Cardiovascular disease is the most common cause of death in developed countries, despite a substantial reduction in age-adjusted rates of death from cardiovascular causes over the past decades. Of all cardiac deaths, approximately 50% are sudden. Sudden cardiac death – cardiac arrest (CA) – is the most common lethal manifestation of heart disease [1-3].

Therapeutic hypothermia (TH) has been shown in randomized trials to improve outcome of patients resuscitated from out-ofhospital ventricular fibrillation [4,5]. Induction of TH after return of spontaneous circulation has been associated with improved functional recovery and a clear benefit in neurologic outcome and mortality [4,5]. Indeed, hypothermia decreases cerebral oxygen demand and thus provides protection from ongoing cerebral ischemia. TH also reduces the glutamate level, the production of free radicals [6], and the intracranial pressure [7].

Today, few methods exist to predict outcome in the early stage after CA. Biochemical markers of brain injury like S-100B and serum neuron-specific enolase (NSE) have been used to predict neurological outcome. However, their predictive value was determined in trials without hypothermia [8-11] and was found to be of limited benefit in patients treated with TH [12,13]. Therefore, there is an unmet clinical need to find new prognostic biomarkers after CA.

We hypothesized that gene expression profiles from circulating blood cells may be used to identify new prognostic biomarkers after CA.

Patients

All adult comatose patients (Glasgow coma score <8) successfully resuscitated from CA admitted to our institution, were included in this study, regardless of the cause of CA and the initial rhythm. Exclusion criteria were age under 18 years, no therapeutic hypothermia performed, terminal disease and traumatic cardiac arrest. The study protocol was approved by the National Ethics Committee and informed consent was obtained from the next of kin. This was a single-centre, prospective, observational study. All data were recorded according to the Utsteinstyle template [14].

Treatment of the patients was performed according to standard protocols. All patients were treated with TH as soon as possible [15] after arrival in the catheterization laboratory or in the intensive care unit (ICU). Target temperature was 33 ± 1°C. Induction of hypothermia was done either by using cold IV fluids [16] or an ivcooling device (Coolgard^® with Icy-cath^® ALSIUS Corp., CA, USA) [17]. After 24 h, rewarming was started at a maximum rate of 0.5°C h^-1. Core temperature was controlled by an indwelling bladder catheter and a second, rectal, probe serving as control. Sedation was done by a standardized protocol using Midazolam (max. 0.2 mg kg^-1 h^-1) and Fentanyl (max. 1.5 μg kg^-1 h^-1).

At the end of the ICU stay, neurological examination was performed using the Glasgow coma scale and the cerebral performance category score (CPC) [18]. CPC1 and CPC2 were regarded as good outcome, whereas CPC3, 4 and 5 were considered as poor outcome. At 6 months, a clinical neurological examination was done by physicians uninvolved in the ICU management of the patient. At the end of hypothermia, blood samples were drawn in PAXgene™ tubes (BD Biosciences, Erembodegem, Belgium) and tubes containing a clot activator (BD Biosciences). Total RNA extracted from PAXgene™ tubes was used to study gene expression by microarrays and PCR. NSE was measured in serum using an electrochemiluminescent immunoassay (ECLIA, Roche, Basel, Switzerland) performed on the on the Cobas e601 analyzer ^® (Roche, Basel, Switzerland) with a detection limit of <0.05ng/ml.

