Associations of osteopontin and NT-proBNP with circulating miRNA levels in acute coronary syndrome.

The genomic regulatory networks underlying the pathogenesis of non-ST-segment elevation acute coronary syndrome (NSTE-ACS) are incompletely understood. As intermediate traits, protein biomarkers report on underlying disease severity and prognosis in NSTE-ACS. We hypothesized that integration of dense microRNA (miRNA) profiling with biomarker measurements would highlight potential regulatory pathways that underlie the relationships between prognostic biomarkers, miRNAs, and cardiovascular phenotypes. We performed miRNA sequencing using whole blood from 186 patients from the TRILOGY-ACS trial. Seven circulating prognostic biomarkers were measured: NH2-terminal pro-B-type natriuretic peptide (NT-proBNP), high-sensitivity C-reactive protein, osteopontin (OPN), myeloperoxidase, growth differentiation factor 15, monocyte chemoattractant protein, and neopterin. We tested miRNAs for association with each biomarker with generalized linear models and controlled the false discovery rate at 0.05. Ten miRNAs, including known cardiac-related miRNAs 25-3p and 423-3p, were associated with NT-proBNP levels (min. P = 7.5 × 10-4) and 48 miRNAs, including cardiac-related miRNAs 378a-3p, 20b-5p and 320a, -b, and -d, were associated with OPN levels (min. P = 1.6 × 10-6). NT-proBNP and OPN were also associated with time to cardiovascular death, myocardial infarction (MI), or stroke in the sample. By integrating large-scale miRNA profiling with circulating biomarkers as intermediate traits, we identified associations of known cardiac-related and novel miRNAs with two prognostic biomarkers and identified potential genomic networks regulating these biomarkers. These results, highlighting plausible biological pathways connecting miRNAs with biomarkers and outcomes, may inform future studies seeking to delineate genomic pathways underlying NSTE-ACS outcomes.

INTRODUCTION

MicroRNAs (miRNAs) are short (~22 nt) noncoding molecules that regulate gene expression through effects on mRNA stability or protein translation (3) and have been implicated in a wide range of developmental and cellular processes (7, 15, 52). A single miRNA may target hundreds of mRNAs, and one mRNA may be regulated by multiple miRNAs (6, 31), yielding an environment in which target gene expression can be finely tuned based on changing cellular conditions. MiRNAs are found both intracellularly and extracellularly and display distinct expression profiles across various tissues, cell types, and body fluids (29, 32, 65). The potential of miRNAs to act as tissue-specific master regulators and the relative ease of assaying circulating miRNA levels make them intriguing candidates as biomarkers to monitor disease processes or drug response. Furthermore, recent technological advances permit large-scale miRNA profiling, yielding the opportunity to better define the genome regulatory networks related to disease processes.

Acute coronary syndrome (ACS) is the sudden, severe expression of coronary heart disease, representing the convergence of varied biological pathways. Agglomerating these diverse etiologies into a relatively nonspecific clinical phenotype may reduce the ability to identify the genomic architecture underpinning the various pathways. An alternative approach is to use circulating protein biomarkers as endophenotypes of ACS, as these circulating proteins may represent more granular intermediate phenotypes of the fundamental biological process. In genetic and genomic studies, the underlying premise behind this approach is that intermediate traits, such as biomarker levels, may reside closer to the underlying etiologic gene and therefore provide a stronger genetic signal for association studies than using a downstream, potentially more heterogeneous, clinical phenotype (1, 10, 50).

Levels of several circulating protein biomarkers change during ACS and improve clinical models used to predict outcomes. For example, NH2-terminal pro-B-type natriuretic peptide (NT-proBNP) levels are powerful diagnostic and prognostic indicators for heart failure (37), and C-reactive protein levels are associated with heart failure and mortality in survivors of acute myocardial infarction (MI) (54). However, the genetic and genomic regulation of prognostic biomarkers in ACS is incompletely understood. Given the ubiquitous nature of miRNA regulation of downstream targets, it is likely that miRNAs play important roles in underlying genomic regulatory networks associated with circulating levels of protein biomarkers. Identifying relationships between miRNAs and prognostic biomarkers may shed further light on genomic determinants of ACS outcomes.

Here, we performed a prespecified substudy of the Targeted Platelet Inhibition to Clarify the Optimal Strategy to Medically Manage Acute Coronary Syndromes (TRILOGY-ACS) clinical trial of prasugrel versus clopidogrel in patients with non-ST-segment elevation acute coronary syndrome (NSTE-ACS). Worldwide, NSTE-ACS is now the predominant form of ACS, and long-term outcomes of NSTE-ACS have been shown to be poorer than outcomes of ST-segment elevation ACS (8). We studied the association of circulating miRNAs and biomarkers of known prognostic value in NSTE-ACS [NT-proBNP, high-sensitivity C-reactive protein (hs-CRP), osteopontin (OPN), myeloperoxidase (MPO), growth differentiation factor 15 (GDF-15), monocyte chemoattractant protein (MCP1), and neopterin]. We applied next-generation sequencing technology to examine the relationship between baseline circulating miRNA profiles and 30-day protein biomarker levels, with the goals of better understanding how genomic regulatory networks activated during NSTE-ACS may be associated with adverse biomarker profiles, and providing a resource of miRNA-biomarker associations that might be used to generate plausible hypotheses in future studies.

