MicroRNA Biomarkers Associated with Type 1 Myocardial Infarction in HIV Positive Individuals

Abstract

Objective

Individuals with HIV suffer a higher burden of cardiovascular diseases. Traditional cardiovascular risk scores consistently underestimate cardiovascular risk in this population. Subsets of microRNAs (miRNAs) are differentially expressed among individuals with cardiovascular disease and individuals infected with HIV. However, no study has clarified whether specific miRNAs may be biomarkers for cardiovascular disease in individuals with HIV.

Design/Methods

We compared the miRNA expression profiles of 34 HIV positive individuals who had experienced clinically adjudicated Type I myocardial infarctions (MI) with the profiles of 76 HIV positive controls matched by traditional cardiovascular risk factors and HIV-specific measures. Using the elastic net algorithm, we selected miRNAs most strongly associated with incident MI and then used conditional Cox proportional hazards regression and cross-validation to evaluate miRNAs and their association with incident MI. We evaluated whether miRNA markers would improve risk classification relative to the Framingham Risk Score.

Results

Higher miR-125a-5p and miR-139-5p expression levels were each associated with increased risk of developing MI after adjustment for Framingham Risk Score and HIV-related factors (HR 2.43, p=0.018; HR 2.13, p=0.048, respectively). Compared to the Framingham Risk Score alone, adding expression levels of miR-125a-5p or miR-139-5p resulted in an integrated discrimination improvement of 10.1% or 5.8% respectively.

Conclusions

MiR-125a-5p and miR-139-5p, transcripts known to be differentially expressed in HIV positive individuals, may serve as unique biomarkers predictive of cardiovascular disease in these patients and may help clarify processes due to HIV infection that contribute to cardiovascular pathology in this population.

Introduction

Individuals with HIV experience high rates of cardiovascular disease even while receiving effective treatment.[1–4] HIV-infected individuals develop greater atherosclerosis and vascular dysfunction and are at increased risk for myocardial infarction (MI), heart failure, and vascular disease.[1,5–7] This higher burden of disease is likely due to host-pathogen interactions that cause derangements in the immune system and vascular endothelium.[8–10] Due to these unique pathophysiological processes, traditional risk calculators perform poorly and underestimate cardiovascular risk in HIV patients.[11,12] There remains a need for identifying HIV-specific biomarkers for predicting which individuals with HIV are most at risk for cardiovascular disease in order to guide preventative efforts and improve outcomes.

Within the last decade, microRNAs (miRNAs) have become recognized as fundamental regulators of almost all physiological processes including those responsible for cardiovascular disease.[13–16] Subsets of miRNAs discriminate coronary from non-coronary chest pain,[17,18] predict acute MI before troponin elevation,[19,20] forecast risk for post-MI complications,[21–23] identify patients with stable coronary artery disease,[24,25] and predict major cardiovascular events.[26,27]

While great progress has been made in clarifying the role of miRNAs in cardiovascular disease in the general population, whether these same miRNAs associate with increased cardiovascular risk in individuals with HIV remains unknown. Our group recently performed one of the first large-scale profilings of HIV-infected individuals.[28] We found several miRNAs that were differentially expressed, including miR-126, miR-125, and let-7c, which are 3 miRNAs previously identified as predictors of cardiovascular disease in the general population. Identifying miRNAs that are specific to HIV infection and cardiovascular disease may help elucidate the unique mechanisms that confer additional cardiovascular disease risk in HIV positive patients. In this nested case-control study, we characterized for the first time, differences in miRNA expression profiles between HIV-infected individuals who experienced a Type I MI (atherothrombotic as opposed to Type II demand related MI) compared with matched HIV-infected individuals who did not.

Patients and Methods

Study Cohort

Cases and controls were selected from the CFAR Network of Integrated Clinical Systems (CNICS), a multi-site network with over 31,000 HIV-infected adults followed in HIV/AIDS clinics.[29] Patients were evaluated according to clinical need (typically every 1 to 3 months) with data derived from routine clinical care as captured by electronic medical records. The CNICS consortium uses validated instruments to monitor data quality and accuracy. Institutional review boards at each site approved the study protocol and all study participants provided informed consent.

