Metabolomic differences in connective tissue disease-associated versus idiopathic pulmonary arterial hypertension in the PVDOMICS cohort

Abstract

Objective:

Patients with connective tissue disease-associated pulmonary arterial hypertension (CTD-PAH) experience worse survival and derive less benefit from pulmonary vasodilator therapies than patients with idiopathic PAH (IPAH). We sought to identify differential metabolism in CTD-PAH versus IPAH patients that might underlie these observed clinical differences.

Methods:

Adult subjects with CTD-PAH (n=141) and IPAH (n=165) from the PVDOMICS (Pulmonary Vascular Disease Phenomics) Study were included. Detailed clinical phenotyping was performed at cohort enrollment, including broad-based global metabolomic profiling of plasma samples. Subjects were followed prospectively for ascertainment of outcomes. Supervised and unsupervised machine learning algorithms and regression models were used to compare CTD-PAH versus IPAH metabolomic profiles and to measure metabolite-phenotype associations and interactions. Gradients across the pulmonary circulation were assessed using paired mixed venous and wedged samples in a subset of 115 subjects.

Results:

Metabolomic profiles distinguished CTD-PAH from IPAH, with CTD-PAH patients demonstrating aberrant lipid metabolism, with lower circulating levels of sex steroid hormones and higher free fatty acids (FA) and FA intermediates in CTD-PAH. Acylcholines were taken up by the right ventricular-pulmonary vascular circulation, particularly in CTD-PAH, while free FAs and acylcarnitines were released. In both PAH subtypes, dysregulated lipid metabolites, among others, were associated with hemodynamic and right ventricular measurements and with transplant-free survival.

Conclusions:

CTD-PAH is characterized by aberrant lipid metabolism that may signal shifted metabolic substrate utilization. Abnormalities in RV-pulmonary vascular FA metabolism may imply reduced capacity for mitochondrial beta oxidation within the diseased pulmonary circulation.

Introduction

Pulmonary arterial hypertension (PAH) is a disease characterized by progressive pulmonary vascular remodeling and increasing pulmonary vascular resistance culminating in right ventricular (RV) failure (1). The over-arching PAH designation encompasses multiple phenotypic subtypes with a multitude of causes that are classified largely according to clinical associations, or, when no clinical association is apparent, as idiopathic (IPAH) (2, 3). In most PAH registries, a plurality of patients is classified as having either IPAH or PAH associated with connective tissue diseases (CTD-PAH), including systemic sclerosis (SSc), mixed connective tissue disease, systemic lupus erythematosus, or Sjogren’s syndrome (4–9). Across epidemiologic studies, CTD-PAH patients experience worse survival, despite generally demonstrating milder hemodynamic impairment at cohort enrollment (10–12). In clinical trials, CTD-PAH patients realize less benefit from PAH-specific therapies compared with IPAH patients (10). We and others have demonstrated specific pathophysiologic differences between CTD-PAH (chiefly SSc-PAH) and IPAH, including depressed sarcomere function and limited RV contractile reserve in SSc-PAH (13, 14). However, the fundamental biology underlying observed clinical differences in CTD-PAH remains poorly understood.

In recent years, advances in high-throughput molecular biology techniques coupled with non-frequentist approaches to high-dimensional data have enabled endo-phenotyping based on various -omics (e.g., proteomics, transcriptomics, metabolomics). It has become clear that PAH is, in part, a disease of disordered metabolism, and the potential benefits of metabolically-targeted therapies are beginning to be explored (15–20). The PVDOMICS (Pulmonary Vascular Disease Phenomics) study group has prospectively enrolled a multi-center cohort of subjects from across the spectrum of pulmonary vascular disease with detailed clinical data and accompanying -omics data, including broad-based metabolic profiling (21–23). In the current study, we leverage the PVDOMICS cohort to investigate the metabolic basis for CTD-PAH versus IPAH differences, and to understand whether metabolite-phenotype associations differ between disease subtypes. Additionally, we hypothesized that metabolic phenotyping would identify distinct metabo-endotypes, irrespective of CTD-PAH versus IPAH clinical classification, that correspond to right ventricular/pulmonary vascular (RV-PV) phenotypes and associate with differences in survival. Such deep phenotyping has the potential to pave a path for precision medicine approaches to tailored, metabolically-targeted therapies.

