Integrative Genomic Analysis Reveals Four Protein Biomarkers for Platelet Traits

Abstract

Rationale:

Mean platelet volume (MPV) and platelet count (PLT) are platelet measures that have been linked to cardiovascular disease (CVD) and mortality risk. Identifying protein biomarkers for these measures may yield insights into CVD mechanisms.

Objective:

We aimed to identify causal protein biomarkers for MPV and PLT among 71 CVD-related plasma proteins measured in Framingham Heart Study (FHS) participants.

Methods and Results:

We conducted integrative analyses of genetic variants associated with PLT/MPV with protein quantitative trait locus variants associated with plasma proteins followed by Mendelian randomization to infer causal relations of proteins for PLT/MPV. We also tested protein-PLT/MPV association in FHS participants. Utilizing induced pluripotent stem cell-derived megakaryocyte (MK) clones that produce functional platelets, we conducted RNA-sequencing and analyzed expression differences between low- and high-platelet producing clones. We then performed small interfering RNA gene knockdown experiments targeting genes encoding proteins with putatively causal platelet effects in MK clones to examine effects on platelet production.

In protein-trait association analyses, ten proteins were associated with MPV and 31 with PLT. MR identified four putatively causal proteins for MPV and four for PLT. Glycoprotein V (GP5), granulin (GRN), and melanoma cell adhesion molecule (MCAM) were associated with PLT, while myeloperoxidase (MPO) showed significant association with MPV in both analyses. RNA-sequencing analysis results were directionally concordant with observed and MR-inferred associations for GP5, GRN, and MCAM. In siRNA gene knockdown experiments, silencing GP5, GRN, and MPO decreased platelet counts. Genome-wide association study results suggest several of these may be linked to CVD risk.

Conclusions:

We identified four proteins that are causally linked to platelet counts. These proteins may also have roles in the pathogenesis of CVD via a platelet/blood coagulation-based mechanism.

Graphical Abstract

INTRODUCTION

Platelets are circulating anucleate cells produced by megakaryocytes (MK) in the bone marrow¹ that initiate thrombus formation at sites of injury². They also function as innate and adaptive immunity regulators through secretion of antibacterial proteins, interactions with leukocytes, and mediation of pro-inflammatory functions of neutrophils and dendritic cells³. Their central roles in thrombus formation and immune response make platelets notable contributors to the pathobiology of atherosclerosis⁴ and arterial and venous thrombosis⁵, including venous thromboembolism (VTE)⁶. When the endothelium of a vessel is damaged, platelets aggregate at the site of injury and can promote atherosclerosis and atherothrombosis⁵, which in turn can cause acute coronary syndromes and strokes⁷. A commonly measured platelet metric from blood count panels, mean platelet volume (MPV)⁸, has been reported to be associated with risk of acute myocardial infarction (AMI) events⁹, coronary artery disease (CAD)¹⁰, and VTE⁶. Prior studies have also shown that higher MPV is associated with greater mortality risk among AMI patients⁹ and patients with sepsis¹¹. Furthermore, greater MPV is also correlated with poor prognosis for patients with infective endocarditis¹² and risk of pulmonary embolism among patients with deep vein thrombosis¹³. Younger, reticulated platelets with higher MPV are suggested to be a more reactive platelet subpopulation¹⁴, and may act against the effects of antiplatelet medications¹⁵. A prospective cohort study by Mayer and colleagues demonstrated that elevated MPV levels were associated with increased risk of major adverse cardiovascular events among patients with asymptomatic carotid atherosclerosis¹⁶. Platelet count (PLT), on the other hand, has a U-shaped association with mortality; high- and low- platelet counts have been reported to be associated with increased risk of coronary heart disease (CHD) and cancer mortality¹⁷ and with overall mortality in people older than 65 years of age^{18, 19}. Therefore, given the clinical importance of platelets vis-à-vis cardiovascular diseases (CVD), identifying putatively causal protein biomarkers associated with MPV and PLT may lead to a better understanding of disease mechanisms and highlight therapeutic targets for the treatment and prevention of CVD.

As part of the Systems Approach to Biomarker Research in Cardiovascular Disease (SABRe CVD) Initiative, we conducted a genome-wide association study (GWAS) to identify genetic loci associated with circulating levels of 71 CVD-related plasma proteins (pQTLs; protein quantitative trait loci) in 7333 Framingham Heart Study (FHS) participants²⁰. In the present study we sought to identify associations between circulating protein levels and platelet phenotypes.

Identifying overlaps between genetic variants associated with circulating protein levels and single nucleotide polymorphisms (SNPs) associated with complex traits from previously published genome-wide association studies (GWAS) can prioritize candidate protein biomarkers that can be further explored as causal proteins for disease and/or drug targets of complex disease phenotypes. We previously demonstrated the utility of this technique by identifying putatively causal protein biomarkers for CVD²⁰ and chronic obstructive pulmonary disease²¹. Because PLT and MPV are highly heritable quantitative traits²² with large published GWAS and exome studies^23–25, we hypothesized that these traits may be amenable to a similar approach to identify causal protein biomarkers. To that end, we conducted Mendelian randomization (MR) analyses to test for putatively causal associations between protein biomarkers and MPV and PLT. Proteins that were associated with MPV or PLT in both FHS data and MR were functionally validated using RNA-sequencing data of human induced pluripotent stem cells (iPSC) MK clones followed by siRNA gene knockdown experiments.

