Insulin Resistance Exacerbates Genetic Predisposition to Nonalcoholic Fatty Liver Disease in Individuals Without Diabetes This is the author manuscript accepted for publication and has undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi: 10.1002/HEP4.1353 This article is protected by copyright. All rights reserved 1 2 DR. LISA VANWAGNER (Orcid ID : 0000-0002-6264-2573) 3 4 5 Article type : Original 6 7 8 [HEP4-18-0312] 9 Insulin resistance exacerbates genetic predisposition to NAFLD in individuals 10 without diabetes 11 Llilda Barata1 (barata@wustl.edu), Mary F. Feitosa1 (mfeitosa@wustl.edu), Lawrence F. Bielak2 12 (lfbielak@umich.edu), Brian Halligan3 (halligan@med.umich.edu), Abigail S. Baldridge4 13 (abigail.baldridge@northwestern.edu), Xiuqing Guo5 (xguo@labiomed.org), Laura M. Yerges- 14 Armstrong6 (laura.m.yerges-armstrong@gsk.com), Albert V. Smith7 (albertvs@umich.edu), Jie 15 Yao5 (jyao@labiomed.org), Nicholette D. Palmer8 (nallred@wakehealth.edu), Lisa B. 16 VanWagner4,9 (lvw@northwestern.edu), J. Jeffrey Carr10 (j.jeffrey.carr@vumc.org), Yii-Der I. 17 Chen5 (ichen@labiomed.org), Matthew Allison11 (mallison@ucsd.edu), Matthew J. Budoff12 18 (mbudoff@labiomed.org), Samuel K. Handelman3 (samuelkh@med.umich.edu), Sharon L.R. 19 Kardia2 (skardia@umich.edu), Thomas H. Mosley Jr.13 (tmosley@umc.edu), Kathleen Ryan6 20 (KRyan@som.umaryland.edu), Tamara B. Harris14 (harris99@nia.nih.gov), Lenore J. Launer14 21 (LaunerL@nia.nih.gov), Vilmundur Gudnason15,16 (v.gudnason@hjarta.is), Jerome I. Rotter5 22 (jrotter@labiomed.org), Myriam Fornage17(Myriam.Fornage@uth.tmc.edu), Laura J. 23 Rasmussen-Torvik4 (ljrtorvik@northwestern.edu), Ingrid Borecki1 (iborecki28@gmail.com), 24 Jeffrey R. O’Connell6 (joconnel@som.umaryland.edu), Patricia A. Peyser2 25 (ppeyser@umich.edu), Elizabeth K. Speliotes*3 (espeliot@med.umich.edu), Michael A. 26 Province*1 (mprovince@wustl.edu) 27 A u th o r M a n u s c ri p t https://doi.org/10.1002/HEP4.1353 https://doi.org/10.1002/HEP4.1353 https://doi.org/10.1002/HEP4.1353 mailto:halligan@med.umich.edu mailto:abigail.baldridge@northwestern.edu mailto:samuelkh@med.umich.edu 2 This article is protected by copyright. All rights reserved 28 1 Division of Statistical Genomics, Department of Genetics, Washington University School of 29 Medicine; St. Louis, MO, USA 30 2 Department of Epidemiology, School of Public Health, University of Michigan, Ann Arbor, MI, 31 USA 32 3 Division of Gastroenterology, Department of Internal Medicine, Department of Computational 33 Medicine and Bioinformatics, University of Michigan, Ann Arbor, Michigan USA 34 4 Department of Preventive Medicine, Northwestern University Feinberg School of Medicine, 35 Chicago, IL USA 36 5 The Institute for Translational Genomics and Population Sciences, LABioMed and the 37 Department of Pediatrics, Harbor-UCLA Medical Center, Torrance, CA, USA 38 6 Department of Medicine, University of Maryland School of Medicine, Baltimore, Maryland USA 39 7 Department of Biostatistics, School of Public Health, University of Michigan, Ann Arbor, 40 MI,USA 41 8 Department of Biochemistry, Wake Forest School of Medicine, Winston-Salem, NC, USA 42 9 Division of Gastroenterology and Hepatology, Northwestern University Feinberg School of 43 Medicine, Chicago, IL USA 44 10 Department of Radiology, Vanderbilt University School of Medicine, Nashville, TN, USA 45 11 Department of Family Medicine and Public Health, University of California, San Diego, 46 CA,USA 47 12 Division of Cardiology, Los Angeles Biomedical Research Institute, Torrance, California. 48 13 Department of Medicine, Division of Geriatrics, University of Mississippi Medical Center, 49 Jackson, MS, USA 50 14 Laboratory of Epidemiology and Population Sciences, National Institute of Aging, Bethesda, 51 MD, USA 52 15 Icelandic Heart Association, Kopavogur, Iceland 53 16 Faculty of Medicine, University of Iceland, Reykjavik, Iceland 54 17 University of Texas Health Science Center, Houston, Texas, USA 55 56 * Michael Province and Elizabeth K. Speliotes equally supervised this work. 57 58 59 Keywords: gene-environment interactions, PNPLA3, hepatic steatosis, metabolic factors, 60 GCKR 61 A u th o r M a n u s c ri p t 3 This article is protected by copyright. All rights reserved 62 63 Corresponding Authors: 64 65 Llilda Barata PhD, MPH 66 67 Department of Genetics 68 Division of Statistical Genomics 69 Washington University School of Medicine 70 660 South Euclid Ave 71 St. Louis, MO, 63108 USA 72 barata@wustl.edu 73 (314) 747-4133 74 75 Elizabeth K. Speliotes, MD PhD MPH 76 77 Division of Gastroenterology 78 Department of Internal Medicine 79 University of Michigan 80 1150 West Medical Center Drive 81 Ann Arbor, MI, 48109 USA 82 espeliot@med.umich.edu 83 (734) 936-4785 84 85 Michael A. Province, PhD 86 87 Department of Genetics 88 Division of Statistical Genomics 89 Washington University School of Medicine 90 660 South Euclid Ave 91 St. Louis, MO, 63108 USA 92 mprovince@wustl.edu 93 (314) 362-3616 94 Abbreviations: NAFLD= Nonalcoholic fatty liver disease; TG=triglycerides; LDL=low-density 95 A u th o r M a n u s c ri p t mailto:barata@wustl.edu mailto:espeliot@med.umich.edu mailto:mprovince@wustl.edu 4 This article is protected by copyright. All rights reserved 96 lipoprotein cholesterol; HDL=high-density lipoprotein cholesterol; BMI=body mass index; 97 98 WHRadjBMI= waist-to-hip ratio adjusted for body mass index; PNPLA3=Patatin-like 99 100 phospholipase domain-containing protein 3 gene; GCKR =Glucokinase regulatory protein gene; 101 102 NCAN =Neurocan gene; TM6SF2 =Transmembrane 6 Superfamily Member 2 gene; 103 104 LYPLAL1=Lysophospholipase-like 1 gene; EA=European ancestry; SNP=single nucleotide 105 106 polymorphism; AA=African ancestry; LA=liver attenuation; HOMA-IR=homeostatic model of 107 108 insulin resistance; AGES=Age, Gene/Environment Susceptibility-Reykjavik; Amish=Old Order 109 110 Amish; CARDIA=Coronary Artery Risk Development in Young Adults; FamHS=Family Heart 111 112 Study; FHS=Framingham Heart Study; GENOA=Genetic Epidemiology Network of Arteriopathy; 113 114 MESA=Multi-Ethnic Study of Atherosclerosis; HU=Hounsfield units; IVN=inverse normal 115 116 transformation; LAivn=Inverse normal-transformed residuals of LA; SD=standard deviation 117 118 119 Financial Support: 120 121 This work was performed under the auspices of the Genetics of Obesity-Related Liver Disease 122 Consortium. Funding for this study was made possible by The Age, Gene/Environment 123 Susceptibility-Reykjavik study (AGES) funded by the National Institute of Health (NIH) contracts 124 N01-AG-1-2100 and 271201200022C, the National Institute of Aging Intramural Research 125 Program, Hjartavernd (the Icelandic Heart Association), and the Althingi (the Icelandic 126 Parliament). The study is approved by the Icelandic National Bioethics Committee, VSN:00- 127 063. The researchers are indebted to the participants for their willingness to participate in the 128 study. The Amish study gratefully acknowledges our Amish liaisons, research volunteers, field 129 workers and Amish Research Clinic staff and the extraordinary cooperation and support of the A u th o r M a n u s c ri p t 5 This article is protected by copyright. All rights reserved 130 Amish community without which these studies would not have been possible. The Amish 131 studies are supported by grants and contracts from NIH, including U01 HL072515, U01 132 HL84756, U01 HL137181 and P30 DK72488. Funding for the Coronary Artery Risk 133 Development in Young Adults Study (CARDIA) is conducted and supported by the National 134 Heart, Lung, and Blood Institute (NHLBI) in collaboration with the University of Alabama at 135 Birmingham (HHSN268201300025C & HHSN268201300026C), Northwestern University 136 (HHSN268201300027C), University of Minnesota (HHSN268201300028C), Kaiser Foundation 137 Research Institute (HHSN268201300029C), Johns Hopkins University School of Medicine 138 (HHSN268200900041C) and Vanderbilt University Medical Center (R01 HL 098445). CARDIA 139 is also partially supported by the Intramural Research Program of NIA and an intra-agency 140 agreement between NIA and NHLBI (AG0005). The National Human Genome Research 141 Institute (NHGRI) supported