Survivin Autoantibodies Are Not Elevated in Lung Cancer When Assayed Controlling for Specificity and Smoking Status

Long-term survival in lung cancer depends mainly on early detection and immediate start of treatment of the tumor (1). During tumor development, tumor-antigen expression elicits cellular and humoral immune responses (2–4). Identifying autoantibodies to tumor-associated antigens (TAA) is thought to be a promising method of early lung cancer diagnosis (5–8), especially because these autoantibodies have been found up to 5 years before CT detection (9, 10).

Antibodies to survivin are one of the autoantibodies described most frequently in lung cancer (5, 11–18). Survivin, also known as baculoviral IAP repeat-containing protein 5 (BIRC5), is a member of the inhibitor apoptosis proteins (IAP) family. It promotes cell proliferation and inhibits apoptosis, thereby favoring the growth and progression of transformed cells and tumors. Although survivin is abundantly expressed in fetal cells, transformed cell lines, and various tumors, it is undetectable in most normal, differentiated adult tissues (19). If overexpressed in lung cancer, it may lead to antibody responses to this protein. Various studies have compared amounts of survivin autoantibodies in lung cancer patients and healthy blood-donor control subjects; the presence of antibodies against survivin in lung cancer sera collected after diagnosis was reported to range between 8% (16) and 58% (15).

In our institute, a well-controlled multicenter population study was conducted aiming at early detection of lung cancer with specific emphasis on smoking habit, the NELSON trial (20, 21). We wanted to test the hypothesis that, in this population, anti-survivin autoantibodies are detectable before the radiologic diagnosis of non–small cell lung cancer (NSCLC) and establish how these autoantibodies may emerge over time in the disease. For that purpose, we collected, NSCLC cases from the NELSON trial with well-matched controls to the NSCLC cases, subdivided into individuals who were currently smoking or were former smokers. In addition, we included late-stage NSCLC patients from the Dutch Association of Pulmonologists for Lung Diseases and Tuberculosis (NVALT)-12 study (22) and a control group of generally healthy blood donors, similar to that used in the publications on survivin autoantibodies (5, 11–18).

All published studies on survivin autoantibodies in lung cancer have used the direct antigen-coating (DAC) form of an ELISA with a recombinant survivin (5, 11–18). However, the DAC-ELISA assay can give false-positive results if antibodies bind to impurities in the antigen preparation. In our study we therefore chose to assess the presence of autoantibodies to survivin with the sandwich ELISA. In this assay, a highly specific capture antibody is absorbed to the solid phase and incubated with an antigen solution. Only autoantibodies to survivin are then detected in the serum samples. At the cost of less signal, this greatly increases the specificity of the assay.

We obtained serum from 50 NSCLC cases (82% adenocarcinoma; 90% stage I/II) before and after diagnosis and from 50 smoking-habit controls drawn from current and former smokers in the NELSON trial, as described previously (20, 21). Smoking-habit controls were matched to the NSCLC cases for age, gender, smoking status, smoking duration, number of cigarettes smoked per day, chronic obstructive pulmonary disease (COPD) status, asbestos exposure, and blood collection center (Supplementary Table S1). For the reference population, serum was derived from 50 healthy nonsmoking blood donors of the Sanquin Blood Supply Rotterdam. These nonsmoking controls were matched to cases and smoking-habit controls for age and gender (Supplementary Table S1). From the NVALT-12 study, we obtained baseline serum from 20 patients (10 men and 10 women; 6 current smokers, 12 former smokers, 2 nonsmokers; mean age, 63.2 ± 9.1 years) with stage IV NSCLC (80% adenocarcinoma; 20% large cell carcinoma). From Sanquin Blood Supply Rotterdam we obtained 7 independent serum samples from healthy nonsmoking controls with a normal serum IgG (mean 10.5 ± 1.78 g/L). Professor Anastasios E. Germenis (16) made available 9 serum samples labeled positive for anti-survivin antibodies by the DAC assay from NSCLC patients (8 men and 1 woman; mean age, 57.6 ± 8.0 years) in his laboratory. Upon diagnosis, these patients—part of a cohort of 117 NSCLC patients (108 males and 9 females; mean age, 64.2 ± 9.3 years)—had been shown by DAC-ELISA to have higher anti-survivin antibody levels than 100 healthy, age- and gender-matched individuals (99 men and 1 woman; mean age, 68.2 ± 5.8 years).

