Genetic risk of fatty liver disease and mortality in the general population: A Mendelian randomization study

Fatty liver disease has been associated with higher all‐cause as well as liver‐related, ischemic heart disease (IHD)‐related and extrahepatic cancer‐related mortality in observational epidemiological studies. We tested the hypothesis that fatty liver disease is a causal risk factor for higher mortality.


| INTRODUC TI ON
Fatty liver disease has become one of the most common diseases worldwide. 1,2 Previously it was an uncommon disease mainly caused by habitual high alcohol consumption. Presently, the obesity pandemic is driving a burgeoning in fatty liver disease, which now affects between 25% and 40% of the adult population in Westernized societies. The first stage of fatty liver disease is hepatic steatosis, which is characterized by an accumulation of excess fat (triglycerides) inside liver cells. A fraction of those with hepatic steatosis develops steatohepatitis, an inflammatory state that promotes the development of hepatic cirrhosis and hepatocellular carcinoma, both severe and irreversible liver diseases.
Apart from its effects on hepatic outcomes, fatty liver disease is associated with extrahepatic disorders, including ischemic heart disease (IHD), type 2 diabetes, extrahepatic cancers, and chronic kidney disease, and with a high rate of all-cause mortality. [3][4][5][6][7][8] These associations, which have been observed in multiple large-scale epidemiological studies, have led to the hypothesis that fatty liver disease might be a causal driver of extrahepatic morbidity and mortality. However, it remains unclear if these associations reflect causal effects of fatty liver disease rather than shared underlying risk factors (confounding), or effects of the extrahepatic disorders on fatty liver disease (reverse causation).
Mendelian randomization is an epidemiological method that uses genetic variants with known effects on an exposure (here: fatty liver disease) to examine the causal relationship between the exposure and an outcome (here: mortality). 9 The method utilizes the fact that genetic variants are generally unconfounded and unaffected by reverse causation. If fatty liver disease is a causal risk factor for extrahepatic and all-cause mortality, genetic variants that cause fatty liver disease should cause a higher rate of extrahepatic and all-cause mortality.
The aim of this study was to use Mendelian randomization to test if the association between fatty liver disease and mortality is causal.
For this purpose, we genotyped seven variants (in PNPLA3, TM6SF2, HSD17B13, MTARC1, MBOAT7, GCKR, and GPAM) known to be associated with fatty liver disease in 110 913 individuals from the Danish general population, of whom 16 119 died during a median follow-up of 9.5 years. We tested whether the genetic variants (individually, combined into risk scores, or used in instrumental variable analyses), were associated with all-cause, liver-related, IHD-related, and extrahepatic cancer-related mortality.

| Population
As previously described, we combined two studies, the Copenhagen City Heart Study (CCHS) (n = 10 061) and the Copenhagen General Population Study (CGPS) (n = 100 852), into one cohort, referred to here as CCHS+CGPS. 10 Table S1. The studies were approved by institutional review boards and Danish ethical committees and were conducted according to the Declaration of Helsinki. Written informed consent was obtained from participants.
A detailed description of the UK Biobank can be found elsewhere. 12 In brief, the cohort consists of approximately 500 000 individuals from the general population in the United Kingdom.
Participants were aged 40 to 69 at inclusion. Comprehensive information on medical outcomes, biochemistry, genotypes, and baseline covariates is available on all participants. Our last update of the UK Biobank was the 1st of July 2021.

| Biochemistry
We measured plasma levels of alanine transaminase (ALT) and other biochemical parameters at the time of study inclusion using standard hospital assays. We dichotomized ALT at 70 U/L for men and 45 U/L for women. These are the sex-specific cut-off values for abnormally high ALT, we currently use in Denmark. 13 In a sensitivity analysis, we used a cut-off value of 32 U/L for women and 44 U/L for men, values found to reflect the upper normal limits in the US population. 14

| Covariates
We calculated body mass index (BMI) as measured weight in kilograms divided by measured height in meters squared. Participants reported current weekly alcohol consumption in units (1 unit = 12 g of alcohol). We defined diabetes mellitus (including both type 1 and type 2) as individuals with one or more of the following ICD codes in the national Danish Patient Registry: International Classification of Diseases (ICD)-10 E10, E11, E13, or E14 and/or ICD-8249 or 250.

