Performing highly parallelized and reproducible GWAS analysis on biobank-scale data.

Genome-wide association studies (GWAS) are transforming genetic research and enable the detection of novel genotype-phenotype relationships. In the last two decades, over 60 000 genetic associations across thousands of traits have been discovered using a GWAS approach. Due to increasing sample sizes, researchers are increasingly faced with computational challenges. A reproducible, modular and extensible pipeline with a focus on parallelization is essential to simplify data analysis and to allow researchers to devote their time to other essential tasks. Here we present nf-gwas, a Nextflow pipeline to run biobank-scale GWAS analysis. The pipeline automatically performs numerous pre- and post-processing steps, integrates regression modeling from the REGENIE package and supports single-variant, gene-based and interaction testing. It includes an extensive reporting functionality that allows to inspect thousands of phenotypes and navigate interactive Manhattan plots directly in the web browser. The pipeline is tested using the unit-style testing framework nf-test, a crucial requirement in clinical and pharmaceutical settings. Furthermore, we validated the pipeline against published GWAS datasets and benchmarked the pipeline on high-performance computing and cloud infrastructures to provide cost estimations to end users. nf-gwas is a highly parallelized, scalable and well-tested Nextflow pipeline to perform GWAS analysis in a reproducible manner.

Meta-GWAS on PCSK9 concentrations reveals associations of novel loci outside the PCSK9 locus in White populations.

BACKGROUND AND AIMS: Proprotein convertase subtilisin/kexin type 9 (PCSK9) is a key regulator of lipid homeostasis. A few earlier genome-wide association studies (GWAS) investigated genetic variants associated with circulating PCSK9 concentrations. However, uncertainty remains about some of the genetic loci discovered beyond the PCSK9 locus. By conducting the largest PCSK9 meta-analysis of GWAS (meta-GWAS) so far, we aimed to identify novel loci and validate the previously reported loci that regulate PCSK9 concentrations. METHODS: We performed GWAS for PCSK9 concentrations in two large cohorts (GCKD (n = 4,963) and KORA F3 (n = 2,895)). These were meta-analyzed with previously published data encompassing together 20,579 individuals. We further conducted a second meta-analysis in statin-naïve individuals (n = 15,390). A genetic risk score (GRS) was constructed on PCSK9-increasing SNPs and assessed its impact on the risk for coronary artery disease (CAD) in 394,943 statin-naïve participants (17,077 with events) of the UK Biobank by performing CAD-free survival analysis. RESULTS: Nine loci were genome-wide significantly associated with PCSK9 concentrations. These included the previously described PCSK9, APOB, KCNA1/KCNA5, and TM6SF2/SUGP1 loci. All imputed SNPs in the PCSK9 locus account for ∼15% of variance of PCSK9 concentrations. We further identified FADS2 as a novel locus that was also found in statin-naïve participants. All imputed SNPs within the FADS2 locus explain ∼1.2% of variance of PCSK9 concentrations. Additionally, four further loci (a region on chromosome 5, SDK1, SPATA16 and HPR) were genome-wide significant in either the main model or the statin-naïve subset. The linear increase in a PCSK9 genetic risk score was associated with 1.41-fold (95%CI 1.16-1.72, p < 0.001) higher risk for incident CAD. CONCLUSIONS: We identified five novel loci (FADS2, SPATA16, SDK1, HPR and a region on chromosome 5) for PCSK9 concentrations that would require further research. Additionally, we confirm the genome-wide significant loci that were previously detected.

KLKB1 and CLSTN2 are associated with HDL-mediated cholesterol efflux capacity in a genome-wide association study.

