A Mendelian randomization study of the role of lipoprotein subfractions in coronary artery disease
Abstract
Recent genetic data can offer important insights into the roles of lipoprotein subfractions and particle sizes in preventing coronary artery disease (CAD), as previous observational studies have often reported conflicting results. We used the LD score regression to estimate the genetic correlation of 77 subfraction traits with traditional lipid profile and identified 27 traits that may represent distinct genetic mechanisms. We then used Mendelian randomization (MR) to estimate the causal effect of these traits on the risk of CAD. In univariable MR, the concentration and content of medium high-density lipoprotein (HDL) particles showed a protective effect against CAD. The effect was not attenuated in multivariable analyses. Multivariable MR analyses also found that small HDL particles and smaller mean HDL particle diameter may have a protective effect. We identified four genetic markers for HDL particle size and CAD. Further investigations are needed to fully understand the role of HDL particle size.
Introduction
Lipoprotein subfractions have been increasingly studied in epidemiological research and used in clinical practice to predict the risk of cardiovascular diseases (CVD) (Rankin et al., 2014; Mora et al., 2009; China Kadoorie Biobank Collaborative Group et al., 2018). Several studies have identified potentially novel subfraction predictors for CVD (Mora et al., 2009; Hoogeveen et al., 2014; Williams et al., 2014; Ditah et al., 2016; Lawler et al., 2017; Fischer et al., 2014) and demonstrated that the addition of subfraction measurements can significantly improve the risk prediction for CVD (Würtz et al., 2012; van Schalkwijk et al., 2014; McGarrah et al., 2016; Rankin et al., 2014). However, these observational studies often provide conflicting evidence on the precise roles of the lipoprotein subfractions. For example, while some studies suggested that small, dense low-density lipoprotein (LDL) particles may be more atherogenic (Lamarche et al., 1997; Hoogeveen et al., 2014), others found that larger LDL size is associated with higher CVD risk (Campos et al., 2001; Mora, 2009). Some recent observational studies found that the inverse association of CVD outcomes with smaller high-density lipoprotein (HDL) particles is stronger than the association with larger HDL particles (Ditah et al., 2016; Kim et al., 2016; McGarrah et al., 2016; Silbernagel et al., 2017), but other studies reached the opposite conclusion in different cohorts (Li et al., 2016; Arsenault et al., 2009). Currently, the utility of lipoprotein subfractions or particle sizes in routine clinical practice remains controversial (Superko, 2009; Mora, 2009; Davidson et al., 2011; Bays et al., 2016), as there is still a great uncertainty about their causal roles in CVD, largely due to a lack of intervention data (Bays et al., 2016).
Mendelian randomization (MR) is an useful causal inference method that avoids many common pitfalls of observational cohort studies (Smith and Ebrahim, 2003). By using genetic variation as instrumental variables, MR asks if the genetic predisposition to a higher level of the exposure (in this case, lipoprotein subfractions) is associated with higher occurrences of the disease outcome (Didelez and Sheehan, 2007). A positive association suggests a causally protective effect of the exposure if the genetic variants satisfy the instrumental variable assumptions (Didelez and Sheehan, 2007; Davey Smith and Hemani, 2014). Since MR can provide unbiased causal estimate even when there are unmeasured confounders, it is generally considered more credible than other non-randomized designs and is quickly gaining popularity in epidemiological research (Gidding et al., 2012; Davies et al., 2018). MR has been used to estimate the effect of several metabolites on CVD, but most prior studies are limited to just one or a few risk exposures at a time (Emdin et al., 2016; Ference et al., 2017).
In this study, we will use recent genetic data to investigate the roles of lipid and lipoprotein traits in the occurrence of coronary artery disease (CAD) and myocardial infarction (MI). In particular, we are interested in discovering lipoprotein subfractions that may be causal risk factors for CAD and MI in addition to the traditional lipid profile (LDL cholesterol, HDL cholesterol, and triglycerides levels). To this end, we will first estimate the genetic correlation of the lipoprotein subfractions and particle sizes with the tradition risk factors and remove the traits that have a high genetic correlation. We will then use MR to estimate the causal effects of the selected lipoprotein subfractions and particle sizes on CAD and MI. Finally, we will explore potential genetic markers for the identified lipoprotein and subfraction traits.
Materials and methods
GWAS summary datasets and lipoprotein particle measurements
Request a detailed protocolTable 1 describes all GWAS summary datasets used in this study, including two GWAS of the traditional lipid risk factors (Willer et al., 2013; Hoffmann et al., 2018), two recent GWAS of the human lipidome (Kettunen et al., 2016; Davis et al., 2017), and three GWAS of CAD or MI (Nikpay et al., 2015; Nelson et al., 2017; Abbott et al., 2018). In the two GWAS of the lipidome (Kettunen et al., 2016; Davis et al., 2017), high-throughput nuclear magnetic resonance (NMR) spectroscopy was used to measure the circulating lipid and lipoprotein traits (Soininen et al., 2009). We investigated the 82 lipid and lipoprotein traits measured in these studies that are related to very-low-density lipoprotein (VLDL), LDL, intermediate-density lipoprotein (IDL), and HDL subfractions and particle sizes. All the subfraction traits are named with three components that are separated by hyphens: the first component indicates the size (XS, S, M, L, XL, XXL); the second component indicates the fraction according to the lipoprotein density (VLDL, LDL, IDL, HDL); the third component indicates the measurement (C for total cholesterol, CE for cholesterol esters, FC for free cholesterol, L for total lipids, P for particle concentration, PL for phospholipids, TG for triglycerides). For example, M-HDL-P refers to the concentration of medium HDL particles.
Information about the GWAS summary datasets used in this article.
The columns are the phenotypes reported by the GWAS studies, the consortium or name of the first author of the publication, PubMed ID, population, sample size, other GWAS datasets with other lapping sample, and URLs we used to download the datasets.
Phenotype | Dataset name | PubMed ID | Population | Sample size | Sample overlap with other datasets | URL to summary dataset |
---|---|---|---|---|---|---|
Traditional lipid traits | GERA | 29507422 Hoffmann et al., 2018 | Multi-ethnic | 94,674 | ftp://ftp.ebi.ac.uk/pub/databases/gwas/summary_statistics/ | |
GLGC | 24097068 Willer et al., 2013 | European | 188,578 | Kettunen, CARDIoGRAMplusC4D | http://csg.sph.umich.edu/abecasis/public/lipids2013/ | |
Lipoprotein subfraction traits | Davis | 29084231 Davis et al., 2017 | Finnish | 8372 | http://csg.sph.umich.edu/boehnke/public/metsim-2017-lipoproteins/ | |
Kettunen | 27005778 Kettunen et al., 2016 | European | 24,925 | GLGC, CARDIoGRAMplusC4D | http://www.computationalmedicine.fi/data#NMR_GWAS | |
Heart disease traits | CARDIoGRAMplusC4D (CAD) | 26343387 Nikpay et al., 2015 | Mostly European | 185,000 | GLGC, Kettunen | http://www.cardiogramplusc4d.org/data-downloads/ |
CARDIoGRAMplusC4D + UK Biobank (CAD) | 28714975 Nelson et al., 2017 | Mostly European | ||||
UK Biobank (MI) | Interim round two release Abbott et al., 2018 | European | 360,420 | http://www.nealelab.is/uk-biobank/ |
Aside from the concentration and content of lipoprotein subfractions, the two lipidome GWAS also measured the traditional lipid traits (TG, LDL-C, HDL-C), the average diameter of the fractions (VLDL-D, LDL-D, HDL-D) and the concentration of apolipoprotein A1 (ApoA1) and apolipoprotein B (ApoB). A full list of the lipoprotein measurements investigated in this article can be found in Appendix 1.
Genetic correlation and phenotypic screening
Request a detailed protocolGenetic correlation is a measure of association between the genetic determinants of two phenotypes. It is conceptually different from epidemiological correlation that can be directly estimated from cross-sectional data. In this study, we applied the LD-score regression (Bulik-Sullivan et al., 2015) to the lipidome GWAS (Kettunen et al., 2016; Davis et al., 2017) to estimate the genetic correlations between the lipoprotein subfractions, particle sizes, and traditional risk factors. We then removed lipoprotein subfractions and particle sizes that are strongly correlated with the traditional risk factors, defined as an estimated genetic correlation > 0.8 with TG, LDL-C, HDL-C, ApoB, or ApoA1 in the GWAS published by Davis et al., 2017. Because these traits are largely co-determined with the traditional risk factors, they do not represent independent biological mechanisms and may lead to multicollinearity issues in multivariate MR analyses. Finally, we obtained an independent estimate of the genetic correlations between the selected traits by applying the LD score regression to the GWAS published by Kettunen et al., 2016. We used Bonferroni's procedure to correct for multiple testing (familywise error rate at 0.05).
Three-sample Mendelian randomization design
Request a detailed protocolFor MR, we employed a three-sample design (Zhao et al., 2019b) in which one GWAS was used to select independent genetic instruments that are associated with one or several lipoprotein measures. The other two GWAS were then used to obtain summary associations of the selected SNPs with the exposure and the outcome, as in a typical two-sample MR design (Pierce and Burgess, 2013; Hemani et al., 2016). More specifically, the selection GWAS was used to create a set of SNPs that are in linkage equilibrium with each other in a reference panel (distance >10 megabase pairs, ). This was done by ordering the SNPs by the p-values of their association with the trait(s) under investigation and then selecting them greedily using the linkage-disequilibrium (LD) clumping function in the PLINK software package (Purcell et al., 2007). To avoid winner's curse, we require the other two GWAS to have no overlapping sample with the selection GWAS.
As the GWAS published by Davis et al., 2017 has a smaller sample size, we used it to select the genetic instruments so the larger dataset can be used for statistical estimation. In univariable MR, associations of the selected SNPs with the exposure trait (a lipoprotein subfraction or a particle size trait) were obtained from the GWAS published by Kettunen et al., 2016 and the associations with MI were obtained using summary data from an interim release of UK BioBank (Abbott et al., 2018). To maximize the statistical power, we used the so-called ‘genome-wide MR’ design. Independent SNPs are selected by using LD clumping, but we do not truncate the list of SNPs by their p-values. More details about this design can be found in a previous methodological article (Zhao et al., 2019b).
To control for potential pleiotropic effects via the traditional risk factors, we performed two multivariable MR analyses for each lipoprotein subfraction or particle size under investigation. The first multivariable MR analysis considers four exposures: TG, LDL-C, HDL-C, and the lipoprotein measurement under investigation. The second multivariable MR analysis replaces LDL-C and HDL-C with ApoB and ApoA1, in accordance with some recent studies (Richardson et al., 2020). SNPs were ranked by their minimum p-values with the four exposures and are selected as instruments only if they were associated with at least one of the four exposures (p-value ). Both multivariable MR analyses used the Davis (Davis et al., 2017) and GERA (Hoffmann et al., 2018) datasets for instrument selection, the Kettunen (Kettunen et al., 2016) and GLGC (Willer et al., 2013) datasets for the associations of the instruments with the exposures, and the CARDIoGRAMplusC4D + UK Biobank (Nelson et al., 2017) dataset for the associations with CAD.
Statistical estimation
Request a detailed protocolFor univariable MR, we used the robust adjusted profile score (RAPS) because it is more efficient and robust than many conventional methods (Zhao et al., 2020; Zhao et al., 2019b). RAPS can consistently estimate the causal effect even when some of the genetic variants violate instrumental variables assumptions. For multivariable MR, we used an extension to RAPS called GRAPPLE to obtain the causal effect estimates of multiple exposures (Wang et al., 2020). GRAPPLE also allows the exposure GWAS to have overlapping sample with the outcome GWAS, while the original RAPS does not. We assessed the strength of the instruments using the modified Cochran's Q statistic (Sanderson et al., 2019). Because many lipoprotein subfraction traits were analyzed simultaneously, we used the Benjamini-Hochberg procedure to correct for multiple testing (Benjamini and Hochberg, 1995) and the false discovery rate was set to be 0.05. More detail about the statistical methods can be found in Appendix 3.
Results of some multivariable Mendelian randomization analyses.
Each row in the table corresponds to a multivariable MR analysis with traditional lipid profile and the specified lipoprotein subfraction or particle size trait. Reported numbers are the point estimates and 95% confidence intervals of the exposure effect.
Trait | Effect of TG | Effect of LDL-C | Effect of HDL-C | Effect of subfraction/particle size |
---|---|---|---|---|
None | 0.19 [0.09,0.29] | 0.38 [0.33,0.44] | −0.053 [-0.13,0.03] | |
M-HDL-P | 0.37 [0.22,0.52] | 0.39 [0.32,0.45] | 0.30 [0.08,0.52] | −0.69 [-1.09,–0.3] |
S-HDL-P | 0.23 [0.12,0.33] | 0.45 [0.38,0.52] | −0.11 [-0.2,–0.02] | −0.33 [-0.52,–0.15] |
HDL-D | 0.11 [0.00,0.22] | 0.42 [0.36,0.49] | −0.44 [-0.69,–0.2] | 0.33 [0.11,0.56] |
Effect of TG | Effect of ApoB | Effect of ApoA1 | Effect of Subfraction/Particle size | |
None | 0.05 [-0.05,0.14] | 0.49 [0.38,0.60] | −0.095 [-0.21,0.02] | |
M-HDL-P | −0.00 [-0.18,0.17] | 0.50 [0.31,0.69] | 0.13 [-0.06,0.32] | −0.47 [-0.80,–0.15] |
S-HDL-P | 0.07 [-0.03,0.17] | 0.53 [0.41,0.65] | −0.13 [-0.25,–0.02] | −0.24 [-0.40,–0.08] |
HDL-D | 0.06 [-0.04,0.15] | 0.61 [0.47,0.76] | −0.46 [-0.73,–0.19] | 0.30 [0.08,0.52] |
Genetic markers for lipoprotein subfractions and CAD
Request a detailed protocolTo obtain genetic markers, we selected SNPs that are associated with the lipoprotein measurements identified in the MR (p-value ) and CAD (p-value ) but are not associated with LDL-C or ApoB (p-value ). To maximize the power of this exploratory analysis, we meta-analyzed the results of the two lipidome GWAS (Kettunen et al., 2016; Davis et al., 2017) by inverse-variance weighting. For the associations with LDL-C and CAD, we used the GWAS summary data reported by the GLGC (Willer et al., 2013) and CARDIoGRAMplusC4D (Nelson et al., 2017) consortia. We used LD clumping to obtain independent markers (Purcell et al., 2007) and then validate the markers using tissue-specific gene expression data from the GTEx project.
Sensitivity analysis and replicability
Request a detailed protocolBecause we had multiple GWAS summary datasets for the lipoprotein subfractions and CAD/MI (Table 1), we swapped the roles of the GWAS datasets in the three-sample MR design whenever permitted by the statistical methods to obtain multiple statistical estimates. These estimates are not completely independent of the primary results, but they can nonetheless be used to assess replicability. As a sensitivity analysis, We further analyzed univariable MR using inverse-variance weighting (IVW) (Burgess et al., 2013) and weighted median (Bowden et al., 2016) and compared with the primary results obtained by RAPS. We also assessed the assumptions made by RAPS using some diagnostic plots suggested in previous methodological articles (Zhao et al., 2019b).
Results
Genetic correlations and phenotypic screening
We obtained the genetic correlations of the lipoprotein subfractions and particle sizes with the traditional lipid risk factors: TG, LDL-C, HDL-C, ApoB, and ApoA1 (Table 1). We found that almost all VLDL subfractions traits (besides those related to very small VLDL subfraction) and the mean VLDL particle diameter have an estimated genetic correlation with TG very close to 1. Most traits related to the large and very large HDL subfractions also have a high genetic correlation with HDL-C and ApoA1.
After removing traits that are strongly correlated with the traditional risk factors, we obtained 27 traits that may involve independent genetic mechanisms. Figure 1 shows the genetic correlation matrix for these traits and the traditional lipid factors. The selected traits can be divided into two groups based on whether they are related to VLDL/LDL/IDL particles or HDL particles. Within each group, most traits were strongly correlated with the others. In the first group, most traits had a positive genetic correlation with LDL-C and ApoB, while in the second group, most traits had a positive genetic correlation with HDL-C and ApoA1. Exceptions include LDL-D, which had a negative but statistically non-significant genetic correlation with LDL-C and ApoB, and S-HDL-P and S-HDL-L, which showed no or weak genetic correlation with HDL-C and ApoA1.
Mendelian randomization
Figure 2 shows the estimated causal effect of the selected lipoprotein measurements on MI or CAD that are statistically significant (false discovery rate = 0.05). The unfiltered results can be found in Appendix 3, which also contains results of the sensitivity and replicability analyses.

Results of the Mendelian randomization analyses (false discover rate = 0.05): Estimated odds ratio [95% confidence interval] per standard deviation increase of the selected lipoprotein measurements on MI or CAD.
The concentration and lipid content of VLDL, LDL, and IDL subfractions showed harmful and nearly uniform effects on MI in univariable MR. However, after adjusting for the traditional lipid risk factors, the effects of these ApoB-related subfractions become close to zero (besides IDL-FC in one multivariable analysis). The mean diameter of LDL particles (LDL-D) showed a harmful effect on MI in univariable MR, though the effect was smaller than those of the LDL subfractions in univariable MR. The estimated effect of LDL-D was attenuated in the multivariable MR analyses.
The concentration and content of medium HDL particles showed protective effects in univariable and multivariable MR analyses. In particular, adjusting for the traditional lipid risk factors did not attenuate the effect of traits related to medium HDL. The concentration of and total lipid in small HDL particles showed protective effects in multivariable MR analyses, though the effect sizes were smaller than those of the medium HDL traits. The mean diameter of HDL particles (HDL-D) had almost no effect on MI in the univariable MR analysis, but after adjusting for the traditional lipid risk factors, it showed a harmful effect.
Table 2 reports the estimated effects of M-HDL-P, S-HDL-P, HDL-D, and traditional lipid traits (TG, LDL-C, HDL-C, ApoB, ApoA1) in the multivariable MR analyses. To better understand the role of HDL subfractions and particle sizes, we also included in the table the results of the multivariate MR analyses for the traditional lipid risk factors only. Those baseline analyses suggested that HDL-C/ApoA1 had a weak, non-significant protective effect on CAD, which is consistent with prior studies (Holmes et al., 2015; Wang et al., 2020). Adding S-HDL-P to the MR analysis did not substantially alter the estimated effects of the traditional lipid traits. However, when M-HDL-P or HDL-D was included in the model, the estimated effects of M-HDL-P and HDL-D changed substantially. In particular, when M-HDL-P was included in the multivariable MR analyses, HDL-C/ApoA1 showed a harmful effect on CAD. When HDL-D was included, HDL-C/ApoA1 showed a protective effect.
Genetic markers associated with HDL subfractions and CAD
We identified four genetic variants that are associated with S-HDL-P, M-HDL-P, or HDL-D, not associated with LDL-C or ApoB, and associated with CAD: rs838880 (SCARB1), rs737337 (DOCK6), rs2943641 (IRS1), and rs6065904 (PLTP) (Figure 3). These SNP-cis gene pairs are also supported by examining expression quantitative trait loci (eQTL) in the tissue-specific GTEx data (Appendix 4). The first three variants were not associated with S-HDL-P. However, they had uniformly positive associations with M-HDL-P, L-HDL-P, XL-HDL-P, HDL-D, ApoA1, and HDL-C, and a negative association with CAD. The last variant rs6065904 had positive associations with S-HDL-P and M-HDL-P, negative associations with L-HDL-P, XL-HDL-P, HDL-D, negative but smaller associations with ApoA1 and HDL-C, and a negative association with CAD.
Sensitivity and replicability analysis
We also investigated the effects of lipoprotein subfractions and particle sizes on MI/CAD using multiple GWAS datasets, MR designs and statistical methods. The results are provided in Appendix 3 and are generally in agreement with the primary results reported above. The diagnostic plots for S-HDL-P and M-HDL-P did not suggest evidence of violations of the instrument strength independent of direct effect (InSIDE) assumption (Bowden et al., 2015) made by RAPS and GRAPPLE (Appendix 4).
Discussion
By using recent genetic data and MR, this study examines whether some lipoprotein subfractions and particle sizes, beyond the traditional lipid risk factors, may play a role in coronary artery disease. We find that VLDL subfractions have extremely high genetic correlations with blood triglyceride level and thus offer little extra value. We find some weak evidence that larger LDL particle size may have a small harmful effect on myocardial infarction and coronary artery disease.
Our main finding is that the size of HDL particles may play an important and previously undiscovered role. Although the concentration and lipid content of small and medium HDL particles appear to be positively correlated with HDL cholesterol and ApoA1, their genetic correlations are much smaller than 1, indicating possible independent biological pathway(s). Moreover, the MR analyses suggested that the small and medium HDL particles may have protective effects on CAD. We also find that larger HDL mean particle diameter may have a harmful effect on CAD. Finally, we identified four potential genetic markers for HDL particle size that are independent of LDL cholesterol and ApoB.
There has been a heated debate on the role of HDL particles in CAD in recent years following the failure of several trials for CETP inhibitors (Barter et al., 2007; Schwartz et al., 2012; Lincoff et al., 2017) and recombinant ApoA1 (Nicholls et al., 2018) targeting HDL cholesterol. Observational epidemiology studies have long demonstrated strong inverse association between HDL cholesterol and the risk of CAD or MI (Miller and Miller, 1975; Lewington et al., 2007; Di Angelantonio et al., 2009), but conflicting evidence has been found in MR studies. In an influential study, Voight and collaborators found that the genetic variants associated with HDL cholesterol had varied associations with CAD and that almost all variants suggesting a protective effect of HDL cholesterol were also associated with LDL cholesterol or triglycerides (Voight et al., 2012). Other MR studies also found that the effect of HDL cholesterol on CAD is heterogeneous (Zhao et al., 2019b) or attenuated after adjusting for LDL cholesterol and triglycerides (Holmes et al., 2017; White et al., 2016).
