The COPI coatomer regulates several steps of HDL metabolism

Abstract

Background:

Reverse cholesterol transport by high density lipoproteins (HDL) is considered as an anti-atherogenic metabolic pathway. Hepatocytes are the main contributors to the efficacy of this pathway by the production of apolipoprotein A-I (apoA-I) and its lipidation by ATP binding cassette transporter A1 (ABCA1), selective uptake of cholesterol via scavenger receptor BI (SR-BI) and uptake of entire HDL particles. The molecular determinants of the latter step are not well understood.

Methods:

We performed a genome-wide RNA interference screen for genes limiting the uptake of HDL fluorescently labeled at its protein moiety into Huh-7 hepatocarcinoma cells. Top hit genes were validated by targeted in vitro experiments and the analysis of associations between their variants and HDL cholesterol (HDL-C) levels in the databases of the Global Lipids Genetics Consortium and UK Biobank as well as inborn errors of metabolism and their respective mouse models.

Results:

The knockdown of 128 genes significantly inhibited HDL uptake. Six of them encode for components of the COPI coatomer, namely COPA, COPB1, COPB2, COPG1, ARCN1, and COPZ1. Knocking down any of them decreased the uptake of both fluorescently labeled proteins and lipids of HDL, the cell surface abundance of SR-BI, as well as APOA1 expression and apoA-I secretion, but increased the cell surface abundance of ABCA1. Common single nucleotide polymorphisms of ARCN1 and COPB1 were associated with significantly higher HDL-C levels in the population while rare COPA and COPG1 variants causing immunopathies were associated with rather lower levels of HDL-C in both affected patients and the corresponding genetically modified mice.

Conclusions:

In hepatocytes, the COPI coatomer regulates HDL holoparticle uptake, selective lipid uptake, apoA-I secretion, and cholesterol efflux, and thereby, it influences plasma levels of HDL-C.

Graphical Abstract

Introduction

Low plasma levels of HDL-C are associated with increased risk of atherosclerotic cardiovascular disease (ASCVD), but unsuccessful drug developments and negative findings of Mendelian Randomization studies have led to doubts about the causal role of HDL-cholesterol in ASCVD¹. Nevertheless, the mediation of reverse cholesterol transport by HDL is still considered as an important anti-atherogenic metabolic pathway. The first steps in this process are well understood, namely the secretion of apoA-I containing HDL precursors, their lipidation by the ATP binding cassette transporter ABCA1 and the subsequent esterification of cholesterol by lecithin:cholesterol acyltransferase (LCAT), as well as the HDL-mediated cholesterol efflux from macrophage foam cells¹. However, the mechanism of the final steps of reverse cholesterol transport, namely the hepatic uptake of HDL, is only partially resolved. Cholesteryl esters of HDL are indirectly delivered to the liver by cholesteryl ester transfer protein (CETP)-mediated transfer to apoB-containing lipoproteins that are ultimately taken up by the low density lipoprotein (LDL) receptor (LDLR)^1,2. The direct delivery involves both selective uptake of lipids independently of HDL’s protein moiety by SR-BI, as well as HDL-holoparticle uptake by a yet little understood mechanism². The latter was suggested to involve high affinity binding of apoA-I, i.e. the main protein constituent of HDL, by ectopic beta-ATPase and thereby generation of ADP, which in turn stimulates the purinergic P2Y13 receptor to signal to a yet unidentified lower affinity HDL binding site, distinct from SR-BI, that mediates the endocytosis of HDL particles^2,3.

To identify genes that limit the endocytosis of HDL holoparticles by the liver, we performed a genome wide siRNA screening in Huh-7 hepatocarcinoma cells. The uptake of HDL fluorescently labeled at its protein moiety was most significantly inhibited by interference with the COPA, COPB1, COPB2, COPG1, ARCN1, and COPZ1 genes which encode for indispensable subunits of the COPI coatomer, namely α-COP, β-COP, β’-COP, γ-COP, δ-COP and ζ-COP, respectively⁴. β-COP, δ-COP, γ-COP and ζ-COP form the F- subcomplex, while α-COP, β’-COP, and the thermosensitive and accessory ε-COP form the B-subcomplex (figure 1A). Together the B- and F-subcomplexes encase vesicles mediating the transport of proteins either from the Golgi apparatus to the endoplasmic reticulum or between the Golgi cisternae, and thereby play an important role in the processing and quality control of proteins⁴. In addition, COPI contributes to the maturation of early endosomes, to lysosomal trafficking and autophagy^5,6. Rare variants of COPA, COPG1 COPZ1, or ARCN1 cause inherited syndromes characterized by autoinflammation, immunodeficiency, neutropenia, and disturbed craniofacial development, respectively^7–11.

Figure 1. Silencing of COPI genes and its effect on the uptake of HDL into Huh-7 hepatocarcinoma cells.

A Illustration depicting the formation of COPI complex as a heterodimer of sub complexes B (comprised of α-, β´- and ε-COP) and F (comprised of β-, γ-, δ-, and ζ-COP). γ2- and ζ2-COP, paralogues of γ- and ζ-COP respectively, are also shown. B-F. Huh-7 cells transfected with the indicated siRNAs were collected 72 hours post transfection. Lysates were used for Western Blotting of γ-COP (C), δ-COP (D), α-COP (E) or ε-COP (F), with TATA binding protein (TBP) (C, D) or β-actin (E, F) serving as the loading control, respectively. The Western Blots are representatives of three (C) or four (D-F) independent experiments. Bold arrows indicate the COPI protein directly targeted by the siRNA while hatched arrows indicate COPI proteins residing in the same F-subcomplex (C, D) or B-subcomplex (E, F) of the coatomer. B shows the quantification of knockdown efficacy by quantification of the targeted protein relative to actin or TBP in the western blots by densitometry. For quantification of all conditions shown in each Western Blotting experiment, please see supplementary figure S2B. G-J. 72 hours after transfection with the indicated siRNAs, Huh-7 cells were incubated with 20 μg/mL Atto655-HDL (G, H) or Dil-HDL (I, J) for 3 hours in the presence or absence of 100-fold excess unlabeled HDL and were then collected for measurement with flow cytometry. The specific cell association was calculated as the difference between the two conditions. The data were normalized to the non-coding control (NC). SCARBI (I) was included as the positive control. G and I show means ± SD of 3 independent experiments for each condition separately. Statistical analysis was performed using Mann-Whitney test between the NC and each targeting siRNA without any correction for multiple testing (B) or Kruskal-Wallis test coupled with Dunn’s test for multiple comparisons between NC and each targeting siRNA (G, I). H and J compare the pooled data on the non-dispensable COPI subunits (red bars, N = 18) with the pooled data on the dispensable or paralogous COPI genes (blue bars, N = 9) by using two-tailed Mann-Whitney test. Except in figure 1B, only statistically significant differences (p<0.05) are indicated.

Since the role of the COPI coatomer in lipoprotein metabolism is little understood, we investigated the cellular mechanisms underlying disturbed HDL uptake upon loss of COPI function, as well as the impact of variants in COPI genes on plasma HDL-C concentrations in humans and mice.

Methods

Data Availability Statement

The authors declare that all supporting data are available within the article and its online supplementary files.

Cell culture

Huh-7 cells (cat. JCRB0403, JCRB Cell Bank, Ibaraki, Japan) were cultured in normal growth medium, comprised of Dulbecco’s Modified Eagle’s Medium (DMEM) (cat. D-5796, Sigma-Aldrich, St. Louis, USA) supplemented with 10% fetal bovine serum (FBS), (cat. 10500056, Thermo Fisher Scientific, Waltham, Switzerland) and 100 U/mL each penicillin/streptomycin (P/S) (cat. 15140122, Thermo Fisher Scientific).

Isolation and labeling of HDL

HDL (1.063<d<1.21 kg/L) was isolated from human plasma by sequential ultracentrifugation and labeled with Atto647, Atto655, or DiI- as reported previously¹² and described in detail in the Expanded Material and Methods section of the Supplement.

siRNA genome-wide screening

The genome-wide siRNA screening for genes limiting HDL uptake into Huh-7 cells was performed together with our previous screening for genes limiting LDL uptake which targeted the 21,584 human genes by three unique non-overlapping siRNAs (Ambion Silencer Select Human Genome siRNA library V4, cat. 4397926, Thermo Fischer Scientific)¹³. As the internal controls, anti-PLK1^13,14 (assay ID s448, cat. 4390824, Thermo Fischer Scientific) and the Silencer^™ Select Negative Control No. 1 siRNA (cat. No. 4390843) were plated in four replicate wells in each of the 192 assay plates. Since no endocytic HDL receptor is known² and our explorative pilot experiments excluded candidate genes (e.g. SCARB1, PDZK1), we could not include any condition to control for specificity. To determine the overall signal-to-noise ratio of our assay, cells transfected with a non-targeting siRNA were incubated in the absence of fluorescent lipoproteins (background control). 72 hours after transfection cells were exposed to a mixture of Atto655-HDL and Atto594-LDL¹³ DMEM, at a final concentration of 33 μg/mL each. Procedures for washing, fixation, automated microscopy and image analysis were described previously¹³ and are summarized in the Expanded Material and Methods section of the Supplement. As for our previous LDL screen¹³, we applied the Redundant siRNA Activity (RSA) analysis to data from the best performing assay feature, namely median cytoplasm intensity for the identification of hit genes. Z’-factor values for median cytoplasm intensity in each assay plate for the background without fluorescent HDL (median 0.79, interquartile range [IQR] 0.68 to 0.90) indicated excellent signal-to-noise ratio.

siRNA transfection in validation experiments

Huh-7 cells were reverse transfected with either siRNAs against the indicated target genes (COPA, COPB1, COPB2, ARCN1, COPG1, COPG2, COPE, COPZ1, COPZ2, LDLR, ABCA1, SCARB1) or non-targeting siRNAs. Details are described in the Expanded Materials and Methods as well as the Major Resources Table. Knockdown efficiency was determined by qRT-PCR and Western Blotting combined with densitometry using ImageJ¹⁵ (RRID:SCR_003070) as described in the Expanded Material and Methods section of the Supplement.