MATERIALS AND METHODS

Study population.

The TRILOGY clinical trial was a multicenter, international phase 3 trial that randomized 9,326 individuals presenting with ACS to prasugrel or clopidogrel (ClinicalTrials.gov identifier NCT00699998) (9). Most TRILOGY participants were medically managed without revascularization during the index hospitalization. Therefore, the majority of recurrent ACS events (96%) were spontaneous and not related to revascularization procedures. The primary trial results showed no significant difference in the primary composite outcome between the two treatment arms (47). Subjects for this analysis were selected from the TRILOGY Advanced Biomarker SubStudy (TRILOGY-ABSS, n = 1,391) and were eligible for inclusion if they had available baseline RNA samples extracted from whole blood and valid plasma and serum biomarker data available from a sample collected 30 days postenrollment. Patients were initially considered as a matched case-control sample: cases were defined as patients who experienced a recurrent event (cardiovascular death, MI, or stroke) within 1 yr of randomization (n = 108 eligible subjects), while controls did not experience an event during study follow-up (n = 637 eligible subjects with at least 1 yr of follow-up). The final cohort was not a matched sample, as different numbers of cases (n = 89) and controls (n = 100) were selected to be sequenced; instead, the groups were broadly matched on the following variables: age at randomization, race, sex, previous clopidogrel use, presentation with unstable angina versus non-ST elevation MI, and treatment group (prasugrel versus clopidogrel). In this analysis, case/control status was considered as a confounder and included as a covariable in the model (see Statistical analysis section). Seven samples were sequenced twice to serve as technical replicates.

Biomarker assays.

Circulating biomarkers were measured at 30 days poststudy enrollment to minimize instability in biomarker levels from the acute presentation of each study subject. Other studies have shown biomarkers that predict long-term events when measured later rather than sooner after ACS (14). Plasma and serum samples were stored at −20°C and shipped monthly on dry ice for central storage at −70°C. Measurement of all circulating peptides and proteins was performed using commercially available enzyme-linked sandwich immunoassays based on chemiluminescent detection. NT-proBNP and hs-CRP were measured in plasma and serum samples respectively, using the IMMULITE 2000 NT-proBNP kit (Diagnostic Products, Los Angeles CA; catalog #L2KNT2) and hs-CRP kit (Diagnostic Products, catalog #L2KCR2) on an IMMULITE 2000 Analyzer (Diagnostic Products) at Quintiles Laboratories (Smyrna GA). Calculated average coefficients of variance (CV) for NT-proBNP and hs-CRP were 4.7 and 5.2%, respectively. OPN (MesoScale, Gaithersburg, MD; catalog #K151HJC-1), myeloperoxidase (MesoScale, catalog #K151EEC-1), and monocyte chemoattractant protein 1 (MesoScale, catalog #K151AYC-1) were measured in plasma samples on a Meso-Scale Discovery Sector Imager 2400 plate reader and analyzed using the Discovery Workbench 3.0 software (Meso-Scale Discovery, Gaithersburg MD). Neopterin (IBL International, Toronto, Canada; catalog #RE59321) and growth differentiation factor 15 (R&D Systems, Minneapolis MN; catalog #DGD150) were measured in plasma on a TECAN plate reader (Tecan). OPN, MPO, GDF-15, MCP1, and neopterin were measured in duplicate at two separate locations on 96-well plates and referenced to standard curves on each plate to assign values. Both biomarker measures were used to determine average reported values and CV. Overall assay precision was determined by averaging measured CV across all assay points of the study. Calculated average CV for OPN, MPO, GDF-15, MCP1, and neopterin assays were 5.5, 12.6, 5.5, 8.4, and 16.8%, respectively.

MiRNA sequencing and quality control.

Total RNA was extracted from frozen whole blood using the PerfectPure RNA blood kit (5Prime, Gaithersburg, MD). Libraries were prepared with the Illumina TruSeq Small RNA sample prep kit according to the manufacturer’s protocol with an ethanol precipitation step. Samples were resuspended in 16 µL of Tris·HCl. Libraries were stored at −80°F after size validation on the Bioanalyzer (Agilent). The libraries were quantified with the KAPA Library Quantification kit and protocol, and library pools were generated by combining (on an equimolar basis) sets of 24 libraries. Each library pool was then loaded into one rapid run sequencing flowcell, and clusters were generated for a single read flowcell. Single-end 50 bp sequencing runs were conducted on an Illumina HiSeq2500 instrument.

