Genetic Mimicry Analysis Reveals the Specific Lipases Targeted by the ANGPTL3-ANGPTL8 Complex and ANGPTL4

Abstract

Angiopoietin-like proteins, ANGPTL3, ANGPTL4, and ANGPTL8, are involved in regulating plasma lipids. In vitro and animal-based studies point to LPL and endothelial lipase (EL, LIPG) as key targets of ANGPTLs. To examine the ANGPTL mechanisms for plasma lipid modulation in humans, we pursued a genetic mimicry analysis of enhancing or suppressing variants in the LPL, LIPG, lipase C hepatic type (LIPC), ANGPTL3, ANGPTL4, and ANGPTL8 genes using data on 248 metabolic parameters derived from over 110,000 nonfasted individuals in the UK Biobank and validated in over 13,000 overnight fasted individuals from 11 other European populations. ANGPTL4 suppression was highly concordant with LPL enhancement but not HL or EL, suggesting ANGPTL4 impacts plasma metabolic parameters exclusively via LPL. The LPL-independent effects of ANGPTL3 suppression on plasma metabolic parameters showed a striking inverse resemblance with EL suppression, suggesting ANGPTL3 not only targets LPL but also targets EL. Investigation of the impact of the ANGPTL3-ANGPTL8 complex on plasma metabolite traits via the ANGPTL8 R59W substitution as an instrumental variable showed a much higher concordance between R59W and EL activity than between R59W and LPL activity, suggesting the R59W substitution more strongly affects EL inhibition than LPL inhibition. Meanwhile, when using a rare and deleterious protein-truncating ANGPTL8 variant as an instrumental variable, the ANGPTL3-ANGPTL8 complex was very LPL specific. In conclusion, our analysis provides strong human genetic evidence that the ANGPTL3-ANGPTL8 complex regulates plasma metabolic parameters, which is achieved by impacting LPL and EL. By contrast, ANGPTL4 influences plasma metabolic parameters exclusively via LPL.

Keywords: angiopoietin-like proteins; cardiovascular disease; dyslipidemias; lipase/endothelial; lipase/hepatic; lipidomics; lipids; lipolysis and fatty acid metabolism; lipoprotein/metabolism; triglycerides.

Fig. 1

ANGPTL3-, ANGPTL4-, andANGPTL8R59W-mediatedgenetic LPL disinhibition shows distinct patterns of LPL mimicry. The similarity of genetic LPL enhancement and ANGPTL3, ANGPTL4, or ANGPTL3-ANGPTL8 complex suppression was compared using data derived from up to 115,078 individuals on 248 metabolic parameters in the UK Biobank (derivation cohort) and validated in up to 24,925 individuals on 122 metabolic parameters in an independent European validation set. A: LPL enhancement versus ANGPTL3 suppression in the derivation cohort. B: LPL enhancement versus ANGPTL4 suppression in the derivation cohort. C: LPL enhancement versus an ANGPTL3-ANGPTL8 complex function altering variant (ANGPTL8 R59W) in the derivation cohort. D: LPL versus ANGPTL3 in the validation set. E: LPL versus ANGPTL4 in the validation set. F: LPL versus ANGPTL8 R59W in the validation set. Each scatter dot represents the effect of the genetic variant on a lipoprotein lipid or plasma metabolite. The gray solid lines starting from each dot indicate the standard error of the estimate. The black dashed line indicates the regression estimate. The 95% CI of the regression is indicated by the gray area adjacent to the line. The gray dashed line indicates a reference regression line with a coefficient = 1. To ease comparisons, each effect estimate for the variant effect on each metabolic parameter was scaled so that they represent the 1-SD effect per 1-SD effect on total plasma TGs. The mimicry of effects was estimated using linear regression. The conditional variance explained (R²) indicates the degree of LPL enhancement that is explained by genetic variation in ANGPTL3, ANGPTL4, or ANGPTL8 R59W.

