Functional Characterization of Lipid Regulatory Effects of Three Genes Using Knockout Mouse Models

25 Integrative analysis that combines genome-wide association data with expression quantitative trait 26 analysis and network representation may illuminate causal relationships between genes and 27 diseases. To identify causal lipid genes, we utilized genotype, gene expression, protein -protein 28 interaction networks, and phenotype data from 5,257 Framingham Heart Study participants and 29 performed Mendelian randomization to investigate possible mechanistic explanations for observed 30 associations. We selected three putatively causal candidate genes ( ABCA6, ALDH2, and SIDT2) 31 for lipid traits (LDL cholesterol, HDL cholesterol and triglycerides) in humans and conducted 32 mouse knockout studies for each gene to confirm its causal effect on the corresponding lipid trait. 33 We conducted the RNA -seq from mouse livers to explore transc riptome-wide alterations after 34 knocking out the target genes. Our work builds upon a lipid- related gene network and expands 35 upon it by including protein -protein interactions. These resources, along with the innovative 36 combination of emerging analytical techniques, provide a groundwork upon which future studies 37 can be designed to more fully understand genetic contributions to cardiovascular diseases. 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 2

61 Cardiovascular disease (CVD) remains the leading cause of death worldwide1. A key goal of CVD 62 research is the identification of specific genes and genetic variants that contribute to the disease2. 63 Genome-wide association studies (GWAS) have revealed numerous inherited DNA sequence 64 variants associated with CVD and its risk factors 3-5. For many known CVD -associated variants, 65 however, the biological mechanisms and underlying causal genes remain largely unknown. The 66 overwhelming majority of CVD-related genetic variants lie within intergenic or noncoding regions 67 of the genome 3, indicating that these single nucleotide polymorphisms (SNPs) impact CVD ris k 68 indirectly, for example, by altering transcription levels of nearby (cis) or remote (trans) genes. The 69 identification of genetic variants that alter gene expression – expression quantitative trait loci 70 (eQTLs) – has paved the way for functional studies linking candidate genes to disease6. 71 We postulated a priori that a network approach could portray biological systems underlying 72 CVD traits and provide insights into disease -related genes and pathways. Network -based 73 approaches have many biological and clinical applications and have yielded promising results in 74 recent studies7, 8. Not only do networks allow for the integration of different types of biological 75 information, but they also allow for unbiased discovery of disease genes that, when integrated with 76 protein-protein interactions (PPI), can provide mechanistic explanations for diseases 9. In a 77 previous study, we linked 21 CVD traits based on their shared SNP associations, and in doing so, 78 recapitulated the clustering of metabolic risk factors observed in epidemiological studies 10. Here, 79 we expand that network model to identify causal genes and pathways for lipid traits and tested 80 three genes for causality. 81 By integrating high- quality PPI data 11 with SNPs associated in GWAS of lipids, gene 82 expression, and fasting blood lipid levels in 5,257 Framingham Heart Study (FHS) participants 83 (Supplementary table 1), we selected ABCA6 , ALDH2, and SIDT2 as candidate causal genes for 84 lipid traits (low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, 85 and triglycerides (TG)). Dysregulation of circulating lipid levels has been implicated in the 86 pathogenesis of CVD12. To confirm that the candidate genes identified from the network model 87 are causal for the corresponding lipid traits, we created mouse knockout (KO) models to assess, in 88 vivo, the consequences of perturbing key genes as well as the networks in which they function. 89 The mouse KO models were therefore phenotyped for the corresponding lipid traits and 90 additionally liver gene expression was measured by RNA -Seq to study perturbations within the 91 network (Figure 1). 92 We posit that creating a network based on the relationships among genes, traits, and PPIs can 93 further our understanding of the genetic basis of lipid traits and their contributions to lipid 94 dysregulation. By combining human genotype data with mouse KO experiments, we also provide 95 a framework for selecting and validating lipid-related candidate genes. 96

97 Identifying lipid trait-related expression quantitative loci 98 We identified 4670 (84%) lipid trait- associated expression quantitative loci (eQTLs; SNPs 99 association with gene expression) by linking 5568 lipid trait GWAS SNPs (Supplementary Table 100 2) that were also associated with gene expression in whole blood in 5,257 Framingham 101 participants13 (Supplementary Table 3). Seventy- two percent of these eQTLs affected the 102 expression of a nearby (cis) transcript (SNP located within 1 Mb of the transcript), suggesting that 103 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 3 they may play a role in regulating gene expression. For these eQTLs, we conducted mediation 104 testing to detect if the association of the GWAS SNP with the corresponding trait was mediated 105 by gene expression. At P<0.005, we identified 464 SNPs with significant mediation effects on 106 lipids traits (Supplementary Table 4). 107 Expanding the candidate gene network 108 To expand the candidate gene network, we built a CVD gene network by incorporating all protein 109 coding genes containing CVD GWAS SNPs. Our CVD-centric network contained binary PPIs for 110 846 CVD-associated proteins (out of ~1,300 tested) extracted from a dataset of ~58,000 binary 111 PPIs among 10,690 human proteins obtained from both systematic binary mapping and lite rature 112 curations14. The 846 CVD -associated proteins and their first -degree neighbors constituted a 113 network of ~8,600 interactions among 4,336 proteins (Supplementary Table 5), including 444 114 intra-CVD PPIs among 349 lipids -associated proteins (Supplementary Table 6). This ex panded 115 network is amenable to investigating disease models by analyzing other features such as shared 116 gene ontology (GO) terms, shared expression profiles, and functional similarity. 