Causal statistical association between remnant cholesterol and coronary heart disease: genetic insights into the PSRC1–CELSR2–SORT1 gene cluster

Abstract

Remnant cholesterol (RC) is increasingly recognized as an independent contributor to coronary heart disease (CHD) risk beyond LDL-C. However, its causal roles, genetic determinants, tissue-specific regulation, and relevance across ancestrally diverse populations remain incompletely characterized. Associations between RC and incident CHD were evaluated in the UK Biobank (UKB) employing Cox regression and restricted cubic splines models, including subgroup analyses by LDL-C levels. Causality was assessed using two-sample Mendelian randomization (MR) and colocalization using genome-wide summary statistics from UKB and FinnGen. Findings were validated in a multiancestry dataset. Genetic regulatory mechanisms were explored using tissue-specific MR integrating expression quantitative trait locus. Lipid-wide MR was used to evaluate gene effects across multiple lipid traits. Comparison between identified genes and established lipid-modifying target genes was conducted. RC showed a size-specific, LDL-C-independent association with CHD, which remained strong among individuals with normal LDL-C (<2.6 mmol/l). Multivariable MR confirmed a robust causal relationship. Colocalization identified shared signals at the PSRC1-CELSR2-SORT1 locus, with lead variants rs12740347 and rs646776. This cluster exhibited liver-specific inverse associations with CHD, pleiotropic effects on lipid traits, and stronger influence on RC than well-known targets such as HMGCR and PCSK9. RC represents a potentially modifiable marker of CHD risk, especially in individuals with normal LDL-C. Hepatic expression of PSRC1-CELSR2-SORT1 protects against CHD via RC reduction, independent of LDL-C, supporting RC as promising target for CHD prevention.

Despite the proven efficacy of current effective lipid-lowering therapies targeting LDL-C, a substantial residual cardiovascular risk persists, indicating additional lipid-related contributors. This residual risk is frequently associated with elevated triglyceride-rich lipoproteins (TRLs) and reduced HDL-C (1). However, clinical trials aimed at pharmacologically increasing HDL-C have failed to demonstrate improvements in cardiovascular outcomes, and genetic studies suggest that low HDL-C is unlikely to be a causal factor in the development of coronary heart disease (CHD) (2, 3, 4). Notably, recent findings from the large-scale prospective Copenhagen General Population Study demonstrates that elevated remnant cholesterol (RC), the cholesterol content in TRLs, contributes to a significant portion of the excess risk of CHD associated with unhealthy lifestyle (5). Together, these findings have redirected scientific focus toward targeting RC as a potentially modifiable risk factor and a key component of primary prevention strategies.

Observational studies have demonstrated an association between RC and increased risk of CHD independent of LDL-C (6, 7, 8), but these studies are inherently limited in their ability to establish causality due to confounding factors and reverse causation. Interpretation is further complicated by the indirect estimation of RC (“calculated RC”) and the absence of standardized covariate adjustment methods (9). These limitations are particularly evident when individuals on lipid-lowering therapy are not analyzed separately, as this may obscure the influence of the treatment on other lipid traits (e.g., LDL-C, HDL-C, and triglycerides). In addition, many studies overlook the heterogeneity of TRLs, which differ in size, density, composition, and atherogenic potential (9, 10).

Establishing a definitive causal relationship between RC and CHD ideally requires randomized controlled trials. However, most existing randomized controlled trials have focused on triglyceride-lowering therapies, which only indirectly affect RC without isolating its specific role (11, 12). In the absence of RC-specific interventions, genetic approaches such as Mendelian randomization (MR) provide valuable tools for causal inference (13, 14). Several MR studies have suggested a causal link between RC and atherosclerotic diseases (15, 16), but many rely on genetic instruments that influence specific lipid traits, raising concerns about horizontal pleiotropy. To strengthen causal interpretation, more robust analytical approaches such as multivariable MR (MVMR) and colocalization analyses are necessary. Furthermore, the predominance of European ancestry in prior studies limits the generalizability of findings, highlighting the need for multiancestry research frameworks.

