Cancer morbidity and mortality remain a major challenge to public health in China, with over two million cancer deaths per year in China1,2. In recent years, precision oncology has enabled individual diagnosis, prognosis and treatment based on increasingly accurate and high-resolution molecular stratification of cancers, largely focused on genomically targeted therapies3,4. Notably, patient ethnicity can also be a factor in cancer diagnostics and treatment, since differences in cancer mutations exist between tumor patient populations of various ethnicities5-7.
To explore the genomic landscape of Chinese patients with solid tumors as encountered in clinical practice, we collected a total of 11,553 tumor specimens and matched peripheral blood specimens from 11,553 individuals encompassing 25 principal tumor types and >100 tumor subtypes. After excluding samples (n = 1359) with insufficient tumor content or DNA yield or subsequent technical failure (Supplementary Fig. 1a), we successfully sequenced 10,194 (88%) tumor samples. Summaries of the clinical characteristics of the patients’ specimens included in the study, and the median sequencing target coverage of samples are presented in Supplementary Table 1-2 and Supplementary Fig. 2-3. A total of 31 ethnicities were presented in our cohort with Han being the most frequent (92%, 9382/10,194). The majority of patients in this study were from eastern and southern provinces in China ("East China" and "South China" in Wikipedia) (41% and 29%, respectively). In terms of tumor stage, 55% (5,652/10,194) of patients had advanced-stage cancers (stage III/IV) while 35% (3,579/10,194) had early-stage cancers (pre-cancers or stage I/II). In our cohort, majorities (76%) of patients were treatment-naïve and the ratio of patients who had received treatment was 16% (Supplementary Fig. 1b). The major tumor types were non-small cell lung cancer (NSCLC; 20%), colorectal carcinoma (CRC; 12%), liver hepatocellular carcinoma (LIHC; 11%), gastric cancer (GC; 8%), esophageal carcinoma (ESCA; 6%), soft tissue sarcoma (STS; 6%), intrahepatic cholangiocarcinoma (ICC; 5%), pancreatic cancer (PAC; 5%), extrahepatic cholangiocarcinoma (ECC; 3%), and breast carcinoma (BRCA; 3%) (Fig. 1a). In general, the distribution of these predominant tumor types such as liver cancer (LIHC, ICC, and ECC) and lung cancer (NSCLC and SCLC) represented the distribution of tumors encountered in clinical practice in China1.
Based on an NGS-based assay with a validated 450-gene panel8, we detected 80,703 single nucleotide variants (SNVs) and insertions and deletions (indels), 19,192 truncations, 17,779 gene amplifications, 1,688 gene homozygous deletions, and 3,111 gene fusions/ rearrangements in the 10,194 cases. We only focused on somatic alterations within tumors in this study, and no germline genetic data. Analysis of significantly mutated cancer related genes in solid tumors found the most frequently altered genes to be TP53 (58% of cases), KRAS (18%), TERT (14%), EGFR (13%), APC (13%), CDKN2A (12%), and PIK3CA (11%). The most common mutations were KRASG12, EGFRL858, and TP53R273 (Fig. 1b; Supplementary Table 3); of note, EGFR and KRAS represented obviously pairwise co-occurring alterations with SNV/indel and copy number variation (CNV) (Supplementary Fig. 4). Subsequent analysis of CNV showed high frequencies of CDKN2A/B deletion, SMAD4 deletion, ERBB2 amplification, EGFR amplification and MYC amplification in metastatic samples, and chromosome 11q13.3 (CCND1/FGF3/FGF4/FGF19) amplification in primary samples at the pan-cancer level; meanwhile ERBB2 amplification and chromosome 11q13 amplification were respectively enriched in breast cancer (BRCA) (24% vs. 2%; FDR = 7.645E-105) and ESCA (43% vs. 4%; FDR = 3.553E-301), compared to other tumor types (Supplementary Fig. 5; Supplementary Table 4-5). In addition, we sought to investigate the features of gene fusions in solid tumor and identified a total of 513 fusion events, including 31 driver genes in our cohort. As shown in Fig. 1c, fusion events in genes such as ALK (n = 139), ROS1 (n = 51), RET (n = 50), FGFR2/3 (n = 50), NTRK1/3 (n = 30) and BRAF (n = 12) occurred widely across tumor types, while others such as EWSR1 and TFE3 were enriched in certain tumor types (sarcomas [soft tissue sarcoma or STS, and bone sarcoma] and KIRC, respectively); PRKACA fusions were only detected in a specific tumor type (LIHC, subsequent diagnosis as fibrolamellar hepatocellular carcinoma [FL-HCC]). Moreover, new fusion partners for driver genes were identified; multiple novel fusions were seen in kinase genes, including GRIK2-ROS1, PARP12-BRAF, KIF13B-MET, and LRRC28-NTRK3; novel fused exons were seen in KIF5B-ALK and EML4-ALK compared to the Quiver database (http://quiver.archerdx.com/) (Fig. 1d, Supplementary Table 6, and Supplementary Fig. 6-7).