Generation of Huh-7 cells overexpressing GFP-SR-BI

The plasmid containing the coding sequence of SCARB1 (NM_005505) carrying an N-terminal GFP tag was purchased from GeneCopoeia (cat. EX-G0782-M29, Rockville, USA). On Day 1, Huh-7 cells were seeded into 6-well plates using normal growth medium. On Day 2, the medium of the cells was changed to normal growth medium without antibiotics (DMEM +10%FBS, - P/S) and the cells were transfected with 3 μg of an empty plasmid (cat. EX-NEG-M02, GeneCopoeia) or the plasmid encoding for GFP-SR-BI using 6 μL of Lipofectamine 2,000 μL medium /well (cat. 11668019, Invitrogen, Waltham, USA) according to the manufacturer’s instructions. 6 hours post transfection, the medium was removed and fresh normal growth medium without antibiotics was added. 24 hours after transfection, the culture medium was replaced with selection medium [(DMEM +10% FBS +1% P/S) supplemented with 750 μg/mL G418 (cat. 10131–027, Gibco, Waltham, USA)]. Cells were selected over 4 passages and maintained under low confluence before being sorted using a BD FACS ARIA III (BD-Biosciences. Allschwil, Switzerland). Single sorted GFP-positive⁺ cells were then cultured in 96-well plates to obtain single-cell derived colonies. Cells transfected with the empty plasmid served as negative control (EV).

Flow cytometry

Flow cytometry was used to record the uptake of Atto655-HDL, or Atto647-HDL, or DiI-HDL as well as the cell surface expression of SR-BI and the overall emitted GFP signal by alive Huh7 cells principally as reported by our lab previously for the uptake of fluorescent LDL and the cell surface expression of LDLR, respectively¹³. Details are described in the Expanded Material and Methods section of the Supplement.

Cell surface biotinylation and enzymatic deglycosylation

The cell surface expression of endogenous SR-BI and ABCA1 was analysed by cell surface biotinlyation principally as reported previously¹² and described in detail in the Expanded Material and Methods section of the Supplement. To identify the mechanism underlying the misglycosylation of SR-BI upon loss of COPI genes 20 μg of Huh-7 cells pelleted and lysed in RIPA buffer were incubated with PNGase F, neuraminidase and/or O-glycosidase (New England Biolabs, Ipswich, USA cat. P0704, cat. P0720, and cat. P0733, respectively) for 9 hours at 37°C, according to the manufacturer’s instructions. Equal amounts from each sample were used for Western Blotting as described in the Expanded Material and Methods section of the Supplement.

Confocal Microscopy

Wild type Huh-7 cells or Huh-7 cells overexpressing or GFP-SR-BI were seeded in 24 well plates containing glass coverslips and transfected with the indicated siRNA. After 72 hours, the cells were fixed in PBS containing 4% paraformaldehyde (PFA) for 20 minutes at room temperature. The cells were then washed 3 times with PBS and incubated with PBS containing 0.1 % saponin (cat. 84510, Fluka, Buchs, Switzerland) for 10 minutes at room temperature. After three washes with PBS, the cells were incubated in PBS containing 5% normal donkey serum (cat. 566460, Sigma- Aldrich) and 1% BSA (cas 9048-46-8, Sigma- Aldrich) for 1 hour. The cells were then incubated with primary antibodies diluted in PBS containing 5% normal donkey serum and 1% BSA overnight at 4°C. After five washes with PBS, the cells were incubated with secondary antibodies. in PBS containing 5% normal donkey serum and 1% BSA for 45 minutes at room temperature. After five additional washes with PBS, the cells were mounted in Prolong^™ Gold Antifade reagent with DAPI (cat. P36935, Invitrogen) and were imaged using a SP8 confocal microscopy (Leica, Weltzar, Germany). For the lysotracker assay, the cells were incubated for 30 minutes with 10 nM Lysotracker^® Red DND-99 (cat. L7528, Molecular Probes, Oregon, USA) diluted in DMEM without FBS before being processed as above with the exception of no permeabilization. Sources and concentrations of both primary and secondary antibodies used are summarized in the Major Resources Table.

Cholesterol efflux and apoA-I secretion

The effect of COPI gene silencing on HDL- and apoA-I -induced cholesterol efflux from Huh-7 cells was recorded by a radiometric assay in the presence of an ACAT inhibitor and an activator of LXRα principally as published by Sun et al¹⁶ and described in detail in the Expanded Material and Methods section of the Supplement. ApoA-I secretion by Huh-7 cells transfected with siRNAs against COPI genes was quantified in the cell culture media using a commercially available ELISA kit (cat. 3710–1HP Mabtech Nacka Strand, Sweden), according to the manufacturers’s instructions. Sample preparation is described in the Expanded Material and Methods section of the Supplement

Quantification of cellular sterols

Forty-eight hours after transfection of Huh7 cells with siRNAs, the medium was aspirated and replaced with DMEM (cat. D-5796, Sigma-Aldrich) supplemented with 0.5% FBS and the indicated combinations of DMSO, 50 μg cholesterol/mL HDL, 10μM LXR agonist or 5 μg/mL ACAT inhibitor. After 24 hours, the cells were harvested for the measurement of sterols by either gas chromatography-mass spectrometry (GC-MS) or thin layer chromatography (TLC). For GC-MS, cholesterol, its precursors (lanosterol, dihydrolanosterol, desmosterol, and lathosterol) as well as phytosterols (campesterol, stigmasterol, and sitosterol) were extracted from dried cells with chloroform overnight and separated and measured after trimethylsilylation as described previously^17,18. The amount of each sterol in the cells was normalized to the actual mass of dry cell pellet. Procedures used for the quantification of unesterified and esterified cholesterol by TLC are described in the Expanded Material and Methods section of the Supplement,

Human genetic studies

The GWAS summary statistics for variants and phenotypes were downloaded from http://csg.sph.umich.edu/willer/public/glgc-lipids2021/results/trans_ancestry/. The data included meta-analysis of associations of 51,735,689 detected and imputed SNPs from approx. 1.65 million subjects from five different ancestry groups as described by the Global Lipids Genetics Consortium¹⁹.

For the comparison of associations with HDL-C versus apoA-I levels in UK Biobank, the GWAS data for variants and phenotypes were downloaded from https://www.nealelab.is/uk-biobank (Version 3). Associations were plotted as effect sizes and p-values of 816 SNPs in the nine COPI genes from 315,133 samples²⁰. In addition we explored the association of rare exome variants in the COPI genes with HDL-C and apoA-I levels in the most recent UK Biobank 500k WGS (v2) dataset which is made publicly in https://www.azphewas.com²¹. The association of rs11216909 with ARCN1 gene expression was analysed by using data of nearly 1000 individuals collected by the Genotype-Tissue Expression (GTEx) project²².

Five patients with the COPA syndrome due to heterozygosity for the rare mutations in the COPA gene presented in figure 6B as well as the patients with combined immunodeficiency due to homozygosity for the rare p.K652E mutation in the COPG1 gene and there heterozygous parents presented in figure 6C were previously described^8,9. Data on demographics and lipids were provided by Drs. Anthony Shum (San Francisco), Raif Geha and Janet Chou (both Children’s Hospital in Boston, USA). Plasma lipids were analyzed by the use of enzymatic photometric assays from Abbott (Architect Chemistry Analyzer and Sigma in San Francisco and Boston, respectively. Because of their different geographic and ethnic origin, we compared HDL-C levels in COPA and COPG1 variant carriers with those in different populations, namely NHANES from the United States²³ for the COPA variants and population studies in adults and children from the United Arab Emirates for the COPG1 variants^24,25.

Figure 6. Associations of SNPs or rare pathogenic variants of COPI genes with HDL-C and triglyceride levels or gene expression.

Volcano plots depicting the association of 5,821 SNPs of the nine COPI genes with HDL-C (A) and triglycerides (B) in 1.65 million individuals aggregated by the Global Lipid Genetics Consortium¹⁹. The x-axis depicts the effect size (beta) and the y-axis the p-value. The dashed horizontal line marks the threshold of statistical significance after Bonferroni correction for multiple testing on GWAS level (p = 9.60*10⁻¹⁰). Violin plots illustrate the associations of the rs11216909 genotypes with normalized ARCN1 gene expression in whole blood (from GTEx Analysis Release V10)²² C). The normalized effect size was −0.11, which was statistically significant between the genotypes p=9.35×10^−9. Tables D and E show HDL-C and triglyceride levels in individuals suffering from COPA syndrome⁸ and COPG1 syndrome⁹, respectively. For comparison with the US-American COPA variant carriers (D), data of 6127 male and 6157 female participants of NHANES²³ are presented (*) as means± standard error. For comparison of the heterozygous (†) and homozygous (††) carriers of the rare COPG1 variant (E) who originate from Oman, data of adults²⁴ (§) and children and adolescents²⁵ (#) from the United Arab Emirates are shown for comparison. Data are means ± SD of 485 men and 492 women (§) or 490 boys and 476 girls (#). HDL-C: HDL-Cholesterol; TG: Triglycerides; SD: Standard deviation.

Animal studies

Heterozygous Copa^wt/E241K and homozygous Copg1^K652E/K652E mice have C57BL/6J and C57BL/6N backgrounds, respectively, and been previously described^9,26. They were maintained in the specific pathogen free facilities at UCSF and Boston Children’s Hospital, respectively. Blood for plasma preparations were collected in the US laboratories and shipped to the laboratories in Zurich and Groningen for quantification of lipids, apolipoproteins and lipoproteins.

Concentrations of cholesterol, triglycerides and HDL-cholesterol were measured in individual mouse plasma samples using photometric assays from Roche diagnostics on the COBAS8000 autoanalyser (Rotkreuz, Switzerland). NonHDL-cholesterol was calculated as the difference between total and HDL-cholesterol.

For the profiling of plasma lipoproteins by fast protein liquid chromatography (FPLC) as described previously²⁷, pools of equal plasma volumes from 3 male or 3 female Copa^wt/E241K and 5 male or 5 female Copg1^K652E/K652E mice were passed through a SuperoseTM 6 Increase 10/300 GL column (GE Healthcare Hoevelaken, Netherlands) at a flow rate of 0.31 mL/minute for lipoprotein fraction separation. Chromatographic profiles of pooled wildtype (C57Bl/6) mice served as a reference standard. Data analysis was performed using GraphPad Prism 10.

Aliquots of the same plasma pools were used to quantify apoA-I concentrations by targeted proteomic assays as previously described²⁷. In short, target peptides were concentrated into synthetic proteins (or ‘QconCATs’; Polyquant GmbH), with ¹³C-labeled arginines and lysines. Concentrations of these isotopically labeled QconCATs served as standards to calculate the concentrations of endogenous equivalents of the peptides. In-gel digestion of one microliter plasma for the samples was performed as described previously²⁷. Isotopically-labeled standard peptides were added prior to the liquid chromatography-mass spectrometry (LC-MS) detection (1 ng digested standard peptides per 0.05 μL plasma digest).

