Epigenetic–smoking interaction reveals histologically heterogeneous effects of TRIM27 DNA methylation on overall survival among early-stage NSCLC patients Xinyu Ji1, Lijuan Lin1,2, Sipeng Shen1,2,3, Xuesi Dong1,2,4, Chao Chen1, Yi Li5, Ying Zhu1, Hui Huang1, Jiajin Chen1, Xin Chen1, Liangmin Wei1, Jieyu He1, Weiwei Duan1,6, Li Su2, Yue Jiang1, Juanjuan Fan1, Jinxing Guan1, Dongfang You1,2, Andrea Shafer7, Maria Moksnes Bjaanæs8, Anna Karlsson9, Maria Planck9, Johan Staaf9,  Aslaug Helland8,10, 11,12,13,14 1,2,3 Manel Esteller , Yongyue Wei , Ruyang Zhang1,2,3 , Feng Chen1,3,15,16 and 2,7 David C. Christiani 1 Department of Biostatistics, Center for Global Health, School of Public Health, Nanjing Medical University, Nanjing, China 2 Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, MA, USA 3 China International Cooperation Center for Environment and Human Health, Nanjing Medical University, Nanjing, China 4 Department of Epidemiology and Biostatistics, School of Public Health, Southeast University, Nanjing, China 5 Department of Biostatistics, University of Michigan, Ann Arbor, MI, USA 6 Department of Bioinformatics, School of Biomedical Engineering and Informatics, Nanjing Medical University, Nanjing, China 7 Pulmonary and Critical Care Division, Department of Medicine, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA 8 Department of Cancer Genetics, Institute for Cancer Research, Oslo University Hospital, Oslo, Norway 9 Division of Oncology and Pathology, Department of Clinical Sciences Lund and CREATE Health Strategic Center for Translational Cancer Research, Lund University, Lund, Sweden 10 Institute of Clinical Medicine, University of Oslo, Oslo, Norway 11 Josep Carreras Leukaemia Research Institute, Badalona,Barcelona, Spain 12 Centro de Investigacion Biomedica en Red Cancer, Madrid, Spain 13 Institucio Catalana de Recerca i Estudis Avancßats, Barcelona, Spain 14 Physiological Sciences Department, School of Medicine and Health Sciences, University of Barcelona, Barcelona, Spain 15 State Key Laboratory of Reproductive Medicine, Nanjing Medical University, Nanjing, China 16 Jiangsu Key Lab of Cancer Biomarkers, Prevention and Treatment, Cancer Center, Collaborative Innovation Center for Cancer Personalized Medicine, Nanjing Medical University, Nanjing, China Keywords Tripartite motif containing 27 (TRIM27) is highly expressed in lung cancer, DNA methylation; interaction; non-small-cell including non-small-cell lung cancer (NSCLC). Here, we profiled DNA lung cancer; overall survival; prognosis; methylation of lung adenocarcinoma (LUAD) and lung squamous cell car- TRIM27 cinoma (LUSC) tumours from 613 early-stage NSCLC patients and evalu- Correspondence ated associations between CpG methylation of TRIM27 and overall R. Zhang, SPH Building Room 406, 101 survival. Significant CpG probes were confirmed in 617 samples from The Longmian Avenue, Nanjing, Jiangsu 211166, Cancer Genome Atlas. The methylation of the CpG probe China cg05293407TRIM27 was significantly associated with overall survival in Tel: +86 025 86868436 patients with LUSC (HR = 1.65, 95% CI: 1.30–2.09, P = 4.52 9 10 5), E-mail: zhangruyang@njmu.edu.cn but not in patients with LUAD (HR = 1.08, 95% CI: 0.87–1.33, and P = 0.493). As incidence of LUSC is associated with higher smoking inten- Y. Wei, SPH Building Room 418, 101 Longmian Avenue, Nanjing, Jiangsu 211166, sity compared to LUAD, we investigated whether smoking intensity China impacted on the prognostic effect of cg05293407TRIM27 methylation in Tel: +86 025 86868436 NSCLC. LUSC patients had a higher average pack-year of smoking Abbreviations CI, confidence interval; FDR, false discovery rate; HR, hazard ratio; LUAD, lung adenocarcinomas; LUSC, lung squamous cell carcinomas; NSCLC, non-small-cell lung cancer; OS, overall survival; QC, quality control; SD, standard deviation; TCGA, The Cancer Genome Atlas; TRIM27, tripartite motif containing 27; TSS, transcription start site. Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 2759 This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. Epigenetic–smoking interaction analysis of TRIM27 X. Ji et al. E-mail: ywei@njmu.edu.cn (37.49LUAD vs 54.79LUSC, P = 1.03 9 10 19) and included a higher propor- tion of current smokers than LUAD patients (28.24%LUAD vs 34.09- Xinyu Ji and Lijuan Lin contributed equally %LUSC, P = 0.037). cg05293407TRIM27 was significantly associated with to this work overall survival only in NSCLC patients with medium–high pack-year of Feng Chen and David C. Christiani are senior authors who supervised the work smoking (HR = 1.58, 95% CI: 1.26–1.96, P = 5.25 9 10 5). We conclude that cg05293407TRIM27 methylation is a potential predictor of LUSC prog- (Received 5 May 2020, revised 27 July nosis, and smoking intensity may impact on its prognostic value across the 2020, accepted 3 August 2020, available various types of NSCLC. online 3 September 2020) doi:10.1002/1878-0261.12785 Furthermore, DNA methylation changes have been 1. Introduction linked to various environmental exposures (e.g., cigarette Lung cancer is the most commonly diagnosed cancer, smoking) and may explain part of the association accounting for 11.6% of total cases in 2018 [1]. More between smoking and cancer recurrence and mortality than 85% of lung cancer cases are non-small-cell lung [15,27,28]. However, LUAD is more common in non- cancer (NSCLC), of which lung adenocarcinoma smokers and long-term former smokers, while most (LUAD) and lung squamous cell carcinoma (LUSC) NSCLC patients among current smokers have LUSC are the most common subtypes [2,3]. Compared to [29,30], indicating substantially different pathology and late-stage patients, NSCLC patients diagnosed at an oncology. Anyway, few study focused on heterogeneous early stage have a better prognosis [4]. However, wide effect of DNA methylation between LUAD and LUSC. heterogeneity in overall survival has been observed Therefore, we utilized a two-stage design to identify even within the same stage of cancer, indicating the NSCLC prognosis associated epigenetic biomarkers in possible existence of prognosis-influencing molecular TRIM27 and further explored the potential reason of mechanisms [5]. Epigenetic alterations such as DNA heterogeneous effect of biomarkers across histology by methylation are considered important representatives performing epigenetic–smoking interaction analysis. of these molecular mechanisms [6]. Meanwhile, the robustly significant biomarkers were inves- Tripartite motif containing 27 (TRIM27) is highly tigated for the associated alterations in gene expression expressed in lung cancer and plays an important role which were also studied for effect on lung cancer survival. in cancer prognosis [7,8] by encoding a member of the tripartite motif (TRIM) family. TRIM family proteins 2. Materials and methods play crucial roles in a wide range of processes, includ- ing cell growth, apoptosis and stem cell differentiation [9]. TRIM27 is an oncogene of various tumour types, 2.1. Study populations