Stromal Cell PD-L1 Inhibits CD8⁺ T-cell Antitumor Immune Responses and Promotes Colon Cancer

Colon cancer development and metastasis is a multistep process, during which accumulation of genetic mutations drives tumor development through acquisition of hallmarks, including the ability to evade immune surveillance (1). Colon tumors are composed of numerous cellular components that can facilitate an immunosuppressive microenvironment, including immune, stromal, and endothelial cells (2–5). In colon cancer, the type, density, and location of immune cells within tumors can predict clinical outcome (6), and colon cancer patients without tumor recurrence are shown to have higher immune cell densities (CD3, CD8, granzyme B) than patients whose tumors recurred after treatment (6). These observations highlight the need to better understand the cellular and molecular mechanisms that dictate an immunosuppressive microenvironment in colon cancer.

In the colon cancer microenvironment, activated T cells traffic from the blood or lymph nodes into the tumor. The stromal cells in the colon are a heterogeneous population (7, 8), with a large proportion being of mesenchymal origin. They are positioned between the epithelial cells and the underlying vasculature and can passively or actively impair immune cell trafficking and activation. Stromal cell influence in the tumor microenvironment has been highlighted in several studies that identify a stromal cell signature from colon tumors that is associated with tumor progression and a poorer prognosis in colon cancer patients (9). However, it remains poorly understood if the prognosis associated with these signatures is due to the inherent tumor-promoting or immunomodulatory potential of mesenchymal cells, which constitute a significant proportion of stromal cells in the intestinal microenvironment. These findings highlight the need to better understand the immunomodulatory role of the tumor stromal compartment to improve the development of effective stroma-targeting strategies.

Inflammation can promote tumorigenesis and is recognized as one of the hallmarks of cancer (1, 10). Inflammatory bowel disease is causally associated with colon cancer promotion (11). The transcription factor NF-κB, a key regulator of both inflammation and cancer, regulates the expression of many proinflammatory chemokines and cytokines and is associated with cancer progression and metastasis (8, 12). Many NF-κB–dependent soluble factors, including TNFα, can directly activate stromal cells to inhibit adaptive and innate immune effector mechanisms (13). Mesenchymal stromal cells (MSC) can suppress effector T lymphocytes, but little is known about how these stromal cell properties are altered in the inflammatory colon tumor microenvironment (14).

One important mechanism by which tumors avoid clearance by the immune system is by upregulating the expression of immunomodulatory ligands. In colorectal cancer, programmed death-ligand 1 (PD-L1) expression on tumor cells and tumor-infiltrating lymphocytes is of particular importance to this immune escape phenotype (15, 16). The receptor for PD-L1, PD-1, is expressed on activated T cells and, when bound by the ligand, results in downregulated proliferation and inhibition of effector function (15). The PD-1/PD-L1 pathway represents an adaptive immune resistance mechanism that is exerted by tumor cells in response to endogenous antitumor activity. An antibody targeting PD-1 has been granted FDA approval for metastatic colon cancer (17). Although the PD-1/PD-L1 signaling axis is important in colorectal cancer, anti–PD-1 therapy is not always effective in this setting, with some patients experiencing disease progression following treatment (18). Efforts to better stratify patients for anti–PD-1 immunotherapy have relied on the measurement of tumor or immune cell PD-L1 expression, but reports show that some patients deemed tumor PD-L1–negative respond positively to anti–PD-1 therapy (19).

The stromal compartment of the tumor microenvironment has received little attention to date in terms of stratifying patients for immunotherapy despite the known immunomodulatory potential of these cells. We, therefore, hypothesized that stromal cells in the inflammatory colon tumor microenvironment can directly modulate T-cell–mediated antitumor immunity through cell contact–dependent mechanisms. We showed that stromal cell PD-L1 expression induced by TNFα signaling in colon cancer cells led to decreased CD8⁺ T-cell proliferation, activation, and promotion of colon cancer growth in vivo.

Eight- to 14-week-old female Balb/c mice were purchased from Charles River. Experimental animals were housed in a specific pathogen-free facility and fed a standard chow diet. All procedures performed were conducted in a fully accredited animal housing facility at NUI Galway under a license granted by the Health Products Regulatory Authority, Ireland, and were approved by the NUI Galway Animal Care Research Ethics Committee.

Mouse colon adenocarcinoma cells CT26, derived from Balb/c mice, were purchased from the European Collection of Cell Cultures (ECACC) and cultured in DMEM (Biosciences-Gibco) supplemented with 10% fetal bovine serum (FBS; Sigma) and 1% penicillin/streptomycin (Sigma). To assess the effects of CT26 NF--κB signaling, stable CT26 clones expressing a mutant form of the human IкB-α with serine 32 and 36 mutated into alanine (IкB-α SR) were generated as previously described (4). Human colon cancer cell lines HCT116 and HT29 were purchased from the ATCC and cultured in McCoys 5A medium (Sigma) supplemented with 10% FBS (Fisher, Hyclone), 1% l-glutamine, and 1% penicillin/streptomycin (Sigma). Mouse and human cell master stocks were authenticated by ECACC/ATCC, confirmed mycoplasma negative (MycoAlert, Lonza), expanded, frozen, and used within 15 passages of testing for all subsequent experiments.

For murine MSC isolation, Balb/c mice were euthanized by CO₂ inhalation, and the femur and tibia were removed, cleaned, and placed in MSC culture medium; MEM-α (Biosciences-Gibco) supplemented with 10% FBS (Fisher, Hyclone), 10% Equine serum (Fisher, Hyclone), and 1% penicillin/streptomycin (Sigma). MSC were flushed from the bones, filtered, and plated at a density of 9 × 10⁵ per cm² in a T175 flask. Cells were incubated at 37°C, 5% CO₂, and nonadherent cells were removed 24 hours later. This process was repeated 3 times per week until the cells reached confluency. MSC were characterized according to the criteria published by the ISCT cell-surface characterization (Supplementary Fig. S1A), and trilineage differentiation was carried out at passage 4 (Supplementary Fig. S1B–S1D). All subsequent experiments were carried out with mouse MSC between passages 5 and 14.

