Introduction

Heart failure (HF) with preserved ejection fraction (HFpEF) is as common and associated with as much morbidity and mortality as HF with reduced ejection fraction (HFrEF).¹

See Clinical Perspective

Exploration of a wide range of biomarkers has contributed fundamentally to the understanding of underlying mechanisms, pathophysiology and prognosis in HFrEF,² which, in turn, has contributed to the development of and successful trials of neurohormonal antagonist drugs. Despite the convincing evidence of prognostic benefit in HFrEF, these pharmacological agents have not been shown to improve outcomes in HFpEF,^3–6 suggesting a different underlying pathophysiology and greater heterogeneity of phenotypes in HFpEF than that in HFrEF.

HFpEF is characterized by cardiac hypertrophy and fibrosis, leading to diastolic dysfunction and impaired systolic strain, increased filling pressures, and classic signs and symptoms of HF, indistinguishable from those in HFrEF. However, patients with HFpEF are older with more noncardiac comorbidities such as diabetes mellitus, obesity, hypertension, and chronic kidney disease, which are associated with and may drive the poor outcomes.⁷ Indeed a new paradigm has emerged proposing noncardiac comorbidities driving HFpEF by inducing a proinflammatory state causing immune cell recruitment, coronary microvascular, and endothelial inflammation resulting in depressed nitric oxide bioavailability and low protein kinase G activity, in turn leading to interstitial fibrosis, remodeling, and high diastolic left ventricular stiffness.⁸

In this context, biomarkers associated with matrix remodeling, inflammation, fibrosis, apoptosis, and cell turnover reflecting potentially key drivers of HFpEF may be of particular interest.^9–11 In HFpEF, associations with diastolic dysfunction and prognosis have been suggested for several markers of fibrosis and neurohormonal activation, but we are not aware of any comprehensive proteomics approach.^12–17

Analyzing a large set of biomarkers in a holistic manner offers a powerful way of contextualizing the biomarkers and improving the understanding of disease-related pathways. We used orthogonal projection to latent structures by partial least square (OPLS) analysis,¹⁸ which identifies the quantitative relationship between all variables in a data matrix (X) and outcome (Y), but also filters out systematic unrelated variation to Y. Furthermore, we combined OPLS analyses of clinical variables and multiplexed biomarkers, with a semiautomated literature search of the scientific literature to identifiy biomarker patterns predicting HF severity and outcome, as well as potential pathophysiological mechanisms preceding outcome in HFpEF.

Materials and Methods

The Karolinska Rennes (KaRen) was a prospective observational multicenter study characterizing patients with HFpEF, described in detail elsewhere.¹⁹ In brief, 539 patients presented to the hospital with acute signs and symptoms of HF according to the Framingham criteria, N-terminal probrain natriuretic peptide (NT-proBNP) of >300 ng/L and a left ventricular ejection fraction (LVEF) of ≥45% were enrolled in French and Swedish centers. The prespecified KaRen biochemistry substudy enrolled 86 patients between 21 May, 2007 and 29 December, 2011 at Karolinska University Hospital, Stockholm, Sweden. Patients returned to the hospital in stable condition 4 to 8 weeks after enrollment for a follow-up visit, including blood sampling and detailed echocardiography, and were thereafter followed until 30 September, 2012 when vital status was assessed by telephone contact or by the Swedish National Patient Register and Population Register. The primary outcome was defined as time to mortality from any cause or first hospitalization due to HF. All HF hospitalizations were adjudicated and defined according to clinical judgment by the local investigator and additionally centrality validated to confirm the presence of heart failure at hospitalization.

The echocardiographic assessment was performed on a ViVid 7 echo-platform (GE VingMed, Horten, Norway) and analyzed in a dedicated core center in Hôpital Pontchaillou, Rennes, France. Each examination was interpreted once and measurements were performed 3× and averaged by an echocardiographist (E.D.) blinded to the specific clinical history of the patient.

At the follow-up visit, blood samples were collected in a fasting state in the morning in EDTA tubes, centrifuged, and plasma was aliquoted and stored in −70°C until analysis.

