The Contribution of GWAS Loci in Familial Dyslipidemias

Familial combined hyperlipidemia (FCH) is a complex and common familial dyslipidemia characterized by elevated total cholesterol and/or triglyceride levels with over five-fold risk of coronary heart disease. The genetic architecture and contribution of rare Mendelian and common variants to FCH susceptibility is unknown. In 53 Finnish FCH families, we genotyped and imputed nine million variants in 715 family members with DNA available. We studied the enrichment of variants previously implicated with monogenic dyslipidemias and/or lipid levels in the general population by comparing allele frequencies between the FCH families and population samples. We also constructed weighted polygenic scores using 212 lipid-associated SNPs and estimated the relative contributions of Mendelian variants and polygenic scores to the risk of FCH in the families. We identified, across the whole allele frequency spectrum, an enrichment of variants known to elevate, and a deficiency of variants known to lower LDL-C and/or TG levels among both probands and affected FCH individuals. The score based on TG associated SNPs was particularly high among affected individuals compared to non-affected family members. Out of 234 affected FCH individuals across the families, seven (3%) carried Mendelian variants and 83 (35%) showed high accumulation of either known LDL-C or TG elevating variants by having either polygenic score over the 90th percentile in the population. The positive predictive value of high score was much higher for affected FCH individuals than for similar sporadic cases in the population. FCH is highly polygenic, supporting the hypothesis that variants across the whole allele frequency spectrum contribute to this complex familial trait. Polygenic SNP panels improve identification of individuals affected with FCH, but their clinical utility remains to be defined.

across the families, seven (3%) carried Mendelian variants and 83 (35%) showed high accumulation of either known LDL-C or TG elevating variants by having either polygenic score over the 90 th percentile in the population. The positive predictive value of high score was much higher for affected FCH individuals than for similar sporadic cases in the population. FCH is highly polygenic, supporting the hypothesis that variants across the whole allele frequency spectrum contribute to this complex familial trait. Polygenic SNP panels improve identification of individuals affected with FCH, but their clinical utility remains to be defined.

Author Summary
Familial combined hyperlipidemia (FCH) is a familial dyslipidemia and the most common familial risk factor for premature coronary heart disease. Its genetic architecture is poorly understood. Rare high-impact variants have been identified in some patients, but have not explained a substantial portion of the trait. FCH has previously been speculated to be a polygenic disorder, but genetic data supporting this hypothesis have so far been incomplete. We provide experimental evidence for the polygenicity and heterogeneity of FCH in a large set of affected families using comprehensive genome-wide variant data. Approximately a third of the affected FCH individuals in our sample had high polygenic burden, and only a minority carried high-impact variants identifiable by genotyping. We show that the polygenic burden of affected FCH family members is comparable to that observed in individuals with similar lipid phenotypes in the general population. Genetic variants identified in large-scale population studies can also underlie the typical phenotypes observed in complex familial diseases such as FCH. Advances in genetic diagnosis based on population samples may thus also benefit FCH families. Families without high polygenic burden are good candidates for sequencing studies to identify rare variants not observable with genotyping.

Introduction
Familial combined hyperlipidemia (FCH), classically defined by elevations in serum total cholesterol (TC), triglycerides (TG), or both, in two or more first degree relatives, displays a prevalence of greater than 1% in Western populations [1,2]. It is the most common familial risk factor for premature coronary heart disease (CHD), occurring in 11-14% of individuals with this condition, and raising by up to five-fold the CHD risk in first-and second-degree relatives of affected individuals [3,4]. In clinical practice, FCH is characterized by elevations of low-density lipoprotein cholesterol (LDL-C), TG, or both [5]. The phenotype within a family shows high inter-and intraindividual variability of lipid values (TG, LDL-C, high-density lipoprotein cholesterol [HDL-C], and apolipoprotein B) and therefore the diagnosis is commonly missed. Despite attempts to identify rare high-impact variants underlying FCH, no such variant has explained a substantial proportion of the trait [6]. Also, so far the genetics of FCH has not been addressed using high-density genotyping panels [7].
Here, we present a comprehensive evaluation of the genetic background of FCH using a dense-marker genotyping panel. We hypothesized that FCH risk could derive in part from a combination of common variants associated, in population cohorts, with LDL-C and/or TG as well as uncommon variants in genes that have been implicated in Mendelian lipid syndromes. To test our hypothesis we evaluated Finnish FCH families to determine if they demonstrated  [8][9][10][11], we measured, in these families, the relative contributions of LDL-C and TG associated polygenic scores to the lipid traits and the FCH phenotype.

