suPAR, CAC, and incident cardiovascular events
Study cohort. MESA is a multicenter observational cohort designed to identify risk factors for the incidence and progression of CVD. A detailed description of the study design and methods has been published previously (76). In summary, 6,814 (3,601 women; 3,213 men) participants aged 45 to 84 years who identified as either White, Black, Hispanic, or Chinese were enrolled between 2000 and 2002 at 6 participating communities across the US. Participants were eligible if they were free of clinical CVD at enrollment. For the present study, we included all participants who provided serum samples for suPAR biomarker measurements at enrollment (n = 5,406).
Measurement of CAC. A detailed description of the methodology for the acquisition and interpretation of CAC scores in MESA has been published previously (77). Briefly, CT scanning of the chest was performed using either electron-beam CT (Chicago, Los Angeles, and New York field centers) or using a multidetector CT system (Baltimore, Maryland, USA; Forsyth County, Georgia, USA; and St. Paul, Minnesota, USA, field centers). CAC scores were calculated using the Agatston score and adjusted with a standard calcium phantom that was scanned with the participant (78). The mean Agatston score (AU) was used in all analyses. Interobserver and intraobserver agreement were high (k statistic = 0.90 and 0.93, respectively). CAC scores were measured at baseline (exam 1: July 2002–August 2002) with initial follow-up measurements performed on half of the cohort at exam 2 (September 2002–January 2004) and the other half at exam 3 (March 2004–July 2005). A quarter of participants were selected for CAC measurement at exam 4 (September 2005–May 2007).
Measurement of suPAR. suPAR was measured using a commercially available ELISA (suPARnostic, ViroGates) in serum samples. The lower limit of detection of the assay is 100 pg/mL; however, all measurements were above the lower limit of detection. The interassay coefficient of variation determined using blinded replicate samples from participants ranged from 8% to 11%, depending on the cohort. suPAR levels are stable in stored serum samples, with levels reproducible in samples stored for over 5 years at –80°C (79).
suPAR and CAC. Clinical characteristics for the cohort are reported stratified by suPAR categories (0–2.0 ng/mL, 2.0–2.5 ng/mL, 2.5–3.0 ng/mL, and > 3.0 ng/mL). We examined the correlation between suPAR and CAC scores at baseline using Spearman’s rank. To determine whether suPAR levels (log-transformed base 2) were independently associated with CAC at baseline, we used linear regression with CAC as the dependent variable adjusted for CVD risk factors including age, sex, race, BMI, history of smoking, eGFR using the Chronic Kidney Disease Epidemiology Collaboration equation (80), LDL levels, HDL levels, C-reactive protein, hypertension (use of antihypertensives or systolic blood pressure ≥140/90 at enrollment), and diabetes mellitus. We then visualized the median CAC scores at baseline and initial follow-up stratified by suPAR categories using bar graphs. Additionally, we examined the adjusted difference in CAC scores between baseline and initial follow-up by calculating the mean predicted change in CAC score for each suPAR category accounting for age, sex, race, BMI, history of smoking, eGFR, LDL levels, HDL levels, C-reactive protein, and diabetes mellitus.
To determine whether suPAR levels at baseline were associated with an increase in CAC over time, we generated generalized estimating equations modeling with CAC as a continuous and longitudinal variable using all CAC scores measured after baseline and examined the interaction term suPAR × follow-up time. The model was adjusted for the aforementioned variables in addition to baseline CAC.
