Targeted Metabolomic Profiling of Dapagliflozin in Heart Failure with Preserved Ejection Fraction: PRESERVED-HF

Abstract

Background:

SGLT2i improve heart failure (HF)-related symptoms and outcomes in HF with preserved ejection fraction (HFpEF), yet underlying mechanisms remain unclear. In HF with reduced EF, dapagliflozin altered ketone and fatty acid metabolites versus placebo, though metabolite signatures of SGLT2i have not been well elucidated in HFpEF.

Objectives:

To assess whether SGLT2i treatment altered systemic metabolic pathways and their relationship to outcomes in HFpEF.

Methods:

Targeted profiling of 64 metabolites was performed from 293 participants in PRESERVED-HF, a 12-week, placebo-controlled trial of dapagliflozin. Linear regression assessed changes in principal components analysis (PCA)-defined metabolite factors with dapagliflozin versus placebo. The relationship between changes in metabolite factors with changes in study endpoints was also assessed.

Results:

The mean age was 70 (11) years, 58% were women, and 29% were Black. There were no significant differences in 12 PCA-derived metabolite factors between treatment arms, including metabolites reflecting ketone, fatty acid, or branched-chain amino acid (BCAA) pathways. Combining treatment arms, changes in BCAAs and branched-chain ketoacids were negatively associated with changes in NT-proBNP, changes in medium/long-chain acylcarnitines were positively associated with changes in NT-proBNP and negatively associated with changes in 6MWT distance, and changes in ketones were negatively associated with changes in weight, without treatment interaction.

Conclusions:

Leveraging targeted metabolomics in placebo-controlled SGLT2i trial of HFpEF, dapagliflozin did not alter systemic metabolic pathways as reflected by circulating metabolites, in contrast with reported effects in HFrEF. Metabolite biomarkers reflecting BCAA, ketone, and fatty acid metabolism were associated with markers of disease severity, suggesting a role for potential novel treatment targets.

CONDENSED ABSTRACT

SGLT2i improve heart failure (HF)-related symptoms and outcomes in HF with preserved ejection fraction (HFpEF), yet underlying mechanisms remain unclear. In HF with reduced EF, dapagliflozin altered ketone and fatty acid metabolites versus placebo, though these relationships have not been studied in HFpEF. Leveraging principal components analysis of 64 metabolites, there were no changes in 12 metabolite clusters between dapagliflozin and placebo, including metabolites reflecting ketone, fatty acid, or branched-chain amino acid pathways. Metabolite biomarkers reflecting BCAA, ketone, and fatty acid metabolism were associated with markers of disease severity, suggesting a role for potential novel treatment targets.

INTRODUCTION

Sodium-glucose cotransporter-2 inhibitors (SGLT2i) are breakthrough therapies for patients with heart failure and preserved ejection fraction (HFpEF). In this high-risk population, several trials have demonstrated that SGLT2i improve HF-related symptoms and physical limitations while robustly reducing the risk of adverse events (1,2). PRESERVED-HF (Dapagliflozin in PRESERVED Ejection Fraction Heart Failure) demonstrated that dapagliflozin increased the Kansas City Cardiomyopathy Questionnaire Clinical Summary Score (KCCQ-CSS) by a mean of 5.8 points over placebo, with dapagliflozin also augmenting submaximal exercise capacity (1). Despite these and other studies to date across the HF spectrum, mechanisms of benefit for SGLT2i remain incompletely understood. Importantly, metabolic and energetic maladaptations are central to the HFpEF syndrome (3,4). Therefore, understanding the effects of SGLT2i on metabolic pathways, and their relationships to clinically meaningful HF outcomes, may help better define the mechanisms of cardiovascular benefit and potentially direct the development of future interventions.

We previously demonstrated the effects of dapagliflozin on metabolic reprogramming in patients with HFrEF from the DEFINE-HF trial (Dapagliflozin Effects on Biomarkers, Symptoms and Functional Status in Patients with HF with Reduced Ejection Fraction) (5). Employing targeted metabolomics, we found that dapagliflozin exerted effects on key metabolic pathways, supporting a role for altered ketone and fatty acid biology with SGLT2i in patients with HFrEF. Leveraging the similarly designed and larger companion trial (PRESERVED-HF), we sought to understand the influence of dapagliflozin on key metabolic pathways and whether these metabolites are associated with HF-related outcomes in HFpEF.

