usual progression is through fatty liver, frequent in high-risk drinkers but reversible with abstinence, to fibrosis and cirrhosis. Some patients will develop alcoholic hepatitis, and some will develop hepatocellular carcinoma (HCC), generally with cirrhosis as a precursor. Therefore, cirrhosis is not only the end-stage of liver damage, but also increases risk for other life-threatening conditions. Apart from abstinence from alcohol, supportive measures, and liver transplantation in selected abstinent patients, current treatment options for alcohol-related cirrhosis are limited.
The relationship between alcohol intake and cirrhosis has been recognised since the late eighteenth century (1), with subsequent efforts to quantify this association made by Pequignot (2) who noted an increased risk of cirrhosis in people drinking more than 40 grams of alcohol per day. It is known that women are more susceptible to liver damage from alcohol than men (3), and larger studies and meta-analyses (4) have refined the threshold for detectable risk from alcohol intake.
It is notable that only a minority of high-risk drinkers develop cirrhosis. It is difficult to find reliable estimates, but in Denmark 7.7% of patients diagnosed with harmful alcohol use and 8.8% of those diagnosed with alcohol dependence developed cirrhosis over the subsequent 15 years (5). Meta-analysis (6) showed that 7-16% of people in alcohol problem cohorts had cirrhosis after 8-12 years. Variation in susceptibility may be due to genetic variation, and/or presence of other environmental and lifestyle risk factors which increase the probability of liver damage. Apart from alcohol intake and gender, obesity (also associated with nonalcoholic liver disease) has the strongest evidence for increasing risk of alcohol-related cirrhosis. For instance, liver biopsy histology showed more severe abnormalities in patients with alcohol use disorders with greater body weight (7); this was confirmed in a subsequent study (8) which showed that being overweight was a risk factor for steatosis, hepatitis and cirrhosis in addition to the effects of age, gender and duration of alcohol abuse. Other studies have also found an association between obesity or body mass index (BMI) and liver disease (9,10), fibrosis (11), alcoholic hepatitis (12) or HCC (13). There is evidence that coffee or tea consumption can reduce risk of liver disease or favourably affect biomarkers associated with liver disease (14)(15)(16)(17). Smoking has been associated with increased risk of alcohol-related cirrhosis and of cirrhosis in general, particularly among women (18). A recent report showed that cannabis use protected against liver disease in patients with alcohol use disorders (19), possibly through effects on inflammation mediated by cannabinoid receptors (20).
There is a lack of hard data from twin or family studies on genetic risk for alcohol-related cirrhosis. Alcohol dependence is partially heritable (21) but twin studies on its consequences such as alcohol-related liver disease (22) have been limited by small numbers and lack of adjustment for heritable effects on alcohol exposure (23). Our earlier report (24) suggested that a history of liver disease in a parent with alcohol problems was associated with increased risk of alcohol-related cirrhosis. The known genetic risk loci for cirrhosis in PNPLA3 and HSD17B13 (25,26) are associated with lipid metabolism and potentially with metabolic changes which accompany obesity.
The GenomALC Consortium (24) was initiated to gather data and samples for identification of risk factors for alcohol-related cirrhosis, including a case-control genetic association study.
In this paper we focus on comparison of case and control groups for potential clinical and phenotype factors that alter disease risk including beverage preference, other substance use, family history, obesity and diabetes. Where we have identified potential risk-altering factors from our data, we have attempted validation using comparable data from the UK Biobank.

GenomALC Study
Recruitment and data collection were based on our published GenomALC protocol (24). Two groups of patients were recruited between 2012 and 2017 in six countries (Australia, France, Germany, Switzerland, UK and USA).. Cases were recruited through hepatology clinics and controls were recruited from psychiatric clinics or detoxification facilities. All participants gave written informed consent. The study was approved by appropriate Ethics Committees or Institutional Review Board at each site and conformed to the ethical guidelines of the 1975 Declaration of Helsinki. Data and samples were identified by a study-specific code with no personal information.
To be confident that participants either had or were at substantial risk of alcohol-related cirrhosis and to minimise the chance that cirrhosis was caused by factors other than alcohol, we recruited patients with alcohol intake of at least 80 grams per day for men and 50 grams per day for women for 10 years or more. Both cases and controls were required to have negative test results (antibody/antigen/viral load) for hepatitis B and C, and no clinical or serological evidence of human immunodeficiency virus (HIV). Unequivocal evidence of cirrhosis in cases was defined as imaging results (sonography, computed tomography, magnetic resonance imaging) compatible with cirrhosis; together with detectable ascites by imaging or paracentesis, and/or grade 2 or higher spontaneous hepatic encephalopathy, and/or moderate or large oesophageal varices on upper gastrointestinal endoscopy. Histological cirrhosis on biopsy was defined as Metavir fibrosis stage F4 or Ishak fibrosis stage 5 or 6.
Liver stiffness (Fibroscan ® ) was accepted as diagnostic for cirrhosis if greater than 22 kPa in the presence of aspartate aminotransferase (AST) less than 100 u/l or ≥30 kPa if AST between 100-200 IU/L (27,28) . Other causes of liver disease, including haemochromatosis, Wilson's Disease, and autoimmune liver disease were excluded by laboratory tests or clinical criteria, and any patient who had received a liver transplant for a condition other than alcohol-related cirrhosis was also excluded. Controls met the alcohol intake criteria but with no evidence or history of liver disease, had normal results for liver function tests (AST, alanine aminotransferase (ALT), bilirubin, albumin, but not necessarily for gammaglutamyl transferase (GGT)), platelet count and International Normalised Ratio (INR), and/or had less than 6 kPa liver stiffness (Fibroscan ® ), while drinking or within seven days of abstinence.
Information was collected on demographics, self-reported ancestry, history of alcohol, tobacco and cannabis use, tea and coffee consumption, clinical symptoms, biopsy results if available, and biochemical and haematological test results. The data collection form (24)  This analysis was restricted to patients whose fathers or mothers were reported to have had 'problems with alcohol', and assumes that death from liver disease in a parent with alcohol problems is due to alcohol-related liver disease (potentially alcoholic hepatitis or HCC, as well as alcohol-related cirrhosis).

