The effect of liver enzymes on adiposity: a Mendelian randomization study

Poorer liver function is positively associated with diabetes in Mendelian randomization (MR) studies. Observationally, adiposity is associated with poorer liver function. To clarify the etiology, we assessed the association of liver enzymes with adiposity observationally and using two-sample MR for validation. In the “Children of 1997” birth cohort, we used multivariable linear regression to assess the associations of alanine transaminase (ALT) and alkaline phosphatase (ALP) at ~17.5 years with body mass index (BMI) (n = 3,458). Using MR, genetic predictors of ALT, ALP and gamma glutamyltransferase (GGT), were applied to genome-wide association studies of BMI (n = 681,275), waist circumference (WC) (n = 224,459) and waist-hip ratio (WHR) (n = 224,459) to obtain unconfounded estimates. Observationally, ALT was positively associated with BMI (0.10 kg/m2 per IU/L, 95% confidence interval (CI) 0.09 to 0.11). ALP was inversely associated with BMI (−0.018 kg/m2 per IU/L, 95% CI −0.024 to −0.012). Using MR, ALT was inversely associated with BMI (−0.14 standard deviation per 100% change in concentration, 95% CI −0.20 to −0.07), but not WC or WHR. ALP and GGT were unrelated to adiposity. Poorer liver function might not cause adiposity; instead higher ALT might reduce BMI, raising the question as to the role of ALT in body composition.

Of the 4 SNPs predicting ALT, rs2954021 (TRIB1) predicted both ALT and ALP, and rs738409 (PNPLA3) is highly associated with NAFLD. Of the 14 SNPs predicting ALP, rs281377 (FUT2) is highly associated with resting metabolic rate; rs579459 is located in the ABO gene. Of the 26 SNPs predicting GGT, rs516246 (FUT2) is associated with obesity-related traits; rs1260326 (GCKR) is associated with Crohn's disease which might be associated with adiposity (Supplemental Table S2). The F statistics and variance explained (r 2 ) were 15 and 0.001 for ALT,  Genetically instrumented ALT with BMI, WC, and WHR. Genetically instrumented ALT was negatively associated with BMI using inverse variance weighting (IVW), which was also evident using a weighted median (WM) and after excluding the potentially pleiotropic SNPs. The negative association was more obvious for women. ALT was not associated with WC or WHR using any method with and without potentially pleiotropic SNPs. Few MR-Egger intercepts differed from the null, giving little indication of pleiotropy. Heterogeneity was almost absent after the pleiotropic SNP were excluded (Tables 3-5).
Genetically instrumented ALP and GGT with BMI, WC, and WHR. Genetically instrumented ALP and GGT were not clearly associated with BMI, WC, or WHR using any method with and without potentially pleiotropic SNPs. Overall, there was no evidence of pleiotropy based on the null values of the MR-Egger intercepts. Heterogeneity was most evident for ALP (Tables 3-5).