METHODS

Study design

PRESERVED-HF was an investigator-initiated, randomized, double-blind, placebo-controlled, multi-center trial that enrolled 324 patients with HFpEF across 26 sites in the United States (1). Patients were required to have a left ventricular ejection fraction (LVEF) ≥45%, New York Heart Association class II-IV symptoms, elevated natriuretic peptides (N-terminal pro-b-type natriuretic peptide [NT-proBNP] ≥225 pg/mL or BNP ≥75 pg/ml for patients in sinus rhythm, with higher thresholds for those in atrial fibrillation), use of diuretic treatment, and estimated glomerular filtration rate (eGFR) ≥20 mL/min/1.73m². Objective evidence of HFpEF included a requirement for either HF hospitalization or urgent HF visit with intravenous diuretic treatment in the past 12 months, documented elevated filling pressures on cardiac catheterization, or echocardiographic evidence of structural heart abnormalities (1). Participants were randomized to treatment with dapagliflozin 10 mg daily or matching placebo for 12 weeks.

The primary outcome was change in HF-disease-specific health status as assessed by KCCQ-CSS (which combines symptoms and physical limitations). The key secondary outcome was the change in 6-minute walk test (6MWT) distance, and other notable secondary outcomes included changes in other KCCQ domains, NT-proBNP and BNP levels, hemoglobin A1c, and weight at 12 weeks.

Institutional review boards approved the PRESERVED-HF study for all sites, and all participants provided informed consent. The metabolomics substudy was approved by the institutional review boards at Saint Luke’s Mid America Heart Institute and Duke University. For the present analyses, we included all participants with available baseline and follow-up blood samples. Of the original 324 participants randomized in PRESERVED-HF, 293 (90%) had both baseline and follow-up samples (288 participants with baseline and 12-week visit samples, while 5 participants did not have a 12-week sample and thus baseline and the premature treatment discontinuation visits were used).

Metabolomic profiling

Quantitative metabolomic profiling of 66 metabolites (15 amino acids, 45 acylcarnitines, 3 branched-chain ketoacids (BCKAs), total ketones, ß-hydroxybutyrate, and non-esterified fatty acids) was performed in thawed, previously frozen fasting plasma samples using methods previously described (Supplementary Methods, Table S1) (5,6). Metabolites with >25% of samples below lower limits of quantification were not further analyzed (1 metabolite, C5-DC acylcarnitine) (6). Quality control measures revealed that aspartate/asparagine (Asx) demonstrated run order differences and was thus not further analyzed. Therefore, 64/66 metabolites were included in these analyses.

Statistical analysis

Baseline characteristics of the study cohort are described using means±standard deviation, medians and interquartile ranges, or percentages as appropriate. Welch’s two sample t-tests (for normally distributed variables) or the Wilcoxon rank sum tests (for non-normally distributed variables) were employed for continuous variables. Chi-squared tests or Fisher’s exact tests, as appropriate, were used for categorical variables. To reduce dimensionality of the metabolomics data and identify shared metabolic pathways, unsupervised machine learning was performed using principal components analysis (PCA) with varimax rotation on baseline metabolite levels (5). The PCA algorithm produces orthogonal, uncorrelated factors (i.e. clusters of metabolites) from correlated data based on the variance observed in the dataset. These factors are weighted linear combinations of levels of correlated metabolites; factors with an eigenvalue > 1 (Kaiser criterion) were retained for analysis. PCA factor weights from baseline values were then projected onto follow-up sample values to create follow-up PCA metabolite factor levels. Metabolites with absolute value of factor load > 0.4 were used to describe each factor given these have the highest importance within each factor (Table S2). Of note, because each PCA factor is composed of a different weighted sum of all metabolites, it is possible that more than one factor is annotated as the same biological pathway based on individual metabolites with the greatest variable importance within that factor.

To identify metabolite factor levels that changed differentially between dapagliflozin and placebo, a linear model of change in factor between baseline and follow-up was regressed on treatment group and baseline metabolite factor level, and then additionally adjusted for age, self-reported race/ethnicity, sex, eGFR, and type 2 diabetes mellitus (concordant with previous analysis) (5). We assessed for interactions by body mass index (BMI) and hemoglobin A1c, modeled as continuous variables, with treatment.

Next, we analyzed the relationship between baseline factor levels and both baseline and change in clinical endpoints (log-transformed NT-proBNP, KCCQ-OSS, KCCQ-CSS, 6MWT distance, and weight) using multivariable linear regression adjusted for age, self-reported race/ethnicity, sex, eGFR, type 2 diabetes mellitus, and corresponding baseline values of the respective endpoint. Finally, to understand the relationship between change in factors and change in endpoints, we performed linear regression adjusting for baseline factor, baseline endpoint, and the same covariates as above. To assess for differences in these relationships by treatment randomization, additional models inclusive of a (treatment) × (change in metabolite factor) interaction term were constructed.

For each of these analyses, the Benjamini-Hochberg false discovery rate (FDR) was used to address multiple comparisons at the level of number of metabolite factors analyzed with an FDR threshold <0.10 defining statistical significance (5). In sensitivity analyses of individual metabolites heavily loaded on each significant factor, tested using analogous models, a 2-tailed α of 0.05 was used to determine statistical significance. Analyses were performed using R Studio v4.1.2.

