IRF1 Inhibits Antitumor Immunity through the Upregulation of PD-L1 in the Tumor Cell

The interferon (IFN) regulatory factors (IRF) are transcription factors involved in cellular stress responses. Depending on the cellular context, specific members of the IRF family are responsible for the induction of IFNs, lymphocyte development, and oncogenic signaling (1–3). Because of the role in inducing type I IFN, which mediates immunosurveillance of tumors, a number of IRFs, such as IRF1, IRF3, and IRF7 have been ascribed as antitumorigenic factors, whereas both pro and antitumor functions have been reported for the other IRFs (4). In fact, in an in vivo genetic screen using the lung metastasis model of mouse B16-F10 melanoma, IRF1 knockout mice were found to have the highest metastasis score; IRF7 knockout mice also had a higher metastatic score than wild-type (WT; ref. 5). IRF1 is lost or reduced in expression in a number of human leukemias (6–8). This and other cellular studies (9, 10) have suggested an antitumorigenic role of IRF1. However, a tumor cell–intrinsic role of IRF1 in solid tumors to affect tumor progression is not clear.

Despite the success of immune checkpoint blockade (ICB) therapy in different cancers, resistance and relapses are common (11, 12). ICB is based on the finding that a majority of intratumoral T cells are ineffective in their effector function due to inhibitory signaling through T-cell receptors such as CTLA4 and PD-1. Therefore, blocking of this inhibitory signaling using neutralizing antibody should reinvigorate the cytotoxic function of the effector T cells to clear the tumor. However, one mechanism of resistance, especially for the ICB therapy targeting the PD-1 axis, is the upregulation of PD-L1, a ligand for the T-cell–inhibitory receptor PD-1. PD-L1 is expressed on tumor cells and tumor-associated macrophages, where its transcription is induced by multiple signals including cytokines such as IFNγ, IFNα/β, TNFα, and other various TLR and oncogenic signals (13). Transcriptional regulation of steady-state PD-L1 mRNA expression is controlled through 3′-UTR–mediated RNA-decay (14, 15). A number of studies have identified correlation between genetic changes in the IFNγ signaling and the ICB therapy resistance (16, 17). However, mechanisms for primary and acquired resistance to PD-1/PD-L1 inhibition are varied and can be both multifactorial and overlapping (18).

IRF1 is an early-target gene downstream of IFNγ signaling and modulates IFNγ-mediated gene induction (19). IRF1 also regulates constitutive and inducible expression of PD-L1 by IFNγ (20–23). This led us to hypothesize that IRF1 might play a different role in tumor cells than in immune cells in determining the outcome of tumor progression. Here, using syngeneic mouse implantable tumor models, we show a tumor cell–intrinsic protumorigenic role of IRF1. IRF1 deficiency in the tumor cell results in reduced tumor progression. We found that IRF1 is necessary for PD-L1 upregulation in tumor cells and tumor progression in vivo. Although IRF1 deficiency did not result in changes in the tumor-infiltrating lymphocyte profiles, the cytotoxicity of the tumor-infiltrating CD8⁺ T cells was increased in tumors arising from IRF1-deficient cells compared with those from the parental cells (here onward designated as WT cells). These results indicate a tumor-enhancing effect of IRF1 in the tumor cell, which is distinct from its role in the host immune cell.

CT26 and B16-F10 cells were purchased from ATCC. Murine colon adenocarcinoma cell line MC38 was a generous gift from Dr. Dario Vignali (University of Pittsburgh, Pittsburgh, PA). All cell lines were used within 10 generations. Cell lines were not authenticated, but checked monthly for Mycoplasma contamination by DAPI staining and commercial PCR (Roche Diagnostics). MC38 and CT26 cells were maintained in DMEM (Corning) supplemented with 10% FBS (Atlanta Biologicals) and penicillin/streptomycin (Lonza). B16-F10 cells were maintained in RPMI (Corning) supplemented with 10% FBS and penicillin/streptomycin.

