Association of lipid-lowering drugs with COVID-19 outcomes from a Mendelian randomization study
Abstract
Background:
Lipid metabolism plays an important role in viral infections. We aimed to assess the causal effect of lipid-lowering drugs (HMGCR inhibitiors, PCSK9 inhibitiors, and NPC1L1 inhibitior) on COVID-19 outcomes using two-sample Mendelian randomization (MR) study.
Methods:
We used two kinds of genetic instruments to proxy the exposure of lipid-lowering drugs, including expression quantitative trait loci of drugs target genes, and genetic variants within or nearby drugs target genes associated with low-density lipoprotein (LDL cholesterol from genome-wide association study). Summary-data-based MR (SMR) and inverse-variance-weighted MR (IVW-MR) were used to calculate the effect estimates.
Results:
SMR analysis found that a higher expression of HMGCR was associated with a higher risk of COVID-19 hospitalization (odds ratio [OR] = 1.38, 95% confidence interval [CI] = 1.06–1.81). Similarly, IVW-MR analysis observed a positive association between HMGCR-mediated LDL cholesterol and COVID-19 hospitalization (OR = 1.32, 95% CI = 1.00–1.74). No consistent evidence from both analyses was found for other associations.
Conclusions:
This two-sample MR study suggested a potential causal relationship between HMGCR inhibition and the reduced risk of COVID-19 hospitalization.
Funding:
Start-up Fund for high-level talents of Fujian Medical University.
Editor's evaluation
There are mixed results from studies of COVID-19 outcomes in patients treated with statins and there are multiple confounders. The authors use two Mendelian randomization methods to explore the association between HMGCoA reductase inhibitors (statins) and other lipid lowering drugs and outcomes and find that increased expression of HMGCoA reductase and HMGCoA reductase mediated LDL cholesterol increase hospitalization risk. This makes it possible but does not prove that statins could improve outcomes which will be of broad interest.
https://doi.org/10.7554/eLife.73873.sa0eLife digest
The virus SARS-CoV-2 has caused millions of infections and deaths during the COVID-19 pandemic, but as of December 2021, no new drugs targeted to SARS-CoV-2 specifically exist. Thus, it is important to identify existing drugs that can reduce the infection and mortality of this virus, since repurposing old drugs is faster and cheaper than developing new ones. Fats, such as cholesterol, can play an important role in viral infections, meaning that drugs intended to lower the levels of fats in the blood could have a protective effect against SARS-CoV-2.
To test this hypothesis, Huang, Xiao, et al. carried out a Mendelian randomization study to investigate if there is a link between drugs that lower fats and outcomes of SARS-CoV-2 infection, including susceptibility, hospitalization, and severe disease. This approach consists on grouping people according to their version of a particular gene, which minimizes the effect of variables that can cause spurious associations, something known as confounding bias. Thus, Mendelian randomization studies allow scientists to disentangle cause and effect.
Using this method, Huang, Xiao, et al. found an association between statins (a type of drug that decreases the levels of bad cholesterol) and a reduced risk of being hospitalized after being infected with SARS-CoV-2.
These findings suggest that statins could benefit patients infected with SARS-CoV-2, and indicate that they should be prioritized in future clinical trials for treating COVID-19.
Introduction
The COVID-19 pandemic has caused millions of infections and deaths, which is caused by a novel coronavirus, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Lacking drugs specifically targeted to SARS-CoV-2 infection has led to a great interest to identify drugs that can be repurposed to reduce the infection and mortality of the disease.
Available studies have suggested an important role of lipid metabolism in viral infections, including in the pathogenesis of SARS-CoV-2 infection (Proto et al., 2021). The plausible mechanisms include the involvement of host lipids in the virus life cycle, the influence of cholesterol on the immune cell functions, interfering with the mevalonate pathway, and so on Proto et al., 2021. Such evidence indicates the potential protective effect of lipid-lowering drugs against COVID-19. HMG-CoA reductase (HMGCR) inhibitors, known as statins, are the most commonly used class of lipid-lowering drugs, which have a couple of predominant merits, such as the well-proven safety, low cost, and pleiotropic effects. Proprotein convertase subtilisin/kexin type 9 (PCSK9) and Niemann–Pick C1-Like 1 (NPC1L1) are proteins playing a crucial role for the circulating level of low-density lipoprotein cholesterol (LDL-C) (Sabatine, 2019; Williams et al., 2020). Both PCSK9 inhibitors (i.e., evolocumab and alirocumab) and NPC1L1 inhibitors (i.e., ezetimibe) are FDA-approved lipid-lowering agents (Sabatine, 2019; Williams et al., 2020). A number of observational studies have investigated the association between lipid-lowering drugs and COVID-19 outcomes, but generated mixed results (Butt et al., 2020; Hariyanto and Kurniawan, 2020; Kow and Hasan, 2020; Zhang et al., 2020; Gupta et al., 2021). What’s more, confounding bias and reverse causation cannot be avoided in most of these studies.
Mendelian randomization (MR) study uses genetic variants as an instrument to perform causal inference between an exposure and an outcome, which could indicate whether an observational association is consistent with a causal effect (Davies et al., 2018). Confounding bias can be minimized in MR study because genetic variants are randomly assigned to the individual at birth. Similarly, reverse causation can be avoided because genetic variants are assigned prior to the development of disease.
Therefore, we performed two-sample MR analysis in this study to test the association of lipid‐lowering drugs (HMGCR inhibitiors, PCSK9 inhibitiors, and NPC1L1 inhibitior) with COVID-19 outcomes (susceptibility, hospitalization and very severe disease).
Materials and methods
Study design
Request a detailed protocolThis two-sample MR study is based on publicly available summary-level data from genome-wide association studies (GWASs) and expression quantitative trait loci (eQTLs) studies (Supplementary file 1—Table 1). All these studies had been approved by the relevant institutional review boards and participants had provided informed consents.
