Functional interrogation of cellular Lp(a) uptake by genome-scale CRISPR screening

Abstract

Background and Aims:

An elevated level of lipoprotein(a), or Lp(a), in the bloodstream has been causally linked to the development of atherosclerotic cardiovascular disease and calcific aortic valve stenosis. Steady state levels of circulating lipoproteins are modulated by their rate of clearance, but the identity of the Lp(a) uptake receptor(s) has been controversial.

Methods:

We performed a genome-scale CRISPR screen to functionally interrogate all potential Lp(a) uptake regulators in HuH7 cells. Screen validation was performed by single gene disruption and overexpression. Direct binding between purified lipoproteins and recombinant protein was tested using biolayer interferometry. An association between human genetic variants and circulating Lp(a) levels was analyzed in the UK Biobank cohort.

Results:

The top positive and negative regulators of Lp(a) uptake in our screen were LDLR and MYLIP, encoding the LDL receptor and its ubiquitin ligase IDOL, respectively. We also found a significant correlation for other genes with established roles in LDLR regulation. No other gene products, including those previously proposed as Lp(a) receptors, exhibited a significant effect on Lp(a) uptake in our screen. We validated the functional influence of LDLR expression on HuH7 Lp(a) uptake, confirmed in vitro binding between the LDLR extracellular domain and purified Lp(a), and detected an association between loss-of-function LDLR variants and increased circulating Lp(a) levels in the UK Biobank cohort.

Conclusions:

Our findings support a central role for the LDL receptor in mediating Lp(a) uptake by hepatocytes.

INTRODUCTION

Lipoprotein(a), or Lp(a), was discovered in 1963 as a unique variant of low-density lipoprotein (LDL)¹. Early cross-sectional and retrospective studies suggested an association between elevated Lp(a) and coronary artery disease (CAD)^2,3 that was confirmed in multiple larger prospective studies^4-8. Lp(a) levels vary from <1 to >200 mg/dL in the general population, with each 2-fold increase associated with an estimated 22% greater risk of myocardial infarction in a study population of European ancestry^7,9. Human genetic studies have established the causal influence of Lp(a) elevation on the development of both atherosclerotic cardiovascular disease^7,10-14 and calcific aortic valve stenosis^15-18, a common cause of heart failure and death. In light of this evidence, Lp(a) has become an attractive target for therapeutic development. While a clinical trial for LPA antisense oligonucleotide-based therapy is eagerly anticipated¹⁹, clinical experience with LDL-targeted treatments has demonstrated the benefits of a multifaceted approach to lowering atherogenic lipoproteins.

Lp(a) levels are highly heritable and largely determined by heterogeneity at the LPA locus, which encodes the apolipoprotein(a), or apo(a), component of Lp(a)^20,21. Copy number variants in the KIV2 domain of LPA are inversely correlated with Lp(a) levels²², presumably due to more efficient maturation and secretion of smaller apo(a) isoforms^23,24. The steady state level of circulating Lp(a) also depends on its rate of clearance, which occurs primarily in the liver^25,26. The molecular basis of Lp(a) clearance, however, remains poorly understood. Multiple studies of the LDL receptor (LDLR) in cells^27-34, animal models^25,32,35,36, and humans^10,28,37-48 have provided inconsistent or conflicting results. Over time, several other candidate Lp(a) receptors have been proposed, including other lipoprotein receptors^49,50, toll-like receptors⁵¹, scavenger receptors^52,53, lectins^26,54, and plasminogen receptors^29,33. It has also been proposed that multiple receptors may contribute to Lp(a) clearance⁵⁵. In general, prior functional studies of Lp(a) uptake have involved a variety of cell types and methods for Lp(a) purification and detection. To date, no study has reported a systematic and unbiased functional interrogation of all potential Lp(a) receptors in the same experimental context.

We previously applied a genome-scale CRISPR screen to successfully identify regulators of cellular LDL endocytosis⁵⁶. We readily detected expected genes encoding the LDL receptor and its canonical transcriptional and posttranscriptional regulators. We also detected novel regulators of LDL uptake that were reproducible across independent replicates and during follow up testing^56,57. Given the excellent technical performance of this approach, in this study we sought to adapt it to identify functional modifiers of Lp(a) uptake. The results of our screen and our subsequent validation, mechanistic investigation, and analysis of human genetic variants all support a primary role for the LDL receptor in mediating Lp(a) uptake by hepatocytes.

MATERIALS AND METHODS

Cell lines and reagents.

HuH7 and HEK-293T cells (ATCC, Manassas VA) were cultured in DMEM supplemented with 10% fetal bovine serum, 10 U/mL penicillin, and 10 μg/mL streptomycin (ThermoFisher Scientific, Waltham MA) in a humidified 5% CO2 chamber at 37°C. Cell lines were periodically tested for mycoplasma contamination and verified by microsatellite genotyping. Generation and genotyping of a LDLR-disrupted HuH7 clonal line was previously described⁵⁶. A LDLR expression construct was prepared by PCR amplification of LDLR cDNA from a plasmid template (GenScript Biotech, Piscataway NJ, #OHu22799) with primers (Integrated DNA Technologies, Coralville IA) designed to provide flanking homology arms for HiFi assembly (New England Biolabs, Ipswich MA) into EcoRI/NotI-digested (New England Biolabs) LeGO-ACE2-IRES-blast⁵⁸. Lentivirus was generated as previously described⁵⁷ and used to either overexpress or rescue LDLR by transduction of HuH7 wild-type or LDLR-disrupted cells, respectively. Commercial Lp(a) preparations (Athens Research and Technology, Athens GA, #12-16-121601) were purified from single donor human plasma by sequential density ultracentrifugation with potassium bromide and size exclusion chromatography using a Sephacryl 400 column. Supplier validation (data not shown) of Lp(a) preparations relative to lipoprotein standards was performed with SPIFE Vis Cholesterol assay (Helena Laboratories, Beaumont TX, #3218), protein gel electrophoresis with Coomassie Blue staining, and Lp(a) immunoturbidometric assay (Diazyme, Laboratories, Poway CA, #DZ131B). Commercial LDL (Athens Research and Technology, #12-16-120412) and HDL preparations (Athens Research and Technology, #12-16-080412) were purified from human plasma and validated by the supplier against lipoprotein standards with SPIFE Vis Cholesterol assay and protein gel electrophoresis with Coomassie Blue staining. We also analyzed apo(a) isoform distribution by immunoblotting with a primary apo(a) antibody (Abcam #ab208184) at 1:1000 dilution as previously described⁵⁷.