Human MSC (hMSC) were isolated from the bone marrow of three healthy volunteers at Galway University Hospital under an ethically approved protocol (NUIG Research Ethics Committee, Ref: 08/May/14) according to a standardized procedure. Written consent was obtained from the volunteers. Briefly, bone marrow cell suspensions were layered onto a Ficoll density gradient, and the nucleated cell fraction was collected, washed, and resuspended in MSC culture medium. After 24 hours of cultivation, nonadherent cells were removed, fresh medium was added, and individual colonies of fibroblast-like cells were allowed to expand and approach confluence prior to passage. hMSC were grown in alpha MEM (Gibco, Invitrogen) supplemented with 10% FBS (Sigma-Aldrich), 1% penicillin–streptomycin, and fibroblast growth factor 2 (FGF2, 1 ng/mL; Sigma-Aldrich).

For MSC tumor conditioning (MSC^TCM), CT26, HCT116, and/or HT29 cells were seeded in a T175 flask (Nunc, Fisher) at a density of 1 × 10⁶ cells per flask in 25 mL of the corresponding tumor cell culture medium. Cells were left to grow at 37°C at 5% CO₂ (normoxia) for a total of 72 hours, at which point conditioned medium was collected, spun at 1,000 × g to pellet any cellular debris, and stored at −80°C. For TNFα conditioning (MSC^TNF-TCM), TNFα (100 ng/mL; PeproTech) was added to tumor cell cultures 24 hours prior to collection of tumor cell medium. For conditioning mouse MSC, cells were seeded at a density of 0.03 × 10⁶ cells per well of a 6-well plate in 2 mL MSC culture medium. Human MSC were seeded at a density of 0.1 × 10⁶ cells per well of a 6-well plate. Twenty-four hours after seeding, the culture medium was removed and replaced with 40% fresh MSC medium and 60% tumor-conditioned medium. Fresh DMEM (MSC^Control) and DMEM with TNFα (100 ng/mL; MSC^TNF) were included as controls. MSC were analyzed at 24 and 72 hours after addition of conditioned medium. In vitro immunosuppression assays and in vivo experiments, described below, were all carried out using MSC that had been conditioned for 72 hours.

The NF-κB luciferase reporters driven by 5 × wild-type (5 × NF-κB-Luc; pNF-κB-Luc plasmid; Stratagene) were used in this study. CT26/EV or CT26/IкB-α SR cells were generated and validated as previously described (4), plated in a 96-well plate, and cotransfected with 500 ng of pNF-κB-Luc or 500 ng of pLuc control vector (Stratagene) and 50 ng of RSV-pRL reporter (Promega Corp.). Twenty-four hours after transfection, cells were treated with TNFα at doses of 10 to 100 ng/mL for 24 hours. NF-κB activity was determined by analysis of luciferase activity with the Dual-Glo Luciferase Reporter Assay System (Promega) to validate reduced NF-κB activity in CT26/IкB-α SR cells (Supplementary Fig. S2A).

Four PD-L1 (CD274) mouse-specific shRNA expression vectors in pGFP-C-shLenti plasmids (CD274 shRNA Lenti-plasmid) and one scrambled negative control noneffective shRNA cassette in pGFP-C-shLenti plasmid (scramble shRNA Lenti-plasmid) were purchased from OriGene (OriGene Technologies; Cat. No. TL503436). Lentiviruses were generated by cotransfection of 293 T cells with either CD274 shRNA or scramble shRNA Lenti-plasmids, as well as with packaging (psPAX2.2), envelope (pMD2.G), and additional (pRSV-rev) plasmids using a standard protocol (Invitrogen). Supernatants were harvested 48 and 72 hours after transfection, combined, filtered through 0.45-μm pore size filters, and frozen at −80°C until required.

Balb/c MSC were removed from liquid nitrogen storage and passaged twice before transduction. Because pGFP-C-shLenti plasmids encode a puromycin resistance gene, MSC were treated with 7 μg/mL of puromycin (Gibco-Biosciences) for 96 hours, followed by 72 hours recovery in fresh MSC media. Following this, cells were imaged by fluorescent microscopy to identify GFP⁺ cells (Supplementary Fig. S4A). GFP expression was observed in both the CD274 shRNA– and scramble shRNA–transduced MSC, demonstrating successful transduction (Supplementary Fig. S4A).

To determine if successful knockdown of PD-L1 was achieved, CD274 shRNA– and scramble shRNA–transduced MSC were centrifuged at 800 × g, supernatants were removed, and RNA was isolated from cell pellets using the Isolate II RNA MiniKit (Bioline) according to the manufacturer's instructions. RNA was resuspended in 40 μL of nuclease-free water and quantified by NanoDrop. cDNA was synthesized using RevertAid H Minus Reverse Transcriptase (Fermentas) with random primers. Two-step qRT-PCR was performed to determine the mRNA expression of CD274 (IDT Primetime qPCR assay: Mm.PT.58.11921659, Integrated DNA Technologies) by comparing Ct values with that of the housekeeping gene GAPDH (Primetime qPCR assay: Mm.PT.39a.1, Integrated DNA Technologies). Quantitative real-time PCR was performed according to the standard program on the ABI Step-one machine (Applied Biosystems). CD274 mRNA expression in CD274 shRNA– and scramble shRNA–transduced MSC are illustrated in Supplementary Fig. S4B.