Statistics

The RNAi screening assay feature data were analyzed as reported previously¹³. In brief, data were first normalized by the median value of each batch, microscope, plate and well and were finally expressed as robust Z-score²⁸ normalized values. The Redundant siRNA Activity (RSA) analysis was performed for each assay feature on the normalized data to rank the genes and detect the top hits defined here as the genes with an RSA p-value of less than 10⁻³. This p- value cutoff was dictated by our ability to verify the results in vitro. Transfection efficiency and the dynamic range of the screening assay were determined by calculating the Z’-factor between positive and negative transfection and assay controls as published before²⁸. Dimensionality reduction across the five main assay features mentioned above was performed using the Locally-Linear-Embedding²⁹ method on log₂-transformed data through the sklearn Python (RRID: SCR_024202) implementation of the method (http://scikit-learn.org/).

Top hits were clustered by function using the online String tool (https://string-db.org/), version 12.0 according to the following settings: the network edges indicate the type of interaction evidence, all active interaction sources (textmining, experiments, databases, co-expression, neighbourhood, gene fusion, co-occurrence) were selected for the analysis, with the minimum interaction score set to 0.400 (medium confidence) and with no network clustering used. Gene Ontology (GO) analysis for statistical overrepresentation of GO-Slim biological process terms was performed using the Panther Gene List Analysis tool version 19.0 (http://www.pantherdb.org; RRID:SCR_004869). Enrichment was calculated by Fisher’s exact test and corroborated by the calculation of the false discovery rate (FDR) according to the Benjamini-Hochberg procedure. GO terms with FDR <0.05 are reported as significant in this manuscript

All experimental data were obtained in an unblinded fashion, without prior power calculation and analyzed using GraphPad Prism version 10.6.0 (RRID:SCR_002798). Data in figure 4E, supplemental figure S5B, and the comparison of pooled HDL-C levels in Copa and Copg1 mutant mice were analyzed by two-way ANOVA after assessing residual diagnostics (residual distribution plots, homoscedasticity plots, and Spearman’s test for heteroscedasticity) and finding no significant evidence of heteroscedasticity. For figure 4E, a full interaction model was used; while for supplemental figure S5B and the pooled HDL-C comparisons, a main-effects-only/additive model was used. For figure 4E and supplemental figure S5B, the indicated comparisons were adjusted using Bonferroni’s multiple comparisons test, while for the pooled HDL-C comparisons, genotype and sex effects were reported directly from the 2-way ANOVA. For supplemental figure S5A residual diagnostics from ordinary 2-way ANOVA indicated unequal variances. Therefore, a two-factor additive linear model (main effects only) was applied, and inference was based on heteroscedasticity-consistent HC3 standard errors, using log2-transformed normalized values. Planned comparisons were adjusted using the Bonferroni method. This analysis related to supplemental figure S5A was performed using ChatGPT (GPT-5.5 thinking) and was independently verified by Gemini (version 3.1 Pro), while the authors reviewed the outputs in both cases. All other experimental data were analyzed with non-parametric tests, namely Mann-Whitney test for one to one comparisons or Kruskal-Wallis test coupled with Dunn’s test for multiple comparisons when comparing several conditions. For graphs in which all depicted conditions were compared to the respective control, only p values lower than the threshold limit mentioned in each legend are depicted. Otherwise, if different conditions were compared with each other, all p values are depicted regardless of statistical significance. All p values were rounded to the third significant decimal digit. The numbers of experiments, the used statistical tests, and if applicable adjustments for multiple testing, are described in the legends of the respective figures and tables.

Figure 4. Effects of silencing COPI genes or ABCA1 on cholesterol efflux (A-D) and HDL uptake (E, F) by Huh-7 cells.

A-D. [³H]-cholesterol efflux was measured 72 hours after transfection of Huh-7 cells with the indicated siRNAs by using lipid-free apoA-I (A, B) or HDL (C, D) as the acceptors. The data are shown as means±SD of 4 (A, B) or 3 (C, D) independent experiments. B and D compare the summarized data on the non-dispensable COPI subunits (red bars) with the data on the dispensable or paralogous COPI genes (blue bars). E and F. 48 hours post transfection and treatment with 10 μM of the LXR agonist T0901317 for 24 hours (E) or 72 hours post transfection (F) the cells were incubated with 20 μg/mL Atto647-HDL (E) or Atto655-HDL (F) for 3 hours. The specific cell association was determined as described in the legend of figure 1. The data are shown as means ± SD of 3 independent experiments. Statistical analysis was performed using either Kruskal-Wallis test coupled with Dunn’s multiple comparisons test between the NC and each targeting siRNA (A,C) or between the indicated conditions (F) or 2-way ANOVA (full interaction model) coupled with Bonferroni’s test for multiple comparisons (E). The pooled data of the non-dispensable COPI gene knock-downs (red bars, N = 12 in B, N = 9 in D) and the pooled data on the dispensable or paralogous COPI gene knock-downs (blue bars, N = 12 in B, N = 9 in D) were compared by using two-tailed Mann-Whitney test. In (A-D) only statistically significant findings with p<0.05 are indicated.

Ethics

The use of clinical data and samples from patients with the COPA syndrome or the COPG1 syndrome for this study was approved by the Institutional Review Boards (IRB) for the protection of human subjects of the University of California in San Francisco (UCSF, IRB protocol 10–02467) and Boston Children’s Hospital (IRB protocol 04-09-113R). All participants provided written informed consent.

The sampling of blood and livers from the mutant COPA and COPG1 mice were performed as approved by the Institutional Animal Care and Use Committees of UCSF in San Francisco (Mouse protocol number and approval number 202539) and Boston Children’s Hospital (Mouse protocol number and approval number 00001617), respectively.

Results

Genome wide RNAi screen identifies several COPI components as limiting factors of HDL endocytosis

We performed a genome-wide siRNA screen in Huh-7 hepatocarcinoma cells for genes that limit the uptake of HDL containing a fluorescently labeled protein moiety. Since no receptor is known to mediate the endocytosis of HDL holoparticle into hepatocytes² and our exploratory pilot experiments also excluded a limiting effect of RNA interference against SCARB1, which was shown to facilitate HDL holoparticle uptake by endothelial cells but not hepatocytes³⁰, we could not include any condition to control for specificity. As in a previous LDL screen¹³, we applied Redundant siRNA Activity (RSA) analysis to data from the best performing assay feature, namely median cytoplasm intensity, for the identification of hit genes. Z’-factor values¹³ for median cytoplasm intensity in each assay plate for the background without fluorescent HDL (median 0.79, interquartile range [IQR] 0.68 to 0.90) indicated excellent signal-to-noise ratio (supplemental figure S1A).

At the RSA p value cut off p < 0.001, silencing of 128 and 44 genes decreased and increased HDL uptake, respectively (table 1 and supplemental table S1). Functional clustering of these genes with the STRING tool revealed four major groups of genes whose loss compromised HDL uptake: the ribosome (N =47), the proteasome (N = 11), the spliceosome (N =20), and vesicle coat (N =8) (Supplementary Figure S1B). Of note, the latter contains CLTC and AP2M1 encoding the clathrin heavy chain and Adaptor Related Protein Complex 2 Subunit Mu 1, both of which are essential for clathrin-mediated endocytosis, as well as the COPI genes investigated in more detail by us. Also, the PANTHER Gene Ontology (GO) enrichment analysis showed significant clustering for genes whose loss of function decreased HDL uptake (Supplemental Table S2). The 44 genes whose knockdown increased HDL uptake did not cluster. (Supplementary Figure S1C).

Table 1.

Genes whose loss-of-function in Huh-7 cells cause either a decrease (left column) or an increase (right column) in HDL uptake.