CRISPR-Cas9–mediated genome editing was carried out as described previously (24–26), with the use of murine IRF1 target sequence GAAGCACGCTGCTAAGCACGG. Briefly, plasmid-encoding gRNA, Cas9-mCherry was transfected into MC38, CT26, and B16-F10 parental cell lines with lipofectamine 3000 (Thermo Fisher Scientific). After 48 hours of transient transfection, mCherry⁺ single cells were sorted into 96-well cell culture plates. Single-cell clones were expanded and transferred into 12-well cell culture plate. Cells were treated with 100 ng/mL mouse IFNγ (BioLegend) for 6 hours before harvesting. We examined the expression of IRF1 via Western blot analysis.

Mouse PD-L1 coding region (PD-L1) was PCR amplified from pUNO mouse CD274 (Addgene Plasmid# 107012) using primers 5′-CACCATGAGGATATTTGCTGGCATTATAT TCAC-3′) and 5′-GAGTTTGGTGACTACATCTTAAGATCTATCATGTCGTC-3, TOPO cloned into pENTR D-TOPO and transferred into pInducer20 vector using Gateway Cloning (Thermo Fisher Scientific). B16-F10 IRF1-KO cells were transduced with lentivirus prepared from pInducer20-PD-L1 and selected by G418 (800 μg/mL) to make a stable cell line. Doxycycline-induced expression of PD-L1 in B16-F10 IRF1-KO cells was confirmed by Western blot analysis.

All mice experiments were conducted in accordance with, and with the approval of the University of Pittsburgh Institutional Animal Care and Use Committee (IACUC). All mice were housed in specific pathogen-free conditions and used at 6 to 8 weeks old. C57BL/6J and BALB/c mice were obtained from The Jackson Laboratory. C57BL/6J mice were subcutaneously inoculated with 10⁶ MC38 cells or intradermally inoculated with 5 × 10⁵ B16-F10 cells. BALB/cJ mice were subcutaneously inoculated with 10⁶ CT26 cells. Each dose of tumor cells was suspended in 100 μL of PBS (Lonza). For CD8⁺ T-cell–depletion studies, mice were intraperitoneally injected with anti-mouse CD8α (Clone YTS 169.4, 250 μg/mouse; Bio X Cell) or rat IgG2b isotype control (Clone LTF-2, 250 μg/mouse; Bio X Cell) 2 days before tumor cell inoculation and then twice a week. For rechallenge studies, after 1 month of complete tumor regression with MC38 IRF1-KO cells, mice were reinoculated with 10⁶ MC38 cells. In all the tumor studies, tumor volumes were measured and calculated twice a week with the formula V = 0.52*width²*length.

To detect tumor-specific memory T cells, spleens, and brachial-draining lymph nodes (dLN) were collected after 1 month of complete tumor regression with CT26 IRF1-KO cells. Naïve mice and CT26 IRF1-KO cell–injected tumor-bearing mice (day 45 of postinjection) were used as negative and positive controls, respectively. Spleens and dLNs were mechanically processed between frosted glass slides to generate single-cell suspensions in RPMI1640 with 10% FBS. Cells were resuspended in flow cytometry staining (FCS) buffer (2% FBS in 1× PBS), blocked with Clear Back FC Receptor Blocking Reagent (MBL Intl.) for 5 minutes at room temperature and stained with PE-conjugated AH1 MHC Class I Tetramer (H-2Ld MuLV gp70 SPSYVYHQF, MBL Intl.) for 30 minutes at 4°C. Antibodies to cell-surface markers [anti-CD8 (clone KT15, MBL Intl.), anti-CD44 (clone IM7, BioLegend), anti-CD62L (clone MEL-14, BioLegend), anti-CD45 (clone 30-F11, BioLegend)] were added to a final concentration of 1:500 and cells were stained for 30 minutes at 4°C. Cells were washed twice with FCS buffer, stained with Fixable Viability Dye eFluor780 (eBioscience) at 1:4,000 dilution in 1× PBS for 10 minutes at 4°C, washed twice with FCS buffer, and fixed in Fixation/Permeabilization Reagent (eBioscience) for 15 minutes at room temperature in the dark. Fixed cells were washed once with 1× Permeabilization Buffer (eBioscience) prior to a final wash with FCS buffer and storage in FCS buffer at 4°C protected from light. A BD LSRFortessa Cytometer and FACSDiva Software (BD Biosciences) were used to collect uncompensated data. Compensation and analyses were performed using FlowJo v10 Software (FlowJo). We designated naïve T cells = CD45⁺ CD44⁻ CD62L⁺, central memory T cells (T_CM) = CD45⁺ CD44⁺ CD62L⁺, and effector memory T cells (T_EM) = CD45⁺ CD44⁺ CD62L⁻.

