The SETDB1–TRIM28 Complex Suppresses Antitumor Immunity

Immunotherapies such as immune checkpoint blockade (ICB) using mAbs targeting the PD-1/PD-L1 axis show striking clinical benefit in several cancer types. Despite this advance, the majority of cancers show low response rates to ICB (1). Thus, new therapeutic strategies are urgently needed to expand the utility of ICB in the treatment of cancer. Combining ICB with other therapeutic approaches is an effective strategy to overcome this challenge (2). For example, genetic alterations that increase PD-L1 expression often predict a better response to anti–PD-1/PD-L1 therapy (3). Likewise, therapeutics that enhance PD-L1 expression typically synergize with anti–PD-1/PD-L1 therapy (4). Chromatin regulators are of emerging interest in this regard because they have a key role in shaping antitumor immunity through cell-intrinsic mechanisms (5, 6).

Pathways that regulated host innate immunity, such as the cyclic GMP–AMP synthase (cGAS)–stimulator of interferon genes (STING) pathway, play an important role in antitumor immunity and represent an attractive target to boost therapeutic responses to immunotherapies, including ICB (7). cGAS is a cytosolic nucleotidyltransferase that binds double-stranded DNA in a sequence-nonspecific manner (8, 9). For example, cGAS senses cytosolic DNA generated by unrepaired DNA lesions and/or mitosis defects during the G₂–M-phase of the cell cycle; such DNA is often in the form of micronuclei (10, 11). Micronuclei are typically positive for both cGAS and DNA damage markers such as γH2AX (10, 11). cGAS activation generates the cyclic dinucleotide cyclic GMP–AMP, which in turn triggers a type I IFN response via the adaptor protein STING (12). Activation of cGAS–STING signaling by micronuclei leads to upregulation of unique type I IFN response genes, including CCL5 and CXCL10 (10).

SET domain bifurcated histone lysine methyltransferase 1 (SETDB1) is a protein lysine methyltransferase that methylates histone H3 at lysine 9 to transcriptionally silence expression of its target gene (13). SETDB1 is a component of the human silencing hub (HUSH) complex that includes tripartite motif–containing 28 (TRIM28, also known as KAP1; ref. 14). TRIM28 plays a key role in the recruitment of SETDB1 to specific genomic loci, and both SETDB1 and TRIM28 are implicated as oncogenes in several cancer types (15, 16). For example, SETDB1 overexpression correlates with a poor prognosis in non–small cell lung cancer, colon cancer, acute myeloid leukemia, and melanoma (17–22). Likewise, TRIM28 overexpression predicts poor prognosis in gastric cancer, ovarian cancer, glioma, and hepatocellular carcinoma (23–28). The SETDB1–TRIM28 complex plays a pivotal role in silencing of transposable elements (TE). Indeed, a recent study shows that SETDB1 loss derepresses latent TE-derived regulatory elements, immunostimulatory genes, and TE-encoded retroviral antigens in these regions, and triggers TE-specific cytotoxic T–cell responses (29). However, the role of SETDB1 in regulating responses to ICB and its underlying mechanism has never been investigated. Here, we show that loss of the SETDB1–TRIM28 complex synergizes with ICB through a two-pronged mechanism: simultaneously upregulating PD-L1 expression and enhancing infiltration of effector CD8⁺ T cells by activating the cGAS–STING innate immune pathway.

ID8 (RRID:CVCL_VA22), UPK10 (gift from J. Conejo-Garcia, Moffitt Cancer Center, Tampa, FL), and HEK293FT (RRID:CVCL_0045) cells were cultured in DMEM (CORNING, #10-013-CM) supplemented with 10% FBS (R&D Systems, #S11510) and 1% penicillin/streptomycin (CORNING, #30-002-CI) at 37°C with 5% CO₂. The human high-grade serous ovarian cancer (HGSOC) cell lines A1847 (RRID:CVCL_9724), PEO4 (RRID:CVCL_2690), OVSAHO (RRID:CVCL_3114), OVCAR10 (RRID:CVCL_4377), OVCAR5 (RRID:CVCL_1628), OVCAR4 (RRID:CVCL_1627), OVCAR3 (RRID:CVCL_0465), CAOV3 (RRID:CVCL_0201), COV362 (RRID:CVCL_2420), COV318 (RRID:CVCL_2419), EF027 (obtained from Gordon Mills' laboratory, MD Anderson Cancer Center, Houston, TX), Kuramochi (RRID:CVCL_1345) were cultured in RPMI1640 (CORNING, #10-040-CM) supplemented with 10% FBS and 1% penicillin/streptomycin at 37°C with 5% CO₂. The human fallopian tube epithelial cells FT246 (RRID:CVCL_UH61) and FT237 (RRID:CVCL_UH59; ref. 30) were gifts from Ronny Drapkin at the University of Pennsylvania (Philadelphia, PA), and they were grown in DMEM/F12 (CORNING, #10-092-CM) with 10% FBS. All the cell lines were authenticated at The Wistar Institute Genomics Facility using short tandem repeat DNA profiling. Mycoplasma testing was performed using LookOut Mycoplasma PCR detection (Millipore Sigma, #MP0035-1KT) every month. Transfection was performed using Lipofectamine 2000 (Invitrogen, #11668027) following the manufacturer's instructions (see “Lentivirus infection” below). ID8 cells were treated with 2.5 μg/mL RU.521 (Invitrogen, #inh-ru521) for 48 hours to inhibit cGAS activity. To activate the cGAS–STING pathway, ID8 cells were transfected with 10 μg IFN stimulatory DNA (InvivoGen, #tlrl-isdn) in a 6 cm dish using lipofectamine 2000 at a ratio of 1 μL lipofectamine to 1 μg DNA following the manufacturer's instructions.