Selection of genetic instruments
View detailed protocolThree classes of FDA-approved lipid-lowering drugs were included as exposures in this study: HMGCR inhibitors, PCSK9 inhibitors, and NPC1L1 inhibitor.
As shown in Table 1, we used available eQTLs for drugs target genes (i.e., HMGCR, PCSK9, and NPC1L1) as the proxy of exposure to each lipid-lowering drug. The eQTLs summary-level data were obtained from eQTLGen Consortium (https://www.eqtlgen.org/) or GTEx Consortium V8 (https://gtexportal.org/), the details of which are presented in Supplementary file 1—Table 1. We identified common (minor allele frequency [MAF] >1%) eQTLs single-nucleotide polymorphisms (SNPs) significantly (p < 5.0 × 10−8) associated with the expression of HMGCR or PCSK9 in blood, and the expression of NPC1L1 in adipose subcutaneous tissue as there are no eQTLs in blood or other tissues available at a significance level for NPC1L1. Only cis-eQTLs were included to generate genetic instruments in this study, which were defined as eQTLs within 1 Mb on either side of the encoded gene.
Information of genetic instruments.
Abbreviations and acronyms: eQTLs, expression quantitative trait loci; GWAS, genome-wide association study; HEIDI, heterogeneity in dependent instruments; HMGCR, HMG-CoA reductase; IVW-MR, inverse-variance-weighted Mendelian randomization; LDL, low-density lipoprotein; MR-PRESSO, Mendelian Randomization Pleiotropy RESidual Sum and Outlier; MAF, minor allele frequency; NPC1L1, Niemann–Pick C1-Like 1; PCSK9, proprotein convertase subtilisin/kexin type 9; SNP, single-nucleotide polymorphism; SMR, summary-data-based Mendelian randomization.
Exposure | Genetic instruments | |
---|---|---|
Genetic variants associated with mRNA expression levels (eQTLs) | Genetic variants associated with LDL cholesterol level | |
HMGCR inhibitors | Nine hundred and twenty-one common cis-eQTLs (MAF >1%) in blood for HMGCR gene (p < 5.0 × 10−8), top SNP: rs6453133 | Seven common SNPs (MAF >1%) in low linkage disequilibrium (r2 < 0.30), associated with LDL cholesterol (p < 5.0 × 10−8), located within ±100 kb windows from HMGCR region |
PCSK9 inhibitors | Twenty-four common cis-eQTLs (MAF >1%) in blood for PCSK9 gene (p < 5.0 × 10−8), top SNP: rs472495 | Twelve common SNPs (MAF >1%) in low linkage disequilibrium (r2 < 0.30), associated with LDL cholesterol (p < 5.0 × 10−8), located within ±100 kb windows from PCSK9 region |
NPC1L1 inhibitors | Eleven common cis-eQTLs (MAF >1%) in adipose subcutaneous tissue for NPC1L1 gene (p < 5.0 × 10−8), top SNP: rs41279633 | Three common SNPs (MAF >1%) in low linkage disequilibrium (r2 < 0.30), associated with LDL cholesterol (p < 5.0 × 10−8), located within ±100 kb windows from NPC1L1 region |
Statistical analyses | ||
Primary analysis | Summary-data-based Mendelian randomization | Inverse-variance-weighted Mendelian randomization |
Sensitivity analyses | F-StatisticPositive control analysis (LDL cholesterol used as outcome)Linkage disequilibrium test: HEIDI testHorizontal pleiotropy test: SMR association between expression of adjacent genes and outcome | F-StatisticPositive control analysis (coronary heart disease used as outcome)Heterogeneity test: Cochran Q testHorizontal pleiotropy test: MR-Egger regression, MR-PRESSO test |
Secondly, to validate the observed association using the eQTLs as an instrument, we additionally proposed an instrument by selecting SNPs within 100 kb windows from target gene of each drug that was associated with LDL cholesterol level at a genome-wide significance level (p < 5.0 × 10−8) to proxy the exposure of lipid-lowering drugs. A GWAS summary data of LDL cholesterol levels from the Global Lipids Genetics Consortium (GLGC) with a sample size of 173,082 were used to identify these SNPs, where only common SNPs (MAF >1%) were included (Willer et al., 2013; Supplementary file 1—Table 1). Seven SNPs within 100 kb windows from HMGCR gene were selected for proxying HMGCR inhibitors, 12 SNPs from PCSK9 gene identified for PCSK9 inhibitors, and 3 SNPs from NPC1L1 gene selected for NPC1L1 inhibitor. To maximize the strength of the instrument for each drug, SNPs used as instruments were allowed to be in low weak linkage disequilibrium (r2 < 0.30) with each other.
Outcome sources
Request a detailed protocolGWAS summary-level data for COVID-19 outcomes were obtained from the COVID-19 Host Genetics Initiative V4 with a sample size of 1,299,010 for COVID-19 susceptibility, 908,494 for COVID-19 hospitalization, and 626,151 for COVID-19 severe disease, respectively (https://www.covid19hg.org/; COVID-19 Host Genetics Initiative, 2020; Supplementary file 1—Table 1). The study population was restricted to individuals with European ancestry, including meta-analyses of GWASs containing up to 22 cohorts from 11 countries. GWAS from these cohorts used a model adjusted for age, sex, age × age, age × sex, genetic principal components, and study-specific covariates. A COVID-19 case was confirmed by lab or self-reported infections, or electronic health records of infections. The susceptibility outcome was measured by comparing COVID-19 cases and controls who did not have a history of COVID-19. The hospitalized outcome was measured by comparing COVID-19 hospitalized cases and controls who were never admitted to the hospital due to COVID-19, including individuals without COVID-19. The severe disease outcome was measured by comparing COVID-19 cases who died or required respiratory support and controls without severe COVID-19, including individuals without COVID-19. We included individuals without COVID-19 as controls for all outcomes to decrease collider bias and allow for population-level comparisons (Griffith et al., 2020; Butler-Laporte et al., 2021).
