2.1 NFKB2 expression is significantly elevated in advanced metastatic CRC and is positively correlated with patient survival

We first performed differential gene expression analysis using the 23 primary colorectal tumor samples and 19 colorectal hepatic metastasis samples from the GSE81558 database to identify genes enriched in CRC patients with hepatic metastasis (Figure S1A). These metastasis-upregulated genes were significantly elevated in the tumor necrosis factor (TNF) and NF-κB signaling pathways (Figure 1A). GSEA results revealed that the NF-κB signaling pathway had the highest enrichment score (Figure 1B). The activation of the NF-κB signaling pathway was also found in the 75 Stage I samples and 51 Stage IV samples of patients in the TCGA-COAD database (Figure S1B,C), suggesting that the abnormal upregulation of the NF-κB signaling pathway is closely correlated with advanced CRC. Further exploration of the expression levels of NF-κB signaling-associated genes demonstrated that NFKB2 was significantly upregulated in metastatic CRCs (Figure 1C). Based on the analysis of a small genetic sample of 10 paired CRC/adjacent non-tumor samples, NFKB2 was among the top-ranked deregulated genes and its expression was significantly upregulated (Figure S1D). NFKB2 showed significantly higher genomic copy numbers in CRC samples compared to those in the nontumor samples (Figure 1D,E), and the mRNA and proteins levels of NFKB2 were also upregulated in CRC samples (Figure 1F and Figure S2A). Moreover, high NFKB2 mRNA levels were correlated with bigger tumor size, poorer differentiation grade, and worse tumor node metastasis (TNM) stage (Table 1 & Table S1). Importantly, distinctly upregulated NFKB2 mRNA levels were observed in CRC samples with advanced hepatic metastases (Figure 1G). Furthermore, NFKB2 showed the highest mRNA levels in the CRC specimen in the hepatic metastasis loci, which was higher than that in the CRC specimen in situ and the non-tumor tissue from the same patient (Figure 1H), suggesting that NFKB2 may be one of the main factors driving CRC metastasis, especially CRC-hepatic metastasis. In particular, high NFKB2 expression levels remarkably correlated with poor clinical outcomes in CRC patients in the TCGA-COAD cohort (Figure S2B).

2.2 NFKB2 promotes CRC tumor formation in multiple mouse tumor models

Next, we further investigated the biological functions of NFKB2 in CRC cell line. The results indicate that the highly expressed NFKB2 in the NFKB2-overexpressing cell lines only exerted a subtle effect on the proliferation of HT29 cells, with an approximate 10% enhancement in proliferation observed on the seventh day (Figure S3A). Consequently, NFKB2 overexpression should have shown its ability on the tumorigenesis of HT-29 cells in vivo, but surprisingly, there was no obvious effects that NFKB2 promote the tumor growth of HT-29 xenografts in nude mouse models (Figure S3B). Considering the role of NFKB2 in the immune response,¹⁹ we further examined the effects of NFKB2 overexpression on CRC cells using syngeneic mouse models. The results revealed that Nfkb2 significantly promoted subcutaneous tumor formation in MC38-C57BL/6 mouse models and CT26-BALB/c mouse models (Figure 2A and Figure S3C). We also evaluated the promoting effects of NFKB2 on cell metastasis by establishing an orthotropic transplantation model and found that NFKB2 significantly promote hepatic metastasis in MC38 and CT26 CRC tumor models (Figure 2B and Figure S3D). On hematoxylin-eosin staining the number of metastatic foci derived from NFKB2 cells dramatically increased in the intestine, liver, and lung sections (Figure 2C–E). Taken together, these findings suggest that NFKB2 acts as an oncogenic driver in CRC development and progression.