Subject Characteristics

We included 117 participants: 36 HIV-infected adults who developed a Type I MI between 2001 and 2012 (hereafter referred to as the “index diagnosis date”) and 81 matched controls. MI events were adjudicated according to prior published procedures.[30] Cases were actively taking antiretroviral therapy, had a plasma HIV RNA level <500 copies/ml, and had no prior history of MI. Up to 3 controls were selected for each case using incidence density sampling from those who had no known history of MI up to the index diagnosis date. All controls were HIV positive and were matched to cases according to age at index diagnosis date, birth sex, CD4+ T cell count, and duration of virologic suppression.

miRNA Profiling

Plasma samples were isolated and stored at 8 different CNICS sites according to institutional protocols. MiRNA expression profiling was completed by the high-throughput Gene Expression Core Laboratory at the University of Massachusetts Medical Center.[31]

Total RNA was obtained from plasma samples using miRNeasy Serum Plasma Kit (Qiagen, Frederick, MD, USA). RNA samples were converted to complementary DNA (cDNA) via reverse transcription using the miScript II RT Kit (Qiagen). The cDNA underwent preamplification using the miScript Microfluidics PreAMP Kit (Qiagen) before undergoing Real-time PCR miRNA profiling on Fluidigm Dynamic Arrays (Fluidigm, San Francisco, CA, USA) using human miRNA assays (Qiagen). Real-time PCR analysis was performed on BioMark Real-Time PCR Analysis software using linear derivative and user global settings (threshold setting 0.002).

We excluded any miRNA if greater than 80% of samples failed to meet quality scores, a sign of either very low miRNA levels and/or failed amplification. We normalized miRNA expression data by subtracting the cycle threshold (Ct) from the global mean across the Ct’s of all miRNAs for that patient, a widely accepted normalization method.[32] Lastly, we excluded any patients if greater than 80% of their samples failed to meet quality scores.

Out of 117 individuals, 7 were excluded for having too many samples that failed to meet quality scores. The final analysis was performed on 34 cases and 76 controls. For miRNAs, 135 of 175 miRNAs had sufficient expression levels. Of these miRNAs, 48 had minimal data loss (missing for <10% of participants), 49 had moderate data loss (missing for 10-50%), and the remaining 38 had substantial data loss (missing for 50-78% of participants). A complete case analysis would lead to biased results and loss of power. The missing not at random (MNAR) assumption was violated so that traditional multiple imputation methods would not be valid. Participants with a greater fraction of missing miRNA data were more often male and had lower BMI. Participants with more missing miRNA data had lower values on average (Supplemental Figure 1). We imputed missing data using a pattern-mixture model approach with the neighboring-case missing values method,[33–35] using 100 imputations to ensure high relative efficiency and reliable estimation. Imputation models included miRNA variables and MI status as well as covariates listed below.

Covariates

Covariates for all multivariable models were age, gender, race/ethnicity, body mass index (BMI), diagnosis of diabetes mellitus, diagnosis of hypertension (defined as blood pressure > 140/90 or taking antihypertensives), CD4 lymphocyte count, and duration of HIV RNA suppression. The Framingham Risk Score (FRS) was calculated for each participant using the ATPIII calculator based on age, gender, total cholesterol, HDL cholesterol, smoking status (smoker being defined as current or prior smoker), systolic blood pressure, and antihypertensive use.[36,37] Missing components of the FRS were imputed as described above. All covariate measurements were taken at the time of the plasma sample for miRNA profiling except for blood pressure readings which were taken within 90 days of the plasma sample acquisition date.

Statistical Analysis

We summarized demographic and clinical characteristics overall and then stratified by MI status. We examined unadjusted Spearman correlations between miRNA markers using Fisher’s z transformation to combine estimates from multiple imputed variables.[38]

To identify miRNA markers associated with developing MI, we modeled miRNA variables in combination using elastic net, a variable selection method that combines Least absolute shrinkage and selection operator (LASSO) and Ridge regression penalties to produce a parsimonious model. The elastic net is well suited to analyses where the number of predictors is large relative to the sample size, and where predictors may be correlated.[39] We standardized each marker (mean=0, SD=1) so that markers with larger variances would not have a greater influence on variable selection. We fit stratified Cox models with elastic net penalties for each of the 100 imputed datasets, using cross-validation to select the optimal L1 (LASSO) and L2 (Ridge) penalties. To handle uncertainty due to imputed missing data, we retained only variables that were selected by >=25% of the models. We averaged the estimated coefficients across the 100 models to compare the strength of the association.