Decreased HDL uptake				Increased HDL uptake
Gene	Assay scoreA avgB	Assay scoreA SDC	RSA p-value	Gene	Assay scoreA avgB	Assay scoreA SDC	RSA p-value
COPA D	−5.040723862	0.844316	1.53*10−10	PET117	2.942280383	1.725334	2.45*10−5
ARCN1	−4.799691625	0.44648	4.11*10−10	PROX1	2.93073874	1.946663	3.49*10−5
COPZ1	−4.527579252	0.295166	4.90*10−10	TGFBR1	2.94828644	0.941197	3.82*10−5
SF3B1	−4.857835852	1.071677	9.22*10−9	CDC37	2.233370577	2.244654	4.15*10−5
SF3A1	−3.559635516	0.088789	2.40*10−8	BRI3	2.001389493	0.213612	5.27*10−5
RPL7A	−3.664671783	0.760184	1.12*10−7	ARMT1	2.564839396	1.809901	5.43*10−5
CHMP2A	−4.067666726	0.817382	1.41*10−7	ZNF84	2.247295577	1.34654	9.52*10−5
COPB1	−3.653164204	0.523153	1.63*10−7	DSN1	2.451608999	3.407179	1.26*10−4
PSMD3	−3.759283965	0.97221	2.00*10−7	ESCO1	2.446763176	0.878105	1.49*10−4
AQR	−3.550087883	0.64399	2.44*10−7	ZNF138	1.899968121	0.332696	1.87*10−4
COPB2	−3.672670636	0.732791	6.00*10−7	SCARB1	2.172749211	1.126483	1.87*10−4
PSMD8	−3.355483695	0.519461	7.04*10−7	LONP1	1.781937945	0.197192	1.93*10−4
SF3B6	−2.974729818	0.115341	1.01*10−6	IDH3A	1.783052371	2.602182	2.12*10−4
RPL5	−3.485321168	0.80216	1.18*10−6	ZC3H4	2.462454841	3.569949	2.23*10−4
RPL7	−3.269064801	0.496465	1.44*10−6	JAG2	1.654988885	1.908713	2.57*10−4
RPL17	−3.545096747	0.716171	1.70*10−6	EGLN2	2.472715402	3.382913	2.68*10−4
RPS12	−2.873592995	0.157598	1.70*10−6	PCDH9	1.911677573	3.624453	3.12*10−4
RPLP2	−3.321066167	0.980756	1.79*10−6	KCNIP2	1.709715844	1.509489	3.92*10−4
RBM8A	−3.343511225	0.637388	1.85*10−6	FSTL5	0.473585051	4.630876	4.02*10−4
RPS2	−3.017633853	0.281376	2.17*10−6	TGIF1	1.569370878	0.122058	4.41*10−4
ISY1	−3.713267960	1.01113	2.27*10−6	ZNF544	2.352498311	3.088487	4.46*10−4
RPL27A	−2.842001078	0.156972	2.28*10−6	ZNF45	1.096681268	2.725219	4.62*10−4
COPG1	−4.360208077	1.504634	2.89*10−6	CCDC77	1.780106315	3.463815	4.91*10−4
SON	−3.255568244	0.641425	3.42*10−6	MAGEB2	1.800177019	1.13101	5.23*10−4
RPL37A	−3.063072627	0.495454	3.59*10−6	MRPL32	1.813121709	1.731127	5.30*10−4
RPS3A	−2.656749844	0.138981	5.20*10−6	PRR3	1.953870066	1.283166	5.54*10−4
EIF3A	−2.759065635	0.344374	5.62*10−6	ENPEP	2.517320194	1.728008	5.56*10−4
EFTUD2	−3.045537656	0.617988	6.29*10−6	C10orf10	2.288876507	1.313241	6.54*10−4
RPS3	−2.971857046	0.477078	6.66*10−6	IL33	2.662264178	3.132637	6.54*10−4
RPS19	−2.466435068	0.00894	7.45*10−6	INTU	1.826282244	0.476045	6.59*10−4
PSMD11	−3.470882770	1.358785	7.79*10−6	SPRYD7	1.536004509	0.257087	6.81*10−4
PSMD2	−2.709936293	0.395144	8.13*10−6	ETS1	1.690318109	0.560675	6.84*10−4
RPL23	−2.844317170	0.393678	8.57*10−6	ENY2	1.980794257	0.604161	6.90*10−4
RPL18A	−2.740824166	0.289759	9.22*10−6	ERN1	1.745020886	1.533007	7.10*10−4
PSMD6	−3.354882032	0.861655	1.05*10−5	ANAPC15	1.607502524	0.317184	7.13*10−4
RPL30	−2.694047310	0.276485	1.08*10−5	PKDCC	2.905300513	1.213858	7.15*10−4
RPL10A	−2.491750667	0.137017	1.19*10−5	YOD1	1.496661306	0.163577	7.71*10−4
RPL35	−3.014740325	0.779897	1.35*10−5	PRDM11	1.144218586	3.523791	8.03*10−4
PSMD1	−3.310762528	0.941473	1.36*10−5	ULBP2	1.864258701	0.569045	8.24*10−4
RPL34	−2.524555701	0.202938	1.53*10−5	KLHL25	1.817920541	2.862492	8.48*10−4
SF3B4	−2.770516260	0.502821	1.56*10−5	PUS7	1.649301427	1.684449	8.57*10−4
SNIP1	−2.005111303	2.881253	2.06*10−5	CCT7	1.333541723	3.181397	8.92*10−4
EIF3B	−2.560075303	0.466985	2.21*10−5	SMCO2	1.990177414	0.72497	8.92*10−4
RPL38	−2.492702772	0.246558	2.26*10−5	PLIN3	2.007664224	2.59357	9.37*10−4
RPS18	−2.977185914	0.873169	2.31*10−5
EIF3C	−2.831263223	0.577086	2.33*10−5
RPS28	−2.242147031	0.038273	2.36*10−5
SF3B2	−2.839604557	0.595351	2.62*10−5
RPL14	−2.541293307	0.546392	3.12*10−5
EIF4A3	−2.785175497	0.57183	3.31*10−5
RPS20	−2.485170409	0.330939	3.69*10−5
RPLP0	−2.432589089	0.439882	3.86*10−5
RPL19	−2.569886065	0.518541	4.06*10−5
RPLP1	−2.450487106	0.317139	4.07*10−5
EIF3E	−2.466991836	0.601668	4.09*10−5
KIF11	−2.302340058	0.313574	5.55*10−5
RPS29	−2.941379189	1.030957	6.43*10−5
SNW1	−2.496862349	0.645152	6.53*10−5
CLTC	−2.272438754	0.387034	6.98*10−5
RNASEK	−2.946825136	1.396363	7.35*10−5
RPS10	−2.188345422	0.331573	7.63*10−5
RPS4X	−2.352833551	0.528522	1.13*10−4
RPL31	−2.278564688	0.355378	1.16*10−4
POLR2A	−2.540993176	1.011829	1.25*10−4
AP2M1	−2.271003604	0.462875	1.28*10−4
PRPF38A	−3.100080397	0.48687	1.34*10−4
NFKB2	−2.672485560	0.914358	1.42*10−4
ATP6V0C	−2.641594386	0.701514	1.45*10−4
RNF115	−2.135247926	1.98335	1.64*10−4
PSMA1	−2.108917373	0.408836	1.68*10−4
RPL9	−2.483195553	0.86102	1.92*10−4
TMEM104	−1.800709365	0.031962	1.93*10−4
SF3A3	−2.330501245	0.725445	1.98*10−4
SUPT6H	−2.162823668	0.406903	2.05*10−4
RPL8	−2.162194321	0.628083	2.16*10−4
UBA52	−2.283373011	0.48158	2.23*10−4
NAPA	−2.836178345	1.493822	2.27*10−4
POMP	−2.111959692	0.666725	2.36*10−4
CWC22	−0.549454817	3.977714	2.68*10−4
MYLK2	−1.477981406	3.009054	2.97*10−4
RPL26	−1.864662427	0.236191	3.09*10−4
RPS24	−2.305809148	0.554555	3.10*10−4
U2AF2	−2.017993637	0.363156	3.10*10−4
PSMD12	−2.114514773	0.487831	3.11*10−4
RPL11	−2.489596610	0.975893	3.13*10−4
RPS6	−2.019226268	0.451242	3.43*10−4
SNU13	−2.007543638	2.581144	3.53*10−4
UBQLN4	−2.037617735	0.537563	3.64*10−4
RDH12	−2.039401012	0.544247	3.67*10−4
RPS23	−1.962882558	0.297086	3.73*10−4
RPL21	−2.320609680	0.616757	3.86*10−4
SOX4	−0.817371199	3.794075	4.02*10-4
MAN2B2	−1.551458039	2.393802	4.04*10−4
RPS9	−1.897714697	0.350598	4.19*10−4
EBF1	−1.838053057	0.382412	4.71*10−4
RPL18	−2.624801830	0.934007	4.86*10−4
P2RY6	−1.138161361	2.725195	4.99*10−4
MYL7	−2.340374930	0.719223	5.24*10−4
RPS13	−2.173910175	0.605976	5.49*10−4
DDX24	−1.865594787	0.5511	5.50*10−4
RPL6	−2.126988850	0.54718	5.57*10−4
RPL24	−2.231338323	0.675468	5.57*10−4
SF3B5	−1.984370978	0.524576	5.61*10−4
ATP6V0D1	−1.755423119	0.275368	5.65*10−4
NUP62	−1.826438123	0.268001	5.66*10−4
AURKA	−1.570182010	0.049012	5.88*10−4
MDGA1	−1.633339990	0.150249	5.93*10−4
COG1	−1.872237673	0.352524	5.96*10−4
EP400	−2.151707922	0.806586	6.13*10−4
RPS27A	−2.306223246	0.811435	6.15*10−4
PSMD7	−2.820224689	1.557067	6.23*10−4
ADAT1	−0.888348000	3.570255	6.25*10−4
ARGLU1	−1.741768395	0.379625	6.39*10−4
SPSB1	−1.752659824	0.258357	6.41*10−4
PSMD14	−1.882027486	0.637177	6.58*10−4
RPTOR	−1.822682082	0.50851	6.66*10−4
CKAP5	−2.164529656	0.732547	6.68*10−4
HINT2	−1.737570477	0.41976	7.40*10−4
RPL32	−1.887042530	1.678868	7.62*10−4
PSMA7	−1.789226078	0.281513	7.70*10−4
RBM25	−2.238222915	0.970273	7.81*10−4
WBP11	−2.006891128	0.733633	8.20*10−4
SNRNP200	−2.226376141	0.749748	8.55*10−4
DCAF1	−1.726121073	0.431739	8.58*10−4
TST	−1.866804160	1.884949	8.76*10−4
RPL39	−1.645854099	0.317614	8.78*10−4
TRPM7	−2.430646722	0.885995	8.84*10−4
AFDN	−1.315363104	1.313086	9.13*10−4

^AAssay score: normalized score for the median cytoplasm intensity assay feature

^BAvg = average.

^CSD: standard deviation.

^Dp-values are not adjusted for multiple testing (p <3.6*10⁻⁶ after Bonferroni adjustment for 14’000 genes with expressed transcripts)

^EThe 6 hit genes identified in the screening and which are members of the COPI complex are highlighted in bold and underlined.

The three top hits for decreased HDL uptake were COPA, ARCN1, and COPZ1, which encode for three of the seven canonical components of the COPI coatomer complex. Silencing of COPB1 (rank 8), COPB2 (rank 11), and COPG1 (rank 23) also significantly decreased the uptake of HDL in Huh-7 cells (table 1). Silencing of COPE, which encodes for the thermosensitive and accessory ε-COP subunit of the COPI coatomer’s B-subcomplex^31,32, also resulted in lower HDL uptake, however, without passing the RSA p-value cut off (supplemental table S1). The knockdowns of COPG2 and COPZ2, which encode for paralogues of γ-COP and ζ-COP, respectively, but are not part of the COPI coatomer^33,34, did not reduce HDL uptake (supplemental table S1). Of note, SCARB1 belonged to the genes whose knockdown increased the uptake of HDL in Huh-7 cells (table 1).