Statistical analyses
Primary MR analysis
Request a detailed protocolSummary-data-based MR (SMR) method was applied to generate effect estimates when using eQTLs as an instrument, which investigates the association between the expression level of a gene and outcome of interest using summary-level data from GWAS and eQTL studies (Zhu et al., 2016). Allele harmonization and analysis were performed using SMR software, version 1.03 (https://cnsgenomics.com/software/smr/#Overview). Inverse-variance-weighted MR (IVW-MR) method was used to combine effect estimates when using genetic variants associated with LDL cholesterol level as an instrument. Allele harmonization and analysis were conducted using the TwoSampleMR package in R software, version 4.1.0.
Sensitivity analysis
Request a detailed protocolThe strength of SNPs used as the instrument was assessed using the F-statistic, and we included SNPs with an F-statistic of >10 to minimize weak instrument bias (Burgess and Thompson, 2011). Positive control analyses were performed for validation of both genetic instruments. Since lowering the level of LDL cholesterol is the well-proven effect of lipid-lowering drugs, we thus examined the association of exposures of interest with LDL cholesterol level as positive control study for the instrument from eQTLs. For the instrument from LDL cholesterol GWAS, we performed positive control study by examining the association of exposures of interest with coronary heart disease because coronary heart disease is the main indication of lipid-lowering drugs.
For SMR method, the heterogeneity in dependent instruments (HEIDI) test was used to test if the observed association between gene expression and outcome was due to a linkage scenario, which was performed in the SMR software (Zhu et al., 2016). The HEIDI test of p < 0.01 indicates that association is probably due to linkage (Chauquet et al., 2021). One SNP could be related to the expression of more than one genes, leading to the presence of horizontal pleiotropy. To assess the risk of horizontal pleiotropy, we identified other nearby genes (within a 1 Mb window), the expression of which was significantly associated with the genetic instrumental variant, and performed SMR analysis to examine if the expression of these genes was related to the COVID-19 outcomes.
For IVW-MR method, we tested the heterogeneity by using a Cochran Q test, where p < 0.05 indicates the evidence of heterogeneity (Higgins et al., 2003). MR-Egger regression and Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) analysis were used to assess the potential horizontal pleiotropy of the SNPs used as instrument variants. In MR Egger regression, the intercept term is a useful indication of directional horizontal pleiotropy, where p < 0.05 indicates the evidence of horizontal pleiotropy (Burgess and Thompson, 2017). MR-PRESSO analysis can identify horizontal pleiotropic outliers and provide adjusted estimates, where p < 0.05 for Global test indicates the presence of horizontal pleiotropic outliers (Verbanck et al., 2018). Besides, a multivariable MR study was further conducted to examine if the observed association was direct association. We first investigated the association of HMGCR-mediated LDL cholesterol with the common risk factors of COVID-19 hospitalization, including body mass index, diabetes, hypertension, and coronary heart disease. After that, a multivariable MR study was performed by adjusting for the factors which showed a significant association. All these analyses were implemented in R software, version 4.1.0.
To account for multiple testing, Bonferroni correction was used to adjust the thresholds of significance level, thus a strong evidence was suggested for p < 0.006 (three exposures and three outcomes) and a suggestive evidence of 0.006 ≤ p < 0.05.
Results
Genetic instruments selection and COVID-19 outcomes
A total of 921,24, and 11 cis-eQTLs were identified from eQTLGen or GTEx Consortium for drugs target gene HMGCR, PCSK9, and NPC1L1, respectively, and the most significant cis-eQTL SNP was selected as a genetic instrument for the target gene of each drug (Table 1, Supplementary file 1—Table 2). A total of 7, 12, and 3 SNPs within or nearby gene HMGCR, PCSK9, and NPC1L1 were selected from a GWAS summary data of LDL cholesterol levels in the Global Lipids Genetics Consortium, respectively (Table 1, Supplementary file 1—Table 3). F-Statistics for all instrument variants were over 30, suggesting that weak instrument bias can be minimized in our study (Supplementary file 1-Tables 2 and 3). Positive control study showed significant associations between exposure to each drug and LDL cholesterol when using eQTLs-proposed instruments (Supplementary file 1—Table 5), as well as between exposure to each drug and coronary heart disease when using LDL cholesterol GWAS-proposed instruments (Supplementary file 1—Table 6), further ensuring the efficacy of the selected genetic instruments.
From COVID-19 GWASs, a total of 14,134 cases and 1,284,876 controls were used to explore the association with COVID-19 susceptibility, 6406 cases and 902,088 controls for COVID-19 hospitalization, and 3886 cases and 622,265 controls for COVID-19 severe disease (Supplementary file 1—Table 1).
Primary analysis
In Figure 1 and Supplementary file 1—Table 2, results from SMR analysis found a suggestive evidence for the association of the increased expression of HMGCR gene in blood (equivalent to a one standard deviation increase) with the higher risk of COVID-19 susceptibility (odds ratio [OR] = 1.30, 95% confidence interval [CI] = 1.05–1.61; p = 0.017) and COVID-19 hospitalization (OR = 1.38, 95% CI = 1.06–1.81; p = 0.019), indicating that HMGCR inhibitors might lower the risk of COVID-19 susceptibility and hospitalization. Suggestive evidence was observed regarding the negative association between PCSK9 expression and risk of COVID-19 susceptibility (OR = 0.84, 95% CI = 0.73–0.97; p = 0.02). No significant association was found between the expression of NPC1L1 and COVID-19 outcomes.

Summary-data-based Mendelian randomization (SMR) association between expression of gene HMGCR, PCSK9, or NPC1L1 and COVID-19 outcomes.
SMR method was used to assess the association.