To determine whether miRNA variables selected by elastic net were independently associated with developing MI, multivariable models were adjusted for demographics, traditional CVD risk factors, and HIV-related risk factors (per the Covariates section) as well as FRS. Our final models were estimated using conditional Cox proportional hazard regression to account for the matched case control design. We combined the results of the analyses of the 100 imputed datasets using SAS Proc MIANALYZE. We evaluated pairwise and three-way combinations of markers that were selected by elastic net to determine if markers were simultaneously associated with risk of MI.

We constructed Kaplan-Meier plots to compare risk of MI by expression tertiles for candidate miRNAs. We used the integrated discrimination improvement (IDI) to evaluate whether miRNA markers could improve risk classification relative to the FRS alone.[40] The IDI was calculated as (p2_E-p1_E) + (p2_NE-p1_NE), where p1_E and p1_NE denote the mean probabilities from the reference model (1) of events (E) and non-events (NE), and p2_E and p2_NE denote the mean probabilities from the new model (2) of events (E) and non-events (NE). Event probabilities were estimated from conditional logistic regression models (to account for the matched case-control design) which controlled for the FRS (reference model 1) or the FRS plus the candidate marker (new model 2).

Elastic net variable selection was performed using the R package penalized for stratified Cox models.[41] All other analyses were performed using SAS version 9.4.

Results

The matched cohort of 34 cases and 76 controls had a median age of 47. By the time of MI event (cases) or censoring (controls), participants had been followed for an average of 7.3 years (SD 3.9). Among cases, the median and mean times between sample acquisition and MI were 4.5 and 3.5 years, respectively. Three-fourths of participants were male. Approximately half were African-American. While hypertension was present in one-third of participants, antihypertensive use was documented in 5%. Only 13% of individuals were taking a statin and 3% were taking clopidogrel, with no significant difference between cases and controls. Diabetes was documented in 5% and 43% of patients were smokers. The median LDL level was 106 mg/dL and median BMI was 28. Median duration of viral suppression was 33 months and median CD4 count was 574 (Table 1).

Table 1:

Summary of demographic and clinical characteristics of MI cases and matched controls in CNICS cohort

Parameter N		Overall 110	Cases 34	Controls 76
Age (y)		47 (43, 55)	47 (44, 55)	46 (43, 56)
Male		82 (75%)	26 (76%)	56 (75%)
Race/ethnicity:
Black		58 (53%)	17 (50%)	41 (55%)
White		49 (45%)	15 (44%)	34 (45%)
Other		2 (2%)	2 (6%)	0
SBP (mmHg)		126 (116, 139)	122 (116, 135)	127 (115, 140)
Antihypertensive use:	Yes	6 (5%)	2 (6%)	4 (5%)
	No	96 (87%)	32 (94%)	64 (84%)
	Unknown	8 (7%)	0	4 (5%)
Hypertension		35 (32%)	11 (32%)	24 (32%)
Statin use		14 (13%)	6 (18%)	8 (11%)
Clopidogrel use		2 (2%)	0 (0%)	2 (3%)
Diabetic		5 (5%)	2 (6%)	3 (4%)
Smoker:	Yes	47 (43%)	16 (47%)	31 (41%)
	No	21 (19%)	2 (6%)	19 (25%)
	Unknown	42 (38%)	16 (47%)	26 (34%)
HDLc (mg/dL)		48 (34, 56)	43 (36, 52)	48 (34, 58)
LDLc (mg/dL)		106 (83, 121)	113 (85, 151)	103 (82, 116)
Total Cholesterol (mg/dL)		191 (161, 221)	185 (162, 240)	193 (156, 215)
BMI (kg/m2)		28 (25, 32)	27 (25, 30)	28 (24, 33)
Months virally suppressed		33 (15, 68)	27 (13, 55)	38 (16, 79)
CD4 count		574 (402, 804)	562 (359, 689)	616 (438, 815)

