Aberrant long-chain fatty acid metabolism associated with evolving systemic sclerosis-associated pulmonary arterial hypertension

Abstract

We sought to investigate differential metabolism in patients with systemic sclerosis (SSc) who develop pulmonary arterial hypertension (PAH) versus those who do not, as a method of identifying potential disease biomarkers. In a nested case-control design, serum metabolites were assayed in SSc subjects who developed right heart catheterization-confirmed PAH (n = 22) while under surveillance in a longitudinal cohort from Johns Hopkins, then compared with metabolites assayed in matched SSc patients who did not develop PAH (n = 22). Serum samples were collected at “proximate” (within 12 months) and “distant” (within 1–5 yr) time points relative to PAH diagnosis. Metabolites were identified using liquid chromatography-mass spectroscopy (LC-MS). An LC-MS dataset from SSc subjects with either mildly elevated pulmonary pressures or overt PAH from the University of Michigan was compared. Differentially abundant metabolites were tested as predictors of PAH in two additional validation SSc cohorts. Long-chain fatty acid metabolism (LCFA) consistently differed in SSc-PAH versus SSc without PH. LCFA metabolites discriminated SSc-PAH patients with mildly elevated pressures in the Michigan cohort and predicted SSc-PAH up to 2 yr before clinical diagnosis in the Hopkins cohort. Acylcholines containing LCFA residues and linoleic acid metabolites were most important for discriminating SSc-PAH. Combinations of acylcholines and linoleic acid metabolites provided good discrimination of SSc-PAH across cohorts. Aberrant lipid metabolism is observed throughout the evolution of PAH in SSc. Lipidomic signatures of abnormal LCFA metabolism distinguish SSc-PAH patients from those without PH, including before clinical diagnosis and in mild disease.

NEW & NOTEWORTHY Abnormal lipid metabolism is evident across time in the development of SSc-PAH, and dysregulated long-chain fatty acid metabolism predicts overt PAH.

INTRODUCTION

Systemic sclerosis (SSc) is frequently complicated by pulmonary arterial hypertension (PAH), a disease of the pulmonary vasculature characterized by progressive vascular remodeling and elevated pulmonary arterial pressures, ultimately resulting in right heart failure (1). PAH development represents an inflection point in the health of SSc patients, as those with PAH have roughly threefold increased risk of mortality compared to those without PAH (2). When compared with idiopathic PAH (IPAH), SSc-PAH patients suffer disproportionately high morbidity and mortality and demonstrate poorer responses to PAH-specific therapies compared with other PAH subgroups (3, 4). It is recommended that SSc patients be screened annually for the development of PAH (5), as earlier detection and treatment may improve outcomes (6). However, classic screening tests such as echocardiography and pulmonary function tests are only abnormal after substantial rises in pulmonary arterial pressure occur.

Dysregulated metabolism is a known feature of PAH pathobiology (7–10). Case-control metabolomics studies have consistently demonstrated altered bioenergetics in PAH, with striking abnormalities in glucose and lipid metabolism and aberrations in the TCA cycle, pentose phosphate pathway, and various other pathways (11–15). However, whether evolving metabolic abnormalities may signal PAH development in surveilled at-risk groups, such as SSc patients, remains unknown. With this study, we first leverage a nested case-control design to examine metabolic evolution over time in SSc patients who developed PAH, compared with those who did not, within a longitudinal surveillance cohort. We then compare our results with those from an analogous cohort examining metabolism in SSc patients with only mildly elevated pulmonary pressures as well as overt PAH. We hypothesized that identifying metabolic abnormalities associated with early PAH development could uncover candidate risk biomarkers and yield insights into the pathobiology underlying PAH development in SSc.

PATIENTS AND METHODS

Cohorts

The Johns Hopkins Scleroderma Center (JHSC) maintains a research registry and companion biorepository. Consenting subjects with SSc contribute serum samples serially to the biorepository during routine clinical care. Within the registry, 22 SSc subjects who developed right heart catheterization (RHC)-confirmed PAH while under surveillance and who contributed at least two serial serum samples (one “proximate” sample within 12 mo of PAH diagnosis, and one prior “distant” sample collected 12 mo to 5 yr before diagnosis) were identified as case subjects. Case subjects were matched one-to-one on age, sex, and disease duration (time from first non-Raynaud’s SSc symptom) to 22 SSc control subjects who also contributed serial serum samples but did not develop PAH under surveillance. SSc diagnosis was determined by expert opinion according to the American College of Rheumatology (ACR)/European League Against Rheumatism (EULAR) classification criteria (16). Screening thresholds for RHC referral at our center have been previously published (17).

SSc patients enrolled in an IRB-approved metabolomics and lipidomics study at the University of Michigan with 1) no PH (n = 22), 2) borderline elevation of pulmonary pressures (n = 21), or 3) overt PAH at RHC (n = 20) formed a second cohort for examination of metabolic differences across the spectrum of PAH evolution. These data are publicly available at the NIH Common Fund’s National Metabolomics Data Repository website (Project ID PR000551).