In Figure 2 and Supplementary file 1—Table 4, IVW-MR analysis also found a suggestive evidence for the association between HMGCR-mediated LDL cholesterol (equivalent to a 1 mmol/l increase) and the risk of COVID-19 hospitalization (OR = 1.32, 95% CI = 1.00–1.74; p = 0.049), further supporting a possible protective effect of HMGCR inhibitors against COVID-19 hospitalization. Strong evidence was observed between NPC1L1-mediated LDL cholesterol and the risk of COVID-19 susceptibility (OR = 2.02, 95% CI = 1.30–3.13; p = 0.002). IVW-MR analysis did not provide any evidence for the association between PCSK9-mediated LDL cholesterol and COVID-19 outcomes.
Sensitivity Analysis
For SMR analysis, HEIDI test suggested that all observed associations were not due to a linkage (p > 0.01), except for the association between HMGCR expression and COVID-19 susceptibility (p = 0.009) (Supplementary file 1—Table 2). We further examine if horizontal pleiotropy was present in the association between HMGCR expression and COVID-19 outcomes by investigating if there was an association between the expression of nearby genes which are significantly associated with the top eQTL SNP (instrument variant) of HMGCR and COVID-19 outcomes. We identified six genes, including HMGCR, the expression of which were associated with the instrument variant (Supplementary file 1—Table 7). Only four genes have available eQTLs at a genome-wide significance level (p < 5.0 × 10−8). Among these four genes, only HMGCR expression was significantly related to COVID-19 susceptibility and COVID-19 hospitalization, suggesting a small role of horizontal pleiotropy in the observed associations (Supplementary file 1—Table 8).
For IVW-MR analysis, Cochran Q test did not find evidence of heterogeneity for all reported results (all p > 0.05; Supplementary file 1—Table 4). Both the intercept term in MR-Egger regression and MR-PRESSO analysis suggested no significant overall horizontal pleiotropy (all p > 0.05; Supplementary file 1—Table 4). A multivariable MR study suggested that BMI and diabetes might play a role in the association between HMGCR-mediated LDL cholesterol level and COVID19 hospitalization (Supplementary file 1—Tables 9 and 10).
Discussion
The present MR study provided a suggestive evidence regarding the positive association of the HMGCR expression and HMGCR-mediated LDL cholesterol level with the risk of COVID-19 hospitalization, both of which together indicated a potential protective effect of HMGCR inhibition against COVID-19 hospitalization (OR instrument 1 = 0.72, 95% CI = 0.55–0.95; OR instrument 2=0.76, 95% CI = 0.57–1.00). We found a suggestive evidence of the negative association between PCSK9 expression and COVID-19 susceptibility, but which was not supported when using LDL cholesterol GWAS as an instrument. A strong evidence was observed for the protective effect of NPC1L1-mediated lower level of LDL cholesterol on COVID-19 susceptibility, but there was no evidence for the association of NPC1L1 expression and COVID-19 outcomes.
Compared to developing a new drug, repurposing an old drug is much more economical and time-saving, in particular, the importance is further highlighted during a pandemic. COVID-19 pandemic has driven a number of studies of drug repurposing (Fajgenbaum and Rader, 2020; Gaziano et al., 2021). The role of lipid metabolism in viral infections has raised interest regarding the possibility of repurposing lipid-lowering drugs as anti-COVID-19 agents (Proto et al., 2021). As one of the most commonly prescribed drugs, statins have received the greatest attention for their pleiotropic effects, including lowering serum cholesterol, anti-inflammatory and immunomodulatory properties, and antithrombotic effect, all of which play a role in viral infections (Fajgenbaum and Rader, 2020; Proto et al., 2021; Rubin, 2021). Emerging observational studies have investigated if statins might benefit patients with COVID-19 (Butt et al., 2020; Hariyanto and Kurniawan, 2020; Kow and Hasan, 2020; Zhang et al., 2020; Gupta et al., 2021). A largest retrospective cohort study including 13,981 patients admitted to hospital due to COVID-19 suggested a significant reduction in 28-day all-cause mortality by 42% in the group with statins than patients without statins (Zhang et al., 2020). A meta-analysis with 8990 COVID-19 patients also found a 30% lower risk of fatal or severe disease (Kow and Hasan, 2020). However, a Danish nationwide cohort study with 4842 COVID-19 patients and a meta-analysis with 3449 COVID-19 patients did not find the association between statins use and improved COVID-19 outcomes (Butt et al., 2020; Hariyanto and Kurniawan, 2020). In addition, in these studies, considerable differences in clinical characteristics cannot be avoided between patients with and without statins, and causal inference is not allowed due to the retrospective nature of observational studies.
As a genetic epidemiological method, MR study could overcome the limitations of traditional observational studies. In this MR study, we used genetic variants related to HMGCR expression or HMGCR-mediated LDL cholesterol as instruments to proxy the exposure of statins. Both analyses found a suggestive evidence that HMGCR inhibition could reduce the risk of COVID-19 hospitalization, rather than COVID-19 susceptibility and very severe outcome. Although strong evidence is lacking, these results provided a causal evidence supporting the finding from the largest cohort study (Zhang et al., 2020), which calls for additional observational studies in different populations, mechanistic studies, and randomized controlled studies to examine its potential effect against COVID-19. Patients with COVID-19 who already take statins or start to take it for the indication of statins were recommended to continue to take it, which might be beneficial to both its original indication and COVID-19 (Rubin, 2021). And statins might be a prioritized drug in future clinical trials for treating COVID-19. Besides, although no association was found between NPC1L1 expression in adipose subcutaneous and COVID-19 outcomes, there was a strong evidence of the association between NPC1L1-mediated LDL cholesterol and COVID-19 susceptibility. The effect of NPC1L1 inhibitor on COVID-19 susceptibility may be worth further studies as well.