Beyond refining causal inference, addressing the RC-CHD relationship at the molecular level also requires deeper exploration of its genetic underpinnings. A critical gap in the current literature is a lack of genetic insights due to limited investigation of risk genes involved in RC metabolism, which has hindered our understanding of the regulatory networks that connect RC to CHD. To bridge this gap, integrating genome-wide association study (GWAS) and expression quantitative trait locus (eQTL) data help unravel genetic mechanisms and identify risk genes for various diseases (17, 18). This approach leverages the concept that GWAS variants may influence the expression of nearby genes, which in turn play a role in disease pathogenesis. Building on these principles, MR-like analyses can be used to test the causal effect of exposures (e.g., APOC3 gene expression) on disease outcomes (e.g., CHD) using genetic variants (e.g., SNPs) as instrumental variables. This methodology has led to identification of novel genes and biological pathways enriched with disease-associated SNPs (19, 20).

In this study, we firstly conducted a prospective investigation of the association between RC subclasses (based on particle size) and CHD using data from the UK Biobank (UKB), followed by subgroup analyses stratified by the LDL-C levels of participants. We then employed two-sample MR and colocalization analyses, leveraging large-scale GWAS data to estimate the causal role of RC in CHD, with validation in multiancestry populations. Finally, we integrated eQTL and GWAS data with tissue-specific MR and lipid-wide MR to identify potential causal genes and to elucidate the genetic regulatory mechanism. The roles of candidate genes in regulating RC were additionally compared with those of currently approved lipid-modifying target genes. Figure 1 shows a schematic representation of the research design.

Fig 1.

Schematic of the study design UKB Basic covariates in Cox regression model included age, sex, assessment center, and NMR spectrometer; Full covariates included ethnic background, smoking status, alcohol status, townsend deprivation index (TPI), physical activity (MET), educational levels, body mass index (BMI), and fasting time. BMI, body mass index; CHD, coronary heart disease; eQTL, expression quantitative trait locus; IDL-C, intermediate density lipoprotein cholesterol; LDL-C, low density lipoprotein cholesterol; L-VLDL-C, large VLDL cholesterol; MR, Mendelian randomization; M-VLDL-C, medium VLDL cholesterol; MVMR: multivariable MR; PPI, protein-protein interaction; RC, remnant cholesterol; SMR, summary-data-based Mendelian randomization; S-VLDL-C, small VLDL cholesterol; TPI, townsend deprivation index; UKB, UK Biobank; UVMR, univariable MR; XL-VLDL-C, very large VLDL cholesterol; XS-VLDL-C, very small VLDL cholesterol; XXL-VLDL-C, extremely large VLDL and chylomicron cholesterol.

Materials and methods

Data sources

We included individual level data (n = 171,559) from the UKB under application ID #4844. GWAS summary data of RC and its subclasses were available from UKB (21) in discovery analysis (n = 115,078) and meta-analysis of multiancestry cohorts (22) in validation analysis (n = 136,016). Genetic summary data with CHD were sourced from UKB (n = 361,194) and FinnGen (n = 412,181) (23). Genetic summary data of LDL-C, HDL-C, total cholesterol (TC), total triglyceride (TG) (n = 115,078) were obtained from MRC IEU OpenGWAS (24). Cis-eQTLs for target genes were obtained from the eQTLGen Consortium (https://eqtlgen.org) and Genotype-Tissue Expression (GTEx, https://gtexportal.org/). Additional details are described in Supplemental Table S1.

Individual-level data

Individual-level data from the UKB population (over 502,000 UK residents of mainly European ancestry) were utilized. After excluding the participants with no metabolomic data measured from NMR (n = 227,926), with lipid lowering therapy (n = 48,770), or being diagnosed with CHD at baseline (n = 3,837), with missing covariates (n = 49,627) or lack of follow-ups (n = 469), we included 171,559 participants in the final analysis (Supplemental Figure S1). The UKB study was approved by North West Multi-Centre Research Ethics Committee (Ref: 21/NW/0157). All participants provided written informed consent prior to enrollment, and the study was conducted in accordance with the principles of the Declaration of Helsinki.

Lipid was measured using the high-throughput NMR spectroscopy profiling platform (Nightingale Health Ltd.) (25). Total RC was regarded as the combination of VLDL, IDL, and chylomicron remnants, as conventionally defined. VLDL was further, according to the particle size, divided into very small VLDL (XS-VLDL-C), small VLDL (S-VLDL-C), medium VLDL (M-VLDL-C), large VLDL (L-VLDL-C), very large VLDL (XL-VLDL-C), and extremely large VLDL and chylomicron (XXL-VLDL-C). Prior to analysis, the levels of each lipid biomarker were transformed to a standard normal distribution using Z-score scaling.