A previously published library of single-guide RNAs (sgRNA) targeting 1,218 genes that encode chromatin regulators was obtained from Kristian Helin (31). ID8 cells were first transduced by adding filtered DMEM (CORNING, #10-013-CM) containing Cas9-expressing lentivirus using a Lenti-EF1a-Cas9-2A-Blast construct (Addgene, #52962, RRID:Addgene_52962). Polybrene (10 μg/mL; Santa Cruz Biotechnology, #sc-134220) was added to enhance lentiviral injection. A single stable Cas9-expressing ID8 clone was selected by seeding single cells into 96-well to grow into clones. A total of 48 × 10⁶ Cas9-expressing ID8 cells were transduced by adding a predetermined amount of filtered DMEM containing virus expressing the mouse epigenetic gDNA library at a multiplicity of infection (MOI) of 0.3 to achieve a library representation of 1,000× (see “Lentivirus infection” below). The established cell library was then treated with 30 ng/mL IFNγ (STEMCELL, #78021) for 24 hours and stained with fluorochrome-conjugated anti–PD-L1 (BioLegend, #124312, RRID:AB_10612741) and sorted with flow cytometry (see “Flow cytometry analysis” below). The top 15% of PD-L1–expressing cells were sorted using flow cytometry. Unsorted bulk cells were used as internal controls.

Genomic DNA was extracted from the PD-L1^high cells and unsorted control cells using salt precipitation. In brief, cell pellets (1–5 × 10⁷ cells) were suspended in 6 mL gDNA lysis buffer (50 mmol/L EDTA, 1% SDS, 50 mmol/L Tris pH 8.0) and 30 μL 20 mg/mL proteinase K (Thermo Fisher Scientific, #25-530-049) was added before the mixture was incubated at 55°C overnight. Then, 30 μL 10 mg/mL RNase A (Thermo Fisher Scientific, #EN0531) was added and the mixture incubated at 37°C for 30 minutes, before cooling on ice. Protein was precipitated by adding 2 mL 7.5 mol/L stock solution of Ammonium Acetate and centrifuging at > 4,000 × g for 10 minutes. The supernatant containing genomic DNA was collected and precipitated by adding 6 mL 100% isopropanol and centrifuging at > 4,000 × g for 10 minutes. The DNA pellet was washed with 70% ethanol, air dried and dissolved in H₂O for PCR amplification.

Library was constructed by a two-step PCR amplification. The first PCR was performed in eight PCR reactions each containing 5 μg gDNA, 1.5 μL 10 μmol/L forward and reverse primers (see below for the sequences), 50 μL NEBNext Q5 HotStart HiFi PCR Mastermix (NEB, #M0543L) for 15 cycles of 98°C, 20 seconds; 60°C, 30 seconds; 65°C, 45°C using a Bio-Rad T100 thermal cycler PCR machine (#1861096). The first PCR product from the eight reactions for each sample was pooled together and purified by AMPure XP beads (Beckman Coulter, #A63880) following the manufacturer's instructions and suspended in 800 μL H₂O. To barcode the samples, for each sample, 3-second PCR reactions containing 10 μL first PCR product, 1 μL 10 μmol/L forward and reverse Illumina primers, 50 μL NEBNext Q5 HotStart HiFi PCR Mastermix for 20 cycles of 98°C, 20 seconds; 60°C, 30 seconds; 65°C, 45°C. The final PCR product was purified by AMPure XP beads (Beckman Coulter) following the manufacturer's instructions and sequenced in a 75 bp single-end run on the Next Seq 500 (Illumina) at the Wistar Genomic facility. The following primers were used: first PCR forward: 5′-AATGGACTATCATATGCTTACCGTAACTTGAAAGTATTTCG-3′; first PCR reverse: 5′-TCTACTATTCTTTCCCCTGCACTGTTGTGGGCGATGTGCGCTCTG-3′; second PCR Universal P7 primer: 5′-CAAGCAGAAGACGGCATACGAGATGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTTCCTTGGTTCTACTATTCTTTCCCCTGCACTGT-3′; second PCR P5 primer for total unsorted sample: 5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTtAAGTAGAGTCTTGTGGAAAGGACGAAACACCG-3′; second PCR P5 primer for PD-L1–high sample: 5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTATACACGATCTCTTGTGGAAAGGACGAAACACCG-3′; second PCR P5 primer for PD-L1 low sample: 5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTGATCGCGCGGTTCTTGTGGAAAGGACGAAACACCG-3′.

pLentiCRISPR v2 (Addgene #52961, RRID:Addgene_52961) and pLentiCRISPR v2-blast (Addgene #83480, RRID:Addgene_83480) were digested with BsmBI (NEB, #R0739) at 55°C for 1 hour and run on a 1% agarose gel. The digested plasmid was cut out and purified using QIAquick gel extraction kit (QIAGEN, #166047244). Each pair of oligos were phosphorylated using T4 PNK (M0201S) in T4 ligation buffer (New England Biolabs, #B0202S) and annealed in a thermocycler at 37°C for 30 minutes, 95°C for 5 minutes, ramped down to 25°C at 5°C/minute. Annealed oligonucleotides were diluted 1:200 in RNase/DNase-free water. Ligation of the annealed oligonucleotide and digested pLentiCRISPR v2 plasmid was performed using Quick Ligase (New England Biolabs, #M2200). The following oligonucleotides were used for cloning: mouse cGas gRNA (5′- CGAGGCGCGGAAAGTCGTAA-3′), mouse Setdb1 gRNA#1 (5′-ACTATTGCAACTCAACCACG-3′), mouse Setdb1 gRNA#2 (5′- ACTCTGGCGCCCGACCGCAA-3′), mouse Trim28 gRNA#1 (5′- ATCCCCCGAATTATTCGCTG-3′), mouse Trim28 gRNA#2 (5′- CGCCGCAGCGAATAATTCGG-3′).

