Human Cardiac Metabolism

Abstract

The heart is the most metabolically active organ in the human body, and cardiac metabolism has been studied for decades. However, the bulk of studies have focused on animal models. The objective of this review is to summarize specifically what is known about cardiac metabolism in humans. Techniques available to study human cardiac metabolism are first discussed, followed by a review of human cardiac metabolism in health and in heart failure. Mechanistic insights, where available, are reviewed. And the evidence for contribution of metabolic insufficiency to heart failure, as well as past and current attempts at metabolism-based therapies, are also discussed.

Introduction and conceptual framework

The heart’s ceaseless demand for robust generation of ATP relies almost exclusively on mitochondrial oxidative phosphorylation because it is ~20 times more efficient at generating ATP from glucose than is glycolysis (~38ATP vs 2ATP per glucose molecule) and is the sole method to generate ATP from fuels such as fatty acids and ketones (see box 1). A full quarter of human cardiac volume is thus comprised of mitochondria, the highest density of all body organs. What metabolic fuels sustain cardiac ATP generation? How do they differ under different physiological conditions? Does aberrant metabolism contribute to heart failure? These questions have been addressed extensively in model organisms, including mice, rats, dog, pigs, and others, and the results discussed in many excellent reviews^4–6. For the non-expert reader, however, it is often unclear which findings in these organisms extrapolate to humans, and which do not. The objective of the current review is therefore to summarize what we know about cardiac metabolism in humans, explicitly excluding data from model organisms. Effort will be made to be quantitative, and to report on original sources, rather than review articles.

Box: ATP and the heart.

The heart consumes ~5% of total body VO₂ (~0.1 ml/min/g cardiac tissue, or ~4 μmoles/min/g) at rest, and can increase consumption up to 10-fold with heavy exercise, despite weighing only ~0.4% of body mass. ~95% of oxygen consumption is used for ATP generation (~25 μmoles/min/g cardiac tissue). There is a strict stoichiometry between fuel use, TCA cycling, oxygen consumption, ATP generation, and ATP consumption, i.e. the heart is highly coupled (Figure). There is therefore normally a linear relationship between cardiac work, oxygen consumption, and fuel consumption. And, importantly, cardiac metabolic activity is demand-side driven, i.e. the rate of mitochondrial oxygen and fuel consumption is driven by ADP generated by cardiac ATPases. Cardiac ATP content is relatively low (~5–8mM¹). Given the high rate of ATP generation, there is thus only a few seconds of reserve high-energy phosphates at any time, and the total ATP pool undergoes turnover in ~10–20 seconds. The levels of ATP and free ADP (<20μM, i.e. >100x less than ATP; note that up to 90% of ADP is bound to actin filaments) are defended by the heart to maintain constant the deltaG of ATP hydrolysis at ~60kJ/mole, the driving force for cardiac work. On average, each cardiac stroke generates ~1J of work, costing ~100μmoles ATP. Contraction reactions (e.g. ATP hydrolysis by myosin) account for ~60% of cardiac ATP consumption. However, Ca⁺⁺ uptake into the endoplasmic reticulum is more sensitive to low ATP/ADP ratios, because the deltaG of SERCA2 at ~52kJ/mole is closest to that of ATP hydrolysis^2,3, thus perhaps explaining why diastolic dysfunction often precedes systolic dysfunction. Studies in rats suggest that high ATP/ADP ratios are also required for actin/myosin crossbridge dissociation, another key step in diastolic relaxation⁷.

Figure from Stanley et al. 2005

In sum, ATP generation in the heart is:

• robust,

• almost entirely mitochondrial,

• highly coupled to cardiac function,

• driven by generation of ADP by ATP hydrolases,

• and working in the context of only seconds of reserves.

Studying human cardiac metabolism is not easy. A strong and opaque rib cage shields the heart from access and view, and subjects are reticent to part with tissue samples for research. We will therefore begin by reviewing the various approaches, invasive and non-invasive, that have commonly been employed to study human cardiac metabolism. We will then discuss what these approaches have taught us of fuel use by the heart, in both health and disease; what potential mechanisms of cardiac dysfunction in heart failure are supported by human data; the few metabolic interventions that have to date been evaluated in human subjects; and finally, a discussion of future directions. We will assume the reader has knowledge of basic biochemical pathways of glycolysis, TCA cycle, etc., and refer the reader to excellent reviews if needed⁸.

Techniques

Arterial-venous sampling

Werner Forssman received the 1956 Nobel Prize in Medicine for developing cardiac catheterization, famously disproving nay-sayers by performing the first one on himself (with a urinary catheter) ⁹. The approach for the first time enabled non-perturbative, in vivo studies of human cardiac function, including metabolism. Researchers, most prominently Richard Bing^10–13, quickly realized that simultaneous blood sampling from catheters placed in the coronary sinus, which drains most of the heart and little else, and in any artery could therefore be used to measure cardiac metabolism, including fuel use and metabolite secretion. The approach (A/CS sampling) was used frequently in the 2^nd half of the 20^th century, using various biochemical methods to measure metabolites. Recently, the approach has had a resurgence, often leveraging modern mass spectrometry that allows for precise quantification of hundreds of metabolites simultaneously^14–16. Advantages of A/CS sampling include the nonperturbative nature of the approach, as well as the ability to measure metabolite fluxes with precision, to simultaneously modulate cardiac work, to acquire multiple sets of samples for kinetic studies, and to infuse isotopically labeled metabolites to quantify cardiac metabolite secretion. Limitations include the lack of tissue sampling for mechanistic investigations, the invasive nature of cardiac catheterization, the inability to query different parts of the heart separately, and the fact that the coronary sinus also drains epicardial fat (but not pericardial fat), which in humans can be consequential, and can change with age, diabetes, and obesity^17,18. A/CS studies are listed in Table S1.

Positron emission tomography (PET)

2-deoxy-2-[¹⁸F]fluoro-D-glucose (¹⁸FDG-PET)

The use of ¹⁸FDG for PET imaging was first introduced in 1978 as a method of measuring myocardial glucose uptake (MGU)¹⁹. ¹⁸FDG is a glucose analog, substituting ¹⁸F at the C-2 position of glucose, normally occupied by a hydroxyl group (Figure 1). ¹⁸FDG is particularly well suited to measure myocardial glucose uptake (MGU). The substitution of ¹⁸F does not prevent FDG uptake into cells through glucose transporters and its phosphorylation by hexokinase to FDG-6-phosphate, in proportion to endogenous glucose uptake. However, FDG-6-P cannot undergo the subsequent step of glycolysis, isomerization to fructose 6-phosphate. Trapped FDG-6-P thus accumulates intracellularly in proportion to uptake and can be quantified by PET. With a half-life of 110 minutes, ¹⁸F decays to ¹⁸O by loss of a positron (Figure 1). The emanated positron collides with an electron usually within a millimeter, yielding 2 photons sent in precisely antiparallel directions. Coincidence detection of these photons obviates the need for collimation and greatly improves sensitivity and resolution over single-photon emission CT (SPECT). The relatively long half-life of ¹⁸F permits time for transport of FDG, removing the need for a cyclotron at every site of use. Ultimately, ¹⁸FDG decay generates glucose-6-phosphate (with a heavy-labeled oxygen at position C-2) that can then undergo standard glycolytic breakdown.

Figure 1:

Methodologies for evaluation of human cardiac metabolism. A: arterio-venous sampling. B: FDG-PET scanning. Inset shows typical short axis FDG-PET image. Below: spontaneous decay of FDG to glucose. C: 4-compartment model of ¹¹C-palmitate uptake and oxidation. D: ³¹P-MRS. Inset shows typical resonance pattern at 3 and 7 tesla (T), revealing prominent peaks for high-energy phosphates PCr, ATP, and 2,3 DPG. Below: PCr shuttle; see text for details. Image created with Biorender.

Cardiac FDG-PET has become a staple of clinical care for various purposes, including detection of viable myocardium and the detection of inflammatory processes, sarcoidosis, and tumors. Most cardiac FDG-PET studies are limited to visualizing and comparing relative uptake of FDG between different regions of the heart, e.g. to identify inflammatory sites, tumors, ischemic areas, or infarcted areas. Such studies do not quantify MGU. To do so requires normalization data, i.e. an arterial input function, and dynamic FDG-PET, i.e. multiple images over time. The simplest way to derive arterial input functions is by serial arterial sampling, but less invasive methods have been developed, using the FDG-PET images themselves and time-activity curves from a region of interest (ROI) composed of blood, usually the aorta or the left ventricle (LV)²⁰. Myocardial extraction of glucose measured by FDG-PET and by A/CS studies correlate well²¹. Dynamic FDG-PET studies can thus provide quantitative MGU noninvasively and will be the focus of this review (Table S2). Such studies are used primarily for research purposes. Of note, the affinities of FDG for glucose transporters and hexokinase is not identical with those of glucose, requiring a correction of 0.67, known as a lumped constant, that may sometimes introduce error²².

