Research Priorities for Heart Failure with Preserved Ejection Fraction: National Heart, Lung, and Blood Institute Working Group Summary

Cardiac pathophysiology

HFpEF was initially conceptualized as an isolated disorder of LV diastolic dysfunction, but numerous studies over the past two decades have revealed that HFpEF is a significantly more complex, highly-integrated multisystem disorder. LV diastolic stiffness and relaxation are clearly abnormal in HFpEF with impaired LV diastolic stiffness and relaxation which result in elevated LV filling pressures at rest and during exercise,¹⁴ and contribute to to symptoms of dyspnea, lung gas exchange abnormalities, impaired aerobic capacity,¹⁵ and the development of pulmonary hypertension, eventually culminating in the increased morbidity and mortality observed in this population. While organ system involvement in HFpEF may display substantial heterogeneity between individuals,¹ elevation in LV filling pressures, at rest or with exercise, represents a unifying hemodynamic derangement.¹⁴

LVEF is normal in HFpEF, but myocardial LV contractility and systolic mechanics are typically impaired.¹⁶ These relatively subtle deficits in resting LV contractile function, which are present along with normal or supra-normal resting systolic LV chamber stiffness, lead to dramatic limitations in systolic reserve with exercise, impairing the ability of the LV to contract to low end-systolic volumes.⁴ This impairs diastolic suction, thereby exacerbating the elevation in filling pressures¹⁵ while simultaneously impairing the ability of the heart to augment stroke volume, and cardiac output during stress.^{4, 15} The chronotropic response to exercise is also impaired in most patients with HFpEF and contributes to reduced exercise cardiac output.^{4, 17}

Coronary microvascular function is impaired in the majority of patients with HFpEF,¹⁸ and coronary microvascular rarefaction was demonstrated in autopsy HFpEF specimens where it was associated with the severity of myocardial fibrosis.¹⁹ Diffuse microvascular inflammation is believed to contribute to microvascular dysfunction and ultimately rarefaction,^{9, 20} but this needs to be further established. Coronary microvascular dysfunction may contribute to myocardial injury at rest and ischemia during exercise and thus cardiac reserve limitations in HFpEF.²¹ Indeed, an overarching theme that applies to the heart, as well as other systems in HFpEF, is the concept that organ level reserve capacity is profoundly diminished in HFpEF.^{4, 15} At the present time, it is unclear what causes these reserve limitations, though impairments in nitric oxide (NO) signaling, metabolism, inflammation, oxidative stress, and myocardial ischemia may each play important roles that require further elucidation.

Systemic vascular dysfunction

The normal reduction in systemic vascular resistance with exercise is blunted in patients with HFpEF,^{4, 15} and arterial stiffness is elevated, which leads to earlier arrival of reflected waves to the heart, increasing pulsatile arterial loading that exacerbates LV remodeling, fibrosis, diastolic dysfunction, and pulmonary venous hypertension.²² Small vessel endothelium-dependent vasodilation is depressed in HFpEF,⁴ which, together with impairments in cardiac output,^{15, 22} further impairs the delivery of oxygen to the tissues. Venous system function is also impaired in HFpEF, and further study is required for better understanding of systemic venous compliance/congestion and its consequences in HFpEF.

Skeletal muscle pathophysiology

Skeletal muscle sarcopenia, reduced capillary density, increased intermuscular fat, reduced sirtuin-AMPK signaling and glucose utilization, and abnormal mitochondrial content and function have been observed in patients with HFpEF.^{23, 24} Skeletal muscle abnormalities and vascular dysfunction impair the appropriate distribution and utilization of oxygen in the tissues, contributing to the severely reduced aerobic capacity in patients with HFpEF.¹⁷ Future studies should integrate detailed skeletal muscle assessment in early phase HFpEF trials; and later-stage trials, which often evaluate exercise capacity, should differentiate between the relative impact of putative therapies on cardiac vs. skeletal muscle limitations.

Pulmonary pathophysiology

Chronically elevated LV filling pressures promote left atrial (LA) remodeling and dysfunction that contribute to increased burden of atrial fibrillation, worsening functional capacity, and increased mortality.^{25, 26} Pulmonary hypertension (PH), typically defined as a mean pulmonary artery pressure ≥25 mm Hg, commonly develops secondary to elevated LA pressures in HFpEF and is associated with increased mortality. In some patients there is progressive increase in pulmonary vascular resistance (PVR) in addition to passive elevation of LA pressures, referred to as combined pre- and post-capillary PH (CpcPH) or HFpEF with pulmonary vascular disease. Although there is controversy rergarding the definition of CpcPH, the presence of pulmonary vascular disease in HFpEF is associated with higher hospitalization and mortality rates.²⁷

Progressive pulmonary vascular disease leads to impaired gas exchange, with low diffusion capacity for carbon monoxide observed in half of patients with HFpEF and PH. Pulmonary venous and small vessel intimal thickening and fibrosis are more severe than arterial intimal thickening in patients with HFpEF, analogous to that seen in pulmonary veno-occlusive disease.²⁸ There is also evidence that lung congestion itself may be associated with pulmonary vascular disease in patients with HFpEF, possibly due to effects of increased lung fluid on vascular resistance. Patients with HFpEF and pulmonary vascular disease display a unique pathophysiology during exercise compared to patients with isolated left heart disease.²⁹ Remodeling and pulmonary vasconstriction limit the ability of the RV to transfer blood into the lungs as venous return increases, causing right heart overload, heightened pericardial restraint, and compromised LV filling through enhanced ventricular interdependence.²⁹

Longstanding PH leads to RV dysfunction, tricuspid regurgitation and right-sided heart failure in many patients with HFpEF, and these abnormalities are associated with poor outcomes in HFpEF.¹³ Atrial fibrillation, coronary disease, and obesity are also important non-hemodynamic factors predisposing to the development of RV dysfunction in HFpEF. RV dysfunction is not exclusive of advanced HFpEF, because limitations in RV reserve are present even in the earliest disease stages, suggesting that systemic processes affecting both ventricles are operative throughout the natural history of HFpEF.¹⁵ Similarly, abnormal right atrial conduit and reservoir function have been reported in early HFpEF without overt PH.

