Human colon carcinoma tumor specimens and matched tumor adjacent non-neoplastic colon tissues were provided by Dr. Shannon McCall, Ms. Aubrey Coulas and Ms. Kellie Soafer at the Cooperative Human Tissue Network Southern Division (BioRepository & Precision Pathology Center, Duke University, Durham, NC). The patient pathological and treatment reports are presented in online supplemental table S1. Human colorectal cancer patient datasets were extracted from UCSC xenabrowser TCGA Colon and Rectal Cancer (COADREAD) database. Survival analyses were calculated using the R survival and survminer packages. Datasets from human patients with melanoma with nivolumab immunotherapy were obtained from the BMS038 trial (https://www.ncbi.nlm.nih.gov/pubmed/29033130, GSE91061). All studies with human specimens were performed under a protocol approved by Augusta University Institutional Review Board (#1354508-1).

C57BL/6J, Nfkb1 KO (p50 KO, B6.Cg-Nfkb1^tm1Bal /J), Rela^fl mice (B6.129S1-Rela^tm1Ukl /J) and Lck-cre (B6.Cg-Tg(Lck-cre)548Jxm/J) were obtained from the Jackson Laboratory (Bar Harbor, ME). WT and p50 KO chimera mice were created by adoptively transferring bone marrow cells from WT and p50 KO mice to lethally irradiated (900 RAD) recipient C57BL/6J mice. All animal studies are approved in advance by Augusta University Institutional Animal Care and Use Committee. The murine colon epithelial cell line MC38 were provided by Dr. Jeffrey Schlom. EL4 T cells were obtained from ATCC.

Mice were injected with azoxymethane (AOM, Sigma-Aldrich, 10 mg/kg body weight) intraperitoneally once, followed by administration of 2% dextran sulfate sodium (DSS; MP Biomedicals, Santa Ana, CA) through the drinking water 1 day later for 1 week. DSS was replaced with water for 2 weeks. The DSS/water cycle was repeated two more times. Mice were maintained with regular drinking water for one more month after the third cycle and then sacrificed for analysis. Colon tissues were dissected from the mice, flushed and cleaned with PBS, and cut open longitudinally to examine tumor nodules. Anti-PD-1 immunotherapy in the MC38 tumor model was performed as previously described.22

Immunohistochemistry was performed essentially as described previously.22 Human and mouse tissue sections were probed with Anti-p50 antibody (Santa Cruz) overnight at 4°C, followed by incubation with HRP Universal Anti-Mouse/Anti-Rabbit IgG Antibody (Vector Laboratories, Burlingame, CA) according to the manufacturer’s instructions.

The subunits of NF-κB complexes have contrasting functions in tumor development.12 14 15 19 24 We first aimed at profiling inflammation-driven NF-κB activation using a DNA probe with κB consensus sequence for all five Rel subunits. In the colitis model, p50 activation was predominant in the inflamed non-neoplastic colon over the other four NF-κB Rel subunits (figure 1A). A similar p50 activation profile was also observed in the colon carcinoma of the AOM–DSS colon cancer model (figure 1B). It seems that p50 also primarily exist as p50–p50 homodimer in the inflamed colon and colon carcinoma since no significant activation of the other four Rel subunits were detected (figure 1A,B). DSS-mediated inflammation induced p50 activation in the colon in a time-dependent manner (figure 1C, online supplemental figure S1). In addition, the degree of colonic inflammation is correlated with p50 activation (online supplemental figure S1). The p50 protein level also increased with time and correlated with increased activation and the degree of colonic inflammation (online supplemental figure S1). The colon-infiltrating immune cells have higher level of p50 protein than the colon epithelial cells (online supplemental figure S1).

Ten matched pairs of human colon carcinoma and adjacent non-neoplastic colon tissues were analyzed by immunohistochemistry. The nuclear staining of p50 was analyzed and used as marker of p50 activation in epithelial, tumor and immune cells. p50 protein level is higher in colon epithelial cells than in tumor cells (figure 1D,E) and also p50 activation level is higher in colon epithelial cells than in the tumor cells (figure 1D,E). Furthermore, immune cells have higher p50 activation than epithelial cells and tumor cells, respectively (figure 1D,E). As a complimentary approach, analysis of p50 mRNA expression level of human colorectal carcinoma and adjacent non-neoplastic colon from the TCGA database also determined that p50 expression level is significantly higher in the adjacent non-neoplastic colon tissue than in colorectal carcinoma in human patients with colorectal cancer (figure 1F).

The p50 NF-κB has a contrasting role in tumor development.12 14 15 17–19 24–27 We next aimed at determining the function of p50 in colon tumorigenesis. Colon tumorigenesis incidence was significantly lower in mice with global p50 deficiency as compared with WT mice (figure 2A). p50 deficiency also resulted in a significant decrease in Methylcholanthrene (MCA)-induced tumorigenesis (figure 2B). However, chimera mice with p50 deficiency only in immune cells also exhibited a significant decrease in colon tumorigenesis incidence (figure 2C). Consistent with p50 intrinsic function in immune cells as a tumor promoter, p50-sufficient colon tumor grew significantly slower in mice with p50 deficiency as compared with WT mice (figure 2D,E). We next took advantage of the existing p65-floxed mice and generated mice with p65 deficiency only in T cells (p65 TKO mice). As observed in the p50 chimera mice, the p65 TKO mice developed significantly less tumor than the control mice (figure 2F).

