Sulindac Modulates the Response of Proficient MMR Colorectal Cancer to Anti–PD-L1 Immunotherapy

Currently, colorectal cancer remains the third most common cancer in both men and women, with an estimated 53,200 deaths per year (1). Although the overall survival of patients with colorectal cancer has improved dramatically over the past few decades, the 5-year survival rate for individuals with stage IV colorectal cancer remains lower than 15% (1). Given this poor overall outcome, there is a greater need to develop more effective and safer treatment options for advanced and metastatic colorectal cancer. Recently, clinical immunotherapy with immune-checkpoint inhibitors (ICIs), such as PD-1 and PD-L1 inhibitors, has shown remarkable inhibitory efficacy in a variety of cancers (2, 3). The PD-1 cell receptor is a T-cell co-suppressor receptor that is expressed primarily on immune cells, such as T cells and macrophages (4, 5); the ligand of PD-1, PD-L1, is mainly expressed on dendritic cells (DCs) and various types of tumor cells. Under human physiologic cellular conditions, it appears that the interactions of PD-1 binding with PD-L1 lead to a protective mechanism against the development of autoimmune types of diseases in the host (6). However, when activated T lymphocytes recognize tumor cells, elevated PD-L1 on the surface of tumor cells can bind to PD-1 of T lymphocytes, blocking the attack of immune cells and resulting in a failure of immune surveillance (7). Therefore, the central concept of ICI immunotherapy is to block the direct binding between PD-1 and PD-L1 and restore the optimal surveillance and attack ability of immune cells on tumor cells (8).

Colorectal cancer is characterized by a malignant transformation of cells involving a progressive accumulation of various genetic alterations, including chromosomal instability (CIN) and microsatellite instability (MSI; ref. 9). The former CIN accounts for approximately 80% of sporadic colorectal cancers with mutations in oncogenes and tumor suppressors, such as K-RAS, APC, and TP53 (10, 11). Furthermore, the mechanism of MSI is characterized by elevated microsatellites with intragenic mutations of short and tandemly repeated DNA sequences, estimated to be present in approximately 15% of sporadic colorectal cancers. In addition, MSI is regarded as a biomarker of deficient DNA mismatch repair (dMMR), an essential mechanism required for cellular repair of damaged DNA (9, 11). Intriguingly, several clinical studies have reported that only patients with colorectal cancer with dMMR, or high microsatellite instability (MSI-H), responded to ICI immunotherapy, specifically anti–PD-1 therapy (12–14). Of significance, ICI immunotherapy dramatically improved overall survival (OS) and progression-free survival (PFS) in patients with stage 4 colorectal cancer with dMMR/MSI-H, which accounts for less than 5% of all colorectal cancer cases. However, it did not appear to have a significant improvement in OS and PFS for those patients with proficient MMR (pMMR) and microsatellite stability (MSS; refs. 15–17). Based on these clinical results, the FDA approved pembrolizumab (anti–PD-1), nivolumab (anti–PD-1), and ipilimumab (anti–CTLA-4) in combination with other agents for the treatment of patients with dMMR/MSI-H metastatic colorectal cancer. Unfortunately, such options are available only for a very small group of patients with metastatic colorectal cancer.

We have a long-standing research interest in studying metastatic colorectal cancer, particularly examining the chemopreventative effects of sulindac, an FDA-approved nonsteroidal anti-inflammatory drug (NSAID) that has been shown to inhibit colorectal cancer development and progression (18). In this study, we utilized low doses of sulindac in combination with PD-L1 antibody (Ab) to treat CT26 syngeneic mice, which are characterized as pMMR colorectal cancer (19). We showed that CT26 mice treated with the combination therapy (sulindac + PD-L1 Ab) displayed an enhanced immune response with subsequent tumor regression, compared with the monotherapy, in terms of PD-L1 Ab or sulindac alone. Of interest, infiltrating CD8⁺ T-cell aggregates were found in tumor tissues of mice treated with sulindac alone or in combination, inversely correlated with PD-L1 expression. Regarding the mechanism of action, we demonstrated that sulindac can downregulate PD-L1 gene expression by blocking NF-κB signaling. Thus, our findings provide evidence that sulindac acts as an immunomodulator to enhance the sensitivity of pMMR colorectal cancer tumors to anti–PD-L1 therapy.