117 Ranking candidate genes 118 Combining GWAS lipid trait SNPs, eQTLs, and medi ation results, we selected seven candidate 119 genes to test against three lipid traits in Mendelian randomization (MR) analyses: DOCK7, 120 TAGLN, SIDT2, ALDH2, SLC44A4, ABCA6, and ATG4C (See Methods for gene selection). We 121 used genetic variants (focusing on those that are eQTLs variants) as instruments to investigate the 122 causal relations between gene expression and lipid phenotypes. The causal effects of the seven 123 genes were further tested by MR using independent GWAS data from the Global Lipids Genetics 124 Consortium (GLGL)15 (Table 1) and filtered based on 1) gene novelty (not previously studied in a 125 mouse model for cardiovascular phenotypes), 2) repository availability of knockout sperm or 126 embryos that have been shown to produce viable animals, and 3) interaction with CVD -related 127 proteins in an integrated network based on PPIs. This approach identified three genes as suitable 128 for detailed mouse KO experiments: ABCA6 for non-HDL, ALDH2 for HDL, and SIDT2 for TG 129 (Table 2). Though ABCA6 was not found to be causally associated with lipids traits by MR. 130 rs740516, a variant in ABCA6, was significantly associated with LDL in GWAS (P=7x10 -9) and 131 was associated with expression of ABCA6 (P=3.3x10-5). SIDT2 has been previously studied in 132 relation to lipid metabolism and homeostasis, our finding is one of the first to utilize an integrative 133 genomics framework to better understand the genetic ba sis of its effect on lipid traits 16, 17. The 134 effects of these genes on secondary lipid traits (ABCA6 on HDL and TG, ALDH2 on non-HDL and 135 TG, SIDT2 on HDL and non-HDL) are also summarized. 136 ABCA6, ALDH2, SIDT2, and the corresponding mouse knockout models 137 ABCA6 Knockout 138 The protein ABCA6, encoded by ABCA6, is a member of the superfamily of ATP-binding cassette 139 (ABC) transporters. ABC proteins transport various substrates, including lipids, peptides, 140 vitamins, and ions18, across extra- and intracellular membranes. ABCA6 is expressed exclusively 141 in multicellular eukaryotes and has been suspected of acting as an intracellular transporter in 142 macrophage lipid homeostasis19. 143 Comparing total cholesterol and non-HDL cholesterol levels (which are primarily comprised of 144 LDL cholesterol) between C57BL/6N- A bca6tm2a(KOMP)Wtsi/TcpRkorJ mice (Abca6 knockout) and 145 WT C57BL/6N mice, we observed a significant increase in total cholesterol in males on the chow 146 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 4 diet at 22 weeks of age and in males on the high- fat diet at both 18 and 22 weeks of age (Figure 147 2). A similar trend was identified in females. 148 In addition, knockout of ABCA6 led to a significant increase in HDL cholesterol in male mice 149 on the chow diet but not on a high- fat diet with a similar but nonsignificant trend observed in 150 females. This finding is supported by previous studies that have shown that ABCA6 is involved in 151 reverse cholesterol transport, specifically by facilitating phospholipid and cholesterol export from 152 the cell19, 20. There were no significant differences in TG levels i n the female mice on both chow 153 and high-fat diet, but the male Abca6-/- mice had lower TG levels on the chow diet. 154 ALDH2 Knockout 155 ALDH2 encodes aldehyde dehydrogenase, the second enzyme of the major oxidative pathway of 156 alcohol metabolism. This gene encodes a mitochondrial isoform, which has a high affinity for 157 acetaldehydes and is localized in the mitochondrial matrix. Deficiencies in ALDH2 have been 158 shown to be correlated with differences in lipid levels due to its role in the metabolism of lipid -159 peroxidation-derived aldehydes 21. 160 ALDH2 was causally related to all three lipids in MR (Table 1). In addition, SNPs in ALDH2 161 were significantly associated with HDL and LDL in GWAS and gene expression in FHS 162 (Supplementary Table 3). Comparing HDL cholesterol levels between B6Dnk;B6N -163 Aldh2tm1a(EUCOMM)Wtsi/IegRkorJ mice (Aldh2 knockout) and WT C57BL/6N mice, we found 164 significantly increased HDL cholesterol levels in both female and male Aldh2 KO mice at several 165 time points compared to controls when fed the chow diet (Figure 3). On a high- fat diet, we no 166 longer observed a difference in males, but there was a significant difference in females at 18 weeks 167 of age. 168 Except for a few specific time points, there were no differences in non -HDL levels for either 169 diet in males or females. There were no differences in TG levels in the female mice, but in the 170 male mice we observed increased TG levels in the KO animals on a high -fat diet. This suggests 171 that the knockout of Aldh2 in males might lead to a decreased risk of CVD on the chow diet, but 172 an increased risk on the high-fat diet. 173 SIDT2 Knockout 174 SIDT2 is a transmembrane protein that predominantly localizes on lysosomes but is also detectable 175 in the plasma membrane of human embryonic kidney cells. Overexpression of SIDT2 in some 176 cells is accompanied by a significant reduction of detectable lysosomes, indicating that the 177 overexpressed protein leads to lysosomal dysfunction. Lysosomes are thought to be the major 178 intracellular compartment for the degradation of macromolecules and play an important role in 179 regulating lipid degradation pathways22, 23. Previous SIDT2-knockout experimentation has shown 180 that SIDT2 plays a major role in regulating lipid autophagy and metabolism and cholesterol and 181 triglyceride transport in mammalian cells, particularly in the liver16, 17. 182 SIDT2 was significant in MR test for all three lipids traits. SNPs in SIDT2 were significantly 183 associated with LDL in GWAS and gene expression in FHS (Supplementary Table 3). We 184 identified SIDT2 in our human cohort to be associated primarily with triglycerides and therefore 185 compared TG levels between B6;129S5-Sidt2tm1Lex/MmucdRkorJ mice (Sidt2 knockout) and WT 186 control littermates. We observed a significant increase in TG levels in the female mice on the chow 187 diet at 18 and 22 weeks of age (Figure 4). Male TG levels were only increased at 8 weeks of age. 188 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 5 In addition, HDL cholesterol levels were significantly decreased in male and female KO mice 189 on the high-fat diet. On a chow diet, HDL cholesterol levels appeared to decrease even further in 190 male mice. HDL cholesterol levels, however, increased slightly in females. Female non -HDL 191 cholesterol levels did not change on the chow diet; in the early stages there was a large decrease 192 in non-HDL cholesterol on the high-fat diet that ultimately dissipated. Male non-HDL cholesterol 193 levels did not change on the high-fat diet and significantly increased on a chow diet. 194 Expression changes after gene knock out 195 To explore transcriptome -wide alterations resulting from knocking out the target genes, we 196 conducted RNA-Seq for ALDH2 and ABCA6 in three male and three female knockout animals and 197 three male and three female control littermates for each of the two diets. We didn’t conduct the 198 RNA-Seq for SIDT2 in light of previously reported studies of this knockout 16, 17. At FDR<0.05 (T 199 test, P<0.014), we identified an average of 238 differentially-expressed genes (ranging from 69 to 200 403 for eight comparisons) after knockout (Supplementary Table 7). The differentially-expressed 201 genes were analyzed through the use of Ingenuity Pathway Analysis (IPA,QIAGEN Inc.,) 24. 202 Function enrichment analyses were conducted on the differentially expressed genes using curated 203 information from the QIAGEN Knowledge Base 24. Though canonical pathways of differentially 204 expressed genes from mice of different diet conditions are enriched in different metabolic and cell 205 signaling pathways, the affected diseases and molecular functions are highly consistent with regard 206 to metabolic disease and lipid metabolism (Supplementary figure 1). The top toxicological 207 functions were liver steatosis and cardiac dysfunction. Network analysis based on Ingenuity 208 Knowledge Base24 revealed that Abca6 affected cholesterol level via cytokine IFNG (interferon 209 gamma) and that Aldh2 affected cholesterol level through enzyme GNMT (glycine N -210 methyltransferase) and acetaldehyde (Supplementary Figure 2). Comparing RNA-seq data from 211 Abca6 and Aldh2 knockout of the same sex and diet, we found 19%-60% overlap of differentially 212 expressed genes. Using IPA functional enrichment at P<0.005, we found that the overlapping 213 genes are enriched in lipid metabolism ( Supplementary Table 8), suggesting a common pathway 214 affected by these two very different genes. 215