For all human samples, a medical ethical statement according to the Declaration of Helsinki was given by the METC Erasmus MC Rotterdam. Written informed consent was obtained from all participants for the use of their samples.

During the participants' visits to the center, one serum gel tube was collected per participant. The venous blood was allowed to clot, and was centrifuged for 10 minutes at 1400× g and 4°C within 2 hours after collection. After centrifugation, the serum was stored immediately in aliquots at −80°C. All samples were blinded and analyzed in random order.

Antibody reactivity against recombinant full-length human survivin fused to calmodulin (CaM)-tag (BIRC5, Abcam) was measured by DAC-ELISA as described previously (16) and by sandwich ELISA. To validate that both assays detected survivin autoantibodies and to assess their dynamic ranges, we used a rabbit monoclonal antibody (mAb) to survivin (0.06–10,00 ng/mL; Abcam) to establish standard curves (Supplementary Fig. S1). The standard curve of rabbit anti-survivin showed a dynamic range from 0.06 to 40 ng/mL in the DAC-ELISA and from 0.06 to 1000 ng/mL in the sandwich ELISA (Supplementary Fig. S1). To determine if these assays could distinguish human autoantibody–positive from –negative sera, we analyzed serum from 7 NSCLC patients that had been previously reported (16) to be positive for survivin autoantibodies and from 7 healthy nonsmoking control subjects, the type of control group used by Karanikas and colleagues (16). The replicability of our sandwich ELISA was established by repeatedly testing a single sample for intra-assay variation, as well as testing inter-assay variation. The intra-assay coefficient of variation (CV) for the optical densities (OD) with survivin of 20 replicates of 1 serum sample from a nonsmoking control was 6.9%, the interassay (plate-to-plate) CV for 6 serum samples from nonsmoking controls ranged from 5.3% to 11.0%.

One half of a microtiter plate (Nunc-Immuno Maxisorp flat bottom, Thermo Scientific) was incubated with 100 μL of the 2 μg/mL recombinant CaM-tagged survivin in 0.05 mol/L carbonate buffer (pH 9.6) and the other half only with 0.05 mol/L carbonate buffer for 20 hours at 4°C. After washing 5 times with phosphate-buffered saline containing 0.05% Tween-20 (PBST), plates were blocked with 200 μL 5% bovine serum albumin (BSA; Sigma) in PBST for 20 hours at 4°C. Plates were washed 5 times with PBST, and wells with and without survivin on the same plate were incubated for 1 hour at room temperature with 100 μL of serum sample either diluted 1:40, as described previously (16), diluted 1:100, or with serial dilutions of rabbit anti-survivin in PBST. After washing as before, plates were incubated for 1 hour at room temperature with 100 μL of horseradish peroxidase (HRP)–conjugated goat anti-human IgG (H+L; 100 ng/mL; Vector) or HRP-conjugated goat anti-rabbit IgG (Fc) (80 ng/mL; Jackson) in 1% BSA-PBST. Plates were washed 6 times, and enzyme activity was visualized by adding 100 μL of 3,3′,5,5′-tetramethylbenzidine substrate (TMB; Sigma-Aldrich). Absorbance at 450 nm was measured after 10 minutes. Absorbance values were corrected by blank subtraction.

Microtiter plates were coated overnight at 4°C with 100 μL of 0.50 μg/mL Cam-tag specific mAb (Santa Cruz Biotechnology) in 0.05 mol/L carbonate buffer. Plates were washed 5 times with PBST and blocked with 200 μL 5% BSA-PBST for 1.5 hours at room temperature. After washing 5 times with PBST, one half of the plate was incubated with 100 μL of 400 ng/mL recombinant CaM-tagged survivin in 5% BSA-PBST, and the other half with only 5% BSA-PBST for 3.5 hours at room temperature. Plates were washed as before, and 100 μL of serum sample diluted 1:100 or serial dilutions of rabbit anti-survivin in PBST were added to wells with and without survivin on the same plate. After 1-hour incubation at room temperature followed by washing, wells were incubated with 100 μL of (HRP)-conjugated goat anti-human IgG (H+L) (100 ng/mL) or (HRP)-conjugated goat anti-rabbit IgG (Fc) (80 ng/mL) in 1% BSA-PBST for 1 hour at room temperature. After washing 6 times, enzyme activity was visualized by adding 100 μL of TMB. After 10-minute incubation, absorbance at 450 nm was measured. Absorbance values were corrected by blank subtraction.

