Associations of High Attenuation Area-related Proteomic Biomarkers with Fibrotic or Subpleural Interstitial Lung Abnormalities

Abstract

Rationale

High attenuation area (HAA) is a computed tomography (CT) tool that correlates with lung inflammation and fibrosis. Systemic molecular correlates of HAA (e.g., plasma proteins) may inform biological process involved in interstitial lung disease (ILD).

Objectives

Identify plasma proteins that associate with HAA and correlate with a higher probability of developing new-onset fibrotic or subpleural interstitial lung abnormalities (ILA).

Methods

Plasma protein levels were measured using a semiquantitative aptamer-based platform in the Multi-Ethnic Study of Atherosclerosis (MESA, n=5486) and Subpopulations and Intermediate Outcome Measures in COPD Study (SPIROMICS, n=1761). Linear regression models identified HAA-associated proteins after adjustment for demographic and socioeconomic factors, CT scanner parameters, study center, and batch. Associations of HAA-related proteins with new-onset fibrotic or subpleural ILA were examined in MESA participants with ILA assessments on full-lung CT 10 years later. Immunohistochemical staining of select proteins was performed in lung tissue from pulmonary fibrosis cases.

Measurements and Main Results

There were 75 proteins detected that were significantly associated with HAA in MESA and SPIROMICS. Gene ontology analysis of these proteins identified processes involved in immune cell chemotaxis and cellular growth and apoptosis. Seven proteins were associated with a higher probability of new-onset fibrotic or subpleural ILA in MESA and two of these, junctional adhesion molecule-like protein and GTP cyclohydrolase 1 feedback regulatory protein, stained in areas of fibrosis in lung tissue from patients with ILD.

Conclusions

Plasma proteins associated with more HAA are involved in immune and cellular processes and associate with new onset fibrotic-subpleural ILA.

INTRODUCTION

Idiopathic pulmonary fibrosis (IPF) is the most well-defined and severest form of interstitial lung disease (ILD) (1). The current approach to IPF clinical trials recruits patients who already have a diagnosis. However, patients often have a significant burden of fibrosis upon enrollment, and this limits the effectiveness of interventions given the irreversible loss of viable lung tissue. Thus, there has been interest in identifying individuals at the earlier stages of IPF in which treatments may have the most impact at preserving lung function (2, 3).

Clinical trials that test therapies to mitigate the risk of disease progression in IPF necessitate cohort enrichment (2). Computed tomography (CT) scans of the lung and detection of features related to pulmonary fibrosis in individuals without known ILD is a promising tool. Interstitial lung abnormalities (ILA), a qualitative assessment of lung parenchymal abnormalities, is the most well-studied CT assessment (4–9). Also, automated CT methods have emerged that can quantify lung inflammation and fibrosis (10). One such method is high attenuation area (HAA), a densiometric-based method that quantifies the proportion of lung with areas of attenuation between −600 to −250 Hounsfield units. More HAA on CT associates with the MUC5B promoter variant (rs35705950), one of the most common genetic risk factors for IPF, and a higher risk of ILD-related death and hospitalization (11–13).

Circulating biomarkers may also help identify individuals at higher risk of developing ILD. This includes blood-based proteins with recent advances in semiquantitative proteomic platforms that can examine thousands of proteins and their associations with clinical phenotypes. Studies have used semiquantitative proteomics to identify proteins that associate with ILA and quantitative interstitial abnormality (QIA) in cigarette smokers and older adults (14, 15). Whether there are overlapping and distinct protein biomarkers that associate with HAA is unknown. Also, these studies have relied on baseline proteomic measurements but there is interest in whether longitudinal changes of proteins correlate with ILD risk which could have implications on assessing treatment responses in trials (16).

We hypothesized proteomic markers related to extracellular matrix maintenance, vascular integrity, and innate immunity would associate with more HAA on CT scan. We leveraged semiquantitative proteomic and radiologic data in the Multi-Ethnic Study of Atherosclerosis (MESA) and Subpopulations and Intermediate Outcome Measures in COPD Study (SPIROMICS). Secondarily, we investigated whether HAA-associated protein markers and their change over time were associated with future development of radiological phenotypes concerning for ILD. Immunohistochemical (IHC) stains of proteins that associate with HAA were performed in lung tissue.

METHODS

Study participants

MESA is a longitudinal U.S.-based cohort study that recruited 6,814 adults between the ages of 45–84 years without known cardiovascular disease at Exam 1 (2000–2002) from six U.S. communities (17). There have been subsequent study visits including the most recent visit at Exam 6 (2016–2018). MESA served as the discovery cohort to identify proteins that associate with HAA. For replication we used the SPIROMICS cohort, a U.S.-based prospective cohort comprised mostly of adults with a self-reported history of 20 cigarette pack-years or more and between the ages of 40–80 years at baseline (2010–2015) (18). We used data from these cohorts because protein measurements were obtained from the same assay and HAA assessments were performed from CT scan images using the same protocol (see Methods below). Also, since cigarette smoking exposure is highly prevalent in IPF and other ILD types (19), this was another justification to use SPIROMICS as a replication cohort. In each study, institutional review board approval was obtained from the cohort study sites and written informed consent was provided by each study participant.