genotyping of CARDIA participants through grants U01-HG- 142 004729, U01-HG-004446, and U01-HG-004424. This manuscript has been reviewed by 143 CARDIA for scientific content. Exome chip genotyping and data analyses were funded in part 144 by grants U01-HG004729, R01-HL093029 and R01-HL084099 from NIH to Dr. Myriam 145 Fornage. Lisa B. VanWagner is supported by NIH grant K23 HL136891. The Family Heart 146 Study (FamHS) was supported by grant R01-DK-089256 from the National Institute of Diabetes 147 and Digestive and Kidney Diseases (NIDDK) and grant R01HL117078 from NHLBI. Elizabeth 148 K. Speliotes and Brian Halligan are supported by NIH grants R01 DK106621, R01 DK107904 149 and The University of Michigan Department of Internal Medicine. Data from the Framingham 150 Heart Study (FHS) came from the database of Genotypes and Phenotypes (dbGaP). Lawrence 151 F. Bielak and Patricia A. Peyser are supported, in part, by NIH grants R01 DK106621 and R01 152 DK107904. Support for the Genetic Epidemiology Network of Arteriopathy (GENOA) study was 153 provided by NHLBI (HL054457, HL054464, HL054481, HL087660, and HL085571) of the 154 National Institutes of Health. The Multi-Ethnic Study of Atherosclerosis (MESA) and the MESA 155 SHARe project are conducted and supported by NHLBI in collaboration with MESA 156 investigators. This research was supported by R01 HL071739, and MESA was supported by 157 contracts N01-HC-95159, N01-HC-95160, N01-HC-95161, N01-HC-95162, N01-HC-95163, 158 N01-HC-95164, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168, N01-HC- 159 95169 from the National Heart, Lung, and Blood Institute, and by grants UL1-TR-000040, UL1- 160 TR-001079, and UL1-RR-025005 from National Center for Research Resources. The provision 161 of exome chip genotyping data was supported in part by the NHLBI contract N02-HL-64278, 162 National Center for Advancing Translational Sciences, CTSI grant UL1TR001881, and the 163 National Institute of Diabetes and Digestive and Kidney Disease Diabetes Research Center A u th o r M a n u s c ri p t 6 This article is protected by copyright. All rights reserved 164 (DRC) grant DK063491 to the Southern California Diabetes Endocrinology Research Center. 165 Funding support for MESA’s NAFLD dataset was provided by grant HL071739-05A2. The 166 views expressed in this manuscript are those of the authors and do not necessarily represent 167 the views of NHLBI, NIA, NIDDK or NIH. 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 ABSTRACT 186 The accumulation of excess fat in the liver (hepatic steatosis), in the absence of heavy alcohol 187 consumption, causes nonalcoholic fatty liver disease (NAFLD), which has become a global 188 epidemic. Identifying metabolic risk factors that interact with the genetic risk of NAFLD is 189 important for reducing disease burden. We tested whether serum glucose, insulin, insulin 190 resistance, triglycerides, low density lipoprotein cholesterol, high density lipoprotein cholesterol, 191 body mass index (BMI), and waist-to-hip ratio adjusted for BMI interact with genetic variants in 192 or near the patatin-like phospholipase domain containing 3 gene (PNPLA3), the glucokinase 193 regulatory protein gene (GCKR), the neurocan gene (NCAN/TM6SF2), and the 194 lysophospholipase-like 1 gene (LYPLAL1) to exacerbate hepatic steatosis, estimated by liver 195 attenuation (LA). We performed association analyses in ten population-based cohorts 196 separately and then meta-analyzed results in up to 14,751 individuals (11,870 of European A u th o r M a n u s c ri p t 7 This article is protected by copyright. All rights reserved 197 ancestry and 2,881 of African ancestry). We found that PNPLA3-rs738409 significantly 198 interacted with insulin, insulin resistance, BMI, glucose, and TG to increase hepatic steatosis in 199 nondiabetic individuals carrying the G-allele. Additionally, GCKR-rs780094 significantly 200 interacted with insulin, insulin resistance and TG. Conditional analyses, using the two largest 201 European ancestry cohorts in the study, showed that insulin levels accounted for most of the 202 interaction of PNPLA3-rs738409 with BMI, glucose, and TG in nondiabetic individuals. Insulin, 203 PNPLA-rs738409, and their interaction accounted for at least 8% of the variance in hepatic 204 steatosis in these two cohorts. 205 Conclusion: Our results suggest that insulin resistance, either directly or via the resultant 206 elevated insulin levels, more than other metabolic traits, amplifies the PNPLA3 rs738409-G 207 genetic risk for hepatic steatosis. These results suggest that improving insulin resistance in 208 nondiabetic individuals carrying PNPLA3-rs738409-G may preferentially decrease hepatic 209 steatosis. 210 211 Nonalcoholic fatty liver disease (NAFLD) is a result of the excess accumulation of lipids in 212 hepatocytes (hepatic steatosis) in the absence of heavy alcohol consumption(1). Hepatic 213 steatosis is also associated with the risk of developing dyslipidemia or dysglycemia(2), as well 214 as cardiovascular disease, which is the number one cause of death in individuals with NAFLD(3, 215 4). Hepatic steatosis may progress to advanced liver disease in the form of nonalcoholic 216 steatohepatitis, fibrosis (cirrhosis), and cancer (hepatocellular carcinoma)(5-7). In the U.S., the 217 prevalence of hepatic steatosis in the adult population is between 10% to 30%; worldwide it is 218 25% to 45%(8). While the pathogenesis of NAFLD is not entirely understood, both genetic 219 factors and metabolic traits increase the risk of hepatic steatosis. 220 221 Heritability of hepatic steatosis ranges from 22 to 38% across all ancestries suggesting that 222 specific genotypes may predispose individuals to NAFLD(1). Previously, the Genetics of 223 Obesity-Related Liver Disease Consortium conducted a genome-wide association study in 224 7,176 individuals of European ancestry (EA) with replication in histology-based samples (9). 225 This study identified that rs738409 (PNPLA3), a missense single nucleotide polymorphism 226 (SNP) first associated with hepatic fat content a decade ago (10), the missense variant 227 rs2228603 (NCAN/TM6SF2) and intronic variants rs12137855 (LYPLAL1) and rs780094 228 (GCKR) were significantly associated with hepatic steatosis(9). We and others have replicated 229 the association of these common variants with hepatic steatosis in other populations and 230 ethnicities (11-13), and the associations are consistent between those of EA and African A u th o r M a n u s c ri p t 8 This article is protected by copyright. All rights reserved 231 ancestry (AA) (direction of effect is similar)(11). Further, the G allele for rs738409 was 232 associated with susceptibility to nonalcoholic steatohepatitis (OR 2.64, 95% CI: 1.85-3.75, p≤ 233 1.0E-04), nonalcoholic steatohepatitis severity (OR 1.85, 95% CI: 1.05-3.26, p≤ 3.5E-02) and 234 fibrosis (OR 1.95, 95% CI: 1.17-3.26, p≤ 1.3E-02) in EA individuals(14). 235 236 Traits that predispose to metabolic syndrome, i.e. higher body mass index (BMI) (15), 237 dyslipidemia, hyperglycemia, and insulin resistance are associated with hepatic steatosis (2, 3, 238 16). Eighty to ninety percent of obese (BMI ≥ 30 kg/m2) adults have hepatic steatosis(17), while 239 20-80% of individuals with hepatic steatosis also have higher levels of triglyceride (TG) and low- 240 density lipoprotein cholesterol (LDL), but lower levels of high-density lipoprotein cholesterol 241 (HDL)(18). Diabetes is also commonly associated with hepatic steatosis(19). How these 242 modifiable metabolic traits interact with genetic variation to influence risk for hepatic steatosis is 243 not known. 244 245 In this cross-sectional study, we tested whether several metabolic traits interact with the four 246 genetic variants previously associated with hepatic steatosis(9) to affect liver attenuation (LA), a 247 computed tomographic quantitative measure that is inversely related to histologically measured 248 liver fat (20). The metabolic traits tested were: insulin resistance (as homeostatic model of 249 insulin resistance (HOMA-IR)), fasting insulin, fasting glucose, BMI, centralized fat deposition 250 measured by waist-to-hip ratio adjusted for BMI (WHRadjBMI), fasting TG, fasting HDL and 251 fasting LDL. We first carried out interaction analyses between each of these traits and each of 252 the genetic variants in ten separate population-based cohorts from seven different studies. Then 253 we meta-analyzed results across cohorts in up to 14,751 individuals (EA, n=11,870 and 254 AA, n=2,881). We then carried out conditional analyses in the two largest EA cohorts in the 255 study to determine the driving metabolic factor. 