Methods

Methods for PVDOMICS cohort enrollment, clinical phenotyping procedures, and data curation have been previously published (21, 22). Plasma samples for metabolomics were collected from fasting subjects during right heart catheterization and shipped and stored by the Data Coordinating Center biorepository in accordance with study protocol. From the complete PVDOMICS cohort, a sub-cohort of metabolically profiled subjects with CTD-PAH (n=141) or IPAH (n=165) was identified for this study. Healthy controls (n=85) and CTD subjects without pulmonary vascular disease (n=38) were utilized for some comparisons. Subjects with significant interstitial lung disease (ILD), defined as having forced vital capacity ≤ 60% and characteristic radiographic features of ILD, were excluded from the CTD-PAH subgroup during phenotype adjudication. Mass spectrometry and data preprocessing was performed by Metabolon, Inc. (Morrisville, NC) as previously described (24–26). Metabolomic data analysis was performed using the MetaboAnalystR package for R (27, 28). Individual metabolite features (974) were log-transformed then re-scaled by mean-centering and dividing by the standard deviation of each feature. Xenobiotics such as drugs, foods, plants and their metabolites were removed. Principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were performed on transformed metabolite data. The Q2 statistic was calculated to validate PLS-DA model predictive performance, and a p-value for permutation tests was calculated for class prediction accuracy. Variable importance in projection (VIP) coefficients were generated for metabolite features driving component projection. To determine metabolite features most important for discriminating CTD-PAH from IPAH, MetaboAnalyst’s support vector machine algorithm generated various multivariable classification models, using univariable receiver operating characteristic (ROC) curves for metabolite feature ranking (29). Multivariable ROC curves were generated to assess model predictive accuracy with Monte-Carlo cross validation using repeated balanced sub-sampling (30). Two thirds of the samples were used for derivation, and 1/3 of the samples were used for model validation. Metabolite set enrichment analysis (MSEA) was performed to contextualize metabolic differences between groups at the pathway level, with the Small Molecule Pathway Database pathway library used as reference and a reference metabolome provided to eliminate platform-specific effects (31).

Age- and sex-adjusted logistic and linear regressions examined associations between individual metabolite abundances and clinical variables. Unknown metabolites (247) were not included in analyses of metabolite-phenotype relationships. Analysis of covariance (ANCOVA) models were used to investigate metabolite-by-subtype interactions. Cox proportional hazard models examined relationships between metabolite features and time to transplant-free survival. Outcomes data were censored as of 10/20/2022. Fold-change calculations and the Wilcoxon signed-rank test were used to analyze paired differences in untransformed metabolite concentrations across the RV-PV circulation (mixed venous versus wedged samples). To adjust for multiple comparisons, a Bonferroni-adjusted p-value of less than 8×10⁻⁵ was taken as significant, except for interaction terms, for which a p value less than 0.05 was considered significant (with p < 8×10⁻⁵ regarded as significant for main effects).

K-means clustering was performed to form new groupings of subjects based on metabolite features alone, without regard to clinical data. The elbow method was utilized to determine the optimal value of k. Differences in survival by cluster membership were examined with the Kaplan-Meier survivor function. To examine clinical differences across clusters, clinical variables were regressed on cluster membership with adjustment for age, sex, and disease subtype. PLS-DA was performed on labeled clusters to understand metabolic differences discriminating cluster membership.

Statistical analyses were performed using the MetaboAnalyst package (32) in R (v4.1.2; R Core Team 2021) and Stata Statistical Software (version 17, College Station, TX, USA).

Results

CTD-PAH versus IPAH clinical differences are shown in Table 1. CTD-PAH patients were older, with lower BMI and a higher proportion of female subjects. Hemodynamically, CTD-PAH patients had lower mean pulmonary arterial pressure (mPAP), lower time constant of the pulmonary circulation (RC-time), and trended toward worse pulmonary arterial (PA) compliance. Despite exclusion of subjects with significant interstitial lung disease (e.g., forced vital capacity ≤ 60% and characteristic radiographic features), CTD-PAH patients had worse pulmonary function than IPAH patients. Conversely, CTD-PAH patients had better RV function, as judged by MRI RV ejection fraction (RVEF), and preserved RV mass and volumes relative to IPAH. Despite lower mPAP and higher RVEF in CTD-PAH, these patients had significantly higher proBNP levels, as expected for this group (33). The vast majority of patients in both subgroups had prevalent disease; on average, IPAH patients had longer disease duration prior to enrollment. Within the CTD-PAH subgroup, most patients had SSc (n=62), followed by systemic lupus erythematosus (n=28) and mixed connective tissue disease (n=21). The demographic and clinical characteristics of healthy controls and CTD subjects without pulmonary vascular disease are shown in Supplemental Table 1.

Table 1.