Lp(a) labeling.

For initial pilot studies, purified Lp(a) preparations were treated with a variety of fluorescent labeling strategies – FITC Conjugation Kit (Abcam, Cambridge UK, #ab188285), Dil (1,1'-Dioctadecyl-3,3,3',3'-Tetramethylindocarbocyanine Perchlorate) Stain (ThermoFisher, #D282), and pHrodo Red Microscale Labeling Kit (ThermoFisher, #P35363) – all per manufacturer’s instructions. Subsequent labeling experiments, including for the genome-scale CRISPR screen, were performed using pHrodo iFL Red Antibody Labeling Kit (ThermoFisher, #P36014) at a lipoprotein to dye molar ratio of 1:28 for 15 mins in the dark at room temperature. Unconjugated free dye was removed from each sample by extensive dialysis against PBS. A portion of labelled and unlabeled lipoprotein samples were analyzed by SDS-PAGE on NuPAGE 3-8% gradient Tris-Acetate gels (ThermoFisher, # EA0375) under reducing and nonreducing conditions followed by SilverQuest Silver Staining (ThermoFisher, # LC6070) per manufacturer’s instructions or by in-gel fluorescence scanning with settings for Red Epifluorescence excitation and 700/50 nm emission using a ChemiDoc MP Imaging System (Bio-Rad Laboratories, Hercules CA).

Analysis of Lp(a) uptake.

Cells were seeded in 6-well or 12-well plates for flow cytometry or ELISA or in Lab-Tek II Chamber Slides (ThermoFisher, #154461) for immunofluorescence and analyzed 2-3 days later when their confluence reached 70-90%. To quantify cellular Lp(a) uptake, monolayers were washed once with serum-free DMEM and incubated in serum-free DMEM containing unlabeled or fluorescent Lp(a) in the presence or absence of competitive inhibitors at the indicated concentrations for the indicated duration at 37°C. For flow cytometry analysis, cells were detached with TrypLE express (ThermoFisher, #12605036), washed twice with ice cold PBS, resuspended in ice-cold PBS, and analyzed on a Ze5 flow cytometer (Bio-Rad Laboratories) with gating and quantification of fluorescence intensity performed using FlowJo. Mean fluorescence intensities were normalized to control wild-type cells and adjusted with subtraction of background autofluorescence for control samples analyzed in parallel in the absence of exogenous Lp(a). For apo(a) ELISA, cells were washed once with PBS, scraped into ice-cold PBS suspensions, and centrifuged at 500 x g. Protein lysates were extracted from cell pellets by resuspension and incubation in RIPA buffer with end-over-end rotation for 15 mins followed by centrifugation at 21,000 x g for 30 min at 4°C and transfer of supernatants to new tubes for immediate analysis or storage at −80°C. Total protein concentrations of lysates were quantified by BCA Protein Assay (ThermoFisher, #23225) and each sample was analyzed in technical duplicates for apo(a) protein concentration by ELISA (Abcam, #ab108878) with detection on a Spark Multimode Microplate Reader (Tecan, Männedorf Switzerland). Apo(a) concentrations were normalized to total protein concentrations of lysate and adjusted with subtraction of nonspecific background signal for control samples analyzed in parallel in the absence of exogenous Lp(a). For immunofluorescence, cells were fixed with pre-chilled methanol for 10 minutes, washed with PBS, incubated with blocking buffer (1% bovine serum albumin in PBS) for 60 mins, and immunostained by overnight incubation at 4°C with an apo(a) biotinylated primary antibody (Abcam, #ab27631) at 1:150 dilution in antibody dilution buffer (1% BSA, 1% glycine, 0.2% sucrose in PBS). Cells were then washed 3 times with wash buffer (0.5% bovine serum albumin in PBS), incubated with a streptavidin-conjugated Alexa Fluor 647 secondary antibody (ThermoFisher, #S32357) at 1:1000 dilution in blocking buffer for 2 hours at room temperature, and washed another 3 times with wash buffer. Coverslips were mounted and nuclei stained with ProLong Diamond Antifade Mountant with DAPI (ThermoFisher, #P36971). Imaging was performed using a Leica DM 5000B fluorescence microscope.

CRISPR screen.