Two additional SSc cohorts were identified for validation of metabolomic results from the JHSC and Michigan cohorts. At the Johns Hopkins Pulmonary Hypertension Program (JHPHP), we maintain continuous enrollment of a prospective cohort of patients referred for RHC based on known or suspected pulmonary hypertension (PH). As part of an IRB-approved research protocol, patients undergo detailed clinical phenotyping and advanced hemodynamic assessment. Blood is collected during hemodynamic assessment for research. From within the JHPHP cohort, we identified blood samples collected from SSc patients consecutively referred; 24 patients were found to have PAH, and 14 had no PH at RHC, comprising an internal validation cohort.

The Pulmonary Vascular Disease Phenomics (PVDOMICS) Cohort is a multicenter observational study that enrolls patients with known PH as well as “comparator” patients with PH-predisposing conditions but no PH. Patients undergo extensive cardiopulmonary phenotyping, including untargeted metabolic profiling. PVDOMICS protocols and descriptions of the cohort have been previously published (18–20). Within the PVDOMICS cohort, 62 subjects have SSc-PAH, and 19 comparators with SSc do not have PH (or interstitial lung disease), comprising a second external validation cohort.

For the present study, PAH was defined as having mean pulmonary arterial pressure (mPAP) ≥25 mmHg, pulmonary capillary wedge pressure (PCWP) ≤15 mmHg, and pulmonary vascular resistance (PVR) ≥3 Wood units, in the absence of significant left heart disease, chronic lung disease, or thromboembolic disease, as this was the definition of disease in place at the time of cohort enrollment (21, 22). Borderline or mildly elevated pressures were defined as mPAP of 20–24 mmHg (23). Subjects with PAH were intentionally compared with subjects with no PH of any kind and without significant interstitial lung disease to avoid the potential for confounding.

Metabolic Profiling

Samples from the JHSC, JHPHP, and PVDOMICS cohorts underwent ultra-high-performance liquid chromatography-tandem mass spectroscopy (UPLC-MS/MS) using the Metabolon platform for global, untargeted metabolic profiling, as previously described (24). Briefly, a Waters Acquity UPLC and a ThermoScientific Q-Exactive high-resolution/accurate mass spectrometer interfaced with a heated electrospray ionization (HESI-II) source and Orbitrap mass analyzer operating at 35,000 mass resolution to analyze extracted and reconstituted samples. Raw data were extracted, peak-identified, and quality control-processed. Peaks were quantified using area under the curve.

Samples from Michigan underwent chromatographic separation on a Shimadzu CTO-20A Nexera X2 UHPLC. For lipid separation, the lipid extract was injected into a Waters Acquity HSS T3 column (Waters, Milford, MA). Data acquisition of each sample was performed in both positive and negative ionization modes, using a TripleTOF 5600 equipped with a Turbo VTM ion source (AB Sciex, Concord, Canada). Complete methodology for the Michigan cohort can be accessed via Project PR000551 DOI: 10.21228/M8FH50, and further methodological details regarding both platforms are in the Supplemental Methods.

Statistical Analysis

Metabolite abundances were normalized and scaled to achieve normality and compared between SSc-PAH and SSc without PH using fold-change (FC) differences and t tests. A Benjamini–Hochberg procedure was performed to account for multiple testing. Orthogonal projections to latent structures discriminant analysis (OPLS-DA) was performed on metabolite data at proximate and distant time points for dimensionality reduction and selection of important individual features using the ropls package for R (version 4.3.2) (25–27). Metabolite set enrichment analysis (MSEA) was performed to contextualize individual feature results at the metabolic pathway level using the MetaboAnalyst package for R (28, 29). A reference metabolome was included for each analysis to account for platform-specific effects. We evaluated the diagnostic accuracy of predictive features for SSc-PAH identified via dimensionality reduction using age- and sex-adjusted receiver operating characteristics (ROC) analysis implemented via the rocreg command in Stata (version 17.0). Areas under the curve (AUC) were calculated to quantify the diagnostic performance of models, and confidence intervals for the AUCs were estimated using bootstrapping with bias correction.

RESULTS

Patient Characteristics: JHSC Discovery Cohort

Thirty-one of the 44 patients in the JHSC cohort had limited cutaneous SSc, and the remaining had diffuse SSc. On average, subjects were (means ± SD) 60 ± 13 yr old with an SSc disease duration of 14 ± 9 yr. Ninety percent were women, and 76% were White. In subjects diagnosed with PAH, mean disease severity was moderate, with mPAP 34 ± 10 mmHg, PCWP 11 ± 4 mmHg, CI 2.4 ± 0.7 L/min/m², and PVR 6.9 ± 4.5 Wood units. Demographic and clinical characteristics are listed in Table 1.

Table 1.

