We have selected a total of 89 treatment naïve lung adenocarcinoma (LUAD) or lung squamous cell carcinoma (LUSC) patients from the Cancer Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College who underwent definitive surgical resection before adjuvant therapy, including chemotherapy or radiotherapy. This study was approved by the Cancer Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, and performed in accordance with the Declaration of Helsinki Principles. All these samples were fresh frozen tissues which were under low temperature conditions (at − 80 °C). After obtaining informed consents, tumor tissues and their matched control were obtained for WES and DNAm profiling. To avoid the contamination of tumor tissue, all the matched normal tissues were collected at lobectomy edge. All samples had been subjected to pathology review for histological subtyping. The detailed clinical characteristics of these 89 Chinese Han population samples are summarized in Additional file 1: Table S1.

Sequencing protocol: DNA libraries for tumor and their matched control samples were prepared with standard protocol using MGIEasy Exome Capture V4 Probe Set capture kit (cat. No: 1000007745, https://en.mgitech.cn/article/detail/v4.html) with the capture region size 36 Mb. Pair-end sequencing (2 × 100 bp) was performed on BGI-Seq 500 instruments. Data processing: Alignment: The raw paired end reads were mapped to the human reference genome (hg19) using bwa-mem (version 0.7.16 with –M option: mark shorter split hits as secondary and the remaining setting was at default). Samtools v1.3.1 was used to sort and merge bam files from the same patient sequenced from different lanes. PCR duplicate read pairs were identified using biobambam (v.0.0.148). The quality control (all the sample QC files were available at https://drive.google.com/open?id=1HggApA8homvpF4xD2YOI3EQ2HsY3hS4S) was generated with FastQC (v0.11.8) and the post-alignment QC metrics information was shown in Additional file 1: Table S2. Variants calling: Variants calling was performed using a modified version of DKFZ-pipeline based on samtools mpileup and bcftools version 0.1.19 (pcawg-dkfz-workflow). Briefly, variants in the tumor sample were initially and used as query in the control sample. The raw calls were then annotated with various publicly available databases, including 1000 Genomes variants, ESP exon database, single-nucleotide polymorphism database (dbSNP), ExAC v.0.3.1 (non-TCGA variants), repeats and other elements. The functional consequence of the variants was predicted using Annovar [22] with UCSC Refseq annotations, followed by assessment of the variants in terms of their confidence and then classified into somatic or non-somatic calls. Only highly confident somatic variants with the following filtering criteria: Read Depth > =10, AF > =5%, Number of reads indicating mutation > = 3, were used for further analysis. TMB level is defined by two ways: one is as number of nonsynonymous coding somatic mutations (NOMs) per tumor, including single nucleotide variation (SNVs) and short insertion/deletion polymorphism (INDELs); the other is the number of mutations is proportion to the size of UCSC Refseq annotations (33.4 Mb). R/Biocondcutor package “maftools” [23] was used for visualization and summarization of MAF files from this study. TCGA WES somatic mutations: Confident somatic mutation calls derived from the WES data of the LUAD and LUSC cohorts were directly downloaded from the TCGA GDC Data Portal (https://portal.gdc.cancer.gov).

Mutational signature analysis was conducted using the deconstructSigs package v1.8.0 [24]. All the detected somatic mutations including synonymous in the cohort were imported for signature analysis. In details, the frequency of 96 possible mutation types in trinucleotide context of each patient were first calculated in somatic mutation dataset. Normalization were then processed, according to the number of times each trinucleotide context is observed in our capture region. Finally, the weights of 30 known cancer mutation signature in COSMIC (https://cancer.sanger.ac.uk/cosmic/signatures) were generated by linear regression based on normalized frequency of each possible mutation types. Each weight indicates that how strongly can mutation signature influence the patient. Hierarchy cluster based on mutation signatures’ weights among patients were drawn by R package ‘pheatmap’ [25].