Notice that the harmful effect of larger HDL particle diameter found in this study relies on including HDL-C or ApoA1 in the multivariable MR analysis. Thus, the role of HDL particles in preventing CAD may be more complicated than, for example, that of LDL cholesterol or ApoB. It is possible that HDL cholesterol, HDL subfractions, and HDL particle size are all phenotypic markers for some underlying causal mechanism. A related theory is the HDL function hypothesis (Rader and Hovingh, 2014). Cholesterol efflux capacity, a measure of HDL function, has been documented as superior to HDL-C in predicting CVD risk (Rohatgi et al., 2014; Saleheen et al., 2015). Recent epidemiologic studies found that HDL particle size is positively associated with cholesterol efflux capacity in post-menopausal women (El Khoudary et al., 2016) and in an asymptomatic older cohort (Mutharasan et al., 2017). However, mechanistic efflux studies showed that small HDL particles actually mediate more cholesterol efflux (Favari et al., 2009; Du et al., 2015). A likely explanation of this seeming contradiction is that a high concentration of small HDL particles in the serum may mark a block in maturation of small HDL particles (Mutharasan et al., 2017). This can also partly explain our finding that small HDL traits have a smaller effect than medium HDL traits, as increased medium HDL might indicate successful maturation of small HDL particles.
Among the reported genetic markers, SCARB1 and PLTP have established relations to HDL metabolism and CAD. SCARB1 encodes a plasma membrane receptor for HDL and is involved in hepatic uptake of cholesterol from peripheral tissues. Recently, a rare mutation (P376L) of SCARB1 was reported to raise HDL-C level and increase CAD risk (Zanoni et al., 2016; Samadi et al., 2019). This is opposite direction to the conventional belief that HDL-C is protective and could be explained by HDL dysfunction. PLTP encodes the phospholipid transfer protein and mediates the transfer of phospholipid and cholesterol from LDL and VLDL to HDL. As a result, PLTP plays a complex but pivotal role in HDL particle size and composition. Several studies have suggested that high PLTP activity is a risk factor for CAD (Schlitt et al., 2003; Schlitt et al., 2009; Zhao et al., 2019a).
Our study should be viewed in the context of its limitations, in particular, the inherent limitations of the summary-data MR design. Any causal inference from non-experimental data makes unverifiable assumptions, so does our study. Conventional MR studies assume that the genetic variants are valid instrumental variables. The statistical methods used by us make less stringent assumptions about the instrumental variables, but those assumptions could still be violated even though our model diagnosis does not suggest evidence against the InSIDE assumption. Our study did not adjust for other risk factors for CAD such as body mass index, blood pressure, and smoking. All the GWAS datasets used in this study are from the European population, so the same conclusions might not generalize to other populations. Furthermore, our study used GWAS datasets from heterogeneous subpopulations, which may also introduce bias (Zhao et al., 2019c). We also did not use more than one subfraction traits as exposures in multivariable MR because of their high genetic correlations. Alternative statistical methods could be used to select the best causal risk factor from high-throughput experiments (Zuber et al., 2019). Finally, as pointed out by revieweres, triglycerides has a greater intra-individual biological variability than HDL particle size. It is likely that triglycerides and HDL size represent a gene/environment interaction with a very large environmental component. Further investigations are needed to fully understand this mechanism.
Recently, a NMR spectroscopy method has been developed to estimate HDL cholesterol efflux capacity from serum (Kuusisto et al., 2019). That method can form the basis of a genetic analysis of HDL cholesterol efflux capacity and may complement the results here. We believe more laboratorial and epidemiological research is needed to clarify the roles of HDL subfractions and particle size in cardiovascular diseases.
Appendix 1
Lipid and lipoprotein traits
Two published GWAS of the human lipidome [Kettunen2016, Davis2017] measured lipoprotein subfractions and particle sizes using NMR spectroscopy. We investigated the 82 lipid and lipoprotein traits measured in these studies that are related to very-low-density lipoprotein (VLDL), LDL, and HDL subfractions and particle sizes. All the subfraction traits are named using three components separated by hyphen: the first indicates the size (XS, S, M, L, XL, XXL); the second indicates the category according to the lipoprotein density (VLDL, LDL, IDL, HDL); the third indicates the measurement (C for total cholesterol, CE for cholesterol esters, FC for free cholesterol, L for total lipids, P for particle concentration, PL for phospholipids, TG for triglycerides). A full list of lipid and lipoprotein traits used in our study can be found in Appendix 1—table 1 below.
All 82 traits included in this study and whether they are measured in the Kettunen and Davis GWAS (NA means not available).
Trait | Description | Kettunen | Davis |
---|---|---|---|
VLDL traits and total triglycerides | |||
TG | Total triglycerides | ||
VLDL-D | VLDL diameter | ||
XS-VLDL-L | Total lipids in very small VLDL | NA | |
XS-VLDL-P | Concentration of very small VLDL particles | ||
XS-VLDL-PL | Phospholipids in very small VLDL | ||
XS-VLDL-TG | Triglycerides in very small VLDL | ||
S-VLDL-C | Total cholesterol in small VLDL | ||
S-VLDL-FC | Free cholesterol in small VLDL | ||
S-VLDL-L | Total lipids in small VLDL | NA | |
S-VLDL-P | Concentration of small VLDL particles | ||
S-VLDL-PL | Phospholipids in small VLDL | ||
S-VLDL-TG | Triglycerides in small VLDL | ||
M-VLDL-C | Total cholesterol in medium VLDL | ||
M-VLDL-CE | Cholesterol esters in medium VLDL | ||
M-VLDL-FC | Free cholesterol in medium VLDL | ||
M-VLDL-L | Total lipids in medium VLDL | NA | |
M-VLDL-P | Concentration of medium VLDL particles | ||
M-VLDL-PL | Phospholipids in medium VLDL | ||
M-VLDL-TG | Triglycerides in medium VLDL | ||
L-VLDL-C | Total cholesterol in large VLDL | ||
L-VLDL-CE | Cholesterol esters in large VLDL | ||
L-VLDL-FC | Free cholesterol in large VLDL | ||
L-VLDL-L | Total lipids in large VLDL | NA | |
L-VLDL-P | Concentration of large VLDL particles | ||
L-VLDL-PL | Phospholipids in large VLDL | ||
L-VLDL-TG | Triglycerides in large VLDL | ||
XL-VLDL-L | Total lipids in very large VLDL | NA | |
XL-VLDL-P | Concentration of very large VLDL particles | ||
XL-VLDL-PL | Phospholipids in very large VLDL | ||
XL-VLDL-TG | Triglycerides in very large VLDL | ||
XXL-VLDL-L | Total lipids in chylomicrons and extremely very large VLDL | NA | |
XXL-VLDL-P | Concentration of chylomicrons and extremely very large VLDL particles | ||
XXL-VLDL-PL | Phospholipids in chylomicrons and extremely very large | ||
XXL-VLDL-TG | Triglycerides in chylomicrons and extremely very large | ||
LDL and IDL traits | |||
LDL-C | Total cholesterol in LDL | ||
ApoB | Apolipoprotein B | ||
LDL-D | LDL diameter | ||
S-LDL-C | Total cholesterol in small LDL | ||
S-LDL-L | Total lipids in small LDL | NA | |
S-LDL-P | Phospholipids in small LDL | ||
M-LDL-C | Total cholesterol in medium LDL | ||
M-LDL-CE | Cholesterol esters in medium LDL | ||
M-LDL-L | Total lipids in medium LDL | NA | |
M-LDL-P | Concentration of medium LDL particles | ||
M-LDL-PL | Phospholipids in medium LDL | ||
L-LDL-C | Total cholesterol in large LDL | ||
L-LDL-CE | Cholesterol esters in large LDL | ||
L-LDL-FC | Free cholesterol in large LDL | ||
L-LDL-L | Total lipids in large LDL | NA | |
L-LDL-P | Concentration of large LDL particles | ||
L-LDL-PL | Phospholipids in large LDL | ||
IDL-C | Total cholesterol in IDL | ||
IDL-FC | Free cholesterol in IDL | ||
IDL-L | Total lipids in IDL | NA | |
IDL-P | Concentration of IDL particles | ||
IDL-PL | Phospholipids in IDL | ||
IDL-TG | Triglycerides in IDL | ||
HDL traits | |||
HDL-C | Total cholesterol in HDL | ||
ApoA1 | Apolipoprotein A1 | ||
HDL-D | HDL diameter | ||
S-HDL-L | Total lipids in small HDL | NA | |
S-HDL-P | Concentration of small HDL particles | ||
S-HDL-TG | Triglycerides in small HDL | ||
M-HDL-C | Total cholesterol in medium HDL | ||
M-HDL-CE | Cholesterol esters in medium HDL | ||
M-HDL-FC | Free cholesterol in medium HDL | ||
M-HDL-L | Total lipids in medium HDL | NA | |
M-HDL-P | Concentration of medium HDL particles | ||
M-HDL-PL | Phospholipids in medium HDL | ||
L-HDL-C | Total cholesterol in large HDL | ||
L-HDL-CE | Cholesterol esters in large HDL | ||
L-HDL-FC | Free cholesterol in large HDL | ||
L-HDL-L | Total lipids in large HDL | NA | |
L-HDL-P | Concentration of large HDL particles | ||
L-HDL-PL | Phospholipids in large HDL | ||
XL-HDL-C | Total cholesterol in very large HDL | ||
XL-HDL-CE | Cholesterol esters in very large HDL | ||
XL-HDL-FC | Free cholesterol in very large HDL | ||
XL-HDL-L | Total lipids in very large HDL | NA | |
XL-HDL-P | Concentration of very large HDL particles | ||
XL-HDL-PL | Phospholipids in very large HDL | ||
XL-HDL-TG | Triglycerides in very large HDL |
Appendix 2
Genetic correlations
We estimated the genetic correlation between lipoprotein subfractions, particle sizes, and traditional lipid risk factors using the LD score regression (Li et al., 2016). Appendix 2—figure 1–3 show the estimated genetic correlation matrix between selected traits using different datasets. Below the figures, Appendix 2—table 1 shows the estimated genetic correlations of the lipoprotein subbfractions with the traditional lipid risk factors using the Davis GWAS. The results in Appendix 2—table 1 were then used to screen the traits as described in Materials and methods.

Genetic correlations computed using the Davis et al., 2017 GWAS summary dataset.

Genetic correlations computed using the Kettunen et al., 2016 GWAS summary dataset.

Genetic correlations computed by meta-analyzing the results in Appendix 2—figures 1 and 2.
Estimated genetic correlation (standard error) of the lipoprotein subfractions with the traditional lipid risk factors using the Davis GWAS.
Bolded estimates are above 0.8 and the corresponding traits were removed in phenotypic screening.
Trait | ApoA1 | ApoB | HDL-C | LDL-C | TG |
---|---|---|---|---|---|
S-HDL-L | 0.31 (0.28) | 0.34 (0.25) | 0.13 (0.26) | 0.27 (0.3) | 0.2 (0.22) |
S-HDL-P | 0.36 (0.24) | 0.27 (0.22) | −0.01 (0.22) | 0.1 (0.31) | 0.48 (0.17) |
S-HDL-TG | −0.13 (0.25) | 0.77 (0.13) | −0.66 (0.15) | 0.13 (0.28) | 1.03 (0.07) |
M-HDL-C | 0.65 (0.14) | −0.18 (0.2) | 0.81 (0.09) | −0.09 (0.25) | −0.34 (0.17) |
M-HDL-CE | 0.68 (0.14) | −0.23 (0.21) | 0.57 (0.12) | −0.24 (0.24) | −0.32 (0.18) |
M-HDL-FC | 0.67 (0.12) | −0.08 (0.21) | 0.83 (0.08) | 0.04 (0.24) | −0.28 (0.18) |
M-HDL-L | 0.71 (0.15) | 0.02 (0.27) | 0.52 (0.17) | −0.03 (0.29) | −0.19 (0.25) |
M-HDL-P | 0.75 (0.12) | 0.15 (0.23) | 0.46 (0.14) | 0.08 (0.26) | 0 (0.19) |
M-HDL-PL | 0.69 (0.13) | 0.04 (0.22) | 0.65 (0.11) | 0.02 (0.25) | −0.04 (0.19) |
L-HDL-C | 0.76 (0.11) | −0.42 (0.13) | 0.95 (0.02) | −0.1 (0.18) | −0.62 (0.09) |
L-HDL-CE | 0.82 (0.1) | −0.4 (0.12) | 0.93 (0.04) | −0.16 (0.17) | −0.62 (0.09) |
L-HDL-FC | 0.66 (0.12) | −0.46 (0.13) | 0.92 (0.03) | −0.13 (0.18) | −0.7 (0.08) |
L-HDL-L | 0.81 (0.11) | −0.29 (0.15) | 0.74 (0.07) | −0.15 (0.18) | −0.56 (0.12) |
L-HDL-P | 0.79 (0.09) | −0.35 (0.13) | 0.82 (0.05) | −0.12 (0.16) | −0.61 (0.09) |
L-HDL-PL | 0.77 (0.09) | −0.34 (0.13) | 0.79 (0.05) | −0.12 (0.17) | −0.61 (0.09) |
XL-HDL-C | 0.75 (0.16) | −0.25 (0.19) | 0.9 (0.1) | 0.4 (0.27) | −0.63 (0.13) |
XL-HDL-CE | 0.82 (0.16) | −0.17 (0.19) | 0.82 (0.09) | 0.41 (0.27) | −0.54 (0.12) |
XL-HDL-FC | 0.72 (0.14) | −0.37 (0.18) | 0.94 (0.08) | 0.17 (0.23) | −0.71 (0.11) |
XL-HDL-L | 0.93 (0.16) | −0.08 (0.25) | 0.68 (0.14) | 0.1 (0.27) | −0.35 (0.2) |
XL-HDL-P | 0.81 (0.13) | −0.32 (0.16) | 0.86 (0.08) | 0.17 (0.21) | −0.69 (0.11) |
XL-HDL-PL | 0.76 (0.12) | −0.41 (0.15) | 0.83 (0.07) | −0.09 (0.18) | −0.7 (0.09) |
XL-HDL-TG | 0.72 (0.13) | 0.49 (0.17) | 0.33 (0.13) | 0.13 (0.26) | 0.3 (0.15) |
HDL-D | 0.7 (0.11) | −0.36 (0.13) | 0.8 (0.06) | −0.08 (0.17) | −0.64 (0.09) |
IDL-C | 0.38 (0.21) | 0.58 (0.19) | 0.07 (0.19) | 0.8 (0.14) | 0.39 (0.17) |
IDL-FC | 0.23 (0.2) | 0.78 (0.12) | −0.05 (0.17) | 0.61 (0.19) | 0.44 (0.15) |
IDL-L | 0.38 (0.23) | 0.65 (0.18) | 0.05 (0.2) | 0.64 (0.2) | 0.47 (0.17) |
IDL-P | 0.31 (0.2) | 0.66 (0.14) | −0.04 (0.17) | 0.82 (0.13) | 0.49 (0.14) |
IDL-PL | 0.25 (0.23) | 0.83 (0.1) | −0.12 (0.19) | 0.7 (0.19) | 0.64 (0.15) |
IDL-TG | 0.22 (0.18) | 0.82 (0.08) | −0.2 (0.13) | 0.56 (0.15) | 0.67 (0.08) |
S-LDL-C | 0.11 (0.28) | 0.66 (0.18) | −0.16 (0.22) | 0.44 (0.34) | 0.58 (0.14) |
S-LDL-L | 0.26 (0.23) | 0.66 (0.17) | −0.06 (0.21) | 0.62 (0.21) | 0.58 (0.13) |
S-LDL-P | 0.34 (0.2) | 0.68 (0.15) | −0.02 (0.19) | 0.63 (0.18) | 0.58 (0.13) |
M-LDL-C | 0.15 (0.26) | 0.63 (0.18) | 0.22 (0.22) | 0.87 (0.08) | 0.13 (0.23) |
M-LDL-CE | 0.3 (0.23) | 0.61 (0.2) | 0.05 (0.21) | 0.65 (0.2) | 0.45 (0.16) |
M-LDL-L | 0.29 (0.22) | 0.63 (0.18) | 0.01 (0.21) | 0.66 (0.19) | 0.5 (0.15) |
M-LDL-P | 0.29 (0.23) | 0.63 (0.18) | −0.01 (0.21) | 0.65 (0.21) | 0.51 (0.15) |
M-LDL-PL | 0.2 (0.24) | 0.69 (0.16) | 0.11 (0.2) | 0.89 (0.06) | 0.18 (0.22) |
L-LDL-C | 0.25 (0.24) | 0.58 (0.21) | 0.25 (0.22) | 0.68 (0.19) | 0.23 (0.21) |
L-LDL-CE | 0.3 (0.23) | 0.58 (0.22) | 0.05 (0.21) | 0.65 (0.21) | 0.41 (0.17) |
L-LDL-FC | 0.31 (0.24) | 0.57 (0.22) | 0.33 (0.23) | 0.7 (0.18) | 0.13 (0.23) |
L-LDL-L | 0.31 (0.23) | 0.61 (0.2) | 0.04 (0.21) | 0.65 (0.21) | 0.44 (0.17) |
L-LDL-P | 0.31 (0.23) | 0.63 (0.19) | 0.02 (0.21) | 0.65 (0.21) | 0.47 (0.16) |
L-LDL-PL | 0.27 (0.25) | 0.61 (0.2) | 0.24 (0.22) | 0.67 (0.2) | 0.27 (0.2) |
LDL-D | −0.33 (0.25) | −0.22 (0.23) | −0.15 (0.21) | −0.15 (0.29) | −0.37 (0.16) |
XS-VLDL-L | 0.25 (0.23) | 0.8 (0.08) | −0.2 (0.17) | 0.61 (0.14) | 0.73 (0.09) |
XS-VLDL-P | 0.17 (0.18) | 0.83 (0.07) | −0.26 (0.13) | 0.57 (0.13) | 0.71 (0.07) |
XS-VLDL-PL | 0.21 (0.19) | 0.78 (0.09) | −0.15 (0.15) | 0.74 (0.14) | 0.57 (0.11) |
XS-VLDL-TG | 0.06 (0.18) | 0.83 (0.08) | −0.37 (0.11) | 0.56 (0.13) | 0.85 (0.04) |
S-VLDL-FC | −0.08 (0.2) | 0.94 (0.05) | −0.49 (0.12) | 0.59 (0.12) | 0.92 (0.03) |
S-VLDL-L | −0.12 (0.24) | 0.7 (0.08) | −0.46 (0.15) | 0.5 (0.14) | 0.8 (0.05) |
S-VLDL-P | −0.09 (0.19) | 0.78 (0.07) | −0.48 (0.11) | 0.5 (0.14) | 0.95 (0.02) |
S-VLDL-PL | −0.03 (0.2) | 0.82 (0.08) | −0.43 (0.12) | 0.44 (0.17) | 0.92 (0.03) |
S-VLDL-TG | −0.1 (0.2) | 0.9 (0.08) | −0.49 (0.11) | 0.49 (0.15) | 0.98 (0.01) |
S-VLDL-C | 0.01 (0.2) | 0.9 (0.06) | −0.39 (0.13) | 0.61 (0.15) | 0.89 (0.05) |
M-VLDL-C | −0.01 (0.2) | 0.8 (0.09) | −0.47 (0.12) | 0.41 (0.18) | 0.95 (0.02) |
M-VLDL-CE | 0.01 (0.19) | 0.78 (0.08) | −0.43 (0.12) | 0.5 (0.15) | 0.9 (0.03) |
M-VLDL-FC | 0 (0.21) | 0.83 (0.09) | −0.48 (0.12) | 0.4 (0.18) | 0.97 (0.01) |
M-VLDL-L | −0.1 (0.24) | 0.66 (0.11) | −0.48 (0.15) | 0.4 (0.18) | 0.8 (0.05) |
M-VLDL-P | −0.06 (0.19) | 0.78 (0.1) | −0.46 (0.12) | 0.43 (0.16) | 0.98 (0.02) |
M-VLDL-PL | 0.03 (0.21) | 0.85 (0.09) | −0.48 (0.12) | 0.4 (0.18) | 0.98 (0.01) |
M-VLDL-TG | −0.02 (0.21) | 0.82 (0.11) | −0.5 (0.13) | 0.33 (0.19) | 0.98 (0.02) |
L-VLDL-C | −0.05 (0.2) | 0.83 (0.12) | −0.55 (0.12) | 0.36 (0.19) | 1 (0.02) |
L-VLDL-CE | 0 (0.19) | 0.78 (0.12) | −0.44 (0.12) | 0.43 (0.19) | 0.93 (0.03) |
L-VLDL-FC | −0.03 (0.2) | 0.84 (0.12) | −0.53 (0.13) | 0.36 (0.19) | 1 (0.02) |
L-VLDL-L | −0.06 (0.24) | 0.66 (0.14) | −0.47 (0.16) | 0.36 (0.2) | 0.86 (0.05) |
L-VLDL-P | −0.02 (0.21) | 0.72 (0.12) | −0.44 (0.13) | 0.33 (0.18) | 0.98 (0.02) |
L-VLDL-PL | 0.01 (0.21) | 0.86 (0.12) | −0.53 (0.13) | 0.3 (0.2) | 1.04 (0.03) |
L-VLDL-TG | −0.06 (0.21) | 0.78 (0.12) | −0.54 (0.13) | 0.26 (0.19) | 1 (0.02) |
XL-VLDL-L | −0.08 (0.24) | 0.7 (0.15) | −0.52 (0.16) | 0.43 (0.2) | 0.85 (0.05) |
XL-VLDL-P | −0.06 (0.2) | 0.76 (0.12) | −0.48 (0.13) | 0.44 (0.18) | 0.95 (0.03) |
XL-VLDL-PL | −0.09 (0.23) | 0.82 (0.13) | −0.62 (0.15) | 0.32 (0.21) | 1.06 (0.04) |
XL-VLDL-TG | −0.14 (0.21) | 0.86 (0.13) | −0.65 (0.13) | 0.34 (0.19) | 1.03 (0.04) |
XXL-VLDL-L | −0.07 (0.25) | 0.65 (0.16) | −0.5 (0.17) | 0.38 (0.22) | 0.83 (0.06) |
XXL-VLDL-P | 0.17 (0.2) | 0.72 (0.15) | −0.3 (0.15) | 0.39 (0.21) | 0.86 (0.07) |
XXL-VLDL-PL | −0.3 (0.24) | 0.66 (0.17) | −0.8 (0.16) | 0.22 (0.21) | 1.06 (0.06) |
XXL-VLDL-TG | −0.21 (0.25) | 0.64 (0.16) | −0.7 (0.15) | 0.22 (0.22) | 1.08 (0.05) |
VLDL-D | −0.22 (0.2) | 0.55 (0.14) | −0.53 (0.12) | 0.12 (0.19) | 0.86 (0.04) |
Appendix 3
Mendelian randomization
We implemented several Mendelian randomization (MR) designs and statistical methods to estimate the causal effect of lipoprotein subfractions and particles sizes on coronary artery disease. In general, we adopted the three-sample summary data MR design described in Zhao et al., 2019b, Wang et al., 2020 and we swapped the roles of the GWAS datasets whenever permitted by the statistical methods. More specifically, the statistical methods we used for univariable MR (RAPS, IVW, weighted median) require that the GWAS datasets for obtaining instruments, SNP effects on the exposure, and SNP effects on the outcome must have no overlapping sample. The multivariable MR method we used (GRAPPLE) allows the exposure and outcome GWAS to be dependent and estimates the proportion of overlapping sample. However, GRAPPLE still requires that the selection GWAS uses an non-overlapping sample.