including colitis-associated cancer, salivary gland intra- ductal carcinoma, colon cancer, uterus cancer and We collected data from early-stage (stage I and II) prostate cancer [7,10,11]. Further, originally identified NSCLC patients from five international study centres. to be involved in oncogenic rearrangements with the Cases from the Harvard, Spain, Norway and Sweden transfection proto-oncogene (RET), TRIM27 is also cohorts were assigned into the discovery phase [31–34], known as RFP (RET finger protein) [12]. RET rear- while cases from The Cancer Genome Atlas (TCGA) rangements were implicated in NSCLC [13]. were assigned into the validation phase. All patients DNA methylation has been recognized as cancer provided written informed consent. The study method- biomarkers and therapeutic targets for NSCLC [14– ologies conformed to the standards set by the Declara- 21], as well as other cancers [22,23], since it is stable to tion of Helsinki and was approved by the local ethics measure but is modifiable with proper interventions committee. [24,25]. Also, aberrant DNA methylation of TRIM27 was identified affecting lung function in monozygotic 2.1.1. Harvard twins [26]. Nevertheless, the association between DNA methylation of TRIM27 and early-stage NSCLC sur- The Harvard Lung Cancer Study cohort was described vival still remains largely unclear. previously [31]. Patients were recruited at 2760 Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. X. Ji et al. Epigenetic–smoking interaction analysis of TRIM27 Massachusetts General Hospital (MGH) since 1992. 2.2. Quality control procedures for DNA All were newly diagnosed and histologically confirmed methylation data as primary NSCLC at the time of recruitment. Snap- frozen tumour samples were taken from patients dur- For each patient, DNA methylation was assessed ing complete resection. The 151 early-stage patients using Infinium HumanMethylation450 BeadChips selected in this study had complete survival informa- (Illumina Inc., San Diego, CA, USA). All centres fol- tion. Tumour DNA was extracted from 5-lm-thick lowed the same quality control (QC) procedures histopathologic sections. Each specimen was evaluated before conducting the association study. GenomeStu- by an MGH pathologist for amount (tumour cellular- dio Methylation Module V1.8 (Illumina Inc.) was ity > 70%) and quality of tumour cells. All specimens used to convert raw image data into beta values (con- were histologically classified using World Health Orga- tinuous numbers ranging 0–1) for background sub- nization criteria. traction and control normalization. Unqualified probes meeting any one of the following criteria were excluded: (a) failed detection (P > 0.05) in > 5% of 2.1.2. Spain samples; (b) coefficient of variance of < 5%; (c) all The study population was described previously [32]. samples methylated or unmethylated; (d) common Tumours were collected by surgical resection from 226 single nucleotide polymorphisms located in the probe patients. DNA extraction was performed on tumour sequence or 10-bp flanking regions; (e) cross-reactive specimens (10 lm thick, tumour cellularity > 50%). probes or cross-hybridizing probes [35]; or (f) did not The study was approved by the Bellvitge Biomedical pass QC in all centres. Samples with > 5% unde- Research Institute Institutional Review Board. tectable probes were excluded. Methylation signals were further processed for quantile normalization (be- taqn function in R package minfi) as well as type I 2.1.3. Norway and II probe correction (BMIQ function in R package Participants were 133 LUAD patients with operable lumi). Data were adjusted for batch effects (ComBat lung cancer tumours seen at Oslo University Hospital, function in R package sva) according to the best pipe- Rikshospitalet, Norway, in 2006–2011 [33]. Tumour line by a comparative study [36]. Details of QC pro- tissues were collected during surgery, snap-frozen in cesses are described in Fig. S1. liquid nitrogen, and stored at 80 °C until DNA iso- lation. All early-stage patients did not receive 2.3. Gene expression data chemotherapy or radiotherapy before surgery. The project was approved by the Oslo University Institu- In TCGA cohort, all of the 281 LUAD and 277 LUSC tional Review Board and the Regional Ethics Commit- cases had complete mRNA sequencing data. Gene tee (S-05307). expression was measured by RNA sequencing. Data processing and QC were done by TCGA workgroup. Raw counts were normalized by RNA-seq expectation 2.1.4. Sweden maximization. Level-3 gene quantification data were Tumour tissue samples were collected from 103 downloaded from TCGA and were further checked for patients with early-stage NSCLC who underwent oper- quality. Expression of TRIM27 was extracted and ation at Skane University Hospital, Lund, Sweden log2-transformed before analysis. [34]. The study was approved by the Regional Ethical Review Board in Lund, Sweden (Registration no. 2.4. Statistical analysis 2004/762 and 2008/702). The study design is shown in Fig. 1. To investigate the association between DNA methylation of TRIM27 2.1.5. TCGA and overall survival, we applied a Cox proportional TCGA dataset included 332 LUAD and 285 LUSC hazards model adjusted for age, sex, smoking status, cases. Overall survival times and common covariates clinical stage and study centre for LUAD and LUSC were included. Level-1 HumanMethylation450 DNA patients, respectively. Proportional hazards assumption methylation data (image data) of each patient were for each CpG probe was also tested. Hazard ratio downloaded from https://portal.gdc.cancer.gov on 1 (HR) and 95% confidence interval (CI) were with October 2015. respect to per 1% level of methylation increment. Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 2761 Epigenetic–smoking interaction analysis of TRIM27 X. Ji et al. Multiple comparisons were adjusted by using false dis- 2.5. Methylation–smoking interaction analysis covery rate method (FDR; measured by FDR-q value) [37] to control the overall false-positive rate at 5% We observed a significant heterogeneous effect of level. CpG probes with FDR-q ≤ 0.05 in the discovery cg05293407TRIM27 across histology, but the distribu- phase were further replicated in the validation phase. tions of cg05293407TRIM27 methylation in LUAD and Robustly significant CpG probes were finally retained LUSC patients were comparable. Meanwhile, com- if they met the following criteria: (a) P ≤ 0.05 in vali- bined prior literature information with our results, all dation phase and (b) consistent effect direction across evidence indicated that heavy smoking was relevant to two phases. For robustly significant CpG probes, LUSC. Therefore, we hypothesized that this hetero- Kaplan–Meier curves were used to compare survival geneity might be explained by methylation–smoking difference between patients with different methylation interaction, which was further tested as a product term levels. (methylation and pack-year of smoking) in a Cox Fig. 1. Flow chart of study design and statistical analyses. 