256 257 POPULATION AND METHODS 258 Ethics Statement 259 The Institutional Review Boards or equivalent committees of all participating studies approved 260 this study. The principal investigator of each institution obtained written consent from 261 participants. 262 Study Description A u th o r M a n u s c ri p t 9 This article is protected by copyright. All rights reserved 263 The study was comprised of up to 14,751 individuals (EA, n=11,870 and AA, n=2,881); 56% of 264 participants were female. The sample derived from seven population-based studies participating 265 in the Genetics of Obesity-Related Liver Disease Consortium: Age, Gene/Environment 266 Susceptibility-Reykjavik (AGES), Old Order Amish (Amish), Coronary Artery Risk Development 267 in Young Adults (CARDIA), Family Heart Study (FamHS), Framingham Heart Study (FHS), 268 Genetic Epidemiology Network of Arteriopathy (GENOA), and Multi-Ethnic Study of 269 Atherosclerosis (MESA). In total, ten cohorts were included in the analysis, as three studies 270 contributed two ethnic groups (AA, EA). Each ethnic group was analyzed separately. CARDIA, 271 MESA, and AGES have unrelated individuals while FHS, Amish, GENOA, and FamHS are 272 family-based. Detailed information about the characteristics and design of each study is 273 provided in Supplementary Table 1. 274 275 Outcome variable and metabolic traits 276 The outcome variable was LA (liver attenuation), measured non-invasively with computed 277 tomography in Hounsfield units (HU) (21). LA is inversely proportional to liver fat, i.e. lower LA 278 values indicate a higher fat content in the liver (more hepatic steatosis)(2). The procedures 279 followed by each cohort to measure LA are described in Supplementary Table 2. Individuals 280 with active malignancies, focal lesions, or other incidental findings on computed tomography 281 were excluded from the studies. 282 283 Metabolic traits of interest were harmonized across cohorts following standard clinical 284 definitions. Overall adiposity was characterized by BMI (kg/m2), and abdominal adiposity by 285 waist-to-hip ratio adjusted for BMI (WHRadjBMI, cm). Since waist-to-hip ratio is correlated with 286 both BMI and visceral fat, we chose to use WHRadjBMI to have a measure that is independent of 287 overall fatness (i.e. BMI), but does reflect visceral adiposity, and is easily measured in the clinic. 288 Fasting insulin (mU/L) and fasting glucose (mmol/L) were measured from plasma or serum 289 using standard laboratory techniques detailed in Supplementary Table 2. When fasting 290 glucose was measured from whole blood, it was converted to plasma glucose using a correction 291 factor of 1.13 (22). HOMA-IR was assessed using fasting glucose (mmol/L) x fasting insulin 292 (mU/L) divided by 22.5 (23). Each cohort assayed fasting TG (mg/dL) and fasting HDL (mg/dL) 293 using methods described in Supplementary Table 2. If fasting LDL (mg/dL) was assayed, it 294 was used. Otherwise, LDL was calculated using the Friedewald formula, LDLF =(Total 295 cholesterol(mg/dL) - HDL(mg/dL) -TG(mg/dL)/5.0), only if TG < 400 mg/dL (24). A u th o r M a n u s c ri p t 10 This article is protected by copyright. All rights reserved 296 297 Alcohol consumption, history of diabetes, and use of lipid lowering medications were acquired 298 by questionnaire. Total alcohol consumption, defined in drinks per week, was calculated from 299 daily intake of beer, wine, and spirits. One drink was defined as a serving of 14 grams of 300 ethanol, the same as a 12 oz. bottle or can of beer, 5 oz. glass of wine, or 1.5 oz. shot of 80- 301 proof spirits such as gin, vodka, or whiskey(25). Heavy drinking was defined as ≥ 8 drinks per 302 week for women and ≥ 15 drinks per week for men (26). Diabetes (Type 1, Type 2) was defined 303 as having fasting plasma glucose levels ≥ 7 mmol/L (126 mg/dL), or self-reporting the use of 304 insulin or oral antidiabetic medications, or having a physician diagnosis of diabetes. The use of 305 statins was assessed from medication questionnaires. 306 307 Genotyping and Imputation 308 309 Four common variants were included in the analyses: rs738409 - a missense variant in the 310 patatin-like phospholipase domain containing 3 gene (PNPLA3); rs780094, an intronic variant 311 within the glucokinase regulatory protein gene (GCKR) that is in high linkage disequilibrium 312 (r2=0.93) with rs1260326, a likely functional missense variant in this gene; rs2228603, a 313 missense variant in the neurocan gene (NCAN) that is in high linkage disequilibrium (r2=0.798) 314 with rs585422926, a likely functional missense variant in the transmembrane 6 Superfamily 315 Member 2 gene (TM6SF2); and rs12137855, an intronic variant in the lysophospholipase-like 1 316 gene (LYPLAL1). These variants were either directly genotyped (allele counts were coded 0, 1, 317 or 2), or dosages were imputed from HapMap II or 1000G. Genotype calling algorithms and 318 imputation methods are detailed in Supplementary Table 3. 319 320 STATISTICAL ANALYSIS 321 322 Cohort-specific analyses 323 Cohorts performed analyses separately in each ancestry group (EA, AA). LA and metabolic 324 traits, used as continuous variables in all analyses, were adjusted for sex, age, principal 325 component estimates of ancestry, and study-specific covariates using linear regression as 326 detailed in Supplementary Table 2. LA was also adjusted for alcohol consumption, a 327 continuous variable (drinks/week), and for scan penetrance using phantom or spleen density. 328 Residuals from adjusted LA and metabolic traits were transformed using inverse normal A u th o r M a n u s c ri p t 11 This article is protected by copyright. All rights reserved 329 transformation (IVN) to reduce the influence of outliers and to standardize the phenotypes 330 across cohorts. Inverse normal-transformed residuals of LA, (LAivn), and each metabolic trait 331 (MTivn) were used to fit the interaction models. 332 333 Each cohort tested for statistical interactions between each variant and each metabolic trait 334 using multivariable linear regression or mixed linear modeling. LAivn was the dependent 335 variable. The independent variables were each SNP and MTivn, plus the interaction: 336 LAivn = α + β1 (SNP) + β2 (MTivn) + β3 (SNP x MTivn) + є. An additive model of inheritance was 337 assumed. Studies with family data (FHS, GENOA, Amish, and FamHS) used linear mixed 338 models to account for family relatedness among participants and computed robust standard 339 errors. Participants with diabetes (Type1 and Type 2) were excluded from the insulin, glucose 340 and HOMA-IR models, and those taking statins were excluded from the LDL model. As a 341 secondary analysis, BMI was included as a covariate in the models to investigate whether the 342 effect of the interaction between each SNP and each metabolic trait on LAivn occurred 343 independent of overall adiposity. Associations were carried out using MMAP(27), R(28), and 344 SAS (29) software. 345 346 Meta-analyses 347 We conducted fixed-effects meta-analyses by ancestry and overall on the parameter estimates 348 (β-coefficients and standard errors) for the main effects and interaction effects. We utilized the 349 inverse variance weighting method implemented in METAL (30). Using Cochran’s Q test (31), 350 we tested for heterogeneity of effects across all analyses. Within ancestries, focusing on 351 interactions, we found evidence of heterogeneity only for the interaction between TG and GCKR 352 in the EA cohorts. We did not find any heterogeneity for the interaction in the meta-analyses 353 between the two ancestry groups (EA vs AA); thus, we report the combined ancestry meta- 354 analyses. To determine the level of statistical significance while accounting for multiple testing, 355 we applied a Bonferroni correction that consisted of grouping correlated traits into three 356 metabolic domains: insulin-glucose, adiposity, and lipids. The critical p-value α=0.05 was 357 divided by 12 (4 variants x 3 metabolic domains) to obtain a corrected p-value. Meta-analyses 358 results and heterogeneity tests were considered significant if the two-tailed p-value was 359 ≤ 4.17E-03. As a secondary analysis, to investigate whether the statistically significant 360 interactions were consistent between genders, we fit the interaction models in men and women 361 separately, and meta-analyzed results within gender. 