ORIGINAL ARTICLE Genome-Wide DNA Methylation Profiling in Early Stage I Lung Adenocarcinoma Reveals Predictive Aberrant Methylation in the Promoter Region of the Long Noncoding RNA PLUT: An Exploratory Study Soo-Zin Kim-Wanner, MD,a,b Yassen Assenov, PhD,a Mridul B. Nair, PhD,a Dieter Weichenhan, PhD,a Axel Benner, PhD,c Natalia Becker, PhD,c Katharina Landwehr, MSc,b Ruprecht Kuner, PhD,d,e Holger Sültmann, PhD,d,f Manel Esteller, MD, PhD,g,h,i,j Ina Koch, PhD,k Michael Lindner, MD,k Michael Meister, PhD,f,l Michael Thomas, MD,f,l Matthias Bieg, PhD,m Ursula Klingmüller, PhD,f,n Matthias Schlesner, PhD,f,m,o Arne Warth, MD, PhD,p Benedikt Brors, PhD,q Erhard Seifried, MD, PhD,b Halvard Bönig, MD, PhD,b Christoph Plass, PhD,a,f Angela Risch, PhD,a,f,r,s,* Thomas Muley, PhDf,l a Division of Epigenomics, German Cancer Research Center (Deutsches Krebsforschungszentrum [DKFZ]), Heidelberg, Germany b German Red Cross Blood Donor Service, Institute of Transfusion Medicine and Immune Hematology of the Goethe University Medical School, Frankfurt, Germany c Division of Biostatistics, German Cancer Research Center (DKFZ), Heidelberg, Germany d Division of Cancer Genome Research, German Cancer Research Center (DKFZ), German Cancer Consortium (Deutsches Konsortium für Translationale Krebsforschung) and National Center for Tumor Diseases, Heidelberg, Germany e Translational Oncology at the University Medical Center of Johannes Gutenberg University, Mainz, Germany f Translational Lung Research Centre Heidelberg, Member of the German Centre for Lung Research (Deutsches Zentrum für Lungenforschung), Heidelberg, Germany g Josep Carreras Leukaemia Research Institute, Badalona, Barcelona, Catalonia, Spain h Institució Catalana de Recerca i Estudis Avançats, Barcelona, Catalonia, Spain i Centro de Investigacion Biomedica en Red Cancer, Madrid, Spain j Physiological Sciences Department, School of Medicine and Health Sciences, University of Barcelona, Barcelona, Catalonia, Spain k Center of Thoracic Surgery, Asklepios Fachkliniken München-Gauting, Ludwig Maximilians University, Gauting; Comprehensive Pneumology Centre Munich, German Centre for Lung Research (Deutsches Zentrum für Lungenforschung), Munich, Germany l Translational Research Unit, Thoraxklinik at University Hospital Heidelberg, Heidelberg, Germany m Division of Theoretical Bioinformatics, and Heidelberg Center for Personalized Oncology, German Cancer Research Center (DKFZ), Heidelberg, Germany n Division Systems Biology of Signal Transduction, German Cancer Research Center (DKFZ), Heidelberg, Germany o Bioinformatics and Omics Data Analytics, German Cancer Research Center (DKFZ), Heidelberg, Germany p Institute of Pathology, University Hospital Heidelberg, Heidelberg, Germany q Division of Applied Bioinformatics, German Cancer Research Center (DKFZ), Heidelberg, Germany nonfinancial support from AstraZeneca, Bristol-Myers Squibb, Roche *Corresponding author. Holdings AG, and Takeda Pharmaceutical; and personal fees and Disclosure: Dr. Bonig reports receiving grants from Bayer, Chugai, nonfinancial support from Boehringer Ingelheim, Celgene, Chugai EryDel, Miltenyi Biotec, Polyphor, Sandoz-Hexal (a Novartis Com- Pharmaceutical, Eli Lilly, Merck Sharp & Dohme, Novartis, and Pfizer pany), Stage (a Celgene Company), Terumo BCT, and uniQure; personal outside the submitted work. The remaining authors declare no fees from Celgene, Chugai Pharmaceutical, Fresenius Medical Care, conflict of interest. kiadis Pharma, Miltenyi Biotec, Novartis, Sandoz-Hexal, Terumo BCT, Address for correspondence: Angela Risch, PhD, Department of Bio- Genzyme, Novartis, Stage, and Terumo BCT; and Royalties and sciences, University of Salzburg, Billrothstr. 11, Salzburg 5020, Austria. License Fees from Medac and Stock in Siemens Healthineers outside E-mail: angela.risch@sbg.ac.at of the submitted work. Dr. Muley reports receiving nonfinancial support from the National Center for Tumor Disease, Heidelberg, ª 2020 International Association for the Study of Lung Cancer. during the conduct of the study; and grants and nonfinancial support Published by Elsevier Inc. This is an open access article under the from Roche Diagnostics GmbH outside of the submitted work. Dr. CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/ Meister reports receiving grants from Heidelberg Center for 4.0/). Personalized Oncology (DKFZ-HIPO) (project H004) during the ISSN: 1556-0864 conduct of the study. Dr. Plass reports receiving grants from the German Cancer Research Center (DKFZ), during the conduct of the https://doi.org/10.1016/j.jtho.2020.03.023 study. Dr. Thomas reports receiving grants, personal fees, and Journal of Thoracic Oncology Vol. 15 No. 8: 1338–50 August 2020 PLUT Promoter Methylation Predicts Survival 1339 r University of Salzburg, Department of Biosciences, Allergy-Cancer-BioNano Research Centre, Salzburg, Austria s Cancer Cluster Salzburg, Universität Salzburg, Salzburg, Austria Received 19 March 2019; revised 27 March 2020; accepted 29 March 2020 Available online - 7 April 2020 ABSTRACT million per annum and a poor prognosis with a 5-year overall survival (OS) of 18%.1,2 Despite tremendous Introduction: Surgical procedure is the treatment of choice in early stage I lung adenocarcinoma. However, a consider- developments in targeted therapies for advanced stages, able number of patients experience recurrence within the the treatment paradigms and prognosis have not first 2 years after complete resection. Suitable prognostic changed in early stage lung adenocarcinoma for the past biomarkers that identify patients at high risk of recurrence decades; recurrence rate remains high at 30% to 50% (who may probably benefit from adjuvant treatment) are still and the prognosis is still guarded.3-6 Adjuvant chemo- not available. This study aimed at identifying methylation therapy is currently recommended for patients with markers for early recurrence that may become important resected stage II and III.7-12 However, a benefit for tools for the development of new treatment modalities. adjuvant chemotherapy after radical resection in stage Ia lung adenocarcinoma could not be reported and remains Methods: Genome-wide DNA methylation profiling was per- controversial for stage Ib.13,14 The heterogeneity of the formed on 30 stage I lung adenocarcinomas, comparing 14 disease, which evolves through a multitude of disrupted patients with early metastatic recurrence with 16 patients with cell control mechanisms that ultimately converge into a long-term relapse-free survival period using methylated- CpG-immunoprecipitation followed by high-throughput next- carcinogenesis, may explain why there are currently no generation sequencing. The differentially methylated regions known driver genes predisposing for early relapse.15,16 between the two subgroups were validated for their prog- Furthermore, outside of nodal involvement and tumor nostic value in two independent cohorts using the Mas- size, a separate (but