The MR designs we implemented in this study are summarized in Appendix 3—table 1. We considered two ways of instrument selection for univariable MR. In ‘traditional selection’, the traditional lipid traits were used to select the instruments for the corresponding subfraction traits. That is, HDL-C was used to select SNPs for HDL subfractions and particle size, LDL-C for IDL and LDL subfractions and particle size, and TG for VLDL subfractions and particle size. This tends to select more instruments because the GWAS for traditional lipid traits had a larger sample size. In ‘subfraction selection’, the instrumental SNPs were selected for each lipoprotein subfraction and particle size using the same or closest trait in the selection GWAS. For example, if the exposure under investigation is S-HDL-L but it is not measured in the Davis GWAS (if it is used for selection), S-HDL-P is used instead for instrument selection.
For multivariable MR, we considered two models with different sets of exposures: TG, LDL-C, HDL-C, and the subfraction/particle size under investigation; TG, ApoB, ApoA1, and the subfraction/particle size under investigation. SNPs were selected as potential instruments if they were associated (p-value ) with at least one of the four exposures. LD clumping was then used to obtain independent instruments, as described in Materials and Methods.
We briefly comment on the statistical methods used in univariable MR. All the three methods we used—RAPS, IVW, weighted median—require that the exposure GWAS and outcome GWAS have non-overlapping samples. RAPS and weighted median can provide consistent estimate of the causal effect even when some of the genetic variants are not valid instruments, provided that the direct effects of the genetic variants are independent of the strength of their associations with the exposure. The last condition is called the Instrument Strength Independent of Direct Effect (InSIDE) assumption in the MR literature [bowden2015mendelian]. RAPS is also robust to idiosyncratically large direct effect (Bowden et al., 2015). Because IVW and weighted median can be severely biased by weak instruments (Zhao et al., 2020), we only used them with the set of SNPs that have genome-wide significant association (p-value ) with the exposure. In comparison, RAPS does not suffer from weak instrument bias and we used it with all the SNPs obtained by LD clumping without any p-value threshold.
Below, Appendix 3—figure 1 shows the MR results for the 27 lipoprotein measurements selected in phenotypic screening. Estimates that are statistically significant at a false discovery rate of 0.05 are shown in Figure 2 of the main paper. Appendix 3—table 2 shows the estimated effect of all the lipoprotein subfractions and particle sizes on myocardial infarction or coronary artery disease in various MR designs. Full results of the multivariable MR analyses, including the estimated effects of the traditional lipid risk factors, can be found in Appendix 3—tables 5 and 6. The results of the univariable MR analyses using IVW and weighted median estimators can be found in Appendix 3—tables 3 and 4.
Three-sample Mendelian randomization designs.
MR design | Selection | Exposure | Outcome | Reported in |
---|---|---|---|---|
Univariable(traditional selection) | GERA | Davis | CARDIoGRAMplusC4D | Appendix 3—table 2–4 |
GERA | Davis | UK Biobank | Appendix 3—table 2–4 | |
GERA | Kettunen | UK Biobank | Appendix 3—table 2–4 | |
GLGC | Davis | UK Biobank | Appendix 3—table 2–4 | |
Univariable(subfraction selection) | Davis | Kettunen | UK Biobank | Figure 2; Appendix 3—figure 1 and Appendix 3—table 2–4 |
Kettunen | Davis | UK Biobank | Appendix 3—figure 1 and Appendix 3—table 2–4 | |
Multivariable | Davis, GERA | Kettunen, GLGC | CARDIoGRAMplusC4D+ UK Biobank | Figure 2, Table 2; Appendix 3—figure 1 and Appendix 3—table 2–4 |
Pooled results

Mendelian randomization results for the 27 lipoprotein measurements selected in phenotypic screening.
In the tables below, Red indicates p-value is significant (at level 0.05) after Bonferroni correction for all the results in the corresponding table and blue indicates p-value ≤ 0.05.
Mendelian randomization results using all selected SNPs (univariable MR using RAPS and multivariable MR using GRAPPLE).
Method: RAPS/GRAPPLE + All SNPs | ||||||||
---|---|---|---|---|---|---|---|---|
Screening | GERA | GERA | GERA | GLGC | Davis | Kettunen | GERA + Davis | GERA + Davis |
Exposure | Davis | Davis | Kettunen | Davis | Kettunen | Davis | GLGC + Kettunen | GLGC + Kettunen |
Outcome | CAD | UKB | UKB | UKB | UKB | UKB | CAD + UKB | CAD + UKB |
Adjusted | HDL-C + LDL-C + TG | ApoA1 + ApoB + TG | ||||||
VLDL traits | ||||||||
TG | NA | NA | NA | NA | ||||
VLDL-D | 0.028 (0.074) | 0.072 (0.073) | 0.116 (0.065) | |||||
XS-VLDL-L | NA | NA | NA | NA | 0.132 (0.119) | 0.084 (0.141) | ||
XS-VLDL-P | 0.118 (0.125) | 0.061 (0.158) | ||||||
XS-VLDL-PL | 0.159 (0.12) | 0.253 (0.135) | ||||||
XS-VLDL-TG | -0.157 (0.187) | -0.248 (0.15) | ||||||
S-VLDL-C | NA | NA | ||||||
S-VLDL-FC | ||||||||
S-VLDL-L | NA | NA | NA | NA | ||||
S-VLDL-P | ||||||||
S-VLDL-PL | ||||||||
S-VLDL-TG | NaN | |||||||
M-VLDL-C | 0.304 (0.078) | 0.312 (0.079) | ||||||
M-VLDL-CE | ||||||||
M-VLDL-FC | ||||||||
M-VLDL-L | NA | NA | 0.311 (0.079) | NA | NA | |||
M-VLDL-P | ||||||||
M-VLDL-PL | ||||||||
M-VLDL-TG | -0.806 (0.455) | |||||||
L-VLDL-C | -0.609 (0.337) | |||||||
L-VLDL-CE | 0.282 (0.072) | |||||||
L-VLDL-FC | ||||||||
L-VLDL-L | NA | NA | NA | NA | -0.532 (0.278) | |||
L-VLDL-P | -1.357 (0.344) | |||||||
L-VLDL-PL | 0.346 (0.089) | 0.397 (0.101) | ||||||
L-VLDL-TG | ||||||||
XL-VLDL-L | NA | NA | NA | NA | ||||
XL-VLDL-P | ||||||||
XL-VLDL-PL | ||||||||
XL-VLDL-TG | ||||||||
XXL-VLDL-L | NA | NA | NA | NA | ||||
XXL-VLDL-P | ||||||||
XXL-VLDL-PL | ||||||||
XXL-VLDL-TG | ||||||||
IDL/LDL traits | ||||||||
LDL-C | NA | 0.319 (0.182) | ||||||
ApoB | NA | |||||||
LDL-D | 0.271 (0.215) | 0.452 (0.299) | 0.831 (0.684) | 0.119 (0.071) | ||||
S-LDL-C | -0.282 (0.152) | 0.08 (0.238) | ||||||
S-LDL-L | NA | NA | NA | NA | -0.251 (0.145) | -0.005 (0.29) | ||
S-LDL-P | -0.266 (0.151) | -0.362 (0.596) | ||||||
M-LDL-C | -0.271 (0.162) | -0.169 (0.909) | ||||||
M-LDL-CE | -0.088 (0.188) | NaN | ||||||
M-LDL-L | NA | NA | NA | NA | -0.069 (0.191) | NaN | ||
M-LDL-P | -0.179 (0.174) | 0.432 (0.31) | ||||||
M-LDL-PL | -0.566 (0.839) | |||||||
L-LDL-C | -0.059 (0.261) | NaN | ||||||
L-LDL-CE | 0.116 (0.321) | |||||||
L-LDL-FC | 0.223 (0.315) | NaN | ||||||
L-LDL-L | NA | NA | NA | NA | 0.167 (0.273) | NaN | ||
L-LDL-P | 0.084 (0.213) | NaN | ||||||
L-LDL-PL | -0.036 (0.195) | NaN | ||||||
IDL-C | 0.192 (0.229) | 0.769 (0.501) | ||||||
IDL-FC | 0.19 (0.156) | |||||||
IDL-L | NA | NA | NA | NA | 0.148 (0.175) | 0.444 (0.328) | ||
IDL-P | 0.153 (0.148) | 0.292 (0.173) | ||||||
IDL-PL | 0.153 (0.18) | |||||||
IDL-TG | 0.11 (0.103) | 0.047 (0.135) | ||||||
HDL traits | ||||||||
HDL-C | -0.045 (0.059) | NA | NaN | |||||
ApoA1 | 0.023 (0.058) | 0.075 (0.064) | -0.13 (0.068) | -0.481 (0.271) | NA | |||
HDL-D | -0.008 (0.027) | 0.004 (0.046) | -0.092 (0.048) | 0.067 (0.045) | 0.007 (0.041) | |||
S-HDL-L | NA | NA | -0.098 (0.095) | NA | -0.037 (0.085) | NA | ||
S-HDL-P | -0.13 (0.092) | -0.053 (0.081) | -0.08 (0.094) | |||||
S-HDL-TG | 0.253 (0.637) | -0.044 (0.466) | ||||||
M-HDL-C | ||||||||
M-HDL-CE | ||||||||
M-HDL-FC | ||||||||
M-HDL-L | NA | NA | NA | NA | ||||
M-HDL-P | -0.307 (0.079) | |||||||
M-HDL-PL | ||||||||
L-HDL-C | -0.049 (0.045) | |||||||
L-HDL-CE | ||||||||
L-HDL-FC | -0.03 (0.047) | |||||||
L-HDL-L | NA | NA | NA | NA | ||||
L-HDL-P | -0.083 (0.05) | |||||||
L-HDL-PL | ||||||||
XL-HDL-C | 0.055 (0.046) | -0.013 (0.068) | 0.11 (0.066) | 0.064 (0.073) | 0.048 (0.069) | 0.112 (0.068) | ||
XL-HDL-CE | 0.064 (0.044) | 0.006 (0.066) | 0.08 (0.07) | 0.057 (0.068) | 0.046 (0.075) | |||
XL-HDL-FC | 0.009 (0.039) | -0.05 (0.059) | 0.066 (0.058) | -0.026 (0.067) | 0.102 (0.06) | 0.049 (0.066) | ||
XL-HDL-L | NA | NA | 0.073 (0.055) | NA | 0.038 (0.058) | NA | ||
XL-HDL-P | 0.038 (0.033) | -0.022 (0.049) | 0.112 (0.057) | 0.017 (0.056) | 0.083 (0.055) | 0.023 (0.057) | ||
XL-HDL-PL | 0.029 (0.031) | -0.031 (0.046) | 0.037 (0.05) | 0.005 (0.055) | 0.038 (0.052) | 0.013 (0.046) | ||
XL-HDL-TG | 0.165 (0.086) |
Univariable MR results
Mendelian randomization results using genome-wide significant SNPs and inverse variance weighted (IVW) estimator.
Method: IVW + Significant SNPs | ||||||
---|---|---|---|---|---|---|
Selection | GERA | GERA | GERA | GLGC | Davis | Kettunen |
Exposure | Davis | Davis | Kettunen | Davis | Kettunen | Davis |
Outcome | CAD | UKB | UKB | UKB | UKB | UKB |
VLDL traits | ||||||
TG | NA | NA | ||||
VLDL-D | 0.044 (0.06) | 0.052 (0.09) | 0.038 (0.102) | 0.118 (0.091) | -0.083 (0.16) | -0.083 (0.138) |
XS-VLDL-L | NA | NA | NA | NA | ||
XS-VLDL-P | ||||||
XS-VLDL-PL | ||||||
XS-VLDL-TG | ||||||
S-VLDL-C | NA | NA | ||||
S-VLDL-FC | ||||||
S-VLDL-L | NA | NA | NA | NA | ||
S-VLDL-P | ||||||
S-VLDL-PL | ||||||
S-VLDL-TG | ||||||
M-VLDL-C | ||||||
M-VLDL-CE | ||||||
M-VLDL-FC | ||||||
M-VLDL-L | NA | NA | NA | NA | ||
M-VLDL-P | ||||||
M-VLDL-PL | ||||||
M-VLDL-TG | ||||||
L-VLDL-C | 0.108 (0.223) | |||||
L-VLDL-CE | 0.182 (0.187) | |||||
L-VLDL-FC | ||||||
L-VLDL-L | NA | NA | NA | 0.125 (0.232) | NA | |
L-VLDL-P | ||||||
L-VLDL-PL | ||||||
L-VLDL-TG | ||||||
XL-VLDL-L | NA | NA | NA | 0.365 (0.286) | NA | |
XL-VLDL-P | 0.346 (0.28) | |||||
XL-VLDL-PL | 0.333 (0.265) | |||||
XL-VLDL-TG | 0.323 (0.272) | |||||
XXL-VLDL-L | NA | NA | NA | -0.165 (0.425) | NA | |
XXL-VLDL-P | -0.12 (0.389) | 0.006 (0.153) | ||||
XXL-VLDL-PL | -0.145 (0.395) | 0.071 (0.191) | ||||
XXL-VLDL-TG | 0.09 (0.36) | 0.349 (0.303) | ||||
IDL/LDL traits | ||||||
LDL-C | ||||||
ApoB | ||||||
LDL-D | 0.217 (0.151) | 0.271 (0.143) | ||||
S-LDL-C | ||||||
S-LDL-L | NA | NA | NA | NA | ||
S-LDL-P | ||||||
M-LDL-C | ||||||
M-LDL-CE | ||||||
M-LDL-L | NA | NA | NA | NA | ||
M-LDL-P | ||||||
M-LDL-PL | ||||||
L-LDL-C | ||||||
L-LDL-CE | ||||||
L-LDL-FC | ||||||
L-LDL-L | NA | NA | NA | NA | ||
L-LDL-P | ||||||
L-LDL-PL | ||||||
IDL-C | ||||||
IDL-FC | ||||||
IDL-L | NA | NA | NA | NA | ||
IDL-P | ||||||
IDL-PL | ||||||
IDL-TG | ||||||
HDL traits | ||||||
HDL-C | -0.085 (0.044) | -0.146 (0.085) | -0.082 (0.159) | -0.015 (0.109) | ||
ApoA1 | -0.072 (0.054) | -0.036 (0.09) | 0.001 (0.192) | 0.066 (0.158) | ||
HDL-D | -0.027 (0.042) | -0.071 (0.058) | -0.052 (0.073) | -0.092 (0.063) | 0.073 (0.098) | 0.074 (0.074) |
S-HDL-L | NA | NA | -0.064 (0.148) | NA | -0.033 (0.092) | NA |
S-HDL-P | -0.117 (0.087) | -0.172 (0.116) | -0.13 (0.146) | -0.033 (0.09) | -0.115 (0.174) | |
S-HDL-TG | 0.286 (0.17) | |||||
M-HDL-C | ||||||
M-HDL-CE | -0.341 (0.238) | |||||
M-HDL-FC | -0.288 (0.218) | -0.278 (0.144) | ||||
M-HDL-L | NA | NA | NA | NA | ||
M-HDL-P | -0.386 (0.307) | -0.18 (0.118) | ||||
M-HDL-PL | -0.419 (0.301) | |||||
L-HDL-C | -0.047 (0.044) | -0.097 (0.059) | -0.124 (0.08) | 0.022 (0.106) | 0.021 (0.105) | |
L-HDL-CE | -0.049 (0.044) | -0.098 (0.059) | -0.12 (0.079) | 0.023 (0.112) | 0.004 (0.106) | |
L-HDL-FC | -0.044 (0.046) | -0.094 (0.062) | -0.106 (0.082) | -0.127 (0.067) | 0.038 (0.103) | 0.017 (0.109) |
L-HDL-L | NA | NA | -0.106 (0.077) | NA | 0.034 (0.102) | NA |
L-HDL-P | -0.045 (0.043) | -0.097 (0.058) | -0.102 (0.077) | 0.009 (0.111) | 0.025 (0.11) | |
L-HDL-PL | -0.054 (0.044) | -0.11 (0.06) | -0.115 (0.079) | 0.006 (0.115) | 0.016 (0.115) | |
XL-HDL-C | 0.03 (0.06) | -0.012 (0.084) | 0.014 (0.099) | -0.05 (0.088) | -0.015 (0.165) | 0.161 (0.101) |
XL-HDL-CE | 0.03 (0.059) | -0.009 (0.081) | 0.025 (0.098) | -0.042 (0.086) | -0.001 (0.166) | |
XL-HDL-FC | -0.003 (0.056) | -0.05 (0.076) | -0.001 (0.089) | -0.077 (0.081) | 0.072 (0.11) | 0.057 (0.092) |
XL-HDL-L | NA | NA | 0.001 (0.085) | NA | -0.009 (0.138) | NA |
XL-HDL-P | 0.015 (0.049) | -0.021 (0.067) | 0.013 (0.088) | -0.042 (0.071) | 0.103 (0.1) | 0.135 (0.093) |
XL-HDL-PL | 0 (0.047) | -0.037 (0.065) | -0.026 (0.079) | -0.055 (0.069) | 0.081 (0.088) | 0.071 (0.069) |
XL-HDL-TG | 0.103 (0.059) | 0.14 (0.075) |
Mendelian randomization results using genome-wide significant SNPs and the weighted median estimator.
