Metabolite signatures of heart failure, sleep apnoea, their interaction, and outcomes in the community

Abstract

Aims

Sleep apnoea and congestive heart failure (CHF) commonly co‐exist, but their interaction is unclear. Metabolomics may clarify their interaction and relationships to outcome.

Methods and results

We assayed 372 circulating metabolites and lipids in 1919 and 1524 participants of the Framingham Heart Study (FHS) (mean age 54 ± 10 years, 53% women) and Women's Health Initiative (WHI) (mean age 67 ± 7 years), respectively. We used linear and Cox regression to relate plasma concentrations of metabolites and lipids to echocardiographic parameters; CHF and its subtypes heart failure with reduced ejection fraction (HFrEF) and heart failure with preserved ejection fraction (HFpEF); and sleep indices. Adenine dinucleotide phosphate (ADP) associated with left ventricular (LV) fractional shortening; phosphocreatine with LV wall thickness; lysosomal storage molecule sphingomyelin 18:2 with LV mass; and nicotine metabolite cotinine with time spent with an oxygen saturation less than 90% (β = 2.3 min, P = 2.3 × 10⁻⁵). Pro‐hypertrophic metabolite hydroxyglutarate partly mediated the association between LV wall thickness and HFpEF. Central sleep apnoea was significantly associated with HFpEF (P = 0.03) but not HFrEF (P = 0.5). There were three significant metabolite canonical variates, one of which conferred protection from cardiovascular death [hazard ratio = 0.3 (0.11, 0.81), P = 0.02].

Conclusions

Energetic metabolites were associated with cardiac function; energy‐ and lipid‐storage metabolites with LV wall thickness and mass; plasma levels of nicotine metabolite cotinine were associated with increased time spent with a sleep oxygen saturation less than 90%, a clinically significant marker of outcome, indicating a significant hazard for smokers who have sleep apnoea.

Introduction

Nutritional deficiency (e.g. thiamine deficiency), oversupply (e.g. obesity), states of metabolic inflexibility (e.g. insulin resistance and type 2 diabetes), and altered metabolic regulation (e.g. thyroid disorders) all confer structural and functional cardiac phenotypes, underscoring the importance of metabolism in maintaining cardiac performance and integrity.

Heart failure with preserved ejection fraction (HFpEF) has become the most common form of heart failure, yet it remains poorly understood compared with its easily recognized counterpart, HFrEF (reduced ejection fraction). It most commonly manifests in the presence of comorbidities such as the metabolic syndrome.¹ An underappreciated comorbidity is sleep apnoea, the obstructive form of which is more common in this patient population.² Central sleep apnoea has long been associated with heart failure, although it is not clear if it is related more to HFrEF or HFpEF. While severe sleep apnoea confers an increased mortality risk, the impact of interaction with heart failure remains unclear.³ Furthermore, there are conflicting data on the impact of moderate sleep apnoea on cardiovascular disease (CVD) and mortality.⁴

Metabolomics is emerging as a useful tool to uncover molecular signatures of heart failure and outcome.⁵ Therefore, to help generate insight into the potential interaction of HFpEF and/or HFrEF with sleep apnoea, we explored the relationships between metabolites and cardiac structure, function, heart failure, and sleep apnoea indices in ~2000 individuals in the community. We performed validation analyses of congestive heart failure (CHF) associations in 1524 participants in the Women's Health Initiative (WHI), as women are increasingly recognized to be disproportionately affected by CHF, particularly the HFpEF phenotype.

Methods

Framingham Heart Study

Participants of the Framingham Heart Study (FHS) Offspring cohort were examined.⁶ Of the 3799 participants enrolled, 1919 attended Examination 5 (1991–95; 53% female) and had blood collected and analysed by metabolomics and lipidomics. The corresponding clinical report at Examination 5 was used to obtain blood pressure, body mass index (BMI), age, gender, and so forth. Echocardiography reports from the Offspring cohort were acquired at Examinations 5, 6, and 8. The investigation conforms with the principles outlined in the Declaration of Helsinki.

Women's Health Initiative

The WHI is a long-term national health study funded by the National Heart, Lung, and Blood Institute, or NHLBI. The original WHI study began in the early 1990s and concluded in 2005, continuing after this as Extension Studies. The blood samples for which the metabolomics was measured were obtained from the substudy Metabolomics of Coronary Heart Disease (CHD) in the WHI. The datasets used for the analyses described in this manuscript were obtained from dbGaP, full details in Supporting Information. The investigation conforms with the principles outlined in the Declaration of Helsinki.

Congestive heart failure diagnosis (heart failure with preserved and reduced ejection fraction)

Congestive heart failure reports were acquired from the Framingham database, performed in 2013 and 2016. Participants deemed to have CHF (either hospitalized or not hospitalized) or to meet the Framingham criteria of CHF (two major criteria or one major and two minor) were identified, and the date of CHF onset was identified. CHF diagnosed at the time of Examination 5 (1991–95) were classified as prevalent CHF, while CHF diagnosed from Examination 6 onwards (1995–2016) were classified as incident CHF for the purposes of correlation to metabolites and lipids. All CHF diagnoses were further identified as HFpEF or HFrEF based on ejection fraction measurements obtained from echocardiography performed closest to the date of diagnosis/onset of CHF. CHF diagnosis in conjunction with ejection fraction of <50%, or graded as mild, moderate, or severe systolic dysfunction, was classified as HFrEF, while ejection fraction of ≥50%, or graded as normal or borderline systolic dysfunction, was graded as HFpEF.