SomaScan proteomics

Proteins were quantified using SomaScan Assay version 4.1, an aptamer-based affinity platform (SomaLogic, Inc., Boulder, CO, USA). The SomaScan platform uses short single-stranded DNA aptamer reagents for high throughput multiplexing of proteins (20, 21). SomaScan version 4.1 measures 7,596 aptamers that map to 6,401 unique proteins . In MESA, proteins were measured from samples collected in up to three batches. Proteins were measured from a single batch in SPIROMICS. For modeling purposes, protein levels were log₂-transformed. The SomaScan platform has shown excellent specificity and precision of measurements with median coefficients of variation ranging from 3.6–7.0% based on prior literature (20–23).

CT scan assessments

In MESA, participants underwent cardiac CT scans at Exam 1 which captures approximately two-thirds of the lower lung fields. At Exam 5 (2010–2012), participants underwent full-lung CT scans. In SPIROMICS, study participants underwent full-lung CT scans at Visit 1 (2010–2015). Full-lung scans in both cohorts used the MESA Lung/SPIROMICS protocol (24).

HAA was defined as the percentage of voxels with a range of attenuation values between −600 and −250 Hounsfield units and measured with the Pulmonary Analysis Software Suite (PASS) by the University of Iowa’s Advanced Pulmonary Physiomic Imaging Laboratory (13). A trained technician semiautomatically segmented and corrected lung images (25).

Based on the 2020 Fleischner Position Statement, ILA was defined as the presence of non-dependent abnormalities that include ground-glass or reticular abnormalities, traction bronchiectasis, non-emphysematous cysts, honeycombing, and lung distortion and involve at least 5% of a lung zone (9). In MESA, ILA was assessed from Exam 1 cardiac scans (26) and Exam 5 full-lung scans (13). Serial ILA assessments were not available in SPIROMICS at the time of this analysis.

Lung tissue immunohistochemical staining

Candidate protein marker distribution in lung tissue of patients with IPF (n=3) was assessed using IHC staining and compared with tissue from patients without known pulmonary fibrosis undergoing surgical resection (n=3) and emphysema cases (n=3). Specimens were obtained from the University of Virginia Biorepository and Tissue Research Facility (BTRF, Charlottesville, VA, USA). Full description of methods is in online supplementary appendix. Two lung pathologists reviewed the immunostaining for candidate proteins associated with HAA.

Statistical analysis

The workflow of HAA-associated proteomic biomarker identification is shown in Figure 1. MESA was the discovery cohort and linear regression models were used to identify proteins that cross-sectionally associate with HAA (continuous variable) at Exam 1. Models were adjusted for potential confounders: age, sex, smoking history, cigarette pack-years, height, weight, waist circumference, education attainment, income, estimated glomerular filtration rate (eGFR), lung volume imaged, percent emphysema, scanner precision variables (scanner model, radiation dose, and voxels), study center, and the first five principal components of genetic ancestry (PCA). We also adjusted for batch in the model as a categorical variable. To account for false-discovery, we used a Benjamini-Hochberg adjusted p-value <0.05. We then determined which of these proteins replicated in the SPIROMICS cohort using a Bonferroni-adjusted p-value to account for the number of significant proteins from the derivation cohort. The same linear regression model from MESA was used for the SPIROMICS cohort except for eGFR which was not measured in SPIROMICS. The R package, “clusterProfiler”, was used to identify canonical pathways related to those proteins that were associated with HAA in the replication cohort by mapping protein IDs to Entrez IDs and performing Gene Ontology (GO) enrichment pathway analysis (27).

Figure 1.

Workflow of analysis

We then examined whether those HAA-associated proteomic markers that had concordant direction of associations in the discovery and replication cohorts were associated with a higher probability of developing fibrotic or subpleural ILA in MESA on follow-up scan at Exam 5. We chose ILA with fibrotic and/or subpleural features as the outcome because they have the strongest associations with shorter survival, the MUC5B promoter variant, and is the closest to resemble IPF on CT compared with other ILA subtypes (28, 29). We excluded participants who had fibrotic or subpleural ILA detected on their Exam 1 scans for this analysis, because we were interested in identifying proteins that may help identify those individuals at higher risk of developing a radiologic phenotype closest to IPF. Logistic regression models with the presence of fibrotic or subpleural ILA on MESA Exam 5 scans as the outcome were used. Due to differences in clinical outcomes by race and ethnicity in ILD, we performed a sensitivity analysis stratified by self-reported race and ethnicity in MESA (2, 30). Due to the exploratory nature of this analysis, p-value <0.05 was considered statistically significant for this analysis.

Of those proteins that were significantly associated with new onset fibrotic or subpleural ILA, we examined their change over time in MESA by taking the difference between Exam 1 and Exam 5 measurements. We used logistic regression models to examine the association of changes in these proteins with Exam 5 fibrotic or subpleural ILA. We then used generalized additive models (GAM) to visualize the continuous association between changes in protein measurement and probability of Exam 5 CT fibrotic or subpleural ILA. Age, sex, smoking history, cigarette pack-years, height, weight, waist circumference, glomerular filtration rate, education attainment, income, study center, PCA, batch, and baseline protein level were adjusted for in models. Participants with missing covariate data were excluded from the final analysis. R version 4.2.2 (R Foundation for Statistical Computing) was used for the analysis.