UK Biobank
Data evaluation of the effects of family history, the possibility of differential transmission of effects to male and female patients was taken into account using patient sex for stratification, testing for heterogeneity of OR across strata with the Breslow-Day test and, if no heterogeneity was found, estimating the common OR. Similarly, for testing whether casecontrol differences were consistent across country of recruitment, countries were treated as the strata and heterogeneity and common odds ratios were evaluated. ancestry, with the highest proportion in Germany (99%) and lowest in the US (88%).

GenomALC participantscase-control comparisons
Cases drank significantly less alcohol per day than controls, but had been drinking for significantly longer. Total lifetime alcohol intake did not differ significantly between male cases and controls, and in female cases was slightly lower than for controls (Table 1). A breakdown by country of recruitment is given in Supplementary Table 2, with comparisons of lifetime alcohol intake in cases and controls by country in Supplementary Figure 1. Controls reported taking a significantly higher proportion of their total alcohol in the form of wine (Table 2), but were less likely to report usually drinking with (rather than between) meals.
Forty eight percent of cases but only 28% of controls were currently living with a spouse or partner. There was no significant difference in years of education. Controls were more likely than cases to have been coffee drinkers during the time they were drinking alcohol heavily, and to have drunk more coffee per day, but there was no significant difference for tea consumption ( Table 2). A slightly higher but statistically significant proportion of controls reported drinking green tea (7% of cases and 9% of controls). Most people in both groups were or had been smokers, but the proportion was significantly higher in controls (83%) than cases (72%). Regular cannabis use was about three times more common among the controls (27%) than cases (9%) ( Table 2) but the proportion decreased with age (in both cases and controls) and the case-control difference was non-significant in patients aged over 60 years ( Figure 1(a)).
Mean BMI was higher among the cases than the controls (Table 2). Because this difference might be secondary to the disease, e.g. through fluid retention in the cases or through inadequate diet in the controls, we also compared patients' pre-morbid BMI. This was estimated from participants' reports on their weight at age 40 (for those over 40) or else at age 20, with the intention of avoiding effects of the disease on BMI. Again, there was a highly significant difference with the cases having a higher mean for this measure of obesity.
A larger proportion of cases, 262 out of 1280, but only 48 out of 734 controls were reported to be diabetic (Odds Ratio 3.68, 95% CI 2.66 to 5.08) ( Table 2). Information about whether reported diabetes was Type 1 or Type 2 was not available. As expected, the prevalence of diabetes increased with age ( Figure 1(b)), and diabetes was significantly associated with cirrhosis risk only in patients aged over 40 years.
We also tested whether the differences between cases and controls showed variation between countries, with results shown in Supplementary Table 4.
When all the risk factors were tested together, using multiple logistic regression to identify independent effects on risk of alcohol-related cirrhosis (Table 3), the most significant effects were from cannabis use (protective), coffee and possibly tea consumption (each decreasing risk to a similar extent). Diabetes and pre-morbid BMI, but not current BMI, were associated with increased risk.

GenomALC participantsfamily history
Among those whose fathers had a reported alcohol problem, 21.5% of cases versus 9.4% of controls reported that their fathers died of liver disease (OR 2.64, 95% CI 1.68 to 4.14).
Among those whose mothers had a reported alcohol problem, 17.9% of cases versus 12.5% of controls reported that their mothers died of liver disease (OR 1.53, 95% CI 0.79 to 2.97).
We also tested for differential effects by sex of the participants, analysing effects on sons and daughters (male and female patients) separately ( Figure 2). Risk of cirrhosis was significantly increased in both male and female patients if the Father was reported as excessive alcohol user and to have died from liver disease. There were trends towards increased risk in both sexes if the Mother was affected, but these did not reach statistical significance. Combining data from all four groups gave an odds ratio of 2.25 (95% CI 1.55 -3.26).

UK Biobankcase-control comparisons
Means and distributions of alcohol-related characteristics for cases and controls from UK Biobank are shown in Supplementary Table 3. Ages were similar, but reported alcohol intake differed substantially, largely because of the minimum current drinking level required for controls but not cases, but perhaps also from reduction or cessation of alcohol intake by cases with poor health.
There were significant differences (Table 4) between cases and controls for prevalence of diabetes, obesity, coffee consumption, and smoking but not for cannabis use. Beverage preferences also differed significantly, with controls taking a higher proportion of their alcohol as wine (32%, against 26% for cases) and cases taking a higher proportion as spirits (15%, against 8% for controls).
To test all potential risk factors simultaneously and attempt to identify independent effects, multivariate logistic regression was performed with results shown in Table 5. Cannabis use was excluded from the multivariate analyses because it was only available for a subset of the UK Biobank participants and its inclusion in an analysis involving listwise deletion greatly reduced the available numbers. Coffee and tea consumption, measures of obesity and prevalence of diabetes were independently significant. When both BMI and WHR were included, their effects were in opposite directions, with higher WHR associated with higher

DISCUSSION
We have a number of important findings about factors associated with alcohol-related cirrhosis in high-risk drinkers. The novelty of the study lies in the fact that we used high-risk drinkers as controls, and well-defined selection of cases and controls allowed evaluation of the factors specifically altering risk for alcohol-related cirrhosis. Importantly, validation in an independent cohort enhances confidence in our results. Unlike previous studies that reported association with individual risk factors for alcohol-related cirrhosis, our study has simultaneously evaluated multiple potential aspects of risk in well characterised large cohorts of high-risk drinkers.