Method: Weighted median + Significant SNPs | ||||||
---|---|---|---|---|---|---|
Selection | GERA | GERA | GERA | GLGC | Davis | Kettunen |
Exposure | Davis | Davis | Kettunen | Davis | Kettunen | Davis |
Outcome | CAD | UKB | UKB | UKB | UKB | UKB |
VLDL traits | ||||||
TG | 0.042 (0.055) | NA | NA | |||
VLDL-D | -0.098 (0.052) | 0.039 (0.095) | 0.057 (0.11) | 0.058 (0.093) | -0.107 (0.099) | -0.052 (0.115) |
XS-VLDL-L | NA | NA | NA | NA | ||
XS-VLDL-P | ||||||
XS-VLDL-PL | ||||||
XS-VLDL-TG | ||||||
S-VLDL-C | NA | NA | ||||
S-VLDL-FC | ||||||
S-VLDL-L | NA | NA | NA | NA | ||
S-VLDL-P | ||||||
S-VLDL-PL | ||||||
S-VLDL-TG | ||||||
M-VLDL-C | ||||||
M-VLDL-CE | ||||||
M-VLDL-FC | 0.081 (0.058) | |||||
M-VLDL-L | NA | NA | NA | NA | ||
M-VLDL-P | 0.047 (0.06) | |||||
M-VLDL-PL | 0.103 (0.056) | |||||
M-VLDL-TG | -0.005 (0.06) | |||||
L-VLDL-C | 0.109 (0.068) | |||||
L-VLDL-CE | ||||||
L-VLDL-FC | 0.045 (0.065) | |||||
L-VLDL-L | NA | NA | NA | NA | ||
L-VLDL-P | 0.041 (0.064) | |||||
L-VLDL-PL | 0.08 (0.063) | |||||
L-VLDL-TG | -0.008 (0.061) | |||||
XL-VLDL-L | NA | NA | NA | NA | NA | |
XL-VLDL-P | -0.026 (0.063) | NA | ||||
XL-VLDL-PL | -0.006 (0.067) | NA | ||||
XL-VLDL-TG | -0.026 (0.064) | NA | ||||
XXL-VLDL-L | NA | NA | NA | -0.156 (0.22) | NA | |
XXL-VLDL-P | 0.091 (0.071) | -0.104 (0.173) | 0.185 (0.098) | |||
XXL-VLDL-PL | 0.153 (0.082) | -0.139 (0.178) | 0.126 (0.124) | |||
XXL-VLDL-TG | 0.126 (0.078) | 0.227 (0.171) | 0.23 (0.123) | |||
IDL/LDL traits | ||||||
LDL-C | ||||||
ApoB | ||||||
LDL-D | 0.271 (0.142) | |||||
S-LDL-C | ||||||
S-LDL-L | NA | NA | NA | NA | ||
S-LDL-P | ||||||
M-LDL-C | ||||||
M-LDL-CE | ||||||
M-LDL-L | NA | NA | NA | NA | ||
M-LDL-P | ||||||
M-LDL-PL | ||||||
L-LDL-C | ||||||
L-LDL-CE | ||||||
L-LDL-FC | ||||||
L-LDL-L | NA | NA | NA | NA | ||
L-LDL-P | ||||||
L-LDL-PL | ||||||
IDL-C | ||||||
IDL-FC | ||||||
IDL-L | NA | NA | NA | NA | ||
IDL-P | ||||||
IDL-PL | ||||||
IDL-TG | ||||||
HDL traits | ||||||
HDL-C | -0.017 (0.04) | -0.096 (0.077) | -0.085 (0.07) | |||
ApoA1 | 0.094 (0.049) | -0.06 (0.076) | -0.069 (0.087) | 0.005 (0.083) | -0.051 (0.121) | |
HDL-D | 0.062 (0.061) | 0.102 (0.064) | 0.088 (0.061) | 0.099 (0.061) | 0.096 (0.058) | |
S-HDL-L | NA | NA | -0.174 (0.113) | NA | NA | NA |
S-HDL-P | 0.018 (0.106) | -0.171 (0.109) | NA | -0.049 (0.108) | ||
S-HDL-TG | ||||||
M-HDL-C | ||||||
M-HDL-CE | ||||||
M-HDL-FC | ||||||
M-HDL-L | NA | NA | NA | NA | ||
M-HDL-P | -0.165 (0.131) | |||||
M-HDL-PL | -0.203 (0.12) | |||||
L-HDL-C | -0.009 (0.066) | 0.031 (0.083) | -0.032 (0.08) | 0.003 (0.09) | 0.006 (0.068) | |
L-HDL-CE | -0.011 (0.067) | 0.075 (0.077) | -0.037 (0.076) | 0.015 (0.091) | -0.006 (0.068) | |
L-HDL-FC | -0.005 (0.067) | 0.079 (0.081) | -0.019 (0.076) | 0.041 (0.078) | 0.027 (0.074) | |
L-HDL-L | NA | NA | 0.074 (0.077) | NA | 0.068 (0.084) | NA |
L-HDL-P | 0.046 (0.062) | 0.075 (0.074) | -0.01 (0.066) | 0.066 (0.07) | 0.078 (0.064) | |
L-HDL-PL | 0 (0.067) | 0.051 (0.082) | -0.021 (0.071) | 0.054 (0.075) | 0.074 (0.071) | |
XL-HDL-C | 0.122 (0.091) | 0.136 (0.087) | 0.132 (0.09) | 0.02 (0.098) | 0.161 (0.096) | |
XL-HDL-CE | 0.106 (0.088) | 0.122 (0.09) | 0.148 (0.085) | 0.038 (0.091) | ||
XL-HDL-FC | 0.065 (0.079) | 0.133 (0.081) | 0.027 (0.077) | 0.052 (0.086) | ||
XL-HDL-L | NA | NA | 0.119 (0.075) | NA | 0.023 (0.078) | NA |
XL-HDL-P | 0.087 (0.07) | 0.12 (0.073) | 0.129 (0.067) | |||
XL-HDL-PL | 0.064 (0.07) | 0.11 (0.072) | 0.121 (0.069) | 0.088 (0.065) | ||
XL-HDL-TG |
Multivariable MR results
Multivariable Mendelian randomization results (adjusted for HDL-C, LDL-C, and TG).
Trait | HDL-C | LDL-C | TG | Subfraction |
---|---|---|---|---|
VLDL traits | ||||
VLDL-D | -0.251 (0.052) | 0.29 (0.037) | 0.6 (0.087) | -0.588 (0.094) |
XS-VLDL-L | -0.086 (0.046) | 0.286 (0.077) | 0.089 (0.099) | 0.132 (0.119) |
XS-VLDL-P | -0.083 (0.045) | 0.299 (0.078) | 0.093 (0.106) | 0.118 (0.125) |
XS-VLDL-PL | -0.083 (0.046) | 0.249 (0.098) | 0.112 (0.076) | 0.159 (0.12) |
XS-VLDL-TG | -0.114 (0.046) | 0.463 (0.079) | 0.286 (0.173) | -0.157 (0.187) |
S-VLDL-C | -0.267 (0.084) | 0.754 (0.112) | 1.033 (0.28) | -1.035 (0.323) |
S-VLDL-FC | -0.195 (0.068) | 0.898 (0.163) | 0.935 (0.26) | -1.027 (0.337) |
S-VLDL-L | -0.25 (0.072) | 0.755 (0.112) | 0.876 (0.233) | -0.898 (0.28) |
S-VLDL-P | -0.31 (0.101) | 0.819 (0.157) | 1.209 (0.4) | -1.245 (0.463) |
S-VLDL-PL | -0.168 (0.051) | 0.673 (0.074) | 0.626 (0.159) | -0.613 (0.182) |
S-VLDL-TG | -0.499 (0.305) | 0.906 (0.34) | 2.532 (1.57) | -2.628 (1.741) |
M-VLDL-C | -0.201 (0.068) | 0.808 (0.127) | 1.472 (0.424) | -1.433 (0.451) |
M-VLDL-CE | -0.168 (0.061) | 0.799 (0.111) | 0.996 (0.249) | -1.035 (0.293) |
M-VLDL-FC | -0.2 (0.072) | 0.658 (0.089) | 1.469 (0.417) | -1.412 (0.444) |
M-VLDL-L | -0.355 (0.139) | 0.602 (0.096) | 1.787 (0.654) | -1.878 (0.75) |
M-VLDL-P | -0.362 (0.124) | 0.569 (0.08) | 1.889 (0.676) | -1.974 (0.745) |
M-VLDL-PL | -0.332 (0.141) | 0.722 (0.159) | 1.996 (0.869) | -2.012 (0.943) |
M-VLDL-TG | -0.408 (0.153) | 0.432 (0.061) | 1.974 (0.772) | -2.133 (0.879) |
L-VLDL-C | -0.216 (0.063) | 0.509 (0.046) | 1.163 (0.254) | -1.254 (0.297) |
L-VLDL-CE | -0.272 (0.072) | 0.465 (0.04) | 1.038 (0.242) | -1.081 (0.282) |
L-VLDL-FC | -0.144 (0.059) | 0.493 (0.044) | 1.233 (0.27) | -1.274 (0.308) |
L-VLDL-L | -0.228 (0.066) | 0.414 (0.045) | 1.17 (0.263) | -1.277 (0.313) |
L-VLDL-P | -0.115 (0.056) | 0.442 (0.046) | 1.351 (0.317) | -1.357 (0.344) |
L-VLDL-PL | -0.221 (0.111) | 0.473 (0.07) | 2.135 (0.948) | -2.316 (1.112) |
L-VLDL-TG | -0.196 (0.066) | 0.355 (0.05) | 1.357 (0.322) | -1.428 (0.372) |
XL-VLDL-L | -0.126 (0.049) | 0.451 (0.04) | 0.896 (0.159) | -1.069 (0.203) |
XL-VLDL-P | -0.127 (0.053) | 0.474 (0.043) | 1.038 (0.183) | -1.209 (0.238) |
XL-VLDL-PL | -0.138 (0.055) | 0.5 (0.044) | 1.052 (0.204) | -1.214 (0.257) |
XL-VLDL-TG | -0.129 (0.049) | 0.424 (0.04) | 0.944 (0.167) | -1.071 (0.205) |
XXL-VLDL-L | -0.228 (0.067) | 0.444 (0.043) | 0.978 (0.207) | -1.355 (0.318) |
XXL-VLDL-P | 0.063 (0.076) | 0.452 (0.05) | 1.371 (0.384) | -1.639 (0.502) |
XXL-VLDL-PL | -0.185 (0.056) | 0.371 (0.042) | 0.997 (0.185) | -1.259 (0.262) |
XXL-VLDL-TG | -0.152 (0.059) | 0.41 (0.04) | 0.966 (0.19) | -1.202 (0.262) |
LDL/IDL traits | ||||
ApoB | -0.084 (0.046) | 0.8 (0.146) | 0.427 (0.101) | -0.532 (0.191) |
LDL-D | -0.057 (0.042) | 0.367 (0.03) | 0.21 (0.053) | 0.145 (0.061) |
S-LDL-C | -0.062 (0.043) | 0.614 (0.126) | 0.261 (0.062) | -0.282 (0.152) |
S-LDL-L | -0.06 (0.044) | 0.584 (0.118) | 0.266 (0.068) | -0.251 (0.145) |
S-LDL-P | -0.033 (0.047) | 0.589 (0.119) | 0.29 (0.078) | -0.266 (0.151) |
M-LDL-C | -0.082 (0.044) | 0.623 (0.146) | 0.203 (0.054) | -0.271 (0.162) |
M-LDL-CE | -0.074 (0.043) | 0.485 (0.167) | 0.169 (0.059) | -0.088 (0.188) |
M-LDL-L | -0.071 (0.044) | 0.444 (0.171) | 0.19 (0.063) | -0.069 (0.191) |
M-LDL-P | -0.054 (0.044) | 0.539 (0.153) | 0.213 (0.063) | -0.179 (0.174) |
M-LDL-PL | -0.081 (0.045) | 0.747 (0.134) | 0.232 (0.062) | -0.407 (0.162) |
L-LDL-C | -0.071 (0.049) | 0.437 (0.242) | 0.167 (0.054) | -0.059 (0.261) |
L-LDL-CE | -0.07 (0.048) | 0.277 (0.301) | 0.149 (0.065) | 0.116 (0.321) |
L-LDL-FC | -0.112 (0.057) | 0.184 (0.304) | 0.163 (0.053) | 0.223 (0.315) |
L-LDL-L | -0.075 (0.049) | 0.229 (0.26) | 0.146 (0.068) | 0.167 (0.273) |
L-LDL-P | -0.083 (0.046) | 0.33 (0.2) | 0.128 (0.064) | 0.084 (0.213) |
L-LDL-PL | -0.101 (0.046) | 0.446 (0.177) | 0.155 (0.057) | -0.036 (0.195) |
IDL-C | -0.108 (0.057) | 0.231 (0.215) | 0.128 (0.064) | 0.192 (0.229) |
IDL-FC | -0.107 (0.05) | 0.23 (0.147) | 0.123 (0.056) | 0.19 (0.156) |
IDL-L | -0.1 (0.05) | 0.274 (0.161) | 0.123 (0.069) | 0.148 (0.175) |
IDL-P | -0.101 (0.047) | 0.269 (0.134) | 0.109 (0.071) | 0.153 (0.148) |
IDL-PL | -0.076 (0.048) | 0.25 (0.162) | 0.134 (0.071) | 0.153 (0.18) |
IDL-TG | -0.083 (0.046) | 0.314 (0.069) | 0.103 (0.089) | 0.11 (0.103) |
HDL traits | ||||
ApoA1 | 0.345 (0.25) | 0.544 (0.081) | 0.334 (0.109) | -0.481 (0.271) |
HDL-D | -0.442 (0.124) | 0.421 (0.033) | 0.111 (0.055) | 0.333 (0.114) |
S-HDL-L | -0.117 (0.046) | 0.488 (0.044) | 0.189 (0.054) | -0.312 (0.106) |
S-HDL-P | -0.112 (0.046) | 0.453 (0.035) | 0.225 (0.056) | -0.331 (0.095) |
S-HDL-TG | 0.002 (0.145) | 0.314 (0.156) | -0.007 (0.469) | 0.253 (0.637) |
M-HDL-C | 0.179 (0.097) | 0.36 (0.038) | 0.147 (0.054) | -0.508 (0.165) |
M-HDL-CE | 0.167 (0.087) | 0.319 (0.036) | 0.166 (0.055) | -0.487 (0.157) |
M-HDL-FC | 0.339 (0.141) | 0.436 (0.04) | 0.247 (0.059) | -0.649 (0.225) |
M-HDL-L | 0.27 (0.108) | 0.362 (0.032) | 0.299 (0.063) | -0.606 (0.188) |
M-HDL-P | 0.302 (0.112) | 0.386 (0.033) | 0.371 (0.075) | -0.694 (0.204) |
M-HDL-PL | 0.311 (0.117) | 0.402 (0.033) | 0.333 (0.07) | -0.632 (0.191) |
L-HDL-C | -0.589 (0.211) | 0.469 (0.039) | 0.146 (0.055) | 0.516 (0.213) |
L-HDL-CE | -0.602 (0.239) | 0.477 (0.042) | 0.137 (0.056) | 0.519 (0.23) |
L-HDL-FC | -0.573 (0.177) | 0.437 (0.034) | 0.171 (0.054) | 0.518 (0.181) |
L-HDL-L | -0.556 (0.193) | 0.437 (0.034) | 0.142 (0.055) | 0.457 (0.189) |
L-HDL-P | -0.515 (0.198) | 0.417 (0.03) | 0.133 (0.056) | 0.422 (0.191) |
L-HDL-PL | -0.53 (0.201) | 0.415 (0.034) | 0.152 (0.055) | 0.443 (0.202) |
XL-HDL-C | -0.447 (0.182) | 0.342 (0.036) | 0.071 (0.079) | 0.474 (0.223) |
XL-HDL-CE | -0.425 (0.146) | 0.366 (0.038) | 0.051 (0.069) | 0.426 (0.177) |
XL-HDL-FC | -0.459 (0.147) | 0.377 (0.031) | 0.097 (0.062) | 0.433 (0.16) |
XL-HDL-L | -0.405 (0.146) | 0.364 (0.031) | 0.077 (0.068) | 0.358 (0.154) |
XL-HDL-P | -0.451 (0.134) | 0.374 (0.03) | 0.078 (0.064) | 0.41 (0.139) |
XL-HDL-PL | -0.422 (0.119) | 0.412 (0.033) | 0.115 (0.055) | 0.343 (0.118) |
XL-HDL-TG | -0.186 (0.073) | 0.336 (0.035) | 0.045 (0.086) | 0.147 (0.074) |
Multivariable Mendelian randomization results (adjusted for ApoA1, ApoB, and TG).
Trait | ApoA1 | ApoB | TG | Subfraction |
---|---|---|---|---|
VLDL traits | ||||
VLDL-D | -0.227 (0.067) | 0.545 (0.092) | 0.208 (0.139) | -0.32 (0.112) |
XS-VLDL-L | -0.123 (0.063) | 0.53 (0.163) | -0.121 (0.085) | 0.084 (0.141) |
XS-VLDL-P | -0.121 (0.064) | 0.553 (0.17) | -0.123 (0.088) | 0.061 (0.158) |
XS-VLDL-PL | -0.147 (0.066) | 0.273 (0.138) | 0.028 (0.05) | 0.253 (0.135) |
XS-VLDL-TG | -0.102 (0.06) | 0.762 (0.168) | 0.069 (0.055) | -0.248 (0.15) |
S-VLDL-C | -0.384 (0.141) | 1.426 (0.354) | 0.606 (0.351) | -1.265 (0.568) |
S-VLDL-FC | -0.188 (0.077) | 1.001 (0.235) | 0.081 (0.053) | -0.489 (0.213) |
S-VLDL-L | -0.46 (0.146) | 1.776 (0.417) | 0.7 (0.316) | -1.629 (0.586) |
S-VLDL-P | -0.494 (0.159) | 1.677 (0.386) | 0.825 (0.372) | -1.644 (0.606) |
S-VLDL-PL | -0.262 (0.097) | 1.41 (0.343) | 0.532 (0.261) | -1.213 (0.478) |
S-VLDL-TG | -0.18 (0.069) | 0.792 (0.121) | 0.078 (0.051) | -0.301 (0.108) |
M-VLDL-C | -0.157 (0.062) | 0.867 (0.132) | 0.085 (0.051) | -0.373 (0.118) |
M-VLDL-CE | -0.221 (0.069) | 1.224 (0.223) | 0.47 (0.21) | -0.995 (0.338) |
M-VLDL-FC | -0.222 (0.074) | 0.902 (0.133) | 0.482 (0.251) | -0.799 (0.311) |
M-VLDL-L | -0.174 (0.065) | 0.76 (0.104) | 0.073 (0.05) | -0.298 (0.098) |
M-VLDL-P | -0.181 (0.065) | 0.764 (0.1) | 0.077 (0.051) | -0.312 (0.096) |
M-VLDL-PL | -0.159 (0.065) | 0.776 (0.116) | 0.08 (0.051) | -0.297 (0.106) |
M-VLDL-TG | -0.263 (0.106) | 0.724 (0.094) | 0.547 (0.406) | -0.806 (0.455) |
L-VLDL-C | -0.218 (0.084) | 0.732 (0.101) | 0.352 (0.278) | -0.609 (0.337) |
L-VLDL-CE | -0.293 (0.079) | 0.781 (0.096) | 0.405 (0.189) | -0.673 (0.217) |
L-VLDL-FC | -0.197 (0.069) | 0.737 (0.094) | 0.365 (0.25) | -0.619 (0.291) |
L-VLDL-L | -0.194 (0.071) | 0.666 (0.087) | 0.289 (0.234) | -0.532 (0.278) |
L-VLDL-P | -0.184 (0.061) | 0.677 (0.086) | 0.415 (0.217) | -0.617 (0.229) |
L-VLDL-PL | -0.155 (0.063) | 0.715 (0.095) | 0.075 (0.051) | -0.287 (0.104) |
L-VLDL-TG | -0.154 (0.062) | 0.67 (0.083) | 0.073 (0.05) | -0.252 (0.091) |
XL-VLDL-L | -0.186 (0.066) | 0.694 (0.088) | 0.263 (0.19) | -0.577 (0.249) |
XL-VLDL-P | -0.167 (0.061) | 0.742 (0.088) | 0.075 (0.05) | -0.373 (0.109) |
XL-VLDL-PL | -0.191 (0.068) | 0.712 (0.092) | 0.271 (0.197) | -0.583 (0.268) |
XL-VLDL-TG | -0.195 (0.068) | 0.666 (0.087) | 0.334 (0.21) | -0.603 (0.248) |
XXL-VLDL-L | -0.173 (0.066) | 0.732 (0.098) | 0.088 (0.052) | -0.402 (0.144) |
XXL-VLDL-P | -0.071 (0.065) | 0.705 (0.097) | 0.607 (0.321) | -1.089 (0.449) |
XXL-VLDL-PL | -0.244 (0.082) | 0.666 (0.091) | 0.414 (0.257) | -0.814 (0.344) |
XXL-VLDL-TG | -0.3 (0.091) | 0.694 (0.095) | 0.627 (0.306) | -1.075 (0.402) |
IDL/LDL traits | ||||
LDL-C | -0.119 (0.062) | 0.247 (0.167) | 0.066 (0.054) | 0.319 (0.182) |
LDL-D | -0.123 (0.06) | 0.544 (0.091) | -0.036 (0.087) | 0.119 (0.071) |
S-LDL-C | -0.097 (0.06) | 0.438 (0.216) | 0.044 (0.051) | 0.08 (0.238) |
S-LDL-L | -0.097 (0.063) | 0.503 (0.268) | 0.043 (0.051) | -0.005 (0.29) |
S-LDL-P | -0.059 (0.103) | 0.932 (0.597) | -0.122 (0.112) | -0.362 (0.596) |
M-LDL-C | -0.099 (0.065) | 0.78 (1.034) | -0.172 (0.425) | -0.169 (0.909) |
M-LDL-CE | -0.157 (0.128) | -0.346 (2.587) | 0.195 (0.855) | 0.854 (2.221) |
M-LDL-L | -0.123 (0.095) | 0.247 (1.479) | -0.001 (0.445) | 0.32 (1.293) |
M-LDL-P | -0.134 (0.07) | 0.13 (0.286) | 0.053 (0.052) | 0.432 (0.31) |
M-LDL-PL | -0.075 (0.077) | 1.165 (0.868) | -0.248 (0.253) | -0.566 (0.839) |
L-LDL-C | -0.855 (1.68) | -5.337 (13.402) | 2.405 (5.735) | 5.257 (11.72) |
L-LDL-CE | -0.151 (0.065) | 0.129 (0.193) | 0.061 (0.052) | 0.461 (0.213) |
L-LDL-FC | -0.397 (0.219) | -1.139 (1.395) | 0.786 (0.711) | 1.531 (1.189) |
L-LDL-L | -0.265 (0.148) | -0.854 (1.42) | 0.41 (0.51) | 1.266 (1.188) |
L-LDL-P | -0.258 (0.153) | -0.607 (1.225) | 0.276 (0.402) | 1.064 (1.029) |
L-LDL-PL | -0.312 (0.187) | -0.741 (1.411) | 0.39 (0.518) | 1.227 (1.245) |
IDL-C | -0.3 (0.123) | -0.334 (0.616) | 0.276 (0.254) | 0.769 (0.501) |
IDL-FC | -0.199 (0.069) | 0.247 (0.118) | 0.044 (0.049) | 0.33 (0.127) |
IDL-L | -0.215 (0.089) | 0.021 (0.409) | 0.101 (0.15) | 0.444 (0.328) |
IDL-P | -0.175 (0.075) | 0.214 (0.172) | 0.04 (0.051) | 0.292 (0.173) |
IDL-PL | -0.183 (0.07) | 0.159 (0.172) | 0.031 (0.049) | 0.406 (0.184) |
IDL-TG | -0.143 (0.075) | 0.565 (0.146) | -0.119 (0.087) | 0.047 (0.135) |
HDL traits | ||||
HDL-C | -1.513 (1.109) | 0.982 (0.314) | 0.27 (0.291) | 1.446 (1.112) |
HDL-D | -0.457 (0.138) | 0.613 (0.073) | 0.056 (0.049) | 0.296 (0.1) |
S-HDL-L | -0.128 (0.059) | 0.524 (0.062) | 0.067 (0.05) | -0.224 (0.087) |
S-HDL-P | -0.132 (0.059) | 0.531 (0.059) | 0.071 (0.05) | -0.24 (0.083) |
S-HDL-TG | -0.11 (0.113) | 0.595 (0.221) | -0.057 (0.297) | -0.044 (0.466) |
M-HDL-C | 0.091 (0.084) | 0.459 (0.101) | -0.1 (0.083) | -0.442 (0.143) |
M-HDL-CE | 0.09 (0.078) | 0.291 (0.083) | 0.082 (0.05) | -0.413 (0.137) |
M-HDL-FC | 0.148 (0.11) | 0.378 (0.063) | 0.066 (0.049) | -0.408 (0.166) |
M-HDL-L | 0.133 (0.091) | 0.491 (0.097) | -0.029 (0.086) | -0.485 (0.155) |
M-HDL-P | 0.129 (0.097) | 0.501 (0.097) | -0.004 (0.09) | -0.472 (0.166) |
M-HDL-PL | 0.162 (0.107) | 0.519 (0.096) | -0.037 (0.087) | -0.486 (0.171) |
L-HDL-C | -0.724 (0.232) | 0.856 (0.132) | 0.032 (0.093) | 0.575 (0.204) |
L-HDL-CE | -0.761 (0.236) | 0.899 (0.145) | 0.004 (0.084) | 0.61 (0.206) |
L-HDL-FC | -0.749 (0.174) | 0.842 (0.102) | 0.094 (0.05) | 0.59 (0.148) |
L-HDL-L | -0.717 (0.217) | 0.815 (0.12) | 0.023 (0.089) | 0.541 (0.184) |
L-HDL-P | -0.653 (0.191) | 0.749 (0.104) | 0.057 (0.049) | 0.476 (0.155) |
L-HDL-PL | -0.679 (0.201) | 0.774 (0.109) | 0.05 (0.049) | 0.51 (0.169) |
XL-HDL-C | -0.639 (0.194) | 0.692 (0.095) | -0.058 (0.086) | 0.565 (0.196) |
XL-HDL-CE | -0.576 (0.2) | 0.667 (0.096) | -0.077 (0.086) | 0.511 (0.206) |
XL-HDL-FC | -0.734 (0.174) | 0.674 (0.073) | 0.094 (0.052) | 0.609 (0.159) |
XL-HDL-L | -0.652 (0.168) | 0.733 (0.097) | -0.06 (0.084) | 0.481 (0.141) |
XL-HDL-P | -0.52 (0.147) | 0.691 (0.094) | -0.075 (0.084) | 0.39 (0.135) |
XL-HDL-PL | -0.652 (0.151) | 0.687 (0.076) | 0.079 (0.051) | 0.466 (0.12) |
XL-HDL-TG | -0.281 (0.111) | 0.539 (0.09) | -0.152 (0.092) | 0.165 (0.086) |
Q-statistics for multivariable Mendelian randomization
Here we provide the list of modified Cochran's Q-statistics for the multivariable MR analyses (Appendix 3—tables 7 and 8).