Key points
Genetic variation that increases the risk of fatty liver disease also increases the risk of liver-related death. This supports that fatty liver disease per se is a causal driver of liver-related mortality.

| Computerized tomography of the liver
We invited a randomly selected subset of participants from the CGPS for a computerized tomography (CT) scan (Aquilion One; Toshiba Medical Systems Corporation, Tokyo, Japan). 10 The examination included a low-dose calcium scan of the thorax and upper abdomen.
CT scans were available in n = 6965. We gauged attenuation in a 1.5 cm 2 diameter area of liver segment 5 on the CT scan of the upper abdomen. CT attenuation is measured in Hounsfield Units(HU) and is inversely associated with liver fat content. 15 Steatosis was defined by hepatic CT attenuation <48 HU because specificity for moderate to severe hepatic steatosis (>30% liver fat) is 100% at this cut-off. 16 In a sensitivity analysis, we used a less specific cutoff of <57 HU, which corresponds to >5% hepatic fat content quantified by magnetic resonance imaging. 16,17

| Genotyping
The variants PNPLA3 p.Ile148Met (rs738409), TM6SF2 p.Glu167Lys (rs1260326), and GPAM T > C (rs58175211) were genotyped using TaqMan assays (ABI PRISM 7900HT Sequence Detection System, Applied Biosystems, Foster City, CA), as previously described. 11,18,19 The rs58175211 variant in GPAM is in high linkage (r 2 = 0.88) with the rs2792751 missense-variant recently found to associate with fatty liver disease. 20 Genotyping in the UK Biobank has been described elsewhere. 19 In the UK Biobank, we selected the rs2792751 variant in GPAM. Genotypes were coded as 0 for reference allele homozygotes, 1 for fatty liver disease risk-increasing allele heterozygotes, and 2 for risk-increasing allele homozygotes.
The genetic risk score was derived from the three variants in PNPLA3, TM6SF2, and HSD17B13 and calculated as the total number of risk-increasing alleles (range: minimum of 0 and maximum of 6). Since there were too few individuals with a score of 6, they were combined with the individuals who had a score of 5 for further analyses. We defined cause-specific death as follows: Liver-related mortality was due to cirrhosis and/or liver cancer; IHD-related mortality was due to chronic angina and/or myocardial infraction; and extrahepatic cancer was any cancer excluding liver cancer. The classification was based on ICD-10 or ICD-8 codes in the registries. We chose the same ICD-10 codes to define liver-related mortality in the UK Biobank. The specific codes are in Table S1.

| Statistical analysis
All analyses were performed using R statistical software version 4.1.0 (R Foundation for Statistical Computing, Vienna, Austria).
Differences in baseline characteristics were tested with the χ 2 test for categorical traits and the Kruskal-Wallis rank sum test for continuous traits. We chose age as the underlying timescale and calculated the cumulative all-cause mortality stratified by hepatic steatosis (<48 HU vs. ≥48 HU) in the subset with available with CT-scans, and the cumulative all-cause and cause-specific mortality in the full cohort stratified by elevated ALT (dichotomized at ≤45 U/L vs. >45 U/L in women and ≤70 U/L vs. >70 U/L in men), genetic risk variants, or genetic risk score, using the Aalen-Johansen estimator. When investigating liver-related, IHD-related, and extrahepatic cancer-related mortality, we entered mortality from other causes as a competing risk into the model. Cox proportional hazard models estimated hazard ratios (HRs) for all-cause and cause-specific mortality with 95% confidence intervals (95% CIs) and P for trend. In the CGPS+CCHS, we adjusted the Cox proportional hazard models for sex and age for the observational analyses and additionally for cohort (CCHS vs. CGPS) in the genetic analyses. We also added adjustments for date of birth, cohort (in the observational analyses), BMI, alcohol consumption, smoking, diabetes, and lipid-lowering therapy in multivariable adjusted models. In the UK Biobank, we adjusted the Cox proportional hazard models for sex, age, and Principal Components 1-10. We included two-factor interaction terms in the model to test for interaction between BMI, alcohol consumption or diabetes and the genetic risk score on mortality in the CGPS+CCHS. Potential interactions were visualized by plotting hazard ratios and 95% CIs stratified by alcohol intake (in quartiles), BMI (in quartiles), or diabetes (yes/no). For age, sex, date of birth, cohort, and smoking we performed similar tests for interaction, and found no evidence for interaction.
We used a Mendelian randomization framework to test whether the effects of the genetic variants on liver-related mortality were consistent with their effects on liver steatosis and plasma ALT. We excluded the risk allele in HSD17B13 from the analyses for steatosis because this allele affects the later stages of fatty liver disease, but not steatosis. 21 We used linear regression, adjusted for age, sex and cohort, to estimate the effect of the individual risk alleles on continuous hepatic CT-attenuation (in HUs) and plasma ALT. All alleles were simultaneously entered into a regression to evaluate the percentage of variance explained (quantified by the adjusted R-squared for the model) and the strength of the genetic instrument measured by the F-statistic (an F-statistic >10 indicates low risk of weak instrument bias). For each variant, we plotted the per-allele beta-coefficients for liver attenuation and ALT against the beta-coefficients from a sex, age, and cohort adjusted Cox proportional hazard model on liver-related mortality and added an inverse variance weighted (IVW) regression line. The slope of the regression line is the causal estimate of a genetically determined 1-unit increase in exposure (here: liver attenuation on CT-scan or plasma ALT) and is referred to as the IVW-estimate. In sensitivity analyses, we recalculated the regression estimate with the weighted median, MR-Egger, and MR-PRESSO methods. These methods differ in their ability to detect various violations of the assumptions underlying the Mendelian randomization method. [22][23][24] The estimates for these different methods are consistent if most of the alleles included in the genetic instrument are valid and no directional pleiotropy is present. We also repeated the above-described analyses using log-transformed ALT and HU, and using effects on ALT and HU derived from controls only (i.e., individuals that did not die of liver disease during follow-up). To validate positive associations identified in the Mendelian randomization analyses in the CCHS+CGPS, we repeated the analyses in the UK Biobank 25 with exposure weights from the CCHS+CGPS (i.e. two-sample Mendelian randomization).