suPAR and cardiovascular events. We then assessed whether suPAR levels were predictive of CVD events. A CVD event was defined in MESA as the composite of myocardial infarction, resuscitated cardiac arrest, angina, revascularization, stroke (excluding transient ischemic attack), or death due to CVD (76, 77). We used stepwise multivariable-adjusted Cox’s proportional hazards modeling to assess the contribution of relevant factors such as eGFR and CAC to the association between suPAR and CVD events. Model 0 (suPAR alone) was unadjusted; model 1 was adjusted for age, sex, race, BMI, history of smoking, LDL, HDL, C-reactive protein, hypertension, and diabetes mellitus; model 2 included all variables in model 1 in addition to baseline eGFR; and model 3 included the variables in model 2 with the addition of baseline CAC. We explored eGFR as a time-varying covariate in a separate model including the covariates from model 3. In MESA, eGFR was measured at baseline and at exam 5 (April 2010–February 2012). suPAR was modeled as a continuous (log-transformed base 2) and categorical variable (0–2.0 ng/mL, 2.0–2.5 ng/mL, 2.5–3.0 ng/mL, and > 3.0 ng/mL) in all models. Additionally, we conducted a sensitivity analysis, further adjusting for baseline high-sensitivity troponin T and NT-proBNP in addition to the variables in model 2. Follow-up time was up to the first CVD event, death, last contact with the research team, or end of study period. Unadjusted and adjusted Kaplan-Meier cumulative incidence curves for CVD events were generated. Adjusted Kaplan-Meier curves were calculated using inverse probability weighting for suPAR categories with propensity scores estimated using generalized boost models adjusted for age, sex, race, BMI, history of smoking, eGFR, LDL levels, HDL levels, C-reactive protein, and diabetes mellitus (81). A complete case analysis was performed. A 2-sided P value of less than 0.05 was used to determine statistical significance. Analyses were performed using R, version 4.1.0, (R Foundation for Statistical Computing).
Genetic determinants of suPAR and the link to atherosclerosis
We measured plasma suPAR levels using immunoassay (suPARnostic,ViroGates) in 4 different cohorts: the Trinity Student Study (TSS) (82), the Genes and Blood-Clotting cohort (GABC) (83), MESA, and the Malmo Diet and Cancer Study (MDCS), totaling 12,937 participants (84). We performed GWAS and meta-analysis to identify genetic determinants of suPAR levels and replicated our findings in 12,177 healthy participants of the DBDS where suPAR levels were measured using the same immunoassay. The top 2 significantly associated missense variants of PLAUR were then expressed in human embryonic kidney cells (HEKs) (CRL-3216; ATCC) and in C57BL/6J mice (000664; Jackson Laboratory) to determine which variants led to significant increases in suPAR levels. We then leveraged the UK Biobank to perform MR and assess for a causal link between genetically determined suPAR levels and CVD (n = 408,894) (85).
GWAS cohorts and analysis. The TSS is a cohort of 2,179 unrelated healthy and ethnically Irish individuals between 21 and 24 years old (59% women, 41% men, all European ancestry) (82). The GABC cohort comprises 931 young and healthy students between 14 and 35 years of age (63% women, 37% men, all European ancestry) (83). The MESA cohort included 5,092 unrelated participants aged 45 to 84 years (53% women, 47% men, 38% European ancestry, 28% African American, 22% Hispanic American, and 11% Chinese American) free from CVD (76). The MDCS is a Swedish population-based cohort that included 4,735 randomly selected unrelated participants between 44 and 73 years of age (59% women, 41% men, all European ancestry) (84). Finally, the DBDS genomic cohort comprises a subset of 12,177 healthy blood donors aged 18 to 66 years (47% women, 53% men, all European ancestry) (36).
Quality control measures were performed to exclude low-quality samples and low-quality variants within each study prior to imputation to reference genomes. In general, samples were excluded if they showed discordance between genetically inferred and reported sex, low call rate, and duplications. Variants were excluded if they deviated from the Hardy-Weinberg equilibrium.
Imputation was done to predict nongenotyped variants. The TSS, GABC, and MESA were imputed using TOPMed Freeze 5b (GRCh 38). The MDCS was imputed using the Haplotype Reference Consortium reference panel (GRCh 37) (86). The build was liftover to GRCh 38 using CrossMap (87). The DBDS was imputed using 1 KG phase 3, HapMap, and a data set consisting of more than 6,000 Danish whole-genome sequences.
