Preacinar Arterial Dilation Mediates Outcomes of Quantitative Interstitial Abnormalities in the COPDGene Study

Abstract

Rationale

Quantitative interstitial abnormalities (QIAs) are a computed tomography (CT) measure of early parenchymal lung disease associated with worse clinical outcomes, including exercise capacity and symptoms. The presence of pulmonary vasculopathy in QIAs and its role in the QIA–outcome relationship is unknown.

Objectives

To quantify radiographic pulmonary vasculopathy in QIAs and determine whether this vasculopathy mediates the QIA–outcome relationship.

Methods

Ever-smokers with QIAs, outcomes, and pulmonary vascular mediator data were identified from the Genetic Epidemiology of COPD (COPDGene) study cohort. CT-based vascular mediators were right ventricle–to–left ventricle ratio, pulmonary artery–to–aorta ratio, and preacinar intraparenchymal arterial dilation (pulmonary artery volume, 5–20 mm² in cross-sectional area, normalized to total arterial volume). Outcomes were 6-minute walk distance and a modified Medical Council Research Council Dyspnea Scale score of 2 or higher. Adjusted causal mediation analyses were used to determine whether the pulmonary vasculature mediated the QIA effect on outcomes. Associations of preacinar arterial dilation with select plasma biomarkers of pulmonary vascular dysfunction were examined.

Measurements and Main Results

Among 8,200 participants, QIA burden correlated positively with vascular damage measures, including preacinar arterial dilation. Preacinar arterial dilation mediated 79.6% of the detrimental impact of QIA on 6-minute walk distance (56.2–100%; P < 0.001). Pulmonary artery–to–aorta ratio was a weak mediator, and right ventricle–to–left ventricle ratio was a suppressor. Similar results were observed in the relationship between QIA and modified Medical Council Research Council dyspnea score. Preacinar arterial dilation correlated with increased pulmonary vascular dysfunction biomarker levels, including angiopoietin-2 and N-terminal brain natriuretic peptide.

Conclusions

Parenchymal QIAs deleteriously impact outcomes primarily through pulmonary vasculopathy. Preacinar arterial dilation may be a novel marker of pulmonary vasculopathy in QIAs.

At a Glance Commentary

Scientific Knowledge on the Subject

Quantitative interstitial abnormalities (QIAs) are measures of early fibrotic and inflammatory injury on computed tomography chest imaging and are associated with increased symptoms and mortality. The relationship of QIAs with the pulmonary vasculature is not known, nor are the physiologic mechanisms underpinning the QIA–outcome relationship. We sought to detect and quantify radiographic pulmonary vasculopathy and its clinical impact in a multicenter cohort with QIAs.

What This Study Adds to the Field

QIA negatively impacts clinical outcomes through associated pulmonary vasculopathy, as opposed to the direct effect of the parenchymal disease itself. We establish that preacinar intraparenchymal arterial dilation on computed tomography chest imaging appears to be a novel, sensitive marker of pulmonary vasculopathy that is corroborated by biochemical evidence of pulmonary vascular dysfunction.

Because of lung inflammation and remodeling, a spectrum of parenchymal diseases may develop in smokers that extends from destruction causing emphysema to fibrosis resulting in interstitial lung disease (ILD). With computed tomography (CT), early or mild ILD may be visually assessed as interstitial lung abnormalities (ILAs). Early fibrotic and inflammatory changes can also be objectively identified and measured on CT and are termed quantitative interstitial abnormalities (QIAs) (1–3). As an automated aggregate of interstitial features present in the lung parenchyma, QIA is notably more sensitive for the detection of interstitial changes in the parenchyma than features that can be visually assessed as ILA. Like the presence of ILA, the extent of QIAs has been shown to correlate with poor clinical outcomes, including increased exercise limitations, symptoms, and mortality in smokers, as previously described in detail (1–8). In people with QIAs, the physiologic mechanisms underlying these relationships have not been explored.

There is growing recognition that pulmonary vasculopathy begins early in the course of parenchymal lung damage and disease (9). Vascular abnormalities develop well before pulmonary arterial pressures reach the current hemodynamic threshold for pulmonary hypertension (PH), correlate with abnormal pulmonary blood flow patterns, and can be detected by circulating proteins associated with activated endothelium (9–11). Distal microvascular destruction may lead to upstream dilation of structures, including the preacinar arteries, main pulmonary artery (PA), and right ventricle (RV), and the occurrence of the latter two features is well-established in PH associated with ILD (12–14). Our group at Brigham and Women’s Hospital has previously shown that distal vascular changes exist even in mild chronic obstructive pulmonary disease (COPD) (15). The presence of radiographic pulmonary vasculopathy in QIAs, however, and its potential role as an intermediary between QIAs and clinical outcomes are not known.