Tumor and lymph nodes were collected on postinjection days 12 and 14 when the tumor sizes between B16-F10 WT and IRF1-KO groups did not show significant difference. Collected tumors were injected with 0.5 to 1 mL/tumor digestion mixture [2 mg/mL collagenase IV (Thermo Fisher Scientific), 200 μg/mL Hyaluronidase V (Sigma-Aldrich), and 4 U/mL DNase I (Sigma-Aldrich)] and incubated at 37°C for 20 minutes. Tumors were dissociated between two frosted glass slides and filtered through 40 μmol/L cell strainer to make single-cell suspensions. Lymph nodes were mechanically smashed through a 40 μmol/L cell strainer to get single-cell suspensions.

Cells from tumors and lymph nodes were resuspended in FCS buffer and stained with the following fluorescence-conjugated Abs: anti-CD8 (clone 53-6.7, BioLegend), anti-Tim3 (clone RMT3-23, BioLegend), anti-CD3 (clone 145-2C11, eBioscience), anti-LAG3 (clone C9B7W, BioLegend), anti-PD-1 (clone 29F.1A12, BioLegend), anti-CD45 (clone 30-F11, BioLegend), anti-PD-L1 (clone 10F.9G2, BioLegend), anti-CD4 (clone RM4-5, eBioscience), anti-CD25 (clone PC61.5, eBioscience), anti-Foxp3 (clone FJK-16s, eBioscience), anti-Ly6c (clone HK1.4, eBioscience), anti-Ly6G (Gr-1) (clone RB6-8C5, eBioscience), anti-CD11b (clone M1/70, eBioscience), anti-CD335 (NKp46) (clone29A1.4, BioLegend), anti-CD19 (clone 6D5, BioLegend), anti-TNFα (clone MP6-XT22, BioLegend), anti-IFNγ (clone XMG1.2, BioLegend). Cells stained for Foxp3 and/or intracellular cytokines (TNFα and IFNγ) were fixed and permeabilized with the Foxp3/Transcription Factor Staining Buffer Set (eBioscience). Flow cytometry data were acquired and analyzed as above. We designated, exhausted T cells = CD3⁺, CD8⁺, PD-1^high, Tim3⁺; T-regulatory cells (Treg) = CD3⁺, CD4⁺, CD25⁺, Foxp3⁺; granulocytic myeloid-derived suppressor cells (GrMDSC) = CD45⁺, CD11b⁺, Gr-1⁺; monocytic MDSC = CD45⁺, CD11b⁺, Ly6c⁺; NK cells = CD3⁻, C19⁺, CD335⁺.

For intracellular cytokine staining, isolated tumor cells and lymph node cells were cultured in 96-well plates with R10 media containing 100 ng/mL PMA and 500 ng/mL Ionomycin (Thermo Fisher Scientific) overnight. The next day, cells were incubated with BD GolgiPlug Protein Transport Inhibitor (1:1,000) for 4 hours. After incubation, cells were harvested for intracellular cytokine staining as described above.

B16-F10 WT and IRF1-KO cells were treated with 100 ng/mL of IFNγ for different time points and then trypsinized for harvesting. Harvested cells were subjected to regular flow staining with anti–PD-L1 and anti-MHC I (H-2Kb) (clone AF6-88.5, BioLegend).