Modified Cochran’s Q-statistics (p-values) for the multivariable Mendelian randomization analyses (adjusted for HDL-C, LDL-C, and TG).
DF is short for degrees of freedom.
Trait | DF | HDL-C | LDL-C | TG | Subfraction |
---|---|---|---|---|---|
VLDL traits | |||||
VLDL-D | 432 | 7640.8 (0) | 1918.9 (7.9e-186) | 877.6 (1.4e-32) | 840.2 (1.6e-28) |
XS-VLDL-L | 436 | 7983.9 (0) | 1104.9 (1.1e-59) | 1935.8 (2.2e-187) | 926 (1.9e-37) |
XS-VLDL-P | 436 | 7927.8 (0) | 1066.6 (1.1e-54) | 1814 (4.8e-167) | 893.6 (9.6e-34) |
XS-VLDL-PL | 435 | 8291.5 (0) | 968.1 (1.4e-42) | 2771.5 (0) | 849.8 (4.3e-29) |
XS-VLDL-TG | 431 | 7549.8 (0) | 894.4 (1.3e-34) | 739.5 (1.3e-18) | 682.5 (1.2e-13) |
S-VLDL-C | 429 | 8598.1 (0) | 652.6 (1.7e-11) | 1220.7 (4.6e-77) | 541.3 (0.00018) |
S-VLDL-FC | 434 | 7861.2 (0) | 576 (5.4e-06) | 519.4 (0.003) | 507.9 (0.0082) |
S-VLDL-L | 438 | 7105.3 (0) | 626 (8.5e-09) | 525.2 (0.0026) | 514.3 (0.0069) |
S-VLDL-P | 438 | 6686.5 (0) | 616.5 (3.6e-08) | 515.6 (0.0061) | 507.3 (0.012) |
S-VLDL-PL | 437 | 7589.1 (0) | 702.8 (1e-14) | 591.5 (1.1e-06) | 555.1 (0.00011) |
S-VLDL-TG | 437 | 7658.7 (0) | 612.7 (5.3e-08) | 498.9 (0.021) | 494.5 (0.03) |
M-VLDL-C | 432 | 9167.8 (0) | 740.8 (1.3e-18) | 558.9 (3.5e-05) | 551.5 (8.3e-05) |
M-VLDL-CE | 432 | 8055.2 (0) | 705.9 (1.6e-15) | 556.6 (4.6e-05) | 539.7 (0.00031) |
M-VLDL-FC | 436 | 8272.8 (0) | 814.8 (2.7e-25) | 528.3 (0.0016) | 519.1 (0.0037) |
M-VLDL-L | 429 | 7109.2 (0) | 1269.2 (5.5e-84) | 532.6 (0.00047) | 515.9 (0.0025) |
M-VLDL-P | 436 | 8260.7 (0) | 2059.5 (2.1e-208) | 527.5 (0.0017) | 516.8 (0.0046) |
M-VLDL-PL | 435 | 6849.2 (0) | 599.6 (2.6e-07) | 496.8 (0.021) | 493.5 (0.027) |
M-VLDL-TG | 436 | 6123.7 (0) | 9854.8 (0) | 532.3 (0.0011) | 521 (0.0031) |
L-VLDL-C | 435 | 8617.2 (0) | 8966 (0) | 654.7 (4.3e-11) | 561.5 (3.9e-05) |
L-VLDL-CE | 434 | 6636.6 (0) | 11134 (0) | 581.6 (2.6e-06) | 539.5 (0.00041) |
L-VLDL-FC | 431 | 7779.6 (0) | 6691 (0) | 595.1 (2.5e-07) | 562.7 (1.9e-05) |
L-VLDL-L | 434 | 8104.9 (0) | 5191.4 (0) | 560.3 (3.9e-05) | 548.6 (0.00015) |
L-VLDL-P | 435 | 2308 (5.1e-252) | 10360.3 (0) | 545.4 (0.00024) | 537.9 (0.00054) |
L-VLDL-PL | 430 | 8155.4 (0) | 1310.8 (8.6e-90) | 491.8 (0.021) | 489.7 (0.024) |
L-VLDL-TG | 438 | 8581.8 (0) | 4800.1 (0) | 569.1 (2.3e-05) | 559.2 (7.5e-05) |
XL-VLDL-L | 437 | 8686.8 (0) | 8322.2 (0) | 674.7 (1.9e-12) | 620.2 (1.7e-08) |
XL-VLDL-P | 431 | 8550.2 (0) | 2459.4 (2e-280) | 608.3 (3.6e-08) | 588.6 (6.3e-07) |
XL-VLDL-PL | 431 | 7478.2 (0) | 5042.5 (0) | 613.3 (1.7e-08) | 591.6 (4.1e-07) |
XL-VLDL-TG | 433 | 8237.3 (0) | 9628.9 (0) | 651.8 (4.6e-11) | 618.3 (1.1e-08) |
XXL-VLDL-L | 439 | 8476.2 (0) | 10436.4 (0) | 652.9 (1.3e-10) | 570.7 (2.2e-05) |
XXL-VLDL-P | 437 | 1291.3 (2.8e-85) | 9987.4 (0) | 540.3 (0.00053) | 529.5 (0.0016) |
XXL-VLDL-PL | 436 | 9631.8 (0) | 11287.1 (0) | 641.6 (4.8e-10) | 595.5 (5.3e-07) |
XXL-VLDL-TG | 429 | 7809.4 (0) | 9476.4 (0) | 595.6 (1.7e-07) | 564 (1.2e-05) |
LDL/IDL traits | |||||
ApoB | 435 | 9220.8 (0) | 550.1 (0.00014) | 1809.7 (1.2e-166) | 535.1 (0.00072) |
LDL-D | 429 | 2909.2 (0) | 3918.8 (0) | 2706 (0) | 1426.1 (2.9e-107) |
S-LDL-C | 431 | 8189.7 (0) | 569.8 (7.8e-06) | 4880.9 (0) | 564.1 (1.6e-05) |
S-LDL-L | 435 | 8403.8 (0) | 574.4 (7.8e-06) | 3931.2 (0) | 564.3 (2.7e-05) |
S-LDL-P | 431 | 7371.4 (0) | 547.1 (0.00012) | 3144.7 (0) | 537.9 (0.00034) |
M-LDL-C | 430 | 9723.7 (0) | 570.9 (5.8e-06) | 6568.6 (0) | 562.9 (1.6e-05) |
M-LDL-CE | 432 | 8442.1 (0) | 558.3 (3.8e-05) | 5773.6 (0) | 549.1 (0.00011) |
M-LDL-L | 430 | 8801.7 (0) | 555.4 (4e-05) | 5176.1 (0) | 548.2 (9.5e-05) |
M-LDL-P | 429 | 8798.9 (0) | 541.6 (0.00018) | 5049.7 (0) | 535.2 (0.00035) |
M-LDL-PL | 436 | 7981.7 (0) | 573.9 (9.6e-06) | 4304.8 (0) | 558.9 (6e-05) |
L-LDL-C | 432 | 8865.2 (0) | 567.7 (1.2e-05) | 6179.8 (0) | 567 (1.3e-05) |
L-LDL-CE | 433 | 8464.3 (0) | 558.7 (4.1e-05) | 5731.3 (0) | 555.6 (5.9e-05) |
L-LDL-FC | 431 | 7481.1 (0) | 580.6 (1.9e-06) | 6760.8 (0) | 580.2 (2e-06) |
L-LDL-L | 433 | 8486.8 (0) | 604.5 (8.9e-08) | 5755.8 (0) | 601.8 (1.3e-07) |
L-LDL-P | 434 | 8310.7 (0) | 592.1 (6.3e-07) | 5553.3 (0) | 584.9 (1.7e-06) |
L-LDL-PL | 435 | 8341.4 (0) | 588.5 (1.2e-06) | 5327.8 (0) | 577.4 (5.3e-06) |
IDL-C | 434 | 7873.9 (0) | 645.5 (1.7e-10) | 6336 (0) | 642.1 (2.9e-10) |
IDL-FC | 432 | 8036 (0) | 729.5 (1.4e-17) | 6630.5 (0) | 725.6 (3e-17) |
IDL-L | 434 | 7869.8 (0) | 694.5 (2.4e-14) | 5198.3 (0) | 689 (7e-14) |
IDL-P | 436 | 9660.5 (0) | 736.7 (9e-18) | 5002 (0) | 726.6 (7.1e-17) |
IDL-PL | 431 | 8432.6 (0) | 680.6 (1.7e-13) | 5023 (0) | 677.4 (3e-13) |
IDL-TG | 436 | 7741.2 (0) | 1077.5 (4.2e-56) | 1992.9 (4.9e-197) | 931.6 (4.4e-38) |
HDL traits | |||||
ApoA1 | 434 | 494.1 (0.024) | 511.5 (0.006) | 932.1 (1.8e-38) | 492 (0.028) |
HDL-D | 438 | 783.5 (6.6e-22) | 8500 (0) | 5713.2 (0) | 860.1 (9.4e-30) |
S-HDL-L | 438 | 3067.3 (0) | 4414.6 (0) | 3763.2 (0) | 882.2 (3.7e-32) |
S-HDL-P | 438 | 2592.4 (1.1e-301) | 7652.1 (0) | 3097.3 (0) | 951.1 (4.9e-40) |
S-HDL-TG | 425 | 896.9 (6.9e-36) | 641.3 (5.2e-11) | 540.1 (0.00013) | 523 (8e-04) |
M-HDL-C | 437 | 957.6 (5.5e-41) | 10172.4 (0) | 4875.5 (0) | 628.3 (4.9e-09) |
M-HDL-CE | 434 | 955.3 (3.2e-41) | 1383.1 (1.7e-99) | 4355.4 (0) | 648.3 (1e-10) |
M-HDL-FC | 432 | 759.4 (2.4e-20) | 2989.1 (0) | 3512.2 (0) | 538.2 (0.00037) |
M-HDL-L | 435 | 914.2 (3e-36) | 11535.3 (0) | 2327.7 (1.7e-255) | 570.3 (1.3e-05) |
M-HDL-P | 434 | 997.6 (2.3e-46) | 10709.6 (0) | 1942.9 (3.2e-189) | 561.3 (3.4e-05) |
M-HDL-PL | 434 | 977.8 (6.3e-44) | 9439.9 (0) | 2566 (1.8e-298) | 581.3 (2.7e-06) |
L-HDL-C | 434 | 580 (3.2e-06) | 1257.1 (4.4e-81) | 4502.7 (0) | 604.3 (1.1e-07) |
L-HDL-CE | 434 | 549 (0.00014) | 930.2 (3e-38) | 5517.2 (0) | 557.2 (5.6e-05) |
L-HDL-FC | 441 | 627.6 (1.2e-08) | 8415.3 (0) | 3594 (0) | 658.4 (7.9e-11) |
L-HDL-L | 434 | 603.6 (1.2e-07) | 6743.8 (0) | 5314.7 (0) | 623.7 (5.7e-09) |
L-HDL-P | 432 | 601.1 (1.2e-07) | 7769.3 (0) | 6024.6 (0) | 633.2 (8.6e-10) |
L-HDL-PL | 434 | 584.5 (1.8e-06) | 9935.5 (0) | 3544.3 (0) | 611.3 (3.8e-08) |
XL-HDL-C | 430 | 732.9 (3.9e-18) | 10426.6 (0) | 2077.7 (1.4e-213) | 686.9 (4e-14) |
XL-HDL-CE | 430 | 771.4 (9.3e-22) | 8564.4 (0) | 2457 (2.2e-280) | 711.4 (3.3e-16) |
XL-HDL-FC | 432 | 761.8 (1.4e-20) | 11265.2 (0) | 2549.4 (3.1e-296) | 770.9 (1.9e-21) |
XL-HDL-L | 429 | 767.6 (1.6e-21) | 11490.7 (0) | 2355.7 (1.2e-262) | 784.6 (3.4e-23) |
XL-HDL-P | 433 | 724.9 (4.6e-17) | 11372.5 (0) | 2539.9 (3.9e-294) | 798.5 (4.8e-24) |
XL-HDL-PL | 443 | 809.7 (7.8e-24) | 10093.1 (0) | 5762 (0) | 895.4 (7.5e-33) |
XL-HDL-TG | 432 | 1849.1 (3.9e-174) | 2635.9 (6.5e-312) | 2240.8 (2.9e-241) | 1267.8 (4.4e-83) |
Modified Cochran’s Q-statistics (p-values) for the multivariable Mendelian randomization analyses (adjusted for ApoA1, ApoB, and TG).
DF is short for degrees of freedom.