In vitro validation of the COPI complex as a rate-limiting factor in HDL endocytosis

To validate these findings, we utilized pools of three siRNAs against each candidate from a different vendor than the siRNA screening library to knockdown target genes in Huh-7 cells. The knockdown reduced the mRNA and protein expression of the targeted genes by 60% (COPZ2) to 90% (ARCN1, COPZ1, COPG2; supplemental figure S2A) and by 76% (COPG1/γ-COP) to 97% (ARCN1/ δ-COP) (figure 1B), respectively. Of note, while non-targeted COPI mRNAs were not suppressed (supplemental figure S2A), the abundance of some non-targeted COPI proteins was substantially decreased (figures 1C–1F, supplemental figure S2B). probably because they form complexes with the targeted protein⁴ (figure 1A). For example, γ-COP is suppressed by the knockdowns of all units of the F-subcomplex but much less so by the knockdowns of the units of the B-subcomplex or the paralogues (figure 1C). The knockdowns of the paralogues show good specificity, i.e. no suppression of any unit of the COPI coatomer (figures 1C to 1F and supplemental figure S2B). However, there are also some exceptions. For example, anti- δ -COP immunoreactivity did not decrease or even increased upon knockdown of COPZ1 and COPG1, respectively (figure 1D). And in some instances, the knockdowns of COPI units of the F-subcomplex also decreased the abundance of units of the B-complex, most prominently α-COP (−57%) a ε-COP (−68%) upon knockdown of COPB1, (figures 1E and 1F, supplemental figure S2B).

We next used flow cytometry to investigate the effects of siRNA interference against each COPI component on the uptake of HDL labelled on either the protein moiety (Atto655-HDL) (figures 1G and 1H) or the lipid moiety with 1,1’-dioctadecyl- 3,3,3’,3’-tetramethylindocarbocyanine perchlorate (DiI-HDL) being fluorescently labeled (figures 1I and 1J). Knocking down the indispensable components of the COPI coatomer (COPA, COPB1, COPB2, ARCN1, COPG1, or COPZ1) reduced the uptake of Atto655-HDL (figures 1G and 1H) and DiI-HDL (figures 1I and 1J) by at least 75% and 79%, respectively, the latter to a similar extent as knocking down SCARB1 that was used as the positive control for selective lipid uptake from HDL². Conversely, knocking down the dispensable COPE as well as the paralogous COPG2 or COPZ2 altered neither Atto655-HDL nor DiI-HDL uptake. Despite these large effect sizes when analyzed individually, the differences measured upon interferences with indispensable COPI subunits did not achieve statistical significance after correcting for multiple testing (figures 1G and 1I). However, the decreases became statistically significant when replotting and comparing the pooled data of the six indispensable COPI versus the three dispensable or paralogous COPI gene knockdowns (figures 1H and 1J, both p < 0.001). Taken together, the loss of indispensable COPI coatomer subunits compromises the uptake of both protein-labeled HDL and fluorescently lipids of HDL by Huh-7 cells.

Glycosylation and cell surface expression of SR-BI are compromised upon silencing of different COPI genes.

As expected because of its canonical role in selective lipid uptake², the loss of SR-BI decreases the uptake of DiI-HDL into Huh-7 cell (figure 1I). However, both in the siRNA screening (table 1) and in the targeted replication experiments (figure 2A), the knockdown of SCARB1 led to increased uptake of protein-labeled Atto655-HDL. Interestingly, this increase was not prevented by the additional knockdown of ARCN1 (figure 2A) underscoring the missing contribution of SR-BI to HDL-holoparticle uptake into Huh-7 cells. To explore the reason for the decreased activity of SR-BI in selective lipid uptake (i.e. DiI-HDL), we investigated the effects of lost COPI function on the expression of SR-BI. Silencing of the different COPI genes did not alter SCARBI mRNA (supplemental figure S3A and S3B). In addition, the electrophoretic mobility of SR-BI was higher after knockdown of COPA, COPB1, COPB2, ARCN1, COPG1, and COPZ1 but unaltered after knockdowns of COPE, COPG2, or COPZ2 (figure 2B), suggesting an influence on post-translation modification. We hypothesized that the increased mobility of SR-BI reflects decreased glycosylation in response to the loss of COPI proteins. Testing this hypothesis we found that the treatment of Huh-7 cells with neuraminidase (that removes sialic acid residues) alone or in combination with O-glycosidase (that removes O-linked disaccharides) but not O-glycosidase alone, resulted in similar changes of SR-BI’s electrophoretic mobility as observed upon knockdown of ARCN1, while treatment with PGnase F (that removes N-linked glycans) led to a much more profound increase of SR-BI’s mobility (figure 2C). In agreement with SR-BI being only N-glycosylated³⁵, our results suggest that the knockdown of indispensable COPI genes interferes with the sialylation of SR-BI.

Figure 2. Effect of silencing COPI genes on the HDL-uptake activity, protein expression, glycosylation and localization of SR-BI.

Wild type Huh-7 cells (A-E) or Huh-7 cells overexpressing GFP-SR-BI (F-I) were transfected with the indicated siRNAs. 72 hours after transfection, the cells were collected and used for functional evaluation. A. The uptake of HDL was recorded as described in the legend of figures 1G and 1H. The data were normalized to the non-coding control (NC) and are shown as means ± SD of 4 independent experiments for each condition separately. B and C show representative Western Blots of SR-BI each out of 3 independent experiments, using lysates immediately after harvesting (B) or after overnight treatment with the indicated combinations of deglycosylating enzymes (C). TATA binding protein (TBP) served as the loading control in (B). Figures D-F show SR-BI cell surface levels of alive Huh-7 cells as determined by flow cytometry with an antibody against an extracellular epitope of SR-BI. The fluorescence emitted by the GFP-SR-BI was recorded as a measure of total cellular SR-BI (G). All data were normalized to the non-coding (NC) control and are shown as means ± SD of 4 (D) or 6 (F, G) independent experiments. E compares merged data on the non-dispensable COPI subunits (red bars) and the dispensable or paralogous COPI genes (blue bars). Figures H and I depict confocal microphotographs on the colocalization of GFP-SR-BI with ZO-1 (H) and LAMP1 (I). The data shown are representative of 2 independent experiments. The scale bar for both microphotographs is 25 μm. Statistical analysis was performed by using Kruskal-Wallis test with Dunn’s multiple comparisons test between the indicated conditions (A) or between the NC and each targeting siRNA (D, F, G). The pooled data of the non-dispensable COPI gene knockdowns (red bars, N = 24) and the pooled data on the dispensable or paralogous COPI gene knock-downs (blue bars, N = 12) were compared by using two-tailed Mann-Whitney test (E). In (D-G) only statistically significant differences (p<0.05) are shown.

Flow cytometry analysis with an anti-SR-BI antibody revealed a strongly reduced cell surface abundance of SR-BI upon knockdown of each indispensable COPI subunit gene similar to the knockdown of SCARB1 (figure 2D). Since western blotting accompanied with cell surface biotinylation after knockdown of SCARB1 detected no or only traces of SR-BI immunoreactivity both in the total cell lysate (figure 2B) and on the cell surface (supplemental figure S3C), we consider the residual signal from the flow cytometry analysis as unspecific background (figure 2D). The statistical analysis of the pooled data revealed a significant decrease of SR-BI cell surface abundance after knockdown of indispensable COPI genes versus dispensable and paralogous COPI genes (p<0.001, figure 2E).

To corroborate that the loss of COPI genes interferes with the trafficking and cell surface expression of SR-BI, we overexpressed SR-BI with an N-terminal GFP-tag in Huh-7 cells and an empty vector (EV) as the control. Western blot analysis with either an anti-SR-BI or an anti-GFP antibody confirmed the successful overexpression of the construct (supplemental figures S3D and S3E). Cells overexpressing GFP-SR-BI (clone “6” in supplemental figures S3D and S3E) emitted fluorescence and displayed more anti-SR-BI-immunoreactivity on the cell surface than the EV cells (supplemental figures S3F and S3G, respectively). Silencing of ARCN1 or COPG1 in Huh-7 cells overexpressing GFP-SR-BI decreased cell surface levels of SR-BI by 80.5% and 77%, but increased the total GFP signal by 41.5% and 43.3% (figures 2F and 2G, respectively), albeit without reaching statistical significance.

We next examined with confocal microscopy the subcellular distribution of the GFP signal in Huh-7 cells overexpressing GFP-SR-BI. Upon silencing of ARCN1, the finely granulated GFP signal disappeared from the ZO-1 positive plasma membrane to a great extent but accumulated in LAMP1-positive late endosomes and lysotracker-positive lysosomes (figure 2H, 2I and supplemental figure S4A–S4C). The GFP signal was co-localized with neither the ER (SEC61A1, supplemental figure S4D) nor the Golgi (GM130, supplemental figure S4E).

Taken together, the loss of indispensable COPI coatomer subunits interferes with the proper glycosylation and cell surface abundance of SR-BI. This explains the disturbed uptake of HDL’s DiI (i.e. selective lipid uptake) but not the decreased uptake of atto655-HDL (i.e. holoparticle uptake) by Huh-7 cells lacking COPI components.

Silencing of COPI genes increases the cell surface abundance of ABCA1 in hepatocytes

In various cell types, cholesterol efflux has been shown to involve endocytosis and re-secretion of HDL². We therefore investigated the effects of lost COPI function on ABCA1 expression and cholesterol efflux. ABCA1 mRNA levels were significantly decreased by 33% upon silencing of indispensable COPI genes compared to non-essential genes (figures 3A and 3B, p < 0.001). Similarly to SR-BI, knockdowns of essential COPI genes increased the electrophoretic mobility of the ABCA1 band (figure 3C), possibly indicating altered glycosylation of ABCA1. To examine whether the lack of essential COPI coatomer affects the cell surface abundance of ABCA1, we carried out cell surface biotinylation experiments in Huh-7 cells after ARCN1 knockdown. Silencing of ARCN1 had no significant effect on ABCA1 levels in total lysates but led to a non-significant two-fold increase of ABCA1 abundance on the cell surface (figures 3D and 3E). Silencing of COPI genes slightly increased cholesterol efflux from Huh-7 cells to lipid-free apoA-I (figures 4A and 4B) and HDL (figures 4C and 4D) by 20% (COPE)-to 105% (COPA) and 25% (COPE) to 110% (COPB2), respectively, but these increases were not statistically significant, even not by comparing data of indispensable vs. non-essential COPI gene knockdowns (figures 4B and 4D). We next investigated whether ABCA1 contributes to the decreased Atto655-HDL uptake observed upon loss of ARCN1 (figures 4E and 4F). Upregulation of ABCA1 by treatment with a Liver X receptor (LXR) agonist increased the uptake of Atto647-HDL by 33%, although not statistically significant (figure 4E). In both unstimulated and stimulated cells, silencing of ARCN1 and ABCA1 decreased the uptake of Atto647-HDL by 67% and 21% and by 69% and 46%, respectively (p < 0.001 for both, figure 4E). The comparison of residual Atto647-HDL uptake between stimulated and non-stimulated cells after knockdown of either ARCN1 or ABCA1 did not reveal any statistically significant difference. When compared to single knockdown of ABCA1, the additional knockdown of ARCN1 did not further decrease the uptake of Atto655-HDL (figure 4F).