216 To represent the complex molecular dynamics that underlie CVD, we constructed a network that 217 integrates SNPs, gene expression, protein- protein interactions, and lipid phenotypes. From this 218 integrated network approach, we identified three genes – ABCA6, ALDH2, and SIDT2 – that were 219 linked to lipid levels and were investigated further through mouse knockout models. The mouse 220 experiments recapitulated network-predicted relations of all three genes to the corresponding lipid 221 traits and provided experimental support for the integrative approach we developed. 222 Disease processes involve many interacting molecules and physical and biochemical processes, 223 and thus the analysis of single data types is often insufficient to explain the etiology of complex 224 traits. For example, carriers of risk alleles do not always have phenotypic consequences as such 225 genetic variants do not necessarily alter the expression of disease- related genes or proteins. 226 Therefore, in order to draw a more comprehensive view of biological processes, high-throughput 227 data from different elements of multidimensional molecular information must be integrated and 228 analyzed. 229

from our mouse knockout models provide proof of principle for the utility of the 230 integrative network approach that we developed. Results of the mouse experiments largely support 231 our network model, with a few exceptions. Knockout of Abca6, which contains GWAS SNPs for 232 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 6 LDL cholesterol and total cholesterol, resulted in higher total cholesterol (which is largely LDL 233 cholesterol) and HDL cholesterol in male mice, suggesting a potential role of this gene in CVD 234 development. ABCA6 plays a key role in macrophage lipid homeostasis by facilitating 235 phospholipid and cholesterol export from the cell 19, 20. ABCA6 dysfunction may cause a buildup 236 of intracellular phospholipids and cholesterol. Knockout of Aldh2, which was chosen for its 237 putatively causal association with HDL cholesterol, resulted in significantly increased HDL 238 cholesterol levels in both female and male mice. Aldehyde dehydrogenase, the protein encoded by 239 ALDH2, is an enzyme in the oxidative pathway and plays a major role the metabolism of lipid 240 peroxidation-derived aldehydes 21. In our knockout models, the absence of Aldh2 resulted in higher 241 lipid levels, which supports the idea that ALDH2 is involved in lipid degradation/clearance. 242 Knockout of Sidt2, which contains GWAS SNPs for HDL cholesterol, resulted in a significant 243 increase in triglyceride levels in the females on the chow diet at 18 and 22 weeks of age. SIDT2 244 plays a crucial role in the uptake and intracellular transport of triglycerides and cholesterol 22, 23. 245 Previous studies16 have shown that the absence of Sidt2 causes increased serum triglycerides and 246 free fatty acids in mice. After analyzing the knockout results, we compared them with expression 247 data for the corresponding human genes. We discovered that while each gene knockout produced 248 some significantly different results in the female and male mice, FHS human gene expression data 249 did not reveal sex differences in lipid effects for any of the genes (P>0.05, data not shown). This 250 may be due to innate differences in lipid metabolism and regulation between the two species. 251 In contrast to similar studies that have relied on public databases to identify eQTLs25, our study 252 takes a direct and thorough approach, analyzing extensive phenotypic datasets in relation to gene 253 expression in human peripheral blood. This method, combined with mediation analysis, allowed 254 us not only to identify significant pathway enrichment from GWAS SNPs, but also to investigate 255 mechanisms underlying lipids traits. Our study is also unique in that we combined several 256 emerging techniques to identify and validate genes; these techniques included the use of functional 257 eQTL studies and the integration of PPI data. Functional studies, especially those involving the 258 use of networks to mirror biological systems, have shown great promise in the ability to identify 259 candidate genes for various diseases 26. In one study, Atanasovska et. al used a network- based 260 approach to prioritize genes for hundreds of cardiometabolic SNPs to identify disease-predisposing 261 genes27. The integration of PPI with eQTL analy ses also has great potential to offer mechanistic 262 explanations for diseases11, 14. 263 One limitation of our study is the use of whole blood for expression profiling. Although the lipid 264 traits were measured in blood, and as such, whole blood-derived eQTLs may be highly relevant to 265 the phenotypes we studied, in some cases, we observed opposite direction between MR predicted 266 effects and mouse knockout (ALDH2 on HDL). Another limitation is that our MR analyses could 267 not infer sex -specific causal effects because the underlying GWAS studies were conducted in 268 pooled-sex analyses, even though our mouse experiments revealed notable sex differences. 269 Differential lipid metabolic responses between male and female mice have also been previously 270 identified in other gene knockout studies28. 271 In summary, our network approach allowed us to identify novel candidate genes that may 272 contribute to CVD via lipid effects. As such, these genes represent attractive targets for the 273 treatment of dyslipidemia and the prevention of CVD. Looking forward, we plan to continue to 274 expand the CVD network and investigate additional causal genes. As we do so, we anticipate that 275 it will increasingly explain genes, pathways, and mechanisms underlying CVD and point toward 276 promising precision drug targets. Our integrativ e network also provides insights into how 277 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 7 dysfunction of novel CVD -associate genes is manifested at the molecular level 29. Approaches 278 similar to those used in this study may not only be expanded upon in relation to CVD but also be 279 used to investigate other diseases impacted by genetic and epigenetic factor s. To make future 280 studies more comprehensive and accurate, it will be important to expand eQTL databases to other 281 disease-related tissues and to include additional clinically important traits as well. 282

