We systematically analyzed NOTCH3 alterations across multiple cancer types from various molecular perspectives. At the transcriptional level, NOTCH3 expression exhibited a heterogeneous pattern among different cancers (Figure 1a) compared to their corresponding normal tissues. Elevated NOTCH3 expression was observed in glioblastoma (GBM), ovarian cancer (OV), pancreatic cancer (PAAD), and gastric cancer (STAD), Diffuse Large B-cell Lymphoma (DLBC), Kidney Renal Clear Cell Carcinoma (KIRC), Liver Hepatocellular Carcinoma (LIHC), Testicular Germ Cell Tumors (TGCT), Thymoma (THYM), while reduced expression was detected in melanoma (SKCM), kidney renal papillary cell carcinoma (KIRP), and uterine cancer (UCEC). In CRC, NOTCH3 expression was not significantly different. Mutation analysis revealed that NOTCH3 mutations were distributed sparsely across the coding region (Figure 1b, left), a pattern often associated with passenger mutations. This trend was also consistent in COAD (CRC, Figure 1b, right). The relationship between DNA CpG methylation (cg06650786) within NOTCH3 and its expression was analyzed, and a significant negative correlation was detected (Figure 1c, p =1.5e-302, Spearman correlation, R=-0.36). Additionally, NOTCH3 expression was also significantly correlated with copy number variations (CNVs, Figure 1d). Integrating CNV and mutational data, we found that UCEC exhibited the highest NOTCH3 alteration frequency (16%, Figure 1e), followed by ovarian cancer, melanoma, and esophageal cancer. In CRC, 7% samples harbored NOTCH3 alterations (mutations and/or CNVs), ranking 7th among all cancer types. In summary, our findings demonstrate that NOTCH3 alterations are prevalent across multiple cancers, with its expression significantly associated with both DNA methylation and copy number variations, suggesting that they may regulate the expression of NOTCH3.

We next analyzed the impact of NOTCH3 on clinical outcomes across different cancers. The prognostic value of NOTCH3 expression for overall and progression-free survival was assessed using Cox univariate regression. Elevated NOTCH3 expression predicted poor overall survival in uveal melanoma, skin melanoma, uterine cancer, gastric cancer, mesothelioma, kidney renal papillary cell carcinoma, colorectal cancer, lower-grade glioma, and bladder cancer, while opposite in thymoma (Figure 2a). A similar trend was observed for progression-free survival (Figure 2b). Patients in each cancer type were equally divided into high- and low-expression groups based on NOTCH3 expression in the corresponding cohort, and survival differences between the groups were evaluated. The high-NOTCH3 expression group exhibited worse overall and progression-free survival (Figures 2c, d), which is similar to Cox univariate regression. The consistent findings demonstrate the prognostic value of NOTCH3, which is robust in specific cancer types—regardless of the statistical algorithms used—yet exhibits heterogeneity across cancer types. Specifically, NOTCH3 mRNA abundance was significantly correlated with both overall and progression-free survival in both Cox regression (Figures 2a, b) and group comparison analyses (Figures 2c–e) in CRC. The predictive performance of NOTCH3 was further evaluated using the area under the receiver operating characteristic curve for one-, three-, and five-year survival across cancer types (Figure 2f). NOTCH3 expression has a favorable predictive value in several cancers. We also examined the impact of its somatic mutations on survival. However, NOTCH3 mutations in most cancers were not significantly associated with overall survival (Supplementary Figure S1; N = 488 for wild-type, N = 38 for mutant), possibly due to limited NOTCH3-mutated sample, or it may influence the prognosis under other conditions (i.e treatment). Furthermore, NOTCH3-mutated cases showed significantly higher microsatellite instability (MSI) scores than wild-type cases (Figure 2g). In summary, NOTCH3 expression serves as a prognostic indicator in multiple cancer types, including CRC.

