Abstract Background Immune checkpoint inhibitors (ICIs) shed new light on triple‐negative breast cancer (TNBC), but only a minority of patients demonstrate response. Therefore, adaptive immune resistance (AIR) needs to be further defined to guide the development of ICI regimens. Methods Databases, including The Cancer Genome Atlas, Gene Ontology Resource, University of California Santa Cruz Genome Browser, and Pubmed, were used to screen epigenetic modulators, regulators for CD8^+ T cells, and transcriptional regulators of programmed cell death‐ligand 1 (PD‐L1). Human peripheral blood mononuclear cell (Hu‐PBMC) reconstruction mice were adopted for xenograft transplantation. Tumor specimens from a TNBC cohort and the clinical trial CTR20191353 were retrospectively analyzed. RNA‐sequencing, Western blotting, qPCR and immunohistochemistry were used to assess gene expression. Coculture assays were performed to evaluate the regulation of TNBC cells on T cells. Chromatin immunoprecipitation and transposase‐accessible chromatin sequencing were used to determine chromatin‐binding and accessibility. Results The epigenetic modulator AT‐rich interaction domain 1A (ARID1A) gene demonstrated the highest expression association with AIR relative to other epigenetic modulators in TNBC patients. Low ARID1A expression in TNBC, causing an immunosuppressive microenvironment, promoted AIR and inhibited CD8^+ T cell infiltration and activity through upregulating PD‐L1. However, ARID1A did not directly regulate PD‐L1 expression. We found that ARID1A directly bound the promoter of nucleophosmin 1 (NPM1) and that low ARID1A expression increased NPM1 chromatin accessibility as well as gene expression, further activating PD‐L1 transcription. In Hu‐PBMC mice, atezolizumab demonstrated the potential to reverse ARID1A deficiency‐induced AIR in TNBC by reducing tumor malignancy and activating anti‐tumor immunity. In CTR20191353, ARID1A‐low patients derived more benefit from pucotenlimab compared to ARID1A‐high patients. Conclusions In AIR epigenetics, low ARID1A expression in TNBC contributed to AIR via the ARID1A/NPM1/PD‐L1 axis, leading to poor outcome but sensitivity to ICI treatment. Keywords: adaptive immune resistance, ARID1A, CD8^+ T cell, PD‐L1, triple‐negative breast cancer __________________________________________________________________ Abbreviations AIR adaptive immune resistance ARID1A AT‐rich interaction domain 1A ATAC Assay for Transposase Accessible Chromatin ATCC American Type Culture Collection Ate Atezolizumab BRG1/ SMARCA4 SWI/SNF related, matrix associated, actin dependent regulator of chromatin, subfamily a, member 4 ICI immune checkpoint inhibitor CCK‐8 cell counting kit‐8 ChIP chromatin immunoprecipitation CN copy number CPS combined positive score DFS disease‐free survival DMEM Dulbecco's Modified Eagle Medium EGFR epidermal growth factor receptor ELISA enzyme‐linked immunosorbent assay ER estrogen receptor FPKM fragments per kilobase million FUSCC Fudan University Shanghai Cancer Center GB granzyme B GEO Gene Expression Omnibus GSEA gene set enrichment analysis HER2 human epidermal growth factor receptor 2 HR hazard ratio HRP horseradish peroxidase Hu‐PBMC human peripheral blood mononuclear cell IFN interferon IHC immunohistochemistry staining IL interleukin ko knockout MACS Model‐based Analysis of ChIP‐Seq min minute nc negative control NPM1 nucleophosmin 1 oe overexpression OS overall survival PBS phosphate buffer saline pCR pathological complete response PD‐1 programmed cell death‐1 PD‐L1 programmed cell death 1 ligand‐1 PIK3CA phosphatidylinositol‐4,5‐bisphosphate 3‐kinase catalytic subunit alpha Pla placebo PMA phorbol 12‐myristate 13‐acetate PTEN phosphatase and tensin homolog PFS progression‐free survival RPMI Roswell Park Memorial Institute s second sgRNA small guide RNA shRNA short hairpin RNA SWI/SNF switch/sucrose non‐fermentable TBST tris buffered saline with Tween‐20 TCGA The Cancer Genome Atlas TIL tumor‐infiltrating lymphocyte TIME tumor immune microenvironment TMA tissue microarray TNBC triple‐negative breast cancer TPS tumor proportion score. UCSC University of California Santa Cruz 1. BACKGROUND Adaptive immune resistance (AIR) is crucial for cancer cells to evade immune attack, leading to tumor growth and metastasis. Triple‐negative breast cancer (TNBC) comprises 15%‐20% of newly diagnosed breast cancer cases and is still the molecular type with the poorest prognosis [[50]1]. Recent studies highlighted the role of AIR in TNBC early metastasis and led to the successful development of immune checkpoint inhibitors (ICIs) targeting programmed cell death 1 (PD‐1) and its ligand 1 (PD‐L1), alleviating immune suppression and boosting the anti‐tumor effect of T cells [[51]2, [52]3]. Pembrolizumab, an ICI targeting PD‐1, has been indicated for early and first‐line metastatic TNBC with promising clinical data [[53]4, [54]5]. Despite the progress, the agent failed to bring survival benefits to most metastatic TNBC patients and heavily treated patients, suggesting the complexity of AIR in TNBC [[55]5, [56]6, [57]7]. Therefore, we need to integrate a holistic view of the TNBC ecosystem, focusing on AIR in the era of ICIs, and identify biomarkers to improve ICI treatment efficacy. AIR is under the regulation of both genetic and epigenetic factors. From the genetic perspective, single‐gene alterations have been found to alter the immune landscape. Tumor protein p53 mutations, for instance, could suppress innate immune signaling and promote immune evasion [[58]8], whereas deficiency of BRCA2 DNA repair associated was associated with both innate and adaptive immune gene signatures [[59]9]. In recent years, increasing studies have focused on the dynamic role of epigenetic factors such as the innate microenvironment, immune modulation [[60]10] and drug exposure [[61]11]. Recent studies uncovered a novel role for the mammalian chromatin remodeling complex switch/sucrose non‐fermentable (SWI/SNF), a macromolecular assembly repositioning nucleosomes and manipulating DNA accessibility, in regulating anti‐tumor immunity and implied therapeutic vulnerabilities underlying aberrations in SWI/SNF core members, such as the AT‐rich interaction domain 1A gene (ARID1A) [[62]12, [63]13, [64]14]. As one of the 4 core subunits of SWI/SNF, ARID1A has been recognized as a tumor suppressor, and its deficiency was related with poor clinical outcomes across multiple cancer types, including breast cancer [[65]15, [66]16, [67]17, [68]18]. In tumors, ARID1A alterations, reaching an approximate proportion of 10%, are mostly inactive mutations leading to loss of ARID1A protein and inability to direct SWI/SNF to target gene promoters, thereby endowing cells with cancerous functions and causing poor prognosis across multiple cancer types [[69]15, [70]16, [71]19, [72]20]. However, certain exceptions have also been reported. In liver cancer, ARID1A demonstrated context‐dependent tumor‐suppressive and oncogenic roles [[73]21]. In TCGA pan‐cancer studies queried in cBioPortal, an ARID1A‐mutant cohort demonstrated a more favorable prognosis than its ARID1A‐wildtype counterpart [[74]22]. Therefore, whether ARID1A functioned as a friend or foe in TNBC is still worth exploring. Our previous results in metastatic TNBC patients demonstrated that ARID1A was a commonly mutated gene with a rate of 7.1% [[75]23], much higher than that in early TNBC [[76]24]. This trend is consistent with findings in matched primary and metastatic estrogen receptor positive (ER+) breast cancer (3.3% vs. 12.9%, P < 0.001) [[77]25]. In heavily treated TNBC patients harboring ARID1A mutations, we also observed a better response to anti‐PD‐1 therapy compared with ARID1A wildtype counterparts (data unpublished). Therefore, we reasonably hypothesized that ARID1A deficiency could induce AIR and predict a better response to ICI treatment relative to ARID1A‐proficient TNBC. However, the exact association between ARID1A and AIR in TNBC, as well as the underlying mechanisms, remains to be explored. In the present study, we evaluated the role of ARID1A among a series of epigenetic modulators in TNBC and explored how low ARID1A expression contributed to AIR. In addition, we assessed the targetability of low ARID1A expression‐promoted AIR by ICIs. 2. MATERIALS AND METHODS 2.1. Cell culture and reagents Most cell lines were obtained from American Type Culture Collection (ATCC, Manassas, VA, USA), while human primary T cells were obtained from Oribiotech (Shanghai, China). All cell lines were authenticated via short tandem repeat genotyping in November 2021. Human TNBC cell line MDA‐MB‐231 and human embryonic kidney cell line HEK293T were grown in Dulbecco's Modified Eagle Medium (DMEM, Invitrogen, Carlsbad, CA, USA). Human T‐cell leukemia cell line Jurkat, human primary T cells and mouse TNBC cell line 4T1 were cultured with Roswell Park Memorial Institute (RPMI)‐1640 (Invitrogen), and human TNBC cell line MDA‐MB‐468 was cultured with Leibovitz's L‐15 (Gibco, Grand Island, NY, USA). All media were supplemented with 10% fetal bovine serum (Invitrogen), 100 U/mL penicillin and 100 μg/mL streptomycin (Invitrogen). MDA‐MB‐231, HEK293T, Jurkat, human primary T cells and 4T1 cells were cultured in a humidified atmosphere of 5% CO[2] at 37°C, while MDA‐MB‐468 cells were cultured under the same conditions except for a CO[2] concentration of 0.03%. 