Author + information
- Received May 29, 2008
- Revision received June 26, 2008
- Accepted July 10, 2008
- Published online October 28, 2008.
- ↵⁎Reprint requests and correspondence:
Dr. James L. Januzzi, Jr., Cardiology Division 32 Fruit Street, Yawkey 5984, Boston, Massachusetts 02114
Objectives The purpose of this study was to examine the patient-specific characteristics of the interleukin-1 receptor family member ST2 in patients with acute heart failure (HF).
Background ST2 signaling is involved in the process of cardiac fibrosis and hypertrophy.
Methods In all, 346 patients with acute HF had ST2 measured. Associations between ST2 and demographics, severity/type of HF, and other biomarkers were examined. Receiver-operator characteristic curves and multivariable Cox proportional hazards analyses evaluated the prognostic ability of ST2.
Results The ST2 values correlated with the severity of HF (p < 0.001), left ventricular ejection fraction (r = −0.134; p = 0.014), creatinine clearance (r = −0.224; p < 0.001), B-type natriuretic peptide (r = 0.293; p < 0.001), amino terminal B-type natriuretic peptide (r = 0.413; p < 0.001), and C-reactive protein (r = 0.429; p < 0.001). ST2 was not associated with age, prior HF, or body mass index. The ST2 levels at presentation were higher among patients who died by 1 year. The area under the receiver-operator characteristic for death at 1 year was 0.71 (p < 0.001). In a multivariable Cox model containing established clinical and biochemical predictors (including natriuretic peptides), ST2 remained a predictor of mortality (hazard ratio: 2.04, 95% confidence interval: 1.30 to 3.24, p = 0.003), and was equally predictive in patients with HF and preserved or impaired systolic function. When both ST2 and natriuretic peptides were elevated, the highest rates of death were observed in cumulative hazard analysis (p < 0.001). In the presence of a low ST2 level, natriuretic peptides did not predict mortality.
Conclusions Consistent with its proposed role in a myocardial-specific response to stretch, ST2 has strong clinical and biochemical correlates in patients with acute HF. Prognostically, ST2 is powerful in acute HF and is synergistic with natriuretic peptides for this use.
The ST2 gene encodes a protein that is a member of the interleukin (IL)-1 receptor family (1,2) and consists of both a transmembrane receptor form (ST2L) and a truncated, soluble receptor form (ST2) that can be detected in serum (3,4). The ST2 gene is markedly up-regulated in cardiac myocytes and fibroblasts subjected to mechanical strain (5), which is important as the functional ligand of ST2 was identified to be IL-33, a cardiac fibroblast product also induced by mechanical strain (6). The ST2/IL-33 signaling is thought to play an important role in regulating the myocardial response to biomechanical overload in stretched cardiac fibroblasts and cardiomyocytes (7–10), in a manner similar to B-type natriuretic peptide (BNP) (11); indeed, knock-out of the ST2 gene leads to a phenotype not unlike that seen in BNP knock-out models, with severe myocyte hypertrophy and interstitial cardiac fibrosis (12).
Clinically, serum levels of ST2 have been reported to increase significantly early after acute myocardial infarction, and they inversely correlate with ejection fraction (5). Among patients with chronic heart failure (HF), serum ST2 levels appear to be associated with adverse outcome (13,14), although few detailed data regarding ST2 and acute HF exist.
We previously showed that higher ST2 levels are associated with a higher risk for death by 1 year among patients with dyspnea in the emergency department setting (15). The purpose of the present study was to derive a better understanding of the patient-specific clinical and biochemical characteristics with effects on ST2 values in acute HF, as well as to further explore the association between ST2 concentrations and outcomes in these patients. To do so, we studied a cohort of patients with acute HF from 2 prospective trials of acute HF with available data regarding ST2 (16,17).