In this study, we sought to detect and quantify radiographic pulmonary vasculopathy and its clinical impact in ever-smokers with mild parenchymal lung disease, utilizing three CT-based vascular measures: RV–to–left ventricle (RV/LV) volume ratio, PA–to–aorta (PA/Ao) ratio, and preacinar intraparenchymal arterial dilation. We hypothesized that 1) vascular damage exists in people with QIAs and is most accurately quantified by preacinar intraparenchymal arterial dilation; that 2) pulmonary vascular dysfunction mediates, or links, QIAs with poor clinical outcomes; and that 3) radiographic vasculopathy correlates with circulating protein biomarkers of pulmonary vascular dysfunction. To explore these hypotheses in detail, we leveraged the radiologic, epidemiologic, and clinical data collected from participants in the Genetic Epidemiology of COPD (COPDGene) study.

Some of these findings were presented in abstract form at the 2024 Pulmonary Vascular Research Institute Annual Congress (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11094575/).

Methods

Data Source and Study Population

The COPDGene study is a prospective observational cohort designed to study phenotypes and genetic risk factors for COPD. The COPDGene study enrolled over 10,000 non-Hispanic White and Black participants ages 45 to 80 years old, with at least 10 pack-years of tobacco smoke exposure across 21 clinical centers (16). Subjects with centrally adjudicated ILD and bronchiectasis were excluded. As part of the COPDGene study protocol, participants underwent pre- and postbronchodilator spirometry; noncontrast high-resolution CT chest scans at full inspiration and relaxed exhalation; 6-minute walk distance (6MWD) testing; and symptom assessment at the baseline, or Phase 1, visit. The COPDGene study was approved by the institutional review boards of participating centers (Mass General Brigham IRB #2007P000554), and patients provided informed consent at enrollment.

In this study, clinical outcomes were defined as 1) exercise capacity as assessed by baseline 6MWD and 2) breathlessness as assessed using the modified Medical Council Research Council (mMRC) Dyspnea Scale (17, 18). Given that an mMRC Dyspnea Scale score of 2 or higher denotes the presence of clinically relevant dyspnea in people with chronic parenchymal lung disease, it was dichotomized by this threshold for analysis (19). In this study, we included ever-smokers from the COPDGene Phase 1 cohort with data for both clinical outcomes, QIA burden, and at least two potential pulmonary vascular mediators (Figure 1).

Figure 1.

CONSORT flow diagram showing patient selection from the Genetic Epidemiology of COPD (COPDGene) study. 6MWD = 6-minute walk distance; mMRC = modified Medical Research Council Dyspnea Scale score; QIA = quantitative interstitial abnormality; TLC_CT = TLC on computed tomography (in liters).

Pulmonary Vascular and Cardiac CT Quantification

Peripheral lung parenchymal disease causes microvascular damage and subsequent upstream structural dilation ranging from the small lung arteries to central cardiac structures (20–23). On the basis of their anatomic relation to the site of parenchymal disease or their established role in PH associated with ILD, we selected three potential pulmonary vascular metrics to study as mediators: RV/LV ratio, PA/Ao ratio, and preacinar intraparenchymal arterial dilation. Sample reconstructions of the pulmonary vascular mediators are shown in Figure 2.

Figure 2.

Example of computed tomography (CT) reconstructions of pulmonary vascular mediators. (A) An automated reconstruction of cardiac chambers used to compute volumes of right (purple) and left (pink) ventricles. Right (light blue) and left (orange) atria are also included. (B) Ratio of pulmonary artery (21.6 mm) to aorta (26.5 mm) (PA/Ao). (C) The reconstructed pulmonary arterial vasculature from the right lung of a nonsmoking COPDGene participant. The preacinar arteries are displayed in yellow in the lung core (inner 80% of lung volume), with the remaining pulmonary tree displayed in blue. The vast majority of the vessels in the peel (the outer 20% of the lung) have a vascular cross-sectional area of less than 5 mm². Asc. Ao = ascending aorta; Desc. Ao = descending aorta; LA = left atrium; LPA = left pulmonary artery; LV = left ventricle; MPA = main pulmonary artery; RA = right atrium; RPA = right pulmonary artery; RV = right ventricle.