The tumor microenvironment, shaped by cell–cell interactions, plays a critical role in CRC development and progression. Therefore, we investigated the association between NOTCH3 and immune infiltration. After estimating immune cell abundances using different algorithms, we compared immune infiltration proportions between NOTCH3-mutant and wild-type samples (Figure 3a). NOTCH3 mutations were significantly associated with altered infiltration of various immune cells. In CRC, NOTCH3 mutation was characterized by increased proportions of central memory CD8⁺ T cells and M1 macrophages (Figure 3b). Cytotoxicity and immune scores were also higher in NOTCH3-mutant samples. The immune infiltration pattern in NOTCH3-low and NOTCH3-high expression samples resembled NOTCH3 mutation (Supplementary Figure S2). We further examined infiltration differences between NOTCH3-low and NOTCH3-high expression samples using CRC single-cell RNA sequencing data. After cell type annotation, cell proportions were analyzed (Figure 3c). NOTCH3-high samples showed a lower proportion of T cells including ZBTB+ T cell (19), CD8+T cell and γδ T cells (Figure 3c). Interestingly, total macrophage proportion was higher. NOTCH3 was primarily expressed in epithelial (cancer) cells (Figure 3d). The cell proportions of NOTCH3-low and NOTCH3-high samples were compared. CD8⁺ T cells and γδ T cells were more abundant in NOTCH3-low than in NOTCH3-high samples (Figure 3e). We also assessed the Spearman correlation between NOTCH3 expression and immune cell proportions. As expected, NOTCH3 expression was significantly negatively correlated with CD8⁺ T cells, γδ T cells, and ZBTB16⁺ T cells, but positively correlated with total macrophages (Figure 3f). In addition, CD8⁺ T cells from NOTCH3-low samples showed higher expression of KLRC2, HSPA6, and SH2D1B—markers associated with NK cells and CD8⁺ T cells—compared to those from NOTCH3-high samples (20) (Figure 3g). PD-L1 and PVR are key genes expression on cancer cells to inhibit the activity of CD8+ T cells, NK cells and γδ T cells. Thus, the correlation between NOTCH3 and these genes were also analyzed. Positive correlation of NOTCH3 and PVR expression in the scRNA-seq data (Figure 3h) was detected, but not PD-L1 (not shown). In summary, NOTCH3 expression is significantly associated with effector immune cell infiltration (CD8⁺ T and γδ T cells) and PVR expression.

We next investigated the impact of NOTCH3 on cell–cell interactions in CRC using scRNA-seq data. Overall, both the number and strength of cell–cell interactions were significantly greater in NOTCH3-high samples compared to NOTCH3-low samples (Figure 4a). Most interactions between cell types were present in both NOTCH3-low and NOTCH3-high samples (Figure 4b), though their interaction strengths and frequencies differed (Figures 4c, d). It is noted that Interactions involving CD8⁺ T cells and ZBTB1+ T cells were decreased in NOTCH3-high samples compared to NOTCH3-low samples (Figures 4c, d). In addition, we identified signaling pathways specifically detected in either NOTCH3-low or NOTCH3-high samples, including RESISTIN, SIRP, EGF, and IL1 pathways (Figures 4e, f). The PVR pathway was detected in NOTCH3-high samples but absent in NOTCH3-low samples (Figure 4f). Since effector cells directly mediate cytotoxicity against cancer cells, we analyzed interactions between cancer cells (epithelial) and natural killer cells, CD8⁺ T cells, macrophages, and γδ T cells. Although their most interaction patterns were similar, the PVR–TIGIT interaction was detected at significant level only in NOTCH3-high samples (Figure 4g). Taken together, NOTCH3 expression remarkably associated with cell-cell interaction profile, particularly interactions between cancer cells and CD8⁺ T cells via the PVR pathway.