2.2. Stable transfection using lentiviral infection To knockout (ko) ARID1A through CRISPR/Cas9, the Lenti‐CAS9‐Puro vector and the GV371‐EGFP vector (Shanghai Genechem, Shanghai, China) were sequentially transfected. The latter carried the small guide RNA (sgRNA) sequence, 5’‐CACCGATGGTCATCGGGTACCGCTG‐3' or 5’‐CACCGCCCCTCAATGACCTCCAGTA‐3' for ARID1A knockout, and 5'‐CGCTTCCGCGGCCCGTTCAA‐3' for negative control (nc). For ARID1A overexpression (oe) using CRISPR/Cas9, a synergistic activation mediator vector (Genechem) was applied which carried the sgRNA sequence 5'‐CACCGGGCGCTCTAGCCGCTCAGTC‐3' or 5'‐CACCGCTTGGGTCGAGGCTGCTGCG‐3', and the empty vector was used as negative control. The GV654 vector (Genechem) was used to clone the short hairpin RNAs (shRNAs) for knocking down PD‐L1 and nucleophosmin 1 (NPM1). The sequences of shPD‐L1 and shNPM1 were 5'CCGGACCATCAAGTCCTGAGTGGTACTCGAGTACCACTCAGGACTTGATGGTTTTTTG‐3' and 5'‐CCGGCTGGAGGTGGTAGCAAGGTTCCTCGAGGAACCTTGCTACCACCTCCAGTTTTTG‐3', respectively, and the empty vector was used as negative control. The GV657 vector (Genechem) was used to clone the coding sequence of PD‐L1 for overexpression, and the empty vector was used as negative control. The plasmids were transfected into HEK293T cells with Lipofectamine 3000 transfection reagent (L3000‐150, Invitrogen), and the virus‐containing supernatant was collected 72 h after the incubation. The virus was then concentrated and transfected into TNBC cells with polybrene (sc‐134220, Santa Cruz Biotechnology, Dallas, TX, USA). The transfected cells were selected 2 days later with neomycin for at least 1 week. 2.3. Western blotting analysis Cells were harvested, and protein was extracted using radio immunoprecipitation assay lysis buffer (PC101, Shanghai Epizyme Biomedical Technology, Shanghai, China). A nitrocellulose membrane (Bio‐Rad Laboratories, Hercules, CA, USA) was used to transfer protein, which was incubated for 1 h at room temperature with nonfat milk. Primary antibodies included ARID1A (1:1,000, 12354, Cell Signaling Technology, Danvers, MA, USA), PD‐L1 (1:1,000, 13684, Cell Signaling Technology), NPM1 (1:1,000, FC‐61991, Invitrogen) and β‐Actin (1:1,000, 3700, Cell Signaling Technology). After three washes with tris buffered saline with Tween‐20 (TBST, PS103, Shanghai Epizyme Biomedical Technology), the membrane was incubated with an IRDye800‐labelled secondary antibody (1:1,000, ab216773, Abcam, Cambridge, UK) for 1 h at room temperature. The blots were developed with an Odissey CLx Imager (Li‐Cor GmbH, Homburg, Germany) and quantified through Image Studio, version 5 (Li‐Cor GmbH). 2.4. RNA extraction and real‐time quantitative PCR (qPCR) assay Total RNA was extracted from MDA‐MB‐231, MDA‐MB‐468 and 4T1 cells by TRIzol reagent (Invitrogen) according to the manufacturer's instructions. RNA was reversely transcribed and subjected to real‐time PCR with SYBR Premix Ex Taq (RR420Q/A/B, TaKaRa Biotechnology Dalian, Dalian, Shandong, China). Thermal cycling was as follows: 94°C for 30 s; 35 cycles of 94°C for 30 s, 55°C for 30 s and 72°C for 30 s; 72°C for 10 min; hold at 4°C. The expression levels were normalized to β‐Actin and calculated using the 2^−ΔΔCt method. The primer sequences (5'–3') were as follows: * ARID1A‐forward: CCAGCAGAACTCTCACGACC; * ARID1A‐reverse: CTGAGCGAAGGACGAAGACG; * PD‐L1‐forward: TGGCATTTGCTGAACGCATTT; * PD‐L1‐reverse: TGCAGCCAGGTCTAATTGTTTT; * NPM1‐forward: GGAGGTGGTAGCAAGGTTCC; * NPM1‐reverse: TTCACTGGCGCTTTTTCTTCA; * β‐Actin‐forward: GCACAGAGCCTCGCCTT; * β‐Actin‐reverse: GTTGTCGACGACGAGCG. 2.5. RNA‐seq library construction and analysis Total RNA (1 μg per sample) was extracted from MDA‐MB‐231 cells. RNA‐seq libraries were prepared using a VAHTS Stranded mRNA‐seq Library Prep Kit for Illumina V2 (NR612‐01, Vazyme, Nanjing, Jiangsu, China) according to the manufacturer's instructions and sequenced after quality inspection by the Illumina sequencing platform (Illumina, San Diego, CA, USA) on a 150 bp paired‐end run. Sequencing reads were processed through FastQC ([78]http://www.bioinformatics.babraham.ac.uk/projects/fastqc), and the 3' end 60 nt was trimmed off to remove low‐quality nucleotide and adaptor sequences originated from the running off of relatively short inserted fragments. The clean reads were aligned using the spliced read aligner HISAT2 ([79]http://daehwankimlab.github.io/hisat2/), which was supplied with the Ensembl human genome assembly (Genome Reference Consortium GRCh38) as the reference genome. Gene expression levels were calculated by fragments per kilobase million (FPKM). 2.6. Analysis of PD‐L1 expression on cell surface MDA‐MB‐231 and MDA‐MB‐468 cells were digested into single cells and washed three times with cold phosphate buffer saline (PBS, Gibco). Cells were then resuspended with 100 μL PBS and incubated for 15 min on ice with CD274‐PE (1:20, 12‐5983‐42, eBioscience, San Diego, CA, USA) or mouse IgG1 kappa isotype control‐PE (1:20, 12‐4714‐82, eBioscience). After washing with cold PBS, PD‐L1 expression on the cell surface was detected by flow cytometry (Thermo Fisher, Boston, MA, USA). 2.7. Cell Counting Kit‐8 (CCK‐8) and colony formation assay For CCK‐8 cell proliferation assay, cells were seeded in 96‐well plates (3,000 cells/well). The absorbance at 450 nm was determined using a microplate reader 2 h after adding 10 μL CCK‐8 solution (HY‐K0301, MedChemExpress, Shanghai, China) to each well. For colony formation assay, cells were seeded in 6‐well plates (1,000 cells/well) and cultured for 10‐14 days. Colonies were fixed and stained with 0.2% crystal violet. A colony was defined as more than 10 cells, and colonies were counted under the microscope. 2.8. Jurkat coculture and enzyme‐linked immunosorbent assay (ELISA) for interleukin (IL)‐2 MDA‐MB‐231 and MDA‐MB‐468 cells were treated with interferon (IFN)‐γ (300‐02, Peprotech, Rocky Hill, NJ, USA), and Jurkat cells were stimulated with 50 ng/L phorbol 12‐myristate 13‐acetate (PMA, P8139, Sigma‐Aldrich, Saint Louis, MO, USA) and 1 μg/mL ionomycin (I3909, Sigma‐Aldrich), both for 24 h. Next, 1 × 10^4 TNBC cells/well were seeded in 96‐well plates. The supernatants were removed after the cells adhered, and Jurkat cells were added to TNBC cells at a ratio of 4:1 in 200 μL media. At 48 h later, the supernatants were collected and examined with a Human IL‐2 Valukine ELISA kit (VAL110, Novus, Littleton, CO, USA). The results were analyzed by ELISACalc V0.1 (Shanghai Bluegene Biotech Co. Ltd, Shanghai, China). 2.9. Primary T cell coculture and IFN‐γ production assay MDA‐MB‐231 and MDA‐MB‐468 cells were treated with IFN‐γ, and primary T cells were stimulated with 50 ng/L PMA and 1 μg/mL ionomycin for 24 h. Next, 2 × 10^4 TNBC cells/well were seeded in 24‐well plates. After the cells adhered, the supernatants were removed, and primary T cells were added to TNBC cells at a ratio of 30:1 in 500 μL media. At 24 h later, primary T cells were collected, treated with protein transport inhibitor cocktail (00‐4980‐93, eBioscience) for 6 h, fixed and permeabilized with intracellular fixation and permeabilization buffer set (88‐8824‐00, eBioscience). Samples were further incubated with IFN‐γ antibody (1:20, 17‐7319‐41, eBioscience) for 30 min and analyzed with flow cytometry. Data were assessed using CytExpert (2.2.0.97) [[80]26, [81]27, [82]28, [83]29]. 2.10. Chromatin immunoprecipitation (ChIP)‐qPCR assay ChIP assay was conducted using SimpleChIP Enzymatic Chromatin IP kit (9002S, Cell Signaling Technology) according to the manufacturer's instructions. In brief, TNBC cells were cultured to approximately 1 × 10^7 and cross‐linked by 1% formaldehyde. Chromatin was digested with micrococcal nuclease (10011, Cell Signaling Technology), and the nuclear membrane was broken through several pulses. The DNA fragment length was between 150‐900 bp. Chromatin was immunoprecipitated by either control IgG, H3K4me3 (1:50, 9751, Cell Signaling Technology), H3K4me1 (2 μg, ab8895, Abcam), H3K27ac (1 μg, C15210016, Diagenode, New Jersey, USA), ARID1A (1:100), SWI/SNF‐related, matrix‐associated, actin‐dependent regulator of chromatin, subfamily A, member 4 (SMARCA4, also known as BRG1) (1:50, 49360, Cell Signaling Technology), or NPM1 primary antibody (1:50, 4TOU‐1B2, Invitrogen) and reversely cross‐linked. The eluted DNAs were processed by a QIAseq ultralow input library kit (180495, QIAGEN, Nordrhein‐Westfalen, German) and quantified by qPCR. Thermal cycling was as follows: 94°C for 30 s; 35 cycles of 94°C for 30 s, 55°C for 30 s and 72°C for 30 s; 72°C for 10 min; hold at 4°C. The primer sequences (5'‐3') were as follows: * PD‐L1‐forward‐1: TTGGGCCCATTCACTAACCC; * PD‐L1‐reverse‐1: AAGAACTTCCCATCCCGAGC; * PD‐L1‐forward‐2: CTAGAAGTTCAGCGCGGGAT; * PD‐L1‐reverse‐2: GGCTGCGGAAGCCTATTCTA; * NPM1‐forward‐1: GGTCTTCAGAACGCCCCAAT; * NPM1‐reverse‐1: GGGAGCGAGCATGGGAATTA; * NPM1‐forward‐2: ACGTTAATTCCCATGCTCGC; * NPM1‐reverse‐2: GGGCCGACTCTGACTTCTTG; * F‐box protein 38 (FBXO38)‐forward: GCATGGCGCTTTAGTGTCAG; * FBXO38‐reverse: GCCTCAAACCCGTGTCCATA; * Nonspecific primer (NSP)‐forward: GGAGTGTACACATCTGACCT; * NSP‐reverse: TCTTCTCGCAGGACACGTCA. 