11C palmitate

¹¹C-palmitate PET is increasingly used to measure FA uptake, oxidation, and esterification. Like ¹⁸F, ¹¹C emits positrons that annihilate with nearby electrons, allowing coincidence detection of generated antiparallel photons. The half-life of ¹¹C is 20 mins, shorter than ¹⁸F but still manageable for clinical studies. Unlike FDG, ¹¹C-palmitate undergoes normal metabolism once in the cell, akin to unadulterated palmitate. Evaluation of its kinetics are fit into a 4-compartment model, including vascular, interstitial, and intracellular spaces (Figure 1C)²³. The initial rapid uptake of ¹¹C-palmitate largely reflects myocardial blood flow (MBF). The clearance of myocardial ¹¹C activity then follows a bi-exponential pattern: the early rapid phase reflects FAO, which liberates ¹¹CO2 that is then lost to the blood, while the subsequent slower phase reflects decay of esterified ¹¹C-palmitate that has been incorporated into the TG pool (Figure). Mathematical modeling using ¹¹C activity clearance rates can thus infer myocardial FAO and esterification rates. Other PET-based probes have also gained traction, including FA analogs such as 14(R,S)-[ ¹⁸F]fluoro-6-thia-heptadecanoicacid (¹⁸F-FTHA) and second-generation ¹⁸F-FT-palmitate (¹⁸F-FTP), which have the interesting property of being trapped in mitochondria after uptake, unable to redistribute into intracellular TG pools, and thus allow for targeted measurement of FAO. Cardiac FA-PET studies are listed in Table S3.

Advantages of PET studies include the noninvasive nature, high sensitivity, and the ability to discern geographic differences. Limitations include the restriction to only one metabolite or fuel, the use of ionizing radiation, the inability to assess endogenous metabolites, and the lack of tissue sampling for mechanistic investigations. The reader is referred to recent excellent reviews for more details on these and other less frequently used approaches for noninvasive metabolic imaging of the human heart.^24,25

Magnetic resonance spectroscopy

Phosphorus Magnetic resonance spectroscopy (³¹P-MRS) permits the non-invasive detection of high-energy phosphorus-containing metabolites and has provided the strongest evidence for the notion that the failing human heart is energetically compromised (see section below). Phosphorus (³¹P) has an odd atomic number and thus has a magnetic moment that can be aligned with a magnet and stimulated with radiofrequency. Its relaxation emission can then be measured by MRS. The sensitivity of MRS is limited, however, allowing detection of only the most abundant phosphorus-containing metabolites, with a limit of detection of ~>0.6mM. In the heart, these are in ATP, phosphocreatine (PCr), and inorganic phosphates. PCr is generally in ~2x excess over ATP. PCr is generated by the equilibrium reaction ATP+creatine⇌PCr+ADP+H, catalyzed be creatine kinase (CK). The reaction is terminal, i.e. PCr has no other fate than to donate its high-energy phosphate back to ADP. PCr is thought to serve at least two main functions in the heart: (1) to facilitate transport of high-energy phosphates from ATP-producing to ATP-consuming sites, because PCr diffuses much more rapidly than ATP (aka the creatine phosphate shuttle, Figure) ^26,27; and (2) to buffer ADP and ATP, maintaining appropriate ATP/ADP ratios near both these sites. The PCr/ATP ratio has widely been used as a surrogate of “energy charge” in the myocardium (although it may in fact more closely reflect creatine concentrations, see below). The measured PCr/ATP ratio in healthy individuals is usually ~2, but it can range widely for various technical and analytic reasons including scan techniques, corrections, and normalizations. Comparing data between centers and studies is therefore often difficult. Nevertheless, noninvasive measures of high-energy phosphates by ³¹P-MRS provide an inestimable advantage over direct tissue measurements, because ATP hydrolysis is so rapid that even seconds of warm ischemia during tissue harvest markedly affects concentrations.

³¹P-MRS can also be used to measure the rate of phosphoryl transfer from PCr to ADP (i.e. the CK flux), using magnetization (aka saturation) transfer techniques, in which the ATP γ-P is selectively “saturated” with a long radiofrequency pulse, rendering it unable to relax and therefore undetectable by MRS. The kinetics of transfer of the “silent” γ-P to creatine can then be measured as reduced PCr signal, representing flux through the CK reaction. CK flux in normal volunteers at rest is ~3 μmoles/g/sec.²⁸ This flux is an order of magnitude faster than ATP production in mitochondria (see box), suggesting that CK flux capacity is not rate-limiting. However, while ATP production is likely relatively constant, ATP consumption is not: ~50% of ATP hydrolysis occurs during the systolic power stroke, which occurs over ~15% of the cardiac cycle, indicating that ATP consumption is ~6-fold higher during peak contraction than during the rest of the cycle. Moreover, the heart can also increase VO₂ substantially with, e.g., exercise. It thus seems likely that CK flux rates can become limiting, particularly during contraction. In this sense, CK flux may be a better representation of cardiac energetics than PCr/ATP. ³¹P-MRS studies are listed in Table S4.

MRS can also be used with other atoms with odd atomic numbers, including ¹H and ¹³C. Given the abundance of H atoms, ¹H-MRS is especially tractable, and has been used to contrast aqueous from nonpolar cardiac regions, e.g. to measure the triglyceride (TG) content of the myocardium, pericardium, and epicardium. ¹H-MRS studies are listed in Table S5. Detection of ¹³C is more difficult, because of its lower magnetic energy and its lower natural abundance (99% of carbons are ¹²C). Hyperpolarization techniques, e.g. with ¹³C-pyruvate, can transiently (seconds) increase sensitivity by >10,000, and are under development for cardiac imaging^29,30.

Tissue acquisition

Studies of metabolism using human heart tissue date back to the 1960s and provide the opportunity for much deeper phenotyping of metabolic pathways than do non-invasive approaches, including evaluations of mitochondria structure and respiration, enzymatic activities, metabolic intermediate concentrations, etc. However, high-quality human cardiac samples are difficult to acquire. Excisional biopsies from hearts in situ can be obtained by percutaneous approach or during cardiac surgery. The former, known as endomyocardial biopsy (EMB), usually takes a right jugular vein approach and removes several small (a few mgs) endomyocardial tissue samples from the right ventricular side of the septal wall. These samples are taken from the heart while it is still active and beating, making it potentially highly representative of physiologic conditions. On the other hand, the processing of these samples can prove a challenge for downstream metabolic analyses because ischemia develops within seconds of extraction, potentially clouding results. Furthermore, cardiac metabolism is spatially organized, and biopsy from a single site can lead to overgeneralization of metabolic findings. In the case of cardiac surgery, biopsies are typically taken from the epicardium of the left ventricle, or from the expendable left atrium appendage, and are generally taken after the heart has undergone cardioplegic arrest. Samples can also be acquired at the time of LV-assist device (LVAD) implantation, which requires removal of a left ventricle apical plug. LVAD plugs can conveniently be paired with repeat sampling from the same patient at cardiac transplantation (see below), although the use of LVAD as bridge therapy is currently in decline. Like EMBs, LVAD plugs generally suffer from periods of warm ischemia.

Explanted hearts provide an alternative source of primary cardiac tissue. These hearts are generally collected from organ donors following brain death or from patients undergoing cardiac transplantation and are also typically arrested using cardioplegia prior to being explanted. While the use of cardioplegia introduces an additional variable into the interpretation of results, this method minimizes tissue ischemic injury, potentially providing a more accurate accounting of in vivo metabolism. The explanted heart also allows for collection of samples across different anatomical regions, allowing for broader spatial examination. However, the availability of explanted hearts is limited, and due to the patient population, are highly enriched for end-stage, poorly compensated heart failure, which may not be representative of milder pathology. In short, there is no perfect method for acquiring cardiac tissue for molecular studies, and care must be taken to acknowledge the caveats of the approach chosen. We recommend that all studies involving cardiac tissue acquisition include in their methods, at a minimum, (1) the anatomical site of tissue acquisition, (2) whether the heart is arrested, (3) which cardioplegia is used (if any), (4) if the tissue is perfused (and with what), (5) and how long the acquired tissue experiences warm ischemia.

What metabolic fuels sustain cardiac ATP generation?

Glucose

A/CS and FDG-PET studies have both conclusively demonstrated uptake of glucose in human hearts. In fasting young healthy individuals, uptake is typically ~0.1–0.2 μm/min/g (Sup tables) (but see discussion below). Little is known about rates of MGU in fed individuals, but MGU in fasting individuals undergoing a hyper-insulinemic euglycemic clamp can range as high as 0.5–0.9 μm/min/g (Sup tables). Of note, A/CS studies may partly underestimate MGU because heparin, usually used in these studies, can suppress MGU³¹ (and see below). FDG-PET studies, in contrast, generally do not use heparin, unless specifically aiming to suppress MGU for clinical indications. Importantly, both FDG-PET and standard A/CS studies measure only glucose uptake and cannot differentiate between the different fates of glucose: anerobic glycolysis leading to secretion of 2 molecules of lactate; entry into the pentose phosphate pathway or other side branch of glycolysis; entry into the glycogen pool; or complete oxidation in mitochondria. A/CS studies performed with simultaneous infusion of labeled glucose and labeled lactate have shown that as much as 50% of glucose taken up by the heart can be secreted as lactate^32–34. Tracer studies in young healthy fasting subjects indicate that only 10–30% of glucose is immediately oxidized, even in the presence of hyperinsulinemia; another ~13% is secreted as lactate, and the remaining majority of glucose extracted likely cycles through the glycogen pool, i.e. the tracer is unaccounted for^32,33 (Figure 2). Importantly, the heart remains a net consumer of lactate, i.e. taking up and oxidizing more lactate than it secretes (see lactate discussion below). When MGU is low, flux of glucose prioritizes replenishing the glycogen pool, and direct oxidation of glucose approaches zero. Similar findings are reported with ¹¹C Glucose PET in patients with obesity or with DCM^35,36. Substrate cycling through the glycogen pool is often ignored, leading to underestimation of glucose oxidation. Flux through the pentose phosphate pathway has not been measured in humans, but it is <5% of glycolytic flux in experimental animals^37,38. In summary, in healthy individuals, there is measurable uptake of glucose by the heart, variably partitioned between immediate oxidation, generation of lactate, and cycling through the glycogen pool (Figure 2).