The study population consisted of patients from 2 previously reported prospective clinical trials of dyspneic emergency department patients from Boston, Massachusetts (16), and Linz, Austria (17). These trials had compatible inclusion/exclusion criteria and obtained similar clinical information including standard demographics, medical history and drug therapy, presenting symptoms and signs (including severity of breathlessness by New York Heart Association [NYHA] functional class), physical examination, results of serum chemistry tests, radiographic studies (typically plain chest radiographs), electrocardiography results, and finally, the results of biomarkers testing including ST2, measures of renal function, troponin T, BNP, amino terminal B-type natriuretic peptide (NT-proBNP), and C-reactive protein (CRP).
For the purposes of this study, a total of 346 patients with acute HF were considered: 209 subjects from Boston, Massachusetts, and 137 from Linz, Austria. Follow-up for vital status among HF subjects was completed at 1 year (18,19).
The ST2 was measured using an enzyme-linked immunosorbent assay (Medical & Biological Laboratories, Woburn, Massachusetts) (20), on blood specimens frozen at −80°C. Assay performance characteristics for the 2 cohorts were previously reported (15,21); we also previously documented the stability of ST2 at −80°C, and after freeze/thaw cycles (21). In addition, NT-proBNP (Elecsys proBNP, Roche Diagnostics, Indianapolis, Indiana) and BNP (Boston subjects: Bayer Advia BNP, Tarrytown, New York; Linz subjects: Abbott BNP, Abbott Park, Illinois) assays were performed, and CRP was also measured using a commercially available high-sensitivity method (Roche Diagnostics).
Demographics and clinical and laboratory variables were generally described as means with SD, and variables with a skewed distribution (identified using assessment of skewness, with a value greater than twice its standard error indicating non-normality) were described with medians and interquartile range (IQR). The analysis of variance was used to compare normally distributed variables, and the Wilcoxon rank-sum test was used to compare continuous variables in states of non-normality. The Kruskal-Wallis test was performed to assess and compare the ST2 concentrations across NYHA functional class and National Kidney Foundation Kidney Disease Outcomes Quality Initiative categories of creatinine clearance (of <30, 30 to 59, 60 to 89, and 90+ ml/min/1.73 m2). Impaired left ventricular systolic function was defined as <50%, as estimated by echocardiography or radionuclide studies.
Univariable Spearman correlation was used to evaluate the magnitude and significance of relationships among continuous variables. After univariable linear regression analysis, multivariable linear regression analyses were then performed, with ST2 concentration as the dependent variable.
In an effort to better understand the characteristics of ST2 concentrations as a predictor of death in acute HF, several methods were employed. Patients were grouped into ST2 deciles, and the frequency of mortality relative to increasing ST2 concentrations was estimated. Receiver-operator characteristic (ROC) analyses with death at 1 year as the reference standard were also performed, and area under the curve (AUC) estimated. Similar ROC curves were plotted for BNP, NT-proBNP, and CRP and compared with ST2 using Z-testing.
Multivariable Cox proportional hazards analyses using forward stepping were then performed, in an effort to identify the independent predictors of death at 1 year for this patient cohort. Variables were retained if their univariable p value was <0.05, and entered into a multivariable model; only those variables with significant p values were retained in the final multivariable model. This analysis was performed with either NT-proBNP or BNP in different models.
Stratified multivariable analyses were done considering the relative importance of ST2 versus natriuretic peptides alone and in combination with each other. To illustrate the concept, cumulative hazard curves were plotted to demonstrate the ability of ST2 to predict hazard, alone or in combination with BNP and NT-proBNP in a multimarker strategy. To do so, patients were categorized on the basis of their concentrations for each marker relative to the median for each group; this yielded 4 groups: low ST2/low NT-proBNP, high ST2/low NT-proBNP, low ST2/high NT-proBNP, and high ST2/high NT-proBNP. Similar groups were selected replacing NT-proBNP with BNP. Interaction tests were done between ST2 and both BNP or NT-proBNP as well as with left ventricular ejection fraction.
For all statistical analyses, either SPSS (SPSS Inc., Chicago, Illinois) or STATA software (Stata Corp., College Station, Texas) was used; all p values are 2-sided, with composite results <0.05 considered significant. ROC curve analysis was performed with Analyse-It software (Analyse-It, Ltd., Leeds, United Kingdom).