To quantify the central structural dilation metrics of the RV/LV and PA/Ao ratios, the heart was automatically fit with a statistical model that has been validated previously with cardiac magnetic resonance imaging and echocardiography (15, 24). The RV/LV ratio was calculated with RV and LV epicardial volumes, which correlate with end-diastolic volume measurements on cardiac magnetic resonance imaging. To calculate the PA/Ao ratio, maximal PA and aorta diameters were measured at the level of the PA bifurcation on axial CT images (13).

Preacinar arterial dilation is a measure of the central intraparenchymal arterial response to distal microvascular disease and was selected as the third mediator. The pulmonary vessels were reconstructed from the inspiratory CT, and the arteries and veins were identified using a deep-learning approach (25, 26). The PA volume between 5 and 20 mm² in a cross-sectional area (BV5–20) was calculated and normalized to total arterial blood volume (TBV; expressed as BV5–20/TBV) (27). Increased values of arterial BV5–20/TBV indicate greater preacinar intraparenchymal arterial dilation.

Parenchymal CT Quantification

On inspiratory CT chest imaging, the individual percentages of parenchymal lung tissue with QIAs and emphysema were classified and quantified as previously described (2, 28). In brief, the lungs were segmented, and a k-nearest neighbors classification algorithm based on local tissue histogram measurements and distance from pleural surface was used to label a particular tissue subtype (QIA [previously called interstitial features], emphysema, and normal) (2). On CT imaging, QIA is a composite measure of reticulations, honeycombing, centrilobular nodules, linear scar, nodular changes, subpleural line, and ground glass opacities. These features are summed and divided by the total lung volume on CT to arrive at a QIA percentage (2, 28, 29). Emphysema percentage was similarly quantified. (For further details on image quantification techniques, see the online supplement.)

Selected Biomarkers of Pulmonary Vascular Disease

We selected candidate biomarkers related to pulmonary vascular disease from a previously run dataset of a representative subset of participants from Phase 1 (n = 1,126) that used the SOMAscan proteomics platform of 1,305 plasma proteins (30). From this panel, we selected four circulating candidate biomarkers a priori for analysis on the basis of their potential role in pulmonary vascular disease. The first candidate, growth differentiation factor-15 (GDF-15), has been shown to be upregulated across the cardiopulmonary system, including in vascular endothelial cells from plexiform lesions of pulmonary arterial hypertension (PAH) and in epithelial cells in lung fibrosis (31–33). We further included two myocardial-derived proteins indicative of cardiac stress: N-terminal brain natriuretic peptide (NT-proBNP) and endostatin. Finally, the vascular endothelial marker angiopoietin-2 was also examined. These latter three proteins are elevated in PH compared with normal controls and correlate with hemodynamic derangements and outcomes (12, 31, 32, 34–36). Protein levels were reported in relative fluorescence units and analyzed as natural log-transformed values. (For additional SOMAscan details, see the online supplement.)

Statistical Analysis

Continuous variables are displayed as median (interquartile range; IQR), and categorical variables are shown as numbers (percentage). Linear regression models were used to evaluate the association of QIA burden with 6MWD and pulmonary vascular metrics. The relationship of QIA burden and mMRC score of 2 or higher was assessed with logistic regression. Logistic regression adjusted for QIA percentage was also used to examine the relationship of protein analyte levels with vascular measures. Residual plots were utilized to validate the model assumptions. Participants with missing data were removed without imputation. Statistical analysis was performed with SAS, Version 9.4. A two-sided P value <0.05 was considered significant.

Causal Mediation Analysis

Causal mediation analysis evaluates the mechanism linking an exposure and outcome (37, 38). We sought to understand whether parenchymal QIAs directly impact clinical outcomes or whether their effect is indirect through associated radiologic pulmonary vascular disease.

Causal mediation analysis derived from the counterfactual framework was performed (39). The exposure was QIA percentage, the mediator was the pulmonary vascular metric, and the outcome was 6MWD or a mMRC Dyspnea Scale score of 2 or higher. To view a directed acyclic graph including confounders, see Figure E1 in the online supplement. Analyses were adjusted for potential confounders including age (in years), sex, race, height (in meters), weight (in kilograms), postbronchodilator FEV₁ (in liters), postbronchodilator FEV₁/FVC, total lung capacity on CT (in liters), current smoking status, pack-year tobacco exposure history, supplemental oxygen requirement, and emphysema percentage (40). An interaction term describing the potential impact of the exposure–mediator relationship on the outcome was also included.

