Genomic and immunological profiles of small-cell lung cancer between East Asians and Caucasian

The Institutional Review Boards (IRBs) of the First Affiliated Hospital of Guangzhou Medical University, the Sun Yat-sen University Cancer Center, and the Zhujiang Hospital of Southern Medical University approved this study. A total of 98 samples from EA SCLC patients were provided by these hospitals under IRB-approved protocols with informed consent. These 98 tumor samples from EA SCLC patients were collected retrospectively from surgical material. We used blood or adjacent normal tissues as a matched control. TNM stage, sex, smoking history, and age were collected. Detailed information is provided in the Additional file: Supplemental methods 16.

Processed WES data and RNA sequencing (RNA-seq) files of 45 Caucasian SCLC patients from a 2015 Nature study were downloaded from cBioPortal (https://​www.​cbioportal.​org/​study/​summary?​id=​sclc_​ucologne_​2015) [1]. The SCLC cell lines described in this study were derived from the Genomics of Drug Sensitivity in Cancer (GDSC) database [12] and had drug sensitivity and WES data.

The CIBERSORT algorithm [13] was supplied with mRNA data of EA and Caucasian SCLC patients. The proportion of twenty-two tumor-infiltrating immune cells was used in downstream analysis. Then, expression values were selected for CIBERSORT analysis using default parameters (perm = 1,000; QN = F). Marker genes [14, 15] related to immune cells, antigen presentation, cytotoxicity, cytokines, and immune checkpoints were collected from previous studies and used to evaluate the immune signatures of SCLC. To compare differentially immune cells and immune-related genes, a linear model of the limma package was supplied.

The Complexheatmap package [16] was used to visualize the waterfall plot of mutations in EA and Caucasian SCLC patients. Nonsynonymous mutation types were determined using the maftools package [17]. The summary plot of the MAF files and figures of somatic interactions were generated by the maftools package. Tumor mutation burden (TMB) values were calculated according to a previous study [18]. A list of hallmark and DNA damage response (DDR) genes was collected from the Molecular Signatures Database (https://​www.​gsea-msigdb.​org/​gsea/​msigdb/​index.​jsp) [19] and used for signaling alteration analysis. The DDR signature scores were calculated using the gene set variation analysis (GSVA) package [20] with the single-sample gene set enrichment analysis (ssGSEA) method. According to the median age or TMB value, SCLC patients were classified into groups: older vs young and high TMB vs low TMB. Driver gene annotations were downloaded from the Network of Cancer Genes (NCG) database [21].

To identify which inhibitors/compounds may be useful for targeting cells with TP53 and RB1 co-mutations, we applied the Broad Institute’s Connectivity Map (CMap) build 02 [22], which is a public online analytical tool (https://​portals.​broadinstitute.​org/​cmap/​) that allows the analyzer to predict potential inhibitors/compounds based on upregulated and downregulated genes in a gene expression signature.

To further discover the mechanism of action (MoA) [23] and inhibitors/compounds, we analyzed them using CMap tools (https://​clue.​io/​). The CMap method is similar to the gene set enrichment analysis (GSEA) algorithm, which can identify similarities and connectivities (range: − 1 to 1) based on differential gene expression data.

All analyses were performed in R (version 3.6.1). The Mann–Whitney U test was used for the comparison of two continuous variables. Fisher’s exact test was supplied with two categorical variables. P values were controlled for false discovery rate (FDR), and an FDR less than 0.05 was considered statistically significant; all statistical tests were two-sided. Survival analysis was performed using the Kaplan–Meier method, and the log-rank test p-value was calculated. A Cox regression model was used in univariable analyses.

With the Mutect2 algorithm [24], 103,100 single-nucleotide polymorphisms (SNPs) and 2926 short insertions/deletions (indels) were identified in the EA cohort (Fig. 1a), and the majority were missense mutations (n = 92,103). We identified 11,029 SNPs and 676 indels in the Caucasian cohort (Fig. 1b). In particular, C > T was identified predominantly in the EA cohort, whereas C > A was the predominant single-nucleotide variant (SNV) in the Caucasian cohort (Fig. 1a,b). The top 20 mutated genes in the EA cohort are shown in Fig. 1c and TP53 (89%), TTN (80%), RB1 (67%), MUC16 (57%), and RYR2 (49%) were the most frequently mutated genes. The top 20 mutations in the Caucasian cohort are shown in Fig. 1d, and TP53 (89%), TTN (73%), RB1 (71%), LRP1B (49%), and MUC16 (49%) were the most frequently mutated genes. We observed higher co-occurrence and exclusiveness gene pairs in the top 20 mutations of the EA cohort than of the Caucasian cohort (Additional file 1: Fig. S1). Oncogenes and tumor suppressor genes (TSGs) play a role in cancer evolution and development. For example, LRP1B and MUC16 exhibited co-occurrence in the EA cohort (Additional file 1: Fig. S1a; Additional file 10: Table S1), while there was no co-occurrence/mutual exclusivity of oncogenes/TSGs in the Caucasian cohort (Additional file 1: Fig. S1b; Additional file 11: Table S2). The TMB was significantly higher in EA SCLC patients than in Caucasian SCLC patients (median 16.75 vs 6.24 per Mb; mean 30.95 vs 7.03 per Mb; FDR < 0.0001; Fig. 1e).

