Molecular Subtypes of Pulmonary Large-cell Neuroendocrine Carcinoma Predict Chemotherapy Treatment Outcome

Large-cell neuroendocrine carcinoma (LCNEC) is a high-grade neuroendocrine carcinoma with non–small cell cytologic features that accounts for 1% to 3% of all lung cancers (1, 2). Similar to small-cell lung cancer (SCLC), LCNEC is a disease with a poor prognosis (2, 3). The diagnosis of LCNEC requires assessing both morphology and neuroendocrine differentiation by IHC (4, 5). Previously, we and others have shown that separation of LCNEC from SCLC and pulmonary carcinoids can be difficult even on resection specimens (6–10). In the current WHO classification, some of the features used to classify a tumor as LCNEC overlap with those applied for SCLC, NSCLC, and carcinoids (8).

To improve the separation of LCNEC from carcinoids on a biopsy specimen, the proliferation marker Ki-67 with a cutoff >20% was proposed (10); however, the differential diagnosis between LCNEC and SCLC remains an issue for pathologists, due to crush artefacts, distorted cytologic features of SCLC on large tissue samples (11), tumor heterogeneity (11), and overlap in cell and nuclear size between LCNEC and SCLC (9). This is further worsened by the fact that, at diagnosis, both SCLC and LCNEC are often metastasized, and commonly only one biopsy specimen is available for diagnosis (12). The identification of diagnostic markers to allow separation of LCNEC from SCLC is therefore an unmet need.

Chemotherapy treatment for LCNEC is a subject of debate since it seems to be less chemosensitive than SCLC. In the American Society of Clinical Oncology (ASCO) guideline, either platinum–etoposide chemotherapy (SCLC-PE) treatment or the same regimen as for non–small cell nonsquamous cell carcinoma is advised for LCNEC (13), although SCLC-PE is considered as the most appropriate (13). Nevertheless, recent studies indicate that patients with LCNEC have a more favorable outcome when treated with platinum–gemcitabine or taxane chemotherapy (NSCLC-GEM/TAX) compared with SCLC-PE (14–16). The molecular characteristics that may explain these differences in the response to different chemotherapies remain unknown.

Several next-generation sequencing (NGS) studies have shown that LCNEC tumors can be further subdivided into two mutually exclusive groups based on their mutational patterns (17, 18): one harboring inactivation of TP53 and STK11 and/or KEAP1 genes, and the other one enriched for inactivation of TP53 and RB1, a hallmark of SCLC. It has been hypothesized that these LCNEC subtypes may require different chemotherapy treatment (17). In this study, we tested whether the described molecular LCNEC subtypes may have an impact on the chemotherapy response.

The study protocol was approved by the medical ethical committee of the Maastricht University Medical Centre (METC azM/UM 14-4-043) and performed according to the Dutch “Federa, Human Tissue and Medical Research: Code of conduct for responsible use (2011)” regulations not requiring patient informed consent.

In this retrospective population-based study, all data were retrieved from the Netherlands Cancer Registry and Netherlands Pathology Registry (PALGA, the nationwide registry of pathology in the Netherlands; ref. 19) as previously described (12). Data managers from the cancer registry retrospectively updated (2015) clinical data of all first-line chemotherapy-treated stage IV LCNEC patients (n = 232, Fig. 1). Available data included clinical characteristics, TNM stage, overall survival (OS), and progression-free survival (PFS) from date of diagnosis until first evidence of progression, death or last day of follow-up, and chemotherapy details. All patients received platinum doublet (cisplatin or carboplatin) chemotherapy treatment, further divided into three groups: “NSCLC-GEM/TAX” including gemcitabine, docetaxel, or paclitaxel; “NSCLC-PEM” including pemetrexed; and “SCLC-PE” including etoposide. NSCLC-PEM chemotherapy was separated from the other NSCLC regimens because of previously reported resistance in (large cell) neuroendocrine carcinomas (14, 20–23).