The total effect of QIAs on the clinical outcome is reported. It is shown decomposed into the natural direct effect (e.g., direct impact of parenchymal QIA on 6MWD) and the natural indirect effect (e.g., indirect impact of parenchymal QIAs on 6MWD through associated pulmonary vascular damage). The percentage mediated is expressed as the natural indirect effect divided by total effect, multiplied by 100. The 95% bias-corrected confidence intervals (CIs) were calculated by bootstrapping on the basis of 10,0000 resamples and were capped at 100% for single-mediator models (41). Causal mediation analysis was performed with PROC CAUSALMED in SAS, Version 9.4, using a normal or binomial distribution–based model, as appropriate. To study the potential effect of pulmonary vascular mediators in anatomic sequence extending from the distal to proximal vasculature, we performed a serial multiple mediation analysis with the SAS PROCESS macro (42). For further details on mediation analysis, including confounder selection and serial technique, see the online supplement. We followed the directions from A Guideline for Reporting Mediation Analyses of Randomized Trials and Observational Studies: The AGReMA Statement (40).

Results

There were 8,200 COPDGene participants eligible for inclusion, with a median age of 59.1 years (IQR = 52.1–66.6). Of the 8,200 participants, 3,775 (46%) were women (Figure 1 and Table 1). The postbronchodilator median FEV₁ was 2.27 L (IQR = 1.61–2.90), and the median FEV₁/FVC was 0.71 (IQR = 0.58–0.79). The median QIA burden was 4.67% (IQR = 2.99–7.48). The median 6MWD was 425.3 m (IQR = 343.8–496.5). Approximately 41% of patients had an mMRC Dyspnea Scale score of 2 or higher.

Table 1.

Baseline Characteristics, CT Measures, and Clinical Outcomes among Ever-Smokers in the COPDGene Study with Complete Data

Variable	Median (IQR) or n (%)
Clinical and epidemiologic characteristics
Age, yr, median (IQR)	59.1 (52.1–66.6)
Female sex, n (%)	3,775 (46.0)
Black race, n (%)	2,539 (31.0)
Height, m, median (IQR)	1.7 (1.6–1.8)
Weight, kg, median (IQR)	82.0 (70.0–95.4)
Body mass index, kg/m2, median (IQR)	28.1 (24.6–32.3)
Use of supplemental oxygen, n (%)	883 (10.8)
Current smoker, n (%)	4,239 (51.7)
Smoking pack-years for ever smokers, median (IQR)*	39.6 (27.1–55.0)
Postbronchodilator spirometry, median (IQR)
FEV1, L	2.3 (1.6–2.9)
FEV1, % predicted	81.2 (60.7–95.3)
FEV1/FVC	0.71 (0.58–0.79)
CT-derived parenchymal metrics, median (IQR)
TLCCT, L	5.4 (4.5–6.5)
QIA, %	4.67 (2.99–7.48)
Emphysema, %	1.95 (0.49–8.86)
CT-derived pulmonary vascular metrics, median (IQR)
Arterial BV5-20/TBV	0.23 (0.20–0.26)
PA/Ao ratio	0.83 (0.76–0.92)
RV/LV ratio	0.51 (0.44–0.57)
Clinical outcomes
6MWD, m, median (IQR)	425.3 (343.8–496.5)
mMRC Dyspnea Scale score ⩾2, n (%)	3,326 (40.6)

Definition of abbreviations: 6MWD = 6-minute walk distance; BV5-20/TBV = pulmonary artery vessel volume of 5–20 mm² in a cross-sectional area normalized to total arterial blood volume; COPDGene = Genetic Epidemiology of COPD; CT = computed tomography; IQR = interquartile range; mMRC = modified Medical Research Council; PA/Ao = pulmonary artery–to–aorta; QIA = quantitative interstitial abnormality; RV/LV = right ventricle–to–left ventricle.

N = 8,200. Missing data: arterial BV5-20/TBV = 1,035, PA/Ao ratio = 26, and RV/LV ratio = 856.

^*Total pack-years for ever-smokers, including former (median = 40.0 yr) and current tobacco users (median = 39.2 yr).

Correlation between QIA Percentage, Outcomes, and Pulmonary Vascular Metrics

QIA percentage correlated with a detrimental effect on both exercise capacity and symptom burden. In multivariable regression, there was a 2.0-m decline in 6MWD for every 1% increase in QIA (Table 2). There was a 6% increase in the odds of an mMRC Dyspnea Scale score of 2 or higher for each 1% increase in QIA burden (Table 2).

Table 2.