The definition of CHD is the occurrence of acute myocardial infarction, ST elevation and non-ST elevation myocardial infarction and their complications (STEMI & NSTEMI), or other ischemic heart diseases, according to ICD-10 codes I21-I24 and I25.2, or coronary artery bypass surgery, i.e., OPSC-4 codes k40-k46.

Prospective observational association study

Cox proportional hazards regression models were employed to assess the association between each RC subclass and CHD within the UKB cohort. Three sets of adjustments were applied:

Model I (basic model): Age, sex, assessment center, and NMR spectrometer were included as covariates.

Model II: Additional adjustments were made for ethnic background, smoking status, alcohol status, Townsend deprivation index, physical activity, educational levels, BMI, and fasting time.

Model III: LDL-C was further adjusted.

Subgroup analyses were conducted among individuals with LDL-C levels categorized as < 2.6, 2.6–3.4, and > 3.4 mmol/l, with a specific focus on the effect of RC in those with normal LDL-C levels. We additionally used restricted cubic splines to flexibly model the dose-response association between RC on a continuous scale and CHD, with three knots at 10^th, 50^th, and 90^th percentiles of RC. The same covariates were adjusted as those in Model II.

Colocalization analysis

The GWAS summary data of RC and its subclasses were obtained from the UKB repositories at OpenGWAS database, comprising 115,078 European ancestry. The summary data of 46, 959 CHD cases (total n = 412,181) were from the FinnGen study (version: R10) to avoid the sample overlapping.

Lead SNPs for total RC and its subclasses were identified using the default clumping method in the SNP2GENE tool of FUMA (26). For each of the final lead SNPs, all SNPs within a 1 Mb window of RC and CHD were extracted from the respective GWAS summary statistics. Colocalization analysis was then performed with a window of 1 Mb, using the R coloc package (version 5.2.3). Posterior probability of hypothesis 4 (PP.H4) ≥ 0.95 was set to indicate the shared causal variant for both traits. Ensemble database (https://useast.ensembl.org/) and GTEx Portal (https://www.gtexportal.org/) were subsequently used to determine the gene symbols that are most functionally associated with each corresponding risk locus.

Two-sample MR analysis

Univariable MR (UVMR) was performed using the TwoSampleMR R package (version 0.5.6) to assess causality between RC subclasses (exposure) and CHD (outcome) based on four methods: inverse variance weighted (IVW), MR Egger, weighted median, and weighted mode, of which we used IVW as the main analysis. According to the three key assumptions (relevance, independence and exclusion restriction), genetic instruments were first identified by clumping genome-wide significant SNPs with P < 5∗10^-8 (Linkage disequilibrium R² < 0.001 within a 10 Mb window). SNPs with F-values greater than 10 were selected as valid genetic instruments to avoid weak instrument bias. Heterogeneity and horizontal pleiotropy were estimated by the Cochran Q-test and MR Egger regression, respectively. Leave-one-out analysis was used to ensure the robustness of the results. MVMR was finally performed to derive the direct effect of RC controlling for LDL-C. During the MR process, we strictly adhered to the Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization (STROBE-MR) guidelines (Supplemental Table S2).

Validation in multiancestry dataset

The RC summary data for validation were derived from a meta-analysis study from 136,016 participants across 33 cohorts (22), including 4,435 individuals of East Asian ancestry, 11,340 of South Asian ancestry, and 120,241 of European ancestry. The GWAS summary data of 10,157 CHD cases (total n = 361,194) were obtained from UKB (ukb-d-I9_CHD, Supplemental Table S1), with no sample overlap between the exposure and outcome datasets. We conducted two-sample MR to infer causality using both UVMR and MVMR analyses. Colocalization analysis was performed to identify shared SNPs as causal variants for RC (PP.H4 > 0.95).

Tissue specific SMR analysis

To prioritize functionally candidate genes from the colocalized GWAS hits, we performed summary-data-based Mendelian randomization (SMR) (27). SMR utilized eQTL summary statistics to investigate the relationship between gene expression and phenotype of interest based on MR analysis. For the three prioritized gene targets, PSRC1, CELSR2, and SORT1, first, tissue cis-eQTLs were identified based on expression weights from GTEx version 8 (n = 838) and blood cis-eQTLs were from eQTLGEN Consortium (n = 31,684). Second, MR analyses were conducted to assess the causality of gene expression levels (mRNA) with CHD using the SMR software (27). We assessed the effect of three gene targets with CHD across eight tissues involved in lipid metabolism, including whole blood, liver, heart artery, heart left ventricle, artery aorta, tibial artery, subcutaneous adipose, and visceral adipose tissues.

