Patient characteristics
A total of 4591 Chinese patients with 18 types of solid tumors were included in this study, and 21.8% of them carried variants in at least one of six SWI/SNF genes (ARID1A, ARID1B, ARID2, SMARCA4, SMARCB1, and/or PBRM1). Among them, 301 SWI/SNF-mutant and 700 SWI/SNF-non-mutant patients had received ICIs, including anti-PD-1, PD-L1, and CTLA4 or their combinations. Further, the SWI/SNF-non-mutant group had a higher proportion of patients with TNM stage Ⅰ than the SWI/SNF-mutant group, while age, sex, smoking status, and ICI type were not markedly different between the two groups (Table 1).
The most common cancer type observed in this study was non-small cell lung cancer (32.3%), followed by colorectal cancer (29.6%) and ovarian and fallopian tube cancer (7.9%). The top five malignancies with the highest SWI/SNF mutation rates were endometrial cancer (54.1%), gallbladder and biliary tract cancer (43.4%), gastric cancer (33.9%), urothelial cancer (30.6%), and ovarian and fallopian tube cancer (23.9%). The SWI/SNF mutation rate corresponding to the “Other” subset, which comprised some relatively uncommon tumors, including skin squamous cancer, urachal cancer, gastrointestinal stromal tumor, glioma, adrenal tumors, and medullary thyroid cancer, among others, was 17.1%.
Spectrum of SWI/SNF complex genomic variations
Among the six SWI/SNF genes, ARID1A and SMARCB1 were respectively, the most and least frequently mutated genes in the majority of the cancer types (ARID1A, 49.3%; SMARCA4, 27.7%; ARID1B, 21.6%; ARID2, 18.3%; PBRM1, 15.9%; SMARCB1, 6.1%; Table 2, Fig. 1a). Notably, SMARCA4 mutations were slightly more common than ARID1A mutations in non-small cell lung cancer, cervical cancer, and melanoma. Interestingly, up to 25.0% of the SWI/SNF-mutant tumors showed genetic aberrations at ≥ two SWI/SNF genes (Table 2).
All the genetic alterations were classified under the following seven types: frameshift indels, in-frame indels, nonsense mutations, missense mutations, splice site mutations, CNVs, and fusions (gene rearrangements). Frameshift indels constituted the most common variation type in ARIDIA, whereas missense mutations were more common in the other five genes (Fig. 1b). The proportions of LOF mutations, generally including frameshift indels, nonsense mutations, and splice site mutations, of the SWI/SNF complex genes were as follows: ARID1A, 69.8%; ARID2, 43.4%; PBRM1, 41.8%; ARID1B, 29.0%; SMARCA4, 23.8%; and SMARCB1, 18.5% (Fig. 1c).
Although most variations were widely distributed across the full length of each gene, a number of frameshift indels (fs) and nonsense mutations (*), which led to the truncation of protein products, were relatively frequently detected. These included D1850fs, G276fs, R1989*, R1276*, and F2141fs of ARID1A (Fig. 2a), R1944* of ARID1B (Fig. 2b), I37fs, R53fs, and p.E71* of ARID2 (Fig. 2c), N258fs, I279fs, p.R146, R710*, and K909fs of PBRM1 (Fig. 2d), P109fs, G271fs, and R1077* of SMARCA4 (Fig. 2e), and R40*, T72fs, and R201* of SMARCB1 (Fig. 2f). In addition, several missense mutations such as A329G of ARID1B, R1192H/L/C and D1177N/Y of SMARCA4, and R366C/H and R377C/H of SMARCB1 were detected in a relatively greater number of cases (Fig. 2b, e, and f).
Co-occurrence and mutual exclusivity
To uncover the potential pattern of SWI/SNF gene mutations, the co-occurrence and exclusivity of the mutations of the six SWI/SNF genes and the top 20 most frequently altered genes across all tumors were explored. Of the top 20 mutated genes excluding the six hub genes, it is well known that APC, KRAS, PIK3CA, EGFR, LRP1B, BRCA2, ATM, and ROS1 are mutated in several cancer types, such as non-small cell lung cancer, colorectal cancer, and endometrial cancer. In this study, we observed that ARID1A was the second most frequently mutated gene following TP53. Further, ARID1A variations were identified in a mutually exclusive pattern with variations in EGFR, TP53, ARID1B ARID2, and SMARCA4, while ARID2 was exclusively mutated with SMARCA4. However, PBRM1 tended to be co-mutated with ARID2 and SMARCB1 (Fig. 3).
Furthermore, it is worth noting that MLL2 (MLL4/KMT2D) and MLL3 (KMT2C), belonging to a family of mammalian histone H3 lysine 4 (H3K4) methyltransferases [39], were frequently co-mutated with SWI/SNF genes (Fig. 3). Reportedly, KMT2D collaborates with the SWI/SNF complex to promote cell type-specific enhancer activation [40], and cancer cells with KMT2C deficiency suffer from higher endogenous DNA damage and genomic instability [41]. The subset carrying both SWI/SNF and MLL2/3 mutations showed higher average TMB values (MLL2, 70.9 mutations/Mb; MLL3, 74.5 mutations/Mb), TMB-H ratios (MLL2, 80.5%; MLL3, 83.6%), and MSI-H ratios (MLL2, 48.6%; MLL3, 46.6%) than the whole SWI/SNF-mutant group (all p < 0.0001).