Results
We evaluated the contribution of lipid-level associated genetic variation to FCH in 715 genotyped individuals (234 of whom were considered affected by FCH; Table 1, S1 Compared to 18,715 random Finnish population samples, alleles that elevated LDL-C and TG levels in the population were overall enriched and alleles that lowered LDL-C and TG levels were depleted in 234 affected FCH individuals (sign test p = 0.0025 for LDL-C and p = 0.0016 for TG). We included in this analysis all SNPs with at least one affected carrier (194 of the 212 SNPs previously associated with either LDL-C or TG in population genome-wide association (GWA) studies or with monogenic dyslipidemias). In total, 60 out of 95 LDL-C elevating SNPs were more common in EUFAM than in population samples, and 57 out of 99 of LDL-C lowering SNPs were less common. Similarly, 61 out of 96 TG elevating SNPs were enriched and 57  We then examined if high-impact variants implicated in monogenic dyslipidemias would be enriched and contribute to FCH. APOA5 rs3135506, which predisposes to hypertriglyceridemia in homozygous form [12], was 1.8-fold more frequent in the affected individuals of the FCH families than in the general population (minor allele frequency [MAF] in affected FCH individuals = 0.11, MAF in population = 0.062). In all FCH family members, there were 105 heterozygotes (43 of them affected) and six homozygotes (five of them affected). Homozygotes had a mean TG level (mmol/l) of 2.61, heterozygotes 1.97 and wild type carriers 1.55 (S4 Fig). APOE variations have a well-established role in dyslipidemias yet incompletely defined contribution to FCH. By genotyping rs7412 and imputing rs429358 we were able to determine the phenotyped apolipoprotein E isoform with a minimum accuracy of 95% ( Table 2). The APOE ε2ε2 haplotype that predisposes to type III hyperlipoproteinemia [13], was observed in three FCH family members (two of them affected), all with elevated cholesterol and TG concentrations in very low-density lipoprotein and intermediate-density lipoprotein fractions (Table 2). Its frequency was 0.0023 in the Finnish population, 0.0045 in all FCH families and 0.0085 among FCH affected individuals. LIPC rs28933094 predisposes to hepatic lipase deficiency, the cardiovascular effects of which are unclear but may be dependent on the underlying lipid phenotype [14,15]. It is 4.8-fold more frequent in Finns compared to other Europeans and additionally 2.6-fold more frequent in affected FCH individuals compared to the Finnish population (MAF in affected FCH individuals = 0.041, MAF in Finnish population = 0.016, MAF in Non-Finnish Europeans = 0.0033). Finally, it is of special interest whether the FCH families carry any of the classical FH variants. In the Finnish population there are five major extensively documented, FH-associated LDLR variants [16]. As the genotyping array does not capture these, we successfully imputed one out of the five variants (FH-Pogosta), but observed no carriers. In summary, variants that are strongly linked to known forms of dyslipidemia (in APOA5 and APOE), could at most explain only 7 (3.0%) of all 234 affected FCH individuals, thus being of minor importance in our FCH family sample.
We then estimated what proportion of the affected FCH individuals had a high cumulative burden of enriched LDL-C and TG elevating variants or carried high-impact variants. Out of the 234 affected FCH individuals, 83 individuals (35%) showed high accumulation of known LDL-C or TG elevating variants, having either one or both polygenic lipid scores over the 90 th percentile in the population (Table 3). Three additional FCH affected individuals carried Mendelian variants and did not have elevated polygenic scores. In 14 out of the 53 (25%) families, over half of the affected individuals had high polygenic scores or carried known Mendelian variants (Fig 3). In six out of the 53 (11%) families, all affected members had high polygenic scores or carried known Mendelian variants. This did not appear to be driven by differences in genetic correlation between affected individuals in the families (S5 Fig). Finally, we evaluated how well the high polygenic scores predicted FCH affectedness in the families or among hyperlipidemic individuals in the population (Table 4). Having a high polygenic score (! the 90 th population percentile) for either LDL-C or TG had a positive predictive value (PPV) of 0.45 in the FCH families and 0.23 in the general population. Negative predictive values (NPV) were 0.71 and 0.89, respectively.   In summary, we observed enrichment of both common and uncommon lipid-elevating variants in the FCH families. Only a minority (7 out of 234, 3%) of affected individuals were carriers of high-impact variants of Mendelian dyslipidemias. The polygenic lipid scores contribute to FCH in over one third of 234 FCH subjects in these families, with considerable heterogeneity between families. In more than half of the families we did not observe either an increased load of common variants or carriers of high-impact variants.