Despite advances in our understanding of pulmonary pathophysiology in HFpEF, several unanswered questions remain, including why some patients with HFpEF develop CpcPH while others do not, and what are the optimal means to prevent and treat of RV and right atrial dysfunction in HFpEF.

Renal pathophysiology

Chronic kidney disease (CKD) is common in HFpEF patients and its severity is correlated with time to cardiovascular events or death.³⁰ Albuminuria is a potent risk factor for HFpEF that is associated with abnormal myocardial mechanics³¹ and may be a novel treatment target, but further investigation is needed. For example, activation of the vitamin D receptor with paricalcitol reduced albuminuria, but did not improve LV diastolic function or LV hypertrophy in patients with CKD.³² There is important bi-directional cross-talk between the heart and kidney. For example, elevated cardiac troponin, which is associated with greater severity of HFpEF,²¹ also relates to CKD progression.³³ Such cross-talk is further supported by data from healthy kidney donors, who often develop LV hypertrophy within 1 year after nephrectomy, even as ambulatory blood pressure remains unchanged.³⁴ In rats with subtotal nephrectomy, cardiomyocyte hypertrophy develops that is not matched by a parallel growth in the capillary network.³⁵ Hypertrophy without proportional angiogenesis is a fundamental characteristic of pathologic (versus physiologic) hypertrophy, and could be a cause of the impaired subendocardial perfusion and abnormal longitudinal strain that in HFpEF.

Venous congestion is common in HFpEF and may contribute to worsening renal function due to renal venous hypertension, which has been shown to impair renal function in animal models.³⁶ Elevated renal vein pressure, particularly in the setting of low systemic blood pressure, can result transglomerular gradient and thereby reduce kidney perfusion. Renal venous hypertension may also impair renal function due to intrarenal congestion which can further impede renal perfusion.

The mechanisms underlying the bi-directional heart-kidney relationships and the cardiorenal syndrome that is present in in HFpEF remain elusive but clearly drive HFpEF pathogenesis. Understanding the kidney’s role in determining why some patients with the myocardial substrate for HFpEF do not develop volume overload while others do could lead to improved strategies to prevent HFpEF and limit HF hospitalizations in patients with prevalent HFpEF.

Adipose tissue pathophysiology

Dysregulation of energy storage (adipose tissue), as occurs in obesity, also plays a key role in HFpEF. Increased metabolic requirements may lead to high output HF with preserved EF in some obese patients, but most obese HFpEF patients do not have high output HF.³⁷ Rather, obese HFpEF patients display greater plasma volume expansion, exaggerated cardiac remodeling, impaired RV-pulmonary arterial interaction, and heightened pericardial restraint owing to cardiomegaly and increased mediastinal and/or chest wall fat.³⁷ Obesity is associated with greater symptom severity in HFpEF, and metabolic changes associated with obesity have been implicated in the pathogenesis of PVD.²⁴ Excess adipose tissue causes systemic inflammation and impaired NO signaling, believed to play key roles in HFpEF.^{1, 9} Skeletal intermuscular adipose is increased and correlates with exercise intolerance,³⁸ and excess intramyocardial fat may also be present. Other abnormalities in fat may contribute to organ dysfunction in HFpEF, but the mechanisms are unclear.

Obesity appears to be a major driver of the HFpEF syndrome; however, diagnosing HFpEF in the setting of obesity can be challenging for a variety of reasons, including lower levels of natriuretic peptides, difficulties in determining volume status on physical examination, and the overlap of symptoms between obesity alone and obesity-related HFpEF. Furthermore, severely obese patients are often excluded from HFpEF clinical trials.³⁹ Future studies of HFpEF should specifically examine the diagnosis, pathophysiology, treatment, and prevention of the obesity HFpEF phenotype, and greater inclusion of morbidly obese HFpEF patients in clinical trials should be made a priority.

Natriuretic peptides

The cardiac natriuretic peptides (NPs) ANP and BNP, which are released from the heart in response to myocardial stress, have diverse physiological actions, mediated by binding to NP receptor A (NPRA), thereby increasing cGMP/PKG signaling.^{41, 42} The NP receptor C (NPRC) functions to clear the NPs from circulation through receptor-mediated internalization.⁴³ The ability of the NPs to elicit a biological response depends on the relative ratio of the functional receptor NPRA to the clearance receptor NPRC.

For any given amount of volume overload and congestion, NPs are lower in HFpEF compared to HFrEF due to lower diastolic wall stress in HFpEF compared to HFrEF. While LV diastolic pressures can be equally high in HFpEF and HFrEF, concentric LV remodeling in HFpEF (compared to eccentric remodeling in HFrEF) translates to lower wall stress and thus a reduced stimulus for BNP secretion from the myocardium, because chamber radius is lower and wall thickness is higher in HFpEF. Other factors associated with reduced circulating NP levels include obesity, insulin resistance, increased androgenicity in women, and African ancestry.^{44, 45}

NPs affect natriuresis and blood pressure control, and likely play a role in metabolism in HFpEF. As detailed above, obesity is a common comorbidity in patients with HFpEF, and it appears to be directly related to HFpEF pathogenesis. NP receptors were first discovered in adipose tissue.⁴⁶ ANP stimulates lipolysis in cultured human adipocytes with potency similar to the β-adrenergic agonist isoproterenol.⁴⁷ NPs can also increase mitochondrial biogenesis, brown fat thermogenesis, and induce the adipose ‘browning’ program in white adipose tissue.^{48, 49} These findings are consistent with clinical studies in humans that report NPs can increase energy expenditure and fat oxidation, independent of the β-adrenergic axis, including both adipose tissue and skeletal muscle.^{50, 51} In the Atherosclerosis Risk in Communities (ARIC) population-based study, a gain-of-function polymorphism in the NPPB gene, which encodes for BNP resulting in higher lifelong BNP levels, was associated with lower BMI, lower blood pressure, and lower risk of incident cardiovascular disease.⁵² Thus, higher NP levels are likely protective against HFpEF, and low NP levels may be major factors involved in the pathogenesis and progression of HFpEF.