METHODS

Data Availability.

The blood cell count data, protein levels and covariates from FHS are available in the Database of Genotypes and Phenotypes (dbGaP) (https://www.ncbi.nlm.nih.gov/gap/). The GWAS and exome chip summary statistic data are available in the GRASP database (https://grasp.nhlbi.nih.gov/FullResults.aspx). The pQTL data is available in the Supplement of its original publication. The RNAseq and cellular experiment are available by request to the authors.

Study Design.

An overall flowchart of our study design is shown in Figure 1. We first conducted protein-trait association analyses using FHS data²⁶ for MPV and PLT. We then performed MR using cis-pQTL variants²⁰ that overlapped with prior MPV/ PLT GWAS^{23, 24} SNPs as instrumental variables to test for putatively causal protein-MPV/PLT associations. Proteins showing significant causal association were further investigated for transcriptome-level association with platelet production using RNA sequencing data from human iPSC MK clones. Finally, we performed gene knockdown experiments in iPSC MK cells to assess if gene silencing altered platelet production.

Figure 1.

Flowchart of Study Design

Protein-Trait Association Analysis.

Study participants:

The study sample consisted of FHS Offspring and Third Generation cohort participants. The FHS Offspring study was comprised of 5124 children or spouses of original FHS cohort participants; enrollment began in 1971^{27, 28}. The Third Generation cohort included 4095 children of Offspring cohort participants who began enrollment in 2002²⁹. Among the Third Generation cohort participants, 3411 attended Examination 2 (2008–2011) and provided blood samples for the measurements of PLT and MPV. Among the Offspring cohort participants, 2430 attended Examination 9 (2011–2014) and provided information on cardiovascular health and blood samples for laboratory tests including PLT and MPV. Of these 5841 Offspring and Third Generation participants with platelet measurements, 5233 provided informed consent and participated in the SABRe CVD Initiative²⁶ for which plasma protein measurements were measured. Plasma proteins were measured in plasma samples obtained at Examination 7 (1998–2001) for the Offspring cohort participants and at Examination 1 (2002–2005) for the Third-Generation cohort participants. Among these 5233 participants, complete information on PLT, MPV, and normalized plasma protein concentration was available in 4349 participants including 2026 Offspring (46.6%) and 2323 Third Generation participants (53.4%). Characteristics of 4348 participants with MPV measurements and 4272 participants with PLT measurements are summarized in Table 1.

Table 1.

Characteristics of FHS Participants with Platelet Measurements

	MPV (N = 4,348)		PLT (N = 4,272)
Characteristic	Mean	SD	Mean	SD
Age (year)	58.1	14.5	57.9	14.4
BMI (kg/m2)	28.3	5.6	28.3	5.6
Mean Platelet Volume (fL)	8.7	0.96	−	−
Platelet Count (103 cells/μL)	−	−	240.7	58.7
Systolic BP (mmHg)	121.5	16.0	121.4	15.9
Diastolic BP (mmHg)	73.3	9.5	73.4	9.5
Total-cholesterol (mg/dL)	185.5	36.1	185.8	36.0
HDL-cholesterol (mg/dL)	60.7	18.6	60.9	18.7
Triglyceride (mg/dL)	115.2	70.0	115.0	70.1
	N	%	N	%
Women	2276	52.4	2256	52.8
Cigarette Smoker	361	8.3	357	8.4
Diabetic	329	7.7	321	7.6
Hypertension Treatment	1560	36.0	1513	35.6
Hyperlipidemia Treatment	1425	32.8	1383	32.4
Anticoagulant Use	307	7.1	294	6.9
Prevalent CVD	392	9.0	371	8.7

Hematological measures.

Blood PLT and MPV were measured using the Coulter HmX hematology analyzer (Beckman Coulter, Inc.) on fasting blood samples⁸.

Protein quantification.

All plasma proteins were measured as part of the SABRe CVD Initiative²⁶. The protein quantification procedure has been described previously²⁰. Briefly, fasting blood plasma samples, which included platelet proteins, were analyzed using a modified enzyme-linked immunosorbent assay sandwich method, multiplexed on Luminex xMAP platform (Luminex, Inc., Austin, TX), for plasma protein levels²⁰.

Statistical methods:

We applied linear mixed effects models (LME) to characterize significant associations between 71 plasma proteins and platelet measurements (MPV and PLT). We adjusted for sex, age at the time of platelet measurement, age squared, and ten genotype-based principal components, and derived the residuals. The residuals were inverse normalized and tested for association with the 71 plasma proteins using LME to adjust for family structure. Bonferroni corrected significance threshold was used to identify significant biomarkers associated with platelet measurements.