2.3 NFKB2 participates in the NF-κB inhibitory signaling pathway and programmed death-1/PD-L1 blockade in CRC cells

NFKB2, a key regulator of intracellular TNFα signaling, has long been reported to be closely associated with immune-regulation,^20-22 which had no effect on HT-29 cell growth in our nude mouse models. As crosstalk between tumor cells and the immune system is generally considered a pivotal factor in CRC development,²³ we hypothesized that the murine immune system may contribute to the tumor-promoting function of NFKB2 in syngeneic mouse models. To explore the mechanisms underlying the tumor proliferative effect of NFKB2 in syngeneic mouse models, we analyzed the transcriptomic profiles of MC38 tumors in C57BL/6 mouse model by RNA-seq assays (Figure S4), and an obvious difference existed between the transcriptomic profiles of xenografts with or without Nfkb2 overexpression (Figure 3A). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis and hierarchical clustering heatmap analysis showed that the NF-κB pathway, PD-L1 expression, programmed death-1 (PD-1) checkpoint pathway, and JAK-STAT pathway were the top-ranked signaling pathways responding to Nfkb2 overexpression (Figure 3B). As previously reported, the mRNA expression of PD-L1 gene could be upregulated by TNFα.²⁴ We analyzed the PD-L1 promoter region in the JASPR database, and identified a recognition site of NFKB2 at the −458 site of PD-L1 promoter region (Figure 3C). Luciferase reporter assay further confirmed the regulation of NFKB2 on PD-L1 promoter that NFKB2 overexpression increased PD-L1 promoter activity and vice versa, NFKB2 knockdown significantly inhibited the promoter activity, in HEK293T and MC38 cells (Figure 3D). Considering the crucial role of PD-L1/PD1 signaling in T-cell exhaustion, the effects of NFKB2 on CD8⁺ T-cell exhaustion were further evaluated. Intrahepatic infiltrating immune cells were isolated from the entire livers from spontaneous MC38 mouse models and profiled using CyTOF, and 33 immune cell subsets with specific marker genes were identified by PhenoGraph analysis of single-cell protein expression profiles (Figure 3E). Pd-1⁺ T cells were significantly increased in the samples from Nfkb2-overexpressed MC38 mouse models, whereas the proportion of Cd8⁺ T and IFN-γ cells was significantly decreased in these samples (Figure 3F). Furthermore, flow cytometry (FACS) assays of the Cd45⁺ cells from in situ transplanted tumors of both MC38 and CT26 mouse models also revealed that Nfkb2 overexpression caused a significant increased proportion of Pd-1⁺ T cells, as well as a decreased proportion of Cd8⁺ T cells and a reduced IFN-γ secretion (Figure 3G). We further examined the impact of NFKB2 on immune cells within the tumor microenvironment. Our findings demonstrated a significant reduction in the expression of CD4, CD8, and granzyme B (GZMB) in Nfkb2-overexpressing MC38 tumor tissues and CRC tumor samples with high NFKB2 expression (Figure 3H). This indicates that NFKB2 suppresses T-cell infiltration and impairs T-cell effector functions in the tumor microenvironment. Although NFKB2 did not exhibit cancer-promoting biological functions in the nude mouse-based in vivo model (as mentioned in Figure S3B), it promotes T-cell exhaustion by regulating PD-L1 expression in CRC cells, thereby facilitating CRC progression.

2.4 STAT2 participates in the NFKB2-modulated PD-L1 expression in CRC cells

To explore the molecular mechanism underlying the biological functions of NFKB2 in CRC tissues, the immunoprecipitation (IP) assay followed by mass spectrometry were performed to identify the protein partners of NFKB2 (Figure 4A). Particularly in the CRC samples, 231 proteins that interacted with NFKB2 could be enriched into Jak/STAT signaling pathway, NF-κB signaling pathway, and innate immune response process by KEGG enrichment analysis (Figure 4B). Among these candidate proteins, STAT2, which had been proposed to play a positive role in the tumorigenesis,²⁵ showed a positive correlation with NFKB2 in the mRNA levels of the TCGA-COAD cohort (Figure 4C). Moreover, the interaction between NFKB2 and STAT2 were further confirmed in HT29 and HEK-293T cells (Figure 4D). To investigate the interaction sites between NFKB2 and STAT2, truncation experiments based on the distinct functional domains of NFKB2 and STAT2 sequences were performed. IP assay with deletion mapping of these two proteins revealed that the SH2 domain of STAT2 (475aa-706aa) was crucial for its interaction with NFKB2 (Figure 4E), and the N-terminal half region of the IPT domain (228aa−327aa) of NFKB2 was essential for its binding to STAT2 (Figure 4F). Then, we isolated the cytoplasm and nucleus protein from MC38 cells and performed IP experiments, respectively, to examine the major forms of NFKB2 that bind to STAT2. The results showed that NFKB2-p100 is the primary form of NFKB2 that binds to STAT2 in the cytoplasm, while NFKB2-p52 is the main form that interacts with STAT2 in the nucleus (Figure 4G). Additionally, the protein primarily involved in the interaction between NFKB2 and STAT2 is the phosphorylated form of STAT2, rather than total STAT2 (Figure 4H). Furthermore, NFKB2 overexpression increased the half-life of the STAT2 protein in HEK-293T and MC38 cells (Figure 4I), and consequently led to changed levels of total STAT2 and Tyr690-phosphorylated STAT2 (Figure 4J).