Baseline Clinical Characteristics

Variable	CTD-PAH (n=141)	IPAH (n=165)	p-value
Age (years)	60.42 (12.18)	52.98 (15.02)	<0.01
BMI (kg/m2)	27.97 (7.04)	30.06 (7.54)	0.02
Sex (n, % female)	107 (79)	112 (67)	0.02
Race (n, %)			<0.01
White	89 (65)	140 (83)
Black or African American	28 (20)	9 (5)
Asian	6 (4)	7 (4)
American Indian/Alaska Native	4 (3)	3 (2)
More than one race	4 (3)	7 (4)
Unknown	5 (4)	2 (1)
6MWD (m)	333.90 (108.98)	425.59 (136.97)	<0.01
Systolic BP (mm Hg)	128.99 (23.26)	119.51 (20.68)	<0.01
Diastolic BP (mm Hg)	74.42 (14.81)	70.71 (12.71)	0.01
HR (beats per min)	77.86 (14.39)	74.86 (13.30)	0.09
RAP (mm Hg)	6.52 (4.83)	7.37 (4.54)	0.15
PA systolic (mm Hg)	63.09 (19.47)	67.43 (23.40)	0.09
PA diastolic (mm Hg)	24.91 (9.49)	27.92 (10.99)	0.01
Mean PA pressure (mm Hg)	39.06 (12.44)	42.54 (14.70)	0.03
PCWP (mm Hg)	10.58 (5.53)	10.99 (5.04)	0.63
PVR (Wood units)	6.44 (4.50)	6.50 (3.84)	0.92
Cardiac output (L/min)	5.17 (1.92)	5.46 (1.72)	0.15
Cardiac index (L/min/m2)	2.82 (0.99)	2.82 (0.85)	0.94
PA compliance	2.06 (1.21)	2.43 (1.94)	0.05
RC time	0.60 (0.21)	0.65 (0.17)	0.01
RVEF (%)	40.05 (13.24)	38.57 (11.22)	0.29
RVEDV (mL)	178.13 (59.32)	203.44 (87.21)	0.01
RVESV (mL)	112.11 (57.51)	130.86 (79.94)	0.03
RV stroke volume (mL)	66.03 (18.24)	72.59 (20.20)	0.01
RV mass (mg)	35.64 (15.42)	41.47 (20.36)	0.01
RV mass to volume ratio	0.20 (0.06)	0.21 (0.07)	<0.01
FEV1 (% predicted)	70.50 (18.53)	79.79 (18.58)	<0.01
FVC (% predicted)	74.30 (18.45)	86.33 (16.47)	<0.01
DLCO (% predicted)	40.49 (17.37)	62.77 (20.15)	<0.01
Hemoglobin (g/dL)	12.85 (1.97)	13.98 (1.77)	0.65
Sodium (mmol/L)	139.74 (3.19)	139.87 (2.63)	0.05
eGFR (ml/min)	76.68 (25.16)	81.93 (22.85)	<0.01
Total bilirubin (mg/dL)	0.5 [ 0.4–0.7]	0.6 [0.4–0.8]	0.02
Alkaline phosphate (uL)	80 [64-109]	75 [59-92]	<0.01
Glucose (mg/dL)	94 [86-105]	98 [88-112]	0.35
Triglyceride (mg/dl)	110 [81-149]	100 [77-136]	0.09
HDL (mg/dl)	49 [40-62]	48 [38-59]	0.20
LDL (mg/dl)	96 [77-126]	92 [70-119]	0.19
TAG/HDL ratio	2.23 [1.43-3.52]	2.15 [1.33-3.22]	0.56
proBNP (pg/mL)	454.7 [161.5-1657.0]	166.9 [81.7-785.0]	<0.01
Years with PH at enrollment	3.71 (4.52)	5.40 (6.15)	0.01
PDE5 at enrollment (n, %)	79 (58)	106 (64)	0.34
ERA at enrollment (n, %)	51 (38)	88 (53)	<0.01
Prostacyclin at enrollment (n, %)	28 (20)	73 (44)	<0.01

Data are displayed as mean (SD), median [25^th %ile – 75^th %ile] or n (%), as appropriate

Definition of abbreviations: BMI: body mass index; 6MWD: six minute walking distance; BP: blood pressure; HR: heart rate; RAP: right atrial pressure; PA: pulmonary arterial; PCWP: pulmonary capillary wedge pressure; PVR: pulmonary vascular resistance; RC: resistance-compliance; RV: right ventricular; RVEF: RV ejection fraction; RVEDV: RV end-diastolic volume; RVESV: RV end-systolic volume; FEV1: forced expiratory volume in 1 second; FVC: forced vital capacity; DLCO: diffusion capacity of the lung for carbon monoxide; eGFR: estimated glomerular filtration rate; HDL: high-density lipoprotein; LDL: low-density lipoprotein; TAG: triacylglycerol; proBNP: brain natriuretic peptide pro-hormone; PH: pulmonary hypertension; PDE5: phosphodiesterase-5 inhibitor; ERA: endothelin receptor antagonist.

Metabolic Differences in the Peripheral Circulation

Partial least squares discriminant analysis (PLS-DA) produced separation in scores space among subjects with CTD-PAH, IPAH, and healthy controls and afforded accurate class prediction (Q2 0.65, p <0.01). There was some overlap between CTD-PAH and IPAH groupings (Figure 1), implying metabolic overlap between the two disease subtypes and metabolic commonalities in comparison to healthy controls. A list of metabolites most important for discriminating among these three clinical classifications, along with variable importance in projection (VIP) scores, is included in Figure 1. Multiple androgenic and pregnenolone sex steroids, markers of oxidative stress, tryptophan, tyrosine metabolites, and the polyamine 4-acetamidobutanoate were among the metabolite features most important for distinguishing clinical groups. As demonstrated by the scores plot and the relative abundances of individual metabolite features, the metabolic profile of IPAH was intermediate between that of CTD-PAH and healthy controls. Because SSc-PAH constituted nearly half of the CTD-PAH subgroup, we analyzed SSc-PAH versus all other CTD-PAH in a sub-analysis to determine if substantial metabolic variability existed. With PCA, there was substantial overlap of these two groups in scores space (Supplemental Figure 1), and in univariate analysis, only two individual metabolite features differed significantly (an unknown metabolite and the histidine metabolite 1-ribosyl-imidazoleacetate). To ensure metabolic differences discriminating CTD-PAH from IPAH were not solely attributable to CTD, we compared CTD-PAH to CTD comparators without pulmonary vascular disease. With this comparison, we found CTD-PAH subjects were discriminated by lower circulating sex hormones, higher polyamines, and higher long-chain acylcarnitine levels (Supplemental Figure 2). Additionally, we analyzed metabolomic similarities and differences across key phenotypic comparisons (CTD comparators without pulmonary vascular disease vs. healthy controls; IPAH vs. healthy controls; CTD-PAH vs. IPAH) to pinpoint where metabolic dysregulation was overlapping versus distinctive. We found substantial overlapping metabolic perturbation across the conditions, such that many features that differ in CTD comparators vs. healthy controls also differ in IPAH vs. healthy controls and in CTD-PAH vs. IPAH (see Venn diagrams in Supplemental Figures 3a and 3b). We also found that many perturbations, particularly involving sex steroids, were unique to PAH and not present when comparing CTD subjects without PAH to healthy controls.