Causal Genetic Polymorphisms

Identifying overlap of pQTL variants with GWAS variants:

We interrogated our pQTL database²⁰ from 6861 participants with 1000 Genomes Project reference panel (1000G build 37 phase 1 v3)³⁰-based imputed dosage data for overlap of cis-pQTL, or a single nucleotide polymorphism (SNP) located within 1 megabase (Mb) range (both upstream and downstream) of the transcription start site of the protein-coding gene²⁰, with SNPs from prior genome-wide association studies (GWAS) of MPV and PLT^{23, 24}. Only cis-pQTLs that passed Bonferroni-corrected (BF) significance threshold at P < 1.25E-07 were considered for overlap analysis²⁰.

MR testing for causal protein-trait association:

MR was conducted using the TwoSampleMR³¹ R package using pruned cis-pQTL variants (linkage disequilibrium [LD] threshold of r² < 0.1) from FHS²⁰ that overlapped at SNP level with published MPV/PLT GWAS results^{23, 24} as instrumental variables. The Wald ratio, which is a ratio of the regression coefficient of the exposure on the instrumental variable to that of the outcome on the instrumental variable³², was used for proteins with one independent cis-pQTL, and inverse-variance weighted regression was used for proteins with multiple independent cis-pQTL variants³³.

RNA Sequencing Analyses.

Generation of human iPSC-derived immortalized megakaryocytic cell line (imMKCL) subclones:

For RNA sequencing, we generated single-cell subclones derived from a single phenotypically heterogeneous imMKCL that were in turn, derived from a single DN-SeV2 iPSC cell line. DNSeV-2 iPSC lines was produced using a Sendai viral vector with neonatal fibroblasts from healthy donors (from M. Nakanishi, National institute of advanced Industrial Science and Technology, Tsukuba, Japan) by Nakamura et al³⁴. The gender of the donor is male. The 3 biological replicates for each of the 8 clones that were used for RNA-Seq analyses were collected by variously passaging the 8 subclones. Passaging refers to 1 doubling cycle of the cells requiring changing of the media and resuspending the cells, which typically takes 2 days. The cells for replicate 1 were collected at passage 5 (labelled as early passage) from various clones, cells for replicate 2 and replicate 3 were collected at passages 16 (labelled as intermediate passage) and 26 (labelled as high passage).

Expansion of imMKCLs:

ImMKCLs were cultured and expanded via doxycycline dependent expression of C-MYC, BMI-1 and BCL-XL³⁴. ImMKCLs were maintained in a humidified incubator at 37°C and 5% CO2 and in IMDM (Sigma-Aldrich) medium supplemented with 15% fetal bovine serum (FBS; Sigma-Aldrich), L-glutamine (Gibco), Insulin-transferrin-selenium (Gibco), 50 mg/mL Ascorbic acid (A4544; Sigma-Aldrich), and 450 mM 1-Thioglycerol (Sigma-Aldrich), 50 ng/mL carrier-free recombinant human stem cell factor (SCF; R&D Systems), 50 ng/mL TPO (R&D Systems), and 5 mg/mL Doxycycline (Clontech).

Differentiation of imMKCLs to generate platelets:

Induction of imMKCL maturation was carried out in a humidified incubator at 37C and 5% CO2 with 6 days in IMDM medium supplemented with 15% FBS consisted of 50 ng/mL SCF, 50 ng/mL TPO, 15 mM KP-457 (Medchemexpress), 0.5 mM GNF-351 (TOCRIS), and 10 mM Y27632 (Medchemexpress). On day 3, imMKCLs were highly polyploid and poised for platelet generation, and platelet generation occurred between day 5 and 6. Since heterogenous (Het) imMKCLs consisted of both non-platelet producing and platelet producing megakaryocytes, the imMKCLs were subcloned to decrease the ratio of non-platelet producing and platelet producing megakaryocytes. Subcloning of imMKCLs resulted in derivation of functionally distinct subclones. The 7 subclones varied in their ability to generate platelets and were labeled as Clone 42, Clone JC, Clone 32, Clone DKO, Clone 38, Clone 62, and Clone 72. To identify the genes and signaling networks that are necessary for platelet generation, we performed transcriptomic analyses of the 8 clones (Het and 7 subclones). Three biological replicates for each of the 8 clones were created, and maturing imMKCLs were harvested on day 3 for RNA extraction and subsequent platelet output measurements of the respective subclones were performed on day 6 via flow cytometry.

RNA sequencing analysis and derived platelet counting from human iPSC-derived immortalized megakaryocytic cell line (imMKCL) subclones:

Total RNA was extracted using miRNeasy Mini Kit (QIAGEN) according to the manufacturer’s instructions. RNA-seq libraries were prepared using the NEB Ultra (PolyA) kit as per manufacturer’s protocol with 50 ng input RNA. RNA-seq libraries were prepared and sequenced using the 200-cycle paired-end kit on the Illumina HiSeq2500 system. RNA-seq reads were analyzed with the Tuxedo Tools following a standard protocol. Reads were mapped with TopHat version 2.1.0 and Bowtie2 version 2.2.4 with default parameters against build hg19 of the human genome and the corresponding RefSeq human genome annotation. Then, association analysis between gene transcript levels (FPKM) and platelet output (number of platelets per megakaryocyte) was performed via LME models to account for the eight clone types.