From all histologic specimens, the original hematoxylin and eosin (HE) and IHC slides were collected. Subsequently, three pathologists (R.J. van Suylen, E. Thunnissen, M. den Bakker), who were blinded for clinical outcome, systematically scored all cases at a multi-head microscope for WHO 2015 criteria. Proliferative activity was evaluated by estimation of MIB1 and mitotic counting (mitoses/2 mm²; ref. 8). The MIB1 (Ki-67) staining was scored (<25%, >25%) when available (10, 24). Either >10 mitosis/2 mm², abundant tumor necrosis, or a Ki-67 staining of >25% of tumor cells was sufficient to score for high-grade tumor (19, 24). Diagnoses were considered as consensus when at least two pathologists agreed, further referred to as panel consensus. All panel consensus LCNEC tumors were included for NGS and IHC staining analysis when formalin-fixed paraffin embedded (FFPE) tissue block(s) were available (n = 109; Fig. 1).

Tumor macrodissection was performed aiming at a tumor cell content of at least 20%. DNA was extracted from four to eight 10-μm slides using the Maxwell FFPE LEV Automated DNA Extraction Kit (Promega Corporation). DNA concentration was measured using the QuantiFluor dsDNA Dye System (Promega Corporation).

One hundred and sixty-nine amplicons of 150 base pairs (bp) in size were designed using the Qiagen GeneRead DNAseq Custom V2 Builder tool reference CNGHS-02445X-169 (GRCh37) covering the following exons: TP53, RB1, STK11, and KEAP1. This custom GeneRead amplicon-based custom panel covered 100% of the coding region (i.e., exonic) of TP53, 95% of RB1, 81% of STK11, and 95% of KEAP1. A validated in-house protocol (IARC) was used to perform multiplex PCR with four separate primer pools. Per pool, 5 μL of DNA diluted to a maximum of 4 ng/μL (0.60–4.0) were dispensed and air-dried. Subsequently, 5 μL of the PCR mix were added (containing 2.5 μL primer, 1 μL PCR mix, 0.34 μL HotStar Taq, and 1.16 μL H₂0) and the DNA was amplified in a 384-well plate as following: 15 minutes at 95°C, and 25 cycles of 15 seconds at 95°C and 4 minutes at 60°C, and 10 minutes at 72°C. After amplification, the PCR products were pooled into a single reaction per sample.

The amplified PCR products were purified using NucleoMag NGS Clean-up and Size Select beads (Macherey-Nagel). Purified PCR products were quantified by Qubit DNA high-sensitivity assay kit (Invitrogen Corporation). A minimum of 100 ng of purified PCR product was included for library preparation with the NEBNext Fast DNA Library Prep Set (New England BioLabs, USA) following an in-house validated protocol (IARC). End-repair was performed and ligated to specific adapters and in-house prepared individual barcodes (Eurofins MWG Operon). Bead purification was applied to clean libraries and amplification was performed. Equimolarly pooled libraries were loaded on a 2% agarose gel for electrophoresis (220 V, 40 minutes). Using the GeneClean Turbo kit (MP Biomedicals) DNA fragments of 110 to 220 bp were recovered from the pooled libraries. Library quality and quantity were assessed on the Agilent 2100 Bioanalyzer on-chip electrophoresis (Agilent Technologies). Sequencing was performed on the Ion Torrent Proton Sequencer (Life Technologies Corp.), aiming for a minimum coverage of 250×, using the Ion PI Hi-Q OT2 200 Kit and the Ion PI Hi-Q sequencing 200 Kit with the Ion PI chip V3 (Life Technologies Corp.), following manufacturer's instructions.