To dissect the relative functions of p50 in tumor cells from immune cells, Nfkb1 gene was knocked out in CT26 cells using CRISPR (online supplemental figure S2A). EMSA confirmed that p50 activation is diminished in p50 KO CT26 cells (online supplemental figure S2A). Interestingly, the p65 NF-κB activation was also diminished in the p50 KO tumor cells (online supplemental figure S2A). In contrast to what was observed in p50 global KO mice, knocking out p50 only in tumor cells significantly increased tumor nodule number in WT mice (online supplemental figure S2B). As in the CT26 cells, p65 activation was also diminished in the p50-deficient MCA tumor cells (online supplemental figure S2C). Again, p50 deficiency only in tumor cells significantly increased tumor growth in vivo (online supplemental figure S2D & 2E). We next blocked the canonical NF-κB activation in CT26 and MC38-met tumor cells using the IκBα-AA super suppressor.28 Blocking NF-κB activation significantly increased tumor nodule numbers in both CT26 and MC38-met experimental metastasis tumor models (online supplemental figure S2F). Taken together, our findings indicate that p50 tumor cell intrinsic signaling acts as a tumor suppressor. p50 immune cell intrinsic function overweighs its intrinsic function in tumor cells.

Our aforementioned findings determined that p50 has opposing functions in T cells and tumor cells in tumor growth control in vivo. To determine the relative contributions of p50 in immune cells and tumor cells, we transplanted WT and p50-deficient tumor cells to NSG mice. As in the immune competent mice (online supplemental figure S2D & 2E), p50-deficient tumor also grew significantly faster than WT tumor in NSG mice (online supplemental figure S3A). However, the degree of increased p50-deficient tumor growth is significantly less in NSG mice as compared with that in C57BL/6 mice (online supplemental figure S3B). This observation indicates that p50 intrinsic function in immune cells overwhelms its intrinsic function in tumor cells in vivo.

CD8⁺ CTLs are the primary T cells that mediate host cancer immunosurveillance. Although p50-deficient mice exhibited a decrease in tumorigenesis and tumor growth, analysis of the tumor tissues revealed that p50 deficiency did not increase tumor-infiltrating CTL level in colon tumor-bearing mice (online supplemental figure S4), suggesting that p50 may negatively regulate tumor-infiltrating CTL function in the tumor microenvironment. To test this hypothesis, tumor tissues were dissected from the WT and p50 KO mice as shown in figure 2A and analyzed by RNA-Seq (figure 3A). The differentially expressed genes were then functionally grouped and genes with known function in immune system process were filtered out (figure 3B,C) and validated by qPCR (figure 3D). We further analyzed these genes in tumor-infiltrating CD45⁺ cells. Five genes were upregulated and four genes were downregulated in p50 KO tumor-infiltrating CD45⁺ cells as compared with WT CD45⁺ cells (figure 3E). One of the upregulated genes is Gzmb (figure 3C–E). p50 negatively regulates granzyme B⁺ tumor-infiltrating CTLs.

Granzyme B is the master CTL effector that executes CTL anti-tumor cytotoxicity.29 The aforementioned observations prompted us to test the hypothesis that p50 negatively regulates Gzmb expression to decrease granzyme B⁺ tumor-infiltrating CTLs to promote tumor development. To test this hypothesis, tumor-infiltrating CD8⁺ T cells were analyzed for granzyme B expression. It is clear that granzyme B⁺ tumor-infiltrating CD8⁺ T-cell level is significantly higher in p50-deficient mice than in WT mice (figure 3F,G). Granzyme B protein level of these CD8⁺ granzyme B⁺ CTLs is also significantly higher in p50-deficient mice than in WT mice (figure 3F,G). In vitro activated p50-deficient CD8⁺ T cells also have a significantly higher level of granzyme B⁺ cells than those of WT CD8⁺ T cells (figure 3H). The granzyme B Mean Fluorescence Intensity (MFI) is also higher in p50-deficient CD8⁺ T cells than that in the WT CD8⁺ T cells (figure 3H). Taken together, our findings indicate that p50 represses Gzmb expression in tumor-infiltrating CTLs.

The previously mentioned observations indicate that p50 may directly repress Gzmb expression in T cells. We screened the murine Gzmb promoter DNA sequence and identified eight putative κB consensus sequence elements (figure 4A). Analysis of DNA–protein interaction using these κB sequence-containing DNA probes and T-cell nuclear extracts determined that p50 binds to one of these eight putative κB DNA elements (figure 4B and online supplemental figure S5). ChIP analysis determined that p50 binds to the chromatin region containing this unique κB DNA sequence at the Gzmb promoter in T cells (figure 4C). Taken together, we have identified a previously uncharacterized κB sequence element in the Gzmb promoter and determined that p50 directly binds to this specific κB DNA sequence at the Gzmb promoter to repress Gzmb expression in T cells.

