Elevated plasma β-hydroxybutyrate predicts adverse outcomes and disease progression in patients with arrhythmogenic cardiomyopathy

See allHide authors and affiliations

Science Translational Medicine  12 Feb 2020:
Vol. 12, Issue 530, eaay8329
DOI: 10.1126/scitranslmed.aay8329

Insight from metabolites

Some individuals with arrhythmogenic cardiomyopathy (AC) can live for decades without symptoms, and reliable and specific markers of disease progression remain lacking. Song and colleagues studied metabolic markers in the plasma and hearts of individuals with AC and their relatives. They found elevated plasma ketones, specifically β-hydroxybutyrate, in affected individuals and relatives suspected of having AC. Metabolomics using patient cells and a mouse model of the disease revealed that cardiomyocytes exhibited de novo ketogenesis with higher fatty acid oxidation that favored medium-chain fatty acid substrates. Individuals with early AC with major adverse cardiac events had elevated plasma β-hydroxybutyrate, suggesting that this may be a useful marker of disease progression.


Sudden death could be the first symptom of patients with arrhythmogenic cardiomyopathy (AC), a disease for which clinical indicators predicting adverse progression remain lacking. Recent findings suggest that metabolic dysregulation is present in AC. We performed this study to identify metabolic indicators that predicted major adverse cardiac events (MACEs) in patients with AC and their relatives. Comparing explanted hearts from patients with AC and healthy donors, we identified deregulated metabolic pathways using quantitative proteomics. Right ventricles (RVs) from patients with AC displayed elevated ketone metabolic enzymes, OXCT1 and HMGCS2, suggesting higher ketone metabolism in AC RVs. Analysis of matched coronary artery and sinus plasma suggested potential ketone body synthesis at early-stage AC, which was validated using patient-derived induced pluripotent stem cell–derived cardiomyocytes (iPSC-CMs) in vitro. Targeted metabolomics analysis in RVs from end-stage AC revealed a “burned-out” state, with predominant medium-chain fatty acid rather than ketone body utilization. In an independent validation cohort, 65 probands with mostly non–heart failure manifestations of AC had higher plasma β-hydroxybutyrate (β-OHB) than 62 healthy volunteers (P < 0.001). Probands with AC with MACE had higher β-OHB than those without MACE (P < 0.001). Among 94 relatives of probands, higher plasma β-OHB distinguished 25 relatives having suspected AC from nonaffected relatives. This study demonstrates that elevated plasma β-OHB predicts MACE in probands and disease progression in patients with AC and their clinically asymptomatic relatives.


Arrhythmogenic cardiomyopathy (AC) is an inherited cardiomyopathy with pathological hallmarks of cardiomyocyte loss and fibrofatty infiltration primarily in the right ventricle (RV) and can involve left ventricles (LVs) with various severities, leading to lethal arrhythmia and cardiac dysfunction (17). More than 50% of patients with AC carry desmosomal mutations, yet phenotypic penetrance and clinical presentations are variable (2, 8). The clinically concealed phase of AC, even in individuals with known mutations, could last 20 to 40 years, and some patients’ relatives could be asymptomatic carriers. Sudden cardiac death (SCD) could be the first symptom of patients with AC. The progression of AC is not predictable and could have flares of “hot” phases with malignant ventricular arrhythmias and heart failure. Thus, establishing a biomarker for early diagnosis and better risk stratification (9) of patients and their relatives (10) is quintessential for preventing lethal consequences of AC. Current diagnosis and risk stratification have focused largely on clinically detectable changes in arrhythmia burden and abnormalities of cardiac structure and function (911). Serum markers, such as C-reactive protein (CRP) and N-terminal pro-B type natriuretic peptide (NT-proBNP), have been reported to associate with AC. However, they are markers used for general evaluation of cardiac dysfunction but are neither specific for AC nor predictive of malignant arrhythmias in patients with AC (12, 13).

Many patients with AC excel in endurance sports during the early and clinically concealed phase of their disease, yet long-term endurance exercises are known to accelerate the onset and increase arrhythmia burdens of patients (14, 15). The precise reasons for early excellence in endurance exercise and later accelerated AC courses remain unclear. In an in vitro model, AC cardiomyocytes derived from patient-specific induced pluripotent stem cells (iPSC-CMs) exhibited increased fatty acid oxidation (FAO) compared to iPSC-CMs derived from individuals without AC at the metabolically mature and nondiseased state and only developed a dysregulated metabolic “burn-out” state with aggressive lipogenesis after induction of AC pathologies (16, 17). In this model, decreasing FAO in AC iPSC-CMs prevented cardiomyocyte apoptosis.

On the basis of these findings, we hypothesized (i) that metabolic markers of increased FAO in early-stage AC hearts may provide clues for early diagnosis of AC and (ii) that markers of FAO-related metabolic abnormalities may be used as predictors for adverse disease progression of AC hearts. Because ketone production and utilization are closely associated with high FAO and extrahepatic lipogenesis (18), and ketone utilization is reported to increase in the heart failure state (1921), we investigated the ketone production and utilization pathways in AC hearts and measured plasma β-hydroxybutyrate (β-OHB) in patients with AC to determine the roles of ketones in disease progression and outcomes.


Characteristics and proteomics profiling of RV heart tissue in the discovery cohort

The schematic diagram of the overall study design is shown in fig. S1. We used a discovery cohort including 13 unrelated patients with AC receiving heart transplantation (HTx; 8 males, mean age of 31.31 ± 3.29 years) who were neither diabetic nor obese [body mass index (BMI) < 30] and who did not have coronary stenosis of >50% in degree. All patients fulfilled the revised 2010 Task Force Criteria (22) for AC diagnosis (table S1), and diagnoses were validated by pathology examinations as previously reported (Fig. 1, A and B) (23). Genetic testing confirmed that seven patients (53.8%) carried mutations in AC-related genes, including two in desmoglein-2 (DSG2), one each in desmoplakin (DSP), desmocollin-2 (DSC2), plakophilin-2 (PKP2), or TMEM43, and one with compound lamin A/C (LMNA) and DSG2 mutations.

Fig. 1 Elevated myocardial ketone metabolism enzymes in AC hearts.