| Baseline characteristics
Baseline characteristics of the 110 913 individuals are shown in

| Hepatic steatosis and mortality
We CT-scanned 6965 individuals (Table S2)  (11%) had hepatic attenuation values of less than 48 HU, consistent with moderate to severe steatosis [17]. Of these, 31 died during follow-up.
Hepatic steatosis was associated with all-cause mortality ( Figure 1, panel A)

| Elevated plasma alanine transaminase (ALT) and mortality
We measured baseline ALT in 109 482 individuals (Table S3)  Elevated ALT was associated robustly with all-cause, liverrelated, and extrahepatic cancer-related mortality, but not with IHD-related mortality. The age and sex adjusted HR for all-cause mortality was 1.26 (95% CI: 1.14-1.40, p-value = 1 × 10 −5 ) for individuals with elevated ALT as compared to those with normal levels ( Figure 1, panel B). The corresponding HRs for liver-related and extrahepatic cancer-related mortality were 9.14 (95% CI: 6.18-13.50; p-value = 1 × 10 −28 ) (Figure 1 Repeating the analyses using less stringent cutoffs for elevated ALT (>32 U/L for women and > 44 U/L for men) yielded similar results, although the HRs were attenuated ( Figure S4). For example, the age and sex adjusted HR for all-cause mortality was 1.08 (95% CI: 1.02-1.15, p-value = 0.01) for individuals with elevated ALT as compared to those with normal levels. The corresponding HR after multifactorial adjustments was 1.19 (95% CI: 1.12-1.27, p-value = 1 × 10 −7 ). were individually associated with liver-related mortality. The statistically strongest association was seen for the PNPLA3 variant.

| Individual risk variants and mortality
As compared to non-carriers those homozygous for the PNPLA3 risk allele had almost 3-fold higher risk (P for trend = 2 × 10 −5 ) ( Figure 2, panel A). Homozygosity for the risk allele in TM6SF2 was associated with a six-fold higher risk of liver-related mortality, the highest risk of any single variant (P for trend = 0.05) ( Figure 2, panel B). The risk allele in HSD17B13 increased the liverrelated mortality risk 1.39-fold in homozygotes (P for trend = 0.02) ( Figure 2, panel C). The MBOAT7 allele which has previously been associated with a minor steatogenic effect 18,26,27 Figure 2, panels E-G; Figure S5). The risk allele in GPAM was associated with a marginal decreased risk of IHD-related (P for trend = 0.04). The six other variants were not associated with IHD-related mortality. None of the seven variants were individually associated with all-cause or extrahepatic cancerrelated mortality ( Figure S5).

TA B L E 1 Baseline characteristics of individuals in the Copenhagen City Heart
Study and the Copenhagen General Population Study.