RESULTS

Baseline characteristics

Baseline characteristics of the participants in the parent PRESERVED-HF trial (N=324) and PRESERVED-HF biomarker subset (N=293) are displayed in Table S3. Baseline characteristics of the substudy participant cohort were comparable to the full parent trial participant cohort. Among the 293 participants included in the present study (Table 1), 148 were randomized to dapagliflozin and 145 to placebo. The mean age was 70±11 years, 58% were women, 68% were White, 29% were Black, 3% were non-White/Black, 56% had type 2 diabetes mellitus, and 55% had been previously hospitalized for HF. The mean LVEF was 59±8% and systolic blood pressure was 136±19 mmHg. The median (25^th, 75^th percentile) body mass index was significantly elevated (34.8 [30.4, 41.1] kg/m²), and despite this, the median [25^th, 75^th percentile] NT-proBNP levels were also significantly elevated (681 [356, 1,299] pg/mL), consistent with trial entry criteria. The mean eGFR was 58±19 ml/min/m², and the mean HbA1c was 6.6±1.5%. Baseline characteristics were balanced between randomized groups, aside from a lower percentage of participants taking diuretics at baseline in the placebo arm (93% vs. 83%, p=0.006), similar to the parent trial (1). Baseline fasting levels of the 64 measured metabolites were comparable between randomized groups (Table S4).

Table 1.

Baseline characteristics of patients in the biomarker subset by treatment groups

Characteristic	Dapagliflozin, N = 148	Placebo, N = 145	p-value
Demographics
Age, y	69.5 (10.8)	69.9 (10.4)	0.70
Female	87 (58.8%)	82 (56.6%)	0.70
Self-reported race			0.60
Black	43 (29.1%)	41 (28.3%)
Other	3 (2.0%)	6 (4.1%)
White	102 (68.9%)	98 (67.6%)
Medical history
Duration of heart failure, y	4.39 (4.54)	4.94 (6.19)	0.38
Previous hospitalization for heart failure	87 (58.8%)	75 (51.7%)	0.22
Time since last hospitalization for heart failure, y	1.47 (2.02)	1.85 (3.34)	0.40
Ejection fraction, %	60 (7)	59 (8)	0.56
Type 2 diabetes mellitus	81 (54.7%)	83 (57.2%)	0.67
Atrial fibrillation	77 (52.0%)	81 (55.9%)	0.51
ICD	7 (4.7%)	9 (6.2%)	0.58
Baseline HF/CV medications
ACEI/ARB	89 (60.1%)	89 (61.4%)	0.83
ARNI	2 (1.4%)	3 (2.1%)	0.68
ß-blocker	112 (75.7%)	105 (72.4%)	0.52
Hydralazine	23 (15.5%)	15 (10.3%)	0.19
Long-acting nitrates	33 (22.3%)	25 (17.2%)	0.28
MRA	48 (32.4%)	62 (42.8%)	0.07
Loop diuretics	138 (93.2%)	120 (82.8%)	0.006
Digoxin	3 (2.0%)	7 (4.8%)	0.21
Lipid-lowering agents	121 (81.8%)	115 (79.3%)	0.60
Anticoagulant agents	66 (44.6%)	76 (52.4%)	0.18
Vital signs
Body mass index (median Q1, Q3)	35.5 (30.8, 41.4)	34.5 (30.0, 40.1)	0.42
Heart rate	70 (12)	69 (12)	0.25
Systolic blood pressure	138 (19)	134 (19)	0.11
Baseline laboratory studies
NT-proBNP, pg/mL (median Q1, Q3)	648 (373, 1,210)	722 (347, 1,460)	0.46
BNP, pg/mL (median Q1, Q3)	137 (84, 219)	154 (91, 258)	0.18
eGFR, mL/min/1.73m^2	58 (18)	58 (21)	0.85
Urine albumin/creatinine ratio, mg/g (median Q1, Q3)	19 (8, 108)	27 (9, 96)	0.94
Hemoglobin A1c, %	6.7 (1.6)	6.5 (1.4)	0.45
Hemoglobin, g/dL	12.7 (1.8)	13.4 (10.7)	0.46
Functional measures
NYHA Class			0.21
Class II	77 (52.0%)	83 (57.2%)
Class III	68 (46.0%)	62 (42.8%)
KCCQ-OSS	63 (21)	63 (21)	0.98
KCCQ-CSS	63 (20)	63 (20)	0.96
6-minute walk distance, meters (median Q1, Q3)	244 (169, 329)	245 (165, 317)	0.82

Data presented as mean (standard deviation), median (25^th, 75^th percentile), or n (%).