For each independent biological replicate, a total of ~60 million HuH7 cells distributed in 10 separate 15 cm diameter cell culture plates at ~30% confluence were transduced with a pooled lentiviral library containing the GeCKOv2 library⁵⁹ at MOI ~0.5. Selection of transduced cells with 2.5 μg/mL puromycin was started 24 hr post-transduction and maintained until no viable cells remained in parallel control plates that were not transduced with lentivirus. Cells were passaged every 2-3 days to maintain logarithmic phase growth with a minimum of ~30 million cells, representing >200X library coverage, at all times. On day 14 post-transduction, pools of edited cells were incubated with 20 μg/mL single donor-derived pHrodo iFL Red-labelled Lp(a) in serum-free DMEM for 2 hrs at 37°C and prepared for flow cytometry as described above. Cell sorting was performed on a FACSAria III instrument (Becton Dickinson, Franklin Lakes NJ) with collection of 10% subpopulations of edited cells with the least and greatest magnitude of Lp(a) fluorescence. Genomic DNA was extracted from sorted populations using a DNeasy kit (Qiagen, Hilden, Germany). A range of 5 to 9 million cells were collected in each bin for each replicate. Amplicon sequencing libraries were prepared as previously described^56,60,61 and sequenced with a 1 x 75 single end read on a NextSeq instrument (Illumina, San Diego CA). Individual gRNA reads were extracted from FASTQ files with PoolQ and analyzed for enrichment using MAGeCK⁶². Genes identified as Lp(a) uptake modifiers with FDR<20% were analyzed for enrichment in Gene Ontology annotations using PANTHER v18.0⁶³ and for gene-gene interactions using STRING v12.0⁶⁴ with visualization of results by Cytoscape v3.9.1⁶⁵.

In vitro binding assays.

Biolayer interferometry experiments were performed by incubating Octet Streptavidin Biosensors (Sartorius, Göttingen Germany, #18-5019) with 10 ug/mL biotinylated recombinant LDLR extracellular domain (BPS Bioscience, San Diego CA, #71206) for 20 mins and transferring LDLR-immobilized or control sensors to multiwell plates containing the indicated lipoprotein concentration in Octet Kinetics Buffer (Sartorius, 1#8-1105) at room temperature for 60 seconds followed by transfer to wells containing buffer only for an additional 180 seconds. Detection was performed with measurement of light interference every 0.2 seconds for 4 minutes on an Octet RED96 System instrument (Forte Biosciences, Dallas TX).

UK Biobank analysis.

The UK Biobank has been previously described⁶⁶. Protocols were approved by the North West Multi-centre Research Ethics Committee and analyses in this project were conducted under UKBB Resource Project 18120 with IRB approval. Variant calling and mapping of the July 2022 release of whole exome sequencing (WES) was performed by the OQFE procedure as described by Krasheninina et al⁶⁷. Quality control of variant call-rate, sample call-rate, read-depth, allelic balance, and Hardy-Weinberg equilibrium was performed as described by Szustakowski et al⁶⁸. LDLR variants were selected with Annovar⁶⁹ using the hg38 refGene database. Analysis was performed controlling for age (Datafield 21003), age², sex (Datafield 22001), and the first 10 genetic principal components (Datafield 22009). A parallel analysis was also conducted that included genotype at the rs10455872 SNP as a covariate. Genetic relatedness was accounted for using array genotypes⁷⁰. Individuals were excluded if they did not contain WES, genotyping array, serum LDL, and serum Lp(a) data. Summary statistics regarding the study population can be found in Supplementary Table 3. A total of 359,090 individuals were considered for the final analysis. Direct LDL cholesterol (Datafield 30780) and Lp(a) levels (Datafield 30790) were initially extracted from both instance 0 and instance 1. The final serum levels per individual were chosen by selecting the earliest instance with both non-missing serum values. Age and medication use (Datafield 20003) were retrieved to match the instance used for the phenotype. LDL levels were adjusted by an increase of 25% for the use of lipid-lowering medications as previously described⁷¹. Lp(a) levels were adjusted with decrease by 11% for the use of statins based on their mean effect size in a recent meta-analysis¹⁹ (or 7% and 14% representing the 5-95% confidence interval of effect size in the same study) and by an increase of 30% for the use of niacin based on its estimated effect⁷². An adjustment for PCSK9 inhibitors was omitted as no individuals in this data release were identified as taking these medications. A list of the included medications and their codes is provided in Supplemental Table 3. Burden testing was performed using a pre-defined set of 61 variants annotated in ClinVar⁷³ (accession SCV004022429.1) as pathogenic based on Clinical Genome Resource (ClinGen) Familial Hypercholesterolemia Variant Curation Expert Panel consensus guidelines⁷⁴. SNPs were tested using Regenie⁷⁰. Association testing was conducted for raw, adjusted, and rank-based inverse normalized transformations of the LDL and Lp(a) levels. We also performed a parallel analysis of Europeans only as determined by Oliveri et al⁷⁵. ClinVar functional annotations of individual variants were extracted from gnomADv4.0⁷⁶.

RESULTS

Development of a flow cytometry-based assay for cellular Lp(a) uptake.

To enable a forward genetic approach, we first set out to develop a method for the sensitive and specific selection of individual cells with aberrant Lp(a) uptake. We obtained Lp(a) from a commercial supplier that had been purified from single donor human plasma by ultracentrifugation and size exclusion chromatography and verified to lack detectable contaminating lipoproteins by agarose gel electrophoresis and total cholesterol staining. We performed immunoblotting of this Lp(a) preparation that revealed a single dominant band with MW >250 kDa (Fig 1A), which may be compatible with two apo(a) isoforms of identical sizes, two apo(a) isoforms with a size difference less than the limited resolution of our assay, or with a second apo(a) isoform that is poorly expressed and/or secreted from hepatocytes. Incubation of this purified Lp(a) preparation with HuH7 cells confirmed its cellular uptake, with immunofluorescence revealing apo(a) localization to punctate structures consistent with endolysosomal trafficking (Fig S1). However, our attempts to apply this same apo(a) antibody or other commercial apo(a) antibodies for flow cytometry failed to demonstrate specific staining (data not shown). We therefore instead shifted our efforts toward direct fluorescent labeling of purified Lp(a) preparations.

Figure 1. Development of a method for fluorescent labeling of purified Lp(a) and quantification of its cellular uptake with flow cytometry.