(A) Gross features and (B) Masson staining of an explanted heart from a patient with AC. (C) Heat map of metabolic pathway–associated protein expression. Pound sign (#) indicates significant difference with fold change > 1.3 and P < 0.05. (D) Representative Western blot and (E) the quantification of enzymes involved in myocardial ketone metabolism in 13 AC and 13 nonaffected hearts. (F) qPCR of enzymes involved in myocardial ketone metabolism in AC RVs (n = 13) compared to healthy donor RVs (n = 13). (G) Northern blot showing mRNA of HMGCS2 in AC and healthy RVs (n = 3 per group). (H) Ketone metabolism gene expression in cardiomyocytes, cardiac fibroblast, and adipocytes by qPCR. (I) Immunofluorescence imaging showing ketogenesis enzyme (HMGCS2) expression in cardiomyocytes (judged by the cell size from phalloidin staining) in AC RV. Data are shown as means ± SEM, and statistical significance was analyzed by Mann-Whitney U test. *P < 0.05; **P < 0.01; ***P < 0.001. VDAC, voltage-dependent anion-selective channel protein.

We performed unbiased and quantitative proteomics on myocardial tissues, comparing differentially expressed proteins between four nondiseased RVs and four explanted AC RVs. Proteins associated with fatty acid transport, FAO, glycolysis/lactate production, ketone metabolism, and fatty acid synthesis were analyzed. In AC RVs, FAO-related proteins were decreased except for medium-chain specific acyl–coenzyme A (CoA) dehydrogenase (ACADM) [a key enzyme catalyzing β-oxidation of medium-chain fatty acid (MCFA)]. Mitochondrial (CPT1β) and sarcolemmal long-chain fatty acid (LCFA) transport protein (FATP6) were down-regulated, consistent with depressed LCFA utilization (P < 0.05). Changes in glucose metabolic enzymes suggested depressed glycolysis and pyruvate oxidation, with a significant decrease in adenosine triphosphate (ATP)–dependent 6-phosphofructokinase, muscle type (PFKM) (P < 0.05), the rate-limiting enzyme in glycolysis (Fig. 1C). These changes are consistent with a metabolic dysregulated state in the failing AC RVs. In contrast, adipogenesis/lipogenesis-related proteins, such as fatty acid binding protein 4 (FABP4), perilipin1 (PLIN1), and fatty acid synthase (FASN), were up-regulated, consistent with increased adipogenesis/lipogenesis in AC hearts (Fig. 1, A to C). As recently reported for failing hearts (1921), the ketone utilization rate-limiting enzyme 3-oxoacid coenzyme A-transferase 1 (OXCT1) was significantly up-regulated in AC RVs compared to nondiseased RVs (P < 0.05) (Fig. 1C), suggesting higher ketone metabolism specifically in failing RVs of end-stage AC hearts. Analysis of metabolic enzyme/protein expression in AC LVs also showed a trend of decreased FAO/glycolysis and increased lipid synthesis enzymes, yet OXCT1 expression was not increased in AC LVs (fig. S2 and data file S1).

To confirm the proteomics results, we compared expression of ketone metabolism–related proteins in 13 nondiseased donor RVs and 13 AC RVs in our discovery cohort using Western blots (Fig. 1, D and E, and fig. S3) and quantitative polymerase chain reaction (qPCR) (Fig. 1F). Similar to the proteomics results, the mRNA expression of solute carrier family 16 member 1 (SLC16A1) (P = 0.001) and member 7 (SLC16A7) (P = 0.003) for ketone transport, OXCT1 for ketone metabolism (P = 0.005), and 3-hydroxy-3-methylglutaryl-CoA synthase 2 (HMGCS2) for ketogenesis (P = 0.028) was significantly higher in AC RVs, suggestive of higher ketone metabolism. Western blot results also confirmed higher OXCT1 and HMGCS2 (as well as less abundant HMGCL for ketogenesis) protein expression in all 13 AC RVs compared to 13 nondiseased RVs, validating the proteomics results in end-stage AC RVs (Fig. 1, D and E, and fig. S3). 3-Hydroxybutyrate dehydrogenase type 2 (BDH2; cytosolic BDH), but not BDH1, protein expression was higher in AC RVs. Northern blot also confirmed elevated expression of full-length transcript for HMGCS2 in AC RVs (Fig. 1G). To determine which cell type was the dominant cell that contributes to the changes in ketone metabolic proteins, we isolated cardiomyocytes, cardiac fibroblasts, and adipocytes from AC RVs. We verified the identity and purity of each isolated cell types through expression of cell type–specific genes (fig. S4). Most of the ketone metabolism genes, except for BDH2 and HMGCL, showed higher expression in cardiomyocytes than in fibroblasts or fat cells (Fig. 1H). Immunostaining revealed that the key ketogenesis enzyme HMGCS2 was expressed in AC RVs and colocalized with mitochondrial protein voltage-dependent anion-selective channel protein (VDAC) in cardiomyocytes (Fig. 1I).

Elevation of plasma ketones in patients with AC

The key ketogenic enzyme HMGCS2 is highly up-regulated in AC RVs, which is in contrast to the down-regulation of HMGCS2 found in failing non-AC dilated cardiomyopathy (DCM) hearts as reported previously (19). Our result suggests that the extrahepatic ketogenic process might be up-regulated specifically in AC heart tissue and may further influence plasma β-OHB. Therefore, we measured and compared β-OHB concentration using blood samples from 13 pre-HTx patients with AC, 13 healthy volunteers, 13 pre-HTx patients with non-AC DCM, and 20 patients with pulmonary arterial hypertension (PAH). The latter three groups were used as control groups including the healthy hearts, non-AC failing hearts, and RV failing hearts, respectively. The baseline characteristics of all four groups are summarized in table S2. Of note, no professional athletes or individuals who partook in frequent, high-intensity physical activities were included in our study. The plasma β-OHB of patients with AC was significantly higher than the other three groups (fold change = 6.03 versus healthy volunteers, P < 0.001; 2.07 versus DCM, P = 0.021; 3.55 versus PAH, P < 0.001; Fig. 2A and table S2). β-OHB of patients with DCM was also higher than that of healthy volunteers but lower than that of patients with AC, although mean LV ejection fractions (LVEFs) of patients with DCM were significantly lower than those of patients with AC (P = 0.006; table S2), suggesting that LVEF is not a major determinant for plasma β-OHB. β-OHB concentrations in explanted AC RV myocardium were significantly higher than those in nondiseased donor hearts (P = 0.008; Fig. 2B), a finding confirmed through quantitative targeted metabolome analysis via liquid chromatography–tandem mass spectrometry (LC-MS/MS) (Fig. 2C and fig. S5), which was in contrast to the decreased myocardial ketone body due to enhanced utilization as previously reported in end-stage DCM (19). In addition, the biochemical indices of liver function were similar between AC and DCM. These results suggest that the differences in β-OHB are likely due to the specific subtype of cardiomyopathy rather than general cardiac dysfunction.