Five hundred nanogram of genomic DNA from each sample was bisulfite converted using the EZ DNA Methylation Kit (Zymo Research, Irvine, CA) and then analyzed on Infinium HumanMethylation 850 K EPIC BeadChip (Illumina, San Diego, CA) following the manufacturer’s instructions. The array features more than 850,000 methylation sites covering 96% CpG islands and 99% gene’s promoters. Raw data were analyzed using the “ChAMP” (Chip Analysis Methylation Pipeline for Illumina HumanMethylation450 and EPIC) package in R [26, 27] and all relevant parameters are default values. The differential methylated probe (DMP) of each sample was identified by the beta value of cancer and matched normal tissue with Benjamini-Hochberg (BH)-adjusted P-value < 0.05. R/Biocondcutor package “ConsensusClusterPlus” [28] was used for consensus clustering of Ilumina EPIC data. Bumphunter algorithms was applied to estimate regions for which a genomic profile deviates from its baseline value. Originally implemented to detect differentially methylated genomic regions between tumors and normal controls. By default, the progress of differential methylation region (DMR) finding was done on normalized beta value. The detected DMR and estimated P value (0.05 as cutoff value) was returned.

The R/Bioconductor package ‘conumee’ [29] was employed to calculate CNAs based on the intensities generated using the EPIC array (using default settings). GISTIC [30] was then used to identified common deleted/amplified regions/genes (using default parameters). GISTIC is a tool that identifies genes targeted by somatic copy-number alterations (SCNAs) that trigger cancer growth. By classifying SCNA profiles as arm-level and focal alterations, this tool calculates the background rates of each category as well as delineates the boundaries of the SCNA regions. Aneuploidy score (AS) was calculated as is reported [31, 32], and the scores of each arm are − 1 if lost, + 1 if gained, 0 if non-aneuploid, and “NA” otherwise. For gene enrichment analysis, the function annotation tool from the DAVID website (https://david-d.ncifcrf.gov/) was used.

All statistical tests were conducted in R version 3.4.1 (The R Foundation for Statistical Computing, Austria). Unpaired t test was performed to evaluate the significance of TMB value between two groups (smoking: non-smoking, TP53+: TP53- and Chinese: TCGA LUAD/LUSC). Pearson’s correlation coefficient was calculated to evaluate the strength of correlation between DNA methylation and TMB levels. * stands for P value < 0.05. ** stands for P value < 0.01. *** stands for P value < 0.001.

These 89 patients included in this study were consist of 65 LUAD as well as 24 LUSC patients. From WES data analysis, only high confidence nonsynonymous somatic mutations (Tumor DP > =10, AF > =5%, NO. of reads indicating mutation > = 3) were processed for TMB assessment. The mean coverage is achieved at 167×, 161× in tumor samples and normal samples, respectively. More than 90% of targeted regions with coverage > 10× were found in 87/89 pair samples. TMB distribution showed a median number of 104 NOMs per tumor, ranging from 37 to 465 (Fig. 1a). Consistent with the approach of the clinical trial CheckMate 026 [33], we classified our cohort by high (139–465), medium (83–136) and low (37–82) NOMs or low (1.1–2.5), medium (2.5–4.1), and high (4.2–13.9) mutations/Mb. In order to further explore the relationship between DNAm and TMB, 13 relatively low (37–79 mutations or 1.1–2.4 mutations/Mb) and 16 relatively high (252–465 mutations or 7.5–13.9 mutations/Mb) TMB samples were selected for subsequent methylation level detection. Due to the insufficient amount of DNA after the WES experiment, these samples were not selected successively. Unless specifically mentioned, the high or low TMB group in the following text represents the relatively high TMB group and the relatively low TMB group.Fig1Fig. 1

The methylome of relatively high TMB lung cancer is unique, and many DMRs are recurrent. a The NOMs for every patient (represented by the x-axis). Red/green lines indicate the high/low TMB cutoff in our cohort; Bar plot (b) and MDS analysis (c) of all CpG sites; d Identification of differences in DNAm between high TMB and low TMB groups. e Scatter plot between methylation changes (delta-beta value, high TMB tumors vs. controls) and corresponding -log10(BH adj. P-value) for total assessed 865,918 sites was shown. CpG sites with deltabeta < 0.2 and -log10(BH adj. P-value) < 2 were defined as MVPs. The upper square indicates hypermethylated MVPs, and the bottom square indicates hypomethylated MVPs compared with controls; f The comparation of differently methylation sites and TMB of NSCLCs (LUAD/LUSC) in our cohort; g The comparation of differently methylation sites and TMB of NSCLCs (LUAD/LUSC) in TCGA dataset; h Consensus clustering of the DNA methylation reveals high and low TMB lung cancer groups of DNA methylation. 293 informative probes with strict screening parameters (s.d. > 0.2 between high and low TMB group, s.d. < 0.1 in high or low TMB group, |delta beta| > 0.2, BH adjusted P value < 0.05) were used for the consensus clustering