Technical duplicates were included for all samples and processed in identical 96- and 384-well plates to prevent PCR errors. Sequencing data were aligned to the hg19 (GRCh37) reference genome and BAM files were generated using the Torrent Suite Software (v4.4.2). For all amplicon positions, the read depth was calculated using SAMtools (25) and samples with a median coverage lower than 250× were excluded. Needlestack (https://github.com/IARCbioinfo/needlestack; ref. 26) was used to call variants with default parameters except for the base-quality and the mapping-quality thresholds (10 and 1, respectively). Annotation was performed with ANNOVAR (27) using the PopFreqAll (popfreq_all_20150413), COSMIC v77, SIFT and Polyphen (dbnsfp30a) databases (28, 29). We only considered those mutations identified by Needlestack in the two technical duplicates. In addition, we excluded the ones with an allelic fraction lower than 5%, a relative-variant strand bias (RVSB) higher than 0.85, or those already reported as germline in any of the ExAC, ESP or 1000G populations with a frequency larger than 0.001 (30–32). In addition, all mutations had to either be (i) reported in the COSMIC database, or (ii) damaging mutations (ii) damaging mutations (stop, indels and splice), or (iii), missense mutations classified as deleterious by SIFT or Polyphen databases (Supplementary Data File S1; NGS data).

The N-terminal and C-terminal regions of the RB1 protein were targeted with antibody 4H1 (1:100, Cell Signaling Technology) and 13A10 (1:100, Leica Biosystems). In addition, the protein P16 was targeted with antibody JC8 (1:400, Santa Cruz Biotechnology). Three-micron-thick FFPE slides were stained using a Dako Autostainer Link 48 system with the EnVision FLEX visualization Kit (DAKO, Agilent) according to the standard protocols. For 13A10 and JC8 high-pH antigen retrieval was used, and for 4H1 low-pH antigen retrieval was used. Tonsillar tissue (control for P16/RB1) and tumor stromal cells (internal control for RB1) were included as positive controls. H-scores were calculated as a total score of the percentage of tumor cells with staining intensity 1 (weak nuclear staining) ×1, intensity 2 (moderate nuclear staining) ×2, and intensity 3 (strong nuclear staining) ×3 with a maximum score of 300. H-scores were evaluated for all RB1 and P16 markers by E-J. M. Speel who was blinded for all clinical, histopathologic, and mutational data.

To detect homozygous deletions of the RB1 gene, FISH was performed with the ZytoLight SPEC RB1/13a12 Dual Color Probe (Zytovision). Three-micron-thick sections were cut from FFPE tumors without a detectable RB1 mutation but with loss of RB1 protein expression. FISH slides were deparaffinized, air dried at room temperature, pretreated with 0.2 mol/L HCL for 20 minutes at room temperature, followed by incubation in 1 mol/L NaSCN for 30 minutes at 80°C. Subsequently, sections were treated with pepsin from porcine stomach mucosa (Sigma-Aldrich; 0.5 mg/mL in 0.14 mol/L NaCl pH 2), postfixed in 1% formaldehyde in PBS for 10 minutes, and dehydrated in 70% to 100% ethanol. The ZytoLight SPEC RB1/13a12 Dual Color Probe was added under a coverslip (undiluted, 6–10 μL). Denaturation of probe and target DNA was carried out simultaneously for 5 min at 85°C before hybridization overnight at 37°C in a humid chamber (Thermobrite). After removing the coverslips, slides were stringently washed to remove unhybridized probe in 2× SSC/0.3% NP40 at pH 7.0 at 73°C for 2 minutes. Slides were dehydrated in an ascending ethanol series (70%–100%) and mounted in 0.2 μg/mL DAPI-Vectashield (Vector Laboratories). Probe signals were scored using a DM 5000 B fluorescence microscope (Leica) with specific filter sets for rhodamine-, fluorescein-, and DAPI.

Each tumor was screened for the presence/absence of RB1 (orange) and 13q12 (green) signals, and the predominant copy number per nucleus was noted (e.g., 1:1, 1:2, 2:2). Normal tonsillar tissue and internal normal tissue parts in the tumor tissue served as internal controls with a predominant 2:2 pattern. A homozygous deletion was considered when no RB1 (orange) signal was observed in the tumor tissue. All cases were evaluated by J. Derks and E.-J. M. Speel.