2.11. ChIP‐PCR The immunoprecipitated DNA prepared as above was quantified by PCR amplification with Taq DNA polymerase (Yeasen Biotech, Shanghai, China). PCR primers were the same as those for ChIP‐qPCR. Thermal cycling was as follows: 94°C for 30 s; 30‐40 cycles of 94°C for 30 s, 52‐58°C for 30 s and 72°C for 30 s; 72°C for 10 min; hold at 4°C. The PCR product was visualized in a 2% agarose gel stained with ethidium bromide dye. Values were normalized to input DNA and performed in technical triplicate. 2.12. ChIP‐seq library construction and analysis The DNA fragments obtained as described above were used for library construction following the manufacturer's instructions. Samples were sequenced in Cloud‐seq (Shanghai, China) by NovaSeq 6000 (Illumina) after quality inspection, and paired‐end reads of 150 nt length were obtained. After the evaluation and cleaning described as above, sequencing reads were aligned to the human genome (UCSC hg19) with bowtie2 (v2.2.4) [[84]30]. In addition, Model‐based Analysis of ChIP‐Seq (MACS, v1.4.3) was used for peak calling, and diffReps was applied to find significantly different peak distributions between ARID1A‐ko and negative control samples [[85]31, [86]32]. 2.13. Dual‐luciferase reporter assay MDA‐MB‐231 cells were seeded in 24‐well plates at 1 × 10^5 cells/well and transfected with 0.5 μg/well luciferase reporter plasmids (Genechem). The cells were cotransfected with 10 ng of pRL‐CMV (Renilla luciferase, E2261, Promega, Madison, WI, USA) for normalization. At 48 h post transfection, the luciferase activity was detected using a Dual‐Luciferase Reporter Assay System Kit (E1910, Promega) following the manufacturer's instructions. 2.14. Assay for transposase accessible chromatin (ATAC)‐seq library construction and analysis MDA‐MB‐231 cells were collected and prepared for ATAC‐seq analysis as previously described [[87]33]. In brief, 5 × 10^4 cells were washed with PBS and lysed in lysis buffer (10 mmol/L Tris‐HCl pH 7.4, 10 mmol/L NaCl, 3 mmol/L MgCl[2], 0.1% Nonidet P‐40). After centrifugation at 800 × g, 4°C for 15 min, the pellets were resuspended and incubated in a transposition mix containing Tn5 transposase (Illumina) for 30 min at 37°C. The purified DNA was subsequently ligated with adapters and amplified for 11 cycles. The libraries were purified using AMPure beads (A63882, Beckman Coulter, Hebron, KY, USA) and sequenced on Illumina X‐Ten system through a 150 bp paired‐end strategy. For data analysis, Trimmomatic (v0.36, [88]http://www.usadellab.org/cms/?page=trimmomatic) was used to trim Illumina adapters and reads of low quality. Total reads in each sample were aligned to human reference genome hg19 using bwa (v0.7.11, [89]https://sourceforge.net/projects/bio‐bwa/files/) with default settings. Samtools (v1.6, [90]https://samtools.sourceforge.net) was used to remove low‐quality reads (mapping quality b20) and reads mapping to mitochondrial DNA, and Picard Tools (v1.4.5, [91]https://broadinstitute.github.io/picard/) was used to exclude duplicates. All mapped reads were offset by +4 bp for the positive strand and ‐5 bp for the negative strand using bedtools (v2.25.0, [92]https://github.com/arq5x/bedtools2). Libraries of the same cell lines were merged for subsequent analyses. MACS2 (v2.1.1.20160309, [93]https://pypi.org/project/MACS2/) with parameter “‐nomodel ‐shift 100 ‐extsize 200” was used for peak calling. Peaks with P < 0.01 were merged for searching differential peaks. Peak annotation and motif analysis were conducted using HOMER (v4.9.1, [94]http://homer.ucsd.edu/homer/). Sequencing signals were generated by transforming the mapping files into bigwig tracks and visualized in Integrated Genomic Viewers (University of California, San Diego, CA, USA). 2.15. Animals and treatment For xenograft mouse model, human peripheral blood mononuclear cell (Hu‐PBMC) reconstruction mice were purchased from Shanghai Model Organisms Center (Shanghai, China). Briefly, 5 × 10^6 MDA‐MB‐231 cells were suspended in normal saline and injected into the mammary fat pads of female NOD.Cg‐Prkdc^scidIl2rg^em1Smoc mice aged 6 weeks. The mice lack mature T, B and NK cells and demonstrate low immune rejection against human cells and tissues [[95]34]. Tumor sizes were monitored every 3 days. Tumor volume was calculated as π/6 × length × width^2 (mm^3). When the tumor volume reached 80‐100 mm^3, 5 × 10^6 Hu‐PBMCs were intravenously engrafted. On average, the mice achieved efficient T cell reconstruction after 10 days and resulted in 15%‐45% human CD45^+ cells in the mouse peripheral blood within 4 weeks, suitable to assess tumor immunotherapy [[96]34]. The control group was treated with PBS and IgG, and the anti‐PD‐L1 treatment group was treated with PBS and atezolizumab (Ate) (HK‐65567, F.Hoffmann‐La Roche Ltd, Basel, Switzerland) 3 mg/kg twice a week for 2 weeks intraperitoneally [[97]35]. For isograft mouse model, 2 × 10^6 4T1 cells were suspended in normal saline and injected into the mammary fat pads of female BALB/c mice (Shanghai Model Organisms Center) aged 6 weeks. Tumor sizes were monitored and calculated as described above. All mice were maintained in specific pathogen‐free cages and provided autoclaved food and water in accordance with the Accreditation of Laboratory Animal Care International Guidelines. Animals were humanely euthanized through CO[2] overdose if the tumor burden was greater than 2,000 mm^3 or at the planned end point of the experiment. Animal experimental protocols were approved by the Animal Care Committee of Fudan University Shanghai Cancer Center (FUSCC, No. 2020FUSCCJS‐237). 2.16. Tumor tissue digestion and flow cytometry Mouse mammary tumors were excised and divided into two parts for immunohistochemistry (IHC) and flow cytometry. For flow cytometry, tumor tissues were cut into small pieces and digested with 1 mg/mL collagenase type IV (C5138, Sigma‐Aldrich) and 0.6 ku/mL DNAse (D5025, Sigma‐Aldrich) for 2.5 h. Samples were then filtrated into single‐cell suspensions and stained with CD45‐FITC (1:20, 11‐0451‐82, eBioscience), CD8‐PE (1:20, 12‐0081‐82, eBioscience), CD107‐APC (1:20, MA5‐28671, eBioscience) and CD69‐PerCP‐Cyanine5.5 (1:20, 45‐0691‐82, eBioscience). The results of flow cytometry were analyzed with FlowJo V10 (Becton Dickinson, Franklin Lakes, NJ, USA). 2.17. Human tissue specimens A set of tissue microarray (TMA) including 258 breast cancer tissue samples was obtained from the recruited breast cancer patients, diagnosed between August 2015 and December 2017, at their first radical surgery with no prior systemic neoadjuvant treatment. We included female patients who had been pathologically diagnosed only with breast cancer in their lifetime. Other inclusion criteria were age 18‐90 years, of TNBC subtype (American Society of Clinical Oncology/College of American Pathologists guidelines [[98]36, [99]37]), and with stage I‐IIIA disease (the 8^th edition of the Union for International Cancer Control TNM staging system [[100]38]). Patients who did not receive standard adjuvant treatment or with history of autoimmune diseases were excluded. All patients were graded according to the Nottingham histologic scoring system [[101]39]. The study design, endpoints and results of the CTR20191353 trial have been reported previously [[102]40]. Resected specimens were macroscopically examined to determine the location and size of a tumor. Specimens for histology were fixed in 10% (v/v) formalin and processed for paraffin embedding. Overall survival (OS) was calculated from the date of surgery to the date of any‐cause death or the last follow‐up, whichever came first. Disease‐free survival (DFS) was defined as the time interval from the date of surgery to the date of breast cancer recurrence or the last follow‐up (June 2021), whichever came first. Progression‐free survival (PFS) was defined as the time interval from the date of treatment with chemotherapy plus pucotenlimab to the date of breast cancer progression or the last follow‐up, whichever came first. PFS‐long and short patients were defined based on a 50% cutoff. Written informed consent was obtained from all patients, and all research protocols were approved by the medical ethics committee of FUSCC. 2.18. Histopathology The primary antibodies against ARID1A (1:1,000, sc‐32761, Santa Cruz Biotechnology), CD8 (1:500, ab217344, Abcam), granzyme B (GB, 1:200, ab255598, Abcam), PD‐L1 (1:60, 22C3, Dako, Hovedstaden, Denmark) or NPM1 (1:1,000) were incubated at 4°C overnight. TMA was scanned by a NanoZoomer HT 1.0 (Hamamatsu Photonics Co. Ltd, Hamamatsu, Shizuoka, Japan) and processed through the VIS DIA VisioMarph system (Visiopharm, Hoersholm, Denmark) according to the manufacturer's instructions. Five views were examined per core, and 100 cells were observed per view at 400 × magnification. TMA cores which contained <100 viable tumor cells were excluded. As immunohistochemical staining of ARID1A and NPM1 was primarily located in the nucleus of tumor cells, their protein levels were scored by the number of positive tumor cells divided by the total number of tumor cells. Level 0 to 3 of ARID1A expression were defined based on score quantiles, with 0 correspongding to the lowermost quantile and 3 correpsonding to the uppermost quantile. ARID1A‐ and NPM1‐high versus ‐low patients were classified based on a 50% cutoff. PD‐L1 tumor proportion score (TPS) was defined as the number of positive tumor cells divided by the total number of viable tumor cells multiplied by 100, and combined positive score (CPS) was the number of positive tumor cells, lymphocytes and macrophages divided by the total number of viable tumor cells multiplied by 100 [[103]41]. The CD8 and GB levels were determined as the number of positive cells divided by the total number of cells since both were mainly found in tumor‐infiltrating lymphocytes (TILs). Tertiary lymphoid structures, necrosis and other immune infiltrates beyond the tumor border were excluded. 