Biomarkers were quantified from single samples by a multiplex immunoassay (Proseek Multiplex^96×96 CVD I v1; Olink Bioscience, Uppsala, Sweden), which is a 96-plex immunoassay (92 analytes plus internal controls) based on proximity extension assay. Proximity extension assay uses target-specific antibody pairs that are linked to DNA strands that, upon simultaneous binding to the target analyte, creates a real-time polymerase chain reaction amplicon in a proximity-dependent manner enabled by the action of a DNA polymerase.²⁰ The intra-assay coefficient of variation (CV) ranges between 5% and 13% (mean 8%), and the interassay CV ranges between 9% and 39% (mean 15%). Further information about reproducibility and validation is found at http://www.olink.com.

NT-proBNP was analyzed by proBNPII (Roche Diagnostics, Bromma, Sweden). Creatinine clearance was calculated according to the Cockcroft–Gault formula.

Statistics

Figure 1 shows an overview of the analyses performed. The respective statistical methods are described below.

• Download figure
• Download PowerPoint

Descriptive data in Table 1 were expressed as median and quartiles (Q1; Q3) or number (%) and divided according to the presence of the composite end point and compared by the Wilcoxon rank-sum test and Fisher’s exact tests as appropriate. The difference in biomarker levels was determined on log2-values (ProSeek biomarkers) or log10-values (NT-proBNP levels by ELISA) by 2-sided t test using GraphPad Prism 6. For visualization (Figure I in the Data Supplement), the data were transformed from linear values. The concentrations of the ProSeek biomarkers are expressed as arbitrary units because of the lack of specific calibrators for the assays.

Analyses of Outcome

Associations with the outcome was determined with Cox proportional hazards models and presented as hazard ratio and 95% confidence interval. In the final multivariable Cox regression model, three clinically significant covariates, age, sex, and NT-proBNP, were included together with each biomarker. All biomarkers were analyzed in log-transformed format.

P values were 2-sided, and statistical significance was set at 0.05. Statistical analyses were performed using SAS software version 9.3 (SAS Institute, Cary, NC, USA).

OPLS Analyses Correlating Biomarker Patterns With NYHA Class and Outcome

A total of 240 clinical and medication variables were collected at enrollment and at the 4- to 8-week follow-up (excluding repeated measures), and outcome data were merged with the 87 biomarker data set (5 of the biomarkers in the 92-plex yielded no or noninformative values and were excluded from the analysis). This generated a data matrix consisting of 327 variables in 86 patients who were subjected to OPLS analyses using Simca 13 (Umetrics, Umeå, Sweden). All data were scaled to unit variance and mean centered. In addition, variables with a max/mean ratio >10 were log-transformed to increase equal leverage of all variables. The OPLS analysis yields a model, in which the contribution of each variable is represented by a loading value (herein normalized to a correlation coefficient in between −1 and 1) to what is predicted. Two OPLS models were generated, each one represented by 326 X-variables predicting 1 Y-variable: In model 1, New York Heart Association (NYHA) Functional class (4 levels) at the 4- to 8-week follow-up was set as the Y-variable, and in model 2, the binary composite outcome variable “heart failure hospitalization or all-cause death” was set as the Y-variable. The validity of the models (Q2 values) was assessed by cross-validation, leaving out and predicting a seventh of the data set in an iterative manner. The cross-validation also yielded 95% confidence intervals for the contribution of each of the variables for the respective Y-variable, as implicated in the Simca software. The OPLS model predicting NYHA class explained 70% (R2Y=0.70) of the variation of NYHA class in the data set, with a predictive (Q2 value) of 0.24 after cross-validation. The corresponding values for outcome were 0.96 (R2Y) and 0.37 (Q2). The confidence intervals were used to determine whether a biomarker significantly predicted the Y-variable (NYHA class and outcome, respectively), adding up to 32 significant biomarkers for NYHA class and 28 significant biomarkers for outcome.