283 Samples and Phenotypes 284 In 1948, the FHS started recruiting participants (original cohort) from Framingham, MA to begin 285 the first round of extensive physical examinations and lifestyle surveys to investigate CVD and its 286 risk factors 30. In 1971 and 2002, the FHS recruited offspring (and their spouses) and adult 287 grandchildren of the original cohort participants into the offspring and third- generation cohorts, 288 respectively 31-33. Plasma total cholesterol, HDL cholesterol, triglycerides, and glucose were 289 measured in the morning after an eight- hour overnight fasting. Body mass index (BMI) was 290 defined as weight (kilograms) divided by height squared (meters2). Current smokers were defined 291 as those who smoked, on a verage, at least one cigarette per day during the year prior to the FHS 292 clinical assessment. Clinical characteristics of the study sample are summarized in Supplementary 293 Table 1. All participants from the FHS gave informed consent for participation in this study and 294 for the collection of plasma and DNA for analysis. The FHS study protocol was approved by 295 Boston Medical Center: Protocol ID: H -27984, Boston University Medical Center IRB (BUMC 296 IRB) Title: FRAMINGHAM HEART STUDY BIOMARKER PROJECT. 297 Genotype data: 5,568 SNPs that were associated with the three lipid traits – HDL cholesterol, LDL 298 cholesterol, and triglycerides – at p≤5×10-8 (in the GRASP database, downloaded in June 2016, 299 Supplementary Table 2) were curated and matched with the FHS 1000 Genomes Project imputed 300 genotype data 34. SNPs with imputed quality score (r 2) <0.3 and minor allele frequency (MAF) 301 <0.01 were excluded, resulting in 4173 genome-wide significant SNPs for eQTL analysis. 302 Gene expression: Whole blood was collected in PAXgene™ tubes (PreAnalytiX, Hombrechtikon, 303 Switzerland) and frozen at −80°C. RNA was extracted using a whole blood RNA System Kit 304 (Qiagen, Venlo, Netherlands) and mRNA expression profiling was assessed using the Affymetrix 305 Human Exon 1.0 ST GeneChip platform (Affymetrix Inc, Santa Clara, CA), which contains more 306 than 5.5 million probes targeting the expression of 17,873 genes. The Robust Multi-array Average 307 (RMA) package 35 was used to normalize the gene expression values and remove any technical or 308 spurious background variation. Linear regression models were used to adjust for technical 309 covariates (batch, first principal component, and residual mean of all probesets). 310 Identification of LIPIDS-associated eQTLs 311 The eQTLs were identified from FHS genotype data and whole blood gene expression as described 312 previously 13. eQTL analyses were conducted in two phases: 1) gene expression residuals were 313 generated after accounting for the effects of sex, age, platelet count, white blood cell count, and 314 imputed differential blood cell counts; analyses were performed using R version 3. 0.1 with a 315 mixed-effect modeling package that adjusted for familial relationships; and 2) underlying 316 confounding factors were accounted for by 20 Probabilistic Estimation of Expression Residuals 317 (PEERs)36 factors that were computed using the residualized expression data. The residualized 318 expression was fit to a linear model using the PEER factors along with sex, age, and effect allele 319 dosages. The algorithm was implemented with Graphical Processing Units (GPUs). cis- eQTLs 320 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 8 were defined as SNPs that reside within 1 Mb up or downstream of the transcription start site. 321 False discovery rate (FDR) computations for cis - and trans-eQTLs were computed separately. 322 SNPs at FDR <0.05 were considered statistically significant eQTLs. 323 Putative causality was first tested by mediation tests, in which the underlying mechanism for a 324 relationship between two variables is tested by introducing a third explanatory variable. The 325 analysis was conducted with the Mediation Package 37 in R where the “exposure” variable 326 represented a SNP, the “mediator” or “explanatory variable” represented gene expression, and the 327 “outcome” variable represented the phenotype. The mediation effect was measured on a scale from 328 0-100% where a 100% mediation effect indicated that the entire relationship between a SNP and 329 a phenotype (direct effect) was explained by changes in gene expression, i.e., the “mediator” or 330 the “explanatory variable.” Significant mediation effects were selected at a permutation p -value 331 <0.005 (based on 1000 permutations). 332 Significant mediation effects were further tested by Mendelian randomization (MR) 38 to 333 determine causal associations between gene expression and phenotype. MR uses common genetic 334 variants with well -understood effects on an exposure as instrumental variables (IV) to infer 335 causality of an exposure to an out come. This approach was applied to the aforementioned three 336 lipid traits (LDL cholesterol, HDL cholesterol, and triglycerides). Here, the sentinel cis-eQTL (top 337 eQTL located within 1 Mb of the tested gene), based on the lowest SNP -gene expression p-value 338 from the 1000G GWAS, was selected as the IV for its corresponding gene in MR analysis. Based 339 on the association between the sentinel cis -eQTL and the summary statistics from the three lipid 340 GWAS of 188,577 individuals obtained from the Global Lipids Genetics Consortium (GLGL) 15, 341 the MR analysis was performed using the MR Base package39 and the two-sample MR method 38. 342 Integration with PPI Networks 343 Binary PPIs were extracted from a systematically generated and literature-curated datasets, which 344 in total contain ~58,000 PPIs among 10,690 human proteins11, 14 . Putatively causal variants from 345 genotyping and gene expression analysis were integrated with PPI networks to assess the extent 346 of potential network perturbations associated with disruption of PPIs or altered expression 9. 347 Network perturbation information on missense alleles for various lipids -associated genes, which 348 assesses the degree to which a mutant protein exhibits an altered spectrum of PPIs relative to the 349 WT protein and all other mutants 11, 14, was also included. For proteins without direct (physical) 350 interactions, we use predicted protein -protein interactions from the STRING database to expand 351 the network40. 352 Identifying Candidate Genes 353 All candidate genes were assessed by each of the each of the following criteria: 1) Does the 354 candidate gene contain a SNP(s) associated with a lipid trait in GWAS (p<5x10 -8) and is the 355 GWAS SNP(s) is also an eQTL? 2) Is expression of the eQTL -associated gene also associated 356 with the same lipid trait? 3) Are genetic effects on the lipid trait mediated by the expression of the 357 eQTL-associated gene? 4) Does the gene test positive in MR (p<0.05)? 5) Have any interactions 358 with CVD related proteins? 359 We then ranked the genes based on how many of the criteria were met and further excluded any 360 genes if the knockout was known to be lethal based on existing literature. Three genes ABCA6, 361 ALDH2, and SIDT2 met more than three of these selection criteria and were selected for further 362 investigation. 