Alcohol use
Aspects of alcohol use, other than quantity, differed significantly between cases and controls and may affect risk of developing cirrhosis. In the GenomALC data, a higher proportion of total alcohol intake as wine was observed in the control group. When considered in the logistic regression model, a higher proportion of alcohol as wine was significantly associated with lower risk of cirrhosis but drinking with or between meals had no significant effect. The differential effect of wine, compared to other alcoholic beverages, is consistent with results of several previous studies (30-32) but we cannot distinguish between direct effects from some components of wine and confounding by other characteristics of drinkers who prefer wine.
Nor can we be sure that we are seeing a protective effect of wine rather than a harmful effect associated with a preference for other beverages, because the UK Biobank data suggest that a higher proportion of alcohol taken as spirits is associated with higher risk of cirrhosis. It would be inappropriate, and potentially harmful, to infer that wine consumption is beneficial.

Tea and coffee
We found replicated evidence for a protective effect of coffee consumption. In the GenomALC case-control comparison ( Table 2) controls were more likely to have been a coffee drinker during the period of excessive drinking and to have drunk more coffee per day.
In the UK Biobank data the number of cups of coffee per day was higher among the controls than cases (Table 4). These results are consistent with the reported protective effects of coffee on liver disease (14,33), on liver function test abnormality (14,34,35), and (at least in moderate amounts) on overall mortality (36). This is the first study to demonstrate an independent association of coffee in subjects with well-characterised alcohol use and cirrhosis directly assessed for this analysis. However, there is still uncertainty about which components of coffee confer protection and whether it is protective after liver damage is already present.
The GenomALC case-control comparison showed marginally significant protective effects of tea consumption when both tea and coffee were included in the multivariate analysis (Table   3). At least among the cases, tea and coffee tended to be alternative beverages; tea drinkers were less likely to drink coffee and vice versa. There were not many users of green tea (<10%) in our cohort, and there was only marginally significant protective effect ( Table 2). In similar UK Biobank comparisons, coffee and tea were each significantly associated with lower risk and had comparable effect sizes (Table 5).

Other substance use
Smoking was more common among controls than cases in the GenomALC participants (Table 2), and the UK Biobank data confirmed this (Table 4) with current smoking being more frequent and never smoking being less frequent in the controls. One interpretation could be that smoking is protective against cirrhosis, but this is contrary to its effects on most diseases and cannot be accepted without other evidence. It is possible that cases had more contact with the healthcare system than controls and had received more intensive and effective counselling about the risks of smoking, but this would not have affected the proportions who had never smoked. Even if smoking were protective against cirrhosis, its adverse impact on cardiovascular and respiratory diseases and cancers would outweigh any benefits.
There has been uncertainty about whether cannabis use is protective or harmful. However, a recent study of over 300,000 people with a past or current history of abusive alcohol use showed that cannabis use was associated with lower ORs for all stages of alcohol-related liver disease (19). Our GenomALC data showing cannabis use was more common among the controls confirms this ( Table 2). In addition, multivariate regression in the GenomALC cohort corroborated the association of cannabis as an independent protective factor for cirrhosis (37)(38)(39). Nevertheless many of the controls were recruited from addiction clinics and may have had other substance use disorders (including for cannabis) that could confound these results. In the UK Biobank, cannabis use had no significant effect but the proportion of participants with information on cannabis use was small. We observed that among GenomALC participants, younger patients were more likely to have used cannabis ( Figure   1(a)) but the ORs associated with reported cannabis use were consistent across age groups.
There is independent evidence for a biological link between liver damage and cannabinoids and/or cannabinoid receptors (37)(38)(39), and for the therapeutic potential of several components of the cannabinoid system against liver cirrhosis (40).

Obesity, diabetes and metabolic risk
Our expectation, based on previous reports, was that obesity would be a risk factor for cirrhosis. This was confirmed in the GenomALC case-control comparison ( Table 2), and when the effects of obesity and diabetes were considered together (Table 3) both were independently significant. Distinction between type 1 and type 2 diabetes was not specifically recorded in our data, but over 90% would be expected to be type 2 given the age range of our Page 20 of 38 study participants (41). Results in the UK Biobank were similar (Tables 4, 5) but waist/hip ratio showed a stronger association than BMI. Prevalence of diabetes increased with age, as expected ( Figure 1(b)), and high-risk drinkers who have diabetes in middle age are particularly likely to progress to cirrhosis. The association between obesity and/or diabetes and risk of cirrhosis, including alcohol-related cirrhosis, has been described in community based cohort studies (42)(43)(44) and may reflect a similarity with non-alcoholic liver disease, which is related to metabolic syndrome and dysregulation of carbohydrate and lipid metabolism.

Family History
Our data show that risk of alcohol-related liver disease is transmitted in families, as we previously reported for a subset of our patients (24). Familial/genetic risk is well-established for excessive alcohol intake or alcohol dependence (21), but not for the medical complications of alcohol use such as cirrhosis. The transmission from fathers to offspring was statistically significant, with a trend for similar risk transmission from mothers ( Figure 2).
This apparent difference in risk transmission from fathers and from mothers is likely due to chance, to lower incidence of cirrhosis in mothers (i.e. insufficient power) and/or recall bias by the study participants. Transmission of risk from parents to offspring is likely to be genetic, given the discovery in recent years of loci associated with alcohol-related cirrhosis (25,26,45). If differential transmission of risk from fathers and mothers is a real phenomenon, it may be mediated through genetic/epigenetic imprinting or other mechanisms of selective transmission from father versus mother; multigenerational epigenetic adaptation to hepatic wound healing response has been elucidated in animal models (46). Confirming or refuting such differential transmission will require replication in other studies with family data, or molecular studies on epigenetic changes in candidate genes (47).