Fig. 2 Plasma and myocardial β-OHB were increased in patients with AC.

(A) Plasma β-OHB concentrations in 13 pre-HTx AC patients, 13 healthy volunteers, 13 non-AC DCM patients, and 20 PAH patients. (B) Myocardial β-OHB in AC RVs and unaffected donor hearts identified by (C) LC-MS/MS (n = 13 per group). (D) Plasma β-OHB in patients with AC before and after HTx (n = 10). (E) Serum NEFA in AC patients before and after HTx (n = 12). (F) Plasma β-OHB in the PAH patients before and after thromboendarterectomy surgery (n = 8). (G) Comparison of plasma β-OHB concentration in matched coronary artery (CA) and sinus (CS) blood among different groups. Negative values in bar plot indicate cardiac utilization/uptake of β-OHB. *P < 0.05 by paired Student’s t test between matched coronary artery and sinus within group. Data are presented as means ± SEM, and statistical significance was analyzed by Student’s t test between two independent groups in (A), (B), and (G) and by paired Student’s t test between matched samples in (D) to (F).

We also obtained blood samples after HTx from 10 of the 13 patients with AC in the discovery cohort. β-OHB decreased significantly from pre-HTx concentrations (29.59 ± 5.09 μM versus 145.00 ± 32.60 μM, respectively; P = 0.003), becoming indistinguishable from β-OHB measured from 13 healthy volunteers (P = 0.241) (Fig. 2D). In contrast, the corresponding concentration of serum nonesterified fatty acid (NEFA), the circulating substrates for ketogenesis in liver, did not change significantly after HTx among the 12 patients with AC (P = 0.481; Fig. 2E). There was no correlation between β-OHB and NEFA in pre-HTx blood samples of all 13 patients with AC (fig. S6). In addition, plasma β-OHB of patients with AC in the discovery cohort also showed no correlation with glycated hemoglobin (HbA1c), liver/kidney function indicators, or major heart failure indicators such as NT-proBNP (P = 0.970), LVEF (P = 0.650), or high-sensitivity CRP (P = 0.992) (Table 1).

Table 1 Correlation of plasma β-OHB and other serum biomarkers of patients with AC in the discovery and validation cohorts.

BMI, body mass index; LVEF, left ventricular ejection fraction; HbA1c, glycated hemoglobin; NT-proBNP, N-terminal pro-B type natriuretic peptide; hsCRP, high-sensitivity C-reactive protein; AST, aspartate aminotransferase; ALT, alanine aminotransferase; NYHA, New York Heart Association; MACE, major adverse cardiac events.

View this table:

Considering that RV failure might influence ketogenesis in liver due to the disturbance in hepatic hemodynamics, we studied plasma β-OHB in 20 patients with PAH and RV dysfunction due to pulmonary embolism. Their plasma β-OHB concentrations before thromboendarterectomy were not significantly different from healthy volunteers (P = 0.227; Fig. 2A) and did not change significantly (P = 0.172) after thromboendarterectomy under cardiopulmonary bypass surgery (Fig. 2F), even with decreased RV failure after the surgery. These results suggest that RV dysfunction is not the direct cause of increased plasma β-OHB in patients with AC and support that AC hearts directly or indirectly (via liver) increase plasma β-OHB using a mechanism distinct from altered utilization of ketones in the failing heart.

To investigate ketone body metabolism in AC hearts, we collected matched coronary artery and sinus plasma from patients without heart failure, non-AC heart failure, as well as early- and end-stage AC. The ketone assay demonstrated that non-AC heart failure showed a significantly enhanced uptake of β-OHB as an alternative substrate (P < 0.05), whereas end-stage AC hearts showed a basic β-OHB uptake similar to non–heart failure (Fig. 2G). Three patients with early-stage AC receiving implantable cardioverter defibrillator (ICD) implantation showed moderately elevated β-OHB in coronary sinus plasma (Fig. 2G). Compared to the baseline consumption in individuals without heart failure, the patients with early-stage AC exhibited significantly increased β-OHB production in heart (P = 0.029). These results indicate that early- and end-stage AC hearts present different metabolic remodeling phenotypes regarding ketone bodies, which are also different from traditional heart failure such as DCM.

De novo ketogenesis in early-stage AC patient–derived iPSC-CMs and a mouse model

Because higher FAO in AC iPSC-CMs has been suggested to be associated with its pathogenesis (16, 17), ketogenesis is highly correlated with FAO (18), the key ketogenic enzyme HMGCS2 is highly up-regulated in AC RVs, and AC hearts show a potential ketogenic signature, we used patient-derived AC iPSC-CMs to simulate this disease model in vitro (Fig. 3, A to D). We treated the iPSC-CMs with “3F” factors (insulin, steroid, and 3-isobutyl-1-methylxanthine) to convert cardiomyocytes to become FAO dominant in their mature state, and further added two peroxisome proliferator–activated receptor γ (PPARγ) agonists, rosiglitazone and indomethacin, to the 3F protocol (full protocol termed “5F”) to induce AC pathologies (16). We measured β-OHB concentrations from media and CM lysates (Fig. 3, A and C). The assay results demonstrated a significant increase in β-OHB production from pathological AC CMs and release into culture medium at early (0.5 week, P < 0.05) and intermediate (1 week, P < 0.05) stages until 2 weeks. The greatest ketogenesis activation occurred at 1 week. Once the CMs progressed to end-stage, shown as impaired FAO at 2.5 weeks (16), the ketogenesis process was diminished owing to the lack of substrates. Lipogenesis also occurred at this time (16) because long-chain acyl-CoAs were not accessible to mitochondria for FAO and thus accumulated in the cytoplasm for synthesis of lipids such as triglyceride. Nondiseased iPSC-CMs did not display ketogenic signs under 3F or 5F induction (Fig. 3, B and D). qPCR revealed high mRNA expression of BDH1, OXCT1, and acetyl-CoA acetyltransferase (ACAT1) in pathologically induced CMs at early-stage AC (Fig. 3E). Ketogenesis gene HMGCS2 was barely detected in 0F and 3F groups (Fig. 3E) but was highly induced under 5F stimulation, consistent with expression in human AC hearts (Fig. 1, D to G). At end stage (3.5 week), we observed increased expression of transporter SLC16A7 and enzymes BDH1, ACAT1, and HMGCL, suggesting a high ketone body metabolic capacity in the failing CMs. This in vitro iPSC-CM model is consistent with human AC heart performance and supports the theory of ketogenesis occurring at early- and intermediate-stage AC.