Demographic and clinical characteristics of JHSC SSc-PAH and SSc-No PH subjects

Demographics	SSc-PAH	SSc-No PH
Subjects	22	22
Age, yr	63 (12)	54 (12)
Sex, n (% female)	20 (91)	20 (91)
Race, n (% white)	15 (68)	19 (86)
BMI, kg/m2	28 (3)	36 (5)
Hemodynamics
mPAP, mmHg	34 (10)	NA
PAWP, mmHg	11 (4.6)	NA
PVR, Wood units	6.9 (4.5)	NA
Cardiac output, L/min	4.0 (1.0)	NA
Cardiac index, L/min/m2	2.4 (0.7)	NA

Values are means (SD) unless otherwise specified. Clinical characteristics reflect the time point nearest to the PAH diagnosis. BMI, body mass index; mPAP, mean pulmonary arterial pressure; NA, data not available; PAH, pulmonary arterial hypertension; PAWP, pulmonary artery wedge pressure; PH, pulmonary hypertension; PVR, pulmonary vascular resistance; SSc, systemic sclerosis.

JHSC Proximate Time Point

Proximate time point samples were obtained within a median of 62 days (interquartile range: 14–208 days) of PAH diagnosis. Of the 1,554 metabolites analyzed, 172 (11.1%) were differentially abundant between SSc patients with and without PAH (Fig. 1A). Downregulated metabolites included many lipid species: predominantly acylcholines, other phospholipids, and some androgenic sex steroids. Multiple acylcholines with large FC differences were intermediates of linoleic acid metabolism, including dihomo-linolenoyl-choline (FC 0.360, P value = 0.00006), linoleoylcholine (FC 0.499, P value = 0.00022), 1-dihomo-linolenoyl-glycerophosphocholine (20:3n3 or 6) (FC 0.614, P value = 0.00142), 1-linolenoyl-glycerophosphocholine (18:3) (FC 0.661, P value = 0.01796), and 1-linoleoyl-glycerophosphocholine (18:2) (FC 0.789, P value = 0.03884). Upregulated metabolites with the largest FC differences included bile acid metabolites, histidine metabolites, tyrosine metabolites, tryptophan metabolites, and polyamines, findings in alignment with previous metabolomics studies in PAH (11–14).

Figure 1.

Volcano plots showing the magnitude (x-axis) and significance (y-axis) of metabolite fold-change differences in Hopkins subjects at proximate time point (A), Hopkins subjects at distant time point (B), Michigan patients with borderline elevated right heart pressures (C), and Michigan patients with overt PAH (D). Dots represent individual metabolite features. Features increased in subjects are represented by red dots, and features decreased are shown with blue dots. The horizontal line indicates statistical significance at P = 0.05. PAH, pulmonary arterial hypertension.

Dimensionality reduction of metabolomic data with OPLS-DA produced separation of SSc-PAH versus without PH at the proximate time point (Fig. 2A). Variable importance in projection (VIP) scores identified several acylcholines with long-chain fatty acid (LCFA) residues as among the most important features in distinguishing the two clinical groups (Fig. 2C).

Figure 2.

Metabolism in SSc with vs. without PAH across time. Scores plot of dimensionally reduced metabolomic data between Hopkins SSc patients with and without PAH at proximate time point (A) and distant time point (B). SSc controls without PAH are represented by red dots, and subjects with PAH are represented by blue dots. Variable importance in projection (VIP) plots for OPLS-DA at proximate time point (C) and distant time point (D). Higher VIP scores indicate greater importance of metabolites in discriminating labeled clinical groups in the dataset and driving separation of subjects in scores space. OPLS-DA, orthogonal projections to latent structures discriminant analysis; PAH, pulmonary arterial hypertension; SSc, systemic sclerosis.

JHSC Distant Time Point

Samples from the distant time point were obtained at a median of 2.02 yr (interquartile range: 1.11–2.40 yr) before PAH diagnosis. Of the 1,547 metabolites analyzed, 155 (10.0%) were significantly different in SSc with versus without PAH (Fig. 1B). Downregulated metabolites were primarily lipids; several of the same long-chain acylcholine phospholipids observed to be lower in SSc-PAH near the time of RHC were already lower in abundance approximately 2 yr before PAH diagnosis. For example, 1-linoleoyl-GPC (18:2) (FC 0.581, P value = 0.00013), dihomo-linolenoyl-choline (FC 0.240, P value = 0.00017), linoleoylcholine (FC 0.281, P value = 0.00026), 1-dihomo-linolenoyl-GPC (20:3n3 or 6) (FC 0.504, P value = 0.00006), and 1-linolenoyl-GPC (18:3) (FC 0.529, P value = 0.00071), all phospholipid species that incorporate the LCFA linoleic acid, were significantly lower in SSc-PAH subjects at both the distant and proximate time points (Supplemental Table S1).

Two years before a clinical PAH diagnosis, dimensionality reduction of metabolomic data produced a clear separation between SSc patients who would develop disease versus those who would not (Fig. 2B). VIP scores identified acylcholines with LCFA residues, including linoleic acid intermediates, as the top features distinguishing subjects who would develop PAH (Fig. 2D).