RESULTS

There were 5,486 MESA participants and 1,761 SPIROMICS participants with proteomic data, baseline HAA assessments, and complete covariate data (Figure 1). MESA had lower proportions of males, self-reported non-Hispanic White individuals, and cigarette smokers compared with SPIROMICS (Table 1). The mean percent HAA was higher in MESA (5.00, SD 2.90) than SPIROMICS (4.38, SD 1.84), while percent emphysema was markedly higher in SPIROMICS than MESA. The number of batches in MESA are shown in E-Table 1. Proteins were measured in a single batch for SPIROMICS.

Table 1.

Baseline characteristics

Characteristics	Discovery Cohort MESA (n=5,486)	Validation Cohort SPIROMICS (n=1,781)
Age, years	62 (10)	63 (9)
Female sex	2846 (51.9%)	822 (46.1%)
Non-Hispanic White	2238 (40.8%)	1327 (74.5%)
Asian	688 (12.5%)	15 (0.8%)
African American	1289 (23.5%)	319 (17.9%)
Hispanic	1271 (23.2%)	80 (4.5%)
Other	0	40 (2.2%)
Never smoker	2480 (45.2%)	0
Former smoker	2233 (40.7%)	1122 (63.0%)
Current smoker	773 (14.1%)	659 (37.0%)
Cigarette pack-years	12 (22)	46 (29)
Height, cm	166 (10)	170 (10)
Weight, kg	79 (17)	81 (18)
Body mass index, kg/m2	28 (5)	28 (5)
Waist circumference, cm	98 (14)	100 (14)
Less than high school	985 (18.0%)	197 (11.1%)
High school	2552 (46.5%)	1080 (60.6%)
College or beyond	1949 (35.5%)	504 (28.3%)
Percent high attenuation areas	5.00 (2.90)	4.38 (1.84)
Lung volume imaged, cm3	2787 (806)	5885(1479)
Percent emphysema	4.17 (4.33)	7.54 (9.96)

Abbreviations: MESA=Multi-Ethnic Study of Atherosclerosis; SPIROMICS=Subpopulations of Intermediate Outcome Measures in COPD Study

Percent emphysema defined as percentage of lung voxels below 950 Hounsfield units.

Continuous variables presented as mean (standard deviation).

Categorical variables presented as no. (percentage).

A subset of MESA participants had both HAA and qualitative assessments of interstitial lung abnormalities from Exam 1 scans (E-Table 2). There was a higher proportion of subjects with ground-glass opacities and fibrotic features visualized on CT in the higher HAA group vs. the lower HAA group.

HAA proteomic biomarkers

In MESA, 432 proteins were significantly associated with HAA using a Benjamini-Hochberg adjusted p-value cutoff <0.05 (online supplementary appendix). From this list of proteins in MESA, 75 were validated in the SPIROMICS cohort using a Bonferroni-corrected p-value cutoff <0.00011574 (0.05/432) (Figure 2A). All the proteins had concordant direction of associations in the two cohorts. Table 2 shows the proteins that were significantly associated with HAA in the discovery and validation cohorts. Several proteins with multiple aptamers were significantly associated with HAA (Table 2). Prior studies from the MESA cohort used ELISA assays to measure MMP-7, IL-6, ITP, and PIINP from stored plasma samples and were associated with HAA (13, 31, 32). Plasma levels of these proteins from SomaScan were positively correlated with levels measured from ELISA assays (E-Figure 1). Although meeting statistical significance, there were weak-to-moderate correlations. Most of the proteins had concordant direction of associations with HAA as the prior studies but were weaker (E-Table 3).

Figure 2.

(A) Volcano plot of proteins associated with high attenuation areas in replication cohort, SPIROMICS. Linear regression model adjusted for age, sex, smoking history, cigarette pack-years, height, weight, waist circumference, education attainment, income, principal components of genetic ancestry, lung volume imaged, percent emphysema, and scanner parameters (scanner manufacturer and radiation dose). Bonferroni-corrected p-value cutoff <0.00011574 used to test for statistical significance. Repeat protein names on volcano plot indicate multiple aptamers. (B) Gene ontology enrichment analysis of validated proteins that associate with high attenuation areas.

Table 2.

Proteins associated with high attenuation areas in discovery and validation cohorts