HEK293FT cells were transfected with target plasmids (see below), packaging plasmids psPAX2 (Addgene #12260, RRID:Addgene_12260) and pCMV-VSV (Addgene #8454, RRID:Addgene_8454) using Lipofectamine 2000 at a ratio of 1 μL lipofectamine to 1 μg DNA for 6 hours, after which the medium was replaced. Lentivirus was harvested and filtered through a 0.45 μm filter 72 hours after transfection. The filtered medium was then used to infect cells for 48 hours by culturing cells using virus-containing medium and fresh medium at a 1:1 ratio. After infection, the cells were selected in medium contains 1 μg/mL puromycin for 1 week. The following target short hairpin RNA–expressing plasmids (available from Sigma-Aldrich) were obtained from the Molecular Screening Facility of Wistar Institute were used: pLKO.1-shTrim28 (mouse, TRCN0000071363 and TRCN0000071364), pLKO.1-shTRIM28 (human, TRCN0000017998 and TRCN0000018001), pLKO.1-shMavs (mouse, TRCN0000124771), pLKO.1-shSETDB1 (human, TRCN0000148112 and TRCN0000276169).

Cells were trypsinized and washed two times with PBS. Protein was extracted by incubation with RIPA lysis buffer (50 mmol/L Tris (pH 8.0), 150 mmol/L NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, and 1 mmol/L phenylmethylsulfonyl fluoride) on ice for 30 minutes. Protein concentration was measured by the BCA assay (Pierce, #23225). Samples were separated by SDS-PAGE and transferred to polyvinylidene fluoride membrane (Millipore). Membranes were blocked with 4% BSA/TBS-T and then incubated with primary antibodies and then secondary antibodies (Cell Signaling Technology, #7074S). The following primary antibodies were used: mouse anti–β-actin (Sigma, #A5316), rabbit anti-TRIM28 (Abcam, #ab10484, RRID:AB_297223), mouse anti-SETDB1 (Santa Cruz Biotechnology, #sc-271553, RRID:AB_10649961), rabbit anti-tubulin (Cell Signaling Technology, #2125S, RRID:AB_2619646), rabbit anti–γ-H2AX (Cell Signaling Technology, #9718S, AB_2118009), rabbit anti–PD-L1 (Cell Signaling Technology, #13684S, RRID:AB_2687655), rabbit anti-cGAS (Cell Signaling Technology, #31659S, RRID:AB_2799008), anti-STING (Cell Signaling Technology, #13647, RRID:AB_2732796), and anti–phospho-IRF3 (Cell Signaling Technology, #4947S, RRID:AB_823547).

Total RNA from three replicates was extracted using RNeasy Kit (Qiagen, #74104) following the manufacturer's instructions. A total of 1 μg of purified RNA was used for reverse-transcriptase PCR (RT-PCR) with High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific, #4374967). Quantitative PCR (qPCR) was performed using iTaq Universal SYBR Green Supermix (Bio-Rad, #1725121) and run on QuantStudio 3 Real-Time PCR System. Gene expression fold change was calculated using the 2^−ΔΔC_T. The sequences of the primers used for qRT-PCR are as follows: mCd274 forward: 5′-GCATTATATTCACAGCCTGC-3′ and reverse: 5′-CCCTTCAAAAGCTGGTCCTT-3′; mCxcl10 forward: 5′-TCAGCACCATGAACCCAAG-3′ and reverse: 5′-CTATGGCCCTCATTCTCACTG-3′; mCcl5 forward: 5′-CCACTTCTTCTCTGGGTTGG-3′ and reverse: 5′-GTGCCCACGTCAAGGAGTAT-3′; mMavs forward: 5′-GCTCCTTGGTCTCAGAACCC-3′ and reverse: 5′-CTGGGGCTTTCGTCTACCTG-3′; mTrim28 forward: 5′-CGCTCACAAGGACCATCAGT-3′ and reverse: 5′-AGCTTCGAACCTCCTTGGTG-3′; hTRIM28 forward: 5′-GGAAGGCTATGGCTTTGGGT-3′ and reverse: 5′-GGGAAGACCTTGAAGACGGG-3′; hCXCL10 forward: 5′-TGCCATTCTGATTTGCTGCC-3′ and reverse: 5′-TGCAGGTACAGCGTACAGTT-3′; hCCL5 forward: 5′-CGTGCCCACATCAAGGAGTA-3′ and reverse: 5′-TCGGGTGACAAAGACGACTG-3′; hGAPDH forward: 5′-GTCTCCTCTGACTTCAACAGCG-3′ and reverse: 5′-ACCACCCTGTTGCTGTAGCCAA-3′; mActb forward: 5′-CTCCTATGTGGGTGACGAGG-3′ and reverse: 5′-ACGGTTGGCCTTAGGGTTC-3′. Human GADPH and mouse Actb were used as an internal control to normalize the expression in human and mouse cells, respectively.

Total RNA of control ID8 and two Setdb1-knockout ID8 cell lines was extracted using RNeasy mini Kit (Qiagen, #74106) according to the manufacturer's instructions and digested with DNase I (Qiagen, #79254). RNA sequencing (RNA-seq) libraries were constructed using ScriptSeq complete Gold kit (Epicentre, #SCL24EP) and subjected to a 75 bp paired-end sequencing run on NextSeq 500, using Illumina's NextSeq 500 high output sequencing kit (#20024906) following the manufacturer's instructions.

For cell proliferation analysis, 20,000 control or Setdb1-knockout ID8 cells were seeded into a 6 cm dish. Cell number was counted every 2 days by hemocytometer. Three repeats were performed, and P values were calculated using a two-way ANOVA.