Moreover, Nfkb2 and Stat2 co-localized in the nucleus of MC38 cells (Figure 5A), and a highly conserved STAT2 recognition site was observed at the −171 site of PD-L1 promoter (Figure 5B). Not surprisingly, the PD-L1 mRNA levels correlated with the mRNA levels of NFKB2, and especially highly correlated with the mRNA levels of STAT2 in the TCGA-COAD cohort (Figure 5C). Considering the bioinformatics prediction of potential binding sites for both NFKB2 and STAT2 within the PD-L1 promoter region, sequences at −171 (Fragment1) and −458 (Fragment2) positions were designed to investigate the binding of STAT2 and NFKB2. ChIP experiments revealed that both NFKB2 and STAT2 complexes were able to bind at the −171 and −458 sites (Figure 5D), suggesting that NFKB2 and STAT2 may form a complex binding within the PD-L1 promoter region. Subsequently, we performed Re-ChIP experiments and observed that after two rounds of ChIP, effective binding was detected at both −171 and −458 sites, indicating stable complex formation of NFKB2 and STAT2 in the PD-L1 promoter region (Figure 5E). Luciferase assays were conducted to analyze whether the binding of NFKB2 and STAT2 as a complex could synergistically enhance PD-L1 transcriptional activity. The results demonstrated that the transfection of NFKB2 alone or STAT2 alone resulted in increased transcriptional activity of PD-L1. However, the co-transfection of NFKB2 and STAT2 showed a more significant enhancement of PD-L1 transcriptional activity, suggesting a synergistic effect between NFKB2 and STAT2 in promoting PD-L1 transcriptional activity (Figure 5F). Additionally, overexpression of STAT2 resulted in an increase in both PD-L1 mRNA levels and protein levels and vice versa (Figure S5A,B). Therefore, these findings confirmed that NFKB2 could interact with STAT2 and result in upregulation of PD-L1. As the amino acid sequences of the crucial regions for the interaction of these two genes are highly conserved in human and murine sources (NFKB2, https://www.ncbi.nlm.nih.gov/gene/4791/ortholog/?scope=7776; STAT2, https://www.ncbi.nlm.nih.gov/gene/6773/ortholog/?scope=32523), the evidence could reasonably be obtained on the immune cells in subcutaneous tumor-bearing models of MC38 and CT26 tumors by FCAS analysis, which confirmed the promoting effects of Nfkb2 on the proportion of Pd-1⁺ T cells, together with the inhibitory effects of Nfkb2 on the proportion of Cd8⁺ T cells and IFN-γ secretion, were remarkably reversed by Stat2 knockout (Figure 5G). In further spatial localization analysis, we observed that there is an increased expression of STAT2 and PD-L1 in tumor tissues overexpressing Nfkb2 as well as in CRC samples with high NFKB2 expression. Additionally, these samples showed decreased infiltration of CD8⁺ T cells (Figure 5H). Importantly, co-expression of NFKB2, STAT2, and PD-L1 was associated with poor prognoses in CRC patients in the TCGA-COAD database (Figure 5I). These results suggested the regulatory effects of the NFKB2-STAT2 protein complex on PD-L1 expression and provided evidence on the contribution of this protein complex on the inflammatory and immune responses in CRC.