The linearity of the sandwich ELISA was determined by diluting the rabbit anti-survivin positive control into 6 nonsmoking control serum samples and 10 NSCLC patient serum samples, to create a standard curve. ODs of serial dilutions of rabbit anti-survivin in human sera were identical to those spiked in diluent. To validate the utility and specificity of our secondary antibody (goat anti-human IgG (H+L)), we designed a similar sandwich ELISA using as a positive control a different human protein and sera that were known to contain autoantibodies to it. Patients with paraneoplastic neurologic syndromes (PNS) can generate autoantibodies against the onconeural antigen HuD (23). Recombinant HuD antigen (ELAV4) was applied to serial dilutions of patient serum samples, known to be positive for anti-HuD antibodies. For this test, one healthy donor sample was used as negative control.

To determine nonspecific reactions, samples were analyzed in wells with and without survivin on the same plate. After blank subtraction, the net absorbance at 450 nm for each sample was calculated using the following equation: Net OD = OD with survivin − OD without survivin.

To determine the specificity of the DAC-ELISA and sandwich ELISA, selected samples diluted 1:200 were analyzed by Western blotting of CaM-tagged survivin (50 ng/lane). Rabbit anti-survivin serially diluted (1.6–200 ng/mL) was analyzed to check the sensitivity of the Western blot. Blots were incubated with a 1:10,000 dilution of IRDye 800CW goat anti-human IgG (H+L) or IRDye 680RD donkey anti-rabbit IgG (H+L), and scanned with an Odyssey Infrared Imager (LI-COR) to simultaneously visualize proteins at anti-human and anti-rabbit dual wavelengths.

Blots were performed using HEK-293 cell lysate as a source of endogenous survivin. For equal loading of proteins, 100 μL HEK-293 cell lysate was loaded into a single-well comb slot for Western blot analysis. After protein transfer, the blot was cut into strips containing equal amounts of protein per centimeter of strip. Antibody response to HEK-293 antigens was analyzed with either serum samples diluted at 1:200 or rabbit anti-survivin diluted at 1:2,500. Proteins were detected by fluorescence as described previously or by chemiluminescence using the ECL Western blotting substrate (Thermo Scientific) according to the protocol provided.

Potentially, posttranslational modification of survivin may be necessary for immune recognition and may be a reason that a positive survivin antibody serum is missed. We prepared human embryonic kidney (HEK)-293 cell lysate that endogenously produces survivin (24). HEK-293 cells were cultured in DMEM with GlutaMax (Life Technologies) supplemented with 10% fetal bovine serum, 1% penicillin, and 1% streptomycin at 37°C in 5% CO₂. The identity and homogeneity of the HEK-293 culture were verified by light-microscopic inspection of cell morphology and growth patterns. No additional authentication was performed. Cells were washed with PBS and lyzed in 1 mL SDS sample buffer (±10⁶ cells/mL). The cell lysate was centrifuged and stored in aliquots at −20°C until analysis.

Proteins in the HEK-293 cell lysate were separated by SDS-PAGE. Protein bands (band 1–5) in the 15- to 16-kDa range were tryptic digested for LC/MS measurement as described previously (25). Tryptic peptides were measured by LC/MS on an Ultimate 3000 nano-RSLC system (Dionex) coupled online to a hybrid linear ion trap/Orbitrap MS (Orbitrap Fusion, Thermo Fisher Scientific). The trap column was then switched online with the analytical column (PepMap C18, 75 μm ID × 500 mm, 3-μm particle and 100 Å pore size; Dionex), and peptides were eluted by a 90-minute gradient from 4% to 38% acetonitrile at a flow rate of 250 nL/minute. For electro-spray ionization (ESI), metal-coated nano ESI emitters (New Objective) were used at a spray voltage of 1.7 kV. Two different methods for MS detection were used. First, all samples were run according to a data-dependent shotgun method whereby MS1 survey scans of 400 to 1,600 m/z were acquired in the Orbitrap at 120,000 resolution. Subsequently, CID MS/MS spectra were acquired in the linear iontrap using the top speed mode of the instrument. Second, in a further measurement of protein band 2, the method described above was adjusted to exclusively trigger MS/MS spectra of precursor masses matched with the predicted tryptic peptides of survivin within a 12-ppm m/z window.