363 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 9 Validation of Candidate Genes in Animal Models 364 To test the hypothesis that key genes identified in the LIPIDS networks can be validated in mouse 365 models for the corresponding traits, we obtained KO mouse strains for the three genes identified 366 as having causal effects on HDL cholesterol, LDL cholesterol, and triglycerides, respectively. 367 C57BL/6N-Abca6tm2a(KOMP)Wtsi/TcpRkorJ, B6Dnk;B6N -Aldh2tm1a(EUCOMM)Wtsi/IegRkorJ, and 368 B6;129S5-Sidt2tm1Lex/MmucdRkorJ KO mice were generated at The Jackson Laboratory using 369 sperm or embryos provided by the International Knockout Mouse Consortium and maintained on 370 a C57BL/6NJ genetic background. All animals were housed at The Jackson Laboratory, which is 371 approved by the A merican Association for Accreditation of Laboratory Animal Care. Animals 372 were kept on a 12-hour (6am-6pm) light/dark cycle with a room temperature between 68 and 72°F, 373 and either fed a chow diet (5K52, LabDiet) or a high-fat diet (TD.06414, Teklad Custom Diet). 374 Cohorts of 20 males and 20 females of both KO strains and C57BL/6NJ control mice were 375 raised on the chow diet. At 12 weeks of age, ten males and ten females from each cohort were 376 switched to the high-fat diet while the other animals continued on the chow diet. Plasma samples 377 were collected at 8, 14, 18, and 22 weeks after a four-hour fast. Total cholesterol, HDL cholesterol, 378 and triglycerides were measured using a Beckman Coulter Synchron CX 5 Delta autoanalyzer. 379 Non-HDL was calculated as the difference between total cholesterol and HDL cholesterol. 380 At 22 weeks of age, animals were euthanized and their livers were snap -frozen. RNA was 381 isolated from liver tissue using the MagMAX mirVana Total RNA Isolation Kit (ThermoFisher) 382 and the KingFisher Flex purification system (ThermoFisher). Tissues were lysed and homogenized 383 in TRIzol Reagent (ThermoFisher). After the addition of chloroform, the RNA-containing aqueous 384 layer was removed for RNA isolation according to the manufacturer’s protocol, beginning with 385 the RNA bead binding step. RNA concentration and quality were assessed using the Nanodrop 386 2000 spectrophotometer (Thermo Scientific) and the RNA 6000 Nano LabChip assay (Agilent 387 Technologies). 388 Libraries were prepared by the Genome Technologies core facility at The Jackson Laboratory 389 using the KAPA Stranded mRNA-Seq Kit (KAPA Biosystems), according to the manufacturer’s 390 instructions. Briefly, the protocol entailed isolation of polyA containing mRNA using oligo -dT 391 magnetic beads, RNA fragmentation, first and second-strand cDNA synthesis, ligation of Illumina-392 specific adapters containing a unique barcode sequence for each library, and polymerase chain 393 reaction (PCR) amplification. Libraries were checked for quality and concentration using the DNA 394 1000 LabChip assay (Agilent Technologies) and quantitative PCR (KAPA Biosystems), according 395 to the manufacturer’s instructions. A total of 36 liver samples were collected (3 male and 3 female 396 Aldh2 knockout mice, 3 male and 3 female Abca6 knockout mice, and 3 male and 3 female 397 wildtype mice, for each of the two diets). A library was made for each sample followed by pooling 398 of 6 libraries and samples were sequenced by the Genome Technologies core facility at The 399 Jackson Laboratory, 125 bp paired-end on the HiSeq 2500 system (Illumina, Inc.; San Diego, CA) 400 using the TruSeq SBS Kit v4 reagents (Illumina, Inc.) with a minimum of 40M reads per sample. 401 A repeated measures two -way ANOVA test with post -hoc pairwise comparison was used to test 402 for d ifferentially expressed genes between wild type and knock- out mouse. A significant 403 difference was determined at a false discovery rate <0.05 to account multiple testing. 404 Data availability 405 The genotype data, gene expression, phenotype data that support the findings from the FHS of this 406 study have been deposited in dbGaP (dbGaP Study Accession: phs000363.v16.p10). 407 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 10 408 409 Supplementary Materials 410 Fig. S1. The functional analyses of differentially expressed genes generated through IPA. 411 Fig.S2. Interaction networks of molecules based on known relationships in the QIAGEN 412 Knowledge 413 Table S1. Clinical Characteristics of the Framingham Heart Study Participants. 414 Table S2. SNPs from Lipids GWAS 415 Table S3. GWAS SNPs associated with gene expression(eQTLs) in FHS 416 Table S4. Mediation effect of SNPs on lipids through gene expression 417 Table S5. CVD proteins and their first-degree interactors from protein-protein interaction 418 network 419 Table S6. Interactions between CVD proteins 420 Table S7. Differentially expressed genes from RNA-Seq after ALDH2 and ABCA6 knockout 421 Table S8. Overlap and molecular functions of differentially expressed genes between Aldh2 and 422 Abca6 knockout mice 423 424

425 426 1. Pagidipati NJ and Gaziano TA. Estimating deaths from cardiovascular disease: a review of global 427 methodologies of mortality measurement. Circulation. 2013;127:749-56. 428 2. Ross S, D'Mello M, Anand SS, Eikelboom J, Consortium CAD, Stewart AF, Samani NJ, Roberts R 429 and Pare G. Effect of Bile Acid Sequestrants on the Risk of Cardiovascular Events: A Mendelian 430 Randomization Analysis. Circulation Cardiovascular genetics. 2015;8:618-27. 431 3. Nikpay M, Goel A, Won HH, Hall LM, Willenborg C, Kanoni S, Saleheen D, Kyriakou T, Nelson CP, 432 Hopewell JC, Webb TR, Zeng L, Dehghan A, Alver M, Armasu SM, Auro K, Bjonnes A, Chasman DI, Chen S, 433 Ford I, Franceschini N, Gieger C, Grace C, Gustafsson S, Huang J, Hwang SJ, Kim YK, Kleber ME, Lau KW, 434 Lu X, Lu Y, Lyytikainen LP, Mihailov E, Morrison AC, Pervjakova N, Qu L, Rose LM, Salfati E, Saxena R, 435 Scholz M, Smith AV, Tikkanen E, Uitterlinden A, Yang X, Zhang W, Zhao W, de Andrade M, de Vries PS, 436 van Zuydam NR, Anand SS, Bertram L, Beutner F, Dedoussis G, Frossard P, Gauguier D, Goodall AH, 437 Gottesman O, Haber M, Han BG, Huang J, Jalilzadeh S, Kessler T, Konig IR, Lannfelt L, Lieb W, Lind L, 438 Lindgren CM, Lokki ML, Magnusson PK, Mallick NH, Mehra N, Meitinger T, Memon FU, Morris AP, 439 Nieminen MS, Pedersen NL, Peters A, Rallidis LS, Rasheed A, Samuel M, Shah SH, Sinisalo J, Stirrups KE, 440 Trompet S, Wang L, Zaman KS, Ardissino D, Boerwinkle E, Borecki IB, Bottinger EP, Buring JE, Chambers 441 JC, Collins R, Cupples LA, Danesh J, Demuth I, Elosua R, Epstein SE, Esko T, Feitosa MF, Franco OH, 442 Franzosi MG, Granger CB, Gu D, Gudnason V, Hall AS, Hamsten A, Harris TB, Hazen SL, Hengstenberg C, 443 Hofman A, Ingelsson E, Iribarren C, Jukema JW, Karhunen PJ, Kim BJ, Kooner JS, Kullo IJ, Lehtimaki T, 444 Loos RJF, Melander O, Metspalu A, Marz W, Palmer CN, Perola M, Quertermous T, Rader DJ, Ridker PM, 445 Ripatti S, Roberts R, Salomaa V, Sanghera DK, Schwartz SM, Seedorf U, Stewart AF, Stott DJ, Thiery J, 446 Zalloua PA, O'Donnell CJ, Reilly MP, Assimes TL, Thompson JR, Erdmann J, Clarke R, Watkins H, 447 Kathiresan S, McPherson R, Deloukas P, Schunkert H, Samani NJ and Farrall M. A comprehensive 1,000 448 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 11 Genomes-based genome-wide association meta-analysis of coronary artery disease. Nat Genet. 