In the WHI, CHF heart failure was defined as a constellation of symptoms and physical signs. Only a hospitalization involving new or worsening CHF was a WHI outcome. A comprehensive description of CHF diagnostic methodology is available elsewhere.⁷ Sufficient echocardiographic data were not available in WHI for a study of CHF subtype.

Metabolomics and lipidomics profiling

Metabolomic and lipidomic profiles of the Offspring cohort at Examination 5 were acquired from the Framingham database. Plasma samples were collected in EDTA coated tubes after an overnight fast, spun at 2000 rcf at 4°C to remove red blood cells, and the plasma stored at −80°C. Metabolomics and lipidomics were acquired as previously described,⁸ described in Supporting Information.

Statistical analysis

All 113 metabolites and 104 lipids were regressed against incident CHF, HFpEF, HFrEF, and echocardiographic, and sleep parameters using age, gender, and BMI adjusted regression.

Linear, logistic, and ordinal logistic regression were used for continuous, two-class, and ordinal measures, respectively.

A Cox proportional hazard regression model was used to establish association of sleep parameters and death due to CVD as well as overall death. Models were adjusted for age, gender, and BMI. For any significant metabolite–sleep association, mediation analysis was performed to identify mediation impact of metabolites. Canonical correlation analysis (CCA) was performed to understand the joint relationship between the set of metabolites and sleep parameters, HFpEF, and HFrEF.

Results

Congestive heart failure diagnosis and its subtypes

Of the 1919 FHS participants studied, 175 had definite CHF according to the Framingham criteria: 7 with prevalent CHF (4 HFpEF and 3 HFrEF) and 152 with incident CHF (131 HFpEF and 21 HFrEF). Due to the small number of prevalent CHF, we focused on incident CHF. Baseline characteristics are provided in Table 1. In the WHI, there were 1295 women with metabolomics data at baseline, 269 of whom had CHF (baseline–prevalent). After the baseline metabolomics samples were taken, there were 10 new diagnoses of CHF before samples were taken again for metabolomics (second timepoint). At the second metabolomics sampling, 1295 samples were again available for metabolomics profiling, 279 of whom had CHF (second timepoint–prevalent). After the second metabolomics sampling, there were 132 incident cases of CHF (second timepoint–incident).

Table 1 Baseline characteristics

FHS	Men (907)	Women (1012)
Age	54 (10)	54 (9)
BMI	28.2 (4)	26.6 (5)
Systolic blood pressure (mmHg)	129 (18)	123 (19)
Diastolic blood pressure (mmHg)	77 (10)	73 (10)
Heart rate (b.p.m.)	63 (11)	67 (10)
Currently smoking (N)	165 (18%)	184 (18%)
Use of antihypertensive medications (N)	194 (21%)	151 (15%)
Diabetes (N)	65 (7%)	45 (4%)
Use of statins (N)	80 (9%)	54 (5%)
Fasting blood glucose level (mg/dL)	103 (26)	97 (24)
Total cholesterol (mg/dL)	202 (35)	208 (38)
Triglycerides (mg/dL)	161 (136)	136 (114)
HDL cholesterol (mg/dL)	43 (11)	56 (16)
LDL cholesterol (mg/dL)	129 (32)	125 (34)
Echocardiographic parameters
Left atrial dimension (cm)	4.0 (0.5)	3.5 (0.4)
Left ventricular diastolic dimension (cm)	5.0 (0.4)	4.6 (0.4)
Left ventricular systolic dimension (cm)	3.2 (0.5)	2.8 (0.4)
Fractional shortening (%)	36 (7)	37 (6)
Left ventricular wall thickness (cm)	2.0 (0.2)	1.8 (0.2)
Aortic root dimension (cm)	3.4 (0.3)	3.0 (0.3)
Interventricular septal thickness diastole (cm)	1.0 (0.1)	0.9 (0.1)
Left ventricular posterior wall thickness diastole (cm)	1.0 (0.1)	0.9 (0.1)
FHS–sleep parameters	Men (198)	Women (192)
Obstructive sleep apnoea	121	66
Mild	72 (60%)	42 (64%)
Moderate	34 (28%)	24 (36%)
Severe	15 (12%)	0 (0%)
Central sleep apnoea	92	46
Mild	67 (73%)	36 (78%)
Moderate	18 (20%)	9 (20%)
Severe	7 (8%)	1 (2%)
WHI	Women (1524)
Age	67 (7)
BMI	28.5 (6)
Systolic blood pressure (mmHg)	131 (18)
Diastolic blood pressure (mmHg)	75 (9)
Smoking
Past smoker	602 (39.5%)
Current smoker	153 (10%)
Hypertension
Untreated hypertension	139 (9%)
Treated hypertension	454 (29.7%)
Diabetes
Ever	154 (10%)
Current	115 (7.5%)
Treated	118 (8%)
Congestive heart failure (CHF)
CHF at baseline	269 (17.5%)
New CHF before Timepoint 2	10 (0.006%)
New CHF after Timepoint 2	132 (9%)

Metabolite and lipid associations with prevalent cardiac structure and function in Framingham Heart Study