2.19. Multispectral fluorescent IHC Multiplexed immunofluorescent staining and multispectral image analysis were performed as previously described [[104]42]. In brief, 4‐mm‐thick formalin‐fixed paraffin‐embedded slides were deparaffinized in a Leica auto‐stainer (Leica Biosystems, Nussloch, Germany), and citrate buffer pH 6.0 (Shanghai Epizyme Biomedical Technology) was used for antigen retrieval. Slides were subsequently blocked with Antibody Diluent (BioGenex Laboratories, Fremont, CA, USA) and incubated with primary antibodies for 30 min at room temperature. Primary antibodies included ARID1A (1:300), CD8 (1:100, C8/1444B, Dako), GB (1:300), PD‐L1 (1:100) and NPM1 (1:200). After washing with TBST, the slides were incubated with horseradish peroxidase‐linked secondary antibodies (Life Technologies, Carlsbad, CA, USA). Afterwards, tyramide‐conjugated fluorophores (Opal, PerkinElmer, Waltham, MA, USA) were added at a 1:50 dilution and incubated for 10 min at room temperature. This process was repeated for all 5 antibodies, and 4’,6‐diamidino‐2’‐phenylindole (Life Technologies) was diluted at 1:500. Finally, slides were cover slipped with VECTASHIELD Hard Mount (Vector Laboratories, Burlingame, CA, USA). Scanning was performed with a Vectra automated multispectral microscope (PerkinElmer), and inForm software (PerkinElmer) was used for analysis. 2.20. Gene set enrichment analysis (GSEA) GSEA was performed using Java software GSEA v4.1.0 based on the Molecular Signature Database v7.2 (M9779, M9774, M1086; [105]https://www.gsea‐msigdb.org/gsea/msigdb/index.jsp) [[106]43, [107]44]. Hits (gene set numbers) and misses (non‐members) were scored with the “weighted” enrichment method, and the metric for ranking genes was the “Signal2Noise” method. All other parameters were set as default. 2.21. Database and computational algorithms A total of 251 epigenetic regulator proteins in Homo sapiens were retrieved from Gene Ontology Resource (GO:0040029, [108]http://geneontology.org), and gene signatures for AIR were utilized to assess expression correlation [[109]45, [110]46, [111]47, [112]48]. All gene expression in TNBC was analyzed using The Cancer Genome Atlas (TCGA) data in cBioPortal ([113]https://www.cbioportal.org), and ARID1A‐high and ‐low patients were categorized based on a 50% cutoff [[114]49]. Immune cell abundance in the breast tumor microenvironment was estimated using CIBERSORT [[115]50], Thorsson [[116]51] and TIMER2.0 computational algorithms [[117]52]. Single‐cell RNA‐seq data for TNBC patients were obtained from the Gene Expression Omnibus (GEO) dataset [118]GSE75688 ([119]https://www.ncbi.nlm.nih.gov/geo/), and cells were profiled using Smart‐seq2 [[120]53]. The Seurat package (3.1.1, [121]https://cran.r‐project.org/web/packages/Seurat/index.html) was adopted for further analysis. Regulators for CD8^+ T cells were derived from Gene Ontology Resource (GO: 2001185, [122]http://geneontology.org). Sequences of PD‐L1 and NPM1 promoters and coding regions were obtained from the University of California Santa Cruz (UCSC) database ([123]https://genome.ucsc.edu), and PD‐L1 and NPM1 enhancers were predicted through EnhancerDB ([124]http://lcbb.swjtu.edu.cn/EnhancerDB/). Transcription regulators for PD‐L1 were predicted through UCSC database (score range set as > 530), identifying 13 transcription regulators, and literature review was performed using PubMed ([125]https://pubmed.ncbi.nlm.nih.gov) with the query “(“PD‐L1” [Title/Abstract] OR “CD274” [Title/Abstract]) AND (TNBC [Title/Abstract] OR “triple negative breast cancer” [Title/Abstract] OR “triple‐negative breast cancer” [Title/Abstract]) AND (transcription [Title/Abstract] OR transcript [Title/Abstract] OR mRNA [Title/Abstract])” and returned 60 results. After detailed reading, we filtered out 47 irrelevant publications and summarized the relevant 13 publications (including 15 transcription regulators, 1 overlapping with UCSC prediction) as Supplementary Table [126]S1. ARID1A and BRG1 ChIP‐seq data were obtained from the GEO datasets [127]GSE72141 and [128]GSE174360 ([129]https://www.ncbi.nlm.nih.gov/gds). All sequencing data were viewed using UCSC genome browser. 2.22. Statistical Analysis All data were analyzed and graphed using GraphPad Prism 8 (Graphpad Software Inc, San Diego, CA, USA) or R 4.0.2 ([130]https://www.r‐project.org). Experimental data are presented as mean ± standard deviation of three independent experiments. Two‐group comparisons were conducted by two‐sided Student's t‐test. The correlation analysis was conducted by Pearson correlation analysis. The Kaplan–Meier method was adopted for analyzing DFS, PFS and OS. All variables showing statistical significance in univariate regression were adopted for multiple regression via the Cox proportional hazards model. P < 0.05 was considered statistically significant. Western blotting was repeated three times independently showing similar results. 3. RESULTS 3.1. Deficiency of the epigenetic modulator ARID1A contributed to AIR in TNBC To explore epigenetic regulation for AIR in TNBC, we used RNA‐seq data from TNBC patients in TCGA to analyze the association between the expression of 251 epigenetic genes and AIR signatures (Supplementary Table [131]S2). The expression of ARID1A demonstrated the highest correlation with TNBC AIR among the epigenetic gene set, showing positive correlations with cytotoxic T cell function, lymphocyte infiltration, antigen presentation and IFN‐γ pathway score and a negative correlation with immune checkpoint score (Figure [132]1A, Supplementary Figure [133]S1A). This suggested that low ARID1A in TNBC patients contributed to AIR, resulting in progressive tumor growth. FIGURE 1. FIGURE 1 [134]Open in a new tab Deficiency of the epigenetic modulator ARID1A in TNBC contributed to AIR. (A) Heat map showing top 30 epigenetic modulators of highest expression correlation with AIR in TNBC. Yellow and blue depict positive and negative correlation, respectively. (B) Schematic diagram of Hu‐PBMC reconstruction mice (n = 5 for each group). Black arrow: injection of MDA‐MB‐231 cells into the mammary fat pads. * Hu‐PBMC engraftment when average tumor volume reaches 100 mm^3 (day 9). (C) Average tumor volume depicted over time in Hu‐PBMC reconstruction mice (n = 5 for each group, left) and tumor image at the end of experiment (right. negative control: day 39; knockout: day 24). (D) Quantification of lung metastatic foci (left) and representative HE staining (right) as in (C). (E) Average tumor volume of 4T1 cells injected into BALB/c mice depicted over time (n = 5 for each group, left) and tumor image at the end of experiment (right, day 30 for all experimental groups). (F) Quantification of lung metastatic foci (left) and representative HE staining (right) as in (E). (G) Average tumor volume of MDA‐MB‐231 cells injected into nude mice depicted over time (n = 5 for each group, left) and tumor image at the end of experiment (right, day 36 for all experiment groups). (H) Quantification of lung metastatic foci (left) and representative HE staining (right) as in (G). (I) Kaplan–Meier analysis for DFS (left) and OS (right) based on ARID1A expression in the FUSCC TNBC cohort TMA. 0 to 3 represent the lowest to highest ARID1A expression groups. ***P < 0.001, ns not significantly different. Abbreviations: ARID1A, AT‐rich interaction domain 1A; AIR, adaptive immune resistance; TNBC, triple‐negative breast cancer; Hu‐PBMC, human peripheral blood mononuclear cell; ko, knockout; oe, overexpression; nc, negative control; HE, hematoxylin‐eosin; DFS, disease‐free survival; OS, overall survival; FUSCC, Fudan University Shanghai Cancer Center; TMA, tissue microarray. To verify this hypothesis, we evaluated the role of ARID1A in tumor progression using mouse models. Through knocking out and overexpressing ARID1A in MDA‐MB‐231 cells (Supplementary Figure [135]S1B), we found that ARID1A knockout significantly increased both tumor growth and the number of lung metastatic foci in Hu‐PBMC reconstruction mice (Figure [136]1B‐D). The tumors grew to the approximate size of 2,000 mm^3 at day 24 and day 39 after TNBC cell injection in knockout and negative control groups, respectively (Figure [137]1C). Interestingly, ARID1A‐oe tumors gradually diminished after Hu‐PBMC engraftment, and few metastatic foci were found in the lungs. For analysis in immunocompetent settings through preclinical mouse models, we further constructed Arid1a‐ko and ‐oe 4T1 cells and performed transplantation using BALB/c mice (Supplementary Figure [138]S1C). Both tumor growth and lung metastasis in recipient mice were found dramatically