Strengths and limitations
Our study design has both strengths and weaknesses. One of the issues to be addressed in planning a case-control study is the choice of appropriate criteria for the two groups. For the GenomALC cases, we restricted our recruitment to patients with alcohol-related cirrhosis and definition of criteria for this did not present any significant difficulty. The choice of controls was more complex; it is necessary to have a control group with alcohol intake which puts them at risk of cirrhosis and with similar lifetime alcohol exposure to the cases. In practice we recruited controls from clinics for treatment of substance use disorders and from detoxification facilities, accepting the risk that these controls might have different pattern of psychiatric comorbidities from the cases. In the data analysis, we sought to overcome the problem of non-causative differences between the GenomALC cases and controls by checking for consistency with results from a population-based second source of data, the UK Biobank.
The recruitment of GenomALC participants in six countries is a source of strength in that it provides diversity and allows comparison of results (see Supplementary Table 4). In general the results do not differ significantly across countries, except for cannabis use and possibly smoking status where heterogeneity is driven by stronger effects in France. The GenomALC participants were mostly of European descent and the extent to which our results can be generalised to other populations remains to be determined.
From the UK Biobank data, diagnoses of alcohol-related cirrhosis or alcohol-related liver disease were based on hospital discharge diagnoses or death certification. For the control group from UK Biobank, we cannot exclude liver disease and if it was present in a substantial proportion of these controls then power to detect effects on risk would be reduced. However, any such reduction in power may be mitigated by the much larger number of controls in the UK Biobank dataset. Reduction in power would lead to a failure to find a true difference Page 22 of 38 between cases and controls (false negative result) rather than producing a significant but false difference (false positive).
The GenomALC study was not prospective as patients were assessed after diagnosis, however the research questions were planned and the data collected were for the purposes of these analyses. The lack of prospective design is not a problem for assessment of genetic risk for which these patients were primarily recruited, but recall may be biased by patients' knowledge of their diagnosis, and some of the postulated risk factors such as BMI may change as a consequence of disease. Case-control differences may be causative but could also be due to modes of recruitment (particularly for other drug use, including smoking). Methods using instrumental variables such as Mendelian Randomisation can address causation, but they depend on genetic association results being available for the postulated causative factors.
Study design included definition of data and samples to be collected, but it is inevitable that questions will arise, often due to other research published during the course of a study, that were not envisaged at the outset. Although we have identified multiple risk factors for development of cirrhosis among high-risk drinkers, there are other factors such as variation in the microbiome (48), perhaps in turn associated with obesity, or infection with hepatotropic viruses other than B or C (49), about which we have no data.
A further limitation, which applies to many epidemiological studies, is that associations with risk may not reflect cause-and-effect relationships. For all risk factors, but particularly for the apparent effects of smoking, cannabis use and beverage preference (wine versus spirits) unmeasured confounders could produce the observed associations and we caution against changes in these areas without further evidence.

Conclusions
Page 23 of 38 We identified significant associations between family history of liver disease; diabetes and obesity; tea, coffee, wine and cannabis consumption, and risk of cirrhosis. Our findings may have public health consequences if the causal relationships can be confirmed; measures such as weight loss, intensive treatment of diabetes or pre-diabetic states, and encouragement of coffee consumption may be useful lifestyle interventions to reduce the risk of alcohol-related cirrhosis.        Table  Legends Outcome data 15* Report numbers in each exposure category, or summary measures of exposure Tables 1-5 Main results 16 (a) Give unadjusted estimates and, if applicable, confounder-adjusted estimates and their precision (eg, 95% confidence interval). Make Tables 1-5 clear which confounders were adjusted for and why they were included    and no other relationships or activities that could appear to have influenced the submitted work.

Study Highlights
What is known: Sustained high alcohol intake, often associated with alcohol dependence, can lead to alcoholrelated liver diseases including cirrhosis. The usual progression is through fatty liver, frequent in high-risk drinkers but reversible with abstinence, to fibrosis and cirrhosis. Some patients will develop alcoholic hepatitis, and some will develop hepatocellular carcinoma (HCC), generally with cirrhosis as a precursor. Therefore, cirrhosis is not only the end-stage of liver damage, but also increases risk for other life-threatening conditions. Apart from abstinence from alcohol, supportive measures, and liver transplantation in selected abstinent patients, current treatment options for alcohol-related cirrhosis are limited.
The relationship between alcohol intake and cirrhosis has been recognised since the late eighteenth century (1), with subsequent efforts to quantify this association made by Pequignot (2) who noted an increased risk of cirrhosis in people drinking more than 40 grams of alcohol per day. It is known that women are more susceptible to liver damage from alcohol than men (3), and larger studies and meta-analyses (4) have refined the threshold for detectable risk from alcohol intake.
It is notable that only a minority of high-risk drinkers develop cirrhosis. It is difficult to find reliable estimates, but in Denmark 7.7% of patients diagnosed with harmful alcohol use and 8.8% of those diagnosed with alcohol dependence developed cirrhosis over the subsequent 15 years (5). Meta-analysis (6) showed that 7-16% of people in alcohol problem cohorts had cirrhosis after 8-12 years. Variation in susceptibility may be due to genetic variation, and/or presence of other environmental and lifestyle risk factors which increase the probability of liver damage. Apart from alcohol intake and gender, obesity (also associated with nonalcoholic liver disease) has the strongest evidence for increasing risk of alcohol-related cirrhosis. For instance, liver biopsy histology showed more severe abnormalities in patients with alcohol use disorders with greater body weight (7); this was confirmed in a subsequent study (8) which showed that being overweight was a risk factor for steatosis, hepatitis and cirrhosis in addition to the effects of age, gender and duration of alcohol abuse. Other studies have also found an association between obesity or body mass index (BMI) and liver disease (9,10), fibrosis (11), alcoholic hepatitis (12) or HCC (13). There is evidence that coffee or tea consumption can reduce risk of liver disease or favourably affect biomarkers associated with liver disease (14)(15)(16)(17). Smoking has been associated with increased risk of alcohol-related cirrhosis and of cirrhosis in general, particularly among women (18). A recent report showed that cannabis use protected against liver disease in patients with alcohol use disorders (19), possibly through effects on inflammation mediated by cannabinoid receptors (20).
There is a lack of hard data from twin or family studies on genetic risk for alcohol-related cirrhosis. Alcohol dependence is partially heritable (21) but twin studies on its consequences such as alcohol-related liver disease (22) have been limited by small numbers and lack of adjustment for heritable effects on alcohol exposure (23). Our earlier report (24) suggested that a history of liver disease in a parent with alcohol problems was associated with increased risk of alcohol-related cirrhosis. The known genetic risk loci for cirrhosis in PNPLA3 and HSD17B13 (25,26) are associated with lipid metabolism and potentially with metabolic changes which accompany obesity.
The GenomALC Consortium (24) was initiated to gather data and samples for identification of risk factors for alcohol-related cirrhosis, including a case-control genetic association study.
In this paper we focus on comparison of case and control groups for potential clinical and phenotype factors that alter disease risk including beverage preference, other substance use, family history, obesity and diabetes. Where we have identified potential risk-altering factors from our data, we have attempted validation using comparable data from the UK Biobank.