Study strengths
The main strength of our study is the use of genetic instruments to proxy drug exposure, which could minimize confounding bias and avoid reverse causation. Besides, we used two different kinds of genetic instruments to proxy the studied drug, which contributes to validating the effect estimates from each other. A number of sensitivity analyses have been performed to test the efficacy of genetic instruments and the assumptions of MR study.
Study limitations
This study has several limitations. Firstly, there are no available eQTLs in blood for NPC1L1, so we were not able to explore the association between NPC1L1 expression in blood and COVID-19 outcomes. Besides, there are no available eQTLs in liver (the main tissue related to lipid metabolism) for these target genes, which might provide more convincing evidence of the observed association. The sample size of eQTL study for PCSK9 and NPC1L1 in GTEx is relatively smaller, which may affect the statistical power for the results of PCSK9 or NPC1L1 inhibition. Secondly, the effect of statins probably varies between subgroups, for example, it may be more effective in patients with chronic diseases (e.g., coronary heart disease). However, the use of summary-level data did not allow us to perform subgroup analyses, so further MR study with individual-level data is needed to provide more detailed information. Thirdly, the Bonferroni correction for multiple tests suggests that we cannot rule out the false-positive possibility for the finding of the protective effect of statins on COVID hospitalization. Fourthly, confounding bias and/or horizontal pleiotropy cannot be completely excluded although we have performed various sensitivity analyses to test the assumptions of MR study. Fifthly, both eQTLs and GWAS data used in this study were predominantly obtained from European population ancestry, thus these findings should be interpreted with caution when generalizing to other populations.
Conclusions
In conclusion, this MR study suggested a causal relationship between HMGCR inhibition and the reduced risk of COVID-19 hospitalization. Clninical trials are called to examine if statins have the protective effect against COVID-19 and further researches are needed to explore the underlying mechanisms.
Data availability
Individual-level data cannot be provided but the raw data of the eQTLGen Consortium, GTEx and COVID-19 Host Genetics Initiative can be acessed at https://www.eqtlgen.org/, https://gtexportal.org/, and https://www.covid19hg.org/, respectively. Summary-level GWAS or eQTL data and code used to produce main results have been uploaded to GitHub (https://github.com/WH57/lipid_covid19.git), (copy archived at https://archive.softwareheritage.org/swh:1:rev:3f6e94c8e0553595f6a011e701b01ec3d5380b72). All MR results and GWAS or eQTL associations of selected SNPs were provided in the Supplementary File 1 - Tables 2 to 4.
-
COVID-19 Host Genetics InitiativeID V4. COVID-19 Host Genetics Initiative V4.
-
Global Lipids Genetics ConsortiumID GLGC. Global Lipids Genetics Consortium.
References
-
Avoiding bias from weak instruments in Mendelian randomization studiesInternational Journal of Epidemiology 40:755–764.https://doi.org/10.1093/ije/dyr036
-
Interpreting findings from Mendelian randomization using the MR-Egger methodEuropean Journal of Epidemiology 32:377–389.https://doi.org/10.1007/s10654-017-0255-x
-
Teaching Old Drugs New Tricks: Statins for COVID-19?Cell Metabolism Vol. 32:145–147.https://doi.org/10.1016/j.cmet.2020.07.006
-
Statin therapy did not improve the in-hospital outcome of coronavirus disease 2019 (COVID-19) infectionDiabetes & Metabolic Syndrome 14:1613–1615.https://doi.org/10.1016/j.dsx.2020.08.023
-
Measuring inconsistency in meta-analysesBMJ Vol. 327:557–560.https://doi.org/10.1136/bmj.327.7414.557
-
Meta-analysis of Effect of Statins in Patients with COVID-19The American Journal of Cardiology 134:153–155.https://doi.org/10.1016/j.amjcard.2020.08.004
-
PCSK9 inhibitors: clinical evidence and implementationNature Reviews. Cardiology 16:155–165.https://doi.org/10.1038/s41569-018-0107-8
-
Discovery and refinement of loci associated with lipid levelsNature Genetics 45:1274–1283.https://doi.org/10.1038/ng.2797
-
Lipid lowering and Alzheimer disease risk: A mendelian randomization studyAnnals of Neurology 87:30–39.https://doi.org/10.1002/ana.25642
Decision letter
-
Edward D JanusReviewing Editor; University of Melbourne, Australia
-
David M SerwaddaSenior Editor; Makerere University School of Public Health, Uganda
-
Edward D JanusReviewer; University of Melbourne, Australia
-
Xia JiangReviewer; Karolinska Institutet, Sweden
Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.
Decision letter after peer review:
Thank you for submitting your article "Association of lipid-lowering drugs with COVID-19 outcomes from a Mendelian randomization study" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, including Edward D Janus as Reviewing Editor and Reviewer #1, and the evaluation has been overseen by David Serwadda as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Xia Jiang (Reviewer #2).
The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.
Essential revisions:
1) HMG-CoA reductase was introduced in a detailed manner in the first paragraph of introduction, while PCSK 9 inhibitor and NPC1L1 inhibitor were not mentioned until the last paragraph of introduction. The abbreviations were not spelt out. Please adjust the flow to make it more smoothly.
2) The authors demonstrate with SMR that a higher expression of HMGCR was associated with a higher risk of COVID-19 hospitalization. With IVW-MR they observed a positive association between HMGCR -mediated LDL cholesterol and COVID-19 hospitalization. For the other two drug classes no clear associations were shown.
This suggests but does not prove that HMGCR drugs ie statins might improve outcomes. For the other two classes its unclear if there would be sufficient users of these drugs in the data sets to provide enough power to adequately address the issue.
3) Overall this is a well conducted study providing important further insights but the final sentence of the abstract as well as the related Discussion section including the limitations should be more expressed more conservatively.