Correlation Analyses With Echocardiographic Measurements

Correlations between the 28 biomarkers predictive of outcome according to the OPLS analysis and the 3 guidelines-specified echocardiographic parameters for the diagnosis of HFpEF; E/e′ >15, left atrial volume index (LAVI) >34 mL/m², and left ventricular hypertrophy defined as left ventricular mass index (LVMI) ≥95 g/m² in women and ≥115 g/m² in men, respectively,²¹ were determined using Spearman’s correlation coefficient. Statistical analyses were performed using SAS software version 9.3 (SAS Institute, Cary, NC, USA).

Identification of Pathophysiological Mechanisms and Pathways Preceding Outcome

To investigate which pathophysiological mechanisms were involved in patients with an event, the outcome-predictive biomarker pattern (the 87 loading scores expressed as correlation coefficients) was compared with annotated data from the public domain using QIAGEN’s Ingenuity Pathway Core Analysis (IPA, QIAGEN Redwood City, www.qiagen.com/ingenuity). The IPA Core analysis compares a biomarker pattern with annotated findings, listing the likelihood of involvement of different upstream regulators, canonical pathways, effector mechanisms, and similarities to annotated mechanisms and diseases. The correlation coefficients were imported into IPA and denoted as log expression ratios and annotated with the official gene symbol for each biomarker. NT-proBNP was not included in analysis as it represents an N-terminal cleavage product of BNP and not a gene.

The Ingenuity Knowledge Base was used as reference set and both direct and indirect experimentally confirmed relationships from all species were included. The analysis as implicated in the software calculates a P value (Fisher exact test and right tailed), quantifying the overlap, and a Z score, quantifying the likelihood and direction (up or downregulated), between the biomarker pattern and published canonical pathways, diseases, and mechanisms as well as upstream regulators. Z scores of >2 or <−2 were considered significant. Details about the statistics used can be found at the following website: http://www.ingenuity.com/products/ipa#/?tab=resources.

Ethical Approval

The KaRen study was conducted according to International Conference on Harmonization and Good Clinical Practice guidelines and the Declaration of Helsinki, and the study was approved by the ethical review board at Karolinska Institutet. Written informed consent was obtained from all patients.

Results

Patient Characteristics

At the follow-up visit, in stable state 4 to 8 weeks after hospital discharge (baseline for this substudy), patients were median 73 (interquartile range [IQR] 66; 79) years old and 51% were female. NT-proBNP was 1000 (469–2330) ng/L, and cardiac function assessed as ejection fraction was 64% (58; 68), E/e′ ratio was 10.8 (8.3; 14.0), and LAVI was 43.3 mL/m² (37.2; 53.8). Of the 86 patients, 23% had E/e′>15, 67% had e′<9, 89% had LAVI>34 mL/m², and 61% had left ventricular hypertrophy. Median follow-up time was 579 days (Q1; Q3: 276; 1178), and no patient was lost to follow-up. The composite end point occurred in 36 patients of which 11 were death.

Characteristics of all 86 patients divided by the presence of the composite end point of HF hospitalization or all-cause death are presented in Table 1. Among patients experiencing an end point, the NT-proBNP was 1375 (574–2565), and 40% had an E/e′>15, 65% had e′<9, 87% had LAVI>34 mL/m², and 69% had left ventricular hypertrophy, and in surviving patients free of HF hospitalization, NT-proBNP was 819 (429–1820) and 13% had an E/e′>15, 68% had e′<9, 91% had LAVI>34 mL/m², and 56% had left ventricular hypertrophy.