Before samples were included in this study, the clinical characteristics of nonsmoking controls, NSCLC cases, and smoking-habit controls were statistically analyzed. To detect statistically significant differences in absorbances (OD) among the data sets, the Mann–Whitney U test (two-tailed) was performed. Pearson correlation and linear regression analysis were performed to assess the relationship between the OD of samples measured in wells with survivin and those measured without survivin. All data were analyzed by SPSS (IBM SPSS Statistics 21). A P value of <0.05 was considered statistically significant.

Before we started our studies on the development of survivin autoantibodies in NSCLC, we tested both the DAC-ELISA and the sandwich ELISA for robustness and specificity.

The antibody response to human survivin was measured in 7 NSCLC serum samples previously found positive for survivin antibodies in the provider's laboratory (16) and in 7 serum samples from healthy nonsmoking controls. Tests were carried out at a 1:40 (Fig. 1A) and a 1:100 dilution (Supplementary Fig. S2). Comparisons between patients and healthy nonsmokers were made. The Mann–Whitney U test showed no significant difference between the NSCLC cases and the healthy nonsmoking controls at two different dilutions (1:40 dilution, P = 1.000; 1:100 dilution, P = 0.620). In addition, no significant differences were found in specific binding, between OD readings in assays that contained the antigen, survivin, and those without antigen, either at a 1:40 (P = 0.535) or at 1:100 (P = 0.535) dilution.

We first determined the replicability of our sandwich ELISA by repeatedly testing a single sample for intra-assay variation, as well as testing inter-assay variation (see Materials and Methods) and found no sample matrix interference in the sandwich ELISA (Supplementary Figs. S3 and S4).

To determine whether our sandwich ELISA technique can detect known human autoantibodies, we tested the assay using the onconeural antigen HuD (ELAV-like protein 4) and serial dilutions of an anti-HuD–positive serum from a patient with PNS. Serum from a healthy donor was used as a negative control. The patient serum gave a linear standard curve at serum dilutions of 1:10 to 1:640 and a relatively high OD (2.378) at the 1:100 dilution that is commonly used for immunohistology (Supplementary Fig. S5). The negative control serum had a low OD (0.147) at a dilution of 1:100. When antigen was omitted from the sandwich ELISA and serum used at a 1:100 dilution, the patient serum had a much lower OD (0.102), but the OD of the negative donor serum was virtually unchanged (0.146). Thus, the anti-human secondary antibody reaction works correctly in demonstrating the presence of immobilized antigen-bound human auto-antibody in our sandwich assay.

Having validated the sandwich ELISA as capable of detecting human autoantibodies, we retested the same 7 NSCLC patient and healthy control samples that we had assayed with the DAC-ELISA, at a serum dilution of 1:100. We found that specific binding was essentially zero, ranging from 0.000 to 0.008. We observed that ODs with and without antigen were significantly higher for patients [median, 0.108; interquartile range (IQR), 0.095–0.121] than for the nonsmoking control subjects (median, 0.075; IQR, 0.065–0.092; Mann–Whitney U test, P = 0.011) and were strongly correlated (Fig. 1B).

We tested sera for the presence of autoantibodies to survivin from 50 NSCLC cases, before and after diagnosis, 50 matched, smoking-habit controls from the NELSON trial, and another 50 nonsmoking controls (n = 200). All samples were randomized before measurement. Specific reactions (subtracting background values obtained in the absence of survivin in the sandwich ELISA) were almost zero, ranging from −0.004 to −0.001 (Fig. 2). However, as noted with the sera from the study by Karanikas and colleagues (16), the background median OD was significantly higher in NSCLC patients (P < 0.001) and smoking-habit controls than in nonsmoking controls (Fig. 2 and Table 1). Thus, the choice of control may be critical for these studies if background is not removed before reporting data.

We also tested for autoantibodies in sera of 20 NSCLC patients from the late-stage NVALT-12 study (median, 0.051; IQR, 0.027), 13 smoking-habit controls (median, 0.059; IQR 0.045; P = 0.253), and 13 nonsmoking controls (median, 0.062; IQR, 0.035; P = 0.113). None of these late-stage NSCLC cases was positive for antibody binding to survivin.

