Systematic elucidation of genetic mechanisms underlying cholesterol uptake

Abstract

Genetic variation contributes greatly to LDL cholesterol (LDL-C) levels and coronary artery disease risk. By combining analysis of rare coding variants from the UK Biobank and genome-scale CRISPR-Cas9 knockout and activation screening, we have substantially improved the identification of genes whose disruption alters serum LDL-C levels. We identify 21 genes in which rare coding variants significantly alter LDL-C levels at least partially through altered LDL-C uptake. We use co-essentiality-based gene module analysis to show that dysfunction of the RAB10 vesicle transport pathway leads to hypercholesterolemia in humans and mice by impairing surface LDL receptor levels. Further, we demonstrate that loss of function of OTX2 leads to robust reduction in serum LDL-C levels in mice and humans by increasing cellular LDL-C uptake. Altogether, we present an integrated approach that improves our understanding of genetic regulators of LDL-C levels and provides a roadmap for further efforts to dissect complex human disease genetics.

Figure 1:

CRISPR screening identifies genes required for LDL-C uptake A: Flow cytometric analysis of LDL-C uptake in HepG2 cells with CRISPR-Cas9 knockout of LDLR (red) and MYLIP (blue) in SER and SS conditions. n = 10. Two-tailed t-test. P-values are represented by asterisks (∗p<0.05, ∗∗p≤0.01, ∗∗∗p≤0.001). B: Schematic of LDL-C uptake CRISPR-Cas9 knockout screening. HepG2 constitutively expressing Cas9NG are incubated with a lentiviral gRNA library, followed by selection for infected cells. Cells are incubated in serum-containing or serum-starved conditions overnight followed by incubation with fluorescent LDL (BODIPY-LDL) for 4–6 hours followed by flow cytometric sorting of cells into 2–4 populations with most and least LDL-C uptake. Genomic DNA from sorted populations is prepared for nextgen sequencing (NGS), and gRNA counts are analyzed by MAGECK to identify genes associated with altered LDL-C uptake. C-E: Spearman correlation analysis of the ratio of representation of all gRNAs (C,D) or gRNAs in the 100 genes with strongest effects on LDL-C uptake (E) in cells with top 37.5% vs. bottom 37.5% LDL-C uptake in KO-Library 1 in SS conditions. F: Volcano plot showing the MAGECK LDL-C uptake log2-fold-change (x-axis) and minimum MAGECK -Log10-p-value (y-axis) for the 522 genes in KO-Library 4 in SS condition. Several significant genes are listed. Permutations = 100,000. G: Comparison of LDL-C uptake log2-fold-change for 20 genes between KO-Library 1 and individual gRNA knockout testing (y-axis).

Figure 2:

CRISPR activation screening reveals tunable regulators of LDL-C uptake A: Schematic of CRISPRa Outcome and Phenotype (COP) screening approach. A gRNA is cloned into a lentiviral vector paired with a 205-nt region of its target promoter driving expression of a barcoded reporter transcript (CD5FLAG2). HepG2 cells expressing all CRISPRa components (see text) are treated with this gRNA library followed by selection. CRISPRa outcome analysis is performed through collection of genomic DNA (gDNA) and RNA from infected cells followed by library prep to assess reporter activation as the ratio of barcode representation in cDNA vs. gDNA. CRISPRa phenotype analysis is performed through fluorescent LDL-C uptake flow cytometric screening followed by MAGECK analysis to identify gRNAs and genes that alter LDL-C uptake. B: Reporter cDNA:gDNA ratio of all promoter-paired gRNAs vs. control gRNAs, normalized by the average cDNA:gDNA ratio of control gRNAs for each gene. Mann-Whitney U-test. P-values are represented by asterisks (∗p<0.05, ∗∗p≤0.01, ∗∗∗p≤0.001). C: Spearman correlation analysis of individual flow cytometric cell surface antibody staining and reporter cDNA:gDNA ratio. n=2–4. D: Volcano plot showing the MAGECK LDL-C uptake log2-fold-change (x-axis) and minimum MAGECK -Log10-p-value (y-axis) for the 200 genes in the CRISPRa library. Several significant genes are listed. E: Comparison of LDL-C uptake log2-fold-change for 200 genes targeted by CRISPR-KO (x-axis) and CRISPRa (y-axis).