MSC were seeded at a concentration of 1 × 10⁵ cells per 6-well in 2 mL of mouse MSC media and allowed to adhere overnight. Lentivirus-containing supernatants from the different preparations were mixed with mouse MSC media at a ratio of 1:5 (mouse MSC media:lentiviral supernatant) to a final volume of 2 mL per well. Plates were incubated at 37°C, 5% CO₂ for 24 hours. After 24 hours, the virus-containing media were removed and replaced with fresh mouse MSC media mixed with conditioned medium from CT26 cells preactivated with TNFα (MSC^TNF-TCM; 40% MSC media and 60% TNFα TCM) and incubated for 24 hours. To test the transduction efficiency, 48 hours after transduction GFP⁺ cells were analyzed by flow cytometry (Supplementary Fig. S4D). In the same experiment, successful inhibition of PD-L1 expression on PD-L1 shRNA MSC was confirmed (Supplementary Fig. S4E and see Supplementary Fig. S4C for gating strategy).

To obtain primary Balb/c lymphocytes, inguinal, mesenteric, and cervical lymph nodes, as well as spleens, were harvested from healthy animals following CO₂ asphyxiation. Single-cell suspensions were obtained by mechanical disruption (using a 1-mL syringe plunger) of the tissue in culture medium: RPMI 1640 (Fisher, Lonza) supplemented with 10% FBS (Sigma), 1% sodium pyruvate, 1% nonessential amino acids, l-glutamine, 1% penicillin/streptomycin (all Sigma), and 0.01% β-mercaptoethanol. Cells were washed, resuspended in PBS, and erythrocytes lysed using ammonium–chloride–potassium lysing buffer (Sigma-Aldrich) for 5 minutes on ice. Cells were resuspended in culture medium, and a combination of 90% lymphocyte/10% splenocyte was used for immunosuppression assays.

To assess proliferation, cells were stained with carboxyfluorescein diacetate succinimidyl ester (CFSE) using the CellTrace cell proliferation kit (Invitrogen). For assessment of effects of shRNA MSC, lymphocytes were stained with cell trace violet (CTV) using a CellTrace Violet Cell Proliferation Kit (Thermo Fisher Scientific) in order to exclude MSC GFP expression from analysis. Cells were counted and resuspended in 1 mL PBS at a concentration of 1 × 10⁷ cells/mL. CFSE and CTV were reconstituted in dimethyl sulfoxide, and 2 μL of reconstituted dye was added for each 1 mL of BSA/PBS. Cells were protected from light and incubated at 37°C for 6 minutes. Ice-cold culture medium was added to neutralize the CFSE, and cells were washed twice more. Cells were then activated using CD3/CD28 Mouse T-Activator Dynabeads (Life Technologies) for the duration of the 72-hour coculture.

CFSE-labeled and activated cells (0.1 × 10⁶) from whole-cell isolates (90% lymphocyte/10% splenocyte) were cocultured with naïve or tumor-conditioned Balb/c MSC or TNFα tumor-conditioned PD-L1 shRNA– and scramble shRNA–transduced MSC at a ratio of 1 MSC:50 or 10 lymphocytes in a 96-well round bottom cell culture plate (Sarstedt). After 72 hours, medium was removed from the cocultures, aliquoted, and stored at −80°C for further analysis by ELISA. Cells were resuspended in FACS buffer (PBS supplemented with 2% FBS and 0.05% sodium azide) and were incubated with CD8a (APC, Clone 53-6.7; BioLegend) or CD4 (PerCP-Clone GK1.5; BioLegend) and/or CD25 (PE-clone PC61; BioLegend) for 15 minutes at 4°C. Cell viability was assessed following resuspension in 100 μL of FACS buffer by addition of the viability dye SYTOX blue (Thermo Fisher, S34857; Supplementary Fig. S2B) according to the product instructions. Cells were washed twice, and proliferation was measured using a FACSCanto cytometer (Becton Dickinson). Data were analyzed using FlowJo software (Version X, TreeStar Inc.). T-cell suppression was calculated according to the formula: 100 − ((% positive control divided/% sample divided) × 100) for all population doublings greater than the second generation (Supplementary Fig. S2C and S2D).

Supernatants from MSC and T-cell cocultures were analyzed using Ready-SET-Go! ELISA kits (Affymetrix, eBioscience) for secretion of IFNγ, TNFα, and granzyme B. Flat-bottom 96-well ELISA plates were coated overnight at 4°C with capture antibody according to the manufacturer's protocol. The following morning, plates were washed with wash buffer (1× PBS with 0.05% Tween-20), and plates were blocked using the supplied assay diluent for 1 hour and then washed. Standards, as supplied with the assays and 100 μL of samples (1:2 dilution), were added to the plates and incubated for 24 hours at 4°C. Following 3 washes, detection antibody (100 μL) was added and incubated at room temperature for 1 hour. Following washing, 100 μL avidin horseradish peroxidase (HRP) was added for 30 minutes and washed a further 3 times. Tetramethylbenzidine substrate solution (100 μL) was then added and left at room temperature for 15 minutes before 50 μL stop solution (2N sulfuric acid) was added, and the plates were read at 450 and 570 nm on a Wallac 1420 plate reader (PerkinElmer).

Tumors were induced in 8- to 14-week-old female Balb/c mice by subcutaneous injection into the left flank of 2 × 10⁵ CT26 cells ± 0.5 × 10⁵ MSC (−/+ in vitro tumor preconditioning MSC^Control/MSC^TNF-TCM) in a total volume of 100 μL PBS. Animals receiving PD-1 antibody therapy received an intraperitoneal injection of 200 μg of anti–PD-1 (clone RMP1-14; Bio X Cell, 2B Scientific) in 100 μL PBS on days 7 and 14 after tumor induction. Tumor growth was monitored daily until sacrifice on day 21. At day 21, animals were euthanized, and tumors were harvested from left flank and, where tumor invasion had occurred as determined by visual tumor deposits, the peritoneal cavity. Tumor measurements were taken using digital calipers, and tumor volume was calculated according to the rational ellipse formula: (M^1⁁2 × M² × π/6). Images of tumors were taken of tumor tissue dissected from animals at day 21.