GenomALC Study
Recruitment and data collection were based on our published GenomALC protocol (24). Two To be confident that participants either had or were at substantial risk of alcohol-related cirrhosis and to minimise the chance that cirrhosis was caused by factors other than alcohol, we recruited patients with alcohol intake of at least 80 grams per day for men and 50 grams per day for women for 10 years or more. This analysis was restricted to patients whose fathers or mothers were reported to have had 'problems with alcohol', and assumes that death from liver disease in a parent with alcohol problems is due to alcohol-related liver disease (potentially alcoholic hepatitis or HCC, as well as alcohol-related cirrhosis).

UK Biobank
Data control differences were consistent across country of recruitment, countries were treated as the strata and heterogeneity and common odds ratios were evaluated. ancestry, with the highest proportion in Germany (99%) and lowest in the US (88%).

GenomALC participantscase-control comparisons
Cases drank significantly less alcohol per day than controls, but had been drinking for significantly longer. Total lifetime alcohol intake did not differ significantly between male cases and controls, and in female cases was slightly lower than for controls (Table 1). A breakdown by country of recruitment is given in Supplementary Table 2, with comparisons of lifetime alcohol intake in cases and controls by country in Supplementary Figure 1. Controls reported taking a significantly higher proportion of their total alcohol in the form of wine (Table 2), but were less likely to report usually drinking with (rather than between) meals.
Forty eight percent of cases but only 28% of controls were currently living with a spouse or partner. There was no significant difference in years of education. Controls were more likely than cases to have been coffee drinkers during the time they were drinking alcohol heavily, and to have drunk more coffee per day, but there was no significant difference for tea consumption ( Table 2). A slightly higher but statistically significant proportion of controls reported drinking green tea (7% of cases and 9% of controls). Most people in both groups were or had been smokers, but the proportion was significantly higher in controls (83%) than cases (72%). Regular cannabis use was about three times more common among the controls (27%) than cases (9%) ( Table 2) but the proportion decreased with age (in both cases and controls) and the case-control difference was non-significant in patients aged over 60 years (Figure 1(a)).
Mean BMI was higher among the cases than the controls (Table 2). Because this difference might be secondary to the disease, e.g. through fluid retention in the cases or through inadequate diet in the controls, we also compared patients' pre-morbid BMI. This was estimated from participants' reports on their weight at age 40 (for those over 40) or else at age 20, with the intention of avoiding effects of the disease on BMI. Again, there was a highly significant difference with the cases having a higher mean for this measure of obesity.
A larger proportion of cases, 262 out of 1280, but only 48 out of 734 controls were reported to be diabetic (Odds Ratio 3.68, 95% CI 2.66 to 5.08) ( Table 2). Information about whether reported diabetes was Type 1 or Type 2 was not available. As expected, the prevalence of diabetes increased with age (Figure 1(b)), and diabetes was significantly associated with cirrhosis risk only in patients aged over 40 years.
We also tested whether the differences between cases and controls showed variation between countries, with results shown in Supplementary Table 4.
When all the risk factors were tested together, using multiple logistic regression to identify independent effects on risk of alcohol-related cirrhosis (Table 3), the most significant effects were from cannabis use (protective), coffee and possibly tea consumption (each decreasing risk to a similar extent). Diabetes and pre-morbid BMI, but not current BMI, were associated with increased risk.

GenomALC participantsfamily history
Among those whose fathers had a reported alcohol problem, 21.5% of cases versus 9.4% of controls reported that their fathers died of liver disease (OR 2.64, 95% CI 1.68 to 4.14).
Among those whose mothers had a reported alcohol problem, 17.9% of cases versus 12.5% of controls reported that their mothers died of liver disease (OR 1.53, 95% CI 0.79 to 2.97).
We also tested for differential effects by sex of the participants, analysing effects on sons and daughters (male and female patients) separately ( Figure 2). Risk of cirrhosis was significantly increased in both male and female patients if the Father was reported as excessive alcohol user and to have died from liver disease. There were trends towards increased risk in both sexes if the Mother was affected, but these did not reach statistical significance. Combining data from all four groups gave an odds ratio of 2.25 (95% CI 1.55 -3.26).

