Proteomic Biomarkers of Quantitative Interstitial Abnormalities in COPDGene and CARDIA Lung Study

Abstract

Rationale

Quantitative interstitial abnormalities (QIAs) are early measures of lung injury automatically detected on chest computed tomography scans. QIAs are associated with impaired respiratory health and share features with advanced lung diseases, but their biological underpinnings are not well understood.

Objectives

To identify novel protein biomarkers of QIAs using high-throughput plasma proteomic panels within two multicenter cohorts.

Methods

We measured the plasma proteomics of 4,383 participants in an older, ever-smoker cohort (COPDGene [Genetic Epidemiology of Chronic Obstructive Pulmonary Disease]) and 2,925 participants in a younger population cohort (CARDIA [Coronary Artery Disease Risk in Young Adults]) using the SomaLogic SomaScan assays. We measured QIAs using a local density histogram method. We assessed the associations between proteomic biomarker concentrations and QIAs using multivariable linear regression models adjusted for age, sex, body mass index, smoking status, and study center (Benjamini-Hochberg false discovery rate–corrected P ⩽ 0.05).

Measurements and Main Results

In total, 852 proteins were significantly associated with QIAs in COPDGene and 185 in CARDIA. Of the 144 proteins that overlapped between COPDGene and CARDIA, all but one shared directionalities and magnitudes. These proteins were enriched for 49 Gene Ontology pathways, including biological processes in inflammatory response, cell adhesion, immune response, ERK1/2 regulation, and signaling; cellular components in extracellular regions; and molecular functions including calcium ion and heparin binding.

Conclusions

We identified the proteomic biomarkers of QIAs in an older, smoking population with a higher prevalence of pulmonary disease and in a younger, healthier community cohort. These proteomics features may be markers of early precursors of advanced lung diseases.

At a Glance Commentary

Scientific Knowledge on the Subject

Quantitative interstitial abnormalities (QIAs) are early parenchymal injuries detected by automated machine learning tools on chest computed tomography scans. QIAs are associated with poor spirometric outcomes, exercise limitations, and increased mortality. In some patients, QIAs are likely precursors to advanced parenchymal diseases. Blood-based proteins may have utility in understanding the biological underpinnings of QIAs.

What This Study Adds to the Field

Using a large-scale plasma proteomics panel, we identified 144 proteomic biomarkers and 49 enriched pathways associated with QIAs shared across two cohorts that spanned middle age to older adulthood. These proteins included novel biomarkers as well as those previously associated with lung injury, inflammation, pulmonary fibrosis, chronic obstructive pulmonary disease, emphysema, and lung cancer, suggesting potential shared pathways between QIAs and advanced parenchymal diseases.

Quantitative interstitial abnormalities (QIAs) are subtle parenchymal changes found on chest computed tomography (CT) scans by automated machine learning algorithms. Prior studies across multiple cohorts have shown that the presence and progression over time of QIAs are clinically meaningful and associated with poorer spirometric outcomes, increased symptoms, shorter 6-minute-walk distance, and higher mortality (1–4). Risk factors for QIAs include smoking, older age, the MUC5B polymorphism, and female gender (2, 4).

QIAs share some radiologic findings, risk factors, and outcomes with advanced pulmonary diseases such as pulmonary fibrosis and chronic obstructive pulmonary disease (COPD). QIAs may therefore be an attractive automated measurement to identify people at risk of progression to advanced parenchymal remodeling such as fibrosis and emphysema, for which current treatments only slow down disease progression and symptoms but do not reverse disease (5–7). However, although QIAs encompass some of the earliest forms of lung injury and impaired respiratory health, they represent a sensitive but nonspecific measure, making QIAs an unclear target for intervention. They represent heterogeneous processes, including the injury that progresses and remodels to emphysema and fibrosis, which is important to capture, as well as transient lung injury that self-resolves.

The NHLBI has stressed the importance of developing strategies for the primary prevention of lung diseases, one of which includes understanding the complex biochemical processes that regulate lung and immune system development and subsequently affect lung health (8). We believe that blood-based proteins have utility in understanding the biological underpinnings of QIAs and have potential as biomarkers for detection, prognostication, monitoring, and therapeutics (9). In our study, we analyzed thousands of proteins using a high-throughput aptamer-based proteomics panel within two multicenter, well-characterized cohorts: COPDGene (Genetic Epidemiology of COPD) and CARDIA (Coronary Artery Disease Risk in Young Adults). Both cohorts come from multiple diverse sites across the United States, are well balanced in terms of sex, and share overlapping age ranges and pulmonary risk factors. The two cohorts vary in that COPDGene is enriched for a mostly ever-smoker population, whereas CARDIA is younger and healthier, representing a population that is further “upstream” of the development of chronic lung disease. In this study of older and middle-aged adults, we hypothesized that we would identify novel protein biomarkers of QIAs and gain molecular insight into this clinically important CT measurement.