BACKGROUND AND AIMS: HDL-mediated cholesterol efflux capacity (CEC) may protect from cardiovascular disease. Thus, we aimed to identify its genetic and non-genetic determinants. METHODS: We measured CEC to 2% apolipoprotein B-depleted serum using BODIPY-cholesterol and cAMP-stimulated J774A.1 macrophages using serum samples from 4,981 participants in the German Chronic Kidney Disease (GCKD) study. Variance of CEC explained by clinical and biochemical parameters in a multivariable linear regression model was calculated by proportional marginal variance decomposition. A genome-wide association study with 7,746,917 variants was performed based on an additive genetic model. The main model was adjusted for age, sex and principal components 1-10. Further models were selected for sensitivity analysis and to reduce residual variance by known CEC pathways. RESULTS: Variables that explained 1% and more of the variance of CEC were concentrations of triglycerides (12.9%), HDL-cholesterol (11.8%), LDL-cholesterol (3.0%), apolipoprotein A-IV (2.8%), PCSK9 (1.0%), and eGFR (1.0%). The KLKB1 (chr4) and APOE/C1 (chr19) loci were genome-wide significantly (p < 5x10-8) associated with CEC in our main model (p = 8.8x10-10 and p = 3.3x10-10, respectively). KLKB1 remained significantly associated after additional adjustment for either kidney parameters, HDL-cholesterol, triglycerides or apolipoprotein A-IV concentrations, while the APOE/C1 locus was not significantly associated anymore after adjustment for triglycerides. Adjustment for triglycerides also revealed an association with the CLSTN2 locus (chr3; p = 6.0x10-9). CONCLUSIONS: We identified HDL-cholesterol and triglycerides as the main determinants of CEC. Furthermore, we newly found a significant association of CEC with the KLKB1 and the CLSTN2 locus and confirmed the association with the APOE/C1 locus, likely mediated by triglycerides.

The kringle IV type 2 domain variant 4925G>A causes the elusive association signal of the LPA pentanucleotide repeat.

Lipoprotein(a) [Lp(a)] concentrations are regulated by the LPA gene mainly via the large kringle IV-type 2 (KIV-2) copy number variation and multiple causal variants. Early studies suggested an effect of long pentanucleotide repeat (PNR) alleles (10 and 11 repeats, PNR10 and PNR11) in the LPA promoter on gene transcription and found an association with lower Lp(a). Subsequent in vitro studies showed no effects on mRNA transcription, but the association with strongly decreased Lp(a) remained consistent. We investigated the isolated and combined effect of PNR10, PNR11, and the frequent splice site variant KIV-2 4925G>A on Lp(a) concentrations in the Cooperative Health Research in the Region of Augsburg F4 study by multiple quantile regression in single-SNP and joint models. Data on Lp(a), apolipoprotein(a) Western blot isoforms, and variant genotypes were available for 2,858 individuals. We found a considerable linkage disequilibrium between KIV-2 4925G>A and the alleles PNR10 and PNR11. In single-variant analysis adjusted for age, sex, and the shorter apo(a) isoform, we determined that both PNR alleles were associated with a highly significant Lp(a) decrease (PNR10: ß = -14.43 mg/dl, 95% CI: -15.84, -13.02, P = 3.33e-84; PNR11: ß = -17.21 mg/dl, 95% CI: -20.19, -14.23, P = 4.01e-29). However, a joint model, adjusting the PNR alleles additionally for 4925G>A, abolished the effect on Lp(a) (PNR10: ß = +0.44 mg/dl, 95% CI: -1.73, 2.60, P = 0.69; PNR11: ß = -1.52 mg/dl, 95% CI: -6.05, 3.00, P = 0.51). Collectively, we conclude that the previously reported Lp(a) decrease observed in pentanucleotide alleles PNR10 or PNR11 carriers results from a linkage disequilibrium with the frequent splicing mutation KIV-2 4925G>A.

The effect of LPA Thr3888Pro on lipoprotein(a) and coronary artery disease is modified by the LPA KIV-2 variant 4925G>A.