		MESA		SPIROMICS
Protein	Aptamer ID	Mean difference in HAA (95% CI)	FDR-adjusted p-value	Mean difference in HAA (95% CI)	Bonferroni-adjusted p-value
SP-B	10672–75	2.65 (1.76 to 3.55)	4.6816e−06	4.58 (3.48 to 5.70)	2.8742e−16
HE4	11388–75	8.81 (7.14 to 10.51)	1.3199e−22	6.94 (5.46 to 8.45)	1.462e−20
H6ST2	13524–25	2.44 (1.29 to 3.60)	0.00276855	3.37 (1.87 to 4.88)	8.7477e−06
BLC	13701–2	3.07 (2.12 to 4.03)	1.7825e−07	2.38 (1.28 to 3.50)	2.1525e−05
C7	13731–14	2.68 (1.26 to 4.12)	0.01067555	4.71 (2.58 to 6.89)	1.2507e−05
Neuropilin-2	15387–44	4.44 (1.87 to 7.07)	0.02099152	7.70 (4.41 to 11.10)	3.002e−06
TIM-4	15449–33	1.98 (0.80 to 3.17)	0.02667164	2.85 (1.48 to 4.24)	4.2474e−05
Macrophage scavenger receptor:ECD	15533–97	1.97 (0.90 to 3.06)	0.01397696	2.38 (1.17 to 3.61)	0.00011387
CSPG3	15573–110	−3.62 (−5.01 to −2.20)	0.0002415	−3.55 (−4.86 to −2.21)	3.041e−07
TREM2	16300–4	1.90 (0.95 to 2.86)	0.00570707	2.39 (1.23 to 3.56)	4.8179e−05
Cadherin-11:ECD	16305–10	1.03 (0.45 to 1.62)	0.01865925	2.71 (1.80 to 3.62)	4.6891e−09
PACAP	16322–10	3.02 (1.94 to 4.10)	1.747e−05	2.58 (1.33 to 3.86)	5.1186e−05
RSPO1	16614–27	2.34 (1.03 to 3.67)	0.0173365	3.63 (1.86 to 5.44)	5.2961e−05
CDCP1	16818–200	3.02 (2.05 to 3.99)	6.9506e−07	3.95 (2.87 to 5.04)	4.4534e−13
SEM4A	16915–153	4.67 (2.98 to 6.39)	1.8868e−05	4.92 (2.73 to 7.16)	8.7763e−06
STCH	17515–6	3.26 (1.51 to 5.05)	0.01169371	5.18 (2.79 to 7.63)	1.8164e−05
SP-D	19590–46	1.50 (0.78 to 2.22)	0.00355089	4.95 (4.00 to 5.90)	6.6734e−25
SCRB1	20534–6	1.71 (0.60 to 2.85)	0.04660857	3.68 (2.26 to 5.12)	3.0479e−07
GFRP	21118–48	2.43 (1.44 to 3.43)	0.0003612	3.42 (2.18 to 4.66)	4.7867e−08
METRL	21705–33	4.06 (1.75 to 6.43)	0.01900421	6.84 (3.64 to 10.13)	2.0547e−05
IGFBP-2	22985–160	1.70 (0.79 to 2.62)	0.01169371	3.28 (2.01 to 4.57)	3.8437e−07
TIMP-1	23173–3	2.62 (1.09 to 4.16)	0.02260558	6.15 (4.18 to 8.15)	5.2641e−10
CLIC3	24693–5	2.06 (0.76 to 3.37)	0.03791474	3.69 (1.88 to 5.52)	5.4582e−05
FUBP2	24926–9	3.61 (2.06 to 5.17)	0.00075956	5.21 (2.62 to 7.87)	6.7498e−05
6Ckine	2516–57	2.24 (1.01 to 3.48)	0.01449739	4.91 (3.29 to 6.55)	1.8347e−09
IGFBP-2	2570–72	1.77 (0.89 to 2.66)	0.00531558	3.04 (1.88 to 4.21)	2.1653e−07
Leptin	2575–5	−4.04 (−4.88 to −3.19)	2.3093e−16	−3.89 (−4.88 to −2.90)	6.9264e−14
ERBB1	2677–1	−5.35 (−8.12 to −2.50)	0.01337982	−7.66 (−10.72 to −4.50)	3.7441e−06
MPIF-1	2913–1	3.27 (1.79 to 4.77)	0.0015716	3.92 (2.11 to 5.77)	1.8583e−05
Growth hormone receptor	2948–58	−3.19 (−4.30 to −2.06)	1.8868e−05	−5.87 (−7.44 to −4.28)	2.0457e−12
MIP-1a	3040–59	2.14 (1.20 to 3.09)	0.00116352	4.23 (2.73 to 5.75)	2.4e−08
PARC	3044–3	3.46 (2.22 to 4.71)	1.7919e−05	5.60 (3.84 to 7.40)	3.0099e−10
BAFF	3059–50	2.19 (0.77 to 3.63)	0.04516731	4.29 (2.13 to 6.49)	8.7223e−05
IGFBP-7	3320–49	3.65 (1.57 to 5.78)	0.01999258	6.21 (3.18 to 9.32)	4.5851e−05
TSP2	3339–33	1.53 (0.59 to 2.49)	0.03328343	2.73 (1.65 to 3.81)	5.9137e−07
BLC	3487–32	3.75 (2.67 to 4.83)	9.1218e−09	2.89 (1.51 to 4.28)	3.473e−05
PAPP-A	4148–49	2.38 (1.44 to 3.34)	0.00021104	2.71 (1.46 to 3.99)	2.1535e−05
