One hundred and seventy subjects were recruited from the Cardiac Aging Study, a prospective study initiated in 2014 that examines characteristics and determinants of cardiovascular function in older adults (20) without known cardiovascular disease. All participants had no self-reported history of physician-diagnosed cardiovascular diseases (such as coronary heart disease, atrial fibrillation), stroke, or cancer. The study adhered to the principles outlined in the Declaration of Helsinki, and the SingHealth Centralised Institutional Review Board approved the study protocol. Participants provided written informed consent upon enrollment.

All participants were examined and interviewed during a single study visit. They completed a standardized questionnaire that included medical history and coronary risk factors. Hypertension was defined by the current use of antihypertensive drugs or physician-diagnosed hypertension. Diabetes mellitus was defined by the current use of anti-diabetic agents or physician-diagnosed diabetes mellitus. Dyslipidemia was defined by the current use of lipid-lowering agents or physician-diagnosed dyslipidemia. Smoking history was categorized as ever smokers (former or current smoking) or never smokers. Sinus rhythm status was ascertained through a resting electrocardiogram. Clinical data were obtained on the same day as the assessment of CMR imaging and serum collection.

A validated non-exercise prediction model was employed to estimate peak oxygen uptake, VO₂ (ml/kg/min) (21). The physical activity questionnaire included the frequency of exercise, length of time, and intensity for each workout. This model is closely linked to specific measures of cardiovascular structure and function (22).

Cine CMR scans were performed using a balanced fast field echo sequence. All subjects were imaged on a 3 T magnetic resonance imaging system (Ingenia, Philips Healthcare, The Netherlands). The CMR acquisition protocol and typical parameters were published previously (23). Standard end-expiratory breath-hold cine images were obtained using a steady-state free precession pulse sequence, retrospective electrocardiographic gating, and a typical temporal resolution of 30 frames per cardiac cycle. These images were acquired in the PA bifurcation, RV outflow tract (RVOT), coronal RVOT, and RV 3-chamber views (Figure 1A). Additionally, end-expiratory breath-hold cine images were acquired in multi-planar long-axis views, including 2-, 3-, and 4-chamber views.

PA GLS was assessed by one operator (H.Z.Z.) using an in-house semi-automatic algorithm (18) while remaining blinded to the clinical characteristics of participants and other CMR measurements. The motion of the PA bifurcation and the pulmonary valve annulus was automatically tracked over the cardiac cycle in PA bifurcation, RVOT, coronal RVOT, and RV 3-chamber views (Figure 1A). The trajectory of the centroid derived from six pulmonary annular points was obtained. The distance (L) from PA bifurcation to the centroid was calculated throughout the cardiac cycle. Longitudinal strain at a time point (t) relative to the initial time point (time 0) at end-diastole was calculated as 100 × (L_t–L₀)/L₀ (Figure 1B), and the PA GLS was defined as the maximal absolute strain value (Figure 1C). Details on the measurement of PA GLS can be found in the Supplementary Material.

Reproducibility was assessed in 20 randomly selected subjects. Inter-observer variability was evaluated by two independent operators (H.Z.Z. & S.L.), while intra-observer variability was assessed twice, with an interval of one month, by the first operator (H.Z.Z.). Correlation and Bland-Altman analyses were conducted to investigate intra- and inter-observer agreement.

RV function was assessed by cardiologists blinded to other CMR measurements. The endocardium of the RV was automatically tracked on short-axis cine images at end-systole and end-diastole to obtain RV end-diastolic volume (EDV) and end-systolic volume (ESV), and to calculate stroke volume (SV) and ejection fraction (EF). Using cine 4-chamber CMR images, the RV endocardium was tracked by one reader (X.D.Z.), who was blinded to all participant characteristics. RV GLS was automatically obtained using dedicated, validated QStrain software (Version 2.0, Medis BV, Leiden, The Netherlands).

Clinical characteristics are presented as mean and standard deviation (SD) for continuous data and frequency and percentage for categorical data. Cardiovascular risk factor 2 (CVRF2) was defined as the presence of any two or more cardiovascular risk factors (hypertension, dyslipidemia, ever smoked).

The association between PA GLS and clinical risk factors was assessed using linear regression. All clinical risk factors that showed an association with PA GLS at P < 0.05 in univariate analysis were entered into a multiple linear regression analysis that related PA GLS to significant clinical risk factors, adjusting for metabolite effects.

The identification of metabolites associated with PA GLS occurred in two steps. First, simple linear regression with PA GLS was conducted to identify significant metabolites (P < 0.05). Second, multiple linear regression was performed on each metabolite associated with PA GLS (P < 0.05 in univariate analysis), adjusting for the effects of identified significant clinical risk factors.

Subjects were assigned to groups with higher or lower than average PA GLS—i.e., “high PA GLS” and “low PA GLS”, respectively—based on sex-specific mean values (16.3% for females and 16.1% for males). A heatmap, visualizing the normalized intensities of metabolites, was generated using Mass Profiler Professional software (Agilent, USA).

All statistical analyses were conducted using STATA 15 (College Station, Texas, USA). For all analyses, a two-tailed P-value of <0.05 was considered significant.

This study lays the groundwork for future clinical applications of PA GLS in assessing PA stiffness. The correlation between PA GLS and clinical cardiovascular risk factors aligns with previous studies on PA stiffness (11, 26).