Figure 2:

Fuel uptake, cycling through intracellular pools, and simultaneous lactate uptake and secretion. A: rates of uptake and oxidation of carbohydrates (glucose and lactate) and of glucose cycling through the glycogen pool, in fasting young healthy male volunteers. Adapted from data in Wisneski et al. 1985. Note that one glucose molecule yields 2 molecules of pyruvate. B: rates of uptake and oxidation of fatty acids from albumin-carried FFA and lipoprotein-carried TG pools, of FA cycling through the intracellular TG pool, and of ketones, in fasting young healthy male volunteers. Adapted from data in Wisneski et al. 1987. All values are micromoles/min/g. Image created with Biorender.

Suppression of glucose uptake by fatty acids and ketones

Glucose uptake by the human heart varies greatly in different individuals and under different conditions, rendering generalized conclusions challenging. With FDG-PET studies, for example, it quickly became clear that overnight fasting suppresses glucose uptake. Today, high-fat low-carb (often narrowly called “ketogenic”) diets are in fact often used clinically to suppress myocardial FDG uptake when the goal is to increase contrast with non-myocardial FDG uptake by inflammatory or cancerous cells. Plasma ketone levels predict the effectiveness of FDG uptake suppression (e.g. nearly zero FDG uptake with plasma ketones >1mM³⁹), and ketone infusions are sufficient to suppress MGU by half⁴⁰. Importantly, however, ketones alone do not suppress FDG uptake as much as do high-fat low-carb diets, implicating other diet-mediated effects⁴¹.

Free fatty acids levels likely explain much of the variation in MGU. Randle first proposed in the 1960s that FA consumption by the heart suppresses glucose uptake⁴², a relationship often called the Randle glucose-fatty acid cycle. Numerous A/CS, FDG-PET, and 11C-glucose PET studies subsequently supported the inverse relationship between glucose and FA uptake, even under hyperglycemic conditions^{32,36,43–48}. Interventional studies further demonstrated that administration of exogenous FAs suppresses MGU^49,50, and conversely that suppression of fasting adipose lipolysis during fasting increases MGU^50–52. At FA concentrations >800μM, glucose uptake is minimal, and glucose oxidation is negligible (Figure)^{32,33,50,53,54}. This probably explains why earlier studies in the 20^th century, often with young volunteers with reported or presumably lower FFA levels, reported fasting MGUs in the range of 0–0.4 μmoles/min/g, representing on average ~20% of myocardial VO₂ (MVO₂) (eg^{11,33,44,55–58}), while more recent A/CS and FDG-PET studies involving older subjects with average BMIs of ~30 and elevated FFAs often well above 1mM, more typical of the modern era, report much lower fasting MGUs, nearing zero (eg^{15,16,54,59–61}). Thus, in contrast to young, metabolically healthy individuals in which glucose oxidation contributes substantially to MVO₂, most current patients have a shift in their substrate utilization towards fatty acids, possibly compromising the metabolic flexibility of the heart.

MGU in response to insulin

The effect of insulin on the heart is difficult to disentangle from systemic effects. Studies in model organisms, reviewed elsewhere^62,63, have supported an important role for insulin signaling directly in cardiac tissue, but there are only limited data in humans. Because fasting MGU is low, the great majority of dynamic FDG-PET studies have used glucose loading or hyper-insulinemic clamps to increase MGU and better visualize and quantify glucose uptake. Both A/CS and FDG-PET studies have shown that cardiac MGU rises substantially in the presence of high insulin concentrations, to as high as 0.7 μmoles/min/g (e.g. ^{53,55,57,64–67}). In nearly all these studies, circulating FFAs are drastically lowered by insulin, explaining the rise in MGU. A few studies, however, have shown that elevating FFAs in the presence of hyper-insulinemia does not suppress MGU, indicating that insulin can also override the suppressive effects by the Randle glucose-fatty acid cycle⁵². Regression studies in 25 healthy volunteers suggest that insulin, FFA levels, and glucocorticoids independently contribute to control of carbohydrate uptake⁶⁸. Furthermore, an acute bolus of insulin increases MGU with faster kinetics than it suppresses plasma FA or cardiac FA uptake⁶⁹, suggesting a direct effect of insulin on the heart. On the other hand, plasma FA levels are the best predictor of MGU in the presence of insulin clamp. Moreover, suppression of adipose lipolysis (i.e., while maintaining insulin low) induces MGU as much as hyper-insulinemic clamp⁵¹. No studies have tested the effect of FFA in the context of low dose, i.e. physiological insulin levels. In sum, it is not clear whether indirect effects of insulin, via suppression of lipolysis, versus direct effects on myocardium, dominate under physiological conditions.

MGU and diabetes

In both subjects with type II diabetes and pre-diabetes exposed to hyper-insulinemia, MGU is reduced and is proportional to whole-body glucose disposal^{12,64,70–78}. Treatment with sulfonylureas improves both systemic glucose disposal and MGU⁷⁹. Cardiac MGU is thus clearly responsive to insulin and reduced in parallel to decreasing whole-body insulin sensitivity. How diabetes suppresses MGU is less clear. The relative contributions of cardiac-intrinsic factors, versus indirect effects of, e.g., elevated FFA from persistent lipolysis in insulin resistant adipose tissue, is not well-established. Generally, insulin promotes glucose uptake in part via PI3K/Akt signaling, promoting GLUT4 translocation to the plasma membrane, although this has never been shown directly in human adult cardiomyocytes or tissue. In a small cohort of subjects with diabetes, plasma membrane GLUT4 was markedly suppressed⁷⁸. Interestingly, however, cardiac PI3K and Akt activity were elevated in these subjects with diabetes in proportion to circulating insulin, implying that signaling through cardiac insulin receptor is maintained, despite systemic insulin resistance, and that the block in GLUT4 translocation occurs downstream of insulin signaling⁷⁸. In general, diminished MGU in subjects with diabetes is seen only when suppression of plasma FFA levels by insulin is impaired^80,81, although this is not always the case⁶⁴. It thus appears that the predominant mechanism of MGU suppression in subjects with diabetes is indirect, via inhibition of adipose tissue lipolysis and persistent elevation of circulating FFAs, but that additional cardiac-intrinsic mechanisms may also exist.

Interestingly, studies in young type I diabetics indicate no difference in MGU compared to control subjects, despite peripheral insulin resistance^48,82, and another study in young lean diabetics (type not defined) also shows no differences in MGU⁸⁰, thus suggesting that impaired insulin signaling in skeletal muscle does not necessarily imply the same in cardiac muscle. Moreover, treatment with metformin⁸³ or SGLT2 inhibitors⁸⁴ appears to reduce MGU while improving whole-body glucose disposal, further indicating that these effects can be separated, and that the relationship between insulin resistance and MGU is complex and multifactorial.

Other modulators of glucose uptake

Unlike A/CS studies, imaging with FDG-PET allows for geographic interrogation of MGU. Dynamic FDG-PET studies, whether fasting or with a hyper-insulinemic clamp, have generally revealed that the RV and LV MGU are comparable, and correlate between individuals^85–87. There also appears to be little spatial variability in MGU across the LV^80,88. MGU doubles with moderate exercise or pacing in healthy subjects, and the portion that undergoes immediate oxidation doubles from ~25% to ~50%^89,90. In subjects with CAD, however, glucose taken up during high-intensity pacing is more likely to undergo anaerobic glycolysis, leading to lactate secretion, likely reflecting tissue hypoxia^91,92. MGU increases slightly with increased cardiac work in response to dobutamine, but not as much as does FAO, reflecting concomitant increases in circulating FFA levels^93,94. MGU rises slightly in older individuals, and becomes resistant to further increase with dobutamine, a block that can be reversed with exercise training^93,94. Some studies report little difference in MGU between functional and hibernating myocardium^95,96, while others do report a difference⁸⁸, and some report that MGU is higher in hibernating myocardium that recovers function after revascularization^97,98. Some studies have reported variability in MGU in areas prior to myocardial reperfusion (usually by CABG), but MGU has largely not been predictive of which reperfused areas go on to regain function⁹⁶. MGU is lower in scar tissue.