Methods

Full details are provided in the online supplement.

Study Populations

Our study included data collected from 4,383 participants from COPDGene at the five-year follow-up visit (visit 2) of the study (2013–2017) and 2,925 participants from CARDIA at the 25-year follow-up visit of the study (2010–2011) with complete inspiratory chest CT scans and proteomics measurements available. Studies were approved by the institutional review boards at all centers. Study participants provided written informed consent at each study visit.

COPDGene is a prospective observational study of 10,198 former and current smokers (ever-smokers) with at least a 10–pack-year smoking history, aged 45–80 years at the baseline visit, with no prior bronchiectasis or interstitial lung disease (ILD) of non-Hispanic White or Black self-reported race, from 21 study centers in the United States (10). At visit 2, participants underwent the collection of questionnaires, inspiratory chest CT scans, and proteomic analysis.

CARDIA is a longitudinal study that commenced in 1985 with 5,115 adults from four communities in the United States, aged 18–30 years at the baseline visit, of self-reported White or Black race (11). At the 25-year follow-up visit, participants underwent the collection of clinical information, questionnaires, inspiratory chest CT scans, and proteomic analysis using standardized assessment (11, 12).

QIA Measurements

On inspiratory volumetric chest CT scans, QIAs were measured using a local histogram classifier, as previously described (1). Briefly, the classifier uses a k nearest neighbors algorithm of the local histogram measurements combined with the distance from the pleural surface to output percentages of CT lung volume occupied by QIAs (a sum of reticulation, subpleural lines, ground-glass opacities, honeycombing, linear scarring, centrilobular nodules, and other nodularity features), which were used as a continuous outcome measure in our study. The classifier also identified quantitative emphysema percentage, which was included in one of the secondary analyses, and normal parenchyma percentage of the lung.

Proteomics Measurements

As previously described, the SomaScan assay (SomaLogic) uses chemically modified single-stranded DNA aptamers called SOMAmers that bind to epitopes on their target proteins (12). Proteomics identification was performed on plasma samples from COPDGene participants using the SomaScan version 4.0 (5K) assay (13), which uses 4,979 SOMAmer aptamers to identify 4,776 unique human proteins, and from CARDIA participants using the SomaScan version 4.1 (7K) assay, which uses 7,335 SOMAmer aptamers to identify 6,609 unique human proteins. For both assays, standard processes including within-plate and per-sample median signal normalization, plate scaling, and calibration were performed independently on all samples (14). Protein concentrations were reported in relative fluorescent units and natural log transformed for analysis. To facilitate validation and comparisons between COPDGene and CARDIA, the aptamers included in our analyses were limited to the 4,979 available in the smaller version 4.0 assay used in COPDGene. There was no missing data for any participant included in the analysis.

Statistical Analyses

To identify the proteins associated with QIA within each cohort, we assessed the association between each log-transformed SomaScan protein concentration as a continuous measure and QIA (percentage QIAs in the lung) as a continuous measure with multivariable linear regression adjusted for age, sex, body mass index, smoking status (current, former, or never), and study center as a random effect to adjust for technical variation (20 centers in COPDGene, 4 centers in CARDIA grouped into two on the basis of principal-component analysis of the proteins [see Figure E1 in the online supplement]). Analyses in COPDGene were also adjusted for platelet count and white blood cell count on the basis of prior internal quality control. A supplemental model was also adjusted for self-reported race (Black or White). We adjusted for multiple testing within each cohort using a Benjamini-Hochberg false discovery rate–corrected P value of ⩽0.05. Analyses were performed using R version 4.2.2 (https://www.r-project.org/) and RStudio version 2022.12.0 + 353 (https://posit.co/download/rstudio-desktop/).

Pathway Enrichment Analysis

We performed protein Gene Ontology (GO) pathway enrichment analysis of the significant proteins using the Database for Annotation, Visualization and Integrated Discovery web platform (15). We used the Entrez Gene identifiers (GeneIDs) linked to each SOMA protein and excluded duplicate GeneIDs. The GeneID list was compared with the Homo sapiens GO database lists for molecular functions, cellular components, and biological processes, using the Fisher exact test and a Benjamini-Hochberg false discovery rate–corrected P value of ⩽0.05.