The link between obesity and low cardiac NPs originated with reports that receptors for ANP and BNP are present in adipose tissue,⁴⁶ and that obese humans often have substantially higher amounts of the NP ‘clearance receptor’ NPRC in adipose tissue and lower circulating NPs. An increase in NPRC relative to the signaling receptor NPRA renders the tissue less responsive to NPs. Elevated NPRC in adipose tissue of obese individuals gave rise to the notion that adipose tissue might be a ‘sink’ for circulating NPs which could contribute to the hypertension that is often associated with obesity, thereby leading to and exacerbating obesity-related HFpEF. Very high levels of NPs, which can be present in advanced HFpEF, may also be problematic. Persistent NP-induced increases in lipolysis and energy expenditure over time could be detrimental. As some have discussed,⁵³ chronically elevated fatty acids can lead to their to their ectopic accumulation in liver, skeletal muscle, heart, and pancreatic β-cells, and the possibility that high levels of NPs contribute to cardiac cachexia is currently being debated.

It may be advantageous to identify NP analogues that preferentially interact with the ‘signaling’ receptor NPRA. Given that obesity is a frequent driver of HFpEF, increasing adipose metabolism, brown adipocytes activity, and energy expenditure—as well as potential increases in muscle fatty acid oxidation—could be beneficial for reducing BMI and increasing insulin sensitivity and attenuating HFpEF (see detailed review of brown adipose tissue and metabolism in humans⁵⁴). Further study should determine whether targeting NP signaling in adipose tissue provides a novel therapeutic approach to HFpEF.

cGMP/PKG signaling

One potentially unifying mechanistic theory for cardiac dysfunction in HFpEF is comorbidity-driven systemic and microvascular inflammation, leading to oxidative stress, leading to reduced NO bioavailability, impaired soluble guanylate cyclase (sGC) activity, reduced cGMP and thus reduced PKG signaling.⁹ Inflammation has also been implicated in increased nitrosative stress in HFpEF, leading to accumulation of unfolded proteins, which damage cardiomyocytes and result in cardiomyocyte dysfunction.⁵⁵

One study reported that PKG activity is depressed and associated with low levels of cGMP in LV myocardial biopsies from HFpEF patients.⁵⁶ However, the sample size was small, controls (patients with HFrEF or aortic stenosis) were less than ideal, and there was no comparison with normal or non-failing tissue. Some limited data suggest increased pro-inflammatory markers in LV tissue^{20, 57} and depressed vascular NO signaling. Whether vascular NO has a primary impact on cardiomyocyte cGMP signaling remains uncertain, as the myocyte has the synthetic machinery to generate cGMP by both NO and NP pathways. A decline in PKG activity is implicated in increased myocardial stiffening due to reduced titin phosphorylation (see below), as exposure to activated PKG can reverse this stiffness.^{56, 58} Hypertension and pressure overload animal models with associated cardiac hypertrophy and fibrosis have shown that PKG activation can reduce abnormal diastolic stiffness.^{59, 60} While these models are not truly HFpEF, they support the concept that components of the disorder—particularly those associated with hypertension, hypertrophy, and fibrosis—may benefit from PKG activation.

PKG activation has been attempted by inhibiting phosphodiesterase type 5 (PDE5) to impede cGMP hydrolysis,⁶¹ stimulating soluble guanylyl cyclase,⁶² or administering inorganic nitrite or organic nitrate. PDE5 inhibition was studied in a multicenter randomized clinical trial and showed no benefit, possibly due to its association with worsening renal function.⁶³ In a phase 2 RCT,⁶² soluble guanylate cyclase stimulation did not lower NPs or LA volume in HFpEF, but may have improved skeletal muscle function and physical activity, a hypothesis which is currently being tested in 2 additional phase 2 HFpEF RCTs. NIH-sponsored cross-over RCTs testing organic nitrate (NEAT-HFpEF) and inhaled inorganic nitrite (INDIE-HFpEF) also did not show benefit for the primary endpoint (physical activity and exercise capacity, respectively).^{64, 65} However, pharmacokinetic characteristics of inhaled nitrite may have caused a lack of response, and trials testing oral formulations of inorganic nitrate/nitrite in HFpEF are currently underway ( NCT02713126, NCT02840799, and NCT02918552).