Figure 1.

Metabolic variability among CTD-PAH, IPAH, and healthy controls. A) scores plot for partial least squares discriminant analysis of metabolic data in CTD-PAH, IPAH and healthy controls. Red dots represent individual subjects with CTD-PAH, green dots represent individual subjects with IPAH, and blue dots represent individual healthy controls. B) Variable importance in projection (VIP) plots for PLS-DA Component 1.

Classification and feature selection for discrimination of CTD-PAH from IPAH in particular is summarized in Figure 2. Tryptophan (lower in CTD-PAH), sex steroids (lower in CTD-PAH), 4-acetamidobutanoate (higher in CTD-PAH), and markers of oxidative stress (higher in CTD-PAH) were among the features that distinguished CTD-PAH from IPAH. A model with 25 metabolite features yielded an area under the curve of 0.79 (95% CI 0.74–0.86) and a predictive accuracy of 72.6% for discriminating CTD-PAH (Figure 2). The individual metabolites selected for the 25-feature predictive model were predominantly androgenic and pregnenolone steroids. MSEA performed to contextualize metabolic differences between CTD-PAH and IPAH identified sex steroid metabolism as among the most enriched metabolic processes in CTD-PAH (Figure 2b).

Figure 2.

CTD-PAH versus IPAH metabolic differences. A) ROC curves for various metabolic models discriminating CTD-PAH from IPAH. B) Dot plot depicting the results of metabolite set enrichment analysis (MSEA) for CTD-PAH versus IPAH differences. Circle size represents the magnitude of the enrichment ratio (observed hits/expected hits),(31) and red-orange spectrum of color represents the significance of the association, with redder circles showing more highly significant associations.

Individual metabolite differences in the peripheral circulation between disease subtypes persisted with adjustment for age and sex, and with additional adjustment for PAH therapies. A complete list of individual metabolites differing in CTD-PAH compared to IPAH is provided in Supplemental Table 2, sorted according to metabolic pathway. The circulating metabolite conferring the greatest adjusted odds of CTD-PAH was the polyamine 4-acetamidobutanoate (OR 2.31, p 2.54 ×10⁻⁷). Androgenic and pregnenolone steroids were uniformly lower in CTD-PAH. Significant differences by subgroup also existed for metabolites involved in tryptophan metabolism, nucleotide metabolism, and the urea cycle. Long and medium chain FAs and their acylcarnitine intermediates tended to be higher in CTD-PAH. We performed sensitivity analyses in which we constructed multivariable logistic regression models adjusting for other factors that might alter lipid profiles, including differences in BMI, race, cardiovascular comorbidities, insulin resistance, inflammatory markers, and duration of PH at the time of cohort enrollment (Supplemental Figures 4a–c). All metabolite features that surpassed the Bonferroni-corrected threshold for significance remained strongly associated with CTD-PAH, even after adjustment for potential confounders. Estimates were particularly robust to adjustment among sex steroid metabolites. In general, estimates of effect size were stable with covariate adjustment for metabolites most discriminating of CTD-PAH versus IPAH differences. For several long-chain FAs, and for other metabolites with point estimates near the null, significance was attenuated with adjustment for some covariates (Supplemental Figures 4a–c).

Circulating metabolite associations with phenotypes and outcomes

All metabolites significantly associated with two or more clinical variables reflective of pulmonary vascular and/or right ventricular (RV-PV) function in either CTD-PAH or IPAH are depicted by magnitude and significance of association in Supplemental Figures 5a and 5b. Results for select metabolite associations with PVR and RVEF that are representative of key findings are plotted in Figure 3. In CTD-PAH, the androgenic steroids androsterone sulfate and epiandrosterone sulfate were significantly associated with lower pulmonary pressures and more favorable RV structural and functional parameters. TCA cycle metabolites (particularly fumarate and malate), and the tyrosine metabolites vanillactate and vanillylmandelate (VMA) were associated with higher pulmonary pressures, RV dilation, and RV dysfunction in CTD-PAH. Associations with adverse hemodynamic parameters also existed for acylcarnitines and purine and pyrimidine nucleotide metabolites.

Figure 3.

Regression analysis of select metabolite relationships with pulmonary vascular resistance and right ventricular ejection fraction. Point estimates for CTD-PAH are depicted in red, while those for IPAH are depicted in blue. Beta coefficients are represented by dots, and 95% confidence intervals are represented by spikes.