Figure 3. Effect of silencing COPI genes on the expression and the cell surface abundance of ABCA1.

Wild type Huh-7 cells were transfected with the indicated siRNAs (A-E). 72 hours after transfection, the cells were harvested and used for the evaluation of mRNA (A, B), protein (C) and cell surface levels (D, E) of ABCA1. A and B. The data were normalized to the non-coding (NC) control and are shown as means±SD of 4 independent experiments. B compares the summarized data on the non-dispensable COPI subunits (red bars) with the data on the dispensable or paralogous COPI genes (blue bars). C shows a representative Western Blot from 3 independent experiments. For the analysis of ABCA1 expression on the cell surface (D, E), the cells were incubated 48 hours after transfection with DMEM supplemented with 0.5% FBS, 1% P/S and 10 μM T0901317 for 24 hours prior harvesting. The western blots were probed with antibodies against ABCA1, TBP (as a control for intracellular protein expression), and Na⁺/K⁺-ATPase (as a control for cell surface protein expression). For details see methods. D shows one representative Western Blot out of 4 independent experiments. E shows the respective quantification of the obtained ABCA1 immunoreactive signal normalized either to Na+/K+-ATPase or to TBP. The data are normalized to the non-coding control (NC) and shown as means ± SD of 4 independent experiments. Statistical analysis was performed using either Kruskal-Wallis test with Dunn’s test for multiple comparisons between the NC and each targeting siRNA (A) or between the indicated conditions (E). The pooled data of the non-dispensable COPI gene knockdowns (red bars, N = 24) and the pooled data on the dispensable or paralogous COPI gene knockdowns (blue bars, N = 12) were compared by using two-tailed Mann-Whitney test (B). In (A, B), only levels of statistical significance with p<0.05 are shown.

Taken together, the loss of indispensable COPI subunits including ARCN1 leads to mis-glycosylation and increased cell surface abundance of ABCA1 but has no significant impact on cholesterol efflux. However, our findings suggest that the compromised handling of ABCA1 contributes to the decreased uptake of protein-labeled HDL by Huh-7 cells upon loss of ARCN1.

Loss of COPI genes impairs the secretion of apoA-I

Since hepatic ABCA1 contributes to the biogenesis of HDL, we tested the effects of loss of COPI gene expression on the secretion of apoA-I. Silencing of the essential COPI genes decreased the expression of APOA1 mRNA by 36% to 57% (figure 5A). After pooling of data, the 52% reduction of indispensable compared to non-essential COPI genes was significant (figure 5B, p < 0.001). The secretion of apoA-I was decreased by 41% and 22% upon knockdown of COPA (p = 0.045) and ARCN1, respectively, but unaltered upon knockdown of COPG1 (−3%) and increased by 33% upon knockdown of COPG2 (figure 5C). The comparison of the former three with the paralogous COPG2 revealed significantly decreased apoA-I secretion upon loss of the indispensable COPI genes (p = 0.002; figure 5D).

Figure 5. APOA1 expression (A, B) and apoA-I secretion (C, D) by Huh-7 cells depleted of different COPI genes.

Huh-7 cells were transfected with the indicated siRNAs. After 72 hours the cells were collected for quantification of APOA1 mRNA levels by qRT-PCR (A, B). For quantification of apoA-I secretion (C, D), the cells were reseeded 48 hours after transfection. 6 hours later the medium of the cells was replaced by DMEM containing neither phenol red nor FBS. After another 48 hours, both cells and media were collected with the former being counted and the latter being used for the measurement of apoA-I by ELISA. ApoA-I concentrations in the media were normalized to the cell number of each condition. The data are shown as means ± SDs of 3 (A, B) or 6 (C, D) independent experiments, each normalized to the NC. B and D compare the data of the non-dispensable COPI subunits (red bars) versus the data of dispensable or paralogous genes (blue bars). Statistical analyses were performed using Kruskal-Wallis test coupled with Dunn’s multiple comparison test (A,C). The pooled data of the non-dispensable COPI gene knockdowns (red bars, N = 18 in B and D) and the pooled data on the dispensable or paralogous COPI gene knockdowns (blue bars, N = 9 in B, N = 6 in D) were compared by using two-tailed Mann-Whitney test. Only statistically significant differences (p<0.05) are shown.

Effect of ARCN1 on cellular sterol homeostasis

We next investigated the effect of ARCN1 silencing on sterol homeostasis in Huh-7 cells in the presence of HDL, and under the two conditions that we used for the measurement of HDL uptake (i.e. without LXR agonist) and cholesterol efflux (in the presence of LXR agonist T0901317 and acyl-coenzyme A (CoA):cholesterol acyltransferases (ACAT) inhibitor Sandoz 58–035), (supplemental table S3). In agreement with reduced selective and HDL holoparticle uptake as well as slightly higher cholesterol efflux, cells lacking ARCN1 had a ~24% lower content of total cholesterol under either condition. The content of cholesteryl esters was increased by the knockdown of ARCN1 so that the decrease of total cholesterol was caused by the decrease in unesterified cholesterol (supplemental figures S5A and S5B). The increase in cholesteryl esters was not prevented by ACAT inhibition so that the loss of ARCN1 appears to affect the hydrolysis rather than production of cholesteryl esters (supplemental figure S5A). The lack of ARCN1 also led to lower cellular contents of most phytosterols (supplemental table S3) and to higher cellular concentrations of cholesterol precursors, but these changes were not statistically significant (supplemental table S3).

Genetic data indicate that the COPI complex regulates HDL-C levels

To explore any limiting effects of the COPI coatomer on HDL metabolism, we tested the associations of 5,821 SNPs in the nine COPI genes with HDL-cholesterol concentration (HDL-C) in the most recent aggregated data sets of the Global Lipids Genetics Consortium using >1.65 million individuals¹⁹.

At the Bonferroni adjusted GWAS threshold of p = 9.60*10⁻¹⁰, and in line with the reduced HDL uptake into Huh-7 cells lacking ARCN1, 42 SNPs of ARCN1 were associated with significantly higher levels of HDL-C (figures 6A) but no difference in triglycerides (figure 6B). 40 of these SNPs are in strong if not complete linkage disequilibrium (LD > 0.9). Among them, the intronic variant rs11216909 which has a minor allele frequency (MAF) of 0.171, has the strongest association with HDL-C (METAL effect size 0.0126 mmol/L = 0.49 mg/dl per allele, p= 1.34*10⁻¹³, figure 6A). In the GTEX databank²² it is associated with lower gene expression of ARCN1 in full blood (beta coefficient = −0.11, p=9.35×10⁻⁹, figure 6C) but not in liver. Moreover the rs11216909 polymorphism showed stronger associations with the expression of RPL5P30 and IFT46 which are neighboring the ARCN1 gene on the forward and the reverse strands, respectively, in various tissues, however neither in full blood nor in liver.

COPB1 harbors four and eleven SNPs, which were associated with significantly higher and lower levels of HDL-C including rs7121538 (MAF = 0.14, METAL effect size 0.0133 mmol/L/ 0.51 mg/dL per allele, p= 1.08*10⁻¹¹), which was already previously associated with higher HDL-C levels³⁶. One SNP in COPG2 was associated with lower levels of HDL-C (MAF = 0.0076, METAL effect size −0.054 mmol/L/ −2.07 mg/dL per allele, p=3.63*10⁻¹⁰; figure 6A) and all five SNPs of COPZ2 were associated with significantly higher levels of HDL-C (METAL effect sizes 0.021 to 0.024 mmol/L/0.81 to 0.92 mg/dL per allele, p < 3.55*10⁻¹⁷, figure 6A) as well as lower levels of triglycerides (figure 6B).

Because the GLGC databank does not report any associations with apoA-I levels, we compared the associations of SNPs of COPI genes with HDL-C and apoA-I levels in the smaller dataset of UK Biobank (N = 315,133; supplemental figure S6)²⁰. Only one SNP of ARCN1 (rs59610752, β = 0.009 mmol/L/0.35 mg/dL per allele) and three SNPs of COPB1 (maximal β = 0.011 mmol/L/0.42 mg/dL per allele) were associated with HDL-C levels (supplemental figure S6A), while 44 SNPs of ARCN1 and none of COPB1 were associated with apoA-I levels (maximal β = 0.7 mg/dL per allele; supplemental figure S6B) at the genome-wide threshold for statistical significance (p < 5*10⁻⁸). The analysis of the Whole Genome Sequencing (WGS) data of nearly 500’000 individuals in the UK Biobank (https://www.azphewas.com: Dataset: UK Biobank 500k WGS (v2) Public)²¹ identified two variants of ARCN1, namely 11–118601639-C-A (ENST00000534182: p.R55S; MAF: 0.1247; N = 395962) and 11–118600708-T-C (ENSP00000264028.4: p.I510I; MAF 0.1247, N = 395953) to be significantly associated> with plasma concentrations of HDL-C (effect size β of both 0.023 mmol/L / 0.88 mg/dL per allele; p = 1.40*10⁻¹¹ and p = 1.49*10⁻¹¹, respectively) as well as apoA-I (effect size β of both 0.022 g/L per allele and both p = 1.31*10⁻¹²). No other variant of any COPI gene had a significant association with HDL-C or apoA-I levels at the threshold p < 10⁻⁸

Finally, we investigated the plasma lipids of five American patients heterozygous for pathogenic variants in COPA causing a type I interferonopathy⁸ (figure 6D) as well as four children homozygous for the pathogenic variant p.K652E in COPG1 causing combined immunodeficiency and their heterozygous parents⁹ (figure 6E). Compared to adults from the NHANES reference population²³, the HDL-C levels of the COPA syndrome patients were either normal or rather low except for one female carrier of the COPA (p.E241K) variant who presented with high levels of HDL-C (figure 6D). Because of their origin from Oman, we compared the HDL-C levels of the homozygous COPG1 variant carriers and their heterozygous parents with those of populations in the United Arab Emirates^24,25. HDL-C levels of affected children were rather low while those of their parents were within the normal ranges for the respective control populations (figure 6E).