no less urgent) problem is the lack sCLEAVE assay, a high-resolution quantitative methylation of established biomarkers13,17 to identify subpopulations analysis. of patients that are at high risk of recurrence, and therefore, might benefit from adjuvant treatment Results: Unsupervised clustering of patients in the discov- modalities. ery cohort on the basis of differentially methylated regions In the search for new biomarkers, aberrant DNA identified patients with shorter relapse-free survival (haz- methylation is gaining importance. Previous studies ard ratio: 2.23; 95% confidence interval: 0.66–7.53; p ¼ describe aberrant DNA hypermethylation in promoter 0.03). In two validation cohorts, promoter hyper- regions as frequent molecular events; they were identi- methylation of the long noncoding RNA PLUT was signifi- fied using the candidate gene approach or using the cantly associated with shorter relapse-free survival (hazard ratio: 0.54; 95% confidence interval: 0.31–0.93; p < 0.026) Infinium HumanMethylation BeadChip16,18-24 with pre- and could be reported as an independent prognostic factor determined CpGs. Studies mapping unbiased genome- in the multivariate Cox regression analysis. wide DNA hypomethylation and hypermethylation in lung adenocarcinoma, which would be reflective of the Conclusions: Promoter hypermethylation of the long non- complex machinery of epigenetic regulation, are largely coding RNA PLUT is predictive in patients with early stage I lacking.16,23-25 Especially, differential epigenetic signa- adenocarcinoma at high risk for early recurrence. Further tures and aberrantly methylated regions in intergenic studies are needed to validate its role in carcinogenesis and regions remain to be identified; we predict that this its use as a biomarker to facilitate patient selection and risk could discriminate between subtypes or identify stratification. deregulated genomic regions, and, in turn, predict  2020 International Association for the Study of Lung response to therapy and prognosis. Therefore, this study Cancer. Published by Elsevier Inc. This is an open access intended to identify differential methylation patterns by article under the CC BY-NC-ND license (http:// analyzing all CpGs of the whole genome using creativecommons.org/licenses/by-nc-nd/4.0/). methylated-CpG-immunoprecipitation (MCIp) after next- generation sequencing. By this approach, a distinct Keywords: PLUT; Methylation profiling; Lung adenocarci- methylation signature associated with early relapse in noma; Prognostic marker; IncRNA stage I lung adenocarcinoma can be detected after potentially curative treatment. We obtained a strong predictive DNA methylation marker that was validated Introduction in two independent cohorts that may provide informa- Lung cancer is the leading cause of cancer death tion for further clinical studies on adjuvant treatment worldwide, comprising a mortality of more than 1.5 modalities and therapeutic targets. 1340 Kim-Wanner et al Journal of Thoracic Oncology Vol. 15 No. 8 Discovery cohort Validation cohort 1 Validation cohort 2 30 patients with lung adenocarcinoma 142 patients with lung 71 patients with lung Stage IA/B , surgery, R0 adenocarcinoma stage I adenocarcinoma stage I 16 patients (No Relapse) 14 patients (Relapse) Recurrence-free and Metastatic recurrence 5-24 alive>60months months after surgery after surgery Methylation analysis (MCIp-Seq) Identification of DMRs distinguishing both cohorts Quantitative Methylation analysis of 27 DMRs (MassARRAY) Identification of DMRs associated with prognosis Validation of 10 best DMRs (MassARRAY) Quantitative Methylation analysis of single CpGs associated with Identification of single CpGs asscociated with prognosis prognosis Identification of aberrant methylated region (DMR) and Validation of DMRs impacting genes significantly impacting prognosis prognosis Figure 1. Overview of the experimental design. In the discovery cohort, genome-wide methylation profiling was performed on 30 stage I lung adenocarcinomas comparing 14 patients with early metastatic recurrence to 16 patients with long-term relapse-free survival using MCIp and high-throughput next-generation sequencing. DMRs were validated using the MassCLEAVE Assay (MassARRAY). The 10 DMRs with the highest methylation differences between the two subgroups were validated in two independent validation cohorts. DMR, differentially methylated region; MCIp, methylated-CpG- immunoprecipitation. Materials and Methods minutes after resection and stored at 80 C. In all An overview detailing the experimental design is samples, the viable tumor content was at least 50% on provided in Figure 1. the basis of the pathologic examination of hematoxylin and eosin–stained slides. To limit variance among the Patient Samples samples owing to heterogeneity between individuals, For the discovery cohort and validation cohort 1, lung adjacent normal lung tissue samples (>2 cm distant to cancer and adjacent normal lung tissues were provided tumor site) were obtained from each patient of the dis- by Lung Biobank Heidelberg, a member of the tissue covery cohort for pairwise comparison. bank of the National Center for Tumor Diseases (Hei- delberg, Germany), and the biobank platform of the Discovery Cohort. Defined selection criteria were German Centre for Lung Research, and the validation applied to stratify two different risk cohorts for further cohort 2 tissues were provided by the Asklepios Biobank comparative methylation analysis. A total of 16 patients for Lung Diseases, a member of the platform biobank of with stage Ia or Ib lung adenocarcinoma were identified the Comprehensive Pneumology Center, German Center for the “no relapse” cohort, fulfilling the criteria of for Lung Research (Deutsches Zentrum für Lungenfor- relapse-free survival (RFS) of more than 60 months. schung [DZL]) (Munich, Germany), after approval by the Patients who had received adjuvant treatment were respective local institutional review boards. Written and excluded from this cohort. Patients were eligible to enter oral informed consent was obtained for all samples from the study as part of the “relapse” cohort if recurrence patients with early stage I to IIb lung adenocarcinoma. had occurred between 5 and 24 months after an oper- Patients were diagnosed and treated at Thoraxklinik ation. The rationale for choosing this arbitrary timeline Heidelberg and Asklepios Clinic Munich-Gauting where was to exclude (R0) surgical failures (>5 mo) and pa- clinical follow-up was pursued and documented. Tissue tients with risks of secondary malignancies (<24 mo). A samples were snap-frozen in liquid nitrogen within 30 total of 14 patients with distant metastasis with or August 2020 PLUT Promoter Methylation Predicts Survival 1341 Table 1. Patient Characteristics Validation Cohort Validation Cohort 2 Discovery Cohort 1 (HB) (DZL) No Relapsea Relapseb Characteristics n % n % n % n % Total 16 53 14 47 142 71 Gender M 8 27 12 40 81 57 31 43.7 F 8 