Cutadapt v1.5 was used to trim adapters (maximum error rate = 5%) and low-quality bases (quality score cut-off = 15) from the 3′ end of the sequence and to retain reads with a minimum postprocessed length of 18 nt (38). The trimmed reads were aligned to the human genome (GRCh38) with bowtie v1.0.1 (33), permitting one mismatch per read and reporting up to five best alignments for each read. Reads that mapped to more than 10 locations were discarded. Aligned reads were then mapped to miRNA primary transcripts from miRBase v21 (n = 2,813 miRNAs) (30) with the coverageBed function from bedtools v2.21.0 (44), which allows potentially ambiguous reads to be counted in each miRNA instead of being discarded. Using this approach, we detected 1,423 miRNAs at a rate of ≥0.1 mapped reads/million aligned reads (rpm) in at least one sample. The mapped reads included a high preponderance of miR-486-5p (approx. 80% of all mapped reads), which is highly expressed by erythrocytes (43) and has been suggested as a marker of hemolysis (27). MiR-486-5p has also been reported to be enriched 50× in samples prepared with the Illumina library prep used here versus two other kits tested (21), thus these reads were removed from the analysis. All remaining detected miRNAs were filtered with an expression level cut-off that maximized the Spearman correlation of miRNA levels within the seven pairs of technical replicates. Using a final cut-off of ≥1 rpm in ≥20 samples, we retained 247 miRNAs for analysis. One sample failed alignment quality control checks, and two had high proportions of rRNA (indicating potential sample quality issues); these three samples were removed, yielding a final analysis set of 186 samples.

Statistical analysis.

A negative binomial generalized linear model implemented in DESeq2 (36) was used to test for differential expression of miRNAs across biomarker levels. All models contained the following covariables: age, sex, ancestry (European versus non-European), pre-event clopidogrel use (Y/N), index event type (unstable angina versus non-ST elevation MI), heparin administration at the time of randomization (Y/N), a binary indicator of recurrent event (case-control) status, and experimental batch. We used an analogous model to test for differential expression of miRNAs between samples with and without a recurrent event, adjusting for all covariables except recurrent event status. All biomarkers were log10-transformed to induce normality, yielding fold change estimates and tests of differential expression based on logarithmic units of the biomarker. Within each biomarker, two-sided P values for miRNAs were adjusted to control the false discovery rate (FDR) at 5% (4). Experimentally validated mRNA targets of significant miRNAs were identified with Tarbase v7.0 (58) and tested for enrichment of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways in target genes using DIANA mirPath v3 (59), with the “pathways union” option and P values derived from the hypergeometric distribution (FDR < 1%). Heat maps were constructed with a Euclidean distance function and Ward’s algorithm to cluster the standardized fold changes for each miRNA over log-transformed biomarker levels. All statistical analyses of miRNAs were conducted by R v3.3.2.

OPN and NT-proBNP were tested for association with time to first recurrent event [cardiovascular (CV) death, MI, or stroke] in all TRILOGY-ABSS participants with OPN measured 30 days postenrollment (n = 1,209). Unadjusted Cox proportional hazards models contained biomarker levels only; adjusted models included the same covariables as the miRNA/biomarker association analyses above. The proportional hazards and linearity assumptions were not found to be violated for NT-proBNP. The proportional hazards assumption was not found to be violated for OPN, but visual evidence suggested a nonlinear U-shaped relationship that did not reach statistical significance. Therefore, the Cox model for OPN was fit with linear splines; knots were selected by using the predicted relationship between OPN and risk based on modeling the relationship using a restricted cubic spline (knots at 30 and 65).

Study approval.

The TRILOGY ACS study was approved by regulatory authorities in all participating countries and by participating sites' institutional review boards. All participants provided written informed consent.

RESULTS

The full TRILOGY cohort, the Advanced Biomarker SubStudy (ABSS) subcohort, and the nested cohort included in this analysis are defined in Fig. 1, with demographic and clinical characteristics of each group described in Table 1. Patients with at least one postenrollment event (CV death, MI, and stroke) were intentionally overrepresented in the analysis sample (see materials and methods), and this is reflected in the table: compared with the overall TRILOGY cohort, subjects included in this analysis are older, have more CV risk factors, and are more likely to have had an index event of non-ST elevation MI versus unstable angina. Patients in the analysis cohort also have higher levels of NT-proBNP and GDF-15 at 30 days postindex event compared with ABSS patients not included in the analysis cohort (P < 0.01).

Fig. 1.

Consort flow diagram for the analysis cohort.

Table 1.

Baseline demographic and clinical characteristics of the TRILOGY cohort, the ABSS subcohort, and the analysis sample