Total RNA was isolated from mouse IFNγ-treated B16-F10 WT and IRF1-KO cells at different time points using TRIzol (Thermo Fisher Scientific). cDNA was synthesized using iScript cDNA Synthesis Kit (Bio-Rad). qRT-PCR using Bio-Rad PrimePCR probes and SsoAdvanced Universal SYBR Green Supermix (Bio-Rad) was conducted in a CFX96 Real Time System (Bio-Rad) following the manufacturer's instructions. All RT-PCR amplification was normalized to GAPDH. The PrimePCR Probe Assay IDs for mouse CD274 (PD-L1), Gbp2, Icam1, and Stat1 are qMmuCEP0052618, qMmuCEP0056450, qMmuCEP0056911, qMmuCEP0054514.

B16-F10 WT and IRF1-KO cells were lysed after mouse IFNγ treatment using cell lysis buffer (150 mmol/L NaCl, 1.5 mmol/L MgCl₂, 2 mmol/L EGTA, 2 mmol/L DTT, 10 mmol/L NaF, 12.5 mmol/L β-Glycerophosphate, 1 mmol/L Na₃VO₄, 1 mmol/L PMSF, 1% Triton-X100, 0.1% Sodium Deoxycholate, and protease inhibitors). The cleared cell lysates were subjected to SDS-PAGE and transferred to nylon membrane. The blots were incubated with anti-mouse PD-L1 (BioLegend) and IRF1 (D5E4) XP Rabbit mAb (Cell Signaling Technology) followed by appropriate HRP-conjugated secondary antibody and visualized by ECL Detection (GE Healthcare).

Pmel T cells were harvested from the spleen of B6.Cg-Thy1^a/Cy Tg(TcraTcrb)8Rest/J mice and stimulated with gp100 peptide. After the stimulation, Pmel CD8⁺ T cells were cultured in R10 media with 50 U/mL of mouse IL2 (PeproTech Inc.) for expansion for 7 days. B16-F10 WT and IRF1-KO cells were plated into 96-well plates at 10,000 cells per well and incubated for 5 hours. Pmel CD8⁺ T cells were prepared in R10 media without IL2 and added into the 96-well plate at T cell: tumor cell ratio of 10:1, 5:1, 2:1, 1:1, and 1:2. Cells were incubated overnight and harvested for cell viability examination using Zombie Aqua Fixable Viability Kit (BioLegend) and anti-CD8 (clone 53-6.7, eBioscience).

Results shown are pooled samples from at least twice repeated in vivo infection studies. For each data point, mean and SEM were plotted. Statistical significance was calculated either by Student t test or two-way ANOVA with Sidak multiple comparison test as appropriate and represented as *, P < 0.03 and ***, P < 0.001.

To investigate the tumor-intrinsic role of IRF1 during tumor progression, we generated several IRF1-deficient (IRF1-KO) syngeneic murine tumor cell lines (MC38, B16-F10, and CT26) via CRISPR/Cas9–mediated genome editing (Supplementary Fig. S1A–S1C) and compared their growth rates with WT cells both in vitro and in vivo. As shown in Fig. 1A–C, there was no difference in growth rates between IRF1-KO and WT cells when cultured in vitro. However, when these cells were injected in syngeneic mice (MC38 and B16-F10 in C57BL/6J and CT26 in BALB/c mice, respectively) the IRF1-KO cells showed significantly reduced tumor growth than WT cells (Fig. 1D–F). In addition, in some mice injected with IRF1-KO cells, tumors disappeared from the injection site around the third week of postinjection. For example, in MC38, B16-F10, and CT26 IRF1-KO cell–injected mice, 2/5, 3/5, and 3/5 mice had complete regression of tumors, respectively (Supplementary Fig. S1D–S1F). Taken together, these results indicate that in tumor cells IRF1 may positively contribute in immune escape of the tumor cells.