CpG hypomethylation status has been reported to be related to genetic instabilities, and global hypomethylation in tumor indicates more genomic instabilities [35]. We checked the copy number variation (CNV) in high TMB and low TMB group using aneuploidy score (AS) and found high TMB lung cancers possess more structural variation of CNV, while low TMB ones appeared to retain more stable genomic structural profile (Fig. 2a, Additional file 1: Table S6). The analysis results of TCGA database are consistent with our cohort study (r = 0.18, P value =1 × 10^− 8, Pearson correlation analysis) as is shown in Additional file 1: Table S7 and Additional file 2: Figure S4. Compared to the low TMB group, the high TMB NSCLC group exhibited more genomic deletions and amplifications (Fig. 2b), particularly a gain in chromosome arm 3q (especially 3q26) and a loss of chromosome 3p (especially 3p12). Frequent localized amplifications within chromosomal regions 8q24, 12p11, and 15q11 loci and deletions within 8p22 and 9p23 were also detected. A total of 1237 genes (Additional file 1: Table S8) were significantly mapped to these amplified regions, whereas no genes could be significantly mapped to the deleted regions (whole chromosomal arm deletions were excluded from the analysis) (Fig. 2c). Several of these recurrent CNAs exhibited high chromosomal instability, possibly lead to TMB value increasing. The 1237 genes in CNA regions were also assessed in terms of Gene Ontology enrichment by DAVID, which revealed that the pathways of Jak-STAT signaling (hsa04630) and cytokine-cytokine receptor interaction (hsa04060) were highly represented in Fig. 2d.Fig2Fig. 2

Numerous copy number amplifications characterize high TMB cancers. a CNA profiles of high TMB and low TMB group: heatmaps of aneuploidy score calculated via the intensities of the EPIC array (each tumor versus average normal). The scores of each arm are − 1 if lost, + 1 if gained, 0 if non-aneuploid, and “NA” otherwise; b Amplifications: q values of amplifications of all tumors of high−/low- TMB lung cancer tumors. Deletions: q values of deletions of all tumors of high−/low- TMB lung cancer tumors; c Confirmed number of genes that map to significantly amplified or deleted regions; d GO enrichment analysis showing the enriched pathways of amplified and deleted genes in high TMB lung cancers

In high TMB group, more significant DMPs was found and Fig. 3a shows the top 6 DMPs (cg16732616/DMRTA2, cg26521404/HOXA9, cg20326647/intergenic region, cg02443967/TLL2, cg09792881/DMRTA2 and cg16928066/EMX1) as representives. We also explored the distribution of the DMPs and found hypermethylated DMPs were located closer to transcription start site (TSS), whereas the hypomethylated DMPs was shifted slightly to upstream of TSS (Fig. 3b). We focused on the MVPs with No. > 3 at the promoter region referring to TSS1,500, TSS200, 5′-UTR, and 1stExon to discover significant differential methylated gene and found 1666 genes, in which HOX family genes (26 out of 39 [36, 37]) were most effected (Additional file 2: Figure S5). In order to further exclude inappropriate genes caused by the number of samples, the same analysis was carried out in the TCGA NSCLC database, and Venn analysis (Fig. 3c, Additional file 1: Table S9) showed that there were 437 genes associated with the state of high TMB. The heatmap plot (Fig. 3d) analyzed all the 8703 probes from the 850 k chip related to these 437 genes, and the results show that they are significantly different in the high TMB group. The same analysis was performed on 4916 probes from 450 k chips in the TCGA database, and the results (Additional file 2: Figure S6) were consistent with our cohort study. To further analyze the relationship between these 437 genes and lung cancer, we used DisGeNET [38], a database of gene-disease associations, to analyze the network of these genes, and found that there were 99 genes, related to “Neoplastic Process” of lung (Additional file 1: Table S10).Fig3Fig. 3