Formal testing of interactions between individual metabolite features and disease subtype using ANCOVA models demonstrated that higher circulating 4-acetamidobutanoate was strongly associated with higher pulmonary vascular resistance (PVR) and lower PA compliance in IPAH, but not in CTD-PAH (p for interaction terms <0.0001); higher circulating 11 beta-hydroxyetiocholanolone glucuronide, a byproduct of testosterone catabolism, was associated with lower mPAP in IPAH, yet higher mPAP in CTD-PAH (p for interaction term 0.001) (supplemental Figure 6). Other trends were suggested by inspection of coefficient plots but did not reach the threshold for statistical significance by ANCOVA: kynurenate, choline, and urate demonstrated PV-RV associations that tended to differ by subgroup. In IPAH, androgenic steroids were associated with RV parameters, but associations with pulmonary vascular function were of borderline significance. Acylcarnitines associated with both pulmonary vascular function and RV function in IPAH were not associated with RV parameters in CTD-PAH.

In survival analysis, for both CTD-PAH and IPAH, higher circulating concentrations of androgenic steroids were associated with improved survival, whereas higher levels of acylcarnitines, FAs, purine and pyrimidine nucleotide metabolites, polyamines (including 4-acetamidobutanoate) and the tyrosine metabolites vanillactate and VMA were associated with worse survival (Supplemental Table 3). Higher concentrations of several metabolites were associated with greater hazard of death or transplant in IPAH as compared to CTD-PAH (e.g., interaction terms were statistically significant), including dihydroxy, monohydroxy, and dicarboxylate FAs; the polyamine 4-acetamidobutanoate; the modified nucleosides N2, N2-dimethylguanosine and 5–6-dihydrouridine; and the tyrosine metabolite VMA.

Metabolic changes across the pulmonary circulation

A subset of 58 subjects with CTD-PAH and 57 subjects with IPAH had paired mixed venous and wedged samples collected. A RV-PV gradient was deemed to exist for metabolites with concentrations that differed significantly (p less than 8×10⁻⁵ in paired analysis) when sampled from the wedged position compared to the mixed venous position. Metabolites with significant RV-PV gradients in CTD-PAH and/or IPAH are listed in Table 2, with metabolites involved in lipid biology accounting for nearly half of features listed. In both disease subtypes, metabolites demonstrating the greatest uptake across the pulmonary circulation (e.g., wedged concentrations were significantly lower than mixed venous concentrations) were acylcholine phospholipids. Reductions in concentration across the pulmonary circuit of approximately 50% occurred for the long-chain acylcholines docosahexaenoylcholine, linoleoylcholine, and oleoylcholine in both CTD-PAH and IPAH. Significant RV-PV uptake of palmitoloelycholine was observed in CTD-PAH (FC −0.66, p 6.74×10⁻⁵) though not in IPAH (FC 0, p 0.0003). RV-PV release of the acylcarnitines palmitoylcarnitine, hexanoylcarnitine, and 5-dodecenoylcarnitine tended to be more pronounced in CTD-PAH compared to IPAH. RV-PV gradients with FC differences of >10% (e.g., RV-PV release) were observed for the FAs dodecadienoate, 3-hydroxydecanoate, and 3-hydroxylaurate in CTD-PAH, however no significant FA gradients were observed in IPAH. Similarly, RV-PV gradients for the tryptophan metabolite kynurenine (FC 0.13, p 3×10⁻⁵), the sterol intermediate of bile acid synthesis 7-HOCA (FC 0.15, p 1.08×10⁻⁸), and its precursor 3-beta-hydroxy-5-cholestenoate (FC 0.17, p 1.75×10⁻⁶) were noted in CTD-PAH, but these gradients were not significant in IPAH. Several phospholipids (e.g., choline, phosphoethanolamine) were variably taken up and released across the pulmonary circulation in both CTD-PAH and IPAH.

Table 2.