Discussion
Our results demonstrate an enrichment of many known lipid-level elevating variants in FCH families. This enrichment was observed for both uncommon and common variants known to affect either LDL-C or TG levels in populations. When the known variants were combined into polygenic lipid scores for each individual, 17% and 25% of the affected individuals had polygenic scores that were higher than the 90 th percentile of the population for LDL-C and TG, respectively. In 13 families three or more affected individuals had high polygenic score and in 22 families none of the affected individuals had polygenic score above the 90 th percentile of the population.
Our results allow us to draw several conclusions about the genetic background of FCH. First, variants across the whole frequency spectrum were enriched and contribute to the high LDL-C and/or TG levels in the affected FCH individuals. This emphasizes the polygenic nature of FCH and is in line with previous results with familial hypercholesterolemia demonstrating that this less prevalent syndrome has a polygenic component [9]. Genes in which we observed enriched variants include APOE, LIPC and APOA5, whose role in dyslipidemias is already established, but also include several genes whose function in lipid metabolism is not yet clear (e.g. UBR1, MTHFD2L and the PIGV-NR0B2 region).
Second, the majority of the enriched variants were originally identified in random population samples. This finding confirms that variants identified in population screens play a considerable role also in the complex familial disease FCH, and highlights the potential of using variants identified in population-based GWA studies to characterize familial dyslipidemia cases genetically.
Third, the observed enrichment was stronger for TG loci than for LDL-C loci. Many of the enriched TG SNPs were located in genes known to contribute to hypertriglyceridemia, such as the APOA1-C3-A4-A5 cluster, showing the central role of genetically driven TG in FCH [7,17].
Fourth, over a third of the affected FCH individuals had a high load (polygenic lipid scores over the 90 th percentile of the population) of either known LDL-C or TG associated variants. This proportion with high polygenic score was comparable to the population samples with similar levels of high LDL-C or TG in the population. This observation suggests that the role of the known variants is similar in familial hyperlipidemias and randomly ascertained high LDL-C and/or TG individuals. However, the accumulation of LDL-C or TG-elevating alleles in families was highlighted by the prediction analysis results. High load of known LDL-C or TG associated variants predicted affectedness in FCH families almost two times better than comparable hyperlipidemia in the population. There were large differences among families in the proportion of affected individuals with high polygenic scores. While in nine families more than two thirds of the affected individuals had high polygenic score, in over a third of the families, none of the cases had high score and the underlying genetic architecture remains unexplained in the majority of families. As genotyping and imputation allow for the evaluation of only a portion of catalogued high-impact variants, future sequencing studies might pinpoint rare causal variants in these families.
Our study supports the hypothesis that FCH is a genetically heterogeneous disease, reflected by its heterogeneous lipid phenotype [18][19][20]. large number of LDL-C associated variants are mainly responsible for the elevation of LDL-C in the FCH lipid profile. In contrast, a handful of low-to moderate-impact TG variants drive the elevation of triglycerides. Our study provides support for the polygenic rather than monogenic nature of FCH, and highlights the central role of genetically driven TG in FCH.

Ethics statement
Written informed consent was obtained from all study participants. All samples were collected in accordance with the Helsinki declaration and study protocols were approved by the ethics committees of the participating centers (The Hospital District of Helsinki and Uusimaa Coordinating Ethics Committee, approval number 184/13/03/00/12).

Subjects and measurements
As part of the European Multicenter Study on Familial Dyslipidemias in Patients with Premature Coronary Heart Disease (EUFAM), the Finnish FCH families were identified from patients admitted to university hospitals with a diagnosis of premature CHD who demonstrated levels of TC, TG, or both that were ! 90 th Finnish age-and sex-specific population percentile (S4 Table) [21,22]. Families that contained at least one other first-degree relative affected with hyperlipidemia (according to the same lipid-level criteria as used for the probands) were included in the study. Additionally, in the included families, at least one of the affected individuals had high TG. All family members with hyperlipidemia were then considered affected by FCH. Probands with a diagnosis of FH (screened with a functional LDL receptor test), and any subjects with diabetes or other chronic diseases were considered unaffected and did not contribute to establishing the family's FCH status (S1 Text).
For analyses of continuous lipid traits, individuals using lipid-lowering or estrogen medication at the time of sampling were excluded. Samples from the Finnish National FINRISK study were used as a Finnish population-specific comparison group, and individuals with known diabetes or cancer were excluded from the analyses (S1 Text). For the FCH families, venous blood samples were obtained after an overnight fast. FINRISK participants were advised to fast for four hours before the examination and avoid heavy meals earlier during the day. For both the EUFAM and FINRISK samples, circulating biochemical markers were measured from the venous blood samples using standard methods, as described in the S1 Text.