siRNA Gene Knockdown Experiments.

imMKCL culture and knockdown:

imMKCLs were generated from iPSCs and maintained in presence of 5 μg/mL doxycycline (DOX) as previously described³⁴. Removal of DOX resulted in imMKCL maturation and the generation of platelets after day 6.

siRNA transfection assay:

The imMKCLs maturing in the differentiation medium (without DOX) on day 3 were seeded in a 24-well plate, 24 hours prior to transfection. The imMKCLs were transfected with either 2ug esiRNA (mixture of siRNA oligos (Sigma-Aldrich)) specific to MCAM, MPO, GRN, and GP5. eGFP and LUC siRNA oligonucleotides were used as scramble control. HDAC7 and PRMT7 siRNA oligonucleotides were used as negative controls using Lipofectamine 2000 (Thermo Fisher Scientific) according to the manufacturer’s instructions. 48 hours post-transfection (i.e. on day 6 in the absence of DOX), the cells were harvested for RT-qPCR and/or in vitro flow cytometry-based platelets counting. Experiments were done at different times for GRN and MPO versus GP5 and MCAM due to oligo availability. To adjust for variation between experiments, we divided each experimental groups’ platelet counts by the mean of GFP knock-down (control) group’s platelet counts to calculate fold-change in platelet counts. We used Shapiro-Wilk test to check for normal distribution of data and performed appropriate statistical test (two sample t-test or Wilcoxon rank-sum test) to compare platelet counts between control (GFP) and experimental iPSC MK clone groups.

RNA extraction, reverse transcription, and RT-qPCR:

RNA extraction was performed using an RNAeasy kit (Qiagen). Reverse transcription was performed using Superscript III (Invitrogen), using Oligo (dT) 15 primer, according to manufacturer’s guidelines (SuperScript™ III First-Strand Synthesis System; 18080051). Quantitative RT-qPCR was performed in triplicate with SYBR Green and CFX96 real-time PCR detection system (Bio-Rad). The RT-qPCR conditions are as follows: Step 1) Heat solution to 95°C for 30 seconds, Step 2) heat solution 95°C for 5 seconds, Step 3) heat solution to 60°C for 5 seconds, go back to Step 2 and repeat Step 2 and Step 3 forty times. Target transcript abundance was calculated relative to GAPDH (reference gene) using the 2-ΔΔCT method. Gene specific primer pairs are present in Online Table I.

Flow cytometric analysis of platelet cell markers:

Flow cytometry-based platelet counting was performed on MCAM, MPO, GRN, GP5, GFP, LUC, HDAC7, and PRMT7 knockdown experiments on day 6 differentiating imMKCL cultures growing in differentiation medium in the absence of DOX. Briefly, 200μL cell aliquots were incubated for 20 min with fluorescently labeled monoclonal antibodies and dyes. The antibodies used were: phycoerythrin (PE)-mouse anti-human anti–CD41a antibody (BD Pharmingen; Cat# 555467), fluorescein isothiocyanate (FITC)- mouse anti-human Annexin-V (biolegend; Cat# 640945) to identify non-apoptotic platelet sized events, and APC-anti-human anti-CD42b (GPIb) monoclonal antibody (BD Pharmingen; Cat# 551061). The Calcein AM blue was used to identify live platelet sized events. Stains were fixed in 0.5% paraformaldehyde solution. Data were collected on a LSRII flow cytometer (BD Biosciences) and analyzed with FlowJo software (FlowJo LLC). Calcein AM blue+ AnnexinV- CD41+CD42b+ events were determined as platelets. As shown in Online Figure I, platelet sized events were identified and segregated from larger imMKCL events via SSC-A vs FSC-A FACS plots. Doublet exclusion was performed on the smaller platelet sized events via sequential FSC-H vs FSC-A and then by SSC-W vs SSC-H. The live (Calcein AM blue positive) singlet platelet sized events were identified followed by identification of non-apoptotic (Annexin-V negative) among live singlet platelet sized events. Lastly, CD41+CD42b+ (double positive) events were identified among live and non-apoptotic singlet platelet sized events. Thus, singlet, live, non-apoptotic CD41+CD42b+ events were labelled as platelets and were counted.

RESULTS

FHS protein-MPV/PLT association analyses.

Among the 71 CVD-related plasma proteins, ten were associated with MPV and 31 were associated with PLT after Bonferroni (BF) correction (P < 0.05/71) (Table 2 and Online Table II). Seven proteins showed significant association with both PLT and MPV: adrenomedullin (ADM), cadherin-13 (CDH13), glyceraldehyde 3-phosphate dehydrogenase (GAPDH), translocation associated notch homolog (Notch1), pro-platelet basic protein (PPBP), soluble CD40 ligand (sCD40L), and stromal cell-derived factor 1 (SDF1).

Table 2.