Each of the aforementioned methods to augment cGMP/PKG signaling involve the NO-sGC pathway, thereby enhancing intracellular cGMP with no effect on circulating levels of cGMP, which may explain their lack of benefit thus far in HFpEF. Another method to augment cGMP/PKG signaling is by stimulating the NP-receptor guanylyl cyclase (rGC) pathway. One method is blocking the peptidase neprilysin, which reduces NP proteolytic cleavage as one of its effects, thereby increasing cGMP/PKG signaling, resulting in increased circulating and urinary levels of cGMP. The combination of sacubitril (neprilysin inhibitor) and valsartan (angiotensin receptor type II blocker), which increases urinary cGMP levelsimprove outcomes in HFrEF, and was recently tested in HFpEF in the large-scale PARAGON trial.⁶⁶ PARAGON narrowly missed its primary endpoint (cardiovascular death and recurrent HF hospitalization), but demonstrated potential efficacy in women and in those with EF below the median (<57%).⁶⁶ An alternative method to stimulate the NP-rGC pathway is inhibiting the breakdown of cGMP via PDE type 9 (PDE9). PDE9 is the most cGMP selective member of the phosphodiesterase superfamily. It is upregulated in human HF, and in mouse models was shown to modulate cGMP generated by the NP pathway.⁶⁷ In the latter model, upregulation of PDE9 compromises PKG activity, and its selective inhibition ameliorates hypertrophy, fibrosis, and dysfunction induced by pressure-overload.⁶⁷ These effects persist despite declines in NO synthesis, unlike the ameliorative benefits from PDE5 inhibition that are NO-synthase dependent. This suggests PDE9 inhibitors may circumvent compromised NO signaling associated with oxidant stress and inflammation, as well as estrogen declines post-menopause. In addition, the absence of vascular effects with PDE9 inhibition could render it more attractive than PDE5 inhibition, and enhanced NP signaling from PDE9 inhibition may also confer beneficial effects on fat metabolism and obesity. PDE9 inhibitors were first developed for neuro-cognitive diseases, but Phase Ib-IIa trials of PDE9 inhibitors in HF are underway.

While deficiency of the cGMP-PKG signaling system, which has diverse beneficial actions in multiple organs and tissues, is likely a key driver of HFpEF pathogenesis, additional investigation is necessary to determine: (1) risk factors for cGMP-PKG deficiency in mid-life that could be targeted for HFpEF prevention; (2) why augmentation of cGMP-PKG in HFpEF has been unsuccessful to date and whether augmenting NP-rGC is more beneficial than augmenting NO-sGC; and (3) other molecular systems that may interact with cGMP-PKG signaling to drive HFpEF pathogenesis.

Titin: a key driver of myocardial stiffness

Although HFpEF is a multi-factorial, multi-organ disease, increased myocardial tissue stiffness is thought to play a key role in HFpEF pathogenesis. Increased LV chamber stiffness, as indicated by an upward- and leftward-shifted LV end-diastolic pressure volume relationship, is associated with worse outcomes in HFpEF,⁶⁸ and is caused by increased cardiomyocyte passive stiffness, myocardial tissue fibrosis, or both. Titin—a giant, sarcomere-spanning protein—is the main determinant of cardiomyocyte passive stiffness. Titin–actin interactions account for ~40% of LV viscosity and relaxation. Several studies have shown that titin’s stiffness is elevated in HFpEF.^{58, 60}

Alternative splicing gives rise to 2 titin isoforms, the larger and more compliant N2BA isoform and the smaller N2B isoform. Adult human myocardium contains approximately equal amounts of N2BA and N2B titin.⁶⁹ Altered titin splicing toward a higher N2BA:N2B ratio reduces passive cardiomyocyte stiffness and has been reported in dilated cardiomyopathy, where it may represent a beneficial adaptation.⁷⁰ Splicing is mediated by the splicing factor, RBM20.⁷¹ By genetically targeting RBM20, compliant titin isoforms can be upregulated in mouse models with HFpEF-like characteristics, ameliorating both diastolic dysfunction and exercise intolerance.^{72, 73}

Titin’s stiffness can also be modified by post-translational modifications. Phosphorylation at different sites within titin allows for rapid and site-specific increases or decreases in titin stiffness. Figure 1 displays several intracellular mechanisms by which titin’s stiffness can be modulated by altered phosphorylation. In mouse models, genetic interventions which increase or decrease the stiffness of titin’s spring are associated with corresponding increases or decreases in LV diastolic stiffness and decreases or increases^{72, 74} in exercise capacity. Thus, titin’s spring region represents an attractive target for improving LV diastolic function and symptoms in HFpEF.

Furture research should examine the molecular mechanisms by which phosphorylation causes stiffness to decrease (N2B phosphorylation) or increase (PEVK phosphorylation), including whether these molecular changes present unique druggable targets, and whether titin splicing can be altered in human myocardium to improve LV diastolic function. Titin’s complexity allows for multiple approaches to reduce stiffness that could be promising for HFpEF therapeutics.

Tissue fibrosis

Fibrosis, defined as excess deposition of extracellular matrix (ECM), including increased collagen accumulation and cross-linking, leads to tissue stiffening and organ dysfunction and is associated with HFpEF progression. Aging is associated with fibroblast activation and collagen accumulation/cross-linking.⁷⁵ Comorbidity-driven myocardial inflammation may also contribute to tissue fibrosis in HFpEF.⁹ However, there is significant heterogeneity in the amount of cardiac and extra-cardiac fibrosis in HFpEF. In a small sample of patient biopsies,⁷⁶ cardiac MRI studies that utilized T1 mapping to detect diffuse interstitial fibrosis,⁷⁷ and an autopsy study,¹⁹ showed varying degrees of myocardial interstitial fibrosis in HFpEF. Although variably present, myocardial interstitial fibrosis is associated with a worse prognosis in HFpEF patients,^{78, 79} though it is unclear to what degree this is independent of standard clinical factors. A large population-based cohort study demonstrated that circulating fibrosis biomarkers were associated with incident HFpEF, but not HFrEF.⁸⁰ Biomarkers of tissue fibrosis in HFpEF patients correlated directly with lack of response to therapy with spironolactone. Thus, tissue fibrosis seems important in HFpEF and its presence may modulate the efficacy of HFpEF treatments.