To reveal further somatic alterations associated with clinical characteristics in Chinese cancer patients, we implemented an integrative analysis across the tumor-type distribution of genomic profile and six clinical features including age, gender, tumor stage, smoking history (only in NSCLC, SCLC and HNC), treatment and sample type (primary vs. metastatic/recurrent) (Supplementary Table 7; Fig. 2a-b). In general, clinical feature-associated genomic differences were observed distributed in CRC and NSCLC. In CRC, the numbers of different mutated genes were respectively 270 and 100 in younger and early-stage patients as compared to older and advanced-stage patients, which could be consistent with the significantly high ratio of hypermutated subtypes, such as microsatellite instability-high (MSI-H) and POLE-associated CRC (with microsatellite stability [MSS], high mutation burden and an inactive POLE mutation) in younger and early-stage CRC (Supplementary Fig. 8, FDR < 0.05). In NSCLC, the mutated frequency of genes was markedly affected by gender and smoking history; of note, gender and smoking history were not independent factors in our cohort, because the majority of nonsmokers were female. It was found that female nonsmokers with early-stage NSCLC harbored more mutations in EGFR; while male smokers with advanced-stage NSCLC were characterized by more mutations in TP53, CDKN2A, PIK3CA and KRAS, consistent with recent reports9. Moreover, younger female gastric cancer patients had more CDH1 mutations; in contrast, older gastric cancer patients tended to have more mutations in TP53, NOTCH1 and FAT4. In addition, younger LIHC, KIRC and bone sarcoma patients harbored respectively TP53, TFE3 and VEGFA mutations, while older LIHC, HNC and STS patients had respectively CTNNB1, TERT and TP53 mutations (Fig. 2b, FDR < 0.05).
To assess the characteristics of cancer genomes from Chinese patients in a global context, we made a comparison of genomic alterations with the largest published cancer genomic study: the Memorial Sloan Kettering Cancer Center (MSK) IMPACT study10, including 10,366 cases, mostly advanced cancer specimens. A total of 266 correspond genes were compared in 15 comparable advanced-stage tumor types between OM cohort (aOM, n = 4699) and MSK cohort (n = 6161). To limit the bias of comparisons, we subdivided NSCLC of the two cohorts into lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC). Overall, only 20 tumor type: gene pairs presented significant differences in the frequency of gene variants between the aOM cohort and the MSK cohort (FDR < 0.05) (Fig. 2c-d and Supplementary Table 8), suggesting frequencies of most significantly mutated genes and the tumor-type distribution in aOM cohort were highly consistent with in the MSK cohort, such as CRC:APC (72% vs. 77%, FDR = 0.27) and SCLC:RB1 (81% vs. 72%, FDR = 1). The significant differences between two cohorts were mainly found in lung adenocarcinoma and hepatobiliary tumors, such as LUAD: EGFR, ICC: KRAS and ECC: IDH1. At the same time variants of TP53 gene exhibited widely differences in multiple tumor types. Furthermore, cohorts were independently and significantly related to most of different tumor type: gene pairs in logistic regression analysis of several clinical features, including primary/metastasis/recurrent tumor specimens, sampling method, gender and smoke (Supplementary Table 8). Moreover, several gene fusions and CNVs also showed differences between the aOM cohort and the MSK cohort. For example, ALK fusions in LUAD (Supplementary Table 6) occurred more frequently in the aOM cohort than that in the MSK cohort (9% vs. 3%, P = 3.913E-10); while the incidences of ROS1 fusions (3% vs. 2%, P = 0.38) and RET fusions (2% vs. 2%, P = 0.65) in LUAD were similar between the two cohorts.