MSC or CT26 cells were trypsinized, counted, and washed in FACS buffer (PBS supplemented with 2% FBS and 0.05% sodium azide) prior to staining. Staining was carried out by incubating the cells with the antibody of interest in FACS buffer at 4°C for 10 minutes. Following this incubation, cells were washed twice in FACS buffer, stained with the viability dye SYTOX blue (Thermo Fisher, S34857), and analyzed straight after or prepared for intracellular staining (protocol below). Antibodies used were as follows: MHC-I (APC, clone SF1-1.1), MHC-II (FITC, clone 39-10-8), CD80 (PE, clone 16-10A1), CD86 (PE, clone PO3), PD-L1 (APC, clone 10F.9G2), PD-L2 (PE, clone TY25), and FasL (biotin with streptavidin–PE). All antibodies were purchased from BioLegend. Streptavidin was purchased from eBioscience. For characterization, 100,000 CT26 or MSC were stained per marker.

To analyze tumor immune cell infiltrates, tumors were digested in 1 mL HBSS (Gibco-Biosciences) containing 150 U/mL collagenase IV (Biosciences) and 200 U/mL DNase (Sigma). Samples were left at 37°C for 2 hours and then filtered through 40-μm cell strainers and washed with PBS. Single-cell suspensions were counted and stained with markers of interest: CD3 (FITC, clone 17A2), CD8 (APC, clone 53-6.7), CD25 (PE, clone PC61), CD4 (APC, clone GK1.5), all BioLegend. Cell-surface staining was carried out as before using 1 × 10⁶ cells per sample and appropriate FMOs were used to ensure staining accuracy.

To stain for intracellular granzyme B (PE, clone REA226; Miltenyi), cells were washed twice after surface staining had been completed and fixed [2% paraformaldehyde (Sigma) for 10 minutes at 4°C] for intracellular staining. Cells were then incubated with granzyme B antibody diluted in permeabilization buffer (1% BSA/PBS with 0.5% saponin) according to the manufacturer's instructions. Samples were again incubated at 4°C for 10 minutes, washed twice in FACS buffer, and analyzed on a FACSCanto II.

Gene-expression profiles from two independent colorectal cancer data sets were downloaded from the NCBI Gene-Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/) under accession numbers GSE35602 and GSE70468. GSE35602 contains microarray profiles separately profiled from laser-capture microdissected stroma or epithelium regions from 13 colorectal cancer primary tumors (20). GSE70468 contains microarray profiles from primary fibroblasts derived using colon primary tumor and morphologically normal colonic mucosa tissue isolated from fresh colorectal cancer resection material. Both studies indicate that they were performed after approval by an institutional review board (IRB), and informed written consent was obtained from the subjects (20, 21).

Partek Genomics Suite software (version 6.6; Partek Inc.) was used for analysis of the independent data sets. For the purpose of clustering, data matrices were standardized to the median value of probe set expression. Standardization of the data allows for comparison of expression for different probe sets. Following standardization, two-dimensional hierarchical clustering was performed (samples × probe sets/genes). Euclidean distance was used to calculate the distance matrix, a multidimensional matrix representing the distance from each data point (probe set-sample pair) to all the other data points. Ward's linkage method was subsequently applied to join samples and genes together, with the minimum variance, to find compact clusters based on the calculated distance matrix.

Statistical analysis was carried out using GraphPad Version X (GraphPad Software). Data were assessed for normal distribution using D’Agostino–Pearson omnibus normality test. Data sets with two groups were analyzed using an unpaired t test. Data sets with more than two groups were analyzed by ordinary one-way ANOVA followed by the Tukey multiple comparisons test. For analysis of correlation, Pearson correlation coefficient was calculated. To test for a trend in tumor invasion data, a χ² test was carried out, and to test for the effect of antibody treatment on individual treatment groups, Fisher exact test was used. Results were considered statistically significant at P < 0.05.

A large proportion of the stromal component of the healthy colon and colon tumor microenvironment are of mesenchymal origin, which is an important barrier to overcome for infiltrating immune cells (22). However, the role of these stromal cells in modulating the immune response in the tumor microenvironment, and the influence of the tumor secretome on this capability, is unknown. To test the functional consequences of tumor conditioning on the ability of stromal cells to inhibit lymphocyte proliferation and effector function, we used an ex vivo syngeneic culture system, whereby primary Balb/c lymphocytes were cocultured with stromal cells conditioned by the tumor secretome in the presence (MSC^TNF-TCM) or absence (MSC^TCM) of inflammation [Fig. 1A(i–iv)]. A dose-dependent induction of NF-κB in CT26 in response to TNFα treatment was confirmed by a luciferase reporter gene assay (Supplementary Fig. S2A). Primary bone marrow stromal cells represent a robust model cell type for investigation of the effects of tumor conditioning on stromal cell components of the tumor microenvironment, considering they share numerous characteristics with intestinal stromal cells in terms of origin, surface marker expression, gene signatures, and immunomodulatory capacity (23–25). As previously documented, control MSC had limited ability to reduce CD4⁺ and CD8⁺ T-cell proliferation relative to positive controls [stimulated T cells and no stromal cells; Fig. 1B(i)]. However, this suppressive capacity was significantly increased by stromal cell exposure to the inflammatory tumor secretome [Fig. 1B(ii–iii)]. To identify the mechanism of stromal cell–mediated suppression, we assessed T-cell viability after coculture. No significant differences were observed in either CD4⁺ or CD8⁺ T cells cocultured with MSC^Control, MSC^TCM, or MSC^TNF-TCM, suggesting induction of T-cell anergy or exhaustion rather than activation-induced cell death (Fig. 1C). In addition to enhanced T-cell suppression, inflammatory tumor-conditioned stromal cells, compared with control or TNFα-primed stromal cells, also significantly reduced lymphocyte activation, measured by TNFα and IFNγ release [Fig. 1D(i) and (ii); ref. 26]. A reduction in the cytolytic capacity of lymphocytes was seen when cultured with inflammatory tumor-conditioned stromal cells, measured by a reduction in granzyme B [Fig. 1D(iii)]. The inhibitory effect exerted by tumor-conditioned stromal cells on T-cell effector phenotype was most potent following stromal cell exposure to the secretome from TNFα pretreated CT26 cells, indicating a requirement for tumor cell inflammatory signaling in dictating stromal cell immunomodulatory function. These data demonstrate an enhanced ability of stromal cells to prevent the proliferation and effector function of T cells following exposure to the inflammatory tumor secretome. Considering the close contact between the tumor cells and the stromal niche in the colon, these data may imply stromal cells utilize the induction of PD-L1 to protect tumor cells from destruction by CD8⁺ cytotoxic T lymphocytes by inhibiting cytotoxicity and inducing T-cell anergy and exhaustion (27).