Fatty acids

Fatty acids are the primary fuel source for the heart, representing ~60–95% of energy consumption. FAs can be acquired from the albumin-bound free fatty acid (FFA) pool, or from fatty acids esterified as triglycerides (TGs), found in circulating lipoprotein particles. Net FFA uptake in fasted individuals ranges 0.05 –0.22 μmoles/min/g, measured by various methods^40,99–106 (Sup tables). Individual FFAs generally do not appear to be taken up differently from each other^16,44, although some early studies report a higher relative uptake of oleic acid, and lower of linoleic acid^67,107,108. FFA uptake is largely, but not entirely, a reflection of availability. As with glucose, there may be a threshold plasma level of FFAs, ~200–300 μM, below which cardiac FAU does not occur^55,107. 10–30% of FAs taken up cycle through the intracellular TG pool (see below), while the rest are immediately oxidized^{43,93,104,109,110}. Increased workload, e.g. with pacing or adrenergic stimulation or exercise, increases FA consumption largely in proportion to cardiac work, and in rough proportion to levels of circulating FFA levels, which also rise in response to these stressors^{93,94,111–114}. Conversely, insulin clamp suppresses FA uptake in approximate proportion to suppression of lipolysis and plasma FA levels^33,115,116, as does direct inhibition of adipose tissue lipolysis, with compensation by increases in MGU and lactate uptake^50,111,114. When circulating FFAs are sufficiently lowered, suppression of FAU can be near complete, with the little FA that is taken up largely esterified to the TG pool⁴⁸. Cardiac FAU is usually found increased in obesity¹⁰⁴, in both type 1 and 2 diabetics^{46,99,100,110}, and in non-diabetics in proportion to glucose intolerance¹¹⁷, although not in all studies^118,119. These increases usually associate with increases in circulating FFA levels^48,104,115, but not always¹¹⁰. FAU and FAO also decline slightly with age¹²⁰. Men treated with metformin increased plasma FFAs and MFAU, while women treated with rosiglitazone did the opposite¹²¹.

Compared to the literature on cardiac FFA uptake, few data exist on uptake from the TG pool in lipoproteins, a process driven by lipoprotein lipase (LPL) on the luminal side of capillaries, in part because lipoprotein TGs are harder to quantify. A small study of 2 patients with LPL deficiency suggested higher fasting MGU than 2 control subjects (but still within normal range), suggesting a compensation for deficient LPL-mediated uptake of FAs from VLDL; subjects were not tested in a fed state, which would interrogate uptake from chylomicrons¹²². Studies using orally administered FA tracers, thereby labeling chylomicrons, suggest that more FAs are extracted from chylomicrons by hearts in subjects with pre-diabetes versus healthy controls, and that this is reversed by weight loss^123,124, but much remains to be studied with this approach. Comprehensive quantification, by LCMS in A/CS studies, of fuel use in fasted subjects failed to account for ~25% of oxygen consumption, which was ascribed to unmeasured extraction from VLDL¹⁶.

A complication of all these studies is introduced by the local liberation of FAs from lipoprotein particles without subsequent uptake of FAs into cardiomyocytes, i.e. “spillover” that increases FA content in coronary sinus. Such spillover can cause cloud efforts to differentiate uptake from the free and esterified FA pools, leading to underestimates of FFA uptake and overestimates of uptake from the TG pool in A/CS sampling studies. ¹⁴C- or ³H-labeled palmitate infusions have demonstrated spillover to be as high as 50% of measured FAU^{43,44,47,114,125}. Heparin, nearly always used in A/CS studies, activates LPL and can worsen the spillover effect⁴³. Spillover appears to decrease to near zero during exercise, suggesting more efficient capture of LPL-liberated FFAs¹²⁶. Glycerol can be expected to be liberated during LPP lipolysis, which the heart can consume⁵³.

The molecular mechanisms of FA transport into cardiomyocytes, as in other cells, remain unclear. Loss-of-function genetic variants in the FA transport facilitator CD36 are frequent in some groups, including ~10% of Japanese populations, and subjects lacking CD36 have reduced cardiac FAU, with compensating increase in MGU, demonstrating a key role for CD36 in this process^127–130. Interestingly, subjects lacking CD36 show no obvious baseline cardiac dysfunction, although an association with the development of HCM has been suggested¹³¹.

In summary, the heart may derive substantial FAs from circulating TGs, especially post-prandially, but circulating FFA levels seem to be the primary driver of cardiac FAO. However, plasma levels of FA do not precisely correlate with FA uptake¹⁶, indicating that other factors are at play as well.

Lactate

The absence of imaging modalities (e.g. PET) for lactate limits available data in humans almost exclusively to A/CS studies. These studies have consistently shown that the heart is a net consumer of lactate, ranging from 0.05–0.2 μmoles/min/g fasting to 0.5 under hyper-insulinemic clamp^{10,16,33,43,59,69,99,100,105} (and Sup Tables). As noted above, the heart both generates and secretes lactate. Up to 50% of secreted lactate is derived from exogenous glucose undergoing glycolysis^32–34, accounting for an average of ~15% of MGU (but the ranges are wide), and increasing with obstructive CAD¹³². The remaining lactate likely derives from cycling through the glycogen pool (see below). The fact that lactate secretion occurs even as it is net consumed by the heart may reflect compartmentalized anerobic glycolysis, or it may simply reflect partial exchange of pyruvate pools via the rapid LDH reaction and lactate transport. The lactate that is not secreted is entirely oxidized^32,89.

As with MGU, MLU in healthy subjects best correlates inversely with plasma FFA levels, consistent with suppression of PDH by FAO^32,68. Insulin suppresses adipose lipolysis and increases plasma lactate, likely by promoting systemic glycolysis, and MLU rises accordingly^12,53,55,69. In contrast, lactate consumption is reduced in subjects with diabetes^12,76. Interestingly some studies report increased anaerobic glycolysis to the point of net secretion of lactate in patients with type 1 diabetes^99,100. In general, increasing metabolic demand, e.g. with pacing or exercise, increases lactate consumption to a point that can exceed that of glucose or even FAs, usually concordant with elevations in plasma lactate^{34,89,99,105,126,133–135}. Secretion of lactate increases mildly during cardiac pacing, but MLU rises more, thus increasing the net uptake of lactate^32,34, a response that is unaffected in the presence of acute moderate hypoxia (15% FiO2)¹³⁴. Thus, a healthy heart, even when strenuously exercising, likely does not experience hypoxia sufficient to trigger substantial anerobic glycolysis. Exercise in the context of more severe hypoxia (12% FiO2), however, elicits substantial lactate release to >50% of lactate uptake, although net uptake remains positive, and the contribution of anerobic glycolysis to ATP production remains small¹³⁵. When obstructive CAD is present, increasing workload or vasodilator-induced vascular mismatch promotes anaerobic glycolysis to the point of net lactate secretion, likely a reflection of hypoxia^{91,92,112,136–138}. Induced global ischemia during cardiac surgery also causes net secretion of lactate¹⁴. Interestingly, pacing subjects with MINOCA does not trigger lactate release, despite electrographic and symptomatic evidence of ischemia, suggesting that oxygen may not be limiting in these subjects^139,140.

Amino acids

AAs contribute only marginally to total cardiac ATP generation. Of all AAs, glutamate is taken up most avidly, though this has only been investigated in fasting subjects. Glutamate uptake ranges 2–7 nmoles/min/g, correlates closely with arterial concentrations, and undergoes a high extraction ratio of 20–50%^16,69,141, although extraction saturates with supraphysiological provision of glutamate¹⁴². Interestingly, myocardial uptake of glutamate increases in ischemic tissue^92,136,143. BCAAs are mildly taken up in fasting subjects, ranging 1–3 nmoles/min/g^16,54,99,144 but are secreted in diabetics⁹⁹. Despite the uptake of glutamate and BCAAs, however, the heart of overnight fasted individuals is in substantial overall negative nitrogen balance¹⁶, reflecting substantial protein breakdown, calculated to be ~10mcg/min/g, or ~4% cardiac mass/day^16,54. BCAA infusion promotes BCAA plasma levels, cardiac uptake, and cardiac protein synthesis, but has little effect on protein breakdown, as measured with labeled phenylalanine release⁵⁴, suggesting that the latter is not regulated by BCAA levels.

In the face of protein breakdown, nitrogen balance is maintained in fasting subjects largely by amino acid secretion. Of these, glutamine is most abundantly secreted^16,92,136. Glutamine synthesis from glutamate, the primary method of ammonia detoxification, thus limits ammonia release to trace amounts^92,136. Glutamate uptake and glutamine release are roughly equal in magnitude, suggesting that glutamate uptake is largely devoted to nitrogen secretion^16,92,136. Alanine is usually also reported as substantially secreted, at a rate of ~2 nmoles/min/g^{16,67,69,92,136,140,141}, and secretion increases with pacing¹⁴⁴. In sum, AAs contribute little to cardiac ATP generation but likely play an important role in nitrogen balance, in part mediated by high glutamate uptake and glutamine secretion.

Ketones

Ketones have garnered much attention recently, in light of their potential involvement in the remarkable benefits of SGLT2 inhibitors (see below), but data on human cardiac ketone use are limited. Cardiac ketone uptake is ~0.05 μmoles/min/g, contributing 3–5% of ATP production^11,16,53,99. Ketones have a high extraction ratio of 20–30% and are taken up by the heart in proportion to plasma levels¹⁶, suggesting the absence of cardiac-intrinsic regulation.