Baseline clinical and laboratory characteristics of study patients
In all, 346 subjects were available for analysis. Table 1 depicts baseline clinical, biochemical, and radiological characteristics for the study patients, and describes a relatively typical population of patients with acute HF. The mean age of study patients was 73 years, and a modest majority was male (68%). Most subjects had NYHA functional class III or IV symptoms, consistent with a population presenting urgently to the emergency department setting. Approximately one-half of the subjects had de novo HF; consistent with that, a similar percentage of these patients were taking medications with mortality benefit in this setting, such as beta-blockers or angiotensin-converting enzyme inhibitors. The mean left ventricular ejection fraction was close to 45%; as is typical among patients in contemporary studies of acute HF, nearly one-half had preserved left ventricular systolic function (42.7%). Atrial fibrillation was present in 36% of patients on presentation.
Consistent with the nature of patients having acute HF, the study patients had elevated values for BNP (494 ng/l; IQR: 203 to 1,180), NT-proBNP (3,578 ng/l; IQR: 1,574 to 9,446), and CRP (5.5 mg/l; IQR: 1.2 to 26.0). The median ST2 value among these subjects was 0.49 ng/ml (IQR: 0.26 to 1.03).
ST2 Levels and Clinical and Biochemical Correlates
Effect of comorbid conditions
When categorized as a function of NYHA functional class, median ST2 concentrations were noted to be higher and with worse symptoms, as demonstrated using the Kruskal-Wallis test (p < 0.001) (Fig. 1). Concentrations of ST2 were not significantly different in patients with HF due to ischemic heart disease versus patients with nonischemic etiologies as assessed using the Wilcoxon rank-sum test (0.43 [IQR: 0.23 to 0.96] ng/ml vs. 0.52 [IQR: 0.28 to 1.20] ng/ml; p = 0.52).
In univariable correlations, several associations were noted between ST2 and clinical variables. Interestingly, significant correlations with both temperature as well as leukocyte count were found. As well, a significant, albeit weak association was detected between ST2 and creatinine clearance (r = −0.224; p < 0.001); this relationship existed both with respect to those dichotomized and analyzed using the Wilcoxon rank-sum test as a function of a creatinine clearance of 60 ml/min/1.73 m2 (with higher values for ST2 in patients with moderate or more renal dysfunction, 0.59 ng/ml vs. 0.40 ng/ml; p < 0.001) or when considered with the Kruskal-Wallis test as a function of National Kidney Foundation Kidney Disease Outcomes Quality Initiative categories (0.79 ng/ml vs. 0.58 ng/ml vs. 0.45 ng/ml vs. 0.35 ng/ml; p < 0.001).
With respect to other biomarkers, significant positive correlations were observed between ST2 and BNP (r = 0.293; p < 0.001), NT-proBNP (r = 0.413; p < 0.001), and CRP (r = 0.429; p < 0.001) (Table 2). Although univariable correlations between ST2 and markers of infection such as white blood cell count or temperature were present, these were no longer significant in multivariable modeling.
In contrast to the natriuretic peptides (Table 2), we detected no significant correlations between ST2 and age (r = −0.035; p = 0.52), hemoglobin (r = −0.033; p = 0.54), or body mass index (r = −0.083; p = 0.13), and no association with gender (p = 0.43), prior HF (p = 0.17), or the presence of atrial fibrillation (p = 0.30).
Cardiac structure and function
Among patients with acute HF, baseline ST2 values weakly correlated with ejection fraction (r = −134; p = 0.014); accordingly, patients with preserved left ventricular ejection fraction had lower ST2 concentrations (Wilcoxon rank-sum test: 0.37 [0.21 to 0.81] ng/ml vs. 0.57 [0.30 to 1.1] ng/ml; p < 0.001). There was no difference in ST2 concentrations among patients with HF due to ischemic versus nonischemic causes.
No significant associations between ST2 and right ventricular systolic pressure (r = −0.018; p = 0.79) were found; neither were associations between ST2 and severity of mitral regurgitation (p = 0.35), tricuspid regurgitation (p = 0.46), or aortic regurgitation (p = 0.60) found.