GWAS analyses. GWAS analyses were performed with natural log suPAR levels adjusted for age, sex, and the first 10 principal components of ancestry followed by inverse-normal transformation within each study and ancestry combination using array data imputed to reference genomes. Single-variant association analyses were performed using linear regression in PLINK, version 2.0 (88), within each study-ancestry combination. For GABC, linear mixed models incorporating a kinship matrix were performed using RVTESTS (89). Overall, our analyses resulted in genome-wide summary data from European ancestry data sets from MDC (n = 4,735), TSS (n = 2,179), MESA (n = 2,024), and GABC (n = 931), and African (n = 1,363), East Asian (n = 623) and Hispanic (n = 1,082) populations from MESA. We performed quality control measures on each of the summary association data sets prior to meta-analysis (90, 91). Within each data set, we filtered out variants with minor allele count of less than 20, Hardy-Weinberg equilibrium P value of less than 5 × 10–6, low imputation quality (INFO < 0.6), multiallelic variants, and palindromic variants (A/T or C/G) with minor allele frequency above 0.4.
Meta-analysis. We performed multi-ancestry and European ancestry–specific inverse-variance weighted fixed effects meta-analyses using METAL software (90). We generated quantile-quantile plots to assess for genomic control and structure within our data (Supplemental Figure 18). To identify leading and independent variants from each meta-analysis, we performed pruning and thresholding using the “clump” flag in PLINK. PLINK implements an iterative multistep process in which variants are sorted by their P values and those in linkage disequilibrium are removed (r2 < 0.05 and within 250 kilobases from the lead variant). The process is repeated until the genome-wide significance threshold of 5 × 10–8 is reached. The PLAUR locus was further finemapped using the SuSie Iterative Bayesian Stepwise Selection procedure (92). Top variants were defined as those with a P value of less than 5 × 10–8 and were independent of each other. We then investigated the identified variants in the DBDS cohort. Functional annotations for top variants were obtained from the Ensemble Variant Effect Predictor (91).
In vitro and in vivo expression of PLAUR missense variants. We generated the PLAUR variants rs2302524 and rs4760 (Supplemental Figure 19) using the GeneArt site-directed mutagenesis system (Thermo Scientific) and WT PLAUR (NCBI’s RefSeqGene LRG_637 and RefSeq NG_032898.1) cloned into a pCMV6-entry vector (Origene).
Equal amounts (12 μg) of plasmid DNA encoding vector control, human reference, or the PLAUR missense variants were transfected into HEK293T cells (CRL-3216; ATCC) using the FuGENE 6 transfection reagent (E2691; Promega). The conditioned media and cells from each plate were harvested 48 hours after transfection for performing the following: (a) assess uPAR distribution with immunofluorescence staining of cells using monoclonal uPAR antibody to uPAR domain 2 (NBP2-62800, 1:400, Novusbio) and membrane marker P-cadherin (ab16505; 1:100, Abcam); (b) quantification of gene expression using real-time quantitative PCR testing; and (c) suPAR measurement in the supernatant using the Human uPAR Quantikine ELISA Kit (DUP00; R&D Systems).
We performed hydrodynamic tail-vein injection of plasmid DNA encoding reference human PLAUR (n = 5), PLAUR variant rs2302524 (n = 9), and PLAUR variant rs4760 (n = 7) in 8-week-old C57BL/6J female mice and measured serum suPAR levels 24 hours after injection using the Human uPAR Quantikine ELISA Kit.
MR analysis. We leveraged the UK Biobank for MR analysis in 408,894 participants of European ancestry (UK Biobank resource, application number 59206) (93). Details of measures for variant and sample quality control have been previously reported (94). We used rs4760, the PLAUR missense variant confirmed to alter suPAR levels in both in vitro and in vivo models, as an instrument for MR analyses of 13 cardiovascular phenotypes from the UK Biobank (Supplemental Table 9). Significant associations were replicated using publicly available summary GWAS data from the CARDIoGRAM C4D consortium for coronary artery disease (60,801 cases and 123,504 controls) and the Million Veterans Program for peripheral arterial disease (31,307 cases and 211,753 controls) (37, 38). Wald ratios were used to derive the odds ratio per 1 SD increments in suPAR levels instrumented by rs4760. Similar analyses were performed using the rs2302524 missense variant as an instrument. Finally, we obtained summary-level data from the CKDGen consortium to perform MR and assessed for a causal link between genetically determined suPAR levels by rs4760 and (a) kidney function as measured by creatinine-derived eGFR (n = 567,460) (39) and (b) CKD (41,395 cases, 439,303 controls), defined as an eGFR of less than 60 ml/min/1.73 m2 (95). The MR was then replicated in the UK Biobank (eGFR, n = 387,937; CKD, 8,031 cases and 400,863 controls) (85). MR analyses were performed using the TwoSampleMR package in R, version 4.0.