ACE-I, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker; BNP, B-type natriuretic peptide; CSS, Clinical Summary Score; eGFR, estimated glomerular filtration rate; ICD, implantable cardioverter-defibrillator; KCCQ, Kansas City Cardiomyopathy Questionnaire; OSS, Overall Summary Score; MRA, mineralocorticoid antagonist; NT-proBNP, N-terminal pro-B-type natriuretic peptide.

Effect of dapagliflozin compared with placebo on metabolite clusters and metabolites

PCA resulted in 12 factors reflecting clusters of metabolites in shared biologic pathways (Table S2). For example, PCA factor 3 was largely composed of branched-chain amino acids (BCAA) and BCKAs, while factor 7 reflected ketone-related metabolites. The treatment effect of dapagliflozin on PCA metabolite factors compared with placebo is shown in Table 2. After adjusting for age, race/ethnicity, sex, eGFR, type 2 diabetes mellitus, and baseline PCA factor level, no metabolite factor changed significantly with dapagliflozin therapy as compared with placebo after accounting for multiple comparisons. Two PCA metabolite factors showed nominally significant changes with dapagliflozin therapy compared with placebo: short-chain dicarboxylacylcarnitines (factor 2) and medium-chain acylcarnitines (factor 5) (nominal p-values=0.04 and 0.01, respectively). In particular, unlike prior results in HFrEF (5), dapagliflozin did not differentially change levels of PCA metabolite factors characterized by ketone-related metabolites.

Table 2.

Effect of dapagliflozin compared with placebo on PCA-derived metabolite factors

Factors	Annotation	Nominal P value	FDR corrected P value
1	Medium- and long-chain acylcarnitines and dicarboxyl acylcarnitines	0.74	0.88
2	Short-chain dicarboxyl acylcarnitines	0.04	0.27
3	Branched-chain amino acids and branched-chain ketoacids	0.33	0.66
4	Miscellaneous amino acids	0.09	0.35
5	Medium-chain acylcarnitines	0.01	0.14
6	Long-chain acylcarnitines	0.86	0.92
7	Ketone-related metabolites	0.57	0.76
8	Miscellaneous amino acids	0.56	0.76
9	Long-chain dicarboxyl acylcarnitines	0.52	0.76
10	Medium-chain acylcarnitines	0.17	0.46
11	Miscellaneous acylcarnitines	0.92	0.92
12	Aromatic amino acids	0.19	0.46

Model includes randomized treatment group, baseline factor level, age, race/ethnicity, sex, estimated glomerular filtration rate, and type 2 diabetes mellitus.

FDR, false discovery rate; PCA, principal components analysis.

To facilitate comparison with our previous findings in HFrEF (5), we detailed several mean metabolite levels during follow-up (Tables S5–6). Ketone levels tended to increase in both groups from baseline to follow-up. Notably, ketosis (follow-up ß-hydroxybutyrate levels >500 μM) was achieved in 9/148 in the dapagliflozin arm vs. 2/145 in the placebo arm (p=0.06). Among those achieving ketosis, 7/9 and 1/2 had type II diabetes mellitus, respectively.

Given differences observed when compared with findings from DEFINE-HF, and the greater burden of cardiometabolic disease in HFpEF, we also assessed for interactions of treatment effects with BMI and hemoglobin A1c on factors representing the ketone/fatty acid pathways. There were no significant treatment interactions by hemoglobin A1c or BMI on these pathways (interaction p-value>0.10 for all comparisons).

Association of baseline metabolite clusters with baseline HF-related endpoints

The relationships between baseline metabolite factor levels with baseline NT-proBNP level, KCCQ-CSS, KCCQ-OSS, weight, and 6MWT distance are displayed in Tables S7–9, combining both treatment arms. After multivariable adjustment, baseline levels of factor 3 (heavily loaded with BCAAs and BCKAs) and factor 9 (long-chain dicarboxyl acylcarnitines) were associated with baseline NT-proBNP (FDR-corrected p-value=0.01 and <0.01, respectively), while baseline levels of factor 9 were associated with baseline weight (FDR-corrected p-value=0.07). Specifically, higher baseline levels of BCAAs/BCKAs were associated with lower NT-proBNP, while higher levels of long-chain dicarboxyl acylcarnitines were associated with higher NT-proBNP and lower weight.

Association of baseline metabolite clusters with changes in HF-related endpoints

The relationships between baseline metabolite factor levels with changes in NT-proBNP level, KCCQ-CSS, KCCQ-OSS, weight, and 6MWT distance are displayed in Table 3 and Figure 1, combining both treatment arms. After multivariable adjustment, baseline levels of factor 3 were associated with change in KCCQ-CSS (FDR-corrected p-value =0.01) and KCCQ-OSS (FDR-corrected p-value <0.01), while baseline levels of factor 5 (heavily loaded on medium-chain acylcarnitines) were associated with 6MWT distance (FDR-corrected p-value =0.096). Specifically, higher baseline levels of individual BCAAs and BCKAs at baseline predicted greater increase (i.e. improvement) in KCCQ scores, and higher levels of individual medium acylcarnitine metabolites predicted greater improvement in 6MWT distance in the overall study population (Tables S10–12).