The murine colon cancer cell line CT26 and the human colon cancer cell lines Colo205, HT29, and Caco-2 were purchased from ATCC. We authenticate our cell lines at regular intervals of 12 months in addition to two authentications at the beginning and end of the project using the services provided by IDEXX BioAnalytics. The authentication includes short tandem repeat profiling, mycoplasma testing, and cross-species contamination checking. In this project, only early passages of the cell lines (<10) were utilized. Cell culture media were purchased from Thermo Fisher Scientific and used for cell culture after mixing with 10% fetal bovine serum (Thermo Fisher Scientific). Cells were cultured at 37°C and in a 5% CO₂ humidified incubator. Sulindac, CMC (carboxymethylcellulose), and NF-κB inhibitor (Bay11-7082) were purchased from Sigma-Aldrich. CMC was used as the vehicle of sulindac for in vivo study. Mouse anti–PD-L1 Ab and isotype control rat IgG2b were purchased from Bio X Cell. Sulindac sulfide was purchased from Alfa Aesar.

BALB/c immunocompetent mice were purchased from Charles River Laboratories. Animals were housed under specific pathogen-free conditions. All animal experiments were performed in accordance with a protocol approved by the Institutional Animal Care and Use Committee of Louisiana State University Health Sciences Center.

Orthotopic colorectal cancer mouse model with intrarectal (IR) injection was established as previously described (20). Briefly, after mice were anesthetized with 2.5% isoflurane, the anal canal was dilated with lubricated blunt-tipped forceps to expose the distal anal and rectal mucosa. Luciferase-labeled CT26 cells were injected into the distal posterior rectal submucosa, 1 to 2 mm above the anal canal, using a sterile 30-gauge removable needle on a 50-μL Hamilton Microliter syringe. A dissecting microscope was used to assist in the injection. We injected each mouse with 2 × 10⁵ luciferase-labeled CT26 cells in a volume of 10 μL.

Mice were monitored for tumor growth biweekly with the noninvasive IVIS Imaging System (PerkinElmer). In brief, mice were injected intraperitoneally with D-luciferin substrate (Promega) in Dulbecco's phosphate-buffered saline at 150 mg/kg. After 10 minutes, mice were anesthetized with 2.5% isoflurane and placed on a warm stage inside the light-tight camera box and continuously exposed to isoflurane. The exposure time was 1 to 5 seconds. The level of light emitted from the bioluminescent CT26 cells was detected, integrated, digitized, and displayed by the IVIS Imaging System. The region of interest from displayed images was quantified as photons per second.

When orthotopic tumors were detectable with IVIS, which was about 10 days after IR injection, 20 male and 20 female mice were randomly divided into four groups: (i) vehicle control treated with CMC + IgG2b; (ii) sulindac (15 mg/kg, p.o., b.i.d.); (iii) PD-L1 Ab (75 μg, i.p. M/W/F); (iv) combination of sulindac and PD-L1 Ab (7.5 mg/kg p.o., b.i.d. + 37.5 μg, i.p. M/W/F). Both female and male mice in each group were used equally. Mice were euthanized on day 27 when some animals showed the symptoms of bowel obstruction. Blood samples were reserved for exosome isolation, and tumor tissue samples were collected and fixed in 10% buffered formalin for paraffin embedding.

Immunofluorescence staining was performed on paraffin-embedded sections of CT26 tumor tissue and pMMR colorectal cancer lines (CT26, Colo205, HT29, and Caco-2). The paraffin-embedded sections were deparaffinized and rehydrated. After antigen retrieval and permeabilization, slides were blocked with 5% horse serum for 1 hour at room temperature and then incubated with the primary antibodies at 4°C overnight. To stain CD8⁺ T cells, we incubated tumor tissue slides with Alexa Fluor 647–conjugated CD8a Ab (BioLegend) overnight. For PD-L1 and NF-κB staining, tumor tissue slides were incubated with primary mouse PD-L1 Ab (Abcam) and primary NF-κB p65 Ab (Cell Signaling Technology), followed by Alexa Fluor 488 and 555–conjugated secondary Ab (Thermo Fisher Scientific) for 1 hour at room temperature, respectively. For immunofluorescence staining of all four pMMR colorectal cancer cell lines, we seeded cells in flow dishes at 37°C overnight and then treated them with 10 μmol/L Bay11-7082 for 2 hours and 25 μmol/L sulindac sulfide (SS) for 12 hours. DMSO at 0.1% was used as vehicle control. Mouse TNFα (Thermo Fisher Scientific) and human TNFα (Millipore) at a concentration of 25 ng/mL were added to the cells for incubation of 20 minutes, respectively. Then, cells were fixed by 4% formaldehyde (Alfa Aesar) for 10 minutes, followed by permeabilization with 1% Triton X-100 (Bio-Rad). After blocking with 1% BSA, cells were incubated with mouse PD-L1 Ab (Abcam), human PD-L1 Ab (Cell Signaling Technology), and NF-κB p65 Ab (Cell Signaling Technology) at 4°C overnight. After washing with PBS, cells were incubated with Alexa Fluor 488 (PD-L1) or 555 (p65)–conjugated secondary Ab (Thermo Fisher Scientific) for 1 hour at room temperature. Finally, after washing and staining with 5 ng/mL DAPI (Sigma-Aldrich), slides and cells were processed with Prolong Diamond Antifade Mountant (Thermo Fisher Scientific) and analyzed by confocal microscopy (Nikon Eclipse Ti2). We used ImageJ v1.48 (21) to quantify the expression of PD-L1 and CD8⁺ T-cell infiltration. For PD-L1, we calculated total corrected cell fluorescence (TCCF) using the formula: TCCF = integrated density − (area of selected cells × mean fluorescence of background readings). For CD8⁺ T-cell infiltration, we calculated the percentage of CD8⁺ T cells in the tumor tissue using the formula of the number of CD8⁺ T cells divided by the total number of surrounding tumor cells plus CD8⁺ T cells.