Microarrays

Transcriptomic analysis of blood cells was performed as previously described [19,20]. Total RNA was extracted from PAXgene™ tubes using the PAXgene™ blood RNA kit (Qiagen, Venlo, Netherlands). RNA was subjected to additional purification and concentration with the RNeasy^® MinElute™ kit (Qiagen). RNA quantity was measured with the ND-1000 spectrophotometer (NanoDrop^® Technologies, Wilmington, USA). RNA quality was evaluated with the 2100 Bioanalyzer^® apparatus (Agilent Technologies, Massy, France) and RNA 6000 Nano chips. Only high quality RNA (OD₂₆₀/OD₂₈₀ > 1.9 and OD₂₆₀/OD₂₃₀ > 1.7) and undegraded RNA was retained for further analysis. The Universal Human Reference RNA (Stratagene Europe, Amsterdam, The Netherlands) was used in conjunction with patient’s RNA to provide an internal reference for comparisons of relative gene expression levels across arrays. Messenger RNAs were amplified with the Amino Allyl MessageAmp™ kit (Ambion^®, Cambridgeshire, United Kingdom), starting with 1μg of total RNA. 5μg of amino allyl aRNA were labeled with Cy³ or Cy⁵ (Amersham, Buckinghamshire, United Kingdom). Dye coupling was evaluated using the ND-1000 spectrophotometer. Coupling yield >5% was a prerequisite for further analysis. 750ng of each amino allyl aRNA labeled Cy³ or Cy⁵ (reference RNA or patient RNA) were mixed and hybridized on long oligonucleotide microarrays covering 25,000 genes [21]. 4 microarray replicates and a dye-swap were performed. Scanning was achieved 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 MAIA^® software [22]. A Lowess non linear normalization was completed and genes not present in at least 3 microarrays of 4 were filtered out. Data are available at the Gene Expression Omnibus (www.ncbi.nlm.nih.gov/ geo/) under the accession number GSE29540.

Prior to statistical analysis, genes not present in at least 50% of the patients were removed. Supervised analysis was achieved with the Significance Analysis of Microarrays (SAM) software [23] using 2-class unpaired test and 100 permutations. Classification value of microarrays data was evaluated using Prediction Analysis of Microarrays (PAM) software [24]. The Database for Annotation, Visualization and Integrated Discovery (DAVID) [25] and the GOMIner resource [26] were used for functional analysis of microarray data. Heat maps were obtained with Cluster 3.0 and Java Treeview [27].

Quantitative PCR

Total RNA extracted from PAXgene™ tubes using the PAXgene™ blood RNA kit (Qiagen) was subjected to a second step of purification and concentration with the RNeasy^® MinElute™ kit (Qiagen). 1μg of total RNA was reverse-transcribed using the Superscript^® II Reverse Transcriptase (Invitrogen, Merelbeke, Belgium). PCR primers were designed using the Beacon Designer software (Premier Biosoft, Palo Alto, USA) and were chosen to encompass an intron. PCR was performed using the iCycler^® and the IQ^TM SYBR^® Green Supermix (Biorad, Nazareth, Belgium). 1/10 dilutions of cDNA were used. PCR conditions were as follows: 3 min at 95°C, 30 sec at 95°C and 1 min annealing (40-fold). Optimal annealing temperature was determined for each primer pair. Melting point analysis was obtained after 80 cycles for 10 sec from 55°C up to 95°C. Each run included negative reaction controls. GAPDH was used as housekeeping genes for normalization. Expression levels were calculated by the relative quantification method (ΔΔCt) using the Genex software (Biorad) which takes into account primer pair efficiency.

Statistical analysis

The primary end-point of the study was the neurological status at 6 months after resuscitation. The secondary end-point was mortality. The ability of CX3CR1 to predict end-points was evaluated using receiveroperating characteristic (ROC) curves with the corresponding area under the curve (AUC). Multiple logistic regression models proceeding by backward elimination were used to identify clinical covariates with predictive value. Logistic regression models proceeding with 10-fold cross-validation were used to determine the added value of multiple clinical covariates.

Kaplan-Meier curves and the LogRank statistic were used for survival analysis, in which patients were distributed in tertiles of biomarker values. To isolate the tertiles that differ from each other, all pairwise multiple comparison procedure was applied (Holm-Sidak method).

Data normality was assessed using the Shapiro-Wilk test. Comparisons between 2 groups were performed using t-test or Mann- Whitney Rank Sum test. Fisher’s exact test was used for categorical variables. Correlations were determined using the Spearman test. Statistical tests were performed with the Sigma Plot v11.0 software.

Patients

An initial cohort, named “microarray cohort”, consisted of 35 patients. At 6 months post resuscitation, 10 patients had a CPC of 1, 11 patients had a CPC of 2, 3 patients had a CPC of 3, 47 patients had a CPC of 4, and 7 patients died during hospital stay or follow-up (CPC 5). Patients were dichotomized into good neurological outcome, as attested by CPC 1-2 (n=21), and poor neurological outcome, as attested by CPC 3-5 (n=14).