To assess whether rare coding variations with damaging consequences on the suPAR protein are associated with ischemic heart disease, we performed a lookup in a previously published exome-sequenced analysis of more than 280,000 UK Biobank participants (http://azphewas.com/). Both rare protein truncating variants and rare damaging missense variants in the PLAUR gene were selected for studying the impact of attenuated PLAUR function on coronary heart disease. In brief, protein-truncating variants are defined as variants that are predicted to truncate a protein and with a maximum minor allele frequency of 0.001. Rare damaging missense variants were defined as those with a REVEL score of 0.25 or more and a maximum minor allele frequency of 0.0005 (96).
suPAR overexpression in a Pcsk9-AAV murine model of atherosclerosis
A total of 39 mice, 12 to 16 weeks of age, including n = 18 C57BL/6J WT mice (000664, Jackson Laboratory), of which 7 were female, and n = 21 suPARTg mice, of which 4 were female, overexpressing the soluble form of mouse full-length suPAR (corresponding to NP_035243, DI-DII-DIII without GPI anchor) in adipose tissue using the adipocyte fatty acid binding protein (AP2) promoter on C57BL/6 background, were used (10). All mice were maintained on a 12-hour light/12-hour dark cycle with free access to food and water.
To induce hypercholesterolemia, we administered an i.p. injection of recombinant AAV8–D377Y–murine Pcsk9 (5 × 106 viral genomes/kg body weight), which was previously described (97). After 1 week, the diet was switched to a Western diet (42% calories from fat, Teklad, catalog 88137) for 10 weeks and all 39 mice completed the study.
Cholesterol and suPAR measurements. Plasma was collected via tail-vein puncture in heparin-coated tubes. Fasting cholesterol levels were measured by colorimetric assay (STA-384; Cell Biolabs). Plasma levels of suPAR were measured using R&D DuoSet ELISA antibodies and Ancillary Reagent Kit 2 for development of a sandwich ELISA (DY531, R&D Systems). The ELISA has a detection range of 78 to 5000 pg/mL.
Atherosclerotic lesion analysis, histology, and immune histochemistry. Mice were euthanized via carbon dioxide overdose. Blood was harvested by right ventricular puncture and the vasculature perfused with ice-cold PBS. The heart and brachiocephalic artery (BCA) were harvested from all 39 mice, placed in 4% paraformaldehyde, and embedded in paraffin. Sixty sections (6 μm each) were cut through the aortic root as the primary site of atherosclerosis, and 30 sections (6 μm each) were cut through the BCA as a secondary anatomic site from each mouse, as recommended (98). For morphometric analysis, 30 sections from the aortic root and 15 sections from the BCA were stained with H&E and assessed for total lesion size and necrotic core size (acellular lesion area) as previously described (99), for a total coverage of 360 μm of the aortic root. Paraffin-embedded sections of the aortic sinus were deparaffinized and rehydrated. After blocking, sections (6 μm each) were incubated at room temperature for 2 hours with Mac2 (sc-81728; Santa Cruz Biotechnology Inc.). Mac2 slides were counterstained with hematoxylin and coverslipped. Images were captured with an Olympus LC30 camera mounted on an Olympus CX41 microscope. For the Mac2+ area, all images were obtained with the same light source at the same time. The Mac2+ area was determined using the threshold function in ImageJ (NIH) and normalized to total nonnecrotic lesion area. Results were reported as percentage of lesion area. Sectioning and staining were performed by the In Vivo Animal Core Laboratory technicians at the Unit for Laboratory Animal Medicine, University of Michigan. Technicians in this laboratory were blinded to experimental identity. Atherosclerotic plaque size was calculated using ImageJ software and graphed by section number.