Figure 3:

Gene module analysis elucidates LDL-C uptake-altering mechanisms A: Co-essential modules with most significant enrichment of CRISPR-KO genes (hypergeometric p-value) labeled with the most significantly associated GO term for each module. B: tSNE plot based on PCA of module 255. Genes that decrease LDL-uptake when knocked out are labeled red, and genes that increase LDL-uptake when knocked out are labeled blue. Genes with pairwise Pearson correlation of >0.2 (purple) or <−0.2 (green) are connected with dashed lines. C: Module 255 contains RAB10 GTPase components including RAB10, the RAB10 holdase/chaperone RABIF, and the RAB10 GEF DENND4C and GAP RALGAPB. It also contains seven members of the exocyst complex that promotes vesicle exocytosis and three SNARE genes involved in vesicle fusion to the membrane. D: Mean control-normalized LDL-C uptake fold-change of HepG2 cells treated with DMSO (control) or one of 10 small molecules. Dunnett’s multiple comparisons test. n=6–26. P-values are represented by asterisks (∗p<0.05, ∗∗p≤0.01, ∗∗∗p≤0.001). E: Comparison of LDL-C uptake log2-fold-change for 522 genes targeted by CRISPR-KO in the presence of DMSO (x-axis) and Rocaglamide A (y-axis). Genes in co-essential modules annotated to be involved in translation initiation are shown in red and those involved in Golgi vesicle transport are shown in blue.

Figure 4:

Coding burden analysis reveals contributions of genes and modules to serum LDL-C A: Venn diagram showing genes with significant coding burden in the UKB cohort using two analysis methods. Both methods were employed genome-wide and restricted to CRISPR-KO-identified LDL-C uptake modulators. B: Correlation within UKB LDLR variant carriers between VEST4 score and change in adjusted LDL-C as compared to non-carriers. VEST4 scores are binned to the nearest 5%, the bubble sizes represent number of carriers within each bin, and a 95% confidence interval around the estimated linear regression line. C: Schematic of an approach to employ a collapsing burden model that accounts for variant deleteriousness to identify genes associated with altered serum LDL-C in the UKB exome cohort. D: Volcano plot showing the mean adjusted LDL-C difference (x-axis, mg/dL) and -Log10-p-value (y-axis) in the VEST4 burden analysis for 417 genes with significant effects on LDL-C uptake in the CRISPR-KO screen as well as 3 additional well-known LDL-C burden genes (gray). Genes significantly associated with increased (red) and decreased (blue) LDL-C levels in burden analyses are highlighted as well as selected non-significant genes (black). E. Bar chart displaying the number of genes significantly associated with serum LDL-C upon 10,000 random selections of 417-gene sets. Only 97/10,000 selections have at least as many significant genes (13) as when the 417 CRISPR-KO-significant genes are analyzed. F: Correlation within UKB RAB10 and RABIF variant carriers between VEST4 score and change in adjusted LDL-C as compared to non-carriers. VEST4 scores are binned to the nearest 5%, the bubble sizes represent the number of carriers within each bin, and a 95% confidence interval around the estimated linear regression line. G: Hazard ratios and 95% confidence intervals from a Cox proportional hazard estimates of all rare missense and LOF variants’ effect on CAD development. H: Volcano plot showing the UKB LDL-C rank co-essential GSEA normalized enrichment score (NES) (x-axis) and -Log10-p-value (y-axis) for 241 co-essential modules with a ClusterOne threshold (d=0.8). The four significant modules (red) and several enriched non-significant modules are highlighted along with all genes included in those modules. I: Dot plot showing the rank in mean carrier adjusted LDL-C for the 75% most deleterious variants for 16,843 genes with 20 or more carriers. Select genes including those in the RAB10-associated co-essential module are highlighted with those significantly associated with hypercholesterolemia shown in red and hypocholesterolemia in blue.