	TRILOGY(n = 9,326)	ABSS(n = 1,391)	Analysis(n = 186)
Demographics
Age, yr	66.0 (59.0, 74.0)	66.0 (59.0, 74.0)	70.0 (63.0, 77.0)
Male	5,676 (60.9%)	859 (61.8%)	127 (68.3%)
European ancestry	6,276 (67.3%)	1,141 (82.0%)	152 (81.7%)
Body mass index	27.1 (24.2, 30.5)	27.7 (25.0, 31.1)	27.6 (25.0, 31.2)
Cardiovascular risk factors and disease history
Creatinine clearance, ml/min	72.7 (54.1, 96.1)	75.9 (56.0, 100.2)	65.9 (50.9, 91.6)
Current or recent smoker	1,715 (18.6%)	258 (18.8%)	38 (20.9%)
Diabetes mellitus	3,539 (38.0%)	500 (36.0%)	72 (38.9%)
Hypertension	7,625 (82.0%)	1,216 (87.8%)	166 (89.7%)
GRACE risk score	121 (105, 139)	124 (106, 144)	135 (116, 152)
Family history of CAD	2,518 (30.4%)	464 (38.0%)	74 (45.4%)
Previous atrial fibrillation	710 (7.8%)	137 (10.0%)	19 (10.4%)
Chronic heart failure	1,629 (17.6%)	374 (27.1%)	49 (26.5%)
Prior MI	3,987 (43.1%)	652 (47.1%)	89 (48.1%)
Concomitant medications
Aspirin	8,572 (91.9%)	1,280 (92.0%)	171 (91.9%)
Beta blocker	7,251 (77.8%)	1,118 (80.4%)	147 (79.0%)
Statin	7,776 (83.4%)	1,164 (83.7%)	161 (86.6%)
Trial information
Prasugrel treatment arm	4,663 (50.0%)	690 (49.6%)	91 (48.9%)
Index event disease classification
Non-ST elevation MI	6,520 (69.9%)	938 (67.4%)	167 (89.8%)
Unstable angina	2,302 (24.7%)	366 (26.3%)	15 (8.1%)
CVD/MI/Stroke within 1 yr of index event*	954 (10.2%)	156 (11.2%)	87 (46.8%)
MI during entire trial follow-up*	737 (7.9%)	123 (8.8%)	64 (34.4%)
Biomarkers at 30 days postrandomization
NT-proBNP, pg/mL	NA	364 (156, 980)	551 (186, 1,792)
hs-CRP, ng/mL	NA	2.1 (1.0, 4.8)	2.4 (1.0, 6.4)
GDF-15, pg/mL	NA	1,058 (744, 1,588)	1,185 (840, 1,848)
MCP1, pg/mL	NA	257 (215, 311)	253 (205, 305)
MPO, ng/mL	NA	34.6 (17.3, 77.7)	41.2 (18.8, 92.5)
Neopterin, nmol/L	NA	8.8 (6.1, 12.2)	9.1 (6.7, 12.5)
OPN, ng/mL	NA	40.9 (29.4, 56.1)	46.2 (32.0, 61.3)

Raw proportions do not take differential follow-up length or censoring into account; thus, they are overestimates of the true event rate.

Continuous values are presented as median (Q1, Q3). Categorical variables are presented as n (%). ABSS, Advanced Biomarker SubStudy; CAD, coronary artery disease; CVD, cardiovascular disease; MI, myocardial infarction; NT-proBNP, NH2-terminal pro B-type natriuretic peptide; hs-CRP, high-sensitivity C-reactive protein; GDF-15, growth differentiation factor 15; MCP1, monocyte chemoattractant protein; MPO, myeloperoxidase; OPN, osteopontin.

A heat map of the estimated miRNA fold change from the association analysis of each miRNA/biomarker pair, shown in Fig. 2, demonstrates the absence of a single subset of measured miRNAs associated with levels of all measured biomarkers. The biomarkers fall into two broad clusters: OPN and NT-proBNP cluster together more tightly than any other pair of biomarkers and are joined by hsCRP in one cluster, while neopterin and GDF-15 exhibit a relatively close relationship, joined more loosely by MCP1 and MPO. Figure 2 displays the nominal −log10 P values for each miRNA/biomarker association test; the majority of significant miRNA associations (P < 0.05) occur with OPN and NT-proBNP. When we controlled the FDR at 5%, only miRNAs associated with OPN (n = 48, min. P = 1.6 × 10−6) or NT-proBNP (n = 10, min. P = 7.5 × 10−4) remained significant (Table 2). Of the identified miRNAs, three were positively associated with NT-proBNP levels (log10 fold change (FC): 1.11–1.13, min. P = 0.0016), seven were negatively associated with NT-proBNP levels (FC: 0.86–0.91, min. P = 7.5 × 10−4), 25 were positively associated with OPN levels (FC: 1.26–1.71, min. P = 1.6 × 10−6), and 23 were negatively associated with OPN levels (FC 0.67–0.83, min. P = 2.0 × 10−5). Four miRNAs were significantly associated with both biomarkers: miR-148b-3p and miR-140-3p are negatively associated with levels of NT-proBNP and OPN, while miR-877-5p and miR-423 are positively associated with both biomarkers. These results were adjusted for age, sex, race, previous clopidogrel use, heparin administration, index NSTE-ACS event type (non-ST-segment elevation MI versus unstable angina), and recurrent event status. Including randomization arm (prasugrel versus clopidogrel) as a covariate in the model did not affect the significance of the results (data not shown). No miRNAs were differentially expressed between samples with and without recurrent events after controlling for multiple comparisons (FDR < 5%), but nominally significant differential expression was detected for miR-6087 [FC: 1.11 (higher in subjects with a recurrent event), P = 0.015], let-7g-5p (FC: 1.10, P = 0.026), and miR-92b-3p (FC: 0.93, P = 0.046).