Having observed the regression of IRF1-KO tumor in mice models, we sought to investigate the contribution of host T cells to the loss of tumorigenicity by IRF1-KO tumor cells. We depleted CD8⁺ T cells in mice inoculated with MC38 IRF1-KO cells by intraperitoneal injection of anti-mouse CD8α and compared the IRF1-KO tumor growth with MC38 WT tumor. We found that the loss of CD8⁺ T cells in mice allowed the IRF1-KO tumor to grow at the same rate as WT tumor, whereas the IRF-1 KO cell–inoculated mice injected with IgG2b isotype control had the same slow tumor growth compared with the other two groups of mice (Fig. 2A). In addition, depletion of CD8⁺ T cells in MC38 WT tumor cell–injected mice, the tumor growth was faster than in normal mice (Fig. 2A). The depletion efficiency of CD8⁺ T cells was tested and confirmed in spleens (Supplementary Fig. S2A). Similar results were obtained when we repeated this experiment with B16-F10 WT and IRF1-KO cells (Supplementary Fig. S2B and S2C). These results indicate that CD8⁺ T cells play a role in the regression of the tumors arising from IRF1-KO cells injection.

We also investigated whether the mice developed immunologic memory following the regression of IRF1-KO tumor cells. After 30 days of complete IRF1-KO MC38 tumor regression, we rechallenged those mice with IRF1-sufficient MC38 tumor cells and recorded the tumor growth. Twelve mice were injected with MC38 IRF1-KO cells. Four out of 12 mice developed palpable tumors on back (red line), whereas the remaining 8 mice had complete tumor regression after postinjection day 17 (green line; Fig. 2B). After rechallenge of WT MC38 tumor cells, 5 of 8 mice had never developed any tumor, 2 of 8 started developing a tumor that completely regressed later, and only 1 mouse developed a tumor with slow progression (Supplementary Fig. S2B).

To determine whether the tumor rejection in those rechallenged mice was related to tumor-specific T cells, we examined the frequencies of tumor-specific memory T cells using AH1 MHC class I tetramer (H-2Ld MuLV gp70 SPSYVYHQF), which could bind to anti-CT26–specific TCR in mice. Because of the lack of tetramer that could recognize anti-MC38 or anti-B16-F10–specific TCR, we used the tetramer that could detect anti-CT26–specific TCR. After 30 days of complete IRF1-KO CT26 tumor regression, spleens and lymph nodes were collected from those mice. The frequencies of CD8⁺ tetramer⁺ cells in either T_CM or T_EM cells in spleen and lymph nodes were assessed by flow cytometry. Naïve unchallenged mice were used as naïve controls. Approximately 8% of T_EM cells were tetramer⁺ in the spleen of IRF1-KO CT26 cell–challenged mice (Supplementary Fig. S2D), whereas less than 1% of tetramer⁺ T_EM cells were found in lymph nodes (dLN) from those mice (Supplementary Fig. S2E). In addition, there were no tetramer⁺ T_CM cells in both spleens and lymph nodes from those mice. The data suggested that IRF1-KO tumor cells caused the development of tumor-specific memory T cells. And those T cells would contribute to the rejection of WT tumor cells in the rechallenge experiment. Taken together, these observations showed that CD8⁺ T cells were responsible for the IRF-1 KO tumor regression and they could develop memory responses.

Having found that CD8⁺ T cells are necessary in IRF1-KO tumor regression, we examined the profile of the tumor-infiltrating lymphocytes (TIL) in our model system. Following injection of the B16-F10 WT and IRF1-KO cells into C57BL/6J mice, we profiled the TIL via flow cytometry. As we have shown that approximately 50% of the mice injected with IRF1-KO cells had tumor regression (Supplementary Fig. S1) and some mice never developed any measurable size of tumor after the injection, we collected the IRF1-KO tumor as early as possible before the tumor regression. In addition, we compared the TIL between WT and IRF1-KO tumors of similar sizes, days 12 to 14 postinjection (Fig. 3A).

