Progress in proteomic probing for pathogenic pathways in heart failure with preserved ejection fraction

This article refers to ‘Proteomics to improve phenotyping in obese patients with heart failure with preserved ejection fraction’ by K.P. Kresoja et al., published in this issue on pages 1633–1644. Heart failure (HF) with preserved ejection fraction (HFpEF), a common HF phenotype with an ominous prognosis and scant evidence-based therapeutic options, may be the consequence of common comorbidities which foster a systemic pro-inflammatory state leading to endothelial dysfunction and coronary microvascular dysfunction with abnormal cardiac structure and function. This contrasts somewhat with HF with reduced ejection fraction (HFrEF) where the single most common triggering insult is an acute myocardial infarction and HF most often ensues as a consequence of a defined focal cardiac injury (Figure 1).1, 2 The spectrum of comorbidities observed in HFpEF is broad and encompasses hypertension, diabetes, renal impairment and obesity. Obesity is present in up to 80% of cases of HFpEF and this comorbidity is known to be associated with increased circulating volume load and more severe abnormalities in both left and right heart structure and function than observed in non-obese HFpEF. Notably, left ventricular filling pressures are directly related to body mass and plasma volume in obese but not non-obese HFpEF.3 An understanding of the molecular links between comorbidities and HFpEF might offer novel therapeutic targets. On this background, in this issue of the Journal Kresoja et al.4 aimed to identify differentially expressed circulating biomarkers in obese HFpEF patients and link them to disease severity and outcomes. Ninety-two circulating serum biomarkers measured by proximity extension assay (PEA) were compared between obese HFpEF, non-obese HFpEF and non-HF participants in the LIFE-Heart study. Adrenomedullin (ADM), galectin-9 (Gal-9), thrombospondin-2 (THBS-2), CD4, and tumour necrosis factor-related apoptosis-inducing ligand receptor 2 (TRAIL-R2) were significantly higher in obese HFpEF as compared with both lean HFpEF and obese non-HF patients. Except for CD4, these proteins were associated with increased left atrial pressure. ADM, TRAIL-R2 and CD4 were associated with increased mortality in obese HFpEF patients. The primary focus is on arbitrary phenotypic categories (left ventricular ejection fraction ≥50% and body mass index >30 kg/m2) but the report is also strengthened by demonstration of graded relationships between the markers highlighted and key indicators of cardiac structure and function assessed as continuous variables. Protein scores derived from ADM, CD4, Gal-9, THBS2 and TRAIL-R2, as well as the individual proteins, were associated with higher HFA-PEFF scores, left atrial dilatation, left ventricular mass index and increased C-reactive protein. Proteins also associated with E/e′ after adjustment for age, sex and renal function. The authors conclude: ‘Obese HFpEF patients exhibit higher circulating biomarkers of volume expansion (ADM), myocardial fibrosis (THBS-2) and systemic inflammation (Gal-9, CD4) compared to obese non-HFpEF or lean HFpEF. These findings support the clinical definition of a distinct obese HFpEF phenotype and might merit further investigation.’ The report relies heavily upon PEA-based proteomic screening: ‘using the Olink (Uppsala, Sweden) Cardiovascular II panel. This panel was chosen as it reflects biomarkers proposed to be associated with inflammatory, immuno–cardiovascular interaction, tissue and vascular neogenesis as well as cardiometabolic interaction.’ Olink panels elegantly combine the strengths of immunoassay and polymerase chain reaction principles and are increasingly recognized as extremely useful screening tools.5 However, their use entails inherent pre-existing biases reflecting the prior ‘proposed’ associations with cardiovascular disease molecular pathology which underly selection of the 92 markers on the Olink panel. Tissues release proteomic components, via endo/exocytosis or secretion, which more or less readily transition through the extracellular space into the circulation. Proteomic data sets including the UNIPARC, Homo sapiens RIKEN, ENSEMBL, IMAGE, RefSeq, UNIPROT, and SwissProt Federated Library indicate at least 157 478 human peptide/protein sequences differing by one or more amino acids are present in humans.6 Extraction of endogenous tryptic peptides from small volumes of human plasma by liquid chromatography followed by nano-electrospray ionization and tandem mass spectrometry can now identify products from at least 14 000 protein-coding genes in human blood.7 PEAs yield highly sensitive and specific detection of multiple analytes — currently extending from a format assessing 92 analytes to as many as 5000 in as little as 1 µL of plasma. However, the 92 analytes underpinning the current report represent a very small proportion of peptide/protein sequences measurable in human plasma. They are pre-selected for inclusion in the PEA panel because of presumed prior knowledge of their possible roles in cardiovascular disease processes. Hence the study uses a tool necessarily subject to prior assumptions on what should be measured coupled with limitation to a very small part of the full proteomic picture. The plasma analyte profile necessarily omits the majority of intracellular signalling pathway components and any extrapolation to emphasize and implicate particular molecular pathways may be distorted by the artefactual presentation of some pathways as relatively activated or suppressed simply because of the measurement of some circulating entities but continued ignorance of the vast majority. Therefore marker profiling/screening by the current generation of Olink and other methods should be viewed as hypothesis-generating. Such data require