Stromal cells have been shown to inhibit T-cell responses by both cell contact–dependent mechanisms, via cell-surface protein expression, and cell contact–independent mechanisms, via release of soluble mediators (28, 29). Our initial finding of an enhanced immunosuppressive stromal cell potential following exposure to the inflammatory tumor secretome was observed following contact-dependent interactions between the stromal cells and lymphocytes, and this prompted a comprehensive analysis of an array of ligands on the stromal cell surface known to have immunomodulatory capacity. A significant increase in MHC-I expression following stromal cell exposure to the tumor secretome was seen, and this effect was significantly enhanced by inflammatory tumor activation [Fig. 2A(i)]. This observation may have important consequences, considering tumor antigens can be presented on MHC-I complexes potentially resulting in dysfunctional activation of antigen-specific CD8⁺ T cells (30). Compared with control stromal cells, tumor conditioning, in the presence or absence of inflammation, had no effect on the surface expression of MHC-II [Fig. 2A(ii)]. CD80 expression decreased, irrespective of inflammation, but with expression remaining high [Fig. 2A(iii)], whereas CD86 expression was unchanged [Fig. 2A(iv)].

PD-L1 was found to be specifically increased following tumor conditioning, which was further increased in the presence of inflammation [Fig. 2B(i); refs. 31, 32]. Expression of PD-L2 and Fas Ligand (FasL), two negative regulators of T-cell function, was found to be low or almost absent, respectively, and remained unchanged following exposure to the tumor secretome [Fig. 2B(ii) and (iii); refs. 33–35]. These findings confirmed our previous observations that T-cell viability remained unchanged following coculture. Compared with controls or TNFα-treated stromal cells, where PD-L1 expression was low, inflammatory tumor-conditioned stromal cells expressed PD-L1 to a higher extent than CT26 tumor cells (Fig. 2C). This finding points to an important role for the stromal niche in dictating the fate of tumor-infiltrating, PD-1 receptor–expressing immune cells. We confirmed that TNFα, when added to MSC^TCM, did not induce stromal cell PD-L1 expression in contrast to MSC^TNF-TCM. This suggests that TNFα-induced NF-κB signaling in CT26 cells is responsible for the PD-L1 induction (Fig. 2D). This finding was further confirmed using conditioned media from TNFα-treated NF-κB–deficient CT26 tumor cells. PD-L1 expression was significantly reduced on MSC^TNF-TCM conditioned with CT26/IκB-α SR cells when compared with CT26/EV control cells [Supplementary Fig. S3(i) and (ii)].

To confirm a definitive role for induced stromal cell PD-L1 in the suppression of T-cell proliferation and activation, we targeted the PD-1/PD-L1 signaling axis using a monoclonal blocking antibody to the PD-1 receptor. We observed high PD-1 expression on CD4⁺ and CD8⁺ T cells (Fig. 3A) ex vivo. Treatment with anti–PD-1 significantly reduced the ability of inflammatory tumor-conditioned stromal cells to suppress the proliferation of CD8⁺ T cells [Fig. 3B(i–ii)]. Following anti–PD-1 treatment, CD8⁺ T-cell suppression elicited by stromal cells exposed to the inflammatory tumor secretome was comparable with that induced by stromal cells exposed to the tumor secretome alone or TNFα-treated MSC [Fig. 3B(ii)]. These findings indicated that the enhanced CD8⁺ T-cell suppressive capacity of these cells is dependent on stromal PD-L1 expression induced by inflammatory signaling in the tumor microenvironment. No significant differences were seen in the capacity of inflammatory tumor-conditioned stromal cells to suppress CD4⁺ T cells, upon treatment with anti–PD-1 [Fig. 3B(iii)]. These findings may indicate a separate unknown mechanism by which stromal cells in the tumor microenvironment modulate CD4⁺ T cells. We further confirmed these findings using shRNA CD274 knockdown MSC. shRNA CD274 and scramble shRNA MSC were extensively characterized for shRNA transduction and knockdown efficiency by microscopy (Supplementary Fig. S4A), mRNA expression (Supplementary Fig. S4B), and flow cytometry (Supplementary Fig. S4C). Quantification of the percentage of GFP⁺ shRNA CD274 and scrambled MSC (Supplementary Fig. S4D) and of the percentage of PD-L1⁺ MSC from the GFP⁺ population (Supplementary Fig. S4E) were assessed prior to coculture with T cells in vitro. Compared with MSC controls, shRNA PD-L1 MSC^TNF-TCM significantly reversed the ability of MSC^TNF-TCM to suppress CD8⁺ T-cell proliferation [Fig. 3C(i)]. Similar to our findings using anti–PD-1, this effect was specific to CD8⁺ T cells, as the observation was not evident in the CD4⁺ T-cell subset [Fig. 3C(ii)]. We also observed a restoration of CD25 expression on CD8⁺ T cells following coculture with shRNA CD274 MSC^TNF-TCM compared with MSC^TNF-TCM [Fig. 3D(i)]. The CD25 restoration was not evident on the CD4⁺ T cells [Fig. 3D(ii)], validating that the PD-L1–mediated effects were predominantly on the CD8⁺ T cells. PD-1 blockade was sufficient to restore lymphocyte activation and cytolytic potential, measured by IFNγ, TNFα, and granzyme B secretion (Fig. 3E). This demonstrates the critical role of the PD-1/PD-L1 signaling axis in the ability of tumor-conditioned stromal cells to inhibit CD8⁺ T-cell–mediated antitumor immune effector functions.