Exhaustion is a state of T cells with decreased effector cytokine production, reduced cytolytic activity, and overexpressed inhibitory receptors (27). Exhausted T cells are usually associated with poor tumor outcome (12). Thus, we examined the percentage of exhausted T cells defined as CD3⁺, CD8⁺, PD-1^hi, and Tim3⁺ and activated CD8⁺ T cells as CD3⁺, CD8⁺, PD-1^int, and Tim3⁻. As shown in Fig. 3B, there was no difference in the percentage of exhausted T cells and activated CD8⁺ T cells between WT and IRF1-KO tumors. As Tregs are associated with tumor progression and poor clinical outcome (28), we next compared the percentages of Tregs between these two groups. The percentage of Tregs in CD4⁺ T cells did not show any difference (Fig. 3C). In addition, the ratio of CD8⁺ T cell to Treg in IRF1-KO tumors was the same as WT tumors (Fig. 3D). Because LAG3 is also a marker to define exhausted T cells, we also assessed the frequencies of PD1⁺, Tim3⁺, and LAG3⁺ cells in the CD8⁺ T-cell population. There was no difference between the two groups (Fig. 3E). Combined with previous CD8⁺ T-cell–depletion experiment data, it seemed that CD8⁺ T cells were indispensable during tumor regression. But the regression was not due to changes in percentages of exhausted T cell, activated T cells, or Tregs. This led us to investigate changes in the other lymphocyte populations, such as myeloid-derived suppressor cells (MDSC). However, we did not observe any statistically significant difference in the percentages of granulocytic MDSC and monocytic MDSC in WT and IRF1-KO tumors (Supplementary Fig. S3A and S3B). In addition, both groups had the same amount of NK cells (Supplementary Fig. S3C). In summary, the results indicated that IRF1 deficiency in tumor did not quantitatively affect the TIL profile.

Because tumor intrinsic IRF1 did not change the percentage of TILs, we sought to investigate whether the loss of IRF1 changed the properties of tumor cells. First, we compared a common set of IFNγ-induced mRNA expression between B16-F10 WT and IRF1-KO cells in vitro. IFNγ treatment induced the mRNA expression of PD-L1, ICAM1, STAT1, and GBP2 in B16-F10 WT in a time-dependent manner, such that expression peaked between 6 and 8 hours (Fig. 4A and B; Supplementary Fig. S4A and S4B). However, a number of genes, for example, PD-L1, STAT1, and GBP2 showed significantly reduced induction in B16-F10 IRF1-KO cells (Fig. 4A; Supplementary Fig. S4A and S4B). Loss of IRF1 did not affect the induction of ICAM1 mRNA (Fig. 4B), Next, we confirmed the loss of IFNγ-induced PD-L1 mRNA by measuring protein. We found almost a complete loss of total (Fig. 4C) as well as cell surface–expressed PD-L1 protein in B16-F10 IRF1-KO cells after IFNγ treatment (Fig. 4D). These results indicated that IRF1 is necessary for IFNγ-induced expression of PD-L1 in tumor cells, whereas IRF1 does not affect ICAM1 induction by IFNγ.

PD-L1 is an inhibitory ligand that binds to the inhibitory receptor PD-1 on T cells and inhibits T-cell function (27). Accordingly, reduced PD-L1 expression on tumor cells may result in the reduction of inhibitory effects on the tumor-infiltrating T cells. Therefore, we examined PD-L1 expression in tumor cells in vivo. We defined CD45⁻ cells as tumor cells and determined the percentage of CD45⁻ PD-L1⁺ cells, as well as the mean fluorescence intensity (MFI) of PD-L1 signal. As shown in Fig. 4E, WT tumors had a significantly higher percentage of PD-L1–expressing cells than IRF1-KO tumors, whereas the MFI of PD-L1 signal was reduced in the IRF1-KO tumor cells compared with the WT cells. Taken together, these results indicate that the loss of IRF1 expression in tumor cells affects PD-L1 expression on the tumor cells leading to tumor regression.