Referring to a published protocol (22), we analyzed the status of T cells in BALB/c splenocytes under different treatments in vitro using flow cytometry. First, we prepared single-cell suspensions from the spleens of BALB/c mice and performed mechanical dissociation. In brief, spleens were washed and minced in warm Dulbecco's Modified Eagle Medium (DMEM) and minced, and then the clumps were removed by passing through a Corning 40-μm cell strainer. The filtrate containing the single-cell suspensions was washed through DMEM and PBS and centrifuged at 300 × g for 10 minutes at room temperature. Next, we seeded 1 × 10⁶ cells per well in a 24-well plate and then treated with 1 μg/mL anti-mouse CD3e Ab (eBioscience) and 1 μg/mL anti-mouse CD28 Ab (eBioscience; positive control), 0.1% DMSO plus 200 nmol/L IgG2b (vehicle control), 25 μmol/L SS, 200 nmol/L anti-mouse PD-L1 Ab (Bio X Cell), and a combination of 25 μmol/L SS and 200 nmol/L anti-mouse PD-L1 Ab for 72 hours. After treatment, cells were blocked with Fc-blocked Ab (BD Biosciences) at 4°C, followed by live-dead fixable viability stain 450 (BD Biosciences) for 20 minutes and flow cytometry antibodies for 30 minutes, including APC anti-mouse CD45 Ab (BD Biosciences), BV711 anti-mouse CD3e Ab (BD Biosciences), PE-CF594 anti-mouse CD8a Ab (BD Biosciences), and APC-Cy7 anti-mouse CD4 Ab (BD Biosciences). Then, intracellular staining was performed using the BD GolgiStop (BD Biosciences) fixation and permeabilization solution kit. FITC anti-mouse granzyme B Ab (eBioscience) was incubated with the cells for 30 minutes. Cells were then washed once with PBS and fixed with 1% paraformaldehyde prior to analysis by flow cytometry.

Exosomes were extracted from mouse plasma samples and purified with the ExoQuick ULTRA EV isolation kit, which was purchased from System Biosciences. According to the manufacturer's instruction, 125 μL of plasma was pretreated with 1 μL of Thrombin to remove fibrin and then incubated with 33.5 μL of ExoQuick exosome precipitation solution for 30 minutes at 4°C. After centrifugation at 3,000 × g for 10 minutes, the exosomes were concentrated at the bottom of the tubes and then resuspended with buffer. Finally, exosome lysates were prepared by adding RIPA buffer (Thermo Fisher Scientific) and analyzed by Western blot.

Membrane proteins were isolated with the Mem-PER Plus Membrane Protein Extraction Kit (Thermo Fisher Scientific), and cytoplasmic proteins were extracted with the NE-PER Nuclear and Cytoplasmic Extraction Kit (Thermo Fisher Scientific). We quantified the proteins with the DC Protein Assay kit (Bio-Rad). Denatured proteins were separated on a 10% SDS-PAGE gel and then transferred to PVDF filters (Bio-Rad). After blocking with 5% non-fat milk-TBST (Bio-Rad), the blots were incubated mouse PD-L1 Ab (Abcam), human PD-L1 Ab (Cell Signaling Technology), β-actin Ab (Cell Signaling Technology), CD63 polyclonal Ab (Abcam), or Calnexin polyclonal Ab (Abcam) at 4°C overnight, respectively. After washing three times with 1× TBST (Bio-Rad), the blots were incubated with peroxidase-linked secondary goat anti-rabbit IgG Ab (Bio-Rad), or goat-anti-mouse IgG Ab (Bio-Rad) were incubated for 1 hour at room temperature. Finally, after incubation with SuperSignal West Pico PLUS chemiluminescent substrate (Thermo Scientific), visual imaging was performed by VersaDoc Imaging System (Bio-Rad). Volume One software (Bio-Rad) was used for the relative quantification of blotted proteins.