Tissue-Specific Deconvolution

We performed tissue-specific deconvolution of the significant proteins using the Genotype-Tissue Expression (GTEx) database, a comprehensive public resource to study tissue-specific gene expression and regulation, from the Human Protein Atlas (16). The gene expression activities linked by unique GeneIDs to the proteins positively associated with QIAs were scored using the R package singscore for each tissue, which were used to rank the 37 tissues from Human Protein Atlas version 22.0 and Ensembl version 103.38 (17, 18).

Cell-Specific Deconvolution

We performed cell-specific deconvolution of the significant proteins using single-cell RNA sequencing profiles of lung tissue samples from 67 individuals with ILD versus 49 control subjects, as previously described (19). We used the unique GeneID of each protein to link them to the log₂ fold-change values of the differentially expressed genes in each of the 43 cell types characterized.

Results

As shown in Table 1 and Figure E2, the 4,383 participants in COPDGene had a mean age of 65.6 ± 8.7 years, 51.7% were women, and 38.4% were current and 61.6% former smokers. Participants had mean QIA occupying 5.7 ± 4.6% of the lung, mean percentage predicted FEV₁ of 77.7 ± 24.4%, and mean percentage predicted FVC of 86.7 ± 17.8%. In CARDIA, the 2,925 participants were younger, with a mean age of 50.2 ± 3.6 years, and were similarly split, with 56.1% women, 26.8% current smokers, 22.3% former smokers, and 50.9% never-smokers. Participants had less mean QIA occupying the lung of 2.6 ± 5.1%, higher mean percentage predicted FEV₁ of 93.6 ± 14.4%, and higher mean percentage predicted FVC of 96.0 ± 13.8%.

Table 1.

Baseline Characteristics

	COPDGene (n = 4,383)	CARDIA (n = 2,925)
	Age, yr, mean ± SD	65.6 ± 8.7	50.2 ± 3.6
Women, n (%)	2,264 (51.7)	1,641 (56.1)
Men, n (%)	2,119 (48.3)	1,284 (43.9)
Self-reported White, n (%)	3,130 (71.4)	1,568 (53.6)
Self-reported Black, n (%)	1,253 (28.6)	1,357 (46.4)
BMI, kg/m2, mean ± SD	29.0 ± 6.4	30.3 ± 7.1
Current smokers, n (%)	1,681 (38.4)	785 (26.8)
Former smokers, n (%)	2,702 (61.6)	651 (22.3)
Never-smokers, n (%)	—	1,489 (50.9)
Pack-years, mean ± SD	44.0 ± 23.9	5.5 ± 10.1
Percentage QIA, mean ± SD	5.7 ± 4.6	2.6 ± 5.1
25th percentile	2.6	0.2
Median	4.4	0.8
75th percentile	7.3	2.7
Percentage quantitative emphysema, mean ± SD	6.4 ± 13.6	1.0 ± 2.1
Percentage predicted FEV1, mean ± SD	77.7 ± 24.4	93.6 ± 14.4
Percentage predicted FVC, mean ± SD	86.7 ± 17.8	96.0 ± 13.8

Definition of abbreviations: BMI = body mass index; CARDIA = Coronary Artery Disease Risk in Young Adults; COPDGene = Genetic Epidemiology of Chronic Obstructive Pulmonary Disease; QIA = quantitative interstitial abnormality.

In multivariable regression in COPDGene, 852 proteins (818 GeneIDs) were significantly associated with QIAs, with 561 proteins positively and 291 negatively associated with QIAs (see Table E1). In CARDIA, 185 proteins (183 GeneIDs) were significantly associated with QIAs, of which 136 were positively and 49 negatively associated (see Table E2). Of these, 144 proteins (143 GeneIDs) had associations with QIAs that overlapped in COPDGene and CARDIA (Figure 1; see Table E3). The most significant positively and negatively associated proteins by pooled analysis (Table E4) are shown in Table 2. All but 1 of the 144 proteins shared directionality and magnitude between the two cohorts, with a Pearson correlation between regression coefficients across proteins of 0.92 (P < 2.2 × 10⁻¹⁶) (Figure 2). Of the 144 proteins, 140 remained significantly associated with QIAs and maintained directionality of association in both cohorts when adjusted for concurrent quantitative emphysema, another smoking-related CT phenotype (see Table E5). In separate analyses, 71 proteins were significantly associated with QIAs after adjustment for self-reported race (see Table E6).

Figure 1.