Having found that the loss of IRF1 expression in tumor cells decreased their PD-L1 expression, we examined whether this would alter the effectiveness of tumor-infiltrating T cells. After collecting and preparing the single-cell suspension of tumors, we stimulated the mixture of tumor cells and immune cells with PMA and ionomycin overnight, then measured intracellular cytokines following GolgiPlug (protein transport inhibitor that contains brefeldin A) treatment for 4 hours. The intracellular cytokine (IFNγ and TNFα) expression was assessed using flow cytometry. We found that there were significantly more IFNγ, TNFα dual-expressing CD8⁺ cells in IRF1-KO tumors than in WT tumors (Fig. 5A).

We next examined whether IRF1-deficient tumor cells were more sensitive to T-cell killing. We utilized activated Pmel T cells specific for gp100 antigen expressed on B16-F10 tumor cells (29). Pmel T cells were mixed with B16-F10 WT and IRF1-KO cells at different ratios followed by the measurement of cell killing. The results showed that there was significantly more cell killing in B16-F10 IRF1-KO cells than WT cells (Fig. 5B). Because IRF1-KO tumor cells had less PD-L1 expression than WT tumor cells, they showed less inhibitory effects on the Pmel T cells and were more vulnerable to T-cell cytotoxic activity compared with WT tumor cells. In summary, the results indicated that loss of IRF1 in tumor cells impaired the PD-L1–mediated inhibitory effects on tumor-infiltrated T cells to increase the vulnerability of the tumor cells to the T-cell–mediated cytotoxicity.

To confirm the role of PD-L1 in tumor progression, we made a tetracycline-inducible PD-L1–expressing cell line in B16-F10 IRF1-KO (called B16-F10 IRF1-KO pInducer20-PD-L1) and monitored the in vivo tumor growth of this tumor cell line. To avoid 3′ UTR–mediated posttranscriptional repression, (14) we used the coding region of PD-L1. Total protein expression of PD-L1 and cell surface expression of PD-L1 were confirmed using Western blot and flow cytometry (Fig. 6A and B). Mice were fed with doxycycline-containing food (200 mg/kg) from 3 days before tumor cells injection. Mice injected with B16-F10 IRF1-KO pInducer20-PD-L1 cell and maintained on doxycycline food had similar tumor progression as B16-F10 WT cell–injected mice. Without the doxycycline food, the tumor growth rate of B16-F10 IRF1-KO pInducer20-PD-L1 tumor cells was the same as B16-F10 IRF1-KO cells (Fig. 6C). The results confirmed that the reduction of PD-L1 expression is the cause of tumor regression in the IRF1-KO tumor cells.

On the basis of studies using IRF1-deficient mice as well as genomic loss observed in human leukemia, IRF1 has been described as a tumor suppressor gene (4). In contrast to this notion, here we describe tumor cell–intrinsic tumor-promoting activity of IRF1. Loss of IRF1 in the tumor cells resulted in loss of in vivo tumor growth without affecting in vitro cell growth in multiple syngeneic tumor models. Mechanistically, we found that the absence of IRF1 led to a loss of PD-L1 transcription, resulting in increased killing of tumor cells by CD8⁺ T cells. The reduced PD-L1 expression on the tumor cells did not change the TIL profile of the IRF1-KO cell–injected tumors compared with the WT. However, the cytotoxic activity of the TILs from IRF1-KO tumors was higher than WT. This indicated that the cell intrinsic loss of PD-L1 can result in loss of tumorigenicity, as reported (30–32). However, similar to previous reports, we also observed occasional escape of IRF1-KO cell–injected tumors, which may be attributed to the PD-L1 expression on the myeloid cells resulting in the inhibition of T-cell–mediated cytotoxicity.