EZ-Magna ChIP kit (Millipore) was purchased for ChIP analysis. We strictly followed the manufacturer's instructions. In brief, 1 × 10⁷ CT26 cells were treated with 25 ng/mL TNFα for 20 minutes and then cross-linked with 1% formaldehyde (Sigma-Aldrich). After fixation and lysis, chromatin was ultrasonically sheared using preoptimized conditions, followed by immunoprecipitation with NF-κB p65 Ab (Cell Signaling Technology) or goat anti-mouse IgG at 4°C overnight. After extraction and purification of DNA fragments from the pulldown complexes, PCR was implemented using a thermocycler in a total volume of 20 μL, and the program included the following steps: 94°C, 2 minutes; 30 cycles including denaturation at 94°C for 20 seconds, annealing at 59°C for 30 seconds, and extension at 72°C for 30 seconds. Finally, PCR products were electrophoresed on a 2% agarose gel. The primer sequences are: forward primer: 5′CACTTCCAGTTCGCAGAA 3′; reverse primer: 5′ CAAGCAAATGACTCAGTTTTA 3′.

Mouse plasma PGE2 was measured with Prostaglandin E2 Parameter Assay Kit (R&D Systems) following the manufacturer's instruction. In brief, plasma samples (50 μL) were appropriately diluted and added to a 96-well polystyrene microplate coated with goat anti-mouse polyclonal antibodies. After incubation with 50 μL of mouse PGE2 monoclonal Ab for 1 hour at room temperature, the PGE2 conjugate was added to each well and incubated for 2 hours with gentle shaking. After washing, the substrate solution was added to each well and incubated for 30 minutes, and then the stop buffer was added to halt the color development reaction. Finally, the absorbance was measured at 450 nm with a microplate reader (Bio-Rad).

GraphPad Prism version 7.0 and SAS (9.4) were utilized for data analysis. Luciferase intensity data (photons/second) acquired from IVIS imaging were exhibited with the mean plus and minus standard error of the mean, and one-way ANOVA along with Tukey method were used to analyze the difference of tumor size at the endpoint (day 27). Student t test was used to evaluate the difference of tumor weight, PD-L1 expression, and infiltrating CD8⁺ T cells between two groups with Bonferroni adjustments. Simple linear regression was used to assess the percentage of CD8⁺ T cells and PD-L1 expression in tumor tissue, as well as the strength and direction of the linear correlation between exosomal PD-L1 expression and tumor weight. The goodness-of-fit measurement, R² was calculated to measure the strength of the predictor in explaining the outcome, and the P value of coefficients in the linear models was used to report the significance of the predictor.

We established a syngeneic mouse model using well-established techniques, specifically intrarectal (IR) injection of luciferase-labeled murine colorectal cancer CT26 cells. We then treated each group of mice with either sulindac alone (15 mg/kg, p.o., b.i.d.), PD-L1 Ab alone (75 μg, i.p., M/W/F), or a combination of sulindac and PD-L1 Ab (7.5 mg/kg p.o., b.i.d. + 37.5 μg, i.p. M/W/F), respectively. Bioluminescence signals of labeled CT26 cells were measured twice per week using a noninvasive IVIS Imaging System. We sacrificed mice on day 27 due to the development of intestinal obstruction in several mice. As shown in Fig. 1A, there was no significant difference in tumor growth in mice treated with vehicle control, sulindac monotherapy, and anti–PD-L1 monotherapy. However, the mice treated with the combination of sulindac and PD-L1 Ab were found to have the lowest luciferase signal levels during treatment, which were significantly different from the other groups. Autopsy of each mouse revealed the orthotopic tumors from each group, which were fully resected, removed, and weighed. Figure 1B shows the average weight of colorectal cancer tumors at the endpoint. We found that the mice treated with the combination of sulindac and PD-L1 Ab had the smallest tumors compared with the vehicle controls, which in turn translated to the lowest tumor weight (P < 0.05, t test). We also measured the total body weight of mice twice per week but did not find any significant differences between groups (Supplementary Fig. S1), supporting the safety of our treatment regimens.

