A Whole Blood Molecular Signature for Acute Myocardial Infarction

Chest pain is a leading reason patients seek medical evaluation. While assays to detect myocyte death are used to diagnose a heart attack (acute myocardial infarction, AMI), there is no biomarker to indicate an impending cardiac event. Transcriptional patterns present in circulating endothelial cells (CEC) may provide a window into the plaque rupture process and identify a proximal biomarker for AMI. Thus, we aimed to identify a transcriptomic signature of AMI present in whole blood, but derived from CECs. Candidate genes indicative of AMI were nominated from microarray of enriched CEC samples, and then verified for detectability and predictive potential via qPCR in whole blood. This signature was validated in an independent cohort. Our findings suggest that a whole blood CEC-derived molecular signature identifies patients with AMI and sets the framework to potentially identify the earlier stages of an impending cardiac event when used in concert with clinical history and other diagnostics where conventional biomarkers indicative of myonecrosis remain undetected.


Circulating microparticle (CMP) isolation and enumeration:
Whole blood in EDTA tubes from healthy control and AMI subjects was centrifuged at 1500 x g with the plasma phase separated and immediately aliquoted and frozen at -80C. Prior to CMP enumeration, 50 uL aliquots were prepared with 5x SYBR Green I double stranded DNA dye, according to labeled instructions. The sample was loaded into a flow cell and the chip was electrified using an AC function generator (Biological Dynamics, San Diego, CA) set at 7Vp-p (peak to peak), 10kHz for 10 min. 1 . An image was acquired of a ~6x6 microelectrode section of the chip (~1.2mm x 1.2mm) after 10 minutes using a fluorescent microscope with a CCD camera.
A total of 100-200 uL of plasma from each patient was run (50 uL per run, 2-4 chips per sample) after which the number of particles was determined from each image. The number of particles was averaged per chip run (50 uL) and then normalized to a particle count per mL of sample.

Microarray sample preparation:
Enriched CEC-derived RNA was isolated using Trizol Reagent (Life Technologies, Carlsbad, CA) according to the manufacturer's instructions.
Glycogen (Life Technologies) was added to each sample during the RNA extraction to assist in visualization of the RNA pellet. RNA was extracted from whole blood samples similarly, however without the addition of glycogen. Isolated RNA from each sample was quantified using a  Figure 1). We then removed probe sets that are up-regulated in inflammatory diseases. The remaining probe sets were mapped to HGNC gene symbols. If multiple probe sets mapped to the same gene symbol, the probe set with the highest inter-quartile distance was kept for further analysis. We then used the discovery set to calculate fold changes for each probe set.
Probe sets with a fold change less than two were removed. We used elastic net regression and the glmnet package in R to build a predictive model for acute myocardial infarction using the microarray data 3 . Parameters for the elastic net were as follows: alpha of 0.5, pmax of 20, binomial family, and a logistic link function. The model was trained using the discovery set and then predictions were made for the independent validation set. The performance of the model on the discovery and validation sets was evaluated using receiver-operator characteristic curves and the pROC package in R 4 . A differential expression analysis was run on the validation and discovery sets using the limma package in R. For each probe set, a linear model was trained to predict acute myocardial infarction. P-values were calculated using an empirical Bayesian method, which were adjusted using the Bonferroni correction 5 . A gene set enrichment analysis was run on the combined set of discovery and validation samples 6 . For the GSEA, each probe's log fold change was used as the ranking statistic, and the GSEA was set to the "classic" mode. All microarray data are available from the Gene Expression Omnibus database City, CA). The cDNA was amplified with the ABI TaqMan PreAmp method (Applied Biosystems) and reagents according to the manufacturer's instructions. The selected candidate genes and the housekeeping control gene (GAPDH) were evaluated using the qRT-PCR assay with the pre-amplified material. PCR amplification was performed on the Bio-Rad real-time PCR Detection system (Life Science Research, Hercules, CA) using the 96-well block format with a 25-µl reaction volume. The concentration of the primers and the probes was 9 and 2.5 µmol/l, respectively. The reaction mixture was incubated at 95˚C for ten minutes to activate AmpliTaq®, followed by 40 cycles at 95˚C for 15 sec for denaturing and at 60˚C for 45 sec for annealing and extension. All TaqMan® Assay primer and probe sets were purchased from Applied Biosystems sequences of which are available by request. PCR data of Ct values were exported for further analysis. The results were considered valid when the Ct value of GAPDH was ≤30 for enriched CEC samples and ≤20 for whole blood samples as well as when no template control had undetectable Ct. By using this threshold, five of the 60 enriched CEC RNA samples (8.3%) and three of the 76 whole blood RNA samples (4%) were excluded from further analysis (GAPDH cut off = mean Ct ± 2SD). ΔCts normalized by GAPDH were applied for all data analysis. An elastic net model was trained using the qPCR data to predict acute myocardial infarction 3 . Parameters for the elastic net were: alpha of 0.5, binomial family, a logistic link function and a lower limit of zero. The performance of the model was evaluated using leave-one-out cross validation and the receiver-operator characteristic curve.