Lipid-wide MR analysis

Lipid-wide MR was conducted to investigate the potential pleiotropic effects of the three candidate genes on various lipid traits, with the gene expression in liver as exposure and HDL-C, LDL-C, ApoB, TC and TG as outcome. SMR method was used to integrate lipid GWAS and eQTLs data to explore causal inference, with P value < 0.05 defined as statistically significant.

Specific role of candidate genes in regulating RC

To investigate the potential influence of candidate genes on RC independent of LDL-C, we performed MR analyses. In this analysis, we used the eQTLs of three genes as the instrumental variables, with circulating RC as outcome, and circulating LDL-C as the covariate. In additon, we obtained 43 target genes associated with approved lipid-modifying drugs identified from the WHOCC (https://www.whocc.no/), DrugBank (https://go.drugbank.com/), and ChEMBL (https://www.ebi.ac.uk/chembl/) (refer to Supplemental Table S3), of which only 30 had significant eQTLs in blood and 5 exhibited significant eQTLs in liver tissue (P < e^-5). SMR was subsequently performed to compare the effect of the three candidate genes against these drug target genes.

PPI network and enrichment analysis

The STRING database (https://string-db.org/) (28) was employed to elucidate the potential interactions among these genes. The Gene Ontology (https://geneontology.org/) and Kyoto Encyclopedia of Genes and Genomes (https://www.genome.jp/kegg/) database were used to help understand the functional characteristics of genes and proteins. Functional enrichment analyses were conducted using Enrichr (https://maayanlab.cloud/Enrichr/).

Results

Associations between RC and CHD independent from LDL-C in observational study

The observational study explored the epidemiological association between RC and CHD using the UKB data. Baseline characteristics of the participants are detailed in Supplemental Table S4. Among the 171,559 participants, 10,685 developed CHD after a median follow-up of 14.85 years. The CHD patients had much higher RC (including its seven subclasses) and LDL-C levels than the non-CHD individuals (P < 0.001). In Model I and Model II, total RC and all subclasses were significantly associated with an elevated risk of CHD (Fig. 2A). In Model III, which further adjusted for LDL-C, total RC demonstrated a linear relationship with CHD risk (Fig. 2B, P for nonlinearity > 0.05), with the HR increasing to 1.31 (95% CI: 1.25–1.38). Subgroup analyses revealed that in individuals with elevated LDL-C (≥2.6 mmol/l), RC was significantly associated with increased CHD risk (Fig. 2B, P < 0.001). Notably, among individuals with optimal LDL-C (<2.6 mmol/l), the relationship between total RC and CHD risk followed a nonlinear, “reverse L-shaped” association. While a visual elevation in the hazard ratio (HR) was noted at the extreme low end of the RC distribution, this was accompanied by markedly wider 95% confidence intervals, indicating statistical nonsignificance. To verify this, a stratified Cox proportional hazards analysis was performed (Supplemental Table S5). Compared to the reference group (RC: 1.10–1.30 mmol/l), individuals with RC levels below 1.10 mmol/L showed no significant increase in CHD risk (HR: 1.03 [0.93–1.14], P = 0.627). In contrast, risk escalated significantly once RC exceeded the 1.22 mmol/l threshold, with level > 1.30 mmol/l associated with a 14% increase in CHD risk (HR: 1.14 [1.06–1.23], P = 0.001). These findings highlighted that when RC was above 1.22 mmol/l, CHD risk increased significantly, even among individuals with well-controlled LDL-C.

Fig 2.

Observational associations between RC and its subclasses and CHD. A: Cox regression analysis among whole population with three models. B: RCS analysis among whole population and subgroups hazard ratio (solid red square) with 95% confidence interval (error bar) from Cox regression were adjusted for: Model I adjusted for sex, age, assessment center, and spectrometer; Model II adjusted for sex, age, assessment center, spectrometer, ethnicity, smoking status, alcohol status, townsend deprivation index (TPI), physical activity (MET), educational levels, body mass index (BMI), and fasting time; Model III adjusted for those covariates in Model II and LDL-C. Solid red square represents the hazard ratio for per SD increment in RC and its subclasses, while the error bar stands for 95% confidence interval. Restricted cubic splines (RCS) with three predefined knots (10^th, 50^th, and 90^th centiles) in Cox proportional hazards models were used to evaluate the nonlinear association between continuous remnant cholesterol levels and the risk of CHD. Hazard ratios (HRs) were indicated by solid red lines and 95% CIs by shaded areas. Reference lines for no associations were shown by the dashed gray lines at a hazard ratio of 1.0. All the analyses were adjusted for sex, age, assessment center, spectrometer, ethnicity, smoking status, alcohol status, townsend deprivation index (TPI), physical activity (MET), educational levels, body mass index (BMI), and fasting time. LDL-C was additionally included as covariate in whole population. CHD, coronary heart disease.