CaM-tagged survivin was used in Western blots for further testing of serum samples from 9 NSCLC patients previously found positive by DAC-ELISA (16), 5 patients before diagnosis of NSCLC, 5 patients after diagnosis of NSCLC, 2 smoking-habit controls, and 2 nonsmoking controls (Supplementary Fig. S6). To check the sensitivity of the Western blot, we analyzed rabbit anti-survivin serially diluted from 1:5,000 to 1:625,000 (1.6–200 ng/mL), finding a sensitivity of at least 1.6 ng/mL for rabbit anti-survivin antibodies. The 20 NVALT-12 study samples were also analyzed as described above. Sandwich ELISA and Western blot analysis both showed all selected human serum samples to be negative.

To rule out the possibility that autoantibodies are not reactive to recombinant survivin but only to endogenous survivin, blots were also performed with HEK-293 cell lysate as the source of endogenous survivin. Proteins in the HEK-293 cell lysate were separated by SDS-PAGE, and bands of interest (bands 1–5) were excised from the gel for MS.

MS/MS database search identified survivin (predicted: 16 kDa) in 16-kDa protein bands 2 and 3 (Fig. 3). The presence of survivin was verified by a targeted MS analysis on protein band 2, which identified four peptides of BIRC5: KKEFEETAK, HSSGCAFLSVK, AIEQLAAMD, and RAIEQLAAMD. Western blot analysis showed that rabbit anti-survivin recognized a band of 15- to 16-kDa protein in the HEK-293 lysate (Fig. 3) that was in agreement with band 3 of the mass spectrometry (MS) analysis. Thus, Western blot and MS both confirmed the presence of survivin in HEK-293 cells.

Antibody response to antigens in HEK-293 cell lysate was analyzed in serum of 7 randomly selected nonsmoking controls and 9 longitudinally collected sample sets, each set consisting of 1 case (OD > median OD + 2 IQR of nonsmoking controls) analyzed before and after diagnosis of NSCLC (89% adenocarcinoma; stage I) and 1 smoking-habit control matched to this case. Immunodetected proteins at 16 kDa were observed for all samples, including smoking-habit controls and nonsmoking controls. Specific individual patterns of antibodies against HEK-293 proteins were identical for cases before and after diagnosis of lung cancer (Fig. 3). Proteins that showed intense bands in the 16-kDa region after Coomassie Brilliant Blue staining were identified by MS as ribosomal proteins (Supplementary Table S2); it is highly likely that some nonspecific binding at these bands occurs. We conclude that no anti-survivin reactivity was detected in any of the patient sera.

No survivin-specific autoantibodies were detected by sandwich ELISA in sera from NSCLC cases, smoking-habit controls, or healthy nonsmoking controls, even though background nonspecific binding was 24% higher (P < 0.001) in NSCLC cases and smoking controls than in healthy nonsmoking controls. Thus, if background is not subtracted, significantly higher OD values for cases and smokers can be misleading when compared only with nonsmoking healthy controls. A study of autoantibodies to a panel of 10 TAAs reported that smokers—lung cancer patients and non–lung cancer patients alike—consistently had significantly higher background binding when assaying for autoantibodies than did healthy nonsmoking controls (18).

Specific antibody reactivity for survivin was no higher in any of the 50 NELSON NSCLC patients than in smoking-habit controls. Positive rabbit control samples gave a strong response in the assay. Thus, the assay was functioning correctly, and survivin in vivo had not induced significant antibody responses before or after diagnosis of NSCLC. This observation is supported by the absence of cytolytic responses against survivin peptides (26). Western blot analysis revealed no antibody reactivity with the recombinant survivin, consistent with these ELISA results.

To our knowledge, the population we used is the best well-controlled population related to smoking habit in lung cancer case–control studies on survivin autoantibodies to date. Our inability to find survivin autoantibodies in NSCLC patients is not in agreement with the results reported by others (11–18), who found survivin autoantibodies with an 8% to 52% prevalence. Although these inconsistent findings may have been due to differences in the type of tumor, the stage of lung cancer, or the source of antigen, the most likely explanation is the difference in assay methodology.

Forty-one (82%) of the 50 NSCLC patients in our study were diagnosed with adenocarcinoma, and 45 (90%) were classified as early stage (I and II). This uneven distribution of pathology might explain the absence of survivin autoantibodies. Antibodies to survivin have nevertheless been found irrespective of tumor type and clinical stage of NSCLC (15, 16). To exclude the possibility that our selection of patients had been unbalanced, we also investigated late-stage (stage IV) NSCLC serum samples. In these, too, sandwich ELISA and Western blot analysis revealed no positive samples.