There are currently only 2 FDA-approved therapies for fibrosis in human disease: pirfenidone (unknown mechanism-of-action) and nintedanib (pan-tyrosine kinase inhibitor) and both are solely indicated for the treatment of idiopathic pulmonary fibrosis; however, their efficacy in pulmonary fibrosis is limited. Pirfenidone is being tested in a HFpEF clinical trial and several preclinical studies suggest potential of myocardial fibrosis inhibition by histone deacetylases, bromodomain proteins, ECM components, autophagy, signal transduction, microRNAs, BCL-interacting protein 3, as well as caloric restriction and several other approaches, though many of these were studied in HFrEF animal models or human myocardial samples and not in HFpEF.

The factors that stimulate fibrosis in HFpEF are likely multifactorial, but the interactions between circulating monocytes (and macrophages), the endothelium, and fibroblasts may play a key role in determining why some patients with HFpEF have excessive tissue fibrosis while others do not.⁸¹ HFpEF is a multi-organ disease; thus, recent advances in attenuating fibrosis in other organs, particularly lung and kidney, may also hold promise for the treatment of HFpEF.

There is an urgent need to better define the severity, mechanisms, biological impact and therapeutic implications of fibrosis in HFpEF using better biomarkers, imaging studies and ideally, myocardial tissue samples. The lack of efficacy of anti-fibrotic therapeutics in humans may be the results of the inability to differentiate reactive vs. reparative fibrosis, as the former is reversible while the latter may not be.

Animal models of HFpEF for the elucidation of HFpEF mechanisms

The ideal HFpEF animal model should mimic the major common features of human HFpEF, including cardiac, hemodynamic, neurohormonal and peripheral changesand should recapitulate the typical HFpEF phenotype: advanced age, female (in addition to male), and multiple comorbidities. HFpEF animal models were recently reviewed by Valero-Munoz et al.³ and Lourenco et al.⁸²

Murine models allow genetic manipulations, but human HFpEF is usually not a genetic disease. Thus, specific genetic manipulations may be of limited use. However, gene-targeted mice can be useful in the dissection of molecular mechanisms for specific HFpEF phenotypes and the development of therapeutics to investigate the role of a molecule/factor in the HFpEF syndrome. Genetic homogeneity of mouse models can also be problematic, as shown in a study that found wide variations in the development of HFpEF and PH in mice from 36 different genetic strains fed a high fat diet.⁸³ Thus, basic studies of HFpEF may benefit from validating findings in diversity outbred mice or mice with a variety of genetic backgrounds.

Purported animal models of HFpEF that progress to HFrEF are not typically representative since this transition rarely occurs in HFpEF patients.¹¹ Therefore, HFpEF models that dilate and reduce their LVEF (such as transverse aortic constriction) are considered poor models. Another major source of confusion, predominantly in the preclinical literature, lies in the distinction between diastolic dysfunction and HFpEF. These two terms have been used interchangeably, but diastolic dysfunction by itself is not enough to produce HFpEF. Furthermore, HFpEF is not simply a disease of aging nor does it occur only in females; thus, animal models should not be limited to aged or gender-specific models though these two critical biological variables require more detailed examination.

Similar to human HFpEF, a “one-size-fits-all” strategy is also unlikely to work in HFpEF animal models. Although comorbidities play a major role in human HFpEF, mouse models are less influenced by comorbidities. This suggests that perhaps large animals, which more readily recapitulate human pathophysiology and have recently been developed for HFpEF,^{84, 85} may be preferable. Nevertheless, ethical issues, difficulty with introducing exogenous genes, cost, and duration of study limit large animal models. Emerging “multi-hit” rodent models appear to more closely recapitulate the HFpEF syndrome compared to prior “single-hit” models, and therefore may be useful in future investigations. Examples include the ZSF-1 diabetic plus spontaneously hypertensive rat treated with a VEGF inhibitor to produce HFpEF with pulmonary vascular disease,²⁴ and the L-NAME plus high fat diet mouse model,⁵⁵ which combines a potent metabolic/obesity stimulus with a hypertensive stimulus, the 2 most common modifiable risk factors for the development of HFpEF. The aforementioned large animal models are also multi-hit models and therefore may also be useful in future investigations of HFpEF,^{84, 85}

In all animal models, extensive and careful characterization of cardiac and non-cardiac contributors to HFpEF phenotypes is essential. These include: hemodynamics; non-invasive imaging by comprehensive Doppler-echocardiography, cardiac MRI, and exercise testing, along with demonstration of congestion. Other key tools include molecular probes to assess fibrosis or changes in metabolism and energetics in multiple tissues, including the skeletal muscles, kidneys, and lungs. Two additional factors should be considered when using animal models to study HFpEF. First, findings in one HFpEF animal model should be tested in a different model (preferably in a different laboratory) to ensure that the findings are not lab- or model-specific. Second, just as human HFpEF is best evaluated with exercise or volume load, these should also be utilized for eliciting the HFpEF phenotype, elucidating molecular mechanisms, and testing novel therapeutics.

Finally, successful translational HFpEF research will require simultaneously utilizing both humans and multiple animal HFpEF models, with stimuli that mimic the comorbidities that are highly associated with human HFpEF.

Serial imaging with automated computational analysis in animals and humans

The heterogeneity of HFpEF complicates the design of clinical trials and makes it difficult to judge the faithfulness of experimental models. Recently, researchers have begun to examine a broader array of computational tools, collectively referred to as “machine learning”, to help address this complexity.^{2, 86}

One challenge regarding the heterogeneity of HFpEF is lack of serial imaging data early in its course when risk factors are apparent yet symptoms have not occurred. At this stage, higher-order disease manifestations may not yet cloud the picture, thereby enabling identification of causative factors. Such early-stage enquiries will require large sample sizes and imaging collected earlier and more often than what typical in clinical practice. Fully automated interpretation of echocardiograms,⁸⁶ focusing on a limited subset of frequently repeated measurements, can facilitate and standardize these measurements and thereby enable longitudinal disease modeling at reduced cost. Similar approaches could be applied to large cohorts of animals with heterogeneous genetic backgrounds.