Protein-MPV/PLT Association Analysis Results from FHS Data

Trait	Protein*	β‡	SE	Z-test p-value§
MPV	sCD40L†	−0.108	0.014	5.81E-14
	SDF1†	−0.108	0.015	2.47E-13
	PPBP†	−0.087	0.015	2.68E-09
	NOTCH1†	0.066	0.015	7.71E-06
	ADM†	0.066	0.015	9.74E-06
	GAPDH†	−0.064	0.014	9.74E-06
	MMP9	0.061	0.015	3.09E-05
	RETN	0.063	0.015	4.71E-05
	MPO	0.058	0.016	2.06E-04
	CDH13†	−0.050	0.015	5.63E-04
PLT	PPBP†	0.217	0.0150	2.80E-47
	sCD40L†	0.183	0.0150	2.99E-34
	PAI1	0.157	0.0151	2.09E-25
	GMP140	0.152	0.0155	9.78E-23
	sRAGE	−0.136	0.0158	8.42E-18
	GP5	0.128	0.0155	1.31E-16
	CNTN1	−0.122	0.0155	2.96E-15
	CDH13†	0.118	0.0152	6.95E-15
	LDLR	0.118	0.0152	9.69E-15
	GAPDH†	0.101	0.0152	2.96E-11
	NRCAM	0.0875	0.0150	5.87E-09
	PMP2	0.0878	0.0152	8.24E-09
	NOTCH1†	−0.0882	0.0155	1.17E-08
	SDF1†	0.0848	0.0156	5.33E-08
	KLKB1	0.0849	0.0157	6.72E-08
	CRP	0.0813	0.0151	6.82E-08
	ADAM15	0.0781	0.0151	2.40E-07
	CD56 (NCAM)	−0.0779	0.0156	6.50E-07
	ANGPTL3	0.0743	0.0152	1.09E-06
	HPX	0.0742	0.0153	1.29E-06
	ApoB	0.0731	0.0154	2.09E-06
	C2	0.0731	0.0154	2.19E-06
	AGP1	0.0713	0.0152	2.84E-06
	MCAM	−0.0715	0.0156	4.37E-06
	CD5L	−0.0640	0.0160	6.44E-05
	ADM†	−0.0615	0.0155	7.51E-05
	BGLAP	−0.0613	0.0155	7.65E-05
	FGG	0.0617	0.0158	9.30E-05
	BCHE	0.0592	0.0156	1.53E-04
	NTproBNP	−0.0591	0.0157	1.70E-04
	A1M	0.0589	0.0161	2.50E-04

^*A full list of proteins showing nominal significance (p < 0.05) is in Online Table II

^†Proteins showing significant associations with both MPV and PLT

^‡Unit: per standard deviation increment in rank-based inverse normal transformed protein level

^§Bonferroni-corrected significance threshold for 71 proteins P = 0.05/71 = 7.04E-04

MR analyses.

We selected 37 proteins with cis-pQTLs that overlapped with MPV- and PLT-GWAS SNPs for MR analysis (Online Tables III and IV). MR analysis revealed four proteins that were causally associated with MPV after BF correction (P < 0.05/37) (Table 3 and Online Table V). Similarly, four proteins were causally associated with PLT after BF correction (Table 3 and Online Table V). SERPINA10, a protein showing significant association with PLT in MR, was tested for horizontal pleiotropy and heterogeneity because it had 28 SNP IVs (Online Table VI); there was no evidence of horizontal pleiotropy (Egger regression intercept: 0.005, P = 0.14), but a forest plot of individual SNP MR effect size and a test based on the Egger method confirmed heterogeneity (P = 0.022) (Online Figures II–III). Removal of seven SERPINA10 cis-pQTL SNPs with negative MR coefficients resulted in a modest improvement of the effect size and significance (IVW MR β: 0.019, SE: 0.0036, P = 1.69E-07) (Online Figure IV). Granulin (GRN) and platelet glycoprotein V (GP5) were causally associated with both MPV and PLT (Table 3).

Table 3.

MR Results for MPV and PLT

Trait	Protein	Number of SNPs	β†	SE	p-value‡
MPV	GP5*	1	−0.112	0.0250	7.57E-06
	MPO	6	0.0407	0.0102	6.57E-05
	CD5L	4	−0.0350	0.00934	1.76E-04
	GRN*	2	−0.183	0.0522	4.63E-04
PLT	GRN*	2	0.148	0.0146	4.38E-24
	MCAM	1	−0.259	0.0289	3.09E-19
	GP5*	1	0.0988	0.0252	9.05E-05
	SERPINA10	28	0.0141	0.00369	1.31E-04

^*Proteins showing significant causal association with both MPV and PLT.

^†Units: per standard deviation increment of rank-based inverse normal transformed protein level

^‡Bonferroni Corrected Significance Threshold P = 0.05/37 = 1.35E-03

Myeloperoxidase (MPO) was positively associated with MPV in both MR and protein-trait association analyses after BF correction, with directional consistency of effect estimates between the two analyses (Figure 2). Melanoma cell adhesion molecule (MCAM/CD146/MUC18) and glycoprotein V (GP5) were associated with PLT in both MR and protein-trait association analyses with directional consistency (Figure 2). GRN was BF significant in MR but nominally significant in protein-PLT analysis with directional consistency. We included GRN in the list of proteins for subsequent analyses because it was the most significant protein associated with PLT in MR.