All analyses were performed using SPSS (version 22 for Windows, Inc.). The χ² and Fisher exact test were used to compare categorical data; the Wilcoxon signed-rank test was used for continuous variables. OS and PFS were analyzed using two-sided log-rank test and survival curves were estimated using the Kaplan–Meier method. To evaluate the predictive role of RB1 mutation and IHC status, a Cox regression model was used including an interaction term for the marker and the chemotherapy treatment. Two-sided P values <0.05 were considered significant.

A panel of three expert pathologists (R.J. van Suylen, E. Thunnissen, and M. den Bakker) reviewed the 232 clinically annotated and initially classified as stage IV LCNEC, available in the Netherlands Pathology Registry. In total, 148 of them were confirmed as panel consensus LCNEC tumors (Fig. 1; Supplementary Data File S1). The fact that tumors reclassified as carcinoids (n = 9) had a longer survival than those confirmed as LCNEC (P = 0.008) supports the value of the pathology revision (Supplementary Fig. S1A and S1B). Of the 148 confirmed LCNECs, 79 tumors passed the quality controls for NGS analyses and were targeted sequenced for the coding regions of TP53, RB1, KEAP1, and STK11 (Supplementary Data File S1).

We obtained a median coverage of 2,850× (range, 261–6,870) per sample. Mutations in TP53 were present in 85% of the cases (n = 67), RB1 in 47% (n = 37), KEAP1 in 18% (n = 14), and STK11 in 10% (n = 8); five samples were wild-type for the four genes analyzed (Fig. 2). RB1 was coaltered with TP53 in 34 of the 37 RB1-mutated tumors; RB1 mutations were mutually exclusive with STK11 P = 0.006 but not with KEAP1 mutations P = 0.71.

The clinical characteristics of the patients, which tumors were sequenced, are shown in Table 1: median age was 64 (range, 51–79), 65% were males, 55% completed first-line chemotherapy (≥4 cycles), and 19% received second-line chemotherapy treatment. When considering only the patients with available data on the subtype of chemotherapy (n = 72, 91%), we observed that those LCNEC tumors that harbored a wild-type RB1 gene showed a significant longer OS when treated with NSCLC-GEM/TAX compared with SCLC-PE [(9.6; 95% confidence interval (CI), 7.7–11.6 versus 5.8 (95% CI, 5.5–6.1) months, P = 0.026] and to NSCLC-PEM chemotherapy [6.7 (95% CI, 5.1–8.2), P = 0.039; Fig. 3A)]. No difference was observed in the case of LCNECs with an RB1 mutation (Fig. 3B). Using a Cox regression model, the presence of a wild-type RB1 was associated with a significant difference (HR, 2.37; 95% CI, 1.09–5.19) favoring NSCLC-GEM/TAX chemotherapy over SCLC-PE treatment. However, comparison of the overall group did not identify a significant interaction between the RB1 mutational status and the chemotherapy treatment (P = 0.35; Fig. 4).

The PFS of RB1 wild-type NSCLC-GEM/TAX–treated patients was also significantly higher compared with treatment with SCLC-PE [6.1 (95% CI, 4.2–8.0) months versus 5.7 (95% CI, 3.9–7.6), P = 0.019] but similar to treatment with NSCLC-PEM [4.7 (95% CI, 3.0–6.4), P = 0.18; Supplementary Fig. S2A]. Similarly to OS, no difference was observed for PFS in RB1-mutated LCNEC patients for the different chemotherapy regimens (Supplementary Fig. S2B). Finally, the mutational status of none the assessed genes (TP53, RB1, STK11, KEAP1) had a prognostic value (Fig. 5A–D).