Furthermore, the majority of RC subclasses demonstrated positive correlations with the CHD risk, independent of LDL-C (Fig. 2A, Model III). Smaller VLDL subclasses demonstrated stronger associations; specifically, the HRs for S-VLDL-C and M-VLDL-C were both 1.27, whereas for the HR for the largest subclass, XXL-VLDL, was more modest at 1.10 (95% CI: 1.08–1.12). These associations remained significant even after stratifying the participants by LDL-C thresholds of <2.6 mmol/l and <3.4 mmol/l (Supplemental Figure S2). However, the association between IDL-C and CHD became statistically insignificant when LDL-C levels are within the normal range (Supplemental Figure S2), likely due to the strong metabolic correlation between IDL and LDL particles, which masks the independent contribution of IDL-C (Supplemental Figure S3).

Genetic evidence for the causal role of RC in CHD in the UKB dataset

Total RC and all its subclasses demonstrated strong evidence of colocalization with CHD, with a PP.H4 being close to one (Supplemental Table S6). Specifically, S-VLDL-C colocalized with six SNPs (Supplemental Table S7): rs1065853, rs11591147, rs12740374 (Fig. 3A), rs1883711, rs247616, and rs964184. The majority of the risk loci are in noncoding regions, which may influence the expression of several genes, including APOE-APOC1, PCSK9, CELSR2, MAFB-LINC01370, HERPUD1-CETP, and ZPR1 (Supplemental Table S7). Among these genes, four have been proven as lipid-modifying targets, including CETP, PCSK9, APOE, and APOC1.

Fig 3.

Genetic associations between RC and its subclasses and CHD. A: Colocalization of S-VLDL-C and CHD at rs12740374 in UKB dataset. B: Colocalization of total RC and CHD at rs646776 in multiancestry dataset. C: Mendelian randomization of RC subclasses and CHD in UKB dataset. D: Mendelian randomization of RC subclasses and CHD in multiancestry dataset for panel (A) and (B), each dot corresponds to a single SNP. The x-axis represents gene coordinates on chromosome (genome build: hg19) and y-axis showed their statistical significance in -log(P-value) with the corresponding phenotype. The r² refers to the strength of the correlation between the lead SNPs (rs12740374 & rs646776) and other genetic loci from linkage disequilibrium. For panel (C) and (D), solid black square represents the effect size (β) for per SD increment in RC and its subclasses, while the error bar stands for 95% confidence interval. MVMR refers to the analysis that incorporates adjustment for LDL-C. CHD, coronary heart disease; RC, remnant cholesterol; UKB, UK Biobank; MVMR, multivariable MR.

Findings from UVMR indicated that genetic predisposition to all RC subclasses were significantly positively correlated with CHD in the IVW models (Fig. 3C, P < 0.05), except XXL-VLDL-C]). Consistent results were observed using MR Egger, weighted median, and weighted mode methods (Supplemental Table S8). Despite the observed heterogeneity, no significant horizontal pleiotropy was detected for most RC subclasses (Supplemental Table S8, P for MR Egger intercept > 0.05). The effect sizes of individual SNPs are summarized in Supplemental Table S9, and leave-one-out analysis further confirmed the robustness of these results (Supplemental Figures S4–S11). Importantly, S-VLDL-C and L-VLDL-C demonstrated an independent effect of LDL-C on increasing CHD, with ORs of 1.24 (94% CI: 1.05–1.48) and 1.16 (95% CI: 1.04–1.29), respectively (Fig. 3C, Supplemental Figure S12).