Figure 2. Comparison of the Effect Estimates of MR and FHS Protein-Trait Association Analyses.

MPO and NTProBNP were associated with MPV (top), while ADM, CNTN1, CRP, GP5, GRN, and MCAM were associated with PLT (bottom). Proteins with P < 0.05 in both protein-trait and MR analyses were included in the plot. Units of effect estimates: per standard deviation increment of rank-based inverse normal transformed protein level. ggplot2 package in R version 3.6.2 was used to create the plot.

Instrumental variable SNPs for GP5, GRN, MCAM, and MPO (Online Table VI), the genes carried forward in experiments, were queried for CHD outcome associations in CARDIoGRAMplusC4D³⁵, UK BioBank GWAS³⁶, the MEGASTROKE Consortium³⁷, and the Japan BioBank³⁸ (Online Table VII). With the exception of the association between MPO and hypertension in UK BioBank GWAS (P = 5.75E-09), all outcomes were nominally significant (P < 0.05) but did not pass the genome-wide significance (P < 5E-08). We focused on checking directional consistency between our protein-MPV/PLT associations and protein-CHD outcomes in public database. Gene expression levels of GP5, GRN, MCAM and MPO were assessed in publicly available data from RNA-sequencing in tissues believed to have most relevance for megakaryocyte and platelet biology and are summarized in Online Table VIII: namely blood cell subtypes including platelets³⁹ and whole blood, bone marrow, intestine, arterial tissue, heart, liver, lung and spleen from Protein Atlas version 19.1^{40, 41}, GTeX version 8, and Eicher et al. 2016³⁹.

RNA sequence analysis of iPSC MK.

To validate our findings from protein-trait association and MR analyses, we utilized human iPSC – derived MK clones that produced variable numbers of functional platelets to conduct a MK RNA-sequencing and analyzed expression differences between low- and high-platelet producing clones. The expression levels of PLT-associated proteins in the iPSC MK clones showed directionally consistent associations with platelet productivity that we expected based on our human population genetics analyses (Table 4, Online Figures V–VII).

Table 4.

Associations Between Transcription Level and Platelet Productivity in iPSC MK Clones

Gene	β*	SE	Z-test p-value†
GP5	0.00164	0.000338	0.00111
GRN	0.0155	0.00223	7.42E-05
MCAM	−0.170	0.0315	4.34E-04

^*Units: platelets per megakaryocyte per FPKM

^†Bonferroni Corrected Significance Threshold P = 0.05/3 = 0.0167

siRNA gene knockdown experiments using iPSC MK.

Motivated by our findings from the RNA sequencing analysis, we performed a siRNA gene knockdown experiment using the iPSC MK clones. Silencing GRN, GP5, and MPO genes in MK clones caused substantial reductions in platelet production (Figure 3, Online Table IX, Online Figure VIII). In order to show that the platelet productivity of MK clones changed due to gene silencing rather than the experimental procedure, we conducted another siRNA experiment with additional negative controls. A scramble siRNA (siLUC), 2 genes (HDAC7, PRMT7) that showed no significant correlation with platelet count in RNA-sequence analysis (Online Figures IX and X), and GFP, respectively, did not significantly alter the platelet productivity (Online Figure XI), further validating the result of our siRNA experiment. Knockdown of MCAM did not result in a statistically significant change in platelet production compared to that of the control (GFP). The gene knockdown experiments provided further support for causal association between GP5 and GRN and PLT. In addition, the experiments indicated that MPO may also be causally associated with PLT.

Figure 3. iPSC Megakaryocyte gene knockdown experiment results.

Silencing GRN, GP5 and MPO significantly decreased platelet count, but silencing MCAM did not result in a significant change in platelet count. Two-sample t-test was used to compare the control (GFP) with GP5 and MCAM groups. Wilcoxon Rank-sum test was used to compare the control with GRN and MPO groups because GRN and MPO group data were not normally distributed (Wilcoxon rank-sum test p-values are marked with an asterisk). Bonferroni-corrected p-value threshold for significance is 0.05/7 = 0.0071 because GFP group value was compared with 7 other groups (LUC, PRMT7, HDAC7, GRN, GP5, MPO, MCAM). All statistical tests were performed using R version 3.6.2. ggplot2 package in R version 3.6.2 was used to create the plot.

DISCUSSION

There have been several large genetic studies of PLT and MPV, epidemiological studies of clinical correlates of PLT and MPV^{23, 24}, and studies of plasma pQTLs^{26, 42, 43}. To our knowledge, this is the first study to integrate plasma protein measurements, pQTL associations, and genetic studies using MR to identify putatively causal biomarkers of PLT and MPV followed by iPSC MK clone gene silencing experiments.