Table 3.

Association of baseline metabolite factor levels with changes in clinical endpoints

Factor	Annotation	NT-proBNP FDR corrected P value	KCCQ-OSS FDR corrected P value	KCCQ-CSS FDR corrected P value	Weight FDR corrected P value	6-Minute Walk Test Distance FDR corrected P value
1	Medium- and long-chain acylcarnitines and dicarboxyl acylcarnitines	0.95	0.59	0.81	0.56	0.89
2	Short-chain dicarboxyl acylcarnitines	0.74	0.53	0.60	0.56	0.82
3	Branched-chain amino acids and branched-chain ketoacids	0.66	<0.01	0.01	0.56	0.89
4	Miscellaneous amino acids	0.66	0.53	0.60	0.56	0.89
5	Medium-chain acylcarnitines	0.73	0.71	0.60	0.88	0.096
6	Long-chain acylcarnitines	0.73	0.84	0.94	0.78	0.89
7	Ketone-related metabolites	0.73	0.95	0.60	0.80	0.89
8	Miscellaneous amino acids	0.73	0.71	0.60	0.94	0.50
9	Long-chain dicarboxyl acylcarnitines	0.66	0.53	0.63	0.80	0.89
10	Medium-chain acylcarnitines	0.73	0.53	0.60	0.56	0.86
11	Miscellaneous acylcarnitines	0.82	0.53	0.60	0.56	0.40
12	Aromatic amino acids	0.66	0.53	0.60	0.94	0.89

Model adjusted for baseline value of the respective outcome, age, race/ethnicity, sex, eGFR, and type 2 diabetes mellitus.

FDR corrected p-values <0.10 were considered statistically significant.

eGFR, estimated glomerular filtration rate; FDR, false discovery rate; KCCQ, Kansas City Cardiomyopathy Questionnaire; CSS, clinical summary score; OSS, overall summary score; NT-proBNP, N-terminal pro B-type natriuretic peptide

Figure 1: Heatmap of Correlation Between Baseline Metabolite Levels and Changes in Trial Endpoints.

Heatmap cluster correlation matrix is depicted for the relationship between changes in several trial endpoints (KCCQ-OSS, KCCQ-CSS, NT-proBNP, weight, and 6-minute walk test distance) with baseline metabolite levels. See Supplementary Table 1 for metabolite abbreviations. CSS, clinical summary score; KCCQ, Kansas City Cardiomyopathy Questionnaire; NT-proBNP, N-terminal pro-B-type natriuretic peptide; OSS, overall summary score.

Association of changes in metabolite clusters with changes in HF-related endpoints

The relationships between changes in metabolite clusters with changes in HF-related endpoints are displayed in Table 4 and Figure 2. Changes in NT-proBNP were significantly associated with Factors 1, 3, 8; changes in weight were associated with Factor 7; and changes in 6MWT distance were associated with Factor 1. Specifically, changes in medium and long-chain acylcarnitines (and their dicarboxylated versions) were positively associated with change in NT-proBNP, while changes in valine and its cognate alpha-ketoacid (KIV) were negatively associated with changes in NT-proBNP (Table S13). Additionally, changes in ketone-related metabolites were negatively associated with change in weight (Table S14), while changes in individual medium/long-chain acylcarnitines (and their dicarboxylated versions) were negatively associated with change in 6MWT distance (Table S15). There were no significant interactions between treatment group and changes in metabolite factor with these outcomes (p-interaction >0.10 for all comparisons after FDR adjustment). For descriptive purposes, these relationships between changes in factors with changes in outcomes are stratified by treatment arm in Tables S16–17.

Table 4.

Association of changes in metabolite factor levels with changes in clinical endpoints