2762 Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. X. Ji et al. Epigenetic–smoking interaction analysis of TRIM27 proportional hazards model adjusted for same covari- (Fig. 3A), indicating a significant heterogeneous effect ates as aforementioned. of cg05293407TRIM27 across NSCLC histology (P = 8.66 9 10 3). Also, the pattern of survival curves for patients with high, medium and low methylation 2.6. Genome-wide methylation-transcription levels based on tertiles of methylation level differed analysis between LUAD and LUSC patients. Survival curves Based on hypothesis of omnigenetic model [38], for were significantly separated only in LUSC patients these identified prognostic CpG probes, we used a lin- (Fig. 3B,C). ear regression model adjusted for the aforementioned However, distribution of cg05293407TRIM27 methyla- covariates to test the association between DNA methy- tion in LUAD and LUSC patients was similar lation and gene expression using transcriptomic data (Fig. 4A) and comparable (P = 0.518) by Wilcoxon from TCGA. Significant genes were identified as FDR- rank-sum test (Fig. 4B). Since nonsmokers and long- q ≤ 0.05 and presented in Circos plot. Then, the asso- term former smokers are more common in LUAD ciation between gene expression and overall survival patients, while the majority of lung cancer patients was further evaluated using Cox models adjusted for who are current smokers have LUSC [29]. Therefore, the same covariates. Genes significantly associated we assumed that there might exist a methylation– with both methylation and NSCLC survival were smoking interaction accounting for the heterogeneous screened out. effect of cg05293407TRIM27 on NSCLC survival across Continuous variables were expressed as mean  s- histology. The smoking-related variables were com- tandard deviation (SD), and categorical variables were pared between the LUAD and LUSC patients expressed in frequency (n) and proportion (%). Statis- (Table S6). Compared with LUAD patients, LUSC tical analysis was performed using R version 3.5.2 (The patients had more pack-year of smoking averagely R Foundation of Statistical Computing, Tsinghua (37.49LUAD vs 54.79LUSC, P = 1.03 9 10 19) (Fig. 4C, University, Beijing, China). D) and a higher proportion of current smokers (28.24- %LUAD vs 34.09%LUSC, P = 0.037) (Table S6). We identified a significant interaction between 3. Results cg05293407TRIM27 and pack-year of smoking in all Demographic and clinical information for patients NSCLC patients (HRinteraction = 1.01, 95% CI: 1.00– with DNA methylation and gene expression data were 1.02, P = 0.034). With increased pack-year of smoking, detailed in Table 1 and Table S1. There were 96 CpG there was an elevated risk for high methylation of probes located in TRIM27 (Table S2). In LUAD cg05293407TRIM27 on NSCLC survival (Fig. 5). There- patients, none of the CpG probes in the discovery fore, pack-year of smoking was a modifier of the asso- phase were identified by the criterion of FDR-q ≤ 0.05 ciation between cg05293407TRIM27 and NSCLC (Table S3). One probe, cg05293407TRIM27, was signifi- survival. cantly associated with LUSC survival in both discov- We also evaluated joint effect of cg05293407TRIM27 ery (HR = 2.10, 95% CI: 1.41–3.12, P = 2.70 9 10 4, methylation level and pack-year of smoking on FDR-q = 0.026) and validation phases (HR = 1.49, NSCLC survival (Table 2). Patients were categorized 95% CI: 1.07–2.07, P = 0.018) and showed robust into three groups (high, medium and low) by tertiles association in combined data (HR = 1.65, 95% CI: of cg05293407TRIM27 methylation level (1.33% and 1.30–2.09, P = 4.52 9 10 5) (Tables S4 and S5). 1.78%) and were also categorized into three groups To further exemplify the effect of cg05293407TRIM27 (high, medium and low) by cut-off values of pack-year on overall survival, patients were categorized into two of smoking (39 and 54). Only for these patients having groups (high vs low) based on median value of > 39 pack-year of smoking, cg05293407TRIM27 was a methylation level. Kaplan–Meier survival curves significant risk factor (Fig. 5). Therefore, 39 was showed significant differences between two groups in defined as a cut-off value of low and medium–high the discovery phase (HRHigh vs Low = 2.10, 95% CI: levels. Further, the median value (54) of pack-year of 1.28–3.45, P = 3.41 9 10 3), the validation phase smoking for LUAD patients having > 39 pack-year of (HRHigh vs Low = 1.92, 95% CI: 1.20–3.07, smoking was used to distinguish medium and high P = 6.36 9 10 3) and combined data levels. We used the best prognosis group (low–medium (HRHigh vs Low = 1.88, 95% CI: 1.34–2.63, P = 2.35 9 methylation of cg05293407TRIM27 and low–medium 10 4) (Fig. 2). However, the effect of pack-year of smoking) as the reference to evaluate cg05293407TRIM27 on LUAD survival was not signifi- effects of high methylation level, high pack-year of cant (HR = 1.08, 95% CI: 0.87–1.33, P = 0.493) smoking and their joint effect, as well as interaction. Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 2763 Epigenetic–smoking interaction analysis of TRIM27 X. Ji et al. Table 1. Demographic and clinical descriptions for early-stage NSCLC patients in five international study centres. Discovery Validation Combined USA Spaina Norway Sweden TCGA Overall Variable (N = 151) (N = 226) (N = 133) (N = 103) All (N = 613) (N = 617) (N = 1230) Age (years) 67.67  9.92 65.67  10.58 65.52  9.34 67.54  9.99 66.44  10.08 66.51  9.47 66.48  9.78 Sex Female 67 (44.37%) 105 (46.46%) 71 (53.38%) 54 (52.43%) 297 (48.45%) 255 (41.33%) 552 (44.88%) Male 84 (55.63%) 121 (53.54%) 62 (46.62%) 49 (47.57%) 316 (51.55%) 362 (58.67%) 678 (55.12%) Smoking status Never 18 (11.92%) 30 (13.57%) 17 (12.78%) 18 (17.48%) 83 (13.65%) 55 (9.18%) 138 (11.43%) Former 81 (53.64%) 120 (54.30%) 74 (55.64%) 54 (52.43%) 329 (54.11%) 376 (62.77%) 705 (58.41%) Current 52 (34.44%) 71 (32.13%) 42 (31.58%) 31 (30.10%) 196 (32.24%) 168 (28.05%) 364 (30.16%) Unknown 0 5 0 0 5 18 23 Clinical stage I 104 (68.87%) 183 (80.97%) 93 (69.92%) 95 (92.23%) 475 (77.49%) 393 (63.70%) 868 (70.57%) II 47 (31.13%) 43 (19.03%) 40 (30.08%) 8 (7.77%) 138 (22.51%) 224 (36.30%) 362 (29.43%) Histology LUAD 96 (63.58%) 183 (80.97%) 133 (100.00%) 80 (77.67%) 492 (80.26%) 332 (53.81%) 824 (66.99%) LUSC 55 (36.42%) 43 (19.03%) 0 (0.00%) 23 (22.33%) 121 (19.74%) 285 (46.19%) 406 (33.01%) Chemotherapy No 142 (94.04%) 177 (90.77%) 102 (76.69%) 67 (90.54%) 488 (88.25%) 194 (76.98%) 682 (84.72%) Yes 9 (5.96%) 18 (9.23%) 31 (23.31%) 7 (9.46%) 64 (11.75%) 58 (23.02%) 123 (15.28%) Unknown 0 31 0 29 60 365 425 Radiotherapy No 132 (87.42%) 184 (94.36%) 132 (99.25%) 74 (100.00%) 522 (94.39%) 239 (94.84%) 761 (94.53%) Yes 19 (12.58%) 11 (5.64%) 1 (0.75%) 0 (0.00%) 31 (5.61%) 13 (5.16%) 44 (5.47%) Unknown 0 31 0 29 60 365 425 Adjuvant therapyb No 127 (84.11%) 168 (86.15%) 101 (75.94%) 67 (90.54%) 463 (83.73%) 187 (74.21%) 650 (80.75%) Yes 24 (15.89%) 27 (13.85%) 32 (24.06%) 7 (9.46%) 90 (16.27%) 