2.5 NFKB2 inhibitor Rg5 impacts the tumorigenesis of CRC cells with blocking NFKB2-STAT2 interaction

Subsequently, specific inhibitors for NFKB2 were screened from ginsenosides (ALK, CK, PPD, Rg3, Rg5, Rb1, Rb2, Rh2, Rh4, Rk1, and CK), which were isolated and identified from ginseng,²⁶ a traditional Chinese medicine with a range of pharmacological efficacy including immune-enhancing effects. Luciferase reporter assays in a Hela-based p52-response-element reporter system showed that Rg5 significantly inhibited NFKB2 activities as a transcriptional factor (Figure S6A). Also, Rg5 significantly reduced the NFKB2 protein levels in CRC cells (Figure S6B). The direct interaction between Rg5 and NFKB2 protein were proved by microscale thermophoresis (MST) assay, with a binding affinity of 5.1 μM for these two molecules (Figure S6C). In particular, the tumor volumes were both significantly lower in the Rg5-administered MC38 tumor-bearing mice than those in the control mice (Figure 6A). Moreover, Rg5 treatment was administered to the humanized NCG mice carrying patient-derived xenograft (PDX) and resulted in the significant inhibition of the growth of these CRC xenografts (Figure 6B). As expected, TNFα significantly increased the transcriptional activity of the PD-L1 promoter. However, the presence of Rg5 led to a significant suppression of this activity. Furthermore, when the mutant NFKB2 recognition site was introduced, it attenuated the inhibitory effect of Rg5 on PD-L1 promoter activity (Figure 6C). Consistently, treatment with Rg5 led to a reduction in NFKB2 protein levels, STAT2 protein levels, STAT2 phosphorylation levels, Cd8 protein levels, and PD-L1 protein levels (Figure 6D). Furthermore, immunoprecipitation results confirmed that Rg5 treatment significantly inhibited the binding of Nfkb2 and Stat2 within MC38 cells (Figure 6E). These findings provide evidence for the regulatory influence of Rg5 on NFKB2 in modulating the transcriptional activity of the PD-L1 promoter. Based on the observed effects of Rg5 on Nfkb2, we hypothesized that Rg5 potentially binds to specific sites within the Nfkb2 protein. Through computational analysis, we identified that Rg5 could target the Arg52, Ser226, Leu249, Asp251, and Tyr285 sites of NFKB2 (Figure 6F). These sites mostly located in the N-terminal half region of IPT domain (228aa−327aa) of NFKB2 were essential for its binding to STAT2. Among these predicted sites, Leu249 and Tyr285 displayed the lowest binding energies with Rg5 by GMX-MM-PBSA analysis (Figure 6G), suggesting greater accessibility to Rg5. Furthermore, the mutation of the L249 site on NFKB2 can increase the transcriptional activity of the PD-L1 promoter more effectively, while the mutation of the Y285 site suppresses NFKB2-induced enhancement of the transcriptional activity of the PD-L1 promoter (Figure 6H). Co-immunoprecipitation assays were performed to confirm the impact of specific mutations in NFKB2 on its binding to biotin-labeled Rg5. The results revealed that mutations in the L249 or Y285 sites of NFKB2 significantly disrupted its association with biotin-labeled Rg5 (Figure 6I). Cumulatively, Rg5, a small-molecule drug targeting NFKB2, could lead to the disassociation of the NFKB2-STAT2 complex, thus interfering with the PD-1/PD-L1 signaling in CRC.

