Circulating biomarkers of airflow limitation across the life span

Abstract

Background: Airflow limitation is a hallmark of chronic obstructive pulmonary disease, which can develop through different lung function trajectories across the life span. There is a need for longitudinal studies aimed at identifying circulating biomarkers of airflow limitation across different stages of life.

Objectives: This study sought to identify a signature of serum proteins associated with airflow limitation and evaluate their relation to lung function longitudinally in adults and children.

Methods: This study used data from 3 adult cohorts (TESAOD [Tucson Epidemiological Study of Airway Obstructive Disease], SAPALDIA [Swiss Cohort Study on Air Pollution and Lung and Heart Diseases in Adults], LSC [Lovelace Smoker Cohort]) and 1 birth cohort (TCRS [Tucson Children's Respiratory Study]) (N = 1940). In TESAOD, among 46 circulating proteins, we identified those associated with FEV₁/forced vital capacity (FVC) percent (%) predicted levels and generated a score based on the sum of their z-scores. Cross-sectional analyses were used to test the score for association with concomitant lung function. Longitudinal analyses were used to test the score for association with subsequent lung function growth in childhood and decline in adult life.

Results: After false discovery rate adjustment, serum levels of 5 proteins (HP, carcinoembryonic antigen, ICAM1, CRP, TIMP1) were associated with percent predicted levels of FEV₁/FVC and FEV₁ in TESAOD. In cross-sectional multivariate analyses the 5-biomarker score was associated with FEV₁ % predicted in all adult cohorts (meta-analyzed FEV₁ decrease for 1-SD score increase: -2.9%; 95% CI: -3.9%, -1.9%; P = 2.4 × 10^-16). In multivariate longitudinal analyses, the biomarker score at 6 years of age was inversely associated with FEV₁ and FEV₁/FVC levels attained by young adult life (P = .02 and .005, respectively). In adults, persistently high levels of the biomarker score were associated with subsequent accelerated decline of FEV₁ and FEV₁/FVC (P = .01 and .001).

Conclusions: A signature of 5 circulating biomarkers of airflow limitation was associated with both impaired lung function growth in childhood and accelerated lung function decline in adult life, indicating that these proteins may be involved in multiple lung function trajectories leading to chronic obstructive pulmonary disease.

Keywords: Airflow limitation; biomarkers; lung function.

Figure 1.

Overall study design. The TESAOD cohort was used for biomarker selection and score development. Cross-sectional associations of the biomarker score with lung function were validated in SAPALDIA and LSC. The biomarker score was then used in longitudinal analyses to predict subsequent decline of lung function in TESAOD and lung function growth from childhood into adult life in TCRS.

Figure 2.

Cross-sectional associations of the biomarker score with FEV₁ % predicted in the discovery TESAOD and validation SAPALDIA cohorts. Figure 2a shows the estimated reduction in FEV₁ % predicted for each SD increase in the biomarker score after adjustment for BMI, smoking, and physician-confirmed asthma in each cohort and from inter-cohort meta-analyses. Figure 2b shows the estimated adjusted reduction in FEV₁ % predicted associated with the second, third, and fourth quartile of the biomarker score (first quartile is the reference) after adjustment for BMI, smoking, and physician-confirmed asthma in each cohort and from inter-cohort meta-analyses. Footnotes: Sex and age were not included as covariates in these models because the standardized biomarkers had been already adjusted for sex and age The biomarker score was computed as the standardized sum of the z-scores of five proteins (Haptoglobin, CEA, ICAM1, CRP, and TIMP1) within each cohort.

Figure 3.

Estimated levels of FEV₁ (3a), FVC (3b), and FEV₁/FVC (3c) % predicted at survey Bio-2 and during the study follow-up across the three biomarker longitudinal categories in the adult TESAOD cohort. Estimates are from random effects models adjusted for sex, age, BMI, smoking, and asthma. The lines represent predicted values for a female former smoker who was 58-year old at the time of biomarker measurements (i.e., at survey Bio-2). Footnotes: At each of the TESAOD Bio-1 and Bio-2 surveys, the biomarker score was computed as the standardized sum of the z-scores of five proteins (Haptoglobin, CEA, ICAM1, CRP, and TIMP1) and categorized in quartiles. Three biomarker longitudinal categories were then generated: “never high” (not being in the highest quartile at either survey); “inconsistently high” (being in the highest quartile at one but not both surveys); and “persistently high” (being in the highest quartile at both surveys). Random effects models included a set of fixed effects (sex, age, BMI, smoking, and asthma), the random intercept for the level of subject, and the random slope for the follow-up survey. An unstructured covariance matrix was used. N participants = 545; N observations = 2,946. P values from random effects models for overall differences across the follow-up period: For FEV₁ % predicted (3a), p=0.03 (Inconsistently high vs. Never high) and p<0.001 (Persistently high vs. Never high) For FVC % predicted (3b), p=0.02 (Inconsistently high vs. Never high) and p<0.001 (Persistently high vs. Never high) For FEV₁/FVC % predicted (3c), p=NS (Inconsistently high vs. Never high) and p=0.003 (Persistently high vs. Never high).

Figure 4.

Levels of FEV₁ (4a), FVC (4b), and FEV₁/FVC (4c) (expressed as sex-, age- and height-adjusted z-scores) from age 11 up to age 36 years among TCRS participants, according to their levels of the biomarker score (above or below median) at age 6 years. Footnotes: At age 6 years, the biomarker score was computed as the standardized sum of the z-scores of five proteins (Haptoglobin, CEA, ICAM1, CRP, and TIMP1) and participants were categorized into two groups based on whether their score levels were below or above the median. P values are from random effects models (age 11-36 years; N=326 participants; N=1,312 observations) with the random intercept for the level of subject. An unstructured covariance matrix was used.