The relationship of HOX gene methylation status and TMB. a Top 6 differential methylation sites in hi-TMB comparing with low TMB group. cg16732616/DMRTA2, cg26521404/HOXA9, cg20326647/intergenic region, cg02443967/TLL2, cg09792881/DMRTA2 and cg16928066/EMX1 were significantly methylated in high TMB tumor tissues; b In hi-TMB lung cancer, CpGs that are hypomethylated are more likely to be found immediately upstream of the TSS and within the 1st exon, CpGs are hypermethylated (P value < 0.01, two-tailed Wilcoxon rank-sum test); c Venn diagram analysis revealed 437 genes associated with high TMB; d The heatmap of the all methylation probes related to 437 genes in high TMB, low TMB NSCLCs samples and the matched normal controls

The TMB distribution of Chinese NSCLCs has not been well reported in the literature, therefore its description may provide insights for pharmaceutical companies or diagnostic industry to adjust their commercial strategy in China. Recent studies had demonstrated that loss of TP53 function increased genomic instability [39, 40]. We further investigate the mechanism of these differences between these two populations based on genetic alterations. An important driver gene of Chinese NSCLCs, EGFR mutations, which are closely related to the efficacy of molecularly targeted therapy (EGFR TKIs), have been reported to negatively correlate with TMB value [41, 42]. Heatmap plot shows that frequently mutated genes such as TP53 gene, which tends to been enriched in high TMB group (top 30 samples, range: 139–465 NOMs) in lung cancer; EGFR mutants in low TMB (bottom 30 samples, range: 37–82 NOMs), and patients with co-existence of TP53 and EGFR mutations in the intermediate TMB level (median 29 samples, range: 83–136 NOMs) (Fig. 4a). Many disease-causing genes in cancer are co-occurring or show strong exclusiveness in their mutation pattern with high TMB. In our study, gene set TP53, CSMD3, GXYLT1, PPP1R13L and TTN shows a strong co-occurrence and gene set EGFR, TTN, MUC2 and HERC2 shows a strong exclusiveness in high TMB group (Fig. 4b). It was been confirmed in our study that the high TMB samples was mostly LUSC with smoking habit. Our study confirm that smoking was also a key factor associated with TMB (Fig. 4c). We evaluated 30 known mutational signatures for different carcinogens in COSMIC database, including UV light or tobacco by calculating the frequency of specific mutation types in trinucleotide [43]. Consistent with previous findings, we observed that high TMB patients displayed distinct mutation signatures comparing to low TMB patients as is shown in Fig. 4d. Signature 4 was the dominant mutation pattern in high TMB patients with smoking history, while high TMB patients without smoking history contained relatively stronger mutation pattern in Signature 3. Signature 3 and signature 12 occurred simultaneously in low TMB patients, regardless of smoking status. Since signature 4 is a well-known tobacco-related signature characterized by transcriptional strand bias in C > A mutations, it matches the phenotype of smoking-history among high TMB patients. The failure of DNA double-strand break-repair in homologous recombination indicated by signature 3 may confer high mutation ability to patients without smoking history. Regarding to low TMB patients, efforts are needed to investigate the etiology of the strong signal in signature 12 with T > C substitutions.Fig4Fig. 4

Comparative analysis between Chinese and TCGA lung cancer populations. a The top 13 genes with most frequent mutations in our cohort with the decrease in NOMs; b Somatic interactions in our cohort. Such mutually exclusive or co-occurring set of genes can be detected using the somaticInteractions function in R/Bioconductor package ‘maftools’, which performs pair-wise fisher’s exact test to detect such significant pair of genes; c Comparison of TMB levels between smoking and non-smoking group. Unpaired t test P value =0.00015, Smoking group: mean = 218, Non-smoking group: mean = 101.40; d Heatmap plot to interpret the possible associations of mutation signature and TMB classification. Generally, high TMB patients with smoking history display a strong signal on signature 4 (the known signature associated with cigarette). Another high TMB group without smoking history display a dominant weight on signature 3 (The signature probably caused by failure of DNA double-strand break-repair in homologous recombination). Signature 3 and signature 12 occurred simultaneously in low TMB patients, regardless of smoking status; e The differential patterns of mutation between Chinese lung cancer population and TCGA LUAD/LUSC; f The comparation of Chinese LUAD/LUSC and TCGA LUSC/LUAD NOMs; g The relationship of TP53 and NOMs in Chinese NSCLSs and TCGA NSCLCs