The clinical characteristics, mutational characteristics, and survival data of CRC patients were recruited for RNF43-mutated analysis from the Cancer Genome Atlas (TCGA) database using the cBio Cancer Genomics Portal (cBioPortal), available at http://www.cbioportal.org ( Supplementary Tables 1 , 2 ) (25). We integrated all the data sets that have been published so far. We performed data consolidation and de-duplication, excluding a total of 1823 patient data, and finally obtained 4028 patients from 11 data sets (coad_cptac_2019,crc_dd_2022,coadread_dfci_2016,bowel_colitis_msk_2022,crc_nigerian_2020,crc_eo_2020,crc_apc_impact_2020,crc_msk_2017,rectal_msk_2019,rectal_msk_2022,coadread_tcga) for this study ( Figure 1 ). We then filtered single-nucleotide variants (SNVs) through the 1021 panel (Geneplus-Beijing, China), a custom-designed biotinylated oligonucleotide probe (Roche NimbleGen, Madison, WI, USA) covering ~1.4 Mbp coding region of genomic sequence of 1,021 cancer-related genes to explore the relationship with tumor genomic characteristics and prognosis ( Supplementary Table 3 ). The TMB was defined as the total number of mutations per megabase (1 Mb) of non-synonymous single-nucleotide variants (SNV), insertion/deletion (Indel), and splice ±2 (26). The upper quartile of tumor mutational burden (TMB) was deemed as high TMB (TMB-H), with a threshold of 8.87 mutations/Mb in this study (27, 28). MSI-H directly used the downloaded label with cBioPortal, and the total number of MSI-H and MSS patients was 296 and 2858, respectively. Before analyzing this study, we calculated the mutation frequencies in 11 cohorts to better understand the reproducibility and limitations of the research, as shown in Supplementary Table 4 . We evaluated the overall research results based on the completeness of data for 100 genes, 200 genes, and 300 genes, with missing rates of 6.09%, 9.09%, and 14.91%, respectively. We believe that these data not only support the overall reliability of the data but also indicate that the data has certain stability and reproducibility. A total of 339 patients with advanced CRC who underwent 1021 panel NGS sequencing served as the validation cohort for this study ( Supplementary Table 5 ) (29). The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by the Ethics Committee of Beijing Hospital (2023BJYYEC-428-02).

All the data were analyzed using the R statistics package (R version 4.2.1, Austria) or GraphPad Prism version 8 (GraphPad Software, CA, USA). Differences between designed groups were analyzed based on the Fisher test or the t-test. Univariate Cox regression and Multivariate Cox regression analysis methods were used to analyze the correlation between mutation characteristics, genomic characteristics, and clinical outcomes. David 6.8 (https://david.ncifcrf.gov/) was used to carry out the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. The log-rank test Kaplan-Meier (KM) survival curve was used to calculate prognostic differences between groups based on RNF43-mutated. Survival curves were calculated by the Kaplan-Meier method, and differences between groups based on RNF43-mutated status were tested by the log-rank test. P values < 0.05 were denoted as statistically significant.

The clinicopathological and molecular characteristics of the enrolled patient population are shown in Table 1 . The mean age of this cohort was 57.94 years, and most of the population (41.91%, N=1688) was between 50 and 70 years old. At the primary tumor site, the RNF43-mutated group had significantly more right-sided patients than left-sided patients (left-sided: 24.47%, N=69; right-sided: 52.13%, N=147), and the RNF43 wild-type cohort data were contrary (left-sided: 54.30%, N=2034; right-sided: 22.02%, N=825). A total of 65% of the patients in the whole cohort were stage III-IV patients, and the tumor grade was mainly moderately differentiated (34.01%, N=1370). The genomic markers TMB-H (198/282, 70.21%, P<0.001) and MSI-H (138/282, 48.94%, P<0.001) were significantly higher in the RNF43-mutated group compared to those in the RNF43 wild-type group. Class 1 BRAF-mutated and RNF43-mutated co-occurred frequently. In the RNF43-mutated group, the proportion of class 1 BRAF-mutated was 34.05%, and that in the RNF43 wild-type group was only 7.13%.

