Transforming growth factor β receptor 1 is a new candidate prognostic biomarker after acute myocardial infarction

Patient selection and characteristics of the test cohort

Transcriptomic analysis of blood cells

Gene expression profiles of whole-blood cells were obtained using 25,000 genes microarrays [14]. 525 genes were found differentially expressed between high and low EF patients with a 1.3-fold change threshold. Of note, the 5% threshold for statistical significance was reached for 54% of these genes. 226 genes were up-regulated in the high EF group and 299 were up-regulated in the low EF group. Using gene set enrichment analysis (GSEA), we retrieved the 50 genes most significantly associated with one or the other group of patients. The heat-map drawn with the expression data of these 50 genes shows that LV function assessed 4-months after MI is associated with a specific biosignature of blood cells obtained the day of MI (Figure 1A).

Angiogenic genes associated with clinical outcome after MI

There is ample evidence that angiogenesis plays a significant role in LV remodeling after MI. We therefore aimed to identify among the 525 genes differentially expressed between high and low EF patients the genes related to angiogenesis. For this purpose, we retrieved from the Entrez Gene database a list of 494 genes known to be related to angiogenesis in humans with the following query: "angiogenesis" AND "homo sapiens". We assigned to these 494 genes the expression values of the test cohort obtained by microarrays. 28 genes were found differentially expressed: 20 genes were up-regulated in the low EF group and 8 genes were up-regulated in the high EF group (Figure 1B).

Building of a network of protein-protein interactions

We implemented an interaction network of proteins know to be involved in angiogenesis, a process involved in LV remodeling. For this purpose, we retrieved from the Human Protein Reference Database all known interactions between the 28 proteins encoded by the 28 genes differentially expressed between the 2 groups of patients from the test cohort (Figure 1B). The network built with these interactions is represented in Figure 2A, and contains 441 nodes (proteins) and 458 edges (interactions). It has to be noted that the proteins PBEF1 and PFKFB3 did not have known protein-protein interaction. Then we used MCODE algorithm to identify clusters of interconnected proteins. This was based on the assumption that highly interacting proteins may have important biological functions. Nine clusters from 9 to 35 proteins were identified (Table 2).

Prognostic performances of clusters of angiogenic genes

Independent validation

To confirm the results obtained from the test cohort, we measured the expression of TGFBR1, PTK2 and ITK by quantitative PCR in blood cells from a validation cohort of 115 acute MI patients. Demographic features of these patients are shown in Table 1. As for the test cohort, blood samples were obtained at presentation and follow-up was performed at 4-months by echocardiography. Of the 3 candidate genes, TGFBR1 was the most robustly associated with LV function. PTK2 and ITK were not significantly correlated with the EF (data not shown). TGFBR1 expression was inversely correlated with the EF (r = -0.28, P = 0.003) and was higher in patients with LV dysfunction (Figure 3A). Linear regression analyses attested that TGFBR1 was associated with LV dysfunction at 4-months (P = 0.003). The serum markers creatine phosphokinase and troponin T predicted TGFBR1 expression (P = 0.03 and P = 0.003 by linear regression, respectively), suggesting that TGFBR1 is responsive to infarct size. All together, these results show that TGFBR1 expression is associated with LV function in an independent cohort.

Prognostic performance of TGFBR1

In the validation cohort, 7 patients (6%) had a previous MI (Table 1). In the 108 patients who presented with a first MI, the predictive value of TGFBR1 had an AUC of 0.72, identical to the AUC reported for the whole validation cohort (Figure 3B). Therefore, although it is expected that a population of patients with first infarcts would better reflect post-MI LV remodeling, the presence of few patients with a second infarct did not affect the predictive value of TGFBR1.

Together, these data confer a significant prognostic value to TGFBR1.

Prognostic performance of TGFB1

We then evaluated the ability of TGFB1, the ligand of TGFBR1, to predict LV dysfunction. TGFB1 was measured in the plasma obtained at presentation in the 115 patients of the validation cohort. As observed for TGFBR1, TGFB1 was inversely correlated with EF (r = -0.20, P = 0.04) and was over-expressed in patients with LV dysfunction (Figure 3D). Linear regression revealed a significant association between TGFB1 and LV dysfunction (P = 0.006) and ROC curves reported a modest prognostic value with an AUC of 0.66 (Figure 3E). Adding TGFB1 to TGFBR1 did not result in a statistically detectable improvement in prediction (not shown). Therefore, TGFB1 is also associated with LV function but does not provide an additive value to TGFBR1.