Correlation Between Biomarker Patterns With NYHA Class and Outcome by OPLS Analyses

In total, 92 biomarkers (Table I in the Data Supplement) were analyzed, 5 of which generated no or low values (Pappalysin-1, β-nerve growth factor, P-selectin glycoprotein ligand 1, Melusin, and interleukin-4 [IL-4]) and were, thus, not included in the analyses. Of the remaining 87 biomarkers, the OPLS analysis model 1 identified 32 significantly correlated with NYHA class (30 positively and 2 negatively; Table II in the Data Supplement). The OPLS model 2 identified 28 biomarkers significantly correlated with the composite outcome (24 positively and 4 negatively) presented in Table 2 together with OPLS loading values and their impact on the composite end point according to Cox regression analysis (levels of arbitrary units presented in Figure I in the Data Supplement). The biomarkers having the highest (positively correlated) and the lowest (negatively correlated) loading values for the composite outcome were growth/differentiation factor (GDF)-15 and TNF-related apoptosis-inducing ligand (TRAIL), respectively. To illustrate the relationship between the NYHA class- and outcome-predicting biomarker patterns, the respective loading scores of all 326 variables from the 2 OPLS models were plotted against each other (Figure 2). Given their position in the upper right quadrant, biomarkers such as GDF-15 and chitinase-3–like protein 1 positively correlated with higher NYHA class (ie, worse disease state) and with increased risk of outcome, that is, the levels of these biomarkers increased with NYHA class and were also elevated in patients eventually experiencing the composite end point. In the lower left quadrant, the biomarkers such as TRAIL, TRANCE, and GAL were all negatively associated with NYHA class and the composite end point. KLK6, however, was negatively associated with outcome but noninformative in relation to NYHA class.

• Download figure
• Download PowerPoint

Correlations With Echocardiographic Measurements

Correlations by Spearman between biomarkers predicting outcome according to the OPLS analysis and measurements of diastolic dysfunction and structural heart disease, assessed as E/e′, LAVI and left ventricular mass index (LVMI), respectively, are presented in Table 3. The majority of the biomarkers correlated with E/e′, with the soluble receptors tumor necrosis factor receptor 1 (TNF-R1), TRAIL-R2, and U-PAR and the cytokines SPON1, VEGF-A, and the protease matrix metalloproteinase-12 being the most strongly correlated. Four of the biomarkers (C–C motif chemokine 20, prolactin, fatty acid–binding protein, and interleukin 1 receptor like 1 [ST2]) and NT-proBNP measured by ELISA correlated significantly to LAVI, whereas none of the biomarkers correlated to LVMI.

Pathway Analysis and Pathophysiological Mechanisms Preceding Outcome

To identify mechanisms preceding poor outcome, the biomarker pattern (represented by the list of OPLS correlation coefficients of the 87 biomarkers predicting outcome) was subjected to an IPA Core Analysis. Such an analysis summarizes the similarity between a pattern of biomarkers with annotated results published in the scientific literature, yielding probability scores that certain upstream regulators are up or downregulated, whether certain signaling (canonical) pathways are likely to be up or downregulated and whether the pattern resembles other functions or diseases. The analysis suggested that the outcome-predictive biomarker pattern was consistent with reported data from 115 different upstream regulators (Table III in the Data Supplement) of which the most prominent (combined P value and Z score) activators were the endotoxin lipopolysaccharide, IL-1β and TNF. Similarly, typical inflammatory signaling pathways, such as IL-6–signaling, IL-8–signaling, and the coagulation system canonical pathways, were predicted to be upregulated (Table IV in the Data Supplement). The comparison to other functions and diseases is illustrated as a heatmap in Figure 3 (see Table V in the Data Supplement for complete list) where the pattern of outcome-predicting biomarkers is compared with annotated mechanisms and diseases. The pattern highlighted increased immunologic and inflammatory processes as main mechanisms associated with outcome. There was also an association between worse outcome and reduced apoptosis and reduced cell death, as illustrated by the inverted (blue) correlation with these features in the “Cell death and Survival”-annotated square.

• Download figure
• Download PowerPoint

Discussion

In this proteomics approach, numerous novel individual biomarkers and biomarker clusters representing inflammation, immune activation, growth factors, and inhibition of apoptosis independently associated with HFpEF severity and outcome, were identified. These may represent pathophysiological pathways implicated in global inflammation and may for prognostication also complement or even be more important than traditional markers, such as NT-proBNP.