For cell-cycle analysis, control and Setdb1-knockout ID8 cells were suspended in 500 μL PBS and fixed by gently adding 5 mL cold 70% ethanol dropwise. Cells were then washed twice with PBS and stained by adding 1 mL propidium iodide (PI) solution [3.8 mmol/L sodium citrate, 50 μg/mL PI (Invitrogen, #P3566) in PBS] and 50 μL RNase A stock solution (Thermo Fisher Scientific, #FEREN0531) at room temperature for 30 minutes. PI staining was detected using a Becton-Dickinson LSR18 machine, and analyzed with FlowJo version 7 software (Tree Star, Inc.)

For immune infiltration analysis, surgically removed tumors were minced into small (1 to 2 mm) pieces and digested with 1 mg/mL collagenase IV (Sigma-Aldrich, #C5138), 0.1 mg/mL Hyaluronidase (Sigma-Aldrich, #H6254) and 0.01 mg/mL deoxyribonuclease I (Sigma-Aldrich, #D5025). The cells were sequentially filtered through 40 μm cell strainers. Analysis of tumor-infiltrating lymphocytes (TIL) involved viability staining (Thermo Fisher Scientific, #L34957), Fc blocking (BD, #553142, RRID:AB_394657) and then surface staining in FACS buffer (3% FBS in PBS) with fluorochrome-conjugated antibodies specific for mouse CD45 (BioLegend, #103147, RRID:AB_2564383), mouse CD3 (BD, #552774, RRID:AB_394460), mouse CD8 (BioLegend, #100708, RRID:AB_312747), and mouse PD-L1 (BioLegend, #124321, RRID:AB_2563635). For intracellular staining, the TILs were cultured in RMPI1640 plus 10% FBS with cell activation cocktail (BioLegend, #423303) overnight, and then were fixed and permeabilized with Cyto-Fast Fix/Perm Buffer Set (BioLegend, #426803) for mouse Granzyme B staining (BioLegend, #515403, RRID:AB_2114575). For PD-L1 expression in cancer cells obtained from the syngeneic model, CD45⁻ cells were surface stained in FACS buffer with fluorochrome-conjugated antibody specific for mouse PD-L1 (BioLegend, #124312, RRID:AB_10612741). All FACS analyses were performed on a BD LSR II or a Canto II Flow Cytometer, and data were analyzed with FlowJo software (Tree Star, Inc., version 7).

Cells on cover slips were fixed with 3% paraformaldehyde in PBS, permeabilized in PBS containing 0.5% Triton X-100, blocked with 3% BSA in PBS and incubated with rabbit anti-cGAS (Cell Signaling Technology, #31659S, RRID:AB_2799008), and anti–γ-H2AX (Cell Signaling Technology, #9718S, AB_2118009) at 4°C overnight. Cells were then washed twice with PBS and incubated with secondary antibodies conjugated with AlexaFluor dyes (Molecular Probes, #A11008, RRID:AB_143165; Molecular Probes, #A11004, RRID:AB_2534072) at room temperature for 1 hour. Slips were then washed with PBS for three times, and incubated with diluted DAPI DNA staining dye in PBS for 15 minutes at room temperature. Stained slides were analyzed using a Leica TCS SP5 II scanning confocal microscope.

HGSOC tumor tissue microarrays (TMA) were kindly provided by Benjamin G. Bitler from The University of Colorado (Aurora, CO; COMIRB #17-7788). Detailed information on the tumors contained on the TMA has been published previously (32, 33). IHC was performed using the Dako EnVision+ system following the manufacturer's instructions. Briefly, the sections were dewaxed, rehydrated, and immersed in 3% hydrogen peroxide in methanol to quench endogenous peroxidase activity. Antigen retrieval was performed in sodium citrate buffer (Thermo Fisher Scientific, #005000) and boiled for 45 minutes. The sections were incubated with blocking buffer for 1 hour, primary antibody against TRIM28 (Abcam, #ab10484, RRID:AB_297223), SETDB1 (Biological Novus, #NBP2-20322) or PD-L1 (Cell Signaling Technology, #13684S, RRID:AB_2687655) at 4°C overnight and secondary antibody for 1 hour (Dako, #K4011). Counterstaining was performed using Mayer's Hematoxylin (Dako, #3309S). Expression of the stained markers was scored using a histologic score (H score). On the basis of the H scores, Kaplan–Meier survival curves were generated for SETDB1 and TRIM28 high and low patient groups using GraphPad Prism 7 (GraphPad) software. P value was calculated by log-rank test.

Animal protocols were approved by the Institutional Animal Care and Use Committee of The Wistar Institute. C57BL/6 mice were purchased from Charles River Laboratories. A total of 1 × 10⁶ ID8 cells were unilaterally injected into the ovarian bursa sac of 6- to 8-week-old female C57BL/6 mice (n = 5 mice per group). Four weeks after injection, mice were treated with IgG or anti–Pd-l1 (BioXcell, B7-H1, RRID:AB_10949073; 10 mg per kg, twice per week) for 2 weeks. After treatment, tumors from 5 mice per group were surgically dissected and tumor weight was measured as a surrogate for tumor burden.

For expression correlation analysis, The Cancer Genome Atlas (TCGA) HGSOC RNA-seq data (PanCancer Atlas, n = 300; ref. 34) was downloaded from cBioPortal (https://www.cbioportal.org/) and Cancer Cell Line Encyclopedia (CCLE, n = 912). RNA-seq data were downloaded from (https://sites.broadinstitute.org/ccle/datasets). Pearson correlation was used for calculating P and r values in Microsoft Excel. Pearson correlations of SETDB1 or TRIM28 with CD8A or GZMB were analyzed in 24 of 33 types of TCGA PanCancer Atlas datasets with more than 100 patients. The analysis was performed on the cBioPortal website.

A melanoma proteomics dataset with clinical response to anti–PD-1 ICB was downloaded from the Supplementary Information section of the original article (35), the protein levels of TRIM28 in responders (CR: complete responder, n = 10 and PR: partial responder, n = 30) and nonresponders (progressive disease, n = 27) was compared.

