Drug-target Mendelian randomization analysis supports lowering plasma ANGPTL3, ANGPTL4, and APOC3 levels as strategies for reducing cardiovascular disease risk Eur Heart J Open. (2024)

Landfors F, Henneman P, Chorell E, Nilsson SK, Kersten S. Drug-target Mendelian randomization analysis supports lowering plasma ANGPTL3, ANGPTL4, and APOC3 levels as strategies for reducing cardiovascular disease risk. Eur Heart J Open. 2024;4(3):oeae035. Published 2024 Apr 30. doi:10.1093/ehjopen/oeae035

 

 

읽은 이유:

분석 방법에 몇 가지 흥미로운-참고할 만한 것들이 있어서 읽음  

 

 

분석 방법에 참고할만한 점 

1. Mendelian randomization 분석을 할 때 LD에 의한 correlation을 해결하기 위해 GLS IVW라는 방법을 사용함. 어떤 방법인지 알아보면 좋을 것 같음.  

The precision of the IVW estimator can be influenced by LD-related correlation between the genetic IV in the drug target genes cis’ position. Therefore, we used a GLS IVW MR estimator to correct for this potential source of bias.19,20 The GLS-corrected MR approach can be conceptualized as combining the independent information of variants near a target gene while maintaining robust standard errors through weighting for their LD-related correlation. Further information regarding Drug-target MR methodology, GLS, LD matrix sensitivity, and sample overlap bias are found in the Supplementary Methods.

 

2. Genetic mimicry analyses

ANGPTL3와 CAD의 관계에 대해 common variant로 분석했을 때 significant association을 발견하지 못함. 

Common variant들이기 때문에 더 강력한 ANGPTL3 inactivation을 반영하지 못하는 것일 수도 있다고 생각해

genetic mimicry 분석이라는 것을 함 (대략 설명하자면, metabolome같이  굉장히 많은 수의 phenotype에 대해서 association profile이 비슷한가? 를 보는 분석). 분석 결과 association profile이 비슷하기 때문에 common variant들도 ANGPQTL의 modest knock-down을 반영한다고 결론을 내림. 

  In line with a previous investigation,46 we found no significant association between ANGPTL3 inactivation via common variants and CAD. Previously, however, evidence was presented that loss-of-function variants in ANGPTL3 are associated with a decreased risk of CAD.47,48 As the common variants adjacent to ANGPTL3 only modestly impacted plasma lipids, it could be argued that they do not accurately reflect the effects of more profound ANGPTL3 inactivation. Therefore, we examined whether the common variants chosen as genetic instrumental variables and were adjacent to ANGPTL3, ANGPTL4, and APOC3 mimicked the effects (i.e. showed the same effect directionality) of protein-truncating variants. The common variants adjacent to ANGPTL3, ANGPTL4, and APOC3 were highly concordant with protein-truncating variants within the same gene (Figure 5A–C). One hundred sixty-seven metabolite associations near ANGPTL3 showed a high concordance metric (R2 ) of 82% between the common variant and protein-truncating variant models. ANGPTL4 common variants were also highly concordant with ANGPTL4 protein-truncating variants, having an R2 of 83%. APOC3 showed a concordance metric R2 of 86%. These results demonstrate that the common genetic variations adjacent to ANGPTL3, ANGPTL4, and APOC3 would be valid genetic instrume s reflecting a modest ‘knock-down’ of each respective gene.

 

genetic mimicry 분석 방법: 

Genetic mimicry analysis was used to compare the metabolic concordance between common and protein-truncating variants adjacent to the ANGPTL3, ANGPTL4, and APOC3 genes. This method uses linear regression to determine the extent of similarity between different variants’ genetic associations in high-dimensional data sets.42,43 The degree of concordance was reported as the coefficient of determination (R2 ). Genetic associations between the common variants and 167 plasma metabolites were measured by drug-target MR with plasma TGs as the exposure using data sets 8 and 10 (see Table 1). Protein-truncating variants were defined as any protein truncating variant with an allele frequency <0.05 (see Supplementary Methods). The effects of the protein-truncating variants were determined by regressing plasma concentration of metabolites on protein-truncating variant carrier status in 181 672 UK Biobank participants (see Supplementary Methods for details).

 

3. Sensitivity analysis using 'only protein truncating variant within gene region' 

Common variant들로는 association을 발견하지 못해서 gene region에 있고 protein truncating 인 variant 만으로 한정해 다시 분석했는데 association이 그래도 안나옴. 그래서 여러 연구들을 모아서 meta 분석을 했고, 앞에서 발견하지 못한 robust association을 발견할 수 있었음. 특히, 연구들을 모을 때 chip이 아닌 DNA sequencing-based 연구들만 모았는데, rare variant들에 대해 measurement error가 있을 확률이 있기 때문임.    

