Blood signature of pre-heart failure: a microarrays study

Abstract

Background: The preclinical stage of systolic heart failure (HF), known as asymptomatic left ventricular dysfunction (ALVD), is diagnosed only by echocardiography, frequent in the general population and leads to a high risk of developing severe HF. Large scale screening for ALVD is a difficult task and represents a major unmet clinical challenge that requires the determination of ALVD biomarkers.

Methodology/principal findings: 294 individuals were screened by echocardiography. We identified 9 ALVD cases out of 128 subjects with cardiovascular risk factors. White blood cell gene expression profiling was performed using pangenomic microarrays. Data were analyzed using principal component analysis (PCA) and Significant Analysis of Microarrays (SAM). To build an ALVD classifier model, we used the nearest centroid classification method (NCCM) with the ClaNC software package. Classification performance was determined using the leave-one-out cross-validation method. Blood transcriptome analysis provided a specific molecular signature for ALVD which defined a model based on 7 genes capable of discriminating ALVD cases. Analysis of an ALVD patients validation group demonstrated that these genes are accurate diagnostic predictors for ALVD with 87% accuracy and 100% precision. Furthermore, Receiver Operating Characteristic curves of expression levels confirmed that 6 out of 7 genes discriminate for left ventricular dysfunction classification.

Conclusions/significance: These targets could serve to enhance the ability to efficiently detect ALVD by general care practitioners to facilitate preemptive initiation of medical treatment preventing the development of HF.

Figure 1. Flow chart of recruitment protocol involving 294 subjects and overall study design.

Healthy volunteers (HI) were recruited from the general population, individuals with cardiovascular risk factors (RF) were from the atherosclerosis prevention center and patients with chronic heart failure (CHF) were recruited from the cardiology department at Rangueil Hospital, Toulouse. All subjects underwent transthoracic echocardiography for left ventricular ejection fraction (LVEF) assessment. We used a threshold value of LVEF<45% to sort individuals into 4 groups: HI (light blue), RF with LVEF≥45% (dark blue), ALVD (red), and CHF (orange) with LVEF<45%. We identified 9 ALVD cases out of the 128 subjects tested with cardiovascular risk factors. We used the set of cardiovascular risk factors (age, gender, arterial hypertension, diabetes, obesity, dyslipidemia and heredity) based on the characteristics of the ALVD subjects to match the study groups (n = 9). White blood cell gene expression profiling was performed using pangenomic microarrays for all 4 groups. Data were statically analyzed using unsupervised primary component analysis (PCA) and by Significance Analysis of Microarrays (SAM software) which defined the false discovery rate. Then to build an ALVD classifier model, we used the nearest centroid classification method (NCCM) with the ClaNC software package. Classification performance was determined using the leave-one-out cross-validation method. Expression levels of 7 genes capable of discriminating ALVD were compared between the 4 groups and each gene's capability to discriminate patients with LVEF<45% was evaluated using Receiver Operating Characteristic (ROC) analysis.

Figure 2. Unsupervised primary component statistical analysis (PCA) of blood transcriptome data reveals a molecular signature for ALVD.

Three dimensional plot of the three first components(PC1, PC2, PC3) of the blood gene expression data from healthy subjects (HI, light blue), cardiovascular risk factor individuals (RF, dark blue), individuals with asymptomatic left ventricular dysfunction groups (ALVD, red) and chronic heart failure patients (CHF, orange). These three components can classify subjects according to their group and distribute the subjects in grouped locations in the defined space. Numbers in colors indicate subjects' identities. The relative expression level used is defined by the ratio obtained with the tested sample to the signal obtained using the common reference, an equimolar mix of all the RNA used to generate a reference signal.

Figure 3. Expression levels for 7 genes discriminate the ALVD group.

A. a–g Relative expression levels of the 7 genes, sorted by the nearest centroid classifier, are assessed for HI (light blue), RF (dark blue), ALVD (red) and CHF (orange) groups. The relative expression level used is defined by the ratio obtained with the tested sample to the signal obtained using the common reference, an equimolar mix of all the RNA used to generate a reference signal. The box plot presents the median, lower and upper quantiles (25^th, 75^th percentiles) lower and upper whiskers represent the 10th and 90th percentiles. * P<0.05 where indicated, estimated by one-way ANNOVA. B. Receiver-operating characteristic (ROC) analysis of ALVD discriminant genes using HI with RF (LVEF≥45% as disease free) and ALVD with CHF (LVEF<45% as left ventricular dysfunction) groups. Area under curve (AUC), confidence interval and P values to find an AUC value of 0.5 (null hypothesis) for each gene are depicted in Table 5 . ROC curves for each of the 7 genes are displayed on a single figure. With the exception of SLC43A2, ALVD discriminant genes are also CHF biomarkers i.e. left ventricular dysfunction biomarkers.