2.6 Synergy between Rg5 administration and PD-1 antibody treatment leads to the combination effects on CRC immunotherapy

Previous studies have reported that non-canonical NF-κB signaling can improve the therapeutic effects of antibodies targeting PD-1.²⁷ Given that Rg5, which directly targets NFKB2, could can significantly suppressed PD-L1 expression in CRC cells, we further evaluated the potential synergistic impacts of Rg5 treatment on the immunotherapy for PD1/PD-L1 checkpoint were further evaluated. Both Rg5 administration and PD-1 antibody treatment exhibited therapeutic effects on subcutaneous MC38 tumors in syngeneic mouse model, and the combination treatment of Rg5 and PD-1 antibody led to a more effective reduction in tumor volume than either monotherapy (Figure 7A). Moreover, the synergic therapeutic effects of Rg5 and PD-1 antibody were strongly illustrated by the survival time of the MC38-bearing mice (Figure 7B), which were much longer than those in either monotherapy group (32 days of median survival time for the control group, 54 days for the PD-1 antibody monotherapy group, 45 days for the Rg5 monotherapy group, and more than 120 days for the combination treatment group). Furthermore, the proportion of intratumor Cd8⁺ T cells and the IFN-γ levels in the xenograft samples from the combination therapy group were also significantly higher than those in the samples from the two monotherapy groups (Figure 7C). Immunofluorescence staining assays revealed that Rg5 treatment led to significant reduced protein levels of Nfkb2 and Pd-l1, in addition to the distinctive effects of PD-1 antibody on Pd-1/Pd-l1 signaling (Figure 7D, the above). Both Rg5 administration and PD-1 antibody treatment enhanced the proportion of intratumor Cd8⁺ T cells and the IFN-γ levels, and the combination therapy enhanced these effects (Figure 7D, the bottom). The reduced tumor volumes, together with the promoted intratumor CD8⁺ T-cell infiltration and IFN-γ levels, were also found in Rg5/PD-1 antibody-treated humanized PDX model in NOD/SCID mice (Figure S7A,B), supporting the immunotherapeutic effects of Rg5-PD-1 combined treatment. To further confirm the involvement of immune cells in the therapeutic effects of Rg5, the activity of Cd8⁺ T cells were nullified in MC38-bearing mice by specific CD8 antibodies. Strikingly, no more distinct changes in tumor volume were found in tumor-bearing mice lacking effective Cd8⁺ T cells, either in Rg5 monotherapy groups or the combination therapy group (Figure 7E), indicating that Cd8⁺ T cells are essential for the therapeutic effects of Rg5. As expected, tumor-bearing mice depleted of Cd8⁺ T cells showed no significant difference from the control mice in median survival time (Figure 7F). Collectively, these results demonstrated that Rg5 can synergistically promote the efficacy of PD-L1 antibodies in CRC treatment with the support of CD8⁺ T cells.

In conclusion, the findings of the present study demonstrated the frequently overexpressed NFKB2 in advanced CRC and lead to a promoting PD-1/PD-L1 signaling, thus promoting the immune escape and metastatic ability of CRC cells. NFKB2 interacts with STAT2, which results in an increase in phosphorylated STAT2 levels and the promotion of PD-L1 expression. Targeting NFKB2 significantly promotes the PD-1 antibody immune checkpoint blockade therapy, with the crucial contribution of CD8⁺ T-cell infiltration (graphic abstract image).

4.1 Cell lines and mice

The CRC cell lines MC38, CT26, and HT-29 were purchased from the American Type Culture Collection (ATCC) and cultured according to the manufacturer's instructions. These cells were characterized using short tandem repeat markers by Genewiz, Inc. and were confirmed to be mycoplasma free (last tested in 2017). Note that 5- to 6-week-old BALB/c nude, C57BL/6, BALB/c, NOD/SCID, and NOD-Prkdc^em26Cd52IL2^rgγem26Cd52/Gpt (NCG) mice were purchased from Gempharmatech and were housed under pathogen-free conditions, in temperature-controlled holding rooms (20−22°C), with 30%−34% humidity, and on a 12 h on (6 a.m.) and off light cycle (6 p.m.).

4.2 Human patient sample collections

We collected 79 paired human CRC tissues and NCTs at the Affiliated Hospital of Jiangnan University, and 86 paired CRC tissues and NCTs at Fudan University Zhongshan Hospital. These samples were the same as those used in our previous study.⁴⁹ In this study, we utilized different sample cohorts. Cohort 1 (Table S1 n = 70) consisted of DNA samples collected from paired human CRC tissues and corresponding non-cancerous tissues (corresponding to Figure 1D,E). Cohort 2 (n = 50) comprised RNA samples collected from paired human CRC tissues and corresponding non-cancerous tissues (corresponding to Figure 1F). Cohort 3 (n = 21) included RNA samples collected from CRC patients with liver metastasis (corresponding to Figure 1G,H). All patient materials were obtained with informed consent, and this study was conducted with the permission of the Clinical Research Ethics Committees.