Biomarkers of Inflammation as Predictors of Prognosis

Systemic inflammation is proposed to be one potential major driver of HFpEF, causing coronary microvascular endothelial inflammation and the development of diastolic dysfunction.^8,22–25 Indeed, numerous single biomarkers of inflammation have previously (eg, GDF-15, galectin-3, ST2, IL-6, and IL-8^{13,16,17,26,27}) and now, to our knowledge, for the first time (eg, chitinase-3–like protein 1 and C–C motif chemokine 20) been found to be associated with HFpEF severity and outcome. GDF-15 was the strongest predictor of outcome among 87 different biomarkers in this study, consistent with previous analyses.^26,28 In addition, we present novel prognostic information on the pseudochitinase chitinase-3–like protein 1 (YKL-40), present in inflammatory cells, such as in granules of neutrophils, which has been associated with the increased risk of atrial fibrillation,²⁹ a common comorbidity in HFpEF. Chitinase-3–like protein 1 is a marker of disease severity in many systemic inflammatory diseases, including asthma,³⁰ and is also a predictor of mortality in HFrEF.³¹

Previous studies have analyzed single or few biomarkers and because of complex relationships between numerous biomarkers, independent roles of the relevant biomarkers have not been established. In this study, our methodology allows simultaneous and comprehensive assessment of a large number of biomarkers as we used the OPLS approach to simultaneously contextualize a large set of biomarker data in relation to all clinical variables recorded. The OPLS yields a pattern of correlation coefficients of all variables represented in X, which in the case of biomarkers can be viewed as a fingerprint for the analyzed diseased state or outcome. The model had a high degree of explanatory power, 96%, as a result of the integration of all the clinical variables in the analysis. There were for instance a handful of variables related to outcome (eg, death and time to death) that contributed to the strong explanation of the model. Removal of these “outcome-related” variables reduced R2Y to 0.41 and Q2 to 0.013. By keeping all available variables in the OPLS, we minimized the risk of bias that could have been a result by hypothesis-driven exclusion of variables. The principal component analysis is built on an integrative analysis of the pattern formed by all of the variables, rather than a compilation of univariate analyses. Thus, although there is a statistical chance that a single variable may covary with the outcome-predicting pattern by chance, the chance of such a spurious correlation having an impact on the overall pattern is minimized, inasmuch as it would be “diluted” by all the other variables. By comparing the generated fingerprint with databases containing annotated data from the public domain, insights into disease biology can be gained in an expedited and unbiased manner as compared to traditional literature searches. As a result, the value of the biomarker, and also the interbiomarker relationship, for predicting HFpEF severity and prognosis could be illustrated.

As NT-proBNP was shown to predict prognosis in HFpEF,³² several biomarkers have been evaluated,^15,27,33,34 but no comprehensive assessments have to our knowledge been performed. When comparing prognostic impact of biomarkers, IL-18 has been suggested to be a superior predictor of mortality in multivariable analyses, including BNP.³⁵ In this study with more data input, IL-18 quantified by ProSeek did not have any independent predictive value for outcome. Interestingly as many as 21 of the biomarkers, among them several inflammatory markers, outperformed NT-proBNP determined by ELISA (Figure 2; Table 2).