The TGFB1-TGFBR1 axis in the heart

To investigate the physiological relevance of our findings, we subjected rats to LAD-occlusion, a standard animal model of MI. The amount of total TGFB1 in the heart was increased 2 months after MI (Figure 4A). However, the amount of activated TGFB1 was not increased (not shown), suggesting that the up-regulation of TGFB1 expression in the heart of MI animals reflects an increase of the latent form. Total TGFB1 expression did not significantly correlate with LV function, as assessed by the EF at 2-months (Figure 4B). Interestingly, there was a positive correlation between TGBF1 and LV remodeling, as assessed by the variation of end-diastolic and end-systolic volumes between 48 hours and 2 months after MI. Rats with the largest degree of remodeling (increased LV volumes) had the highest levels of TGFB1 (Figure 4C-D).

Faintly expressed in the heart of healthy animals (sham), TGFBR1 was up-regulated in the border zone but not in the remote zone 2-months after MI (Figure 4E). TGFBR1 was highly expressed 2 weeks after MI, but not after 2 days (Figure 4F). Trichrome Masson staining showed the presence of collagen deposition and fibrosis in the necrotic part of the heart after 2 months (Figure 4G). Therefore, the TGFB1-TGFBR1 axis is activated in cardiac tissue after experimental MI in rats and correlates with LV remodeling.

Patients

Patients with acute MI were enrolled in a national MI registry and treated with primary percutaneous coronary intervention. Acute MI was defined by the presence of chest pain < 12 hours, significant ST elevation (> 1 mm), completely occluded major coronary artery (TIMI O flow in LAD, CX or RCA), peak (≈ 24 hours) creatine phosphokinase (CPK) level > 600 units/L and peak troponin T (TnT) level > 0.03 ng/mL. Blood samples obtained shortly (5 min) after mechanical reperfusion, via an arterial catheter and into PAXgene™ tubes (BD Biosciences, Erembodegem, Belgium), were used for determination of TGFBR1 expression. TGFB1 was measured in plasma extracted from citrated tubes collected after mechanical reperfusion. Time to reperfusion was recorded by dedicated nurses as the delay between chest pain onset evaluated by the patient and presentation. CPK activity was assessed with a Roche IFCC method on a Cobas c501 instrument (Roche, Prophac, Luxembourg). TnT was assessed with a 4^th generation assay from Roche performed on a Cobas e601 equipment (Roche). TGF-B1 was measured with the human TGF-B1 ELISA kit (ref DB100B, R&D Systems, Oxon, UK). LV function was assessed by echocardiography 4 months after MI, and LV remodeling was defined by a LV ejection fraction (EF) ≤ 40%. The protocol has been approved by the local ethics committee (National committee of ethics in research of the Grand-Duchy of Luxembourg) and informed consent has been obtained from all subjects.

Microarrays

Transcriptomic analysis of blood cells was performed as previously described [7]. Total RNA was extracted from whole blood cells of MI patients using the PAXgene™ blood RNA kit (Qiagen, Venlo, Netherlands). A second purification and concentration step was performed with the RNeasy^® MinElute™ kit (Qiagen). RNA quantity was measured using the ND-1000 spectrophotometer (NanoDrop^® Technologies, Wilmington, USA). RNA quality was assessed using the 2100 Bioanalyzer^® apparatus (Agilent Technologies, Massy, France) with the RNA 6000 Nano chips. Only high quality RNA (OD₂₆₀/OD₂₈₀ > 1.9 and OD₂₆₀/OD₂₃₀ > 1.7) and un-degraded RNA was considered for further analysis. A common reference RNA (Universal Human Reference RNA, Stratagene Europe, Amsterdam, The Netherlands) was used in conjunction with patient's RNA to provide an internal reference standard for comparisons of relative gene expression levels across arrays. Messenger RNAs were amplified using the Amino Allyl MessageAmp™ kit (Ambion^®, Cambridgeshire, United Kingdom), starting with one μg of total RNA. Five μg of each amino allyl aRNA were labeled with Cy3 or Cy5 (Amersham, Buckinghamshire, United Kingdom). Dye coupling to RNA was measured using the ND-1000 NanoDrop^® spectrophotometer. Coupling yield > 5% was a prerequisite for further analysis. 750 ng of each amino allyl aRNA labeled Cy3 or Cy5 (reference RNA or patient RNA) were combined and hybridized on oligonucleotide microarrays representing 25,000 genes [14]. Four replicates and a dye-swap were performed. Scanning was performed with an Axon 4000B scanner and data were acquired with the GenePix Pro 6^® software (Molecular Devices, Berks, UK). Spot finding and raw data quantification were performed using the MAIA^® freeware. A Lowess non linear normalization was performed and genes that were not present in at least 3 microarrays of 4 were filtered out. Data are available at the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/) under the accession number GSE11947.