Ex vivo aorta culture and CCL2 measurement. Thoracic aortas from C57BL/6J WT mice and suPARTg mice were excised and cultured in DMEM plus 10% fetal bovine serum with 1% penicillin-streptomycin solution (P4333; Sigma-Aldrich) at 37°C for 24 hours. Conditioned culture supernatants were collected and stored at –80°C. CCL2 levels in conditioned media were measured using ELISA (88-7391-22, Thermo Fisher Scientific).
Flow cytometry of aortic cell suspension and circulating cells. Approximately 50 to 100 μl whole blood was harvested via tail vein, and the red blood cells were lysed using red blood cell lysis buffer (420302, BioLegend). Cells were then centrifuged for 5 minutes at 400g, and the supernatant was poured off.
Thoracic aortas were harvested into 1× PBS on ice, then minced and digested in 1× HBSS containing 450 U/mL collagenase I (SCR103), 250 U/mL collagenase XI (C7657), 120 U/mL hyaluronidase (H3506) (Sigma-Aldrich), and 120 U/mL DNAse I (10104159001, Roche) for 45 minutes, followed by quenching with RPMI 1640 plus 10% fetal bovine serum, after which they were passed through a 70 μm cell strainer. Pellets were washed again then resuspended with LIVE/DEAD Aqua Stain (L34957; Thermo Fisher Scientific), followed by addition of FCγR block (101320; BioLegend), flow cytometry buffer (FACS buffer), 1× PBS (Ca2+ and Mg2+ free) containing 5% FBS, and 5 mM EDTA for 15 minutes, followed by addition of antibody cocktail on ice for 30 minutes.
Cells were fixed and permeabilized (554714, BD Bioscience) to stain intracellular antigens according to the manufacturer’s instructions. Cells were washed with FACS buffer 2 times and then resuspended in 200 μl FACS buffer. Flow cytometry was performed on a Bio-Rad Ze5 equipped with 405 nm, 488 nm, 561 nm, and 640 nm lasers using Everest software, version 2.
Data and compensation were analyzed with FlowJo software (FlowJo 10.8.1, BD). Antibodies used in flow cytometry were as follows: FITC anti-CD45 (103108, 10 μg/mL), PE-Cy7 anti-CD11b (101216, 5 μg/mL), BV421 anti–Ly-6C (128031, 4 μg/mL), BV605 Ly-6G (127639, 6 μg/mL), APC-Fire750 anti-CCR2 (150630, 8 μg/mL), BV785 anti-F4/80 (123141, 5 μg/mL), AF700 anti-MHCII (107622, 4μg/mL) (all from BioLegend), and PE anti-uPAR (FAB531P, Bio-Techne, 1 μg/mL).
Monocyte migration assay. The spleens of C57BL/6J WT mice and suPARTg mice were mechanically disrupted through a 70 μm cell strainer, and splenic monocytes were isolated using The Mouse Monocyte Negative Selection Kit (19861; STEMCELL). The chemotaxis ability of isolated monocytes was assessed by cell migration assay (CBA-105; Cell BioLabs) according to the product manual. Briefly, the monocyte suspension was added to the upper membrane chamber. The bottom tray contained chemoattractant solution and RPMI media with or without 1000 ng/ml CCL2 (PHC1011; Life Technologies Corp.). After 4 and 8 hours, both the cells adherent to the membrane and cells in the bottom tray were collected and stained by CyQUANT dye (C7026, Thermo Fisher Scientific). Fluorescence measurement was performed with a 485/538 nm filter set and a 530 nm cutoff.
Statistics. All results are presented as mean ± SEM. Comparisons between multiple groups were performed with Student’s t test, 1-way ANOVA, and 2-way ANOVA with Tukey’s multiple comparison test, where appropriate. A 2-tailed P value of less than 0.05 was considered significant. GraphPad Prism was used to perform statistical analysis and to generate figures.
Study approval. All participants gave written informed consent for their respective studies, and the study protocols were approved by the Institutional Review Board at each participating Clinical Coordinating Center. Animal experiments were carried out with approval of the University of Michigan Institutional Animal Care and Use Committee.