Volcano plot of proteins significantly associated with quantitative interstitial abnormalities independently in COPDGene (Genetic Epidemiology of Chronic Obstructive Pulmonary Disease) on the left and CARDIA (Coronary Artery Disease Risk in Young Adults) on the right. AFM = afamin; APOM = apolipoprotein M; ATP1B1 = Sodium/potassium-transporting ATPase subunit β-1; B2M = β-2-microglobulin; BAGE3 = B melanoma antigen 3; CAPG = macrophage-capping protein; CCDC126 = coiled-coil domain-containing protein 126; CCL18 = C-C motif chemokine 18; CDCP1 = CUB domain-containing protein 1; CDH11 = cadherin-11: extracellular domain; COPDGene = Genetic Epidemiology of Chronic Obstructive Pulmonary Disease; CXCL16 = C-X-C motif chemokine 16; EGFR = epidermal growth factor receptor; F2 = thrombin; FDR = false discovery rate; GDF15 = growth differentiation factor 15; LUM = lumican; MXRA8 = matrix remodeling–associated protein 8: extracellular domain; NPS = neuropeptide S; PLA2G12B = group XIIB secretory phospholipase A2-like protein; PROC = vitamin K–dependent protein C; RFU = relative fluorescent units; SVEP1 = sushi, von Willebrand factor type A, EGF and pentraxin domain-containing protein 1: EGF-like domains 4–6; THBS2 = thrombospondin-2; WFDC2 = WAP four-disulfide core domain protein 2; ZG16 = zymogen granule membrane protein 16.

Table 2.

Significant Multivariable Associations of Proteomics with Quantitative Interstitial Abnormalities in the Genetic Epidemiology of Chronic Obstructive Pulmonary Disease and Coronary Artery Disease Risk in Young Adults Cohorts