Prekallikrein	4152–58	−3.43 (−5.13 to −1.70)	0.00747677	−7.47 (−10.38 to −4.47)	2.0524e−06
Spondin-1	4297–62	3.44 (1.62 to 5.29)	0.01022454	9.72 (6.90 to 12.62)	4.1589e−12
MIC-1	4374–45	2.56 (1.42 to 3.72)	0.00137208	3.84 (2.55 to 5.15)	4.1587e−09
MMP-12	4496–60	2.24 (1.25 to 3.24)	0.00132756	2.11 (1.04 to 3.18)	0.0001021
Lysozyme	4920–10	2.34 (1.12 to 3.57)	0.00902411	3.82 (2.24 to 5.42)	1.7318e−06
CAPG	4968–50	1.87 (0.88 to 2.86)	0.01030245	2.75 (1.36 to 4.16)	9.9535e−05
sICAM-5	5124–62	3.18 (1.77 to 4.62)	0.00132756	6.78 (4.83 to 8.76)	4.1388e−12
sICAM-5	5124–69	2.79 (1.40 to 4.21)	0.00569007	5.29 (3.66 to 6.95)	1.303e−10
FABP	5437–63	−2.94 (−4.11 to −1.75)	0.00038657	−3.35 (−4.79 to −1.88)	1.0104e−05
Spondin-1	5496–49	3.88 (1.91 to 5.89)	0.0065124	8.77 (5.99 to 11.62)	2.5089e−10
TREM2	5635–66	1.96 (0.96 to 2.96)	0.00714912	3.00 (1.67 to 4.36)	8.9614e−06
TMX3	5654–70	3.01 (1.46 to 4.59)	0.00801151	4.86 (2.85 to 6.90)	1.6714e−06
Tenascin	5675–6	1.79 (0.74 to 2.84)	0.02441723	2.94 (1.65 to 4.25)	6.854e−06
CECR1	6077–63	2.12 (1.30 to 2.94)	0.00011056	2.33 (1.19 to 3.49)	6.0324e−05
ELA1	6107–3	2.00 (0.75 to 3.26)	0.0359327	2.60 (1.28 to 3.94)	0.00010839
QSCN6	6217–23	3.51 (1.37 to 5.70)	0.03028889	10.34 (7.14 to 13.63)	7.059e−11
Secretoglobin family 3A member 1	6252–62	2.65 (1.88 to 3.42)	1.5644e−08	2.87 (1.91 to 3.85)	4.8217e−09
Tenascin	6260–14	2.37 (0.91 to 3.86)	0.03257182	7.81 (5.70 to 9.95)	1.2986e−13
C1QT1	6304–8	4.20 (2.26 to 6.17)	0.00196403	7.60 (4.45 to 10.83)	1.39e−06
FAIM3	6574–11	1.65 (0.74 to 2.58)	0.01571336	2.69 (1.51 to 3.88)	7.3708e−06
F176C:CD	7008–13	2.27 (0.89 to 3.66)	0.02916174	3.63 (2.16 to 5.13)	1.1517e−06
Tenascin	7131–207	2.16 (0.73 to 3.60)	0.04965436	4.55 (2.79 to 6.34)	3.038e−07
Amyloid-like protein 1	7210–25	−2.65 (−3.82 to −1.46)	0.00183128	−3.12 (−4.42 to −1.79)	5.1654e−06
ITI heavy chain H1	7955–195	−4.18 (−6.16 to −2.17)	0.00451628	−8.19 (−11.19 to −5.08)	5.3426e−07
EPHB2	8225–86	2.87 (1.15 to 4.62)	0.02791634	4.71 (2.47 to 6.99)	3.2058e−05
JAML1	8232–90	2.74 (1.62 to 3.86)	0.00034816	3.82 (2.41 to 5.25)	9.0507e−08
sICAM-5	8245–27	3.19 (1.81 to 4.58)	0.00090128	8.60 (6.65 to 10.58)	1.046e−18
TNF sR-II	8368–102	2.29 (0.93 to 3.65)	0.02521438	3.70 (1.84 to 5.60)	8.7753e−05
RSPO4	8464–31	2.32 (0.93 to 3.73)	0.02803871	7.07 (4.69 to 9.50)	3.1943e−09
IGFBP-2	8469–41	1.74 (0.86 to 2.63)	0.00693943	3.11 (1.89 to 4.33)	4.627e−07
Leptin	8484–24	−3.90 (−4.69 to −3.09)	4.7716e−17	−3.66 (−4.57 to −2.75)	1.2512e−14
IGFBP-2	8819–3	1.70 (0.86 to 2.56)	0.00531558	3.02 (1.87 to 4.18)	2.0183e−07
sTREM-1	9266–1	4.58 (3.01 to 6.17)	5.1876e−06	4.16 (2.50 to 5.85)	7.5401e−07
SDF-1	9278–9	5.32 (3.47 to 7.21)	7.3726e−06	5.15 (2.92 to 7.43)	4.9054e−06
PTPRD	9296–15	−3.14 (−4.87 to −1.36)	0.02033881	−3.36 (−4.92 to −1.78)	3.9093e−05
MANS4	9578–263	1.71 (0.83 to 2.60)	0.00813013	2.49 (1.24 to 3.75)	8.45e−05
L-plastin	9749–190	4.69 (2.58 to 6.84)	0.00137881	7.23 (4.48 to 10.06)	1.5856e−07
SYWC	9870–17	3.40 (1.94 to 4.88)	0.0007642	6.27 (4.21 to 8.37)	1.4812e−09