Trait | DF | ApoA1 | ApoB | TG | Subfraction |
---|---|---|---|---|---|
VLDL traits | |||||
VLDL-D | 297 | 1194.1 (9.1e-108) | 550 (2.4e-17) | 573.7 (8.2e-20) | 606.7 (2.1e-23) |
XS-VLDL-L | 295 | 1185.1 (6.7e-107) | 927 (2e-66) | 1151.3 (2.2e-101) | 887.9 (1.1e-60) |
XS-VLDL-P | 295 | 1194.9 (1.7e-108) | 900 (1.9e-62) | 895.5 (8.7e-62) | 826.7 (6.4e-52) |
XS-VLDL-PL | 296 | 1148.5 (1.2e-100) | 973.9 (3.2e-73) | 2104.2 (1.4e-269) | 961.4 (2.5e-71) |
XS-VLDL-TG | 302 | 1263.7 (1.1e-117) | 757.9 (4.7e-41) | 1308.1 (4.4e-125) | 976.5 (4.6e-72) |
S-VLDL-C | 290 | 988.8 (4.4e-77) | 394 (4.5e-05) | 459.8 (7.8e-10) | 402.6 (1.3e-05) |
S-VLDL-FC | 296 | 1092 (1.4e-91) | 904 (8.6e-63) | 1238.7 (2.1e-115) | 1010.4 (8.1e-79) |
S-VLDL-L | 301 | 1107.9 (1.1e-92) | 412.3 (2.1e-05) | 420.8 (5.9e-06) | 384.7 (0.00078) |
S-VLDL-P | 301 | 1116.6 (4.6e-94) | 424.8 (3.3e-06) | 401.3 (9.4e-05) | 380.6 (0.0013) |
S-VLDL-PL | 299 | 1096 (2.3e-91) | 428.9 (1.2e-06) | 446 (7.1e-08) | 432.1 (7.1e-07) |
S-VLDL-TG | 300 | 1152.4 (4.3e-100) | 908.5 (1.8e-62) | 1453.4 (1.8e-150) | 1303.1 (7.1e-125) |
M-VLDL-C | 298 | 1171.2 (1e-103) | 824 (7.3e-51) | 1480 (8.9e-156) | 1212.5 (1.8e-110) |
M-VLDL-CE | 298 | 1185.4 (4.9e-106) | 564.4 (1.1e-18) | 468.9 (9.2e-10) | 431.6 (6.3e-07) |
M-VLDL-FC | 298 | 1190.4 (7.4e-107) | 899.8 (1.1e-61) | 415.2 (8.1e-06) | 398.8 (8.4e-05) |
M-VLDL-L | 298 | 1144.1 (2.4e-99) | 869.8 (2.4e-57) | 1381 (1e-138) | 1237.4 (1.4e-114) |
M-VLDL-P | 297 | 1121.3 (5.7e-96) | 821.1 (1.1e-50) | 1250.5 (4.6e-117) | 1206.7 (8.1e-110) |
M-VLDL-PL | 298 | 1149.9 (2.8e-100) | 843.2 (1.5e-53) | 1391.8 (1.5e-140) | 1226.3 (9.8e-113) |
M-VLDL-TG | 296 | 1187.4 (5.8e-107) | 717.3 (5.8e-37) | 366.3 (0.0033) | 360.6 (0.006) |
L-VLDL-C | 295 | 1196.5 (9.1e-109) | 820 (5.6e-51) | 462.5 (1.5e-09) | 376.9 (0.00088) |
L-VLDL-CE | 302 | 1183.1 (1.8e-104) | 844.6 (7.4e-53) | 541.8 (7.2e-16) | 441.7 (2.6e-07) |
L-VLDL-FC | 295 | 1172.3 (8.2e-105) | 851.6 (1.9e-55) | 460.8 (2.1e-09) | 406.2 (1.8e-05) |
L-VLDL-L | 295 | 1163.6 (2.2e-103) | 797 (8.8e-48) | 406.5 (1.7e-05) | 391.5 (0.00014) |
L-VLDL-P | 293 | 1160.2 (2e-103) | 809.5 (5.9e-50) | 420.2 (1.5e-06) | 407.9 (1e-05) |
L-VLDL-PL | 296 | 1292 (2.6e-124) | 833.4 (1.3e-52) | 1216.5 (9.7e-112) | 1098.9 (1.1e-92) |
L-VLDL-TG | 294 | 1150.8 (1.3e-101) | 1213.6 (7e-112) | 1262.6 (5.2e-120) | 1162.8 (1.5e-103) |
XL-VLDL-L | 294 | 1196 (5.4e-109) | 829.4 (1.6e-52) | 442 (4.9e-08) | 423.6 (1.1e-06) |
XL-VLDL-P | 294 | 1265.9 (1.4e-120) | 1180.9 (1.6e-106) | 1202.2 (5.2e-110) | 982.1 (5.4e-75) |
XL-VLDL-PL | 296 | 1199.1 (6.9e-109) | 874.2 (1.9e-58) | 421.2 (2.3e-06) | 405.6 (2.3e-05) |
XL-VLDL-TG | 296 | 1184.3 (1.8e-106) | 828.6 (5.9e-52) | 430.8 (4.9e-07) | 430.1 (5.5e-07) |
XXL-VLDL-L | 304 | 1119.2 (1.2e-93) | 1041.9 (1.6e-81) | 900.9 (2e-60) | 699.6 (3.2e-33) |
XXL-VLDL-P | 303 | 1148 (1.7e-98) | 876.4 (4e-57) | 382.2 (0.0013) | 366 (0.0076) |
XXL-VLDL-PL | 303 | 1203 (2.1e-107) | 775.1 (4e-43) | 438.1 (5.8e-07) | 376.5 (0.0025) |
XXL-VLDL-TG | 303 | 1183 (3.7e-104) | 881.8 (6.6e-58) | 393.7 (0.00034) | 372.7 (0.0039) |
LDL/IDL traits | |||||
LDL-C | 293 | 1198.7 (9.6e-110) | 938.8 (1.1e-68) | 1060.2 (2.1e-87) | 917.6 (1.5e-65) |
LDL-D | 296 | 1325.2 (6.7e-130) | 747.9 (5.9e-41) | 879.1 (3.7e-59) | 1163.5 (4.6e-103) |
S-LDL-C | 296 | 1195.3 (2.9e-108) | 706 (1.6e-35) | 1426 (4.1e-147) | 686.4 (4.8e-33) |
S-LDL-L | 296 | 1054.7 (1.1e-85) | 608 (1e-23) | 1519.6 (2.2e-163) | 586.4 (2.5e-21) |
S-LDL-P | 297 | 852.9 (3.6e-55) | 438.7 (1.6e-07) | 954.7 (4.5e-70) | 440.1 (1.3e-07) |
M-LDL-C | 296 | 1210.9 (8e-111) | 396.2 (8.6e-05) | 409 (1.4e-05) | 398.9 (6e-05) |
M-LDL-CE | 295 | 1204.3 (4.8e-110) | 350.8 (0.014) | 361.7 (0.0048) | 351.3 (0.013) |
M-LDL-L | 296 | 1212 (5.3e-111) | 370 (0.0022) | 392.3 (0.00015) | 371.6 (0.0019) |
M-LDL-P | 297 | 1125.4 (1.2e-96) | 623.9 (2.3e-25) | 911.4 (1.3e-63) | 582.4 (9.6e-21) |
M-LDL-PL | 299 | 1172.5 (1.2e-103) | 399.3 (9.1e-05) | 434.9 (4.5e-07) | 396.2 (0.00014) |
L-LDL-C | 300 | 1174.6 (1.1e-103) | 325.5 (0.15) | 325.5 (0.15) | 325.5 (0.15) |
L-LDL-CE | 299 | 1179.5 (9e-105) | 769.8 (3e-43) | 902.5 (7.7e-62) | 743.8 (8.4e-40) |
L-LDL-FC | 295 | 1161 (5.8e-103) | 322.4 (0.13) | 323.2 (0.12) | 322.3 (0.13) |
L-LDL-L | 300 | 1172.3 (2.6e-103) | 336.9 (0.07) | 349.6 (0.026) | 340.3 (0.055) |
L-LDL-P | 300 | 1185.4 (2e-105) | 352.1 (0.021) | 378.4 (0.0014) | 355.4 (0.015) |
L-LDL-PL | 296 | 1155.2 (9.8e-102) | 343.2 (0.031) | 360.1 (0.0063) | 344.5 (0.027) |
IDL-C | 296 | 1181.7 (4.9e-106) | 426.5 (9.8e-07) | 427.6 (8.3e-07) | 427.7 (8.1e-07) |
IDL-FC | 298 | 1096.5 (9.9e-92) | 986.9 (1.1e-74) | 1075.8 (1.9e-88) | 975.4 (6.1e-73) |
IDL-L | 296 | 1176.1 (4e-105) | 516.7 (3.3e-14) | 531 (1.4e-15) | 521.4 (1.2e-14) |
IDL-P | 297 | 1094.8 (9.5e-92) | 910.9 (1.5e-63) | 1103.9 (3.5e-93) | 890.2 (1.6e-60) |
IDL-PL | 297 | 1107.8 (8.3e-94) | 798.9 (1.3e-47) | 931.6 (1.3e-66) | 785.6 (8.6e-46) |
IDL-TG | 302 | 1060.8 (5.4e-85) | 1052.1 (1.2e-83) | 1092.6 (5.6e-90) | 1118.3 (4.7e-94) |
HDL traits | |||||
HDL-C | 298 | 318.7 (0.2) | 336.3 (0.063) | 329.1 (0.1) | 318.6 (0.2) |
HDL-D | 300 | 637.4 (1.9e-26) | 1156.6 (9.1e-101) | 2305.2 (1.3e-305) | 1183.8 (3.5e-105) |
S-HDL-L | 299 | 1597.7 (4.8e-176) | 1222.5 (8.2e-112) | 1916.4 (1.5e-233) | 1057 (3.1e-85) |
S-HDL-P | 299 | 1666.8 (2.5e-188) | 1249.4 (2.9e-116) | 2146.5 (3.4e-276) | 1103.3 (1.6e-92) |
S-HDL-TG | 299 | 899 (2.5e-61) | 464.9 (2.4e-09) | 464.5 (2.6e-09) | 457.6 (9.2e-09) |
M-HDL-C | 299 | 1145.2 (3.2e-99) | 768.2 (4.9e-43) | 951.8 (4e-69) | 786.8 (1.5e-45) |
M-HDL-CE | 299 | 1201.9 (2e-108) | 1183.9 (1.7e-105) | 2139.7 (6.4e-275) | 843.9 (1.9e-53) |
M-HDL-FC | 298 | 881.1 (5.6e-59) | 1252 (5.5e-117) | 1989.1 (2.4e-247) | 660.1 (1.8e-29) |
M-HDL-L | 299 | 1059 (1.5e-85) | 766.4 (8.7e-43) | 920.6 (1.7e-64) | 672.5 (8.6e-31) |
M-HDL-P | 298 | 990.2 (3.5e-75) | 760.4 (3.4e-42) | 1027.6 (6.2e-81) | 613.7 (4.7e-24) |
M-HDL-PL | 295 | 929.5 (8.3e-67) | 763.9 (2.7e-43) | 1057.2 (2.3e-86) | 588.3 (1.1e-21) |
L-HDL-C | 299 | 579.3 (4.1e-20) | 623.2 (5.7e-25) | 639.6 (7.3e-27) | 617.8 (2.3e-24) |
L-HDL-CE | 299 | 612.2 (1e-23) | 650.7 (3.6e-28) | 690.4 (5.5e-33) | 644 (2.2e-27) |
L-HDL-FC | 308 | 581.7 (4.4e-19) | 857.5 (2.6e-53) | 1213.3 (1.4e-107) | 915.8 (1.3e-61) |
L-HDL-L | 299 | 655.9 (8.7e-29) | 747.7 (2.6e-40) | 670.7 (1.4e-30) | 713.2 (7.5e-36) |
L-HDL-P | 298 | 591.3 (1.5e-21) | 934 (9.9e-67) | 1269.7 (6.2e-120) | 956.8 (3.9e-70) |
L-HDL-PL | 299 | 580 (3.4e-20) | 863.5 (3.3e-56) | 1262.4 (2.1e-118) | 891.8 (2.8e-60) |
XL-HDL-C | 298 | 475.3 (2.7e-10) | 734 (1e-38) | 976.1 (4.9e-73) | 554 (1.3e-17) |
XL-HDL-CE | 299 | 472.9 (5.4e-10) | 736.9 (6.7e-39) | 1117.4 (9e-95) | 517.5 (6.5e-14) |
XL-HDL-FC | 295 | 527.8 (2.1e-15) | 1182.8 (1.6e-106) | 2169.4 (3.1e-282) | 677.3 (4.3e-32) |
XL-HDL-L | 298 | 555.2 (9.6e-18) | 701.2 (1.6e-34) | 1014 (7.9e-79) | 775.3 (3.4e-44) |
XL-HDL-P | 300 | 578.9 (6.3e-20) | 744.5 (1.1e-39) | 1015.5 (1.6e-78) | 751.3 (1.4e-40) |
XL-HDL-PL | 306 | 604.9 (7.8e-22) | 1153.9 (1.4e-98) | 1899 (1.5e-227) | 909.3 (3.7e-61) |
XL-HDL-TG | 300 | 702.2 (2.8e-34) | 779.8 (2.2e-44) | 1140.8 (3.2e-98) | 1399.2 (3.7e-141) |
Appendix 4
Diagnostic plots and the genetic markers
As mentioned above, RAPS is more robust against invalid instruments than other statistical methods for univariable MR, but it still needs the InSIDE assumption to be approximately satisfied. Zhao et al., 2019b described two diagnostic plots RAPS that checks whether there is clear evidence that the InSIDE assumption is violated. Here, we report these plots for HDL-C and M-HDL-P in different studies (Appendix 4—figures 1 and 2). Notice that a lack of evidence to falsify the InSIDE assumption does not mean that it is true.
S-HDL-P
M-HDL-P
Genetic markers for M-HDL-P and S-HDL-P
We can further assess the validity of the InSIDE assumption for M-HDL-P and S-HDL-P but examining the associations of their genetic instruments with the traditional lipid risk factors and other subfraction traits. We meta-analyzed the summary results in the two lipidome GWAS (Davis and Kettunen) and obtained SNPs that are associated with S-HDL-P and M-HDL-P (p-value ; the results are LD-clumped). The next two Tables show some information about these genetic markers and their associations with other traits (Appendix 4—table 1 and 2).
Appendix 4—figures 3 and 4 shows how adjusting for LDL-C and TG changes the effects of the selected SNPs for S-HDL-P and M-HDL-P on CAD. The adjusted effect on CAD is obtained by original effect on CAD – 0.45 * effect on LDL-C – 0.25 * effect on TG. After the adjustment, the associations of the genetic variants with CAD generally became closer to the fitted lines that correspond to the estimated effects of S-HDL-P and M-HDL-P.
List of SNPs associated with M-HDL-P.
SNP | Chr | Gene | S-HDL-P | M-HDL-P | L-HDL-P | XL-HDL-P | HDL-C | LDL-C | TG | CAD |
---|---|---|---|---|---|---|---|---|---|---|
rs11208004 | 1 | DOCK7 | -0.039 ** | -0.075 *** | -0.015 | -0.002 | -0.015 ** | -0.050 *** | -0.069 *** | -0.012 |
rs4846913 | 1 | GALNT2 | -0.000 | -0.061 *** | -0.062 *** | -0.023 . | -0.055 *** | -0.006 | -0.044 *** | -0.025 . |
rs2126259 | 8 | LOC157273 | -0.066 *** | -0.082 *** | -0.063 ** | -0.025 . | -0.075 *** | -0.063 *** | -0.016 . | -0.004 |
rs2083637 | 8 | LPL | -0.001 | -0.058 *** | -0.092 *** | -0.053 ** | -0.105 *** | -0.008 | -0.108 *** | -0.047 ** |
rs10468017 | 15 | ALDH1A2/LIPC | -0.096 *** | -0.060 *** | -0.209 *** | -0.202 *** | -0.118 *** | -0.002 | -0.038 *** | -0.013 |
rs247616 | 16 | CETP | -0.058 *** | -0.121 *** | -0.198 *** | -0.129 *** | -0.243 *** | -0.055 *** | -0.039 *** | -0.044 ** |
rs1943973 | 18 | LIPG | -0.022 | -0.108 *** | -0.104 *** | -0.078 *** | -0.077 *** | -0.024 ** | -0.009 | -0.016 |
rs737337 | 19 | DOCK6 | -0.047 . | -0.087 *** | -0.081 ** | -0.058 * | -0.056 *** | -0.007 | -0.011 | -0.038 . |
rs769449 | 19 | APOE | -0.016 | -0.078 *** | -0.071 *** | -0.015 | -0.064 *** | -0.214 *** | -0.042 *** | -0.085 *** |
rs7679 | 20 | PCIF1/PLTP | -0.188 *** | -0.071 *** | -0.129 *** | -0.152 *** | -0.059 *** | -0.009 | -0.051 *** | -0.025 . |
List of SNPs associated with S-HDL-P.
SNP | Chr | Gene | S-HDL-P | M-HDL-P | L-HDL-P | XL-HDL-P | HDL-C | LDL-C | TG | CAD |
---|---|---|---|---|---|---|---|---|---|---|
rs780094 | 2 | GCKR | -0.074 *** | -0.034 * | -0.04 ** | -0.034 * | -0.011 . | -0.021 ** | -0.110 *** | -0.005 |
rs10935473 | 3 | ST3GAL6-AS1 | -0.052 *** | -0.014 | -0.029 . | -0.031 * | -0.009 . | -0.003 | -0.005 | -0.007 |
rs4936363 | 11 | SIK3 | -0.064 *** | -0.046 ** | -0.019 | -0.006 | -0.034 ** | -0.018 . | -0.043 *** | -0.022 |
rs2043085 | 15 | ALDH1A2/LIPC | -0.092 *** | -0.056 *** | -0.202 *** | -0.197 *** | -0.106 *** | -0.003 | -0.033 *** | -0.008 |
rs1800588 | 15 | ALDH1A2/LIPC | -0.106 *** | -0.050 ** | -0.215 *** | -0.212 *** | -0.114 *** | -0.002 | -0.044 *** | -0.015 |
rs289714 | 16 | CETP | -0.077 *** | -0.122 *** | -0.162 *** | -0.102 *** | -0.214 *** | -0.036 *** | -0.035 *** | -0.012 |
rs6065904 | 20 | PLTP | -0.171 *** | -0.060 *** | -0.127 *** | -0.149 *** | -0.052 *** | -0.008 | -0.040 *** | -0.022 . |

Scatter-plots for S-HDL-P with the effects on CAD adjusted for LDL-C and TG.
Red lines correspond the fitted effects of S-HDL-P in multivariable MR.
Gene expression
Here we provide evidence of variant-gene associations from Quantatitive Trait Locus (QTL) analyses in the GTEx project (Appendix 4—table 3).
Tissue-specific gene expressions associated with the 4 discovered genetic markers in the GTEx project.
SNP.Id | Type | Gene.Symbol | Variant.Id | p value | Effect | Tissue |
---|---|---|---|---|---|---|
rs838880 | eQTL | SCARB1 | chr12_124777047_C_T_b38 | 1.5E-08 | -0.20 | Cells - Cultured fibroblasts |
rs838880 | sQTL | SCARB1 | chr12_124777047_C_T_b38 | 4.1E-06 | -0.34 | Testis |
rs737337 | sQTL | DOCK6 | chr19_11236817_T_C_b38 | 3.8E-43 | 0.99 | Artery - Tibial |
rs737337 | sQTL | DOCK6 | chr19_11236817_T_C_b38 | 6.4E-35 | 0.93 | Adipose - Subcutaneous |
rs737337 | sQTL | DOCK6 | chr19_11236817_T_C_b38 | 6.4E-35 | 0.93 | Adipose - Subcutaneous |
rs737337 | sQTL | DOCK6 | chr19_11236817_T_C_b38 | 1.6E-27 | 0.95 | Esophagus - Muscularis |
rs737337 | sQTL | DOCK6 | chr19_11236817_T_C_b38 | 3.2E-20 | 1.10 | Colon - Sigmoid |
rs737337 | sQTL | DOCK6 | chr19_11236817_T_C_b38 | 1.1E-17 | 0.93 | Esophagus - Gastroesophageal Junction |
rs737337 | sQTL | DOCK6 | chr19_11236817_T_C_b38 | 1.8E-09 | 0.81 | Artery - Coronary |
rs737337 | sQTL | DOCK6 | chr19_11236817_T_C_b38 | 1.2E-07 | -0.49 | Thyroid |
rs737337 | sQTL | KANK2 | chr19_11236817_T_C_b38 | 4.4E-07 | 0.43 | Artery - Tibial |
rs737337 | sQTL | KANK2 | chr19_11236817_T_C_b38 | 3.5E-06 | 0.55 | Heart - Left Ventricle |
rs2943641 | eQTL | IRS1 | chr2_226229029_T_C_b38 | 1.4E-16 | -0.30 | Adipose - Subcutaneous |
rs2943641 | eQTL | IRS1 | chr2_226229029_T_C_b38 | 6.1E-12 | -0.23 | Adipose - Visceral (Omentum) |
rs2943641 | eQTL | RP11-395N3.2 | chr2_226229029_T_C_b38 | 3.5E-09 | -0.23 | Adipose - Subcutaneous |
rs2943641 | eQTL | RP11-395N3.1 | chr2_226229029_T_C_b38 | 2.1E-07 | -0.23 | Adipose - Subcutaneous |
rs2943641 | eQTL | RP11-395N3.2 | chr2_226229029_T_C_b38 | 2.3E-06 | -0.19 | Adipose - Visceral (Omentum) |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 4.4E-22 | -0.27 | Muscle - Skeletal |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 1.6E-16 | -0.27 | Adipose - Subcutaneous |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 1.2E-15 | -0.28 | Adipose - Visceral (Omentum) |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 3.2E-15 | -0.42 | Heart - Atrial Appendage |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 7.2E-14 | -0.25 | Artery - Tibial |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 1.8E-12 | -0.27 | Nerve - Tibial |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 7.3E-12 | -0.26 | Esophagus - Muscularis |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 2.0E-11 | -0.29 | Colon - Transverse |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 4.1E-11 | -0.32 | Colon - Sigmoid |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 1.2E-09 | -0.26 | Artery - Aorta |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 4.2E-09 | -0.29 | Heart - Left Ventricle |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 5.0E-09 | -0.22 | Thyroid |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 1.7E-08 | -0.29 | Stomach |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 4.3E-08 | -0.24 | Lung |
rs6065904 | eQTL | NEURL2 | chr20_45906012_G_A_b38 | 6.6E-08 | -0.26 | Adipose - Subcutaneous |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 6.8E-08 | -0.33 | Liver |
rs6065904 | eQTL | CTSA | chr20_45906012_G_A_b38 | 4.0E-07 | -0.14 | Nerve - Tibial |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 5.3E-07 | -0.37 | Spleen |
rs6065904 | eQTL | NEURL2 | chr20_45906012_G_A_b38 | 5.6E-07 | -0.26 | Adipose - Visceral (Omentum) |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 8.9E-07 | -0.46 | Small Intestine - Terminal Ileum |
rs6065904 | eQTL | RP3-337O18.9 | chr20_45906012_G_A_b38 | 1.8E-06 | -0.22 | Adipose - Subcutaneous |
rs6065904 | eQTL | WFDC3 | chr20_45906012_G_A_b38 | 2.9E-06 | -0.31 | Nerve - Tibial |
rs6065904 | eQTL | DNTTIP1 | chr20_45906012_G_A_b38 | 3.1E-06 | -0.17 | Artery - Tibial |
rs6065904 | eQTL | WFDC3 | chr20_45906012_G_A_b38 | 4.5E-06 | -0.27 | Skin - Sun Exposed (Lower leg) |
rs6065904 | eQTL | SNX21 | chr20_45906012_G_A_b38 | 4.8E-06 | -0.15 | Esophagus - Muscularis |
rs6065904 | eQTL | WFDC3 | chr20_45906012_G_A_b38 | 8.9E-06 | -0.27 | Skin - Not Sun Exposed (Suprapubic) |
rs6065904 | eQTL | DNTTIP1 | chr20_45906012_G_A_b38 | 1.0E-05 | -0.14 | Nerve - Tibial |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 1.1E-05 | -0.27 | Prostate |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 1.3E-05 | -0.26 | Pituitary |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 1.4E-05 | -0.21 | Esophagus - Gastroesophageal Junction |
rs6065904 | eQTL | SNX21 | chr20_45906012_G_A_b38 | 1.5E-05 | -0.16 | Esophagus - Mucosa |
rs6065904 | eQTL | SNX21 | chr20_45906012_G_A_b38 | 1.7E-05 | -0.23 | Colon - Sigmoid |
rs6065904 | eQTL | SNX21 | chr20_45906012_G_A_b38 | 1.7E-05 | -0.17 | Thyroid |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 2.6E-05 | -0.21 | Breast - Mammary Tissue |
rs6065904 | eQTL | WFDC3 | chr20_45906012_G_A_b38 | 2.9E-05 | -0.23 | Artery - Tibial |
rs6065904 | eQTL | NEURL2 | chr20_45906012_G_A_b38 | 3.2E-05 | -0.21 | Thyroid |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 3.7E-05 | -0.17 | Testis |
rs6065904 | eQTL | CTSA | chr20_45906012_G_A_b38 | 4.4E-05 | -0.11 | Skin - Not Sun Exposed (Suprapubic) |
rs6065904 | eQTL | WFDC3 | chr20_45906012_G_A_b38 | 5.8E-05 | -0.23 | Muscle - Skeletal |
rs6065904 | eQTL | NEURL2 | chr20_45906012_G_A_b38 | 8.2E-05 | -0.27 | Heart - Atrial Appendage |
rs6065904 | eQTL | SNX21 | chr20_45906012_G_A_b38 | 8.4E-05 | -0.17 | Artery - Aorta |
rs6065904 | eQTL | NEURL2 | chr20_45906012_G_A_b38 | 9.5E-05 | -0.24 | Artery - Aorta |
rs6065904 | eQTL | WFDC3 | chr20_45906012_G_A_b38 | 9.5E-05 | -0.31 | Artery - Aorta |
rs6065904 | eQTL | RP3-337O18.9 | chr20_45906012_G_A_b38 | 9.5E-05 | -0.29 | Heart - Atrial Appendage |
rs6065904 | eQTL | PLTP | chr20_45906012_G_A_b38 | 1.2E-04 | -0.15 | Skin - Sun Exposed (Lower leg) |
rs6065904 | eQTL | WFDC13 | chr20_45906012_G_A_b38 | 1.5E-04 | 0.28 | Esophagus - Muscularis |
rs6065904 | eQTL | DNTTIP1 | chr20_45906012_G_A_b38 | 2.1E-04 | -0.12 | Cells - Cultured fibroblasts |
rs6065904 | sQTL | ZNF335 | chr20_45906012_G_A_b38 | 3.3E-11 | -0.65 | Testis |
rs6065904 | sQTL | ACOT8 | chr20_45906012_G_A_b38 | 1.3E-09 | 0.58 | Heart - Left Ventricle |
rs6065904 | sQTL | PLTP | chr20_45906012_G_A_b38 | 4.5E-08 | -0.32 | Whole Blood |
rs6065904 | sQTL | PLTP | chr20_45906012_G_A_b38 | 4.8E-08 | 0.53 | Spleen |
rs6065904 | sQTL | ACOT8 | chr20_45906012_G_A_b38 | 1.3E-07 | 0.42 | Esophagus - Mucosa |
rs6065904 | sQTL | ACOT8 | chr20_45906012_G_A_b38 | 2.6E-07 | 0.49 | Heart - Atrial Appendage |
rs6065904 | sQTL | CTSA | chr20_45906012_G_A_b38 | 1.0E-06 | -0.41 | Artery - Aorta |
rs6065904 | sQTL | ACOT8 | chr20_45906012_G_A_b38 | 1.2E-06 | 0.33 | Nerve - Tibial |
rs6065904 | sQTL | ACOT8 | chr20_45906012_G_A_b38 | 1.2E-06 | 0.67 | Brain - Spinal cord (cervical c-1) |
rs6065904 | sQTL | TNNC2 | chr20_45906012_G_A_b38 | 2.1E-06 | 0.54 | Brain - Cerebellum |
rs6065904 | sQTL | ACOT8 | chr20_45906012_G_A_b38 | 2.1E-06 | 0.54 | Brain - Cerebellum |
rs6065904 | sQTL | WFDC3 | chr20_45906012_G_A_b38 | 5.5E-06 | 0.23 | Skin - Sun Exposed (Lower leg) |
rs6065904 | sQTL | WFDC3 | chr20_45906012_G_A_b38 | 9.4E-06 | -0.28 | Skin - Not Sun Exposed (Suprapubic) |
Data availability
GWAS data used in the data are publicly available. Details can be found in Table 1.