Associations of SWI/SNF mutations with TMB and MSI
Previous studies have revealed the existence of a potential linkage between the SWI/SNF chromatin remodeling complex and DNA repair, TMB, and MSI [6]. Thus, in this study, these relationships were further analyzed. Our results in this regard indicated that the average TMB value corresponding to SWI/SNF-mutant tumors was markedly higher than that corresponding to SWI/SNF-non-mutant tumors, regardless of the cancer type (25.8 vs. 5.6 mutations/Mb, p < 0.001). Further, the TMB-H and MSI-H ratios corresponding to SWI/SNF-mutant tumors were also significantly higher than that corresponding to the SWI/SNF-non-mutant tumors (TMB-H ratio: 41.2% vs. 8.5%, p < 0.001; MSI-H ratio: 16.0% vs. 0.9%, p < 0.001), even though the differences were not significant for certain malignancies, such as pancreatic cancer, prostate cancer, and urothelial cancer. SWI/SNF-mutant endometrial cancer, colorectal cancer, gastric cancer, ovarian and fallopian tube cancer, and soft tissue sarcoma exhibited both higher TMB-H and MSI-H ratios than their SWI/SNF-non-mutant counterparts (Table 3). Furthermore, the patient group with mutations at ≥ two SWI/SNF genes had significantly higher TMB values (69.0 vs. 11.3 mutations/Mb, p < 0.0001), TMB-H ratios (79.6% vs. 28.4%, p < 0.0001), and MSI-H ratios (48.0% vs. 5.3%, p < 0.0001) than those with mutations in a single SWI/SNF gene (Table 3).
ICI treatment outcomes of patients with SWI/SNF mutations
Over the past few years, pre-clinical and clinical evidence has implicated the SWI/SNF complex as a potential predictor of response to ICIs [6]. For the ICI-treated patients, we observed that the presence of SWI/SNF variants was significantly associated with a longer PFS [not reached (NR) vs. 29.9 months, HR = 0.52 (0.41–0.66); p < 0.0001], regardless of the presence of LOF or non-LOF variants [NR vs. NR, HR = 0.98 (0.57–1.67); p = 0.9305, Fig. 4a]. Specifically, patients carrying mutations at ≥ two SWI/SNF genes did not show a superior PFS than those single gene mutation carriers [NR vs. NR, HR = 0.85 (0.51–1.42), p = 0.7585; Fig. 4b]. Additionally, the exploration of the predicting significance of each SWI/SNF gene mutation showed that PBRM1 mutations were associated with a relatively better outcome of ICI treatments than the other SWI/SNF gene mutations [NR vs. NR, HR = 0.21 (0.12–0.37), p = 0.0007; Fig. 4c]. Notably, the prediction value of the SWI/SNF variants increased considerably when the TMB-H status was also considered. In particular, we observed that the SWI/SNF-mutant + TMB-L cohort showed a longer PFS than the SWI/SNF-non-mutant + TMB-L cohort [NR vs. 28.0 months, HR = 0.65 (0.46–0.92), p = 0.0322], and that the SWI/SNF-mutant + TMB-H cohort showed an even longer PFS than the SWI/SNF-non-mutant + TMB-L cohort [NR vs. 28.0 months, HR = 0.42 (0.31–0.55), p < 0.0001; Fig. 4d].
Furthermore, regardless of the cancer type, patients in the SWI/SNF-mutant group showed both higher ORR values (3.32% vs. 0.43%, p = 0.0002) and DCR values (80.07% vs. 65.57%, p <0.0001) than their counterparts in the SWI/SNF-non-mutant group. For individual cancer types, SWI/SNF-mutant colorectal cancer (86.27% vs. 67.83%, p = 0.0014), gastric cancer (83.33% vs. 55.77%, p = 0.0222), and non-small cell lung cancer (85.07% vs. 71.58%, p = 0.0324) showed significantly higher DCR values in immunotherapy than their SWI/SNF-non-mutant counterparts (Table 4).
Synthetic lethality involving SWI/SNF members
In recent years, synthetic lethality has attracted considerable attention in oncology, as it may explain the sensitivity of cancer cells to certain inhibitors and provide a new angle for drug development. The previously reported synthetic lethal pairs and effective inhibitors in SWI/SNF-deficient cancers are summarized in Table 5. These synthetic lethal interactions could be classified under four main categories: (a) Two subunits within the SWI/SNF complex, e.g., the BRD2 inhibitor, JQ1 can suppress ARID1A-deficient ovarian clear cell cancer cells because BRD2 inhibition decreases ARID1B transcription [42]. (b) One SWI/SNF subunit with its competitor. Contrary to the chromatin relaxation-inducing function of the SWI/SNF complex, polycomb repressive complex 2 (PRC2), whose enzymatic catalytic subunit is the methyltransferase, EZH2, promotes chromatin compaction via histone H3 K27 trimethylation (H3K27me3). Thus, the inhibition of EZH2 using Tazemetostat or GSK126 causes synthetic lethality in ARID1A-, SMARCA4-, SMARCB1-, PBRM1-deficient cancers [43–48]. (c) Targeting the functions of the SWI/SNF complex. The SWI/SNF chromatin remodeling complex functions in DNA double-strand break repair, transcription, replication, chromosomal segregation, and several metabolic pathways. Therefore, SWI/SNF-deficient cancers are vulnerable to the inhibition of homologous recombination repair factor, PARP1 [20,49], cell cycle regulator, cyclin-dependent kinase (CDK)4/CDK6 [25,50], DNA replication checkpoint factor, ATR [51], chromosomal segregation factor, Aurora kinase A [52], and oxidative phosphorylation [53] and glutathione [54] pathways. (d) Others. PD-1/PD-L1 inhibitors had synthetic lethal effects in ARID1A- and PBRM1-deficient cancers [21,55].