Age has been positively associated with PA PWV (11) and PA pressures (2, 5) in healthy subjects. Similarly, our study demonstrated that, like PWV and pressure, PA GLS decreased with age in a community cohort of older adults, indicating that aging is associated with longitudinal length changes in the PA.

Elevated heart rate is another risk factor associated with adverse cardiovascular events (27). In a porcine model, PA pressure was found to increase with heart rate, which was attributed to increase in flow and/or downstream flow resistance (28). A study of patients with pulmonary hypertension showed positive linear relationship between mean PA pressure and heart rate (29). Similar effects of heart rate on atrial distensibility were observed in earlier studies (30–32). PA distensibility was lower among patients with left heart disease and pulmonary hypertension vs. no pulmonary hypertension, and negatively correlated with heart rate (33). PA stiffness, assessed by the pulse pressure to stroke volume index ratio, has been shown to increase with heart rate (r²= 0.15), resulting in an increased risk of incident idiopathic PAH (26). Our study demonstrated an inverse association between PA GLS and heart rate (r = −0.24), which is consistent with the literature.

Smoking and dyslipidemia are crucial atherogenic risk factors. Smokers with chronic obstructive pulmonary disease often exhibit extensive atherosclerosis, leading to arterial stiffness (34, 35). Studies have shown that cigarette smoke induces increased wall stiffness in the PA of rats (36). Additionally, arterial stiffness has been directly associated with a 7-year increase in high-density lipoprotein, low-density lipoprotein, and triglycerides (37). Our study found that smoking and dyslipidemia were negatively associated with PA GLS, despite a small number of participants being at risk for dyslipidemia due to smoking.

We integrated advanced CMR imaging with serum metabolomic signals to evaluate clinical features of PA stiffness and metabolite patterns among this aging cohort. GLS assessed stiffness representing the PA trunk, while serum metabolomics distinguished older adults with high vs. low PA stiffness. In our study, we observed an association between PA GLS and circulating amino acids and acylcarnitines.

Both alanine and proline were negatively associated with PA GLS in both univariate analysis and multivariable analysis adjusted for clinical CVRF2. Alanine, a non-essential amino acid, is strongly linked to glucose metabolism. It plays a prominent role in the glucose-alanine shuttle between muscle and liver. A small percentage of individuals older than 85 years and the overall population have reported echocardiographic signs suggestive of pulmonary hypertension, indicating elevated PA stiffness (38, 39). Studies have shown that alanine is positively correlated with the pulmonary arterial medial thickness index—a histologic marker of PAH (16), with higher levels in blood samples from the PA of systemic sclerosis patients with PAH (17). Increased alanine levels in these studies were attributed to reduced tissue alanine aminotransferase levels (17). Proline contributes to pulmonary arterial remodeling in PAH rats (40). Rafikova et al. found high circulating levels of alanine and proline in early-stage PAH patients (41).

Long-chain acylcarnitines increase with age in healthy individuals (42). Our previous study showed that medium- and long-chain acylcarnitines were independently associated with arterial stiffness (20), similar to left atrial strain patterns obtained via CMR (23). This aligns with other studies in patients with pulmonary vascular disorders that observed higher levels of circulating long-chain acylcarnitines in patients with PAH, including idiopathic PAH, chronic thromboembolic pulmonary hypertension, and pulmonary hypertension associated with left heart disease (43, 44). By demonstrating an association between long-chain acylcarnitines and PA strain in otherwise healthy adults, our observations suggest that such a metabolic underpinning may be present upstream in the absence of disease. These findings may support broader validation in other non-disease cohorts, providing strategies for screening or charting pulmonary vascular diseases.

Overall, our integrated data provide a fresh approach to phenotyping human cohorts with omics (45), allowing more in-depth characterization of pulmonary arterial function. In addition to omics data, other strengths of our study include the prospective nature of participant recruitment, reducing biases related to recall while facilitating the simultaneous acquisition of clinical, CMR, and metabolomics data. Our study was prespecified to characterize older adults without clinical cardiovascular disease; therefore, circulating biomarkers represent the aging process rather than disease processes.

This study has limitations. Firstly, we did not acquire velocity-encoded flow measurements at the pulmonary trunk in this study, and are thus neither able to determine pulmonary vascular resistance noninvasively nor perform computational hemodynamic modeling for estimation of PA pressure. Secondly, compared to the current targeted approach, an untargeted metabolomics approach employing more metabolites may broaden the metabolic maps associated with PA GLS. While our targeted approach allowed precise quantification of identified metabolites, untargeted approaches could uncover pathways represented by these metabolites. Thirdly, all measurements and clinical variables originated from a single cross-sectional evaluation. Future studies may find it interesting to associate metabolomic findings at baseline with prospective changes in PA GLS and outcome data. Fourthly, while we did not explicitly exclude participants with preexisting respiratory disease, the calculated VO₂ measurements of participants enrolled into our current study were generally within normal ranges, which suggests that the impact of clinically significant respiratory disease is likely to be low in this community cohort. Fifthly, the effect of physical activity on cardiorespiratory status impacting PA GLS measurements is uncertain in the studied subjects, although aerobic capacity was generally satisfactory among the subjects. Lastly, non-fasting serum samples (used in our study) may introduce analytic differences compared to other cohorts that use fasting samples. However, our practice aligns with other cohort studies that found fasting did not contribute to variability in most metabolite measurements (46).