HGSOC tumor TMAs were kindly provided by Benjamin G. Bitler from The University of Colorado (Aurora, CO; COMIRB #17-7788). The TMAs contain 131 patient tumor samples. In addition, data were available for the quantification of infiltrated immune cell populations by flow cytometry such as CD8⁺ T cell, CD8⁺/Granzyme B⁺ T cells, and regulatory T cells for 120 of the 131 patients. The percentages of indicated infiltrated immune cell populations in SETDB1 or TRIM28 high (H score ≥ 200) and low groups (H score ≤ 200) were compared.

For analysis of the CRISPR–Cas9 screen, the raw .fastq file was parsed to obtain sgRNA abundance values using regular expression (stagger+BARCODE)(primer)(sgRNA), where stagger+BARCODE sequences were (atAAGTAGAG|gatAAGTAGAG) corresponding to PD-L1 and control samples correspondingly, primer sequence TCTTGTGGAAAGGACGAAACACCG and sgRNA sequence of 20 bp length from the Epigenetic sgRNA library (31). Only perfect matches were considered for calculating sgRNA counts. Significance of difference between PD-L1 versus control samples on gene level was estimated using Model-based Analysis of Genome-wide CRISPR–Cas9 Knockout (MAGeCK) software (36) using “mageck test -k counts.txt -t PD-L1 -c ctr -n GeneRankOutput” command. Genes that passed FDR < 5% cutoff were considered significant.

For RNA-seq, data were aligned using bowtie2 (37) against mm10 version of the mouse genome and RSEM v1.2.12 software (38) was used to estimate raw read counts and reads per kilobase per million mapped reads (RPKM) values using Ensemble transcriptome. Abundance of TEs in samples was estimated using TEtranscripts software with default parameters (39). DESeq2 (40) was used to estimate significance of differential expression between groups pairs and calculate normalized counts. Overall gene expression changes were considered significant if passed FDR < 5% thresholds unless stated otherwise. TE expression changes were considered significant if P < 0.05 and FDR < 5%. The significantly changed TEs were shown as a heatmap and volcano plot using ggplot2 package (v3.3.3; Wickham, 2016). Enrichment analysis was performed using gene set enrichment analysis (GSEA; ref. 41) using MSigDB Hallmark gene sets (42) and results passing FDR < 5% were reported.

For statistical analysis, experiments were repeated at least three times unless otherwise stated. Statistical analysis was performed using the GraphPad Prism 7 (GraphPad) software. Quantitative data are expressed as mean ± SEM unless otherwise stated. For all statistical analysis, the cutoff for significance was set at 0.05. For correlation studies, Pearson correlation was used for calculating P and r values in Microsoft Excel. Animal experiments were randomized. Combination index (CI; ref. 43) for PD-L1 treatment and Setdb1 knockout was calculated as follows: CI = [(A+B) − A*B]/AB. A or B is the effect of the single condition, AB is the effect of the combination of A and B. Effect was calculated by the decreased percentage of the tumor weight: (1- treated tumor/control tumor) * 100%.

Next-generation sequencing data for the CRISPR screen and RNA-seq have been deposited in the National Center for Biotechnology Information's Gene Expression Omnibus (GEO) and are accessible through GEO series accession number GSE182198 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE182198).

To systematically explore chromatin regulators of PD-L1 expression, we performed a CRISPR–Cas9 screen in the mouse ovarian cancer cell line ID8 using an sgRNA library targeting 1,218 genes that encode chromatin regulators, with typically 10 sgRNAs per gene (Fig. 1A). IFNγ was used to stimulate PD-L1 expression. To limit experimental variations, we sorted the 15% of ID8 cells with highest PD-L1 expression and compared the sgRNA profile in these cells with that in the bulk, unsorted cell population to ensure identical background and conditions for comparison. MAGeCK analysis revealed a list of sgRNAs that were either downregulated or upregulated in the PD-L1^high cell population (Fig. 1B; Supplementary Table S1). Genes that were known regulators of PD-L1 were successfully identified as top ranked hits in our screen. For example, both Jak2 and Trp53 were identified as hits whose targeting sgRNAs were downregulated in the PD-L1^high cell population (Fig. 1B). This is consistent with previous reports that both Jak2 and Trp53 are required for IFNγ-induced PD-L1 expression (44, 45).

We focused on our analysis on genes whose targeting sgRNAs were enriched in PD-L1 expression because we hypothesized that their inhibition could be used to enhance the therapeutic response to anti–PD-1/PD-L1 (46, 47). The list of top hits whose sgRNAs were enriched in the PD-L1^high cell population included Leo1, Arid1a, Eed, Trim28, Ints12, Kdm1a, and Sin3b (Fig. 1C). To prioritize our functional studies, we examined the correlation between these genes and CD274, which encodes PD-L1, in two publicly available databases, the Broad Institute CCLE and TCGA ovarian serous cystadenocarcinoma. The analysis revealed that LEO1, ARID1A, TRIM28, and KDM1A negatively correlated with CD274 expression in the CCLE database (Supplementary Fig. S1A), and that ARID1A, TRIM28 and KDM1A negatively correlated with CD274 expression in TCGA ovarian cancer database (Supplementary Fig. S1B). By overlapping these analyses, we identified three hits from the screen that negatively correlated with CD274 expression in both datasets: ARID1A, TRIM28, and KDM1A (Supplementary Fig. S1C). ARID1A inactivation is associated with high PD-L1 expression and better response to ICB, including as anti–PD-L1 treatment (48). Likewise, KDM1A ablation enables ICB by stimulating antitumor immunity (49). However, the role of TRIM28 in regulating PD-L1 expression and antitumor immunity has not been explored. Accordingly, we focused our validation and functional studies on the TRIM28 gene.