Models including nine circulating acylcholines with LCFA residues [arachidonoylcholine (FA 20:4), dihomo-linolenoylcholine (FA 20:3n-6), docosahexaenoylcholine (22:6n-3), eicosapentaenoylcholine (20:5n-3), linoleoylcholine (FA 18:2n-6), oleoylcholine (FA 18:1n-9), palmitoloelycholine (FA 16:1n-7), palmitoylcholine (FA 16:0), and stearoylcholine (FA 18:0)] as predictor variables (Supplemental Table S2) distinguished PAH from no PH with excellent discrimination at both time points. In ROC analysis, at the proximate time point, the AUC for acylcholine predictors was 0.91 [95% confidence interval (CI): 0.81–0.98] (Fig. 3A). At the distant time point, acylcholine predictors distinguished subjects who would be diagnosed with PAH 2 yr later with excellent discrimination (AUC: 0.93, 95% CI: 0.87–0.99) (Fig. 3B). Models testing 11 circulating linoleic acid metabolites [linoleoylethanolamide, dihomo-linolenoylcarnitine (C20:3n3 or 6), dihomo-linoleoylcarnitine (C20:2), linolenoylcarnitine (C18:3), linolenoylcarnitine (C18:2), dihomo-linolenoylcholine, linoleoylcholine, N-linoleoylglycine, dihomolinoleate (20:2n6), linoleate (18:2n6), and 1-linoleoyl-GPG (18:2)] as predictor variables (Supplemental Table S3) produced perfect discrimination at the distant time point (AUC: 1.0) and excellent discrimination at the proximate time point (AUC: 0.99, 95% CI: 0.98–1.0).

Figure 3.

Receiver operating characteristics curves for logistic regression models of metabolic predictor variables. Receiver operating characteristic curve of acylcholine metabolites to predict PAH vs. no PH at proximate time point (A) and distant time point (B) in the JHSC cohort. Receiver operating characteristic curve of linoleic acid metabolites to predict PAH vs. no PH at proximate time point (C) and distant time point (D) in the JHSC cohort. Sensitivity is plotted on the y-axis, and 1-specificity is plotted on the x-axis. JHSC, The Johns Hopkins Scleroderma Center; PAH, pulmonary arterial hypertension; PH, pulmonary hypertension.

Michigan Cohort

At both time points in our JHSC discovery cohort, we demonstrated lower circulating abundances of acylcholines containing oleic, linoleic, linolenic, and palmitic fatty acid (FA) residues in SSc-PAH subjects. Given that our discovery data set gave a strong signal for aberrant LCFA metabolism, we next examined the Michigan lipidomics data set for comparison with our findings. Because the Michigan lipidomics data set was derived from a different LC-MS platform than our JHSC data set, the individual metabolite features captured varied; thus, comparisons between the JHSC and Michigan cohorts were made on lipid class and lipid pathway levels. Demographic and clinical characteristics for the Michigan cohort are in Table 2. We first compared SSc subjects with borderline pulmonary pressure elevation to low-risk SSc subjects without PH. We found that LCFAs, including FAs corresponding to differentially abundant acylcholines in the discovery dataset, were higher in subjects with borderline pressures compared with low-risk subjects (Supplemental Table S4), including linoleic acid (FC 2.019, P value = 0.00105), arachidonic acid (FC 1.926, P value = 0.00768), eicosadienoic acid (FC 1.910, P value = 3.2 × 10⁻⁶), palmitic acid (FC 1.160, P value = 0.00067), and oleic acid (FC 1.637, P value = 6.42 × 10⁻⁵) (Fig. 1C). Multiple triglycerides were significantly lower in subjects with borderline pressures. A similar pattern of higher LCFAs with lower triglycerides was demonstrated when we analyzed SSc subjects with overt PAH compared with low-risk subjects without PH (Fig. 1D).

Table 2.

Demographic and clinical characteristics of Michigan overt PAH, borderline elevated pressure, and low-risk subjects

Demographics	Overt PAH	Borderline Elevated Pressure	Low-Risk
Subjects	20	22	22
Age, yr	63 (9)	61 (11)	56 (10)
Sex, n (% female)	17 (85)	19 (86)	20 (91)
Race, n (% white)	15 (75)	18 (82)	21 (95)
BMI, kg/m2	28 (6)	26 (5)	28 (5)
Laboratory values
NTproBNP, pg/mL	1,163 (2,632)	237 (199)	105 (59)
Hemodynamics
RAP, mmHg	9 (3)	7 (2)
mPAP, mmHg	34 (9)	23 (1)
PAWP, mmHg	14 (5)	12 (2)
PVR, Wood units	4 (3.3)	2.0 (0.6)
Cardiac output, L/min	5.5 (1)	5.6 (1.1)
Pulmonary function tests
FEV1, % predicted	87 (25)
FVC, % predicted	80 (24)
TLC, % predicted	95 (24)
DLCO, % predicted	47 (16)

Values are means (SD) unless otherwise specified. BMI, body mass index; DLCO, diffusing capacity of the lung for carbon monoxide; FEV1, forced expiratory volume in 1 s; FVC, forced vital capacity; mPAP, mean pulmonary arterial pressure; NTproBNP, brain natriuretic peptide pro-hormone N-terminal; PAH, pulmonary arterial hypertension; PAWP, pulmonary artery wedge pressure; PH, pulmonary hypertension; PVR, pulmonary vascular resistance; RAP, right atrial pressure; TLC, total lung capacity.

