Background
Among cardiovascular diseases (CVDs), acute coronary syndrome (ACS) represents the most common cause of emergency hospital admission and it is associated with the highest mortality and morbidity [
1,
2]. The prognosis is directly associated with timely initiation of revascularization, and misdiagnosis or late diagnosis may have unfavorable clinical implications. Established risk stratification tools such as the Global Registry of Acute Coronary Events (GRACE) and the Thrombolysis In Myocardial Infarction risk scores are derived from demographic, clinical, laboratory, and electrocardiogram-related variables [
3,
4]. These do not incorporate the use of newer biomarkers, which could represent different pathophysiologic processes and provide complementary prognostic information, thereby improving risk stratification beyond traditionally used variables.
Several studies have evaluated the potential clinical usefulness of new biomarkers able to identify patients who had a poor outcome. In particular, high levels of inflammatory markers such as C-reactive protein and interleukin-8 had long-term prognostic utility in patients with ACS that undergone coronary revascularization [
5,
6]. However, no conclusive and consistent data about the prognostic utility of measuring inflammatory markers in the early phase of ACS are available in the literature.
A number of studies have evidenced that a global approach, such as genomics, proteomics, or metabolomics, may represent a valid strategy for improving current knowledge about pathophysiological mechanisms and for identifying ACS patients at high risk of secondary atherothrombotic events or premature death.
Metabolomics is the accepted name for the -omic science that deals with the characterization of the metabolome, in turn defined as the whole set of metabolites in a certain biological system, such as a cell, tissue, organ, or entire organism [
7]. The two leading analytical techniques used to perform metabolomics are mass spectrometry (MS) and nuclear magnetic resonance (NMR) spectroscopy. Both techniques yield information about many different molecules in a single measurement, and can be used to determine structures and concentrations of metabolites [
8]. Nevertheless, each technique has its own strengths and limitations. MS overshadows NMR in terms of numbers of compounds resolved (of the order of 10
3 [
9]), with a sensitivity down to the picomolar and requiring a very small volume of the biospecimen; however, reproducibility is still a limitation of MS, which must be overcome by an extensive and time-consuming use of standards and quality control samples. NMR analysis is high-throughput [
7], and NMR data are highly reproducible [
10] and intrinsically quantitative over a wide dynamic range, as demonstrated by numerous ring trials performed by many different NMR laboratories [
10]. NMR gives immediately qualitative and quantitative information on around 10
2 different small molecules present in a biological sample [
11], and has already provided a global picture of a wide range of metabolic processes underlying complex and multifactorial diseases such as ACS.
Recently, the metabolomic approach has been applied to identify a risk profile in heart failure patients [
12‐
14], atrial fibrillation patients [
15], and diabetic patients [
16]. In the setting of ACS, studies have characterized the metabolic biosignature of myocardial ischemia [
9,
17‐
20], identified altered signatures in lipid metabolism in patients with angina or myocardial infarction with respect to control subjects [
21], and identified microbial metabolites in urine associated with coronary heart disease [
22].
Risk stratification should identify individuals at high risk who require more intensive therapy, or, conversely, help avoid drug overuse and associated side effects in patients with a favorable prognosis. In this framework, the aim of the present study was to evaluate the impact of the metabolomic fingerprint on the occurrence of cardiovascular death in acute myocardial infarction (AMI) patients after percutaneous coronary intervention.
Discussion
After the acute phase of an AMI, for which management is strictly defined by the European Society of Cardiology Guidelines [
43,
44], patients remain at increased risk of secondary atherothrombotic events, including recurrent ACS events and stroke, and continue to face a high risk of premature death not only in the immediate future but also in the following years [
45,
46]. For these reasons, the identification of a metabolomic fingerprint able to identify patients who are at increased risk of death might allow clinicians to tailor medical treatments and interventions according to patients’ overall risk: high-risk patients could be targeted with more intensive pharmacological treatments (that is, with the highest tolerated statin dosage or more aggressive antiplatelet treatments), and more intensive follow-up programs could be planned with clinical reevaluation at shorter time intervals (that is, monthly instead of the standard visits at 1, 6, and 12 months). In AMI patients enrolled in the AMI-Florence 2 study, we found a metabolic fingerprint which was able to discriminate patients who died within 2 years from the cardiovascular event from survivors with high accuracy (AUC 0.859), and this result was duplicated in a validation set (AUC 0.801). We also built sex-specific RF models and found that the male model was better able to predict outcomes (male: AUC 0.834; female: AUC 0.786), which was confirmed in the validation set. To the best of our knowledge, this is the first study to assess the capability of a metabolomic assay to predict mortality in the setting of AMI.
A metabolic fingerprint can be deemed as a holistic super-biomarker with a discriminative and predictive power undoubtedly higher than that of the sum of the few quantified metabolites [
47]. AMI, as with the majority of human diseases, has a multifactorial etiology and a complex physiopathology that concurrently alters several metabolic pathways [
48]. Therefore, the metabolic fingerprint, composed by superimposing all the visible signals of the low and high molecular weight endogenous metabolites, represents an optimal level at which to analyze pathological changes in biological systems [
49]; indeed, it takes into account all metabolite variations, even slight ones.
The NOESY RF score, based on the metabolic fingerprint here presented, was independent from the classical clinical parameters and the widely used GRACE score, and achieved better results in predicting all-cause death within 2 years after AMI when considering both Cox models and ROC analyses. It is worth of mentioning that age was a very good predictor of mortality in our dataset; in particular, it proved to be better than even the GRACE score in our training set. Older patients showed the worst clinical conditions, with a higher percentage of heart failure, atrial fibrillation, previous cerebrovascular diseases, and diabetes. These parameters are not included in the GRACE score model, and could explain why age performed very well as a predictor in our cohort.