Factor	Annotation	NT-proBNP FDR corrected P value	KCCQ-OSS FDR corrected P value	KCCQ-CSS FDR corrected P value	Weight FDR corrected P value	6-Minute Walk Test Distance FDR corrected P value
1	Medium- and long-chain acylcarnitines and dicarboxyl acylcarnitines	<0.01	0.11	0.28	0.55	0.07
2	Short-chain dicarboxyl acylcarnitines	1.00	0.49	0.80	0.55	0.31
3	Branched-chain amino acids and branched-chain ketoacids	0.09	0.33	0.80	0.78	0.91
4	Miscellaneous amino acids	0.33	0.11	0.28	0.36	0.95
5	Medium-chain acylcarnitines	0.44	0.33	0.80	0.21	0.91
6	Long-chain acylcarnitines	1.00	0.33	0.86	0.15	0.91
7	Ketone-related metabolites	0.55	0.84	0.80	0.01	0.65
8	Miscellaneous amino acids	0.08	0.33	0.45	0.26	0.91
9	Long-chain dicarboxyl acylcarnitines	1.00	0.94	0.80	0.15	0.91
10	Medium-chain acylcarnitines	0.86	0.56	0.80	0.15	0.92
11	Miscellaneous acylcarnitines	0.44	0.56	0.80	0.21	0.91
12	Aromatic amino acids	0.43	0.33	0.80	0.16	0.65

Model adjusted for baseline factor level, baseline value of the respective outcome, age, race/ethnicity, sex, eGFR, and type 2 diabetes mellitus.

FDR corrected p-values <0.10 were considered statistically significant.

eGFR, estimated glomerular filtration rate; FDR, false discovery rate; KCCQ, Kansas City Cardiomyopathy Questionnaire; CSS, clinical summary score; OSS, overall summary score; NT-proBNP, N-terminal pro B-type natriuretic peptide.

Figure 2: Heatmap of Correlation Between Changes in Metabolites and Changes in Trial Endpoints.

Heatmap cluster correlation matrix is depicted for the relationship between changes in several trial endpoints (KCCQ-OSS, KCCQ-CSS, NT-proBNP, weight, and 6-minute walk test distance) with changes in metabolites from baseline to follow-up. See Supplementary Table 1 for metabolite abbreviations. CSS, clinical summary score; KCCQ, Kansas City Cardiomyopathy Questionnaire; NT-proBNP, N-terminal pro-B-type natriuretic peptide; OSS, overall summary score.

DISCUSSION

Leveraging targeted, quantitative metabolomic profiling from nearly 300 participants enrolled in a randomized, placebo-controlled trial of dapagliflozin in HFpEF, we highlight several important findings regarding metabolic pathways using circulating biomarkers (Central Illustration). First, contrary to prior findings observed in patients with HFrEF (5), dapagliflozin therapy did not alter several key metabolic pathways assayed in patients with HFpEF. Specifically, and in contrast to observed effects in other disease states (5,7,8), SGLT2 inhibition did not augment ketone-related pathways, nor did it have appreciable effects on metabolites reflecting BCAA or fatty acid metabolism. Interestingly, while effects on ketone levels were not significantly different between treatment arms, there was a trend toward greater ketotic response as a categorical change with dapagliflozin. Second, higher baseline levels of metabolites reflecting BCAA metabolism (BCAA and BCKA, byproducts of BCAA mitochondrial catabolism) and mitochondrial fatty acid oxidation (medium and long-chain acylcarnitines) predicted improvement in measures of quality of life and exercise capacity in the entire cohort. Finally, increases in BCAA/BCKA and ketone-related metabolites and decreases in medium/long-chain acylcarnitines over 12 weeks were associated with improved HF-related outcomes in the overall cohort. Collectively, these findings demonstrate that changes in key fuel substrate-related metabolic pathways central to healthy and unhealthy myocardial and skeletal muscle function are not mechanistically related to the known treatment benefit of dapagliflozin, though we identify biomarkers associated with clinical features in HFpEF highlighting keys roles in BCAA, ketone, and fatty acid metabolism.

Central Illustration: Targeted Metabolites, Cardiometabolic Endpoints, and Influence of Dapagliflozin in HFpEF.

AC, acylcarnitines; BCAA, branched-chain amino acids; HFpEF, heart failure with preserved ejection fraction; CSS, clinical summary score; KCCQ, Kansas City Cardiomyopathy Questionnaire; NT-proBNP, N-terminal pro-B-type natriuretic peptide; OSS, overall summary score; 6MWD, 6-minute walk distance; PRESERVED-HF, Dapagliflozin in PRESERVED Ejection Fraction Heart Failure. Created with BioRender.com.

Effects of SGLT2i on metabolism in HFpEF

A better understanding of how changes in metabolism are related to the therapeutic benefits of SGLT2i in HF is important, both because it can better direct future research and identify potential targets for future, even more efficacious therapies (5,9,10). As one potential avenue of benefit, SGLT2i facilitate shifts in cardiac metabolism toward ketone, fatty acid, and BCAA utilization in HFrEF (10). The “thrifty substrate” hypothesis further speculates that the mild hyperketonemia during SGLT2i therapy may underpin the HF benefits (5,11). These transitions in fuel may be important, since changes in metabolism (both at the cardiac and systemic levels) are fundamental to HF pathobiology. Indeed, the HFrEF is characterized by increasing ketone reliance (when available) in the context of decreasing fatty acid, BCAA, and glucose oxidation (12,13). On the other hand, HFpEF myocardium resembles a state of “starvation in the midst of plenty”, with substantial deficits in cardiac substrate uptake and oxidation (3). Given our limited understanding of mechanisms behind the benefits of SGLT2i in HFpEF and the relevance of systemic metabolism to the HFpEF syndrome (14), we leveraged metabolomic profiling to help fill these gaps.