4) Choosing the right tissue is very important for eQTL analysis. While a wide number of tissues are available from GTEx consortium, the authors focused only on the blood or adipose subcutaneous tissues. Please clarify the rationale. How about cardiovascular tissues, liver and other organs? have the authors considered other relevant tissues available in GTEx?
5) As mentioned in the methods, "only cis-eQTLs were included to generate genetic instruments in this study, which were defined as eQTLs within 1 Mb on either side of the encoded gene" were included. Were these SNPs independent of each other? Have the authors considered linkage disequilibrium, and if so, what was the threshold? Including correlated instruments might bias the results (false positive).
6) Please clarify the ethnicity of population involved in the current study as well as the generalizability of results.
7) The method section lacked proper description on the GWAS data used for LDL and COVID19 outcome.
8) As mentioned by the authors, "we additionally proposed an instrument by selecting SNPs within 100 kb windows from target gene of each drug that were associated with LDL cholesterol level at a genome-wide significance level (p<5.0 × 10-8) to proxy the exposure of lipid-lowering drugs" – it means the authors intersect the eQTL with LDL-SNPs? What is the purpose of doing so, in other words, by choosing only part of the SNPs, would the MR assumption still hold?
9) The causal association between HMGCR and COVID19 susceptibility was suggestive, so was the causal association between PCSK9 and COVID19 susceptibility, while the estimates were in opposite direction (1.30 vs. 0.84) – how to interpret such results?
10) Have the authors considered confounding effect from, for example, obesity, to the identified putative causal relationship?
https://doi.org/10.7554/eLife.73873.sa1Author response
Essential revisions:
1) HMG-CoA reductase was introduced in a detailed manner in the first paragraph of introduction, while PCSK 9 inhibitor and NPC1L1 inhibitor were not mentioned until the last paragraph of introduction. The abbreviations were not spelt out. Please adjust the flow to make it more smoothly.
Thank you for the suggestions, we have added descriptions of PCSK 9 inhibitor and NPC1L1 inhibitor following HMG-CoA reductase inhibitors in the first paragraph of introduction and full names at the place when first appeared in the manuscript. #Page4
2) The authors demonstrate with SMR that a higher expression of HMGCR was associated with a higher risk of COVID-19 hospitalization. With IVW-MR they observed a positive association between HMGCR -mediated LDL cholesterol and COVID-19 hospitalization. For the other two drug classes no clear associations were shown.
This suggests but does not prove that HMGCR drugs ie statins might improve outcomes. For the other two classes its unclear if there would be sufficient users of these drugs in the data sets to provide enough power to adequately address the issue.
We agree that the current evidence is not enough to prove that HMGCR inhibitors might improve outcomes. We thus did some revisions on both abstract and discussion to interprete the results in a more conservative way #Page 3,11. We agree that the sample size of eQTL study for PCSK9 and NPC1L1 in GTEx is relatively smaller, which may affect the statistical power for the results of PCSK9 or NPC1L1 inhibition. We have added related discussions in the limitation section. #Page10
3) Overall this is a well conducted study providing important further insights but the final sentence of the abstract as well as the related Discussion section including the limitations should be more expressed more conservatively.
It is an important point. As our answer to the previous question, we have revised the abstract and related Discussion section accordingly.
4) Choosing the right tissue is very important for eQTL analysis. While a wide number of tissues are available from GTEx consortium, the authors focused only on the blood or adipose subcutaneous tissues. Please clarify the rationale. How about cardiovascular tissues, liver and other organs? have the authors considered other relevant tissues available in GTEx?
We agree that choosing the right tissue is important for eQTL analysis. Liver is the main tissue related to lipid metabolism, unfortunately there are no available eQTLs in liver for these target genes. Although a wide number of tissues are available from GTEx consortium, the eQTLs at a genome-wide significance level (p<5.0 × 10-8) are limited. For example, for HMGCR, there are only eQTLs identified in blood from eQTLGen and eQTLs identified in muscle skeletal tissue from GTEx. But sharing eQTLs across tissues is very common, particularly for top SNPs. For example, previous study found that genetic effects at the top cis-eQTLs were highly correlated between independent brain and blood samples 1. And eQTLs data from blood is most comprehensive due to the easier access to blood sample as compared to other specific tissues. We thus first focused on eQTLs from blood, which we believe could provide some clues. When the blood eQTLs data is unavailable, we then focused on adipose tissue, because it is another tissue related to lipid metabolism.
5) As mentioned in the methods, "only cis-eQTLs were included to generate genetic instruments in this study, which were defined as eQTLs within 1 Mb on either side of the encoded gene" were included. Were these SNPs independent of each other? Have the authors considered linkage disequilibrium, and if so, what was the threshold? Including correlated instruments might bias the results (false positive).
As we mentioned in the method section, HEIDI test was used to test if the observed association between gene expression and the outcome was due to a linkage disequilibrium. The HEIDI test of P<0.01 indicates that association is probably due to linkage. Briefly, if gene expression and a trait share the same causal variant, the bxy values calculated for any SNPs in LD (using the default value of r2 > 0.05 and also r2 < 0.9 to avoid issues of collinearity) with the causal variant should be identical. Therefore, testing against this null hypothesis of a single causal variant is equivalent to testing for heterogeneity in the bxy values estimated for the SNPs in the cis-eQTL region. The detailed information of the results from HEIDI test is presented in Supplementary file 1-Table 2.
6) Please clarify the ethnicity of population involved in the current study as well as the generalizability of results.
Thank you for the suggestion. The discussion about the ethnicity of population and the generalizability of results has been added in the limitation section. #Page11
7) The method section lacked proper description on the GWAS data used for LDL and COVID19 outcome.