(A) Immunoblotting for apo(a) isoform distribution in the purified Lp(a) preparation. (B) Overt sample precipitation during attempted fluorescent labeling of purified Lp(a) preparations with pHrodo Red. (C) SDS-PAGE and silver straining of purified Lp(a) preparations before and after attempted NHS-ester fluorescent labeling with FITC and pHrodo Red and lipid labeling with DiI. Upper arrow corresponds to precipitated protein remaining in loading well. Lower arrow corresponds to expected size (~500 kDa) of co-migrating apo(a) and apolipoprotein B proteins present in unlabeled Lp(a) preparations and absent after fluorescent labeling. (D) SDS-PAGE under reducing conditions of purified Lp(a) and LDL preparations before and after STP-ester fluorescent labeling with pHrodo iFL Red followed by silver staining. Arrow corresponds to expected comigration of noncomplexed apo(a) and apolipoprotein B under reducing conditions. (E) SDS-PAGE under nonreducing conditions of the same samples as in (D) followed by silver staining. Upper arrow corresponds to the expected migration of disulfide-linked apo(a)-apolipoprotein B complexes in Lp(a); lower arrow corresponds to the expected migration of free apolipoprotein B in LDL. (F) SDS-PAGE under reducing conditions of unlabeled and pHrodo iFL Red-labeled Lp(a) followed by in gel fluorescence scanning. Arrow corresponds to the migration observed for the dominant band after silver staining in (D). (G) Time course analysis of HuH7 cellular uptake of 10 μg/mL pHrodo iFL Red-labeled Lp(a). (H) Concentration-dependence of HuH7 cellular uptake of pHrodo iFL Red-labeled Lp(a) over 2 hrs. (I) Dose-dependent competitive inhibition of HuH7 cellular fluorescent Lp(a) uptake by co-incubation with a molar excess of unlabeled Lp(a). Representative flow cytometry plots are provided in Supplemental Figure 2.

Our initial attempts to fluorescently label Lp(a) with either NHS-ester conjugation of primary amines (FITC, pHrodo Red) or lipid labeling (DiI) were complicated by sample precipitation, which was apparent both by gross visual inspection (Fig 1B) and by SDS-PAGE showing an inability of apolipoproteins to enter the stacking gel (Fig 1C). This effect was not observed in parallel labeling reactions of LDL, suggesting that the apo(a) component of Lp(a) sensitized samples to precipitation. In an attempt to reduce the hydrophobicity of our labeling approach, we next tested the fluorophore pHrodo iFL Red, which utilizes an STP-ester amine-labeling strategy with increased water solubility⁷⁷. This approach circumvented lipoprotein precipitation, as labeled samples exhibited the same solubility and electrophoretic mobility as unlabeled samples (Fig 1D). After pHrodo iFL Red labeling and extensive dialysis to remove free dye, we recovered >90% of the original Lp(a) preparation with no disruption of sample purity or disulfide linkage between apo(a) and apolipoprotein B, as reflected by a shift in protein electrophoretic mobility of the dominant band under nonreducing conditions (Fig 1E). Incorporation of fluorophore into Lp(a) was confirmed by in-gel fluorescent scanning (Fig 1F) and by flow cytometry revealing a highly sensitive, time-dependent, and dose-dependent uptake of labeled Lp(a) by HuH7 cells (Fig 1G-H). Consistent with receptor-mediated endocytosis, coincubation with a molar excess of unlabeled Lp(a) led to a dose-dependent competitive inhibition of fluorescent Lp(a)uptake, albeit with incomplete inhibition suggestive of nonsaturating Lp(a) concentrations (Fig 1I). Together, these findings support the suitability of our fluorescent Lp(a) labeling and cellular uptake strategy for high-throughput genetic screening.

Genome-scale CRISPR screen for Lp(a) uptake regulators.

Informed by the optimization experiments described above, we next scaled our cellular Lp(a) uptake assay to query the same GeCKOv2 genome-wide CRISPR knockout library⁵⁹ we tested in our prior screen of LDL uptake⁵⁶ (Figs 2A, S2). For each of 3 independent biologic replicates, we transduced ~60 million HuH7 cells at multiplicity of infection (MOI) ~0.5, passaged cells for 14 days to allow for target site editing and turnover of residual protein, incubated pools of edited cells with fluorescent Lp(a), sorted 10% subpopulations of cells with the least and greatest Lp(a) uptake, and quantified gRNA abundance in each sample by massively parallel sequencing.

Figure 2. A genome-scale CRISPR screen for modifiers of HuH7 cellular Lp(a) uptake.

(A) Schematic overview of screening strategy, with pools of HuH7 cells transduced with the GeCKOv2 genome-wide CRISPR knockout library at low MOI, passaged for 14 days, incubated with fluorescent Lp(a) for 2 hrs, and sorted into Lp(a)^low and Lp(a)^high subpopulations for extraction of genomic DNA and quantification of individual gRNA abundance by next-generation sequencing (NGS). (B) Volcano plot of aggregate gene-level Robust Rank Aggregation (RRA) scores (y-axis) relative to log2 fold change of normalized gRNA counts in Lp(a)^low relative to Lp(a)^high subpopulations. Genes whose disruption was associated with a significant (FDR<5%) decrease or increase in Lp(a) uptake are labeled and highlighted in blue and red, respectively. (C) Analysis of genes previously identified in our analogous prior screen of cellular LDL uptake, with the aggregate log2 fold change for gRNAs targeting each gene in LDL^low relative to LDL^high cells (x-axis) in that screen plotted relative to gRNA log2 fold change in Lp(a)^low relative to Lp(a)^high cells in this study. (D) Mean and 95% confidence intervals of aggregate gRNA log2 fold enrichment in Lp(a)^low relative to Lp(a)^high cells across 3 independent biologic replicates for genes encoding receptors previously proposed as mediators of cellular Lp(a) uptake. RRA scores, aggregate log2 fold change, and FDR values were calculated by MAGeCK; 95% confidence intervals of independent replicates were calculated using GraphPad Prism. Source data are provided in Supplemental Tables 1 and 2.