			COPDGene		CARDIA
Protein Name	Entrez Gene Symbol	UniProt ID	Percentage QIA per ln(RFU), Value (95% CI)	FDR-Corrected P Value	Percentage QIA per ln(RFU), Value (95% CI)	FDR-Corrected P Value
Positively associated with QIAs
Lumican	LUM	P51884	2.87 (2.21 to 3.54)	7.36 × 10−15	2.56 (1.55 to 3.58)	2.70 × 10−4
WAP four-disulfide core domain protein 2	WFDC2	Q14508	2.74 (2.35 to 3.13)	1.53 × 10−38	1.96 (1.26 to 2.67)	4.16 × 10−5
C-X-C motif chemokine 16	CXCL16	Q9H2A7	2.62 (1.95 to 3.28)	2.57 × 10−12	2.58 (1.62 to 3.54)	1.04 × 10−4
Leukocyte immunoglobulin-like receptor subfamily A member 5	LILRA5	A6NI73	2.28 (1.72 to 2.83)	1.49 × 10−13	1.53 (0.74 to 2.31)	8.95 × 10−3
Serine/arginine-rich splicing factor 6	SRSF6	Q13247	2.14 (1.64 to 2.65)	2.75 × 10−14	1.66 (0.94 to 2.37)	1.25 × 10−3
C-C motif chemokine 18	CCL18	P55774	2.06 (1.64 to 2.48)	9.18 × 10−19	0.99 (0.38 to 1.61)	4.51 × 10−2
β-2-microglobulin	B2M	P61769	1.97 (1.52 to 2.43)	1.09 × 10−14	1.78 (1.11 to 2.45)	1.09 × 10−4
CUB domain-containing protein 1	CDCP1	Q9H5V8	1.78 (1.5 to 2.06)	1.54 × 10−31	1.64 (1.17 to 2.11)	3.92 × 10−8
Triggering receptor expressed on myeloid cells 1	TREM1	Q9NP99	1.76 (1.33 to 2.19)	1.85 × 10−13	1.39 (0.78 to 2)	1.64 × 10−3
Sushi, von Willebrand factor type A, EGF and pentraxin domain-containing protein 1:EGF-like domains 4-6	SVEP1	Q4LDE5	1.62 (1.24 to 2)	1.98 × 10−14	1.73 (1.07 to 2.39)	1.46 × 10−4
Growth/differentiation factor 15	GDF15	Q99988	1.61 (1.31 to 1.91)	1.54 × 10−22	0.84 (0.33 to 1.35)	3.80 × 10−2
C-X-C motif chemokine 13	CXCL13	O43927	1.35 (1.03 to 1.68)	9.68 × 10−14	1.42 (0.81 to 2.04)	1.24 × 10−3
Thrombospondin-2	THBS2	P35442	1.34 (1.08 to 1.61)	8.28 × 10−20	0.91 (0.42 to 1.41)	1.41 × 10−2
Macrophage-capping protein	CAPG	P40121	1.29 (0.91 to 1.66)	1.82 × 10−9	1.66 (1.11 to 2.21)	5.08 × 10−6
Cadherin-11: extracellular domain	CDH11	P55287	0.94 (0.72 to 1.15)	1.86 × 10−15	0.75 (0.47 to 1.02)	1.04 × 10−4
Zymogen granule membrane protein 16	ZG16	O60844	0.22 (0.16 to 0.29)	7.92 × 10−9	0.21 (0.14 to 0.29)	1.56 × 10−5
Negatively associated with QIAs
Thrombin	F2	P00734	−1.39 (−1.75 to −1.03)	6.92 × 10−12	−1.53 (−2.29 to −0.77)	5.66 × 10−3
Interleukin-6 receptor subunit α	IL6R	P08887	−1.42 (−1.92 to −0.91)	1.61 × 10−6	−1.45 (−2.17 to −0.72)	6.36 × 10−3
A disintegrin and metalloproteinase with thrombospondin motifs 13	ADAMTS13	Q76LX8	−1.58 (−2.08 to −1.09)	2.60 × 10−8	−1.25 (−1.97 to −0.52)	2.72 × 10−2
Sodium/potassium-transporting ATPase subunit β-1	ATP1B1	P05026	−1.63 (−2.02 to −1.23)	1.64 × 10−13	−1.09 (−1.62 to −0.57)	3.69 × 10−3
Collagen α-1(XIII) chain	COL13A1	Q5TAT6	−1.89 (−2.52 to −1.26)	2.21 × 10−7	−2.28 (−3.32 to −1.23)	2.47 × 10−3
Afamin	AFM	P43652	−2.06 (−2.78 to −1.34)	9.49 × 10−7	−2.53 (−3.67 to −1.4)	1.80 × 10−3
Coiled-coil domain-containing protein 126	CCDC126	Q96EE4	−2.29 (−2.83 to −1.75)	3.55 × 10−14	−1.27 (−2.01 to −0.53)	2.81 × 10−2
Ciliary neurotrophic factor receptor subunit α	CNTFR	P26992	−2.32 (−3 to −1.64)	2.39 × 10−9	−1.79 (−2.75 to −0.83)	1.28 × 10−2
Group XIIB secretory phospholipase A2-like protein	PLA2G12B	Q9BX93	−2.33 (−2.82 to −1.84)	1.16 × 10−17	−1.2 (−1.91 to −0.5)	2.83 × 10−2
Contactin-1	CNTN1	Q12860	−2.36 (−3.11 to −1.61)	3.95 × 10−8	−1.8 (−2.88 to −0.72)	3.36 × 10−2
Matrix-remodeling-associated protein 8: extracellular domain	MXRA8	Q9BRK3	−2.39 (−3.04 to −1.74)	7.97 × 10−11	−2.07 (−3 to −1.15)	1.67 × 10−3
B melanoma antigen 3	BAGE3	Q86Y29	−2.76 (−3.56 to −1.96)	1.47 × 10−9	−2.29 (−3.47 to −1.12)	8.36 × 10−3
Vitamin K–dependent protein C	PROC	P04070	−2.79 (−3.51 to −2.07)	5.26 × 10−12	−2.09 (−3.32 to −0.86)	2.96 × 10−2
Apolipoprotein M	APOM	O95445	−2.85 (−3.43 to −2.28)	1.87 × 10−19	−1.49 (−2.43 to −0.55)	4.90 × 10−2
Neuropeptide S	NPS	P0C0P6	−3.28 (−4.12 to −2.44)	4.39 × 10−12	−2.68 (−4.05 to −1.31)	8.36 × 10−3
Epidermal growth factor receptor	EGFR	P00533	−4.35 (−5.26 to −3.44)	4.73 × 10−18	−3.4 (−4.8 to −2)	5.98 × 10−4

Definition of abbreviations: CARDIA = Coronary Artery Disease Risk in Young Adults, COPDGene = Genetic Epidemiology of Chronic Obstructive Pulmonary Disease; FDR = Benjamini-Hochberg false discovery rate; ID = identifier; QIA = quantitative interstitial abnormality; RFU = relative fluorescent units.

The top 16 positively associated and top 16 negatively associated proteins with the strongest significance by pooled analysis FDR-corrected P value are shown. All associations are adjusted for age, sex, body mass index, smoking status, and study center as a random effect. COPDGene also adjusted for white blood cell count and platelet count.

Figure 2.