Pathway Analyses in JHSC and Michigan Cohorts

To contextualize individual metabolites identified as important features at the pathway level, we performed MSEA in each data set to assess overarching metabolic differences between the clinically defined groups. Enrichment of linoleic acid metabolism was observed at the distant time point in the metabolomes of JHSC patients who would develop PAH (Fig. 4), and in the metabolomes of Michigan patients with mildly elevated pressures (Fig. 5). Enrichment results corroborated abnormalities in FA metabolism more broadly, identifying other metabolite sets associated with β-oxidation of FA as enriched (Fig. 5). To contextualize our results in terms of FA carbon chain length, we plotted –log₁₀ (P value) for significant features in descending order, then listed corresponding metabolite class and number of carbons. Supplemental Fig. S1 demonstrates that for both cohorts, features with the most significant FC differences contain LCFAs.

Figure 4.

Enrichment analysis of metabolomes among the JHSC cohort across time. Dot plot depicting results of metabolite set enrichment analysis (MSEA) for SSc-PAH vs. SSc-no PH differences in JHSC cohort at proximate time point (A) and JHSC cohort at distant time point (B). Circle size represents the magnitude of the enrichment ratio (observed hits/expected hits), and the orange spectrum of color represents the significance of the association with more orange circles showing more highly significant associations. JHSC, The Johns Hopkins Scleroderma Center; PAH, pulmonary arterial hypertension; PH, pulmonary hypertension; SSc, systemic sclerosis.

Figure 5.

Enrichment analysis of metabolomes among Michigan cohort across time. Dot plot depicting results of metabolite set enrichment analysis (MSEA) for SSc-PAH vs. SSc-no PH differences in Michigan subjects with mildly elevated pressures (A) and Michigan subjects with overt PAH (B). Circle size represents the magnitude of the enrichment ratio (observed hits/expected hits), and the orange spectrum of color represents the significance of the association with more orange circles showing more highly significant associations. PAH, pulmonary arterial hypertension; PH, pulmonary hypertension; SSc, systemic sclerosis.

Validation Cohorts

We next examined our models of linoleic acid metabolites and long-chain acylcholines as predictors of SSc-PAH in the JHPHP (Supplemental Table S5) and PVDOMICS (Supplemental Table S6) cohorts, which, along with the JHSC cohort, were analyzed using the Metabolon global untargeted platform and therefore captured the same individual metabolite features. Demographic and clinical characteristics are listed in Tables 3 and 4 for JHPHP and PVDOMICS cohorts, respectively. In the JHPHP cohort, the combination of acylcholine predictors demonstrated good discrimination of SSc-PAH, with AUC 0.84, 95% CI 0.69–0.94 (Supplemental Fig. S2A). The combination of linoleic acid predictors also provided good discrimination of SSc-PAH, with AUC 0.89, 95% CI 0.77–0.98 (Supplemental Fig. S2B). In the PVDOMICS cohort, acylcholine predictors provided good discrimination of SSc-PAH, with AUC 0.76, 95% CI 0.62–0.86 (Supplemental Fig. S3A); similarly, linoleic acid predictors provided good discrimination of SSc-PAH in PVDOMICS, with AUC 0.81, 95% CI 0.68–0.91 (Supplemental Fig. S3B).

Table 3.

Demographic and clinical characteristics of JHPHP SSc-PAH and SSc-No PH subjects

Demographics	SSc-PAH	SSc-No PH
Subjects	24	14
Age, yr	62 (11)	58 (14)
Sex, n (% female)	21 (88)	11 (79)
Race, n (% white)	18 (75)	11 (79)
BMI, kg/m2	27 (6)	27 (7)
Laboratory values
proBNP, pg/Ml	732 (964)	267 (275)
Hemodynamics
RAP, mmHg	7 (4)	4 (1)
mPAP, mmHg	35 (12)	18 (4)
PAWP, mmHg	9 (4)	9 (3)
PVR, Wood units	6.3 (3.9)	2 (1.4)
Cardiac output, L/min	4.6 (1.1)	5.1 (1.2)
Cardiac index, L/min/m2	2.6 (0.6)	2.7 (0.5)

Values are means (SD) unless otherwise specified. BMI, body mass index; JHPHP, The Johns Hopkins Pulmonary Hypertension Program; mPAP, mean pulmonary arterial pressure; PAH, pulmonary arterial hypertension; PAWP, pulmonary artery wedge pressure; PH, pulmonary hypertension; proBNP, brain natriuretic peptide pro-hormone; PVR, pulmonary vascular resistance; RAP, right atrial pressure; SSc, systemic sclerosis.

Table 4.