References
-
WebsiteRound 2 {GWAS} results of thousands of phenotype inthe {UK} {BioBank}Accessed August 31, 2018.
-
Effects of torcetrapib in patients at high risk for coronary eventsNew England Journal of Medicine 357:2109–2122.https://doi.org/10.1056/NEJMoa0706628
-
National lipid association annual summary of clinical lipidology 2016Journal of Clinical Lipidology 10:S1–S43.https://doi.org/10.1016/j.jacl.2015.08.002
-
Controlling the false discovery rate: a practical and powerful approach to multiple testingJournal of the Royal Statistical Society: Series B 57:289–300.https://doi.org/10.1111/j.2517-6161.1995.tb02031.x
-
Mendelian randomization with invalid instruments: effect estimation and Bias detection through egger regressionInternational Journal of Epidemiology 44:512–525.https://doi.org/10.1093/ije/dyv080
-
Mendelian randomization analysis with multiple genetic variants using summarized dataGenetic Epidemiology 37:658–665.https://doi.org/10.1002/gepi.21758
-
Lipids, lipoproteins, and metabolites and Risk of Myocardial Infarction and StrokeJournal of the American College of Cardiology 71:620–632.https://doi.org/10.1016/j.jacc.2017.12.006
-
Mendelian randomization: genetic anchors for causal inference in epidemiological studiesHuman Molecular Genetics 23:R89–R98.https://doi.org/10.1093/hmg/ddu328
-
Mendelian randomization as an instrumental variable approach to causal inferenceStatistical Methods in Medical Research 16:309–330.https://doi.org/10.1177/0962280206077743
-
Cholesterol efflux capacity and subclasses of HDL particles in healthy women transitioning through menopauseThe Journal of Clinical Endocrinology & Metabolism 101:3419–3428.https://doi.org/10.1210/jc.2016-2144
-
Phenotypic characterization of genetically Lowered Human Lipoprotein(a) LevelsJournal of the American College of Cardiology 68:2761–2772.https://doi.org/10.1016/j.jacc.2016.10.033
-
Mendelian randomization of blood lipids for coronary heart diseaseEuropean Heart Journal 36:539–550.https://doi.org/10.1093/eurheartj/eht571
-
Mendelian randomization in Cardiometabolic disease: challenges in evaluating causalityNature Reviews Cardiology 14:577–590.https://doi.org/10.1038/nrcardio.2017.78
-
Small dense low-density lipoprotein-cholesterol concentrations predict risk for coronary heart disease: the atherosclerosis risk in communities (ARIC) studyArteriosclerosis, Thrombosis, and Vascular Biology 34:1069–1077.https://doi.org/10.1161/ATVBAHA.114.303284
-
Direct estimation of HDL-Mediated cholesterol efflux capacity from serumClinical Chemistry 65:1042–1050.https://doi.org/10.1373/clinchem.2018.299222
-
Residual risk of atherosclerotic cardiovascular events in relation to reductions in Very-Low-Density lipoproteinsJournal of the American Heart Association 6:e007402.https://doi.org/10.1161/JAHA.117.007402
-
Evacetrapib and cardiovascular outcomes in High-Risk vascular diseaseNew England Journal of Medicine 376:1933–1942.https://doi.org/10.1056/NEJMoa1609581
-
Efficient design for mendelian randomization studies: subsample and 2-sample instrumental variable estimatorsAmerican Journal of Epidemiology 178:1177–1184.https://doi.org/10.1093/aje/kwt084
-
PLINK: a tool set for whole-genome association and population-based linkage analysesThe American Journal of Human Genetics 81:559–575.https://doi.org/10.1086/519795
-
HDL and cardiovascular diseaseThe Lancet 384:618–625.https://doi.org/10.1016/S0140-6736(14)61217-4
-
HDL cholesterol efflux capacity and incident cardiovascular eventsNew England Journal of Medicine 371:2383–2393.https://doi.org/10.1056/NEJMoa1409065
-
Association of HDL cholesterol efflux capacity with incident coronary heart disease events: a prospective case-control studyThe Lancet Diabetes & Endocrinology 3:507–513.https://doi.org/10.1016/S2213-8587(15)00126-6
-
An examination of multivariable mendelian randomization in the single-sample and two-sample summary data settingsInternational Journal of Epidemiology 48:713–727.https://doi.org/10.1093/ije/dyy262
-
High plasma phospholipid transfer protein levels as a risk factor for coronary artery diseaseArteriosclerosis, Thrombosis, and Vascular Biology 23:1857–1862.https://doi.org/10.1161/01.ATV.0000094433.98445.7F
-
Effects of dalcetrapib in patients with a recent acute coronary syndromeNew England Journal of Medicine 367:2089–2099.https://doi.org/10.1056/NEJMoa1206797
-
'Mendelian randomization': can genetic epidemiology contribute to understanding environmental determinants of disease?International Journal of Epidemiology 32:1–22.https://doi.org/10.1093/ije/dyg070
-
Discovery and refinement of loci associated with lipid levelsNature Genetics 45:1274.https://doi.org/10.1038/ng.2797
-
Comparative proteome analysis of epicardial and subcutaneous adipose tissues from patients with or without coronary artery diseaseInternational Journal of Endocrinology 2019:6976712.https://doi.org/10.1155/2019/6976712
-
Powerful three-sample genome-wide design and robust statistical inference in summary-data mendelian randomizationInternational Journal of Epidemiology 48:1478–1492.https://doi.org/10.1093/ije/dyz142
-
Two-Sample instrumental variable analyses using heterogeneous samplesStatistical Science 34:317–333.https://doi.org/10.1214/18-STS692
-
Statistical inference in two-sample summary-data mendelian randomization using robust adjusted profile scoreThe Annals of Statistics 48:1742–1769.https://doi.org/10.1214/19-AOS1866
Decision letter
-
Edward D JanusReviewing Editor; University of Melbourne, Australia
-
Matthias BartonSenior Editor; University of Zurich, Switzerland
-
David SullivanReviewer
-
Jie ZhengReviewer; University of Bristol, United Kingdom
In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.
Acceptance summary:
This paper is a welcome extension of the use of Mendelian Randomization in the evaluation of the role of lipoprotein sub-fractions. It is necessarily complex. The confirmation of the atherogenicity of summary estimates of LDL and VLDL is useful as are new insights into the protective role of HDL sub-fractions.
Decision letter after peer review:
Thank you for submitting your article "A Mendelian randomization study of the role of lipoprotein subfractions in coronary artery disease" for consideration by eLife. Your article has been reviewed by 4 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Matthias Barton as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: David Sullivan (Reviewer #1); Jie Zheng (Reviewer #4).
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
As the editors have judged that your manuscript is of interest, but if as described below additional experiments are required before it is published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)
Summary:
Zhao et al. apply an array of MR approaches to attempt to disentangle the contributions of different lipoprotein subclasses to coronary artery disease risk. They also try to answer the debate about "what matters the most", the lipid content of the lipoproteins or the number of particles. They conclude that LDL and VLDL sub-fractions have a universally adverse effect on coronary artery disease and myocardial infarction and that small dense LDL is not more atherogenic. HDL sub-fractions are heterogeneous in their effects with medium sized particle being protective and this supports the HDL function hypothesis.
Essential revisions:
There are aspects of the study design that could be improved and discussed more critically because some of the interpretations are not straightforward. Overall comprehension of the paper would be greatly aided by a more detailed consideration of the relationship between the different lipoprotein classes and subclasses, and a clear explanation of parameters of interest.
Essential are a substantially clearer description of the results (requiring major revision of the manuscript), stronger justification of the statistical approach taken and explanation as to why results for some fractions are not presented. A preferred alternative is to present results for all fractions even if only in supplementary tables.
1. In the abstract and text please modify the use of the terms positive and negative effects. Clinicians interpret positive as meaning beneficial when referring to outcomes eg in RCTs and negative as ineffective or harmful. In this paper positive and negative appropriately refer to the directions of the associations found. To remove any potential ambiguity for readers please consider using positive (harmful) and negative (protective) or a similar rewording of your choice when these terms first appear in the abstract and main text.
2. The analyses presented are laudably thorough in detail, using different combinations of datasets in which variants to be used for analysis are identified in different summary statistics from those used for the MR itself. There does not appear to have been a prior hypothesis to this work. Rather, it seems that all possible analyses have been conducted and the results then mined for potentially noteworthy findings. While such analyses are useful in themselves the conclusions drawn are suspect and, in places, contradicted by other parts of the authors' own results. The main results also rest on a new method which has yet to be established as valid or workable in this context – GRAPPLE has only recently been described in a bioRXiV preprint, applied to traits with well over 100 independent genome-wide associations, and separate publication resting almost entirely on results using it, applied to traits with much less well-powered GWAS, is premature at best.
3. Genetic correlation analysis – The genetic correlation analysis seems to be stand alone to other MR analyses. The listed motivation of this analysis is to check whether the MR findings are independent to each other. However, it is clear that all these lipid sub-fractions are highly correlated to each other (genetically and observationally), especially for VLDL and LDL sub-fractions. So the rationale of linking TG with VLDL was not tested (although biologically VLDL is the main particle carrying TG in fasting). Even in a clustering point of view, it is hard to split the lipid sub-fraction into three groups: VLDL-TG, IDL/LDL-LDL-C and HDL-HDL-C.
Also some of the rg estimates are missing in Figure 1, e.g. TG is related to the key findings but not included in this genetic correlation analysis.
Also, in a method point of view there are some new genetic correlation methods coming out recently https://www.nature.com/articles/s41588-020-0653-y, which could be considered.
4. Instrument selection – From a practical point of view, It is tricky to jointly model TG with VLDL subtypes characteristics in MVMR because it is unlikely that one would have genetic instruments that strongly predict each exposure once adjusting for the others. Therefore, weak instrument bias is probably an issue here. This is worrying as weak instrument bias in MVMR is not necessarily conservative. At a minimum please provide the conditional F statistics for the MVMR model for each exposure.
5. The authors present a large amount of results for (unadjusted) univariate MR. Given the known highly pleiotropic nature of variants affecting lipoproteins and their sub-fractions, presentation of these in the main figure adds little and obscures the main results of potential note. The full results could be presented as supplementary data, but the main figures should be restricted to multivariable MR.
6. Independent effects of lipid sub-fractions? multivariable MR (MVMR)
One of the key issue for lipids sub-fractions (and other metabolites such as fatty acids) is all the sub-fractions are highly correlated to each other. It is almost impossible to include all of the sub-fractions in the same MVMR model. Some methods considering reduce the dimensionality to solve this issue. In this study, the author used a different approach, which split the lipid sub-fractions into three classes, HDL, LDL and TG related. In this way, the MVMR could be conducted with limited number of variables. However, a lot of SNPs are associated with two or more lipid phenotypes, e.g. the CETP example the author provided in Table 3, the SNP is associated with HDL-C, LDL-C and TG. Even in a MVMR model controlling for LDL-C and TG, it is still hard to prove that the highlighted M-HDL-P finding showed an independent effect on CAD.
Also, more and more studies considered APOB and APOA1 in the MVMR model (e.g. https://journals.plos.org/plosmedicine/article/comments?id=10.1371/journal.pmed.1003062), so at least worth including APOB/APOA1 in the MVMR model.
7. VLDL MR results – For VLDL results, some of them showed negative effects (CIs did not cross OR=1) in the MVMR model but positive effect in UVMR model, which seems complex and did not fit with the simple interpretation that "VLDL subfraction traits had uniformly positive effect on coronary artery disease" mentioned in the abstract. Can the authors discuss this in a bit more details and try to find the reason?
8. There are multiple lipidomics measures which are available from the NMR platform, but which are not analyzed, including critical ones such as M-HDL-TG. This needs explanation and/or rectification, since complete data would address multiple points. For instance, the authors dismiss the positive association of S-HDL-TG as being confounded by correlation with VLDL fractions, despite VLDL fractions not having any independent positive effect on CAD in the multivariable analysis. Results for M-HDL-TG would be very informative, and would avoid the possibility that erroneous conclusions are drawn simply because contrary data is missing.
9. It is unclear that it is appropriate to adjust for overall LDL-C when analyzing HDL sub-fractions. Attempting to analyze sub-fractions of one class alongside an aggregate measure of another (which is correlated with aggregate measures of the first) seems suspect and at the very least requires detailed and careful justification.
10. The introduction anticipates the resolution of conflicting findings in relation to LDL size. This is dealt with in discussion, but it needs to be be highlighted beforehand in results. Since levels of different sized LDL and VLDL particles are strongly correlated, a statement that different sizes are not differently atherogenic is not supported by the data. To draw this conclusions would require multivariable MR in which the different size classes were adjusted for each other. In fact, the statement is directly contradicted by the clear associations of VLDL and (particularly) LDL diameter with CAD.
11. Similarly the statement that medium and small HDL particles are protective is contradicted by the fact that HDL particle diameter shows no association at all with CAD risk.
12. Selection of lipid sub-fractions – 82 lipid sub-fractions were selected but it is not directly clear why these fractions (but not others) were selected. It will be helpful to have a DAG to explain the inclusion and exclusion criteria. There is an even more fundamental question: Does this data justify a re-classification of NMR lipoprotein subclasses? Most people would agree that 82 is a few too many. Can the authors nominate a rational condensation of NMR lipoprotein sub-fraction data into a handful of independently predictive parameters with putative mechanisms? Ideally, these would be targets for therapy and indicators of response.
13. HDL MR results – For S-HDL-TG, this is a very complex case and it is doubtful if the current setting can distinguish its effect driven by HDL from its effect driven by TG. I personally think its strong genetic correlation with VLDL (Figure 1) implies that the effect of S-HDL-TG on CAD is driven by TG. However, the MVMR adjusted TG showed similar effect estimate, which suggest the effect is independent to TG (as mentioned about, the reviewer is not sure the MVMR model can prove independent effect). The authors may need to consider integrating some biological information of each instruments been used here to dig into the button of this case study. For example, whether each of instrument colocalized with the expression level of the cis gene in multiple tissues. For the colocalized genes, are they related to HDL pathway or TG pathway etc. This could be very complex so the reviewer suggests to explain the results with caution.
For M-HDL-P, it showed some correlations with HDL-C and APOA1, so hard to say it is not correlated with HDL-C at all.
14. The identification of a leading role for mid-sized HDL particles presents a tantalizing opportunity to link the latter finding with recent NMR studies of cholesterol efflux, but this is not pursued. It would be worth flagging this opportunity at least in the discussion if you cannot easily address it. How do the results (M-HDL-P etc) relate to NMR assessments of cholesterol efflux capacity (eg Direct estimation of HDL-mediated cholesterol efflux capacity from serum. Sanna Kuusisto, Michael V. Holmes, Pauli Ohukainen, Antti J. Kangas, Mari Karsikas, Mika Tiainen, Markus Perola, Veikko Salomaa, Johannes Kettunen, Mika Ala-Korpela doi: https://doi.org/10.1101/396929, now published in Clinical Chemistry doi: 10.1373/clinchem.2018.299222).
15. The multiple versions of the analysis using different combinations of datasets do provide worthwhile technical replication. However, these do not provide wider replication of the findings since any shortcomings in one analysis (e.g. in failing to fully address pleiotropy) will inevitably be present in another.
16. A failure to find evidence against the INSIDE assumption, on the basis of only a small number of SNPs, is a very weak basis for making a claim that there is not horizontal pleiotropy affecting LDL-C or TG. The arguments made by the authors on this point are very weak.
17. The issue of horizontal pleiotropy, especially as it applies to m-HDL-P in relation to the inSIDE assumptions for TG and LDL-C, is justified in discussion and acknowledged in limitations, but it is very difficult follow. For example, from table 3, TG is strongly associated with all traits apart from LOC157273, LIPG and DOCK6. Accordingly, the simple clinician is tempted to regard m-HDL-P as a surrogate for TG. Perhaps there needs to be greater clarity concerning the concepts of genetic correlation and "weak instrument bias".
18. Discussion – "Our results for the HDL sub-fractions support the conclusion that HDL-C is not the causally relevant biomarker." The MR estimate of S-HDL-TG, M-HLD-C and M-HDL-L showed the complexity of HDL-C on CAD. So this claim seems against the main finding of the manuscript.
[Editors' note: further revisions were suggested prior to acceptance, as described below.]
Thank you for resubmitting your work entitled "A Mendelian randomization study of the role of lipoprotein sub-fractions in coronary artery disease" for further consideration by eLife. Your revised article has been evaluated by a Reviewing Editor and a Senior Editor.
The manuscript has been improved but there are some remaining issues noted by the reviewers that need to be addressed, as outlined below:
You have made substantial efforts to restructure the manuscript (e.g. using genetic correlation as a filtering step in the new version), highlighted the key findings of HDL particle size on CAD as well as identified four potential causal genes linking HDL particle size with CAD. We are in general more convinced that the new findings from this study will bring good value for the existing argument for effect of HDL on CAD.
1. The statistical superiority of HDL over TG is largely a reflection of the greater intra-individual biological variability of the LATTER. This is really the crux of our concern. It seems likely that TG and HDL size represent a gene/environment interaction with a very large environmental component. We are concerned that environmental factors could distort a Mendelian Randomization perspective of this analysis in a way in which the statistics used are valid, but conclusions are not generalizable. This can be addressed in the discussion maybe as a limitation.
2. A sentence in the abstract which summarizes the key findings and potential value of the study would be helpful.
3. Methods, genetic correlation. The authors claim that genetic correlation "is generally different from epidemiological correlation that is estimated from cross-sectional data." By concept, the genetic correlation and phenotypic correlation are different. But in the metabolites case, they have very similar estimates! So please refine this statement to make sure you are just talking about the concept.
4. A very brief introduction about genome-wide MR will be helpful. e.g. did you do LD pruning or any other selection.
5. The genetic marker section. This is value added but the subtitle "genetic markers" is a slightly underselling the value here. You are trying to map variants to genes to inform causal genes that are linking HDL-C size with CAD. It may worth stating this in the subtitle. Also, to make this section more informative, you can try a formal Wald ratio + colocalization analysis to estimate the putative causal effects of these HDL size related genes on CAD (rather than just did a SNP lookup in different GWASs).
6. Table 2, better to show 95%CI rather than just SE.
7. Discussion "small and medium HDL particles appear to be positively correlated with HDL cholesterol and ApoA1, their genetic correlations are much smaller than 1, indicating the existence of independent biological pathways." This statement is too strong by just using a statistical approach. Better to say "indicating possible independent biological pathway(s)."
https://doi.org/10.7554/eLife.58361.sa1Author response
Essential revisions:
There are aspects of the study design that could be improved and discussed more critically because some of the interpretations are not straightforward. Overall comprehension of the paper would be greatly aided by a more detailed consideration of the relationship between the different lipoprotein classes and subclasses, and a clear explanation of parameters of interest.
Essential are a substantially clearer description of the results (requiring major revision of the manuscript), stronger justification of the statistical approach taken and explanation as to why results for some fractions are not presented. A preferred alternative is to present results for all fractions even if only in supplementary tables.
We would like to thank the referees and editors for carefully reviewing our manuscript and giving the constructive comments. We have implemented new analyses and substantially revised our manuscript. The main changes include:
1. We have made it clear in the Introduction that the goal of this study is to use genetic data to "discover lipoprotein subfractions that may be causal risk factors for CAD and MI in addition to the traditional lipid profile". To this end, we used the estimated genetic correlation with the traditional lipid traits to screen the subfraction measurements. This allows us to identify lipoprotein subfractions that may involve independent biological mechanisms before we look at the data about coronary artery disease.