A recent study reported that CD8⁺ T lymphocytes mediated the inhibitory activity of sulindac upon tumor cell growth, specifically in the syngeneic breast cancer (4T1) animal model (23). We collected tumor tissues to analyze CD8⁺ T-lymphocyte infiltration. Utilizing immunofluorescence imaging, we visualized infiltrating T lymphocytes (green) and stained PD-L1 with Alexa Fluor 488–conjugated secondary antibodies (red). As shown in representative merged confocal images (Fig. 2A; unmerged images are exhibited in Supplementary Fig. S2), the combination group treated with sulindac and PD-L1 Ab were found to have the most infiltrating CD8⁺ T lymphocytes (green) in the tumor tissues, compared with vehicle control and PD-L1 Ab–treated mice. Mice treated with sulindac alone showed an eightfold and a threefold increase of infiltrating CD8⁺ T lymphocytes compared with vehicle control and PD-L1 Ab–treated mice, respectively. These results support the notion that sulindac can facilitate T-lymphocyte infiltration in pMMR colorectal cancer tumors. Intriguingly, when analyzing PD-L1 expression in the same set of tumor samples, we found that PD-L1 expression is inversely correlated with infiltrating CD8⁺ T lymphocytes in all groups, with the vehicle control and combination groups displaying a statistical significance (P < 0.05; Fig. 2B).

Although our results show that sulindac monotherapy facilitates CD8⁺ T-lymphocyte infiltration and reduces PD-L1 expression in tumor tissues, comparing the results of in vivo inhibitory activity, as shown in Fig. 1B, it is unlikely that sulindac monotherapy exhibits compelling inhibition on tumor growth to the same extent as combination therapy. We speculated whether there were differences in the status of infiltrating CD8⁺ T lymphocytes induced by sulindac monotherapy and combination therapy. Utilizing flow cytometry, we tested our hypothesis and found that anti–PD-L1 monotherapy significantly activated CD8⁺ T lymphocytes (49.80% GranzB⁺), compared with sulindac monotherapy (14.71% GranzB⁺); the combination therapy showed a synergistic effect of two agents (72.19% GranzB⁺), as shown in Fig. 2C. A previous study reported that PD-L1 Ab or PD-1 Ab could increase mTOR signaling and granzyme B expression in virus-specific CD8⁺ T lymphocytes, resulting in faster clearance of lymphocytic choriomeningitis mammarenavirus (LCMV) infection (24). This study supports the notion that PD-L1 Ab can activate CD8⁺ T lymphocytes. Thus, we conclude that the synergistic effect of PD-L1 Ab and sulindac on CD8⁺ T-lymphocyte activation resulted in the most significant inhibitory efficacy of the combination therapy.

To further investigate the relationship between sulindac and PD-L1 Ab, we determined the inhibitory effect of sulindac on PD-L1 expression using mouse CT26 cells, as well as human colorectal cancer Colo205, HT29, and Caco-2 cells. All of these cell lines were characterized as pMMR colorectal cancer models (19, 25). After isolation of membrane and cytoplasmic proteins from all four cell lines, we compared the PD-L1 expression levels after the treatment of vehicle control and SS, the active metabolite of sulindac using Western blot assay. As shown in Fig. 2D, SS could effectively downregulate PD-L1 expression in the cell membrane and cytoplasm. Taken together, our results demonstrate that sulindac can facilitate the infiltration of CD8⁺ T lymphocytes in colorectal cancer tumors by downregulating PD-L1 expression.

We have previously demonstrated a novel mechanism by which sulindac inhibits tumor cell motility by blocking NF-κB signaling downregulation of selected oncogenic miRNAs (26). Recent studies have also reported the ability of NF-κB to regulate PD-L1 in different cancer cells (27–30). Because our in vivo data demonstrated that sulindac could downregulate PD-L1 expression in tumor tissues, we investigated if NF-κB signaling might also be involved in sulindac's inhibitory effect on PD-L1 expression in pMMR colorectal cancer.