362 A u th o r M a n u s c ri p t 12 This article is protected by copyright. All rights reserved 363 Conditional Analyses in FamHS and FHS 364 To determine whether the interaction of BMI, glucose or TG with PNPLA3-rs738409 was 365 independent of insulin, we analyzed each trait’s interaction effect before and after including 366 insulin in the model. The analyses were performed with EA individuals in FamHS and replicated 367 in FHS. We chose these two cohorts because they are the two largest cohorts in the study; 368 together they represent more than 1/3 of our total sample. Individuals with diabetes and/or 369 missing information for the metabolic traits of interest were excluded resulting in a sample of 370 2,280 individuals in FamHS and 2,581 in FHS. After adjusting LA for phantom in both cohorts, 371 and for field centers in FamHS, LA residuals were transformed using inverse normal 372 transformation to approximate normality. LA transformed residuals (LAivn) were used as the 373 dependent variable. Using linear mixed models, we first regressed LAivn on either BMI, glucose, 374 or TG, and their interaction with PNPLA3-rs738409 (Supplementary Text). We then added 375 insulin to the models and its interaction with PNPLA3-rs738409 and the metabolic trait (either 376 BMI, glucose, or TG). Insulin and TG were log-transformed due to the presence of influential 377 outliers. Models were adjusted for age, sex, and alcohol consumption (drinks/week), and for 378 genotype batch effects in FamHS. Results from conditional analyses in each cohort were then 379 meta-analyzed. 380 381 The conditional models included principal components to adjust for population stratification. 382 Because the principal components were not associated with LAivn in either cohort, and their 383 inclusion in the conditional models did not change the inferences, we present the models 384 without them. We also performed conditional analyses after excluding individuals from FamHS 385 (n=231), and FHS (n=371) who reported heavy alcohol use (≥ 8 drinks per week for women, 386 and ≥ 15 drinks per week for men (Supplementary Tables 10-12) (26). Since the inferences 387 were unchanged, to increase power, we included all individuals, and adjusted for alcohol as a 388 covariate. Additionally, we conducted the conditional analyses with log-transformed HOMA-IR 389 instead of log-transformed insulin (Supplementary Tables 13-15). Insulin and HOMA-IR 390 provided similar inferences. Because glucose explains significantly less of the variation in LAinv, 391 we focused on insulin over HOMA-IR since there was no added benefit of measuring glucose on 392 variance explained by HOMA-IR than with just measuring insulin. 393 394 Illustration in FamHS of the interaction between insulin and PNPLA3-rs738409 in 395 individuals without diabetes A u th o r M a n u s c ri p t 13 This article is protected by copyright. All rights reserved 396 To assess the interaction effect of insulin with PNPLA3-rs738409 on hepatic steatosis 397 prevalence in FamHS, we plotted the percentage of individuals with LA ≤ 60 HU per PNPLA3- 398 rs738409 genotype by the lowest and highest quartile of insulin. Individuals with diabetes 399 and/or missing information for insulin were excluded and ancestries were combined to obtain a 400 sample of n=2,725. LA and insulin were not adjusted or transformed. The LA cut point of ≤ 60 401 HU, which corresponds to a liver/spleen ratio of 1.1, has previously been shown to identify 402 individuals with moderate to severe macrovesicular steatosis (≥ 30% of the liver parenchyma 403 with fat) at histology with a high diagnostic accuracy (32). In the literature, ≥ 30% liver fat 404 suggests moderate to severe hepatic steatosis (33). 405 406 RESULTS 407 Demographics and clinical characteristics across the study cohorts are presented in Table 1. 408 The mean age ± standard deviation (SD) across cohorts ranged from 49.47±3.86 to 76.38±5.46 409 years old. All cohorts included more women than men. The mean±SD of LA across cohorts 410 ranged from 55.05±12.28 HU to 65.40±9.83 HU. Mean±SD of fasting insulin levels in non- 411 diabetics ranged from 8.30±5.73 to 13.02±10.22 mU/L and fasting blood glucose levels ranged 412 from 4.90±0.58 to 5.49±0.50 mmol/L. The lowest mean±SD for HOMA-IR in non-diabetics was 413 1.99±1.27 and the highest was 3.14±2.69. The mean±SD of BMI ranged from 27.00±4.49 to 414 32.71±7.37 kg/m2. Several cohorts reported mean fasting TG >100 mg/dL. Mean±SD for 415 fasting LDL cholesterol in non-statin users was borderline high in Amish (141.31±8.66 mg/dL) 416 and AGES (146.84±5.73 mg/dL). Across cohorts, the range of fasting HDL was within the 417 recommended limit of ≥ 40 mg/dL. Heavy drinking varied among studies with GENOA having 418 the lowest percentage (0%) and CARDIA the highest (37%). 419 420 PNPLA3-rs738409 and GCKR-rs780094 interact with several metabolic traits 421 We found significant interactions for PNPLA3-rs738409 and GCKR-rs780094 with several 422 metabolic traits in combined ancestries after adjusting for multiple comparisons (Table 2, 423 Supplementary Table 4). PNPLA3-rs738409 interacted with insulin (p= 4.79E-14), HOMA-IR 424 (p= 4.68E-15), glucose (p= 1.26E-03), BMI (p= 8.13E-08) and TG (p=2.95E-03). As each of 425 these metabolic traits increased, a decrease in LAivn (i.e. higher fat content in the liver) became 426 more pronounced in presence of the G allele at PNPLA3-rs738409 as compared to the 427 presence of the C allele. Additionally, GCKR-rs780094 interacted with insulin (p= 4.57E-04), 428 HOMA-IR (p= 1.32E-03), and TG (p= 4.17E-03). As levels of insulin, HOMA-IR, and TG 429 increased, a decrease in LAivn (i.e. higher fat content in the liver) became more pronounced in A u th o r M a n u s c ri p t 14 This article is protected by copyright. All rights reserved 430 the presence of the T allele at GCKR-rs780094, compared to the C allele. All interactions 431 remained significant after adjusting for BMI (Supplementary Table 5) suggesting that overall 432 adiposity did not alter these effects. We did not find evidence of significant interactions between 433 any of the four genetic variants and WHRadjBMI, LDL, or HDL. Although the interaction between 434 WHRadjBMI and PNPLA3 did not reach the Bonferroni significance level, it was borderline 435 significant. This suggests that a larger sample size may be needed to detect an interaction. 436 Alternatively, the lack of statistical significance could be because WHRadjBMI does not represent 437 overall fatness to the extent that BMI or other anthropometric measurements do. 438 439 We also carried out meta-analyses in men and women separately to investigate possible gender 440 differences focusing only on the statistically significant interactions with PNPLA3-rs738409 and 441 GCKR-rs780094 (Supplementary Table 6). Women made up 56% of our study sample. 442 The interaction effects of insulin and HOMA-IR with PNPLA3-rs738409 did not differ between 443 men and women, and both reached statistical significance (women= p=3.24E-11, men=7.24E- 444 05; and women: p=1.62E-11, men: p=2.88E-05, respectively). For glucose, the interaction 445 effect was slightly less in men than in women (beta smaller), and did not reach significance in 446 men. These results suggest that gender did not alter the interactions between PNPLA3- 447 rs738409 and insulin/HOMAIR and the interaction effect of glucose was still present only in 448 women in the present study. Further, the interaction effects of BMI with PNPLA3-rs738409 449 were similar between men and women, and reached significance in both (p=1.20E-03 and 450 p=3.39E-05, respectively). The interaction effect of TG with PNPLA3- rs738409 did not reach 451 statistical significance in either gender. Moreover, the interaction effects of both insulin and 452 HOMA-IR with GCKR-rs780094 reached significance only in women (p=1.02E-03 and 453 p= 6.46E-04, respectively). Similarly, the interaction of TG with GCKR-rs780094 was significant 454 only in women (p=8.71E-04). Stratifying by gender substantially reduced our sample size, and 455 as a result power. 