Recent findings have highlighted the information we obtain from markers of inflammation and oxidative stress for the development of the HFpEF syndrome^17,36 and a specific role of TNFα and its receptors and IL-16 correlating with myocardial fibrosis, and stiffening has been demonstrated in patients with HFpEF compared with HFrEF and controls.^37,38 Cytokines signaling inflammation, such as IL-8, IL-6, and PTX3 (a marker for the primary local activation of innate immunity), and low levels of matrix metalloproteinase-1 (indicating a decrease in the major collagenase system) are all suggested to be correlated with diastolic dysfunction in HFpEF.^12,16,17,39 In this study, we confirm correlation with diastolic dysfunction expressed as E/e′ for several inflammatory markers, such as IL-6, GDF-15, and C–C motif chemokine 20. However, when analyzed individually, the correlation with structural heart disease was less prominent, indicating that functional class and prognosis in HFpEF may be better reflected by markers of inflammation and apoptosis, or that the echocardiographic measurements remain poor markers of the severity of the HFpEF syndrome. This is supported in a recent study where miRNA combinations, including lower expression of miR-146a, associated with TNFα and IL6, differentiated HFpEF from HFrEF but did not correlate with LAVI.⁴⁰ Inflammatory biomarkers such as cytokines have pleiotropic effects, which depend on signaling, cellular, organ and disease context.⁴¹ When comparing the outcome-predicting biomarker pattern with previously published data compiled in the IPA knowledge base, inflammation, hyper-proliferation, angiogenesis, and anti-apoptotic activity appeared as processes likely to be associated with poor prognosis. This pattern is in agreement with recent findings, suggesting impaired myocardial oxygen delivery possibly relating to microvascular dysfunction as a possible mechanism driving HFpEF.^25,42,43 This hypothesis is also consistent with histopathologic findings of more prominent myocardial fibrosis and coronary capillary rarefaction in this patient group.⁴⁴

Biomarkers as Prognostic Protective Predictors

We found the TNF superfamily member (TRAIL) to be negatively associated with outcome, with a reciprocal pattern of the soluble form of its receptor TRAIL-R2 (which was positively associated with outcome). By activating cell death–related pathways, members of the TNF superfamily can induce a variety of effects through the TNF receptors on cells, for exmple, within the myocardium, including TRAIL-mediated apoptosis. Not only lower levels of the ligand but also increased levels of the decoy receptor were associated with outcome, suggesting a beneficial role of TRAIL signaling in patients with HFpEF. In the myocardium, this would seem plausible given that myocardial hypertrophy is a major component of the HFpEF syndrome.⁴⁵ Cardiomyocyte cell death has been associated with worse prognosis in HFrEF⁴⁶ but is less prominent in HFpEF. However, although cardiomyocyte apoptosis may be disadvantageous in HFrEF, with reduced contractility, it is conceivable that global systemic apoptosis may be advantageous, for example, by limiting inflammation and downstream remodeling. Notably, higher levels of TRAIL have previously been associated with a better prognosis in HFrEF,⁴⁷ as observed here for HFpEF perhaps also reflecting the need for TRAIL-induced apoptosis to resolve inflammation. Another member of the TNF superfamily, TRANCE, was also negatively associated with outcome. Similar to the TRAIL-R2 pattern, the corresponding soluble decoy receptor for TRANCE (OPG) was also positively associated with outcome, although this did not reach statistical significance in the OPLS analysis. TRANCE has previously been suggested to be implicated in myocardial remodeling,⁴⁸ and the ratio between OPG and TRAIL has been shown to be associated with HF after myocardial infarction.⁴⁹ In addition to being a regulator of calcification and immunity, the TRAIL–OPG axis is also involved in apoptosis, for exmple, by the OPG-binding and neutralization of TRAIL.⁵⁰

Pathophysiological Pathways

Compared to the use of single biomarkers in the characterization of pathophysiology and prognosis, multiplexed analyses combine numerous biomarkers creating patterns of up and downregulated processes to more broadly shed light on potential underlying pathophysiology and to identify the most prominent pathways associated with both severity and prognosis and also potentially with pathophysiology. This methodology is particularly relevant now when large biomarker data sets are readily generated through novel technologies, for example, by next-generation sequencing, by mass cytometry, or by multiplexed protein analysis. Using multivariate projection methods, dissecting patterns of data in large data sets, identifies dimensions describing the majority of data in a data set. Recently, Zordoky et al⁵¹ investigated the metabolomic profile of HFpEF by analyzing 180 different endogenous metabolites. They reported higher serum concentration of long-chain acylcarnitines in patients with HFpEF compared with both patients with HFrEF and non-HF controls and identified a set of 4 biomarkers (2-hydroxybutyrate, octadecenoylcarnitine [C18:1], hydroxyprionylcarnitine [C3–OH], and SM [C24:1]) that without NT-proBNP produced a receiver operating characteristic area under the curve value of 0.908 to differentiate between HFpEF and HFrEF patients.⁵¹ Our data in HFpEF patients with a poor outcome identified a biomarker pattern consistent with an upregulated innate immune response but also impaired apoptosis. Leukocyte apoptosis is a key process for control of inflammation, which is a regulated rather than a passive process.⁵² Examples of resolution factors are lipoxins, the paucity of which results in chronic inflammation.⁵³ Recently, it was shown that the reduced levels of lipoxins predict poor outcome in chronic HF patients with reduced LVEF, suggesting together with our present data that improperly regulated chronic inflammatory reactions may be a common feature of chronic HF, irrespective of subtype.