A total of 1870 individuals with metabolomic data also had echocardiographic data available at Examination 5. A total of 113 metabolites and 104 lipids (Supporting Information, Table S1) were profiled and then associated. After multivariable adjustment, cross-sectional association analysis of prevalent echocardiographic parameters at Exam 5 revealed 104 significant metabolite–echocardiographic parameter associations at nominal significance levels and 20 at false discovery rate (FDR)-adjusted threshold (Table 2). Remarkably, energetic metabolite adenine dinucleotide phosphate (ADP) was positively associated (β = 0.04, P = 1.3 × 10⁻⁴) with fractional shortening, indicating that this circulating energy metabolite can report on cardiac systolic function, an energy-intensive process (Figure 1). As reported previously,⁵ kynurenine was inversely associated with left ventricular (LV) end-diastolic dimension (β = −0.7, P = 8.98 × 10⁻⁵), as were branched chain amino acids and cardiac energy substrates leucine (β = −0.7, P = 8.98 × 10⁻⁵) and valine (β = −1.0, P = 3.1 × 10⁻⁴); glycine was positively associated (β = 0.54, P = 8.98 × 10⁻⁵). LV end-systolic dimension was inversely associated with niacinamide (β = −0.34, P = 2.2 × 10⁻⁴), the precursor of nicotinamide adenine dinucleotide, known to decrease with cardiac hypertrophy and cardiac dilatation.⁹ Inositol was significantly associated with aortic dimension (β = 0.4, P = 4.3 × 10⁻⁴), which may reflect the relationship of inositol and inositol-3-phosphate receptors with hypertension,¹⁰ a known cause of aortic dilation.

Table 2 Framingham Heart Study: Metabolites and lipids associated with prevalent cardiac structure/function

Parameter	Metabolite/lipid	β coefficient	R2	P value	Adj P value (FDR)
LVIDd (cm)	Kynurenine	−0.741	0.231	9.0E-05	0.005
LVIDd (cm)	Leucine	−1.007	0.231	9.1E-05	0.005
LVIDd (cm)	Glycine	0.538	0.230	2.4E-04	0.007
LVIDd (cm)	Aminoadipate	−0.524	0.230	2.6E-04	0.007
LVIDd (cm)	Valine	−1.011	0.230	3.1E-04	0.007
LVFS (%)	C38_6_PC	0.963	0.359	1.3E-04	0.007
LVIDd (cm)	C58_9_TAG	0.474	0.359	1.7E-04	0.007
LVIDs (cm)	C58_6_TAG	0.450	0.359	1.9E-04	0.007
LVIDs (cm)	C58_12_TAG	−0.181	0.133	8.2E-05	0.009
LVIDs (cm)	C58_8_TAG	0.505	0.358	4.9E-04	0.013
LVIDs (cm)	ADP	0.040	0.023	1.3E-04	0.015
LVIDs (cm)	Isoleucine	−0.724	0.229	7.7E-04	0.015
LVIDs (cm)	C56_8_TAG	0.449	0.358	8.1E-04	0.016
LVIDs (cm)	C20_5_CE	0.311	0.358	9.1E-04	0.016
LVIDs (cm)	Niacinamide	−0.338	0.132	2.2E-04	0.023
LVIDs (cm)	Alpha-ketoglutarate	−0.311	0.131	7.5E-04	0.023
LVIDs (cm)	Aminoadipate	−0.514	0.131	7.6E-04	0.023
LVIDs (cm)	ADP	−0.214	0.131	9.6E-04	0.023
LVWTd (cm)	Kynurenine	−0.653	0.130	1.2E-03	0.023
IVSd (cm)	Lactate	−0.724	0.130	1.4E-03	0.023

[IMAGE OMITTED. SEE PDF]

Metabolite and lipid associations with incident cardiac structure and function in Framingham Heart Study

A total of 1711 and 1698 individuals with metabolomic data also had incident echocardiographic data at Examinations 6 and 8, respectively. We also determined associations of metabolites and lipids with incident cardiac structural and functional parameters, for potential insight into early pathogenic changes underpinning cardiac remodelling. There were 440 metabolite–incident echocardiographic associations at nominal significance, with 51 meeting FDR-adjusted significance (top 20 metabolite and lipid associations shown in Table 3). The strongest metabolite–cardiac structure association was phosphocreatine, the phosphorylated creatine molecule that serves as a reserve of high-energy phosphates, and was inversely associated with incident LV wall thickness (β = −0.3, P = 7.2 × 10⁻⁷); this relationship was previously seen using magnetic resonance spectroscopy.¹¹ Cotinine was associated with several incident cardiac structural and functional parameters including positive associations with wall motion, heart rate, mitral valve E-point septal separation, and LV internal diameter and negative associations with LV ejection fraction and LV fractional shortening. Plasma glucose was associated with incident left atrial internal dimension (β = 3, P = 9.4 × 10⁻⁵). Ornithine was inversely associated with LV percentage fractional shortening (β = −7.11, P = 1 × 10⁻⁴). Glutamine was inversely associated with heart rate (β = −13, P = 1 × 10⁻⁴). As previously reported for prevalent association,⁵ we also found asparagine to be associated with LV internal dimension.