A second group of patients, named “PCR cohort”, was formed with the 35 patients of the microarray cohort and 10 additional patients (n=45): 24 patients had a CPC of 1, 3 patients had a CPC of 2, 2 patients had a CPC of 3, 1 patient had a CPC of 4 and 15 patients died during hospital stay or during the follow-up period. Thus, 27 patients had a favorable neurological outcome at 6-months and 18 patients had a poor outcome.

Detailed demographic and key Utstein data are shown in Table 1. Both cohorts were comparable regarding age and gender, but men appeared to have better outcome. In both groups, a shockable rhythm favored good outcome. Not surprisingly, the time to return of spontaneous circulation (ROSC) was significantly higher in bad outcome groups and the percentage of patients in shock was higher in the PCR cohort.

Table 1: Main patient characteristics and Utstein data.

Gene biosignature after CA

Gene expression profiles of whole blood cells obtained at the end of hypothermia in the 35 patients of the microarray cohort were characterized by microarrays. Among the 25000 genes present on the arrays, 14986 were retained for analysis using the selection criteria described in the Methods section. Statistical analysis of microarrays using SAM software identified 582 genes differentially expressed between patients with favorable outcome and patients with poor outcome with a fold-change >1.3 and a false discovery rate <5%. 70 genes were up-regulated in poor outcome group and 512 genes were upregulated in good outcome group (Figure 1). Therefore, neurological outcome after CA is associated with specific gene expression profiles in blood cells.

Figure 1: Transcriptomic biosignature after CA. Microarrays were used to compare the gene expression profiles of blood cells from 21 patients with good neurological outcome (CPC 1-2) and 14 patients with poor neurological outcome (CPC 3-5) 6 months after CA. Represented is a heat-map of 582 genes differentially expressed between good outcome and poor outcome groups with a fold-change above 1.3 and a false discovery rate <5%. 70 genes were up-regulated in poor outcome group and 512 genes were up-regulated in good outcome group. Differential expression was determined using SAM software.

Functional analysis

The functional classification tools GOMiner and DAVID were used to retrieve the biological significance beyond microarray data. Several biological processes linked to apoptosis, cell death and neuronal death were enriched by genes up-regulated in poor outcome group (Supplementary Table 1). On the other hand, genes up-regulated in the good outcome group enriched biological processes related to energy production (Supplementary Table 2). Pathway analysis using DAVID revealed an enrichment of the “apoptosis” pathway by genes up-regulated in the poor outcome group (Supplementary Table 3). The pathway “oxidative phosphorylation” was enriched by genes upregulated in the good outcome group (Supplementary Table 4). These observations strengthened our hypothesis that gene expression of blood cells may reflect neurological damage induced by CA.

Selection of genes with predictive value

We used the PAM software to identify genes and groups of genes able to predict neurological outcome after CA. Several groups of genes were identified and one gene was included in all groups, CX3CR1 (Table 2). CX3CR1 is the chemokine (C-X3-C motif) receptor 1, a member of the superfamily of chemokines and their receptors involved in leukocyte migration and adhesion. CX3CR1 provided a high sensitivity (90%) and a low specificity (14%) (Table 2). We focused further investigations on CX3CR1.

Table 2: Prediction capacities of groups of genes.

Prognostic value of CX3CR1 in the microarray cohort

Prognostic performance of CX3CR1 was first evaluated with microarray data from 35 patients. As shown in Figure 2A (left panel), CX3CR1 mRNA expression was higher (2-fold) in patients with favorable neurological outcome (CPC 1-2) compared to patients with poor outcome (CPC 3-5). SAM software attributed a q value of 0% to this difference, and a standard t-test (no correction for multiple testing) generated a P value <0.001. ROC curves analysis reported an AUC of 0.92 (95% CI [0.83; 1.00]) for CX3CR1 to predict outcome (Figure 2A right panel). A cut-off value of 1.6 predicted good outcome with a sensitivity of 95% (95%CI [76,100]) and a specificity of 64% (95%CI [35, 87]).