Abbreviation: CI=confidence interval; FDR=false discovery rate; HAA=high attenuation area; MESA=Multi-Ethnic Study of Atherosclerosis; SPIROMICS=Subpopulations of Intermediate Outcome Measures in COPD Study

Linear regression model adjusted for age, sex, smoking history, cigarette pack-years, height, weight, waist circumference, education attainment, income, principal components of genetic ancestry, lung volume imaged, percent emphysema, study center, batch, and scanner parameters (scanner model, radiation dose, and voxels). Estimated glomerular filtration rate adjusted for in MESA.

Results reported per 1-unit increment in log₂-transformed protein level.

Bonferroni-adjusted p-value cutoff of 0.00011574.

We mapped HAA-associated proteins from Table 2 with their respective encoding gene using the Entrez identifier (E-Table 4). Pathway analysis of the HAA-associated proteins showed enrichment of immune cell and vascular adhesion-related processes (Figure 2B). Regulation of the mitogen activated protein kinase (MAPK) cascade was also identified which regulates cellular processes related to apoptosis and proliferation.

Associations of HAA proteomic markers with fibrotic and subpleural ILA

We then examined whether the 75 validated HAA-associated proteins were associated with the probability of developing fibrotic or subpleural ILA in the MESA cohort. After excluding participants who had fibrotic or subpleural ILA at baseline and missing covariate data, there were 1,766 participants with Exam 1 SomaScan protein measurements and serial Exam 1 and Exam 5 ILA assessments (E-Figure 2). There were 102 individuals (5.8%) with new-onset fibrotic or subpleural ILA. Characteristics of this group are shown in E-Table 5. Descriptively, those MESA participants with repeat ILA assessment at Exam 5 were younger, had a higher proportion of women, and lower proportion of current smokers and below high school education than the overall cohort. Participants with new-onset fibrotic or subpleural ILA at Exam 5 were older at baseline, comprised of more smokers, and had a higher cigarette pack-year history than those participants without fibrotic-subpleural ILA. Of the proteins that were associated with HAA at baseline, seven of them were associated with a higher probability of developing fibrotic or subpleural ILA 10 years later at Exam 5 (E-Table 6). These proteins were human epididymis protein 4 (HE4), CUB domain-containing protein 1 (CDCP1), macrophage inhibitory cytokine-1 (MIC-1), junctional adhesion molecule-like (JAML), GTP cyclohydrolase 1 feedback regulator (GCHFR), surfactant protein-B (SP-B), and tumor necrosis factor soluble receptor II (TNF sR-II). Stratified analysis by self-reported race and ethnicity showed the proteins had concordant direction of associations with Exam 5 fibrotic or subpleural ILA, except for the Asian subgroup (E-Table 7).

Longitudinal changes in proteomic biomarkers

We then examined whether changes in plasma levels over time of these seven HAA-associated proteins also correlated with the future development of fibrotic or subpleural ILA. In MESA, SomaScan protein measurements were available from Exam 1 (2000–2002) and Exam 5 (2010–2012) plasma samples. After excluding participants with baseline fibrotic or subpleural ILA, there were 1,731 with serial SomaScan measurements and ILA assessments at Exams 1 and 5, of whom 98 (5.7%) had new-onset subpleural or fibrotic ILA. Increased plasma levels in 5 of the 7 proteins were associated with new onset fibrotic or subpleural ILA (HE4, CDCP1, MIC-1, GCHFR, SP-B) (Table 3). Associations were linear for HE4, MIC-1, GCHFR, and SP-B (Figure 3).

Table 3.

Associations of changes in HAA-derived proteins with new-onset subpleural and fibrotic interstitial lung abnormalities

Protein	Aptamer ID	Odds ratio for new-onset subpleural or fibrotic ILA per 1-unit increase in log-transformed protein level (95% CI)	P-value
HE4	11388–75	1.31 (1.03–1.67)	0.03
CDCP1	16818–200	2.15 (1.16–4.00)	0.02
GCHFR	21118–48	2.34 (1.38–3.96)	0.002
MIC-1	4374–45	1.75 (1.02–3.01)	0.04
JAML	8232–90	2.11 (1.00–4.42)	0.049
SP-B	10672–75	2.41 (1.61–3.61)	1.92e−05
TNF sR-II	8368–102	1.81 (0.87–3.78)	0.11

Logistic regression model adjusted for age, sex, smoking history, cigarette pack-years, height, weight, waist circumference, education attainment, income, estimated glomerular filtration rate, study center, batch, principal components of genetic ancestry, and baseline plasma protein level.

Results reported per 1-unit increase in log₂-transformed protein level between Exam 1 (2000–2002) and Exam 5 (2010–2012).

Figure 3.

Plots of changes in plasma proteomic biomarkers between Exam 1 (2000–2002) and Exam 5 (2010–2012) and probability of Exam 5 fibrotic or subpleural interstitial lung abnormalities in MESA. Generalized additive models used to plot associations with adjustment for baseline age, sex, smoking history, cigarette pack-years, height, weight, waist circumference, glomerular filtration rate, education attainment, income, study site, principal components of genetic ancestry, and batch, and Exam 1 protein level. X-axis is the difference in log₂-transformed protein levels between Exam 1 (2000–2002) and Exam 5 (2010–2012). One participant was removed from analysis due to outlier measurements in all six proteins.