The heart does not secrete ketones⁹⁹. Elevated ketone uptake is seen in subjects with diabetes, reflecting elevated plasma levels^76,99. Infusion of ketones to supraphysiologic levels (4mM) suppresses both MGU (by as much as 50%), and FAO^40,145. Insulin completely suppresses circulating ketones and thus cardiac uptake^53,99.

Stored fuels

Myocardial TG content

Isotope tracing studies have shown that 10–30% of FFAs taken up by the heart are channeled to the TG pool, rather than FAO. The dynamics of the TG pool cannot be measured over small periods of time, which means that measured FAO with ¹¹C palmitate PET studies may at times underestimate true FAO, i.e. it does not measure FAO derived from these intracellular stores. Myocardial TG content (myoTG, not to be confused with pericardial and epicardial fat) can be evaluated noninvasively by ¹H (proton) MRS, and is normally in the range of 0.5 % fat/water, or ~20μmoles/g, though it can double in the context of metabolic derangements (Sup Tables). The turnover of myocardial TG pools is thus likely on the order of hours. TG content correlates with markers of metabolic syndrome, and most closely with obesity and visceral adiposity^146–153. It also increases substantially with other contexts of high FFA delivery, such as 48hrs fasting^154–156, or in short-term very low calorie diet (VLCD) in healthy men¹⁵⁷, despite lower liver TG content in this context. 3 days of VLCD in subjects with diabetes also increases myoTG, and the increase is prevented by blocking adipose tissue lipolysis, indicating that FFA delivery is the primary driver of myoTG¹⁵⁸. Consistent with this, myoTG does not increase after a single high-fat meal¹⁵⁶. Interestingly, myoTG does not decrease after bariatric surgery, despite dramatic loss of adipose burden elsewhere, including epicardial fat^159,160. In contrast, a 16 week diet that reduces BMI by 25% in obese diabetics also decreases myoTG by 25%¹⁶¹. Diabetes and age, independently of BMI, also promote myoTG accumulation, and in both cases myoTG correlates with markers of diastolic dysfunction (e/a ratio) or concentric remodeling^{150,151,158,162–165}. In healthy volunteers, the accumulation of myoTG with 48hrs of fasting less clearly associates with diastolic dysfunction¹⁵⁵. The flux out of the cardiac TG pool likely increases under some circumstances, e.g. exercise^114,126, but data on this question are scant. In summary, TGs provide a substantial source of fuel in the heart, probably accounting for 10–30% of FAO, and excess TGs, which accumulate in the context of excess plasma FFA levels, generally correlate with evidence of diastolic dysfunction.

Glycogen content

As noted above, A/CS tracer studies show that 50–75% of MGU likely cycles through the glycogen pool, even under artificial hyperglycemic conditions^33,43. Atrial and papillary muscle biopsies reveal glycogen content in the ranges of 25–40 μmoles/g tissue, or ~0.5–1% of cardiac mass^166–168, theoretically sufficient to fuel cardiac contraction for ~30 minutes in the absence of other fuels. Based on ¹⁴C-glucose and lactate tracing studies, Wisneski et al. calculated glycogen synthesis rates of 0.1 μmoles/min/g (glycosyl unit equivalents) under fasting conditions, which matched or even exceeded calculated breakdown, indicating that, unlike in skeletal muscle and especially liver, fasting promotes glycogen accumulation in the heart³³. Induced hyperglycemia or insulinemia further increases calculated glycogen synthesis out of proportion to calculated breakdown^33,55. Glycogen is thus presumably reserved for acute demand, perhaps especially when oxygen is lacking. Consistent with this notion, the observed elevated MGU during pacing persists during recovery, suggesting depletion and subsequent replenishment of glycogen stores⁹⁰. Rates of cardiac glycogen use in human have not been reported, however.

Does aberrant metabolism contribute to heart failure?

Metabolic insufficiency can undeniably cause heart failure. Approximately 5% of cardiomyopathy cases in pediatric cohorts are caused by inherited metabolic disorders, driven largely by glycogen storage diseases and defects in fatty acid oxidation and oxidative phosphorylation. In general, these are most likely caused by impaired cardiac metabolism, although indirect contributions by systemic metabolic insufficiencies cannot always be ruled out. The reader is referred to excellent reviews on this wide range of disorders ^169–173. These disorders demonstrate conclusively that metabolic derangements are sufficient to cause human heart failure. It cannot be concluded, however, that such derangements are necessary for heart failure to develop, or found in more common forms of HF. The metabolic perturbations in inherited disorders are often profound, generated by complete enzyme loss, in contrast to often only mild aberrancies detected in patients with heart failure but without inherited metabolic disorders. Moreover, it is notable that extensive genome-wide association studies (GWAS) have to date identified few SNPs that associate with HF or DCM and that reside near genes with known function in cardiac metabolic pathways¹⁷⁴, although these studies are limited by the heterogeneity and multifactorial nature of HF. It is worth asking, therefore what evidence supports the existence of metabolic derangements in patients with heart failure who do not have inherited metabolic disorders.

Is the failing human heart energy starved?

The notion that heart failure stems in part from an energetic deficiency has long been proposed, and the failing heart has been called “an engine out of fuel”¹⁷⁵. It should be noted, however, that as HF develops and compensatory hypertrophy occurs, myocardial oxygen consumption does rise, largely in proportion to LV mass. Thus, even though the heart becomes a less efficient user of fuel, it does not at baseline appear fuel-limited. On the other hand, ATP production and energy reserves may be hindered. Interrogating ATP levels and production has almost exclusively been accomplished by ³¹P-MRS, because the high metabolic activity of the heart renders high-energy phosphates too short-lived for evaluation by tissue acquisition, and ³¹P-MRS is the only technique capable of reliably measuring high-energy phosphates noninvasively. The PCr/ATP ratio has widely been used as a surrogate of “energy charge” in the myocardium, because PCr and ATP are most abundant and easy to quantify, and a reduction in PCr can indirectly reflect a rise in ADP (i.e. a decline the ΔG of ATP hydrolysis). However, it should be noted that this conclusion assumes a constant level of Cr, which is likely true acutely, but chronic changes in PCr/ATP may instead reflect reduced levels of Cr (see below).

Cardiac PCr/ATP remain stable with increased workload in healthy subjects, e.g. with low-dose leg exercise or with adrenergic stimulation^164,176,177, and studies in dogs have shown that this is true over a wide range of workloads^178,179, reflecting the remarkable dynamic range of cardiac oxidative capacity (in contrast to skeletal muscle, where PCr/ATP drops with exercise). In contrast, cardiac PCr/ATP is reduced in numerous chronic states, even in the absence of HF (Sup Tables). It is reduced 15–25% in subjects with DM, either type I or II, independently of BMI, and is further reduced w/mild exercise (unlike control subjects), unveiling a metabolic insufficiency^{164,180–184}. Cardiac PCr/ATP is also reduced 20% in subjects with pre-diabetes or with liver steatosis, in absence of overt cardiac dysfunction^185,186. PCr/ATP is also reduced 15% in obesity¹⁷⁶, reversed by weight loss^176,187, and reported to reduce further with increasing cardiac workload by some¹⁸⁸ but not others¹⁷⁶. PCr/ATP may also be reduced with sedentary lifestyle¹⁸⁹ and after a few days of high-fat/low-carbohydrate diet¹⁹⁰. Finally, PCr/ATP reduces linearly with age, at the rate of ~0.02/yr^191,192. In sum, PCr/ATP is normally defended by the heart under a range of workloads acutely, but numerous conditions of systemic metabolic dysregulation chronically lower PCr/ATP.

Cardiac PCr/ATP has consistently been found reduced in HFrEF, whether ischemic, valvular, or other etiology^28,193–200, with few exceptions²⁰¹. The ratios are also reduced 15–20% in HCM, independent of causal mutation^{177,201–204}. In both cases, reduced PCr/ATP is further reduced w/mild exercise, again demonstrating poor reserves. Importantly, reduced PCr/ATP in patients with DCM strongly correlates with mortality at 2.5 yrs¹⁹⁸. Reduced PCr/ATP is partially reversed with heart failure therapy or valve replacement and improved NYHA class in some studies¹⁹⁷, but not all²⁰⁵. Cardiac PCr/ATP is also reduced 25% in the context of severe aortic stenosis (AS) and partial resolves after AVR^199,206,207. In general, PCr/ATP ratios correlate better to NYHA class than to LVEF, and to diastolic parameters such as peak LV filling rate or E/A on imaging. Consistent with these observations, cardiac PCr/ATP is also reduced in HFpEF in proportion to severity^208–211. In sum, PCr/ATP reductions are seen in all forms of HF, independent of EF or underlying metabolic conditions.

The notion that PCr/ATP reductions reflect increased cardiac ADP (and therefore impaired energy charge), is based on the assumption that reduced PCr levels result in a lower PCr/total Cr ratio. However, creatine content in myocardium, normally ~100 nmol/mg, is reduced as much as 50% in tissue from patients with both hypertrophic and dilated cardiomyopathy^212–214, and noninvasive measures of cardiac creatine correlate positively with LVEF and inversely with BNP^153,215,216. Reduced PCr/ATP in HF thus may reflect low Cr rather than high ADP. The mechanism for decreased Cr is unclear. One hypothesis has been that PCr decreases when energy consumption exceeds supply, and the creatine pool shrinks in compensation to maintain PCr/Cr, and thus ATP/ADP²¹⁷.