Table 3 Framingham Heart Study: Metabolites and lipids associated with incident cardiac structure/function

Parameter	Metabolite/lipid	β coefficient	R2	P value	Adj P value (FDR)
LVmWT (cm)	Phosphocreatine	−0.301	0.344	7.2E-07	8.1E-5
Wall Motion Score Index	Cotinine	0.027	0.042	4.0E-06	4.5E-4
LVEF (%)	Cotinine	−1.329	0.087	1.1E-05	0.001
Heart rate	Cotinine	1.718	0.041	8.6E-06	0.001
LVFS (%)	Cotinine	−0.996	0.077	1.7E-05	0.002
LV mass (g)	C18_1_SM	−70.679	0.488	2.3E-05	0.002
MV E septal separation (cm)	Cotinine	0.045	0.134	2.8E-05	0.003
MV annulus diameter (cm)	C34_2_PC	−2.298	0.285	7.1E-05	0.004
MV annulus diameter (cm)	C36_2_PC	−1.457	0.285	6.0E-05	0.004
Heart rate	Glutamine	−12.981	0.038	1.0E-04	0.005
Heart rate	Glycerol	8.480	0.038	1.2E-04	0.005
Wall Motion Score Index	C18_2_LPC	−0.293	0.040	5.1E-05	0.005
Heart rate	C38_6_PC	−22.949	0.038	1.1E-04	0.006
Heart rate	C22_6_CE	−9.210	0.038	8.1E-05	0.006
LVFS (%)	Ornithine	−7.110	0.053	1.0E-04	0.007
LV mass (g)	C16_1_SM	−104.258	0.457	7.3E-05	0.008
LVIDs (cm)	Cotinine	0.067	0.222	8.3E-05	0.009
LVEDV (mL)	C36_2_PC	−104.064	0.280	9.3E-05	0.010
LV mass (g)	C22_1_SM	−101.958	0.456	3.9E-04	0.010
LV mass (g)	C18_1_LPC	−55.068	0.456	3.7E-04	0.010

There were 579 lipid–incident echocardiographic associations at nominal significance, with 59 reaching FDR-adjusted significance (Table 3 shows top 20 metabolite and lipid associations). Sphingomyelin 18:2 was the top associated lipid, showing an inverse association with LV mass (β = −70.1 mg per SD increment, P = 2.3 × 10⁻⁵), which may report on the association of sphingolipids to lysosomal storage and cardiac hypertrophy in metabolic disease.¹²

Metabolite and lipid associations with incident congestive heart failure

After adjustment for age, gender, and BMI, 11 metabolites and 14 lipids were significantly associated with incident CHF at nominal significance (P < 0.05), with none after FDR adjustment (Supporting Information, Table S2).

The top associated metabolite with incident CHF was ornithine (β = 3.6, P = 2.8 × 10⁻³), consistent with previous reports of elevated plasma ornithine in CHF patients, possibly reflecting perturbed arginine and nitric oxide metabolism.¹³ The second most associated was glycerol (β = 2.3, P = 5.7 × 10⁻³). Not widely appreciated, glycerol is a significant cardiac energy substrate that serves as a key regulator of cardiomyocyte lipid metabolism and energy balance.¹⁴

Metabolite and lipid associations with incident heart failure with preserved ejection fraction

We subsequently performed analyses on CHF subtypes: HFpEF and HFrEF. There were 11 metabolites associated with incident HFpEF after multivariable adjustment (age, gender, and BMI) at nominal significance (P < 0.05) (Supporting Information, Table S3). The top associated metabolite with incident HFpEF was ornithine (β = 3.6, P = 2.8 × 10⁻³), which as above, may reflect perturbed arginine and nitric oxide metabolism¹³; N^G-monomethyl-l-arginine (L-NMMA), competitive inhibitor of nitric oxide synthase (NOS), an important risk marker of CVD that reports on abnormal NO-mediated vascular tone in CHF patients,¹⁵ was also significantly associated (β = 3.96, P = 0.04) with incident HFpEF. The second most associated metabolite with incident HFpEF was cardiac energy substrate glycerol (β = 2.3, P = 0.006).¹⁴ The next associated metabolite was CVD mortality risk marker dimethylglycine (β = 2.1, P = 0.01), followed by nicotine metabolite cotinine (β = 0.35, P = 0.02).

After multivariable adjustment, only one lipid species, phosphatidylcholine 36:4 (β = −6.2, P = 0.001), was associated with incident HFpEF, an inverse association (Supporting Information, Table S3).

Metabolite and lipid associations with incident heart failure with reduced ejection fraction

We found only one metabolite associated with incident HFrEF after multivariable adjustment, the nicotine metabolite cotinine (β = 0.6, P = 0.045) (Supporting Information, Table S4). Aetiologically, this makes sense, as the most common cause of HFrEF is ischaemic heart disease that is causally linked to smoking.

Metabolite–congestive heart failure associations in the Women's Health Initiative

As the most common form of CHF has higher prevalence in women, we next examined metabolite–CHF associations in ~1300 female participants of the WHI. There were 100 metabolite/lipid-prevalent CHF associations at FDR-adjusted significance (top 20 shown in Supporting Information, Table S5). There were several baseline metabolite–prevalent CHF associations in the WHI that validated associations seen in FHS. For example, at FHR-adjusted significance after adjustment for age and BMI, choline (β = 3.8, P = 1.3 × 10⁻⁵), dimethylglycine (β = 2.1, P = 9.2 × 10⁻⁶), and ornithine (β = 1.4, P = 6.7 × 10⁻³) were all significantly associated with prevalent CHF in WHI; the choline and dimethylglycine associations persisted at Timepoint 2. Therefore, amongst the validated associations, choline and its one-carbon metabolite dimethylglycine had the most significant associations with CHF and warrant further investigations as to underlying pathology.

In the WHI, the strongest three prevalent associations, adjusted for age and BMI, at both baseline and the second timepoint had a diabetic/insulin resistance signature (Supporting Information, Tables S5 and S6). At baseline, glucose (β = 6.7, P = 2.9 × 10⁻⁹), C4-acylcarnitine (β = 1.3, P = 1.8 × 10⁻⁷), and isoleucine (β = 4.9, P = 2.2 × 10⁻⁷) were the top 3 associations, with similar association metrics at the second timepoint. Unsurprisingly, creatinine was significantly associated with prevalent CHF at both timepoints in WHI, reporting on the known association between reduced renal function and CHF, with impaired renal function conferring a worse outcome in CHF.¹⁶ There were no second timepoint metabolite–incident CHF associations after FDR adjustment in the WHI.