Proteins significantly associated with quantitative interstitial abnormalities in CARDIA (Coronary Artery Disease Risk in Young Adults) and COPDGene (Genetic Epidemiology of Chronic Obstructive Pulmonary Disease), with β coefficients of associations in CARDIA on the x-axis and in COPDGene on the y-axis. ADAMTS13 = A disintegrin and metalloproteinase with thrombospondin motifs 13; AFM = afamin; AGER = receptor for advanced glycation end-products; APOM = apolipoprotein M; BAGE3 = B melanoma antigen 3; CCDC126 = coiled-coil domain-containing protein 126; CCL18 = C-C motif chemokine 18; CHI3L1 = chitin-3-like protein 1; CNTFR = ciliary neurotrophic factor receptor subunit α; CNTN1 = contactin-1; COL13A1 = collagen α-1(XIII) chain; CXCL16 = C-X-C motif chemokine 16; EGFR = epidermal growth factor receptor; F2 = thrombin; IL6R = IL-6 receptor subunit α; LILRA5 = leukocyte immunoglobulin-like receptor subfamily A member 5; LUM = lumican; MXRA8 = matrix remodeling–associated protein 8: extracellular domain; NPS = neuropeptide S; PLA2G12B = group XIIB secretory phospholipase A2-like protein; PROC = vitamin K–dependent protein C; SPON1 = spondin-1; WFDC2 = WAP four-disulfide core domain protein 2; ZG16 = zymogen granule membrane protein 16.

Pathway enrichment analysis showed overrepresentation of proteins involved in multiple pathways, including GO biological processes in inflammatory response, cell adhesion, immune response, ERK1/2 regulation, and signaling; cellular components in extracellular regions, spaces, exosome, and matrix; and molecular functions including calcium ion and heparin binding (Figure 3; see Table E7).

Figure 3.

Pathway enrichment analysis of proteins associated with quantitative interstitial abnormalities shared in COPDGene and CARDIA participants. CARDIA = Coronary Artery Disease Risk in Young Adults; COPDGene = Genetic Epidemiology of Chronic Obstructive Pulmonary Disease.

Tissue-specific deconvolution analysis of the proteins showed that the gene transcripts of the proteins positively associated with QIAs were most commonly expressed in the lungs, further suggesting that plasma proteomic biomarkers may be beneficial in understanding pulmonary disease processes (Figure 4). Cell-specific deconvolution analysis using a single-cell dataset comparing human lung tissue in individuals with ILD versus control subjects showed that the proteins associated with QIAs are associated with perturbations in gene expression across numerous cell types, including the pulmonary vasculature and capillaries, adventitial and alveolar fibroblasts, pericytes, profibrotic epithelial cells, and interstitial macrophages (see Figure E3).

Figure 4.

Results of the tissue-specific deconvolution of the significant proteins using the Genotype-Tissue Expression database. The gene expression activities of the proteins positively significantly associated with quantitative interstitial abnormalities were scored using the R package singscore for each tissue, which were used to rank the 37 tissues in COPDGene (left) and CARDIA (right). CARDIA = Coronary Artery Disease Risk in Young Adults; COPDGene = Genetic Epidemiology of Chronic Obstructive Pulmonary Disease.

For sensitivity analysis, given that smoking is an important risk factor for QIAs and many advanced lung diseases and that COPDGene is an older, ever-smoking cohort, we restricted the CARDIA population to the 876 ever-smokers (former and current smokers) who were at least 50 years old (the minimum recruitment age of COPDGene; see Table E8) and ran the analyses on the 842 proteins associated with QIAs in COPDGene. After adjusting for multiple testing in the smaller sample, 201 proteins (195 GeneIDs) were independently associated with QIAs (see Table E9). Next, as QIAs are present even in participants without histories of cigarette use, we ran the analyses restricted to the subset of 1,490 never-smokers in CARDIA and found that 49 of those proteins (48 GeneIDs) were independently associated with QIAs (see Table E10). Of these, 19 proteins (19 GeneIDs) were shared between the older ever-smoker subset and the never-smoker subset; the magnitudes of effects were greater in the older ever-smokers, demonstrating that smoking status and advanced age are important effect modifiers in the associations of each of the 19 proteins with QIAs (see Table E11).

Last, analysis of the full CARDIA 7K assay, which contains 2,356 additional unique aptamers not available in the COPDGene 5K assay, yielded 68 additional unique proteins (67 GeneIDs) that were significantly associated with QIA after adjustment for covariates (see Table E12). These associations were not entered into the pathway enrichment analysis, as they need to be validated in other cohorts.

Discussion

This is the first study using a large-scale proteomics panel to uncover novel protein biomarkers associated with QIAs, a quantitative CT-based measurement of early lung injury. Prior studies of other quantitative measures of lung injury have shown associations with injury biomarkers, adhesion molecules, collagen, and adipokines, but they focused on a limited number of targeted proteins (20–23). Using a large global proteomics panel, we found 144 unique proteins associated with QIAs shared across two cohorts that spanned middle age to older adulthood. These proteins included novel biomarkers as well as those previously associated with lung injury, inflammation, pulmonary fibrosis, COPD, emphysema, and lung cancer, as described below, thus suggesting potential shared pathways between QIA and advanced parenchymal diseases. Many of these associations were present even in the younger community-based cohort of adults with varied smoking histories, including never-smokers.