Associations between QIA Percentage, Clinical Outcomes, and Pulmonary Vascular Metrics

Outcome Measure	Associations between QIA Percentage and 6MWD
	Unadjusted Values		Adjusted Values*
	β Coefficient (95% CI)	P	β Coefficient (95% CI)	P
6MWD, m	−6.47 (−7.09, −5.86)	<0.001	−2.00 (−2.61, −1.39)	<0.001

	Odds Ratios for the Association between QIA Percentage and mMRC Dyspnea Scale Scores ⩾2 compared with Scores <2
Outcome Measure	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P
QIA, %	1.06 (1.05, 1.07)	<0.001	1.03 (1.02, 1.05)	<0.001

	Associations between QIA Percentage and Pulmonary Vascular and Right-Heart Metrics
Outcome Measure	β Coefficient (95% CI)	P	β Coefficient (95% CI)	P
Arterial BV5-20/TBV	0.005 (0.004, 0.005)	<0.001	0.004 (0.003, 0.004)	<0.001
PA/Ao ratio	0.004 (0.004, 0.005)	<0.001	0.002 (0.001, 0.002)	<0.001
RV/LV ratio	0.001 (0, 0.001)	0.019	0.001 (0, 0.002)	0.009

Definition of abbreviations: 6MWD = 6-minute walk distance; BV5-20/TBV = pulmonary artery vessel volume of 5–20 mm² in a cross-sectional area normalized to total arterial blood volume; CI = confidence interval; mMRC = modified Medical Research Council; PA/Ao = pulmonary artery–to–aorta; QIA = quantitative interstitial abnormality; RV/LV = right ventricle–to–left ventricle.

^*Values were adjusted for the following variables at Phase-1 enrollment: age, sex, race, height (in meters), weight (in kilograms), current smoking status, pack-years of tobacco exposure, postbronchodilator FEV₁ (in liters) and FEV₁/FVC, TLC_CT (in liters), emphysema percentage, use of supplemental oxygen, and scanner manufacturer.

Increasing QIA percentage was associated with dilation of pulmonary vascular structures (Table 2). The effect of QIA burden was strongest on preacinar intraparenchymal arterial dilation, less significant at the PA/Ao ratio, and weakest on the RV/LV ratio.

Dilated Pulmonary Vascular Structures Mediate the QIA–Outcome Relationship

We sought to explain and measure the mechanisms underlying the association of QIAs and worse clinical outcomes, with the hypothesis that pulmonary vascular damage—not parenchymal disease—is the major causal pathway.

In causal mediation analysis of the effect of QIAs on 6MWD, the mediator strength of pulmonary vascular metrics corresponded to the anatomic distance from the primary site of parenchymal disease. The RV/LV ratio had a suppressor effect, such that it mediated −1.81% of the effect of QIAs on 6MWD (95% CI = −4.73, −0.37; P = 0.050; Figure 3 and Table 3). The PA/Ao ratio accounted for a small but statistically significant 3.15% of the QIA–6MWD relationship (95% CI = 1.23, 7.05%; P = 0.009). Finally, the intraparenchymal measure of preacinar arterial dilation mediated 79.6% of the total impact of QIAs on 6MWD (95% CI = 56.2, 100; P < 0.001).

Figure 3.

Direct acyclic graphs (DAGs) displaying the relationship between exposure, outcome, and mediator. Left: the proposed DAGs for the outcome of 6-minute walk distance (6MWD). Right: the actual relationships determined through causal mediation analysis. The PM shows the percentage of the total effect mediated through the indirect pathway. The weight of the arrow approximates the amount of the effect of QIA on 6MWD through the respective pathway. BV5-20/TBV = arterial volume of vessels 5–20 mm² in cross-sectional area normalized to total arterial blood volume; LV = left ventricle; PA/Ao ratio = pulmonary artery–to–aorta ratio; PM = percentage mediated; QIA = quantitative interstitial abnormality; RV = right ventricle.

Table 3.

Adjusted Causal Mediation Analyses of Pulmonary Vascular Metrics in the Relationship of QIA Burden with 6MWD as a Continuous Variable

Variable	Arterial BV5-20/TBV		PA/Ao Ratio		RV/LV Ratio
	Estimate (95% CI)	P	Estimate (95% CI)	P	Estimate (95% CI)	P
Total effect	−2.15 (−2.98, −1.36)	<0.001	−1.83 (−2.51, −1.18)	<0.001	−1.96 (−2.67, −1.22)	<0.001
Natural direct effect	−0.44 (−1.26, 0.34)	0.266	−1.77 (−2.44, −1.11)	<0.001	−2.00 (−2.71, −1.25)	<0.001
Natural indirect effect	−1.71 (−2.01, −1.43)	<0.001	−0.06 (−0.11, −0.02)	0.004	0.04 (0.01, 0.08)	0.036
Percentage mediated	79.6 (56.2, 100)	<0.001	3.15 (1.23, 7.05)	0.009	−1.81 (−4.73, −0.37)	0.050
Percentage due to interaction	−0.35 (−1.29, 0.49)	0.360	0.21 (−0.40, 1.51)	0.540	0.04 (−0.22, 0.81)	0.739
Percentage eliminated	79.9 (56.6, 100)	<0.001	2.94 (0.88, 6.72)	0.014	−1.85 (−4.78, −0.36)	0.050