65 (25.79%) 155 (19.25%) Unknown 0 31 0 29 60 365 425 Survival yearc Median (95% 6.66 (5.41– 7.12 (5.06– 7.36 (6.77– 7.39 (4.98–9.12) 7.39 (6.50– 4.54 (3.68– 6.60 (5.84–7.35) CI) 7.87) 9.63) 7.95)* 8.23) 5.41) Censoring rate 19.21% 55.31% 68.42% 43.69% 43.31% 76.99% 62.20% a Spain centre is a collaborative study centre, containing samples from Spain, Italy, UK, France and United States. b Adjuvant therapy included chemotherapy and/or radiotherapy. c Restricted mean survival time was given because median was not available; proportion of samples lost to follow-up or alive at end of study. In the combined dataset, the main effect of high pack- survival was evaluated in patients with low, medium year of smoking was HR = 1.12 (95% CI: 0.84–1.49; and high levels of pack-year of smoking. The effect of P = 0.438), and the main effect of high methylation of cg05293407TRIM27 varied across patients with different cg05293407TRIM27 was HR = 1.10 (95% CI: 0.83–1.45; pack-year of smoking. For LUAD patients with a high P = 0.502). However, the joint effect was HR = 2.00 level of pack-year of smoking, high methylation of (95% CI: 1.43–2.79; P = 5.00 9 10 5), which was cg05293407TRIM27 had significantly worse survival greater than the product of the two individual risk (HRHigh vs Low = 1.88, 95% CI: 1.07–3.32, P = 0.029) effects (1.2320 = 1.12 9 1.10), indicating a synergistic (Fig. 6A,B). In LUSC and overall NSCLC patients, we interaction between high methylation of observed similar significant results in patients with both cg05293407TRIM27 and high pack-year of smoking medium (HRHigh vs Low = 2.51, 95% CI: 1.10–5.73, (HRinteraction = 1.62; 95% CI: 1.04–2.53; P = 0.029 in LUSC patients; HRHigh vs Low = 1.89, P = 3.21 9 10 2). 95% CI: 1.17–3.06, P = 9.35 9 10 3 in overall To illustrate the modification effect by pack-year of patients) and high (HRHigh vs Low = 2.55, 95% CI: smoking, effect of cg05293407TRIM27 on NSCLC 1.43 4.55, P = 1.49 9 10 3 in LUSC patients; 2764 Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. X. Ji et al. Epigenetic–smoking interaction analysis of TRIM27 Fig. 2. Kaplan–Meier survival curves of LUSC patients. High and low methylation groups were defined according to median value in (A) discovery phase (N = 121, median value: 1.72%), (B) validation phase (N = 285, median value: 1.48%) and (C) combined data (N = 406, median value: 1.55%). HR, 95% CI, and P value were derived from a Cox proportional hazards regression model adjusted for age, sex, smoking status, clinical stage and study centre. Fig. 3. Methylation and histology interaction on survival of NSCLC patients. (A) Forest plot of the effects of cg05293407TRIM27 in LUAD and LUSC populations. HR, 95% CI, and P value were derived from a Cox proportional hazards regression model adjusted for age, sex, smoking status, clinical stage and study centre. PHeterogeneity was used to evaluate heterogeneity across both groups. (B) Kaplan– Meier survival curves of LUAD patients (N = 824) with high, medium and low methylation categorized by tertiles (1.38% and 1.84%) of cg05293407TRIM27. (C) Kaplan–Meier survival curves of LUSC patients (N = 406) with high, medium and low methylation categorized by tertiles (1.32% and 1.79%) of cg05293407TRIM27. HRHigh vs Low = 1.94, 95% CI: 1.32–2.84, sensitivity analysis based on smoking status also indi- P = 7.51 9 10 4 in overall patients) levels of pack-year cated an upward trend (PTrend = 0.022) in effect size of of smoking (Fig. 6C–F). Our results indicated that cg05293407TRIM27 from never smokers (HR = 0.89), cg05293407TRIM27 influenced NSCLC survival actually former smokers (HR = 1.23) to current smokers regardless of histology, but only among these patients (HR = 1.88) in overall population, even not taking exposed to relatively heavy smoking. Since the pack- pack-year of smoking into account (Fig. S4). year of smoking might be bimodal distributed due to Further, because cg05293407TRIM27 maps to plenty of zero values from never smokers, we also per- TRIM27, the association between TRIM27 expression formed sensitivity analysis by testing the methylation– and cg05293407TRIM27 methylation was evaluated in smoking interaction in NSCLC patients excluding the TCGA population. In NSCLC patients, there was never smokers and still observed the significant interac- a significant association (r = 0.13, P = 1.38 9 10 3) tion (Fig. S2) and same pattern (Fig. S3). Another (Fig. 7), suggesting that methylation of Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 2765 Epigenetic–smoking interaction analysis of TRIM27 X. Ji et al. Fig. 4. Distribution of cg05293407TRIM27 and pack-year of smoking in LUAD (N = 824) and LUSC patients (N = 406). (A) Distribution of cg05293407TRIM27. (B) Box plot of cg05293407TRIM27. (C) Distribution of pack-year of smoking. And, 39 and 54 are the two cut-off values of the three groups of pack-year of smoking (low, medium and high). (D) Box plot of pack-year of smoking in LUAD patients (N = 652) and LUSC patients (N = 342). Wilcoxon rank- sum test was used to estimate the P value. cg05293407TRIM27 upregulated TRIM27 expression. the two NSCLC types [41]. We only observed one However, TRIM27 expression had no significant asso- probe, cg05293407TRIM27, exclusively associated with ciation with overall survival (HR = 0.84, 95% CI: early-stage LUSC prognosis in stratified analysis by 0.62–1.14, P = 0.270) (Fig. S5). However, genome- histology, whereas no promising CpG probes were wide methylation-transcription analysis showed that observed for LUAD, possibly due to underlying epige- expression of 29 genes was associated with netic heterogeneity between LUAD and LUSC. Fur- cg05293407TRIM27 (Table S7 and Fig. S6A). Among ther, LUSC is more strongly associated with smoking them, expression of three genes was associated with than LUAD, suggesting different causes for their overall survival: ubiquitin-specific protease 26 induction as well [30]. In addition, a methylation– (USP26), gap junction protein gamma 3 (GJC3), and smoking interaction may potentially provide interpre- N-acetylated alpha-linked acidic dipeptidase 2 (NAA- tation of the heterogeneous effect of LAD2) (Fig. S6B–D). cg05293407TRIM27. The tumour-specific shift to transcriptional repres- sion is associated with DNA methylation at TSSs in 4. Discussion multiple tumour types [42]. Generally, hyper-methyla- We performed a two-stage study and integrative analy- tion blocks transcription initiation and reduces gene sis of DNA methylation of TRIM27 and gene expres- expression [43]. However, a small proportion of sion in early-stage NSCLC patients. The CpG probe, methylation surrounding the TSS region upregulates cg05293407TRIM27, located at the 200 kb transcription gene expression, indicating that DNA methylation reg- start site (TSS) region of TRIM27, was identified as an ulation may be more complex [44]. In our study, DNA exclusive biomarker of early-stage LUSC prognosis. methylation at cg05293407 in the 200 kb TSS region Further, the heterogeneous effect of