Diagnostic performance of metabolites and lipids compared with gold standard, BNP, and mediation analyses of risk factors in heart failure with preserved and reduced ejection fraction

Brain natriuretic peptide (BNP) is the current gold standard heart failure biomarker. In a separate analysis, we combined all CHF cases in Examinations 5 and 6 with the closest measured metabolites (Exam 5) and BNP (Exam 6). We then performed receiver-operator characteristic (ROC) curve analysis to obtain an approximate comparison of the ability of metabolites and BNP to distinguish between CHF and no CHF. As can be seen in Supporting Information, Figure S1, the top performing metabolite ornithine [area under the curve (AUC) = 0.77] had almost equal discriminatory capacity for CHF as gold standard biomarker BNP (AUC 0.78).

Using metabolite association, we next performed a metabolite mediation analysis between HFpEF and risk factors known to be associated with HFpEF (hypertension and diabetes) and between HFpEF and structural characteristics (left atrial size and LV wall thickness). Although no metabolites appeared to mediate the hypertension association with HFpEF, two metabolites mediated the association of diabetes with HFpEF: ornithine 3.4% (P < 0.05) and asparagine 5% (P < 0.05) (Figure 1C; Supporting Information, Table S7).

We next determined if any metabolites mediated the relationship between cardiac remodelling changes and HFpEF, results summarized in Supporting Information, Table S8. Intriguingly, the α-amino acid asparagine used in the biosynthesis of proteins partly mediated several cardiac structural parameters. Asparagine negatively mediated the association between left atrial dimension in systole and HFpEF by −4.8% (P = 0.004) and in diastole by −3.7% (P = 0.012), positively mediated the association between LV systolic wall thickness and HFpEF by 4.2% (P = 0.036) and LV posterior wall thickness by 4.8% (P = 0.02). Hydroxyglutarate mediated LV systolic wall thickness–HFpEF by 5% (P < 0.01) and LV posterior wall thickness–HFpEF by 3.6% (P < 0.05) (Supporting Information, Table S8). This is intriguing, as elevations in this enzyme are seen in d-2-hydroxyglutarate aciduria syndrome that includes cardiac hypertrophy as part of the syndrome, and preclinical work demonstrated it leads to up-regulation of pro-hypertrophic genes¹⁷ and post-translational modifications in cardiac hypertrophy.^17,18

Metabolite signatures of sleep apnoea indices and their relationship to heart failure and mortality

As per Table 1, there were 390 individuals (198 male and 192 female) with sleep study data at Exam 5. We first determined the association between sleep study indices including obstructive-apnoea hypopnoea index (obAHI), central-AHI (cAHI), average O₂ saturation, minimum O₂ saturation, and time spent with O₂ saturation below 90% and the following outcomes: CHF; HFrEF; HFpEF and death due to any cause; and death due to CVD. The only significant association was between cAHI and CHF (P = 0.03) and between cAHI and HFpEF (P = 0.031) (Supporting Information, Table S9), and sleep indices were not associated with HFrEF, overall mortality, or mortality from CVD. We next performed a mediation analysis and found that glycine (P = 0.01) and ribose (P = 0.02) significantly mediated the association between central sleep apnoea and HFpEF (Supporting Information, Table S10). Ribose is an intermediate in the pentose phosphate pathway, implicating this pathway as a mediator. Glycine, inversely associated with insulin resistance, may confer protection in both diseases.

We next performed association of metabolites with sleep apnoea parameters. We also examined the association of metabolites with average oxygen saturation, minimum oxygen saturation, and time spent with oxygen saturations less than 90%, the latter being one of the strongest predictors of cardiovascular outcome.¹⁹ Cotinine, the nicotine metabolite, was the strongest associated metabolite with each parameter, and each SD increase in cotinine was associated with 2.3 min spent with an oxygen saturation less than 90% (β = 2.3 min, P = 2.3 × 10⁻⁵) (Table 4). To our knowledge, this is the first quantitative report of the relationship between plasma concentrations of a nicotine metabolite with oxygen saturations in people with sleep apnoea.

Table 4 Framingham Heart Study: Metabolites associated with sleep study indices

Parameter	Metabolite	β coefficient	R2	P value	Adj P value (FDR)
T < 90%	Cotinine	2.252	0.144	1.1E-04	0.013
T < 90%	Anthranilic acid	−4.729	0.128	6.3E-03	0.271
T < 90%	Indoxyl sulfate	−8.824	0.126	8.9E-03	0.271
T < 90%	Xanthosine	−14.425	0.126	9.8E-03	0.271
T < 90%	Deoxycholates	−6.219	0.125	1.2E-02	0.271
T < 90%	Glucose	44.189	0.124	1.5E-02	0.282
obAHI	ADP	−9.822	0.154	5.6E-03	0.396
obAHI	Ribose phosphate + ribulose phosphate	−14.128	0.153	6.9E-03	0.396
T < 90%	Phosphoglycerate	8.879	0.122	2.7E-02	0.400
T < 90%	3-OH-anthranilic acid	10.068	0.121	3.4E-02	0.400
T < 90%	Creatinine	−23.998	0.121	3.7E-02	0.400
T < 90%	Glutamine	10.928	0.121	3.9E-02	0.400
T < 90%	Adipate	−12.207	0.120	4.0E-02	0.400
T < 90%	Triiodothyronine	−14.808	0.120	4.4E-02	0.400
T < 90%	Glutamic acid	−3.379	0.120	4.6E-02	0.400
obAHI	3-OH-anthranilic acid	17.022	0.149	2.2E-02	0.610
obAHI	Glyco deoxycholates	−8.688	0.148	2.5E-02	0.610
obAHI	Salicylurate	−2.158	0.148	2.7E-02	0.610
cAHI	Ribose	−9.216	0.188	6.0E-03	0.678
Av O2 sat	Cotinine	−2.139	0.029	6.2E-03	0.678