Analysis of gRNA enrichment across screen replicates revealed that the most significant decrease in Lp(a) uptake was conferred by disruption of LDLR while the most significant increase was conferred by disruption of the LDLR negative regulator MYLIP (Fig 2B, Supplemental Tables 1-2, Fig S3). Using a stringent cutoff for screen analysis (false discovery rate, or FDR < 5%), we identified no other genes with a significant role in Lp(a) uptake. Using a more liberal significance threshold (FDR < 20%), we identified another 8 genes whose disruption was associated with decreased Lp(a) uptake. This group included the canonical LDLR regulators SCAP and MBTPS2 and was significantly enriched for gene-gene interactions and functional annotations related to cholesterol metabolism (Fig S4). Among genes not classically associated with LDLR regulation but identified in our prior analogous screen of LDL uptake, we observed a significant correlation between each gene’s influence on LDL and Lp(a) uptake (Fig 2C, r = 0.63, p < 0.001). In a focused analysis of specific receptors previously proposed to mediate Lp(a) uptake, we did not identify a significant effect in our screen for any aside from LDLR (Fig 2D).

LDLR expression modulates HuH7 Lp(a) uptake.

Prior studies by us and others have established that LDLR is expressed and localized to the cell surface of HuH7 cells^56-58,78-80. To validate the functional influence of LDLR on Lp(a) uptake by HuH7 cells, we next generated and analyzed HuH7 cells with intact or genetically disrupted endogenous LDLR in the presence or absence of heterologous LDLR cDNA expressed from a lentiviral construct (Fig 3A). Because of the potential for contaminating LDL in Lp(a) preparations to confer a false positive dependence on LDLR, in our validation experiments we measured Lp(a) uptake with a secondary method of detection that was specific for Lp(a) (an ELISA for the apo(a) component of Lp(a) that is absent in LDL particles). To exclude the possibility of the fluorescent label mediating the interaction between Lp(a) and LDLR, we performed validation testing with native, unlabeled Lp(a) preparations. Consistent with our screen findings, we found that disruption of endogenous LDLR abrogated cellular Lp(a) uptake; this phenotype was not due to a CRISPR off-target effect since it was rescued by heterologous expression of LDLR cDNA (Fig 3B). Further supporting the role of LDLR in Lp(a) uptake, we also found that overexpression of LDLR on a wild-type background significantly augmented cellular uptake of Lp(a) (Fig 3B), while transduction with empty lentiviral vector alone had no significant effect (Fig S5). Additionally, we found that coincubation with purified LDL competitively inhibited the uptake of fluorescent Lp(a) by HuH7 cells (Fig S6). Together, these findings validate the functional influence of LDLR on HuH7 cellular Lp(a) uptake.

Figure 3. Analysis of Lp(a) uptake in LDLR-disrupted and LDLR-overexpressing HuH7 cells.

(A) Immunoblot analysis of LDLR protein abundance in wild-type HuH7 cells and a clonal HuH7 cell line harboring a homozygous frameshift-causing insertion in exon 6 of LDLR, each with and without transduction of a lentiviral expression construct of a LDLR cDNA. (B) Quantification of cellular apolipoprotein(a) internalization for the cell lines indicated in (A) after incubation with 50 μg/mL Lp(a) for 3 hrs. Individual data points represent the mean of technical duplicates for each independent biologic replicate. Asterisks indicates p<0.05 by Student’s t-test and error bars depict standard deviation.

Lp(a) directly binds to the LDL receptor extracellular domain.

The functional influence of LDLR on HuH7 Lp(a) uptake might have been mediated by direct binding to Lp(a) or by an indirect effect of LDLR perturbation such as an alteration of cellular metabolism or the expression of other genes. To evaluate for the potential of Lp(a) to directly bind the LDL receptor, we used bio-layer interferometry, a label-free method in which binding to a ligand is detected by a change in the refractive index for white light⁸¹. We immobilized biotinylated recombinant LDLR extracellular domain to streptavidin-coated sensors and incubated with a range of concentrations of Lp(a) or the positive and negative controls LDL and HDL, respectively. As expected, we observed a concentration-dependent shift in refracted light for LDL with LDLR-immobilized sensors but not with control streptavidin-coated sensors alone, while parallel reactions with HDL revealed no evidence of LDLR binding (Fig 4A, C). Lp(a) similarly exhibited a concentration-dependent and LDLR-dependent shift in refracted light, albeit of lesser magnitude than that observed for LDL (Fig 4B, C). The amount of Lp(a) binding to LDLR was nonsaturating at the concentrations tested, similar to the nonsaturating binding of Lp(a) to hepatoma cells in a prior study⁸² and of the nonsaturating fluorescent Lp(a) uptake by HuH7 cells in our study (Fig 1I).

Figure 4. In vitro binding between Lp(a) and the LDLR extracellular domain.

(A) Biolayer interferometry measurements for LDLR binding with LDL at the indicated concentrations. (B) Biolayer interferometry measurements for LDLR binding with Lp(a) at the indicated concentrations. (C) Comparison of biolayer interferometry measurements for LDLR binding to LDL, Lp(a), and HDL each at 100 nM. For all panels, solid lines indicate mean values and shaded regions indicate 1 standard deviation above and below the mean. Sensors were incubated with lipoprotein suspensions at t = 0 seconds and transferred to buffer only at t = 60 seconds, indicated by the dashed line.

Humans with pathogenic LDLR alleles have elevated levels of circulating Lp(a).