Right ventricular-pulmonary vascular metabolite gradients in CTD-PAH and IPAH

Metabolite	Pathway	Fold-change CTD-PAH	p value	Fold-change IPAH	p value
4-hydroxybenzoate	Benzoate Metabolism	0.57	1.84E-07	0.71	1.25E-07
methyl-4-hydroxybenzoate sulfate	Benzoate Metabolism	0.03	9.60E-03	0.22	2.78E-05
o-cresol sulfate	Benzoate Metabolism	0.14	1.56E-05	0.43	1.28E-05
ADSGEGDFXAEGGGVR*	Fibrinogen Cleavage Peptide	4.06	3.65E-07	0.00	1.64E-03
DSGEGDFXAEGGGVR*	Fibrinogen Cleavage Peptide	7.32	1.14E-07	0.00	6.39E-05
chain, or cyclopropyl 12:1 fatty acid*	Partially Characterized Molecules	0.13	1.68E-06	0.06	5.40E-04
hypoxanthine	Purine Metabolism, (Hypo)Xanthine/Inosine containing	0.42	2.68E-07	0.18	3.69E-06
inosine	Purine Metabolism, (Hypo)Xanthine/Inosine containing	0.24	1.04E-04	0.33	1.65E-06
cytidine	Pyrimidine Metabolism, Cytidine containing	−0.04	4.16E-01	−0.22	2.99E-06
cortisone	Corticosteroids	0.18	1.06E-07	0.20	3.38E-07
hexanoylcarnitine (C6)	Fatty Acid Metabolism (Acyl Carnitine, Medium Chain)	0.17	5.96E-05	−0.01	3.24E-03
5-dodecenoylcarnitine (C12:1)	Fatty Acid Metabolism (Acyl Carnitine, Monounsaturated)	0.25	8.59E-04	0.09	3.60E-05
arachidonoylcholine	Fatty Acid Metabolism (Acyl Choline)	−0.31	1.40E-05	−0.54	2.55E-08
dihomo-linolenoyl-choline	Fatty Acid Metabolism (Acyl Choline)	−0.34	1.35E-05	−0.65	2.79E-07
docosahexaenoylcholine	Fatty Acid Metabolism (Acyl Choline)	−0.52	2.27E-06	−0.48	8.08E-08
linoleoylcholine*	Fatty Acid Metabolism (Acyl Choline)	−0.44	2.32E-06	−0.51	1.58E-08
oleoylcholine	Fatty Acid Metabolism (Acyl Choline)	−0.48	2.41E-06	−0.48	2.80E-08
palmitoloelycholine	Fatty Acid Metabolism (Acyl Choline)	−0.66	6.74E-05	0.00	2.85E-04
palmitoylcholine	Fatty Acid Metabolism (Acyl Choline)	−0.34	5.92E-06	−0.48	4.59E-08
stearoylcholine*	Fatty Acid Metabolism (Acyl Choline)	−0.38	1.17E-05	−0.52	2.21E-07
dodecadienoate (12:2)*	Fatty Acid, Dicarboxylate	0.12	2.45E-05	0.02	1.75E-02
3-hydroxydecanoate	Fatty Acid, Monohydroxy	0.11	1.24E-05	0.06	6.13E-02
3-hydroxylaurate	Fatty Acid, Monohydroxy	0.20	4.74E-07	0.02	4.74E-02
glycerol 3-phosphate	Glycerolipid Metabolism	0.06	1.71E-03	0.19	2.61E-06
acetoacetate	Ketone Bodies	−0.18	6.49E-06	−0.31	1.75E-07
choline	Phospholipid Metabolism	0.11	8.67E-06	0.13	2.09E-05
glycerophosphoethanolamine	Phospholipid Metabolism	−0.17	4.09E-04	−0.23	1.50E-06
glycerophosphorylcholine (GPC)	Phospholipid Metabolism	−0.12	3.85E-04	−0.16	1.90E-08
phosphoethanolamine (PE)	Phospholipid Metabolism	0.30	5.35E-07	0.13	1.87E-02
pregnenetriol sulfate*	Pregnenolone Steroids	−0.05	2.04E-01	−0.11	6.89E-06
3beta-hydroxy-5-cholestenoate	Sterol	0.17	1.75E-06	0.14	3.05E-04
7-HOCA	Sterol	0.15	1.08E-08	0.06	7.23E-04
lactate	Glycolysis, Gluconeogenesis, and Pyruvate Metabolism	0.19	5.94E-09	0.11	1.21E-07
glutamate	Glutamate Metabolism	−0.12	1.53E-04	−0.12	6.76E-06
S-1-pyrroline-5-carboxylate	Glutamate Metabolism	−0.10	1.25E-04	−0.16	4.35E-05
fructosyllysine	Lysine Metabolism	0.10	1.56E-05	0.21	3.24E-03
alpha-ketobutyrate	Methionine, Cysteine, SAM and Taurine Metabolism	−0.01	4.83E-06	−0.27	5.10E-06
kynurenine	Tryptophan Metabolism	0.13	3.50E-05	−0.03	1.02E-03

Unsupervised clustering of the metabolome

Given the metabolic overlap between CTD-PAH and IPAH, we performed k-means clustering of all subjects with CTD-PAH and IPAH on the basis of metabolite features alone, without regard to clinical data, which produced three groupings of subjects (Figure 4a). K-means is an unsupervised clustering method that does not consider class labels, in contrast to PLS-DA, a supervised dimensionality reduction method which seeks to optimize separation between labeled groups. K-means clusters 1 and 3 exhibited distinct metabolic differences, while Cluster 2 demonstrated an intermediate metabolic profile. Within each metabolic cluster, membership included subjects with both CTD-PAH and IPAH (Supplemental Figure 7). Cluster membership predicted significant differences in survival times (p<0.0001) (Figure 4a) and was associated with significant differences in pulmonary hemodynamics and RV function (Figure 4b), even with adjustment for disease subtype. Consistently, Cluster 1 subjects demonstrated the most favorable clinical profile, Cluster 3 demonstrated the most adverse clinical profile, and Cluster 2 demonstrated an intermediate clinical profile. PLS-DA of labeled clusters demonstrated that the most important individual metabolite features driving cluster differences were dicarboxylate and monohydroxy FAs, acylcarnitine intermediates of FA metabolism, and the ketone bodies 3-hydroxybutyrate and acetoacetate (Supplemental Table 4).

Figure 4.

Unsupervised clustering of the metabolome. A) Kaplan-Meier plot depicting survival from the time of cohort enrollment by cluster. B) Box plots demonstrating differences in RV and pulmonary vascular variables by cluster. The center line denotes the median value (50^th percentile) and upper and lower hinges denote 75^th and 25^th percentiles of data respectively. Upper and lower fences denote 1.5 times the 75^th percentile and 25^th percentile respectively. Individual data points beyond upper and lower fences are presented as dots. Asterisks denote significance with linear regression p-values <0.05 in comparison to cluster 1.