Limitations

Caution should be taken when interpreting concentrations of plasma biomarkers in the description of pathophysiological mechanisms in specific tissues as they may not necessarily reflect processes in that tissue, for example, in the myocardium. However, the use of biomarkers provides us with pathophysiological clues on how to proceed and validate our findings further, especially when the evaluation of biomarker patterns are used instead of single biomarker approaches.

Of the 87 analyzed ProSeek biomarkers, there were 3 for which the platform displayed a considerable interassay variability (CV≥25%). One of these was significantly associated with outcome (matrix metalloproteinase-12), whereas the others (matrix metalloproteinase-1 and tumor necrosis factor ligand superfamily member 14) were not. Caution should be taken regarding these biomarkers, inasmuch as the high-assay variability may both yield falsely positive and negative associations.

The NT-proBNP criterion of >300 ng/L was used to select patients with a reasonable likelihood of heart failure and discriminate from other causes of dyspnea. It is increasingly recognized that NT-proBNP may be elevated in the absence of HF in patients with combinations of female sex, higher age, lower body mass index, renal insufficiency, and atrial fibrillation. However, the inclusion criteria were similar to other trials and studies of HFpEF, and the diagnosis was verified by Framingham criteria.

The LVEF inclusion criterion was set at 45%, which was a cutoff used in many of the HFpEF trials.^4,6 In the Karen biomarker substudy presented here, only two patients had an LVEF of <50%, thus the vast majority of the population did in this aspect comply with the present criteria of HFpEF as recommended by the American College of Cardiology Foundation and American Heart Association and the European Society of Cardiology guidelines (LVEF ≥50%).^21,54

A limitation of this study is that the data are based on a relatively small single cohort, which may represent a bias in terms of geographical and ethnical aspects.

Conclusions

In HFpEF, novel biomarkers reflecting inflammation and hyper-proliferative, proangiogenetic, and anti-apoptotic activity and their interrelation are linked to disease severity and prognosis. This may support the hypotheses of global inflammation, implicating microvascular inflammation, which may be a potential treatment target in patients with HFpEF. Furthermore, the prognostic information may complement or even be more important than for traditional markers, such as NT-proBNP. The validity of these findings for the overall population with HFpEF will require the replication of the data in ethnically and geographically distinct cohorts.

Acknowledgments

We are grateful to Gunilla Förstedt at Karolinska University hospital for blood sampling and patient care, Kambiz Shahgaldi and Maria Westerlind for echocardiogram assessments and to Eva Wallgren for laboratory analysis.

Sources for Funding

This work was supported by grants from Center for Gender Medicine Karolinska Institutet, Stockholm, Sweden (to Dr Hage) and The Prospective Karolinska Rennes (KaRen) study was supported, in part, by grants from Fédération Française de Cardiologie/Société Française de Cardiologie, France, and Medtronic Bakken Research Center, Maastricht, The Netherlands. Dr Lund was supported by the Swedish Research Council (grant 2013-23897-104604-23) Swedish Heart Lung Foundation (grants 20080409 and 20100419), and the Stockholm County Council (grants 00556-2009 and 20110120). Dr Linde was supported by the Swedish Heart Lung Foundation (grants 20080498 and 20110406) and the Stockholm County Council (grants 20090376 and 20110610). No funding agency had any role in the design and conduct of the study, in the collection, management, analysis, or interpretation of the data.

Footnotes