To further confirm the similarities and differences between the OM and MSK studies observed in advanced cancers, we also compared the aOM data with genomic data of advanced-stage cases from The Cancer Genome Atlas studies (aTCGA). Because of heterogeneous methodologies (including detecting platform, algorithm and report criteria of variants), substitutions, indels and truncations mutations were considered in the comparison. In 9 comparable tumor types and 266 genes, we identified a total of 26 tumor type: gene pairs with significant differences between the aOM cohort (n = 3505) and the aTCGA cohort (n = 2065) (FDR < 0.05) (Fig. 2e-f and Supplementary Table 9), of which 7 different tumor type: gene pairs presented consistently changed trends with those in the comparison between the aOM cohort and the MSK cohort, including higher frequencies of LIHC: TP53, BRCA: TP53, LUAD: EGFR and lower frequencies of LUAD: KEAP1, LUAD: KRAS, HNC: TP53, HNC: PIK3CA in the aOM cohort, compared with other two cohorts. Altogether, these multiple comparisons revealed at the greatest extent the similarity and distinctive of genomic alteration across these cohorts.
In addition to targeted therapy, the recent clinical success of immune checkpoint blockade11-14 makes the comparison of immunotherapy related mutations and signatures across cancers from patients in different countries another important question. Hence, we analyzed the distribution of tumor mutational burden (TMB) within tumor types. Even though an algorithm to evaluate TMB in routine clinical practice has not yet reached a consensus15, an individual TMB has been shown to predict patient outcome after immunotherapy11-14. Here, we identified as TMB high (TMB-H) and TMB low (TMB-L) according to the TMB-high status definition from KEYNOTE-158 study (the value ≥ 10 or not)14. As shown in Fig. 3a and Supplementary Table 10, median TMB values in most tumor types in the aOM cohort were different compared with the MSK cohort (Supplementary Fig. 9). Overall, the whole pattern of TMB distribution in the aOM cohort was similar to that in the MSK cohort, characterized by a “tail” that includes 119 samples with TMB ≥ 40 (Fig. 3b). We further analyzed the distribution of 186 samples harboring MSI-H in our cohort and found that the overall proportion of patients with MSI-H was 2% and was mainly in CRC (55%, 102/186) (Fig. 3c). Furthermore, because MSI-H and TMB-H have recently been recognized as biomarkers for response to immune checkpoint blockades (anti-PD-1/PD-L1)11,14, we evaluated the combined association of TMB and MSI with PD-L1 expression evaluated by immunohistochemical (IHC) staining in 2,723 tumors of OM cohort. The overall proportion of samples with at least one of MSI-H, TMB-H, or PD-L1 positive was 30.3% (891/2,723) and the proportion of such samples is highest in SCLC (48%; 24/50), followed by NSCLC (46%; 298/648), and ESCA (34%; 80/235) (Fig. 3d, Supplementary Fig. 10), suggesting the possibility of a high proportion of Chinese patients with lung cancer benefitting from immunotherapy.