Alterations in RB1 and TP53 occurred in approximately 65% and approximately 90% of the SCLC patients, respectively [1]. We identified that 59% of patients in the EA cohort had co-occurrent TP53 and RB1 mutations; 28% had only a TP53 mutation, 6% had only an RB1 mutation, and 7% had no alterations in TP53 or RB1 (Fig. 2a). The Caucasian cohort harbored more patients with alterations in both TP53 and RB1 (67%), although there was no significant difference between cohorts. In the EA cohort (Fig. 2b), multiple missense mutations occurred in the DNA-rich and tetramerization domains, while truncating mutations in RB1 affected the DUF3452 and RB B, RB A and RB C domains. Similarly, RB1 and TP53 mutations affected the same domain in the Caucasian cohort (Fig. 2c). Subsequently, we explored the association between RB1 and TP53 mutations and the survival of SCLC patients. In the EA cohort, TP53, RB1, and TP53/RB1 co-mutations exhibited no significant associations with a survival benefit (Fig. 2d). In addition, there were no correlations between survival and alterations in RB1, TP53 or both RB1 and TP53 in the Caucasian cohort (Fig. 2e).

For each SCLC patient, we calculated the frequency of patients harboring at least 1 alteration in each of the 28 signaling pathways in both the EA and Caucasian cohorts (Fig. 3a–b). Epithelial-mesenchymal transition (EMT) was the signaling pathway that was most frequently mutated (10.2% of all alterations) across all patients in the EA cohort, followed by E2F targets, KRAS signaling, P53 signaling, IL2/STAT5 signaling, hypoxia, adipogenesis, interferon-gamma response, and inflammatory response pathways, which were altered in 8.9%, 7.4%, 7.4%, 7.0%, 6.6%, 6.5%, 6.4% and 6.2% of all alterations, respectively. In contrast, reactive oxygen species pathways harbored the lowest number of alterations (Fig. 3a). Alterations in EMT signaling were identified most predominantly in the Caucasian cohort, while only some genes were altered in the angiogenesis and reactive oxygen species signaling pathways. More detailed information on the mutation frequencies of hallmark pathways is shown in Fig. 3b. Subsequently, with annotations from the Network of Cancer Genes (NCG) database, we compared the differences in frequencies of driver genes between the EA and Caucasian cohorts. Of the 42 significantly mutated genes between the two cohorts (all p < 0.05; Fig. 3c), TSGs were commonly involved (up to 54.8%). For alterations commonly found in the EA cohort, TSGs, such as FAT1, NCOR2, SMARCA4, UBR5, and CREBBP, showed higher alteration rates, while ARHGEF10 exhibited a lower alteration rate than those found in the Caucasian cohort. Additionally, a well-known oncogene (EGFR) had higher numbers of alterations (mainly missense mutations) in the EA cohort than in the Caucasian cohort, followed by other oncogenes (TRRAP, MTOR, TNC, DNMT1, and RET). For mutual exclusivity and co-occurrence driver analyses, alterations in MTOR, SPEN, NCOR1, BRCA2, POLE, EGFR, ARID1B, and some other genes frequently co-occurred in the EA cohort (Additional file 2: Fig. S2a; Additional file 12: Table S3), whereas there were no significant mutual exclusivity gene pairs. In contrast, there were a few comutated gene pairs, namely, BRCA2, FANCA and ZMYM3, in the Caucasian cohort (Additional file 2: Fig. S2b; Additional file 13: Table S4).