Data are presented as Median (IQR) or numbers (percent)

Using conditional Cox regression models, we examined associations of miRNAs selected by elastic net with incident MI (Table 2, Supplemental Figure 2). Higher miR-125a-5p expression levels were associated with an increased risk of developing MI in both unadjusted analysis (HR 1.94 per 1 SD increase, p=0.027) (Supplemental Figure 3) and after adjustment for traditional CVD risk factors and HIV-related factors (HR 2.18, p=0.026). In models that adjusted for FRS alone or for FRS in combination with HIV-related factors, miR-125a-5p remained associated with a 2-fold increased risk of MI. Separately, higher levels of miR-139-5p were also associated with a 2-fold increased risk of MI in models that adjusted for FRS alone or FRS in combination with HIV-related factors (HR 1.94 p=0.046, HR 2.13 p=0.048, respectively). None of the other selected markers showed statistically significant associations with MI risk.

Table 2.

MiRNA markers selected by elastic net and their associations with incident type 1 myocardial infarction in HIV patients

miRNA	Unadjusteda HR (95%CI)	Model 1 HR (95%CI)	Model 2 HR (95%CI)	Model 3 HR (95%CI)
25_5p	0.87 (0.63, 1.20), p=0.39	0.84 (0.55, 1.28), p=0.41	0.87 (0.61, 1.24), p=0.44	0.86 (0.56, 1.32), p=0.49
590_5p	1.06 (0.82, 1.37), p=0.65	1.08 (0.79, 1.49), p=0.62	1.04 (0.81, 1.34), p=0.76	1.06 (0.79, 1.42), p=0.71
183_5p	1.27 (0.73, 2.23), p=0.40	1.38 (0.71, 2.70), p=0.34	1.27 (0.72, 2.24), p=0.41	1.33 (0.71, 2.49), p=0.38
34b_3p	1.16 (0.92, 1.46), p=0.20	1.12 (0.86, 1.47), p=0.40	1.18 (0.93, 1.50), p=0.17	1.13 (0.87, 1.47), p=0.35
125a_5p	1.94 (1.08, 3.50), p=0.027	2.18 (1.10, 4.32), p=0.026	2.21 (1.18, 4.14), p=0.013	2.43 (1.17, 5.04), p=0.018
331_5p	0.91 (0.66, 1.25), p=0.55	0.92 (0.62, 1.37), p=0.69	0.93 (0.67, 1.29), p=0.68	0.96 (0.65, 1.40), p=0.81
16_1_3p	1.05 (0.74, 1.48), p=0.79	0.96 (0.59, 1.57), p=0.87	0.99 (0.69, 1.42), p=0.94	0.88 (0.51, 1.55), p=0.67
1291	0.82 (0.56, 1.19), p=0.30	0.73 (0.42, 1.24), p=0.24	0.76 (0.48, 1.21), p=0.24	0.70 (0.37, 1.33), p=0.27
127_3p	0.72 (0.44, 1.18), p=0.19	0.68 (0.40, 1.17), p=0.16	0.75 (0.45, 1.27), p=0.29	0.71 (0.41, 1.25), p=0.24
7_1_3p	1.03 (0.76, 1.38), p=0.86	1.02 (0.71, 1.46), p=0.92	1.02 (0.75, 1.40), p=0.88	1.00 (0.71, 1.41), p=0.99
30d_3p	0.92 (0.57, 1.50), p=0.74	0.90 (0.48, 1.70), p=0.75	0.89 (0.51, 1.54), p=0.67	0.89 (0.46, 1.71), p=0.73
194_5p	1.16 (0.76, 1.78), p=0.49	1.19 (0.70, 2.01), p=0.52	1.15 (0.75, 1.76), p=0.53	1.16 (0.70, 1.92), p=0.55
206	0.88 (0.52, 1.51), p=0.65	0.86 (0.42, 1.75), p=0.67	0.86 (0.49, 1.50), p=0.60	0.83 (0.41, 1.68), p=0.60
143_3p	1.11 (0.83, 1.47), p=0.49	1.12 (0.82, 1.54), p=0.49	1.08 (0.81, 1.44), p=0.58	1.09 (0.81, 1.49), p=0.56
140_5p	1.02 (0.77, 1.35), p=0.90	1.03 (0.74, 1.44), p=0.86	1.04 (0.78, 1.40), p=0.77	1.06 (0.75, 1.51), p=0.74
1180	0.88 (0.56, 1.38), p=0.58	0.83 (0.48, 1.43), p=0.51	0.87 (0.52, 1.45), p=0.59	0.84 (0.45, 1.56), p=0.58
101_3p	0.95 (0.68, 1.33), p=0.77	0.95 (0.64, 1.40), p=0.78	0.94 (0.66, 1.33), p=0.73	0.95 (0.64, 1.40), p=0.78
15a_5p	1.11 (0.66, 1.89), p=0.69	1.16 (0.56, 2.39), p=0.69	1.18 (0.66, 2.14), p=0.57	1.22 (0.58, 2.56), p=0.61
139_5p	1.81 (0.98, 3.35), p=0.059	2.04 (0.99, 4.18), p=0.053	1.95 (1.01, 3.76), p=0.046	2.13 (1.01, 4.50), p=0.048
409_3p	0.85 (0.57, 1.27), p=0.43	0.78 (0.49, 1.24), p=0.30	0.87 (0.56, 1.34), p=0.52	0.82 (0.50, 1.34), p=0.43
378a_3p	1.14 (0.76, 1.73), p=0.52	1.14 (0.68, 1.93), p=0.62	1.22 (0.78, 1.89), p=0.38	1.25 (0.73, 2.13), p=0.41
155_5p	0.87 (0.55, 1.36), p=0.54	0.83 (0.48, 1.42), p=0.50	0.86 (0.54, 1.38), p=0.53	0.85 (0.51, 1.44), p=0.55
144_5p	0.90 (0.57, 1.44), p=0.67	0.87 (0.50, 1.51), p=0.61	0.96 (0.58, 1.59), p=0.87	0.94 (0.53, 1.66), p=0.82
192_5p	0.93 (0.60, 1.43), p=0.73	0.94 (0.57, 1.53), p=0.80	0.91 (0.58, 1.43), p=0.69	0.95 (0.58, 1.56), p=0.84
342_3p	0.76 (0.47, 1.23), p=0.26	0.80 (0.46, 1.38), p=0.42	0.80 (0.49, 1.33), p=0.39	0.83 (0.48, 1.45), p=0.52
146a_3p	1.02 (0.76, 1.36), p=0.91	1.00 (0.71, 1.41), p=0.99	1.03 (0.75, 1.43), p=0.85	1.01 (0.71, 1.43), p=0.96
132_3p	1.08 (0.71, 1.64), p=0.73	1.08 (0.64, 1.80), p=0.78	1.07 (0.69, 1.66), p=0.77	1.07 (0.64, 1.78), p=0.80