Fig. 2.

Heat maps of differential microRNA (miRNA) expression across biomarker levels. A: miRNA fold change estimates were standardized across each biomarker independently: hs-CRP, high-sensitivity C-reactive protein; OPN, osteopontin; NT-proBNP, NH2-terminal pro B-type natriuretic peptide; GDF-15, growth differentiation factor 15; neopterin; MCP1, monocyte chemoattractant protein; and MPO, myeloperoxidase. Individual miRNAs are displayed on the y-axis. B: unadjusted –log10 P values for each generalized linear model association analysis are shown. MiRNAs are in the same order as in A.

Table 2.

Differentially expressed miRNAs across biomarkers

miRNA	P Value	FDR-adjusted P value	Fold Change*
NT-proBNP
miR-25-3p	0.00075	0.036	0.90
miR-148b-3p	0.0010	0.036	0.86
miR-186-5p	0.0015	0.036	0.90
miR-423-3p	0.0016	0.036	1.11
miR-140-3p	0.0020	0.036	0.90
miR-451a	0.0022	0.036	0.87
miR-328-3p	0.0035	0.046	1.12
miR-425-5p	0.0036	0.046	0.91
miR-96-5p	0.0041	0.046	0.88
miR-877-5p	0.0047	0.048	1.13
OPN
miR-6734-5p	0.0000016	0.00039	1.71
miR-378a-3p	0.000020	0.0025	0.76
miR-4732-5p	0.00010	0.0051	1.38
miR-423-5p	0.00014	0.0051	1.36
miR-3184-3p	0.00014	0.0051	1.36
miR-6842-3p	0.00014	0.0051	0.67
miR-140-3p	0.00016	0.0051	0.76
miR-6777-3p	0.00017	0.0051	1.55
miR-939-5p	0.00031	0.0086	1.57
miR-320a	0.00036	0.0090	1.28
miR-142-5p	0.00051	0.011	1.55
miR-99b-5p	0.00085	0.015	0.69
miR-148b-3p	0.00089	0.015	0.68
miR-320d	0.00090	0.015	1.50
miR-744-5p	0.00091	0.015	1.32
miR-1976	0.0012	0.019	1.35
miR-320b	0.0015	0.022	1.29
miR-181a-2-3p	0.0019	0.025	0.72
miR-190b	0.0020	0.025	0.68
let-7g-5p	0.0021	0.025	0.74
miR-27b-3p	0.0022	0.025	0.81
miR-106b-3p	0.0026	0.027	1.32
miR-93-5p	0.0026	0.027	0.78
miR-1224-5p	0.0027	0.027	1.46
miR-199a-3p	0.0027	0.027	0.69
miR-505-3p	0.0031	0.027	0.71
miR-1229-3p	0.0031	0.027	1.43
miR-125a-5p	0.0032	0.027	0.76
miR-6087	0.0032	0.027	1.45
miR-324-3p	0.0035	0.029	1.26
miR-1180-3p	0.0036	0.029	1.29
miR-199b-3p	0.0039	0.030	0.70
miR-3200-3p	0.0041	0.031	0.71
miR-92b-3p	0.0044	0.032	1.28
miR-6793-3p	0.0047	0.033	1.41
miR-877-5p	0.0050	0.034	1.31
miR-23a-3p	0.0052	0.034	0.79
miR-1307-5p	0.0052	0.034	0.70
miR-1306-5p	0.0055	0.035	1.34
miR-20b-5p	0.0057	0.035	0.78
miR-155-5p	0.0065	0.039	0.77
miR-21-5p	0.0069	0.040	0.78
miR-6749-3p	0.0071	0.041	1.35
miR-29c-5p	0.0073	0.041	0.71
miR-342-5p	0.0081	0.045	1.31
miR-148a-3p	0.0086	0.046	0.78
miR-30e-5p	0.0092	0.048	0.83
miR-3940-3p	0.0095	0.049	1.29

Fold change indicates the increase/decrease in a given miRNA per log10 unit increase in NT-proBNP or OPN.

miRNA, microRNA; FDR, false discovery rate; NT-proBNP, NH2-terminal pro-B-type natriuretic peptide; OPN, osteopontin.

A total of 27 Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were significantly enriched (FDR-adjusted P < 0.01) in the predicted mRNA targets of either the OPN- or NT-proBNP-associated miRNAs (Supplemental Table S1; all supplemental material is available at https://figshare.com/s/59b3060f2bfb2bd5d6a1). The top 10 enriched pathways for NT-proBNP and OPN are shown in Table 3, and the putative target genes of associated miRNAs in pathways common to OPN and NT-proBNP are given in Supplemental Tables S2–S6, along with median expression levels in heart tissue, coronary artery, and whole blood as reported in the Genotype-Tissue Expression project (GTEx). One of the most significantly enriched pathways for each biomarker is prion diseases: three NT-proBNP-associated miRNAs have potential downstream mRNA targets in this pathway, while six OPN-associated miRNAs have targets in the pathway (both P < 10−325). Generally, the top results include several pathways involved in metabolism (fatty acid biosynthesis, fatty acid metabolism, steroid biosynthesis, lysine degradation), along with others related to cell signaling, apoptosis, and proliferation (viral carcinogenesis, p53, and hippo signaling pathways).