increased in Arid1a‐ko group, whereas Arid1a‐oe 4T1 cells failed to form tumors and developed few lung metastatic foci (Figure [139]1E‐F). Notably, in nude mice without Hu‐PBMC reconstruction, no difference in tumor growth or lung metastasis was observed among ARID1A‐ko, ‐oe and negative control groups engrafted with MDA‐MB‐231 cells (Figure [140]1G‐H). We also compared tumor growth in Hu‐PBMC and non‐Hu‐PBMC conditions (Supplementary Figure [141]S1D). For negative control MDA‐MB‐231 cells, Hu‐PBMC reconstruction inhibited tumor growth relative to nude mice. For ARID1A‐ko MDA‐MB‐231 cells, tumors in Hu‐PBMC reconstruction mice demonstrated significantly faster growth as well as more frequent lung metastasis than in nude mice. For further validation in vitro, we found that ARID1A status did not affect colony formation, cell proliferation or apoptosis of two TNBC cell lines, MDA‐MB‐231 and MDA‐MB‐468, in the absence of lymphocytes (Supplementary Figure [142]S1E‐H). These results indicated immunity as the determinant for low ARID1A expression‐triggered TNBC progression. To validate the role of ARID1A in AIR using our own cohort, we included a 258‐TNBC patient cohort from FUSCC. Through analyzing associations between ARID1A expression and tumor malignancy and prognosis of TNBC, we found that patients with lower ARID1A demonstrated larger tumor size, more frequent lymph node invasion, higher KI67 and more advanced tumor grade than patients with higher ARID1A (Table [143]1, Supplementary Figure [144]S1I). In survival analyses, low ARID1A in TNBC was identified as an independent poor prognostic factor for both DFS and OS (Figure [145]1I, Supplementary Tables [146]S3‐S4). TABLE 1. Characteristics of 258 patients with TNBC in the FUSCC cohort. Characteristic Total population [cases (%)] ARID1A‐high [cases (%)] ARID1A‐low [cases (%)] P [147]^a Total 258 (100) 128 (49.4) 130 (50.2) Age at diagnosis 18‐34 years 17 (6.6) 4 (3.1) 13 (10.0) 0.042 ≥ 35 years 241 (93.4) 124 (96.9) 117 (90.0) T stage T1 106 (41.1) 62 (48.4) 44 (33.8) 0.023 T2‐3 152 (58.9) 66 (51.6) 86 (66.2) Lymph node metastasis Negative 132 (51.2) 75 (58.6) 57 (43.8) 0.019 Positive 126 (48.8) 53 (41.4) 73 (56.2) KI67 expression[148] ^b Low 57 (22.1) 36 (28.1) 21 (16.2) 0.024 High 201 (77.9) 92 (71.9) 109 (83.8) Histological grade I 17 (6.6) 13 (10.2) 4 (3.1) 0.025 II 118 (45.7) 62 (48.4) 56 (43.1) III 123 (47.7) 53 (41.4) 70 (53.8) Vascular invasion Negative 158 (61.2) 81 (63.3) 77 (59.2) 0.446 Positive 100 (38.8) 47 (36.7) 53 (40.8) [149]Open in a new tab ^^a Differences between ARID1A‐high and ARID1A‐low were compared using the Pearson's Chi‐square test. ^^b KI67 low ≤ 20%, high > 20% We also assessed the relationship between the protein level and mutations and copy number alterations of ARID1A in TNBC using the data from TCGA database. We found that no patient possessed high‐level copy number amplification (CN2) or homozygous deletion (CN‐2). Meanwhile, 20.2% (24/119), 36.1% (43/119) and 43.7% (52/119) patients showed low levels of copy number gain (CN1), hemizygous deletion (CN‐1) and no change (CN0), respectively. In addition, we found that the protein levels of ARID1A in both CN0 (P = 0.007) and CN1 patients (P = 0.006) were significantly higher than that in CN‐1 patients (Supplementary Figure [150]S1J). Based on a 50% cutoff for protein expression, 64.3% CN‐1, 45.8% CN1 and 37.2% CN0 patients were in ARID1A‐low group (CN‐1 vs. CN1, P < 0.001; CN‐1 vs. CN0, P < 0.001; CN1 vs. CN0, P = 0.075). In addition, among the 119 TNBC patients, only 2 (1.68%) harbored ARID1A mutations. One was Y2031*, categorized as nonsense, and the other was ARID1A X1136_splice. It was noteworthy that the latter patient also demonstrated ARID1A CN‐1, and both patients were categorized as ARID1A‐low based on the 50% cutoff for protein expression. Therefore, we consider that categorizing TNBC patients by ARID1A protein level has taken into account the impact of ARID1A mutation and copy number alteration. Taken together, low expression of the epigenetic modulator ARID1A contributed to AIR in TNBC, leading to tumor progression. 3.2. ARID1A deficiency inhibited CD8^+ T cells to enhance AIR in TNBC To further explore the role of ARID1A in AIR regulation, we focused on the relationship between ARID1A and immune cells in the tumor immune microenvironment (TIME) through CIBERSORT computational algorithm. We found that ARID1A expression in TNBC tissue was positively associated with infiltration of CD8^+ T cells, T follicular helper cells and M1 macrophages, and negatively correlated with infiltration of regulatory T cells and M2 macrophages. Compared with other immune cell types, CD8^+ T cells were found to be the most significantly correlated with ARID1A expression in TCGA (Figure [151]2A). This was further supported by single‐cell RNA‐seq in 5 TNBC patients, as well as Thorsson and TIMER computational algorithms (Figure [152]2B‐C, Supplementary Figure [153]S2A‐B). GSEA also showed CD8^+ T cell inactivation to be significantly associated with low ARID1A expression (Figure [154]2D). Therefore, we hypothesized that CD8^+ T cells might be the key factor for low ARID1A expression‐induced AIR in TNBC. FIGURE 2. FIGURE 2 [155]Open in a new tab TNBC ARID1A deficiency inhibited CD8^+ T cells to enhance AIR. (A) CIBERSORT analysis showing correlation of ARID1A with immune cell abundance in TCGA. Red and green depict lower and higher P values, respectively, and circle size represent the percentage of immune cells. (B) Bar plot showing the fraction of immune cells in 5 TNBC patients using single cell RNA‐seq. The patients were numbered according to ARID1A expression in TNBC. Yellow triangle (down) represented lower to higher ARID1A expression in patient Number I to V. (C) Scatter plot depicting the relationship between ARID1A mRNA level as measured by RNA‐seq and CD8^+ T cell percentage in TIME in TCGA. (D) GSEA plot showing low ARID1A expression‐induced genes enriched for CD8^+ T cell inactivation in TNBC patients. (E) Comparison of CD8^+ T cell markers CD8 and GB by ARID1A expression in the FUSCC TNBC cohort TMA. (F) Quantification (left) and representative images (right) of CD8^+ T cell markers CD8 and GB IHC in Hu‐PBMC tumor tissue. (G) Jurkat cells were cocultured with MDA‐MB‐231 cells and IL‐2 secretion was detected by ELISA. Primary T cells were cocultured with MDA‐MB‐231 cells and IFN‐γ‐producing T cells were detected by flow cytometry. ***P < 0.001. Abbreviations: ARID1A, AT‐rich interaction domain 1A; AIR, adaptive immune resistance; TCGA, The Cancer Genome Atlas; TNBC, triple‐negative breast cancer; RNA‐seq, RNA sequencing; TIME, tumor immune microenvironment; GSEA, gene set enrichment analysis; GB, granzyme B; FUSCC, Fudan University Shanghai Cancer Center; TMA, tissue microarray; IHC, immunohistochemistry staining; Hu‐PBMC, human peripheral blood mononuclear cell; ko, knockout; nc, negative control; ELISA, enzyme linked immunosorbent assay; FPKM, fragments per kilobase million; NES, normalized enrichment score; FDR, false discovery rate. We further explored the relationship between ARID1A expression and CD8^+ T cells in tumor samples from the FUSCC TNBC cohort and from the above‐described mouse models through analyzing CD8 and GB expression. In accordance with transcriptome data, the protein levels of CD8 and GB were much lower in ARID1A‐low tumors, indicating inhibited CD8^+ T cell infiltration and activation (Figure [156]2E‐F, Supplementary Figure [157]S2C). To verify the function of ARID1A in vitro, two types of T cells, Jurkat and primary T cells, were cocultured with TNBC cells differentially expressing ARID1A. ELISA showed that the secretion of IL‐2 by Jurkat cells was dramatically decreased when cocultured with ARID1A‐ko TNBC cells compared to those cocultured with negative control TNBC cells. Similarly, ARID1A silencing in TNBC cells significantly decreased the percentage of IFN‐γ‐producing primary T cells. Opposite effects of ARID1A overexpression on T cells were found (Figure [158]2G, Supplementary Figure [159]S2D). Taken together, TNBC cells with ARID1A low expression inhibited CD8^+ T cells in TIME. 3.3. PD‐L1 mediated ARID1A low expression‐induced CD8^+ T cell inactivation To probe the mechanisms underlying low ARID1A expression‐induced CD8^+ T cell malfunction and AIR, we focused on 21 CD8^+ T cell regulators derived from the Gene Ontology Resource (GO:2001185), including 12 positive and 9 negative regulators. Expression of these genes was examined through RNA‐seq in TCGA TNBC patients (ARID1A‐low versus ‐high) and MDA‐MB‐231 cells (ARID1A‐ko versus negative control) by calculating the fold change (Figure [160]3A, Supplementary Figure [161]S3A). Except for SH3 domain containing ring finger 1 (SH3RF1) and mitogen‐activated protein kinase 8 interacting protein 1 (MAPK8IP1), 11 positive and 8 negative regulators were significantly decreased and increased in ARID1A‐low patients and ARID1A‐ko cells, respectively. After mapping RNA‐seq q values of the 19 candidate genes in TCGA patients to the x‐axis, and those of MDA‐MB‐231 cells to the y‐axis, PD‐L1 was identified to be the candidate gene furthest from the origin (Figure [162]3B). FIGURE 3. FIGURE 3 [163]Open in a new tab PD‐L1 mediated ARID1A low expression‐induced