UK Biobankcase-control comparisons
Means and distributions of alcohol-related characteristics for cases and controls from UK Biobank are shown in Supplementary Table 3. Ages were similar, but reported alcohol intake differed substantially, largely because of the minimum current drinking level required for controls but not cases, but perhaps also from reduction or cessation of alcohol intake by cases with poor health.
There were significant differences (Table 4) between cases and controls for prevalence of diabetes, obesity, coffee consumption, and smoking but not for cannabis use. Beverage preferences also differed significantly, with controls taking a higher proportion of their alcohol as wine (32%, against 26% for cases) and cases taking a higher proportion as spirits (15%, against 8% for controls).
To test all potential risk factors simultaneously and attempt to identify independent effects, multivariate logistic regression was performed with results shown in Table 5. Cannabis use was excluded from the multivariate analyses because it was only available for a subset of the UK Biobank participants and its inclusion in an analysis involving listwise deletion greatly reduced the available numbers. Coffee and tea consumption, measures of obesity and prevalence of diabetes were independently significant. When both BMI and WHR were included, their effects were in opposite directions, with higher WHR associated with higher

DISCUSSION
We have a number of important findings about factors associated with alcohol-related cirrhosis in high-risk drinkers. The novelty of the study lies in the fact that we used high-risk drinkers as controls, and well-defined selection of cases and controls allowed evaluation of the factors specifically altering risk for alcohol-related cirrhosis. Importantly, validation in an independent cohort enhances confidence in our results. Unlike previous studies that reported association with individual risk factors for alcohol-related cirrhosis, our study has simultaneously evaluated multiple potential aspects of risk in well characterised large cohorts of high-risk drinkers.

Alcohol use
Aspects of alcohol use, other than quantity, differed significantly between cases and controls and may affect risk of developing cirrhosis. In the GenomALC data, a higher proportion of total alcohol intake as wine was observed in the control group. When considered in the logistic regression model, a higher proportion of alcohol as wine was significantly associated with lower risk of cirrhosis but drinking with or between meals had no significant effect. The differential effect of wine, compared to other alcoholic beverages, is consistent with results of several previous studies (30-32) but we cannot distinguish between direct effects from some components of wine and confounding by other characteristics of drinkers who prefer wine.
Nor can we be sure that we are seeing a protective effect of wine rather than a harmful effect associated with a preference for other beverages, because the UK Biobank data suggest that a higher proportion of alcohol taken as spirits is associated with higher risk of cirrhosis. It would be inappropriate, and potentially harmful, to infer that wine consumption is beneficial.

Tea and coffee
We found replicated evidence for a protective effect of coffee consumption. In the GenomALC case-control comparison ( Table 2) controls were more likely to have been a coffee drinker during the period of excessive drinking and to have drunk more coffee per day.
In the UK Biobank data the number of cups of coffee per day was higher among the controls than cases (Table 4). These results are consistent with the reported protective effects of coffee on liver disease (14,33), on liver function test abnormality (14,34,35), and (at least in moderate amounts) on overall mortality (36). This is the first study to demonstrate an independent association of coffee in subjects with well-characterised alcohol use and cirrhosis directly assessed for this analysis. However, there is still uncertainty about which components of coffee confer protection and whether it is protective after liver damage is already present.
The GenomALC case-control comparison showed marginally significant protective effects of tea consumption when both tea and coffee were included in the multivariate analysis (Table   3). At least among the cases, tea and coffee tended to be alternative beverages; tea drinkers were less likely to drink coffee and vice versa. There were not many users of green tea (<10%) in our cohort, and there was only marginally significant protective effect (Table 2). In similar UK Biobank comparisons, coffee and tea were each significantly associated with lower risk and had comparable effect sizes (Table 5).

Other substance use
Smoking was more common among controls than cases in the GenomALC participants (Table 2), and the UK Biobank data confirmed this (Table 4) with current smoking being more frequent and never smoking being less frequent in the controls. One interpretation could be that smoking is protective against cirrhosis, but this is contrary to its effects on most diseases and cannot be accepted without other evidence. It is possible that cases had more contact with the healthcare system than controls and had received more intensive and effective counselling about the risks of smoking, but this would not have affected the proportions who had never smoked. Even if smoking were protective against cirrhosis, its adverse impact on cardiovascular and respiratory diseases and cancers would outweigh any benefits.
There has been uncertainty about whether cannabis use is protective or harmful. However, a recent study of over 300,000 people with a past or current history of abusive alcohol use showed that cannabis use was associated with lower ORs for all stages of alcohol-related liver disease (19). Our GenomALC data showing cannabis use was more common among the controls confirms this ( Table 2). In addition, multivariate regression in the GenomALC cohort corroborated the association of cannabis as an independent protective factor for cirrhosis (37)(38)(39). Nevertheless many of the controls were recruited from addiction clinics and may have had other substance use disorders (including for cannabis) that could confound these results. In the UK Biobank, cannabis use had no significant effect but the proportion of participants with information on cannabis use was small. We observed that among GenomALC participants, younger patients were more likely to have used cannabis ( Figure   1(a)) but the ORs associated with reported cannabis use were consistent across age groups.
There is independent evidence for a biological link between liver damage and cannabinoids and/or cannabinoid receptors (37)(38)(39), and for the therapeutic potential of several components of the cannabinoid system against liver cirrhosis (40).

Obesity, diabetes and metabolic risk
Our expectation, based on previous reports, was that obesity would be a risk factor for cirrhosis. This was confirmed in the GenomALC case-control comparison (Table 2), and when the effects of obesity and diabetes were considered together (Table 3) both were independently significant. Distinction between type 1 and type 2 diabetes was not specifically recorded in our data, but over 90% would be expected to be type 2 given the age range of our study participants (41). Results in the UK Biobank were similar (Tables 4, 5) but waist/hip ratio showed a stronger association than BMI. Prevalence of diabetes increased with age, as expected (Figure 1(b)), and high-risk drinkers who have diabetes in middle age are particularly likely to progress to cirrhosis. The association between obesity and/or diabetes and risk of cirrhosis, including alcohol-related cirrhosis, has been described in community based cohort studies (42)(43)(44) and may reflect a similarity with non-alcoholic liver disease, which is related to metabolic syndrome and dysregulation of carbohydrate and lipid metabolism.