To examine the physiologic relevance of our in vitro findings, we analyzed Lp(a) levels and their relationship to LDLR genotype among individuals in the UK Biobank cohort⁶⁶. We identified 359,090 individuals with whole exome sequencing and serum Lp(a) levels, including 225 carriers of 25 different LDLR alleles with the highest level of evidence for pathogenicity as determined by an expert panel⁷⁴. Each of these 225 individuals were heterozygous for the pathogenic allele. We adjusted lipoprotein levels for the use of lipid-lowering medications and controlled for age, sex, and ancestry (Supplemental Table 3). As expected, carriers of LDLR pathogenic alleles were found to have LDL cholesterol levels that were significantly greater than noncarriers (32.3% increase, p = 4.1 x 10⁻¹¹⁶, Fig 5A and Supplemental Table 4). Likewise, pathogenic LDLR allele carriers exhibited significantly higher Lp(a) levels than noncarriers, though with a smaller magnitude of increase (9.2% increase, p = 6.9 x 10⁻³, Fig 5B and Supplemental Table 4). This relative increase in Lp(a) levels was statistically significant and of a comparable effect size whether analyzing raw or adjusted Lp(a) levels, untransformed or rank-based inverse-normalized values, all ancestries in the cohort or Europeans only, or with or without adjustment for genotype at the rs10455872 SNP that tags a narrower range of LPA KIV2 copy number variants and represents one of the top associations with Lp(a) levels in multiple genome-wide association studies^10,46-48 (Fig S7 and Supplemental Table 4). We also performed a sensitivity analysis using different magnitudes of Lp(a) adjustment for statin usage; the association between Lp(a) levels and LDLR genotype remained statistically significant after conservatively adjusting for statin effects based on the boundaries of the 95% confidence interval for their effect size in a recent meta-analysis of 5256 individuals¹⁹ (Fig S8 and Supplemental Table 4).

Figure 5. Analysis of Lp(a) levels for carriers of LDLR pathogenic alleles in the UK Biobank cohort.

(A-B) Adjusted LDL and Lp(a) levels aggregated for 225 carriers and 358,405 noncarriers of 25 expert-curated pathogenic LDLR alleles in UK Biobank. Statistical analysis for (A) and (B) was performed with the burden testing framework in Regenie. (C) Correlation between LDL and Lp(a) beta coefficients for all single LDLR variants with the indicated annotations and a minor allele count of at least 25 in UK Biobank. Individual data points represent single variants and are colored by functional annotation. Regression analysis was performed with GraphPad Prism. (D) Lp(a) beta coefficients for single variants represented by at least 25 individuals in UK Biobank, grouped by ClinVar annotation with statistical analysis between groups performed by Student’s t-test assuming equal variance. Raw data is provided in Supplemental Tables 4 and 5. For all panels, horizontal lines represent mean values and error bars indicate standard error of the mean.

We also analyzed the relationship of Lp(a) levels to a broader set of LDLR variants with different functional annotations (Supplemental Table 5). To avoid increased variability in small sample sizes, we focused our analysis on LDLR alleles represented by at least 25 carriers. We found that relative to alleles annotated as “Benign” or “Likely Benign”, those annotated as “Pathogenic” or “Likely Pathogenic” were associated with significantly increased Lp(a) levels (Fig 5C, p = 0<0.001). Among these single variants, there was a significant correlation (r = 0.43, p <0.001) between the change in associated LDL and Lp(a) levels (Fig 5D). There was no significant increase in Lp(a) levels for carriers of LDLR variants lacking a functional annotation or annotated with “uncertain significance” or “conflicting interpretations of pathogenicity” (Fig 5C).

DISCUSSION

A role for the LDL receptor in Lp(a) clearance has long been suspected due to the structural similarities between Lp(a) and LDL particles. Lp(a) is comprised of an LDL-like moiety which includes the LDLR ligand apolipoprotein B covalently bound to a single apo(a) molecule via disulfide linkage. Indeed, multiple studies in cells have provided support for an interaction between Lp(a) and LDLR^27-30. However, other studies have found no influence of LDLR on cellular Lp(a) binding or uptake in vitro^31-34. Likewise, while one study in humans found a defect in Lp(a) catabolism for patients with Familial Hypercholesterolemia (characterized by a defect in LDLR-mediated LDL uptake)²⁸, other studies did not^37-39. Although mice do not produce endogenous Lp(a), studies of their catabolism of exogenous human Lp(a) have also yielded mixed results, with some demonstrating LDLR-dependent^35,36 and some LDLR-independent^25,32 clearance. The basis for these inconsistent findings, both in relation to each other and to our findings, remains unclear but may be related to differences in experimental design, including methods of Lp(a) purification and measurement. In this context, we now report the first forward genetic screen in human cells for cellular regulators of Lp(a) uptake, controlling for technical differences by functionally testing all potential Lp(a) receptors simultaneously. Overall, our screen results indicate a primary role for the LDL receptor in mediating cellular Lp(a) uptake under these conditions, with LDLR and its negative regulator MYLIP representing the top hits influencing Lp(a) uptake and no other potential receptor demonstrating a significant effect.

In addition to the functional data supporting a role for LDLR in Lp(a) uptake, we also found evidence of direct binding between Lp(a) and the LDLR extracellular domain. The amount of Lp(a) binding at the time points and concentrations we tested, however, was qualitatively less than we observed for LDL binding and did not saturate. Other studies have also demonstrated reduced Lp(a) cellular binding relative to LDL for primary human hepatocytes, HepG2, and Hep3B cells^82,83, with the caveat that each study varied in its particular experimental conditions. We were unable to calculate an exact binding affinity for the interaction between Lp(a) and LDLR, a known limitation of BLI for moderate to weak affinity interactions and analyte concentrations below the Kd⁸⁴. Although our findings indicate direct binding between Lp(a) and LDLR in vitro, we cannot conclude whether this interaction is sufficient for physiologically relevant binding in vivo or if cooperative interactions with other molecules at the cell surface might have a role in increasing the affinity of Lp(a)-LDLR binding. It is also possible that the increased affinity of LDLR for LDL relative to Lp(a) may result in the former outcompeting the latter when the number of available receptors is limited relative to the abundance of both circulating lipoproteins.