Metabolite signatures sleep parameters and association of signature with congestive heart failure and mortality

We next sought to derive baseline metabolite signatures of heart failure subtype and five indices of sleep (average O₂ saturation, minimum O₂ saturation, time spent with O₂ saturation below 90%, obAHI, and cAHI). Figure 2A shows that the first three metabolite variates were significant. Figure 2B shows correlations of these three metabolite variates with HFpEF, HFrEF, and the five sleep indices. Metabolite Variate 1 was equally and inversely associated with both HFpEF and HFrEF. Metabolite Variate 2 was most strongly inversely associated with HFrEF. Metabolite Variate 3 was most strongly positively associated with HFpEF. Figure 2C illustrates the associations of each metabolite variate with overall mortality and CVD mortality.

[IMAGE OMITTED. SEE PDF]

Discussion

Although there have been previous efforts to identify circulating metabolite biomarkers of sleep apnoea,²⁰ to our knowledge, we are the first to use metabolites to determine the relationship between sleep disordered breathing, cardiac remodelling, heart failure risk, mediation of risk, and canonical variates related to outcome.

We confirm several previous metabolite associations with cardiac remodelling parameters and heart failure. Increased α-ketoglutarate correlated with increased LV wall thickness, while increased asparagine demonstrated an association with increased LV diastolic dimension.^5,13 Higher levels of ornithine, which has a role in the urea cycle, were associated with CHF.⁵ We report an intriguing inverse association between baseline levels of phosphocreatine, the major cardiac reservoir of high-energy phosphates, and future LV wall thickness. Likewise, we found a strong inverse association between baseline (lysosomal storage) sphingomyelins and future LV mass. This may suggest that individuals storing more energy and lipids in LV myocardium at baseline are most prone to subsequent loss of LV wall thickness/mass; however, this needs further study to confirm.

Central sleep apnoea is known to be associated with CHF, but the association with HFpEF rather than HFrEF was unexpected. Ribose partly mediated this association. Ribose is an intermediate in the pentose phosphate pathway and is a critical component of adenosine and essential for maintaining ATP levels. In a clinical trial, administration of ribose improved diastolic function and quality of life in CHF patients.²¹ The relationship to diastolic, rather than systolic, function is unclear but is consistent with the relationship we found to HFpEF rather than HFrEF.

Our mediation analyses offer insight into the cardiac remodelling process in HFpEF. Both asparagine and 2-hydroxyglutarate partly mediated the association of LV wall thickness with HFpEF. LV wall thickening is a cardinal feature of HFpEF, and changes in the proteogenic amino acid asparagine may relate to the structural composition of a thickening heart. 2-Hydroxyglutarate is associated with cardiac hypertrophy and activates several processes that are pro-hypertrophic including glycogen biosynthesis, histone modification, and protein acetylation.^17,18 Myocardial hypoxia from sleep apnoea likely serves as the stimulus for up-regulation of 2-hydroxyglutarate in this context.²²

Choline and its one-carbon metabolite dimethylglycine were both associated with CHF in both FHS and WHI. The associations in WHI were much more significant, which likely represents the closer temporal relationship of metabolite sampling to CHF diagnosis in this cohort. Although both metabolites are reported risk factors for CVD²³ or CV²⁴ mortality, neither have been previously shown to have direct associations with CHF. However, preclinical work has shown that mice ingesting diets supplemented with choline had worse myocardial fibrosis and worsened haemodynamic and anatomic parameters after trans-aortic constriction (TAC).²⁵ Choline is a precursor of trimethylamine-N-oxide (TMAO), and there are several reports of clinical associations between TMAO and CHF²⁶ and diastolic dysfunction²⁷; one study even demonstrated greater predictive utility of TMAO compared with BNP in HFrEF patients.²⁸

That nicotine metabolite cotinine was significantly associated with several sleep study indices was unsurprising, but the strong association with time spent with oxygen saturation less than 90%, a major predictor of cardiovascular outcome,¹⁹ is an important finding. We found that each SD increment in cotinine levels was associated with a 2.3 min increase in time spent with oxygen saturations below 90%. To our knowledge, we are the first to quantify the effect of smoking on sleep apnoea. This result underscores the risk of smoking in those with sleep apnoea, a very common disorder estimated to affect up to one-third of the adult population or 1 billion people worldwide.²⁹