cg05293407TRIM27 of TRIM27 upregulated gene expression in tumour tis- across histology may be explained by a methylation– sues, which was consistent with previous reports smoking interaction. [45,46]. This phenomenon may be mediated by affect- As LUAD and LUSC differ in the origin and histol- ing the binding activity of upstream transcription fac- ogy, the mechanism of occurrence and progression tors [47]. However, further functional studies are may be different at a molecular level [39,40]. For warranted to elaborate the possible mechanism. example, both mutated genes and recurrent somatic In LUSC patients, the methylation level of copy number alterations are largely distinct between cg05293407TRIM27 ranged from 0.62% to 4.09% and 2766 Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. X. Ji et al. Epigenetic–smoking interaction analysis of TRIM27 Fig. 5. Methylation–smoking interaction on survival of NSCLC patients (N = 994). HR of interaction term, 95% CI, and P value were derived from a Cox proportional hazards regression model adjusted for age, sex, smoking status, clinical stage, cancer type and study centre. HR of methylation estimated based on level of pack-year of smoking. Shallow area represented 95% CI. Histogram on the top shows the distribution of pack-year of smoking. Table 2. Joint effect and interaction of elevated methylation and pack-year of smoking on prognosis of early-stage NSCLC. Effect type a High methylationb High pack-year of smoking Number Death Crude mortality HR (95% CI) a Pa No No 486 158 32.51% Reference Main effect 1 No Yes 177 77 43.50% 1.12 (0.84, 1.49) 0.438 Main effect 2 Yes No 229 94 41.05% 1.10 (0.83, 1.45) 0.502 Joint effect Yes Yes 102 56 54.90% 2.00 (1.43, 2.79) 5.00 9 10 5 Interaction c 1.62 (1.04, 2.53) 3.21 9 10 2 a Patients were categorized into three groups (high vs medium–low) by tertiles of cg05293407TRIM27 methylation level (1.33% and 1.78%) and were categorized into three groups (high vs medium–low) by cut-off value of pack-year of smoking (39 and 54). The risk effect of cg05293407TRIM27 on survival was significant only for these patients having > 39 pack-year of smoking. And, 54 was the median value of pack-year of smoking for LUAD patients having > 39 pack-year of smoking. b Main effects of elevated methylation and pack-year of smoking as well as their joint effect and interaction were derived from Cox propor- tional hazards regression model adjusted for age, sex, clinical stage, histology, study centre, and stratified by cancer type. c Interaction = Joint effect  (Main effect 1 9 Main effect 2). 1.62  2.00  (1.12 9 1.10). its median value was 1.48%, indicating a narrow range TRIM27 belongs to the TRIM family, an extended and low average level. As shown in Fig. S7, these family of proteins with a common denominator of a maximum values of all 311 891 CpG probes followed tripartite combinatorial motif encompassing RING fin- a bimodal distribution with the first peak around 5%. ger, B-box, and coiled-coil domain homologies [50]. Furthermore, there were 9341 (2.99%) CpG probes TRIM27 is an important positive regulator of signal with even lower maximum value than that of transducer and activator of transcription 3 (STAT3) cg05293407TRIM27 indicating its narrow range was rea- activation. TRIM27, located at retromer-positive struc- sonable. Meanwhile, plenty of studies have revealed tures, can recruit STAT3 after IL-6 stimulation and that aberrant DNA methylations of these hypo-methy- lead to improved STAT3 activation [11]. STAT3 activ- lated CpG probes were also involved in diseases (e.g., ity plays important roles in pathogenesis of many can- female panic disorder risk associated cg07308824HECA cers, including breast, head and neck, prostate and and paediatric medulloblastoma prognosis associated brain cancers [51]. Further, STAT3 is overexpressed in cg02257300ERCC2) [48,49]. NSCLC tumour samples, and sorafenib can inhibit Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 2767 Epigenetic–smoking interaction analysis of TRIM27 X. Ji et al. Fig. 6. Kaplan–Meier survival curves of LUAD, LUSC and overall NSCLC patients. Kaplan–Meier curves of low and high methylation groups and forest plots of the effects of cg05293407TRIM27 were stratified by populations having low, medium and high pack-year of smoking for (A, B) LUAD patients (NLow = 370, NMedium = 143, NHigh = 139), (C, D) LUSC patients (NLow = 106, NMedium = 96, NHigh = 140) and (E, F) overall patients (NLow = 476, NMedium = 239, NHigh = 279). HR, 95% CI, and P value were derived from a Cox proportional hazards regression model adjusted for age, sex, smoking status, clinical stage and study centre. PHeterogeneity was used to evaluate heterogeneity of HRs across groups. STAT3 activation to produce anticancer effects in patients with high pack-year of smoking, heavy expo- NSCLC [52]. Combined with our results, these data sure to cigarette smoking may strongly inhibit PTEN suggested that high methylation of cg05293407TRIM27 expression through ROS and relate to poor NSCLC might promote TRIM27 expression, further leading to prognosis (Fig. 8). STAT3 activation and poor prognosis (Fig. 8). Moreover, PTEN is an essential modulator of Smoking is associated with several genetic alter- STAT3-mediated pathways. Although STAT3 is a ations in NSCLC [53] and has been well-established as downstream target of PTEN, STAT3 also reversely a relevant factor of lung cancer risk as well as progno- inhibits PTEN expression by directly activating miR- sis [15]. Cigarette smoke contains reactive oxygen spe- 21, which is part of the epigenetic switch linking cies (ROS), which inhibit phosphatase and tensin inflammation to cancer [57]. Therefore, STAT3 activa- homolog (PTEN) expression by phosphorylating the tion can downregulate PTEN expression (Fig. 8). In ROS-dependent Src/EGFR-p38MAPK pathway [54]. terms of the cg05293407TRIM27 and smoking interac- PTEN inhibits glycolysis in brain tumour cells by tion, high methylation was associated with poor prog- directly interacting with phosphoglycerate kinase 1 nosis in NSCLC patients with medium–high pack-year (PGK1) [55]. Further, PTEN inhibits cancer cells by of smoking rather than low pack-year of smoking, moderating signalling through the PI3K pathway. possibly because high activation of STAT3 and low PTEN is lowly expressed in NSCLC tumour samples expression of PTEN may only occur in patients with and is more prevalent in LUSC [56]. Therefore, for medium–high methylation of cg05293407TRIM27. 