JAML and GCHFR stain in IPF lung tissue

To our knowledge, there is a paucity of data whether JAML and GCHFR stains in fibrotic lung tissue. We obtained lung tissue samples from patients with IPF and performed immunohistochemical stains for JAML and GCHFR (online supplementary appendix). Tissue samples from patients without known pulmonary fibrosis who underwent surgical resection for nodules and emphysematous lungs were also stained for comparison (E-Table 8). Staining of control tissue is shown in E-Figure 3. Isotype antibody control staining of lung tissue for JAML and GCHFR is shown in E-Figure 4A-B. In IPF lung tissue, JAML stained most prominently in areas of fibrosis and particularly in early and late fibroblastic foci (Figure 4A-C). In tissue samples from non-IPF lungs and emphysema cases (E-Figure 4C-D), JAML stained in similar regions of the lung but less so, due to the absence of fibrosis. Compared with non-IPF lung and emphysema cases (E-Figure 4E-F), GCHFR stained more so in lymphocytes and alveolar macrophages in IPF lung tissue (Figure 4D-F). Overall, the staining of JAML and GCHFR was most prominent in areas of fibrosis in IPF and minimal staining in geographically heterogenic non-fibrotic lung tissue.

Figure 4.

Immunohistochemistry (IHC) staining of JAML in IPF lung tissue at (A) 12x, (B) 40x, and (C) 400x magnification. GCHFR immunostaining of IPF lung tissue at (D) 12x, (E) 40x, and (F) 400x magnification.

DISCUSSION

We identified plasma proteins that were associated with high attenuation areas on CT lung images in two independent cohorts. Pathway analysis revealed biological processes involved in regulation of immune cells, vascular adhesion, cellular apoptosis and proliferation. Several HAA-associated proteins at baseline and their increase over time were associated with a higher probability of developing fibrotic or subpleural interstitial lung abnormalities and stained in IPF lung tissue.

HAA is a CT-based assessment of increased attenuation areas in lungs that can be prone to confounding effects of body habitus, artifacts, and abnormalities unrelated to ILD that can influence HAA. Thus, we aimed to identify plasma proteins that associate with HAA and therefore may reflect systemic biological processes that HAA of the lungs capture. Several of the HAA-associated proteins we identified also associate with ILA, QIA, and IPF survival, which include ICAM5, CDCP1, TREM1, MIC-1 (GDF15), and HE4 (WFDC2) and further supports their potential involvement in subclinical lung injury and fibrosis (14, 15, 33). We also identified other proteins unique to HAA and as we hypothesized, several of them are involved in immune cell and vascular processes and supports prior studies that have implicated these pathways in pulmonary fibrosis (34–39).

Several plasma proteins associated with lower HAA including leptin and growth hormone receptor, both of which are involved in adipogenesis and body habitus development. We expected a positive association with HAA and leptin due to possible confounding effects of adiposity on HAA. Furthermore, leptin binds to receptors in the lung and concentrations of leptin correlate between blood and bronchoalveolar lavage fluid specimens with preclinical experiments demonstrating conflicting roles of leptin in fibrogenesis (40–43). Further mechanistic work targeting these proteins will elucidate their therapeutic potential.

Due to interest in identifying individuals at-risk of developing future disease, we examined whether the HAA-associated proteins were associated with a higher probability of future ILD using fibrotic and subpleural ILA subtypes as radiologic surrogates. We were interested in those individuals without fibrosis and subpleural involvement on their baseline CT scans, because this subgroup may help with future trial enrichment to test preventative therapies. We found higher plasma levels of HE4, CDCP1, MIC-1, JAML, GCHFR, TNF sR-II, and SP-B were associated with new onset fibrotic or subpleural ILA. Likewise, associations of the proteins had concordant direction of associations in an analysis stratified by self-reported race and ethnicity, except for the Asian subgroup which may be due to the smaller sample size. Proteins were strongly associated with a higher probability of fibrotic or subpleural ILA in the self-reported African American and Black subgroup. This is notable in light of studies demonstrating key differences in clinical outcomes by self-reported race and ethnicity (30, 44, 45) and lack of identified common genetic variants related to ILD risk in underrepresented minority groups which would hinder enrollment in future precision medicine trials (2, 46). Our study demonstrates blood-based proteins may bypass inherent limitations of genomic studies that are not adequately powered to identify common variants in populations outside of non-Hispanic White of European ancestry. We did not have an independent diverse population of adults to replicate our findings for fibrotic and subpleural ILA. A major next step based on our findings and other work (14, 15) is to derive and validate prediction models that incorporate clinical, genomics, imaging, and proteomic biomarkers to predict the future development of ILD. Doing so will be critical for cohort enrichment for future trials that test preventative therapies, particularly in at-risk individuals like 1^st-degree relatives of patients with pulmonary fibrosis (47). Also, this approach may identify at-risk individuals from whom pathobiological inferences can be made and identify molecular mechanisms that are unique to the early development of ILD and lead to therapeutic developments.