In this study, 375 RNF43 variants were detected in 282 RNF43-mutated patients. The most common variant was frameshift, which occurred in codon 659-mutated of exon 9, including RNF43 p. G659Vfs*41 (N=116) and RNF43 p. G659Sfs*87 (N=2), as shown in Figure 2A , Supplementary Table 6 . The distribution range of RNF43-mutated varies (1.89%-28.13%) in 11 cohorts. The frequency of RNF43-mutated in the vast majority of cohorts was between 7.34% and 17.27%. Additionally, RNF43 codon 659-mutated has a relatively high proportion in all queues, including coadread_dfci_2016, cro_eo_2020, rectal_msk_2022, and coad_cptac_2019 ( Supplementary Figure 1 ). In RNF43-mutated cohorts, the most commonly mutated genes were ARIDIA (59%), CIC (45%), PIK3CA (43%), PTPRS (43%), APC (42%), FAT1 (40%), POLE (40%), NOTCH3 (39%), SPEN (39%), BRAF (38%) ( Figure 2B ). In contrast, the top10 mutated genes in RNF43 wild-type tumors were APC (74%), TP53 (71%), KRAS (41%), PIK3CA (17%), FBXW7 (14%), SMAD4 (13%), TCF7L2 (10%), SOX9 (10%), ARID1A (8%), BRAF (8%) ( Supplementary Figure 2A ). Survival analysis showed that RNF43-mutated had worse progression-free survival (PFS, P=0.0048) and overall survival (OS, P=0.18) ( Supplementary Figures 2B, C ). Considering the limitations of single-mutation data, we conducted a joint analysis using 106 mRNA data from the coad_cptac_2019 dataset. We found that the expression level of RNF43 in the RNF43-mutated was significantly lower than that in the RNF43 wild-type (P < 0.001, Supplementary Figure 2D ). Meanwhile, we found that there were also cases of low RNF43 expression levels within the RNF43 wild-type. Given the relatively small size of the study cohort, we integrated two groups: the RNF43-mutated with expression values < 0, and the RNF43 wild-type with expression values > 0. We found that the RNF43-mutated/expression < 0 shared a similar mutation spectrum with the RNF43-mutated, and RNF43 wild-type/expression > 0 had a similar mutation spectrum to the RNF43 wild-type ( Supplementary Figure 2E ). This finding implies that DNA combined with RNA-based approaches for precise prognostic stratification represents a more optimal choice in the future.

In the validation cohort, we found consistent results in the RNF43-mutated: KRAS (32% vs 43%), APC (42% vs 35%), ARID1A (59% vs 35%), and NF1 (35% vs 28%). However, there were differences in TP53. The abundance in the validation cohort is as high as 70%. The top mutations of the RNF43 wild-type showed high consistency in both the analysis cohort and the validation cohort ( Supplementary Figures 3A, B ). Differential gene analysis showed that the RNF43-mutated group had significantly higher mutation frequency ( Supplementary Table 7 ). The mutation differences between RNF43-mutated and RNF43 wild-type in validation cohort was also analyzed. We founf that NF1, ARID1A, BRAF, B2M, WRN were significantly enriched in RNF43-mutated group, while APC was significantly enriched in RNF43 wild-type, and the RNF43-mutated group had significantly higher mutation frequency ( Supplementary Table 8 , Supplementary Figure 3C ). KEGG pathway enrichment analysis showed that hsa05206: MicroRNAs in cancer, and hsa04151: PI3K-Akt signaling pathway were significantly enriched in the RNF43-mutated group ( Figure 2C ). GO enrichment analysis showed that the RNF43-mutated group had higher enrichment of proliferative signaling pathway (GO: 0016310-Phosphorylation, GO: 0008284~positive regulation of cell population proliferation, GO: 0043066-negative regulation of apoptotic process, etc) ( Figure 2D ). The results of KEGG and GO enrichment analyses of the verification cohort were consistent ( Supplementary Figures 3D, E ).