Mutations in DDR pathways have been identified to affect the efficacy of platinum-based chemotherapy and immunotherapy for SCLC. For each DDR pathway, we computed the alteration frequencies of SCLC samples with at least 1 mutation in each of 8 signaling pathways (Fig. 4a–b). Homologous recombination (HR) was the pathway harboring the highest alteration frequencies among total mutations (32.2%), followed by single-strand breaks (SSB; 29.5%) and nucleotide excision repair (NER; 28.3%). However, nonhomologous end joining (NHEJ) pathways harbored the lowest fraction of alteration frequencies in the EA cohort (Fig. 4a). Next, we applied the same analytic pipeline in the Caucasian cohort. Genomic alterations in SSB signaling were highest, while mutations in base excision repair (BER) were the lowest (Fig. 4b). Additionally, some pathways, such as SSB and NHEJ, had mutations distributed among each SCLC patient in either the EA or Caucasian cohort (Fig. 4a, b). Additionally, the number of alterations in each DDR signaling pathway was higher in the EA cohort than in the Caucasia cohort (all adjusted p < 0.05; Fig. 4c). Next, we discovered that there was a significantly positive correlation between TMB and each DDR signaling pathway mutations (p < 0.05; Fig. 4d). By contrast, we discovered that only NER and DDR mutations were significantly related to higher TMB in the Caucasian cohort (Fig. 4e). Based on the ssGSEA method, none of the DDR signaling pathway alterations were significantly correlated with TMB in both the EA and Caucasian cohorts (Additional file 3: Fig. S3a–b).

The top twenty altered driver genes of the EA and Caucasian cohorts are characterized in Fig. 5a–b. We discovered up to 60% TSGs were driver genes in the EA cohort (Additional file 4: Fig. S4a), especially the top 5 altered mutations (TP53, RB1, CSMD3, and LPR1B). Up to 25% of oncogenes were commonly altered in the Caucasian cohort (Additional file 4: Fig. S4b), including MUC16 (50%), PREX2 (16%), ZNF521 (14%), ALK (11%), CTNND2 (11%) and MUC4 (11%). Particularly interesting alterations across the EA and Caucasian cohorts were TP53 and RB1. The main alteration type of TP53 in the EA cohort was missense mutations (66.0%), followed by nonsense mutations (16.0%) and frameshift mutations (9.6%). In the Caucasian cohort, TP53 was commonly altered as missense mutations (69.0%) and frameshift mutations (16.7%). In the EA cohort, RB1 was frequently altered as frameshift mutations (31.5%), followed by nonsense mutations (28.8%), splice sites (23.3%), and missense mutations (16.4%). RB1 had nearly equal distribution of the three mutation types (37.5% nonsense mutations, 31.3% spice site mutations and 28.1% frameshift mutations) in the Caucasian cohort. Through exploring significant alterations of key pathways, we combined several hallmark pathways and their genes to map nonsynonymous mutations. Multiple genes of each of the five pathways (DNA repair, EMT, G2M checkpoint, hypoxia, and KRAS signaling pathways) were significantly mutated in the EA and Caucasian cohorts (Additional file 4: Fig. S4c). For example, in the G2M checkpoint signaling pathway, mutations in TSGs, including POLE (26% vs 10%), BRCA2 (19% vs 3%) and STAG1 (19% vs 3%), occurred more commonly in the EA cohort than in the Caucasian cohort (all Fisher’s exact test p < 0.05). We identified EGFR, a well-known oncogene, as being mutated at a higher rate in EA patients than in Caucasian patients (25% vs 10%, Fisher’s exact test p < 0.05). IGFBP2, a protein in the EMT signaling pathway, was significantly more mutated in the Caucasian cohort than in the EA cohort (20% vs 4%, p < 0.05). Subsequently, we analyzed the potential association among driver genes, clinical phenotypes, and survival of SCLC patients using univariable Cox regression models. In the EA cohort, alterations in APC, NSD3, KDM5C, CNTRL, GRM3, CTNND1, FANCG, MET, and SRGAP3 were associated with a significantly poor survival, but mutations in TP53 or RB1 conferred no survival benefits (Additional file 5: Fig. S5a). Subsequently, we identified two different driver genes (FCRL4 and PTPRT) to stratify the Caucasian SCLC patients with the same analytical model (Additional file 5: Fig. S5b). However, alterations in TP53 or RB1 exhibited no potential associations with patient survival in the Caucasian cohort. Furthermore, we discovered four alterations with totally different prognosis values in the EA and Caucasian cohorts. For instance, driver gene mutations in OTOF, ANKRD30B, and TECPR2 correlated with significantly shorter survival in the Caucasian cohort (Additional file 5: Fig. S5c), but these mutations showed no survival benefits in the EA cohort (Additional file 5: Fig. S5d). In contrast, alterations in COL6A6 were associated with poor OS in EA SCLC patients (Additional file 5: Fig. S5d), while this mutation had no correlation with OS in Caucasian patients (Additional file 5: Fig. S5c).