Funding

This work was supported by the Sydney Medical School Foundation (JFOS); NSW Ministry of Health (JFOS DOH1003; DOH1006); and the National Heart Foundation of Australia (JFOS NHF104853). The Framingham Heart Study is conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with Boston University (Contract Nos N01-HC-25195 and HHSN268201500001I). This manuscript was not prepared in collaboration with investigators of the Framingham Heart Study and does not necessarily reflect the opinions or views of the Framingham Heart Study, Boston University, or NHLBI. Funding for SHARe Affymetrix genotyping was provided by NHLBI Contract N02-HL-64278. SHARe Illumina genotyping was provided under an agreement between Illumina and Boston University. Funding for Affymetrix genotyping of the FHS Omni cohorts was provided by Intramural NHLBI funds from Andrew D. Johnson and Christopher J. O'Donnell. Funding support for the Framingham Targeted and Untargeted Metabolomics – HILIC – Installment 1 dataset was provided by Massachusetts General Hospital Departmental funding. Funding support for the Framingham Metabolomics (HILIC – Installments 1, 2, and 3) datasets, Framingham Central Metabolomics – HILIC – Installments 1 and 2, and Lipid Platform – Installments 1 and 2 were provided by National Institutes of Health Grant R01 DK081572. The Women's Health Initiative (WHI) programme is funded by the NHLBI, National Institutes of Health, and U.S. Department of Health and Human Services through Contracts HHSN268201600018C, HHSN268201600001C, HHSN268201600002C, HHSN268201600003C, and HHSN268201600004C. Likewise, this manuscript was not prepared in collaboration with investigators of the WHI, has not been reviewed and/or approved by the WHI, and does not necessarily reflect the opinions of the WHI investigators or the NHLBI. Metabolomic analyses were funded by NHLBI Contract HHSN268201300008C. The datasets used for the analyses described in this manuscript were obtained from dbGaP at through dbGaP accession phs000200.v11.p3.

Conflict of interest

The authors report no conflicts of interest.

Author contributions

J.F.O.S. conceived and designed the study, interpreted the results, and wrote the manuscript. S.D. and J.Y. designed the statistical plans and performed the statistical analyses with Y.W. and wrote the manuscript. D.L., A.W., and M.I. retrieved and organized the relevant data from the databases and wrote the manuscript. Y.C.K. gave intellectual input to the metabolomic analysis and interpretation. S.L. gave intellectual input to the heart failure results and interpretation and wrote the manuscript. H.D. retrieved and analysed sleep data. P.C. gave intellectual input on sleep data and reviewed the manuscript. I.W. gave intellectual input on heart failure–sleep interaction and reviewed the manuscript. K.C. and M.F. gave intellectual oversight to the sleep study data and offered interpretation. All authors reviewed the manuscript and have approved the final version of the manuscript.

References

Mishra S, Kass DA. Cellular and molecular pathobiology of heart failure with preserved ejection fraction. Nat Rev Cardiol 2021; 18: 400–423.

Romero‐Corral A, Caples SM, Lopez‐Jimenez F, Somers VK. Interactions between obesity and obstructive sleep apnea. Chest 2010; 137: 711–719.

Bradley TD, Douglas Bradley T, Floras JS. Sleep apnea and heart failure. Circulation 2003; 107: 1822–1826.

Bixler EO, Vgontzas AN, Gaines J, Fernandez‐Mendoza J, Calhoun SL, Liao D. Moderate sleep apnoea: a ‘silent’ disorder, or not a disorder at all? Eur Respir J. 2016; 47: 23–26.

Andersson C, Liu C, Cheng S, Wang TJ, Gerszten RE, Larson MG, Vasan RS. Metabolomic signatures of cardiac remodelling and heart failure risk in the community. ESC. Heart Fail 2020; 7: 3707–3715.

Kannel WB, Feinleib M, McNamara PM, Garrison RJ, Castelli WP. An investigation of coronary heart disease in families. The Framingham offspring study. Am J Epidemiol 1979; 110: 281–290.

Rautaharju PM, Kooperberg C, Larson JC, LaCroix A. Electrocardiographic predictors of incident congestive heart failure and all‐cause mortality in postmenopausal women. Circulation 2006; 113: 481–489.

Wang TJ, Larson MG, Vasan RS, Cheng S, Rhee EP, McCabe E, Lewis GD, Fox CS, Jacques PF, Fernandez C, O'Donnell CJ, Carr SA, Mootha VK, Florez JC, Souza A, Melander O, Clish CB, Gerszten RE. Metabolite profiles and the risk of developing diabetes. Nat Med 2011; 17: 448–453.

Diguet N, Trammell SAJ, Tannous C, Deloux R, Piquereau J, Mougenot N, Gouge A, Gressette M, Manoury B, Blanc J, Breton M, Decaux J‐F, Lavery GG, Baczkó I, Zoll J, Garnier A, Li Z, Brenner C, Mericskay M. Nicotinamide riboside preserves cardiac function in a mouse model of dilated cardiomyopathy. Circulation 2018; 137: 2256–2273.

Eid AH, El‐Yazbi AF, Zouein F, Arredouani A, Ouhtit A, Rahman MM, Zayed H, Pintus G, Abou‐Saleh H. Inositol 1,4,5‐trisphosphate receptors in hypertension. Front Physiol 2018; 9: 1018.

Lamb HJ, Beyerbacht HP, van der Laarse A, Stoel BC, Doornbos J, van der Wall EE, de Roos A. Diastolic dysfunction in hypertensive heart disease is associated with altered myocardial metabolism. Circulation 1999; 99: 2261–2267.

Rijzewijk LJ, van der Meer RW, Smit JWA, Diamant M, Bax JJ, Hammer S, Romijn JA, de Roos A, Lamb HJ. Myocardial steatosis is an independent predictor of diastolic dysfunction in type 2 diabetes mellitus. J Am Coll Cardiol 2008; 52: 1793–1799.

Hakuno D, Hamba Y, Toya T, Adachi T. Plasma amino acid profiling identifies specific amino acid associations with cardiovascular function in patients with systolic heart failure. PLoS One 2015; 10: [eLocator: e0117325].