To date, the method used most widely to detect survivin autoantibodies involved direct adsorption assays with immobilized recombinant survivin. If the antigen solution is not absolutely pure, contaminating antigens may be coimmobilized in much the same way as Escherichia coli proteins, which are very often copurified with the recombinant protein (27). Due to the high prevalence of E. coli infections in humans, background responses to the E. coli bacteria used in producing recombinant proteins have been found to be a major problem (28, 29). We also detected these antibody responses to E. coli proteins in a Western blot of E. coli extract with sera of lung cancer patients (data not shown). Although the specificity of the survivin recognition in the DAC-ELISAs was confirmed by decreased reactivity after preabsorption of sera with soluble survivin, reactivity is also likely to be reduced by nonspecific reactions with impurities in the survivin solution.

In solid-phase immunoassays, nonspecific binding for autoantibodies has been found to correlate with increased concentrations of IgG and other inflammatory mediators (30). Despite the fact that the DAC-ELISA showed higher apparent anti-survivin levels than the sandwich ELISA, in our hands this assay showed no significant difference between lung cancer patients previously reported (16) to be positive for anti-survivin and healthy nonsmoking control subjects—in addition, any differences disappear when background is subtracted. In general, DAC-ELISA may incorrectly assess antibodies to survivin and, presumably, to other tumor antigens, because it does not compensate for nonspecific binding. Much more than DAC-ELISA, sandwich ELISA makes it possible to significantly reduce—and also monitor—nonspecific binding. Due to better specificity, absorbance in the sandwich ELISA was an order of magnitude lower than that in the DAC-ELISA.

To test if any differences in posttranslational modifications of endogenous survivin, compared with recombinant survivin, could be the cause of a lack of anti-survivin reactivity to recombinant survivin, we also used HEK-293 cell lysate as the source of endogenous survivin antigen. Although Western blot analysis with HEK-293 cell lysate showed no autoantibody reactivity to endogenous survivin (16 kDa), it showed that the individual patterns of antibodies against HEK proteins before and after the diagnosis of lung cancer were almost identical in each NSCLC cancer patient.

In conclusion, we demonstrated that survivin autoantibody reactivity is not present in sera from NSCLC cases, smoking-habit matched controls, and healthy nonsmoking controls. Higher apparent survivin autoantibody reactivity in the serum samples of smokers than in nonsmokers that has been previously reported is likely the result of nonspecific binding in smokers. In general, autoantibodies to lung tumor antigens should be investigated using sandwich ELISA or another well-characterized technology in a well-controlled population, stratified for at least smoking habit.

No potential conflicts of interest were disclosed.

Conception and design: I. Broodman, L.J.M. Dekker, R.J. van Klaveren, T.M. Luider

Development of methodology: I. Broodman, M.M. VanDuijn, L.J.M. Dekker, H.J. de Koning, R.J. van Klaveren, T.M. Luider

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): I. Broodman, C. Stingl, A.E. Germenis, R.J. van Klaveren, J.G. Aerts

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I. Broodman, M.M. VanDuijn, C. Stingl, L.J.M. Dekker, A.E. Germenis, H.J. de Koning, R.J. van Klaveren, J.G. Aerts, J. Lindemans, T.M. Luider

Writing, review, and/or revision of the manuscript: I. Broodman, M.M. VanDuijn, C. Stingl, L.J.M. Dekker, H.J. de Koning, R.J. van Klaveren, J.G. Aerts, J. Lindemans, T.M. Luider

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): I. Broodman, C. Stingl, R.J. van Klaveren

Study supervision: I. Broodman, J. Lindemans, T.M. Luider

Other (provided data from the NELSON study): H.J. de Koning

The authors thank Frank Santegoets and Roel Faber (Dept. of Public Health, Erasmus MC, Rotterdam, the Netherlands) for their contributions and maintenance of the database of the NELSON trial. They thank Pauline de Goeje (Department of Pulmonology, Erasmus MC) for providing the samples from the NVALT-12 study. They also thank David Alexander (lecturer in English Biomedical Writing, Erasmus MC) for his careful reading of the manuscript.

This work was supported by a Zenith grant (93511034 to I. Broodman, M.M VanDuijn, and L.J.M. Dekker) from the Netherlands Organization for Scientific Research (NWO), which had no role in the study design, collection, analysis, or interpretation of the data, and no role in the preparation, review, or approval of the manuscript.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.