Prior human genetic studies on the association between LDLR alleles and Lp(a) levels have also produced conflicting results. Some studies of individuals with Familial Hypercholesterolemia (FH) have reported an increase in Lp(a) levels^40-42, but others have not^43-45 and suggested that the former may have been confounded by diagnostic ascertainment bias and/or unequal distributions of LPA risk alleles between FH and non-FH populations^44,45. Multiple genome-wide association studies (GWAS) did not find an association between Lp(a) levels and common variants at the LDLR locus^10,46,47, but a recent GWAS of 371,212 individuals in the UK Biobank did⁴⁸. Analyzing the same cohort, we now report an association between rare pathogenic LDLR alleles and increased Lp(a) levels, both by an aggregate analysis of 25 expert-curated pathogenic alleles with the highest level of evidence and by single variant analysis of a broader set of LDLR alleles. We analyzed for potential confounding variables that might drive this association. We found the association between LDLR genotype and Lp(a) levels remained significant after accounting for the rs10455872 genotype that tags a narrower range of LPA alleles¹⁰, but we could not directly assess LPA KIV2 copy number variants in the UKBB cohort to assess for a difference between groups. Statin usage has been associated with increased circulating Lp(a) levels and we did find increased rates of this medication among pathogenic LDLR allele carriers. However, the association between these alleles and Lp(a) remained significant after adjusting Lp(a) levels for statin usage across the entire 95% confidence interval for this effect¹⁹. Intriguingly, another study reported that statin usage may decrease rather than increase Lp(a) levels among FH patients⁸⁵. If true, this effect could mask a more pronounced Lp(a) elevation among pathogenic LDLR allele carriers than we detected in our study.

While we did find a significant association between LDLR genotypes and Lp(a) levels, the magnitude of this effect was notably smaller than that observed for LDL levels, with the former increased by ~9% and the latter ~32% among pathogenic LDLR allele carriers. This finding is consistent with our lipoprotein uptake studies that revealed a less pronounced defect in LDLR-disrupted cells for Lp(a) uptake (~61% reduction) compared to LDL uptake (~87% reduction⁵⁶), and with our in vitro binding studies showing a reduced affinity for LDLR binding to Lp(a) relative to LDL. It is also possible that the aggregate effect of these pathogenic LDLR alleles on Lp(a) may be blunted by the inclusion of individual variants with discordant effects on LDL and Lp(a) uptake. Further investigations will be necessary to establish the structural basis of the Lp(a)-LDLR interaction and its sensitivity to LDLR variants and different Lp(a) components including apo(a).

Modulation of LDLR activity has been a highly successful strategy for LDL-lowering and the prevention and treatment of atherosclerotic cardiovascular disease. Paradoxically, while PCSK9 inhibitors (which upregulate LDLR) have been found to significantly lower Lp(a), statins (which also upregulate LDLR by a different mechanism) do not⁸⁶. A variety of models have been proposed to reconcile these findings, including the possibilities that Lp(a)-lowering by PCSK9 inhibitors may be mediated by receptors other than LDLR⁸⁷, by an effect on Lp(a) synthesis rather than clearance^34,88, or by a dependence of Lp(a) clearance on the expression level of LDLR and the relative concentrations of LDL and Lp(a)^89,90. Intriguingly, a recent study also found that statin treatment caused a significant increase in LPA expression in HepG2 cells¹⁹. Although a quantitative proteomic assessment of HepG2 cells did not find detectable apo(a) expression in this cell line⁹¹, if it is true that statin levels increase LPA mRNA expression in vivo, it is possible that this concurrent Lp(a)-increasing effect of statins may offset and mask any Lp(a)-lowering resulting from increased LDLR-mediated clearance.

There are intrinsic limitations to our CRISPR screening approach that warrant consideration. First, a gene knockout screen such as ours may not detect an effect for essential genes, as cells harboring gRNAs targeting these genes are expected to become progressively depleted in pooled cultures prior to selection. Second, although HuH7 cells have been shown to serve as excellent models for lipoprotein uptake by hepatocytes⁹², the expression of any single gene may vary from its in vivo state in hepatocytes, potentially leading to a false negative result if a putative Lp(a) receptor is poorly expressed. Third, although the GECKOv2 library is optimized for activity and tests multiple gRNAs per target gene, stochastic variation in gRNA efficiency may lead to inefficient editing for a given target that could similarly cause a false negative result. Fourth, pooled CRISPR screens are only able to identify gene perturbations that lead to a phenotype through a cell autonomous mechanism. For example, individual cells with disrupted PCSK9 will remain sensitive to the extracellular PCSK9 secreted by neighboring cells, precluding their development of a LDLR-mediated phenotype in this study and in prior pooled CRISPR screens for LDL uptake or LDLR abundance^56,80,93. Fifth, the lack of a statistically significant effect for any particular gene disruption is not definitive evidence for it playing no role in Lp(a) uptake, as false negatives may arise either due to compensatory changes in the cells or due to limited power to detect small effect sizes. Finally, loss-of-function CRISPR screens have limited ability to detect functionally redundant genes. For example, if a Lp(a) receptor were encoded by two different paralogues, then the intact function of one gene may compensate for loss of the other. These caveats notwithstanding, our screen findings together with our subsequent validation testing, in vitro binding studies, and human genetic analysis all lend support to the central importance of LDLR in mediating Lp(a) uptake by hepatocytes.

Supplementary Material

1

Figure S1. Detection of cellular Lp(a) uptake by immunofluorescence. Fluorescence microscopy of HuH7 cells incubated with and without exogenous Lp(a), fixed, and immunostained for apo(a) or secondary antibody only.