F I G U R E 1
Hepatic steatosis, elevated ALT, and mortality in the general population. (A) Cumulative all-cause mortality as a function of age, stratified by liver steatosis (marked by liver tissue attenuation and measured in HUs on CT-scans of the liver). HU < 48 is consistent with moderate to severe steatosis. (B) Cumulative all-cause mortality as a function of age, stratified by liver cell damage(steatohepatitis) marked by baseline plasma ALT. An ALT value >70 U/L for men and >45 U/L for women was defined as elevated (C) Cumulative liver-related mortality as a function of age stratified by liver cell damage(steatohepatitis) as in (B), with death due to other causes entered as a competing risk factor in the model. For all panels, the curves were estimated with the Aalen-Johansen method while hazard ratios and 95% CI were calculated with a sex and age adjusted Cox regression model. ALT, alanine transaminase; CI, confidence interval, HU, Hounsfield units.

| Genetic risk score and mortality
We combined the individual variants in PNPLA3, TM6SF2 and HSD17B13 into a genetic risk score, ranging from zero to six for number of risk-increasing alleles, as previously shown in the CCHS+CGPS [20].
An increasing genetic risk score was associated with higher liverrelated but not with all-cause mortality (Figure 3), IHD-related, or cancer-related mortality. Liver-related mortality increased stepwise with higher genetic risk score, reaching a 10-fold higher risk for carriers of 5 or 6 risk alleles as compared to those with 0 risk alleles ( Figure 3, panel B). Adding the risk alleles at MBOAT7, MTARC1, GCKR, or GPAM to the score did not improve the associations with allcause or cause-specific mortality. Weighting the alleles at PNPLA3, TM6SF2, and HSD17B13 by their individual effects on ALT before calculating the score yielded similar results for the association with liver-related mortality and did not change the null associations for all-cause, IHD-related, or cancer-related mortality (Table S11).
Adiposity, heavy alcohol consumption and diabetes are all known to amplify the genetic risk of fatty liver disease. 11,19,28 We therefore tested for interaction between these risk factors and the genetic risk score on liver-related mortality. There were no interactions ( Figure S6).

| Correlation between genetic effects on hepatic steatosis and mortality
The effects of the genetic variants on higher liver fat content in an independent cohort, we repeated the analysis in UK Biobank.
The estimates from UK Biobank were very similar to those in the CCHS+CGPS ( Figure S7). In sensitivity analyses, we repeated the instrumental variable analysis using log-transformed HU, or with HU-associations derived from persons who did not die of liver disease during follow-up. The results remained unchanged (Tables S12 and S13).

| Correlation between genetic effects on plasma ALT and mortality
The effects of the genetic variants on higher plasma ALT correlated with their effects on higher liver-related mortality (Figure 4, panel B).
In instrumental variable analysis, a genetically proxied 1 U/L higher ALT associated with a 1.34-fold increase in liver-related mortality  Figure S7). We also repeated the instrumental variable analysis using log-transformed ALT, or with ALT-associations derived from individuals who did not die of liver disease during follow-up. The results were similar (Tables S12 and S13).