27 2 6.7 61 43 40 56.3 Age Median [yrs] 64 62 66 65 Range [yrs] 38-74 43-75 38-85 43-85 Stage TNM 6th (7th) 6th (7th) 6th (7th) 7th Ia 5 (5) 17 2 (2) 6.7 30 (30) 21 34 47.9 Ib 11 (7) 37 12 (5) 40 112 (88) 79 37 52.1 IIa 0 (3) 0 (6) 0 (16) 0 IIb 0 (1) 0 (1) 0 (8) 0 Tumar diameter Median (range) [cm] 3.8 (1.4-11) 4.05 (1.9-16) 3.3 (1-17) 2.75 (0.6-5) Histological subtype Acinary 9 41 30 Lepidic 3 18 3 Micro papillary 0 2 0 Papillary 3 12 18 Solid 7 23 16 Fetal 0 1 0 Mixed Types 7 40 4 Adenocarcinoma w/o further information 2 5 0 Smoking history 20 Smoker 6 20 4 13 49 35 43 60.6 Former smoker 7 23 8 6.7 22 15 18 25.4 Non-smocker 2 6.7 2 6.7 17 12 10 14.1 NA 1 3.3 0 0 1 0.7 0 0 Adjuvant chemotherapy 0 0 4 13 23 16 3 4.23 Relapse free duration [months] 0 9.1 (5.2-16) 41.3 (0.4-107) 34.9 (0-88.9) Death 0 0 13 93 38 27 9 12.7 Median observation time (range) 72 (60-107) 35 (7.3-37.8) 49 (0.4-108.5) 36.2 (0-88.9) [months] a Patients without relapse after R0-resection of the tumor and alive for more than 60 months. b Patients with relapse after 5 to 24 months after R0-resection of the tumor. DZL, Deutsches Zentrum für Lungenforschung; F, female; HB, Heidelberg Biobank; M, male. without local recurrence were enrolled for the relapse of patients with lung adenocarcinoma stage Ia or Ib cohort. Diagnosis and tumor resection were carried out (TNM seventh edition) from the Asklepios Biobank from 2002 to 2010. The clinical characteristics of the Munich-Gauting from 2009 to 2015. Ethical approvals of patients are summarized in Table 1. the respective local institutional review boards were in place. Patient characteristics are summarized in Table 1. Validation Cohort 1 and Validation Cohort 2. For validation cohort 1 (Heidelberg Biobank), 142 samples DNA and RNA Isolation of patients with lung adenocarcinoma stage Ia or Ib DNA and RNA of the tumor and adjacent lung tissues (TNM sixth edition) were obtained from the Lung bio- were extracted using the QIAamp DNA mini kit and bank, Heidelberg. Diagnosis and tumor resections were RNeasy mini kit (Qiagen, Hilden, Germany), respectively. carried out from 2002 to 2011. Reassessment of stages Quality control activities were performed using gel according to TNM seventh edition revealed 128 stage Ia electrophoresis (DNA) and Agilent Bioanalyzer 2100 or Ib and 24 stage IIa or IIb. All patients were nodal (RNA) (Agilent, CA). For reverse transcriptase polymer- negative. Validation cohort 2 (DZL) included 71 samples ase chain reaction (PCR) analysis, RNA was reverse 1342 Kim-Wanner et al Journal of Thoracic Oncology Vol. 15 No. 8 transcribed into complementary DNA using the Super- reads were converted to Sequence Alignment Map or Bi- Script II (Invitrogen, Carlsbad, CA). Reverse transcriptase nary Alignment Map formats using SAMtools. Read counts PCR was performed using complementary DNA equiva- of each sample were normalized for total read length and lent to 10 ng RNA by adding a mixture of 5.5 mL Probes the number of sequencing reads (reads per kilobase per 480 Master Enzyme Mastermix (Roche, Mannheim, Ger- million mapped reads; RPKMs). Peak calling was per- many), 0.11 mL of 20 mM primer pairs, 0.11 mL Universal formed using the software HOMER (v4.4). A differentially Probe Library probe (UPL, Roche: BTG2: UPL1; FYN: methylated region (DMR) of one peak was defined if at UPL29; PDX1: UPL78; RPTOR: UPL23), and 0.28 mL H2O least a twofold difference in tag densities between the on a LightCycler 480 PCR system (Roche, Mannheim, tumor and adjacent normal tissue of one patient sample Germany). Crossing point values were adjusted to the set was detected. DMRs enriched in the relapse subcohort housekeeping gene GAPDH and relative expression was compared with the no relapse subcohort were calculated calculated as given by Livak and Schmittgen.26 Primer using Fisher’s exact test and p values were corrected by information is given in Supplementary Figure 1A. multiple testing (Benjamini-Hochberg). The second list of DMRs was obtained by comparison of the mean RPKMs of each cohort (no relapse versus relapse) using the Wil- MCIp and High-Throughput Next-Generation coxon ranked sum test (p < 0.05) after multiple testing Sequencing correction (Benjamini-Hochberg). For candidate marker Global methylation analysis was performed using selection and further quantitative methylation analysis, MCIp followed by high-throughput next-generation we focused on the nonrepetitive overlapping DMRs ful- sequencing, as described earlier.27,28 First, 60 mg of filling both criteria. methyl-CpG-binding domain Fc-protein was coupled to 40 mL of protein A-coated magnetic beads (Diagenode, Liege, Belgium) and bound to 3 mg of sonicated genomic DNA (Bioruptor NextGen, Diagenode) with a median Quantitative Methylation Analysis fragment length of 145 base pairs (bp) as confirmed by Quantitative methylation values for candidate gene Agilent’s BioAnalyzer high-sensitivity DNA platform. regions were technically validated using matrix-assisted Methylated DNA was then eluted by the variable strin- laser desorption and ionization time-of-flight mass gency of the magnetically labeled protein using the SX- spectrometry (MALDI-TOF MS), as described before.30 8G IP-Star robot (Diagenode) by increasing the salt The 10 best-validated amplicons were further assessed concentration (300–1000 mM). Highly methylated DNA for clinical validation in the independent sample cohort (1000 mM eluate) was desalted with MinElute columns (Heidelberg Biobank). Genomic DNA was bisulfite- (Qiagen) and enrichment was quantified by real-time converted using the EZ DNA methylation kit (Zymo PCR using the imprinted gene SNRPN. MCIp-enriched Research, Orange, CA) in accordance with the manu- methylated DNA fragments were submitted to the facturer’s protocol. Defined genomic regions covering DKFZ Genomics and Proteomics Core Facility for library the candidate gene regions of bisulfite-converted DNA preparation and next-generation sequencing. Afterward, were PCR-amplified, followed by in vitro transcription fragmented DNA was end-repaired and ligated to and base-specific cleavage. Methylation standards (0%, Illumina-paired end adaptors using NEBnext DNA 20%, 40%, 60%, 80%, 100%) were used to control for Library Prep Master Mix Set (New England Biolabs) in the dynamic range of measurements. DNA methylation accordance with the manufacturer’s instructions. status was quantitatively assessed by MALDI-TOF MS Adapter ligated libraries were directly amplified by 14 using the Sequenom (Sequenom, San Diego, CA) (primer cycles of PCR with the standard Illumina index primers information is provided in Supplementary Table 1A). and distributions were validated using the Agilent Bio- Besides the methylation value of single CpG units, the analyzer before it was quantified by a Qubit fluorometer mean methylation value of overall CpG units per (Invitrogen). The libraries were sequenced on the Illu- amplicon was used for further analysis. The samples of mina HiSeq 