Table 3.

Top 10 KEGG pathways enriched in downstream target genes of miRNAs associated with NT-proBNP and OPN

KEGG pathway	P Value	Target Genes, n	Associated miRNAs, n
NT-proBNP
Prion diseases	<1 × 10−325	6	3
Viral carcinogenesis	1.1 × 10−10	73	4
Fatty acid biosynthesis	3.1 × 10−10	2	1
p53 signaling pathway	1.4 × 10−8	37	5
Proteoglycans in cancer	2.9 × 10−8	69	5
Adherens junction	3.6 × 10−7	33	4
ECM-receptor interaction	5.4 × 10−7	13	2
Protein processing in endoplasmic reticulum	5.6 × 10−6	73	4
Cell cycle	1.2 × 10−5	58	4
Steroid biosynthesis	2.9 × 10−5	6	4
OPN
Prion diseases	<1 × 10−325	12	6
ECM-receptor interaction	<1 × 10−325	35	7
Fatty acid biosynthesis	<1 × 10−325	6	14
Fatty acid metabolism	<1 × 10−325	26	15
Lysine degradation	<1 × 10−325	31	16
Proteoglycans in cancer	<1 × 10−325	132	16
Hippo signaling pathway	6.2 × 10−12	78	8
Viral carcinogenesis	6.9 × 10−10	116	12
Glioma	2.0 × 10−7	41	13
Chronic myeloid leukemia	2.7 × 10−6	48	12

KEGG, Kyoto Encyclopedia of Genes and Genomes; miRNAs, microRNA; NT-proBNP, NH2-terminal pro-B-type natriuretic peptide; OPN, osteopontin.

NT-proBNP was significantly associated with time to first recurrent event (CV death/MI/stroke) in both the unadjusted [hazard ratio (HR) (95% CI) = 1.23 (1.17–1.29), P < 0.001] and adjusted [HR (95% CI) = 1.20 (1.13–1.26), P < 0.001] Cox models. Using a linear spline model, we found OPN to be significantly associated with time to event for OPN values between 30 and 65 ng/mL in the unadjusted model [HR (95% CI) = 1.68 (1.22–2.32), P = 0.002]. This association was attenuated in the adjusted model [HR (95% CI) = 1.35 (0.96–1.90), P = 0.08], suggesting a relationship between the risk explained by OPN levels and the adjustment variables used in the model.

DISCUSSION

As more cardiovascular biomarkers are identified with prognostic value in NSTE-ACS, it becomes increasingly important to understand their genetic and genomic underpinnings to sift out biomarkers and pathways of high biological relevance. Here, we explored associations of miRNAs identified through large-scale sequencing with seven protein biomarkers of known prognostic value in an unbiased fashion, with the goal of identifying miRNAs representing genomic networks regulating these biomarkers along with miRNA-biomarker associations that may tag biologically plausible pathways connecting miRNAs, proteins, and cardiovascular outcomes. Using this approach in a subpopulation of subjects from a clinical trial of medically managed NSTE-ACS, we have identified a set of miRNAs associated with OPN and NT-proBNP levels that meet rigorous adjustment for multiple comparisons. Because persistent elevations of OPN and NT-proBNP after NSTE-ACS are associated with increased mortality, these miRNAs have plausible biological roles in determining outcomes after NSTE-ACS.

OPN is a matricellular protein that regulates inflammation and biomineralization, making it an obvious candidate as a biomarker for cardiovascular phenotypes, as atherosclerosis is an inflammatory process (49), and active biomineralization and microcalcification are key features of unstable coronary plaques (22). While OPN is expressed in a variety of tissues, including immune cells, bone, kidney, and gallbladder, it is only found at low levels in cardiac tissue under normal physiological conditions. Expression of OPN is increased under various pathological conditions such as ACS (53), heart failure (51), and aortic stenosis (35), and elevated OPN levels strongly predict mortality in patients with heart failure (48). Several of the miRNAs found to be associated with OPN levels in this study have previously reported associations with cardiac phenotypes, and the associations with OPN levels provide plausible hypotheses for intermediate steps in the pathway between miRNA expression and disease. For example, one of the miRNAs most strongly associated with OPN levels in our study, miR-378 (P = 2.04 × 10−5), has been reported to be significantly upregulated in mobilized CD34+ progenitor cells in patients with ST-elevation MI compared with healthy patients (55). Both in vitro and in vivo experiments reveal miR-378 as a key regulator of the proangiogenic capacity of CD34+ cells, which aids the preservation and recovery of left ventricular function in patients with MI, but the exact mechanism of this effect is unclear. As OPN has been widely implicated in angiogenesis (11, 63), the strong association between miR-378 and OPN levels may suggest a plausible pathway by which miR-378 exerts its proangiogenic activity. Another of the top miRNAs associated with OPN levels in this study is miR-320a (P = 3.63 × 10−4), which has previously been associated with cardiac injury following ischemia/reperfusion in mice (45), reduced ejection fraction (2), and heart failure (17). The association between miRNA-320 levels and OPN suggests one potential mechanism underlying these phenotypic associations: TLR4 is also associated with reduced ejection fraction (2), and OPN has been reported to negatively regulate TLR4-mediated immune responses. Several other miRNAs associated with OPN have relevant reported associations for which our results may provide plausible biological pathways: miR-320b is released from activated platelets in patients with ACS and has been shown in vitro to have downstream effects on ICAM-1 expression in endothelial cells (16); miR-99b-5p has been associated with endothelial cell differentiation from pluripotent human embryonic stem cells (26); and miR-155 is downregulated in coronary atherosclerotic plaques (19, 62). Taken together, these previous associations and our results suggest potential roles for miRNAs and OPN in the regulation of disease progression in NSTE-ACS.