We applied the CMap algorithm for identifying associations among different groups and conditions to discover potential inhibitors/compounds targeting signaling pathways correlated with co-mutations in TP53 and RB1 (Fig. 5a; Additional file 14: Table S5, Additional file 15: Table S6). The adrenergic receptor antagonist terazosin, the ATP channel activator pinacidil, the topoisomerase inhibitors mitoxantrone and irinotecan, the heat shock protein (HSP) inhibitor alvespimycin, and the PARP inhibitor (PARPi) NU − 1025 showed significant correlations with TP53 and RB1 co-mutations in the EA cohort. We identified two compounds, the NFκB pathway inhibitor parthenolide and the ATPase inhibitor thapsigargin, that were significantly enriched in TP53 and RB1 co-mutations in the Caucasian cohort, but these inhibitors exhibited significantly negative correlations with TP53 and RB1 co-mutations among the EA cohort. Subsequently, applying CMap MoA analysis to the EA (Fig. 5b) and Caucasian cohorts (Fig. 5c), we discovered 17 mechanisms shared by 19 inhibitors/compounds in the EA cohort (Fig. 5b). Two compounds (irinotecan and mitoxantrone) shared MoAs of topoisomerase inhibitor. We identified SC − 560 and indometacin as cyclooxygenase inhibitors. In the Caucasian cohort (Fig. 5c), we found 34 mechanisms shared by 51 inhibitors/compounds, such as dopamine receptor antagonists (thioridazine, promazine, and prochlorperazine), ATPase inhibitors (proscillaridin, thapsigargin, ouabain, digoxin, helveticoside, and digitoxigenin), MTOR inhibitors (sirolimus and LY-294002) and an NFκB pathway inhibitor (parthenolide). Additionally, we calculated the drug sensitivity associated with more than two mutations from the top 20 mutated genes in the EA and Caucasian cohorts (Additional file 6: Fig. S6a, b).

CIBERSORT, an algorithm to evaluate the fractions of 22 immune cells, was applied to characterize the proportion of tumor-infiltrated immune cells in the EA and Caucasian cohorts. In total, 16 immune cells among the EA and Caucasian cohorts were characterized to be significantly different (Fig. 6a). For example, several immune cells, such as naïve B cells, naïve CD4 + T cells, resting memory CD4 + T cells, resting natural killer cells (NKs), monocytes, activated dendritic cells (DCs), eosinophils, and neutrophils, were significantly enriched in the EA populations, while some cells, such as plasma cells, CD8 + T cells, activated memory CD4 + T cells, activated NKs, M1-type macrophages, M2-type macrophages, resting DCs and activated mast cells, accounted for significantly more immune cells in the Caucasian cohort than in the EA cohort. In addition, we compared the differences in the immune-infiltrated signature between the EA and Caucasian cohorts. The mean differences (log fold change) in the 22 immune cells between clinical and genomic features in the EA and Caucasian cohorts are shown in Additional file 7: Fig. S7a, b. Figure 6b shows the mean differences (log fold change) in immune-related mRNA expression levels between EA and Caucasian SCLC patients. Several inhibitory mediators, such as VEGFA, TGFB1, and FOXP3, were significantly upregulated in EA SCLC patients, but some antigen presentation genes, such as MICA, MICB, and TAP1, were significantly downregulated in the EA cohort. Additionally, chemokines (CXCL9 and CXCL10) and cytolytic activity-related genes (GZMB) were commonly downregulated in the EA cohort. The mean differences (log fold change) in the clinical and genomic features of immune-related genes in the EA and Caucasian cohorts are shown in Additional file 8: Fig. S8a, b. Notably, among the immune checkpoint-related genes in both the EA and Caucasian cohorts, we discovered that two genes (PDCD1 and HAVCR2) were significantly enriched in the Caucasian cohort compared with the EA cohort (Fig. 6c). The correlation analysis between the TMB, DDR, and immune cells in the EA and Caucasian cohort was shown in Additional file 9: Fig. S9. In the Caucasian cohort, we found that there was a positive correlation between the proportions of the activated NK cells and the DDR mutation counts or TMB. Similarly, the abundance of the activated DCs was positively associated with the DDR or, SSB, or MMR mutation counts. The proportion of the follicular helper T cells was positively correlated with the TMB (Additional file 9: Fig. S9a). In the EA cohort, there was a negative correlation between the proportion of the monocytes and the HR, DSB, NHEJ, FA, or DDR mutation counts. Additionally, we found that the abundance of the resting NK cells was positively correlated with the BER mutation counts (Additional file 9: Fig. S9b).