2000 sequencer (50 bp, single read 50 bp) validation cohort 1 were divided into two groups (high using standard Illumina protocols. and low methylation) for each DMR using the median methylation value of the whole cohort as the cutoff. The Data Analysis and Candidate Gene Selection mean methylation values of the amplicons were used for Sequencing reads were aligned to the hg19 genome survival analysis (Mantel-Cox test). The methylation assembly of the human reference genome using the status of the predictive amplicons was tested in the in- Burrows-Wheeler Alignment tool.29 Aligned reads were dependent validation cohort (DZL) and correlated further processed by merging lane-level data and with clinical outcomes to validate the prognostic removing duplicates. The remaining uniquely mapped significance. August 2020 PLUT Promoter Methylation Predicts Survival 1343 Reporter Gene Assay The reporter gene assay was performed as previously described.31 The promoter region of the long noncoding RNA (lncRNA) PLUT was amplified from genomic DNA using phusion polymerase (Thermo Scientific) and inserted in the luciferase vector pGL4.10 (Promega). Luciferase activity was measured 48 hours after trans- fection into HEK-293 cells and normalized to the activity of a cotransfected cytomegalovirus promoter–driven b- galactosidase vector. Three independent experiments were performed in duplicates. Error bars show the SD. Transfections were performed with metafectene ac- cording to the manufacturer’s instructions. Statistical Analysis The primary end point of the statistical analysis was RFS, defined as the time from surgical procedure to tu- mor recurrence or death from any cause. The secondary end points were OS, defined as the time from surgical procedure to death from any cause, and disease-free survival (DFS). Events for DFS are the same as for RFS but, in addition, include any signs or symptoms of the Figure 2. Methylation analysis and identification of DMRs. (A) disease, and time to event was also measured (whatever Unsupervised hierarchical clustering of the RPKMs (0 to >250) of all DMRs derived by pairwise comparison between occurs first). Patients without an observed event were peak regions of tumor and adjacent normal tissue samples censored at the last available follow-up time point. separates two clusters. Each cluster harbored tumor samples Distributions of survival times were estimated by and their adjacent normal samples, although the tumor and using the method of Kaplan and Meier. The distribution normal samples separated in both clusters. (B) Kaplan-Meier estimates of both clusters revealed significant (p ¼ 0.0304) of follow-up times was estimated by the reverse Kaplan- differences in relapse-free survival between both clusters. Meier estimate. (C) Schematic overview of two peak regions derived from the To investigate the influence of the DMRs on the sur- MCIp-seq analysis with differences in the “relapse” and “no vival end points (RFS, DFS, OS), multivariable Cox relapse” cohorts. Amplicons (MA-amplicon) for quantitative methylation analysis using the MassCLEAVE assay (MassAR- regression models were applied. A backward variable RAY) and the results for each CpG unit are reported in a selection was used to recursively reduce the model size heatmap and a scattergram for the relapse cohort (pink) and and keep only DMRs with p values less than 0.5 using the the no relapse cohort (blue). CI, confidence interval; DMR, pooled residual chi-square as the stopping rule. differentially methylated region; HR, hazard ratio; MCIp-seq, methylated-CpG-immunoprecipitation sequencing; RPKMs, For illustration purposes, Kaplan-Meier curves for reads per kilobase per million mapped reads. cohorts of patients with low- and high-methylation levels were computed for each DMR. Groups of low- and high- All statistical tests were two-sided. Values of p less than methylation levels were defined using the median 0.05 were considered statistically significant. All calcula- methylation level (0.3) of validation cohort 1 as the cutoff. tions were done using the statistical software environment Prediction error curves were calculated using the R, version 3.1.3 (R packages: pec_ 2.4.7, prodlim_1.5.1, time-dependent Brier score, the quadratic difference rms_4.3-1, survival_2.38-1, and glmnet_2.0-2). between the 0 to 1 response status, and the predicted survival probability at time point t.32 The 0.632þ boot- strap method was applied to reduce the bias in esti- Results mating the predictive accuracy. The integrated Brier Distinct Genome-Wide Methylation Patterns score (IBS) over time was used as a statistic for com- Between the Early and Late Relapse Groups in parison of models, including the model for the average the Discovery Cohort methylation of the whole amplicon, the model using all Genomic DNA of 30 paired samples of bronchial single CpGs, and the model of CpGs after backward se- adenocarcinoma and adjacent normal lung tissue were lection. The empty model, that is, the Kaplan-Meier es- analyzed for genome-wide methylation using MCIp. timate was used as a reference. Analysis by real-time PCR revealed that methylation 1344 Kim-Wanner et al Journal of Thoracic Oncology Vol. 15 No. 8 Figure 2. (continued). enrichment failed for six samples. The recovered DNA of of 99 DMRs were significantly different between the the remaining 54 samples was sequenced using the relapse cohort and the no relapse cohort (p < 0.05; Illumina HiSeq 2000 platform. Six libraries required Fisher’s exact test; Benjamini-Hochberg). For further resequencing to reach a saturation greater than 95%. candidate marker selection, we focused (after multiple One tumor sample was excluded from the analysis owing testing corrections) on the nonrepetitive DMRs that were to suspected sample mix-up. The total number of also significantly different in RPKMs between the two sequence reads, mapped reads, and unique reads for subcohorts (p < 0.05; Wilcoxon rank sum test; each sample are represented in Supplementary Table 2. Benjamini-Hochberg). A total of 27 DMRs fulfilled the For the final analysis, sequence reads of 24 paired selection criteria. The distribution of the DMRs among sample sets and additional three tumor samples and two the genomic features is shown in Supplementary normal lung samples were available. In total 572.354 Figure 2. peaks could be identified in 200 bp genomic windows after normalization. On comparing the peaks in each Quantitative Methylation Analysis Reveals DMRs paired sample for regions found to be differentially Between the Two Subcohorts methylated in at least one tumor when compared with For technical validation of the results obtained with its lung counterpart, 238.331 DMRs between normal and MCIp sequencing, the same genomic DNA of the dis- tumor tissue were observed. Unsupervised hierarchical covery cohort was assessed for quantitative methylation clustering of the RPKMs of all 238.331 DMRs separated in the 27 candidate marker regions