CD8^+ T cell inactivation. (A) Scatter plot showing expression variation of CD8^+ T cell‐modulating genes in ARID1A‐low versus high TNBC patients in TCGA (x‐axis), as well as in ARID1A‐ko versus negative control MDA‐MB‐231 cells (y‐axis). (B) A log10‐qValue plot of screening results in (A). The RNA‐seq q value of the 19 candidate genes in ARID1A‐low versus high TCGA TNBC patients was mapped to the x‐axis and that of ARID1A‐ko versus negative control MDA‐MB‐231 cells was mapped to the y‐axis. (C) qPCR for PD‐L1 following ARID1A‐ko and ‐oe in MDA‐MB‐231 cells. (D) Western blotting for PD‐L1 following ARID1A knockout and overexpression in MDA‐MB‐231 cells. (E) Quantification (left) and representative images (right) of PD‐L1 CPS and TPS by ARID1A expression in the FUSCC TNBC cohort TMA. (F) Quantification (left) and representative images (right) of PD‐L1 IHC in Hu‐PBMC tumor tissue. (G) Average tumor volume of MDA‐MB‐231 cells with different ARID1A and PD‐L1 expression status injected into Hu‐PBMC reconstruction mice depicted over time (n = 5 for each group, up) and tumor image at the end of experiment (down left, day 30 for all experimental groups), with quantification of lung metastatic foci (down right). (H) Quantification of ARID1A, PD‐L1, CD8 and GB IHC in tumor tissue from (H). **, P < 0.01, ***P < 0.001, ns not significantly different. Abbreviations: PD‐L1, programmed cell death 1 ligand 1; ARID1A, AT‐rich interaction domain 1A; TNBC, triple‐negative breast cancer; TCGA, The Cancer Genome Atlas; ko, knockout; oe, overexpression; sh, short hairpin RNA‐mediated knockdown; nc, negative control; qPCR, quantitative real‐time PCR; CPS, combined positive score; TPS, tumor proportion score; FUSCC, Fudan University Shanghai Cancer Center; TMA, tissue microarray; IHC, immunohistochemistry staining; Hu‐PBMC, human peripheral blood mononuclear cell; GB, granzyme B; FC, fold change; vs., versus. Next, qPCR, Western blotting and flow cytometry all confirmed notably high PD‐L1 expression on the TNBC cell membrane in the ARID1A‐ko group, while the opposite variations were observed in ARID1A‐oe TNBC cells (Figure [164]3C‐D, Supplementary Figure [165]S3B‐D). In the FUSCC TNBC cohort, ARID1A‐low patients showed higher PD‐L1 levels both in TPS and CPS compared with ARID1A‐high counterparts (Figure [166]3E). In addition, we found that the protein level of PD‐L1 was higher in ARID1A‐ko tumor samples from MDA‐MB‐231 Hu‐PBMC mice and 4T1 BALB/c mice than that of negative control groups (Figure [167]3F, Supplementary Figure [168]S3E). To further assess whether PD‐L1 mediated low ARID1A expression‐induced CD8^+ T inactivation, we knocked down PD‐L1 in ARID1A‐ko TNBC cells (Supplementary Figure [169]S3F) and performed coculture assays with T lymphocytes. The activity of both Jurkat and primary T cells, though dampened by ARID1A‐ko TNBC cells relative to negative control TNBC cells, were rescued by ARID1A‐ko/shPD‐L1 TNBC cells (Supplementary Figure [170]S3G). Further, we overexpressed PD‐L1 in ARID1A‐oe MDA‐MB‐231 (Supplementary Figure [171]S3H) and evaluated the role of PD‐L1 in rescuing tumor growth in vivo under ARID1A‐oe condition (Figure [172]3G). Similar to previous observations, AIRD1A‐ko tumors demonstrated faster growth as well as a higher frequency of lung metastasis compared with negative control. However, in ARID1A‐ko tumors coharboring shPD‐L1, both tumor growth and lung metastasis were significantly decreased. While ARID1A‐oe tumors failed to maintain growth after Hu‐PBMC engraftment and demonstrated few lung metastatic foci, ARID1A‐oe/PD‐L1‐oe tumors demonstrated faster growth as well as higher frequency of lung metastasis (Figure [173]3G, Supplementary Figure [174]S4A). We also found that CD8 and GB were decreased in ARID1A‐ko tumors but were rescued by shPD‐L1 (Figure [175]3H, Supplementary Figure [176]S4B). The above in vitro and in vivo results indicated that PD‐L1, induced by low ARID1A expression, inhibited CD8^+ T cells and led to AIR. 3.4. Low ARID1A expression activated PD‐L1 transcription, while ARID1A did not bind to PD‐L1 promoter or enhancer To elucidate how ARID1A inhibited PD‐L1 expression, ChIP‐seq was performed with antibodies against H3K4me3 (indicative of active promoters). In the promoter region of PD‐L1, signals were markedly increased in ARID1A‐ko cells (Figure [177]4A), which was further confirmed by ChIP‐qPCR (Figure [178]4B). Dual luciferase reporter assay demonstrated corresponding variations in PD‐L1 promoter activity (Figure [179]4C‐D), with an irrelevant promoter (activin A receptor type 1, FOP) as the negative control to confirm specificity in ARID1A's regulatory activity (Supplementary Figure [180]S5A). These results led to the conclusion that ARID1A deficiency activated PD‐L1 transcription. FIGURE 4. FIGURE 4 [181]Open in a new tab Low ARID1A activated PD‐L1 transcription, while ARID1A did not bind to PD‐L1 promoter or enhancer. (A) Signal track of H3K4me3 ChIP‐seq for PD‐L1 in ARID1A‐ko and negative control MDA‐MB‐231 cells. (B) H3K4me3 ChIP‐qPCR for PD‐L1. Primer1 and 2 are located according to arrows in (A). NSP, nonspecific primer at the open reading frame‐free region. (C) PD‐L1 promoter activity measured by dual‐luciferase assay in ARID1A‐ko versus negative control human TNBC cells. (D) PD‐L1 promoter activity measured by dual‐luciferase assay in ARID1A‐oe versus negative control human TNBC cells. (E) ARID1A (down) and BRG1 (up) ChIP‐seq for PD‐L1 promoter. (F) ATAC‐seq profiles for PD‐L1 in ARID1A‐ko and negative control MDA‐MB‐231 cells showing similar chromatin accessibility. (G) Pathway enrichment plot for H3K4me3 ChIP‐seq (left) and ATAC‐seq (right) in ARID1A‐ko versus negative control MDA‐MB‐231 cells. ***P < 0.001, ns not significantly different. Abbreviations: ARID1A, AT‐rich interaction domain 1A; PD‐L1, programmed cell death 1 ligand 1; ko, knockout; oe, overexpression; nc, negative control; TNBC, triple‐negative breast cancer; ChIP, chromatin immunoprecipitation; qPCR, quantitative real‐time PCR; BRG1, SWI/SNF related, matrix associated, actin dependent regulator of chromatin, subfamily a, member 4; ATAC, assay for transposase accessible chromatin; KEGG, Kyoto Encyclopedia of Genes and Genomes. We next examined whether ARID1A directly binds to the PD‐L1 promoter to regulate its expression and analyzed ChIP‐seq data for ARID1A and BRG1, another key component of SWI/SNF, in MDA‐MB‐231 cells. However, we failed to identify ARID1A‐ or BRG1‐binding sites in the PD‐L1 promoter (Figure [182]4E), while FBXO38, a positive control, demonstrated obvious peaks in its promoter region (Supplementary Figure [183]S5B). This negative result was also confirmed by ChIP‐qPCR and ChIP‐PCR (Supplementary Figure [184]S5C‐D), while binding sites in FBXO38 promoter could be confirmed (Supplementary Figure [185]S5E). We also checked PD‐L1 enhancers (Supplementary Table [186]S5) for ARID1A and BRG1 binding and found only 1 was bound by BRG1, while no ARID1A‐binding peak was observed (Supplementary Figure [187]S6A). In addition, ChIP‐qPCR and ChIP‐PCR for ARID1A and BRG1 failed to validate direct binding for this enhancer in TNBC cell lines (Supplementary Figure [188]S6B). We further performed ATAC‐seq to assess the chromatin accessibility of PD‐L1 in MDA‐MB‐231 cells and found no significant alteration induced by ARID1A‐ko (Figure [189]4F). To explore the potential regulatory mechanism underlying low ARID1A expression‐induced PD‐L1 transcription activation, we also performed ChIP‐seq for enhancer‐specific histone markers H3K4me1 and H3K27ac in ARID1A‐ko and negative control MDA‐MB‐231 cells, and conducted pathway analyses in H3K4me3, H3K4me1 and H3K27ac ChIP‐seq as well as ATAC‐seq. We found that the differential H3K4me3 signal distribution between ARID1A‐ko and negative control cells were enriched in AIR‐related pathways, including the “T cell receptor signaling pathway” and “antigen processing and presentation”, both ranked top 10 among a series of pathways. As for ATAC‐seq, top 10 enriched pathways also included the “T cell receptor signaling pathway” (Figure [190]4G). In H3K4me1 and H3K27ac ChIP‐seq, however, AIR was not among the top‐ranked enriched pathways (Supplementary Figure [191]S6C). This further suggested that ARID1A may regulate AIR in TNBC predominantly through promoter‐binding and chromatin remodeling. Taken together, the stimulatory effect of low ARID1A expression on PD‐L1 transcription may be achieved indirectly and possibility associated with promoter‐binding and chromatin remodeling. 