Fig. 2

ANGPTL3 suppression adjusted for the effects of theANGPTL8R59W coding mutation shows a remarkably high degree of LPL mimicry. A: Multivariable model of ANGPTL3 rs11207977-T and ANGPTL8 rs2278426 [R59W] in the UK Biobank cohort (N = 110,058–115,078, NMR parameters = 248) showed a mimicry of 92% (R²). B: The pattern of increased explained variance was concordant with the European validation set (N = 13,171–24,925, NMR parameters = 122), with 72% conditional variance explained (R²). C: The directed acyclic graph (DAG) summarizes the causal model for the regression analysis. In ANGPTL3, there was variance that is related to LPL enhancement (X^R) and variance unrelated to LPL enhancement (X^U). D: Bar plot comparing the explained variance of each genetic instrument. X^R in ANGPTL3 (instrumented through decreased ANGPTL3 transcription) was 55–58%, whereas the X^R of altered ANGPTL3-ANGPTL8 complex function was 0% (instrumented through the ANGPTL8 R59W mutation). However, when correcting for the ANGPTL3-ANGPTL8-mediated LPL disinhibition in an ANGPTL3 + ANGPTL8 model, the proportion of conditional variance explained increased to 72% and 92%. E: The coefficient of ANGPTL3 suppression accounting for ANGPTL8 R59W increased from 0.90 to 1.45 in the derivation cohort. F: The pattern was replicated using the same method in the validation set (unconditional coefficient = 1.05, conditional coefficient = 1.44). Together, this suggested that ANGPTL8 rs2278426-T [R59W] acts as a statistical suppressor (73). This means that it is correlated to, and corrects for, the unrelated variance (X^U) in the relationship between ANGPTL3 and LPL. In this case, X^U should be read as the systemic effects of decreased ANGPTL3 transcription via pathways unrelated to LPL (and random error).

Fig. 3

The ANGPTL8 R59W missense mutation shifts the inhibitory action of ANGPTL3-ANGPTL8 complexes from LPL to EL (LIPG). A: EL versus ANGPTL8 R59W in the European validation set (N = 110,058–115,078, NMR parameters = 248). The systemic effects of the ANGPTL8 R59W substitution strongly resemble the effects of EL suppression (R² = 94%). B: EL versus ANGPTL8 R59W in the European validation set (N = 13,171–24,925, NMR parameters = 122). The coefficients, but not the conditional variance explained (R² = 38%), were replicated in the validation set analysis. The EL validation set results should be interpreted with caution. There were only up to 66–175 LIPG rs77960347-G carriers per metabolic parameter (for allele frequency, see Table 2), which could lead to statistical imprecision. Since EL inhibition had a small effect on total TG levels (Table 2), the effect estimates were scaled so that they represent the 1-SD effect per 1-SD total cholesterol (TC) change (in contrast to Figs. 1 and 2). The different scaling does not influence the conditional explained variance (R²).

Fig. 4

AnANGPTL8PTV is highly LPL specific. Data on the very rare (allele frequency = 0.04%) ANGPTL8 Q121X coding variant was obtained from an exome sequencing analysis of the derivation cohort (N = 110,058–115,078). PTVs introduce a premature stop codon that disrupts transcription and causes the translation of a shortened protein. The figure shows the effects of an LPL-enhancing eQTL versus the ANGPTL8 Q121X PTV on 248 metabolic parameters. The degree of LPL mimicry was remarkedly high (R² = 97%) and comparable to the LPL specificity of ANGPTL4 (Fig. 1B, E). These results strongly suggest that ANGPTL3-ANGPTL8 complexes act on plasma lipids through inhibition of LPL.

Fig. 5

Genetic evidence for ANGPTL3 inhibition of both LPL and EL. The similarity of EL (LIPG) suppression was compared with the residuals from the LPL versus ANGPTL3 model (Fig. 1A, D) in the derivation (N = 110,058–115,078, NMR parameters = 248) and validation sets (N = 13,171–24,925, NMR parameters = 122). A: EL (LIPG) suppression explains the variance in ANGPTL3 that is unrelated to LPL enhancement in the derivation set (X^U, Fig. 2), with an explained conditional variance (R²) of 93%. B: The EL versus ANGPTL3 conditional on LPL results were replicated in the European validation set; however, with a smaller R² of 69%. These data confirm that ANGPTL3 affects plasma lipids through both LPL and EL.

Fig. 6

HL (LIPC) and EL (LIPG) inhibition show minimal concordance with ANGPTL4 suppression. HL activity was instrumented through LIPC rs1800588-T, a promoter variant associated with decreased LIPC promoter activity and lower postheparin HL activity. EL inhibition was instrumented using LIPG rs77960347-G, a missense mutation (N396S) leading to decreased EL activity. A: HL versus ANGPTL4 suppression in the derivation cohort using data on 248 NMR metabolic parameters derived from 110,058 to 115,078 UK Biobank individuals. B: EL versus ANGPTL4 suppression in the derivation cohort. C: HL versus ANGPTL4 suppression using data on 122 NMR metabolic parameters derived from 13,171 to 24,925 in the non-UK European validation set. D: EL versus ANGPTL4 suppression in the validation set. Because EL had a small effect on TG levels (Table 2), the EL effect estimates were scaled so that they represent 1-SD effect per 1-SD total cholesterol (TC) change (in contrast to HL and LPL). The different scaling has no effect on the conditional explained variance (R²).