Discussion

Our results demonstrate differential metabolism in the peripheral circulation as well as across the RV-PV circulation that may align with the disparate outcomes observed in CTD-PAH versus IPAH. Androgenic and pregnenolone sex steroid metabolites were associated with better RV function and lower pulmonary pressures, yet these favorably-associated features were markedly lower in CTD-PAH compared to IPAH. FAs and their metabolic intermediates were associated with worse RV-PV function, and these metabolites (among others) tended to be higher in CTD-PAH. We make several other novel observations here: transpulmonary sampling allowed us to demonstrate striking acylcholine uptake across the pulmonary circulation in both PAH subgroups, as well as acylcarnitine and FA release from the pulmonary circulation that appeared more pronounced in CTD-PAH. Similar to prior work comparing PAH subjects to non-PAH subjects (18, 34), our analyses also identified abnormalities in tryptophan metabolism, polyamine metabolism, nucleotide metabolism, and tyrosine metabolism that differed in CTD-PAH versus IPAH.

Despite these metabolic differences, the CTD-PAH and IPAH subgroups shared many metabolic similarities that distinguished PAH from healthy controls. From among all PAH patients, we were able to identify three groupings of subjects, based on differential metabolism alone, that exhibited clear differences in RV-PV clinical phenotypes. Cluster differences were not driven solely by CTD-PAH versus IPAH subgrouping. Cluster 3 subjects, with the most adverse clinical phenotypes, demonstrated the most striking abnormalities in lipid metabolism, with the highest circulating levels of acylcarnitines and free FAs. In our cohort, cluster membership was also significantly associated with survival.

The relevance of lipid biology to PAH and a role for lipotoxicity in RV dysfunction have been well-described. Under normal conditions, the myocardium is dependent on FA beta oxidation for cellular respiration and energy production, and free FAs are transported into mitochondria by the carnitine shuttle in the form of acylcarnitines. A metabolomic analysis of BMPR2-mutated human pulmonary microvascular endothelial cells (EC) revealed reduced levels of carnitine and acylcarnitine metabolites (relative to native ECs) and decreased gene expression of carnitine palmitoyltransferases and carnitine/acylcarnitine translocases (16). RV lipid deposition has been identified in RV cardiomyocytes in BMPR2 mutated mice stressed with PA banding and in autopsy specimens from humans with heritable PAH (HPAH) (35). In human HPAH, RV lipid deposition is accompanied by transcriptionally-identified defects in RV FA oxidation.

In vivo myocardial triglyceride content, as assessed by proton magnetic resonance spectroscopy, is increased in human subjects with multiple forms of PAH. Ceramides (mediators of lipotoxicity) and long-chain FAs are increased and long-chain acylcarnitines are reduced in the human PAH RV, while circulating long-chain FAs and long-chain acylcarnitines are elevated, a pattern suggestive of either failure of FA transport into the mitochondria or of failed utilization by the mitochondria for beta oxidation (15). In contrast to gene expression data from the pulmonary vasculature, there are no differences in expression of carnitine palmitoyltransferases in PAH versus control RV myocardium. However, RVs from BMPR2-mutated mice failed to augment oxygen consumption after supplementation with the long-chain acylcarnitine palmitoylcarnitine, suggesting an inability of RV mitochondria to utilize this particular fuel source for oxidative metabolism (15). Recently published data suggest that in experimental PAH, supplementation with l-carnitine may overcome this apparent dysfunction, leading to augmented mitochondrial FA oxidation, reduced RV lipid accumulation, and improved RV function, implying a relative carnitine deficiency in some PAH phenotypes that could be targetable (20).

In our study, circulating abundances of long-chain FAs and long-chain acylcarnitines were higher in CTD-PAH than in IPAH, and higher circulating FAs were associated with worse survival in both subgroups. Interestingly, long-chain acylcarnitines were associated with PV-RV dysfunction in CTD-PAH, while medium-chain acylcarnitines were associated with RV-PV dysfunction in IPAH. FA metabolism appeared impaired or incomplete across the RV-PV circulation: RV-PV release of the acylcarnitines hexanoylcarnitine (medium chain) and 5-dodecenoylcarnitine (long chain) was observed in CTD-PAH in particular, suggesting a mismatch between FA supply to the RV-PV circulation and the capacity of mitochondrial beta oxidation. These observations suggest more severe mitochondrial dysfunction may exist in CTD-PAH than in IPAH. Free FAs and acylcarnitines were also among the metabolites most important in discriminating phenotypically-distinct metabo-endotypes. The observation that free FAs, acylcarnitines, and the ketone bodies beta-hydroxybutyrate and acetoacetate are among the metabolic features that discriminate severe phenotypes supports the notion that disease severity in PAH is linked with metabolic shifts and altered fuel substrate utilization by the heart.