All 10 sgRNAs targeting Trim28 showed a consistent enrichment in PD-L1^high cells (Fig. 1D). TRIM28 is a component of the HUSH complex with SETDB1 as its catalytic subunit (14, 50). We validated that Trim28 knockdown in ID8 cells increased Cd274 expression both at the basal level and in response to IFNγ stimulation (Supplementary Fig. S1D and S1E). We also observed an increase in PD-L1 expression by flow cytometry analysis (Fig. 1E and F). Similar observations were made in the human OVCAR3 ovarian cancer cell line (Supplementary Fig. S1F). Indeed, the negative correlation between TRIM28 and CD274 was observed in a statistically significant majority of the cancer types in the 24 TCGA datasets with more than 100 patients (Fig. 1G). Because SETDB1 is the catalytic subunit of the HUSH complex, we examined whether SETDB1 loss phenocopies TRIM28 loss. Toward this goal, we knocked out Setdb1 in ID8 cells using CRISPR (Supplementary Fig. S1G). Indeed, Setdb1 knockout increased PD-L1 expression both at the basal level and in response to IFNγ stimulation (Fig. 1H and I). Similar observations were also made in human OVCAR3 cells (Supplementary Fig. S1H and S1I). In addition, SETDB1 expression negatively correlated with CD274 expression in the ovarian cancer TCGA dataset (Fig. 1J; Supplementary Fig. S1J). Finally, a negative correlation between SETDB1 and CD274 was observed in a statistically significant majority of the cancer types in the 24 TCGA datasets with more than 100 patients (Fig. 1K).

We next sought to determine the correlation between SETDB1–TRIM28 and PD-L1 expression in a tumor microarray consisting of 131 cases of human HGSOC. Both SETDB1 and TRIM28 expression negatively correlated with PD-L1 expression (Fig. 2A–C; Supplementary Table S2). In addition, SETDB1 and TRIM28 expression positively correlated (Fig. 2D; Supplementary Table S2). Compared with normal fallopian tube epithelial cells, SETDB1 and TRIM28 were found to be overexpressed in several ovarian cancer cell lines and there was concordance between SETDB1 and TRIM28 expression in these cell lines (Supplementary Fig. S2A). This is consistent with previous reports that SETDB1 is often overexpressed in human cancers (51). Indeed, SETDB1 was amplified/overexpressed in approximately 26% of HGSOC in TCGA dataset and its amplification correlated with a significantly higher mRNA expression (Supplementary Fig. S2B and S2C).

Immune cell infiltration profiles are available on 120 of the 131 TMA cases we examined (Supplementary Table S2). Thus, we correlated with SETDB1 and TRIM28 expression with immune infiltration profile. SETDB1 expression negatively correlated with tumor infiltrated CD8⁺ T cells and activated CD8⁺ T cells, as evidenced by Granzyme B⁺/CD8⁺ T cells (Fig. 2E and F; Supplementary Table S2). Similar trends were also observed between TRIM28 expression and infiltration of CD8⁺ and Granzyme B⁺/CD8⁺ T cells (Supplementary Fig. S2D and S2E; Supplementary Table S2). In contrast, there was no correlation between expression of SETDB1 or TRIM28 with immune suppressive populations such as regulatory T cells (Supplementary Table S2). Next, we sought to validate our findings in the ovarian cancer TCGA dataset. We observed a negative correlation between SETDB1 expression and CD8A expression and GZMB expression (Fig. 2G and H). Similar findings were also made between TRIM28 expression and CD8A expression and GZMB expression in the ovarian cancer TCGA dataset (Supplementary Fig. S2F and S2G). Finally, the negative correlation between SETDB1 or TRIM28 expression and CD8A expression and GZMB expression was observed in a statistically significant majority of the cancer types in the 24 TCGA datasets with more than 100 patients (Fig. 2I and J; Supplementary Fig. S2H and S2I).

We next sought to determine whether SETDB1 and/or TRIM28 expression predicted prognosis, as determined by overall survival in the 131 TMA cases examined. Consistent with previous reports that high CD8⁺ T-cell infiltration correlates with an increase in overall survival (52, 53), we observed a negative correlation between SETDB1 expression and overall survival of patients with ovarian cancer (Fig. 2K), which was validated in an independent, public ovarian cancer microarray dataset from 1,287 patients (Supplementary Fig. S2J; ref. 54). A similar trend between TRIM28 expression and overall survival was observed, albeit the statistical analysis did not reach significance (Supplementary Fig. S2K). Reanalysis of a publicly available melanoma proteomics dataset with clinical response to anti–PD-1 ICB revealed that TRIM28 was expressed at significantly higher levels in patients who did not respond to anti–PD-1 treatment compared with those who had a clinical response (Fig. 2L; Supplementary Table S3; ref. 35). SETDB1 was not detected in the proteomics dataset. These findings are consistent with the notion that SETDB1–TRIM28 represses PD-L1 expression and inhibits the infiltration of effector CD8⁺ T cells.