Definition of abbreviations: 6MWD = 6-minute walk distance; BV5-20/TBV =  pulmonary artery vessel volume of 5–20 mm² in a cross-sectional area normalized to total arterial blood volume; CI = confidence interval; PA/Ao = pulmonary artery–to–aorta; QIA = quantitative interstitial abnormality; RV/LV = right ventricle–to–left ventricle.

Values were adjusted for the following variables at Phase-1 enrollment: age, sex, race, height (in meters), weight (in kilograms), current smoking status, pack-years of tobacco exposure, postbronchodilator FEV₁ (in liters) and FEV₁/FVC, TLC_CT (in liters), emphysema percentage, use of supplemental oxygen, and scanner manufacturer. The model also included an interaction term between QIA burden and arterial BV5-20/TBV.

Similar results were seen in the causal mediation analysis of the effect of QIAs on mMRC Dyspnea Scale scores of 2 and higher (Table 4). The RV/LV ratio was not a significant intermediary variable, and the PA/Ao ratio was a weak mediator. Preacinar arterial dilation mediated approximately 50% of QIA effect on dichotomized mMRC score.

Table 4.

Adjusted Causal Mediation Analyses of Pulmonary Vascular Mediators in the Relationship of QIA Burden with mMRC Dyspnea Scale Scores ⩾2 compared with Scores <2

Variable	Arterial BV5-20/TBV		PA/Ao Ratio		RV/LV Ratio
	Estimate (95% CI)	P	Estimate (95% CI)	P	Estimate (95% CI)	P
Odds ratio total effect	1.04 (1.02, 1.07)	<0.001	1.03 (1.02, 1.05)	<0.001	1.03 (1.01, 1.05)	<0.001
Odds ratio natural direct effect	1.02 (1.00, 1.04)	0.023	1.03 (1.02, 1.05)	<0.001	1.03 (1.01, 1.05)	<0.001
Odds ratio natural indirect effect	1.02 (1.01, 1.03)	<0.001	1.00 (1.00, 1.00)	0.023	1.00 (1.00, 1.00)	0.138
Percentage mediated	48.7 (29.0, 90.0)	<0.001	3.26 (0.82, 8.73)	0.044	−1.52 (−6.08, 0.02)	0.167
Percentage due to interaction	−3.75 (−10.6, 0.67)	0.091	−0.11 (−4.36, 4.04)	0.941	0.62 (−1.03, 5.93)	0.548
Percentage eliminated	46.3 (27.3, 86.2)	<0.001	3.13 (0.16, 10.6)	0.124	−0.75 (−6.38, 1.67)	0.559

Definition of abbreviations: BV5-20/TBV = pulmonary artery vessel volume of 5–20 mm² in a cross-sectional area normalized to total arterial blood volume; CI = confidence interval; mMRC = modified Medical Research Council; PA/Ao = pulmonary artery–to–aorta; QIA = quantitative interstitial abnormality; RV/LV = right ventricle–to–left ventricle.

Values were adjusted for the following variables at Phase-1 enrollment: age, sex, race, height (in meters), weight (in kilograms), current smoking status, pack-years of tobacco exposure, postbronchodilator FEV₁ (in liters) and FEV₁/FVC, TLC_CT (in liters), emphysema percentage, use of supplemental oxygen, and scanner manufacturer. The model also included an interaction term between QIA burden and arterial BV5-20/TBV.

A serial multiple mediation analysis was performed to study the impact of the sequential pathway from distal (preacinar arterial dilation) to proximal (PA/Ao ratio) vasculature in the QIA–outcome relationship. The indirect path through preacinar arterial dilation alone remained the major mechanism linking QIA with worse clinical outcomes (see Tables E1 and E2). Conversely, there was minimal indirect effect of the sequential mediator path from preacinar arterial dilation through the PA/Ao ratio.