ANGPTL3와 CAD의 연관성에 대해 서로 상충되는 연구결과들이 있었는데 이런 meta 분석을 통해 ANGTPL3가 CAD에 protective하다는 것을 밝혀낼 수 있었음. (이에 대해 discussion에서는 다음과 같이 설명함: broader case definition, a stricter definition of controls, and the meta-analysis, which incorporated evidence from previous studies.> strengthened statistical power in our study)

 Two previous studies found that loss-of-function variants in ANGPTL3 protected against CAD.3,47 In an effort to reproduce these findings, we performed a sensitivity MR analysis of CAD and limited the selection of genetic instruments to functional variants. ... Considering the beneficial effects of ANGPTL3, ANGPTL4, and APOC3 on plasma lipids, it was expected that genetic inactivation of these proteins would confer protection against CAD. However, the ANGPTL3 MR analyses focusing on common variants and MR of functional variants (identified through DNA microarrays) did not support this hypothesis. Therefore, we pursued a meta-analysis of DNA sequencing-based studies that studied the effect of ANGPTL3, ANGPTL4, and APOC3 protein-truncating variants on CAD. The rationale for excluding DNA microarray and exome bead chip-based studies was the potential risk of introducing measurement error for rare variants,51,52 leading to bias towards the null hypothesis. DNA-sequencing-based substudies from previous papers,3,47,53,54 were extracted and analysed together with genetic association analyses conducted in the UK Biobank. ...

 

방법: 

We performed sensitivity MR analyses of ANGPTL3, ANGPTL4, APOC3, LPL, and LIPG on CAD by restricting the genetic instrument selection to variants within these target genes predicted to have functional impacts. This strategy aimed to mitigate potential biases arising from common non-coding small-effect variants outside the target genes, which could be confounded due to linkage disequilibrium with other genes in the same genomic region. Ensembl Variant Effect Predictor (VEP) version 10944 was used to annotate variants within 2.5 Kb of the target gene associated (P ≤ 0.01) with target protein levels and plasma triglycerides. Non-coding variants outside of the 5′ untranslated region (UTR), 3′ UTR, or splice site regions were filtered out and excluded from further analysis, as were missense variants lacking SIFT deleterious or PolyPhen likely or probably damaging annotations. MR was conducted for single variants using the Wald ratio estimator, and meta-analysis was performed using a random-effects IVW estimator

 

4. Adverse side effect를 판단하기 위해 possible side effect와 관련된 phenotype들을 outcome으로 medelian randomization 분석을 함.    

- 혹시 treatment의 outcome이나 side effect와 관련된 trait들 만으로 GWAS 분석을 한 논문도 있을까?

- 여기서 선택한 side effect 관련 trait들이 완전히 이해가 가지는 않아서.. 다른 종류의 질환들의 side effect를 판단할 때도 비슷하게 분석하면 될지 잘 모르겠음 

 

Discussion 

위와 같은 결과들에 기반해, mendelian randomization 분석에서 (DNA sequencing 방법에 기반해 찾은) rare functional variant를 사용하는 것이 중요하다고 주장하고 있음. 하지만 rare variant는 통계적으로 정확하지 않을 수도 있다는 한계점이 있기 때문에, common variant와 rare variant 결과를 종합해 판단해야 한다고 함.  

Interestingly, the association of ANGPTL3 inactivation with CAD was only present for rare functional variants when the carrier status was determined by DNA sequencing. This exposes the limitations of drug-target MR studies using DNA micro-array-based GWAS. When rare variants are incorrectly imputed, this typically introduces a onesided loss of information that biases toward the null hypothesis, leading to falsely negative findings.52 Even though the imputation quality score (e.g. ‘INFO’) reports an imputation quality metric, this metric does not really measure the true imputation accuracy.61 The imputation accuracy can only truly be determined if variant carrier status is called by genotyping.
However, studying rare variants in genetic association studies is not without drawbacks. An important limitation of rare variants is statistical imprecision simply due to their rarity.62 Rare variants also often emerged relatively recently and consequently are more susceptible to confounding by enrichment in specific geographical regions, families, or socioeconomic strata.63 Even if appropriate model adjustments are applied, subtle differences in population structure could cause a small number of extra alleles to be present in the control (or case) group. This can lead to biased estimates when the rare alternative allele is present in ten, or hundred individuals in total, which is often the case for rare variant studies even when the total sample size is above hundreds of thousands. Overall, our findings underscore the importance of combining evidence from rare loss-of-function and common variants in genetic association studies of complex disease phenotypes.

 

*Discussion의 기타 내용들.. 

The complexity of the APOA1-APOA5-APOC3 locus and the potential confounding due to LD poses significant challenges in separating the genetic association signals. The use of APOC3 c.55 + 1G > A as a genetic instrument was justified because of its independence from common variants within this region, making it ideal for studying APOC3 inactivation specifically. On the other hand, the analyses of APOC3 inactivation that did not include the c.55 + 1G > A variant should be interpreted with caution.

Compared to clinical trials, MR analysis can exaggerate the magnitude of the effect of inactivating a gene/protein.64 Cis-pQTL MR utilizing protein-coding variants warrants extra carefulness due to the possibility of epitope-binding artefacts, which may complicate the precise interpretation of effect sizes. The p.E40 K coding variant was the only variant qualifying as a cis-pQTL in the ANGPTL4 region in the Steps 1–2 analyses. While our validation analyses suggested this specific association was not attributable to epitope-binding artefacts, we still advise caution when extrapolating effect sizes from the analyses.

Additionally, MR and other genetic association studies estimate lifelong exposure to changed gene function, while drug trials typically last 2–5 years in late adulthood. If the treatment effect multiplicatively interacts with time, MR may exaggerate it. This constraint should be considered when translating MR findings to predict the results of clinical trials.

 

 

흥미로운 점 

- Mendelian randomization 분석에서 rare functional variant를 사용하면 common variant들로는 발견하지 못한 결과들을 발견할 수 있다는 점