We employed Gene Set Enrichment Analysis (GSEA) to assess the pathway-level differences between NOTCH3-wild-type and NOTCH3-mutant samples from The Cancer Genome Atlas (TCGA, Figure 5a). In most cancer types, NOTCH3 mutations were associated with elevated immune-related pathways, including the T cell receptor, B cell receptor, antigen processing and presentation pathways, and natural killer cell-mediated cytotoxicity in most cancers; this was also observed in CRC (Figure 5b). Our previous results have shown that NOTCH3 alterations (expression and mutation) are associated with survival, immune infiltration, and CD8⁺ T cell-related PVR pathways. Furthermore, scRNA-seq data revealed that the PVR pathway activity was significantly different in NOTCH3-low and NOTCH3-high samples. Thus, we hypothesized that NOTCH3 may regulate PVR expression to modulate CD8⁺ T cell cytotoxicity in CRC, as PVR is a classical immune checkpoint pathway that regulates CD8⁺ T cell activity (21). Both NOTCH3 and PVR expression were elevated in CRC compared to normal tissue (Figure 5c). PVR expression was significantly correlated with NOTCH3 expression in TCGA dataset (Figure 5d). Moreover, NOTCH3-mutant CRC samples showed significantly lower PVR levels in the TCGA dataset (Figure 5e). NOTCH3 R1669H mutation (the most frequent mutation site in the TCGA dataset, Figure 1b, Supplementary Figure S3) significantly reduced PVR expression compared to the wild-type (Figure 5f). We knocked down NOTCH3 in HCT116 cells and found that PVR expression was downregulated (Figure 5g). We then assessed CD8⁺ T cell cytotoxicity using an LDH cytotoxicity assay. When co-cultured with HCT116 cells overexpressing NOTCH3-WT, CD8⁺ T cell cytotoxicity was significantly decreased (Figure 5h). However, no significant difference in CD8⁺ T cell cytotoxicity was detected in the NOTCH3-MT group compared to the control. Cell viability analysis yielded results consistent with the LDH assay (Figure 5i). Knocking down PVR restored the effect of NOTCH3 (Supplementary Figure S4). In conclusion, NOTCH3 upregulates PVR expression and reduces CD8⁺ T cell cytotoxicity, while its mutation (R1669H) disrupts this function.

Since the activation of canonical NOTCH pathway initiates expression of downstream genes, we suspect that NOTCH3 also regulates PVR expression by binding to transcription factors. We screened the predicted transcription factors of PVR using the TFlink database (22) and NOTCH3-interacting genes from the STRING database (23). Five genes were identified in the intersection (Figure 6a), including HEY1, NOTCH1, RBPJ, RELA, and TP53. Among these genes, RBPJ is a classical downstream effector of the NOTCH family. Thus, we assayed whether NOTCH3 promotes the transcription of PVR via RBPJ. The co-immunoprecipitation (co-IP) assay revealed that NOTCH3 binds to RBPJ in HCT116 cell line (Figure 6b). Transfection of NOTCH3-WT significantly increased nuclear NOTCH3 levels compared to NOTCH3-MT (Figure 6c). Consistently, the overexpression NOTCH3-WT significantly also enhanced the nucleus RBPJ level, while not NOTCH3-MT (Figure 6d), suggesting that NOTCH3 binds to RBPJ and facilitate RBPJ for nucleus translocation. NOTCH3-WT overexpression also up-regulated the canonical downstream gene HEY1 and HES1 (Supplementary Figure S5) relative to NOTCH3-MT. We next analyzed whether RBPJ binds to the promoter region of PVR and initiates its transcription, as a transcription factor. The binding motif was retrieved from the JASPAR database and PVR promoter harbors a binding site for RBPJ (Figure 6e). Luciferase assay revealed that introduction of the wild-type PVR promoter region significantly enhanced luciferase activity in presence of RBPJ, while the mutant promoter showed no significant difference (Figure 6f). Meanwhile, overexpression of RBPJ also enhanced the expression level of PVR, on both mRNA and protein levels (Figure 6g). In addition, ChIP-PCR revealed that the PVR promoter region was significantly enriched in the presence of RBPJ (Figure 6h) compared to IgG. Furthermore, the presence of both RBPJ and NOTCH3-WT significantly enhanced PVR expression levels (Figure 6i). Knocking down of RBPJ significantly reduced the NOTCH3 induced PVR up-regulation (Supplementary Figure S6). In summary, NOTCH3 interacts with RBPJ, facilitate its nucleus translocation. RBPJ binds to the promoter of PVR and initiates PVR expression.