As discussed above, CK flux is an alternative method to evaluate cardiac energetic capacity by ³¹P-MRS and is not identical to PCr/ATP, although it similarly interrogates the metabolic reserve capacity of the heart. For example, PCr/ATP is reduced in obesity while CK flux is higher, and increasing cardiac workload in these individuals fails to increase CK flux, in contrast to lean controls, again demonstrating a loss of metabolic flexibility¹⁷⁶. CK flux is reduced by 50% in DCM²⁸, post-MI²¹⁸, and in HCM²¹⁹. It is also reduced by 30% in LVH without clinical heart failure, and 65% in LVH with heart failure (all with preserved EF), and reductions in CK flux correlate best with NYHA class²²⁰ and cardiac work²²¹. Importantly, akin to PCr/ATP ratios, low CK flux predicts HF-related events at 4–5yrs²²² and preserved CK flux correlates with improved EF over a median 7 years²²³. Finally, independently of PCr/ATP or CK flux, reduced baseline levels of ATP (determined by ³¹P-MRS) correlate with ICD firings and cardiac death over subsequent 10 years²²⁴, suggesting that metabolic insufficiency also underpins arrhythmias, but perhaps by different mechanisms than mechanical insufficiency.

In sum, despite the differences between PCr/ATP and CK flux, both are reduced in multiple forms of HF, and both correlate with adverse clinical outcomes (although much more data is available for PCr/ATP). In both cases, defects are also detected in subjects with aspects of metabolic syndrome but without apparent cardiac dysfunction, suggesting that cardiac metabolic alterations precede functional impairments. Precisely what metabolic alterations lead to these observations (e.g. mitochondrial insufficiency versus loss of creatine versus altered CK enzymes), and whether these defects contribute to HF, remains uncertain.

Are mitochondrial energetics defective in heart failure?

The answer is remarkably unsettled, reflecting the multitude of approaches taken to ask the question. Many studies measure enzyme activities, usually spectrophotometrically, in tissue homogenates or isolated mitochondria, but these approaches lack the ability to evaluate integrated ATP generation by the intact system. The latter can be achieved by measuring oxygen consumption, with polarographic (Clark) electrodes, of permeabilized fibers or isolated mitochondria (each of which has limitations, e.g. selection bias toward more healthy mitochondria during the isolation process), but there are few such studies. Mitochondrial volume, evaluated by EM, is often reported to be diminished in HFrEF, with fragmented mitochondria, loss of normal cristae architecture, and decreased matrix density^225–227, but not always so²²⁸. Citrate synthase (CS) activity, a good correlate with mitochondrial quantity, is also often reported to be reduced in DCM samples^229–232, but not always^{228,233–236}. Similarly, mitochondrial DNA (mtDNA) content is usually reported reduced²²⁹ but not always²³⁷. One study found mtDNA suppression out of proportion to other markers of mitochondrial function, suggesting a mtDNA replication defect²²⁹. Expression of genes encoding mitochondrial proteins, including of FAO, are also often reported reduced, but not always^214,238, and sometimes reported to reverse with LVAD unloading²³⁹. Most studies report reduced levels of TCA intermediates in DCM^{214,227,239,240}. Respirometry with permeabilized fibers report 20–50% reductions in activities of all mitochondrial complexes²³³, specifically CI²³⁰, or CI and CIV²²⁷, the latter interestingly not reversed by LVAD unloading despite improvements in cardiac function. Spectrophotometric assays yield equally variable results, including 20–50% reductions in CII and CIII, but not CI and CIV²³⁵, or reduced CIII and CIV but not CII²⁴¹. Other studies, including polarographic studies of isolated mitochondria with Clark electrodes, have indicated no significant changes²²⁸ or compensatory increases in mitochondrial activity in cardiomyopathy^236,242,243. All these studies are typically carried out with relatively low numbers of patients (as low as 2 controls vs 3 CHF²³⁰). In sum, the nature of mitochondrial energetic insufficiency in heart failure, if any, remains uncertain, reflecting the limitations of small studies, the variable quality of tissue acquisition, and the divergent methods to measure mitochondrial quantity and function. The interested reader is referred to an excellent review on mitochondria in the human heart²⁴⁴.

Altered fuel use in heart failure

Glucose

Preclinical studies in mice, rats, dogs, and pigs almost universally reveal an increase in MGU in models of HFrEF, at the expense of FAO, often termed a “switch” to glycolysis. The evidence that this occurs in humans is mixed, however, and complicated by multiple factors. First, as noted above, MGU is reduced in subjects with diabetes and pre-diabetes, conditions that are increasingly prevalent among HF cohorts. Moreover, HFrEF itself can promote systemic insulin resistance^245–248, and MGU is usually seen to track with systemic and skeletal muscle glucose disposal. Systemic metabolic alterations may thus counteract a HFrEF-induced myocardial-specific increase of MGU²⁴⁹. In parallel, different pathophysiologic processes, e.g. ischemic insults vs genetic deficiencies, likely have different impacts on cardiac metabolism. Different medications can also affect cardiac metabolism differently. And lastly, cardiac metabolic abnormalities likely differ substantially between different stages of HF, e.g. early vs terminal, and different demands, e.g. rest vs exercise or pacing.

The first A/CS study in HF, comparing 20 patients with low-output failure (no echocardiography was available at the time) to 11 subjects without heart disease, found nearly no differences in contributions of substrates to cardiac metabolism¹¹. A subsequent A/CS study found no differences in MGU in patients with CHF at baseline but revealed reduced MGU under hyper-insulinemic clamp, in the context of demonstratable systemic insulin resistance²⁴⁶. Similarly, another study showed an increase in baseline fasting MGU but a severely diminished ability to further recruit MGU in response to pacing⁵⁹. FDG-PET studies in subjects with DCM have often not included controls, but based on appropriate comparisons (i.e. comparable BMI, age, FFA levels) the studies similarly suggest elevated fasting MGU in patients with reduced EF^95,250,251 but suppressed MGU when insulin is high²⁵². ¹¹C-glucose studies also revealed a doubling of fasting MGU in patients with DCM at baseline⁵⁸. In contrast, neither patients with reduced LVEF nor controls revealed appreciable MGU in recent A/CS studies with older subjects and higher circulating FFA levels ¹⁶. Myocardial glycogen content does not appear to be altered in HFrEF²⁵³. In cardiac hypertrophy with preserved EF secondary to AS, fasting MGU does not change, but MGU in response to hyperinsulinemia is decreased compared to controls²⁵⁴. MGU is reported to decrease in HCM²⁵⁵ but increase in proportion to BP in patients with hypertension^65,66. Together, though not conclusive, these studies most consistently reveal an elevated baseline MGU in HFrEF if circulating FFAs are sufficiently low, but a lack of metabolic flexibility in response to insulin or metabolic demand.

Tissue-based studies have not, on balance, convincingly indicated molecular changes in glucose transport in human HF. GLUT1 and GLUT4 are widely accepted as the predominant cardiac glucose transporters, but this has not been validated in the human heart. GLUT4 mRNA and protein expression are generally reported as unchanged in HFrEF, especially in larger cohorts^214,226,256, though in smaller cohorts sometimes reported as elevated^78,257 or suppressed ^77,258,259. In one study of hypertrophy in response to AS, GLUT4 protein expression correlated with LVMI, and inversely with CD36 expression²⁶⁰, although this is not always seen^254,257. GLUT4 protein is in part regulated by translocation to the plasma membrane (PM) by Insulin/PI3K/Akt signaling, although this has never been shown directly in human adult cardiomyocytes or tissue. In a small cohort of subjects with HFrEF, PI3K activity and GLUT4 at the PM were elevated compared to controls, in proportion to circulating insulin, suggesting intact signaling through cardiac insulin receptor despite systemic insulin resistance⁷⁸. Others, however, report reduced²⁴⁸ or no change²¹⁴ in Akt activation in HFrEF. GLUT1 expression is alternatively reported as increased^226,257, decreased^258,259,261, or in a large well-controlled cohort reduced at mRNA but unchanged at protein level²¹⁴.

Fatty acids

FAU and FAO are nearly always found to be suppressed HF, regardless of the etiology of HF or the study modality. For example, FAU is reduced by 30% in DCM by A/CS studies⁵⁹; 30–50% by ¹¹C-palmitate PET studies, despite similar FFA levels (~600μM)^58,101,103; 20% by FTHA PET studies in LVH, despite similar FFA levels, with an inverse association between FAO and LV mass^101,102,262; and a range of decreases using the non-hydrolysable FA analog 123I-BMIPP SPECT (akin to FDG), with FAU correlating with LVEF, and inversely with PCWP and LVEDP^216,263. However, as with glucose uptake, interpretations of FAU and FAO can be complicated by the fact that HFrEF can independently promote systemic insulin resistance^245–248 which may promote FAU and in part counteract HFrEF-induced suppression of FAU. HFrEF-induced adrenergic stress can also promote lipolysis and have similar effects²⁶⁴. Subject with low EF and insulin resistance can thus be found to “pseudonormalize” their cardiac FAU¹⁰³. Suppression of FAU may be more pronounced in men than women³⁵. In sum, there is strong support for the notion that FAU and FAO are substantially reduced in HF of various origins, although it should be noted that the suppression is not complete, and FAO generally remains the major source of cardiac fuel even in HFrEF.