Demographic and clinical characteristics of PVDOMICS SSc-PAH and SSc-No PH subjects

Demographics	SSc-PAH	SSc-No PH
Subjects	62	19
Age, yr	65 (10)	67 (8)
Sex, n (% female)	45 (72)	17 (89)
Race, n (% white)	50 (81)	16 (84)
BMI, kg/m2	26 (5)	26 (6)
Laboratory values
proBNP, pg/mL	2,105 (4,721)	412 (1,097)
Hemodynamics
RAP, mmHg	6 (5)	4 (2)
mPAP, mmHg	38 (11)	17 (3)
PAWP, mmHg	10 (5)	7 (3)
PVR, Wood units	6 (4)	2 (1)
Cardiac output, L/min	5.1 (1.9)	5.1 (1.9)
Cardiac index, L/min/m2	2.8 (1)	2.8 (1)
Pulmonary function tests
FEV1, % predicted	76 (16)	97 (22)
FVC, % predicted	77 (18)	98 (20)
DLCO, % predicted	35 (13)	64 (21)

Values are means (SD) unless otherwise specified. BMI, body mass index; DLCO, diffusing capacity of the lung for carbon monoxide; FEV1, forced expiratory volume in 1 s; FVC, forced vital capacity; mPAP, mean pulmonary arterial pressure; PAH, pulmonary arterial hypertension; PAWP, pulmonary artery wedge pressure; PH, pulmonary hypertension; proBNP, brain natriuretic peptide pro-hormone; PVDOMICS, Pulmonary Vascular Disease Phenomics; PVR, pulmonary vascular resistance; RAP, right atrial pressure; SSc, systemic sclerosis.

Metabolite Meta-Analysis

As a final analytic step, we analyzed FC differences for all individual features found in common across the four datasets produced by the Metabolon platform. We performed direct merging of datasets as well as integration of datasets with calculation of P values by Stouffer’s method, which incorporates weights based on sample sizes of the individual datasets. Eighty-eight features demonstrated significant FC differences between SSc-PAH and SSc across all four datasets (Supplemental Table S7). Among these 88 features were the acylcholines docosahexaenoylcholine [combined log (FC) –0.54638, P value = 0.005], palmitoloelycholine [combined log(FC) –0.4109, P value = 0.036], and dihomo-linolenoyl-choline [combined log(FC) –0.39, P value = 0.044]. Linoleic acid metabolites included linoleolycarnitine (18:2) [combined log (FC) 0.49, P value = 0.009] and dihomo-linolenoyl-choline. Other lipid species with LCFA residues that demonstrated significant FC differences across the datasets were 1-palmitoyl-GPG (16:0) and arachidonoylcarnitine (20:4).

DISCUSSION

Our results show that signatures of abnormal LCFA metabolism, particularly abnormal linoleic acid metabolism, distinguish SSc-PAH patients from those without PH across the spectrum of PAH evolution. Such a signature is detectable at least 2 yr before PAH diagnosis in our JHSC cohort and is present in the Michigan cohort in subjects with only mildly elevated pulmonary pressures at RHC. Similar lipid aberrations are observed in patients with overt PAH at RHC from both Hopkins and Michigan. The various cohorts included in our study serve complementary purposes: the JHSC cohort captures metabolism before PAH diagnosis, the Michigan cohort captures metabolism in early, mild disease, and the JHPHP and PVDOMICS cohorts capture overt PAH. In all instances, abnormal linoleic acid metabolism, at the pathway level, distinguishes the presence of, or impending development of, pulmonary vasculopathy. Furthermore, across the three datasets generated by Metabolon’s LC-MS platform (JHSC, JHPHP, and PVDOMICS), specific acylcholine features with LCFA residues robustly distinguish SSc-PAH from SSc without PH at different stages of the disease.

Linoleic acid is a polyunsaturated omega-6 fatty acid that is considered an essential fatty acid. Linoleic acid is the metabolic precursor of arachidonic acid (AA), which can be converted into various eicosanoids, prostaglandins, leukotrienes, and other bioactive lipid mediators. Prostaglandins are derived from AA through the cyclooxygenase pathway and play a crucial role in vascular biology. Prostacyclin (PGI2) is a prostaglandin with vasodilatory and antiproliferative effects (30). Its production is known to be reduced in PAH, and therapies aimed at increasing prostacyclin levels have been used to treat PAH for decades (31). Other oxidized lipids downstream of linoleic acid and AA metabolism, such as hydroxyeicosatetraenoic acids (HETEs) and hydroxyoctadecadienoic acids (HODEs), are known to directly impact pathogenic pulmonary vascular remodeling (32). Furthermore, several LCFAs, including linoleate, are differentially abundant in the human PAH RV, though the pathobiological mechanisms underlying this finding remain undefined (15).