Independent Predictors of ST2 Concentration in Acute HF
Multivariable linear regression
In a multivariable linear regression model, the following, listed in descending order, were found to be predictors of ST2 concentration in acute HF patients: CRP (T = 6.79; p < 0.001), temperature (T = 5.66; p < 0.001), pulse (T = 4.44; p < 0.001), BNP (T = 3.0; p = 0.003), blood urea nitrogen (T = 2.76; p = 0.006), smoking history (T = 2.32; p = 0.021), and systolic blood pressure (T = 2.01; p = 0.045).
ST2 levels as a function of mortality
Concentrations of ST2 at presentation with acute dyspnea were significantly higher among patients who died by 1 year (n = 97; 28%) compared with patients who were alive (0.87 ng/ml vs. 0.40 ng/ml; p < 0.001); in addition, decile analysis of ST2 concentrations examined as a function of mortality rates at 1 year revealed that there was a graded increase in mortality with rising concentrations of the marker (Fig. 2).
ROC analyses performed for predicting death at 1 year demonstrated an AUC of 0.71 for ST2 (95% confidence interval [CI]: 0.66 to 0.77; p < 0.001), similar to BNP (AUC = 0.66, 95% CI: 0.60 to 0.73; p = 0.17 for difference with ST2) and NT-proBNP (AUC = 0.68, 95% CI: 0.62 to 0.74; p = 0.30 for difference with ST2) and statistically superior to CRP (AUC = 0.61; 95% CI: 0.55 to 0.68; p = 0.001 for difference with ST2) (Fig. 3).
Considered as a function of the median ST2 value of 0.49 ng/ml, rates of death at 1 year were comparable for patients above versus below the median ST2 value, irrespective of the presence of HF with systolic dysfunction (44.0% vs. 12.9%; p < 0.001) or HF with preserved left ventricular systolic function (31.0% vs. 17.9%; p < 0.001); no significant interaction term was found between ST2 and ejection fraction in these mortality analyses (p = 0.08).
Using the median value for the group as a predictor of mortality, ST2 had 72% sensitivity (95% CI: 62% to 81%), 56% specificity (95% CI: 49% to 62%), a positive predictive value of 39%, and a negative predictive value of 84%. Using a cut-point of 0.20 ng/ml identified in prior analyses (15), a negative predictive value of 96% for mortality was observed.
In a multivariable Cox model containing ST2 concentrations, clinical variables, and biomarkers such as measures of renal function, CRP, NT-proBNP, and BNP, ST2 levels higher than the median were independently predictive of mortality (Table 3); combining values of ST2 with NT-proBNP or BNP incremented the AUC for death from 0.71 to 0.74 and 0.73, respectively—both significant changes—and adding clinical variables including renal function and CRP incremented the AUC to 0.80, again a significant increase. The prognostic ability of ST2 was consistent when the cohorts were examined as a function of data source, with preserved prognostic ability for death when examining patients from either Linz or Boston.
Of interest, in a stratified multivariable analysis examining patients with high and low ST2 values, baseline levels of BNP, NT-proBNP, and CRP did not contribute independent information in the presence of ST2 levels below the median; no interaction term between ST2 and either natriuretic peptide was detected (p = 0.30 and 0.10, respectively, for NT-proBNP and BNP).
The potential enhanced value from the combined use of ST2 with a natriuretic peptide for prognostication is detailed in the cumulative hazards curves in Figure 4. The lowest mortality rates were observed (10%) for patients with values for both biomarkers below the median (n = 114), followed by the mortality rates for patients with isolated elevation of natriuretic peptide and for patients with isolated elevation of ST2, respectively. Importantly, the 1-year mortality rate observed among the 116 patients who had elevation in both biomarkers was considerably higher than in any other group (42%).