^aUnadjusted hazard ratios (with 95% confidence intervals) from Cox proportional hazards regression models stratified by matched case-control pair.

Model 1: adjusted for age, race, BMI, DM, HTN, CD4, and months of viral suppression and stratified by matched case-control pair.

Model 2: adjusted for Framingham risk score and stratified by matched case-control pair.

Model 3: adjusted for Framingham risk score, race, BMI, DM, CD4, viral suppression, and stratified by matched case-control pair

We created Kaplan-Meier plots for miR-125a-5p and miR-139-5p (Figure 1). For miR-125a-5p, MI risk was greatest for those in the highest expression tertile (HR 5.04 for tertile 3 vs. 1, p=0.022), while there was little difference between the middle and lower tertiles (HR 1.22, p=0.74). Similarly, those in the highest expression tertile of miR-139-5p appeared to have an increased risk of MI relative to the lowest tertile, although the difference did not reach statistical significance (HR 2.93, p=0.17). We were also able to show that miR-125a-5p and miR-139-5p improved risk prediction when added to the Framingham Risk Score model. Compared to FRS alone, adding expression levels of miR-125a-5p improved the overall risk classification by 10.1% by appropriately estimating a lower probability of MI in those who did not end up experiencing an MI and estimating a higher probability of MI in those who did eventually experience an MI (Figure 2). Risk classification improved by 5.8% when miR-139-5p expression levels were added to the FRS.

