Value of metalloproteinases in predicting COPD in heavy urban smokers

We enrolled 316 study subjects from January 1, 2010, to December 31, 2015. Subjects were excluded if CT-scans indicated other disease findings, were irretrievable from outside hospitals or no blood samples were available (See Fig. 1). Two hundred and fifty six of these remaining subjects were > 50 years of age, most with smoking histories of over 20 pack-years who consented to participate in the lung cancer screening protocol of the NYU Lung Cancer Biomarker Center. Among those 256 subjects, 34 samples were excluded due to technical errors that occurred during assay processing. The distribution of the demographics and clinical characteristics of these 34 subjects were similar to the remaining subjects in the dataset. After excluding these patients, samples from the remaining 222 patients were analyzed. This study was approved by the NYU Institutional Review Board. Each participant completed a low-dose computerized tomography (CT) of the chest, a baseline demographic questionnaire, and a peripheral blood draw. We utilized the American Thoracic Society (ATS) respiratory questionnaire which included questions on demographic characteristics, tobacco use, occupation, occupational exposures, alcohol use, family history, and medical history (Supporting Document). Spirometry was performed by a trained technician according to ATS standards.

CT-scans, the gold standard for detecting emphysema, utilized 64 multi-detector scanners obtaining 5 mm–thick images; images were reconstructed every 5 mm using 1 mm collimation. This technique allowed simultaneous prospective reconstruction of contiguous 1 mm images for high-resolution detailed analysis. CT-scans were read by a thoracic radiologist and confirmed by two additional pulmonologists for the presence of emphysema using a qualitative assessment.

The Luminex bead multiplex assay (Austin, Texas, USA) was utilized for measuring serum MMPs -1, -2, -7, -9, -10 and TIMPs -1 and -2, as per manufacturer’s instruction. The patients were classified into two groups: smokers with emphysema (n = 101), and smokers without emphysema (n = 121).

In addition to detecting emphysema we sought to evaluate whether a panel of biomarkers (MMPs -1, -2, -7, -9, and -10, and TIMPs -1 and -2) could distinguish emphysema from no emphysema among heavy smokers. Distributions of patients and disease characteristics including clinical features, smoking history, pulmonary function, and other variables were compared using frequency distributions and proportions using Fisher’s exact tests for categorical variables.

Seventy-eight of the 222 eligible subjects (35%) had undetectable MMP-10 measurements below the limits of detection. Therefore, we used binary classes (detectable, undetectable: detectable if value ≥27 pg/ml and undetectable if value < 27 pg/ml) to explore the potential utility of this marker.

Biomarker levels were measured in two batches because limited laboratory personnel and assay kits were available to pull the sequentially identified samples for thaw and aliquot. Batch 1 had 44 smokers with emphysema and 68 without emphysema, while in Batch 2 there were 57 with emphysema and 53 non-emphysema. The analyses of the batch effects using Fisher’s exact test suggested that the distributions of subjects with and without emphysema did not differ significantly between batches (56% of batch 1 and 44% of batch 2 had no emphysema (Supplement Table 1, p = 0.08). Nonparametric Wilcoxon rank-sum tests were used to evaluate differences in the distributions of individual biomarkers between the two batches with respect to emphysema and non-emphysema separately. The results indicated that all biomarkers had significantly different distributions by batch with the exception of MMP-1 (Supplement Table 2). Batch effects were adjusted with a log₂ transformation to each biomarker to remove skewness and the biomarker levels were standardized within each batch. After standardization within batches, the Wilcoxon rank sum test indicated that the distributions of the biomarkers of emphysema (and of smokers) did not differ significantly between the two batches (Supplement Table 2).

We first used stepwise multiple logistic regression with bidirectional variable selection to distinguish smokers with emphysema from those without emphysema, including only their clinical characteristics (Model 1: Age, BMI, pack-years, FEV1/FVC ratio). Next, we used the same regression method and included both clinical variables and biomarkers (MMP1, MMP2, MMP7, MMP9, TIMP1 and TIMP2) to obtain Model 2. An additional model that included only the set of biomarkers (MMPs/TIMPs) was also evaluated (Model 3). Receiving Operating Characteristic (ROC) curves were plotted from these logistic models; Areas Under the Curve (AUC) and Akaike Information Criterion (AIC) [9] for the resulting models were calculated under these three scenarios for comparison of these models. The Youden Index [10] was used to identify the optimal cut points for the models and estimates of sensitivity and specificity of each model provided at the cut point. All data analyses were performed with R version 3.4.3, MASS package version 7.3-48, and pROC package 1.15.0.

The FEV₁/FVC ratio was lower in those with radiographic emphysema (emphysema 68 ± 11%(SD) versus non-emphysema 76 ± 8%(SD); p < 0.0001). A higher percentage of emphysema subjects had symptoms of cough (emphysema 29%, non-emphysema 17%; p = 0.04) and reports of phlegm (emphysema 14%, non-emphysema 7%; p = 0.11) than those without emphysema. Next, three stepwise multiple logistic models were conducted to explore the extent to which MMP biomarkers could help identify emphysema among smokers. Table 2 shows the results of Model 1, a stepwise multiple logistic regression with bidirectional elimination using clinical characteristics including age, smoking duration, pack-years, cough and FEV1/FVC ratio. With the exception of smoking duration, the other 4 variables were all highly associated with emphysema among smokers.

In order to explore the potential ability to predict emphysema using biomarker levels, we added biomarkers log₂ (TIMP-1), log₂ (TIMP-2), log₂ (MMP-1), log₂ (MMP-2), log₂ (MMP-7) and log₂ (MMP-9) and obtained Model 2 (Table 3). Age, FEV1/FVC and pack-years remained statistically significant, but none of the biomarkers showed a significant association with emphysema. We further compared the AIC of these models. The AIC of Model 1 is 238.1 and the AIC of Model 2 is 237.93 with the addition of the biomarkers as predictors. The ROC curves for Models 1 and 2 are shown in Fig. 2, with AUC as 0.788 and 0.789 respectively. The optimal cut-point based on the Youden Index for Model 1 is 0.405 (sensitivity 0.789 and specificity 0.703), and for Model 2 is 0.500 (sensitivity 0.674 and specificity 0.802) (Fig. 2). These results suggest that the serum-based MMP biomarkers do not add additional value to clinical characteristics predicting emphysema in smokers.