456 457 Conditional analyses suggest that insulin may mediate the interaction effect of BMI, TG 458 and glucose on LAivn in individuals without diabetes 459 We observed that the interaction of insulin with PNPLA3-rs738409 had a greater effect on LAivn 460 (hepatic steatosis defined by liver attenuation) than that of BMI, TG, or glucose. To determine if 461 the interaction of BMI, TG, or glucose with PNPLA3-rs738409 was independent of insulin, we 462 carried out conditional analyses in FamHS and FHS, and meta-analyzed results. We found that 463 the interaction of BMI (p=7.57E-02), TG (p=3.49E-01), or glucose (p=9.09E-01) with PNPLA3- A u th o r M a n u s c ri p t 15 This article is protected by copyright. All rights reserved 464 rs738409 was no longer statistically significant after including insulin as a main effect and 465 interactor with PNPLA3-rs738409 and the respective metabolic trait in the models 466 (Supplementary Tables 7-9). In contrast, the interaction of insulin with PNPLA3-rs738409 467 remained significant after controlling for BMI, TG, or glucose (pinsulin-BMI= 4.04E-04; pinsulin-TG= 468 3.24E-06; pinsulin-glucose= 8.40E-08), although the effect sizes and p-values were attenuated. 469 These results suggest that insulin may account for most of the interaction effect of BMI, glucose, 470 and TG with PNPLA3-rs738409 on LAivn. Previously, we reported that PNPLA3-rs738409 471 explained 2.4% of the variance in hepatic steatosis, estimated by LA, in EA individuals (11). In 472 the present study, PNPLA3-rs738409, insulin and their interaction together explain as much as 473 8% of the variance in hepatic steatosis in the two largest EA cohorts excluding individuals with 474 diagnosed diabetes. This suggests that insulin levels/insulin resistance may be a key 475 contributor to NAFLD. Excluding heavy drinkers from the conditional analyses did not change 476 our inferences regarding PNPLA3-rs738409 (Supplementary Table 10-12). We were not 477 powered to carry out these analyses for GCKR-rs780094. 478 479 Interaction effect of insulin with PNPLA3 on hepatic steatosis prevalence in FamHS 480 We also assessed the interaction effect of insulin with PNPLA3-rs738409 on hepatic steatosis 481 prevalence in individuals without diabetes (Figure 1). In the lowest quartile of insulin levels 482 (≤ 5.20 mU/L), the percentage of individuals with ≥ 30% liver fat (i.e. moderate to severe hepatic 483 steatosis) was 23.42%, 35.81%, and 39.47% for CC, CG and GG individuals, respectively. In 484 the highest quartile of insulin levels (≥ 13.06 mU/L), the percentage of individuals with ≥ 30% 485 liver fat was 54.44%, 76.32% and 95.29% for CC, CG and GG individuals, respectively. The 486 data show that as insulin levels increase the percentage of individuals with moderate to severe 487 hepatic steatosis increases. However, among those with the GG genotype, this effect is 488 magnified. The difference in the percentage of individuals with moderate to severe hepatic 489 steatosis increases by 55 percentage points between the lowest and highest insulin quartiles 490 among those with GG genotype, and increases by 41 percentage points among heterozygotes, 491 while that difference increases only by 31 percentage points among those with the CC 492 genotype. These data suggest that insulin has a strong effect on exacerbating the accumulation 493 of liver fat in individuals without diabetes who have 1 or 2 G- alleles at PNPLA3-rs738409. 494 495 DISCUSSION 496 In a sample of 14,751 EA and AA individuals, we found interactions between PNPLA3-rs738409 497 and insulin, HOMA-IR, BMI, glucose, and TG on LAinv (hepatic steatosis) after adjusting for A u th o r M a n u s c ri p t 16 This article is protected by copyright. All rights reserved 498 differences in age, sex, and alcohol consumption. We also found interactions between GCKR- 499 rs780094 and insulin, HOMA-IR, and TG on LAinv. Conditional analyses in more than 5,000 EA 500 individuals suggest that insulin, more than glucose, BMI, or TG drive the interaction with 501 PNPLA3-rs738409 to affect LAinv in non-diabetics. We did not see significant interactions 502 between PNPLA3-rs738409 and BMI, TG or glucose once insulin was accounted for, whereas 503 the reverse was not true. That is, there was still evidence for an interaction between PNPLA3- 504 rs738409 and insulin even after accounting for the other metabolic traits. These results persist 505 after accounting for alcohol intake, gender and overall adiposity. We estimated in FamHS and 506 FHS that as much as 8% of the variance in hepatic steatosis is explained by PNLPA3-rs738409, 507 insulin and their interaction in non-diabetic EA individuals. In our previous study, PNPLA3- 508 rs738409 alone explained only 2.4% of hepatic steatosis variance in EA individuals (11). 509 510 Our findings suggest that non-diabetic individuals with PNPLA3-rs738409-G and high insulin 511 levels may have a particularly high risk for hepatic steatosis. The PNPLA3 gene encodes 512 adiponutrin, an enzyme found on the membrane of lipid droplets within hepatocytes (34). Its 513 function may be to break down TG stored in the droplets, helping regulate hepatic TG content 514 (34, 35). The missense polymorphism rs738409 (C > G) in PNPLA3 substitutes the amino acid 515 isoleucine for methionine at residue 148 (I148M), changing the configuration of adiponutrin’s 516 catalytic site, and rendering the enzyme inactive (10, 36). The accumulation of the inactive 517 enzyme on lipid droplets is associated with TG buildup in hepatocytes (36). Humans and mice 518 carrying one or two copies of the I148M mutation (rs738409 CG or GG genotype) accumulate 519 excess TG in lipid droplets, and show more pronounced hepatic steatosis and NAFLD than 520 those without the mutation(35, 36). 521 522 It is possible that having high insulin levels in addition to the PNPLA3-rs738409 G allele may 523 result in a strong synergistic effect that exacerbates the accumulation of fat in the liver of non- 524 diabetic individuals, predisposing them to NAFLD. Insulin resistance stimulates the hydrolysis 525 of TG in adipose tissue releasing fatty acids in the bloodstream, which are taken up by the liver 526 in an unregulated manner promoting the accumulation of TG in hepatocytes (37). Higher insulin 527 levels also activate fatty acid synthesis in the liver further driving the formation and storage of 528 TG (34). In addition, insulin resistance elevates plasma glucose, which is sequestered by the 529 liver, phosphorylated, and metabolized to make glycerol and acetyl-CoA, the building blocks for 530 the synthesis of TG (34,38). In this context, it is possible that increased lipid synthesis and fatty 531 acid delivery to the liver may combine with the inability of hepatocytes to dispose of TG from A u th o r M a n u s c ri p t 17 This article is protected by copyright. All rights reserved 532 lipid droplets, due to the presence of PNPLA3-rs738408-G, and lead to increased hepatic 533 steatosis. High insulin levels and PNPLA3-rs738409-G may also be involved in molecular 534 feedback loops that increase hepatic steatosis. Insulin resistance and increased insulin levels 535 augment the activity of transcription factors such as SREBP-1c (39). These transcription factors 536 may promote TG synthesis in the liver and up-regulate the expression of PNPLA3 I148M by 537 binding to its promoter in a positive feedback loop (39). In this way, insulin and PNPLA3 I148M 538 may synergize to promote hepatic steatosis. This conjecture is also consistent with the 539 enhanced risk of steatosis and liver damage as evident by elevated liver enzymes and liver fat 540 content seen with liver directed long-acting insulin analogues in type 2 diabetics carrying the 541 PNPLA-3 variant (40). 