Three proteins (GP5, GRN, MCAM) were associated with PLT in FHS protein-trait, MR, as well as iPSC MK clone gene expression analyses (Table 2–4). The causal roles of GP5 and GRN on PLT were further supported by iPSC MK gene knockdown experiments (Figure 3). GP5 is a transmembrane glycoprotein expressed on the surface of platelets. It has a soluble extracellular domain, which is cleaved by thrombin during thrombin-induced platelet activation⁴⁴. Mutations in GP5 are not known to cause Bernard-Soulier syndrome (enlarged platelets associated with bleeding and thrombocytopenia) unlike other gene members of the GPIb-V-IX complex. However, prior and recent work has suggested that alloantibodies targeting GP5⁴⁵, as well as pediatric varicella⁴⁶, gold-triggered autoimmune responses in rheumatoid arthritis⁴⁷, and quinidine-related platelet directed antibodies⁴⁸ may trigger immune thrombocytopenia via GP5. These studies suggest that GP5 alterations may trigger platelet clearance, and that maintenance of normal GP5 function or high levels on platelets could preserve platelet numbers in circulation. We observed consistent association between higher levels of plasma GP5 and PLT in our analyses. When we knocked down GP5 in the iPSC MK system, we observed reduced platelet production.

Granulin (GRN/epithelin) is a group of protein growth factors with homologous protein domain structures^{49, 50}. Granulin’s precursor, progranulin (PGRN), consists of a paragranulin domain and seven GRN domains that can be cleaved and released by proteases including MMP9, MMP12, and ADAMTS-7⁵¹. PGRN and GRN have opposite effects on cells; GRN B promotes secretion of interleukin 8 (IL-8) by epithelial cells, but PGRN does not cause such response⁴⁹. IL-8 has been shown to hyper-activate platelets and induce procoagulant behavior⁵². The opposing effects of GRN/PGRN on IL-8 release was also observed in human aortic smooth muscle cells⁵³. PGRN appears to be anti-inflammatory^{53, 54} while GRN shows pro-inflammatory behavior⁴⁹. PGRN has previously been implicated in blood lipid effects on CVD through its connection to sortilin (SORT1). Sortilin binds to PGRN and is responsible for endocytosis and transportation of PGRN to lysosomes⁵⁵. A strong trans-pQTL for PGRN (rs646776), located on chromosome 1p13, was associated with CELSR2 (cadherin EGF LAG seven-pass G-type receptor 2), PSRC1 and SORT1 (sortilin) mRNA expression⁵⁶. SNP rs646776 was also associated with blood low-density lipoprotein (LDL) cholesterol levels and myocardial infarction risk⁵⁷. GRN was associated with an increased risk of CVD death in a previous study²⁰. Our MR analysis results using two cis-pQTLs of GRN (rs35203463, rs850733) and the siRNA experiment showed that silencing GRN expression leads to a decreased platelet count, suggesting a novel pathway potentially linking GRN and CVD through platelet effects. Gene expression levels of GRN are found widely across blood cell types and tissues (Online Table VIII), though there appears to be moderate enrichment in megakaryocytes and platelets, and MK/platelet-related tissues (e.g., bone marrow, spleen, lung), suggesting GRN may have functional roles influencing CVD in multiple cell types including platelets.

MPO is a heme-enzyme and a bactericidal protein found in neutrophils, specifically in azurophilic granules⁵⁸. In addition to its role in immune response to pathogens, MPO has been suggested to play a role in inflammatory conditions and CVD⁵⁸. MPO shows nominally significant, positive associations with coronary heart disease (CHD) events risk and CVD death in FHS data, and a positive association with CHD risk in MR²⁰. In queries of publicly available GWAS data, we found consistent associations of MPO-increasing alleles with CHD outcomes (Online Table VII). MPO partially activates platelets, enhances their actin cytoskeletal rearrangement resulting in increased cytoplasmic calcium levels, and increases platelets’ surface expression of P-selectin and PECAM-1 receptors as well as the frequency of aggregate formation with neutrophil granulocytes^{59, 60}. In our study, MPO was positively associated with MPV in both FHS protein-trait and MR analyses after correction for multiple testing (Tables 2 and 3). MPO also showed a positive, nominally significant (P = 0.03), association with PLT in FHS data (Online Table II); furthermore, MPO knockdown MK clones showed a statistically significant decrease in MK derived platelet count, demonstrating a positive correlation between the protein and platelet production (Figure 3). Gene expression analysis shows MPO expression levels highest in bone marrow cells, with lower RNA levels in megakaryocytes, platelets, spleen, lung and other platelet-related tissues, suggesting the effects of MPO may be mediated in early stages of platelet production, or via effects at the post-translational level (Online Table VIII). MPO’s positive correlation with both MPV and PLT is not concordant with the inverse relationship between MPV and PLT shown in prior literature⁶¹ and in our protein-trait association analysis.