The ST2 gene is a member of the IL-1 receptor family (1,2); the protein products of ST2 gene, ST2L and soluble ST2, are biomechanically induced in cardiomyocytes (5). The functionally active transmembrane form, ST2L, plays a role in modulating responses of T-helper type 2 cells (6) and is involved in the development of immunologic tolerance (22), but also has a prominent role in ameliorating fibrosis in the context of cardiovascular stretch via binding of IL-33. In this context, a coordinated reduction in atrial natriuretic peptide and BNP gene expression and adaptive resistance to severe pressure overload develops (12), a reduction that is blocked by knocking out the ST2 gene or administering large amounts of soluble ST2 to compete for IL-33. This finding raises the possibility that concentrations of the truncated soluble ST2 may represent a “decoy” receptor, acting as a biologic “off switch” for ST2/IL-33 signaling. Accordingly, imbalances in this system would theoretically be functionally and prognostically important in the context of acute HF; therefore, ST2 represents an intriguing candidate biomarker for patients with heart disease. Supporting this possibility, concentrations of soluble ST2 have been reported to be measurable in the blood of patients with HF, and concentrations of ST2 may parallel the presence of HF, representing a prognostically meaningful biomarker in this setting. Little is known about the patient-specific factors that determine ST2 concentrations in HF, however, and the role of this marker to specifically prognosticate in patients with destabilized HF (a population at high risk for adverse short- and intermediate-term outcomes) is less defined.
Among 346 patients with acutely destabilized HF from centers in Boston, Massachusetts, and Linz, Austria, ST2 concentrations segregated with more severe HF, and had several independent associations with variables such as left ventricular ejection fraction and renal function, findings reminiscent of natriuretic peptides. However, these associations were considerably weaker than those reported seen with natriuretic peptides; in addition, several important covariates for natriuretic peptides, such as age, gender, heart rhythm, and body mass index did not have a significant effect on ST2 concentrations either in univariable or multivariable linear regression analyses. This relative independence from prevalent comorbidities in patients with HF might represent a potential advantage of ST2 for prognostication over other widely used markers such as the natriuretic peptides. Lastly, and of interest, associations between measures of inflammation, such as with temperature, leukocyte count, and CRP, were present between ST2 but not natriuretic peptides.
With respect to prognosis, ST2 concentrations at presentation strongly predicted mortality as early as a few months from presentation and out to at least 1 year, and did so at least as well as natriuretic peptides. Furthermore, ST2 was superior for prognostication than a high-sensitivity method for CRP measurement. The ability of ST2 to predict hazard was further delineated in a multivariable model containing established clinical and biochemical variables significant for mortality, including natriuretic peptides, CRP, and renal function. It is worth pointing out that the ability of ST2 to prognosticate was present in patients with HF and preserved and impaired left ventricular systolic function, although interaction testing was suggestive of a trend toward interaction with systolic function. Notably, in multivariable models, in the presence of a low ST2 value, natriuretic peptides were not prognostically meaningful, whereas the conjoined use of ST2 plus a natriuretic peptide best identified patients at highest risk for death, arguing for the concomitant use of the peptides to identify patients at particularly low risk (i.e., <10% mortality at 1 year) versus those at highest risk (i.e., >40% mortality at 1 year). Interestingly, patients with a low natriuretic peptide but high ST2 value (who would otherwise be viewed as being “low risk” based on conventional means of risk assessment) actually had death rates that were more consistent with a higher risk presentation.
Why ST2—a biomarker with a putatively beneficial role in the cardiovascular system to counterbalance fibrosis—has such strong associations with risk remains hypothetical. As noted, ST2/IL-33 signaling plays a key cardioprotective role in regulating the myocardial response to biomechanical overload in cardiac fibroblasts and cardiomyocytes (12), a benefit that can be abrogated either by interrupting the ST2 gene or by administering large amounts of soluble ST2 (12,23). We hypothesize that intact ST2 signaling is important to synergize with IL-33 to protect from excessive pressure overload and fibrosis, but that patients with excessively increased soluble ST2 levels may paradoxically inhibit binding of IL-33 to the functionally active ST2L, resulting in a maladaptive situation that may lead to excessive cardiac fibrosis, and worse outcomes in HF. Such a finding has been observed in cardiac specimens of mice with ST2−/− phenotypes, in which excessive hypertrophy, cardiac fibrosis, ventricular chamber dilation, worse systolic function, and reduced survival were all noted (12). Further investigation into ST2 and IL-33 as novel targets for modifying prognosis in patients with HF is justified, given the value of this pathway for predicting prognosis.