542 543 When taken together, results show evidence that insulin and PNPLA3-rs738409 interact to have 544 an important role in hepatic steatosis, and as a result NAFLD. Consequently, lowering the risk of 545 hepatic steatosis and its liver complications in individuals with PNPLA3-rs738409-G may be 546 achieved by reducing insulin resistance and concomitant high levels of insulin. One way to 547 accomplish this could be through lifestyle changes that include increased exercise, weight loss, 548 and better nutrition (41). For example, decreasing exposure to carbohydrate rich diets, which 549 adversely increase insulin levels, may mitigate risk (42, 43). Also, treatments that target insulin 550 resistance may be of greater benefit for preventing or treating hepatic steatosis than drugs that 551 simply lower glucose. For example, insulin sensitizing medications such as pioglitazone may be 552 an option; it has already been shown to improve NAFLD, although at the expense of weight gain 553 (44). More studies are warranted to better understand the effect of the relationship between 554 insulin levels and PNPLA3-rs738409-G on hepatic steatosis in different populations. 555 556 We also observed significant interactions of PNPLA3-rs738409 with BMI, glucose, and TG. Our 557 results support the findings of Stender et al. who reported that high BMI augmented the effect of 558 PNPLA3-rs738409-G on hepatic steatosis conferring susceptibility to NAFLD (45). Graff et al. 559 also showed an interaction effect between PNPLA3-rs738409 and visceral fat content, a 560 measure of metabolic dysfunction (46). However, we found that the effect of BMI in 561 exacerbating hepatic steatosis in the presence of PNPLA3-rs738409-G is attenuated by 562 controlling for insulin levels in the model. We made the same observation for glucose and TG 563 suggesting that insulin/insulin resistance in the presence of PNPLA3-rs738409-G may confer 564 most of the risk for hepatic steatosis on its own or through other metabolic intermediates. 565 A u th o r M a n u s c ri p t 18 This article is protected by copyright. All rights reserved 566 Studies have reported an association between LDL and hepatic steatosis (47, 48). However, 567 our study did not find an interaction between any of the genetic variants considered and LDL. 568 This suggests that for individuals carrying PNPLA3-rs738409-G, reducing insulin levels or 569 insulin resistance may have a greater effect on reducing the risk of hepatic steatosis than 570 reducing LDL. 571 572 In addition to PNPLA3, we found that GCKR interacts with insulin resistance to increase 573 susceptibility to hepatic steatosis. GCKR encodes the glucokinase regulatory protein, which 574 has an important role in glucose metabolism(49). The glucokinase regulatory protein binds to 575 the glucose metabolizing enzyme, glucokinase, to inhibit its role in the uptake and storage of 576 dietary glucose via stimulating de novo lipogenesis(49). The variant rs780094/rs12060326 in 577 the glucokinase regulatory protein reduces its ability to inhibit glucokinase (49). This results in 578 an increased activity of glucokinase in the liver, which promotes de novo lipogenesis. When this 579 mutation is combined with insulin resistance, it may amplify de novo lipogenesis to promote 580 hepatic steatosis. We did not replicate the interaction between TM6SF2 and BMI reported by 581 Stender et al. (45); however, our results show a similar trend. The interaction was borderline 582 non-significant in the combined ancestry meta-analyses (Bint= -0.05, p=5.89E-02). Some 583 differences between Stender et al. and this study may explain why we did not detect a 584 statistically significant interaction. First, Stender et al. used proton magnetic resonance 585 spectrometry to measure steatosis, which is a more sensitive measure than computed 586 tomography. Second, they used the genotyped missense variant, rs58542926; we used the 587 proxy, imputed variant, rs2228603. The two variants are in high linkage disequilibrium 588 (D’=0.926, r2=0.798). Third, Stender et al. combined the heterozygotes (EK), and homozygotes 589 (KK), and compared them to those without the risk allele (EE). These three differences may 590 have increased their power to see the weak effect they reported. 591 592 Our study has several limitations. It is a cross-sectional design that cannot prove temporal 593 causality of insulin exposure on increasing hepatic steatosis. Because we used population- 594 based cohorts that lacked biopsy information, we do not know whether we included individuals 595 with advanced stages of NAFLD such as nonalcoholic steatohepatitis, fibrosis, or cirrhosis. We 596 also could not differentiate peripheral insulin resistance from hepatic insulin resistance with our 597 data. Moreover, even though in euglycemic individuals HOMA-IR was highly correlated to a 598 single value of insulin (r2=0.98), we do not have direct measures of dynamic glucose regulation. 599 Therefore, functional studies are needed to gain more insight into the biological processes A u th o r M a n u s c ri p t 19 This article is protected by copyright. All rights reserved 600 driving our observations. Finally, our study did not include the genetic variant MBOAT7 601 (rs641738), which has been associated with hepatic fat accumulation (50). In our prior 602 association analyses (11), we did not see an association between MBOAT7 and LA 603 (Beta= -0.03, p=0.15). Because our inclusion criteria for variants was that they needed to be 604 associated with LA, and we could not substantiate the association of MBOAT7 in our sample, 605 we excluded it. 606 607 In conclusion, to our knowledge, this is the largest study examining the interaction between 608 multiple metabolic traits and four genetic variants on hepatic steatosis in multiple cohorts 609 representing two different ancestry groups. Our findings suggest that insulin levels/insulin 610 resistance more than other correlated metabolic traits including glucose, TG, and BMI interact 611 with genetic variants in PNPLA3 to promote hepatic steatosis. Through conditional analyses, 612 we show that insulin levels explain the interactions observed between PNPLA3-rs738409 and 613 BMI, as well as the interactions between PNPLA3-rs738409 and glucose and TG, in almost 614 5,000 nondiabetic, EA individuals. Our work suggests that improving insulin resistance and 615 reducing insulin levels in pre-diabetic individuals carrying fatty liver promoting alleles at 616 PNPLA3-rs738409 may offer preferential benefit and mitigate their risk of developing NAFLD. 617 Although PNPLA3 genotype information is not currently used to make clinical decisions, it may 618 be helpful in the future not only to risk stratify individuals, but also to tailor their treatment. Our 619 work contributes to the understanding of the pathophysiology of NAFLD, and informs further 620 interventional research to better diagnose and/or treat individuals with increased risk of NAFLD. 621 622 623 624 625 626 627 628 629 REFERENCES 630 631 1. Kahali B, Halligan B, Speliotes EK. Insights from Genome-Wide Association Analyses of 632 Nonalcoholic Fatty Liver Disease. Semin Liver Dis 2015;35:375-391. A u th o r M a n u s c ri p t 20 This article is protected by copyright. All rights reserved 633 2. Speliotes EK, Massaro JM, Hoffmann U, Vasan RS, Meigs JB, Sahani DV, et al. Fatty 634 liver is associated with dyslipidemia and dysglycemia independent of visceral fat: the 635 Framingham Heart Study. Hepatology 2010;51:1979-1987. 636 3. Gaggini M, Morelli M, Buzzigoli E, DeFronzo RA, Bugianesi E, Gastaldelli A. Non- 637 alcoholic fatty liver disease (NAFLD) and its connection with insulin resistance, dyslipidemia, 638 atherosclerosis and coronary heart disease. Nutrients 2013;5:1544-1560. 639 4. Hassan K, Bhalla V, El Regal ME, HH AK. Nonalcoholic fatty liver disease: a 640 comprehensive review of a growing epidemic. World J Gastroenterol 2014;20:12082-12101. 