We examined the plasma lipoprotein profile of mice with heterozygosity for the Copa^E241K(Copa^E241K/+) mutation (figures 7A and 7B) or homozygosity for the Copg1^K652Emutation (Copg1^{K652E/ K652E}) (figures 7C and 7D), which mimic the immunological phenotypes of the COPA and COPG1 syndromes described above^9,26. FPLC profiling indicated lower levels of HDL-C in both Copa^E241K/+ and Copg1 ^{K652E/ K652E} mice of both sexes compared to wild type mice. Upon direct quantification, HDL-C levels were 60% and 10% lower in male and female Copa^E241K/+ mice, respectively (figures 7A and 7B), and 19% and 21% lower in male and female Copg1 ^{K652E/ K652E} mice, respectively (figures 7C and 7D), however without reaching statistical significance. To increase the statistical power, we combined the data of female and male mice and corrected for sex. Using two-way ANOVA with genotype and sex as additive factors, HDL-C levels were significantly lower in Copa^E241K/+ mice than in wild type controls (51.3 ± 18.0 vs. 83.2 ± 16.2 mg/dL; genotype effect, p = 0.013). HDL-C levels were also significantly lower in Copg1^K652E/K652E mice than in wild type controls after correction for sex (53.6 ± 18.1 vs. 67.0 ± 12.0 mg/dL; genotype effect, p = 0.036). Sex was not a significant factor in the Copa cohort (p = 0.749), whereas it was significant in the Copg1 cohort (p = 0.015). The quantification of apoA-I-derived peptides by mass spectrometry did not reveal consistently different concentrations in the pooled plasma of male and female mutant mice compared to their sex-specific littermate controls (supplemental table S4).

Figure 7: Lipoprotein profiles of plasmas from mice expressing pathogenic variants of COPA (A, B) or COPG1 (C, D) and their littermates.

Lipoproteins were fractionated by fast protein liquid chromatography column (FPLC) of pooled plasmas from three male (A) and female (B) mice heterozygous for Copa^E241K or five male (C) and female (D) mice homozygous for Copg1^K652E (dashed lines) and the respective control littermates of the same sex (complete lines). Cholesterol in the fractions was measured by an enzymatic photometric test. In all graphs, the x-axis describes the time at which the different fractions were collected, while the y-axis describes the absorbance of the different fractions measured at 505 nm. a.u: arbitrary units. The lipid levels presented in the tables below the FPLC profiles were measured by photometric tests on the COBAS8000 autoanalyzer from Roche diagnostics (Rotkreuz, Switzerland). The values are presented as the means± SD of three (A, B) or five (C,D) animals. The statistical analyses were performed using a 2-tailed Mann-Whitney (M-W) test.

In summary, HDL-C levels were higher in carriers of common polymorphic ARCN1 and COPB1 alleles, but normal or rather low in both humans and mice carrying rare immunopathogenic variants of COPA or COPG1.

Discussion

The molecular mechanism of HDL holoparticle removal by the liver is an unresolved part of HDL metabolism. Its elucidation is not only of biological interest but may also have clinical relevance. Like LDL-R dysfunction impairs LDL removal and thereby leads to LDL-hypercholesterolemia, disturbed HDL removal is expected to increase HDL-C levels. In fact, the plasma concentration of inhibitory factor IF1, which inhibits ecto-F1ATPase^2,3 and thereby HDL holoparticle uptake, correlates with HDL-C³⁷. Likewise, defective selective lipid uptake due to variants of SCARB1 is associated with increases in HDL-C³⁸. However, it is uncertain how this will affect the risk of ASCVD. HDL-C levels are inversely associated with the risk of ASCVD, however, only within the range of the lower 6 or 7 deciles¹. Moreover, large epidemiological studies have found high HDL–C levels associated with increased risks of total and cardiovascular mortality¹. Molecular understanding of HDL removal may help to resolve the question of whether the blockage of HDL removal is beneficial by maintaining potentially protective particles in the circulation, or deleterious, for example, by delaying reverse cholesterol transport or increasing the lifetime of HDL and thereby the likelihood of potentially harmful particle modifications.¹

To identify genes that limit the uptake of atto655-labeled HDL holoparticles into hepatocytes, we performed a genome-wide RNA interference screen. This approach was already successfully undertaken by our lab and other labs to identify regulators of hepatic or endothelial LDL uptake^13,39,40, but, to the best of our knowledge, has never been applied to HDL. None of the 128 genes identified by our screen as limiting the uptake of atto655-HDL by Huh-7 cells encoded for any of the proteins previously suggested to limit HDL uptake (vigilin⁴¹, HDL binding protein 2⁴², CD36⁴³, ecto-F1ATPase³, P2Y13³, megalin⁴⁴, cubilin⁴⁴ ) or any bona fide endocytic receptor. The knockdown of P2YR6, which encodes the purinergic receptor P2Y6 is noteworthy, because the activation of purinergic receptors P2Y13 and P2Y1 by binding of apoA-I to ectoATPase and subsequent ADP generation was previously shown to promote the HDL holoparticle uptake into hepatocytes and endothelial cells, respectively, by an as yet unresolved endocytic pathway³. However, the metabolic effects of P2Y6 have been characterized in skeletal muscle, adipocytes, and hypothalamus rather than in liver⁴⁵, possibly because, although expressed, its activation in Huh-7 and HepG2 cells did not result in any Ca²⁺ mobilization⁴⁶, a typical second messenger response to the activation of purinergic receptors. Also of note, the reduced atto655-HDL uptake upon interference with CLTC and AP2M1 supports the finding of previous microscopic studies that hepatocytes endocytose HDL holoparticles via clathrin-coated pits². Last but not least, the knockdown of SCARB1, which encodes the mediator of selective lipid uptake from HDL^2,38, increased the uptake of HDL holoparticles.

Our screening identified six of the seven canonical components of the COPI coatomer as limiting factors of HDL uptake, namely COPA, COPB1, COPB2, ARCN1, COPG1, and COPZ1. By using other siRNAs and flow cytometry as an alternative recording method, we confirmed that the loss of each of these six indispensable components of the COPI coatomer results in reduced HDL holoparticle uptake into Huh-7 cells. The only component not identified as a contributor by our screening – COPE – was neither limiting HDL uptake in our targeted verification experiments. Of note, COPE encodes for the least evolutionarily conserved COPI subunit, namely ε-COP whose main role is to stabilize α-COP³². In the mutant CHO cell line ldlF, the lack of COPE interferes with the trafficking of cargo at the non-permissive temperature (39.5°C)⁴⁷. All our HDL uptake experiments were carried out at lower temperature (37°C) although still higher than the permissive one (34°C), suggesting that under these conditions ε-COP function might be dispensable for the function of the COPI complex. Interestingly and in agreement with our findings, Steiner and colleagues found that the knockdowns of COPA, COPG1, and ARCN1 but not COPE led to aberrant activation of the inflammatory cGAS/STING signaling⁴⁸. The two other COPI genes COPG2 and COPZ2, which did not limit HDL uptake either in the screening or in the validation experiments, are paralogs of COPG1 and COPZ1, respectively. Their functions are not well understood, but they do not appear to limit the functionality of the COPI coatomer^33,34. Of note, we also found that the knockdowns of each of the six indispensable subunits, but neither of COPE nor the paralogous COPG2 and COPZ2, compromised LDL uptake⁴⁹.

We identified SR-BI, ABCA1, and apoA-I as three targets, whose gene expression (APOA1 and ABCA1 but not SCARB1), cellular localization (SR-BI and ABCA1), functionality (SR-BI and ABCA1), or secretion (apoA-I) are altered by the loss of COPI subunits and hence coatomer function, but with directionally different effects. Compared to the knockdown of non-essential COPI genes, the interference with indispensable COPI genes led to a significant 30–50% reduction in the mRNA expression of APOA1 and ABCA1 but no significant change in SCARB1 expression. While western blotting analysis did not indicate any obvious differences in the total protein levels of ABCA1, the lower mRNA expression of APOA1 may contribute to the lower apoA-I secretion recorded by us. Mechanistically, it is known that the COPI coatomer contributes to the retrograde transport of transcription factors or their activators from the Golgi to the ER, for example, of uncleaved SREBP-SCAP complexes or the ARA160 coactivator of the androgen receptor^50,51. It may therefore be that the various transcription factors and cofactors contributing to gene expression of APOA1, ABCA1, and SCARB1 genes differ in their susceptibility to disturbances of retrograde transport.

In addition to retrograde transport from the Golgi to the endoplasmic reticulum, the COPI coatomer plays an important role for the anterograde transport of proteins through the Golgi cisternae, where glycosylation of cargo proteins takes place⁴. Of note, a de novo mutation in ARCN1 was previously identified as the cause of a transient N-glycosylation deficiency during episodes of acute illness⁵². In our hands, SR-BI and ABCA1 showed altered electrophoretic mobility upon knockdown of all six indispensable COPI genes but not the three dispensable genes. By in vitro treatment with neuraminidase, we mimicked the altered mobility of SR-BI. Moreover, the altered electrophoretic properties of SR-BI upon loss of COPI genes resembled those previously found upon treatment with sialidase or substitution of the N-glycosylated amino acid residues of SR-BI³⁵. As SR-BI is known to be only N-glycosylated^35,53, we conclude that the loss of each indispensable COPI subunit interferes with the sialylation of SR-BI’s N-glycans. Likewise, the slightly increased electrophoretic mobility of the band corresponding to ABCA1 is also in agreement with defective N-glycosylation. ApoA-I proteoforms include only a quantitatively minor fraction of a diglycan-proteoform, with one mole of Hex and HexNAc⁵⁴ each, which cannot be distinguished from the unglycosylated proteoforms by SDS-PAGE.