Family History
Our data show that risk of alcohol-related liver disease is transmitted in families, as we previously reported for a subset of our patients (24). Familial/genetic risk is well-established for excessive alcohol intake or alcohol dependence (21), but not for the medical complications of alcohol use such as cirrhosis. The transmission from fathers to offspring was statistically significant, with a trend for similar risk transmission from mothers ( Figure 2).
This apparent difference in risk transmission from fathers and from mothers is likely due to chance, to lower incidence of cirrhosis in mothers (i.e. insufficient power) and/or recall bias by the study participants. Transmission of risk from parents to offspring is likely to be genetic, given the discovery in recent years of loci associated with alcohol-related cirrhosis (25,26,45). If differential transmission of risk from fathers and mothers is a real phenomenon, it may be mediated through genetic/epigenetic imprinting or other mechanisms of selective transmission from father versus mother; multigenerational epigenetic adaptation to hepatic wound healing response has been elucidated in animal models (46). Confirming or refuting such differential transmission will require replication in other studies with family data, or molecular studies on epigenetic changes in candidate genes (47).

Strengths and limitations
Our study design has both strengths and weaknesses. One of the issues to be addressed in planning a case-control study is the choice of appropriate criteria for the two groups. For the GenomALC cases, we restricted our recruitment to patients with alcohol-related cirrhosis and definition of criteria for this did not present any significant difficulty. The choice of controls was more complex; it is necessary to have a control group with alcohol intake which puts them at risk of cirrhosis and with similar lifetime alcohol exposure to the cases. In practice we recruited controls from clinics for treatment of substance use disorders and from detoxification facilities, accepting the risk that these controls might have different pattern of psychiatric comorbidities from the cases. In the data analysis, we sought to overcome the problem of non-causative differences between the GenomALC cases and controls by checking for consistency with results from a population-based second source of data, the UK Biobank.
The recruitment of GenomALC participants in six countries is a source of strength in that it provides diversity and allows comparison of results (see Supplementary Table 4). In general the results do not differ significantly across countries, except for cannabis use and possibly smoking status where heterogeneity is driven by stronger effects in France. The GenomALC participants were mostly of European descent and the extent to which our results can be generalised to other populations remains to be determined.
From the UK Biobank data, diagnoses of alcohol-related cirrhosis or alcohol-related liver disease were based on hospital discharge diagnoses or death certification. For the control group from UK Biobank, we cannot exclude liver disease and if it was present in a substantial proportion of these controls then power to detect effects on risk would be reduced. However, any such reduction in power may be mitigated by the much larger number of controls in the UK Biobank dataset. Reduction in power would lead to a failure to find a true difference Page 22 of 38 between cases and controls (false negative result) rather than producing a significant but false difference (false positive).
The GenomALC study was not prospective as patients were assessed after diagnosis, however the research questions were planned and the data collected were for the purposes of these analyses. The lack of prospective design is not a problem for assessment of genetic risk for which these patients were primarily recruited, but recall may be biased by patients' knowledge of their diagnosis, and some of the postulated risk factors such as BMI may change as a consequence of disease. Case-control differences may be causative but could also be due to modes of recruitment (particularly for other drug use, including smoking). Methods using instrumental variables such as Mendelian Randomisation can address causation, but they depend on genetic association results being available for the postulated causative factors.
Study design included definition of data and samples to be collected, but it is inevitable that questions will arise, often due to other research published during the course of a study, that were not envisaged at the outset. Although we have identified multiple risk factors for development of cirrhosis among high-risk drinkers, there are other factors such as variation in the microbiome (48), perhaps in turn associated with obesity, or infection with hepatotropic viruses other than B or C (49), about which we have no data.
A further limitation, which applies to many epidemiological studies, is that associations with risk may not reflect cause-and-effect relationships. For all risk factors, but particularly for the apparent effects of smoking, cannabis use and beverage preference (wine versus spirits) unmeasured confounders could produce the observed associations and we caution against changes in these areas without further evidence.

Conclusions
Page 23 of 38 We identified significant associations between family history of liver disease; diabetes and obesity; tea, coffee, wine and cannabis consumption, and risk of cirrhosis. Our findings may have public health consequences if the causal relationships can be confirmed; measures such as weight loss, intensive treatment of diabetes or pre-diabetic states, and encouragement of coffee consumption may be useful lifestyle interventions to reduce the risk of alcohol-related cirrhosis.        Table  Legends Outcome data 15* Report numbers in each exposure category, or summary measures of exposure Tables 1-5 Main results 16 (a) Give unadjusted estimates and, if applicable, confounder-adjusted estimates and their precision (eg, 95% confidence interval). Make Tables 1-5 clear which confounders were adjusted for and why they were included (b) Report category boundaries when continuous variables were categorized N/A, except for Figure 1 (see Figure) (   *The 'Drink with meals' variable was recoded as 'Mostly drink with meals' = 1, 'Mostly drink between meals' = 0, 'Both' = 0.5. Figure 1. Estimated lifetime alcohol intake in Cases and Controls, by country of recruitment.

RESPONSE TO EDITORS' AND REVIEWERS' COMMENTS
We thank the editors and reviewers for their comments, and appreciate the opportunity to submit a revised version of this paper.
In the text below, the editors' or reviewers' comments are in italics and our responses are indented below each comment. Page numbers, where mentioned below, refer to the revision.