Gambert S, Héliès‐Toussaint C, Grynberg A. Extracellular glycerol regulates the cardiac energy balance in a working rat heart model. Am J Physiol Heart Circ Physiol 2007; 292: H1600–H1606.

Ishibashi Y, Shimada T, Sakane T, Takahashi N, Sugamori T, Ohhata S, Inoue S, Katoh H, Sano K, Murakami Y, Hashimoto M. Contribution of endogenous nitric oxide to basal vasomotor tone of peripheral vessels and plasma B‐type natriuretic peptide levels in patients with congestive heart failure. J Am Coll Cardiol 2000; 36: 1605–1611.

McAlister FA, Ezekowitz J, Tonelli M, Armstrong PW. Renal insufficiency and heart failure: prognostic and therapeutic implications from a prospective cohort study. Circulation 2004; 109: 1004–1009.

Akbay EA, Moslehi J, Christensen CL, Saha S, Tchaicha JH, Ramkissoon SH, Stewart KM, Carretero J, Kikuchi E, Zhang H, Cohoon TJ, Murray S, Liu W, Uno K, Fisch S, Jones K, Gurumurthy S, Gliser C, Choe S, Keenan M, Son J, Stanley I, Losman JA, Padera R, Bronson RT, Asara JM, Abdel‐Wahab O, Amrein PC, Fathi AT, Danial NN, Kimmelman AC, Kung AL, Ligon KL, Yen KE, Kaelin WG, Bardeesy N, Wong KK. D‐2‐hydroxyglutarate produced by mutant IDH2 causes cardiomyopathy and neurodegeneration in mice. Genes Dev 2014; 28: 479–490.

Karlstaedt A, Zhang X, Vitrac H, Harmancey R, Vasquez H, Wang JH, Goodell MA, Taegtmeyer H. Oncometabolite d‐2‐hydroxyglutarate impairs α‐ketoglutarate dehydrogenase and contractile function in rodent heart. Proc Natl Acad Sci U S A 2016; 113: 10436–10441.

Kendzerska T, Gershon AS, Hawker G, Leung RS, Tomlinson G. Obstructive sleep apnea and risk of cardiovascular events and all‐cause mortality: a decade‐long historical cohort study. PLoS Med 2014; 11: [eLocator: e1001599].

Humer E, Pieh C, Brandmayr G. Metabolomics in sleep, insomnia and sleep apnea. Int J Mol Sci 2020; 21: 7244.

Omran H, Illien S, MacCarter D, St Cyr J, Lüderitz B. D‐ribose improves diastolic function and quality of life in congestive heart failure patients: a prospective feasibility study. Eur J Heart Fail Wiley 2003; 5: 615–619.

Intlekofer AM, Dematteo RG, Venneti S, Finley LWS, Lu C, Judkins AR, Rustenburg AS, Grinaway PB, Chodera JD, Cross JR, Thompson CB. Hypoxia induces production of L‐2‐hydroxyglutarate. Cell Metab 2015; 22: 304–311.

Roe AJ, Zhang S, Bhadelia RA, Johnson EJ, Lichtenstein AH, Rogers GT, Rosenberg IH, Smith CE, Zeisel SH, Scott TM. Choline and its metabolites are differently associated with cardiometabolic risk factors, history of cardiovascular disease, and MRI‐documented cerebrovascular disease in older adults. Am J Clin Nutr 2017; 105: 1283–1290.

Svingen GFT, Schartum‐Hansen H, Ueland PM, Pedersen ER, Seifert R, Ebbing M, Bønaa KH, Mellgren G, Nilsen DWT, Nordrehaug JE, Øyen J, Nygård O. Elevated plasma dimethylglycine is a risk marker of mortality in patients with coronary heart disease. Eur J Prev Cardiol 2015; 22: 743–752.

Organ CL, Otsuka H, Bhushan S, Wang Z, Bradley J, Trivedi R, Polhemus DJ, Tang WHW, Wu Y, Hazen SL, Lefer DJ. Choline diet and its gut microbe‐derived metabolite, trimethylamine N‐oxide, exacerbate pressure overload‐induced heart failure. Circ Heart Fail 2016; 9: [eLocator: e002314].

Tang WHW, Wang Z, Fan Y, Levison B, Hazen JE, Donahue LM, Wu Y, Hazen SL. Prognostic value of elevated levels of intestinal microbe‐generated metabolite trimethylamine‐N‐oxide in patients with heart failure: refining the gut hypothesis. J Am Coll Cardiol 2014; 64: 1908–1914.

Trøseid M, Ueland T, Hov JR, Svardal A, Gregersen I, Dahl CP, Aakhus S, Gude E, Bjørndal B, Halvorsen B, Karlsen TH, Aukrust P, Gullestad L, Berge RK, Yndestad A. Microbiota‐dependent metabolite trimethylamine‐N‐oxide is associated with disease severity and survival of patients with chronic heart failure. J Intern Med 2015; 277: 717–726.

Schuett K, Kleber ME, Scharnagl H, Lorkowski S, März W, Niessner A, Marx N, Meinitzer A. Trimethylamine‐N‐oxide and heart failure with reduced versus preserved ejection fraction. J Am Coll Cardiol 2017; 70: 3202–3204.

Lyons MM, Melanie Lyons M, Bhatt NY, Pack AI, Magalang UJ. Global burden of sleep‐disordered breathing and its implications. Respirology 2020; 25: 690–702.