Genetic causality validated via multiancestry datasets

We repeated the genetic analysis with a total of 136,016 participants from East Asian, South Asian, and European ancestries. We identified five causal SNPs with PP.H4 > 0.95 (Supplemental Table S10). Notably, one SNP (rs646776) in the intergenic region of PSRC1 and CELSR2 was identified with a posterior probability of 0.99 shared by RC and CHD (Fig. 3B). Genetically predisposed RC and its subclasses showed strong causal effects on increasing the risk of CHD across all models of UVMR analyses (Fig. 3D, Supplemental Table S11). Total RC, IDL-C, XS-VLDL-C, and S-VLDL-C remained significant after adjusting for LDL-C (Fig. 3D).

Three candidate genes as potential target and their liver-specific roles

Given that variants rs12740374 and rs646776 are identified as eQTLs for the genes PSRC1, CELSR2, and SORT1 (as shown in Supplemental Figure S13), we conducted a downstream SMR analysis to explore the associations between these three genes and CHD across eight different tissues. As shown in Fig. 4A & Supplemental Table S12, PSRC1 displayed a statistically significant inverse association with CHD in liver tissue, with OR of 0.94 (95% CI: 0.92–0.95). Similarly, both CELSR2 and SORT1 were inversely associated with CHD in liver, with an OR of 0.93 (95% CI: 0.91–0.95) and 0.94 (95% CI: 0.92–0.95), respectively. In contrast, the associations observed in blood samples were more varied: PSRC1 and SORT1 exhibited negative associations with CHD, while CELSR2 showed a positive association.

Fig 4.

PSRC1-CELSR2-SORT1 negatively regulate CHD and multiple lipid traits. A: SMR results of candidate genes with CHD among different tissues. B: SMR results of candidate genes in liver with HDL-C, LDL-C, TC, TG, and ApoB SMR: Summary-data-based Mendelian randomization; Solid red and blue square represent the odds ratio for per 1-SD increase in gene expression, while the error bar stands for 95% confidence interval in SMR analysis. CHD, coronary heart disease; SMR, summary-data-based Mendelian randomization.

In view of their liver specific effect, we further investigated the potential causal relationships between the hepatic expression levels of the three genes and various lipid traits (Fig. 4B). The analysis demonstrated that elevated mRNA levels of all three genes in the liver were significantly linked to increased levels of HDL-C, and decreased levels of ApoB, LDL-C, TC, and TG.

Comparison of PSRC1- CELSR2- SORT1 in regulating RC with lipid-modifying target genes

The PSRC1-CELSR2-SORT1 gene cluster in the liver exhibited a negative association with RC (Supplemental Figure S14), a relationship that persisted even after adjusting for LDL-C levels, as illustrated in Fig. 5A (P < 0.001). Given the three genes were strongly interconnected with PCSK9 (Fig. 5B) and predominantly enriched in pathways involving cholesterol metabolism (Supplemental Figure S15), we compare the effect of the three candidate genes against the current drug target genes. The three genes PSRC1, CELSR2, and SORT1 had a similar effect on RC as the well-known lipid-modifying drug target genes HMGCR (the target of statins) and PCSK9 (the target of PCSK9 inhibitors), with statistically significant associations (Figure 5C and D, P < 0.05 after FDR correction). Importantly, these three genes showed much stronger correlations with RC levels compared to other common lipid-modifying drug target genes, such as LPL and MMP25.

Fig 5.

Roles of PSRC1-CELSR2-SORT1 in regulating RC. A: MR of three genes with RC with adjustment of LDL-C levels. B: PPI network. C: Comparison of three genes’ effect on RC with approved lipid modifying targets genes in blood. D: Comparison of three genes’ effect on RC with approved lipid modifying targets genes in liver. RC, remnant cholesterol.

Discussion

More recent prospective cohort studies have reinforced the link between elevated RC and other cardiovascular events including atherosclerotic cardiovascular disease and strokes across age (8) and sex (29) groups, extending the evidence beyond CHD alone. In this study, we integrated observational and large-scale genomic data to investigate associations between RC subclasses and residual cardiovascular risk, and to examine genetic regulatory mechanisms. We demonstrated the role of RC and its subclasses as independent causal determinants of CHD, particularly under conditions of optimal LDL-C. Our MR analysis of eQTL data provided strong evidence for PSRC1-CELSR2- SORT1 in regulating RC and CHD, with a more pronounced effect than that of current lipid-modifying genes. This highlights the promise of these genes as therapeutic targets for reducing TRLs.