Cardiac MRI and CT also provide high-resolution 4D data that could provide valuable insights into structural and functional alterations in HFpEF. While cost and scanning time are decreasing, the time for data analysis and interpretation also need to improve. Unsupervised machine learning techniques such as statistical shape modeling are now being used to derive 3-D “atlases” that can capture almost all of the variation between patients in ventricular shape and systolic function with as few as ten parameters or modes each,⁸⁷ and therefore may be useful to capture variation across the phenotypic spectrum of HFpEF.

Biomarker discovery for HFpEF

Circulating biomarkers (such as BNP) facilitate screening, diagnosis, and/or risk stratification. Biomarkers may fill a particularly important need in the diagnosis and management of HFpEF, given the phenotypic and etiologic heterogeneity of this disorder. Selected biomarker profiles could be combined with clinical characteristics to assign HFpEF patients to disease subtypes² which might be used to select appropriate therapies.

Relatively few HF biomarker studies have focused specifically on HFpEF, or on the differences between HFpEF and HFrEF. De Boer and colleagues used data from four longitudinal, community-based cohorts to identify biomarkers of HFpEF and HFrEF.⁸⁸ They focused on established biomarkers, including BNP, C-reactive protein, troponin, and others. While a large number of biomarkers predicted incident HFrEF, only a few were associated with incident HFpEF. The strongest predictive biomarkers for HFpEF were BNP and urinary albumin excretion, although both were still more strongly associated with HFrEF than HFpEF. Plasminogen activator inhibitor-1 (PAI-1) —which has been implicated in aging and insulin resistance in addition to its role in coagulation⁸⁹—was associated with HFpEF but not HFrEF. More recently, Tromp and colleagues performed a proteomic analysis of HF, stratified by EF in a derivation and validation group, and found pathways—including cytokine response, extracellular matrix organization, and inflammation—that were more closely associated with HFpEF than HFrEF.⁹⁰

A limitation of most biomarker studies in HF is the focus on candidate biomarkers from established pathways such as neurohormonal activation. Although these pathways have a prominent role in HF syndromes, they do not necessarily distinguish different types of HF. Also, measurement of multiple biomarkers from the same pathway is unlikely to improve discrimination of risk because they are highly correlated with each other.⁹¹ Evaluation of biomarkers at a single timepoint, as done in most studies, is also problematic. Evolution of biomarkers over time may provide critical insight into the development and progression of HFpEF.

Newer metabolomic and proteomic platforms allow global analysis of hundreds or thousands of analytes from a single blood specimen. Such technologies should facilitate the discovery of “uncorrelated” (i.e., “orthogonal”) biomarkers, because they remove the need to focus on specific pathways or candidate biomarkers. This unbiased approach to biomarker discovery should accelerate the identification of clinically-valid biomarkers in the same way that genome-wide arrays have transformed the identification of genetic markers of disease. However, such studies pose unique challenges with regard to sample size, validation, and statistical analyses, which will require close collaboration across multiple individuals and institutions.

HFpEF tissue biorepositories

The current molecular understanding of HFpEF (e.g., cGMP/PKG signaling), derived from human HFpEF myocardium and not animal models, has come from only a few laboratories. All have reported results from RV septum or LV myocardium, some epicardial, some endocardial, but the analysis has not necessarily overlapped, so independent corroboration of key findings remains limited. Moreover, the types of patients and settings have varied⁵⁸ raising questions about representativeness to a general HFpEF population. Given the growing acceptance of HFpEF as a multi-system disease, with the heart being a component but not always the pivotal one, much broader research into myocardial and non-myocardial abnormalities at a tissue level in carefully phenotyped HFpEF subgroups is needed.

From a cardiac perspective alone, several questions remain unanswered. Is there a subset of patients with excess contractility but insufficient diastolic filling period who would benefit from a negative inotropes? Which patients, other than diabetics, have too much fibrosis, and is fibrosis more common in the presence of metabolic syndrome/obesity even in the absence of diabetes? Is fibrosis associated with elevated inflammatory biomarkers? Is PKG signaling really abnormal, and if so what is driving it? If NO signaling is critically suppressed, why have the responses to various therapies that enhance NO signaling proved disappointing thus far? Is it possible that more prominent benefits may be obtained by targeting the NP signaling pathway? Aside from questions about the heart in HFpEF, there are numerous other questions that would benefit from analysis of extracardiac tissues in HFpEF patients and controls with comorbidities but not the HFpEF syndrome.

The understanding of human HFrEF was accelerated by provision of tissue from cardiac transplant and ventricular assist device treatments. However, neither of these procedures is typically used in HFpEF, and access to tissue is too limited for a disease that impacts >50% of HF patients. There is a critical need to establish a multicenter tissue biobank with uniform tissue processing and storage procedures to formally test key hypotheses in adequate number of samples to provide power for subgroup analyses to assess HFpEF sub-phenotypes. While this will require considerable effort—and biopsy samples may be limited in quantity—prospective, systematic consenting and collection of autopsy tissue samples may alleviate this problem. Developing a large multicenter tissue bank will be a key step towards enabling precision therapy, an important strategy for this heterogeneous syndrome.