2

Figure S2. Detection of cellular uptake of labeled Lp(a) by flow cytometry. (A) Representative gating strategy for detection of HuH7 cells incubated with pHrodo iFL Red-labeled Lp(a). (B) Representative cell sorting strategy for the subpopulations of HuH7 cells with the greatest and least uptake of labeled Lp(a). Gray and red histograms represent cells incubated in the absence or presence of labeled Lp(a), respectively.

3

Figure S3. CRISPR screen quality control. (A) Cumulative distribution functions of normalized read counts for each gRNA in Lp(a)^high and Lp(a)^low populations for each of 3 independent biologic replicates. (B-C) Q-Q plots of observed versus expected −log(p-value) for gene perturbations associated with decreased (B) or increased (C) Lp(a) uptake. Observed p-values were calculated by MAGeCK (Supplemental Table 1).

4

Figure S4. CRISPR screen analysis. (A) Network analysis of Lp(a) uptake modifiers identified in the genome-wide CRISPR screen with FDR < 20%. The borders of individual nodes are weighted by each gene’s −log(RRA score), and the lines connecting nodes are weighted by the strength of the protein–protein interaction within the STRING database. Genes whose disruption was associated with a decrease or increase in Lp(a) uptake are shaded in blue or red, respectively. The significance of the number of detected interactions relative to a randomly selected gene set was calculated by STRING. (B) Gene Ontology annotations for biologic processes enriched among the 10 genes identified in the Lp(a) uptake screen with FDR < 20% relative to the human genome. When multiple annotations within the same hierarchy were identified, the annotation with the most significant enrichment was selected for display. Calculation of p-values was performed by Fisher’s exact test using PANTHER.

5

Figure S5. Analysis of Lp(a) uptake in HuH7 cells transduced with empty lentiviral backbone. For each of 2 independent biologic replicates, wild-type HuH7 cells were compared to HuH7 cells transduced with lentivirus from the same backbone as for LDLR heterologous expression in Fig 3, but lacking the cDNA insert. Exogenous Lp(a) was applied and internalization of apo(a) quantified by ELISA as in Fig 3.

6

Figure S6. Competitive inhibition of HuH7 cellular Lp(a) uptake by coincubation with LDL. Relative uptake of fluorescent Lp(a) by HuH7 cells in the absence or presence of a 20-fold molar excess of unlabeled Lp(a) or LDL.

7

Figure S7. UK Biobank analysis of Lp(a) levels by LDLR genotype. Comparison of adjusted Lp(a) levels between carriers and noncarriers of 25 pathogenic LDLR alleles, as performed in Fig 5 with the following variations. (A) Comparison of raw Lp(a) levels. (B) Comparison of rankbased inverse-normalized adjusted Lp(a) levels. (C) Comparison of adjusted Lp(a) levels for the subset of individuals of European ancestry. (D) Comparison of adjusted Lp(a) levels with inclusion of rs10455872 genotype as a covariate.

8

Figure S8. Sensitivity analysis of different Lp(a) adjustments for statin usage. Comparison of adjusted Lp(a) levels between carriers and noncarriers of 25 pathogenic LDLR alleles, as performed in Figs 5 and S6, with varying magnitudes of adjustment of Lp(a) levels for individuals taking statin medications. Analyses were derived from the range of effects seen in a recent meta-analysis¹⁹, including the lower (A) and upper (B) limits of the 95% confidence interval in comparison to the mean (C) for the observed effect size. Raw data is provided in Supplemental Table 4. For all panels, horizontal lines represent mean values and error bars indicate standard error of the mean.

9

Supplemental Table 1. Gene-level CRISPR screen results for modifiers of HuH7 cellular Lp(a) uptake. MAGeCK analysis of aggregate gene-level gRNA-level enrichment or depletion in Lp(a)^high relative to Lp(a)^low cells. Positive log2 fold change and RRA scores correspond to a gene’s disruption decreasing Lp(a) uptake; negative values correspond to a gene’s disruption increasing Lp(a) uptake.

10

Supplemental Table 2. Individual gRNA-level CRISPR screen results for modifiers of HuH7 cellular Lp(a) uptake. MAGeCK output for individual gRNA-level enrichment or depletion in Lp(a)^high relative to Lp(a)^low cells. Positive log2 fold change and RRA scores correspond to a gene’s disruption decreasing Lp(a) uptake; negative values correspond to a gene’s disruption increasing Lp(a) uptake.

11

Supplemental Table 3. Study characteristics of UK Biobank cohort. Demographic data, lipoprotein levels, and medication usage by individuals in the UK Biobank analyzed in this study.

12

Supplemental Table 4. Aggregate analysis of LDL and Lp(a) levels for carriers of 25 different pathogenic LDLR alleles. LDL and Lp(a) analysis for carriers of 25 expert-curated pathogenic LDLR alleles, individually and in aggregate. Different tabs include data for raw lipoprotein levels, adjusted lipoprotein levels including different magnitudes of Lp(a) adjustment for statin usage, rank based-inverse normalized values, and adjusted lipoprotein levels accounting for rs10455872 genotype as a covariate.

13

Supplemental Table 5. Single variant analysis of LDL and Lp(a) levels according to functional annotation. LDL and Lp(a) analysis for carriers of LDLR variants of the indicated functional annotations and allele frequencies in the UK Biobank cohort.

Figure 6. Graphical abstract.

Conceptual overview of the approaches used in this study.

Highlights.

• The top hits from a genome-wide CRISPR screen for modifiers of HuH7 cellular Lp(a) uptake were LDLR and its negative regulator MYLIP.

• Purified Lp(a) directly bound to the LDL receptor extracellular domain in vitro.

• Individuals in the UK Biobank with loss-of-function mutations in LDLR exhibited elevated levels of circulating Lp(a).