3.5. Low ARID1A expression increased NPM1 to activate PD‐L1 transcription in TNBC We thus focused on the transcriptional regulators of PD‐L1. Through literature and UCSC Genome Browser, we identified 27 transcription regulators (Supplementary Table [192]S1). Except for interferon regulatory factor 7 (IRF7), all were reported to elevate PD‐L1 expression. Among the positive regulators, only 4 were increased in ARID1A‐ko MDA‐MB‐231 cells compared to negative control group, while the expression of IRF7 decreased in ARID1A‐ko cells. Among these 5 candidates, only NPM1 possessed both ARID1A‐ and BRG1‐binding sites at the promoter region according to ARID1A and BRG1 ChIP‐seq (Figure [193]5A, Supplementary Figure [194]S7A). Furthermore, TCGA TNBC data showed a negative correlation between NPM1 and ARID1A mRNA (Supplementary Figure [195]S7B), and GSEA demonstrated enrichment of the NPM1 neighborhood in ARID1A‐ko MDA‐MB‐231 cells relative to negative control (Supplementary Figure [196]S7C). Therefore, we hypothesized that ARID1A directly bound to the NPM1 promoter and inhibited its transcription, further resulting in a PD‐L1 decrease. FIGURE 5. FIGURE 5 [197]Open in a new tab ARID1A low expression increased NPM1 to activate PD‐L1 transcription in TNBC. (A) Transcription regulators for PD‐L1. Locations in Venn diagram indicate promoter binding by ARID1A and BRG1. Colors denote expression variation induced by ARID1A‐ko in MDA‐MB‐231 cells. (B) Western blotting for NPM1 following ARID1A‐ko (up) and ‐oe (down) in human TNBC cells. (C) Quantification (left) and representative images (right) of NPM1 IHC in the FUSCC TNBC cohort TMA. (D) Signal track of H3K4me3 ChIP‐seq for NPM1 in ARID1A‐ko and negative control MDA‐MB‐231 cells. (E) NPM1 promoter activity measured by dual‐luciferase assay in ARID1A‐ko (left) and ‐oe (right) MDA‐MB‐231 cells. (F) ATAC‐seq profiles for NPM1 in ARID1A‐ko and negative control MDA‐MB‐231 cells showing altered chromatin accessibility. (G) Western blotting (left) and qPCR (right) for PD‐L1 in ARID1A‐ko and shNPM1 MDA‐MB‐231 cells. (H) PD‐L1 promoter activity in ARID1A‐ko and shNPM1 human TNBC cells measured by dual‐luciferase assay. (I) Coculture and cytokine secretion assays using ARID1A‐ko and shNPM1 human TNBC cells as described in Figure [198]2 G. ***P < 0.001, ns not significantly different. Abbreviations: ARID1A, AT‐rich interaction domain 1A; NPM1, nucleophosmin 1; PD‐L1, programmed cell death 1 ligand 1; BRG1, SWI/SNF related, matrix associated, actin dependent regulator of chromatin, subfamily a, member 4; ko, knockout; oe, overexpression; sh, short hairpin RNA‐mediated knockdown; nc, negative control; IHC, immunohistochemistry staining; FUSCC, Fudan University Shanghai Cancer Center; TNBC, triple‐negative breast cancer; TMA, tissue microarray; ChIP, chromatin immunoprecipitation; ATAC, assay for transposase accessible chromatin; qPCR, quantitative real‐time PCR. Indeed, we found a remarkable NPM1 increase in ARID1A‐ko cells through both qPCR and Western blotting, while a NPM1 decrease was observed in ARID1A‐oe cells (Figure [199]5B, Supplementary Figure [200]S7D‐E). Higher NPM1 expression was also found in ARID1A‐low TNBC patients compared to ARID1A‐high counterparts (Figure [201]5C), as well as in ARID1A‐ko tumors relative to negative control from MDA‐MB‐231 Hu‐PBMC mice and 4T1 BALB/c mice (Supplementary Figure [202]S7F‐G). To further assess whether ARID1A downregulation activated NPM1 transcription, we analyzed H3K4me3 ChIP‐seq data and found enriched signals in the NPM1 promoter region in ARID1A‐ko cells (Figure [203]5D). H3K4me3 ChIP‐qPCR (Supplementary Figure [204]S7H), as well as luciferase reporter assays targeting the NPM1 promoter (Figure [205]5E, Supplementary Figure [206]S7I), also showed enhanced transcriptional activity. Using antibodies against ARID1A and BRG1, we confirmed direct binding of ARID1A to the NPM1 promoter (Supplementary Figure [207]S8A). ATAC‐seq profiles showed that chromatin accessibility at NPM1, particularly in its promoter, was increased in ARID1A‐ko MDA‐MB‐231 cells compared with negative control (Figure [208]5F). We also checked whether ARID1A or BRG1 bound to NPM1 enhancers but found no binding peaks through ChIP‐seq (Supplementary Table [209]S6). In addition, the binding of NPM1 to the PD‐L1 promoter was significantly elevated in ARID1A‐ko cells relative to negative control (Supplementary Figure [210]S8B). Finally, the increase in PD‐L1 expression in ARID1A‐ko cells was reversed by shNPM1 (Figure [211]5G‐H, Supplementary Figure [212]S8C), and impaired T cell activity induced by ARID1A‐ko TNBC cells compared to negative control was restored through shNPM1 in TNBC cells (Figure [213]5I). Taken together, ARID1A directly bound to the NPM1 promoter and its deficiency activated NPM1 transcription, promoting PD‐L1 expression and inhibiting CD8^+ T cells. 3.6. ARID1A low expression‐induced AIR and PD‐L1 elevation in TNBC could be targeted by ICI To explore the clinical significance of ARID1A deficiency‐induced PD‐L1 elevation and AIR, we treated the Hu‐PBMC reconstruction mice with Ate, a PD‐L1 antibody (Figure [214]6A). Again, ARID1A‐ko led to faster tumor growth and more frequent pulmonary metastasis relative to negative control. However, after being treated with Ate twice a week for two doses, dramatic reductions in both tumor volume and pulmonary metastasis were observed for the ARID1A‐ko group, while negative control group showed no sign of remission (Figure [215]6B‐C, Supplementary Figure [216]S9A). This suggested that ARID1A‐deficient tumors, though demonstrating a more aggressive nature, responded much better to ICI treatment compared with ARID1A‐proficient tumors. FIGURE 6. FIGURE 6 [217]Open in a new tab TNBC AIR and PD‐L1 contributed by low ARID1A expression could be targeted by ICI. (A) Schematic diagram of the dosage regimen (n = 5 for each group). Black arrow: tumor cell injection. * Hu‐PBMC engraftment when average tumor volume reaches 100 mm^3 (day 9). # atezolizumab (Ate) or placebo (Pla) administration. (B) Average tumor volume depicted over time. (C) Quantification of lung metastatic foci (up) and representative HE staining (down). (D) Kaplan–Meier analysis for PFS based on ARID1A and PD‐L1 expression in CTR20191353. (E) Quantification (up) and representative multispectral fluorescent IHC (down) of ARID1A, CD8^+ T cell markers, PD‐L1 and NPM1 by PFS in CTR20191353. ***P < 0.001, ns not significantly different. Abbreviations: TNBC, triple‐negative breast cancer; AIR, adaptive immune resistance; PD‐L1, programmed cell death 1 ligand 1; ARID1A, AT‐rich interaction domain 1A; ICI, immune checkpoint inhibitor; Hu‐PBMC, human peripheral blood mononuclear cell; HE, hematoxylin‐eosin; PFS, progression‐free survival; IHC, immunohistochemistry staining; NPM1, nucleophosmin 1; GB, granzyme B; ko, knockout; oe, overexpression; nc, negative control. To confirm the status of the PD‐1/PD‐L1 axis and CD8^+ T cells in response to Ate, TILs in the Hu‐PBMC model were analyzed by flow cytometry (Supplementary Figure [218]S9B). After being gated by the leukocyte common antigen CD45, the percentage of CD8^+ T cells was shown to be lower in ARID1A‐ko tumors compared to negative control. Since activated CD8^+ T cells coexpress CD69 and CD107, we further analyzed CD8^+CD69^+ and CD8^+CD107^+ cell populations and found, in accordance with CD8^+ T cells, that both were lower in ARID1A‐ko tumors relative to negative control (Supplementary Figure [219]S9C). In addition, IHC confirmed decreased CD8^+ T cell markers for anti‐tumor immune response along with elevated PD‐L1 and NPM1 in ARID1A‐ko tumors relative to negative control (Supplementary Figure [220]S9D). Ate significantly increased CD45^+ CD8^+ T cell population in ARID1A‐deficient tumors but failed to achieve this in the control group. This was also the case with CD107^+ and CD69^+ T cells. IHC also confirmed a higher ratio and enhanced activity of T cells in ARID1A‐ko tumors compared to negative control tumors after treatment with Ate (Supplementary Figure [221]S9C‐D). To further explore whether AIR and PD‐L1 elevation induced by ARID1A low expression could pose a therapeutic vulnerability, tumor tissues from patients with metastatic TNBC enrolled in the CTR20191353 clinical trial were analyzed. In total, 31 patients received pucotenlimab (a humanized anti‐PD‐1 antagonist IgG4 monoclonal antibody) plus gemcitabine and cisplatin treatment. Patients with low ARID1A expression in TNBC demonstrated significantly longer PFS. Notably, the longest PFS of 615 days was found in ARID1A‐low/PD‐L1‐high group (Figure [222]6D). Next, multispectral fluorescent IHC showed that patients with longer PFS demonstrated lower ARID1A expression and CD8^+ T cell infiltration compared to those with shorter PFS, whereas the expression of both PD‐L1 and NPM1 were significantly higher (Figure [223]6E). Taken together, the above results indicated that AIR and PD‐L1 expression in TNBC contributed by low level of ARID1A could be potentially reversed by ICIs, such as Ate and pucotenlimab (Figure [224]7). FIGURE 7. FIGURE 7 [225]Open in a new tab Schematic diagram of the working model. Abbreviations: TNBC, triple‐negative breast cancer; ARID1A, AT‐rich interaction domain 1A; NPM1, nucleophosmin 1; PD‐L1, programmed cell death 1 ligand 1; IFN, interferon; GB, granzyme B; ICI, immune checkpoint inhibitor. 