Since the PVR–TIGIT interaction is a key immune checkpoint for CD8⁺ T cell cytotoxicity, and the goal of immunotherapy is to activate CD8⁺ T cells by inhibiting PD-1/PD-L1/CTLA4 (also checkpoint genes), we investigated whether NOTCH3 blockade enhances the efficacy of immunotherapy in CRC. To test this hypothesis, we injected the CRC cell line HCT116 into C57BL/6 mice (proficient immune system) and treated them with IgG, anti-PD-L1, and/or anti-NOTCH3 antibody (Figure 7a; n = 5 per group). Tumor size was evaluated after 5 weeks. Anti-NOTCH3 antibody did not alter the mRNA expression of NOTCH3 or RBPJ as expected, while significantly reduced PVR expression in vivo (Figure 7b). Compared to the IgG group, anti-PD-L1 therapy reduced tumor size and volume. Moreover,​the combination of anti-PD-L1 and anti-NOTCH3 further enhanced the effect of anti-PD-L1 therapy (Figures 7c, d), on both tumor weight and volume. Consistently, Cd8a expression (marker for CD8+ T cell) was significantly increased in the combination group compared to single PD-L1 treatment (Figure 7e). In summary, NOTCH3 depletion enhances the efficacy of anti–PD-L1 treatment in a CRC mouse xenograft model.

We also analyzed the impact of NOTCH3 alterations (expression and mutation) on immunotherapy outcomes. Using the TISIDB database (a pan-cancer cohort of immunotherapy-treated patients), we compared survival between NOTCH3-high and NOTCH3-low patients (Figure 8a, N = 954). NOTCH3-low patients showed better survival than NOTCH3-high patients. We also evaluated survival differences in an independent cohort of 109 immunotherapy-treated patients (16 with NOTCH3 mutations) from the MSKCC immunotherapy cohort, comparing NOTCH3-mutant and NOTCH3-wild-type groups. The NOTCH3-mutant group demonstrated significantly prolonged survival compared to the wild-type group (Figure 8b, p<0.01). Cox multivariate regression confirmed NOTCH3 status as an independent prognostic factor (Figure 8c). We also retrospectively analyzed 102 immunotherapy-treated patients from our institution and other centers. NOTCH3 expression was quantified using immunohistochemistry on tissue microarrays, which revealed overexpression in tumor tissue compared to normal tissue (Figure 8d), consistent with previous findings. Although NOTCH3 expression was not significantly associated with clinical indicators such as age, sex, or stage (Figure 8e), Cox multivariate regression also identified it as an independent predictor of survival in immunotherapy-treated CRC patients (Figure 8f). When patients were stratified into NOTCH3-high and NOTCH3-low groups, the NOTCH3-high group showed significantly worse survival (Figure 8g). Together, these results demonstrate that NOTCH3 serves as a predictive biomarker for immunotherapy response in CRC.

This study was approved by the Ethics Committee of Henan Cancer Hospital conducted under Declaration of Helsinki. We enrolled CRC patients hospitalized at our institution between 2014 and 2022 in this study. Written informed consent was obtained from all participating individuals. All samples involved are MSI CRC according to the treatment guideline, using Candonilimab (dual anti-PD-1/CTLA-4). In the MSKCC dataset, the samples are also mostly MSI, with differed treatment drugs (anti-PD-(L1) and/or anti-CTLA4). Multiple ‘omics’ data and phenotype data was obtained from the cBioPortal database (24). Gene expression data were log2-transformed, and synonymous mutation data were filtered from the genomic data. For single-cell sequencing data, raw count data from GSE178341 (19) was downloaded via the Gene Expression Omnibus (GEO) database along with the annotated information. In this analysis, only tumor samples were used. Samples originating from different sections of the same patient were treated as distinct samples.