Figure 3A exhibited a putative binding site for NF-κB (GGGAAGTTCT) within the promoter region of the PD-L1 gene, as reported previously (31). We further determined the binding of NF-κB to the promoter of PD-L1 using the chromatin immunoprecipitation (ChIP) assay. In brief, NF-κB p65 Abs were used to immunoprecipitate the sheared chromatin fractions extracted from CT26 cells. After purifying the DNA fragments from the pulldown complexes, we performed a PCR analysis with the designated primers to detect the target DNA fragments containing the putative NF-κB binding site (GGGAAGTTCT). As shown in Fig. 3B, we detected a band with the expected size using DNA templates extracted from the p65 immunoprecipitated complex. In addition, we used a dual-luciferase reporter assay to confirm the ChIP results. In brief, wild-type or mutated promoter fragments of the mouse PDL-1 gene were cloned into pEZX-LvPG04 vectors, and the dual-reporter system used Gaussia Luciferase (GLuc) as the promoter reporter and secreted alkaline phosphatase (SEAP) as an internal control for signal normalization. After transfection of these constructs into CT26 cells, we measured the signals of GLuc and SEAP and calculated the relative light unit (RLU) ratio. As shown in Supplementary Fig. S3, the RLU of the mutated constructs is much lower than that of the wild-type constructs upon TNFα treatment or NF-κB/p65 cotransfection (P < 0.05). Together, these results provide strong evidence that NF-κB binds directly to the promoter of PD-L1 in CT26 cells.

Furthermore, we utilized immunofluorescence imaging to analyze the regulation of PD-L1 expression following sulindac treatment. Using our published protocol, TNFα and Bay11-7082 were included to stimulate and block NF-κB signaling, respectively (26). All four colorectal cancer cell lines were utilized in this study. As shown in Fig. 3C and Supplementary Fig. S4–S7, TNFα treatment promoted translocation of NF-κB p65 into the nucleus, which in turn upregulated PD-L1 expression. Additionally, treatment with SS can effectively attenuate the induction of TNFα and blocked NF-κB signaling, which was comparable with that of the NF-κB inhibitor Bay11-7082. We also extended our study by examining NF-κB p65 and PD-L1 in tumor tissues that we analyzed above. As expected, mice treated with sulindac alone, or in combination with PD-L1 Ab, showed reduced NF-κB signaling in the nucleus along with PD-L1 downregulation (Fig. 3D; Supplementary Fig. S8), which is consistent with the in vitro results.

A recent study reported that exosomal PD-L1 could significantly affect the efficacy of PD-L1 Ab (32). After extracting exosomes from mouse plasma samples, we examined exosomal PD-L1 expression in mice from each treatment group. The exosome marker CD63 was used as the endogenous control to normalize the expression of PD-L1. As shown in Fig. 4A, the expression of exosomal PD-L1 was decreased in all treated mice compared with the vehicle control group, and the lowest plasma exosomal PD-L1 expression was observed in the combination group. Supplementary Fig. S9 exhibits the results of Western blot and relative quantitation of exosomal PD-L1 in mice. We further analyzed the correlation between exosomal PD-L1 and tumor progression using a linear model. The values of tumor weight were used to indicate tumor progression. As shown in Fig. 4B, only the combination group revealed a positive correlation, which was statistically significant (R² = 0.81; P = 0.01). Intriguingly, the PD-L1 Ab group showed a decrease in exosomal PD-L1, but a weak correlation to tumor weight (R² = 0.23; P = 0.23). This result implies that PD-L1 Ab may be competitively occupied by circulating exosomal PD-L1, leading to a lack of sufficient amount of PD-L1 Ab available to effectively block immune checkpoints by binding to PD-L1 on the surface of colorectal cancer tumor cells. In combination therapy, sulindac can downregulate PD-L1 expression, ultimately resulting in less PD-L1 being released into circulation via exosomes and relatively more PD-L1 Ab being available to approach colorectal cancer tumor cells, not only to bind to PD-L1 on the cell surface but also to activate CD8⁺ T cells, as shown in Fig. 2C.

The FDA-approved dose of sulindac for human use is 300 to 400 mg/day. Utilizing a published protocol (33), we calculated equivalent doses for the small animal study, which are 61.5 to 82.0 mg/kg. In this project, we only utilized doses of 15 and 7.5 mg/kg to treat mice, which are considered low doses of sulindac. To evaluate the systematic toxicity of sulindac, specifically the inhibition of PGE2, we utilized plasma samples collected from CT26 mice and assayed the concentration of PGE2 by ELISA. As shown in Fig. 5, there was no significant difference between the vehicle control and the drug-treated mice. These results demonstrate that low-dose sulindac is a safe agent to modulate the response of pMMR colorectal cancer to ICI therapy, especially to anti–PD-L1 immunotherapy. 