Machine learning and predictive multi-scale computational modeling

The variety of etiologies, underlying mechanisms, and phenotypes in human HFpEF necessitates the use of integrative research approaches to elucidate interactions between systems and subsystems such as the interactions between: the heart, vasculature and other organs; microvascular dysfunction, structural remodeling, and inflammation; mechanics and energetics with metabolism and regulatory mechanisms; myocytes, fibroblasts, and extracellular matrix. Computational modeling can analyze interactions and elucidate integrative mechanisms in HF by extracting patterns, relationships and biomarkers from large data sets (bioinformatics), by analyzing interacting cellular and physiological systems and subsystems (systems biology), and by relating structural alterations to functional defects (multiscale modeling) (Supplemental Figure 1).⁹² While all of these approaches, including multi-scale, patient-specific modeling, have been applied successfully in elucidating integrative arrhythmia and HFrEF mechanisms, there has been almost no comparable work on HFpEF.

Existing computational models are few and fall into the following categories: statistical and data models including machine learning from clinical data; lumped parameter hemodynamic models of circulatory dynamics; biophysical models of specific myocyte phenotypes in HFpEF; shape and kinematic models of imaging data; and multi-scale biomechanics models of heart and circulation. Of particular relevance are new computational modeling approaches to cardiac mechanobiology. These include systems models of myocyte and fibroblast stretch signaling, agent-based models of fibroblast-mediated fibrosis, and continuum models of ventricular hypertrophy and remodeling.⁹³ However, all such existing mechanistic models have actually been models of diastolic dysfunction or pressure overload hypertrophy. Most multi-scale systems models of HF have been based on detailed biophysics of excitation-contraction coupling and biomechanics, and the reason that these models have focused on HFrEF perhaps is because the prevailing paradigm for HFrEF has been that cardiac dysfunction is primary and leads to systemic effects. In contrast, in most cases of HFpEF, the systemic abnormalities are primary and lead to secondary effects on cardiac structure and function.

Machine learning can also be used to identify potentially distinct disease subclasses which may share common pathophysiology.⁹⁴ This approach, termed unsupervised learning, involves finding recurring patterns of data. For HFpEF, such data may include a combination of imaging, circulating biomarkers, electrocardiographic information, and common laboratory values.² Such data-driven groups will ideally reflect greater homogeneity of aberrant biological pathways. Unsupervised learning methods can naturally handle multiple correlated covariates and may thus identify categories that do not align with simple binary classification schemes based on comorbidities such as HFpEF with/without obesity or with/without diabetes mellitus. Mechanistic computational modeling can then start to map the mechanisms identified in animal models of specific HFpEF phenotypes back to these newly identified clinical subclasses.

Imaging and other phenotyping methods for HFpEF

The most widely utilized cardiac imaging techniques have been transthoracic echocardiography and cardiac magnetic resonance imaging (CMR). Transthoracic echocardiography provides an assessment of LV structure and function, including LV hypertrophy, remodeling, EF, and diastolic function, LA structure and function; RV structure and function; and estimations of ventricular filling pressures and PA pressures,² and has the advantage of being relatively low cost and amenable to serial measurements. Echocardiography can also be used to sub-phenotype the HFpEF syndrome and determine the extent of myocardial involvement based on severity of abnormalities in longitudinal strain or tissue velocities and other variables.^{95, 96} In addition, myocardial strain can be useful in identifying specific etiologies of HFpEF such as cardiac amyloidosis and hypertrophic cardiomyopathy.⁹⁵

Echocardiography can be utilized for exercise studies (typically using supine or semi-supine bicycle protocols), in which cardiac output reserve and vasodilatory reserve can be assessed.⁹⁷ When used in combination with expired gas analysis (peak VO₂), it can also provide estimations of the arterio-venous O₂ difference based on the Fick equation, which has proven useful to assess the cardiac vs. peripheral effects of interventions such as exercise training,⁹⁸ drugs,^{97, 99} and devices. When combined with arterial tonometry, echocardiography can provide a detailed characterization of LV afterload via pressure-flow relations. This approach is superior to the pressure-volume plane to characterize pulsatile load,¹⁰⁰ particularly late systolic load from wave reflections, which lead to abnormal ventricular-arterial interactions and may contribute to diastolic dysfunction, maladaptive LV remodeling and fibrosis, atrial dysfunction and exercise intolerance.¹⁰¹

CMR can provide a wide range of assessments, including myocardial mechanics (including ventricular and atrial deformation) and tissue characterization (particularly the assessment of diffuse myocardial fibrosis via T1 mapping post-gadolinium contrast, measured as extracellular volume [ECV]). CMR techniques have shown that, consistent with autopsy studies,¹⁹ ECV is increased in HFpEF, but ECV exceeds the upper limit observed in control populations in only 23–30% of HFpEF patients.^{102, 103} Patients who have HFpEF and increased ECV have been shown to have predominantly increased passive myocardial stiffness¹⁰⁴ and a worse prognosis,^{79, 105} whereas patients with normal or near normal ECV demonstrated predominant abnormalities in LV relaxation and LV afterload during exercise.¹⁰⁴ CMR can also interrogate myocardial vasodilatory reserve,¹⁰⁶ which has been shown to be abnormal in HFpEF and may underlie myocardial ischemia during exercise. Due to much lower variability and higher evaluability rates, CMR can assess LV hypertrophy/mass with 90% smaller sample sizes than echocardiography.¹⁰⁷ and can also assess aortic distensibility/stiffness, a major determinant of LV afterload that is a potential contributor to exercise intolerance in HFpEF. Novel CMR techniques can interrogate myocardial mitochondrial function,¹⁰⁸ myocardial fiber orientation, myocardial inflammation, and intramyocardial fat accumulation, but these techniques have not yet been applied broadly to HFpEF. In the future, multi-organ magnetic resonance imaging (i.e., kidneys, liver, adipose tissue, and skeletal muscle in addition to the heart) could also be used to examine the systemic nature of HFpEF and to systematically phenotype individual patients.