by MALDI-TOF MS. the samples into two clusters. Clustering was performed For all selected regions, we confirmed differences be- using the package “pheatmap” in R Bioconductor tween normal and tumor tissue and at least a trend (distance ¼ average, k-means ¼ 1000). The normal tis- between the relapse and no relapse cohorts. Significant sue samples clustered separately to their respective differences between the subgroups were detected for the paired tumor tissue sample in each cluster (Fig. 2A), as DMRs annotated to the closest genes BTG2 and OSTM1- expected. Kaplan-Meier estimates of the two clusters AS (t test, p < 0.05). In some cases, an individual CpG with distinct methylation patterns revealed significant unit within an analyzed amplicon revealed significantly differences in RFS (hazard ratio [HR]: 2.23; 95% confi- stronger methylation differences between the two co- dence interval [CI]: 0.66–7.53; p ¼ 0.03) (Fig. 2B). A total horts, compared with the average methylation of all August 2020 PLUT Promoter Methylation Predicts Survival 1345 annotated to HERC5, Loc648987, PAX3, SEC14L5, AF524019, BTG2, KIF1A, RPTOR, FYN, and PLUT obtained from the quantitative methylation analysis (MALDI-TOF MS) (Fig. 4) were further investigated in 142 samples of validation cohort 1. The influence of the 10 candidate DMRs on clinical outcome was evaluated by multivariate Cox regression analysis. After backward selection, the average methylation of the amplicons PLUT (p ¼ 0.01) and FYN (p ¼ 0.03) revealed a significant prognostic impact on RFS. Figure 5A shows the corresponding Kaplan-Meier curves for PLUT using a median split (HR: 0.54; 95% CI: 0.31–0.93; p ¼ 0.026). After backward selection methylation of the single CpGs of PLUT_CpG17 (p < 0.0002) and FYN_CpG13.14 (p < 0.01), FYN_CpG6 (p ¼ 0.01) and FYN_CpG9 (p ¼ 0.01) could be identified as significantly predictive of RFS (data not shown). The predictive accuracy was estimated using the 0.632þ bootstrap method. Prediction error curves are shown in Supplementary Figure 3. The IBS over 59 Figure 3. Analysis of the DMR upstream of the lncRNA PLUT. months of follow-up revealed improvement of the model (A) Schematic overview of the DMR 22 bp upstream of the using selected CpGs (IBS ¼ 0.180) compared with the lncRNA. The DMR (MassARRAY amplicon) was quantitatively Kaplan-Meier reference (IBS ¼ 0.185) and using the analyzed for methylation using the MassCLEAVE Assay and is annotated to a CpG island region enriched for H3K4Me1 mark model with the average methylation of the amplicon (ChiP-seq ENCODE) for the transcription factor EZH2 (TF (IBS ¼ 0.198). Regarding DFS and OS, average methyl- ChiP-seq ENCODE). A total of three CpGs (cg01193690, ation of the DMR KIF1A (DFS: p ¼ 0.02; OS: p ¼ 0.01), cg10034364, and cg15992563) within the amplicon are pre- and again, PLUT (DFS: p ¼ 0.004; OS: p ¼ 0.005) sent on the Illumina Human Methylation 450kBeadChip. (B) The region (gene reporter amplicon) was cloned in the revealed a statistically significant prognostic impact pGL4.10 luciferase vector and transfected in HEK-293 T cells. (data not shown). The transfection of the empty pGL4.10 vector served as a control. Luciferase promoter assay was performed in three independent experiments. bp, base pair; ChiP-seq, chro- matin immunoprecipitation sequencing; DMR, differentially Aberrant Methylation of the Promoter Region of methylated region; ENCODE, encyclopedia of DNA elements; lncRNA PLUT Predicts Clinical Outcome in an lncRNA, long noncoding RNA. Independent Validation Cohort The prognostic model including the DMRs PLUT and CpGs of the whole amplicon (Fig. 2C). None of the top 10 FYN was further tested in a second independent vali- DMRs with the highest differences were located in pro- dation cohort with stage Ia or Ib (patient characteristics moter regions of coding regions. Two DMRs were located are shown in Table 1). Validation cohort 2 included 71 in promoter regions of lncRNAs: PLUT and LOC648987 patients with a median follow-up of 36.2 months (95% (uncharacterized RNA gene). The aberrant methylated CI: 31.1–47.1). The methylation levels of the same 10 region 22 bp upstream of the lncRNA PLUT (NR_ DMRs that were tested in validation cohort 1, including 047484.2) is located in CpG islands within silico bind- PLUT and FYN, were analyzed by MALDI-TOF MS using ing of H3K4me3 and revealed promoter activity in the the same amplicons and primers as for the discovery reporter gene assay (Fig. 3). Peak regions of the 10 best cohort and validation cohort 1. After backward selec- DMRs derived from MCIp sequencing and amplicon in- tion, methylation of PLUT could be validated with formation with annotations to genomic features are respect to RFS (HR; 0.32; 95% CI: 0.12–0.86; p ¼ 0.024) summarized in Supplementary Tables 1A and 1B. (Fig. 5B). The DMR FYN was not significantly associated with RFS in validation cohort 2. The prediction error estimation revealed an IBS over 87.7 months of follow- DMRs Correlate With Clinical Outcome in up within the same range for the reference model Patients With Stage I Lung Adenocarcinoma (IBS ¼ 0.151), the average methylation model (IBS ¼ To discover markers prognostic of risk of relapse in 0.166), and the model including the selected CpGs early stage lung adenocarcinoma, 10 candidate DMRs (IBS ¼ 0.160). 1346 Kim-Wanner et al Journal of Thoracic Oncology Vol. 15 No. 8 1.0 0.8 ** * Methylation 0.6 0.4 0.2 0.0 BTG2 AF524019 PLUT Herc5 FYN PAX3 Sec14L5 RPTOR Loc648987 KIF1A R R R R _R R _R _R _R R R R No Relapse Relapse Figure 4. Quantitative methylation analysis of DMRs in the discovery cohort (n ¼ 29) using MassARRAY. Quantitative methylation was analyzed in 29 samples of the discovery cohort using the MassCLEAVE Assay. The average methylation of the amplicons annotated to the genes BTG2, HERC5, FYN, PAX3, and Sec14L5 significantly differed between the “relapse” and “no relapse” cohorts. DMR, differentially methylated region. Aberrant Methylation of BTG2, FYN, RPTOR, and the transcriptional regulation of their annotated genes PLUT Correlates With Expression in the BTG2, FYN, RPTOR, and PDX1, mRNA expression was Discovery Cohort determined in the two subgroups (no relapse versus To investigate whether the four detected DMRs with relapse) of the discovery cohort and six additional the highest ability to predict prognosis were involved in samples. Expression levels for BTG2, FYN, and RPTOR A validation cohort 1 100 PLUT low Relapse Free Survival [%] PLUT high HR, 0.5389 (95% CI, 0.3126 to 0.9290) p=0.0304 50 0 0 50 100 150 months validation cohort 2 B 100 PLUT low Relapse Free Survival [%] PLUT high HR, 0.3219 (95%CI, 0.12 to 0.8637) p=0.0244 50 0 0 20 40 60 80 100 months Figure 5. Kaplan-Meier survival estimates on the basis of promoter methylation of lncRNA PLUT. Relapse-free survival curves with respect to methylation of the promoter region of PLUT are reported for (A) validation cohort 1 and (B) validation cohort 2. Subgroups of low- and high-methylation