| CON CLUS IONS
In this study, we used genetics to study the causal relationship between fatty liver disease and mortality. The study has two principal findings. First, we found that genetic variants that associate with hepatic steatosis and/or increased plasma ALT also associated with F I G U R E 3 Genetic risk score for fatty liver disease and all-cause and liver-related mortality in the general population. (A) Cumulative allcause mortality as a function of age, stratified by genetic risk score for fatty liver disease. (B) Cumulative liver-related mortality as a function of age and genetic risk score for fatty liver disease, with death due to other causes entered as a competing risk factor in the model. The genetic risk score was calculated by counting the total number of risk-increasing alleles in PNPLA3, TM6SF2, and HSD17B13 for each person. The score ranged from a minimum of 0 to a maximum of 6 for risk-increasing alleles. The curves were estimated with the Aalen-Johansen method while hazard ratios and 95% confidence intervals were calculated with a sex-, age-, and cohort-adjusted Cox regression model.
increased liver-related mortality in a dose-dependent manner. This supports that fatty liver disease is a causal risk factor for liver-related mortality. Second, the genetic variants were not robustly associated with cardiovascular, extrahepatic cancer-related, or all-cause mortality. This suggests that in the general population, the contribution of fatty liver disease to all-cause mortality is minimal compared to that of other diseases.
In support of our findings, fatty liver disease was associated with increased liver-related mortality in previous observational epidemiological studies. [4][5][6][7] For example, Simon and colleagues reported an 18-fold higher risk of death due to cirrhosis in Swedish individuals with biopsy-confirmed nonalcoholic fatty liver disease as compared to age and sex matched population controls. 4 Likewise, Unalp-Arida and Ruhl found a seven-fold higher risk of liver-related death in participants from the Third National Health and Nutrition Examination Survey (NHANES III) with a high baseline risk for nonalcoholic fatty liver as compared to those with a low baseline risk. 6,29 In agreement, we found a nine-fold higher risk of liver-related death in individuals with elevated plasma ALT (a marker of liver cell damage) at baseline as compared to those with normal baseline plasma ALT.
In theory, shared underlying risk factors might have confounded the strong observational association between fatty liver disease and liver-related mortality which we and others observed. To avoid the risk of confounding, we turned to Mendelian randomization, a method that uses naturally occurring genetic variation as unconfounded proxies for an exposure of interest (here: fatty liver disease).
We found a causal association between genetic risk of fatty liver disease and risk of liver-related death when we examined seven genetic variants robustly associated with fatty liver disease. This supports the hypothesis that fatty liver disease causes liver-related mortality with cirrhosis and liver cancer likely responsible for most of the risk. liver-related mortality. However, studies focused on the PNPLA3 variant have found it to be associated with markedly higher liverrelated death, 6,30 in agreement with the data reported here. It remains unclear if hepatic steatosis per se is a causal driver of liverrelated mortality, independent of the later stages of fatty liver disease (steatohepatitis, fibrosis, cirrhosis, and HCC). This question is hard to answer using Mendelian randomization, because most genetic variants that associate with steatosis also associate with steatohepatitis and biomarkers of liver injury in a dose-dependent manner. 27,31,32 The variants with the largest effects (e.g., in PNPLA3 and TM6SF2) also associate robustly with fibrosis, cirrhosis, and HCC. These correlations with outcomes throughout the spectrum of fatty liver disease make it difficult to isolate the effects of steatosis from those of steatohepatitis and/or chronic liver disease in a Mendelian randomization setting.
Fatty liver disease is associated with an approximately two-fold higher rate of all-cause mortality in several previous observational studies. [4][5][6] In agreement, we observed that baseline hepatic steatosis and elevated plasma ALT were associated with a 1.84-fold and Observationally, elevated ALT at baseline was associated with a 1.25-fold higher extrahepatic cancer-related mortality in our study.
Previous studies found similar, or slightly higher risk estimates for the observational associations between fatty liver disease and extrahepatic-cancer related mortality. 5 Little genetic evidence exists on an association between fatty liver disease and extrahepatic cancer. The genetic variants in PNPLA3 and TM6SF2 were not associated with cancer-related mortality in NHANES 36 or in the UK Biobank. 33 In accordance, the genetic variants were not associated with death due to extrahepatic cancer in our study, suggesting that fatty liver disease is not a cause of extrahepatic cancer-related mortality, but that fatty liver disease and cancer-related mortality are associated in observational studies because they share unhealthy lifestyle factors.
The following limitations to our study should be addressed.
Hepatic CT-attenuation and plasma ALT are imperfect proxies for hepatic steatosis and steatohepatitis, respectively. Liver biopsy is the diagnostic gold standard, but it is unfeasible and unethical to biopsy almost 111 000 individuals from the general populationmost of whom are unsuspected of liver disease. A key point of our study is that the genetic variants we use here are known to have robust effects on biopsy-proven fatty liver disease, ensuring that they are valid proxies for fatty liver disease. We only included variants in seven loci with known robust effects on fatty liver disease.
Recent studies have identified several additional variants associated with this phenotype. 31,37 Future Mendelian randomization studies of fatty liver disease will benefit from including more of these newly identified variants as genetic instruments. Another potential limitation is that, despite the large overall sample size, some of the stratified analyses may have lacked power due to low number of events and/or moderate effect sizes. For example, we were not able to detect interactions between the genetic risk variants and obesity or alcohol affecting mortality. Previous studies have reported such interactions affecting the full spectrum of fatty liver disease. 11,19,38 We hypothesize that similar gene-environment interactions affect liver-related mortality, but that larger sample sizes will be required to detect them. Lastly, we only studied white individuals of Danish and British descent, potentially limiting the generalizability to other ethnicities.
In conclusion, while fatty liver disease is observationally associated with all-cause, liver-related, IHD-related, and extrahepatic cancer-related mortality, this study only supports a causal link between fatty liver disease and liver-related mortality.

CO N FLI C T O F I NTER E S T S TATEM ENT
None of the authors had any conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the authors, upon reasonable request.

E TH I C S S TATEM ENT
The studies were approved by institutional review boards and Danish and English ethical committees and were conducted according to the Declaration of Helsinki. Written informed consent was obtained from participants.