641 5. McPherson S, Hardy T, Henderson E, Burt AD, Day CP, Anstee QM. Evidence of 642 NAFLD progression from steatosis to fibrosing-steatohepatitis using paired biopsies: 643 implications for prognosis and clinical management. J Hepatol 2015;62:1148-1155. 644 6. Pais R, Charlotte F, Fedchuk L, Bedossa P, Lebray P, Poynard T, et al. A systematic 645 review of follow-up biopsies reveals disease progression in patients with non-alcoholic fatty 646 liver. J Hepatol 2013;59:550-556. 647 7. Wong VW, Wong GL, Choi PC, Chan AW, Li MK, Chan HY et al. Disease progression of 648 non-alcoholic fatty liver disease: a prospective study with paired liver biopsies at 3 years. Gut 649 2010;59:969-974. 650 8. Younossi ZM, Koenig AB, Abdelatif D, Fazel Y, Henry L, Wymer M. Global epidemiology 651 of nonalcoholic fatty liver disease-Meta-analytic assessment of prevalence, incidence, and 652 outcomes. Hepatology 2016;64:73-84. 653 9. Speliotes EK, Yerges-Armstrong LM, Wu J, Hernaez R, Kim LJ, Palmer CD, et al. 654 Genome-wide association analysis identifies variants associated with nonalcoholic fatty liver 655 disease that have distinct effects on metabolic traits. PLoS Genet 2011;7:e1001324. 656 10. Romeo S, Kozlitina J, Xing C, Pertsemlidis A, Cox D, Pennacchio LA, et al. Genetic 657 variation in PNPLA3 confers susceptibility to nonalcoholic fatty liver disease. Nat Genet 658 2008;40:1461-1465. 659 11. Palmer ND, Musani SK, Yerges-Armstrong LM, Feitosa MF, Bielak LF, Hernaez R, et al. 660 Characterization of European ancestry nonalcoholic fatty liver disease-associated variants in 661 individuals of African and Hispanic descent. Hepatology 2013;58:966-975. 662 12. Hernaez R, McLean J, Lazo M, Brancati FL, Hirschhorn JN, Borecki IB, Harris TB, et al. 663 Association between variants in or near PNPLA3, GCKR, and PPP1R3B with ultrasound- 664 defined steatosis based on data from the third National Health and Nutrition Examination 665 Survey. Clin Gastroenterol Hepatol 2013;11:1183-1190 e1182. A u th o r M a n u s c ri p t 21 This article is protected by copyright. All rights reserved 666 13. Lin YC, Chang PF, Chang MH, Ni YH. Genetic variants in GCKR and PNPLA3 confer 667 susceptibility to nonalcoholic fatty liver disease in obese individuals. Am J Clin Nutr 668 2014;99:869-874. 669 14. Liu YL, Patman GL, Leathart JB, Piguet AC, Burt AD, Dufour JF, et al. Carriage of the 670 PNPLA3 rs738409 C >G polymorphism confers an increased risk of non-alcoholic fatty liver 671 disease associated hepatocellular carcinoma. J Hepatol 2014;61:75-81. 672 15. Locke AE, Kahali B, Berndt SI, Justice AE, Pers TH, Day FR, et al. Genetic studies of 673 body mass index yield new insights for obesity biology. Nature 2015;518:197-206. 674 16. Chatrath H, Vuppalanchi R, Chalasani N. Dyslipidemia in patients with nonalcoholic fatty 675 liver disease. Semin Liver Dis 2012;32:22-29. 676 17. Bellentani S, Scaglioni F, Marino M, Bedogni G. Epidemiology of non-alcoholic fatty liver 677 disease. Dig Dis 2010;28:155-161. 678 18. Zhang QQ, Lu LG. Nonalcoholic Fatty Liver Disease: Dyslipidemia, Risk for 679 Cardiovascular Complications, and Treatment Strategy. J Clin Transl Hepatol 2015;3:78-84. 680 19. Petaja EM, Yki-Jarvinen H. Definitions of Normal Liver Fat and the Association of Insulin 681 Sensitivity with Acquired and Genetic NAFLD-A Systematic Review. Int J Mol Sci 2016;17. 682 20. Limanond P, Raman SS, Lassman C, Sayre J, Ghobrial RM, Busuttil RW, et al. 683 Macrovesicular hepatic steatosis in living related liver donors: correlation between CT and 684 histologic findings. Radiology 2004;230:276-280. 685 21. Graffy PM, Pickhardt PJ. Quantification of hepatic and visceral fat by CT and MR 686 imaging: relevance to the obesity epidemic, metabolic syndrome and NAFLD. Br J Radiol 687 2016;89:20151024. 688 22. D'Orazio P, Burnett RW, Fogh-Andersen N, Jacobs E, Kuwa K, Kulpmann WR, et al. 689 Approved IFCC recommendation on reporting results for blood glucose (abbreviated). Clin 690 Chem 2005;51:1573-1576. 691 23. Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC. 692 Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma 693 glucose and insulin concentrations in man. Diabetologia 1985;28:412-419. 694 24. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density 695 lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem 696 1972;18:499-502. 697 25. National Institute of Alcohol Abuse and Alcoholism . What is a Standard Drink? Available 698 at https://www.niaaa.nih.gov/alcohol-health/overview-alcohol-consumption/what-standard-drink. 699 Accessed 2017 A u th o r M a n u s c ri p t 22 This article is protected by copyright. All rights reserved 700 26. USDA. Dietary Guidelines for Americans 2015-2020 (Appendix 9: Alcohol). Available at 701 https://health.gov/dietaryguidelines/2015/guidelines/appendix-9/ (Accessed 2017). 702 27. O'Connell J. MMAP:Mixed Model Analysis for Pedigrees and Populations. Available: 703 https//mmap.github.io (Accessed 4 April, 2017). In; 2013. 704 28. R Development Core Team (2008). R: A language and environment for statistical 705 computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0 URL 706 http://www.R-project.org/ 707 29. SAS Institute Inc; SAS Software, Version 9.4, Cary, NC: SAS Institute Inc. 2011. 708 30. Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta-analysis of genomewide 709 association scans. Bioinformatics 2010;26:2190-2191. 710 31. Cochran WG. The combination of estimates from different experiments. Biometrics 711 1954;10:101-129. 712 32. Park SH, Kim PN, Kim KW, Lee SW, Yoon SE, Park SW et al. Macrovesicular hepatic 713 steatosis in living liver donors: use of CT for quantitative and qualitative assessment. Radiology 714 2006;239:105-112. 715 33. Zeb I, Li D, Nasir K, Katz R, Larijani VN, Budoff MJ. Computed tomography scans in the 716 evaluation of fatty liver disease in a population based study: the multi-ethnic study of 717 atherosclerosis. Acad Radiol 2012;19:811-818. 718 34. Chamoun Z, Vacca F, Parton RG, Gruenberg J. PNPLA3/adiponutrin functions in lipid 719 droplet formation. Biol Cell 2013;105:219-233. 720 35. Li JZ, Huang Y, Karaman R, Ivanova PT, Brown HA, Roddy T, et al. Chronic 721 overexpression of PNPLA3I148M in mouse liver causes hepatic steatosis. J Clin Invest 722 2012;122:4130-4144. 723 36. Smagris E, BasuRay S, Li J, Huang Y, Lai KM, Gromada J, et al. Pnpla3I148M knockin 724 mice accumulate PNPLA3 on lipid droplets and develop hepatic steatosis. Hepatology 725 2015;61:108-118. 726 37. Johnson AM, Olefsky JM. The origins and drivers of insulin resistance. Cell 727 2013;152:673-684. 728 38. Saponaro C, Gaggini M, Gastaldelli A. Nonalcoholic fatty liver disease and type 2 729 diabetes: common pathophysiologic mechanisms. Curr Diab Rep 2015;15:607. 730 39. Dubuquoy C, Robichon C, Lasnier F, Langlois C, Dugail I, Foufelle F, et al. Distinct 731 regulation of adiponutrin/PNPLA3 gene expression by the transcription factors ChREBP and 732 SREBP1c in mouse and human hepatocytes. J Hepatol 2011;55:145-153. A u th o r M a n u s c ri p t 23 This article is protected by copyright. All rights reserved 733 40. Pillai S, Duvvuru S, Bhatnagar P, Foster W, Farmen M, Shankar S, Harris C, et al. The 734 PNPLA3 I148M variant is associated with transaminase elevations in type 2 diabetes patients 735 treated with basal insulin peglispro. Pharmacogenomics J 2018;18:487-493. 736 41. Maglio C. The PNPLA3 I148M variant and chronic liver disease: When a genetic 737 mutation meets nutrients. Food Research International 2014;63:293-243. 738 42. Davis JN, Le KA, Walker RW, Vikman S, Spruijt-Metz D, Weigensberg MJ, et al. 739 Increased hepatic fat in overweight Hispanic youth influenced by interaction between genetic 740 variation in PNPLA3 and high dietary carbohydrate and sugar consumption. Am J Clin Nutr 741 2010;92:1522-1527. 742 43. Stojkovic IA, Ericson U, Rukh G, Riddestrale M, Romeo S, Orho-Melander M. The 743 PNPLA3 Ile148Met interacts with overweight and dietary intakes on fasting triglyceride levels. 744 Genes Nutr 2014;9:388. 745 44. Sanyal AJ, Chalasani N, Kowdley KV, McCullough A, Diehl AM, Bass NM, 746 Neuschwander-Tetri BA, et al. Pioglitazone, vitamin E, or placebo for nonalcoholic 747 steatohepatitis. N Engl J Med 2010;362:1675-1685. 