Preclinical studies have typically revealed mechanistic defects in FAO in various heart failure models. Data from human cardiac tissue, however, tend instead to primarily reveal defects in FAU, thereby indirectly reducing FAO. Circulating plasma acyl-carnitines, intermediaries of FAO, are elevated in HF, generally reversed by LVAD implantation, and correlate with adverse outcomes²⁶⁵, which some have interpreted as evidence of defects in cardiac FAO. However, A/CS studies reveal little to no secretion of acyl-carnitines in HF^15,16, and intermediates of FAO, including acyl-carnitines, acyl-CoAs, and FAs, are found reduced in cardiac tissue of patients with DCM, HCM, and even HFpEF with markedly elevated BMIs^{214,239,240,266–268}. These observations are thus most consistent with a defect in FA import, upstream of FAO, indicating that circulating acyl-carnitines must arise elsewhere. The fact that reduced FAO intermediates are seen in all cases, including end-stage disease and biopsies from compensated HF, suggests that defects of FAU are common to multiple forms of HF, and occur relatively early in disease progression. The mechanisms underlying FA import remain poorly understood. CD36 plays an important role (see above) and its expression is reduced in hypertrophy in response to AS²⁶⁰. The expression of other regulators of FA import, including ACSL1, are reduced in DCM²¹⁴. FAO genes are also often reported suppressed in HFrEF tissue, though not always^{214,225,248,256,258}, indicating that the FAO program may also be affected.

Some studies report that myoTG levels are decreased in DCM^248,269, consistent with decreased FAU in this condition, while others report lower myoTG in HCM but not DCM, and one study reported higher myoTG in DCM, mostly driven by obesity and diabetes^153,270. As discussed above, obesity associates with elevated myoTG, often called a “fatty heart”²⁷¹, and myoTGs correlate with diastolic dysfunction. Moreover, prolonged obesity associates with the development of reduced systolic and diastolic function and heart failure^272,273. However, whether the accumulated myoTG contributes to cardiac dysfunction remains uncertain. The observation that tissue acyl carnitines are reduced even in patients with BMI>40²⁶⁸ suggest that accumulated myoTG may in fact reflect a compensatory mechanism to remove potentially cardiotoxic FA intermediates. Other incompletely esterified FA intermediates, including DAGs and ceramides have been reported elevated in HF tissue in some studies^248,274 but reduced in others²¹⁴. In summary, elevated myoTGs reflect excess FAU by the heart and are likely a protective compensatory response.

Lactate, ketones and AAs

In contrast to the sometimes equivocal findings for MGU, net lactate consumption is consistently found to ~double in HF, independently of plasma levels^15,16,59. A/CS studies with lactate tracers suggest that the increase in net lactate uptake largely reflects decreased lactate secretion, i.e. reduced anerobic glycolysis and relatively more glucose oxidation, in addition to increased lactate uptake⁵⁹. This contrasts with animal studies that have suggested increased anaerobic glycolysis in HFrEF. The mechanism driving increased lactate oxidation is likely two-fold. First, tissue studies have revealed a profound reduction of PDK4 protein in failing heart, with consequent dephosphorylation and activation of PDH²¹⁴, thus derepressing entry of pyruvate (and thus lactate) carbons into the TCA cycle. Second, the reduced FA influx (see above) likely also relieves PDH inhibition mediated by the Randle glucose-fatty acid cycle. Expression of the mitochondrial pyruvate carrier (MPC) has been reported decreased in HF^275–277 (though not always²¹⁴), indicating either that MPC levels are not limiting for pyruvate entry or that other methods of lactate entry into mitochondria exist²⁷⁸, as suggested by ¹³C tracer studies in mice lacking cardiac MPC²⁷⁹. In sum, even in the absence of appreciable MGU, lactate consumption increases in HFrEF¹⁶.

Ketone uptake, measured by A/CS sampling, has also consistently been reported increased in patients with reduced EF or AS^15,16. The higher uptake is likely driven primarily by mass action, as uptake correlates with plasma ketone levels, which are elevated in HF in proportion to disease severity²⁸⁰. Expression of ketone catabolic enzymes, BDH and OXCT1, may also be mildly increased in HFrEF^214,240. Administration of ketone ester drink increases plasma ketone levels and is sufficient to increase cardiac ketone use, at the expense of both FA and lactate uptake, in both control and subjects with HF⁶¹. Cardiac consumption of ketones in heart failure is widely thought to be adaptive and has been proposed as a mechanism for the dramatic benefits of SGLT2 inhibitors (see below). However, PET and A/CS hemodynamic studies have shown that acute ketone infusion in normal subjects and patients with HFrEF has consequential vasoactive effects, nearly doubling MBF⁴⁰ as well as CO, HR, EF, and SV, likely driven by reduced SVR^40,281. Putative benefits of ketones may thus largely occur via vasorelaxation, rather than direct effects on cardiac metabolism, as shown recently in rodents²⁸².

Glutamine secretion, alanine secretion, and net negative nitrogen balance are accentuated in fasting subjects with reduced EF¹⁶ or critical CAD^92,136,144, and metabolic demand by pacing stresses the cardiac ability to maintain nitrogen balance in these patients, increasing ammonia release in inverse proportion to extraction of glutamate, the substrate for fixing ammonia to glutamine^92,136. HFrEF hearts thus appear to have a predilection for high protein breakdown, although this hasn’t been proven. Tissue levels of amino acids are typically reported as reduced in HFrEF, with the exception of elevated BCAAs and BCKAs^{214,227,283,284}. The latter are sometimes accompanied by suppressed expression of BCAA catabolic genes^214,283,284, which has suggested that failing hearts harbor a BCAA catabolic defect that contributes to HF. A/CS studies, however, do not reveal decreased oxidation of BCAAs in HF, despite increased protein catabolism and secretion²⁸⁴. Interestingly, GWAS studies have associated increased, rather than decreased, BCKDHA expression with HF²⁸⁵, and LVAD-mediated unloading increases, rather than decreases, BCAA concentrations²²⁷. The role of BCAAs, and of increased protein breakdown, in human HF thus remains uncertain.

HFpEF

HFpEF is now as common as HFrEF in the United States, with similar prognosis. The syndrome has variable presentations²⁸⁶ and is increasingly associated with metabolic derangements, including obesity and diabetes, suggesting the possible presence of unique cardiac metabolic changes. However, very little is known about cardiac metabolism in HFpEF. No A/CS studies have been reported in HFpEF. As in HFrEF, cardiac PCr/ATP is reduced in proportion to severity^208–211. MRS and invasive approaches have reported microvascular and coronary flow reserve insufficiency^287–290, consistent with reduced microvascular density found on autopsy samples²⁹¹, findings that are not generally reported in HFrEF. MyoTG assessed by proton NMR appears increased in HFpEF, in inverse proportion to PCr/ATP and to diastolic function^208,292. This observation also contrasts with the typical reduction in myoTG seen in HFrEF^248,269 and may reflect the high prevalence of insulin resistance and diabetes in patients with HFpEF. Despite this increase in myoTG, however, endomyocardial biopsies in obesity-associated HFpEF revealed strong reductions in levels of acyl carnitines, akin to the reductions found in HFrEF²⁶⁸. These findings suggest that, as in HFrEF, FAU is limiting in HFpEF, despite being associated with elevated myoTG. No PET studies with FA tracers have addressed this question to date.

Therapeutic interventions: failures and unexpected successes

To date, no approved therapy exists designed to target cardiac metabolism, despite much effort. However, an argument can be made that most existing therapies, largely involving neurohormonal blockade, indirectly benefit cardiac metabolism by reducing preload, afterload, and chronotropy. More recent therapies such as SGLT2 inhibitors may also impact cardiac metabolism (see below). However, to what extent the improved cardiac metabolism contributes to adaptive remodeling, if at all, remains unclear in all these cases.