levels were generated using the median methylation level in validation cohort 1 as a cutoff. CI, confidence interval; HR, hazard ratio; lncRNA, long noncoding RNA. August 2020 PLUT Promoter Methylation Predicts Survival 1347 Table 2. Multivariate Regression Analysis Reveals were limited to analyses of promoter hypermethylation Significant Correlation With RFS for PLUT, Sex, and Stage of CpG island methylation.18,19,33-35 Other DNA methyl- ation profiles were approached with a lower resolution HR 95% CI P value in small cohorts and not selective for stage I cancers.22,34 PLUT methylation 5.53 1.33–22.92 0.018 Suggested surrogate markers like p16, APC, CDH13, and Sex 2.19 1.22–3.95 0.008 RASSF1A promoter hypermethylation were found to be Age (y) 1 0.97–1.03 0.99 P-stage (TNM seventh edition) 2.25 1.23–4.12 0.008 associated with early recurrence36 but not validated in CI, confidence interval; HR, hazard ratio; lncRNA, long noncoding RNA; RFS, independent cohorts. Also, a signature on the basis of relapse-free survival. promoter hypermethylation of the five genes HIST1H4F, PCDHGB6, NPBWR1, ALX1, and HOXA9 was recently reported to determine two subgroups associated significantly differed between the two groups. PLUT with survival33; besides lung adenocarcinoma, the expression analysis could not be performed owing to study also included SCLC with a different molecular lack of sufficient material and technical reasons. Never- pathomechanism. theless, expression of PDX1 as a target gene of PLUT Whereas it is well accepted that promoter hyper- revealed a trend without reaching significance, but methylation and silencing of tumor suppressor genes are owing to assay performance, quantitative PCR could not involved in carcinogenesis, DNA methylation is involved be performed for all samples (22 of 36) (Supplementary in different epigenetic processes genome-wide.37 Altered Fig. 1B). transcription and gene regulation may also be influenced by DNA methylation of regions other than promoter re- Multivariate Regression Analysis Identifies gions of coding genes.19 Furthermore, the spread of DNA Aberrant Promoter Methylation of PLUT as an methylation and its interaction with chromatin remod- Independent Prognostic Marker eling also play important roles.38,39 The methylation Because clinical parameters are known to influence patterns unraveled in this study are in line with these the outcome, we performed a multivariate analysis in the observations, in that, just 10% of DMRs differentiating 142 patients of validation cohort 1 to test whether the the two subgroups are annotated in transcription start DMR PLUT may serve as an independent prognostic site regions (Supplementary Fig. 2), as opposed to more marker for early recurrence. Using Cox regression, be- than 40% of DMRs mapping to intergenic regions. None sides methylation status (binary) of PLUT, age, stage, and of the previously suggested candidate genes in promoter sex were included, and hypermethylation of PLUT could regions were among the best 100 DMRs, underscoring be detected as an independent unfavorable prognostic the importance of sequencing-based genome-wide factor (HR: 5.53; 95% CI: 1.3–22.9) (Table 2). Further- methylation analyses, whereas the Illumina Human- more, a significant difference in outcome regarding sex Methylation 450k BeadChip just covers 1.7% of all CpGs was apparent in both validation cohorts. As illustrated in the whole genome and is focused on promoter and by the Kaplan-Meier curves, women revealed signifi- CpG island regions. cantly better survival than men (Supplementary Fig. 4A). In this study, promoter CpG island hypermethylation Men with promoter hypermethylation of PLUT had the of the lncRNA PLUT was found to correlate significantly worst outcome with an estimated DFS of 33% after 60 with early recurrence. This region has a high density of months compared with 76% for women with hypo- transcription factor binding sites and is enriched for methylation of the promoter region of PLUT (p ¼ 0.001) H3K4me1 histone marks in the encyclopedia of DNA (Supplementary Fig. 4B), the other subcohorts revealing elements data (ENCODE) set. Owing to the shortness of intermediate outcomes. tissue material, the expression of the lncRNA PLUT could not be investigated and correlated to the methylation Discussion status of its promoter region in this study. However, Clinical management in lung cancer will be greatly aberrant methylation of this genomic region is present in aided by the discovery of reliable prognostic biomarkers all disease stages of patients with lung adenocarcinoma to guide treatment decisions. Stratification of patients by when looking at publicly available The Cancer Genome the need, or lack of it, for (neo)adjuvant treatment is Atlas Program (TCGA) Infinium HumanMethylation450 highly warranted. We report that the methylation status BeadChip data, implying a role in carcinogenesis of the promoter of lncRNA PLUT distinguishes between (Supplementary Fig. 5). However, its possible role and patients at risk and not at risk for early relapse. downstream molecular mechanism in lung adenocarci- This study represents the most comprehensive anal- noma remains to be determined. ysis of genome-wide DNA methylation in primary lung Methylation analysis at individual CpG sites within adenocarcinoma. Previous DNA methylation studies the DMRs revealed variations in methylation and 1348 Kim-Wanner et al Journal of Thoracic Oncology Vol. 15 No. 8 detected single CpGs discriminating the two subgroups downstream molecular mechanisms are involved is not with a higher statistical significance than the whole known. The idea of using PDX1 or other PLUT-induced amplicon. The functional regulation of single CpG sites is genes as therapeutic molecular targets is certainly not yet well-known and needs further investigation, compelling. though the importance of single CpG sites was previously However, the tremendous advances made in deci- commented on.18,40 However, functional studies and phering the molecular mechanisms involved in lung causal explanations of aberrant single CpG methylation tumorigenesis have not yet translated into better prog- in lung cancer are rare. nosis for patients, especially those with early stage tu- LncRNAs are believed to play critical regulatory roles mors for whom cure is most feasible. Although new as tumor suppressors or oncogenic drivers.41,42 In lung sensitive screening methods like low-dose computed cancer, MALAT1, HOTAIR, HNF1A-AS1, and linc00673 tomography may increase the numbers of patients are believed to increase proliferation and invasiveness diagnosed in early stage I disease,57 the need for prog- by regulating protoadhesion factors (e-cadherin), tran- nostic biomarkers to identify newly diagnosed patients scription factors, and reorganization of chromatin and by with a high risk of recurrence will gain further impor- acting as oncogenic lncRNAs.43-47 Conversely, the tance. The results of ongoing studies