Eicosanoid metabolites were recently examined at high resolution in 482 participants with chronic dyspnea undergoing invasive hemodynamic assessment; 200 were found to have resting PH, defined broadly as mPAP >20 mmHg (33). Forty-eight eicosanoids were associated with PH, including prostaglandin (11β-dhk-PGF2α), linoleic acid, arachidonic acid, and their derivatives. Many unnamed eicosanoids (identified only by mass/charge ratios) demonstrated associations with hemodynamic measurements, and 14 PH-associated eicosanoids demonstrated gradients across the pulmonary vasculature (e.g., differential abundance in samples measured from the pulmonary artery vs. the radial artery). These findings are generally in alignment with our results and bolster the likelihood of physiological relevance for these molecules. However, this study was limited to a cross-sectional analysis of associations in patients who were well-phenotyped hemodynamically but lacked clear clinical PH diagnoses. Furthermore, investigators studied eicosanoids only (as opposed to lipid intermediates more generally), and among the 200 patients with PH, only 14 (7.1%) had connective tissue disease (CTD).

Long-chain acylcholine metabolites were robustly predictive of SSc-PAH in our study. Acylcholines are molecules consisting of a choline moiety and a fatty acid (also known as an acyl group) linked by an ester bond. Acylcholines are being increasingly identified in metabolomics studies using ultra-high-performance liquid chromatography-tandem mass spectroscopy (34, 35), yet their endogenous functions have not been fully elucidated (36). These molecules are broadly classified as phospholipids, with most classified as phosphatidylcholines with a glycerol and phosphate backbone. Depending on the biological context, phosphatidylcholines can serve as sources of free fatty acids and other bioactive lipid molecules. Our results show that long-chain acylcholines are lower in PAH subjects compared with subjects without PAH, whereas corresponding free LCFAs are higher in PAH. This pattern suggests that long-chain acylcholines may represent a reservoir for FAs that becomes depleted as free LCFAs are used by a diseased RV and/or pulmonary circulation experiencing a shift in fuel substrates. We have previously shown that in both connective tissue disease-associated PAH and IPAH, acylcholines are taken up across the pulmonary circulation, with significantly lower concentrations when sampled from the wedged position compared with the mixed venous position (24). Conversely, free FAs were released from the pulmonary circulation in our prior study, with significantly higher concentrations in the wedged position compared with the mixed venous position. Taken together, this constellation of findings supports the possibility that acylcholines may function as an endogenous source of LCFAs for meeting increased metabolic demands of the right ventricular-pulmonary vascular circuit in PAH.

Our previous study also demonstrated the release of choline itself across the pulmonary circulation, with an 11% step-up in the wedged position (vs. mixed venous) in CTD-PAH patients, and a 13% step-up in IPAH patients (24). Phosphatidylcholines are synthesized from choline and other constituent precursors in the liver, with the liver dependent on a source for choline (37). Indeed, deficiencies in dietary choline have been linked to liver dysfunction and liver disease in humans (38). The observations from this study and our previous study, taken together, further corroborate a liver-lung axis in PAH that is becoming increasingly evident. It is well established that patients with PAH develop liver dysfunction, generally via congestive hepatopathy (39). Conversely, patients with liver cirrhosis can develop PAH (e.g., portopulmonary hypertension), which is occasionally resolved with liver transplantation (40). Portopulmonary hypertension liver tissue demonstrates a pattern of differential gene expression that is well aligned with pathways known to be dysregulated in PAH, including differential expression of BMPR2 and genes involved in estrogen and TGF-β signaling (41). Furthermore, our group has previously demonstrated metabolic interactions across tissue beds, including lung-liver cross talk and interactions with pulmonary hypertensive phenotypes, in the Sugen hypoxia animal model of PH (42). Future studies are needed to comprehensively examine the possibility of altered lipid metabolism in other metabolically active sites in PAH, such as the liver or skeletal muscles, as well as the potential for contributions to systemic lipid metabolism from the gut microbiome, which is also implicated in PAH (43–45).

Our results show that a parsimonious combination of circulating long-chain acylcholines discriminates subjects with SSc-PAH from SSc patients without PH across multiple cohorts. In our JHSC surveillance cohort, circulating acylcholines predicted SSc-PAH ∼2 yr before patients received a clinical PAH diagnosis. Circulating linoleic acid metabolites also provided good discrimination of SSc-PAH across all cohorts. However, the relatively modest sample sizes across our cohorts, representative of population sampling in a rare disease, raise questions of model overfitting and generalizability to other forms of PAH, questions we cannot address with the data presently available. We therefore cannot draw firm conclusions about the biomarker utility of these metabolites. Additional study is required to examine absolutely quantified molecules in larger cohorts comprising diverse PAH subtypes to better understand the predictive power of these multimetabolite models and, if warranted, to determine appropriate thresholds for discrimination.

Our study does not shed light on whether the phospholipid abnormalities that we demonstrate evolve alongside disease development, or represent pre-existent abnormalities in metabolism that are intrinsic to individuals and confer disease susceptibility. In subjects who developed PAH while under surveillance, there were no FC reductions in acylcholine concentrations from the distant to proximate time point. This observation stands in contrast to several amino acid metabolites, bile acids, polyamines, and nucleotides shown to be higher in SSc-PAH, which exhibited FC increases over time as PAH evolved. It is provocative to speculate that these lipid aberrations may represent intrinsic metabolic risk factors that are modifiable, via metabolically targeted medical therapies or perhaps dietary interventions, to reduce PAH risk. Future prospective studies will be required to address these questions.