The defective N-glycosylation appears to have opposite effects on the cell surface trafficking of SR-BI and ABCA1. While SR-BI levels are reduced at the cell surface, possibly by increased lysosomal degradation as indicated by our microscopic co-localization experiments, the cell surface abundance of ABCA1 is increased. The localization in the lysosomes rather than in the Golgi indicates that the compromised COPI function and sialyation of SR-BI interferes with the cycling of SR-BI between its two prevailing cellular pools in the plasma membrane and endosomes/lysosomes⁵⁵, for example by the Wiskott Aldrich Syndrome protein and scar homologue (WASH) complex⁵⁶. The loss of function of SR-BI and the resulting decrease in selective lipid uptake explains the decreased uptake of the fluorescent lipid DiI⁵⁷, but not the decreased uptake of the fluorescently labeled protein moiety analyzed by both our screening and validation experiments. Rather by contrast, knockdown of SCARB1 was found by us to promote the uptake of atto655-labeled HDL, ie. HDL holoparticles, both in our screening and targeted experiments. Similarly, in vivo and in vitro experiments in mice and primary hepatocytes with Scarb1 knock-out, as well as in hepatocyte-like cells grown from patients with a loss-of-function mutation in SCARB1, only found reduced uptake of HDL-lipids but not HDL holoparticles. We assume that the enhanced holoparticle uptake upon loss of SCARB1 observed by us is a counter-regulatory response to compensate for reduced selective uptake. However, this appears to be a cell-specific phenomenon of Huh-7 cells as it was not described for primary hepatocytes^38,43,58,59. Contrasting previous and our findings, loss of GPR146 in primary murine hepatocytes increased the cell surface abundance of SR-BI as well as HDL holoparticle uptake, possibly because GPR146 also regulates yet unknown proteins in addition to SR-BI which promote HDL holoparticle uptake⁶⁰. In this regard, it is important to note that the loss of SR-BI in endothelial cells and non-hepatic cancer cells also compromises HDL holoparticle uptake^2,43. As discussed by us in detail previously², the cell-specific role of SR-BI for HDL holoparticle uptake may depend on differential splicing or interaction with co-receptors or downstream signaling molecules.

In parallel to enhancing the cell surface abundance of ABCA1, the knockdowns of the essential COPI genes led to a modest increase in cholesterol efflux from Huh-7 cells, which did not reach statistical significance. However, altered glycosylation and cell surface abundance of ABCA1 function may contribute to the decreased HDL uptake upon loss of COPI function, as HDL uptake by LXR-agonism was lost upon knockdown of either ABCA1 or ARCN1. Principally, the increased cell surface abundance of ABCA1 upon loss of COPI subunits may result from facilitated anterograde transport of ABCA1 from Golgi or endosomal vesicles to the plasma membrane, for example by interfering with suppressors of this traffic such as serine palmitoyltransferase enzyme 1 (SPTLC1), Rab4A and Rab4B^61,62, or from decreased removal from the cell surface, for example by calpain- or ubiquitin-mediated degradation^63,64. With respect to anterograde transport it is important to note that in contrast to our findings, the knockdown of COPB1, the inhibition of ADP-ribosylation factor 1 (ARF1) with brefeldin A, and the depletion of brefeldin A- Inhibited Guanine Nucleotide-Exchange Protein (BIG1) were previously found to decrease cholesterol efflux from THP1 macrophages or HepG2 cells^65–67. However, although ARF1 is considered as a limiting factor for the formation of COPI coatomers⁶⁸, it is increasingly recognized that ARF1 also plays important roles in the regulation of exocytosis, endo-lysosomal trafficking and other coatomer-independent actions⁶⁹. In fact, in HepG2 cells, BIG1 depletion compromised the internalization and recycling of cell surface ABCA1 by interfering with ARF-dependent guanine nucleotide-exchange activity⁵³.

The decreases in HDL holoparticle and selective lipid uptake would deprive Huh-7 cells from cholesterol if not compensated by reduced cholesterol efflux and increased de novo synthesis. In fact, upon knockdown of ARCN1 and incubation with HDL with or without LXR activation, the cellular concentrations of total and unesterified cholesterol in Huh-7 cells were decreased, while the cellular concentrations of cholesteryl esters were increased. Cholesteryl esters remained increased upon loss of ARCN1 even in the presence of an ACAT inhibitor, suggesting disturbed cholesteryl ester hydrolysis rather than enhanced cholesterol ester formation in the absence of functional COPI coatomers. In view of the important role of cholesterol for the regulation of morphogenesis, it is interesting to note that de novo mutations in ARCN1 have been identified as the cause of craniofacial dysplasia and microcephaly¹⁰, i.e. similar phenotypes as observed in Smith-Lemli-Opitz syndrome⁷⁰, which is caused by defects in cholesterol biosynthesis.

By limiting HDL holoparticle uptake as well as SR-BI-mediated selective lipid uptake, the loss of COPI coatomer function is expected to increase HDL-C levels, while the decreased apoA-I secretion could counteract these effects. Our analysis of the GWAS data of more than 1.65 million persons published by GLGC¹⁹ and UK Biobank²¹ identified several common variants in ARCN1 as well as the previously reported rs7121538 polymorphism of COPB1 to be associated with higher plasma levels of HDL-C. In GLGC¹⁹, common SNPs in COPZ2 are also associated with HDL-C, although loss of COPZ2 affects neither the glycosylation of SR-BI and ABCA1, nor selective lipid uptake nor HDL particle uptake. It thus appears that COPZ2 regulates HDL metabolism by other mechanisms. In view of the metabolic interaction of HDL and triglyceride-rich lipoproteins, it is interesting to note that, in contrast to the SNPs of ARCN1, all SNPs of COPZ2, are associated with both higher levels of HDL-C and lower levels of triglycerides but not with apoA-I levels. It may therefore be that COPZ2 regulates HDL metabolism indirectly via the metabolism of triglyceride-rich lipoproteins.

In contrast to the results in these population-based genetic studies, human carriers of rare variants in COPA and COPG1, which cause inherited immunopathological syndromes, have normal or rather low plasma levels of HDL-C. We also found 15% lower levels of HDL-C in six carriers of other no immunopathological COPA or COPG1 variants compared to 86 hypercholesterolemic non-carriers of the same cohort³⁷. In line, the knock-in of the immunopathological COPA or COPG1 variants in mice led to significantly lower levels of HDL-C than in their littermates. The similar 15 to 20% mean decreases of HDL-C in humans and mice suggests that the presence and absence of CETP in humans and mice, respectively, does not grossly change the effects of the COPA and COPG1 variants on HDL metabolism. The discrepancy between higher HDL-C levels in carriers of common ARCN1 variants and lower HDL-C levels in carriers of rare COPA or COPG1 variants may be due to different effects, for example on apoA-I secretion and hence HDL production than on the catabolism of HDL by selective uptake and HDL holoparticle uptake. Moreover, disturbed COPI function may also affect other regulators of HDL metabolism which were not investigated here. In this regard, it is important to note that the rather low plasma levels of triglycerides in both humans and mice carrying variants of COPA or COPG1 which argue against any secondary lowering of HDL-C in response to hypertriglyceridemia, which is considered as the basis of low HDL-C in many individuals of the general population¹. In view of the discrepant associations of SNPs in ARCN1, COPB1 and COPZ2 with HDL-C and apoA-I levels discussed above as well as the rather indirect information of the biomarker HDL-C, it will be interesting to analyse the characteristics of HDL particles in plasmas of COPI variant carriers. SCARB1 variants compromising selective uptake as well as plasma levels of IF1, which inhibits HDL holoparticle uptake are associated with similar increases of HDL-C and apoA-I^37,38. Conversely, carriers of GPR146 variants which enhance both selective lipid and holoparticle uptake of HDL as well as carriers gain of function SCARB1 variants, present with similar decreases of HDL-C and apoA-I, due to the loss of (very) large and medium HDL-particles but no significant change in small HDL-particles⁶⁰.

In conclusion, we identified components of the COPI complex as limiting factors of important steps in HDL metabolism in hepatocytes and, probably thereby, determinants of HDL-C levels in the human population.

Supplementary Material

• Expanded Materials & Methods

• Supplemental Figures S1–S6

• Data Set (supplemental tables S1 – S5)

• Uncropped versions of Western blots

• Major Resources Table

What are the Clinical Implications?

Plasma levels of high density lipoprotein cholesterol (HDL-C) have inverse, positive, or parabolic associations with various clinical end points, including cardiovascular diseases, diabetes, chronic kidney disease, infections, specific cancer types, autoimmune and neurodegenerative diseases, and overall mortality. Towards therapeutic exploitation, it is important to understand the molecular and metabolic basis of both low and high HDL-C as well as the dynamics of HDL metabolism. By producing apoA-I, the major protein constituent of HDL, and by removing HDL-associated cholesteryl esters and HDL particles by selective uptake through scavenger receptor BI (SR-BI) and holoparticle uptake by a little understood mechanism, respectively, the liver is the major determinant of HDL-C levels. Our genome-wide siRNA screening and the subsequent validation by in vitro experiments in human hepatocarcinoma cells showed that the COPI coatomer modulates both the production of apoA-I as well as the SR-BI mediated selective lipid uptake and holoparticle uptake. Our subsequent human and mouse genetic studies suggest that the COPI coatomer contributes to the determination of HDL-C levels, however variably depending on the underlying genetic alteration. Future studies should unravel the impact of the COPI coatomer on the dynamics of HDL metabolism and its role for HDL-associated diseases.

Nonstandard Abbreviations and Acronyms

RNAi

RNA interference

ABCA1

ATP binding cassette subfamily A member

ACAT

acyl-CoA:cholesterol acyltransferase

APOA1, apoA-I

Apolipoprotein A1

ASCVD

atherosclerotic cardiovascular disease

CETP

cholesteryl ester transfer protein

COPI

Coat protein

COPA, α-COP

COPI coat complex subunit alpha

COPB1, β-COP

COPI coat complex subunit beta 1

COPB2, β’-COP

COPI coat complex subunit beta 2

ARCN1, δ-COP

COPI coat complex subunit delta

COPE, ε-COP

COPI coat complex subunit epsilon

COPG1, γ-COP

COPI coat complex subunit gamma 1

COPG2, γ’-COP

COPI coat complex subunit gamma 2

COPZ1, ζ-COP

COPI coat complex subunit zeta 1

EV

empty vector

FDR

false discovery rate

FPLC

Fast protein liquid chromatography

GC-MS

Gas chromatography mass spectrometry

GLGC

Global Lipids Genetics Consortium

GO

Gene Ontology

GTEx

Genotype-Tissue Expression

GWAS

Genome wide association study

LCAT

lecithin:cholesterol acyltransferase

LC-MS

liquid chromatography–mass spectrometry

LDL-C

Low-density lipoprotein cholesterol

LDLR

low-density lipoprotein receptor

LXR

Liver-X-receptor

MAF

minor allele frequency

NC

non-targeting control

NHANES

National Health and Nutrition Examination Survey

RSA

Redundant siRNA Activity

SCARB1, SR-BI

Scavenger receptor class B type I

siRNA

small interfering RNA

SNP

single-nucleotide polymorphism

TLC

Thin layer chromatography

TG

triglycerides