Although our study has several strengths, including the number of patients studied, the long-term follow-up (2 years), and the analysis replication in a validation set, some limitations should also be mentioned. First, sample collections were done exclusively in the acute phase of the disease, impairing the acquisition of data correlated to the biochemical mechanisms of the transition to the quiescent phase. Owing to the importance of this aspect, further efforts in this direction are required. Second, even though the data were replicated in a validation set, a totally independent cohort for validation is lacking. However, before attempting to replicate these findings in very large multicenter studies, common standard operating procedures are required for sample collection and storage, otherwise samples collected in different centers will not be comparable. Our group is strongly committed to this, and have contributed to the development of the optimal pre-analytical procedures for metabolomics [
25]. Finally, NMR is less sensitive than MS (although it is more suitable for metabolic fingerprinting), and thus only a limited number of metabolites have been found to be statistically significant. Specifically, we found that patients who died were characterized by significantly higher levels of 3-hydroxybutyrate, proline, creatinine, acetate, acetone, formate, and mannose, and significantly lower levels of valine and histidine.
Lifestyle and the medication administered to a patient influence the molecular signatures in plasma and serum samples, and the relative metabolite concentrations reflect tissue lesions and organ dysfunctions. In this framework, previous studies have underlined the usefulness of metabolomics of serum and plasma in determining the individual’s disease risk, prognosis, and therapeutic options in different clinical settings [
29,
30,
50,
51]. For instance, sera of heart failure patients carries a strong signature of the disease, allowing the estimation of heart failure-related metabolic disturbance and possessing a better prognostic value than conventional biomarkers [
13,
51].
In a prospective study of three population-based cohorts from Finland that were free of CVD at baseline [
52], a metabolomic analysis evidenced that circulating phenylalanine, monounsaturated fats, and polyunsaturated fatty acids were as strongly predictive of cardiovascular risk as the conventional lipid risk factors, and were markers of CVD onset during a long-term follow-up (more than a decade). In our study, we did not find any role for these metabolites in predicting mortality. However, the different study populations (CVD- free vs. AMI patients) with different lifestyle and dietary habits may explain the different molecular signatures.
The usefulness of the metabolomic approach was also demonstrated in a general cohort of patients at risk for cardiovascular events undergoing cardiac catheterization [
53]; at baseline, plasma metabolomic profiles independently predicted cardiovascular death after adjustment for multiple clinical covariates. In this study, a significant predictive role was demonstrated for five metabolite factors (medium-chain acylcarnitines, short-chain dicarboxylacylcarnitines, long-chain dicarboxylacylcarnitines, branched-chain amino acids, and fatty acids). We consistently found in the present study that higher levels of valine were a protective factor in AMI patients. At variance with our study, increased concentrations of branched-chain amino acids levels have been shown in coronary artery patients compared with control subjects [
54], and high levels of these essential amino acids significantly correlated with the severity of coronary artery disease (CAD) [
55]. However, these studies were case-control studies and did not evaluate CAD patients in the acute phase of the disease. The pathways of the branched amino acids in humans are very complex and it is likely that in ACS patients an altered metabolic pathway for these amino acids represents a prognostic risk factor.
Our study, performed in a large sample population of AMI patients followed for 2 years, provided new data about the role of high 3-hydroxybutyrate circulating levels on post-AMI mortality. A previous study that evaluated the ketone bodies in the urine of five ACS patients demonstrated that ketone bodies, and particularly 3-hydroxybutyrate, were altered during the acute event [
56]. Furthermore, high levels of 3-hydroxybutyrate have been associated with high prevalence of heart failure and diabetes [
57]. In insulin deficiency, when the release of free fatty acids from adipose tissues exceeds the capacity of the tissues to metabolize them, severe and potentially fatal diabetic ketoacidosis can occur in which levels of 3-hydroxybutyrate in the blood can reach up to 25 mM [
58]. A recent study has also demonstrated a significant increase of serum ketone bodies in response to angioplasty-induced ischemia performed in patients with stable angina, and it has been hypothesized that these metabolic changes could be a response to reperfusion oxidative stress and may play a key role in free radical homeostasis during ischemia-reperfusion injury [
59]. However, it remains unclear whether the elevation of ketone bodies represents an adaptive mechanism required to maintain cell metabolism or if it actually contributes to disease progression and, thus, the worsening of the prognosis.
Furthermore, it is worth of mentioning that formate has already been proposed as a possible biomarker of ACS [
60].
As expected, patients with worst prognosis showed higher level of serum creatinine, a well-known marker of renal insufficiency. It has already been demonstrated that elevated serum levels of creatinine on admission are associated with impaired myocardial flow and poor prognosis for 1-year mortality [
61,
62].
Previous studies evidenced an interaction between sex and adverse cardiovascular events in CAD patients [
63‐
65], with myocardial infarction morbidity and mortality higher in women than in men. Consistently, our results demonstrated a higher prevalence of mortality in women than in men (19.4% vs. 12.5%). The excess risk of mortality in women could be ascribed to the differences in cardiovascular risk factor, that is, age, hypertension, diabetes, and co-morbidities prevalence. Accordingly, in our study, women showed a higher median age and higher prevalence of hypertension, atrial fibrillation, and heart failure with respect to men.
Building sex-specific models enabled us to improve the outcome prediction in the male cohort, but not in the female one. Thus, the male metabolic fingerprint seems to show a higher association with cardiovascular mortality than the female one. However, the female cohort was smaller than the male cohort and this may have affected the predictive capability of the model; therefore, larger cohorts of patients are needed to build robust sex-specific models.