4.3 CRC xenograft tumor model

For in vivo tumor formation assays, 2 × 10⁶ HT-29 cells were suspended in 200 μL serum-free DMEM and subcutaneously injected into the flank of each nude mouse (female BALB/c nude, five per group). Tumor sizes were measured twice a week as soon as the tumors were measurable, and tumor volumes were calculated using the following formula: V (mm³) = width² (mm²) × length (mm)/2.

4.4 CRC syngeneic mouse model and hepatic metastasis model

For the in vivo tumor formation assays, 5 × 10⁵ MC38 cells were suspended in 200 μL serum-free DMEM medium and subcutaneously injected into the flank of each C57BL/6 mouse (female , five per group). Note that 1 × 10⁶ CT26 cells were suspended in 200 μL serum-free DMEM medium and subcutaneously injected into the flank of each BALB/c mouse (female BALB/c, five per group). In Rg5 separate treatment assay, Rg5 was administered by gavage once daily at a dose of 30 mg/kg from the seventh day on.

For Rg5 and PD-1 antibody combination therapy, 7 days later, Rg5 was administered by gavage once daily at a dose of 30 mg/kg, and 100 μg/mouse anti-PD-1 (pembrolizumab, BE0273, BioX cell) was injected into the abdomen every 2 days. To deplete mCd8⁺ T cells in C57BL/6 mice, 200 μg/mouse anti-Cd8 depletion mAb (BE0117, BioX cell) was administrated intraperitoneally to C57BL/6 mouse 1 day before pembrolizumab treatment, followed by weekly intraperitoneal injections of anti-CD8 mAb for 25 days. IgG2b mAb (BE0086, BioX cell) was injected intraperitoneally into control C57BL/6 mice on the same days as the control.

In the orthotopic tumor model, female BALB/c or C57BL/6 mice aged 6−8 weeks were initially subcutaneously injected with either 1 × 10⁶ CT26-luciferase cells or 5 × 10⁵ MC38-luciferase cells into the right flank. Upon reaching a tumor size of approximately 500 mm³, the mice were euthanized, and the tumors were excised, cut into 1–2 mm fragments, and collected in cold PBS after removal of necrotic areas. Anesthesia was administered to the mice before surgery, and a tumor fragment was implanted onto the serosal layer of the cecum through a laparotomy procedure. Closure of the muscle and skin was performed using 5-0 absorbable suture and 5-0 monofilament nylon suture, respectively. The mice were administered an intraperitoneal injection of d-luciferin (2.5 mg/100 μL in PBS) for in vivo bioluminescent imaging using an IVIS system (PerkinElmer).

4.5 Humanized PDX NCG mouse models

Female NCG mice (5−6 weeks old) received myeloablation using busulfan (30 mg/kg, i.p.; B2635, Sigma-Aldrich) at the day before CRC-PDX transplantation. The injection of PBMCs (5 × 10⁶ cells/mouse, i.v. Schbio, PBMNC050C) was administrated on day 7. The engraftment levels of hCD45⁺ cells were determined 2 weeks post-PBMCs transplantation by flow cytometric quantification of peripheral blood hCD45⁺ cells. For humanized PDX (Hu-PDX) tumor experiments, Rg5 treatments were started when implanted PDX tumors reached a volume of 100–200 mm³. The control group received intraperitoneal injections of PBS. Rg5 was administered once daily at a dose of 30 mg/kg, and 100ug anti-PD-1 was injected into the abdomen every 2 days.

4.6 Statistical analysis

All results are presented as mean values ± SEM. The Student's t-test or one-way analysis of variance was performed to evaluate the differences between two groups or more than two groups, respectively, followed by Dunnett's multiple comparisons tests. Wilcoxon tests were used to analyze Mrna levels in paired human samples, and Mann–Whitney tests were used to analyze Mrna levels in grouped human samples. The Kaplan–Meier method and log-rank test were used to determine the differences in survival rates between the NF-κB signaling pathway in the low- and high-expression groups. All statistical analyses were performed using R v3.5.1. p-Values <  0.05 were considered statistically significant.

Detailed Materials and Methods are available in the Supplementary Information.