with EGFR tyrosine lncRNAs MEG3 and PANDAR are believed to inhibit tu- kinase inhibitors (ALCHEMIST; NCT02201992, mor growth by regulating TP53 or BCL2 or by inhibition NCT02193282) or immunotherapy (PEARLS/KEYNOTE- of epithelial-mesenchymal transition.45,48,49 09, NCT02504372) are eagerly awaited; however, these lncRNA PLUT(O) (PDX1-associated lncRNA, up- will only be applicable for a subgroup of patients. This regulator of transcription) was recently found to regu- study reveals that promoter hypermethylation of the late the transcription of PDX1 by promoting contacts lncRNA PLUT significantly correlates with early recur- between the PDX1 promoter region and enhancer clus- rence, especially in male patients. Whether the methyl- ters in human pancreatic b-cells.50 In this study, PDX1 ation status of this genomic region will serve as another expression in the discovery cohort revealed differences; puzzle piece for implementation in personalized medi- however, sample numbers were too small for correlation cine as a biomarker for adapted treatment modalities in analysis with methylation data of the lncRNA PLUT. early stage lung cancer to overcome bad prognosis has to PDX1 is a homeobox transcription factor and a key be proved in further studies. regulator in the embryonic development of the pancreas and pancreatic b-cell function and also in the differen- tiation of the gastric mucosa and pseudopyloric Acknowledgments glands.51,52 For pancreatic cancer, there is evidence that This study was funded by the Heidelberg Center for PDX1 acts as an oncogene and regulates cell proliferation Personalized Oncology (DKFZ-HIPO) (project H004) and and invasion45,46; its overexpression is associated with partly funded by the German Center for Lung Research poor survival.39,53 In addition to pancreatic cancer, (DZL) 82DZL00402. The authors thank the Genomics overexpressed PDX1 is associated with several solid tu- and Proteomics Core Facility of the German Cancer mors, including breast, colon, prostate, and kidney car- Research Center for providing and performing the high- cinoma.54 Its role in tumorigenesis remains unclear and throughput next-generation sequencing; Oliver Mücke, may differ among tissues. In gastric cancer, the expres- Marion Bähr, Stefanie Herkt, Elisabeth Ehrend, Anne- sion of PDX1 is aberrantly reduced by promoter hyper- karin Meyer for their technical assistance; and Jörn methylation and histone deacetylation, and PDX1 is, Lausen for providing the plasmids. therefore, considered a tumor suppressor.55,56 We postulate that the transcription of PLUT among others is regulated by methylation in lung adenocarcinoma. Supplementary Data Hypermethylation of its promoter region resulted in Note: To access the supplementary material accompa- reduced expression, and subsequent translation of the nying this article, visit the online version of the Journal of gene PDX1 was unhindered, resulting in a higher Thoracic Oncology at www.jto.org and at https://doi. expression in patients with early relapse in our discov- org/10.1016/j.jtho.2020.03.023. ery cohort. To our knowledge, this is the first study revealing an association of the methylation status of the promoter region of PLUT with lung adenocarcinoma. References 1. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2017. CA Whether aberrant methylation of this gene region me- Cancer J Clin. 2017;67:7–30. diates tumorigenesis in lung adenocarcinoma mainly by 2. Jemal A, Ward EM, Johnson CJ, et al. Annual report to transcriptional regulation of PLUT and subsequent the nation on the status of cancer, 1975-2014, featuring translational modulation of PDX1 or whether other survival. J Natl Cancer Inst. 2017;109:djx030. August 2020 PLUT Promoter Methylation Predicts Survival 1349 3. Office for National Statistics. Cancer survival by stage at 17. Butts CA, Ding K, Seymour L, et al. Randomized phase III diagnosis for England: adults diagnosed 2012, 2013 and trial of vinorelbine plus cisplatin compared with obser- 2014 and followed up to 2015; 2016. https://www.ons.gov. vation in completely resected stage IB and II non-small- uk/peoplepopulationandcommunity/healthandsocialcare/ cell lung cancer: updated survival analysis of JBR-10. conditionsanddiseases/bulletins/cancersurvivalbystageatdia J Clin Oncol. 2010;28:29–34. gnosisforenglandexperimentalstatistics/adultsdiagnosed2 18. Buckingham L, Penfield FL, Kim A, et al. PTEN, RASSF1 0122013and2014andfollowedupto2015. Accessed January and DAPK site-specific hypermethylation and outcome in 19, 2019. surgically treated stage I and II nonsmall cell lung cancer 4. Nesbitt JC, Putnam JB Jr, Walsh GL, Roth JA, patients. Int J Cancer. 2010;126:1630–1639. Mountain CF. Survival in early-stage non-small cell lung 19. Lokk K, Vooder T, Kolde R, et al. Methylation markers of cancer. Ann Thorac Surg. 1995;60:466–472. early-stage non-small cell lung cancer. PLoS One. 5. Berger AH, Brooks AN, Wu X, et al. High-throughput 2012;7:e39813. phenotyping of lung cancer somatic mutations. Cancer 20. Guo Y, Zhang R, Shen S, et al. DNA methylation of Cell. 2016;30:214–228. LRRC3B: a biomarker for survival of early-stage non- 6. Imielinski M, Berger AH, Hammerman PS, et al. Mapping small cell lung cancer patients. Cancer Epidemiol Bio- the hallmarks of lung adenocarcinoma with massively markers Prev. 2018;27:1527–1535. parallel sequencing. Cell. 2012;150:1107–1120. 21. Han L, Xu G, Xu C, Liu B, Liu D. Potential prognostic 7. Keller SM, Adak S, Wagner H, et al. A randomized trial of biomarkers identified by DNA methylation profiling postoperative adjuvant therapy in patients with analysis for patients with lung adenocarcinoma. Oncol completely resected stage II or IIIA non-small-cell lung Lett. 2018;15:3552–3557. cancer. Eastern Cooperative Oncology Group. N Engl J 22. Selamat SA, Chung BS, Girard L, et al. Genome-scale Med. 2000;343:1217–1222. analysis of DNA methylation in lung adenocarcinoma and 8. Artal CA, Calera UL, Hernando CJ. Adjuvant chemo- integration with mRNA expression. Genome Res. therapy in non-small cell lung cancer: state-of-the-art. 2012;22:1197–1211. Transl Lung Cancer Res. 2015;4:191–197. 23. Lenka G, Tsai MH, Lin HC, et al. Identification of 9. Douillard JY, Rosell R, De Lena M, et al. Adjuvant methylation-driven, differentially expressed STXBP6 as a vinorelbine plus cisplatin versus observation in patients novel biomarker in lung adenocarcinoma. Sci Rep. with completely resected stage IB-IIIA non-small-cell 2017;7:42573. lung cancer (Adjuvant Navelbine International Trialist 24. Wei Y, Liang J, Zhang R, et al. Epigenetic modifications in Association [ANITA]): a randomised controlled trial. KDM lysine demethylases associate with survival of early- Lancet Oncol. 2006;7:719–727. stage NSCLC. Clin Epigenetics. 2018;10:41. 10. Felip E, Rosell R, Maestre JA, et al. Preoperative 25. Carvalho RH, Haberle V, Hou J, et al. Genome-wide DNA chemotherapy plus surgery versus surgery plus adjuvant methylation profiling of non-small cell lung carcinomas. chemotherapy versus surgery alone in early-stage non- Epigenetics Chromatin. 