Surprisingly, we did not find that dapagliflozin changed levels of any metabolite factors including those representing ketones, which are signature effects of SGLT2i in diabetes and HFrEF (5,7,8). Using a similarly designed trial in patients with HFrEF (DEFINE-HF), we recently showed that dapagliflozin increased ketone-related metabolites and several short/medium-chain acylcarnitines, but these signatures were not evident in HFpEF (Figure 3). Indeed, ketone levels increased in the dapagliflozin arm (to a similar extent as in DEFINE-HF) (5), though this was nearly matched by the placebo arm (underscoring the importance of the placebo arm in our analysis). While average ketone levels were similar with dapagliflozin and placebo, a small proportion of patients experienced an exaggerated ketotic response, as shown by the slightly greater proportion of patients achieving ketosis (defined as beta-hydroxybutyrate>500μM) in the dapagliflozin versus placebo arm. A smaller trial that randomized HFpEF patients without significant coronary artery disease to empagliflozin vs. placebo also found no significant difference in several metabolites studied, though these metabolites largely reflected different processes than studied here (Krebs cycle and glycolytic intermediates, in addition to ketones) and no differences in myocardial energetics were simultaneously observed (15).

Figure 3: Differences in Metabolites Levels with Dapagliflozin versus Placebo in DEFINE-HF and PRESERVED-HF.

The placebo-adjusted, 12-week differences in metabolite levels that were differentially affected by dapagliflozin treatment in DEFINE-HF are shown in comparison with PRESERVED-HF. Depicted values include the mean placebo-adjusted differences in metabolite levels with the corresponding 95% confidence intervals for each study. DEFINE-HF, Dapagliflozin Effects on Biomarkers, Symptoms and Functional Status in Patients with HF with Reduced Ejection Fraction; PRESERVED-HF, Dapagliflozin in PRESERVED Ejection Fraction Heart Failure. Created with BioRender.com.

It is possible that the greater degree of cardiometabolic disease in patients with HFpEF who were enrolled in PRESERVED-HF (compared with those with HFrEF enrolled in DEFINE-HF) may have blunted the metabolic treatment effects. For example, since peripheral muscle amount generally correlates with body mass index, and muscle is important for clearance of ketones, the ketogenic effect may have been blunted (12,16). Additionally, higher glycogen stores in obese individuals, who are more frequently represented in HFpEF populations (17), might have buffered the increase in ketones. However, BMI did not modify the ketotic response. Another reason for the overall lack of metabolism-related treatment effects may relate to a limited power to detect important differences, though PRESERVED-HF is approximately 25% larger than DEFINE-HF (where significant treatment effects were observed). Finally, differing degrees of central versus peripheral muscular dysfunction may result in differential effects of SGLT2i on systemically measured metabolites (16,18). Overall, it appears that systemic ketosis is not a key mechanism of benefit with SGLT2i in HFpEF. Further, the lack of systematic difference in ketone levels also underscores the fact that compliance with SGLT2i cannot be confirmed by assays of ketone levels. An important clinical contribution of our results is reassurance to clinicians that supraphysiologic ketosis (as seen in diabetic ketoacidosis) is very rarely observed in trials of HFpEF (2). Finally, these findings may be relevant when counseling patients who may be embarking on, or currently undergoing, a low carbohydrate diet (19).

The overall lack of change in circulating metabolite clusters with treatment must be interpreted in the context of systemic versus local metabolic regulation. There may be uncoupling of systemic metabolites compared with cardiac metabolites, as recently demonstrated in HFpEF where metabolomics were assayed from the peripheral blood and myocardial tissue (3). This does not argue against the importance of circulating metabolites in HFpEF, which have prognostic utility (6), but simply underscores that circulating levels may not reflect specific tissue bed activity. Therefore, SGLT2i may still have important metabolic effects in the myocardium (as well as other organ beds, including adipose tissue and skeletal muscle), though their relationship to cardiac energetics (which does not appear to be altered with SGLT2i) would need further clarification (15).