Genotyping and imputation
All FCH individuals with DNA available (715 out of 1161 from all 53 FCH families) and the FINRISK samples (n = 20,626) were genotyped with the HumanCoreExome BeadChip (Illumina Inc., San Diego, CA, USA) using standard methods (S1 Text). Additionally, over nine million variants were imputed using a combined reference panel of 1000 Genomes Phase I integrated haplotypes and 1943 Finnish genomes (S1 Text).

Polygenic lipid score calculation
To construct a polygenic lipid score, we catalogued and used 212 SNPs, representing either the lead SNPs for genome-wide significant associations to LDL-C or TG from GWA studies of these traits (171 SNPs), or SNPs catalogued in the Online Mendelian Inheritance in Man (OMIM) database as located in genes implicated in primary and secondary monogenic dyslipidemic syndromes (44 SNPs, four of which overlap with the GWA lead-SNPs, S2 Table) [23][24][25][26]. The SNPs were assigned to LDL-C and TG scores based on their previously reported associations (S1 Text). The scores were calculated as the sum of the risk alleles weighted by their effect estimates drawn from a multiple linear model estimating all SNP effects on the trait (LDL-C or TG) at the same time in the FINRISK samples (S1 Text). The FINRISK samples used to estimate the weights were independent from the FCH samples.
We applied linear mixed models to test for differences in metabolic and clinical characteristics between FCH affected and unaffected individuals (S1 Text). An empirical genetic correlation matrix between individuals was included as the covariance structure of a random effect. Linear mixed models were applied with MMM (version 1.01) [27], and the other statistical analyses were performed using R (version 3.2.1) [28].
Supporting Information S1 Text. Supplementary Materials and Methods. The recruitment, assessment, and genotyping of study subjects, the calculation of polygenic lipid scores, and the statistical analyses   Table. LDL-C or TG associated variants included in the study. NFE, non-Finnish Europeans; FINRISK, The National FINRISK Study; FCH, familial combined hyperlipidemia. Ã Effect allele. † Global minor allele. ‡ MAF in non-Finnish Europeans calculated from the public ExAC data set (version 0.3) for coding SNPs (those present in the ExAC data set), and from the public 1000 Genomes Phase 3 data set for non-coding SNPs (S1 Text). § MAF in the FINRISK cohort after excluding those with known diabetes or cancer. k Common: MAF > 5%, low-frequency: 0.5% < MAF 5%, rare: MAF 0.5%. ¶ SNPs were selected based on OMIM entries for genes implicated in monogenic dyslipidemia, and recent genome-wide association studies (S1 Text). ♦ Effect allele weight not estimated due to only one copy of minor allele present in FINRISK. (PDF) S3 Table. Enrichment ratios and observed p-values for all score SNPs with at least one affected carrier. Under the null hypothesis of no enrichment, we established the null distribution for enrichment testing by calculating the enrichment statistic (enrichment ratio) for all variants with effect allele (minor allele in the FINRISK cohort) frequency > 0.001% across the genome excluding the loci of the 212 SNPs. Enrichment ratios were calculated for each of the 212 SNPs if at least one carrier was identified (194 SNPs for affected individuals, 191 SNPs for probands), and one-tailed p-values were estimated from the null distribution. Enrichment ratios are presented for the affected individuals. The carrier counts were calculated from all genotyped FCH family members who passed exclusion criteria and did not have diabetes or other relevant confounders (n = 661), and did not account for non-independence within family. The + and-symbols denote elevating and lowering effects on the lipid in question, respectively. FCH, familial combined hyperlipidemia; FINRISK, The National FINRISK Study. J.A.K, J. A. Kuivenhoven; C.J.W., C. J. Willer; T.M.T., T. M. Teslovich; I.S., I. Surakka (S1 Text).
† Effect allele frequencies in non-Finnish Europeans were calculated from the public ExAC data set (version 0.3) for coding SNPs (those present in the ExAC data set), and from the public 1000 Genomes Phase 3 data set for non-coding SNPs (S1 Text). ‡ Effect allele frequencies in the FINRISK cohort after excluding those with known diabetes or cancer. § Enrichment ratio is the ratio of the effect allele frequencies in affected individuals (n = 234) to the Finnish population (n = 18,715). (PDF) S4 Table. The 90 th age-and sex-specific Finnish population percentiles for total cholesterol and triglycerides. In the EUFAM study, dyslipidemia was established based on levels of total cholesterol, triglycerides, or both that were ! 90th Finnish age-and sex-specific population percentile. The population percentiles were derived from FINMONICA, a large population survey performed in 1992. The percentile estimates and a description of the polynomial regression analyses used to establish them have been reported in detail previously [21]. (PDF)