After data retrieval and pre-processing, quality control was performed using the R package ‘Seurat’ (25). Cells were clustered and annotated according to previously reported methods. Pseudo-bulk RNA-seq was performed using Seurat::AggregateExpression to generate a raw count matrix. After transforming the raw count matrix to counts per million (CPM) values, samples were stratified into NOTCH3-low and NOTCH3-high cohort based on the median CPM value of NOTCH3. Cell proportions were calculated according to the number of each cell type in the corresponding samples. Cell proportion differences were evaluated using the Wilcoxon test. The correlation of NOTCH3 CPM values and cell proportions was calculated using Pearson correlation. For cell-cell interaction analyses, the interaction number and strength between NOTCH3-low and NOTCH3-high samples were compared using the R package “CellChat” (26), following the recommended steps and default parameters.

For immune cell infiltration estimation, log2-transformed TPM values from the TCGA dataset were analyzed using the TIMER2 website (27) which incorporates multiple estimation algorithms (CIBERSORT (28), XCELL (29), QUANTISEQ (30), etc.), yielding a pan-cancer immune infiltration matrix. Infiltration differences between groups were compared. For GSEA analyses, samples for each cancer type were stratified into NOTCH3 wild-type (NOTCH3-WT) and NOTCH3 mutant (NOTCH3-MT) groups based on non-synonymous mutation status. Log2-transformed fold changes and corresponding p-values between groups were calculated. Following gene ranking by log2 fold change, Gene Set Enrichment Analysis (GSEA) was carried out and visualized utilizing R packages ‘clusterProfiler’ (31).

To construct a NOTCH3 (R1669H) mutant (NOTCH3-MT) overexpression plasmid, we first obtained a wild-type (WT) NOTCH3 overexpression plasmid (pcDNA3.1) as a template. Complementary primers for site-directed mutagenesis were designed with the mutation centered within 35 bp of the primer sequences. Amplification was implemented with DNA polymerase. The product was then transformed into competent E. coli cells, plated on antibiotic-containing media, and positive colonies were screened. The successful introduction of the R1669H mutation was confirmed by Sanger sequencing of the target region, and plasmid integrity was verified through restriction enzyme digestion analysis. For functional studies, the confirmed mutant plasmid was transfected into target cells for expression analysis.

For WB, cells were lysed and centrifuged, 150 μL of Dilute Buffer was added to the tube, which was vortexed, incubated for 5 min, centrifuged (12,000, 15 min, 4 °C). The clarified supernatant (500 μL) was decanted into new tubes, mixed with 500 μL of pre-cooled isopropanol, and vigorously vortexed for 15 s, and incubated for 10 min. It was centrifuged, and the upper phase was removed. It was washed using pre-cooled 75% ethanol, mixed 5 times, and centrifuged. The supernatant was discarded, and the pellet was washed again with 75% ethanol. The RNA pellet was dried in a 37 °C heating block for 2 min and dilute to ~500 ng/μL, followed by additional drying at 37°C for 5 min. RNA concentration was measured. A 1% agarose gel was prepared, and 4 μL of 1 kb Plus DNA ladder was loaded. Electrophoresis was performed at 120 V for 25 min. Total RNA (1 μg) was reverse transcribed, and 140 μL of nuclease-free water was added. Primers were diluted to 10 μM. qPCR was performed to quantify reference and target genes, and gene expression value was normalized based on the endogenous control.

For protein analysis, separating and stacking gels were prepared, loaded, and electrophoresed at 80 V, followed by an increase to 120 V. The gel was rinsed, and wet transfer was performed at 80 V in ice-water for 1.5–2 hours. The PVDF membrane was removed and stained with Ponceau S, and washed. Blocking was performed with 20 mL of 5% BSA for 1.5 h. After removal the solution, the membrane was briefly washed. BSA diluted primary antibody was added and incubated for 2 hours. Secondary antibody diluted in 5% BSA was added and incubated for 1.5 h. The secondary antibody was discarded, and the membrane was placed in a washing box, rinsed three times with 20 mL of 1× TBST (15 min per wash). The membrane was then transferred to a chemiluminescence imager, and luminescent signals were acquired after adding the substrate.