Sensitivity Analyses of Pulmonary Vascular Mediators

Sensitivity analyses of causal mediation models were performed to examine 1) mMRC Dyspnea Scale score as a continuous variable and 2) removal of the QIA–mediator interaction term. Preacinar arterial dilation continued to mediate a large proportion of the continuous mMRC Dyspnea Scale score (see Figures E2 and Table E3). Separately, the mediator strengths of the pulmonary vascular metrics were similar after the removal of the interaction term (see Tables E4 and E5).

Pulmonary Vascular Dysfunction Biomarkers Correlate with Radiographic Vasculopathy

There were 789 people with Phase-1 plasma protein analysis and CT reconstructions. A subset of 394 participants had a high QIA burden as defined by the median of ⩾4.48%.

The high-QIA subset was partitioned by the presence of preacinar arterial dilation defined by the median value of ⩾0.23 (versus absent when <0.23; n = 197 for each). The QIA percentage differed between subsets with present or absent preacinar arterial dilation: 7.10% (IQR = 5.75, 10.0) versus 6.49% (IQR = 5.32, 8.76), P = 0.004.

Among participants in the high-QIA subset, the presence of preacinar arterial dilation correlated with an increase in selected circulating biomarkers of endothelial and myocardial dysfunction, including angiopoietin-2, endostatin, GDF-15, and NT-proBNP (see Table E6). Within this subset, there was an increased likelihood of preacinar dilation with higher levels of pulmonary vascular dysfunction biomarkers, even after adjusting for QIA percentage (Figure 4).

Figure 4.

OR for the presence of preacinar dilation by pulmonary vascular biomarker level within the subgroup with high quantitative interstitial abnormality (QIA) percentage (defined as ⩾4.48%). Expressed per unit change of 1 in natural log-transformed relative fluorescence units. *Adjusted for continuous QIA percentage. CI = confidence interval; GDF-15 = growth differentiation factor-15; NT-proBNP = N-terminal brain natriuretic peptide; OR = odds ratios.

Discussion

In this analysis, we sought to determine the mechanism of reduced exercise capacity and breathlessness in smokers with mild parenchymal lung disease. As an automated radiographic measure of early parenchymal disease, QIA burden correlates with greater CT-based pulmonary vasculopathy and worse clinical outcomes. Importantly, causal mediation analysis suggested that the direct impact of the parenchymal QIAs themselves contributes minimally to this relationship. Instead, the indirect effect of associated pulmonary vascular disease was the major mechanism that linked QIAs with worse outcomes. Moreover, we observed that pre-acinar arterial dilation is a novel and sensitive CT marker of pulmonary vasculopathy in early lung disease. This radiographic feature—corroborated by the presence of protein biomarkers that are elevated with PH presence and worse outcomes—mediated the relationship between QIA burden and worse clinical outcomes to a stronger degree than traditional metrics used to screen for PH in lung disease, such as the PA/Ao and RV/LV ratios.

Despite the established association between QIA burden and worse outcomes, the culprit mechanisms underpinning this relationship are not well characterized (1, 3, 7, 29). Here, we demonstrate that, even after adjusting for factors such as supplemental oxygen use, the associated intraparenchymal vasculopathy—not the lung tissue damage itself—is the major factor underlying this relationship, suggesting that clinically relevant pulmonary vascular dysfunction exists on a spectrum that begins well before hemodynamic thresholds for PH are met (10, 43–45). Compared with the weak effects of PH screening metrics like increased RV/LV and PA/Ao ratios, preacinar arterial dilation was responsible for the majority of QIA-related reduction in 6MWD, with a similar, albeit lesser, role in clinically relevant breathlessness reflected by mMRC scores of 2 or higher (19).

Our results suggest that arterial dilation nearest to the site of injury is a critical arbitrator of the clinical impact of QIAs. Direct vessel injury, abnormal cytokine signaling, endothelial and perivascular dysfunction, and hypoxic vasoconstriction promote microvascular disease and upstream structural dilation. Our group showed that small arterial dilation occurs in PH associated with ILD, and previous histologic studies demonstrated an increase in the number and diameter of preacinar vessels in PAH (46–49). As it occurs in close anatomic proximity to lung parenchymal injury, it is possible that preacinar arterial dilation reflects the compensatory response to increased vascular resistance caused by damage to the lung tissue.

The utility of preacinar arterial dilation as a novel marker of pulmonary vasculopathy is further supported by a concurrent biologic protein signal of PH. In participants with high QIA burdens, the presence of preacinar arterial dilation correlated with higher levels of angiopoietin-2. Similar findings were seen with GDF-15, even after adjustment for QIA burden. Prior work has demonstrated that both biomarkers are increased in PAH compared with healthy control subjects, with expression primarily localized to the pulmonary vascular endothelial cells, and that high circulating levels correlate with worse outcomes (31, 32, 36). Similarly, preacinar dilation also correlated with increased NT-proBNP and endostatin levels that, although nonspecific for PH, are elevated in this population and associated with worse outcomes (12, 35).