Fig. 3 Validation of ketogenesis in patient-derived iPSC-CMs and in vivo in a mouse model of AC.

(A) β-OHB detected in media from AC iPSC-CMs exposed to the 3F or 5F protocol (*P < 0.05 between 5F versus 0F groups by Student’s t test). (B) β-OHB detected in media from healthy iPSC-CMs. (C) β-OHB expression in AC iPSC-CMs under 3F or 5F induction (*P < 0.05 between 5F versus 0F groups by Student’s t test). (D) β-OHB expression in healthy iPSC-CMs. (E) Expression of ketone metabolism transporters and enzymes by qPCR at early stage (1 week) and advanced stage (3.5 weeks) among 3F, 5F, and non-induced AC iPSC-CMs. Western blot and quantitation of ketone body metabolic enzyme alterations in (F) heart and (G) liver of Myh6-Cre: Dspw/f mice at early stage (4 months, n = 5 per group), as well as (H) liver at end-stage mice (12 months, n = 6 per group). (I) Western blot and quantitation of PPARγ in RV of nondiseased human donors (n = 13) and AC patients (n = 13). (J) ChIP assay of PPARγ binding to transcription sites at promoter region of HMGCS2 in human AC heart. Data are shown as means ± SEM, and statistical significance was analyzed by Student’s t test between two independent groups. *P < 0.05, #P < 0.1.

To further confirm hepatic or extrahepatic ketogenesis in vivo, we used a previously reported mouse model of AC (Myh6-Cre: Dspw/f) to measure alterations in ketone body enzymes during disease progression. This model presented no heart dysfunction until 6 months and progressed to apparent heart failure at 12 months (fig. S7). At 4 months, without apparent cardiac dysfunction, the early-stage AC mice showed significantly increased protein expression of the ketogenic enzyme HMGCS2 (P < 0.01) and moderate increase of HMGCL (P < 0.1) in heart tissue but not in liver (Fig. 3, F and G, and fig. S7A). At 12 months, however, the end-stage failing AC mice presented with enhanced hepatic ketogenesis, shown as increased HMGCS2 and HMGCL in the liver (Fig. 3H and fig. S7B). These results validate that ketogenesis is occurring within the heart during early-stage AC, whereas the liver may also be involved in ketone production during end-stage AC with heart failure.

Considering that PPARγ activation had been identified as one of the main metabolic drivers of pathogenesis in AC and was necessary for ketogenesis of AC iPSC-CMs in vitro, we speculated that PPARγ might be the transcriptional regulator of ketogenic enzyme expression in human hearts with AC. We observed significantly increased expression of PPARγ in human AC RV (P < 0.001; Fig. 3I), and chromatin immunoprecipitation (ChIP) assay indicated PPARγ bound to the transcriptional promoters of HMGCS2 (Fig. 3J). Collectively, our results demonstrate the ketogenic capacity of AC cardiomyocytes.

Metabolic deregulation favoring MCFA utilization rather than ketone bodies in end-stage human hearts with AC

To clarify the discordance of increased ketogenic enzymes and low β-OHB in end-stage AC in vitro with increased plasma β-OHB in patients with end-stage AC, we performed targeted metabolomics analysis using LC-MS/MS on AC RV tissue to gain insights into metabolic changes. Acyl-carnitine species detected by LC-MS/MS showed a moderate decrease in long-chain acyl-carnitine (C14-22) and a significant decrease in medium-chain acyl-carnitine (C8-12) and short-chain acyl-carnitine (≤C6) (Fig. 4A). Acyl-CoA profile demonstrated the accumulation of long-chain acyl-CoA but a decrease in medium-chain acyl-CoA (Fig. 4B). mRNA expression by qPCR showed decreased sarcolemmal LCFA transport proteins (SLC27A1 and SLC27A6) and decreased enzymes for LCFA and short-chain fatty acid β-oxidation (ACADVL, ACADL, and ACADS), but the key enzyme for MCFA β-oxidation (ACADM) and cytoplasm LCFA accumulation–related gene (FABP4) were increased (Fig. 4C). Western blots also showed decreased mitochondrial LCFA transport protein CPT1β (Fig. 4D), which could explain the decrease in long-chain acyl-carnitine (impaired entrance in mitochondria) and accumulation of long-chain acyl-CoA in the cytoplasm in hearts with end-stage AC. These results are consistent with impaired LCFA utilization and elevated utilization of MCFA as the main fuel in end-stage AC RVs because MCFA does not need CPT1β for mitochondrial entry.

Fig. 4 Metabolic dysregulation and increased MCFA use in RV tissue from patients with end-stage AC.

(A) Concentration of short-, medium-, and long-chain acyl-carnitine in terminal AC RV versus nondiseased donor RV (n = 13 per group). (B) Concentration of long-chain (C14-C18), medium-chain (C8-C12), and short-chain (C4) acyl-CoA in terminal AC versus nondiseased donor RV (n = 13 per group). (C) qPCR of genes involved in glucose and fatty acid metabolism in end-stage AC versus nondiseased donor RV (n = 13 per group). (D) Protein expression of LCFA mitochondrial transport protein CPT1β in end-stage AC versus nondiseased donor RV (n = 13 per group). (E) Concentration of ketone-related metabolic intermediates in terminal AC versus nondiseased donor RV (n = 13 per group). (F) Summary of ketone metabolism in end-stage AC RV. Data are presented as means ± SEM, and statistical significance was analyzed by Mann-Whitney U test between two groups. #P < 0.1; *P < 0.05; **P < 0.01; ***P < 0.001. NS, not significant.

Analysis of ketone body–related intermediate metabolites showed decreased acetyl-CoA, C2-carnitine, and C4-carnitine, suggesting a burned-out state in these hearts with end-stage AC (Fig. 4, B and E). This could explain the lack of β-OHB production despite high expression of ketogenic enzymes in end-stage AC iPSC-CMs in vitro. In contrast to the increased ketone utilization in end-stage heart failure as previously reported (19), our AC hearts showed that C4-OH-carnitine and β-hydroxybutyl-CoA (β-OHB-CoA) were decreased to different degrees, whereas β-OHB was accumulated, suggesting that the AC RVs did not use ketone bodies as an alternative energy substrate. This result is consistent with the matched coronary artery and sinus blood ketone assay, which showed lack of enhanced ketone utilization in end-stage AC (Figs. 2G and 4E). In combination with the results obtained from the AC mouse model, Fig. 4F presents a simplified diagram illustrating ketone body metabolism in end-stage AC RV.