For luciferase assays, cells were properly seeded in medium. For transfection, 1 µL of TSnanofect V1 was mixed with 49 µL of antibiotic-free, serum-free medium or PBS to prepare a 50 µL solution, which was then combined with 50 µL of plasmid DNA and incubated for 5 minutes. Subsequently, 100 µL of the transfection mixture was added to each well, and the total volume was adjusted to 500 µL with fresh medium. After 48 hours of incubation at 37°C, gene expression levels were analyzed. The culture medium was aspirated, and cells were lysed by adding 100 µL of passive lysis buffer (PLB), followed by gentle agitation for 15 minutes. A 20 µL aliquot of cell lysate was transferred to a luminometer tube for luciferase activity measurement. Firefly luciferase luminescence signals were recorded, after which 100 µL of Stop & Glo^® reagent was used to assay Renilla luciferase activity.

For ChIP-PCR, cells were cultured to appropriate confluence, and PCR primers were designed to flank the RBPJ binding sites. Following the provided manual for the Simple ChIP^® Enzymatic Chromatin IP Kit (Agarose Beads) (Cell Signaling, USA), cells were cross-linked using formaldehyde, and chromatin was prepared. Nuclei were isolated and chromatin was enzymatically digested to optimal fragment sizes. Chromatin was immunoprecipitated overnight with specific antibodies bound to protein G agarose beads. After elution and reversal of cross-links, DNA was purified and verified with agarose gels.

The Animal Care and Use Committee of The Affiliated Cancer Hospital of Zhengzhou University has approved this work. Immune proficient male mice (5 weeks, ​C57BL/6) mice were acquired from Shanghai Model Organisms and housed in the experimental animal center of our institution. Mice were subcutaneously implanted with HCT116 CRC cells, followed by treatment with anti-PD-L1 antibody (10 mg/kg, administered every other day; Bio X Cell, USA) and/or NOTCH3 antibody (10mg/kg, 55114-1-AP, injected every the other day, proteintech, China) beginning two week post-injection. Upon completion of the treatment protocol, all animals were humanely euthanized via controlled CO2 asphyxiation at a flow rate of 30% of the chamber volume per minute (AVMA guideline) to facilitate surgical resection of xenograft tumors for subsequent analysis.

For NOTCH3 quantification, tissue microarrays (TMAs) were generated internally following established protocols. Immunohistochemical analysis was performed on 7-μm tissue sections obtained from formalin-fixed paraffin-embedded tissue. Initial processing included sequential deparaffinization and rehydration using a graded ethanol series. After antigen retrieval, peroxidase was blocked with 3% H₂O₂ for 20 minutes. Primary antibody against NOTCH3 was applied and incubated at 4 °C overnight. Sections were then incubated with secondary antibodies, and three 5-minute washes with PBS. Diaminobenzidine (DAB) substrate (5 mL) was applied for 10 minutes for chromogenic development. Microscopic evaluation was conducted using a Leica DMRXA2 system with Leica Application Suite software. Staining intensity was quantified using ImageJ with the IHC Profiler plugin.

For LDH detection, cytotoxicity was measured in cell-free co-culture supernatants using the Lactate Dehydrogenase Cytotoxicity Assay Kit (Beyotime, C0016) fallowing the manual. NC (negative control), NOTCH3-WT, and NOTCH3-MT cells were co-cultured with CD8⁺T cells (IPHASE, 071A403.11). After 3 hours of co-culture, LDH concentration was determined following the manufacturer’s protocol.

We carried out all statistical analyses with R software (version 4.4.1). For comparisons between categorical and continuous variables, Student’s t-test and Wilcoxon rank-sum test were implemented according to variance homogeneity. Correlation analyses were conducted using Pearson correlation when parametric assumptions were satisfied; or Spearman’s rank correlation was applied. Three replicates were set for the experiments results. Throughout this study, a p-value < 0.05 was considered statistically significant.