Figure 1.

Kaplan-Meier plots showing association of miRNA markers with incident type 1 myocardial infarction in HIV patients

T1, T2, and T3 denote tertiles of miRNA expression with T3 having highest levels of miRNA expression.

T2 vs. T1 and T3 vs. T1 are unadjusted hazard ratios (with 95% confidence intervals) from Cox proportional hazards regression models stratified by matched case-control pair.

Figure 2.

Improvement in risk classification when adding miRNA expression levels to the Framingham Risk Score.

IDI = Integrated discrimination improvement. Event probabilities were estimated from conditional logistic regression models (to account for the matched case-control design) which controlled for the FRS or the FRS plus the miRNA marker.

We tested combinations of selected markers to determine whether markers were independently and simultaneously associated with MI status. We identified several pairwise combinations of miRNAs involving miR-125a-5p, miR-139-5p, miR-409-3p, miR-342-3p, and miR-382-5p that were simultaneously significant (Figure 3). For example, when miR-125a-5p and miR-342-3p were modeled together, each standard deviation increase in miR-125a-5p expression was associated with a 4.9-fold increased risk of MI (p=0.003), while each standard deviation increase in miR-342-3p was associated with a 68% decreased risk of MI (p=0.012). Higher miR-125a-5p expression and lower miR-382-5p expression were simultaneously associated with MI risk (HR 3.62, p=0.005; HR 0.38, p=0.022, respectively). MiR-139-5p also remained significantly associated with MI when controlling for miR-342-3p (HR 2.90, p=0.013), but the effect of miR-342-3p lost statistical significance when controlling for miR-139-5p (HR 0.53, p=0.07). No three-way or higher order combinations of markers were simultaneously statistically significant.

Figure 3:

MiRNA combinations associated with incident MI in HIV patients

Unadjusted hazard ratios (with 95% confidence intervals) from Cox proportional hazards regression models stratified by matched case-control pair.

Models 1-3: Markers are included individually, not simultaneously. Model 1: adjusted for age, race, BMI, DM, HTN, CD4, and months of viral suppression and stratified by matched case-control pair. Model 2: adjusted for Framingham risk score and stratified by matched case-control pair. Model 3: adjusted for Framingham risk score, race, BMI, DM, CD4, viral suppr, and stratified by matched case-control pair.

We examined 6 miRNAs that had been previously identified as associated with cardiovascular disease (Supplemental Table 1).[25–27] Of these markers, only one (miR-155-5p) was chosen by elastic net, but was not statistically associated with MI risk after multivariate adjustment. Two of the miRNAs, miR-126-3p and miR-223-3p, were strongly correlated with miR-139-5p (r=0.85, r=0.79 respectively) and miR-125a-5p (r=0.69, r=0.67 respectively).

Since platelets can be an important source of miRNAs and platelet number variation can therefore affect miRNA levels, we compared miRNA levels to platelet counts using a limited subset of 51 patients that had platelet counts within 90 days of miRNA measurement.[26] We found no significant correlation between platelet counts and expression levels of miR-125a-5p, miR-139-5p, miR-409-3p, miR-342-3p, or miR-382-5p.

Discussion

In our matched case-control cohort, expression levels of miR-125a-5p and miR-139-5p were significantly associated with incident MI in HIV infected individuals independent of traditional CVD and HIV-related risk factors. Individuals with the highest tertile of miR-125a-5p expression had a five-fold increase in risk for MI when compared to those in the lowest tertile. Accounting for miR-125a-5p and miR-139-5p expression resulted in improvements in risk classification for MI when added to the Framingham Risk Score. Their associations with incident MI were further strengthened when combined with expression levels of miR-409-3p, miR-342-3p, and miR-382-5p. These findings are important given that recent studies continue to show poor prediction of cardiovascular disease in HIV infected individuals when using traditional risk scores.[11] To our knowledge this is one of the first studies to evaluate the role of miRNAs in predicting cardiovascular disease in this population.