Elevated plasma β-OHB predicts occurrence of MACE among AC probands

We hypothesized that the higher plasma β-OHB in patients with AC compared to patients with DCM without AC might reflect the specific pathogenic process of AC and may predict adverse progression of AC rather than being a general marker of heart failure. We recruited and screened a validation cohort of 65 probands hospitalized and diagnosed with AC, their 94 first- and second-degree relatives (for a total of 159 individuals), and 62 healthy volunteers (fig. S1 and tables S1 and S3). Plasma β-OHB of AC probands (77.55 ± 12.02 μM) was significantly higher than healthy volunteers (22.41 ± 0.86 μM) (fold change = 3.46, P < 0.001; Fig. 5A and table S3). The receiver operating characteristic (ROC) analysis indicated that plasma β-OHB could distinguish patients with AC from nonaffected individuals with the area under curve (AUC) of 0.862 (P < 0.001) at a cutoff of 32.74 μM and a sensitivity of 96.77% and specificity of 73.58% (Fig. 5B). The Spearman correlation analysis showed that plasma β-OHB (not displaying the Gaussian distribution) in 65 patients with AC weakly correlated with New York Heart Association (NYHA) classes (P = 0.025) and NT-proBNP (P = 0.025) but not LVEF and hepatic/renal function (Table 1). Plasma β-OHB of patients with AC was correlated with major adverse cardiac events (MACEs) (Spearman ρ = 0.617, P < 0.001; Table 1).

Fig. 5 Plasma β-OHB in the validation cohort.

(A) Plasma β-OHB of 65 AC probands and 62 healthy volunteers. (B) ROC curve with area under curve (AUC) analysis associated with (A), using a cutoff of 32.74 μM by Youden’s index. (C) Plasma β-OHB in AC probands with MACE (n = 34) versus those without MACE (n = 31). (D) ROC curve with AUC analysis associated with (C), using a cutoff of 43.41 μM by Youden’s index. (E) Logistic regression to determine odds ratio (OR) with 95% confidence interval (CI) for MACE with increasing plasma β-OHB, adjusted for age, gender, and BMI. (F) Logistic regression adjusted for NYHA and RVID. Data are presented as means ± SEM, and statistical significance was analyzed by Mann-Whitney U test between two groups.

We therefore examined whether patients with AC with MACE had further elevated plasma β-OHB. In terms of MACE in 65 AC probands, 11 had past syncope, 10 had sustained ventricular tachycardia (VT) and/or received catheter-based ablation, 5 had received at least one appropriate ICD therapy during 2 years before and 6 months after the time of blood sample collections, and 8 had HTx during follow-up. These 34 patients were defined as positive for MACE (table S1). Of note, eight HTx patients in the validation cohort were not included in the discovery cohort because they were not in the waiting list for HTx at the time of blood collections. These 34 probands with MACE had significantly (P < 0.001) higher β-OHB (108.93 ± 20.26 μM) than those without MACE (35.48 ± 4.11 μM, n = 31; Fig. 5, C and D). We also compared other clinical characteristics of patients with AC with MACE to those without MACE (table S4), noting that gender proportion and NYHA classes were different between the two groups, whereas right ventricular inner diameter (RVID) showed borderline changes. Plasma β-OHB remained significantly different between AC patients with MACE and those without MACE after adjusting for age, gender, and BMI [P < 0.001, odds ratio (OR) = 7.349; Fig. 5E] or RVID and NYHA (P < 0.001, OR = 6.431; Fig. 5F) using a logistic regression analysis. This suggests that β-OHB could be an independent predictor of adverse events in AC. To further determine whether plasma β-OHB could predict MACE among patients with AC, we performed ROC analysis. When the cutoff value was set to >43.41 μM, plasma β-OHB could predict MACE with a sensitivity and specificity of 77.42 and 82.35%, respectively (AUC = 0.857, P < 0.001; Fig. 5D). Thirty-nine of 65 patients with AC carried AC-related gene mutations (fig. S8A); however, plasma β-OHB was similar regardless of the presence of mutations (fig. S8B) or the specific type of mutation (fig. S8C).

Plasma β-OHB distinguishes individuals with early-phase AC from unaffected relatives of AC probands

Considering that AC CMs could be ketogenic at early stage in vitro, we determined whether plasma β-OHB of the first- and second-degree relatives of 65 AC probands could be used as an early diagnostic marker of AC. Relatives of these 65 AC probands with AC-related syncope or at least one positive AC Task Force Criteria in echocardiography or electrocardiography (ECG) studies were classified as suspected AC relatives or early-phase patients (n = 25; table S5). We found that plasma β-OHB of relatives with suspected AC (36.47 ± 2.98 μM, n = 25) was higher than that of unaffected relatives (24.04 ± 2.00 μM, n = 69) (fold change = 1.52, P < 0.001; Fig. 6A), whereas unaffected relatives had the same plasma β-OHB as healthy volunteers (P = 0.953; table S3). ROC analysis showed that plasma β-OHB could distinguish suspected AC relatives from nonaffected relatives with a sensitivity of 56.00% and a specificity of 93.94% using a cutoff value of 30.76 μM (AUC = 0.783, P < 0.001; Fig. 6B).

Fig. 6 Plasma β-OHB in relatives and probands correlates with adverse AC disease progression.

(A) Plasma β-OHB of suspected AC relatives (n = 25) versus unaffected relatives (n = 69). (B) ROC curve with AUC analysis associated with (A), using a cutoff value 30.76 μM by Youden’s index. (C) Plasma β-OHB stratified by progression of cardiac dysfunction (healthy volunteers, AC without ventricular alteration, AC with isolated RV dysfunction, and AC with LV involvement). (D) Plasma β-OHB stratified by five progressive stages according to clinical presentation and MACE events (unaffected relatives, suspected AC, AC without MACE, non-HTx MACE, and HTx). Data are presented as means ± SEM, and statistical significance was analyzed by Mann-Whitney U test between two groups.

We next investigated whether the presence of heart failure or history of arrhythmia correlated with plasma β-OHB in patients with AC. In 65 AC probands of the validation cohort, plasma β-OHB of 12 patients who had biventricular heart failure (134.12 ± 48.46 μM), 30 patients who had isolated RV (no LV involvement) failure (56.40 ± 10.73 μM), and 23 patients without structural alteration but suffering arrhythmias (including VT and ventricular extrasystoles as shown in table S1, 75.62 ± 16.55 μM) was higher than β-OHB of 56 healthy volunteers (Fig. 6C). No cardiac dysfunction–dependent elevation of β-OHB was observed. These results support that plasma β-OHB in patients with AC is elevated early and before RV or LV failure and that elevation of plasma β-OHB is not a compensatory response of heart failure but rather a specific metabolic signature in patients with AC. Thus, plasma β-OHB in patients with AC and their relatives might be an early biomarker of adverse pathological progression in AC.