A novel component of our study was the availability of serial protein measurements in MESA. We found that increasing levels of several proteins including HE4, SP-B, GCHFR, CDCP1, MIC-1, and JAML were associated with a future probability of developing fibrotic or subpleural ILA. Our finding provides insight on the temporal relationship between these proteins and the spectrum of ILD and that higher levels may reflect disease progression. Furthermore, tools to identify and monitor progression of pulmonary fibrosis stages remains a significant barrier to designing clinical trials that test preventative therapies. Our findings suggest the plausibility of blood-based protein markers to complement other tools like imaging to assess disease progression for staging purposes and treatment responses in clinical trials. It remains uncertain which of the plasma proteins associates with short-term and long-term development of lung fibrosis. This is pertinent in light of recent studies that have shown some individuals will have more rapid radiologic progression of ILA in 3–5 years (48). More frequent blood sampling and serial imaging with progression assessments is a future research area.

We acknowledge that several of the identified proteins may not be specific to the early pathogenesis of ILD but rather systemic inflammatory and fibrosing processes that affect multiple organs. Of the seven HAA-associated proteins that correlated with new onset fibrotic-subpleural ILA, we determined whether JAML and GCHFR were expressed in IPF lung tissue because they had not been well-studied in IPF. JAML has been implicated in mucosal integrity and delays intestinal epithelial repair during acute inflammation and may have a similar role in the lung (49). A recent study demonstrated increased mRNA expression of JAML in fibrotic lung tissue (50). GCHFR stained heavily in alveolar macrophages and lymphocytes, thus suggesting an immunomodulatory function in IPF and has been implicated in IPF risk (51). Importantly, our immunohistochemical stains of IPF lungs show that JAML and GCHFR localize in and around areas of fibrotic lung. Follow-up studies that examine histopathological correlations with HAA and preclinical experiments that explore the function of these proteins in lung fibrosis will be informative.

A major limitation of our study and other semiquantitative proteomic work is the lack of full quantification using validated assays. Therefore, we could not confirm the direction of associations between the proteins and HAA nor concentration cutoff levels that demonstrate adequate performance characteristics at predicting future development of ILD. We did not have more frequent paired plasma samples with proteomic measurements and could not determine the incremental rate of change in the proteins that strongly associates with developing subsequent ILD. Also, while an advantage of aptamer-based proteomic platforms is the breadth of proteins in their panels which facilitates discovery, their specificity to the actual protein target is lower compared to other approaches (52). We performed a sensitivity analysis and found that 4 proteins measured by ELISA assays and previously associated with HAA in MESA correlated with the aptamer-based protein measurements in our study. None of the proteins were associated with HAA or replicated in SPIROMICS and may be due to differences in sample sizes and the fact that our study used semiquantitative plasma levels. Applying other semiquantitative proteomic platforms and full quantification with ELISA or mass spectrometry will be important for validation of our findings. Selection bias may have influenced our findings, particularly related to subsequent development of ILD on CT scan, which required serial blood samples and CT scans over a span of 10 years. Indeed, the prevalence of fibrotic or subpleural ILA of 5.8% on repeat CT scans in MESA may be higher than expected given prevalence estimates of ILDs are much lower (53). It is possible that some ILA cases may regress while others progress to other fibrotic ILD subtypes besides IPF and this is pertinent given recent studies that suggest shared genomic and histopathologic features across different ILD subtypes (54). Additionally, the MESA cohort was comprised of older adults at recruitment with a minimum age cutoff of 45 years, which may have influenced the prevalence estimates. Conversely, given that ILD remains underrecognized and contributes to significant morbidity and mortality (55, 56), we cannot completely rule out there were MESA participants who developed ILD but did not participate in follow-up visits or had died without ILD adjudication. Thus, this may have biased our results towards the null. More prospective cohorts with serial imaging and omics data and ILD event adjudication will be informative. We acknowledge the SPIROMICS cohort was comprised of individuals with a much stronger smoking history than MESA and may have affected our findings. However, we still detected 75 proteins that replicated across the two cohorts and suggest generalizability of our findings in both older adults and those with a significant smoking history. Similar to prior studies, our findings support the notion that there are molecular markers that correlate with early lung inflammation and scarring that is not entirely explained by smoking exposure. Another future research area is to derive and validate blood-based protein biomarkers that predict development of clinically diagnosed ILD. This requires independent cohorts with many years of follow-up data and adjudication of an ILD diagnosis that is harmonized to proteomic and clinical data. Imaging and lung physiologic surrogates for ILD that complement physician-adjudicated and coded diagnoses is an ongoing area of investigation that could be helpful. Moreso, our study supports the possibility of generating integrative multi-omics prediction models for new-onset lung fibrosis which may help with cohort enrichment for trials that test therapies aimed at preventing lung fibrosis. We did not replicate our longitudinal analyses in independent cohorts and this highlights ongoing needs for collaboration and data harmonization.

In closing, plasma proteins involved in immune and cellular processes were associated with HAA on lung CT and new-onset lung fibrosis on CT scan. Further validation and full quantification are needed of these proteins for future implementation to identify individuals at risk for lung fibrosis.

Supplementary Material

This article has an online data supplement, which is accessible at the Supplements tab.

At a Glance Commentary.

Current Scientific Knowledge on the Subject:

More high attenuation areas (HAA) on lung scans may indicate inflammation and fibrosis but their systemic molecular correlates are unknown.

What This Study Adds to the Field:

Several plasma proteins correlated with more HAA on lung scans and a subset of them were associated with new-onset fibrotic or subpleural interstitial lung abnormalities on subsequent scans. Integration of imaging and omics may help identify adults at risk of developing pulmonary fibrosis.