Endogenous acylcholines are choline molecules acylated with unsaturated FAs that have been discovered recently with broader use of untargeted metabolomics (36). In our cohort, higher circulating acylcholines (e.g., dihomo-linolenoyl-choline, stearoylcholine) were nominally associated with more favorable outcomes. Significant RV-PV uptake of several long chain acylcholines was observed in both CTD-PAH and IPAH, but differential RV-PV uptake was noted for palmitoloelycholine, which was taken up across the RV-PV circulation markedly in CTD-PAH, but not in IPAH. Preclinical data suggest that long-chain acylcholines may function as endogenous modulators of acetylcholine signaling, however in vivo function in the pulmonary vasculature remains undefined and should be the subject of future study (37). The pattern of acylcholine uptake and acylcarnitine release by the RV-PV circulation demonstrated by our study suggests a mechanistic role for acylcholine metabolism in pulmonary vascular disease that warrants further investigation.

The sex steroid metabolome robustly distinguished disease subtypes in our cohort, with circulating sex hormone levels reduced in all PAH compared to healthy controls, and in CTD-PAH compared to IPAH. Higher circulating concentrations of androgenic and pregnenolone steroids were associated with longer survival times in both CTD-PAH and IPAH (even with adjustment for sex), and higher circulating sex steroid levels were associated with more favorable RV function. These observations are generally in alignment with prior work showing that women taking estrogen therapy have higher RVEFs compared to men and compared to women not taking estrogen therapy, and that higher estradiol levels are associated with higher RVEF (38). A cross-sectional analysis of various measures of RV function demonstrated that RV-pulmonary arterial (RV-PA) coupling, the gold standard measure of RV function obtained invasively through RV pressure-volume measurements (PV loops), is superior in women with PAH compared to men (39). Women demonstrate improvements in RVEF when treated with PAH-specific therapy that are not seen in men (40). Epidemiologic studies have consistently demonstrated that women are at higher risk of PAH development than men, yet despite this predisposition, paradoxically, women enjoy a significant survival benefit relative to men (40, 41). It is biologically plausible that differential sex hormone metabolism underlies differential RV performance: in an animal model of PAH, supplementation with dehydroepiandrosterone (DHEA), an endogenous sex hormone precursor, reduced pulmonary arterial pressures, preserved RV contractile function and cardiac output, and improved oxidative stress (42). Taken together, these observations suggest that a depressed sex hormone axis, as observed in our CTD-PAH subgroup, may contribute, at least in part, to worse outcomes in CTD-PAH.

Whether reprogrammed lipid metabolism demonstrated here is a standalone phenomenon, independently linked to phenotypes and outcomes, or a consequence of other metabolic changes already known to exist in PAH is uncertain. The Warburg effect, characterized by a shift in energy production from mitochondrial oxidative phosphorylation to aerobic glycolysis, is known to occur in PAH (16, 43), and we re-demonstrate this in the current study (Figure 2b). It could be that increased circulating FA levels in the periphery and FA release across the pulmonary circulation are attributable entirely to decreased mitochondrial beta oxidation in Warburg physiology. Indeed, lactate, one end-product of Warburg effects, was also released across the RV-PV circulation in our study. Interestingly, however, in additional sensitivity analyses utilizing multivariable regression models adjusted for metabolic end-products of Warburg physiology (lactate, citrate, and pyruvate), the associations between FAs and sex steroids and phenotypes and survival persisted, suggesting independent associations with lipid dysregulation, and supporting the existence of separate metabolic phenomena that may be worth targeting therapeutically.

This study is limited by its cross-sectional nature and the sample sizes available for these subgroups of patients with a rare disease. We may be underpowered to detect other metabolic differences of pathobiologic importance. That said, with a sample size of greater than 300 patients, this is one of the largest PAH metabolomics studies to date. Despite the sensitivity analyses designed to address confounding, there may be residual confounding that is unaccounted for, for instance by the presence or absence of immunomodulatory therapies, which are not tracked by PVDOMICS. We are also unable to conclude whether the differential metabolism observed here is cause or consequence of CTD-PAH versus IPAH clinical differences. Our work is strengthened by the multi-center nature of our cohort, expert disease phenotyping at experienced centers, longitudinal follow-up of patients, and our ability to assess metabolism in the periphery as well as across the pulmonary circulation. In particular, findings regarding subtype differences in RV-PV gradients and the insights yielded by unsupervised clustering of the metabolome are novel aspects of our work.

In conclusion, these data show that CTD-PAH is characterized by aberrant lipid metabolism, with reduced sex steroids and increased circulating FAs and acylcarnitines. Many of these abnormalities are found in common with IPAH, therefore our results suggest that CTD-PAH patients experience more extreme lipid dysregulation than IPAH patients. We speculate that these observations may result from shifted substrate utilization to meet myocardial energy demands, and, in part, reduced capacity for mitochondrial beta oxidation across the RV-PV circulation. More extreme metabolic dysregulation exhibited in CTD-PAH may account for these patients’ shorter survival and unfavorable therapeutic responses. Future work is needed to test the utility of interventions targeted to modulation of lipid metabolism in PAH. The presence of a metabolically and clinically intermediate subgroup within PAH (as defined here by unsupervised clustering) should prompt design of future studies that test early metabolic interventions. Findings such as ours lay a groundwork for investigation of tailored approaches to therapy selection that may eventually replace the “one size fits all” paradigm currently employed across PAH subgroups.