2012;5:9. small-cell lung cancer. J Clin Oncol. 2010;28:3138–3145. 26. Livak KJ, Schmittgen TD. Analysis of relative gene 11. Gilligan D, Nicolson M, Smith I, et al. Preoperative expression data using real-time quantitative PCR and the chemotherapy in patients with resectable non-small cell 2(-Delta C(T)) method. Methods. 2001;25:402–408. lung cancer: results of the MRC LU22/NVALT 2/EORTC 27. Sonnet M, Claus R, Becker N, et al. Early aberrant DNA 08012 multicentre randomised trial and update of sys- methylation events in a mouse model of acute myeloid tematic review. Lancet. 2007;369:1929–1937. leukemia. Genome Med. 2014;6:34. 12. Pisters KM, Vallieres E, Crowley JJ, et al. Surgery with or 28. Heilmann K, Toth R, Bossmann C, Klimo K, Plass C, without preoperative paclitaxel and carboplatin in early- Gerhauser C. Genome-wide screen for differentially stage non-small-cell lung cancer: Southwest Oncology methylated long noncoding RNAs identifies Esrp2 and Group Trial S9900, an intergroup, randomized, phase III lncRNA Esrp2-as regulated by enhancer DNA methylation trial. J Clin Oncol. 2010;28:1843–1849. with prognostic relevance for human breast cancer. 13. Strauss GM, Herndon JE 2nd, Maddaus MA, et al. Adju- Oncogene. 2017;36:6446–6461. vant paclitaxel plus carboplatin compared with obser- 29. Li H, Durbin R. Fast and accurate long-read alignment vation in stage IB non-small-cell lung cancer: CALGB with Burrows-Wheeler transform. Bioinformatics. 9633 with the Cancer And Leukemia Group B, Radiation 2010;26:589–595. Therapy Oncology Group, and North Central Cancer 30. van den Boom D, Ehrich M. Mass spectrometric analysis Treatment Group Study Groups. J Clin Oncol. of cytosine methylation by base-specific cleavage and 2008;26:5043–5051. primer extension methods. Methods Mol Biol. 2009; 14. Pignon JP, Tribodet H, Scagliotti GV, et al. Lung adjuvant 507:207–227. cisplatin evaluation: a pooled analysis by the LACE 31. Kuvardina ON, Herkt S, Meyer A, et al. Hematopoietic Collaborative Group. J Clin Oncol. 2008;26:3552–3559. transcription factors and differential cofactor binding 15. Lawrence MS, Stojanov P, Polak P, et al. Mutational regulate PRKACB isoform expression. Oncotarget. heterogeneity in cancer and the search for new cancer- 2017;8:71685–71698. associated genes. Nature. 2013;499:214–218. 32. Graf E, Schmoor C, Sauerbrei W, Schumacher M. 16. Cancer Genome Atlas Research Network. Comprehensive Assessment and comparison of prognostic classification molecular profiling of lung adenocarcinoma. Nature. schemes for survival data. Statistics in Medicine. 2014;511:543–550. 1999;18:2529–2545. 1350 Kim-Wanner et al Journal of Thoracic Oncology Vol. 15 No. 8 33. Sandoval J, Mendez-Gonzalez J, Nadal E, et al. 46. Ma C, Wu G, Zhu Q, et al. Long intergenic noncoding RNA A prognostic DNA methylation signature for stage I non- 00673 promotes non-small-cell lung cancer metastasis by small-cell lung cancer. J Clin Oncol. 2013;31:4140–4147. binding with EZH2 and causing epigenetic silencing of 34. Sato T, Arai E, Kohno T, et al. DNA methylation profiles at HOXA5. Oncotarget. 2017;8:32696–32705. precancerous stages associated with recurrence of lung 47. Wu Y, Liu H, Shi X, Yao Y, Yang W, Song Y. The long non- adenocarcinoma. PLoS One. 2013;8:e59444. coding RNA HNF1A-AS1 regulates proliferation and metas- 35. Pfeifer GP, Rauch TA. DNA methylation patterns in lung tasis in lung adenocarcinoma. Oncotarget. 2015;6:9160– carcinomas. Semin Cancer Biol. 2009;19:181–187. 9172. 36. Brock MV, Hooker CM, Ota-Machida E, et al. DNA 48. Lu KH, Li W, Liu XH, et al. Long non-coding RNA methylation markers and early recurrence in stage I lung MEG3 inhibits NSCLC cells proliferation and induces cancer. N Engl J Med. 2008;358:1118–1128. apoptosis by affecting p53 expression. BMC Cancer. 37. Feinberg AP, Tycko B. The history of cancer epigenetics. 2013;13:461. Nat Rev Cancer. 2004;4:143–153. 49. Sun M, Liu XH, Wang KM, et al. Downregulation of BRAF 38. Berman BP, Weisenberger DJ, Aman JF, et al. Regions of activated non-coding RNA is associated with poor prog- focal DNA hypermethylation and long-range hypo- nosis for non-small cell lung cancer and promotes methylation in colorectal cancer coincide with nuclear metastasis by affecting epithelial-mesenchymal transi- lamina-associated domains. Nat Genet. 2011;44:40–46. tion. Mol Cancer. 2014;13:68. 39. Quint K, Stintzing S, Alinger B, et al. The expression 50. Akerman I, Tu Z, Beucher A, et al. Human pancreatic b pattern of PDX-1, SHH, patched and Gli-1 is associated cell lncRNAs control cell-specific regulatory networks. with pathological and clinical features in human Cell Metab. 2017;25:400–411. pancreatic cancer. Pancreatology. 2009;9:116–126. 51. Larsson LI, Madsen OD, Serup P, Jonsson J, Edlund H. 40. Claus R, Lucas DM, Stilgenbauer S, et al. Quantitative Pancreatic-duodenal homeobox 1 - role in gastric endo- DNA methylation analysis identifies a single CpG dinu- crine patterning. Mech Dev. 1996;60:175–184. cleotide important for ZAP-70 expression and predictive 52. Offield MF, Jetton TL, Labosky PA, et al. PDX-1 is of prognosis in chronic lymphocytic leukemia. J Clin required for pancreatic outgrowth and differentiation Oncol. 2012;30:2483–2491. of the rostral duodenum. Development. 1996;122:983– 41. Tsai MC, Spitale RC, Chang HY. Long intergenic noncoding 995. RNAs: new links in cancer progression. Cancer Res. 53. Koizumi M, Doi R, Toyoda E, et al. Increased PDX-1 2011;71:3–7. expression is associated with outcome in patients with 42. Gibb EA, Brown CJ, Lam WL. The functional role of long pancreatic cancer. Surgery. 2003;134:260–266. non-coding RNA in human carcinomas. Mol Cancer. 54. Wang XP, Li ZJ, Magnusson J, Brunicardi FC. Tissue 2011;10:38. MicroArray analyses of pancreatic duodenal homeobox-1 43. Gupta RA, Shah N, Wang KC, et al. Long non-coding RNA in human cancers. World J Surg. 2005;29:334–338. HOTAIR reprograms chromatin state to promote cancer 55. Ma J, Chen M, Wang J, et al. Pancreatic duodenal metastasis. Nature. 2010;464:1071–1076. homeobox-1 (PDX1) functions as a tumor suppressor in 44. Qiu M, Xu Y, Yang X, et al. CCAT2 is a lung gastric cancer. Carcinogenesis. 2008;29:1327–1333. adenocarcinoma-specific long non-coding RNA and pro- 56. Ma J, Wang JD, Zhang WJ, et al. Promoter hyper- motes invasion of non-small cell lung cancer. Tumour methylation and histone hypoacetylation contribute to Biol. 2014;35:5375–5380. pancreatic-duodenal homeobox 1 silencing in gastric 45. Han L, Zhang EB, Yin DD, et al. Low expression of long cancer. Carcinogenesis. 2010;31:1552–1560. noncoding RNA PANDAR predicts a poor prognosis of non- 57. Jemal A, Fedewa SA. Lung cancer screening with low- small cell lung cancer and affects cell apoptosis by dose computed tomography in the United States-2010 regulating Bcl-2. Cell Death Dis. 2015;6:e1665. to 2015. JAMA Oncol. 2017;3:1278–1281.