We next sought to determine the mechanism by which SETDB1–TRIM28 regulates PD-L1 expression and immune cell infiltration. Toward this goal, we performed RNA-seq analysis in two independent Setdb1-knockout ID8 clones (Supplementary Fig. S1G). Analysis of differentially expressed genes (DEG) revealed that 6,005 genes were significantly changed by Setdb1 knockout compared with controls (Supplementary Fig. S3A). GSEA of the DEGs revealed that G₂–M checkpoint and inflammatory response pathways were among the top pathways enriched by the analysis (Fig. 3A and B). Micronuclei-induced ISGs such as Ccl5 and Cxcl10 were among the inflammatory genes upregulated by Setdb1 knockout (Fig. 3C; ref. 10). We validated the upregulation of specific micronuclei-induced ISGs such as Ccl5 and Cxcl10 in both Setdb1- and Trim28-knockout ID8 cells (Fig. 3D; Supplementary Fig. S3B and S3C). Similar results were also obtained in Setdb1-knockout UPK10 mouse ovarian cancer cells (Supplementary Fig. S3D and S3E). Likewise, TRIM28 knockdown in OVCAR3 human ovarian cancer cells upregulated the expression of CCL5 and CXCL10 genes (Supplementary Fig. S3F). In addition, we validated that Setdb1 knockout increased G₂–M-phase of the cell cycle (Supplementary Fig. S3G and S3H). Given the fact that micronuclei are generated as a result of mitotic defects during the G₂–M-phase of the cell cycle (11) and Setdb1 knockout increased the expression of micronuclei-induced ISGs such as Ccl5 and Cxcl10, we examined whether Setdb1 loss induced micronuclei formation. Toward this goal, we stained control and Setdb1-knockout ID8 cells using markers of micronuclei such as cGAS and γH2AX (Fig. 3E). Indeed, Setdb1 knockout significantly increased micronuclei formation (Fig. 3F; refs. 10, 11). Similar observations were made in Setdb1knockout UPK10 mouse ovarian cancer cells (Supplementary Fig. S3I). Consistent with the notion that micronuclei resulted from damaged DNA (10, 11), the marker of DNA damage γH2AX was increased in Setdb1-knockout ID8 cells (Supplementary Fig. S3J). Furthermore, Trim28 knockout induced a similar increase in micronuclei formation in ID8 cells (Fig. 3G and H), suggesting that the observed effects are dependent on the SETDB1–TRIM28 complex.

Because upregulation of micronuclei-induced ISGs depends on cGAS signaling (10, 11), we sought to determine whether inhibition of cGAS was sufficient to block the observed increase in micronuclei-induced ISGs such as Ccl5 and Cxcl10. Toward this goal, we inhibited cGAS activity either by a small molecule inhibitor RU.521 (55) or genetically by knocking out cGas in Setdb1-knockout ID8 cells (Supplementary Fig. S3K). Indeed, both RU.521 and cGas knockout significantly suppressed the upregulation of Ccl5 and Cxcl10 observed in Setdb1-knockout cells (Fig. 3I and J). Similar findings were also made in UPK10 Setdb1-knockout cells (Supplementary Fig. S3L). Consistently, Sting dimerization and the increase in phospho-Irf3 induced by Setdb1-knockout cells were both blocked by cGas knockout (Fig. 3K). Furthermore, the upregulation of PD-L1 induced by Setdb1 loss was completely eradicated by cGas knockout in Setdb1-knockout cells both in vitro and in vivo in an orthotopic mouse model (Fig. 3L; Supplementary Fig. S3M and S3N). This suggests that upregulation of PD-L1 induced by Setdb1 loss is mediated by cGas signaling. Consistently, it has previously been shown that upregulating the cGas–Sting pathway in PD-L1–low mouse cells is sufficient to upregulate PD-L1 expression (56). These findings suggest a model whereby SETDB1 or TRIM28 loss activates the cGAS–STING pathway to simultaneously upregulate PD-L1 expression and increase immune cell infiltration through micronuclei-induced ISGs.

The best characterized function of the SETDB1–TRIM28 complex is in silencing TEs (57). Indeed, a very recent study shows that SETDB1 loss derepresses latent TE-derived regulatory elements to trigger TE-specific cytotoxic T–cell responses (29). Consistent with these data, we observed an extensive upregulation of TEs in Setdb1-knockout ID8 cells (Supplementary Fig. S4A and S4B). To determine whether TEs contribute to the observed upregulation of inflammatory genes such as Ccl5 and Cxcl10, we knocked down Mavs expression in Setdb1-knockout cells (Supplementary Fig. S4C). Mavs functions downstream of the essential adaptor proteins Rig-I and Mad-5, which mediate the immune response, such as induction of ISGs, triggered by double-stranded RNAs in response to derepressed TEs in Setdb1-knockout cells (57, 58). In contrast to cGas knockout, Mavs knockdown did not affect the expression of Ccl5 or Cxcl10 (Supplementary Fig. S4D). Likewise, Mavs knockdown did not affect either Sting dimerization or phospho-Irf3 expression in Setdb1-knockout cells (Fig. 3K). Together, we conclude that Setdb1-induced micronuclei formation and the associated cGas–Sting signaling are independent of the depression of TEs that is induced by Setdb1 knockout.

Setdb1 loss simultaneously upregulates PD-L1 expression and increases infiltration of CD8⁺ T cells whose presence in the tumor microenvironment is a prerequisite for response to ICBs such as anti–PD-1/PD-L1. Thus, we hypothesized that Setdb1 loss could enhance the antitumor effects of anti–PD-L1 treatment. To test this hypothesis, we utilized a mouse orthotopic syngeneic ID8 ovarian cancer model (Fig. 4A). Mice receiving Setdb1-knockout ID8 cells had significantly reduced tumor burden, as determined by a decrease in tumor weight, compared with mice receiving control ID8 cells (Fig. 4B and C). Setdb1 knockout did not significantly affect the growth of cancer cells in vitro (Supplementary Fig. S5). Consistent with these observations, Setdb1 knockout was shown to reduce tumor burden in immunocompetent syngeneic, but not in immunocompromised xenograft, mouse models of both melanoma and lung cancer (29). Together, these findings suggest the reduction in tumor burden observed as a result of Setdb1 knockout in the immunocompetent syngeneic mouse model was due to Setdb1-induced antitumor immunity, rather than intrinsic growth inhibition of cancer cells. Anti–PD-L1 and Setdb1 loss were synergistic in reducing the burden of established orthotopic tumors (Fig. 4C). To determine whether the observed synergy depended on cGas signaling, we performed the same experiments in parallel using Setdb1 and cGas double-knockout cells. Indeed, cGas knockout completely eradicated the antitumor effects of anti–PD-L1 and the synergy between anti–PD-L1 and Setdb1 loss (Fig. 4B and C).