A strength of our study is that we identified and described the proteomic biomarkers of QIAs that are present both in an older, smoking population with a higher prevalence of pulmonary disease and in a younger, healthier, community-based cohort that represents an upstream population earlier in the life course. All but 1 of the 144 unique shared proteins had the same direction and relative magnitude of associations between the two cohorts, showing that these proteins have similar associations with QIAs, even though the CARDIA cohort is younger and less enriched for smoking exposure and parenchymal disease. Such associations with QIAs suggest that insults to the lung and disease pathogenesis start years before the manifestation of chronic lung disease and that systemic biological markers of injury may be detectable and important even without the clinical or radiographic presence of advanced, severe disease. These protein biomarkers bring initial biological insight into QIAs as imaging precursors to chronic lung diseases. Our results also highlight the crucial need to study these biomarkers in younger populations in whom prevention, disease interception, and even reversal of progression from lung health to disease may be more advantageous and feasible (8, 24).

Of particular interest are the 49 proteins that were associated with QIAs in the never-smoker subset. These associations suggest that although smoking is an important risk factor for QIA, it is not the only source of injury and inflammation and further support the idea that QIAs represent a heterogeneous condition encompassing several subtypes and pathways of progression. Protein biomarkers may be especially valuable in the detection and stratification of patients with QIAs who do not have the classic risk factors of heavy smoking use or advanced age but are nonetheless at high risk of advanced lung disease. In addition, we found 184 proteins that may have more important associations with QIAs among older ever-smokers by validating these associations within the two cohorts.

Pathway analysis demonstrated high overrepresentation of proteins in the extracellular region and space. The pulmonary extracellular matrix (ECM) contains an estimated 300 proteins, including the collagens, elastin, fibrin, fibronectin, laminin, proteoglycans, and glycoproteins that compose the structural framework; the integrins and other cell adhesion receptors that create feedback connections between pulmonary cells and the ECM; growth factors, cytokines, and chemokines for which the ECM serves as a reservoir; and the metalloproteinases that regulate modification and degradation (25). Injury and aging cause modification and remodeling of a number of these proteins, subsequently leading to chronic lung diseases such as pulmonary fibrosis and emphysema.

Some of the extracellular proteins positively associated with QIAs have previously been associated with the progression of COPD and pulmonary fibrosis and suggest shared progression pathways with QIA. For example, a recent study showed that SPON1 (spondin-1), which is believed to be involved in cell adhesion, was one of the 17 proteomic biomarkers of progressive fibrosing ILD (26). MMP12 (matrix metalloproteinase; also known as macrophage metalloelastase) concentrations are higher in patients with idiopathic pulmonary fibrosis (IPF) and associated with more radiographic fibrosis and lower diffusion capacity (27). MMP12 has also been shown to be required for the development of emphysema in mouse models exposed to smoke, and patients with COPD compared with control subjects have higher serum MMP12 (28, 29). Circulating collagens have been suggestive of ECM turnover in COPD and IPF (30). However, given collagen diversity, specific protein species associated with QIAs need to be studied further. For example, COL6A3 (collagen VI) is associated with pulmonary fibrosis, while COL28A1 (collagen XXVIII) is a relatively novel species found to be elevated in patients with lung cancer but not COPD (31). GDF15 (growth differentiation factor 15) is a ubiquitous aging-related protein that, in lung diseases, has been shown to be an important biomarker of cellular stress and mediator of senescence across the life span (32). GDF15 is upregulated in COPD and pulmonary fibrosis (33, 34). Last, several proteins associated with QIAs have previously been associated with visual interstitial lung abnormalities, including GDF15, CRP (C-reactive protein), TNFRSF1B (tumor necrosis factor receptor 2), and WFDC2 (WAP four-disulfide core domain protein 2), suggesting some shared pathways between QIAs and visual early pulmonary fibrosis (35, 36).

B2M (β-2-microglobulin), a component of major histocompatibility complex class I molecules and proaging protein, is particularly interesting to consider. Prior work showed that patients with emphysema have higher plasma B2M concentrations; the presence of B2M protein caused alveolar epithelial senescence and inhibition of proliferation, which was proposed to be a mechanism for the development of emphysema (37). A separate study demonstrated that among patients with COPD, those with relatively higher B2M concentrations had lower diffusing capacity and, on lung resection pathology, had thickened alveolar walls with increased B2M expression, suggestive of epithelial-to-mesenchymal transition and concluding that B2M may be a biomarker for the development of pulmonary fibrosis in patients with COPD (38). B2M may thus indicate a shared pathogenic pathway between fibrosis and emphysema and even with combined pulmonary fibrosis and emphysema; the association of B2M with QIAs raises the possibility that the origin of pathogenesis of these advanced diseases starts with QIAs.