2768 Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. X. Ji et al. Epigenetic–smoking interaction analysis of TRIM27 Fig. 7. Association between DNA methylation of cg05293407 and expression of corresponding gene TRIM27 in TCGA using 277 biologically independent samples. Correlation coefficients and hypothesis tests were based on Pearson correlation test. Gene expression was log2-transformed before analysis. Histogram on top shows the distribution of cg05293407 methylation; histogram on side shows the distribution of TRIM27 expression. We observed three genes associated with besides the significant statistical interaction observed cg05293407TRIM27: GJC3, NAALAD2 and USP26. on a population level, we experimentally elaborated on GJC3 is one of the genes coding for connexin (CX) a plausible functional interaction between two path- proteins and is reported to be associated with nonsyn- ways based on literature evidence. Third, by control- dromic hearing loss [58]. Further, patients with low ling false positives, our two-stage study and the GJC3 expression had a better prognosis in our study. sensitivity analysis provided robustness to our results. NAALAD2 encodes human prostate-specific membrane Fourth, we performed integrative analysis of DNA antigen (PSM), which is a marker of prostatic carcino- methylation and gene expression and systematically mas and is the first shown to possess NAALADase evaluated associated genes of cg05293407TRIM27 on activity [59]. Similarly, LUSC patients with lower genome-wide scale. NAALAD2 expression had higher survival in our We also acknowledge some limitations. First, study. USP26 is associated with Sertoli cell-only syn- though three genes associated with cg05293407TRIM27 drome and male infertility in both European and Chi- further affected lung cancer prognosis in our study, nese men [60,61]. Further, our study showed consistent there was no explicit evidence of their mechanisms. results in LUSC patients. Although these three genes Therefore, these associations should be interpreted lack explicit evidence of association with LUSC, their with caution. Second, with a high censoring rate in the relationship to cg05293407TRIM27 and LUSC survival TCGA cohort, the statistical power might be limited. may inspire functional studies of these potential genes Anyway, the association between cg05293407TRIM27 and further help elucidate the mechanistic pathway of and prognosis remained significant in TCGA, indicat- cg05293407TRIM27 on LUSC survival. ing our results were conservative and roust. Third, the Our study has several strengths. First, to our knowl- positive association between cg05293407TRIM27 and edge, this is the first multicentre study of interaction TRIM27 expression was not reported by the other lit- between DNA methylation of TRIM27 and smoking, eratures yet. Further functional experiments are war- which attempted to interpret the effect of DNA methy- ranted to confirm our results. Finally, as the majority lation that varied by NSCLC histology. Second, of our population was Caucasian (89.19%), Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 2769 Epigenetic–smoking interaction analysis of TRIM27 X. Ji et al. Fig. 8. Diagram for pathway of DNA methylation–smoking interaction effect on survival for LUSC patients. generalization of the results to the other ethnicity Higher Education Institutions (PAPD). RZ was par- groups should be cautioned. tially supported by the Outstanding Young Teachers Training Program of Nanjing Medical University. 5. Conclusion Conflict of interest In summary, our study identified cg05293407TRIM27 as a potential biomarker for LUSC prognosis and laid The authors declare no conflict of interest. out a case that the methylation–smoking interaction may account for heterogeneous effects of Author contributions cg05293407TRIM27 across histology. Our findings pro- vide a potential dynamic and reversible therapeutic XJ, LL, RZ, YW, FC and DCC contributed to the target for NSCLC patients. study design. RZ, LS, AS, MMB, AK, MP, JS, AH,  ME and DCC contributed to data collection. XJ, LL, SS, XD, CC, YL, RZ, YW, FC and DCC performed Acknowledgements statistical analysis and interpretation and drafted the The authors thank TCGA for contributing clinical, manuscript. YZ, HH, XC, JC, LW, JH, WD, LS, YJ, DNA methylation and RNA sequencing data, as well as JF, JG and DY revised the manuscript. All authors all study subjects who participated in the five study contributed to critical revision of the manuscript and cohorts. This study was funded by the National Natural approved its final version. Financial support and study Science Foundation of China (81530088 to FC, supervision were provided by RZ, YW, FC and DCC. 81973142 to YW), National Key Research and Develop- ment Program of China (2016YFE0204900 to FC), Nat- Data accessibility ural Science Foundation of Jiangsu Province (BK20191354 to RZ), Natural Science Foundation of The DNA methylation image data of Harvard, Spain, the Jiangsu Higher Education Institutions of China Norway and Sweden study cohort can be requested (18KJB310011 to RZ), China Postdoctoral Science from DCC, ME, AH  and JS, respectively. Alterna- Foundation (2018M633767 to RZ), US National Insti- tively, it can be retrieved from gene expression omni- tutes of Health (CA209414, CA092824, and bus database (GSE39279, GSE66836 and GSE56044). U01CA209414 to DCC and YL; AG056764 to YL), and TCGA: https://tcga-data.nci.nih.gov; now hosted at Priority Academic Program Development of Jiangsu GDC: https://portal.gdc.cancer.gov. 2770 Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. X. Ji et al. Epigenetic–smoking interaction analysis of TRIM27 References define a unique molecular and clinicopathologic subtype of non-small-cell lung cancer. J Clin Oncol 30, 4352– 1 Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre 4359. LA & Jemal A (2018) Global cancer statistics 2018: 14 Zhang R, Chen C, Dong X, Shen S, Lai L, He J, You GLOBOCAN estimates of incidence and mortality D, Lin L, Zhu Y, Huang H et al. (2020) Independent worldwide for 36 cancers in 185 countries. CA Cancer J validation of early-stage NSCLC prognostic scores Clin 68, 394–424. incorporating epigenetic and transcriptional biomarkers 2 Chen Z, Fillmore CM, Hammerman PS, Kim CF & with gene-gene interactions and main effects. Chest 158, Wong KK (2014) Non-small-cell lung cancers: a 808–819. heterogeneous set of diseases. Nat Rev Cancer 14, 535– 15 Zhang R, Lai L, Dong X, He J, You D, Chen C, Lin 546. L, Zhu Y, Huang H, Shen S et al. (2019) SIPA1L3 3 Molina JR, Yang P, Cassivi SD, Schild SE & Adjei AA methylation modifies the benefit of smoking cessation (2008) Non-small cell lung cancer: epidemiology, risk on lung adenocarcinoma survival: an epigenomic- factors, treatment, and survivorship. Mayo Clin Proc smoking interaction analysis. Mol Oncol 13, 1235–1248. 83, 584–594. 16 Zhang R, Lai L, He J, Chen C, You D, Duan W, 4 Rami-Porta R, Asamura H, Travis WD & Rusch VW Dong X, Zhu Y, Lin L, Shen S et al. (2019) EGLN2 (2017) Lung cancer – major changes in the American DNA methylation and expression interact with HIF1A Joint Committee on Cancer eighth edition cancer to affect survival of early-stage NSCLC. Epigenetics 14, staging manual. CA Cancer J Clin 67, 138–155. 118–129. 