In addition, recent evidence has suggested somatic amplification in the gene for programmed cell death ligand 1 (PD-L1/CD274) as a response biomarker to immunotherapy in solid tumors, even in the absence of MSI-H, PD-L1 overexpression or TMB-H16. Herein, we identified a total of 85 (1%) tumors with CD274 amplification (copy number ≥ 6) in the OM cohort, a proportion consistent with a previous study17(Supplementary Fig. 11a). Furthermore, in 30 evaluable tumors with CD274 amplification tested for PD-L1 expression, the PD-L1 positive rate was 70% (Supplementary Fig. 11b). Subsequently, we also examined the mutational landscape of the 85 samples with CD274 amplification and found the co-occurrence of CD274 amplification with adjacent PDCD1LG2 and JAK2 amplification (89% and 82% respectively), which are nearby genes in chromosome 9p24.3-9p22.2, associated with advanced stage and poorer outcome17. A high frequency of TP53 mutations (78%) was also observed in these tumors (Supplementary Fig. 11c).
Finally, to assess the potential clinical impact of the somatic alterations found in our cohort, we used the MSK criteria10,18 to systematically evaluate actionable variants identified in solid tumors from Chinese patients in our cohort, using the OncoKB (http://oncokb.org/, v3.6) knowledge base. Patients who harbored potential actionable variants in their tumors were classified into different evidence levels of predictive biomarkers. As shown in Fig. 4a and Supplementary Fig. 12a, 64% of patients (n = 6, 498 harbored at least one genomic variants with a variable highest level of clinical evidence (Level 1, 32%; Level 2, 1%; Level 3A, 1%; Level 3B, 13%; Level 4, 16%), including TMB-H as a predictive biomarker of response to immunotherapy14. To further investigate whether the remaining 3, 696 patients without OncoKB Level 1-4 variants in the OM cohort had an actionable biomarker, we analyzed PD-L1 expression. We found that 4% of these patients exhibited at least PD-L1 positive (Supplementary Fig. 12b), suggesting those patients could be candidates for treatment with immune checkpoint inhibitors even if their tumors did not meet the criteria for Level 1-4. A higher ratio of Level 1 was observed mainly in NSCLC, BRCA, SCLC, and UC, compared to that in other cancer types (Fig. 4b). Level 1 was predominantly represented by TMB-H and EGFR mutations in NSCLC, including EGFR L858R (20%; the ratio of variant to the total number of samples with the tumor type), exon 19 deletion (19%) and G719 (3%) mutations. Others included ALK (7%) fusions in NSCLC, PIK3CA mutations (31%), and ERBB2 amplification (24%) in BRCA and MSI-H in CRC (8%) (Fig. 4c). In terms of population-level mutation of actionable variants, KRAS, EGFR and PIK3CA substitution/indels, ERBB2 amplification, and ALK fusions were most common, which was consistent with reports in the MSK cohort (Supplementary Fig. 12a). Interesting, in NSCLC, TMB-H is negatively associated with fusion positive (3% vs. 13% fusion frequency in TMB-H cohort and TMB-L cohort, respectively, P = 1.31E-11), mostly from ALK gene; in contrast, MSI-H shows positive association with fusion positive (6% vs. 1% fusion frequency in MSI-H cohort and MSS cohort, respectively, P = 0.04), mostly from NTRK gene, which hint that clinical benefit of patients from the combination of fusion-based targeted therapy and immunotherapy is different in different types of cancers and the finding requires more studies to confirm in the future. All these findings suggested the relevance of treatment to the mutational landscape of Chinese tumor patients.
In conclusion, we report herein the somatic mutation landscape of over 10,000 solid tumors in Chinese patients. To our knowledge, this is the largest and most comprehensive mutational landscape analysis of solid tumors in an Asian population. This report provides a highly reliable dataset and resource for cancer medicine. More importantly, this population-level comparative analysis has comprehensively revealed similarities and differences between somatic alterations and actionable variants between Chinese and other ethnic populations with solid tumors, and has an important implication for the selection of patients to clinical trials with molecularly targeted therapies.
Online content
Data summary in this study is accessible on cbioportal (https://www.cbioportal.org/study/summary?id=pan_origimed_2020), and software in our FTP (http://ftp.origimed.com).