We have added brief descriptions on the GWAS data and presented the detailed information in Supplementary file 1-Table 1. #Page12-13
8) As mentioned by the authors, "we additionally proposed an instrument by selecting SNPs within 100 kb windows from target gene of each drug that were associated with LDL cholesterol level at a genome-wide significance level (p<5.0 × 10-8) to proxy the exposure of lipid-lowering drugs" – it means the authors intersect the eQTL with LDL-SNPs? What is the purpose of doing so, in other words, by choosing only part of the SNPs, would the MR assumption still hold?
We apologize for the misunderstanding, we did not mean to intersect the eQTL with LDL-SNP. One genetic instrument in our analysis is SNPs located in the target gene of each drug that was associated with LDL cholesterol, because LDL-C is the most important cholesterol in disease development and it is well-proved that these lipid-lowering drugs are effective in lowering the plasm level of LDL-C. This approach has been applied in several previous studies 2, 3. The other instrument is SNPs associated with the expression of target gene of each drug (i.e., eQTLs) by using SMR method, which is an analysis to test if the effect size of SNP on the phenotype is mediated by gene expression. This approach has been also applied in previous studies 4, 5.
9) The causal association between HMGCR and COVID19 susceptibility was suggestive, so was the causal association between PCSK9 and COVID19 susceptibility, while the estimates were in opposite direction (1.30 vs. 0.84) – how to interpret such results?
We found a negative association between PCSK9 and COVID19 susceptibility in the SMR study by using SNPs associated expression of target genes as instrument. However, we know that PCSK9 needs to bind to the LDL receptor to adjust LDL cholesterol level, thus its functions might be affected by the expression of the LDL receptor. Additionally, no association was observed when using SNPs as the instrument of LDL cholesterol, suggesting lowering LDL cholesterol by targeting PCSK9 might not affect the outcomes of COVID-19. As for HMGCR, we found the associations using SNPs as the instruments for both the expression and LDL cholesterol with COVID-19 hospitalization. As HMGCR is the rate-limiting enzyme, it can directly affect the synthesizing of cholesterol, thus we conclude that lowering LDL cholesterol by targeting HMGCR might affect COVID-19 hospitalization.
10) Have the authors considered confounding effect from, for example, obesity, to the identified putative causal relationship?
We agree that risk factors of COVID-19 may have an impact on identifying a putative causal relationship, such as obesity, chronic diseases (eg., diabetes, hypertention, coronary heart disease). As suggested in previous MR study 6, the impact of proposed risk factor may be due to pleiotropic or confounding (Author response image 1). We thus added multivariable MR analyses for the observed association in the IVW-MR study by including covariates that were associated with HMGCR-mediated LDLC, (Supplementary file 1-Table 9,10) #Page 15. Such analyses could not distinguish between the pleiotropic and confounding, but it can provide an estimate of the direct association between HMGCR-mediated LDLC and COVID19 hospitalization.
References
1. Qi T, Wu Y, Zeng J, Zhang F, Xue A, Jiang L, Zhu Z, Kemper K, Yengo L, Zheng Z, e QC, Marioni RE, Montgomery GW, Deary IJ, Wray NR, Visscher PM, McRae AF and Yang J. Identifying gene targets for brain-related traits using transcriptomic and methylomic data from blood. Nat Commun. 2018;9:2282.
2. Yarmolinsky J, Bull CJ, Vincent EE, Robinson J, Walther A, Smith GD, Lewis SJ, Relton CL and Martin RM. Association Between Genetically Proxied Inhibition of HMG-CoA Reductase and Epithelial Ovarian Cancer. Jama. 2020;323:646-655.
3. Ference BA, Ray KK, Catapano AL, Ference TB, Burgess S, Neff DR, Oliver-Williams C, Wood AM, Butterworth AS, Di Angelantonio E, Danesh J, Kastelein JJP and Nicholls SJ. Mendelian Randomization Study of ACLY and Cardiovascular Disease. The New England journal of medicine. 2019;380:1033-1042.
4. Chauquet S, Zhu Z, O'Donovan MC, Walters JTR, Wray NR and Shah S. Association of Antihypertensive Drug Target Genes With Psychiatric Disorders: A Mendelian Randomization Study. JAMA Psychiatry. 2021;78:623-631.
5. Gaziano L, Giambartolomei C, Pereira AC, Gaulton A, Posner DC, Swanson SA, Ho YL, Iyengar SK, Kosik NM, Vujkovic M, Gagnon DR, Bento AP, Barrio-Hernandez I, Ronnblom L, Hagberg N, Lundtoft C, Langenberg C, Pietzner M, Valentine D, Gustincich S, Tartaglia GG, Allara E, Surendran P, Burgess S, Zhao JH, Peters JE, Prins BP, Angelantonio ED, Devineni P, Shi Y, Lynch KE, DuVall SL, Garcon H, Thomann LO, Zhou JJ, Gorman BR, Huffman JE, O'Donnell CJ, Tsao PS, Beckham JC, Pyarajan S, Muralidhar S, Huang GD, Ramoni R, Beltrao P, Danesh J, Hung AM, Chang KM, Sun YV, Joseph J, Leach AR, Edwards TL, Cho K, Gaziano JM, Butterworth AS, Casas JP and Initiative VAMVPC-S. Actionable druggable genome-wide Mendelian randomization identifies repurposing opportunities for COVID-19. Nat Med. 2021;27:668-676.
6. Kar SP, Brenner H, Giles GG, Huo D, Milne RL, Rennert G, Simard J, Zheng W, Burgess S and Pharoah PDP. Body mass index and the association between low-density lipoprotein cholesterol as predicted by HMGCR genetic variants and breast cancer risk. Int J Epidemiol. 2019;48:1727-1730.
https://doi.org/10.7554/eLife.73873.sa2Article and author information
Author details
Funding
Start-up Fund for high-level talents of Fujian Medical University (XRCZX2021026)
- Wuqing Huang
The funder had no role in study design, data collection, and interpretation, or the decision to submit the work for publication.