BACKGROUND AND AIMS: High lipoprotein(a) [Lp(a)] concentrations are associated with increased coronary artery disease (CAD) risk. Lp(a) is regulated mainly genetically by the LPA gene but involved genetic variants have not been fully elucidated. Improved understanding of the entanglements of genetic Lp(a) regulation may enhance genetic prediction of Lp(a) and CAD risk. We investigated an interaction between the well-known LPA missense SNP rs41272110 (known as Thr3888Pro) and the frequent LPA splicing mutation KIV-2 4925G>A. METHODS: Effects on Lp(a) concentrations were investigated by multiple quantile regression in the German Chronic Kidney Disease (GCKD) study, KORA-F3 and KORA-F4 (ntotal = 10,405) as well as in the UK Biobank (UKB) 200k exome dataset (n = 173,878). The impact of the interaction on CAD risk was assessed by survival analysis in UKB. RESULTS: We observed a significant SNP-SNP interaction in all studies (p = 1.26e-05 to 3.03e-04). In quantile regression analysis, rs41272110 as a predictor shows no impact on Lp(a) (ß = -0.06 [-0.79; 0.68], p = 0.879), but in a joint model including both SNPs as predictors, rs41272110 is associated with markedly higher Lp(a) (ß = +9.40 mg/dL [6.45; 12.34], p = 4.07e-10). Similarly, rs41272110 shows no effect on CAD in UKB (HR = 1.01 [0.97; 1.04], p = 0.731), while rs41272110 carriers not carrying 4925G>A show an increased CAD risk (HR = 1.10 [1.04; 1.16], p = 6.9e-04). This group corresponds to 4% of the population. Adjustment for apolipoprotein(a) isoforms further modified the effect estimates markedly. CONCLUSIONS: This work emphasizes the complexity of the genetic regulation of Lp(a) and the importance to account for genetic subgroups in Lp(a) association studies and when interpreting genetic cardiovascular risk profiles.

Lipoprotein(a) and SARS-CoV-2 infections: Susceptibility to infections, ischemic heart disease and thromboembolic events.

BACKGROUND: Comorbidities including ischemic heart disease (IHD) worsen outcomes after SARS-CoV-2 infections. High lipoprotein(a) [Lp(a)] concentrations are a strong risk factor for IHD and possibly for thromboembolic events. We therefore evaluated whether SARS-CoV-2 infections modify the risk of high Lp(a) concentrations for IHD or thromboembolic events during the first 8.5 months follow-up of the pandemic. METHOD: Cohort study using data from the UK Biobank during the SARS-CoV-2 pandemic. Baseline Lp(a) was compared between SARS-CoV-2 positive patients and the population controls. RESULTS: SARS-CoV-2 positive patients had Lp(a) concentrations similar to the population controls. The risk for IHD increased with higher Lp(a) concentrations in both, the population controls (n = 435,104) and SARS-CoV-2 positive patients (n = 6937). The causality of the findings was supported by a genetic risk score for Lp(a). A SARS-CoV-2 infection modified the association with a steeper increase in risk for infected patients (interaction p-value = 0.03). Although SARS-CoV-2 positive patients had a five-times higher frequency of thromboembolic events compared to the population controls (1.53% vs. 0.31%), the risk was not influenced by Lp(a). CONCLUSIONS: SARS-CoV-2 infections enforce the association between high Lp(a) and IHD but the risk for thromboembolic events is not influenced by Lp(a).

Frequent LPA KIV-2 Variants Lower Lipoprotein(a) Concentrations and Protect Against Coronary Artery Disease.