An important strength of our study is our use of a nested case-control design within the larger longitudinal JHSC surveillance cohort. This design allows for the assessment of temporality and examination of metabolic profiles before PAH diagnosis in this at-risk group. Our work is strengthened by the use of complementary cohorts representative of SSc-PAH across the spectrum of disease, with subjects captured prediagnosis in the JHSC surveillance cohort and with only mildly elevated pressures in the Michigan cohort. Another important study strength is our ability to validate the main findings in additional cohorts assayed using the same LC-MS platform, including external validation in the multicenter PVDOMICS cohort. The very good AUCs calculated from our metabolite models rival those of clinical prediction tools that require complex multiparameter assessment across different testing modalities (46, 47).

Limitations of the present study include its modest sample size, which derives from the rarity of SSc and PAH. For the JHSC surveillance cohort, serum was obtained by convenience sampling, and as such, there is interindividual variation in time intervals between blood draws and diagnosis and limited contemporaneous clinical data. Metabolon’s untargeted metabolomics platform provides only relatively quantified metabolite abundances, limiting our ability to examine thresholds, and calibrate models across cohorts.

In sum, metabolic differences exist in SSc patients who will develop PAH up to 2 yr before clinical diagnosis, including notable abnormalities in LCFA metabolism, particularly linoleic acid metabolism. Future rigorous biomarker studies are needed to test the discrimination and calibration of combinations of LCFA-containing metabolites identified as important for SSc-PAH prediction in this study. Moving forward, high-resolution lipidomics studies may be useful for further dissecting abnormal lipid metabolism in PAH and identifying specific LCFAs with high biomarker potential. Reversal or remediation of lipidomic aberrations could be considered as novel therapeutic approaches worthy of future investigation.

DATA AVAILABILITY

Data are publicly available at the NIH Common Fund’s National Metabolomics Data Repository website (Project ID PR000551; https://doi.org/10.21228/M8FH50). Additional data are available from the corresponding authors upon reasonable request.

SUPPLEMENTAL MATERIAL

Supplemental Tables S1–S7 and Supplemental Figs. S1–S3: https://doi.org/10.6084/m9.figshare.25511572.v2.

GRANTS

The Johns Hopkins Scleroderma Center Research Registry and Biorepository are supported by NIH/National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) P30 AR070254, the Johns Hopkins inHealth initiative, the Donald B. and Dorothy L. Stabler Foundation, the Nancy and Joachim Bechtle Precision Medicine Fund for Scleroderma, the Sara and Alex Othon Fund, and the Manugian Family Scholar. C. E. Simpson, A. A. Shah, L. K. Hummers, S. C. Mathai, P. M. Hassoun, and R. L. Damico report research grants from the National Institutes of Health. A. A. Shah reports support from NIH/NIAMS K24 AR080217. M. Naranjo reports funding from 1F32HL159917-01A1. C. E. Simpson reports research grants from the National Scleroderma Foundation and the Pulmonary Hypertension Association. S. C. Mathai reports research grants from the U.S. Department of Defense.

DISCLOSURES

L. K. Hummers reports clinical trial grants from Prometheus, Cumberland Pharmaceuticals, Mitsubishi Tanabe, Horizon Pharmaceuticals, Arena Pharmaceuticals, Medpace LLC, and Kadmon Corporation that are unrelated to the present work. A. A. Shah reports clinical trial grants from Eicos Sciences, Arena Pharmaceuticals, Kadmon Corporation, and Medpace LLC that are unrelated to the present work. S. C. Mathai reports fees from Actelion, United Therapeutics, Janssen, Merck, and Clinical Viewpoints. R. L. Damico has received payments for expert testimony concerning unrelated matters. S. C. Mathai has served on an Advisory Board for Bayer and reports a leadership/fiduciary role with the Patient-Centered Outcomes Research Institute. P. M. Hassoun serves on a scientific advisory steering committee for MSD, an activity unrelated to the current work.

AUTHOR CONTRIBUTIONS

S.C.M., R.L.D., and C.E.S. conceived and designed research; J.C.C., T.T., M.N., A.W., L.K.H., A.A.S., K.S., S.H.V., S.C.M., P.M.H., R.L.D., and C.E.S. analyzed data; J.C.C., T.T., M.N., A.W., L.K.H., A.A.S., K.S., S.H.V., S.C.M., P.M.H., R.L.D., and C.E.S. interpreted results of experiments; J.C.C. and C.E.S. drafted manuscript; J.C.C., T.T., M.N., A.W., L.K.H., A.A.S., K.S., S.H.V., S.C.M., P.M.H., R.L.D., and C.E.S. edited and revised manuscript; J.C.C., T.T., M.N., A.W., L.K.H., A.A.S., K.S., S.H.V., S.C.M., P.M.H., R.L.D., and C.E.S. approved final version of manuscript.