5 Tang S, Pan Y, Wang Y, Hu L, Cao S, Chu M, Dai J, 17 Guo Y, Zhang R, Shen S, Wei Y, Salama SM, Shu Y, Xu L, Chen J et al. (2015) Genome-wide Fleischer T, Bjaanaes MM, Karlsson A, Planck M, Su association study of survival in early-stage non-small L et al. (2018) DNA methylation of LRRC3B: a cell lung cancer. Ann Surg Oncol 22, 630–635. biomarker for survival of early-stage non-small cell lung 6 Heyn H & Esteller M (2012) DNA methylation cancer patients. Cancer Epidemiol Biomarkers Prev 27, profiling in the clinic: applications and challenges. Nat 1527–1535. Rev Genet 13, 679–692. 18 Wei Y, Liang J, Zhang R, Guo Y, Shen S, Su L, Lin 7 Zoumpoulidou G, Broceno C, Li H, Bird D, Thomas G X, Moran S, Helland A, Bjaanaes MM et al. (2018) & Mittnacht S (2012) Role of the tripartite motif Epigenetic modifications in KDM lysine demethylases protein 27 in cancer development. J Natl Cancer Inst associate with survival of early-stage NSCLC. Clin 104, 941–952. Epigenetics 10, 41. 8 Iwakoshi A, Murakumo Y, Kato T, Kitamura A, Mii 19 Shen S, Zhang R, Guo Y, Loehrer E, Wei Y, Zhu Y, S, Saito S, Yatabe Y & Takahashi M (2012) RET finger Yuan Q, Moran S, Fleischer T, Bjaanaes MM et al. protein expression is associated with prognosis in lung (2018) A multi-omic study reveals BTG2 as a reliable cancer with epidermal growth factor receptor mutations. prognostic marker for early-stage non-small cell lung Pathol Int 62, 324–330. cancer. Mol Oncol 12, 913–924. 9 Ozato K, Shin DM, Chang TH & Morse HC III (2008) 20 Dong X, Zhang R, He J, Lai L, Alolga RN, Shen S, TRIM family proteins and their emerging roles in Zhu Y, You D, Lin L, Chen C et al. (2019) Trans- innate immunity. Nat Rev Immunol 8, 849–860. omics biomarker model improves prognostic prediction 10 Skalova A, Vanecek T, Uro-Coste E, Bishop JA, accuracy for early-stage lung adenocarcinoma. Aging Weinreb I, Thompson LDR, de Sanctis S, Schiavo- (Albany NY) 11, 6312–6335. Lena M, Laco J, Badoual C et al. (2018) Molecular 21 Shen S, Wei Y, Zhang R, Du M, Duan W, Yang S, profiling of salivary gland intraductal carcinoma Zhao Y, Christiani DC & Chen F (2018) Mutant-allele revealed a subset of tumors harboring NCOA4-RET fraction heterogeneity is associated with non-small cell and novel TRIM27-RET fusions: a report of 17 cases. lung cancer patient survival. Oncol Lett 15, 795–802. Am J Surg Pathol 42, 1445–1455. 22 Joo JE, Dowty JG, Milne RL, Wong EM, Dugue PA, 11 Zhang HX, Xu ZS, Lin H, Li M, Xia T, Cui K, Wang English D, Hopper JL, Goldgar DE, Giles GG & SY, Li Y, Shu HB & Wang YY (2018) TRIM27 Southey MC (2018) Heritable DNA methylation marks mediates STAT3 activation at retromer-positive associated with susceptibility to breast cancer. Nat structures to promote colitis and colitis-associated Commun 9, 867. carcinogenesis. Nat Commun 9, 3441. 23 Widschwendter M, Zikan M, Wahl B, Lempiainen H, 12 Takahashi M, Ritz J & Cooper GM (1985) Activation Paprotka T, Evans I, Jones A, Ghazali S, Reisel D, of a novel human transforming gene, ret, by DNA Eichner J et al. (2017) The potential of circulating rearrangement. Cell 42, 581–588. tumor DNA methylation analysis for the early 13 Wang R, Hu H, Pan Y, Li Y, Ye T, Li C, Luo X, detection and management of ovarian cancer. Genome Wang L, Li H, Zhang Y et al. (2012) RET fusions Med 9, 116. Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 2771 Epigenetic–smoking interaction analysis of TRIM27 X. Ji et al. 24 Egger G, Liang G, Aparicio A & Jones PA (2004) evaluation of analysis pipelines for DNA methylation Epigenetics in human disease and prospects for profiling using the Illumina HumanMethylation450 epigenetic therapy. Nature 429, 457–463. BeadChip platform. Epigenetics 8, 333–346. 25 Feinberg AP & Tycko B (2004) The history of cancer 37 Benjamini Y & Hochberg Y (1995) Controlling the false epigenetics. Nat Rev Cancer 4, 143–153. discovery rate: a practical and powerful approach to 26 Bolund ACS, Starnawska A, Miller MR, Schlunssen V, multiple testing. J R Stat Soc Series B Stat Methodol Backer V, Borglum AD, Christensen K, Tan Q, 57, 289–300. Christiansen L & Sigsgaard T (2017) Lung function 38 Boyle EA, Li YI & Pritchard JK (2017) An expanded discordance in monozygotic twins and associated view of complex traits: from polygenic to omnigenic. differences in blood DNA methylation. Clin Epigenetics Cell 169, 1177–1186. 9, 132. 39 Devarakonda S, Morgensztern D & Govindan R (2015) 27 Lee KW & Pausova Z (2013) Cigarette smoking and Genomic alterations in lung adenocarcinoma. Lancet DNA methylation. Front Genet 4, 132. Oncol 16, e342–e351. 28 Shui IM, Wong CJ, Zhao S, Kolb S, Ebot EM, 40 Cancer Genome Atlas Research Network (2012) Geybels MS, Rubicz R, Wright JL, Lin DW, Klotzle B Comprehensive genomic characterization of squamous et al. (2016) Prostate tumor DNA methylation is cell lung cancers. Nature 489, 519–525. associated with cigarette smoking and adverse prostate 41 Campbell JD, Alexandrov A, Kim J, Wala J, Berger cancer outcomes. Cancer 122, 2168–2177. AH, Pedamallu CS, Shukla SA, Guo G, Brooks AN, 29 Hecht SS (2012) Lung carcinogenesis by tobacco Murray BA et al. (2016) Distinct patterns of somatic smoke. Int J Cancer 131, 2724–2732. genome alterations in lung adenocarcinomas and 30 Wingo PA, Ries LA, Giovino GA, Miller DS, squamous cell carcinomas. Nat Genet 48, 607–616. Rosenberg HM, Shopland DR, Thun MJ & Edwards 42 Ando M, Saito Y, Xu G, Bui NQ, Medetgul-Ernar K, BK (1999) Annual report to the nation on the status of Pu M, Fisch K, Ren S, Sakai A, Fukusumi T et al. cancer, 1973–1996, with a special section on lung cancer (2019) Chromatin dysregulation and DNA methylation and tobacco smoking. J Natl Cancer Inst 91, 675–690. at transcription start sites associated with 31 Asomaning K, Miller DP, Liu G, Wain JC, Lynch TJ, Su transcriptional repression in cancers. Nat Commun 10, L & Christiani DC (2008) Second hand smoke, age of 2188. exposure and lung cancer risk. Lung Cancer 61, 13–20. 43 Baylin SB & Jones PA (2011) A decade of exploring the 32 Sandoval J, Mendez-Gonzalez J, Nadal E, Chen G, cancer epigenome – biological and translational Carmona FJ, Sayols S, Moran S, Heyn H, Vizoso M, implications. Nat Rev Cancer 11, 726–734. Gomez A et al. (2013) A prognostic DNA methylation 44 Wagner JR, Busche S, Ge B, Kwan T, Pastinen T & signature for stage I non-small-cell lung cancer. J Clin Blanchette M (2014) The relationship between DNA Oncol 31, 4140–4147. methylation, genetic and expression inter-individual 33 Bjaanaes MM, Fleischer T, Halvorsen AR, Daunay A, variation in untransformed human fibroblasts. Genome Busato F, Solberg S, Jorgensen L, Kure E, Edvardsen Biol 15, R37. H, Borresen-Dale AL et al. (2016) Genome-wide DNA 45 van Eijk KR, de Jong S, Boks MP, Langeveld T, Colas methylation analyses in lung adenocarcinomas: F, Veldink JH, de Kovel CG, Janson E, Strengman E, association with EGFR, KRAS and TP53 mutation Langfelder P et al. (2012) Genetic analysis of DNA status, gene expression and prognosis. Mol Oncol 10, methylation and gene expression levels in whole blood 330–343. of healthy human subjects. BMC Genom 13, 636. 