Acknowledgements
We thank the patients and investigators who contributed to the eQTLGen Consortium, GTEx Consortium, COVID-19 Host Genetics Initiative, Global Lipids Genetics Consortium, and CARDIoGRAMplusC4D Consortium.
Ethics
This two-sample MR study is based on publicly available summary-level data from genome-wide association studies (GWASs) and expression quantitative trait loci (eQTLs) studies. All these studies had been approved by the relevant institutional review boards and participants had provided informed consents.
Senior Editor
- David M Serwadda, Makerere University School of Public Health, Uganda
Reviewing Editor
- Edward D Janus, University of Melbourne, Australia
Reviewers
- Edward D Janus, University of Melbourne, Australia
- Xia Jiang, Karolinska Institutet, Sweden
Version history
- Preprint posted: July 24, 2021 (view preprint)
- Received: September 14, 2021
- Accepted: November 22, 2021
- Accepted Manuscript published: December 6, 2021 (version 1)
- Version of Record published: December 24, 2021 (version 2)
Copyright
© 2021, Huang et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Metrics
-
- 2,658
- Page views
-
- 289
- Downloads
-
- 13
- Citations
Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.
Download links
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Further reading
-
- Biochemistry and Chemical Biology
- Epidemiology and Global Health
Background: High levels of circulating adiponectin are associated with increased insulin sensitivity, low prevalence of diabetes, and low body mass index (BMI); however, high levels of circulating adiponectin are also associated with increased mortality in the 60-70 age group. In this study, we aimed to clarify factors associated with circulating high-molecular-weight (cHMW) adiponectin levels and their association with mortality in the very old (85-89 years old) and centenarians.
Methods: The study included 812 (women: 84.4%) for centenarians and 1,498 (women: 51.7%) for the very old. The genomic DNA sequence data were obtained by whole genome sequencing or DNA microarray-imputation methods. LASSO and multivariate regression analyses were used to evaluate cHMW adiponectin characteristics and associated factors. All-cause mortality was analyzed in three quantile groups of cHMW adiponectin levels using Cox regression.
Results: The cHMW adiponectin levels were increased significantly beyond 100 years of age, were negatively associated with diabetes prevalence, and were associated with SNVs in CDH13 (p = 2.21 × 10-22) and ADIPOQ (p = 5.72 × 10-7). Multivariate regression analysis revealed that genetic variants, BMI, and high-density lipoprotein cholesterol (HDLC) were the main factors associated with cHMW adiponectin levels in the very old, whereas the BMI showed no association in centenarians. The hazard ratios for all-cause mortality in the intermediate and high cHMW adiponectin groups in very old men were significantly higher rather than those for all-cause mortality in the low level cHMW adiponectin group, even after adjustment with BMI. In contrast, the hazard ratios for all-cause mortality were significantly higher for high cHMW adiponectin groups in very old women, but were not significant after adjustment with BMI.
Conclusions: cHMW adiponectin levels increased with age until centenarians, and the contribution of known major factors associated with cHMW adiponectin levels, including BMI and HDLC, varies with age, suggesting that its physiological significance also varies with age in the oldest old.
Funding: This study was supported by grants from the Ministry of Health, Welfare, and Labour for the Scientific Research Projects for Longevity; a Grant-in-Aid for Scientific Research (No 21590775, 24590898, 15KT0009, 18H03055, 20K20409, 20K07792, 23H03337) from the Japan Society for the Promotion of Science; Keio University Global Research Institute (KGRI), Kanagawa Institute of Industrial Science and Technology (KISTEC), Japan Science and Technology Agency (JST) Research Complex Program 'Tonomachi Research Complex' Wellbeing Research Campus: Creating new values through technological and social innovation (JP15667051), the Program for an Integrated Database of Clinical and Genomic Information from the Japan Agency for Medical Research and Development (No. 16kk0205009h001, 17jm0210051h0001, 19dk0207045h0001); the medical-welfare-food-agriculture collaborative consortium project from the Japan Ministry of Agriculture, Forestry, and Fisheries; and the Biobank Japan Program from the Ministry of Education, Culture, Sports, and Technology.
-
- Epidemiology and Global Health
Background:
Machine learning (ML) techniques improve disease prediction by identifying the most relevant features in multidimensional data. We compared the accuracy of ML algorithms for predicting incident diabetic kidney disease (DKD).
Methods:
We utilized longitudinal data from 1365 Chinese, Malay, and Indian participants aged 40–80 y with diabetes but free of DKD who participated in the baseline and 6-year follow-up visit of the Singapore Epidemiology of Eye Diseases Study (2004–2017). Incident DKD (11.9%) was defined as an estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m2 with at least 25% decrease in eGFR at follow-up from baseline. A total of 339 features, including participant characteristics, retinal imaging, and genetic and blood metabolites, were used as predictors. Performances of several ML models were compared to each other and to logistic regression (LR) model based on established features of DKD (age, sex, ethnicity, duration of diabetes, systolic blood pressure, HbA1c, and body mass index) using area under the receiver operating characteristic curve (AUC).
Results:
ML model Elastic Net (EN) had the best AUC (95% CI) of 0.851 (0.847–0.856), which was 7.0% relatively higher than by LR 0.795 (0.790–0.801). Sensitivity and specificity of EN were 88.2 and 65.9% vs. 73.0 and 72.8% by LR. The top 15 predictors included age, ethnicity, antidiabetic medication, hypertension, diabetic retinopathy, systolic blood pressure, HbA1c, eGFR, and metabolites related to lipids, lipoproteins, fatty acids, and ketone bodies.
Conclusions:
Our results showed that ML, together with feature selection, improves prediction accuracy of DKD risk in an asymptomatic stable population and identifies novel risk factors, including metabolites.
Funding:
This study was supported by the National Medical Research Council, NMRC/OFLCG/001/2017 and NMRC/HCSAINV/MOH-001019-00. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.