To explore whether plasma β-OHB could serve as an indicator of disease progression over the entire AC disease spectrum, we categorized AC pathological progression and corresponding plasma β-OHB into five progressively worsening stages according to clinical presentation and MACE events, from completely nonaffected relatives (24.04 ± 2.00 μM, n = 69), suspected AC (36.47 ± 2.98 μM, n = 25), early AC [AC patients without MACE (35.48 ± 4.11 μM, n = 31)], progressive AC [AC probands with MACE but no HTx (89.82 ± 15.43 μM, n = 26)], to end-stage AC (13 AC patients in discovery cohort and 8 in validation cohort who received HTx during follow-up, 165.60 ± 30.23 μM, n = 21). Plasma β-OHB increased progressively with the severity of the disease (Fig. 6D), supporting that progressively rising β-OHB in patients with AC and their relatives may indicate underlying adverse disease progression.


The clinical course of AC and phenotypic penetration of pathology among relatives of affected individuals is unpredictable (6, 10, 11). An early biomarker to predict subclinical adverse disease progression in AC would enable early therapeutic interventions to prevent SCD or ventricular dysfunction. Here, we performed proteomics and mRNA and protein expression analysis using explanted AC hearts in a discovery cohort and showed that two ketone metabolic enzymes, OXCT1 and HMGCS2, and a key MCFA β-oxidation enzyme, ACADM, are up-regulated with increased β-OHB in AC RVs. Plasma β-OHB in patients with AC was higher than in patients with non-AC DCM, PAH with RV failure, and healthy volunteers, suggesting that elevated ketone bodies might be a signature in AC. Using a plasma ketone assay of matched coronary artery and sinus blood, we found that there was no increase in ketone uptake in hearts from patients with end-stage AC compared to those with non-AC heart failure and those not undergoing heart failure, and AC hearts might produce ketone bodies at early disease stage. Using patient-derived iPSC-CMs, we revealed production of ketone bodies with up-regulation of ketogenic enzymes at early disease stage, progressing to a burned-out phase without ketogenesis after 2 weeks. Metabolome profiling demonstrated impaired utilization of LCFA and enhanced MCFA oxidation in failing AC hearts; in contrast to alterative ketone utilization in advanced heart failure caused by DCM, β-OHB was not used but accumulated in AC RVs (19).

We found that patients with AC had higher plasma β-OHB than healthy volunteers, and AC probands with MACE had higher plasma β-OHB than those without MACE in a validation cohort. Plasma β-OHB in relatives of individuals with AC suspected of having AC was higher than that in nonaffected relatives. Plasma β-OHB in patients and their family members correlated with worsening disease progression.

Clinical predictors and risk factors for poor outcomes in AC, including malignant arrhythmias and SCD, have been well established (6, 9, 11). Our recent study also reveals that ECG parameters could reflect myocardial fibrofatty involvement and predict HTx events (23). Nonetheless, these predictors are late AC manifestations, and patients may not survive these adverse events. Inflammatory activities, mediators, and cytokines have been suggested to influence disease progression in AC (24). Elevated plasma inflammatory cytokines, such as interleukin-1β (IL-1β), IL-6, and tumor necrosis factor-α, and myocardial inflammation in the RV (assessed by 67Ga scintigraphy) have been reported in patients with AC (25, 26). Anti-inflammation was recently reported to be an effective therapy for patients with AC (27). Hyperactivation of complement systems, such as C5aR (CD88) signaling, had been shown to mediate AC-like cardiomyopathy in desmin-null mice and found in patients with failing AC hearts (28, 29). A plasma biomarker to link this overactivation of CD88 in AC hearts has not been reported. Most recently, we reported that sex hormones were different between AC probands with or without MACE, and testosterone showed a good performance for predicting MACE in male patients (30). However, this biomarker could not distinguish early-stage AC from healthy individuals. Meanwhile, the autoantibody anti-DSG2 was also reported as a potential diagnostic biomarker based on work in a small cohort; however, its correlation with MACE or disease progression was not reported (31).

Patients with clinically diagnosed AC can have lethal ventricular arrhythmia and heart failure during the so-called active phases but have no symptom in between these active phases. Also, SCD from sustained ventricular arrhythmia sadly may occur as the first symptom, which could precede the structural abnormalities (1, 5). A new clinical prediction model has recently been developed to predict the ventricular arrhythmias for definitely diagnosed AC patients (11). To AC gene mutation carriers, relatives of AC probands, and mildly symptomatic AC patients, the common clinical risk predictors are usually absent and they are not helpful for determining which asymptomatic AC relatives or suspected AC patients would exhibit disease progression and adverse outcomes. Now, there is no early biomarker that can predict the disease progression or symptomatic onset of the at-risk family members of AC patients or mutation carriers. As such, many relatives of AC patients, AC gene mutation carriers, and AC patients with mild symptoms may receive unnecessary exercise restriction, treatment with beta blockers, or ICD implants to prevent SCD. AC is a progressive disease, yet the disease progression does not occur at a consistent pace or intensity (regarding symptoms and pathological presentation). A biomarker that correlates with AC pathology progression and/or arrhythmia incidence at the pre- and early symptomatic stages of AC would facilitate the early implementation of life-saving (6) or disease-slowing therapies (32) as well as avoid unnecessary procedures, drugs, or lifestyle restrictions in AC relatives or mutation carriers.

Blood ketones had been shown to correlate inversely with LVEF (21). Recent studies suggest that ketone utilization is increased in the failing hearts and that enhanced hepatic ketogenesis from increased circulating substrates (NEFA) provides the ketones for failing hearts (19, 20). Myocardial expression of BDH1 and OXCT1 is increased, and HMGCS2 is decreased in the LVs of patients with non-AC DCM (19). In these patients, serum β-OHB was higher and LV myocardial β-OHB was lower than in unaffected donors. In our discovery cohort, however, we found that OXCT1 (a reversible ketosis enzyme) and HMGCS2 (a ketogenic enzyme) expression were up-regulated with accumulation of β-OHB in AC RV tissue, suggesting a specific metabolic signature in the AC heart.