We assessed the RB1 protein expression levels by IHC in all of the 79 tumors sequenced (Fig. 2) and found that 92% of the RB1-mutated LCNEC tumors had an RB1 H-score of 0 [the median was 0 (range, 0–200) compared with the internal controls; Supplementary Fig. S3A]. The three cases that retained RB1 expression carried two splice mutations and one single nucleotide variation, all reported in COSMIC, respectively (Supplementary Data File S1). No difference was observed regarding the C- and N-terminal RB1 protein staining when stratifying by type of mutation (splice/indel/single nucleotide variants).

In LCNEC tumors with RB1 wild-type, the median H-score was 50 (range, 0–200; Fig. 2; Supplementary Data File S1, Supplementary Fig. S3C). In 18 (43%), the H-score was 0. To test whether a copy-number alteration was the underlying mechanism for this loss of expression, we performed FISH in the 16 cases for which tissue was available (Supplementary Data File S1). However, no homozygous deletion was found in any of the samples suggesting alternate mechanisms for RB1 inactivation, such as genomic rearrangements or large deletions detectable with the techniques used in this study.

Considering the reported interobserver variation in the diagnosis of SCLC and LCNEC, and the fact that virtually all SCLCs carry an inactivated RB1 (33), we evaluated whether there was a correlation between the panel diagnosis and the loss of RB1 protein expression. In total, we performed IHC analyses on 109 panel consensus diagnosed LCNEC. For 98 of them, the panel unanimously diagnosed the tumors as LCNEC but for 11, one of the pathologists classified the tumor as SCLC. We did not observe a significant difference in the prevalence of RB1 expression in these 11 cases (27% with an H-score >50) when comparing with the 98 LCNEC unanimously classified (29% with an H-score >50, P = 0.93).

Previous studies have identified a correlation between RB1 inactivation, loss of expression, and, P16 expression (34, 35). We, therefore, analyzed our series of samples for P16 protein expression and found that, while the median P16 H-score in the RB1 wild-type LCNEC group was 180 (range, 0–300), this value went up to 300 (range, 0–300) in the RB1-mutated cases (Fig. 2; Supplementary Fig. S3B). In total, 91% of the LCNEC tumors with a P16 H-score <50 harbored a wild-type RB1 gene (Fig. 2; Supplementary Fig. S3D).

Patients with LCNEC showing an RB1 H-score ≥50 had a significant longer OS when treated with NSCLC-GEM/TAX than with SCLC-PE [9.6 (95% CI, 7.4–11.8) vs. 1.9 (95% CI, 1.7–2.1) months, P = 0.001] or NSCLC-PEM [4.8 (95% CI, 3.9–5.7) months, P = 0.007; Fig. 3C]. Cox regression analysis confirmed the predictive value of RB1-positive staining on NSCLC-GEM/TAX versus SCLC-PE chemotherapy outcome (HR 4.96; 95% CI, 1.79–13.74, P_interaction = 0.002; Fig. 4). PFS was also significantly longer for NSCLC-GEM/TAX versus SCLC-PE–treated patients [5.5 (95% CI, 1.9–9.0) and 1.7 (95% CI, 0.0–4.8) months (P = 0.023), respectively] but not versus NSCLC-PEM [4.1 (95% CI, 4.0–4.2) months, P = 0.21; Supplementary Fig. S2C]. No statistically significant difference in response to different chemotherapy treatments was observed in patients with LCNEC tumors with an RB1 H-score <50, independently of their mutational status (Fig. 3D; Supplementary Fig. S2D).

P16 IHC H-score <50 was correlated with improved OS for NSCLC-GEM-TAX versus SCLC-PE chemotherapy (P = 0.028; Fig. 3D) but it had no impact on PFS (P = 0.24; Supplementary Fig. S2D). Combined evaluation of RB1 H-score ≥50 and/or P16 <50 showed identical results for OS (P = 0.002; Fig. 3F) and PFS (P = 0.027; Supplementary Fig. S2F). Similarly to the mutational status, none of the RB1 and P16 H-scores had prognostic value (Fig. 5E and F).