4. DISCUSSION In the present study, ARID1A was found among a series of epigenetic modulators to be highly related to AIR in TNBC. We observed that low ARID1A expression inhibited CD8^+ T cell infiltration and activation and led to early progression in both clinical samples and xenograft models. PD‐L1 upregulation was found to promote AIR in TNBC with low ARID1A expression. While no direct interaction was observed between ARID1A and PD‐L1 promoter, ARID1A was demonstrated to directly bind to the promoter of NPM1, further activating PD‐L1 transcription. This ARID1A/NPM1/PD‐L1 axis was consolidated in both early and metastatic TNBC samples. Finally, we assessed the targetability of AIR in TNBC with low level of ARID1A by ICI treatment including Ate and pucotenlimab. Notably, ARID1A‐low TNBC demonstrated a better response to ICI treatment relative to ARID1A‐high counterparts. Therefore, we propose that although low ARID1A expression provided survival advantage for TNBC cells, it simultaneously left the Achilles' heel, which could be targeted with ICI. This phenomenon was similar to that of patients with human epidermal growth factor receptor 2 (HER2)‐positive breast cancer. Although characterized by poorer survival compared with HER2‐ counterparts, HER2+ patients could profoundly benefit from HER2‐targeted treatment and showed better survival [[226]54, [227]55, [228]56]. It is widely accepted that immune responses triggered by tumor antigens participate in tumor clearance. However, the execution and restimulation of this response are often impaired in cancer through AIR, further limiting effective anti‐tumor immunity and leading to continued tumor progression [[229]57]. PD‐L1 expression induced by IFN‐γ is the most classical mechanism for AIR induction [[230]58]. In the present study, we observed a negative correlation between ARID1A and PD‐L1 expression in primary and metastatic TNBC. This is consistent with findings in ovarian [[231]12], gastric [[232]59] and colorectal cancers [[233]60], but the underlying mechanism remained open for discussion. In gastric cancer, for instance, ARID1A loss promoted PD‐L1 through the phosphatidylinositol 3‐kinase (PI3K)/protein kinase B (AKT) pathway [[234]59]. Other regulatory mechanisms for TIME components have also been reported. In prostate cancer, ARID1A loss decreased infiltration of total T cells and IFN‐γ^+CD8^+ T cells in TIME through recruiting myeloid‐derived suppressor cells, leading to an inhibitory microenvironment and disease progression [[235]61]. In the present study, we also found low ARID1A expression in TNBC was associated with reduced CD8^+ T cell infiltration and inactivation. Our results further expanded the understanding of ARID1A‐mediated TIME regulation in TNBC and provided mechanistic evidence for new treatment strategies. As a well‐established chromatin remodeler, ARID1A exerted bilateral regulation on chromatin accessibility. Schick et al. [[236]62] characterized a panel of isogenic HAP1 cell lines with individual knockouts of 22 BAF subunits, including ARID1A. ATAC‐seq in ARID1A‐ko and negative control cells revealed that although the proportion of regions with reduced accessibility was approximately 4.5 times that of regions with increased accessibility, both tendencies existed. In addition, they also showed that gene transcriptomic changes was associated with altered chromatin accessibility [[237]62]. Similar cases have also been reported in colorectal cancer HCT116 cells where ARID1A‐ko led to increased accessibility at 1.9% sites and decreased accessibility at 6.8% sites (fold change: 3.6) compared with control cells [[238]63]. However, ARID1A deficiency in lung tumors significantly increased global ATAC peaks, indicating an overall elevation in chromatin accessibility [[239]64]. In the present study, ATAC‐seq showed that ARID1A‐ko in MDA‐MB‐231 cells significantly increased chromatin accessibility at 109,858 sites while 37,668 sites showed significantly decreased chromatin accessibility. Also, pathway enrichment analysis of ATAC‐seq showed that AIR‐related pathways were enriched in ARID1A‐ko cells compared with negative control. In the context of the above global variations, low ARID1A expression‐promoted NPM1 chromatin accessibility increase and subsequent PD‐L1 elevation played a pivotal role in AIR in TNBC. Still, further study is needed to explore why NPM1, in particular, increases chromatin accessibility upon ARID1A‐ko. NPM1 demonstrated a high mutation rate in hematological diseases, and high NPM1 expression has been reported in solid tumors such as gastric, thyroid and liver cancers, further regulating DNA replication and transcription [[240]65]. Specifically, NPM1 has been shown to directly bind to the G‐rich repetitive sequence TTAGGG in the PD‐L1 promoter to endogenously activate its transcription in TNBC [[241]66]. In the present study, we observed enrichment of the NPM1 neighborhood in ARID1A‐ko MDA‐MB‐231 cells relative to negative control. For GSEA, neighborhood sets are defined by expression neighborhoods centered on 380 cancer‐related genes [[242]43]. Eukaryotic chromosomes could be organized into territories spanning several megabases, and the physical interactions between different genomic segments resulted in chromosomal loops and bridges, contributing to the transcriptional silencing and activation of genes within the three‐dimensional context of the nuclear architecture [[243]67]. As a chromatin remodeler, therefore, ARID1A may alter gene expression by influencing the entire neighborhood. Our finding that the NPM1 neighborhood was enriched upon ARID1A‐ko further supported that ARID1A may directly bind to NPM1 to remodel this chromatin region. One of the major barriers of ICI treatment is that only a minority of advanced/metastatic TNBC patients demonstrated response [[244]5, [245]6, [246]7]. Therefore, it is of vital importance to identify those who are likely to benefit from ICI treatment. Our study pinpointed low ARID1A expression in TNBC as a novel biomarker for ICIs such as Ate and pucotenlimab from the epigenetic perspective. ARID1A deficiency‐induced AIR, as well as PD‐L1 expression elevation, could be targeted in both xenograft models and patients through Ate and pucotenlimab, respectively. Previous studies in ovarian [[247]12] and lung cancers [[248]68] also found ARID1A deficiency to be associated with better ICI treatment effects compared with ARID1A‐proficient tumors. Therefore, low ARID1A expression demonstrated the potential to be a universal marker. Although being an approved marker for ICI efficacy, PD‐L1 positivity still demonstrated limited predictive value due to heterogeneity [[249]69, [250]70]. Patients bearing a constitutive expression of PD‐L1 induced by genetic alterations, such as phosphatase and tensin homolog (PTEN) deletions, phosphatidylinositol‐4,5‐bisphosphate 3‐kinase catalytic subunit alpha (PIK3CA) and/or AKT mutations, epidermal growth factor receptor (EGFR) mutations and MYC overexpression, failed to respond to ICIs in the absence of immune infiltration [[251]71, [252]72, [253]73]. The Future‐C‐Plus trial showed that metastatic TNBC patients with CD8 and/or PD‐L1 positivity benefited more from the PD‐1 monoclonal antibody camrelizumab relative to CD8/PD‐L1 double negative patients [[254]74]. Based on the classical AIR theory, ICIs should be effective in type II AIR manifested by both PD‐L1‐positive tumor cells and PD‐1‐positive T cells, especially CD8^+ T cells [[255]57]. On the contrary, our results showed that TNBC with low ARID1A and high PD‐L1 expression demonstrated better response to anti‐PD‐1/PD‐L1 therapy Ate and pucotenlimab relative to ARID1A‐high TNBC in clinical trials and mouse models despite insufficient anti‐tumor immune infiltrates, especially CD8^+ T cells. The exciting finding implied that among TNBC with deficient CD8^+ T cell infiltration, previously not considered candidates for ICIs, there existed an immunological subgroup characterized by low ARID1A and high PD‐L1 expression that could benefit from ICIs. One limitation of our research is the failure to demonstrate reciprocal interactions between immune cells and metastatic TNBC cells harboring ARID1A deficiency. There is an ethical issue here, and it remained difficult to distinguish ARID1A‐related AIR from AIR induced by other causes. Nevertheless, further research on tumor‐immune coevolution in TNBC is ongoing to reverse low ARID1A expression‐promoted AIR. 5. CONCLUSIONS Our study identified low expression of the epigenetic modulator ARID1A in TNBC as a novel subtype of AIR through increasing NPM1 chromatin accessibility and subsequent PD‐L1 transcription. This immunological subgroup of TNBC defined by low ARID1A and further characterized by high PD‐L1 expression could potentially benefit from ICI. DECLARATIONS AUTHOR CONTRIBUTIONS ZHT, XCH and XYC outlined the manuscript. ZHT, XCH, JZ, BYW and ZMS contributed to the preparation of clinical data. ZHT, XYC, TN and SLH performed literature search. XYC, BL, TN, YY, LHH, YW, JJ and MDY contributed to data collection, analysis and interpretation. ZHT, XYC and BL provided the figures and drafted the manuscript, with additional input from all authors. ZHT, XCH, TN, SLH and XYC contributed to manuscript review and editing. All authors read and approved the final manuscript. CONFLICT OF INTEREST STATEMENT The authors declare no conflict of interest. ETHICS APPROVAL AND CONSENT TO PARTICIPATE All research protocols were approved by the medical ethics committee of Fudan University Shanghai Cancer Center (050432‐4‐2108*). Written informed consent was obtained from all patients. CONSENT FOR PUBLICATION Not applicable. Supporting information Supporting information [256]Click here for additional data file.^ (28.4KB, xlsx) Supporting information [257]Click here for additional data file.^ (12.9MB, docx) ACKNOWLEDGMENTS