The clinical benefit of checkpoint immunotherapy is well documented, with improvements in both long-term and OS seen primarily in the treatment of metastatic melanoma and lung cancer, and recently extended to metastatic triple-negative breast cancer (TNBC; ref. 34). However, clinical studies have shown that the advantage of ICI immunotherapy for colorectal cancer is only confined to a small subset of patients with dMMR or MSI-H, representing less than 15% of patients with colorectal cancer (12–14). Unfortunately, for most patients with colorectal cancer carrying pMMR or MSI-L/MSS, current ICI regimens offer little clinical benefit in terms of patient survival.

Long-term use of NSAIDs has been shown to significantly reduce the incidence and risk of death from colorectal cancer and other cancers in both clinical and preclinical settings (35–39). In particular, sulindac is highly effective for treating precancerous lesions in patients with familial adenomatous polyposis and reducing the size and number of polyps by approximately 60% to 70% (18). Our previous studies have also exclusively investigated the anticancer activity of sulindac and several derivatives (26, 40–45). In this study, we aimed to determine if sulindac could modulate the response to ICI immunotherapy of pMMR colorectal cancer. We chose CT26, a murine colorectal cancer cell line that was characterized as pMMR (19), to establish a syngeneic animal model using immunocompetent mice, which was shown by others to reduce immunogenicity to ICI treatment (46, 47). After the intrarectal injection of luciferase-labeled CT26 cells, we treated mice with sulindac, PD-L1 Ab, and a combination of sulindac and PD-L1 Ab, respectively. Noninvasive bioluminescence imaging (IVIS) was utilized to monitor the orthotropic growth of CT26 tumors. Our results clearly demonstrate that the combination of sulindac and PD-L1 Ab was the most effective therapy in our observation, with a significant effect on tumor weight, size, and regression.

In view of a previous study reporting the involvement of CD8⁺ T lymphocytes in the anticancer activity of sulindac against breast cancer (23), we examined the infiltration of CD8⁺ T lymphocytes in tumors using immunofluorescence imaging. Our results showed a significant increase of infiltrating CD8⁺ T lymphocytes in the tumor tissues from the animals treated with sulindac monotherapy and combination therapy, with the latter group clearly exhibiting a more intensive infiltration of T cells. We also examined PD-L1 expression in the same tissue sets, given that previous studies have reported that elevated PD-L1 expression is associated with poor prognosis, reduced survival, and lymph node metastasis of colorectal cancer (48–50). Higher expression of PD-L1 in tumor cells has also been suggested to be one of the drivers of tumor immune escape (8). We found that PD-L1 expression was inversely correlated with the number of infiltrating CD8⁺ T lymphocytes, indicating that sulindac is able to target and inhibit PD-L1 expression in tumor cells, leading to the infiltration of CD8⁺ T lymphocytes. Of interest, we further demonstrated that sulindac and PD-L1 Ab have a synergistic effect on activation of CD8⁺ T lymphocytes (GranzB⁺), supporting our in vivo results showing that the combination therapy has an improved inhibition on tumor growth, compared with the monotherapy.

Furthermore, we measured PD-L1 expression in the cytoplasm and cell membrane after sulindac treatment using four pMMR colorectal cancer cell lines, including CT26, Colo205, HT29, and Caco-2 cells. Our results demonstrate that sulindac can significantly downregulate intracellular and extracellular PD-L1 levels. We further explored the molecular mechanism by which sulindac inhibits PD-L1. Recently, others have reported that NF-κB is involved in the regulation of PD-L1 expression in cancer cells (27–30), and our previous studies have also shown that sulindac can block NF-κB signaling (26, 43). Following our established protocol, we searched for NF-κB binding sequences and identified a putative locus in the PD-L1 gene promoter. Utilizing ChIP and luciferase reporter assays, we identified direct binding between NF-κB and the PD-L1 promoter. Recent studies have also reported that TNFα can induce PD-L1 expression through the NF-κB pathway (30). Therefore, we used TNFα as an inducer to stimulate NF-κB signaling and then determine whether sulindac treatment could effectively alter PD-L1 expression in all four pMMR colorectal cancer cell lines. As expected, our results support the notion that NF-κB signaling is a key mechanism by which sulindac downregulates PD-L1 through transcriptional regulation. We further examined the translocation of NF-κB in the tumor tissues collected from in vivo studies. Consistently, mice treated with sulindac, either alone or in combination with PD-L1 Ab, show a far less intensity of NF-κB signals within the cell nucleus, concomitant with downregulation of PD-L1.