validation by quantitative assays of the identified markers in independent cohorts in addition to appropriate basic science experimentation to elucidate the mechanisms linking a particular marker set to the disease state of interest. However, proteomic and metabolomic screening techniques are becoming increasingly comprehensive, producing more powerful insights that can more rapidly and accurately focus subsequent research with a corresponding accelerated elucidation of molecular pathologies. Early examples of such next-generation studies include the recently published SomaScan (DNA aptamer-array proteomic screening) search for predictors of post-myocardial infarction adverse ventricular remodelling and HF amongst 1300 circulating analytes measured by this platform.8 More pertinent to the current report, the PROMIS-HFpEF study deployed three Olink panels to measure 248 unique circulating proteins.9 Interestingly, PROMIS-HFpEF highlighted biomarkers which clustered in panels plausibly mediating the adverse effects of comorbidities (including obesity, hypertension, coronary artery disease, diabetes and others) upon cardiac structure and function in HFpEF. TRAIL-R2, CD4 and ADM, identified within weighted coexpression network analyses (WCNA) of the PROMIS-HFpEF data, overlap with the markers of note presented by Kresoja et al., although the PROMIS-HFpEF study did not explore any relationship specifically to obese HFpEF. Beyond uncertainties with respect to the comprehensiveness or relevance of the profiling conducted when attempting to ascertain markers pertinent to a condition of interest, interpretation of data from complex data sets relies on selection of appropriate analyses. In the current report, analyses were limited to conventional biostatistics. Associations between markers and the obese HFpEF phenotype, and with mortality in this population, were adjusted in multivariable modelling for age, sex, presence of diabetes, creatinine, presence of coronary artery disease, white blood cell count and haematocrit: ‘as those variables might influence circulating biomarker levels.’4 In cross-sectional associative cohort studies, multivariable adjustments are a valuable tool to address possible confounding in the attempt to ascertain true independent associations between marker(s) and selected conditions/outcomes of interest. The ever-present challenge is in knowing whether such adjustments are adequate or complete. For example, other potential confounders which may alter circulating biomarker levels, not incorporated in the models presented by Kresoja et al., might include: (i) comorbidities with an inflammatory component; (ii) drugs prescribed differentially between phenotypes; (iii) blood pressure with its associated all-pervasive effects on arterial tone, vascular resistance and local tissue perfusion, and (iv) effects of undocumented inter-group differences in exercise and diet, which may well differ between obese and non-obese populations. These potential confounders can only be addressed by further research in sizable clinical cohorts with more fine-grained annotation. As proteomic profiling by either targeted (Olink, SomaLogic) or assorted unbiased mass spectroscopy techniques are applied to generate ever-larger and more complex data sets, our analytical and bioinformatics approach will have to become correspondingly more sophisticated incorporating adequately detailed consideration of background context (including comorbidities). Management of multi-dimensional data requires techniques such as principal component analysis and WCNA. In the evolving world of ‘omics’-grade complexity with concurrent access to thousands of analyte results per biological sample, the provision of a common ‘language’ among investigators to standardize terms such as ‘inflammation’ is needed to progress collaborations and allow valid comparisons between data sets. This will arise over time through establishment of commonly accepted scores derived from defined combinations of inflammation-related markers integrated into intuitively accessible forms (scores) by means of principal component analysis, WCNA and other bioinformatic tools. Notably the study design, in common with many other efforts to study HFpEF, omit any data from patients suffering HFrEF.4, 9 Therefore we cannot rule out the possibility that the combination of obesity and HF may result in an adverse synergism, reflected in the marker profile the authors have reported, quite independent of whether left ventricular ejection fraction is preserved or not. Conceivably, the markers highlighted are even more profoundly perturbed and more sharply prognostic in obese HFrEF than in obese HFpEF. This is a clear knowledge gap to be addressed in ongoing research. Notwithstanding consideration of the issues discussed above, Kresoja et al. have taken us one further step towards elucidating the pathophysiology underlying clinically important subsets of HF. The findings encourage further exploration of the roles, whether culpable or compensatory, of ADM, CD4, Gal-9, THBS-2 and TRAIL-R2 in the evolution of HFpEF compared with HFrEF in the presence and absence of obesity. In addition to further observational cohort studies, this might entail interrogation of pertinent in vitro and in vivo models in which these candidate marker entities, or components of the molecular pathways in which they lie, either singly or in combination, are augmented or blocked in resting and challenged states using transgenic and gene knock-out methods and/or siRNA/antibody based manipulations in preclinical obese and non-obese control and HF models. In this way increasingly powerful screening tools, such as PEAs, inform and direct definitive, focussed research elucidating specific mechanisms leading to novel, personalized therapies in HF and other complex syndromes. Conflict of interest: none declared.