We next profiled changes in immune cell infiltration to correlate the observed antitumor effects with the mechanisms we have revealed. Indeed, both Setdb1 loss and anti–PD-L1 treatment increased the infiltration of CD8⁺ and Granzyme B⁺ CD8⁺ T cells (Fig. 4D and E). Moreover, a combination of anti–PD-L1 and Setdb1 loss was synergistic in increasing the infiltration of CD8⁺ T cells (Fig. 4D and E). Although cGas loss completely eradiated the antitumor effects of anti–PD-L1, Setdb1 loss and their combination, its loss only partially blocked the infiltration of CD8⁺ and Granzyme B⁺ CD8⁺ T cells (Fig. 4D and E). Together, we conclude that Setdb1 loss and anti–PD-L1 are synergistic in reducing tumor burden, and this correlates with changes in infiltration of CD8⁺ and CD8⁺/Granzyme B⁺ T cells. In addition, cGas is required for the observed synergy, but only partially contributes to the observed changes in immune infiltration.

Here we identify Trim28 as a regulator of PD-L1 expression in an unbiased CRISPR–Cas9 epigenome library screen. Although Setdb1 sgRNAs were included in the library, Setdb1 was not among the top hits (Supplementary Table S1). A possible reason for this might be variation in knockout efficacy by these sgRNAs and/or low infection efficacy by some of the sgRNAs. Consistent with this hypothesis, although 6 of 10 individual Setdb1 sgRNAs were scored as positive hits, Setdb1 as a gene did not reach statistical significance in our screen. Regardless, our validation studies clearly show that Setdb1 loss phenocopies Trim28 loss in regulating PD-L1 expression.

We show that the synergy in reducing tumor burden observed by combining anti–PD-L1 and Setdb1 knockout is cGas dependent because cGas knockout blocked the observed antitumor effects. However, cGas knockout was not sufficient to block the infiltration of CD8⁺ T cells induced by either anti–PD-L1 alone or a combination of Setdb1 loss and anti–PD-L1. This is consistent with the notion that immune infiltration induced by a cell-intrinsic mechanism caused by Setdb1 loss is cGas dependent, whereas infiltration of CD8⁺ T cells induced by anti–PD-L1 occurs through a non–cell-intrinsic mechanism and is independent of cGas. In addition, while we were preparing this article, it was reported that Setdb1 loss triggers TE-specific cytotoxic T–cell responses (29). In addition, it was shown that disruption of Setdb1 or its interacting partner Atf7ip augments tumor immunogenicity concomitant with elevated endogenous retroviral antigens and message RNA intron retention (59). Thus, the role of Setdb1 in regulating CD8⁺ T cells is likely multifaceted and context dependent.

Our results show that SETDB1–TRIM28 loss simultaneously increases PD-L1 expression and activates cGas–Sting signaling via micronuclei formation. In addition, our results further support a model whereby cGas activation mediates PD-L1 upregulation induced by SETDB1–TRIM28 loss. These results are important because SETDB1–TRIM28 loss bridges innate immunity and adaptive immunity to synergize with ICB in reducing tumor burden. The synergy is achieved by at least two mechanisms. First, Setdb1 loss increases PD-L1 expression, the target of anti–PD-L1 whose upregulation by both genetic alterations and therapeutic treatment has been linked to improved response to ICB. Consistent with this, we show that in a cohort of anti–PD-1–treated patients with melanoma, TRIM28 is expressed at significantly lower levels in responders compared with nonresponders. Second, Setdb1 loss increases infiltration of effector CD8⁺ T cells in the tumor microenvironment, a prerequisite for response to ICB. Our findings raise the possibility that SETDB1–TRIM28 complex may serve as a biomarker to predict the therapeutic response to ICB. In addition, they suggest a need to develop therapeutic approaches to target the complex to boost antitumor immunity. A limitation of our study is the fact that the ID8 cells that we used lack a TP53 genomic alteration. However, we show comparable results using TP53 inactivated UPK10 cells and TP53-mutant human ovarian cancer cells. In summary, our results establish the SETDB1–TRIM28 complex as an attractive therapeutic target to boost antitumor immunity alone or in combination with ICB such as anti–PD-L1 treatment.

R. Drapkin reports personal fees from Repare Therapeutics, Boehringer Laboratories, and TwoXAR, Inc. outside the submitted work. B.G. Bitler reports grants from NCI/NIH, ACS, and DOD during the conduct of the study. K. Helin reports grants from Novo Nordisk Foundation during the conduct of the study, as well as personal fees from MetaboMed Inc. and Inthera Bioscience AG, and other support from Hannibal Innovation outside the submitted work. R. Zhang reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

J. Lin: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. D. Guo: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. H. Liu: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. W. Zhou: Data curation, investigation. C. Wang: Data curation, investigation, writing–original draft. I. Müller: Resources. A.V. Kossenkov: Formal analysis, methodology, writing–original draft. R. Drapkin: Resources. B.G. Bitler: Resources, writing–original draft. K. Helin: Resources. R. Zhang: Conceptualization, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

This work was supported by NIH grants (R01CA202919, R01CA239128, and P50CA228991, to R. Zhang), the U.S. Department of Defense (OC180109 and OC190181, to R. Zhang), the Honorable Tina Brozman Foundation for Ovarian Cancer Research and the Tina Brozman Ovarian Cancer Research Consortium 2.0 (to R. Zhang), and the Ovarian Cancer Research Alliance (Collaborative Research Development Grant #596552, to R. Zhang; Ann and Sol Schreiber Mentored Investigator Award #649658, to J. Lin). Support of Core Facilities was provided by Cancer Centre Support Grant (CCSG) CA010815 to The Wistar Institute.

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.