In the context of immunotherapy, biologics targeting PD-L1 and PD-1 are showing great promise, but only in a subset of patients. A number of clinical trials have relied on measurements of PD-L1 expression on tumor cells or tumor-infiltrating immune cells in an effort to predict those patients who will respond best, but this method of stratification has had only limited success, with evidence now showing that some patients deemed PD-L1 negative will respond to this therapy (19, 36). Our findings may highlight an important role for stromal cell PD-L1 expression in modulating antitumor T-cell responses and in selecting optimal immunotherapeutic strategies for patients.

The effect of stromal cells on the progression of colon cancer has been tested in vivo, but much of this work has relied on the use of immunocompromised mice, limiting the ability to assess the antitumor immune response (37, 38). In order to test the functional consequences of stromal PD-L1 expression in vivo, we used a fully immunocompetent, syngeneic Balb/c subcutaneous tumor model, whereby the immunogenic CT26 cell line was administered with or without control or inflammatory tumor-conditioned stromal cells (39). Because our in vitro analysis showed that increased PD-L1 expression was unique to stromal cells exposed to the inflammatory tumor secretome, we assessed this group compared with control MSC in vivo (Fig. 4A). Untreated MSC were used as controls as PD-L1 expression on these cells was not significantly different to MSC^TNF-α or MSC^TCM (Fig. 2). The preliminary experiments we carried out as part of this study showed that control stromal cells induced increased tumor growth compared with that of CT26 tumor cells alone (Supplementary Fig. S5A), and tumor burden was not different from that induced by MSC^TNF-α or MSC^TCM (Supplementary Fig. S5B). The use of untreated MSC, thereby, allowed us to minimize the number of variables tested in our in vivo system.

Evidence exists to suggest that tumor-infiltrating lymphocytes contribute significantly to the prognosis, responsiveness to treatment, and likelihood of disease relapse in cancer patients (6). Our analysis of tumor immune cell infiltrates showed no change in the frequency of CD3⁺, CD3⁺CD4⁺, or CD3⁺CD8⁺ cells [Fig. 4B(i–iii); Supplementary Fig. S6]. There was a trend toward reduced frequency of granzyme B⁺ CD8⁺ T cells [Fig. 4B(iv)], but this did not reach statistical significance. Considering the previously published observation of increased granzyme B expression in colon tumors correlating with favorable patient outcomes, we measured granzyme B expression on CD8⁺ T cells by median fluorescent intensity (MFI) and found a significant reduction in the expression in tumors formed by the coadministration of CT26 with inflammatory tumor-conditioned stromal cells compared with CT26 cells alone. This result was dependent on the exposure of the stromal population to the inflammatory tumor secretome and, thus, may indicate a role for stromal PD-L1 in this effect. Cytotoxic CD8⁺ T lymphocytes are important in antitumor immune responses, and a reduction in granzyme B expression could result in an inhibition of their tumor cell clearing capacity (40).

In addition to altering the tumor immune cell infiltrate, stromal cells have been shown to promote the growth and invasion of tumors (41, 42). In our model, injection of CT26 alone induced subcutaneous tumor formation with no evidence of tumor cell invasion and metastasis. In 50% of animals receiving control stromal cells, tumors were observed to have invaded tissue surrounding the primary tumor site and metastasized (Fig. 4C). This effect was significantly potentiated by the coadministration of stromal cells exposed to the inflammatory tumor secretome (Fig. 4C). In the clinical setting, an invasive or metastatic tumor is associated with a much less favorable outcome for patient survival. This holds true in the setting of colorectal cancer, where the emergence of metastasis is particularly grave (43). A significant increase in tumor volume was observed upon the coinjection of stromal cells, with inflammatory tumor-conditioned stromal cells inducing greater increases in volume than CT26 alone (Fig. 4D). These results highlight a role for inflammatory tumor signaling in altering the functional characteristics of the tumor stroma. The observations of smaller, though significant, increases in invasiveness and volume upon administration of control stromal cells may be indicative of in situ induction of PD-L1 expression on control stromal cells when in contact with tumor cells. The enhanced increases observed following inflammatory tumor-conditioned stromal cells point to an important role for early polarization of the stromal cell phenotype prior to or at the beginning of tumor formation.

To confirm the immunomodulatory role of inflammatory tumor-conditioned stromal cells and to test the effect of increasing stromal cell number, a second group of animals was coadministered CT26 with double the number of stromal cells used previously (0.5 × 10⁵ vs. 1 × 10⁵). This was investigated using both control stromal cells and those exposed to the inflammatory tumor secretome. No differences were found in tumor invasiveness or tumor volumes between animals that had received the lower or higher stromal cell number (Supplementary Fig. S7). Hence, all further experiments and analysis were carried out using the lower number of MSC, in consideration of physiologic levels of MSC in the TME. These data confirmed that the presence of an immunosuppressive stromal niche in a tumor is sufficient to significantly affect ability of immune cells to access and eliminate that tumor. We have shown that the presence of this stromal population caused biologically relevant and potent effects on tumor promotion, even when present in low numbers.