To determine whether the aforementioned findings can be translated to human cancer, we analyzed the genomic dataset in human TCGA database. Because Gzmb is repressed by p50 in mouse colon carcinoma, we focused on p50-repressed genes. We extracted the top upregulated genes in p50 KO tumor from our RNA-Seq dataset (figure 3A) and filtered out these genes against the known human NF-κB target genes.30 Twenty-one human genes were generated and defined as p50 signature (figure 5A,B). We defined p50 score as the level of upregulation of these 21 p50 signature genes in p50 KO tumor. The lower the p50 signature gene expression, the higher p50 activation/p50 score. As observed in mouse models, p50 activation is higher in tumor adjacent non-neoplastic colon tissues as compared with the tumor tissues in human patients with colorectal cancer since p50 score is lower in colorectal carcinoma than in adjacent non-neoplastic colon tissues (figure 5C). Furthermore, p50 score is inversely correlated with GZMB expression level in human colorectal carcinoma (figure 5D), indicating that p50 represses GZMB expression in human colorectal carcinoma.

To determine whether p50 regulates T-cell tumor infiltration in human patients with colorectal cancer, we used the CIBERSORT algorithm to deconvolute CD8⁺ T-cell infiltration.31 p50 score high human colorectal carcinomas have a lower T-cell infiltration and p50 score low human colorectal carcinoma exhibit a higher T-cell infiltration (figure 5D). This finding indicates that p50 suppress T-cell tumor infiltration in human colorectal carcinoma.

Our previously mentioned observations suggest that p50 activation may negatively regulate cancer patient response to anti-PD-1 immunotherapy. To test this hypothesis, we used WT and p50 KO mice to test the efficacy of anti-PD-1 immunotherapy in the MC38 tumor model. MC38 is a type of tumor that is responsive to anti-PD-1 immunotherapy.32 As expected, anti-PD-1 immunotherapy significantly suppressed the MC38 colon tumor growth in WT mice (figure 6A,B). p50 KO mice also showed a significantly decreased tumor growth as compared with the WT mice (figure 6A,B). However, anti-PD-1 immunotherapy did not further suppress the established MC38 tumor growth (figure 6A,B).

A statistically significant two-factor interaction between mouse strains and treatment was found for tumor volume (F(1,16)=11.45, p=0.0038) (online supplemental table S4). Tumor volumes were higher with IgG treatment than with anti-PD-1 immunotherapy in WT mice, but in p50 KO mice tumor volumes were slightly lower with IgG treatment than with anti-PD-1 immunotherapy. The Tukey multiple comparison post hoc tests indicated that IgG control mice had significantly higher tumor volumes than mice with anti-PD-1 immunotherapy (p=0.0034) and IgG control p50 KO mice (p=0.0192). There were no other significant differences (WT mice with anti-PD-1 vs p50 KO mice with anti-PD-1 (p=0.4958), WT mice with anti-PD-1 vs p50 KO mice with IgG (p=0.8299), WT mice with IgG vs p50 KO mice with anti-PD-1 (p=0.0604), p50 KO mice with anti-PD-1 vs p50 KO mice with IgG (p=0.9345)) (figure 6B).

A statistically significant two-factor interaction between strain and treatment was found for tumor weight (F(1,16)=18.21, p=0.0006) (online supplemental table S4). Tumor weight was higher in IgG control WT mice than with anti-PD-1-treated WT mice, but in p50 KO mice tumor weight was slightly lower with IgG than with anti-PD-1 treatment. The Tukey multiple comparison post hoc tests indicated that IgG control WT mice had significantly higher tumor weight than WT mice with anti-PD-1 immunotherapy (p=0.0003), p50 KO mice with anti-PD-1 immunotherapy (p=0.0225) and IgG control p50 KO mice (p=0.0075). There were no other significant differences (WT mice with anti-PD-1 vs p50 KO mice with anti-PD-1 (p=0.1586), WT mice with anti-PD-1 vs IgG control p50 mice (p=0.3637), p50 KO with anti-PD-1 vs IgG control p50 KO mice (p=0.9475)) (figure 6B).

There is a rich literature of immune checkpoint immunotherapy datasets from human patients with melanoma.33 34 To determine the role of p50 in human patients who respond to anti-PD-1 immunotherapy, we analyzed p50 activation and GZMB expression in human melanoma patient specimens taken pre-immunotherapy and during immunotherapy. Nivolumab immunotherapy significantly decreased p50 activation and increased GZMB expression in human patients with melanoma (figure 6C). Furthermore, GZMB expression level is inversely correlated with p50 activation level (figure 6D), and nivolumab-induced increase in GZMB expression level is significantly correlated with decreased p50 activation in human patients with melanoma (figure 6E).

In summary, our data indicate that inflammation in the tumor microenvironment activates p50 NF-κB that represses GZMB expression in tumor-infiltrating CTLs to induce CTL exhaustion/dysfunction. Anti-PD-1 immunotherapy may act through unleashing the p50-suppressed dysfunctional CTL subsets of tumor-infiltrating CTLs to suppress tumor growth.