In combination with myocardial metabolomics, matched coronary artery and sinus plasma ketone assay, and in vitro AC iPSC-CM model, we determined that extrahepatic ketogenesis occurs in early AC. Elevated expression of ketogenesis enzymes HMGCS2 and HMGCL in AC RVs, moderate release of ketone bodies in early-stage AC, and de novo ketogenesis of AC iPSC-CMs under pathological induction by 5F protocols containing PPARγ agonists support this conclusion. Although the liver is usually considered the major source of ketones found in blood, a series of recent studies in murine models suggest that extrahepatic ketogenesis in the kidney (33) as well as the heart occurs (34, 35). Streptozotocin-induced and high-fat diet–induced diabetic mouse and rat models showed increased expression of HMGCS2 in the heart but not in the liver (34, 36). Moreover, PPARγ activation was demonstrated to be a regulatory factor responsible for ketogenesis gene expression in the diabetic heart using a PPARγ−/− murine model (37). Similarly, our in vitro study indicated that ketogenesis of AC CMs was observed only under PPARγ activation induced by the 5F protocol. We also observed significantly increased expression of PPARγ in human AC RVs (29) and demonstrated that PPARγ bound to transcriptional promoters of HMGCS2 by ChIP assay. Thus, it is reasonable to conclude that ketogenesis in AC RVs is PPARγ dependent.

At the same time, AC hearts also undergo metabolic remodeling from early to advanced stages of disease (fig. S9). During early stage, high FAO in AC contributes to extrahepatic ketogenesis; the β-OHB produced by AC cardiomyocytes is released into the blood, causing a slight increase in plasma β-OHB among AC patients and their relatives under concealed or early phase. Because of the high flow capacity of the coronary circulation (~300 ml/min, 5% of whole circulation), it is reasonable that the minimal increase in β-OHB in the coronary sinus could result in a slight increase in β-OHB in the whole blood circulation. When progressing to heart failure, LCFA oxidation was inhibited and MCFAs became alterative substrates, whereas ketone utilization did not increase. Meanwhile, ketone production in the CMs was also diminished because of the lack of substrates for ketogenesis in the burned-out heart. Because hepatic ketogenesis was activated in heart failure (according to data from the AC mouse model) and cardiac utilization was not enhanced, the concentration of blood ketone bodies was highly increased in end-stage AC patients compared to healthy volunteers, as well as non-AC heart failure and PAH patients.

In a previously published in vitro model of AC (16), AC cardiomyocytes were found to have higher FAO and lipid synthesis, which were essential for eventual AC cardiomyocyte apoptosis. Higher FAO and lipid synthesis in nonfailing AC CMs would lead to higher ketone generation in AC cardiomyocytes before cardiomyocyte apoptosis. Therefore, we hypothesized that early and high ketone production from high FAO in AC cardiomyocytes in presymptomatic AC hearts might provide the earliest signal for detecting adverse disease progression in AC. Our results in this study support that elevated plasma β-OHB in patients and their relatives displays a positive correlation with adverse progression of AC, regardless of their heart failure status. We show that plasma β-OHB could distinguish individuals with suspected AC from unaffected relatives of AC probands with a moderate sensitivity of 56.00% and an excellent specificity of 93.94% using a cutoff value of 30.76 μM. This suggests that plasma β-OHB could be used to monitor underlying disease progression in AC and can potentially prevent SCD with properly timed therapeutic measures.

Although this study included only 59 and 221 individuals in the discovery and validation cohorts, respectively, we have done our best to minimize several limitations of a study with such an intermediate size of patient populations. First, ketone metabolism is affected by diet and exercise (38). In this study, we reduced confounding factors through strict sample collection protocols, such as (i) collection of blood samples from participants at nonfasting states more than 2 hours after eating a meal, to minimize the changes in plasma β-OHB induced by fasting or food consumption, and (ii) the exclusion of professional athletes and heavy manual laborers. We collected patients’ blood samples in a resting state (without any intense workout within the previous 24 hours) so as to limit the effects of strenuous exercises. Second, note that this study used retrospective cohorts; it was not conducted as a prospective study. Hence, further large-scale, prospective, and longitudinal follow-up studies are needed to confirm the predictive value of blood β-OHB on the incidence and progression of AC pathology and on ventricular arrhythmia in AC patients and their relatives.

In conclusion, here, we demonstrate that cardiac ketogenesis occurs in AC. Plasma β-OHB may be used as a potential biomarker to predict not only MACE in AC probands but also disease progression in AC patients and their family members, particularly useful for individuals at the concealed phase.


Study design

This study was designed to identify metabolic biomarkers of AC and validate their roles in predicting adverse progression of AC (fig. S1). We conducted proteomic and gene expression analysis of 13 explanted AC hearts and 13 nondiseased donor hearts in a discovery cohort and identified two elevated key ketone metabolic enzymes in AC RV (fig. S1A). We then compared plasma β-OHB concentrations in 13 AC patients, 13 healthy volunteers, 13 non-AC cardiomyopathy patients, and 20 PAH patients with RV failure. Myocardial metabolomics of explanted heart and ketone assays of matched coronary artery and sinus blood were used to illuminate the ketone body consumption in early- and end-stage AC. We developed an AC iPSC-CM disease model and Myh6-Cre: Dspw/f mice model to verify the ketone metabolism from early to advanced phases (fig. S1B). Translationally, we included 65 AC probands, their 94 relatives, and 62 healthy volunteers in a separate validation cohort to correlate plasma β-OHB with clinical outcomes and MACE (fig. S1C). The samples from human, animal, and iPSC-CM were assigned to different groups based on their disease state or genotype. Blinding approaches were used in ketone measurement for human blood in discovery and validation cohort. The sample size in animal studies was five or six per group on the basis of previous study reports, and in vitro experiments were conducted three or four times. Primary data are reported in data file S2. Materials and Methods are detailed in the Supplementary Materials.

Assessment of baseline characteristics and MACE in AC probands and their relatives

Clinical information for AC patients and their relatives were collected during hospitalization to Fuwai Hospital in Beijing, China, during family member screening, and at the time of blood collections. We also collected MACE data from 2 years before to 6 months after the blood sample collections. MACE was defined as cardiac death, sustained VT, SCD, aborted SCD, cardiac syncope, appropriate ICD therapy, or HTx. Researchers who collected the clinical data had no previous knowledge of β-OHB assay results. Clinical diagnosis of AC was based on revised Task Force Criteria in 2010 (22).