Discussion
This novel study used two different approaches, an observational study and an MR study, with different data sources, assumptions, and different unrelated sources of bias to assess the role of liver enzymes in adiposity. We found the clearest evidence for ALT being inversely associated with BMI, perhaps particularly among women. We used an observational design to assess the association of liver function, indicated by liver enzymes, with adiposity indices in young people aged 17.5 years and an MR design in adults. However, limitations exist in both study designs. First, liver enzymes represent different aspects of liver function: ALT is a marker of hepatocyte integrity, which is relatively more specific for liver pathology than other indices. ALP and GGT are markers of cholestasis. ALP is not liver specific and also originates from other tissues especially from bone. As such, ALT, ALP, and GGT may not completely or only represent liver function 18 . Second, the two study designs may not be completely comparable. Specifically, the observational study pertains to Chinese, while the MR study mainly pertains to people of European ancestry because a suitably large GWAS of Chinese people is not publicly available. However, causes are usually consistent although not relevant in all contexts 19 . Finally, the two study designs have contrasting limitations.
The conventional observational study is open to residual confounding by factors such as diet, lifestyle, and physical activity, which are hard to measure precisely and eliminate, although smoking is rare, alcohol use is low, and adiposity is not strongly associated with SEP in Hong Kong 20,21 , which may reduce confounding. However, it is difficult to disentangle correlated factors reliably in such observational studies. ALT was also lower than 10 IU/L (n = 254) for 7.3% of the participants in "Children of 1997" and was fixed at 5 IU/L, which was unlikely to affect the estimates, because it was only below the limit of detection for a relatively small proportion of observations. Follow-up in "Children of 1997" was incomplete, however, no major differences were found between the participants with and without adiposity indices (Supplemental Table S1). As such, selection bias from loss-to-follow-up is unlikely.
MR studies are more robust to confounding than conventional observational studies but have strong assumptions. Specifically MR studies rely on the assumptions that the genetic instruments predict the exposure reliably, are independent of confounders of the exposure-outcome association, and are only associated with the outcome via the exposure. The F statistics were all larger than 10, which reduces the risk of weak instrument bias. Pleiotropic effects are possible, but estimates were similar after excluding potentially pleiotropic SNPs, such as rs738409 (PNPLA3) predicting ALT and MR-Egger did not provide statistical evidence of pleiotropy. Although some of the I 2 were large, after excluding potential pleiotropic SNPs, in most cases, the I 2 became smaller. Estimates for ALP showed some heterogeneity although the MR-Egger regression did not show directional pleiotropy. The GWAS for liver enzymes overlapped slightly (~17%) with the GWAS of adiposity indices from the GIANT consortium but is unlikely to cause bias. We assessed sex differences on the assumption that genetic predictors of liver function are similar for women and men, which we could not test empirically. Finally, MR provides an estimate of the effect of life time exposure rather than indicating the exact size of the corresponding intervention, as such it indicates an etiological pathway.    such studies have been conducted. However, some estimates differed between the observational and MR designs, probably because of the difficulty of distinguishing correlated measures of liver function, the possibility of confounding, and/or observational studies not reflecting life-long effects. To our knowledge, only one small MR study has assessed the association of liver function with adiposity, and found no association of ALT with BMI 7 . One possible explanation for ALT potentially reducing BMI, but not WHR or WC, is that ALT, acting via sex hormones, reduces muscle mass rather than or as well as fat mass. ALT reducing muscle mass would be consistent with ALT increasing the risk of diabetes 3 , because low muscle mass is a potential cause of diabetes 29 . www.nature.com/scientificreports www.nature.com/scientificreports/ Overall, this study suggests that ALT reduces BMI. To further clarify the role of liver function in metabolic conditions whether ALT reduces specifically muscle mass, and thereby causes diabetes should be investigated, because it would mean that muscle mass could be an attractive target of intervention to prevent diabetes.

Methods
The "Children of 1997" birth cohort. The "Children of 1997" birth cohort is a population-representative Chinese birth cohort (n = 8,327) which recruited 88% of all births in Hong Kong in April and May 1997 30    the Maternal and Child Health Centers in Hong Kong. Parents of all newborns were encouraged to attend to obtain free preventive care and vaccinations for their child. Parental and infant characteristics were obtained from a self-administered questionnaire in Chinese at recruitment and subsequent routine visits. In 2007, contact was re-established followed by three postal/telephone questionnaire surveys. From 2013 to 2016 a Biobank Clinical follow-up at 16-18 years was conducted, when liver enzymes were assessed. As a compromise between cost and comprehensiveness, liver enzymes were assessed from plasma ALT and plasma ALP analyzed using the Roche Cobas C8000 System, a discrete photometric chemistry analyzer, with International Federation of Clinical Chemistry standardized method with pyridoxal phosphate and substrates of L-alanine and 2-oxoglutarate for ALT, and an optimized substrate concentration and 2-amino-2-methyl-1-propanol as buffer plus the cations magnesium and zinc for ALP. These analyses were conducted at an accredited laboratory serving a teaching hospital in Hong Kong. Height, weight, and waist and hip circumference were measured using standard protocols.
Children of 1997. Exposure -liver enzymes. Liver function at ~17.5 years was assessed from plasma ALT (IU/L) and plasma ALP (IU/L).
Outcome -Adiposity. Adiposity was assessed from BMI (kg/m 2 ), WC (cm), and WHR, which represent different aspects of adiposity. Although these are not completely normally distributed, we present them in natural units for ease of interpretation, given interpretation was similar using a gamma distribution.
Mendelian randomization. Genetic associations with liver enzymes. SNPs associated with plasma log transformed ALT, ALP, and GGT at genome-wide significance (p-value < 5 × 10 −8 ) adjusted for age and sex were obtained from the largest available GWAS of plasma levels of liver enzymes comprising 61,089 adults (~86% European, mean age 52.8 years, 50.6% women) 14 . For SNPs in linkage disequilibrium (R 2 > 0.01), we retained SNPs with the lowest p-value using the Clumping function of MR-Base (TwoSampleMR) R package, based on the 1000 Genomes catalog 31 . Whether any of the selected SNPs was related to adiposity directly rather than through liver enzymes (pleiotropic effects) was assessed from their known phenotypes obtained from comprehensive curated genotype to phenotype cross-references, i.e., Ensembl (http://www.ensembl.org/index.html) and the GWAS Catalog (https://www.ebi.ac.uk/gwas/). We also identified SNPs from highly pleiotropic genes, such as ABO and GCKR, whose full functionality is not yet clearly understood.