Figure 2: Prognostic value of CX3CR1. Expression levels of CX3CR1 were determined by microarrays in 35 patients (A) and by quantitative PCR in 45 patients (B). Left panels: box-plots showing CX3CR1 expression in patients with favorable neurological outcome (CPC 1-2) and patients with poor outcome (CPC 3-5). The lower boundary of the box indicates the 25th percentile, the line within the box marks the median, and the upper boundary of the box indicates the 75th percentile. Whiskers (error bars) above and below the box indicate the 90th and 10th percentiles. Indicated P values are from t-tests. Right panels: ROC curves showing the value of CX3CR1 to predict neurological outcome. AUCs are indicated.

Prognostic value of CX3CR1 in the PCR cohort

Then we measured CX3CR1 expression in 45 patients by quantitative PCR. As shown in Figure 2B, CX3CR1 expression was increased in patients with good outcome (1.6-fold, P=0.003) and predicted outcome with an AUC of 0.76 (95% CI [0.60; 0.92]). A cutoff value of 0.52 (arbitrary PCR units) predicted good outcome with a sensitivity of 85% (95%CI [66; 96]) and a specificity of 72% (95%CI [46; 90]). These results confirm the prognostic value of CX3CR1 observed in the microarray cohort.

Figure 3: Survival analysis. Kaplan-Meier curves showing the mortality of patients from the PCR cohort (n=45) according to CX3CR1 expression, as dichotomized by the cut-off value of 0.52 determined by ROC curves (A) or as divided in tertiles (B). P values are indicated.

Prediction of survival

The ability of CX3CR1 to predict survival after CA was evaluated using Kaplan-Meier curves. Patients with CX3CR1 values below the cut-off value of 0.52 had an elevated risk of death (P<0.001, Figure 3A). When divided into tertiles of CX3CR1 values, patients within the first tertile had an elevated risk of mortality (Figure 3B). No patients within the second tertile died. Interestingly, patients within the third tertile had an intermediate risk of death.

Clinical predictors of survival

A multiple logistic regression proceeding by backward elimination was used to identify clinical covariates that predict survival. The following covariates were tested: CX3CR1, neuron-specific enolase (NSE), age, gender, SAPS II, time CA to ROSC, history of coronary artery disease, heart failure, hypertension, and acute myocardial infarction (MI) at the time of CA. Gender, history of coronary disease, heart failure, and hypertension were rapidly removed from the model and did not predict survival. Age, SAPS II and time CA to ROSC were moderately associated with survival. CX3CR1, NSE and acute MI at the time of CA were retained in the final model and were independent predictors of survival (P=0.02, P<0.001 and P=0.02, respectively). Patients presenting with acute MI had a 2-fold increased risk of death compared to patients with CA not consecutive to acute MI.

Additive prognostic value of CX3CR1

We addressed the ability of CX3CR1 to improve the prognostic value of NSE and acute MI at the time of CA. Using logistic regression, we observed an incremental improvement of the prediction of survival when the covariate acute MI at the time of CA and CX3R1 were added to NSE (Table 3). Combined use of the 3 covariates yielded an AUC of 0.84, with a true positive rate of 82% and a false positive rate of 26%.

Table 3: Performance of combined clinical covariates to predict survival.

Correlation between CX3CR1 and inflammation

Finally, we determined the correlation between CX3CR1 expression and markers of inflammation. There was no significant correlation with white blood cell count (R= -0.01, P=0.96) and procalcitonin (R=-0.02, P=0.90). CX3CR1 was inversely correlated with C-reactive protein concentration (R=-0.31, P=0.04).

In this preliminary study, we report for the first time the biosignature of circulating blood cells after CA. This biosignature allowed identifying CX3CR1 as a potential prognostic biomarker.

Transcriptomic analysis of blood cells from CA patients indicated that patients with favorable neurological outcome may be distinguished from patients with poor outcome. Data-mining approaches revealed significant associations between genes up-regulated in blood cells from patients with poor neurological outcome and biological processes and pathways linked to the death of neurons and cells in general. In addition, genes up-regulated in blood cells from patients with favorable outcome were associated with energy production.