Interestingly, we observed that the mediator strength of pulmonary vascular metrics in the relationship of QIA and outcomes followed a logical anatomic order, attenuating on the basis of the distance from the primary parenchymal site of disease. The PA/Ao ratio weakly mediated the clinical outcomes of QIA, and although preacinar arterial dilation had a strong impact in isolation, its effect was minimal when serially mediated through the proximal vasculature. This decrement in mediation is not surprising, as radiologically detectable changes in the PA caliber likely require significant distal vascular destruction, a process only found in the most advanced lung disease. Although statistically significant, the RV/LV ratio actually acted as a suppressor in mediation analysis, such that increasing values counteracted the detrimental effect of QIAs. Although we adjusted for emphysema and FEV₁, this finding may be related to the alterations in RV and LV volume seen in chronic lung disease (15, 50). Despite their traditional role in PH screening, increased PA/Ao and RV/LV ratios appear to be less useful in assessing the physiologic impact of parenchymal disease. Instead, our results highlight the importance of subtle but clinically significant radiographic vasculopathy—which likely develops before overt hemodynamic derangement—and the utility of CT imaging in its detection and quantification.

Our study has limitations. Because this work was cross-sectional, we do not know how the reported findings change over time or with the progression of parenchymal disease. Although the COPDGene study contains specific self-reported cardiac comorbidities, Phase-1 participants did not routinely undergo echocardiography for left-heart evaluation, which could influence both endostatin and NT-proBNP. Therefore, we cannot exclude that part of the correlation between these variables and radiographic vasculopathy was related to left-heart disease and volume overload. Similarly, right-heart catheterization is not part of the COPDGene protocol, so our cohort likely had a range of pulmonary hemodynamics extending from normal values to PH. That noted, pulmonary vasculopathy begins well before the criteria for PH are met on right-heart catheterization and may not be detectable by traditional hemodynamic measures (43, 51, 52). Our group has previously demonstrated that CT-based surrogates for vasculopathy correlate with histologic measures of pulmonary vascular remodeling, such as vessel wall thickness, in people with parenchymal lung disease. Furthermore, this association was strongest in the subset with echocardiographic PH (53). Separately, we have also shown that greater preacinar arterial dilation occurred in patients with ILD with PH compared with those with normal hemodynamics (49). From the parenchymal perspective, we examined QIA in this analysis, and, although it is an objective measure of interstitial features and parenchymal injury, our results may not extrapolate to ILAs, to ILD, or to non–early vasculopathy.

In terms of statistical limiters, COPDGene Phase-1 data contain only variables collected at enrollment, and precise, nuanced information on tobacco exposure, dose, and timing is absent. The subgroup of Phase-1 participants who underwent circulating protein sampling was small, and dichotomization can lead to misclassification. Although a more comprehensive assessment of the proteome is needed, our results demonstrate that biologically relevant proteins associated with PH are elevated in conjunction with prognostically significant CT markers of vasculopathy. We applied broad yet purposeful criteria to identify and include potential important confounders for mediation analysis, but it is possible that we did not account for unmeasured variables (37, 54). From a technical perspective, there are limits to the sensitivity of CT to detect vasculopathy. That noted, the level of the pulmonary vascular endothelial marker angiopoietin-2 correlated with a higher QIA burden even among the subgroup without preacinar arterial dilation. Recalibration of the disease definition to reflect the presence of early vasculopathy may be beneficial to guide therapy, predict progression, and inform clinical trial design in cohorts with mild parenchymal lung disease (43).

In conclusion, we demonstrate that the negative impact of QIA on clinical outcomes appears to occur primarily through their indirect pulmonary vascular damage, as opposed to the direct effect of parenchymal disease itself. Preacinar intraparenchymal arterial dilation appears to represent a novel and sensitive radiographic marker of pulmonary vasculopathy, with biochemical evidence corroborating the presence of pulmonary vascular endothelial and myocardial dysfunction in this population. Notably, the traditional but more proximal PA/Ao and RV/LV ratios provided a near-negligible link between QIA burden and clinical outcomes. Our results highlight the spectrum of pulmonary vascular dysfunction in parenchymal lung disease and underscore the need for reconsideration of appropriate definitions reaching beyond hemodynamic criteria in this population.