To confirm a role for PD-L1 in the enhanced immunosuppressive and tumor-promoting ability of coadministered stromal cells observed in vivo, three additional groups of animals were treated with anti–PD-1 at 7 and 14 days after tumor induction (Fig. 5A). In line with previous data demonstrating only mild responses to PD-1 monotherapy, in CT26 derived tumors, PD-1 antibody treatment had no significant effect on the immune cell infiltrate of tumors, and anti–PD-1 treatment also did not alter the frequency of CD3⁺, CD3⁺CD4⁺, or CD3⁺CD8⁺ T cells infiltrating coinjected CT26 and stromal cell tumors [Fig. 5B(i–iii)]. However, a trend toward restoration of the cytolytic capacity of intratumoral CD8⁺ T cells was seen, measured by granzyme B expression, in tumors formed by coadministration of CT26 with inflammatory tumor-conditioned stromal cells [Fig. 5B(iv); refs. 44, 45]. The immune cell checkpoint inhibition induced by anti–PD-1 was sufficient to reverse the increased tumor invasiveness that had been observed with the coadministration of inflammatory tumor-conditioned stromal cells (Fig. 5C). This suggests that stromal cell PD-L1 expression may have a central role in enabling the tumor to inhibit CD8⁺ T-cell–mediated inhibition of metastatic tumor spread. A similar significant reduction in tumor volume was observed upon treatment with anti–PD-1 (Fig. 5D). Images of tumors excised from treated animals indicated the increased number of invasive nodules on tumors in the CT26 coinjected with MSC^TNF-TCM group compared with either CT26 alone or CT26 coinjected with MSC^Control. This observation of enhanced invasiveness of tumors was reversed after treatment with anti–PD-1 therapy (Fig. 5E). Tumor volume in animals administered inflammatory tumor-conditioned stromal cells treated with PD-1 monoclonal antibody was similar to that of animals administered CT26 alone. These data point to a central role for stromal cell PD-L1 expression in obstructing the activity of antitumor CD8⁺ immune responder cells, allowing tumors to grow through immune evasion.

Because our experiments to date had been carried out using murine tissue, we next wanted to confirm the phenomenon of stromal cell PD-L1 induction on stromal tissue in the human tumor microenvironment. We examined transcriptional profiles of colorectal cancer resection samples (data set GSE35602), which had been laser-capture microdissected (LCM) to isolate the stromal and epithelial fractions prior to microarray profiling (20). Assessment of PD-L1 gene expression (CD274) indicated significantly higher expression of PD-L1 in the stromal compartment of CRCs compared with the epithelial cells [Fig. 6A(i)]. Assessment of EpCam (EPCAM), E-caderhin (CDH1), as markers of epithelial lineage, and α-SMA (ACTA2) and vimentin (VIM), as markers of mesenchymal lineages, further confirmed the purity of the samples following LCM [Fig. 6A(ii)]. Assessment of PD-L1, PDGFR-α (PDGFRA), CD90 (THY1), and CD105 (ENG) indicated that gene expression of these markers was associated with the stromal compartment in colorectal cancer tissue samples. Because of this observation, we next investigated if exposure to the tumor secretome would induce human stromal cell PD-L1 expression in a manner similar to that seen for our murine cells. Using primary human bone marrow stromal cells from healthy donors, which we showed were CD90, CD105, and PD-L1–positive [Fig. 6B(i–iii)], and the secretome from microsatellite (HCT116) or chromosomal (HT29) unstable cells, we repeated the conditioning protocol outlined in Fig. 1A (46). We observed a significant increase in PD-L1 expression upon exposure to the inflammatory tumor secretome, and this induction occurred irrespective of tumor cell microsatellite instability (HCT116) or chromosomal instability [HT29; Fig. 6B(iii–v); ref. 46].

Our data in both mouse and human models have indicated that PD-L1 expression is elevated in MSC following exposure to conditioned media from inflammatory stimulated colon cancer cells. Cancer-associated fibroblasts (CAF) represent a major component of colorectal cancer stroma and provide a suitable model to test our findings in colorectal cancer clinical samples. Therefore, we utilized gene-expression profiles (data set GSE70468) derived from patient-matched primary fibroblasts (n = 14 samples from 7 patients) isolated from within colorectal cancer tissue (CAFs), to represent MSC following cancer cell exposure, or from adjacent, distant, normal mucosal tissue (normal), to represent unconditioned MSC (20). Analysis of overall gene-expression profiles indicated a nonsignificant trend (P 0.10; range, 1.15–1.08-fold change) toward increased PD-L1 gene expression in CAFs compared with normal fibroblasts (Fig. 6C). A pairwise analysis of patient-matched samples further confirmed this finding, with 6 of the 7 matched CAFs displaying an increase in PD-L1 gene expression compared with their patient-matched normal fibroblast (Supplementary Fig. S8). This analysis in two sets of independent colorectal cancer clinical tissue cohorts supports our observations in the mouse model and adds further weight to our suggestion of an important role for stromal cell PD-L1 induction in colorectal cancer.