NT-proBNP is a 32-amino acid polypeptide with diuretic and natriuretic effects and is released from the myocardium in response to mechanical stretch or endocrine signaling (18). Higher levels of NT-proBNP are strongly associated with heart failure and mortality after ACS (13), while therapies that lower NT-proBNP levels are associated with improved prognosis (25). Recently, a compound that inhibits neprilysin, which breaks down biological active natriuretic peptides, was shown to improve outcomes in heart failure (39). Several miRNAs associated with NT-proBNP levels in our analysis have been previously reported to be associated with other cardiovascular phenotypes. MiR-25 shows the strongest association with NT-proBNP levels in our analysis (P = 7.5 × 10−4), and its expression has been shown to differentiate patients presenting with ST elevation versus non-ST elevation MI (64) and is also increased in response to oxidative stress (41). Expression of another top result, miR-186-5p (P = 0.0015), has been associated with early stages of acute MI (61). In our analysis, NT-proBNP levels and expression of both miR-25 and miR-186-5p were negatively correlated, and the significance of this inconsistent direction of effect remains to be explored. MiR-328-3p (P = 0.0035) has also been shown to be associated with acute MI and subsequent heart failure and mortality (20). Both miR-25 and miR-328-3p are known to regulate calcium reuptake in cardiomyocytes via repression of the sarcoplasmic reticulum calcium uptake pump SERCA2a. MiR-25 was shown to delay calcium uptake kinetics in vivo, while antisense miR-25 rescued the resulting heart failure phenotype in a mouse model (60). MiR-328 has been shown to promote cardiac hypertrophy by directly repressing SERCA2a expression in transgenic mice overexpressing miR-328 in the heart (34). Interestingly, BNP decreases the expression of SERCA2a (28, 56), so the reported miRNA/biomarker associations here may help provide a fuller context for the relevant mechanisms.

Although we are not aware of any genome-wide association studies (GWAS) for osteopontin levels, several GWAS have identified genetic loci associated with NT-proBNP levels: NPPB, NPPA, MTHFR, CLCN6, KLKB1, SLC39A8, and the POC1B-GALNT4 region (12, 24, 40). This raises the question of whether the mRNA targets of the NT-proBNP-associated miRNAs identified here are recapitulating associations discovered in these genetic studies. One of the 10 NT-proBNP-associated miRNAs reported here, miR-148b-3p, is known to be associated with one of the NT-proBNP-associated loci (CLCN6) (5, 42). This suggests that our results are outside the realm of germline variation that affects NT-proBNP levels, and the genomic regulatory architecture of the relationship between miRNAs and NT-proBNP cannot solely be described through germline genetic variation.

The relatively tight cluster relationship of OPN and NT-proBNP in these samples (Fig. 2, Pearson’s correlation = 0.33) and the fact that they are the only biomarkers significantly associated with differential miRNA expression in this analysis raise the question of whether these associated miRNAs are reporting on pathways common to both biomarkers. In fact, four miRNAs were significantly associated with both OPN and NT-proBNP (miR-148b-3p, miR-423-3p/-5p, miR-140-3p, and miR-877-5p), suggesting that there may be some shared genetic regulatory elements. In addition to these overlapping miRNAs, several shared KEGG pathways showed enrichment of downstream mRNA targets, including the prion disease pathway (the strongest results for both biomarkers), fatty acid biosynthesis, proteoglycans in cancer, viral carcinogenesis, and ECM-receptor interaction. Several of the genes in the prion disease pathway are associated with cardiovascular phenotypes [e.g., heat shock protein 70 (23) and Notch1 (57)], but the predicted targets in this pathway are broadly involved in cellular processes such as apoptosis and proliferation, suggesting that this pathway may be reporting on common cellular processes also implicated in atherosclerosis and cardiac function, as opposed to processes specific to the brain or neural tissue. The other pathways enriched for both biomarkers show similar features: genes with relatively high expression in multiple tissues, often targets of both OPN- and NT-proBNP-associated miRNAs, implicated in a range of cellular and metabolic processes. More work is needed to determine whether the associations reported here reflect miRNAs and potential regulatory pathways specific to individual biomarkers, as opposed to more global regulators.