BACKGROUND: Lipoprotein(a) (Lp(a)) concentrations are a major independent risk factor for coronary artery disease (CAD) and are mainly determined by variation in LPA. Up to 70% of the LPA coding sequence is located in the hypervariable kringle IV type 2 (KIV-2) region. It is hardly accessible by conventional technologies, but may contain functional variants. OBJECTIVES: This study sought to investigate the new, very frequent splicing variant KIV-2 4733G>A on Lp(a) and CAD. METHODS: We genotyped 4733G>A in the GCKD (German Chronic Kidney Disease) study (n = 4,673) by allele-specific polymerase chain reaction, performed minigene assays, identified proxy single nucleotide polymorphisms and used them to characterize its effect on CAD by survival analysis in UK Biobank (n = 440,234). Frequencies in ethnic groups were assessed in the 1000 Genomes Project. RESULTS: The 4733G>A variant (38.2% carrier frequency) was found in most isoform sizes. It reduces allelic expression without abolishing protein production, lowers Lp(a) by 13.6 mg/dL (95% CI: 12.5-14.7; P < 0.0001) and is the strongest variance-explaining factor after the smaller isoform. Splicing of minigenes was modified. Compound heterozygosity (4.6% of the population) for 4733G>A and 4925G>A, another KIV-2 splicing mutation, reduces Lp(a) by 31.8 mg/dL and most importantly narrows the interquartile range by 9-fold (from 42.1 to 4.6 mg/dL) when compared to the wild type. In UK Biobank 4733G>A alone and compound heterozygosity with 4925G>A reduced HR for CAD by 9% (95% CI: 7%-11%) and 12% (95% CI: 7%-16%) (both P < 0.001). Frequencies in ethnicities differ notably. CONCLUSIONS: Functional variants in the previously inaccessible LPA KIV-2 region cooperate in determining Lp(a) variance and CAD risk. Even a moderate but lifelong genetic Lp(a) reduction translates to a noticeable CAD risk reduction.

Investigation of a nonsense mutation located in the complex KIV-2 copy number variation region of apolipoprotein(a) in 10,910 individuals.

BACKGROUND: The concentrations of the highly atherogenic lipoprotein(a) [Lp(a)] are mainly genetically determined by the LPA gene locus. However, up to 70% of the coding sequence is located in the complex so-called kringle IV type 2 (KIV-2) copy number variation, a region hardly accessible by common genotyping and sequencing technologies. Despite its size, little is known about genetic variants in this complex region. The R21X variant is a functional variant located in this region, but it has never been analyzed in large cohorts. METHODS: We typed R21X in 10,910 individuals from three European populations using a newly developed high-throughput allele-specific qPCR assay. R21X allelic location was determined by separating the LPA alleles using pulsed-field gel electrophoresis (PFGE) and typing them separately. Using GWAS data, we identified a proxy SNP located outside of the KIV-2. Linkage disequilibrium was determined both statistically and by long-range haplotyping using PFGE. Worldwide frequencies were determined by reanalyzing the sequencing data of the 1000 Genomes Project with a dedicated pipeline. RESULTS: R21X carriers (frequency 0.016-0.021) showed significantly lower mean Lp(a) concentrations (- 11.7 mg/dL [- 15.5; - 7.82], p = 3.39e-32). The variant is located mostly on medium-sized LPA alleles. In the 1000 Genome data, R21X mostly occurs in Europeans and South Asians, is absent in Africans, and shows varying frequencies in South American populations (0 to 0.022). Of note, the best proxy SNP was another LPA null mutation (rs41272114, D' = 0.958, R2 = 0.281). D' was very high in all 1000G populations (0.986-0.996), although rs41272114 frequency varies considerably (0-0.182). Co-localization of both null mutations on the same allele was confirmed by PFGE-based long-range haplotyping. CONCLUSIONS: We performed the largest epidemiological study on an LPA KIV-2 variant so far, showing that it is possible to assess LPA KIV-2 mutations on a large scale. Surprisingly, in all analyzed populations, R21X was located on the same haplotype as the splice mutation rs41272114, creating "double-null" LPA alleles. Despite being a nonsense variant, the R21X status does not provide additional information beyond the rs41272114 genotype. This has important implications for studies using LPA loss-of-function mutations as genetic instruments and emphasizes the complexity of LPA genetics.

Correction to: New clinical manifestation of COVID-19 related Guillain-Barrè syndrome highly responsive to intravenous immunoglobulins: two Italian cases.

The above article was published online with inverted given and family names. The correct presentation has been corrected above.