449 2015;47:1121-1130. 450 4. Nelson CP, Hamby SE, Saleheen D, Hopewell JC, Zeng L, Assimes TL, Kanoni S, Willenborg C, 451 Burgess S, Amouyel P, Anand S, Blankenberg S, Boehm BO, Clarke RJ, Collins R, Dedoussis G, Farrall M, 452 Franks PW, Groop L, Hall AS, Hamsten A, Hengstenberg C, Hovingh GK, Ingelsson E, Kathiresan S, Kee F, 453 Konig IR, Kooner J, Lehtimaki T, Marz W, McPherson R, Metspalu A, Nieminen MS, O'Donnell CJ, Palmer 454 CN, Peters A, Perola M, Reilly MP, Ripatti S, Roberts R, Salomaa V, Shah SH, Schreiber S, Siegbahn A, 455 Thorsteinsdottir U, Veronesi G, Wareham N, Willer CJ, Zalloua PA, Erdmann J, Deloukas P, Watkins H, 456 Schunkert H, Danesh J, Thompson JR, Samani NJ and Consortium CACD. Genetically determined height 457 and coronary artery disease. N Engl J Med. 2015;372:1608-18. 458 5. Do R, Willer CJ, Schmidt EM, Sengupta S, Gao C, Peloso GM, Gustafsson S, Kanoni S, Ganna A, 459 Chen J, Buchkovich ML, Mora S, Beckmann JS, Bragg-Gresham JL, Chang HY, Demirkan A, Den Hertog 460 HM, Donnelly LA, Ehret GB, Esko T, Feitosa MF, Ferreira T, Fischer K, Fontanillas P, Fraser RM, Freitag DF, 461 Gurdasani D, Heikkila K, Hypponen E, Isaacs A, Jackson AU, Johansson A, Johnson T, Kaakinen M, 462 Kettunen J, Kleber ME, Li X, Luan J, Lyytikainen LP, Magnusson PK, Mangino M, Mihailov E, Montasser 463 ME, Muller-Nurasyid M, Nolte IM, O'Connell JR, Palmer CD, Perola M, Petersen AK, Sanna S, Saxena R, 464 Service SK, Shah S, Shungin D, Sidore C, Song C, Strawbridge RJ, Surakka I, Tanaka T, Teslovich TM, 465 Thorleifsson G, Van den Herik EG, Voight BF, Volcik KA, Waite LL, Wong A, Wu Y, Zhang W, Absher D, 466 Asiki G, Barroso I, Been LF, Bolton JL, Bonnycastle LL, Brambilla P, Burnett MS, Cesana G, Dimitriou M, 467 Doney AS, Doring A, Elliott P, Epstein SE, Eyjolfsson GI, Gigante B, Goodarzi MO, Grallert H, Gravito ML, 468 Groves CJ, Hallmans G, Hartikainen AL, Hayward C, Hernandez D, Hicks AA, Holm H, Hung YJ, Illig T, Jones 469 MR, Kaleebu P, Kastelein JJ, Khaw KT, Kim E, Klopp N, Komulainen P, Kumari M, Langenberg C, Lehtimaki 470 T, Lin SY, Lindstrom J, Loos RJ, Mach F, McArdle WL, Meisinger C, Mitchell BD, Muller G, Nagaraja R, 471 Narisu N, Nieminen TV, Nsubuga RN, Olafsson I, Ong KK, Palotie A, Papamarkou T, Pomilla C, Pouta A, 472 Rader DJ, Reilly MP, Ridker PM, Rivadeneira F, Rudan I, Ruokonen A, Samani N, Scharnagl H, Seeley J, 473 Silander K, Stancakova A, Stirrups K, Swift AJ, Tiret L, Uitterlinden AG, van Pelt LJ, Vedantam S, 474 Wainwright N, Wijmenga C, Wild SH, Willemsen G, Wilsgaard T, Wilson JF, Young EH, Zhao JH, Adair LS, 475 Arveiler D, Assimes TL, Bandinelli S, Bennett F, Bochud M, Boehm BO, Boomsma DI, Borecki IB, Bornstein 476 SR, Bovet P, Burnier M, Campbell H, Chakravarti A, Chambers JC, Chen YD, Collins FS, Cooper RS, Danesh 477 J, Dedoussis G, de Faire U, Feranil AB, Ferrieres J, Ferrucci L, Freimer NB, Gieger C, Groop LC, Gudnason 478 V, Gyllensten U, Hamsten A, Harris TB, Hingorani A, Hirschhorn JN, Hofman A, Hovingh GK, Hsiung CA, 479 Humphries SE, Hunt SC, Hveem K, Iribarren C, Jarvelin MR, Jula A, Kahonen M, Kaprio J, Kesaniemi A, 480 Kivimaki M, Kooner JS, Koudstaal PJ, Krauss RM, Kuh D, Kuusisto J, Kyvik KO, Laakso M, Lakka TA, Lind L, 481 Lindgren CM, Martin NG, Marz W, McCarthy MI, McKenzie CA, Meneton P, Metspalu A, Moilanen L, 482 Morris AD, Munroe PB, Njolstad I, Pedersen NL, Power C, Pramstaller PP, Price JF, Psaty BM, 483 Quertermous T, Rauramaa R, Saleheen D, Salomaa V, Sanghera DK, Saramies J, Schwarz PE, Sheu WH, 484 Shuldiner AR, Siegbahn A, Spector TD, Stefansson K, Strachan DP, Tayo BO, Tremoli E, Tuomilehto J, 485 Uusitupa M, van Duijn CM, Vollenweider P, Wallentin L, Wareham NJ, Whitfield JB, Wolffenbuttel BH, 486 Altshuler D, Ordovas JM, Boerwinkle E, Palmer CN, Thorsteinsdottir U, Chasman DI, Rotter JI, Franks PW, 487 Ripatti S, Cupples LA, Sandhu MS, Rich SS, Boehnke M, Deloukas P, Mohlke KL, Ingelsson E, Abecasis GR, 488 Daly MJ, Neale BM and Kathiresan S. Common variants associated with plasma triglycerides and risk for 489 coronary artery disease. Nature genetics. 2013;45:1345-52. 490 6. Rockman MV and Kruglyak L. Genetics of global gene expression. Nature reviews Genetics. 491 2006;7:862-72. 492 7. Weile J, Sun S, Cote AG, Knapp J, Verby M, Mellor JC, Wu Y, Pons C, Wong C, van Lieshout N, 493 Yang F, Tasan M, Tan G, Yang S, Fowler DM, Nussbaum R, Bloom JD, Vidal M, Hill DE, Aloy P and Roth FP. 494 A framework for exhaustively mapping functional missense variants. Mol Syst Biol. 2017;13:957. 495 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 12 8. Menche J, Sharma A, Kitsak M, Ghiassian SD, Vidal M, Loscalzo J and Barabasi AL. Disease 496 networks. Uncovering disease-disease relationships through the incomplete interactome. Science. 497 2015;347:1257601. 498 9. Sharma A, Halu A, Decano JL, Padi M, Liu YY, Prasad RB, Fadista J, Santolini M, Menche J, Weiss 499 ST, Vidal M, Silverman EK, Aikawa M, Barabasi AL, Groop L and Loscalzo J. Controllability in an islet 500 specific regulatory network identifies the transcriptional factor NFATC4, which regulates Type 2 Diabetes 501 associated genes. NPJ systems biology and applications. 2018;4:25. 502 10. Yao C, Chen BH, Joehanes R, Otlu B, Zhang X, Liu C, Huan T, Tastan O, Cupples LA, Meigs JB, Fox 503 CS, Freedman JE, Courchesne P, O'Donnell CJ, Munson PJ, Keles S and Levy D. Integromic analysis of 504 genetic variation and gene expression identifies networks for cardiovascular disease phenotypes. 505 Circulation. 2015;131:536-49. 506 11. Rolland T, Tasan M, Charloteaux B, Pevzner SJ, Zhong Q, Sahni N, Yi S, Lemmens I, Fontanillo C, 507 Mosca R, Kamburov A, Ghiassian SD, Yang X, Ghamsari L, Balcha D, Begg BE, Braun P, Brehme M, Broly 508 MP, Carvunis AR, Convery-Zupan D, Corominas R, Coulombe-Huntington J, Dann E, Dreze M, Dricot A, 509 Fan C, Franzosa E, Gebreab F, Gutierrez BJ, Hardy MF, Jin M, Kang S, Kiros R, Lin GN, Luck K, MacWilliams 510 A, Menche J, Murray RR, Palagi A, Poulin MM, Rambout X, Rasla J, Reichert P, Romero V, Ruyssinck E, 511 Sahalie JM, Scholz A, Shah AA, Sharma A, Shen Y, Spirohn K, Tam S, Tejeda AO, Trigg SA, Twizere JC, Vega 512 K, Walsh J, Cusick ME, Xia Y, Barabasi AL, Iakoucheva LM, Aloy P, De Las Rivas J, Tavernier J, Calderwood 513 MA, Hill DE, Hao T, Roth FP and Vidal M. A proteome-scale map of the human interactome network. Cell. 514 2014;159:1212-1226. 515 12. Dalal JJ, Padmanabhan TN, Jain P, Patil S, Vasnawala H and Gulati A. LIPITENSION: Interplay 516 between dyslipidemia and hypertension. Indian J Endocrinol Metab. 2012;16:240-5. 517 13. Joehanes R, Zhang X, Huan T, Yao C, Ying SX, Nguyen QT, Demirkale CY, Feolo ML, Sharopova NR, 518 Sturcke A, Schaffer AA, Heard-Costa N, Chen H, Liu PC, Wang R, Woodhouse KA, Tanriverdi K, Freedman 519 JE, Raghavachari N, Dupuis J, Johnson AD, O'Donnell CJ, Levy D and Munson PJ. Integrated genome-wide 520 analysis of expression quantitative trait loci aids interpretation of genomic association studies. Genome 521 biology. 2017;18:16. 