Our observational analysis of the UKB cohort demonstrated that RC and its subclasses serve as independent atherogenic factors for CHD even after adjusting for LDL-C. This aligns with previous research highlighting RC as a potent predictor of cardiovascular events alongside LDL-C and ApoB in primary prevention settings (30, 31, 32). In addition, our study specifically addressed the contribution of RC to residual CHD risk among individuals who have achieved optimal LDL-C targets. Our stratified analysis revealed that RC remained significantly associated with CHD risk among participants with “low” LDL-C levels (<2.6 mmol/L), characterized by a nonlinear “reverse L-shaped” relationship with a distinct inflection point at 1.22 mmol/L. This aligns with observations from the Korean diabetic cohort, which reported a 10%–14% increased risk of myocardial infarction and stroke in patients with RC > 0.8 mmol/l and LDL-C < 2.6 mmol/l (33). Our results extend these findings to the general population, identifying a specific RC threshold for residual risk. Although current European Society of Cardiology guidelines recommend using non-HDL cholesterol (non-HDL-C) as an additional target alongside LDL-C with thresholds set at < 2.2, 2.6 and 3.4 mmol/L for very-high-, high-, and moderate-risk groups (34), our findings suggest that RC, a main component of non-HDL-C, should be maintained at a stricter level of less than 1.22 mmol/l.

To assess the causal relationship between genetically predicted RC and CHD, we employed a polygenic MR framework, leveraging all informative SNPs identified from large-scale GWAS summary statistics as candidate instrumental variables. This genome-wide approach reduces susceptibility to horizontal pleiotropy, in contrast to previous MR studies that relied on candidate instrumental variants within a limited number of lipid metabolism genes, such as APOA5 and LPL (16, 35). Our findings demonstrate a robust causal link between genetic predisposition to elevated RC and increased CHD risk, with most RC subclasses showing significant positive associations. Notably, in the large-scale multiancestry dataset, MR analyses revealed a consistent trend where the causal effect decreased as particle size increased. This provides genetic support for prior observational findings (36) which showed that larger VLDL particles had a lower independent impact on CHD risk. Indeed, the independent casual associations for the larger subfractions (XL-VLDL-C and XXL-VLDL-C) were attenuated to nonsignificant after adjusting for LDL-C in both datasets, whereas smaller RC particles retained independent effects. These results align with previous studies that S-VLDL significantly increase myocardial infarction risk (37). Our findings supported the model that smaller RC particles possess greater atherogenic potential and contribute to residual cardiovascular risk independent of LDL-C burden. Larger RC particles, such as nascent chylomicrons and XXL-VLDL, are generally considered too large to cross the vascular wall, resulting in a comparatively lower atherogenic profile (38), while smaller RC particles may more readily penetrate the arterial wall and provoke local inflammatory responses that induce atherosclerotic plaque formation (39).

Using MR analysis integrated with eQTL data, we observed that increased hepatic expression of PSRC1, CELSR2, and SORT1 was associated with lower RC levels and reduced CHD risk. These genes form a cluster on chromosome 1p13.3, indicating a shared genetic architecture for the regulation of RC metabolism and cardiovascular outcomes. Notably, PSRC1, CELSR2, and SORT1 reside in the same linkage disequilibrium block and exhibit consistent MR estimates specifically in liver tissue, even after adjusting for LDL-C. The 1p13.3 locus has previously been reported to be associated with LDL-C levels across diverse populations, with liver-specific regulatory functions (40, 41, 42). The meta-analysis of GWAS studies has further validated that rs646776 and rs12740374 within the gene cluster were associated with an enhanced response to statin (43). Our findings extend the understanding of the PSRC1-CELSR2-SORT1 gene cluster, demonstrating its capacity to simultaneously regulate RC, HDL, TC, TG, and ApoB, beyond LDL. While these genes reside within a high linkage disequilibrium block, making the isolation of their independent causal effects challenges, our analysis highlights this region as a coregulated gene cluster for lipid metabolism. Within this cluster, SORT1 is the most extensively studied. It encodes the sortilin 1 protein, whose primary function involves binding to ApoB100 and facilitating the secretion of VLDL (main component of RC) (44). PSRC1, a key regulator of microtubule destabilization, inhibits macrophage inflammation and foam cell formation. Its deficiency exacerbates atherosclerosis by driving the production of trimethylamine N-oxide, which promotes plaque lipid deposition and macrophage infiltration while simultaneously impairing hepatic cholesterol transport and systemic lipid homeostasis (45). PSRC1 is also associated with medium HDL subfractions and is considered a high-priority therapeutic target for CHD management (46). CELSR2 plays a critical role in cell adhesion and ligand-receptor interactions. While implicated in a broad spectrum of pathologies, ranging from neurodevelopmental disorders (47, 48) to cancer (49), CELSR2 has been strongly linked to atherosclerotic cardiovascular disease development (50, 51, 52). Our findings regarding the PSRC1-CELSR2-SORT1 gene cluster are further supported by clinical evidence demonstrating that the 1p13 locus modulates postprandial lipemia. Carriers of the protective allele at rs646776 exhibit significantly lower fasting and postprandial VLDL particle numbers compared to risk-allele homozygotes (53). This suggests that the coregulated gene cluster identified in our study may confer protection by reducing the daily duration and intensity of exposure to atherogenic remnants, thereby mitigating long-term plaque progression. Notably, these three genes exhibit stronger correlations with RC than most genes currently targeted by lipid-modifying therapies, suggesting that current pharmacological interventions may be limited in their ability to address RC-mediated residual risk. While no approved therapies currently target the three genes directly (Supplemental Table S13), this cluster represents a promising therapeutic target to regulate RC metabolism due to its capacity to simultaneously regulate multiple lipid traits. Our findings support the potential of PSRC1-CELSR2-SORT1 as a broad-spectrum regulator of lipid homeostasis for cardiovascular disease prevention (54).