CHI3L1 (chitin-3-like protein 1), also known as YKL-40 in humans, is another notable protein belonging to the extracellular pathways. Cigarette smoking and aging induce CHI3L1 expression in alveolar macrophages, airways, and pulmonary epithelial cells (39). CHI3L1 has been shown to protect against emphysematous alveolar apoptosis and destruction, but augments fibrotic responses (39, 40). In lung cancers, CHI3L1 is expressed by tumor cells and promotes angiogenesis, invasion, and migration (41, 42). Thus, CHI3L1 appears to have both protective and pathogenic roles in advanced lung diseases. The elevated CHI3L1 concentrations associated with QIAs confirm the presence of inflammatory and injury responses, and future work should explore whether CHI3L1 is beneficial or detrimental in QIA progression.

Of note, several other biomarkers of lung cancers were associated with QIAs. In particular, CDCP1 (CUB domain-containing protein 1) is a marker of inflammation and a transmembrane protein overexpressed in many cancers, believed to lead to a loss of cell adhesion and increased metastatic risk (43). A recent study showed that dysregulation of circulating CDCP1 can be seen years before the diagnosis of lung cancer, independent of cigarette smoking (44). The associations of QIAs with CDCP1 were consistent across both of our cohorts, as well as within the older ever-smoker and never-smoker subsets, suggesting that QIAs are associated with systemic inflammation and increased lung cancer risk.

Our analysis showed several C-C and C-X-C motif chemokines are significantly associated with QIAs, and they represent diverse roles, including neutrophil, lymphocyte, monocyte, and eosinophil chemotaxis. Chemokines broadly have been shown to be important in the pathogenesis of advanced lung diseases such as COPD and pulmonary fibrosis, and although understanding the downstream effects will require further mechanistic studies, as a whole these results suggest that aberrant inflammatory and immune responses contribute to QIA pathogenesis.

One of the proteins negatively associated with QIA was RAGE (receptor for advanced glycation end-products; also called AGER). Although RAGE is a proinflammatory signaling receptor, and soluble RAGE (sRAGE) acts as a competitive receptor and has a protective role. Prior work have proposed sRAGE as a biomarker of COPD, as lower sRAGE concentrations are associated with clinically stable COPD, as well as spirometric and emphysema progression (45). Studies have also shown that lower concentrations of RAGE are associated with progressive fibrosing ILD and with IPF compared with control subjects (26, 33). More generally, sRAGE has been shown to have a protective role in lung function in both smokers and nonsmokers in causal analysis (46). Thus, the negative associations of sRAGE with QIAs may reflect participants’ susceptibility to pulmonary decline, and it may be an important biomarker in identifying those participants that have future QIA progression.

Overall, our analyses demonstrated associations of QIAs with proteins that have important roles in inflammation, chemotaxis, fibroblast interaction, response to cellular stressors, cell apoptosis and senescence, and tumor activity, which are often seen with advanced pulmonary diseases and lung cancers. Deconvolution using reference tissue single-cell RNA sequencing showed that these proteins are associated with aberrant expression not only in the epithelium, fibroblasts, and immune cells but also in the pulmonary capillaries and pericytes, suggesting that epithelial and vascular injury and abnormal inflammatory responses may be present in early lung diseases such as QIAs. Deconvolution using GTEx mapping showed that the plasma proteins positively associated with QIA were most highly expressed in the lungs, providing further support for the idea that biological markers of early lung injury can be systemically detectable in the plasma.

There are several limitations to our study. First, we used a high-throughput aptamer-based assay. Although these results give us biological insights, we cannot elucidate mechanisms, and they should be confirmed using complementary assays such as liquid chromatography–mass spectrometry or ELISAs. Because of the cross-sectional nature of our data, we can conclude associations about protein biomarkers and QIAs but not causality. Although we adjusted for several important covariates in our analyses, significant proteins may nonetheless be shared with other lung diseases or conditions, limiting the specificity of associations. Although the background datasets used in pathway enrichment analysis and GTEx deconvolution are large, they are nonetheless incomplete.

Conclusions

We identified 144 proteomic biomarkers and 49 enriched pathways associated with QIAs in an older smoking population and a younger community-based population. The identified proteins and pathways give us initial biological insight into QIA, and they may be useful as future biomarkers to improve detection of early precursors of advanced parenchymal diseases and potential therapeutic targets for disease interception.