Nuclear imaging techniques evaluate myocardial ischemia and coronary flow reserve, which is often a clinical concern in HFpEF. Nuclear imaging techniques can also assess myocardial metabolism and sympathetic cardiac innervation. Of particular interest is the clinical value of bone scintigraphy for the diagnosis of transthyretin (TTR) cardiac amyloidosis,¹⁰⁹ given that TTR amyloidosis underlies a small but significant proportion of HFpEF cases, and is now amenable to specific pharmacologic therapy. An important area of HFpEF research is whether advances in nuclear imaging of cardiac or coronary vascular inflammation used in diagnosis of sarcoidosis or coronary plaque will permit study of myocardial and microvascular inflammation in HFpEF.

Imaging techniques can also be used for assessments of skeletal muscle. As mentioned above, patients with HFpEF exhibit sarcopenia and fat infiltration of skeletal muscle which can be imaged well with magnetic resonance. Magnetic resonance can also be used to assess skeletal muscle mitochondrial function utilizing ³¹P magnetic resonance spectroscopy.

Imaging in HFpEF can enable earlier diagnosis and improved phenotyping to better guage prognosis and assign patients to optimal interventions. Therefore, the use of imaging to elucidate specific etiologies and pathophysiologies underlying HFpEF will be essential. Just as in TTR cardiac amyloidosis where there are specific patterns on echocardiography, CMR, and nuclear imaging, it is hoped that future studies will discover specific causes of HFpEF that can be diagnosed using imaging techniques and then treated with specific medications in order to improve outcomes.

Therapeutic targets in HFpEF

Treating comorbid conditions which contribute to the development and severity of HFpEF and propagate organ level dysfunction in HFpEF may be fruitful. Table 1 lists potential therapeutic targets, rationale for these targets, and scientific gaps and unmet needs in each therapeutic area.¹ Aside from individual targets, a phenotype-specific approach to the heterogeneous HFpEF syndrome may be valuable in order to match the right treatment to the right patient with HFpEF. Supplemental Table 1 lists strategies for the classification of HFpEF, incluiding in sub-phenotypes that may benefit from particular therapies. Ultimately, targeted therapeutics based on the underlying biological basis of disease subtypes within HFpEF will provide the best opportunity for a precision medicine approach. Successful development of non-invasive diagnosis and disease-modifying treatment of TTR cardiac amyloidosis (Figure 2) provides hope that targeted treatment of HFpEF is possible.

HFpEF prevention

Compared to HFrEF, HFpEF has greater incidence, comparable lifetime costs of care,¹¹⁰ and similarly poor long-term outcomes. In large cohort studies, the strongest modifiable risk factors for developing HFpEF are obesity, systemic hypertension, and diabetes mellitus, suggesting potential prevention strategies.^{111, 112} Several issues complicate assessment of preventive strategies for HFpEF. First, it is uncertain how early in life HFpEF truly begins, and its putative fundamental trigger(s) have yet to be fully elucidated. Most studies that assess incident HFpEF use hospitalization for HF as the primary determinant. While this approach reduces risk of misclassification, it may overlook earlier stages of HFpEF development. In addition, many registries and clinical trials do not adjudicate incident HF cases by LVEF. Nevertheless, prevention of HFpEF can be inferred if overall incident HF is reduced in a cohort with low prevalence of LV systolic dysfunction and high risk for developing HFpEF, such as older women, in whom >80% of incident cases tend to be HFpEF. Despite these limitations, several randomized trials have demonstrated reductions in the incidence of HF in general, and HFpEF in particular. Table 2 lists strategies to prevent HFpEF, their rationale, the evidence that exists for preventive interventions, and the gaps and unmet needs in each of these areas.

Innovative approaches to HFpEF clinical trials

The heterogeneity of the HFpEF syndrome and the use of conventional randomized controlled trial (RCT) designs are possible reasons underlying the disappointing results from RCTs to date in HFpEF. There are several factors—including the widespread adoption of electronic health records, decreasing costs of obtaining high-dimensional data, and the availability of a wide variety of potential therapeutics—that have evolved to enable more innovative clinical trial designs in HFpEF. Several innovative RCT designs could be implemented in HFpEF, including enrichment trials, adaptive trials, umbrella trials, basket trials, and machine learning-based trials, as outlined in Table 3.¹¹³

Examples of ongoing enrichment trials include the REDUCE LAP-HF II pivotal trial of an InterAtrial Shunt Device, which specifically enrolls patients with a left heart (LA overload)-predominant form of HFpEF, whereas the SERENADE (macitentan, endothelin receptor antagonist) trial is enrolling patients with HFpEF who have characteristics consistent with combined post- and pre-capillary PH. The inclusion/exclusion criteria for these types of trials are designed to enrich the trial with the patients most likely to benefit from the tested therapy. Given the large number of HFpEF therapeutics currently being tested, an umbrella design is attractive because it allows large HFpEF centers to enroll patients into a variety of different trials based on their sub-phenotype and helps matching the right patient to the right trial (Supplemental Figure 2).

Although they can be complex to design, adaptive trials allow for increased flexibility and may allow for increased return on large investments made by funding bodies. Registry-based trials are also being conducted in HFpEF (e.g., spironolactone in the SPIRRIT trial); these pragmatic trials seek to test existing therapies in large numbers of HFpEF patients. While these trials are appealing due to their low costs and wide clinical applicability, for most therapeutics being tested, this design approach is at odds with the prevailing theory that the broad definition of HFpEF makes the syndrome heterogeneous. Thus, for most novel therapeutics, pragmatic trials may not be appropriate and may increase risk by exposing undifferentiated patients to therapies that may potentially worsen their clinical condition.