As the incidence of codon 659 mutation accounted for nearly half of the total RNF43-mutated and had unique clinical significance in predicting the efficacy of anti-BRAF/EGFR combinatory therapies (30), the p.G659Vfs*41 and p.G659Sfs*87 was defined as the RNF43 codon 659-mutated group and the other mutation types were defined as the RNF43 Non-codon 659-mutated group in this study. The clinicopathological and molecular characteristics of the two groups were different from the total population ( Table 2 ). The proportion of RNF43 codon 659-mutated patients aged over 70 years was higher (38.98% vs. 29.88%, P=0.2249). The RNF43 Non-codon 659-mutated occurs most frequently in 50-70 years. There was no difference in gender between the two groups. RNF43 codon 659-mutated occurred frequently in right-sided CRC (59.32%, N=70, P<0.0001), and rarely in the left-sided (11.02%, N=13), while the left and right-sided were more balanced in the RNF43 Non-codon 659-mutated (34.15%, N=56; 46.95%, N=77). In terms of TNM stage, RNF43 codon 659-mutated mainly appeared in TNM II-III (66.1%, N=78), which was inconsistent with the total group staging concentrated in III-IV (65.85%, N=2652).

Subsequently, we analyzed the mutation differences between RNF43 codon 659-mutated and RNF43 Non-codon 659-mutated and pathway enrichment results. CIC, ARID1A, PTCH1, SMARCA4, FLT4 were significantly enriched in RNF43 codon 659-mutated group, while TP53 was significantly enriched in RNF43 Non-codon 659-mutated ( Supplementary Table 8 ). The differences in core gene mutation of major signaling pathways (PI3K-Akt signaling pathway, MicroRNAs pathway, DNA damage repair, and tumor suppressor genes) are shown in Figure 2E . Except for TP53 has the highest frequency in the RNF43 wild-type, other frequencies are significantly higher in RNF43-mutated. Considering that the population of RNF43 codon 659-mutated in the validation cohort is relatively small (only three cases), we are temporarily unable to carry out the validation work for this part.

Based on the differences in clinical features and mutational characteristics exhibited by RNF43 codon 659-mutated and RNF43 Non-codon 659-mutated. Our next step aims to screen biomarkers that predict prognosis. We conducted univariate Cox analysis and multivariate Cox analysis based on OS as clinical outcomes. Study results are shown in Figure 3A . We found age, sample_type (left-sided or right-sided), stage_TNM, MSI, KRAS, APC, and BRAF_V600E (Class1 BRAF-mutated) were biomarkers with significant prognostic differences. Subsequently, factors with P<0.05 were included in multivariate analysis, and it was found that MSI-H was the most significantly different biomarker for better prognosis (P=0.004, HR=3, CI 1.4-6.4), and Class 1 BRAF V600E was the most different biomarker for worse prognosis (P<0.001, HR=0.3, CI 0.21-0.42). We also found that KRAS-mutated was the second-highest predictor of poor prognosis (P<0.001, HR=0.68, CI 0.57-0.81).