In both unadjusted and adjusted Cox regression analyses higher expression of miR-125a-5p was associated with a greater risk of MI. MiR-125a-5p is thought to regulate macrophage activation and lipid metabolism, critical processes in cardiovascular disease. Activated macrophages take up oxidized low density lipoprotein (oxLDL) and become foam cells which form atherosclerotic plaques.[42] MiR-125a-5p has been shown to downregulate the expression of scavenger receptor class B type 1 (SR-BI) and thereby reduce selective high-density lipoprotein (HDL) cholesteryl ester (CE) transport which is necessary for lowering cholesterol levels via cholesterol excretion in bile.[43] Contrastingly, other studies show that miR-125a-5p encourages the polarization of macrophages from the classical pro-inflammatory activation M1 pathway to the alternative anti-inflammatory M2 pathway; one study found that miR-125a decreases macrophage uptake of oxidized low density lipoprotein (oxLDL) and the secretion of inflammatory cytokines.[44,45] While there is evidence that miR-125a-5p transcripts may play both pro- and anti-atherosclerotic roles, our results suggest that the imbalance in expression levels seen in HIV may increase the risk of incident MI. MiR-125a-5p’s role in modulating macrophages is especially pertinent in HIV given that macrophages play a key role in HIV disease pathogenesis.[46] The proinflammatory monocyte (CD14+/CD16+) subpopulation is involved in the development of atherosclerosis in HIV,[47] with studies demonstrating that circulating monocyte levels are associated with coronary calcium progression, inflammation, and atherosclerotic plaques in HIV-infected cohorts.[48,49] Our group also recently reported that activated monocytes were positively correlated with arterial inflammation as assessed by FDG-PET CT in treated HIV.[50]

After adjusting for cardiovascular risk factors and HIV-related factors, miR-139-5p was also predictive of incident MI in HIV positive individuals. While miR-139-5p has been well described as a regulator of cancers,[51,52] its relationship to cardiovascular disease has not been well described. Studies have shown that it may be involved in responses to ischemia reperfusion injuries in both cardiac myocytes and neurons.[53–56] It may represent a novel marker for MI in HIV infected individuals.

We also discovered that additional miRNAs, miR409-3p, miR-382-5p, and miR-342-3p, when simultaneously modeled with miR-125a-5p or miR-139-5p both strengthened the association of miR-125a-5p and miR-139-5p with incident MI and were also independently associated with MI. Both miR-409-3p and miR-382-5p exhibited this relationship when modeled simultaneously with miR-125a-5p. Neither transcript has robust evidence to date for roles in ischemic heart disease although miR-409-3p has been shown to decrease fibrinogen production, a marker which is predictive of both higher risk of cardiovascular disease as well as increased mortality among HIV infected persons.[57,58] MiR-382-5p has been shown to be involved in the regulation of cholesterol homeostasis and inflammatory responses.[59] Lower expression levels of miR-342-3p was significantly associated with higher incident MI when modeled with either miR-125a-5p or miR-139-5p. This may be consistent with descriptions of reduced miR-342-3p levels in patients with angiographically defined stable CAD.[60] Interestingly, like miR-125a-5p, miR-342-3p’s counterpart transcript miR-342-5p is also thought to play a role in macrophage activation and atherosclerosis.[61]

Prior studies have identified miR-126, miR-197, and miR-223 as well as miR-145, miR-155, and let-7c as biomarkers predictive of MI and coronary disease in the general population.[24–27] Although they were not found to be significant predictors of MI in our analysis, many of these miRNAs were strongly correlated with at least one of the other significant miRNAs in our analysis. This could mean that they are still significant predictors that were merely obscured due to their colinearity with other miRNA predictors. However, it also may be because they are less involved in the processes of atherosclerosis in HIV. Atherosclerosis has a distinctive phenotype in HIV as evidenced by more extensive and calcified coronary artery plaques and higher baseline arterial inflammation.[49,50,62] HIV infected individuals presenting with ACS also have significantly different clinical characteristics.[4] Compared to the general population of individuals with cardiovascular disease, our cohort is younger (median age of 47) with low rates of diabetes and hypertension.