Limitations of our study include the less well understood biologic role of ST2 in the heart; advances in the understanding of ST2/IL-33 interactions in the heart will further address this issue, one that remains somewhat elusive. Indeed, if one assumes that IL-33 and ST2 comprise a critical biomechanically-induced and cardioprotective signaling system, and that an excess of soluble ST2 may abrogate this protection by interrupting the interaction of IL-33 with the ST2 transmembrane receptor form, then the measurement of IL-33 may provide interesting complementary information in terms of quantitative balance. Unfortunately, our attempts thus far to measure IL-33 in these same blood samples have not shed light on the question. While our data show powerful associations between presenting ST2 concentrations and outcomes, serial ST2 concentrations may be more useful for predicting hazard than a single measurement (14). Indeed, this finding has also been preliminarily found in a distinct population of patients with acute HF (24). Thus, it may be that while a single presenting value for ST2 is powerfully prognostic, serial measurements of the marker may offer even more prognostic value. Although ST2 may be secreted by mechanically overloaded cardiac myocytes, elevations in serum ST2 may be seen in asthma (25), and autoimmune diseases such as systemic lupus erythematosus (26), malignancy (27), and sepsis (28). These observations do not diminish our findings of strong associations between the presence of ST2 and severity of HF (15), or the profound prognostic value of the marker in our cohort, and we note that other “cardiac” markers such as troponins and natriuretic peptides have similar associations between the presence and severity of many of these illnesses (29–34).
In summary, ST2 concentrations in acute HF represent a marker of disease severity, myocardial stretch, and inflammation, and portend a powerfully negative prognosis, independent of established clinical and biochemical predictors in this setting. More exploration of this novel and potentially important biomarker in HF is warranted.
Dr. Rehman is supported by the Dennis and Marilyn Barry Fellowship in Cardiovascular Medicine, and Dr. Januzzi is supported in part by grants from the Balson Scholarship Fund.
- Abbreviations and Acronyms
- area under the curve
- B-type natriuretic peptide
- confidence interval
- C-reactive protein
- heart failure
- amino terminal B-type natriuretic peptide
- New York Heart Association
- receiver-operator characteristic
- interleukin receptor family member
- transmembrane receptor form of interleukin family member ST2
- Received May 29, 2008.
- Revision received June 26, 2008.
- Accepted July 10, 2008.
- American College of Cardiology Foundation
- Townsend M.J.,
- Fallon P.G.,
- Matthews D.J.,
- Jolin H.E.,
- McKenzie A.N.
- Weinberg E.O.,
- Shimpo M.,
- De Keulenaer G.W.,
- et al.
- Manabe I.,
- Shindo T.,
- Nagai R.
- Shimpo M.,
- Morrow D.A.,
- Weinberg E.O.,
- et al.
- Weinberg E.O.,
- Shimpo M.,
- Hurwitz S.,
- Tominaga S.,
- Rouleau J.L.,
- Lee R.T.
- Januzzi J.L. Jr..,
- Peacock W.F.,
- Maisel A.S.,
- et al.
- Mueller T.,
- Gegenhuber A.,
- Poelz W.,
- Haltmayer M.
- Gegenhuber A.,
- Mueller T.,
- Dieplinger B.,
- Poelz W.,
- Pacher R.,
- Haltmayer M.
- Mueller T.,
- Dieplinger B.,
- Gegenhuber A.,
- Poelz W.,
- Pacher R.,
- Haltmayer M.
- Miller A.M.,
- Xu D.,
- Asquith D.L.,
- et al.
- ↵Fitzgerald RL, Boisot S, Chiu A, et al. Serial sampling of ST2 predicts 90 day mortality of CHF patients (abstr). Paper presented at: Annual Meeting of the European Society of Cardiology; September 2, 2007; Vienna, Austria.