2. We have redesigned our multivariable MR analyses, in which all the tradition lipid risk factors were included as exposures. Following the suggestion of a referee, we now considered two multivariable MR designs:
– In the first design, the exposures are TG, LDL-C, HDL-C, and the subfraction measurement under investigation;
– In the second design, the exposures are TG, ApoB, ApoA1, and the subfraction measurement under investigation.
The results of these two multivariable MR designs were largely comparable.
1. To make the results easier to interpret, we have moved the results of most of the univariable MR analyses to the Online Supplement. This also makes the Materials and methods in the main manuscript more straightforward.
2. We have revised the procedure of identifying genetic markers of interest. Because the main conclusion of our study is that HDL particle size may play a role in coronary artery disease, we selected SNPs that are associated with small or medium HDL subfractions and with CAD, but are not associated with LDL cholesterol or ApoB. Using this procedure, we identified four genetic markers for HDL size.
3. We have rewritten the Results section and the interpretation of the results in Discussion.
1. In the abstract and text please modify the use of the terms positive and negative effects. Clinicians interpret positive as meaning beneficial when referring to outcomes eg in RCTs and negative as ineffective or harmful. In this paper positive and negative appropriately refer to the directions of the associations found. To remove any potential ambiguity for readers please consider using positive (harmful) and negative (protective) or a similar rewording of your choice when these terms first appear in the abstract and main text.
Thank you. We have revised the usage of positive and negative effects in the manuscript.
2. The analyses presented are laudably thorough in detail, using different combinations of datasets in which variants to be used for analysis are identified in different summary statistics from those used for the MR itself. There does not appear to have been a prior hypothesis to this work. Rather, it seems that all possible analyses have been conducted and the results then mined for potentially noteworthy findings. While such analyses are useful in themselves the conclusions drawn are suspect and, in places, contradicted by other parts of the authors' own results. The main results also rest on a new method which has yet to be established as valid or workable in this context – GRAPPLE has only recently been described in a bioRXiV preprint, applied to traits with well over 100 independent genome-wide associations, and separate publication resting almost entirely on results using it, applied to traits with much less well-powered GWAS, is premature at best.
Thank you for the feedback. We had suspected that HDL particles may have a complicated and heterogeneous role prior to this work. For example, in a prior study (Zhao et al., Int J Epidemiol. 2019 Oct;48:1478-92), we concluded that "Further investigations are needed to demystify the observational and genetic associations between HDL-C and CAD." However, we did not want to restrict to this single hypothesis, because more information can be obtained and no real extra effort is needed by considering all lipid subfraction traits simultaneously. By being more agnostic, we also avoid the perils of selection/publication bias. This makes the current study slightly less powerful as we need to adjust for multiple testing, but we thought it is a price worth paying.
It is true that the GRAPPLE paper has not been published in a journal. On the other hand, we truly believe GRAPPLE has solved several issues in existing methods for multivariable MR and is more powerful. You are correct that the GWAS in this study are less well-powered, which is exactly the situation GRAPPLE is superior over other methods because it utilizes weak instruments efficiently and avoids weak instrument bias. So it would be a pity if we don’t use GRAPPLE here for this study. But we understand the concerns of relying the results on an unpublished methodological article and would understand if the editors want to put the publication of this manuscript on hold before the paper describing GRAPPLE finishes the peer-review process.
3. Genetic correlation analysis – The genetic correlation analysis seems to be stand alone to other MR analyses. The listed motivation of this analysis is to check whether the MR findings are independent to each other. However, it is clear that all these lipid sub-fractions are highly correlated to each other (genetically and observationally), especially for VLDL and LDL sub-fractions. So the rationale of linking TG with VLDL was not tested (although biologically VLDL is the main particle carrying TG in fasting). Even in a clustering point of view, it is hard to split the lipid sub-fraction into three groups: VLDL-TG, IDL/LDL-LDL-C and HDL-HDL-C.
Also some of the rg estimates are missing in Figure 1, e.g. TG is related to the key findings but not included in this genetic correlation analysis.
Also, in a method point of view there are some new genetic correlation methods coming out recently https://www.nature.com/articles/s41588-020-0653-y, which could be considered.
We have thought about the role of the genetic correlation analysis and redesigned our study. The estimated genetic correlations are now used to screen the lipid subfractions and remove those who are highly genetically correlated with the traditional lipid profile. Those subfractions are uninteresting because they do not seem to involve independent biological mechanisms and make the results of the multivariable MR analyses unstable.
We now included the genetic correlations with TG in the results. We found that all VLDL traits (besides a few related to very small VLDL) have extremely high (close to 1) genetic correlations with TG. In consequence, they were excluded from the MR analysis. We no longer group the lipid subfractions in the main manuscript, although the clustering was kept in some tables in the Online Supplement to make the results more organized.
Thanks for suggesting the new method for estimating genetic correlation via the high-definition likelihood. We tried it with a few traits and found the software is much slower than that of LD score regression. Because we needed to compute a very large number of genetic correlations and the precision of LD score regression seemed to be enough for our purposes, we did not switch to the high-definition likelihood approach in this study.
4. Instrument selection – From a practical point of view, It is tricky to jointly model TG with VLDL subtypes characteristics in MVMR because it is unlikely that one would have genetic instruments that strongly predict each exposure once adjusting for the others. Therefore, weak instrument bias is probably an issue here. This is worrying as weak instrument bias in MVMR is not necessarily conservative. At a minimum please provide the conditional F statistics for the MVMR model for each exposure.
Thank you for bringing up this valuable point. This is actually why we did not adjust for TG in multivariable MR analyses of VLDL subfractions in the original submission, but that seemed to have made the results more difficult to interpret. In the revision, most of the VLDL subfractions were excluded from the MR analysis due to their high genetic correlation with TG (see reply to point 3 above). We also decided to use the same set of traditional lipid risk factors in multivariable MR for different subfractions. So for VLDL subfractions, we adjusted for TG, LDL-C, and HDL-C (or TG, ApoB, and ApoA1). In Online Supplement C.4, we reported the conditional Cochran’s Q statistics for the multivariable MR models as described in the following article: Eleanor Sanderson, George Davey Smith, Frank Windmeijer, Jack Bowden, An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings, International Journal of Epidemiology, Volume 48, Issue 3, June 2019, Pages 713727, https://doi.org/10.1093/ije/dyy262.
5. The authors present a large amount of results for (unadjusted) univariate MR. Given the known highly pleiotropic nature of variants affecting lipoproteins and their sub-fractions, presentation of these in the main figure adds little and obscures the main results of potential note. The full results could be presented as supplementary data, but the main figures should be restricted to multivariable MR.
We have moved most of the univariable MR results to the Online Supplement. In Figure 2 of the main manuscript, we now report the results of one univariable MR analysis and two multivariable MR analyses.
6. Independent effects of lipid sub-fractions? multivariable MR (MVMR)
One of the key issue for lipids sub-fractions (and other metabolites such as fatty acids) is all the sub-fractions are highly correlated to each other. It is almost impossible to include all of the sub-fractions in the same MVMR model. Some methods considering reduce the dimensionality to solve this issue. In this study, the author used a different approach, which split the lipid sub-fractions into three classes, HDL, LDL and TG related. In this way, the MVMR could be conducted with limited number of variables. However, a lot of SNPs are associated with two or more lipid phenotypes, e.g. the CETP example the author provided in Table 3, the SNP is associated with HDL-C, LDL-C and TG. Even in a MVMR model controlling for LDL-C and TG, it is still hard to prove that the highlighted M-HDL-P finding showed an independent effect on CAD.
Also, more and more studies considered APOB and APOA1 in the MVMR model (e.g. https://journals.plos.org/plosmedicine/article/comments?id=10.1371/journal.pmed.1003062), so at least worth including APOB/APOA1 in the MVMR model.
Thank you for the suggestions. We have used genetic correlation analysis to pre-screen the lipid subfractions and reduce the dimensionality. Our purpose was not to identify the single "most causal" subfraction. Rather, we were generically interested in whether the lipid subfractions provide any additional value on top of the traditional risk factors. We hope this becomes clearer in the revision.
It is true that many SNPs are still associated with the lipid subfractions under investigation and some traditional lipid risk factors. We have implemented a new procedure to identify genetic markers for medium HDL and HDL size that are not associated with LDL-C or ApoB. In the Discussion, we made it clear that "the role of HDL particles in preventing CAD may be more complicated than, for example, that of LDL cholesterol or ApoB."
Thank you for the pointer to using ApoB and ApoA1 in MVMR. This is now included in our study as well. The results are generally not too different from using LDL-C and HDL-C, although ApoB does seem to have a slightly larger estimated effect than LDL-C and seems to "take away" some of the estimated effect of TG (see Table 2).
7. VLDL MR results – For VLDL results, some of them showed negative effects (CIs did not cross OR=1) in the MVMR model but positive effect in UVMR model, which seems complex and did not fit with the simple interpretation that "VLDL subfraction traits had uniformly positive effect on coronary artery disease" mentioned in the abstract. Can the authors discuss this in a bit more details and try to find the reason?
Thank you for pointing this out. The VLDL subfractions had very high genetic correlations with TG (and to some extent, with ApoB, see Supplement Table B2). So it is very unlikely that a genetic analysis can differentiate the VLDL subfractions from the traditional risk factors, if there is any. Due to this reason, we now exclude most of them from the MR analysis (see also the reply to point 3 above).
8. There are multiple lipidomics measures which are available from the NMR platform, but which are not analyzed, including critical ones such as M-HDL-TG. This needs explanation and/or rectification, since complete data would address multiple points. For instance, the authors dismiss the positive association of S-HDL-TG as being confounded by correlation with VLDL fractions, despite VLDL fractions not having any independent positive effect on CAD in the multivariable analysis. Results for M-HDL-TG would be very informative, and would avoid the possibility that erroneous conclusions are drawn simply because contrary data is missing.
In the genetic correlation analysis, it seems that S-HDL-TG is highly correlated with TG and behave more like a VLDL subfraction than an HDL subfraction (and thus is removed from the main MR analysis, see reply to point 3 above). The other triglyceride measurement for a HDL subfraction, XL-HDL-TG, had a much weaker genetic correlation with TG (see Table B2). We agree with you that an analysis for M-HDL-TG could be quite informative, but unfortunately data about M-HDL-TG were not reported in the two lipidomic GWAS available to us.
9. It is unclear that it is appropriate to adjust for overall LDL-C when analyzing HDL sub-fractions. Attempting to analyze sub-fractions of one class alongside an aggregate measure of another (which is correlated with aggregate measures of the first) seems suspect and at the very least requires detailed and careful justification.
Thank you for this interesting point. As explained in our reply to point 6, our purpose was not to identify the single "most causal" subfraction. Rather, we were generically interested in whether the lipid subfractions provide any additional value on top of the traditional risk factors. Relatedly, any subfraction discovered in the MR analysis is not necessarily the only causal agent. In the Discussion, we made it clear that "it is possible that HDL cholesterol, HDL subfractions, and HDL particle size are all phenotypic markers for some underlying causal mechanism." We then discussed possible connections with the HDL function hypothesis and the cholesterol efflux capacity.
10. The introduction anticipates the resolution of conflicting findings in relation to LDL size. This is dealt with in discussion, but it needs to be be highlighted beforehand in results. Since levels of different sized LDL and VLDL particles are strongly correlated, a statement that different sizes are not differently atherogenic is not supported by the data. To draw this conclusions would require multivariable MR in which the different size classes were adjusted for each other. In fact, the statement is directly contradicted by the clear associations of VLDL and (particularly) LDL diameter with CAD.
Thank you. We now made it clear in the Results section that "The mean diameter of LDL particles (LDL-D) showed a harmful effect on MI in univariable MR, though the effect was smaller than those of the LDL subfractions in univariable MR. The estimated effect of LDL-D was attenuated in the multivariable MR analyses." In the Discussion, we concluded that "we find some weak evidence that larger LDL particle size may have a small harmful effect on myocardial infarction and coronary artery disease." We were a little conservative in this conclusion because there was not a very convincing evidence from the multivariable MR analyses. The individual LDL subfractions had the right trend in multivariable MR (see Figure C4), but the confidence intervals were quite wide due to strong genetic correlation with LDL-C/ApoB. In one multivariable MR, LDL-D showed a protective effect, but it was barely statistically significant.
11. Similarly the statement that medium and small HDL particles are protective is contradicted by the fact that HDL particle diameter shows no association at all with CAD risk.
Thank you for this observation. HDL-D was not associated with CAD risk in univariable MR, but did show a positive effect in multivariable MR analyses now (notice that HDL-C/ApoA1 is now included in the multivariable MR for HDL-D, but was not included in the original submission). It turns out that adjusting for HDL-C/ApoA1 is quite important here, see Table 2. We do not interpret this as a necessary contradiction to the findings about small and medium HDL particles, but do acknowledge that the mechanism involving HDL particle size may be quite complicated. In the Discussion, we stated the following: "Notice that the harmful effect of larger HDL particle diameter found in this study relies on including HDL-C or ApoA1 in the multivariable MR analysis. Thus, the role of HDL particles in preventing CAD may be more complicated than, for example, that of LDL cholesterol or ApoB. It is possible that HDL cholesterol, HDL subfractions, and HDL particle size are all phenotypic markers for some underlying causal mechanism."
12. Selection of lipid sub-fractions – 82 lipid sub-fractions were selected but it is not directly clear why these fractions (but not others) were selected. It will be helpful to have a DAG to explain the inclusion and exclusion criteria. There is an even more fundamental question: Does this data justify a re-classification of NMR lipoprotein subclasses? Most people would agree that 82 is a few too many. Can the authors nominate a rational condensation of NMR lipoprotein sub-fraction data into a handful of independently predictive parameters with putative mechanisms? Ideally, these would be targets for therapy and indicators of response.
We did not select the lipid subfractions. The 82 subfractions were those reported in the lipidome GWAS that are related to VLDL, LDL, IDL, and HDL. This is now made clear in the beginning of Materials and methods.
Thank you for the suggestion about condensing the subfraction data. As described in the reply to point 3, we now use genetic correlation to screen the subfractions and remove the ones with highly genetic correlations with the traditional lipid risk factors (TG, LDL-C, HDL-C, ApoB, or ApoA1). This left us with 27 traits. Although one may argue that is still a few too many, we wanted to be more agnostic about generating hypotheses and let the data to speak for themselves. Relatedly, the purpose of this study was not to identify the single "most causal" subfraction. Rather, we were generically interested in whether the lipid subfractions provide any additional value on top of the traditional risk factors. In the Discussion, we pointed that the role of HDL particle size may be complicated and avoided making simple-minded conclusions like "increasing M-HDL-P by any means is causally protective, period".
13. HDL MR results – For S-HDL-TG, this is a very complex case and it is doubtful if the current setting can distinguish its effect driven by HDL from its effect driven by TG. I personally think its strong genetic correlation with VLDL (Figure 1) implies that the effect of S-HDL-TG on CAD is driven by TG. However, the MVMR adjusted TG showed similar effect estimate, which suggest the effect is independent to TG (as mentioned about, the reviewer is not sure the MVMR model can prove independent effect). The authors may need to consider integrating some biological information of each instruments been used here to dig into the button of this case study. For example, whether each of instrument colocalized with the expression level of the cis gene in multiple tissues. For the colocalized genes, are they related to HDL pathway or TG pathway etc. This could be very complex so the reviewer suggests to explain the results with caution.
For M-HDL-P, it showed some correlations with HDL-C and APOA1, so hard to say it is not correlated with HDL-C at all.
We agree that it is impossible to make any conclusions about S-HDL-TG because of its high genetic correlation with TG. In fact, S-HDL-TG has been excluded from the main MR analysis in our new design with phenotypic screening. It is true that M-HDL-P is genetically correlated with HDL-C and ApoA1, but adjusting for HDL-C and ApoA1 did not alter the estimated effect of M-HDL-P (the story is different for HDL-D). In summary, we think there is enough evidence to suggest that S-HDL-P and MHDL-P are risk factors that are reasonably independent of the traditional HDL-C and ApoA1. However, we also made it clear in the Discussion that the role of HDL particle size may be quite complicated.
Thank you for the suggestion about integrating biological information. We have used a new procedure to identify instruments associated with HDL particle sizes but not with LDL-C or ApoB. We have also included the cis genes that show up in eQTL data.
14. The identification of a leading role for mid-sized HDL particles presents a tantalizing opportunity to link the latter finding with recent NMR studies of cholesterol efflux, but this is not pursued. It would be worth flagging this opportunity at least in the discussion if you cannot easily address it. How do the results (M-HDL-P etc) relate to NMR assessments of cholesterol efflux capacity (eg Direct estimation of HDL-mediated cholesterol efflux capacity from serum). Sanna Kuusisto, Michael V. Holmes, Pauli Ohukainen, Antti J. Kangas, Mari Karsikas, Mika Tiainen, Markus Perola, Veikko Salomaa, Johannes Kettunen, Mika Ala-Korpela
doi: https://doi.org/10.1101/396929, now published in Clinical Chemistry doi: 10.1373/clinchem.2018.299222.
Thank you. We agree this is a tantalizing opportunity and a MR study for cholesterol efflux capacity can complement the results here. This is mentioned in the last paragraph of Discussion as a future work.
15. The multiple versions of the analysis using different combinations of datasets do provide worthwhile technical replication. However, these do not provide wider replication of the findings since any shortcomings in one analysis (e.g. in failing to fully address pleiotropy) will inevitably be present in another.
We agree with your assessment and have moved the results of the replication analyses to the Online Supplement.
16. A failure to find evidence against the INSIDE assumption, on the basis of only a small number of SNPs, is a very weak basis for making a claim that there is not horizontal pleiotropy affecting LDL-C or TG. The arguments made by the authors on this point are very weak.
We agree, although this is the best we could do to assess the InSIDE assumption. We have deemphasized this in the paper.
17. The issue of horizontal pleiotropy, especially as it applies to m-HDL-P in relation to the inSIDE assumptions for TG and LDL-C, is justified in discussion and acknowledged in limitations, but it is very difficult follow. For example, from table 3, TG is strongly associated with all traits apart from LOC157273, LIPG and DOCK6. Accordingly, the simple clinician is tempted to regard m-HDL-P as a surrogate for TG. Perhaps there needs to be greater clarity concerning the concepts of genetic correlation and "weak instrument bias".
Thank you for the suggestion. Although TG is strongly associated with several instruments for M-HDL-P, their associations were actually in different directions (see Table D9 in the Supplement) and are consistent with the InSIDE assumption. So M-HDL-P should not be regarded as a surrogate for TG. We have moved the diagnostics plots for M-HDL-P to the Online Supplement and only presented the genetic markers that are not associated with LDL-C or ApoB in the main paper. The four genetic markers we identified had no or weak associations with TG.
18. Discussion – "Our results for the HDL sub-fractions support the conclusion that HDL-C is not the causally relevant biomarker." The MR estimate of S-HDL-TG, M-HLD-C and M-HDL-L showed the complexity of HDL-C on CAD. So this claim seems against the main finding of the manuscript.
We meant HDL-C is not causal in the narrow sense that any intervention that increases HDLC will protect against CAD. Essentially we were trying to convey that the role of HDL is complicated, but we agree that this statement is not accurate. In the revised Discussion, we removed this sentence and made it clear that "Thus, the role of HDL particles in preventing CAD may be more complicated than, for example, that of LDL cholesterol or ApoB. It is possible that HDL cholesterol, HDL subfractions, and HDL particle size are all phenotypic markers for some underlying causal mechanism."
[Editors' note: further revisions were suggested prior to acceptance, as described below.]
The manuscript has been improved but there are some remaining issues noted by the reviewers that need to be addressed, as outlined below:
You have made substantial efforts to restructure the manuscript (e.g. using genetic correlation as a filtering step in the new version), highlighted the key findings of HDL particle size on CAD as well as identified four potential causal genes linking HDL particle size with CAD. We are in general more convinced that the new findings from this study will bring good value for the existing argument for effect of HDL on CAD.
We appreciate the additional comments and suggestions by the reviewers. We have revised the manuscript accordingly.
The only suggested change we did not implement is the second part of issue 5:
Also, to make this section more informative, you can try a formal Wald ratio + colocalization analysis to estimate the putative causal effects of these HDL size related genes on CAD (rather than just did a SNP lookup in different GWASs).
This is an interesting suggestion, but we did not make the change because we think the Wald ratio estimator might not provide a full picture of the potential mechanisms involved. In particular, the genetic markers in Figure 3 are not only related to HDL particle size but also the overall HDL cholesterol. The last two markers also have a small association with triglycerides. We are worried that reporting the Wald ratio estimates may provide an oversimplified summary of the information in Figure 3 to the reader. That being said, we are also submitting the raw data for Figure 3. So if some reader is interested in knowing the putative causal effect, they can also take a ratio easily themselves.
https://doi.org/10.7554/eLife.58361.sa2Article and author information
Author details
Funding
No external funding was received for this work.
Senior Editor
- Matthias Barton, University of Zurich, Switzerland
Reviewing Editor
- Edward D Janus, University of Melbourne, Australia
Reviewers
- David Sullivan
- Jie Zheng, University of Bristol, United Kingdom
Publication history
- Received: April 28, 2020
- Accepted: April 23, 2021
- Accepted Manuscript published: April 26, 2021 (version 1)
- Version of Record published: May 28, 2021 (version 2)
Copyright
© 2021, Zhao 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
-
- 1,714
- Page views
-
- 208
- Downloads
-
- 15
- Citations
Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.