Once diagnosed, LCNEC is frequently treated with SCLC-PE chemotherapy, with poor responses (13, 14, 36, 37). Recently, we provided evidence that NSCLC-GEM/TAX chemotherapy may perform better than SCLC-PE chemotherapy on LCNEC tumors (14). Here we have tested whether the molecular characteristics of the LCNEC tumors might explain these differences.

In line with what has been already reported (17, 18, 38), TP53 was found mutated in 85% of our LCNEC cases, RB1 in 47%, KEAP1 in 18%, and STK11 in 10%. RB1 was coaltered with TP53 in 92% of the RB1-mutated tumors, and mutations in RB1 and STK11 occurred in a mutually exclusive way. The frequency of STK11 mutations was lower than expected, probably due to the fact that with our approach we could only cover 60% of the coding region of this gene. Although the mutational status of the genes analyzed did not have a prognostic value, we observed that patients with RB1 wild-type LCNECs showed a significant longer OS when treated with NSCLC-GEM/TAX compared with SCLC-PE and to NSCLC-PEM chemotherapies.

RB1 is inactivated in virtually all SCLCs but only in half of LCNECs (17, 33, 38–40). Similarly, RB1 protein expression is more frequently lost in SCLC (∼90%) than in LCNEC (45%–67%; refs. 17, 18, 35, 38, 41). However, its clinical relevance has not been assessed thoroughly but for a recent prospective study on SCLC reporting that patients with wild-type RB1 but loss of the protein expression showed inferior OS and PFS when treated with SCLC-PE chemotherapy (42). In our study, we found that RB1 expression was completely lost in almost all RB1-mutated LCNECs, but also in 47% of the wild-type cases. Homozygous gene deletions measured by FISH did not explain the loss of expression in the wild-type samples. This suggests that protein expression might be a more reliable measurement of RB1 inactivation than mutation. In total, RB1 protein expression was strongly downregulated or completely lost in 72% of the LCNEC tumors analyzed, similarly to what has been reported in recent studies (17, 38). Similarly to RB1 wild-type LCNEC patients, those with RB1-expressing tumors also showed a significant longer OS when treated with NSCLC-GEM/TAX than with SCLC-PE or NSCLC-PEM. But, in addition, Cox regression analysis confirmed the predictive value of RB1-positive staining on NSCLC-GEM/TAX versus SCLC-PE chemotherapy outcome, and PFS was also significantly longer for NSCLC-GEM/TAX versus SCLC-PE–treated patients.

P16 (CDKN2A) functions as an inhibitor of cyclin D-dependent kinases (CDK4/6) that phosphorylates RB1 enabling cell proliferation (43). CDKN2A (P16) and RB1 inactivation seem to be mutually exclusive in LCNEC (18, 44), and their expression is strongly correlated; previous studies have indicated that combined low RB1 and high P16 proteins expression is observed in 45% to 78% of LCNEC and almost always in SCLC (>90%; refs. 35, 38, 41, 43). In our data, we have validated this patter, and we have additionally found that combined RB1 expression and loss of P16 expression are predictive for improved outcome on NSCLC-GEM/TAX chemotherapy treatment, although P16 staining did not show additive value to the RB1 staining alone.

Overall, we found that RB1 mutational status and RB1/P16 protein expression are predictive markers for chemotherapy response and may aid to guide therapeutic decisions in advanced LCNEC disease. This is the largest (population-based) study evaluating chemotherapy outcome related to mutational patterns in panel-reviewed LCNECs. Although these results are of great interest for the clinical management of LCNEC, the few limitations of our study due to its retrospective design encourage the replication of these results in a prospective randomized clinical trial that stratifies LCNEC based on genomic subtypes and by RB1/P16 protein expression, and investigate outcome to NSCLC-GEM/TAX and SCLC-PE chemotherapy subtypes. In addition, these markers could be tested in biopsy specimens of high-grade neuroendocrine carcinomas to evaluate whether they could help in the differential diagnosis of LCNEC and SCLC.