The chemokine CX3CL1 and its receptor CX3CR1 are constitutively expressed by the microglial tissue and contribute to its protection following ischemic brain injury [28]. Our results showing elevated expression of CX3CR1 in patients with favorable neurological outcome also support a neuroprotective role of CX3CR1 after CA. The exact mechanisms of this effect are unknown.

Since CX3CR1 expression was measured in blood cells, the differential expression of CX3CR1 could be consecutive to changes in the inflammatory status of patients. There was no significant correlation between CX3CR1 and white blood cell count, suggesting that CX3CR1 expression is not directly dictated by the number of blood cells. However, CX3CR1 expression was higher in patients with favorable outcome and inversely correlated with levels of C-reactive protein. This indicates that the inflammatory response which is activated after CA and cardiopulmonary resuscitation [29,30] (“sepsis-like syndrome” [31]) may be detrimental.

The chemokine (C-X3-C motif) receptor 1 is a G protein-coupled receptor of the superfamily of chemokine receptors. Two forms of its ligand, CX3CL1 or fractalkine, exist. One form is soluble and resembles a conventional chemokine with chemotactic activity on monocytes, natural killer cells and T cells, and the other form is bound to the membrane and acts as adhesion molecule promoting the retention of leukocytes. The complex CX3CR1/CX3CL1 is involved in inflammatory diseases (glomerulonephritis, rheumatoid arthritis) [32,33], human immunodeficiency virus infection [34], cardiac allograft rejection [35]. A common variant of CX3CR1 is associated with a reduced risk for coronary events [36]. The role of CX3CR1 in the pathogenesis of heart failure has been suggested [37]. There is therefore strong evidence that activation of CX3CR1 contributes to several cardiac and noncardiac diseases. To our knowledge, this study is the first to report an association between CX3CR1 and outcome after cardiac arrest.

We investigated whether CX3CR1 may provide an additive prognostic value to NSE. Indeed, NSE is altered after TH and its use as a prognostic biomarker in CA patients treated by TH is questionable [12,13]. We observed that CX3CR1 improves the prediction of survival by NSE. Among the clinical covariates tested, acute MI was a significant predictor of outcome. Thus, the combination of high levels of NSE, the presence of an acute MI and low levels of CX3CR1 were predictive of a high likelihood of death.

The identification of biomarkers capable of predicting survival and good neurological recovery after CA could be significant for the management of these critically ill patients. Indeed, in the era of personalized medicine, it is generally accepted that patients may benefit from the use of biomarkers. Nevertheless, it is recommended that management decisions should not rely on a single prognostic parameter [38]. Although the benefit of multi-marker panels has been demonstrated in several acute cardiac conditions, such as acute myocardial infarction [39], the added value of multiple biomarkers over single biomarkers has not been thoroughly investigated in CA patients. Identifying new biomarkers and combining their measurements using multi-marker panels may improve the accuracy of existing markers. Almost all current biomarkers after CA are elevated in patients with poor outcome. The added value of CX3CR1 may in part be due to the fact that it is elevated in patients with good outcome.

To become clinically relevant, biomarkers have to be easily and quickly measurable. In the present study, CX3CR1 was quantified by PCR, a technique that is time-consuming but which may become faster and automated in the future.

Future studies on large populations of cardiac arrest patients are needed to determine: (1) the kinetics of biomarkers after hypothermia, (2) the effect of hypothermia on biomarker levels, (3) the cut-offs for optimal prediction, (4) the added value of combined biomarkers, (5) the effect of biomarker determination on patient care and outcome.

This study is limited by a small population size. In addition, the PCR cohort used for validation purposes was not an independent cohort since it enrolled the patients of the microarray cohort and 10 additional patients.

In conclusion, we have identified CX3CR1 as a new candidate prognostic biomarker after CA which may be further evaluated in larger patient populations.

We thank Bernadette Leners, Malou Gloesener and Loredana Jacobs for expert technical assistance. This study was supported by grants from the Society for Research on Cardiovascular Diseases, the Ministry of Culture, Higher Education and Research, and the National Funds of Research of Luxembourg. M.K. was recipient of a fellowship from the National Funds of Research of Luxembourg (grant # TR-PDR-BFR07-145).