Editor/Editorial Board Comments:
The editorial board asks the authors to address the following points: 1. Adjust for duration of alcohol use, and not only for accumulated quantity of alcohol consumed.
We assume the concern arises from the difference in years of drinking between cases and controls in the GenomALC cohort, and the possibility that duration of alcohol use might have effects independent of lifetime quantity.
We have re-run the logistic regression in Table 3 with additional independent variables. These are sex, age, alcohol intake in grams/day, duration of excessive alcohol use, but not lifetime quantity because this is the product of quantity and duration. Duration of excessive alcohol use is significant (consistent with the results in Table 1). This makes little difference to the HR estimates and p-values for the other variables, but we have changed Table 3 and its legend to reflect the duration-of-alcohol-use effect.
For consistency, we also repeated the logistic regression for UK Biobank data with addition of sex, and age at assessment. It was not appropriate to include alcohol intake (grams/day) because we only have data on consumption at the time of assessment and this is low in the cases (Supplementary Table 3) because many participants declared that they had reduced their intake compared to 10 years previously; and it was not possible to include duration of excessive intake because this was not available. This made no difference to HR estimates for the other variables, but we have changed Table 5 and its legend to be as consistent as possible with Table 3.

Revise title to make it more clear what the protective effects were in this study.
We have done this, by changing from 'Factors that alter risk for alcoholrelated cirrhosis in high-risk drinkers' to 'Obesity, diabetes, coffee, tea and cannabis use alter risk for alcohol-related cirrhosis in two large cohorts of high-risk drinkers'.

Discuss the possible effects of the microbiome on ALD in individuals at risk due to heavy alcohol use.
We have no data or relevant samples to allow us to do more than mention this in the Discussion. We have expanded the section discussing the strengths and limitations of our study by noting that there are potential risk factors which we did not include in our protocol and cannot provide evidence about. These include the microbiome (Bajaj JS, Nat Rev Gastroenterol Hepatol 2019;16 235-246) and no doubt other factors.

Make clear to the readers that the data is NOT about promoting wine over other alcoholic beverages when it comes to ALD.
We agree that the associations between beverage preference and cirrhosis risk do not necessarily mean there is a causal relationship. There are many potential confounders, and we have now given additional emphasis to this in the Discussion (pages 17 and 22).

Provide a country-wise comparison of the GenomALC data.
We have given information about variation in recruitment, and possible heterogeneity of results, between countries in Supplementary Tables 2 and 4 and Supplementary Figure 1 of the original submission. These give the numbers of cases and controls by country, lifetime alcohol intake by country, and test for heterogeneity of effect sizes by country. We have added a comment to the Discussion about possible differences between countries on page 21.
6. Recognize the limitations of the data analysis regarding the lack of information on noncirrhotic ALD among the control groups. Even though these heavy drinkers didn't develop cirrhosis (as defined by the authors), they could still have developed other forms of ALD or certainly have had other alcohol-related illnesses that were not captured in this study.
At least for the GenomALC participants, we are comfortable that controls did not have non-cirrhotic ALD, had not had it in the past, and (because they were matched for age with the cases) had reached an age and lifetime alcohol exposure where they would probably have had it if they were going to. Even if some controls did have undetected ALD (which is more likely for the UKB participants), this would minimise the difference between cases and controls and reduce power rather than producing an apparently significant but untrue difference. A comment to this effect has been added to the Discussion (pages [21][22].

Reviewer #1: Comments to the Authors:
The authors investigated a multi-center case-control study comparing 1293 cases and 754 controls. The authors identified significant associations between family history of liver disease; diabetes and obesity; tea, coffee, wine and cannabis consumption, and risk of cirrhosis. It is an interesting study though, there are some issues that need to be clarified.
1) Some people say that "pure non-alcoholic steatohepatitis "patients are very limited population as most of so-called NASH patients have some history of alcohol consumption and its amount is not enough to be diagnosed as Alcoholic steatohepatitis. The authors' data mention that cases had a higher prevalence of diabetes and pre-morbid BMI than controls.
Since these parameters are considered as metabolic factors, I would like to know the data of lipid metabolism and hypertension in both groups.
We accept that the boundaries between NASH and alcohol-related cirrhosis may be blurred and that risk can be additive or worse across these causes. As far as lipids and hypertension go, we only have data on lipids from the time of recruitment and this most likely reflects changes caused by liver disease rather than reflecting lipid-related risk. The same applies to prevalence of hypertension or to mean blood pressure. Future work incorporating genetic information (risk scores for lipids or blood pressure) may allow testing for such effects but that is beyond the scope of this paper.

2) Was hepatitis E checked in both cases and controls? This is because hepatitis E infection is more frequent in cirrhotic patients and this may cause additional liver dysfunction.
We did not gather data on hepatitis E so we cannot address this directly. We note the recent report of a high proportion of patients with alcohol-related cirrhosis being seropositive for hepatitis E (Fantilli et al. Unexpected high seroprevalence of hepatitis E virus in patients with alcohol-related cirrhosis. PLoS One. 2019 Oct 24;14(10):e0224404), and we cannot exclude the possibility that previous hepatitis E infection may increase the probability of cirrhosis among excessive drinkers. This is now mentioned in the Discussion (page 22). .

3) I would like to know if there is any difference regarding the incidence of satellitosis (neutrophil infiltration confirmed by liver biopsy) in both groups.
Biopsies had only been carried out in patients for whom this had been considered clinically necessary (30% of cases and 1% of controls) and we did not collect detailed information on the liver biopsy appearances. Therefore we cannot address this question, of whether there was a case-control difference for satellitosis.

Reviewer #2:
Thank you to the authors for their work. Two questions come to mind: 1. The GenomALC data includes population from the UK. In your paper the population