Multiple signaling pathways regulate PD-L1 expression both through transcription and mRNA stability (33). IFNγ and the transcription factor IRF1 are established as the primary drivers for PD-L1 expression (20–22). This study establishes that IRF1 was crucial for tumor cell–intrinsic expression of PD-L1 in vivo. We investigated the induction of PD-L1 in tumor cell lines after treatment with several cytokines that have been reported to induce PD-L1. However, only IFNγ significantly induced PD-L1in the B16-F10 cell line. In this case, loss of IRF1 resulted in almost complete loss of PD-L1 induction.

IRF1 is an inducible but ubiquitously expressed transcription factor. In many cell types, IRF1 transcription is induced strongest by IFNγ and less so with IFNβ (34). In addition, IRF1 transcription has been reported downstream of NF-κB activation in a number of cell types (35). Besides transcription, the steady-state amount of IRF1 protein is also regulated through posttranslational modification and degradation (19). A number of IFNγ-inducible genes require IRF1 in addition to STAT1 for sustained induction (36). Our results indicate PD-L1 to be one such gene in vivo in the context of tumor cells. Conversely, IRF2 is a negative regulator of IRF1-mediated transcriptional activation (9) and has been reported to inhibit PD-L1 expression and tumorigenesis (37). Although a role of IRF1 in mediating the PD-L1 regulation by IRF2 was not demonstrated in this report (30), this observation again points toward a role of IRF1 in PD-L1 induction and tumorigenicity.

The primary and acquired resistance to PD-1/PD-L1–targeted therapy can arise due to multiple factors. Among these, poor tumor immunogenicity, T-cell exclusion and tumor cell–intrinsic resistance to IFNγ result in the nonresponder phenotype. IFN resistance can evolve through acquisition of loss-of-function mutations in the JAK/STAT pathway (16–18). In both cases, loss of IFNγ-mediated tumor cell growth inhibition is the primary driver of the resistance (16, 17). However, our results imply a negative role for IFNγ signaling and induction of IRF1-mediated PD-L1 upregulation in the tumor cell resulting in reduced tumor cell killing. In fact, similar to our results, IFNGR loss was found to decrease tumor growth in mice (38). This again points to the complex interaction that exists in the tumor microenvironment. Nonetheless, it may still be possible to modulate the IRF1/IRF2 axis in the tumor cell for targeted tumor therapy.

H.M. Zarour reports receiving a commercial research grant from BMS and Tesaro. G.M. Delgoffe is the founder and scientific consultant at TTMS, Inc. and consultant at Pieris Pharmaceuticals and Western Oncolytics and reports receiving a commercial research grant from Pfizer, Bluebirdbio, and TCR2 Therapeutics and has ownership interest (including stocks and patents) in TTMS, Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Shao, W. Hou, S.H. Thorne, V. Hornung, C.J. Bakkenist, S.N. Sarkar

Development of methodology: W. Hou, V. Hornung, H.M. Zarour, G.M. Delgoffe, S.N. Sarkar

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Shao, W. Hou, N.E. Scharping, F.P. Vendetti, R. Srivastava, C.N. Roy, A.V. Menk, Y. Wang, J.-M. Chauvin, H.M. Zarour, C.J. Bakkenist, G.M. Delgoffe, S.N. Sarkar

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Shao, F.P. Vendetti, S.N. Sarkar

Writing, review, and/or revision of the manuscript: L. Shao, W. Hou, J.-M. Chauvin, S.H. Thorne, C.J. Bakkenist, G.M. Delgoffe, S.N. Sarkar

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Shao, W. Hou, C.N. Roy, P. Karukonda

Study supervision: G.M. Delgoffe, S.N. Sarkar

Others (assisted in writing methods for "Detection of tumor specific memory T cell"): F.P. Vendetti

Others (helped L Shao in most of the experiments): R. Srivastava

This work was supported, in part, by AI118896 (to S.N. Sarkar), CA178766 (to S.N. Sarkar), and CA204173 (to C.J. Bakkenist) from NIH. This project used a number of UPCI core facilities that are supported by award P30CA047904 and R50CA211241.

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.