Human heart tissue acquisition and blood sample collection

Human heart specimens and strict blood sample collection protocols were detailed in Supplementary Materials and Methods and were approved by the Institutional Review Board of Fuwai Hospital. All participants signed a written informed consent to partake in this study. The informed consent for research use of donor and explanted AC hearts was obtained before harvesting donor hearts and HTx.

Proteomics and metabolomics profiling, Western blotting, Northern blotting, qPCR, ChIP assay, and immunofluorescence assay

qPCR, Western blots, Northern blots, and immunofluorescence were used to confirm the mRNA expression, protein contents, and distribution of metabolic enzymes of interest. The qPCR primers were shown in table S6. ChIP assay was used to validate the binding relationship between transcription factor and target gene. Proteomics and metabolomics profiling based on LC-MS/MS was performed in Tsinghua University in Beijing, China, according to the standard protocols shown in Supplementary Materials and Methods, table S7, and data files S1 and S2.

Ketone and NEFA assay

Ketones (mainly β-OHB) were measured in plasma or cardiac tissues using the Beta-Hydroxybutyrate Assay Kit (MAK041, Sigma-Aldrich), and NEFA was measured in plasma by the immunoturbidimetry method using AU5400 Automation Chemistry System Instrument (Olympus) according to the manufacturer’s instructions, respectively.

In vitro model of AC using patient-derived iPSC-CM

A previously published AC iPSC-CM model was used to study the metabolic remodeling in AC cardiomyocytes (16). We used the 3F [0.5 μM dexamethasone, insulin (1 μg/ml), and 0.25 mM 3-isobutyl1-methylxamethasone] protocol to induce FAO-dominant metabolic maturation CMs and added two PPARγ agonists (5 μM rosiglitazone and 200 μM indomethacin) to the 3F protocol, termed the 5F protocol, to induce pathological phenotypes. Culture media and cardiomyocytes were collected at different times over 3.5 weeks to measure β-OHB concentrations and ketone body–related gene expression.

In vivo model of AC mice

A previously published AC model using cardiomyocyte conditional knockout of DSP (Myh6-Cre: Dspw/f) (39) was used to validate the ketone body metabolism enzymes alterations during AC progression in vivo. All animal protocols were approved by Fuwai Hospital Animal Care and Use Committee.

Statistical analysis

Measurement data were shown as means ± SEM, and category data were presented as frequency (percentage). Gene expression, protein quantification, ketone detection, and clinical characteristics between two groups were analyzed using Student’s t or Mann-Whitney U test according to the data distribution and Fisher’s exact test. A paired Student’s t test was used to compare blood biomarker before and after HTx, as well as ketone concentration between coronary artery and sinus blood. Comparison of three groups was performed by one-way analysis of variance (ANOVA) followed by post hoc analysis. Pearson correlation analysis was used to assess the correlation between two continuous variables with Gaussian distributions, whereas Spearman correlation analysis was used when continuous variables were not in Gaussian distributions. ROC with AUC analyses were used to assess the prediction value, and Youden’s index was used to determine the cutoff values with optimal sensitivity and specificity. Binary logistic regression was used to calculate OR with 95% confidence intervals ( of β-OHB and other covariates for MACE events, and we restricted a maximum of four variables in our multivariable analysis to avoid overfitting. A P value of <0.05 in two-sided testing was considered a statistically significant difference. SPSS 19.0 software (IBM) was used for statistical analysis, and GraphPad Prism 5 (GraphPad Software) was used for the preparation of statistical figures.


Materials and Methods

Fig. S1. Schematic diagram of the study design.

Fig. S2. Alterations of metabolic substrate utilization–related enzymes and transporters in AC versus nondiseased donor human heart identified by proteomic profiling.

Fig. S3. Western blots of ketosis enzymes and CPT1β in human RV tissue from AC (n = 13) and nondiseased donor (n = 13) individuals.

Fig. S4. Cardiac cell type isolation and validation by immunofluorescence and qRT-PCR.

Fig. S5. Identification of β-OHB concentration in human myocardium by LC-MS/MS.

Fig. S6. Correlation between plasma β-OHB and NEFA concentrations among 13 patients with AC in the discovery cohort evaluated by Pearson correlation.

Fig. S7. Phenotypic examination of mice.

Fig. S8. Plasma β-OHB concentrations of AC patients with different genotypes.

Fig. S9. Graphical abstract showing ketone body metabolism and FAO remodeling in early and advanced stages of AC.

Table S1. Clinical diagnosis and characteristics of patients with AC in the discovery cohort and validation cohort.

Table S2. Baseline characteristics of individuals included in the comparison of plasma β-OHB concentrations in the discovery cohort.

Table S3. Baseline characteristics of individuals in the validation cohort.

Table S4. Baseline characteristics of AC probands (n = 65) and association with MACE.

Table S5. Characteristics of relatives of AC probands who have Task Force Criteria for suspected AC in the validation cohort.

Table S6. Sequences of primer pairs used for qPCR.

Table S7. Description and calibration curve parameters of metabolites screened by targeted metabolomics.

Data file S1. Proteomics data.

Data file S2. Primary data.


Acknowledgments: We thank all patients for their participation in this study; H. Deng, X. Liu, and F. Yang (Metabolomics Center in National Protein Science Technology Center, Tsinghua University, Beijing, China) for their kind assistance in proteomics analysis; B. Qi (Shanghai Metabolome Institute, Wuhan, China) for metabolomics analysis; and D. Kelly and P. Crawford (Sanford-Burnham-Prebys Medical Discovery Institute, Orlando, FL, USA) for their comments on this manuscript. Funding: This study was supported by the National Natural Science Foundation of China (81970321) and CAMS Innovation Fund for Medical Sciences (2016-I2M-1-015) and PUMC Youth Fund and the Fundamental Research Funds for the Central Universities (3332018140), as well as NIH grant (RO1 HL105194), California Institute of Regenerative Medicine (CIRM) grant (RB4-06276), and start-up funds from KIC and School of Medicine, Indiana University. Author contributions: S.-S.H., H.-S.V.C., J.-P.S., and L.C. designed this study. L.C. conducted this study and collected the main results. L.C. and H.-S.V.C. performed statistics and prepared the manuscript. H.-S.V.C. and T.T. performed in vitro experiments. X.C., Z.-L.H., and N.-N.Z. contributed to sample and clinical data collection. J.R., W.H., and Y.-R.H. contributed to the blood collection and detection. H.-R.T. performed metabolomics analysis. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. The iPSC cell lines are available under a material transfer agreement with Indiana University (H.-S.V.C.).

Stay Connected to Science Translational Medicine

Navigate This Article