E.M. Speel reports receiving other commercial research support from Bristol-Myers Squibb. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.L. Derks, E. Thunnissen, H.J.M. Groen, E.F. Smit, L. Fernandez-Cuesta, E.-J.M. Speel, A.-M.C. Dingemans

Development of methodology: J.L. Derks, E. Thunnissen, M. Foll, J.D. McKay, L. Fernandez-Cuesta, E.-J.M. Speel, A.-M.C. Dingemans

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.L. Derks, N. Leblay, E. Thunnissen, M. den Bakker, R. Damhuis, E.C. van den Broek, A. Chabrier, J.D. McKay, E.-J.M. Speel, A.-M.C. Dingemans

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.L. Derks, N. Leblay, E. Thunnissen, R.J. van Suylen, H.J.M. Groen, R. Damhuis, M. Foll, L. Fernandez-Cuesta, E.-J.M. Speel, A.-M.C. Dingemans

Writing, review, and/or revision of the manuscript: J.L. Derks, E. Thunnissen, R.J. van Suylen, M. den Bakker, H.J.M. Groen, E.F. Smit, R. Damhuis, E.C. van den Broek, A. Chabrier, M. Foll, L. Fernandez-Cuesta, E.-J.M. Speel, A.-M.C. Dingemans

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.L. Derks, E. Thunnissen, R.J. van Suylen, R. Damhuis, E.C. van den Broek, A. Chabrier, A.-M.C. Dingemans

Study supervision: L. Fernandez-Cuesta, E.-J.M. Speel, A.-M.C. Dingemans

Other (specimen review): M. den Bakker

We would like to thank G. Roemen and A. Haesevoets of the Department of Pathology of the Maastricht University Medical Centre (Maastricht, the Netherlands) for their help with performing the FISH analyses.

PALGA-group co-authors: L. Arensman, Meander Medisch Centrum Klinische Pathologie; F.E. Bellot, Klinische Pathologie Hoofddorp; J.E. Broers, Isala Klinieken; C.M. van Dish; Klinische pathologie Groene Hart Ziekenhuis; K.E.S.Duthoi, Amphia Ziekenhuis; M.J. Flens, Symbiant; J.M.M. Grefte, Gelre Ziekenhuis Klinische Pathologie; M.C.H. Hogenes, Labpon; R. Natté, Haga Ziekenhuis; A.F. van Hamel, Pathologie SSZOG; P.J.J.M. Klinkhamer, Stichting PAMM; J.W.R. Meijer, Ziekenhijs Rijnstate; J.C.C.van der Meij, Pathologie Friesland; F.H. van Nederveen Laboratorium voor Pathologie (PAL); E.W.P. Nijhuis, Onze Lieve Vrouwe Gasthuis; M.F.M. van Oosterhout, St. Antonius Ziekenhuis; S.H. Sastrowijoto, Pathologie Sittard; K. Schelfout, Stichting Pathologisch en Cytologisch Laboratorium West-Brabant; J. Sietsma, Pathologie Martini Ziekenhuis; F.M.M. Smedts, Reinier Haga MDC; M.M. Smits, Laboratorium voor Klinische Pathologie; J. Stavast, Laboratorium Klinische Pathologie Centraal Brabant; W. Timens, Pathologie UMCG; M.L. van Velthuysen, NKI-AVL; A. Vink, Universitair Medisch Centrum Utrecht; C.C.A.P. Wauters, Canisius Wilhelmina Ziekenhuis; S. Wouda, VieCuri Medical Centre.

This study was supported by grants from the Dutch Cancer Society (UM-2014-7110, to A-M. C. Dingemans). J.L. Derks is the recipient of an ERS/EMBO Joint Research Fellowship (STRTF 2016) 7178. The research leading to these results has received funding from the European Respiratory Society (ERS) and European Molecular Biology Organization (EMBO)”.

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.