Statistical analyses.
In the "Children of 1997" birth cohort, baseline characteristics of cohort participants who were included and excluded were compared using Cohen effect sizes 32 . Cohen effect sizes indicate the magnitude of the difference independent of sample size. They are usually categorized as 0.10 for small, 0.30 for medium, and 0.50 for large when considering categorical variables 32 . The associations of adiposity indices with potential confounders were assessed using independent t-test or analysis of variance (ANOVA).
We assessed the associations of liver enzymes with adiposity indices adjusted for potential confounders, i.e., household income, highest parental education, type of housing, highest parental occupation, second-hand and maternal smoking, and sex, using multivariable linear regression. We also assessed whether associations differed by sex from the relevant interaction terms.
In the Mendelian randomization study, the strength of the genetic instruments was indicated by the F-statistic 33 . A higher F-statistic indicates lower risk of weak instrument bias 33 . We aligned SNPs for exposure and outcome on allele and effect allele frequency to ensure all SNPs, in particular palindromic SNPs, were aligned correctly. SNPs that could not be unequivocally aligned were replaced by proxies or dropped. SNPs predicting liver enzymes that were not available for adiposity indices were replaced by highly correlated proxies (R 2 > 0.9). Potential proxy SNPs were obtained from the GWAS 14 and their correlations with other SNPs were obtained using LDlink 34,35 .
Unconfounded estimates of the effects of liver enzymes on adiposity indices overall and by sex were obtained by meta-analyzing SNP-specific Wald estimates (SNP-outcome association divided by SNP-exposure association) using IVW with random effects for 4+ SNPs, which assumes that balanced pleiotropy, and with fixed effects for 3 SNPs or fewer. We repeated the analysis excluding pleiotropic SNPs that might be associated with the relevant outcome directly rather than via liver enzymes. As sensitivity analyses, WM and MR-Egger regression were used. The WM may generate correct estimates when >50% of weight is contributed by valid SNPs 36 . MR-Egger generates correct estimates even when all the SNPs are invalid instruments as long as the instrument strength independent of direct effect assumption is satisfied 37 . A non-null intercept from MR-Egger indicates potential directional pleiotropy and invalid IVW estimates 36 . Heterogeneity was assessed using the I 2 statistic 37 . Power calculations were performed using the approximation that the sample size for an Mendelian randomization equates to that of the same regression analysis with the sample size divided by the r 2 for genetic variant on exposure 38 .

Ethics approval and informed consent.
Ethical approval for the study, including comprehensive health related analyses, were obtained from Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster (HKU/HA HKW IRB). Informed written consent was obtained from the parents/guardians or participant if 18 years or older before participation in the Biobank Clinical Follow-up.
The MR study only uses published or publicly-available data. No original data were collected for the MR study. Ethical approval for each of the studies included in the investigation can be found in the original publications (including informed consent from each participant).

Data availability
Data are available upon request from the "Children of 1997" data access committee: aprmay97@hku.hk. The volume and complexity of the data collected preclude public data deposition, because the participants could be identifiable from such extensive data which would comprise participant privacy. Data of the MR study are publicly available summary data.