In this article, we described a central role for stromal cell PD-L1 expression in the tumor microenvironment in aiding the tumor in avoiding immune cell–mediated clearance. This dampening of the antitumor immune responses by the stromal compartment resulted in enhanced tumor burden and invasiveness (Fig. 7). The stromal compartment is located adjacent to both the cancerous epithelial cells and the colonic vasculature and lymphatic network, representing a physical barrier to entry for immune cells in response to inflammation or epithelial transformation (47). In addition to this structural role, these cells have been shown to produce soluble mediators and express proteins that can influence immune and epithelial cells in diseased tissue (48). We have now shown that PD-L1, induced in response to activation of inflammatory signaling in the tumor, is a critical mediator of stromal cell immunomodulatory capacity. More specifically, we identified tumor cell NF-κB as an important signaling pathway in the induction of a tumor-promoting stromal cell phenotype, demonstrated by a lack of induced stromal cell PD-L1 expression in the absence of TNFα-induced NF-κB activation in CT26 cells. Given the inflammatory nature of the colon tumor microenvironment and the important role for NF-κB signaling in the pathogenesis of colon cancer, these results may highlight a mechanism by which NF-κB signaling in colon tumor cells enables communication with the adjacent stromal compartment, thereby dictating their immunomodulatory function (49, 50). Interestingly, a study by Lakins and colleagues has shown that CAFs have similar properties as those described here, in terms of their ability to suppress antitumor responses. However, the effects highlighted in their study were shown to be dependent on T-cell death mediated by FasL and PD-L2. In contrast, we highlighted a significant role for tumor-induced stromal cell PD-L1 in cell death–independent suppression of CD8⁺ T-cell proliferation and activation, indicating induction of anergy. These results may highlight the importance of the nature of the tumor microenvironment on the induction of stromal cell–mediated immunosuppression (i.e., the presence of inflammation or hypoxia). Increasing our understanding of the impact of stromal cell functional status, as well as other tumor-associated immunosuppressive cell populations in different tumor microenvironments, will be essential to improving the efficacy of immunotherapies in the future.

By utilizing an immunocompetent, syngeneic murine model, we could overcome the limitations in assessing immune cell infiltrate associated with the use of xenografts in immunocompromised mice. This allowed us to identify CD8⁺ T-cell granzyme B as a potential effector molecule inhibited by the stromal cell barrier in the tumor microenvironment, which was restored by anti–PD-1 therapy. With specific regard to colon cancer, higher intratumoral expression of granzyme B has been shown to correlate with improved survival in patients (51). Collectively, these data demonstrate the importance of the stromal compartment in modulating the CD8⁺ T-cell–mediated antitumor immune responses in a PD-L1–dependent manner.

Expression of PD-L1 has been primarily detected on the surface of epithelial neoplastic cells in a number of cancers. However, in colorectal carcinoma, IHC-based studies of small cohorts have detected high PD-L1 expression in the stromal and immune compartments (18, 52, 53). In light of studies showing positive responses to PD-1 immunotherapy in patients whose tumors have been deemed PD-L1–negative, we provide a rationale for the assessment of stromal cell PD-L1 expression in order to better stratify patients for immunotherapy (19).

We also showed the phenomenon of inflammatory tumor-induced PD-L1 expression on human stromal cells in response to the secretome from microsatellite- or chromosomal instable human colon tumor cells. This is particularly important in light of data showing a better response to PD-1 immunotherapy in MSI-high colon cancer patients. We suggest that this is, as was discussed by Huang and Wu, a result of a lower mutational load attracting much fewer tumor-infiltrating CD8⁺ T cells (54). In this setting, PD-1 inhibition is of limited use because a paucity of T cells, whose effector function can be disinhibited by such therapy, exists. However, as the authors suggest, treatment with a second immune-stimulatory agent, such as MEKi, could lead to CD8⁺ T-cell infiltration and, thus, synergistic and favorable responses in these patients (54, 55). We validated our in vitro and in vivo findings in two independent colorectal cancer clinical resection cohorts. Using transcriptional profiles specific for the stromal or epithelial fractions of overall colorectal cancer tissue (20), we identified significantly higher gene expression for PD-L1 in the stroma, and we also identified increased PD-L1 gene expression in CAFs compared with patient-matched normal colon fibroblasts.

The precise molecular mechanisms underpinning this enhanced PD-L1 expression on stromal cells remain to be elucidated and represent an important avenue of pursuit for the future in our laboratory. Identification of the factors released from tumor cells under conditions of inflammation, and the signaling pathways activated in MSC, may facilitate the development of novel adjuvants for immunotherapy to enhance clinical therapeutic effects. In summary, targeting stromal cell PD-L1 may be key to breaking the cycle of immune evasion and immunosuppression established by the stromal compartment of the tumor microenvironment, leading to more favorable and durable outcomes for patients.

A.E. Ryan reports receiving commercial research support from Janssen Pharmaceuticals and Bristol-Myers Squibb and has ownership interest in U.S. Patent app 15/798,670. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Ritter, L.J. Egan, A.E. Ryan

Development of methodology: G. O'Malley, K. Lynch, S.D. Naicker, A.E. Ryan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. O'Malley, O. Treacy, K. Lynch, S.D. Naicker, N.A. Leonard, P. Lohan, A.E. Ryan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. O'Malley, K. Lynch, S.D. Naicker, P.D. Dunne, T. Ritter, L.J. Egan, A.E. Ryan

Writing, review, and/or revision of the manuscript: G. O'Malley, K. Lynch, T. Ritter, L.J. Egan, A.E. Ryan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. O'Malley, O. Treacy, K. Lynch, S.D. Naicker, N.A. Leonard, P.D. Dunne, A.E. Ryan

Study supervision: A.E. Ryan

The authors would like to acknowledge financial support from the Irish Cancer Society (CRF12RYA) and Science Foundation Ireland (15/SIRG/3456) and Galway University Foundation to Dr. A.E. Ryan, a postgraduate scholarship from the Irish Research Council (GOIPG/2013/998) to G O’Malley, and a grant from Science Foundation Ireland (12/IA/1624) to T. Ritter.

The authors thank Professor Matthew Griffin, Professor Rhodri Ceredig, and Dr. Declan McKernan for helpful discussions; Ms. Georgina Shaw, Dr. Joana Cabral, and Ms. Athina Rigalou for technical assistance; and Dr. Shirley Hanley in the NUIG Flow Cytometry core facility for technical expertise.

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.