522 14. Luck K, Kim DK, Lambourne L, Spirohn K, Begg BE, Bian W, Brignall R, Cafarelli T, Campos-Laborie 523 FJ, Charloteaux B, Choi D, Cote AG, Daley M, Deimling S, Desbuleux A, Dricot A, Gebbia M, Hardy MF, 524 Kishore N, Knapp JJ, Kovacs IA, Lemmens I, Mee MW, Mellor JC, Pollis C, Pons C, Richardson AD, 525 Schlabach S, Teeking B, Yadav A, Babor M, Balcha D, Basha O, Bowman-Colin C, Chin SF, Choi SG, 526 Colabella C, Coppin G, D'Amata C, De Ridder D, De Rouck S, Duran-Frigola M, Ennajdaoui H, Goebels F, 527 Goehring L, Gopal A, Haddad G, Hatchi E, Helmy M, Jacob Y, Kassa Y, Landini S, Li R, van Lieshout N, 528 MacWilliams A, Markey D, Paulson JN, Rangarajan S, Rasla J, Rayhan A, Rolland T, San-Miguel A, Shen Y, 529 Sheykhkarimli D, Sheynkman GM, Simonovsky E, Tasan M, Tejeda A, Tropepe V, Twizere JC, Wang Y, 530 Weatheritt RJ, Weile J, Xia Y, Yang X, Yeger-Lotem E, Zhong Q, Aloy P, Bader GD, De Las Rivas J, Gaudet S, 531 Hao T, Rak J, Tavernier J, Hill DE, Vidal M, Roth FP and Calderwood MA. A reference map of the human 532 binary protein interactome. Nature. 2020;580:402-408. 533 15. Willer CJ, Schmidt EM, Sengupta S, Peloso GM, Gustafsson S, Kanoni S, Ganna A, Chen J, 534 Buchkovich ML, Mora S, Beckmann JS, Bragg-Gresham JL, Chang HY, Demirkan A, Den Hertog HM, Do R, 535 Donnelly LA, Ehret GB, Esko T, Feitosa MF, Ferreira T, Fischer K, Fontanillas P, Fraser RM, Freitag DF, 536 Gurdasani D, Heikkila K, Hypponen E, Isaacs A, Jackson AU, Johansson A, Johnson T, Kaakinen M, 537 Kettunen J, Kleber ME, Li X, Luan J, Lyytikainen LP, Magnusson PKE, Mangino M, Mihailov E, Montasser 538 ME, Muller-Nurasyid M, Nolte IM, O'Connell JR, Palmer CD, Perola M, Petersen AK, Sanna S, Saxena R, 539 Service SK, Shah S, Shungin D, Sidore C, Song C, Strawbridge RJ, Surakka I, Tanaka T, Teslovich TM, 540 Thorleifsson G, Van den Herik EG, Voight BF, Volcik KA, Waite LL, Wong A, Wu Y, Zhang W, Absher D, 541 Asiki G, Barroso I, Been LF, Bolton JL, Bonnycastle LL, Brambilla P, Burnett MS, Cesana G, Dimitriou M, 542 Doney ASF, Doring A, Elliott P, Epstein SE, Ingi Eyjolfsson G, Gigante B, Goodarzi MO, Grallert H, Gravito 543 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 13 ML, Groves CJ, Hallmans G, Hartikainen AL, Hayward C, Hernandez D, Hicks AA, Holm H, Hung YJ, Illig T, 544 Jones MR, Kaleebu P, Kastelein JJP, Khaw KT, Kim E, Klopp N, Komulainen P, Kumari M, Langenberg C, 545 Lehtimaki T, Lin SY, Lindstrom J, Loos RJF, Mach F, McArdle WL, Meisinger C, Mitchell BD, Muller G, 546 Nagaraja R, Narisu N, Nieminen TVM, Nsubuga RN, Olafsson I, Ong KK, Palotie A, Papamarkou T, Pomilla 547 C, Pouta A, Rader DJ, Reilly MP, Ridker PM, Rivadeneira F, Rudan I, Ruokonen A, Samani N, Scharnagl H, 548 Seeley J, Silander K, Stancakova A, Stirrups K, Swift AJ, Tiret L, Uitterlinden AG, van Pelt LJ, Vedantam S, 549 Wainwright N, Wijmenga C, Wild SH, Willemsen G, Wilsgaard T, Wilson JF, Young EH, Zhao JH, Adair LS, 550 Arveiler D, Assimes TL, Bandinelli S, Bennett F, Bochud M, Boehm BO, Boomsma DI, Borecki IB, Bornstein 551 SR, Bovet P, Burnier M, Campbell H, Chakravarti A, Chambers JC, Chen YI, Collins FS, Cooper RS, Danesh 552 J, Dedoussis G, de Faire U, Feranil AB, Ferrieres J, Ferrucci L, Freimer NB, Gieger C, Groop LC, Gudnason 553 V, Gyllensten U, Hamsten A, Harris TB, Hingorani A, Hirschhorn JN, Hofman A, Hovingh GK, Hsiung CA, 554 Humphries SE, Hunt SC, Hveem K, Iribarren C, Jarvelin MR, Jula A, Kahonen M, Kaprio J, Kesaniemi A, 555 Kivimaki M, Kooner JS, Koudstaal PJ, Krauss RM, Kuh D, Kuusisto J, Kyvik KO, Laakso M, Lakka TA, Lind L, 556 Lindgren CM, Martin NG, Marz W, McCarthy MI, McKenzie CA, Meneton P, Metspalu A, Moilanen L, 557 Morris AD, Munroe PB, Njolstad I, Pedersen NL, Power C, Pramstaller PP, Price JF, Psaty BM, 558 Quertermous T, Rauramaa R, Saleheen D, Salomaa V, Sanghera DK, Saramies J, Schwarz PEH, Sheu WH, 559 Shuldiner AR, Siegbahn A, Spector TD, Stefansson K, Strachan DP, Tayo BO, Tremoli E, Tuomilehto J, 560 Uusitupa M, van Duijn CM, Vollenweider P, Wallentin L, Wareham NJ, Whitfield JB, Wolffenbuttel BHR, 561 Ordovas JM, Boerwinkle E, Palmer CNA, Thorsteinsdottir U, Chasman DI, Rotter JI, Franks PW, Ripatti S, 562 Cupples LA, Sandhu MS, Rich SS, Boehnke M, Deloukas P, Kathiresan S, Mohlke KL, Ingelsson E, Abecasis 563 GR and Global Lipids Genetics C. Discovery and refinement of loci associated with lipid levels. Nature 564 genetics. 2013;45:1274-1283. 565 16. Chen X, Gu X and Zhang H. Sidt2 regulates hepatocellular lipid metabolism through autophagy. J 566 Lipid Res. 2018;59:404-415. 567 17. Meng Y, Wang L and Ling L. Changes of lysosomal membrane permeabilization and lipid 568 metabolism in sidt2 deficient mice. Exp Ther Med. 2018;16:246-252. 569 18. Dean M and Annilo T. Evolution of the ATP-binding cassette (ABC) transporter superfamily in 570 vertebrates. Annu Rev Genomics Hum Genet. 2005;6:123-42. 571 19. Kaminski WE, Wenzel JJ, Piehler A, Langmann T and Schmitz G. ABCA6, a novel a subclass ABC 572 transporter. Biochemical and biophysical research communications. 2001;285:1295-301. 573 20. Pohl A, Devaux PF and Herrmann A. Function of prokaryotic and eukaryotic ABC proteins in lipid 574 transport. Biochimica et biophysica acta. 2005;1733:29-52. 575 21. Wada M, Daimon M, Emi M, Iijima H, Sato H, Koyano S, Tajima K, Kawanami T, Kurita K, Hunt SC, 576 Hopkins PN, Kubota I, Kawata S and Kato T. Genetic association between aldehyde dehydrogenase 2 577 (ALDH2) variation and high-density lipoprotein cholesterol (HDL-C) among non-drinkers in two large 578 population samples in Japan. J Atheroscler Thromb. 2008;15:179-84. 579 22. Gao J, Zhang Y, Yu C, Tan F and Wang L. Spontaneous nonalcoholic fatty liver disease and ER 580 stress in Sidt2 deficiency mice. Biochemical and biophysical research communications. 2016;476:326-581 332. 582 23. Mendez-Acevedo KM, Valdes VJ, Asanov A and Vaca L. A novel family of mammalian 583 transmembrane proteins involved in cholesterol transport. Scientific reports. 2017;7:7450. 584 24. Kramer A, Green J, Pollard J, Jr. and Tugendreich S. Causal analysis approaches in Ingenuity 585 Pathway Analysis. Bioinformatics. 2014;30:523-30. 586 25. Ongen H, Brown AA, Delaneau O, Panousis NI, Nica AC, Consortium GT and Dermitzakis ET. 587 Estimating the causal tissues for complex traits and diseases. Nature genetics. 2017;49:1676-1683. 588 26. Musunuru K, Strong A, Frank-Kamenetsky M, Lee NE, Ahfeldt T, Sachs KV, Li X, Li H, Kuperwasser 589 N, Ruda VM, Pirruccello JP, Muchmore B, Prokunina-Olsson L, Hall JL, Schadt EE, Morales CR, Lund-Katz 590 S, Phillips MC, Wong J, Cantley W, Racie T, Ejebe KG, Orho-Melander M, Melander O, Koteliansky V, 591 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 14 Fitzgerald K, Krauss RM, Cowan CA, Kathiresan S and Rader DJ. From noncoding variant to phenotype via 592 SORT1 at the 1p13 cholesterol locus. Nature. 2010;466:714-9. 593 27. Atanasovska B, Kumar V, Fu J, Wijmenga C and Hofker MH. GWAS as a Driver of Gene Discovery 594 in Cardiometabolic Diseases. Trends in endocrinology and metabolism: TEM. 2015;26:722-732. 595 28. Jones NS, Watson KQ and Rebeck GW. Metabolic Disturbances of a High-Fat Diet Are Dependent 596 on APOE Genotype and Sex. eNeuro. 2019;6. 597 29. Yadav A, Vidal M and Luck K. Precision medicine - networks to the rescue. Curr Opin Biotechnol. 598 2020;63:177-189. 599 30. Oppenheimer GM. Becoming the Framingham Study 1947-1950. Am J Public Health. 600 2005;95:602-10. 