34 Karlsson A, Jonsson M, Lauss M, Brunnstrom H, 46 Zhang D, Cheng L, Badner JA, Chen C, Chen Q, Luo Jonsson P, Borg A, Jonsson G, Ringner M, Planck M W, Craig DW, Redman M, Gershon ES & Liu C & Staaf J (2014) Genome-wide DNA methylation (2010) Genetic control of individual differences in gene- analysis of lung carcinoma reveals one neuroendocrine specific methylation in human brain. Am J Hum Genet and four adenocarcinoma epitypes associated with 86, 411–419. patient outcome. Clin Cancer Res 20, 6127–6140. 47 Maurano MT, Wang H, John S, Shafer A, Canfield T, 35 Chen YA, Lemire M, Choufani S, Butcher DT, Lee K & Stamatoyannopoulos JA (2015) Role of DNA Grafodatskaya D, Zanke BW, Gallinger S, Hudson TJ methylation in modulating transcription factor & Weksberg R (2013) Discovery of cross-reactive occupancy. Cell Rep 12, 1184–1195. probes and polymorphic CpGs in the Illumina Infinium 48 Iurato S, Carrillo-Roa T, Arloth J, Czamara D, Diener- HumanMethylation450 microarray. Epigenetics 8, 203– H€olzl L, Lange J, M€ uller-Myhsok B, Binder EB & 209. Erhardt A (2017) DNA Methylation signatures in panic 36 Marabita F, Almgren M, Lindholm ME, Ruhrmann S, disorder. Transl Psychiatry 7, 1287. Fagerstrom-Billai F, Jagodic M, Sundberg CJ, Ekstrom 49 Banfield E, Brown AL, Peckham EC, Rednam SP, TJ, Teschendorff AE, Tegner J et al. (2013) An Murray J, Okcu MF, Mitchell LE, Chintagumpala 2772 Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. X. Ji et al. Epigenetic–smoking interaction analysis of TRIM27 MM, Lau CC, Scheurer ME et al. (2016) Exploratory 61 Lee IW, Kuan LC, Lin CH, Pan HA, Hsu CC, Tsai analysis of ERCC2 DNA methylation in survival YC, Kuo PL & Teng YN (2008) Association of USP26 among pediatric medulloblastoma patients. Cancer haplotypes in men in Taiwan, China with severe Epidemiol 44, 161–166. spermatogenic defect. Asian J Androl 10, 896–904. 50 McNab FW, Rajsbaum R, Stoye JP & O’Garra A (2011) Tripartite-motif proteins and innate immune regulation. Curr Opin Immunol 23, 46–56. Supporting information 51 Turkson J (2004) STAT proteins as novel targets for Additional supporting information may be found cancer drug discovery. Expert Opin Ther Targets 8, online in the Supporting Information section at the end 409–422. of the article. 52 Wang CY, Chao TT, Tai WT, Chang FY, Su WP, Fig. S1. Quality control processes for DNA methyla- Chen YL, Chen PT, Weng CY, Yuan A, Shiau CW tion chip data. et al. (2014) Signal transducer and activator of Fig. S2. Methylation–smoking interaction on survival transcription 3 as molecular therapy for non-small-cell of LUSC patients excluding never smokers. lung cancer. J Thorac Oncol 9, 488–496. Fig. S3. Kaplan–Meier overall survival (OS) curves of 53 Huang YT, Lin X, Liu Y, Chirieac LR, McGovern R, LUAD, LUSC and overall NSCLC patients excluding Wain J, Heist R, Skaug V, Zienolddiny S, Haugen A never smokers. (A,B) LUAD patients, (C,D) LUSC et al. (2011) Cigarette smoking increases copy number alterations in nonsmall-cell lung cancer. Proc Natl Acad patients and (E,F) overall patients. Hazard ratio (HR) Sci USA 108, 16345–16350. and P value were derived from a Cox proportional 54 Barbieri SS, Ruggiero L, Tremoli E & Weksler BB hazards regression model adjusted for age, sex, smok- (2008) Suppressing PTEN activity by tobacco smoke ing status, clinical stage, and study centre. PHeterogeneity plus interleukin-1beta modulates dissociation of VE- was used to evaluate heterogeneity of HRs across cadherin/beta-catenin complexes in endothelium. groups. Arterioscler Thromb Vasc Biol 28, 732–738. Fig. S4. Kaplan–Meier overall survival (OS) curves of 55 Qian X, Li X, Shi Z, Xia Y, Cai Q, Xu D, Tan L, Du LUAD, LUSC and overall NSCLC patients. (A,B) L, Zheng Y, Zhao D et al. (2019) PTEN suppresses LUAD patients, (C,D) LUSC patients and (E,F) over- glycolysis by dephosphorylating and inhibiting all patients. Hazard ratio (HR) and P value were autophosphorylated PGK1. Mol Cell 76, 516–527.e7. derived from a Cox proportional hazards regression 56 Cumberbatch M, Tang X, Beran G, Eckersley S, Wang model adjusted for age, sex, clinical stage, pack-year X, Ellston RP, Dearden S, Cosulich S, Smith PD, of smoking and study centre. PTrend was used to evalu- Behrens C et al. (2014) Identification of a subset of ate trend of HRs across groups. human non-small cell lung cancer patients with high Fig. S5. Kaplan–Meier overall survival (OS) curves of PI3Kbeta and low PTEN expression, more prevalent in TCGA cases by low or high TRIM27 expression. The squamous cell carcinoma. Clin Cancer Res 20, 595–603. gene expression divided into low and high groups by 57 Iliopoulos D, Jaeger SA, Hirsch HA, Bulyk ML & median value (10.26). Hazard ratio (HR) and P value Struhl K (2010) STAT3 activation of miR-21 and miR- were derived from a Cox proportional hazards regres- 181b-1 via PTEN and CYLD are part of the epigenetic sion model adjusted for age, sex, smoking status, clini- switch linking inflammation to cancer. Mol Cell 39, cal stage, and study centre. 493–506. Fig. S6. Genome-wide methylation transcription analy- 58 Su CC, Li SY, Yen YC, Nian JH, Liang WG & Yang sis of LUSC patients from the TCGA cohort. (A) Cir- JJ (2013) Mechanism of two novel human GJC3 missense mutations in causing non-syndromic hearing cos plot of genome-wide gene expression. For plots in loss. Cell Biochem Biophys 66, 277–286. B–D, left panels show correlation of (B) USP26, (C) 59 Okada Y, Hirota T, Kamatani Y, Takahashi A, GJC3 or (D) NAALAD2 expression (X-axis) with Ohmiya H, Kumasaka N, Higasa K, Yamaguchi- methylation level at cg05293407TRIM27 (Y-axis). Right Kabata Y, Hosono N, Nalls MA et al. (2011) panels show Kaplan–Meier survival plots of gene Identification of nine novel loci associated with white expression divided into low and high groups by med- blood cell subtypes in a Japanese population. PLoS ian value. Genet 7, e1002067. Fig. S7. Distribution of maximum value of all 311891 60 Aston KI, Krausz C, Laface I, Ruiz-Castane E & CpG probes in non-small-cell lung cancer (NSCLC) Carrell DT (2010) Evaluation of 172 candidate patients. polymorphisms for association with oligozoospermia or Table S1. Demographic and clinical characteristics of azoospermia in a large cohort of men of European early-stage NSCLC patients with gene expression data descent. Hum Reprod 25, 1383–1397. derived from TCGA. Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 2773 Epigenetic–smoking interaction analysis of TRIM27 X. Ji et al. Table S2. Annotation information for 96 CpG probes Table S6. Comparison of smoking-related characteris- located in TRIM27. tics of former and current smokers between early-stage Table S3. Results of association analysis of 96 DNA LUAD and LUSC. methylation probes of TRIM27 in LUAD samples. Table S7. Results of genome-wide methylation tran- Table S4. Results of association analysis of 96 CpG scription analysis of 29 genes significantly associated probes of TRIM27 in LUSC samples. with cg05293407 in TCGA LUSC samples. Table S5. Results of proportional hazards test for 96 CpG probes of TRIM27 in LUSC samples. 2774 Molecular Oncology 14 (2020) 2759–2774 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd.