Frequent Loss of IRF2 in Cancers Leads to Immune Evasion through Decreased MHC Class I Antigen Presentation and Increased PD-L1 Expression

The importance of adaptive immunity in preventing cancer was revealed through studies in which immunodeficient animals, such as those lacking IFN-γ, perforin, or RAG-2, were found to have a marked increase in spontaneous and mutagen-induced tumors (1–3). In addition, tumors derived from such immunodeficient animals grew when transplanted into other immunodeficient hosts but were rejected when placed into immunocompetent hosts (3–5), providing further evidence that the immune system recognized such tumors and could reject them. In contrast, many tumors arising in immunocompetent animals grew after being transplanted into immunocompetent hosts (3–5), thereby showing that cancers that arise and successfully progress in the face of the immune system have undergone immunoediting to escape from immune control. This immunoediting process is thought to be why many cancers express low levels of MHC class I (MHC-I) and upregulate certain inhibitory molecules (6). The underlying molecular mechanisms responsible for these changes are poorly understood but have obvious potential impact on tumor progression and immunotherapy (7, 8).

The MHC-I presentation pathway is critical for immune recognition and elimination of tumors by CD8⁺ T cells. In this process, a fraction of peptides that are generated by proteasomal degradation of cellular proteins are transported by the TAP transporter into the endoplasmic reticulum (ER), wherein they can be further trimmed by the aminopeptidase, ERAP1 (9–11). Subsequently, peptides of the correct length and sequence bind to MHC-I molecules and these complexes are then transported to the cell surface for display to CD8⁺ T cells. This allows activated CD8⁺ T cells to identify and kill cells that are presenting tumor-specific peptides (e.g., from mutant proteins) on their MHC-I (12).

We performed an unbiased, forward genetic screen in human cervical carcinoma HeLa H1 cells to identify genes whose loss downregulated the MHC-I pathway. In this screen, the second strongest hit, second only to β2-microglobulin (the MHC-I L chain), was IRF2, an IFN regulatory transcription factor that had not been previously recognized to positively regulate this pathway. In this article, we show that under basal conditions, not only does IRF2 positively regulate the MHC-I pathway by transcriptionally activating genes necessary for peptide transport and processing, but it also transcriptionally represses the expression of programmed death-ligand 1 (PD-L1), an important immune checkpoint molecule. These results, together with our findings that many human cancers have downregulated IRF2 expression and that tumor cells lacking IRF2 are more difficult to kill, demonstrate that loss of IRF2 by cancers is a common immune evasion mechanism and this has obvious therapeutic implications.

DC3.2 is a J2 virus-immortalized dendritic cell (DC) line (13). A particular DC3.2 clone (with Renilla luciferase) was used for all experiments in this study, as this clone has very strong cross-presentation and MHC class II (MHC-II) presentation, as compared with other clones. RF33.70 is a T cell hybridoma that recognizes the OVA peptide OVA_257–264 in the context of H2-K^b (14). MF2.2D9 is a T cell hybridoma that recognizes OVA_258–276 in the context of I-A^b (13). RF33.70 and MF2.2D9 were transduced with lentivirus containing NFAT-luciferase. NIH-3T3 cells were stably transfected with the mouse H2-K^b molecule. A549 and MCF7 were kindly provided by Leslie Shaw (University of Massachusetts), and the D53m and H50m mouse MCA-induced sarcoma lines were kindly provided by R. Schreiber (Washington University in St. Louis). The MCA-induced sarcoma lines were grown in R10 medium, and all other cell lines were grown in RPMI 1640 (Life Technologies) supplemented with 10% FBS (Hyclone), 1% NEAA (Life Technologies), 1% HEPES (Life Technologies), 1% Antibiotic-Antimycotic (Life Technologies), and 5 × 10⁻⁵ M 2-ME (Sigma-Aldrich). The MCF7 growth medium was also supplemented with 10 μg/ml insulin. Antibiotic selection for CRISPR-Cas9 knockout cells was done for 2 wk in medium containing 5 μg/ml blasticidin (InvivoGen). All cells were grown in a 10% CO₂ atmosphere at 37°C.

The LentiCRISPRv2 plus blasticidin selection plasmid (15, 16) was acquired from Addgene (83480) and, unmodified, is the same as the “no short guide RNA (sgRNA)” plasmid. The plasmids used to target mouse B2m, TAP1, TAP2, ERAP1, IRF2, or IRF1 or to target human IRF2 were constructed by inserting the following sgRNA sequences, respectively, into the LentiCRISPRv2 plasmid: mouse B2m: 5′-AGTATACTCACGCCACCCAC-3′; mouse TAP1: 5′-ACTAATGGACTCGCACACGT-3′; mouse TAP2: 5′-ATTACACGACCCGAATAGCG-3′; mouse ERAP1: 5′-TGCAGCATCCAGAGCATAAT-3′; mouse IRF2: 5′-TCCGAACGACCTTCCAAGAA-3′; mouse IRF1: 5′-CTCATCCGCATTCGAGTGAT-3′; human IRF2: 5′-TGCATGCGGCTAGACATGGG-3′. To insert these sgRNA sequences into LentiCRISPRv2, two primers were created for each sgRNA sequence: 1) a forward primer wherein CACCG preceded the aforementioned sgRNA sequence; and 2) a reverse primer where the reverse complement of the aforementioned sgRNA sequence was flanked by AAAC at the 5′ end and C at the 3′ end. Then, these 100 μM primer sets were annealed and diluted 1:50. Three micrograms of LentiCRISPRv2 plasmid was digested for 3 h at 55°C with BsmBI (New England Biolabs [NEB]), and removal of the 2-kb filler sequence was confirmed by gel electrophoresis. The larger m.w. band was gel extracted and quick ligated with the diluted annealed oligos according to the manufacturer’s instructions (NEB). Stable competent Escherichia coli (NEB) was then transformed with 2 μl of the quick ligation product according to the manufacturer’s instructions and grown overnight at 30°C on Luria broth + ampicillin (100 μg/ml) plates. Plasmids were isolated (Clontech) from individual colonies and sequenced (Genewiz) using the primer hU6-F: 5′-GAGGGCCTATTTCCCATGATT-3′ to confirm proper insertion of the sgRNA into LentiCRISPRv2. In addition, sgRNAs were checked for high indel efficiencies in transduced cells by TIDE analysis (17).

Rescue plasmids were constructed by inserting mouse IRF2, TAP2, or ERAP1 cDNA or human IRF2 cDNA into the constitutive expression vector, pCDH-CMV (Addgene), with a modified multiple cloning site. Overlapping PCRs were run using the primers below on either mouse cDNA from DC3.2 cells or human cDNA from HeLa H1 cells to create IRF2 cDNA sequences containing six synonymous mutations within the IRF2 sgRNA target site. The mouse IRF2 K78R sequence was constructed by further mutating A to G at nt 233. The wild-type (WT) TAP2 and ERAP1 cDNA sequences were of C57BL/6 origin. All plasmids were sequenced to confirm correct sequences and reading frames. The primer sets for cloning the rescue/overexpression constructs were as follows: human IRF2 AgeI forward: 5′-GACTACCGGTATGCCGGTGGAAAGGATGCGCATG-3′; human IRF2 sgRNA mut reverse: 5′-GCCGTGCCTCGCTGCGTGCATCCAGGGGATCTGAAAAATCTTCTTTTCCTTG-3′; human IRF2 sgRNA mut forward: 5′-GATGCACGCAGCGAGGCACGGCTGGGATGTGGAAAAAGATGCACCACTCTTTAGAAA-3′; human IRF2 MluI reverse: 5′-GATCACGCGTTTAACAGCTCTTGACGCGGGCCTGG-3′; mouse IRF2 AgeI forward: 5′-GATCACCGGTATGCCGGTGGAACGGATGCGAATG-3′; mouse IRF2 sgRNA mut reverse: 5′-CCTTTTTTCGAGGGGCGCTCTGATAAGGGCAGCATCCGGTAGACTCTGAAGGCG-3′; mouse IRF2 sgRNA mut forward: 5′- CTTATCAGAGCGCCCCTCGAAAAAAGGAAAGAAACCAAAGACAGAAAAAGAAGAGAG-3′; and mouse IRF2 MluI reverse: 5′-GATCACGCGTTTAACAGCTCTTGACACGGGCCTGG-3′.

Where indicated, mouse cells were blocked with 2.4G2 and stained for surface MHC-I levels with anti-K^b–allophycocyanin (AF6-88.5.5.3; eBioscience), MHC-II levels with anti–I^A/I^E-PECy7 (M5/114.15.2; BioLegend), PD-L1 levels with anti–PD-L1-PE (10F.9G2; BioLegend), or with isotype controls (eBioscience mouse IgG2a-allophycocyanin eBM2a, eBioscience rat IgG2b κ-PE eB149) at 1:200 dilutions. Where indicated, human cells were stained for surface MHC-I levels with W6/32. W6/32 staining was performed either by two-step labeling with W6/32 hybridoma supernatant followed by 1:500 donkey–anti-mouse Alexa 647 (Life Technologies) or by one-step labeling with 1:200 FITC-conjugated W6/32 (eBioscience). Where indicated, human cells were stained for surface PD-L1 levels with 1:200 rabbit anti–PD-L1 (28-8; Abcam), followed by 1:500 donkey–anti-rabbit Alexa 647 (Life Technologies). Normalized mean fluorescence intensity (MFI) was computed by dividing the geometric MFI of each knockout cell line by the geometric MFI of the WT (no sgRNA) cell line.

Cross-presentation and MHC-II presentation were measured by coculturing DC3.2 lines in the presence of the indicated concentrations of OVA-coated iron-oxide beads (Polysciences) and RF33.70-Luc CD8⁺ T cells or MF2.2D9-Luc CD4⁺ T cells, respectively, for 16–18 h. Then, One-Glo luciferase substrate (Promega) was added and luciferase activity quantified by an EnVision plate reader (Perkin Elmer). Rescue experiments were performed by adding the OVA beads and RF33.70-Luc cells 48 h posttransduction of the DC3.2 lines. Data from a representative cross-presentation and MHC-II presentation experiment (of n = 4) are shown where points represent mean ± SD of technical duplicates. Normalized CD8⁺ T cell activation was calculated for rescue experiments by dividing the CD8⁺ T cell activation (relative luminometer units of luciferase) at each point by the CD8⁺ T cell activation of the DC3.2 no sgRNA line transduced with empty vector (EV) (WT + EV).

10⁴ HEK293T or HeLa H1 cells were transfected with 10 nM Silencer Select siRNA (Invitrogen) and 0.3 μl Lipofectamine RNAiMAX (Invitrogen) per well in flat-bottom 96-well plates. Individual small interfering RNAs (siRNAs) used were negative control no. 1 (4390843) human β2m (s1854), human TAP1 (s13778), human IRF2 no. 1 (s7506), and human IRF2 no. 2 (s7504). After 72 h, adherent cells were trypsinized, washed in PBS supplemented with 2% FBS, and stained with the surface MHC-I pan–HLA-A/B/C Ab, W6/32. Normalized MFI was computed by dividing the geometric MFI of each experimental siRNA by the geometric MFI of the negative control siRNA.

A total of 10⁴ 3T3-Kb cells were transfected with 100 ng of various pTracer-CMV2 plasmids (Invitrogen) containing SIINFEKL precursors (18) and 0.4 μl Lipofectamine 2000 (Invitrogen) in flat-bottom 96-well plates. After 72 h, cells were stained with 25-D1.16 (specific for the combination of SIINFEKL and H-2K^b) (19), followed by donkey–anti-mouse Alexa 647 (Life Technologies), and analyzed by flow cytometry. Transfected cells were identified by GFP expression and the MFI of 25-D1.16 staining was measured on the gated transfected cells. Transfection efficiency, based on GFP expression, ranged from 5 to 20%, depending on the vector. The normalized MFI for each experiment (with technical duplicates) was calculated by dividing the MFI of each knockout line by the MFI of the WT (no sgRNA) line.

RNA was extracted using the RNeasy kit (Qiagen) after 2 h of stimulating the DC3.2 lines with 5000 U/ml mouse IFN-α, 2 ng/ml mouse IFN-γ, or medium alone. A standard library preparation protocol was used with 50 ng of total RNA as starting material. Libraries were checked for appropriate fragment size traces by Bioanalyzer (Agilent), and concentrations were determined to achieve similar sequencing depth per library. Libraries were run on NextSeq 500/550 high-output and mid-output kits (Illumina) and all libraries had at least 10⁷ reads with single index paired-end sequencing. Trimmomatic-0.32 (20) was used to remove 5′ or 3′ stretches of bases having an average quality of <20 in a window size of 10. Only reads longer than 36 bases were kept for further analysis. RSEM v1.2.28 (21) was used to estimate gene expression, with parameters -p 4–bowtie-e 70–bowtie-chunkmbs 100–strand-specific. Gene quantification was run on the transcriptome (RefSeq v69 downloaded from UCSC Table Browser) (22). Genes with more than 15 TPM in any time point were considered expressed, and genes that did not achieve this threshold were removed from further analysis. Batch effects were observed between samples from different replicates. We used the log-transformed TPM normalized expression values as input to ComBat (package sva version 3.18.0) (23, 24) with default parameters and a model that specified different replicates as batches. Corrected TPM values were transformed back to read counts using the expected size of each transcript informed by RSEM. We only considered genes with at least 15 TPMs in at least one replicate at any time point. The expressed gene list was filtered to include only genes with homologs as defined by the previous step. We used the batch corrected counts per gene to identify differentially expressed genes by at least 2-fold between unstimulated cells (time 0) and 2 h following stimulation with IFN-α or IFN-γ and whose change in expression was significant (p-adjusted < 0.05) according to the package DESeq2 (v1.10.1) (25) in R (v3.5.1). Because of the large transcriptional changes observed in this system, we turned off the fold change shrinkage in DESeq2 with betaPrior = FALSE and we added a pseudocount of 32 to all timepoints to avoid spurious large fold change estimates from lowly abundant genes. RNA sequencing (RNA-seq) data have been deposited to the Gene Expression Omnibus (accession number GSE133089; www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE133089).

Twenty-seven non–small cell lung cancers (NSCLCs) were acquired from the University of Massachusetts Pathology archive of formalin-fixed and paraffin-embedded patient samples. PD-L1 expression was determined at the time of diagnosis by immunohistochemistry (22C3 pharmDx assay; Agilent) by a surgical pathologist. RNA was extracted from the archival material and analyzed by RT-PCR (see below).

RNA was extracted from cell lines using the RNeasy kit (Qiagen) or from formalin-fixed paraffin-embedded NSCLCs using the RNeasy FFPE kit (Qiagen) and reverse transcribed to cDNA using EcoDry premix random hexamers (Clontech). Fifty nanograms of cDNA was used per well (done in triplicate) with the indicated TaqMan probes (Applied Biosystems), according to the manufacturer’s instructions. The TaqMan probes used are as follows: mouse β-actin, 4352933; mouse β2m, Mm00437762_m1; mouse H2-K1, Mm01612247_m1; mouse H2-Ab1, Mm00439216_m1; mouse Tap1, Mm00443188_m1; mouse Tap2, Mm01277033_m1; mouse Erap1, Mm00472842_m1; mouse ERp57, Mm00433130_m1; mouse tapasin, Mm00493417_m1; mouse calnexin, Mm00500330_m1; mouse calreticulin, Mm00482936_m1; mouse Tapbpr, Mm00520408_m1; mouse Irap, Mm00555903_m1; mouse Psme1, Mm00650858_g1; mouse Psme2, Mm01702833_g1; mouse Psmb8, Mm00440207_m1; mouse Psmb9, Mm00479004_m1; mouse Psmb10, Mm00479052_g1; mouse PD-L1, Mm03048248_m1; human GAPDH, Hs02758991_g1; human IRF2, Hs01082884_m1; human TAP2, Hs00241060_m1; and human ERAP1, Hs00429970_m1. mRNA expression levels in the IRF2-knockout (IRF2-KO) DC3.2 were compared with those in the WT DC3.2 by first normalizing to the mRNA expression of β-actin (mouse) in each sample (2^{^−ΔΔCt}, where Ct is the cycle threshold). Statistical analysis was done by comparing the expression of a given gene to that of H2-Ab1. Relative mRNA expression levels in each human lung tumor were measured by first normalizing to the mRNA expression of GAPDH (human) in each specimen (2^{^−ΔCt}). Results are displayed after further normalization to one of the tumors. Statistical analysis was done using two-tailed Mann–Whitney U tests and linear regression models with R² for goodness of fit.

Chromatin immunoprecipitation (ChIP) procedure generally followed the Thermo Fisher Scientific ChIP protocol. In short, 10⁷ DC3.2-no sgRNA cells were stimulated for 2 h with 2 ng/ml IFN-γ or medium alone, harvested and fixed with 1% formaldehyde, quenched with glycine, and washed in cell lysis buffer then nuclear lysis buffer. Chromatin was sheared by sonication for 20 min, and fragment size (∼300 bp) was determined by gel electrophoresis. Immunoprecipitations (IP) of the sheared chromatin were done using 2 μg of primary Ab—normal rabbit IgG (sc-2027; Santa Cruz Biotechnology), rabbit anti-IRF1 (ab186384; Abcam), or rabbit anti-IRF2 (B-80 H53L46; Invitrogen)—by incubating for 1 h at room temperature (RT) then overnight at 4°C. The next day, 25 μl of prewashed Protein A/G beads (Pierce) were added to each of the samples and incubated at RT for 30 min, then 90 min at 4°C. After washing sequentially with low-salt buffer, high-salt buffer, LiCl buffer, and TE buffer, the DNA was eluted from the beads. All immunoprecipitated samples were treated with RNase A (Qiagen) and Proteinase K (Qiagen) and then column purified (Clontech). ChIP–quantitative PCR (qPCR) was performed in triplicate wells using SYBR Green (Bio Rad) and unique primer sets (below) flanking the IRF1/2-binding site within the gene’s promoter (26–28), according to the manufacturer’s instructions. Data shown as fold enrichment (2^{^−ΔCt}) over the normal rabbit IgG control IP (n = 2). The primer sets for ChIP-qPCR were as follows: mouse TAP2 forward: 5′-CAAATTGACAGGCGCCATCT-3′; mouse TAP2 reverse: 5′-GCTTCTTCTCAAACTGGATCTCC-3′; mouse ERAP1 forward: 5′-CTTAGGCTTGCTCTCTTTTAGCG-3′; mouse ERAP1 reverse: 5′-GACTCCTGCTCCCGATCCTC-3′; mouse PD-L1 forward: 5′-CAAGAAAGCTAATGCAGGTTTCAC-3′; and mouse PD-L1 reverse: 5′-CCTGCGGATGACTTTAGAGTC-3′.

Whole-cell lysates were prepared in RIPA buffer with protease inhibitor (Pierce), protein concentrations were determined by BCA assay (Pierce), and 10 μg of denatured samples were run on 10% reducing gels (GenScript). After transfer, PVDF membranes (MilliporeSigma) were blocked with TBS–Tween 1× + 5% milk and then blotted with rabbit anti-IRF2 (ab124744; Abcam) or rabbit anti-IRF1 (ab186384; Abcam) in TBS–Tween 1× + 2% milk overnight at 4°C. The following day, membranes were washed three times with TBS–Tween 1×, goat-anti-rabbit HRP (MilliporeSigma) was added for 1 h at RT, membranes were washed 3×, and HRP substrate (MilliporeSigma) was added. Following exposure, membranes were stripped (MilliporeSigma), blocked, and reblotted with mouse anti–β-actin (sc-47778; Santa Cruz Biotechnology) in TBS–Tween 1× + 2% milk overnight at 4°C. The following day, membranes were prepared as above except anti-mouse HRP (Pierce) was used instead.

OT-I CD8⁺ T cells were preactivated by coculturing them for 4–5 d with irradiated, SIINFEKL-pulsed WT LPS-stimulated B cell blasts in T cell medium containing 30 ng/ml IL-2. After this time, OT-I were checked for CD8 expression and upregulation of CD27 and CD44. WT or IRF2-null RMA cells were counted, subdivided into respective tubes, SIINFEKL-pulsed at the indicated concentrations or kept unpulsed, and CFSE-labeled as high (1 μM) for cells pulsed with SIINFEKL and CFSE-labeled as low (0.1 μM) for unpulsed cells. Pulsed and unpulsed cells were recounted and mixed 1:1, and 10⁵ cells total were plated per well in U-bottom 96-well plates. A total of 5 × 10⁴ OT-I were added to the respective wells and incubated at 37°C for 4 h. Live RMA cells were gated by flow cytometry, and CFSE levels analyzed. Specific killing was determined for each RMA line by calculating 100 × (1 – [(% targets / % bystanders)/(% control targets / % control bystanders)]).

When HeLa H1 cervical carcinoma cells were immunoselected for MHC-I low variants in a genome-wide CRISPR-Cas9 screen, IRF2 scored as the second most targeted gene, with six out of six independent IRF2 guide RNAs hitting. To validate this hit, we tested whether IRF2 affects surface MHC-I levels in human cells by knocking out IRF2. HeLa H1 cells and HEK293T kidney cells were transduced with vectors expressing either Cas9 and a sgRNA targeting human IRF2 or Cas9 alone, and surface MHC-I levels were checked by flow cytometry (Fig. 1A). The IRF2-KO HeLa H1 and HEK293T had significantly lower surface MHC-I levels than their WT controls (Fig. 1B). To further validate this finding with another technique, we silenced IRF2 expression with two independent siRNAs and found that both siRNAs similarly decreased surface MHC-I levels in HeLa H1 and HEK293T (Fig. 1C), confirming the phenotype. The magnitude of the decrease in MHC-I was similar to that observed when the expression of the TAP transporter was silenced (Fig. 1C). To determine whether IRF2 also affects surface MHC-I levels in mouse cells, we transduced NIH-3T3 fibroblasts stably transfected with H2-K^b (3T3-K^b) and DC3.2 DCs (13) with vectors expressing either Cas9 and a sgRNA targeting mouse IRF2 or Cas9 alone. The IRF2-KO mouse fibroblasts and DC cells also had significantly reduced surface MHC-I levels (Fig. 1D). Disruption of the IRF2 gene was validated by TIDE analysis (17), and loss of IRF2 expression was confirmed by Western blot (Fig. 1E). DC3.2 cells also express MHC-II molecules, and we found no change in surface MHC-II levels in the IRF2-KO cells (Fig. 1D), which demonstrates that IRF2 is selectively affecting the MHC-I pathway; further evidence supporting this conclusion will be described below.

To investigate the functional consequence of this reduction in MHC-I levels, we evaluated the importance of IRF2 for MHC-I cross-presentation (i.e., the presentation of peptides derived from exogenous Ag on MHC-I). The IRF2-KO DCs cross-presented more poorly than their WT controls (Fig. 1F), demonstrating that IRF2 positively regulates MHC-I Ag presentation. In the same experiments, MHC-II presentation was unaffected, again showing selectivity in IRF2 effects (Fig. 1G). Finally, to confirm that IRF2 is responsible for these differences, we overexpressed IRF2 in the IRF2-KO DC3.2 line and found that it completely restored surface MHC-I levels (Fig. 1H) and the ability of these cells to cross-present Ag (Fig. 1I). Evidence that loss of IRF2 also compromises the MHC-I presentation of endogenous cellular Ags will be described below.

To determine whether IRF2 was exerting its function as a transcription factor, we introduced a lysine to arginine point mutation at position 78, which prevents acetylation at this site and thereby prevents IRF2 from binding its DNA target sequences (29). Overexpressing this mutant IRF2 did not restore function in the IRF2-KO DC3.2 cells (Fig. 1I). Therefore, IRF2 is necessary for optimal transcriptional regulation of MHC-I Ag presentation pathways.

Because IRF2 is functioning as a transcription factor, we next sought to determine what genes are regulated by IRF2 and could be responsible for producing the low MHC-I phenotype observed in the IRF2-KO cells. RNA-seq was performed on the WT and IRF2-KO DC3.2 lines (Fig. 2A, Supplemental Table I). Surprisingly, relatively few genes were differentially expressed by >2-fold (19 decreased and 33 increased). Of these 52 differentially expressed genes, we identified TAP2, ERAP1, and the immunoproteasome subunit PSME1 as potential contributors to the decreased MHC-I levels. To further confirm this result, we performed qPCR in these cell lines to check expression of all MHC-I pathway genes (Fig. 2B), which showed that the mRNA levels of TAP2, ERAP1, TAPBPR, and PSME9 (another immunoproteasome subunit) were significantly downregulated in the IRF2-KO DC3.2; PSME1 levels were reduced but this decrease did not achieve statistical significance. Interestingly, the mRNA levels of the MHC H chain (H2-K1) and MHC L chain (β2-microglobulin; β2m) were unaffected (Fig. 2B), indicating that IRF2 is not required for synthesis of the MHC-I heterodimer. ChIP-qPCR experiments confirmed that IRF2 regulates TAP2 and ERAP1 mRNA expression by directly binding to their promoters (Fig. 2C). To test whether functional IRF2 is needed for TAP2 and ERAP1 mRNA expression, we transduced WT IRF2 or the IRF2-K78R mutant in the DC3.2 IRF2-KO cells and found that the TAP2 and ERAP1 mRNA levels were increased in the cells expressing WT but not mutant IRF2 (Supplemental Fig. 1).

To more closely examine the functional effects of IRF2 on peptide transport and trimming, 3T3-K^b IRF2-KO cells were transfected with various pTracer plasmids, each of which contained a GFP reporter and a minigene encoding a peptide that could be processed down to the mature epitope (i.e., SIINFEKL/S8L). The cells were surface stained with the Ab 25-D1.16, which recognizes H2-K^b-S8L complexes (19), and transfected cells (GFP-positive) were analyzed. In each case, β2m-knockout cells were included to show MHC-I dependence and the maximum inhibition of S8L presentation, and TAP1-, TAP2-, and ERAP1-knockout cells were included as positive or negative controls, depending on the minigene examined. IRF2-KO cells presented fewer H2-K^b-S8L complexes than WT cells when given the TAP-dependent, ERAP1-dependent Ags CD16-OVA (full-length OVA protein) (30), N25-S8L (a S8L precursor extended by 25 aa on the N terminus), or N5-S8L (a S8L precursor extended by 5 aa on the N terminus) (Fig. 3A). We also tested the presentation of a version of the precursor peptide with five extra N-terminal residues that was targeted into the ER by a colinear signal sequence (ss-N5-S8L). Because the signal sequence allows this peptide to enter the ER through SEC61 instead of TAP, its presentation is TAP independent but still dependent on ERAP1 to remove the extra N-terminal residues (Fig. 3B). IRF2-KO cells also presented fewer H2-K^b-S8L complexes than WT cells when transfected with ss-N5-S8L (Fig. 3B). However, IRF2-deficient cells were equally capable of presenting H2-K^b-S8L complexes when given a TAP-independent, ERAP1-independent peptide (S8L with no extra N-terminal residues that was targeted into the ER via a colinear signal sequence; ss-S8L) (Fig. 3C), demonstrating that although IRF2 affects the transport and processing of MHC-I epitopes, it does not affect the ability of such peptides to be loaded onto MHC-I.

We also compared the MHC-I phenotype in IRF2-KO 3T3-K^b to that observed in the same cells with knockouts of TAP2 or ERAP1. The magnitude of the reduction in MHC-I levels on IRF2-KO cells was in between that of the TAP2- and ERAP1-knockout cells (Fig. 3D), which is consistent with our findings that IRF2 positively regulates TAP2 and ERAP1 but their expression is not entirely lost in IRF2-KO cells. Finally, we performed rescue experiments wherein we overexpressed TAP2 and/or ERAP1 in the IRF2-KO DC3.2 and checked surface MHC-I levels 2 d after transduction (Fig. 3E). Although the double rescue partially restored MHC-I levels, it was not complete, suggesting that other genes regulated by IRF2 also contribute to surface MHC-I expression.

One of the upregulated genes in the IRF2-KO cells was Cd274 (Supplemental Fig. 1), also known as PD-L1, which is often upregulated in certain cancers (e.g., NSCLC) and functions as a checkpoint inhibitor to suppress Ag-specific CD8⁺ T cell effector function (31, 32). To validate our RNA-seq finding with an independent technique, we performed qPCR on the IRF2-KO and WT DC3.2 lines and found that the IRF2-KO cells expressed approximately twice as much PD-L1 mRNA as the WT controls (Fig. 4A). Additionally, ChIP-qPCR revealed that IRF2 regulates PD-L1 mRNA expression by directly binding the PD-L1 promoter (Fig. 4B). Therefore, IRF2 acts as a transcriptional repressor of PD-L1 in these cells. To evaluate the extent to which a roughly 2-fold increase in PD-L1 mRNA translates to surface PD-L1 expression, we analyzed these cells by flow cytometry (Fig. 4C). This analysis reveals that the surface PD-L1 levels also increased by roughly 50% in the IRF2-KO cells (Fig. 4D). Taken together, these results demonstrate that IRF2 plays a role in repressing PD-L1 expression under basal conditions.

Because IRF2 is an IFN regulatory transcription factor, we wanted to see how IFN induction would affect IRF2’s regulation of the MHC-I pathway and PD-L1 expression. Interestingly, stimulation with either IFN-γ or IFN-α restored the surface MHC-I expression in the IRF2-KO DC3.2 (Fig. 5A, Supplemental Fig. 2). This is an important finding because it indicates that impairment of the MHC-I pathway from loss of IRF2 is reversible.

IRF1 and IRF2 recognize the same IFN-stimulated response element (ISRE) (33–35) and, whereas IRF2 is constitutively expressed and minimally affected by IFN induction, IRF1 is significantly upregulated in response to IFN (Fig. 5B, Supplemental Fig. 2) (36–38). Such upregulation of IRF1 causes IRF1 to compete with IRF2 for binding to the TAP2, ERAP1, and PD-L1 promoters, ultimately displacing it from them (Fig. 5C). Consistent with the literature that IRF1 positively regulates both MHC-I and PD-L1 expression under IFN-stimulated conditions (39–41), we found that knocking out IRF1 decreased both surface MHC-I and PD-L1 levels after IFN-γ stimulation (Fig. 5A). Interestingly, when we examined the dual effects of IRF1 and IRF2 on surface MHC-I and PD-L1 levels using single or double knockout DC3.2 lines (Fig. 5A, Supplemental Fig. 2), we found that 1) IRF1/2 double knockouts have a larger reduction in surface MHC-I than is observed in either single knockout; and 2) IRF2-KOs have a larger effect than the IRF1-knockouts on both MHC-I and PD-L1 expression under basal conditions. This suggested that, although these two IRFs recognize the same ISRE, the subset of genes they each primarily regulate differs and that a cell’s dependence on IRF2 versus IRF1 for any given gene may vary depending on the cues (e.g., IFN), which that cell receives from its environment. To better characterize these expression changes more globally, we performed RNA-seq on the single and double knockout DC3.2 lines under basal conditions (Fig. 5D) or after adding IFN-γ (Fig. 5E) or IFN-α (Supplemental Fig. 2). From this analysis, several genes known to be important for Ag presentation and immune cell function were identified (Supplemental Table I). Furthermore, this analysis showed that genes influenced by IRF1 and IRF2 could be grouped into multiple classes. Under basal conditions, there was a subset of Ag presentation–related genes whose expression was activated by IRF2 (e.g., TAP2, ERAP1), others that were repressed by IRF2 and remained so in the double knockouts (e.g., H2-T9), and yet others that were repressed by IRF2 but did not remain so in the double knockouts (e.g., PSMB8, PSMB10) (Fig. 5D). After stimulating with IFN-γ, some genes were primarily activated by IRF1 (e.g., PSME1, PSME2, PSMB9), others were primarily repressed by IRF2 (e.g., PD-L1), and yet others were activated by both IRF1 and IRF2 (e.g., TAP2, ERAP1) (Fig. 5E). Collectively, these studies demonstrate that whereas some genes are acted on antagonistically by IRF1 and IRF2, other genes are regulated synergistically by these two transcription factors and that the relative contributions of IRF1 versus IRF2 in mediating these expression changes varies depending on the inflammatory state of the cell.

Given that experimentally induced loss of IRF2 both compromises MHC-I presentation and increases PD-L1 expression, it was of interest to see how often IRF2 is downregulated in primary cancers. We used the TIMER bioinformatics tool (42) to mine publicly available databases for IRF2 expression in primary human cancers. Remarkably, IRF2 was downregulated in several kinds of human cancers and the overall reductions were highly statistically significant (Fig. 6A). For each of the IRF2-low cancers, which included invasive breast carcinoma, cholangiocarcinoma, colon adenocarcinoma, liver hepatocellular carcinoma, lung adenocarcinoma and squamous cell carcinoma (NSCLCs), prostate, rectum and stomach adenocarcinomas, and uterine corpus endometrial carcinoma, a subset of patients had very low levels of IRF2. We chose one of the IRF2-low cancers, NSCLC, to determine whether IRF2 levels were functionally limiting in primary cancers. At our institution, NSCLCs are screened for PD-L1 expression by immunohistochemistry at the time of diagnosis, which enabled us to randomly select tumors spanning a spectrum of PD-L1 expression. We extracted RNA from archival patient biopsy material and quantified expression of IRF2, TAP2, and ERAP1 by qPCR. In these lung cancers, IRF2 mRNA levels and PD-L1 immunohistochemistry status were significantly inversely correlated (Fig. 6B). Additionally, consistent with our cell line findings, TAP2 and ERAP1 mRNA levels positively and significantly correlated with IRF2 mRNA levels (Fig. 6C). To formally test cause and effect for these correlations, we analyzed a human NSCLC cell line, A549, that is IRF2-low relative to other NSCLCs tested in the NCI-60 panel (43). A549 cells are also MHC-I low and PD-L1 positive (44, 45). Eliminating the residual IRF2 in A549 cells by CRISPR-Cas9–mediated knockout further decreased surface MHC-I but did not further increase surface PD-L1 (Fig. 6D). In contrast, restoring IRF2 by transfection repressed surface PD-L1 expression and increased surface MHC-I expression (Fig. 6D). Stimulation of A549 with IFN-γ augmented both surface MHC-I and PD-L1 expression, as expected, but transfection of IRF2 still had the same pattern of effects as without IFN (repressing PD-L1 and further increasing MHC-I expression) (Fig. 6E). To further generalize these findings to other cancers, we analyzed two human breast cancers (chosen because breast cancers are often IRF2 low) (Fig. 6A) as well as two mouse sarcomas (chosen because sarcomas are known to undergo immunoediting) (3) and a mouse lymphoma (chosen for use in CD8⁺ T cell killing assays) and found similar results (Fig. 6F, Supplemental Fig. 3).

The effects of IRF2 loss on Ag presentation and checkpoint inhibition are predicted to make it harder for CD8⁺ T cells to kill IRF2-low cells. To test this and quantify the magnitude of the effect, we analyzed mouse WT versus IRF2-KO lymphoma (RMA) cell lines. RMA cells were chosen because they are known to be very good targets for cytotoxic T cell killing assays (and are, therefore, a stringent test) and express H2-K^b, which allows the use of potent CD8⁺ T cell effectors from the H2-K^b-S8L–specific TCR transgenic OT-I model. Preactivated OT-I effectors were cultured with pairs of WT or IRF2-KO cells that were S8L-pulsed or not and labeled with different amounts of the dye CFSE. After 4 h, specific killing was quantified by flow cytometry and the dose-response curve for the IRF2-KO cells was shifted up ∼3-fold (which equates to a decreased killing efficiency of ∼67%) in the IRF2-KO RMA, as compared with the WT RMA (Fig. 6G), demonstrating that tumor cells lacking IRF2 are harder for CD8⁺ T cells to eliminate. Taken together, these findings show that IRF2 downregulation leads to immune evasion and that there are several types of human cancers that may use this escape mechanism.

Our findings that IRF2 both positively regulates the MHC-I pathway and negatively regulates PD-L1 expression have implications for cancer progression and immunotherapy. Cancers need to evade the immune system and, if they are immunogenic, require editing to escape and progress (6). One of the ways that murine tumors can evade immune elimination is by downregulating the MHC-I pathway (3). Similar immunoediting of this pathway occurs in humans as it has been found that tumors, which were predominantly MHC-I positive at early stages, subsequently become homogeneously MHC-I deficient (46, 47), and progressing cancers are frequently MHC-I deficient (48, 49). Another way tumors can evade immune elimination is by expressing checkpoint inhibitors. Indeed, many human cancers upregulate PD-L1, and those that do tend to be more aggressive and have fewer T cells present in the tumor (50). Blocking PD-L1 or its receptor (PD-1) can lead to tumor rejection in both mice and humans, proving that this is an important immune evasion mechanism (51, 52). The molecular mechanisms that allow cancers to exploit these immune evasion strategies are incompletely understood but are important to elucidate for understanding pathogenesis, how they affect prognosis and immunotherapy, and how they might be reversed to improve therapy.

Our studies have uncovered an immune evasion mechanism that is relatively frequent. Many cancers of diverse origins (NSCLC, breast, colorectal, liver, stomach, prostate, uterine) downregulate IRF2 expression, which is a molecule that is not essential for viability in fully differentiated cells. In silico analysis through cBioPortal (53, 54) revealed that IRF2 is rarely mutated in cancers and, therefore, the loss of expression must be epigenetic. As a result of losing IRF2, tumor cells decrease their surface MHC-I expression and increase their surface PD-L1 expression. Thus, loss of a single gene can generate a “double whammy” for the immune system, as it enables tumor cells lacking IRF2 to become both harder to identify (loss of MHC-I Ag presentation) and better able to suppress T cell–mediated elimination (increased checkpoint inhibition). As shown in our in vitro cytotoxicity assay, the loss of IRF2 renders such tumor cells more difficult for CD8⁺ T cells to kill.

Analysis of IRF2-KO cells revealed that PD-L1 was highly upregulated under basal conditions. It is well established that IRF1 can promote IFN-γ–inducible activation of PD-L1 (40, 41). Recently, Dorand et al. (55) found that hyperphosphorylation of IRF2BP2 (an IRF2-binding protein) could lead to decreased PD-L1 expression after IFN-γ stimulation. Additionally, Wu et al. (56) recently discovered that a loss of IRF2BP2 can lead to enhanced IRF2 binding to the PD-L1 promoter and that IFN-γ stimulation releases IRF2 from the PD-L1 promoter. However, the role of IRF2 on PD-L1 expression had not been directly examined. Our ChIP-qPCR data shows that IFN-γ stimulation leads to both increased IRF1 binding and decreased IRF2 binding to the PD-L1 promoter. In addition, we show in this study for the first time, to our knowledge, that, under basal conditions, a loss of IRF2 significantly increases PD-L1 mRNA and surface expression, and IRF2 overexpression produces the opposite effects. Collectively, these results establish the repressive role of IRF2 on PD-L1 expression in the absence and presence of IFN-γ and reveal that unstimulated cells can upregulate PD-L1 expression if upstream regulatory factors, such as IRF2, are defective/absent.

The transcriptional control of many MHC-I pathway components, particularly under basal conditions, has not been well defined (57). In this study, we identified IRF2 as a novel positive regulator of MHC-I Ag presentation and show that, under basal conditions, it is important for both classical MHC-I presentation and cross-presentation. IRF2 binds the promoters of TAP2 and ERAP1 and transcriptionally activates their expression. In the absence of IRF2, cells express less TAP2 and ERAP1 and, because of the consequent defects in Ag transport and processing, present fewer peptide-MHC-I complexes at the cell surface. Additionally, IRF2 regulates the expression of many other genes, as demonstrated by others (58, 59) and shown in this study, such as immunoproteasome subunits, which likely also influence the MHC-I presentation in IRF2-deficient cells (60). Interestingly, in one of these earlier studies, IRF2 was reported to bind to the promoter of TAP1 as assessed by EMSAs (59); however, we found no evidence that IRF2 was needed for transcription of endogenous TAP1 by either RNA-seq or qPCR. Most likely, this discrepancy is due to the nature of the different assays (e.g., extract binding to exogenous probes versus endogenous promoter activity).

In the DC line where we focused most of our studies and in other cell lines described in the literature, IRF2 is constitutively expressed, relatively stable with a t_1/2 of 8 h, and minimally affected by IFN (36–38). However, in these same sets of cells, IRF1 is minimally expressed under basal conditions but strongly induced in the presence of IFN and more short-lived than IRF2, with a t_1/2 of only 0.5 h (36). Because of these differences between IRF2 and IRF1, even though they both recognize and bind to the same ISRE (33–35), we hypothesized that the relative contributions of IRF2 and IRF1 to certain signaling pathways varies depending on the inflammatory state of the cell. Although IRF1 and IRF2 were originally characterized as an activator and repressor of IFN-α/β expression, respectively (35), several reports since then have highlighted that IRF1 and IRF2 do not always act as such (37, 61, 62). Yet, to the best of our knowledge, there have been no global differential expression analyses in IRF1- and/or IRF2-lacking cells of the same cell type to better understand the synergistic versus antagonistic roles these transcription factors perform. Here, we conducted RNA-seq on IRF1- and/or IRF2-KO DCs in the absence or presence of IFN to help fill this void. Globally, we found that the expression profile of the double knockout DCs differed from that of either IRF single knockout and that the relative contributions of IRF1 versus IRF2, in terms of their ability to positively or negatively regulate certain genes, varied depending on the inflammatory state of the cell. Transcriptional control of genes can be highly cell-type dependent, and therefore, it is possible that in other cells, the contribution of IRF2 on surface MHC-I and PD-L1 expression may vary. However, our concordant observations in multiple mouse and human cell lines and primary tumors suggest that the number of cell types in which IRF2 regulates these genes in this manner is quite large.

Immune evasion due to tumor PD-L1 upregulation can be reversed by blocking the PD-L1/PD-1 interaction, which is the basis of targeted checkpoint blockade immunotherapy. However, any natural or invigorated (from checkpoint blockade) CD8⁺ T cell response to kill tumors that have downregulated their MHC-I expression will continue to be impaired. In this context, it is of considerable interest and potential importance that the downregulation of the MHC-I pathway from the loss of IRF2 is reversible. When IRF2-deficient cells were treated with IFN, MHC-I levels were restored, likely because of induction of IRF1 and possibly some other transcriptional activators. These findings suggest that IFNs (which are Food and Drug Administration approved for other indications) or IFN-inducing agents could be used to restore MHC-I Ag presentation in IRF2-low tumors. This would be predicted to enhance the effects of immunotherapies, such as checkpoint blockade, that are ultimately dependent on TCR recognition of tumor MHC-I presentation. Currently, checkpoint therapy is effective in only some patients and, based on the mechanisms we have uncovered, it is conceivable that reversing the IRF2 defects might increase the number of patients that can benefit. In addition, because checkpoint blockade is an extremely expensive therapy and one that can have serious side effects, there is a need for good biomarkers to identify those patients that would be more likely to benefit from this type of therapy. Therefore, it will be of interest to examine in future studies whether expression of IRF2 and its downstream target genes (e.g., TAP2 and ERAP1) could be used as biomarkers to help identify checkpoint blockade-responsive patients.

We thank Kate Fitzgerald for helpful discussions, Janice BelleIsle for technical assistance, and Leslie Shaw and Robert Schreiber for sharing cells and reagents.

The authors have no financial conflicts of interest.