MSI-H is a molecular marker that is included in the guidelines and serves as a biomarker indicating a favorable prognosis for CRC. Therefore, further exploration on the value of RNF43-mutated and RNF43 wild-type based on the MSI-H is necessary. Thus, we conducted two groups: MSI-H and RNF43-mutated (N=138), and MSI-H and RNF43 wild-type (N=158). We first conducted a statistical analysis of clinical information ( Table 3 ). No significant differences in age, gender, stage, and tumor grade. However, the proportion of left-sided tumors in the MSI-H and RNF43-mutated group was significantly lower than that in the MSI-H and RNF43 wild-type group (15.22% vs. 28.48%, P=0.0235), and the TMB-H proportion in the MSI-H and RNF43-mutated group was higher (95.65% vs. 87.97%, P=0.0196). Next, we found the overall mutation frequency of MSI-H and RNF43-mutated was higher than that of MSI-H and RNF43 wild-type, with a difference in the distribution of mutations ( Supplementary Table 9 ). In the MSI-H and RNF43-mutated group, the top 5 mutations were ARID1A, PTPRS, FAT1, PIK3CA, and SPEN ( Supplementary Figure 4A ). The top 5 mutations in the MSI-H and RNF43 wild-type group were APC, ARID1A, PIK3CA, PTPRS, and BRAF ( Supplementary Figure 4B ). Furthermore, we analyzed mutations specifically in the MSI-H and RNF43 codon 659-mutated group compared to the MSI-H and RNF43 Non-codon 659-mutated group. The results revealed that the high-frequency mutations in the MSI-H and RNF43 codon 659-mutated group included ARID1A, CIC, PTPRS, FAT1, and PIK3CA ( Supplementary Figure 4C ), with a higher mutation frequency than observed in the MSI-H and RNF43 Non-codon 659-mutated group ( Supplementary Figure 4D ). Finally, we conducted a prognostic analysis; the OS of the MSI-H and RNF43-mutated group was better than that of the MSI-H and RNF43 wild-type group, but there was no significant difference (P = 0.2, Figure 3B ). The OS of the MSI-H and RNF43 codon 659-mutated group was better than that of the MSI-H and RNF43 Non-codon 659-mutated group, and there was also no significant difference (P = 0.61, Figure 3C ). This lack of significant difference may be attributed to the high proportion of poorly differentiated individuals within the MSI-H and RNF43-mutated cohort.

Further cluster analysis was performed for the prognostic markers BRAF and MSI, identified by Cox analysis before. We found high co-occurrence of the RNF43 mutation subtype and BRAF mutation, as well as strong associations with TMB and MSI. We conducted a multi-index UPSET correlation analysis. RNF43 codon 659-mutated was found to overlap with TMB-H, MSI-H, and BRAF V600E. The overlap degree of RNF43 Non-codon 659-mutated with TMB, MSI, and BRAF V600E is lower than that of RNF43 codon 659-mutated ( Figure 4A ). Class1 BRAF-mutated and RNF43 codon 659-mutated, RNF43 Non-codon 659-mutated, and RNF43 wild-type were 35.59%, 28.83% and 4.54%, respectively (P<0.0001). It is also worth noting that the incidence of TMB-H in the RNF43 codon 659-mutated group was 93.22% (110/118), and MSI-H was 78.81% (93/118). This suggests that the co-occurrence of RNF43 with TMB-H or MSI-H is mainly caused by RNF43 codon 659-mutated ( Figure 4B , Table 2 ). Combined with the literature reporting that MSI-H is one of the factors with better prognosis in CRC (21, 22), we believe that multi-indicator association analysis may suggest a better prognosis CRC subgroup.

We then performed a joint analysis of the two indicators, starting with RNF43 combined with BRAF. A total of 282 patients with RNF43-mutated were enrolled and divided into four groups: G1 (N=42): RNF43 codon 659-mutated and Class 1 BRAF-mutated; G2 (N=76): RNF43 codon 659-mutated and Non-class 1 BRAF-mutated; G3 (N=44): RNF43 Non-codon 659-mutated and Class 1 BRAF-mutated; G4 (N=120): RNF43 Non-codon 659-mutated and Non-class 1 BRAF-mutated. The clinicopathological and molecular characteristics of G1-G4 groups are shown in Supplementary Table 10 . Survival analysis of PFS results showed that the G1 group (P=0.0494) and G2 group (P=0.0051) had significantly better prognosis compared with G3 ( Figure 4C ), and OS analysis results showed that only the G2 group and G3 group had significant differences (P=0.0081, Figure 4D ). Patients with RNF43 codon 659-mutated and Non-class 1 BRAF-mutated were found to have a better prognosis. Next, we analyzed the mutation difference between the G2 group and G3 group, and the TOP mutations of the two groups were shown in Supplementary Figure 5A (G2 group) and Supplementary Figure 5B (G3 group). The volcano map of mutation difference analysis showed that ARID1A, CIC, and other genes were significantly enriched in the G2 group ( Supplementary Figure 5C ), and the pathway enrichment results suggested that the microRNAs pathway, DNA damage repair, and tumor suppressive gene mutations were significantly enriched in G2 group: RNF43 codon 659-mutated and Non-class 1 BRAF-mutated. This is consistent with the results of the RNF43-mutated vs RNF43 wild-type analysis, suggesting that the combined detection of RNF43 and BRAF can help predict a better prognosis ( Supplementary Figures 5D-F ).