Figure 5:

Disrupting the RAB10 vesicle transport pathway impedes LDLR membrane trafficking A: Median anti-LDLR flow cytometric staining intensity for HepG2 cells with gene KO as indicated. Two-tailed t-test. n=2. B: Fluorescent microscopic images of HepG2 cells with inducible LDLR-mCherry expression and gene KO as indicated. LDLR-mCherry expression was induced 6 hours prior to imaging. C: Flow cytometric expression of LDLR-mCherry in HepG2 cells with inducible LDLR-mCherry expression and gene KO as indicated. LDLR-mCherry expression was induced 24 hours prior to measurement in the indicated media condition. n=2–4. Two-tailed t-test. P-values are represented by asterisks (∗p<0.05, ∗∗p≤0.01, ∗∗∗p≤0.001). D: Bar chart showing co-essential module enrichment among sgDENND4C RNA-seq upregulated genes.

Figure 6:

Mouse knockdown of Rabif, Csk, and Otx2 alter serum cholesterol levels A: Comparison of the effects of CRISPR-Cas9 gene KO (x-axis) and ectopic ORF expression (y-axis) for CSK, OTX2, RABIF, and non-targeting control. Error bars represent standard error. B-C: Scatter plots showing UKB mean carrier adjusted LDL-C (x-axis) vs. mean carrier HDL-C (B, y-axis) and mean carrier triglycerides (C, y-axis) for all genes. Benchmark genes and novel genes are highlighted. D: HepG2 control or OTX2 KO RNA-seq reads per million (RPM) for the 100 genes that are most robustly upregulated in HepG2 cells upon serum starvation. Paired Mann-Whitney U-test. E: RNA-seq RPM for OTX1 in HepG2 cultured in serum-containing or serum-starved conditions. n=46. Two-tailed t-test. P-values are represented by asterisks (∗p<0.05, ∗∗p≤0.01, ∗∗∗p≤0.001). F: Schematic of mouse AAV shRNA experiment. Retro-orbital AAV8 shRNA injection is performed on 8-week old male mice followed by blood collection 2, 4, and 6 weeks post-injection. Western diet feeding begins 4 weeks post-injection, and endpoint analysis is performed 6 weeks post-injection. Cohort sizes for each treatment group are listed. G: Average percent change in total cholesterol from pre-treatment measurement in mice treated with the designated AAV8 shRNA at the designated timepoints. H: Average LDL/VLDL (mg/dL) in mice treated with the designated AAV8 shRNA at the designated timepoints. Average difference compared to shScrambled and p-value are noted. I: Average HDL (mg/dL) in mice treated with the designated AAV8 shRNA at the designated timepoints. Average difference compared to shScrambled and p-value are noted.

Figure 7:

LDL uptake-altering genes may underlie GWAS loci A: Fine-mapping of the GWAS signal at the RAB10 locus (black: LDL-GWAS variants, red: 95% confidence set, teal: 99% confidence set). Signal tracks represent chromatin accessibility from DNase-seq or H3K27me3 histone modifications from ChIP-seq. Blue bars represent ABC predicted enhancer connections with RAB10. B: Flow cytometric analysis of HepG2 cells with expression of a GFP-RAB10 fusion protein expression construct containing the full RAB10 3’ UTR with rs142787485 major (blue) or minor (red) allele. C: GFP-RAB10 3’ UTR mean GFP flow cytometry. n=6. Two tailed t-test. P-values are represented by asterisks (∗p<0.05, ∗∗p≤0.01, ∗∗∗p≤0.001). D: Fine-mapping of the GWAS signal at the DENND4C locus (black: LDL-GWAS variants, red: 95% confidence set, teal: 99% confidence set). Signal tracks represent chromatin accessibility from DNase-seq or H3K27me3 histone modifications from ChIP-seq. Blue bars represent ABC predicted enhancer connections with DENND4C.