Several limitations of our study deserve comment. While many of the miRNAs identified in this unbiased analysis are extremely plausible candidates based on previous studies, these results should be validated in independent cohorts. Additionally, it is important to note that the current results are only associations, cannot be used to confirm biological relationships, and may not reflect causality: a recent study showed that OPN regulates miR-21 expression in mice (35), underscoring the fact that the regulatory pathways involved are likely complex and may involve multiple layers of feedback. We have tested for associations between miRNAs measured at the time of ACS presentation and protein biomarkers measured 30 days later but do not have longitudinal measurements that might allow us to look at time profiles of the miRNAs and biomarkers. Additionally, the associations reported here are between circulating miRNAs and protein biomarkers, which may not reflect levels found in cardiac tissue. Even with these caveats, the convergence of many of our findings with previous studies suggesting a role for these miRNAs in ACS lends important evidence for the validity of our results. Our study extends the results of these prior studies by elucidating the potential biological pathways regulated by these miRNAs through integration with protein biomarkers.

In conclusion, this application of a systems biology approach in a well-phenotyped clinical trial cohort serves as proof of principle that treating biomarkers as quantitative endophenotypes, and investigating the associations between these endophenotypes and miRNAs, can yield insight into the biological pathways relevant to recurrent cardiovascular events in NSTE-ACS. Among the miRNAs identified here, many have been previously reported to be associated with a wide range of relevant phenotypes and processes including heart failure, cardiac contractility, cardiac remodeling, apoptosis in cardiomyocytes, and acute MI. This work extends the current understanding of the role of these miRNAs to suggest that their associations with cardiovascular phenotypes may be mediated through pathways that also involve NT-proBNP or OPN, and additionally provides an atlas of miRNA-protein biomarker associations that may serve as a valuable resource of hypotheses for future work delineating the pathways that connect known cardiovascular-related miRNAs to clinical phenotypes and outcomes. Additionally, we have novel findings for miRNAs not previously reported to be associated with cardiovascular phenotypes. These novel miRNAs are therefore good candidates for future work, based on their association with established prognostic protein biomarkers. Further studies, including deeper functional evaluations, may give insight into how genetic and genomic networks converge with functional protein pathways to cause cardiovascular events.

GRANTS

This work was supported by a Singapore National Medical Research Council Clinician-scientist award (NMRC/CSA-INV/0001/2016).

DISCLOSURES

The TRILOGY-ACS Advanced Biomarker SubStudy was supported by Eli Lilly and Company (Indianapolis, Indiana). M. Chan reports receiving a research grant from Eli Lilly. M. T. Roe reports receiving research grants from Eli Lilly, Sanofi-Aventis, Daiichi Sankyo, Amgen, and the Familial Hypercholesterolemia Foundation, and consulting/advisory board/other payments from Eli Lilly, Janssen, Elsevier, AstraZeneca, Merck, Amgen, and Bristol-Myers Squibb. S. G. Goodman reports receiving research grants from Daiichi Sankyo, Eli Lilly, AstraZeneca, Bristol-Myers Squibb, and Sanofi-Aventis, and consulting payments/honoraria from Eli Lilly, AstraZeneca, and Sanofi-Aventis. K. A. A. Fox reports receiving research grants from AstraZeneca and the British Heart Foundation. P. W. Armstrong reports receiving research grants from Boehringer Ingelheim, Merck Sharp & Dohme, GlaxoSmithKline, Amylin, Merck, Sanofi-Aventis, and Regado, and consulting/advisory board payments from AstraZeneca, Boehringer Ingelheim, GlaxoSmithKline, Merck, Axio/Orexigen, Eli Lilly, Bayer, and Hoffman-La Roche. H. D. White reports receiving consulting/advisory board payments from AstraZeneca, Merck Sharpe & Dohme, Roche, and Regado Biosciences, and research grants from Sanofi-Aventis, Eli Lilly, The Medicines Company, NIH, Roche, Merck Sharpe & Dohme, AstraZeneca, GlaxoSmithKline, and Daiichi Sankyo Pharma Development. E. M. Ohman reports receiving research grants from Daiichi Sankyo, Eli Lilly, and Gilead, and consulting/advisory board payments from AstraZeneca, Daiichi Sankyo, Eli Lilly, Gilead, Pozen, The Medicines Company, WebMD, Abiomed, Sanofi- Aventis, and Janssen. L. M. Bowsman, J. V. Haas, and K. L. Duffin are employees of Eli Lilly. All other authors report no conflicts of interest.

AUTHOR CONTRIBUTIONS

S.G.G., M.Y.C., and S.H.S. conceived and designed research; E.G. and L.M.B. performed experiments; L.C.K., M.L.N., and J.V.H. analyzed data; K.L.D., M.Y.C., and S.H.S. interpreted results of experiments; L.C.K. and M.L.N. prepared figures; L.C.K., M.Y.C., and S.H.S. drafted manuscript; M.L.N., E.G., M.T.R., E.M.O., K.A.F., H.D.W., P.W.A., J.V.H., and K.L.D. edited and revised manuscript; L.C.K., M.Y.C., and S.H.S. approved final version of manuscript.