Utility of targeted metabolomics to provide insight into HFpEF pathobiology

In contrast with HFrEF, less has been published on the diagnostic and prognostic utility of peripheral metabolomics in HFpEF. Smaller studies generally highlight a HFpEF metabolic signature (distinguished from controls) of elevated levels of acylcarnitines, which reflect fatty acid metabolism, and increases in BCAAs (3,14,20). Generally speaking, the accumulation of acylcarnitines is thought to reflect mitochondrial dysfunction, though also may directly promote cellular stress and mitochondrial inflammation, and interfere with insulin sensitivity (6,21). However, within HFpEF, the association of metabolites with specific HFpEF phenotypes is less well understood. One previous study identified that a short-chain dicarboxylated acylcarnitine was associated with a worse composite clinical rank score, which may reflect dysregulated fatty acid oxidation in the endoplasmic reticulum and peroxisomes (6).

In that context, we identified several novel relationships between circulating metabolites, both at baseline and during follow-up, and HF-related outcomes. For example, higher BCAA/BCKA and medium/long-chain acylcarnitines at baseline predicted greater improvement in KCCQ scores and 6MWT distance, respectively. Similarly, the changes in several metabolites were prognostically relevant. Specifically, decreasing medium/long-chain acylcarnitines/dicarboxylated acylcarnitines as well as increasing BCAA/BCKA and ketone-related metabolites were associated with more favorable cardiometabolic and HF-specific features. The relationship between higher levels of acylcarnitines and worse HF features has been noted in previous analyses, including in HFrEF (from DEFINE-HF) and also in another trial in HFpEF (5,6). In HFrEF populations, plasma BCAAs positively correlate with skeletal muscle mass (22). Because sarcopenia in HF populations correlates with worse clinical outcomes (23), higher BCAAs in HFpEF may reflect more favorable muscle composition (less sarcopenia) and thus improved HF features. Moreover, others have shown that higher BCAA predict improved diastolic function (24). Whether body mass composition in HFpEF accounts for the association of BCAAs with improved outcomes is an important question for future studies. Of note, there were no significant interactions by treatment arm in our study, confirming the utility of using these metabolites for prognostic purposes in the modern treatment era.

Limitations and Strengths

Our findings should be interpreted in the context of several limitations. First, the lack of change in fasting circulating metabolites with SGLT2i does not rule out localized effects on metabolism, nor does it comment on substrate flux or non-fasted metabolism. Second, we assayed several, but not all, metabolic pathways that may be affected by SGLT2i. Third, we assayed changes in metabolites between baseline and up to 12 weeks, which may not capture relevant early or late treatment effects. Strengths of our study include the relatively large sample size for metabolomic analysis, placebo control, characterization of a significant number of relevant metabolites (including BCKA, which are less frequently characterized from peripheral blood in human HF), and inclusion of several relevant HF endpoints.

Conclusions

In summary, using an accurate, quantitative, targeted metabolomics platform for assaying a broad set of metabolites in a placebo-controlled trial of SGLT2i in HFpEF, we did not find that dapagliflozin differentially affected circulating metabolic clusters compared with placebo, including those reflecting ketone, fatty acid, and BCAA metabolism. However, metabolites reflecting these pathways were significantly predictive of changes in HF-related health over 12 weeks. These findings advance our current understanding of the mechanisms of benefit of SGLT2i in HFpEF by demonstrating that changes in relevant systemic metabolic pathways may not underpin the treatment effects. Further, these analyses identify novel metabolites that can serve as biomarkers and potential treatment targets.

CLINICAL PERSPECTIVES.

Competency in Medical Knowledge:

SGLT2 inhibitors are standard of care in HFpEF, but underlying mechanisms remain unclear. Our study shows that changes in systemic metabolism reflected by circulating metabolites do not appear to underpin the benefit. These results may reassure clinicians that supraphysiologic ketosis does not generally occur with SGLT2i.

Translational Outlook 1:

Further research is needed to define why HFpEF is not associated with changes in systemic metabolism that are observed in HFrEF with SGLT2 inhibitors, and whether greater burden of cardiometabolic disease may account for this lack of effect.

Translational Outlook 2:

Whether altering pathways related to ketone, amino acid, or fatty acid metabolism provides benefit in HFpEF warrants further study.

ACRONYMS AND ABBREVIATIONS

6MWT

6-minute walk test

BCAA

branched-chain amino acids

BCKA

branched-chain keto acids

DEFINE-HF

Dapagliflozin Effects on Biomarkers, Symptoms and Functional Status in Patients with HF with Reduced Ejection Fraction

eGFR

estimated glomerular filtration rate

HFpEF

heart failure with preserved ejection fraction

HFrEF

heart failure with reduced ejection fraction

KCCQ-CSS

Kansas City Cardiomyopathy Questionnaire Clinical Summary Score

KCCQ-OSS

Kansas City Cardiomyopathy Questionnaire Overall Summary Score

PCA

principal components analysis

PRESERVED-HF

Dapagliflozin in PRESERVED Ejection Fraction Heart Failure

NT-proBNP

N-terminal pro-B-type natriuretic peptide