748 45. Stender S, Kozlitina J, Nordestgaard BG, Tybjaerg-Hansen A, Hobbs HH, Cohen JC. 749 Adiposity amplifies the genetic risk of fatty liver disease conferred by multiple loci. Nat Genet 750 2017. 751 46. Graff M, North KE, Franceschini N, Reiner AP, Feitosa M, Carr JJ, et al. PNPLA3 gene- 752 by-visceral adipose tissue volume interaction and the pathogenesis of fatty liver disease: the 753 NHLBI family heart study. Int J Obes (Lond) 2013;37:432-438. 754 47. Sun DQ, Liu WY, Wu SJ, Zhu GQ, Braddock M, Zhang DC, et al. Increased levels of 755 low-density lipoprotein cholesterol within the normal range as a risk factor for nonalcoholic fatty 756 liver disease. Oncotarget 2016;7:5728-5737. 757 48. Papandreou D, Karabouta Z, Rousso I. Are dietary cholesterol intake and serum 758 cholesterol levels related to nonalcoholic Fatty liver disease in obese children? Cholesterol 759 2012;2012:572820. 760 49. Raimondo A, Rees MG, Gloyn AL. Glucokinase regulatory protein: complexity at the 761 crossroads of triglyceride and glucose metabolism. Curr Opin Lipidol 2015;26:88-95. 762 50. Mancina RM, Dongiovanni P, Petta S, Pingitore P, Meroni M, Rametta R, Boren J, 763 et al. The MBOAT7-TMC4 Variant rs641738 Increases Risk of Nonalcoholic Fatty Liver Disease 764 in Individuals of European Descent. Gastroenterology 2016;150:1219-1230 e1216. A u th o r M a n u s c ri p t 24 This article is protected by copyright. All rights reserved 765 766 Author names in bold designate shared co-first authorship. 767 768 ACKNOWLEDGMENT 769 We thank the reviewers for their helpful suggestions, which have improved this paper. 770 Conflict of Interest 771 Drr Laura Yerges-Armstrong is a current employee stockholder for GlaxoSmithKline, however, 772 the current work was conducted while at University of Maryland School of Medicine. Dr. Ingrid 773 Borecki owns stock in Regeneron Pharmaceuticals. Dr. Jeffrey R. O'Connell was a consultant 774 for Regeneron Pharmaceuticals for a period of time during this study. 775 776 777 778 779 780 781 Figure 1. Shown is the percentage of non-diabetic individuals in FamHS with ≥ 30% fat in the 782 liver (moderate to severe hepatic steatosis) per PNPLA3-rs738409 genotype in the lowest and 783 highest quartile of insulin levels. As the level of insulin increases, the percentage of individuals 784 with ≥ 30% fat in the liver increases more markedly with increasing copies of the G risk allele 785 (non-parallel lines show interaction). Among those with the GG genotype, the difference (∆) in 786 the percentage of individuals with moderate to severe liver fat increases by 55 percentage 787 points between the lowest and highest insulin quartiles. In contrast, this difference is lower 788 among those with the CG genotype (41%), and CC genotype (31%).A u th o r M a n u s c ri p t 1 This article is protected by copyright. All rights reserved Table 1. Demographic and Characteristics of Study Participants in each Cohort by Ancestry AGES Amish CARDIA FamHS FHS MESA CARDIA FamHS GENOA MESA Demographic European Ancestry (11,870) African Ancestry (2,881) N=14,751 2,865 541 1,282 2,684 2,966 1,532 642 620 560 1,059 Age 76.38 ± 5.46 56.84 ± 12.81 50.74 ± 3.33 57.14 ± 13.28 50.54 ± 10.14 63.05 ± 10.49 49.47 ± 3.86 53.35 ± 10.82 68.86 ± 8.01 63.17 ± 10.00 Men (6,444) 1,139 (40%) 252 (47%) 595 (46%) 1,207 (45%) 1,454 (49%) 746 (49%) 233 (36%) 212 (34%) 141 (25%) 465 (44%) Women (8,307) 1,726 (60%) 289 (53%) 687 (54%) 1,477 (55%) 1,512 (51%) 786 (51%) 409 (64%) 408 (66%) 419 (75%) 594 (56%) Characteristics Liver Attenuation (HU) ¥ 59.22 ± 8.64 63.05 ± 7.76 55.05 ± 12.28 59.14 ± 11.19 65.40 ± 9.83 59.33 ± 12.43 56.38 ± 10.86 59.52 ± 9.23 60.10 ± 9.39 61.18 ± 9.06 Insulin (mU/L) 9.22 ± 6.39 11.75 ± 6.22 9.26 ± 6.77 9.88 ± 7.16 9.05 ± 7.38 8.90 ± 4.94 11.60 ± 8.29 13.02 ± 10.22 8.30 ± 5.73 9.61 ± 5.53 HOMA-IR ょ 2.31 ± 1.76 2.69 ± 1.63 2.20 ± 1.77 2.37 ± 1.87 2.32 ± 2.31 1.99 ± 1.27 2.75 ± 2.14 3.14 ± 2.69 2.02 ± 1.46 2.19 ± 1.41 Glucose (mmol/L) 5.49 ± 0.50 4.94 ± 0.52 5.18 ± 0.50 5.25 ± 0.53 5.47 ± 1.12 4.90 ± 0.58 5.17 ± 0.54 5.25 ± 0.57 5.37 ± 0.50 5.03 ± 0.59 BMI (kg/m 2 ) 27.00 ± 4.49 27.72 ± 4.85 28.50 ± 6.18 28.86 ± 5.69 27.51 ± 5.22 28.06 ± 5.05 31.94 ±7.48 32.71 ± 7.37 32.71 ± 7.27 29.95 ± 5.77 Obese ゆ 618 (22%) c(((22%) 155 (29%) 425 (33%) 972 (36%) 769 (26%) 449 (29%) 352 (55%) 377 (61%) 332 (59%) 465 (44%) WHR (cm) § nval 0.87 ± 0.07 0.85 ± 0.10 0.91 ± 0.10 0.94 ± 0.08 0.93 ± 0.09 0.85 ± 0.08 0.92 ± 0.07 0.89 ± 0.08 0.92 ± 0.08 TG (mg/dL) 106.48 ± 59.06 90.42 ± 57.45 121.64 ± 85.07 144.03± 94.05 126.11 ± 88.07 136.55 ± 99.31 101.55 ± 73.24 111.82 ± 80.09 100.28 ± 62.67 103.82 ± 60.61 LDL (mg/dL) 146.84 ± 35.73 141.31± 38.66 116.27 ± 30.15 112.9 ± 34.22 117.70 ± 31.71 120.24 ± 30.42 112.59 ± 33.83 115.39 ± 36.05 123.85 ± 33.59 118.39 ± 32.87 HDL (mg/dL) 61.75 ± 17.31 57.05 ± 15.37 58.43 ± 18.42 48.82 ± 14.37 54.16 ± 16.77 51.68 ± 15.59 57.59 ± 16.70 53.55 ± 15.41 57.31 ± 16.52 52.39 ± 15.14 Alcohol (drinks/week) 1.09 ± 2.37 nval 5.73 ± 10.07 2.98 ± 7.10 5.39 ± 7.88 5.06 ± 8.40 3.86 ± 10.60 3.24 ± 9.45 0.28 ± 1.18 3.86 ± 8.89 Heavy drinkers* 17 (0.59%) nval 470 (37%) 152 (6%) 424 (14.3%) 335 (22%) 144 (22%) 69 (11%) 0 139 (13%) Statistics are presented as mean ± standard deviation (SD), or as n (%). The table includes individuals with liver attenuation and genetic information from each cohort that were included in analyses. LA and metabolic traits were not adjusted for covariates. The sample size for each trait varied from N depending on the data available. Summary statistics for fasting insulin, HOMA-IR and fasting glucose excludes diabetics; fasting LDL excludes statin users. ¥ Raw liver attenuation measured in Hounsfield units. ‡ Calculated as [fasting insulin (mU/L) x fasting glucose (mmol/L)/22.5]; † Defined as BMI ≥ 30 kg/m2; § not adjusted for BMI; nval= not available in A u th o r M a n u s c ri p t 2 This article is protected by copyright. All rights reserved cohort. *Defined as ≥ 8 drinks per week for women and ≥ 15 drinks per week for men. The Amish do not consume alcohol. Units in the table are HU=Hounsfield units; mU/L=milliunits per liter; mmol/L=millimoles per liter; kg/m 2 = kilograms divided by height in meters squared; cm=centimeters; mg/dL=milligram per deciliter. Table 2. Meta-analyses results for interactions between four SNPs and inverse normal-transformed residuals of metabolic traits on LAivn in combined ancestries. rs738409 rs780094* rs2228603* rs12137855 Gene Chr Alleles (Ref/O) Ref AF Gene Chr Alleles (Ref/O) Ref AF Gene Chr Alleles (Ref/O) Ref AF Gene Chr Alleles (Ref/O) Ref AF PNPLA3 22 G/C 0.24 GCKR 2 T/C 0.39 NCAN/ TM6SF2 19 T/C 0.13 LYPLAL1 8 C/T 0.79 (SNP x Metabolic Traits) Metabolic Traits βint SE P-value N βint SE P-value N βint SE P-value N βint SE P-value N Insulin -0.11 0.02 4.79E-14 12,651 -0.04 0.01 4.57E-04 12,651 -0.06 0.03 4.37E-02 12,651 -0.02 0.02 1.55E-01 12,651 HOMA-IR -0.12 0.02 4.68E-15 12,554 -0.04 0.01 1.32E-03 12,554 -0.06 0.03 3.63E-02 12,554 -0.02 0.02 1.38E-01 12,554 Glucose -0.05 0.02 1.26E-03 12,742 -0.01 0.01 4.37E-01 12,742 -0.06 0.03 7.41E-02 12,742 -0.02 0.02 1.91E-01 12,742 BMI -0.08 0.01 8.13E-08 14,693 -0.03 0.01 6.31E-03 14,693 -0.05 0.03 5.89E-02 14,693 -0.02 0.01 8.18E-01 14,693 WHRadjBMI -0.05 0.02 7.59E-03 10,051 -0.04 0.02 1.32E-02 10,051 -0.08 0.03 1.26E-02 10,051 0.01 0.02 7.76E-01 10,051 TG -0.05 0.02 2.95E-03 14,551 -0.04 0.01 4.17E-03 14,551 0.00 0.03 9.77E-01 14,551 -0.03 0.02 5.75E-02 14,551 LDL 0.00 0.02 7.94E-01 12,123 0.00 0.01 9.33E-01 12,123 -0.06 0.03 5.50E-02 12,123 0.02 0.02 2.29E-01 12,123 HDL -0.04 0.02 1.41E-02 14,543 0.03 0.01 3.72E-02 14,543 0.00 0.03 9.55E-01 14,543 -0.01 0.02 3.72E-01 14,543 Chr, Chromosome; Ref/O, Reference/Other allele (reference allele is the effect allele of each SNP); Ref AF, Reference allele frequency; βint, interaction effect size; SE, standard error. P-values that reached significance threshold (P≤ 4.17E-03) are in bold; N is the highest sample size in meta-analyses. *rs780094 is in LD (r 2 =0.93) with rs1260326, a functional missense variant in GCKR; *rs2228603 is in LD (r 2 =0.79) with rs58542926, a functional missense variant in TM6SF2. A u th o r M a n u s c ri p t hep4_1353_f1.tif This article is protected by copyright. All rights reserved A u th o r M a n u s c ri p t