601 31. Dawber TR, Meadors GF and Moore FE, Jr. Epidemiological approaches to heart disease: the 602 Framingham Study. American journal of public health and the nation's health. 1951;41:279-81. 603 32. Feinleib M, Kannel WB, Garrison RJ, McNamara PM and Castelli WP. The Framingham Offspring 604 Study. Design and preliminary data. Preventive medicine. 1975;4:518-25. 605 33. Splansky GL, Corey D, Yang Q, Atwood LD, Cupples LA, Benjamin EJ, D'Agostino RB, Sr., Fox CS, 606 Larson MG, Murabito JM, O'Donnell CJ, Vasan RS, Wolf PA and Levy D. The Third Generation Cohort of 607 the National Heart, Lung, and Blood Institute's Framingham Heart Study: design, recruitment, and initial 608 examination. Am J Epidemiol. 2007;165:1328-35. 609 34. Benjamin AM, Suchindran S, Pearce K, Rowell J, Lien LF, Guyton JR and McCarthy JJ. Gene by sex 610 interaction for measures of obesity in the framingham heart study. J Obes. 2011;2011:329038. 611 35. Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B and Speed TP. Summaries of Affymetrix 612 GeneChip probe level data. Nucleic acids research. 2003;31:e15. 613 36. Stegle O, Parts L, Piipari M, Winn J and Durbin R. Using probabilistic estimation of expression 614 residuals (PEER) to obtain increased power and interpretability of gene expression analyses. Nat Protoc. 615 2012;7:500-7. 616 37. Imai K, Keele L and Tingley D. A general approach to causal mediation analysis. Psychological 617 methods. 2010;15:309-34. 618 38. Hartwig FP, Davies NM, Hemani G and Davey Smith G. Two-sample Mendelian randomization: 619 avoiding the downsides of a powerful, widely applicable but potentially fallible technique. International 620 journal of epidemiology. 2017. 621 39. Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V, Baird D, Laurin C, Burgess S, Bowden J, 622 Langdon R, Tan VY, Yarmolinsky J, Shihab HA, Timpson NJ, Evans DM, Relton C, Martin RM, Davey Smith 623 G, Gaunt TR and Haycock PC. The MR-Base platform supports systematic causal inference across the 624 human phenome. Elife. 2018;7. 625 40. Szklarczyk D, Gable AL, Lyon D, Junge A, Wyder S, Huerta-Cepas J, Simonovic M, Doncheva NT, 626 Morris JH, Bork P, Jensen LJ and Mering CV. STRING v11: protein-protein association networks with 627 increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic 628 Acids Res. 2019;47:D607-D613. 629 630 631 632 633 634 Acknowledgments: We gratefully acknowledge the contribution of the Reproductive Science, 635 Histopathology, and Genome Technologies Services at The Jackson Laboratory for expert 636 assistance with the work described in this publication. Funding: The Framingham Heart Study is 637 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 15 funded by National Institutes of Health contract N01-HC-25195. The analytical component for 638 this investigation was funded by the Division of Intramural Research, National Heart, Lung, and 639 Blood Institute, National Institutes of Health, Bethesda, MD (D. Levy, Principal Investigator). 640 The laboratory work of this project was funded by American Heart Association (AHA) 641 Cardiovascular Genome-Phenome Study (CVGPS) grant 15CVGPS23430000. Author 642 contributions: Conceptualization, C.Y. and D.L. Methodology and data analysis, C.Y. Mouse 643 knockout experiments, H.S.,Y.T.,R.K. Protein interaction network, T.H.,W.B.,D.E.H, M.V. 644 Writing, C.Y., R.K.,D.L .Competing interests: No competing interests to declare for all authors. 645 Data and materials availability: The SNP, gene expression and protein phenotype data that 646 Support the findings from the FHS of this study have been deposited in dbGaP (dbGaP 647 Study Accession: phs000363.v16.p10). 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 Figures: 672 Figure 1. Study Design 673 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 16 674 Schematic diagram of stepwise approach to identify target genes by utilizing human lipid GWAS 675 datasets, gene expression, protein interaction network and phenotype data , cross -reference with 676 mouse knockout database, prioritize genes, and validate candidate genes. 677 678 679 680 681 682 683 684 685 Figure 2. Comparison between Abca6 knockout (solid line) and wildtype (dotted line) mice 686 in females and males on regular chow and high -fat diet. Ten mice for each genotype and sex 687 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 17 per diet group. Asterisk indicates a significant difference (P<0.05) between knockout and wildtype 688 at the specific time point. 689 Female 690 691 692 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 18 Male 693 694 695 696 697 698 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 19 Figure 3. Comparison between Aldh2 knockout (solid line) and wildtype (dotted line) mice 699 in females and males on regular chow and high -fat diet. Ten mice for each genotype and sex 700 per diet group. Asterisk indicates a significant difference (P<0.05) between knockout and wildtype 701 at the specific time point. 702 Female 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 20 Male 721 722 723 724 725 726 727 728 729 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 21 Figure 4. Comparison between Sidt2 knockout (solid line) and wildtype (dotted line) mice in 730 males and females on regular chow and high-fat diet. Ten mice for each genotype and sex per 731 diet group. Asterisk indicates a significant difference (P<0.05) between knockout and wildtype at 732 the specific time point. 733 Female 734 735 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 22 Male 736 737 738 739 740 741 742 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 23 Table 1 . MR tests for candidate genes 743 Exposure Outcome Instrumental Beta SE P Value DOCK7 HDL cholesterol rs10889354 0.22 0.0821 0.0067 LDL cholesterol rs10889354 0.75 0.086 2.22E-18 Triglycerides rs10889354 1.15 0.078 1.56E-48 TAGLN HDL cholesterol rs3736120 -0.18 0.067 0.0091 LDL cholesterol rs3736120 -0.14 0.074 0.064 Triglycerides rs3736120 -0.34 0.066 2.90E-07 SIDT2 HDL cholesterol rs12420127 0.27 0.11 0.013 LDL cholesterol rs12420127 0.32 0.12 0.0065 Triglycerides rs12420127 0.32 0.11 0.0027 ALDH2 HDL cholesterol rs10744777 0.20 0.09 0.017 LDL cholesterol rs10744777 0.49 0.093 1.28E-07 Triglycerides rs10744777 -0.23 0.083 0.0056 SLC44A4 HDL cholesterol rs535586 0.079 0.055 0.15 LDL cholesterol rs535586 -0.23 0.059 0.00011 Triglycerides rs535586 -0.38 0.054 2.56E-12 ABCA6 HDL cholesterol rs918167 0.10 0.093 0.27 LDL cholesterol rs918167 0.061 0.10 0.55 Triglycerides rs918167 -0.16 0.091 0.079 ATG4C HDL cholesterol rs7540030 0.14 0.19 0.46 LDL cholesterol rs7540030 0.17 0.20 0.40 Triglycerides rs7540030 0.10 0.18 0.56 744 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint 24 Table 2. Criteria of candidate gene selection 745 746 Causal effects (MR P<0.05) Novelty Availability of knockout mice Associated with CVD- proteins in PPI network DOCK7 3 lipids traits No Yes Yes TAGLN 2 lipids traits No Yes Yes SIDT2 2 lipids traits No Yes Yes ALDH2 3 lipids traits Yes Yes Yes SLC44A4 2 lipids traits Yes No No ABCA6 0 lipids traits Yes Yes Yes ATG4C 0 lipids traits Yes Yes No 747 748 749 750 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprintthis version posted July 2, 2021. ; https://doi.org/10.1101/2021.07.01.21259304doi: medRxiv preprint