In order to match the clinical guidelines recommended, we only compared RNF43 combined with MSI and did not perform RNF43 combined with TMB. Therefore, four groups are assigned. G1 (N=93): RNF43 codon 659-mutated and MSI-H; G2 (N=25): RNF43 codon 659-mutated and Non-MSI-H; G3 (N=45): RNF43 Non-codon 659-mutated and MSI-H; G4 (N=119): RNF43 Non-codon 659-mutated and Non-MSI-H. This part also included 282 RNF43-mutated patients. The clinicopathological and molecular characteristics of the four groups are shown in Supplementary Table 11 . As expected, G4 had worse PFS and OS (G1/G4-PFS: P=0.0054; G1/G4-OS: P=0.0024). The top mutations of group G1 and group G4 were shown in Supplementary Figure 6A (G1) and Supplementary Figure 6B (G4). Volcanic map analysis of mutation differences showed that ARID1A, CIC, PTPRS, PTCH1, and other genes were significantly enriched in group G1, and TP53 was significantly enriched in group G4 ( Supplementary Figure 6C ). This is consistent with the conclusion that CRC patients carrying TP53 mutations have a worse prognosis. The enrichment results were consistent with the results of RNF43-mutated vs RNF43 wild-type and RNF43 combined BRAF analysis ( Supplementary Figures 6D–F ).

Subsequently, we integrated the three indicators of RNF43, BRAF, and MSI found above for integrated analysis, to find the population with the best prognosis. Different from the previous analysis process, we also included RNF43 wild-type in this part, and a total of 3937 CRC patients with survival data were recorded and screened, which were divided into three groups: G1: RNF43 codon 659-mutated, Non-class 1 BRAF-mutated, and MSl-H, G3: RNF43 Non-codon 659-mutated (including RNF43 wild-type), Class 1 BRAF-mutated, and Non-MSl-H; G2: Non-G1 and Non-G3. The study found that the G1 had a better PFS and the G3 had a worse PFS (G1/G3: P=0.0005; G1/G2: P=0.3062; G2/G3: P<0.0001, Figure 5A ). The results of the OS survival analysis were more significant: the OS of G1 was 100%. The G3 has the worst OS in this cohort (G1/G2: P=0.0155; G1/G3: 0.0022; G2/G3: 0.0267, Figure 5B ). RNF43 codon 659-mutated, Non-class 1 BRAF-mutated, and MSl-H were found to have the best prognosis in the population.

Meanwhile, we performed clinicopathological analysis and mutation characteristic analysis, as shown in Table 4 . Compared with the G2 and G3 groups, the age of G1 was higher than that of the Elder group (64.71%; G1/G2: P<0.0001; G1/G3: P<0.0001). There was no significant difference in gender among the three groups. In terms of primary tumor location, contrary to previous conclusions, G1 was significantly enriched on the right side (88.24%; G1/G2: P<0.0001; G1/G3: P<0.0001). G1 was significantly enriched in stage II colorectal cancer (52.94%; G1/G2: P<0.0001; G1/G3: P = 0.0087). In terms of Tumor_Grade, the frequency of poorly differentiated tumors was higher in G1 (38.24%, N=13), but there was no statistical difference in G1/G3 group (P=0.0767). Finally, we show the mutation landscape of G1 ( Figure 5C ) and G3 ( Figure 5D ) and carry out mutation difference analysis ( Supplementary Table 12 ) and pathway enrichment ( Figures 5E–F ). The enrichment results of major pathways were consistent with the results of RNF43-mutated/RNF43 wild-type and RNF43-mutated and BRAF/or MSI analysis (as shown in Figures 5E–F ). The mutation frequency of core gene mutations of major signaling pathways (PI3K-Akt signaling pathway, MicroRNAs pathway, DNA damage repair, and tumor suppressor genes) in G1 was significantly higher than that in G3 ( Figure 5G ).