The transcripts that we have highlighted in this analysis overlap greatly with miRNAs previously described in the host response to HIV infection. In fact, all 5 miRNA transcripts (miR-125a-5p, miR-139-5p, miR-409-3p, miR-382-5p, miR-342-3p) identified in this paper as predictive of MI in HIV infected individuals have been found by our group or others to be differentially expressed in HIV positive individuals compared with HIV negative individuals (Supplemental Table 2).[28,63,64] Intriguingly, both miR-125a and miR-382-5b are included in a small family of miRNAs that are believed to target the 3’UTR of HIV-transcripts and inhibit viral translation.[65,66] Thus, the same miRNAs that are part of the host response to HIV may also increase cardiovascular risk.

There are several limitations to our study. Our sample size was small, which would limit the ability to detect small differences in miRNA expression and could magnify differences that occurred by chance. However, the matched case-control design, which factored in HIV-related factors including duration of viral suppression, augmented predictive power. Additionally, the quality of the cohort data was high with good follow-up and rigorously adjudicated clinical outcomes. One aberrancy was the low rate of hypertensive use in this cohort. This could be due to a lack of antihypertensive prescribing by providers and/or lack of adherence by patients given that patients were young and did not have many comorbidities. However, we cannot determine provider intentions or patient behavior from our data. Ultimately, data on antihypertensive use was not critical in influencing our conclusions since it was only used in FRS calculations and had a very small contribution to risk. [36,37]

As this is an observational study, causality cannot be inferred from the relationships we have highlighted. Elastic net and other penalized regression methods do not guarantee control of the type 1 error rate or false discovery rate. We used cross-validation to select the optimal penalty, but it remains possible that our method retained miRNA markers that are falsely associated with the outcome. Therefore, while the markers we identified are hypothesis generating for risk prediction, further exploration on a molecular level will be required to effectively identify which markers may be important to target for risk modification. Utilizing an additional cohort of HIV positive individuals with serum samples and similarly adjudicated clinical outcomes to validate our findings would be ideal but has not been possible at this time given the in-depth characterization and length of follow up required in a unique study population.

A significant proportion of samples had undetectable miRNA levels. This may be due to experimental procedure versus low levels of miRNA expression. For the former, we attempted to minimize miRNA profiling errors by having samples analyzed by a high-throughput gene expression laboratory with significant experience in these procedures. The missing data more likely reflects lower expression values as miRNAs that were missing data tended to have a lower expression value on average. Missing miRNA data was also associated with sex and BMI, demographics which are known to affect miRNA levels.[67] Studies that profiled miRNAs using similar techniques have also frequently found low miRNA levels in blood samples resulting in the exclusion of a significant proportion of samples and miRNAs from final analyses.[26,27,60] We attempted to mitigate the effects of missing data using a form of multiple imputation that accommodates violations of the missing not at random (MNAR) assumption, and we generated a large number of imputations to increase efficiency and reliability of the estimates. Additional prospective studies with larger numbers of patients would be useful to confirm our findings.

Conclusion

We identified miRNAs that were differentially expressed between HIV infected individuals who experienced Type 1 MIs compared to HIV infected individuals who did not. MiR-125a-5p and miR-139-5p were significantly associated with incident MI and provided improvement in risk prediction when compared to using the Framingham Risk Score alone. MiR-409-3p, miR-342-3p, and miR-382-5p were also significant markers when modeled simultaneously with miR-125a-5p and/or miR-139-5p. MiR-125a-5p has been previously characterized as being involved in the activation and regulation of macrophages, which is critical in the development of atherosclerotic plaques and is also implicated in cardiovascular disease development in HIV. The other miRNAs identified in this study have less well described relationships with ischemic heart disease and may be novel markers predictive of MI in HIV positive individuals. All 5 markers are differentially expressed in HIV positive individuals compared with HIV negative individuals, and both miR-125a-5p and miR-382-5p are known to possibly play a role in host suppression of HIV viral replication. Our findings demonstrate that miRNAs can potentially serve as biomarkers predictive of Type I MI risk in HIV infected individuals and may be involved in unique processes due to HIV infection that contribute to the development of cardiovascular disease in this population.