Melanoma cell adhesion molecule (MCAM/CD146/MUC18) is expressed on the cell surface and contributes to binding of a cell to other cells or to the extracellular matrix⁶². MCAM plays a crucial role in macrophage foam cell formation and retention of foam cells in atherosclerotic plaque⁶³. Further, MCAM expression is positively correlated with necrotic core area as well as development of unstable plaques⁶⁴. MCAM was negatively correlated with CHD risk in a prior MR study²⁰. The MCAM instrumental variable SNP (rs11217234) in our study was nominally associated with (P = 0.027) with decreased risk of a second myocardial infarction event in the UK BioBank (Online Table VII). MCAM is an important contributor to development of atherosclerosis; however, its relationship to PLT or MPV has not been studied extensively. In our study, MCAM was causally associated with decreased PLT in MR, but this inferred causality was not supported by the iPSC MK gene silencing experiment (Figure 3). Gene expression analysis suggests a potential reason for this; expression levels of MCAM are highest among endothelial cell beds (Online Table VIII). Thus, circulating or endothelial MCAM may influence platelet through effects in circulation on platelet turnover.

Our study has several limitations. The first limitation is that MPV and PLT are not perfect surrogates of platelet function in relation to CVD. Second, MPV and its relationship to CVD may be confounded in several ways. It displays a strong association with diabetes status⁶⁵, and several studies have suggested that MPV may be influenced by medications including statins, diuretics and anti-diabetic treatments including sulfonylureas, insulin and metformin^65–67. The third limitation is the time gap between plasma protein and platelet measurements in FHS participants. FHS Offspring cohort participants’ protein measurement (Exam 7) and platelet measurement (Exam 9) were about 13 years apart and FHS Third Generation cohort participants’ protein measurement (Exam 1) and platelet measurement (Exam 2) were about 6 years apart. These time differences can weaken the association between protein levels and platelet traits. Fourth, the plasma samples and MPV/PLT measurements were taken at only one time point, which might not account for variability of MPV and PLT that can arise from measurement process²². Fifth, FHS Offspring and Third Generation cohorts consist of people of European descent, potentially limiting the generalizability of study findings. The sixth limitation is that the proteins measured in FHS were from plasma, which do not fully represent the protein expression levels in platelets. Seventh, our study focused on 71 CVD-related plasma proteins as candidate biomarkers, which might not have included other platelet-associated biomarkers. Lastly, although we adjusted for multiple testing at each section, we did not adjust for multiple testing across the entire manuscript.

In this study, we combined pQTL data of 71 CVD-associated proteins, platelet measurements, and published platelet GWAS to identify 4 proteins causally associated with MPV and/or PLT. MPO was causally associated with MPV and three proteins (GP5, GRN, and MCAM) were causally associated with PLT in population genomics analyses. GP5 is a well-known receptor in platelets, but MCAM and GRN are not well-characterized platelet proteins. The results of iPSC-derived MK RNA-sequencing analysis were consistent with our population-based causal associations; GRN and GP5 were over-expressed, and MCAM down-regulated, respectively, in high-platelet producing clones. Gene knockdown experiments confirmed the causal associations for GP5 and GRN in addition to suggesting that MPO might also be causally implicated in platelet production. In total, the results suggest that these proteins are causally linked to platelet generation or turnover and may play important roles in CVD via a platelet-based mechanism. Additional research on these proteins such as functional studies using mice or zebrafish may elucidate specific mechanisms through which those proteins contribute to platelet counts, potentially identifying a platelet-mediated pathway from these proteins to CVD.

NOVELTY AND SIGNIFICANCE.

What Is Known?

• Select platelet measurements are associated with cardiovascular disease (CVD) and mortality risk.

• 71 plasma proteins have been previously linked with CVD in the Framingham Heart Study (FHS).

What New Information Does This Article Contribute?

• Myleoperoxidase (MPO), granulin (GRN), and glycoprotein V (GP5) were causally associated with platelet count, a measure of platelets per unit volume of blood.

• Consistent with the association with platelet count, silencing of MPO, GRN, and GP5 genes in platelet progenitor cells and megakaryocytes resulted in decreased platelet productivity.

Platelets are implicated in development of CVD through atherosclerotic lesion formation and thrombosis and previous studies have defined a genetic link with platelet count and production. We have previously identified genetic variants associated with circulating levels of 71 CVD-related plasma proteins (pQTL: protein quantitative trait loci) at FHS. In this study, we combined this pQTL data with published genome-wide association studies of mean platelet volume and platelet count to identify putatively causal proteins for platelet phenotypes. We then recapitulated these protein-platelet relationships through a gene knock-down experiments of megakaryocyte clones. Our study integrated population genetics and a functional study to identify novel protein biomarkers of platelet counts. Further work may reveal the specific mechanism and clinical importance of these interesting regulatory pathways of platelet levels.

Nonstandard Abbreviations and Acronyms:

AMI

acute myocardial infarction

BF

Bonferroni-corrected

CAD

coronary artery disease

CHD

coronary heart disease

CVD

cardiovascular disease

DOX

doxycycline

FHS

Framingham Heart Study

GWAS

genome-wide association study

imMKCL

immortalized megakaryocytic cell line

iPSC

induced pluripotent stem cell

LD

linkage disequilibrium

LME

linear mixed effects model

Mb

megabase

MK

megakaryocyte

MPV

mean platelet volume

MR

Mendelian randomization

PLT

platelet count

pQTL

protein quantitative trait loci

SABRe CVD

systems approach to biomarker research in cardiovascular disease

SNP

single nucleotide polymorphism

VTE

venous thromboembolism