Taken together, this study supports a causal role for RC in CHD, based on consistent findings from observational epidemiologic data and complementary genetic analyses across diverse populations. Our results indicate that smaller RC particles exert stronger atherogenic effects than larger ones, reinforcing their role in residual risk. Integration of MR with colocalization, eQTL analyses strengthens causal inference and provides further mechanistic insights, identifying hepatic regulators such as the PSRC1-CELSR2-SORT1 cluster as a potential strategy to coregulate LDL, HDL, and TG levels for lipid homeostasis. Furthermore, the inclusion of large-scale summary statistics (>12 million for SNPs, >110,000 participants for RC and >360,000 for CHD), along with validation across multiple ancestral populations, ensured robust statistical power and extended generalizability of our findings. This highlights that RC is not only a “marker but a maker” of cardiometabolic disturbance but may also represent a modifiable target with therapeutic implications beyond LDL-C lowering.

Limitations

First, while we cannot entirely exclude the possibility that pleiotropic effects influenced our findings, the consistency across multiple statistical methods used to cross-validate the results suggests this is unlikely, and the MR-Egger intercept test revealed no evidence of significant horizontal pleiotropy. Second, while we have identified the PSRC1-CELSR2-SORT1 cluster as a high-priority regulator, we acknowledge that these conclusions are derived from statistical inferences of biological processes. More importantly, we are not able to disentangle the independent functional contributions of these genes or confirm their biological interactions, as they are coregulated and in strong linkage disequilibrium at the 1p13.3 locus. Consequently, further investigations, specifically functional biological assays such as knockdown or overexpression experiments of each individual gene are necessary to fully elucidate their independent roles in the pathogenesis of atherosclerosis.

Conclusion

This large integrated study establishes a robust causal association between RC and CHD, highlighting the importance of RC-lowering strategies, especially among individuals who have achieved optimal LDL-C levels. Our findings identify a distinct clinical threshold of <1.22 mmol/l for RC, below which cardiovascular risk remains stable. Furthermore, we characterize the PSRC1-CELSR2-SORT1 gene cluster as a high-priority regulatory region and a promising therapeutic target for the simultaneous modulation of multiple lipid traits. While our genomic analyses provide strong evidence for the three genes in RC metabolism, further functional validation is required to fully elucidate the independent molecular mechanisms of these genes in the pathogenesis of atherosclerosis.

Data availability

UKB data are available by application to the UK Biobank research team. UKB summary statistics of lipid traits and CHD were downloaded at IEU OpenGWAS Project (https://gwas.mrcieu.ac.uk/) with IDs met-d-Remnant_C, met-d-HDL_C, met-d-LDL_C, met-d-Total_C, met-d-Total_TG, and met-d-ApoB. Full GWAS summary statistics of RC for validation are available through the NHGRI-EBI GWAS Catalog (https://www.ebi.ac.uk/gwas) under study accession numbers GCST90301975, GCST90302019, GCST90302057, GCST90302074, GCST90302104, GCST90302138, GCST90302150, and GCST90302162. Access to the FinnGen GWAS summary data of CHD can be found at https://r10.finngen.fi/(I9_CHD).

Supplemental data

This article contains supplemental data.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.