One therapeutic approach specifically targeted to cardiac metabolism has been to suppress FAO in order to promote glucose utilization. Compared to glucose, fatty acids generate 10–30% less ATP per oxygen consumed, known as the ATP/Oxygen (P/O) ratio²⁹³. FAO inhibitors (FAOi) have thus long been proposed as anti-anginal agents. Perhexiline, a reversible inhibitor of CPT1, was a popular and effective treatment for angina in the 1970s (before, in fact, it was known to inhibit FAO)²⁹⁴. However, by the late 1980s, most countries removed perhexiline from the market due to complications of on-target severe hepatotoxicity and neurotoxicity. The variable rates of toxicity likely relate to polymorphisms in CYP2D6, the primary enzyme that metabolizes perhexiline²⁹⁵. More recent trials have revisited the use of perhexiline with level-adjusted dosing, with some evidence of efficacy, but the drug remains unapproved^296–298. Likewise, studies with etomoxir, an irreversible inhibitor of CPT1, were halted prematurely due to liver toxicity²⁹⁹. CPT1 therapy thus seems too aggressive. Trimetazidine is thought to target long-chain 3-ketoacyl Coa thiolase (LC3KAT), which catalyzes the last step of the FA beta-oxidation cycle. Interestingly, ¹¹C-palmitate PET data suggest that trimetazidine suppresses FAO primarily from endogenous TG pools³⁰⁰. Somewhat surprisingly, however, MGU does not appear to increase with treatment^300,301. Trimetazidine has been studied since the 1960s³⁰² and has shown benefit in angina, NYHA class, and LVEF in patients with ischemic heart disease in various studies, several of which are randomized and placebo-controlled^303–310. In contrast, benefits were mild in a single-blind RCT³¹¹ or negative in a larger DBPCRT in patients with non-ischemic cardiomyopathy³⁰¹, and were similarly negative in a HFpEF DBPCRT³¹². Trimetazidine is approved in France, but not in the US. Ranolazine, approved by the FDA in 2006 for the treatment of angina, was originally thought also to target LC3KAT but is now generally recognized to primarily function by inhibiting late I_Na. In summary, FAOi have shown some efficacy for treating angina, but in general not for heart failure. In retrospect, these results are consistent with the fact that there is little evidence that myocardial blood flow is limiting in most forms of heart failure. Changing the P/O ratio, a strategy devised to treat ischemia, is thus not likely useful for HF in the absence of obstructive vascular disease.

Other metabolic strategies, beyond FAOi, have been entertained. Inhibiting PDK, thereby activating PDH and promoting pyruvate entry into the TCA, is a potentially more direct way to boost carbohydrate use by the heart. Dichloroacetic acid (DCA), a pan-PDK inhibitor, has been shown to have acute benefits^313,314, but its substantial toxicity has precluded long-term studies. Moreover, the already suppressed expression of PDK4 in heart failure (see above) may render further PDK suppression moot. There is also great interest in ketone supplementation as a potential therapy. A recent randomized double-blind crossover study in 24 patients with HFrEF showed that oral supplementation with ketone esters improved cardiac hemodynamics, including resting and exercise CO and PCWP, although, as noted above, the effects may largely be secondary to improving vasorelaxation rather than direct effects on cardiac metabolism³¹⁵. Supplementation with NAD precursors is also being tested in early clinical studies, based on preclinical work indicating benefit, and has shown effects on circulating leukocytes in patients with HF³¹⁶, but data on cardiac function in patients with HF are pending^317,318. No impact on cardiac function was seen with supplementation of NAD precursors in healthy overweight individuals³¹⁹. Systemic activation of BCAA catabolism is also being tested in early clinical studies³²⁰, although the mechanism of action may primarily be outside the heart²⁸⁴. Coenzyme Q₁₀ (CoQ₁₀), or ubiquinone, is a critical electron carrier in mitochondria and has been suggested to be deficient in some forms of HF. One randomized placebo-controlled double-blinded trial showed improved outcomes, including reduced mortality, with CoQ₁₀ supplementation in patients with HF, but a recent Cochrane meta-analysis concluded that there is currently no convincing evidence to support or refute the use of CoQ₁₀ for HF³²¹, and CoQ₁₀ supplementation is not currently widely used³²². Elamipretide, a tetrapeptide that interacts with cardiolipin in mitochondria to improve bioenergetics, also had promising pre-clinical findings, but a randomized blinded phase 2 trial showed no improvement in cardiac function at either low or high doses of drug³²³. In general, studies that aim to target mitochondria are hampered by the absence of reliable biomarkers reflecting cardiac mitochondrial function³²⁴.

New agents, including SGLT2 inhibitors (SGLT2i) and GLP-1 receptor agonists (GLP1RA), have proven efficacious agents in the treatment of HF, and have been suggested to do so in part through effects on cardiac metabolism. SGLT2i dramatically improved outcomes in HFrEF and HFpEF in large phase-III DBPCRTs^325–328. How these drugs, which target SGLT2 expressed in the kidney, benefit HF is not clear. One hypothesis invokes increased ketogenesis in response to glucose wasting in the urine, resulting in beneficial increased use of ketones by the heart. Acute administration of ketones IV does markedly improve hemodynamics in HFrEF, including increasing SV and CO, although these benefits are likely mediated indirectly by decreasing SVR, rather than directly acting on cardiac metabolism²⁸¹. Moreover, ketosis is not always seen with SGLT2 inhibition, and SGLT2i did not improve PCr/ATP ratios in patients with HFrEF (although the baseline PCr/ATP ratios were already high in this study)³²⁹. GLP1RAs have recently shown dramatic improvements in weight loss in subjects with obesity. These agents have also shown benefit in HFpEF³³⁰, but in HFrEF have shown a trend to more adverse events^331–333. To what extent these effects stem from weight loss per se versus direct effects on cardiac metabolism is not resolved. Treatment of subjects with HFrEF with GLP-1 agonist does not increase MGU, despite improvements in systemic IR³³⁴. In sum, these new metabolic agents have potent effects on cardiac outcomes, but whether they do so via affecting cardiac metabolism is not clear.

Summary and future opportunities

Despite decades of research, our understanding of human cardiac metabolism remains limited. Extensive research using animal models has greatly advanced our understanding of cardiac metabolism, including deep insight into metabolic pathways, fuel use, and metabolic signaling. But extrapolation from animals models to humans is not always supported by data from human studies, especially in the context of heart failure pathogenesis (Figure 4). For example:

Figure 4:

Comparing pre-clinical and human alterations of fuel use and mitochondrial function in heart failure. See Lopaschuk et al. 2021 for pre-clinical details. G6P: glucose-6-phosphate. PDH: pyruvate dehydrogenase. FAO: fatty acid oxidation. TCA: tricarboxylic acid. Image created with Biorender.

• The notion that energetic deficits contribute to contractile dysfunction and heart failure, well-established in model organisms, remains unproven in humans. PCr/ATP studies provide the most robust evidence for cardiac metabolic insufficiency in human HF, but the reductions are not unique to HF, and likely reflect loss of creatine as much as reductions in ATP/ADP ratios; new methods to interrogate cardiac energetics are needed.

• Mitochondrial dysfunction underlies cardiac dysfunction in several pre-clinical models, but the data on mitochondrial function in human heart failure are remarkably conflicting. Many studies have reported some mitochondrial defects (though not always the same ones), while others have found no changes or even compensatory increases in mitochondrial activity in human HF. Standardized approaches, methodologies to assess mitochondrial function in vivo, and comprehensive evaluations are needed.

• Most pre-clinical models of HF have revealed a metabolic switch from FAO to glycolysis, usually ascribed to defects in FAO, and usually thought to be adaptive. Reductions in FA consumption and increases in carbohydrate use are also seen in human HF, but in general to a lesser extent than in preclinical models. Moreover, evidence for a FAO defect in human HF is limited, and efforts to suppress FAO have not benefited patients with HF. In fact, tissue studies suggest that FA import may be limiting in HF, and future studies aimed to boost instead of inhibiting FAU may prove fruitful.

Several additional important questions remain unaddressed. For example, the great majority of cardiac metabolic studies have been performed with fasting subjects, sometimes with additional glucose or with hyper-insulinemic clamps; little is known of human cardiac fuel use during feeding. Additionally, very little is known of cardiac metabolism in patients with HFpEF, despite the rapidly rising prevalence of this syndrome. In general, a major impediment to progress is the limited number of modalities available to interrogate human cardiac metabolism, especially non-invasively. For example, as noted above, new methods to examine cardiac energetics are needed to define more clearly the relationship between heart failure and energetic dysfunction. The development of novel imaging modalities, e.g. MRS with hyperpolarized ¹³C-labeled substrates, is likely to be impactful. The increasing use of extracorporeal normothermic perfusion systems for transport of hearts prior to transplantation^335,336 also affords new opportunities to quantify metabolism in metabolically active (but unloaded) ex situ human hearts.

Despite these limitations, cardiac metabolism remains an attractive potential therapeutic target in heart failure. There are several promising approaches currently in pre-clinical or early clinical testing, including supplementation of NAD precursors, systemic activation of BCAA catabolism, stimulating ketogenesis or ketone supplementation, or promoting cardiac carbohydrate use. Furthermore, systemic metabolic interventions with SGLT2i and GLP1RAs are revolutionizing care of patients with HF; understanding if, and how, these agents directly affect cardiac metabolism will also likely offer new insights and therapeutic avenues.

In summary, the notion that aberrant cardiac metabolism contributes to human heart failure remains plausible and appealing as a potential therapeutic target, but conclusive evidence is lacking. Much has been learned in nearly 70 years of investigations in human cardiac metabolism, but much more needs to be uncovered.

Figure 3:

Estimated fuel contribution to the generation of cardiac ATP in fasting individuals, compared to fed and fasted mice. Left: healthy 18-year-old male in 1950’s, BMI 25. Middle: average 60-year-old American in 2020, BMI 30. Right: average American in 2020 with HFrEF. Data collated from Bing et al. 1954, Most et al. 1969, Wisneski et al. 1985, Murashige et al. 2020, Hui et al. 2020. Image created with Biorender.