Exosomes are 30- to 100-nm–sized membrane vesicles released by many cell types during physiologic processes. Importantly, exosomes serve as a means of cell–cell communication because they can easily fuse with recipient cells and release their contents (51). Tumor cells typically produce far more exosomes than their normal noncancerous counterparts, and their exosomes are often referred to as tumor-derived exosomes (TDEs; ref. 52). TDEs contain key oncogenic elements, including a variety of miRNAs, mRNA, proteins, and lipids, which can initiate signaling pathways required for the tumor progression and metastatic processes (53). Recent studies have reported that exosomal PD-L1 is able to attenuate the inhibitory effect of PD-L1 Ab (32). Utilizing the blood samples collected from our in vivo studies, we analyzed exosomal PD-L1 in mice treated with sulindac (alone), PD-L1 Ab (alone), and the combination (sulindac + PD-L1 Ab). We found that all treatment regimens significantly reduced exosomal PD-L1 expression, while the combination group is associated with the most reduction of exosomal PD-L1 and tumor weight. We further analyzed the correlation between exosomal PD-L1 expression levels and tumor progression in mice treated with monotherapy and combination therapy. The values of tumor weight were utilized as variables to indicate tumor progression.

Intriguingly, PD-L1 Ab–treated mice showed a significant reduction in exosomal PD-L1, but it poorly correlated with tumor weights (R² = 0.22; P = 0.23). These results suggest that exosomal PD-L1 readily binds to and consumes PD-L1 Ab in circulation, leading to a reduction in PD-L1 Ab that can approach and target tumor cells. For the combination of sulindac and PD-L1 Ab, exosomal PD-L1 displayed a tight inverse correlation with tumor progression (R² = 0.81; P = 0.01), due to the ability of sulindac to downregulate PD-L1 at the transcriptional level, and less total intracellular PD-L1 could lead to less exosomal PD-L1 being secreted into the circulation. Ultimately, PD-L1 Ab in combination therapy can be relatively “enriched” around tumor cells compared with treatment with PD-L1 Ab monotherapy. Thus, although our results show that all treatments can reduce the levels of exosomal PD-L1 in circulation, the amount of “free” PD-L1 Ab that can engage tumor cells is the crucial parameter to interpret the difference of tumor inhibitory effect between different treatment arms, and our correlation analysis supports such a conclusion (Fig. 4B). When excess “free” PD-L1 Ab reaches tumor cells, they can not only turn off immune checkpoints by binding to PD-L1 on the cell surface, but also activate CD8⁺ T lymphocytes to attack tumor cells.

Sulindac, an FDA-approved NSAID, is a nonselective cyclooxygenase-1 and -2 (COX-1 and COX-2) inhibitor. COX-2 inhibition can reduce the synthesis of prostaglandin 2 (PGE2). Therefore, long-term administration of NSAIDs, including higher doses of sulindac, is approached with particular caution regarding the possible human side effects associated with PGE2 reduction upon the gastrointestinal, renal, and cardiovascular systems (54). It has also been reported that COX-2 is capable of impairing tumor immunity and that inhibition of COX-2 has the potential to improve immunotherapeutic responses in a variety of tumor models (55–57). Therefore, if the systemic toxicity due to PGE2 inhibition is minimized, sulindac holds great promise as a very safe addition to ICI therapy. In our in vivo study, we used low doses of sulindac, which are equivalent to approximately ¼ of the FDA-approved dose for human usage. When testing PGE2 in the blood samples collected from mice, we did not see any significant difference in side effects or end-organ function among treatment groups, demonstrating that sulindac is relatively safe and effective at the low doses we chose.

We conclude that sulindac can effectively modulate the response of pMMR colorectal cancer to anti–PD-L1 immunotherapy. The mechanism of action involves sulindac downregulating PD-L1 through the blockade of NF-κB signaling. Given that patients with pMMR colorectal cancer represent the majority of the colorectal cancer population but benefit little from current ICI regimens, our study provides a unique insight into the development of sulindac as part of a safe and effective combination immunotherapy for the treatment of colorectal cancer.

A.C. Ochoa reports grants from NIH during the conduct of the study. Y. Xi reports grants from NIH/NCI during the conduct of the study; grants from NIH/NCI outside the submitted work. No disclosures were reported by the other authors.

B. Yi: Data curation, formal analysis, methodology, writing–original draft. H. Cheng: Data curation, validation, and investigation. D. Wyczechowska: Data curation and methodology. Q. Yu: Data curation and formal analysis. L. Li: Resources and investigation. A.C. Ochoa: Resources and supervision. A.I. Riker: Conceptualization, writing–review and editing. Y. Xi: Conceptualization, resources, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

This work was supported in part by the NIH–NCI grants (R01CA192395 and a pilot award of P20CA233374) to Y. Xi. We also appreciate Z. Liang and Y. Xun for their technical support.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.