Before statistical analysis, genes not present in at least 50% of the patients were filtered out. Supervised analysis was performed using the Significance Analysis of Microarrays software using 2-class unpaired test and 100 permutations. Heat maps were drawn using the Gene Set Enrichment Analysis (GSEA) software [37].

Quantitative PCR

The expression of mRNAs in blood cells of MI patients obtained upon admission was determined by quantitative PCR. Total RNA was extracted from blood cells collected in PAXgene^® blood RNA tubes. PCR was performed using a CFX96 device and the IQ™ SYBR^® Green Supermix (BioRad, Nazareth, Belgium). SF3A1 was used for normalization.

Network of protein-protein interactions

Genes known to be associated with angiogenesis in humans were retrieved from the Entrez-Gene database [38] with the query: "angiogenesis" AND "homo sapiens". The Human Protein Reference Database (release 9) [39] was then interrogated to identify all annotated protein-protein interactions associated with these genes. These interactions were used as inputs to build a network, which was visualized and analyzed with Cytoscape [40]. Clusters of highly interconnected proteins were identified by the MCODE plug-in network clustering algorithm [41].

In vivo experiments and biochemical analyses

Myocardial infarction was induced in 17 anesthetized Wistar rats (Charles Rivers Laboratories; 8-11 weeks; 310-380 g) by permanent occlusion of the left anterior descending (LAD) coronary artery as previously described [42, 43]. An additional 4 rats were sham-operated. All rats were monitored in vivo 48 hours after surgery and after 2 months with the use of a high resolution dedicated small animal positron emission tomography (PET) system with¹⁸F-FDG (IBA, Nancy, France) as a tracer and a premedication by acipimox, as previously described [44]. LV end-diastolic volume (LVEDV) was obtained on contiguous gated short-axis slices by the Quantitative Gated SPECT software [45]. The extent of remodeling was assessed by the change in LVEDV between 48-hours and 2-months after surgery (Δ LVEDV). At sacrifice, blood samples were collected, hearts were harvested, snap frozen and embedded in OCT. Parts of cardiac tissue were rinsed in NaCl and stored at -80°C for protein extraction. These experiments were conducted in accordance with the regulations of the Animal Welfare Act of the National Institutes of Health Guide for the Care and Use of Laboratory Animals (NIH Publication No. 85-23, revised 1996) and protocols were approved by local Ethics Committee and by the Regional Veterinary Department.

For immunohistochemistry and collagen-specific Masson Trichrome staining (Dako, Leuven, Belgium), 8 μm frozen heart sections were generated. Primary antibody used was a rabbit polyclonal anti TGFBR1 antibody (Abcam, Cambridge, UK). Alexa Fluor^®568-coupled donkey anti-rabbit antibodies (Jackson ImmunoResearch, UK) were used as secondary antibodies. Images were recorded on a confocal microscope (Zeiss Laser Scanning Microscope LSM 510) with a 40× objective using the LSM 510 META software. DAPI (blue) staining was used to reveal nuclei.

The TGFB1 ELISA kit (ref MB100B, R&D Systems) was used to measure TGFB1 in cardiac tissue. 50 μg of total proteins from each heart were assayed. This assay detects activated TGFB1 with a sensitivity of 4.6 pg/mL. To measure total TGFB1, latent TGFB1 is activated by acidification with hydrochloric acid prior to assay. Coefficient of variation intra-assay is < 4% and coefficient inter-assay is < 9%.

Statistical analysis

Comparisons between means of two groups were performed with the Mann-Whitney test. Categorical variables were compared using the Fisher exact test. Correlation between biomarker levels and the EF class was estimated with the Spearman test. All tests were two-tailed and were preceded by a normality test (Shapiro-Wilk). A P value < 0.05 was considered statistically significant. The SigmaPlot (v. 11.0) was used to carry-out statistical tests.

Prognostic performances of single genes were evaluated using receiver operating characteristic (ROC) curves and the area under the ROC curves (AUC), as well as linear regression models. Logistic regression models with ridge parameter of 1.0E-8 used to evaluate the predictive performances of multiple markers were implemented and tested using 10-fold cross-validation on the Weka (v. 3.4) data mining platform [46].

Reclassification analyses have been performed to evaluate the additive value of TGFBR1 to TnT and to a mixed clinical model. The net reclassification index (NRI) was calculated as the difference in proportions moving up and down among cases and controls: [Pr(up|case)-Pr(down|case)]-[Pr(up|control)-Pr(down|control)].