Elevated CHAF1A Suppresses Type I Interferon Production and Causes Immunotherapy Resistance in Esophageal Squamous Cell Carcinoma

Resistance to immunotherapy is common in esophageal squamous cell carcinoma (ESCC). The authors demonstrate that CHAF1A suppresses tumor-intrinsic innate immunity in ESCC and that targeting CHAF1A with baimaside offers a strategy to overcome immunotherapy resistance.

Abstract

In esophageal squamous cell carcinoma (ESCC), chemoradiotherapy potentiates the effects of immune checkpoint inhibitors by activating the tumor-intrinsic innate immune response. However, ESCC cells frequently suppress this activation, which contributes to the high rates of immunotherapy resistance (70%–80%) observed clinically. Thus, identifying intracellular suppressors of this innate immune response remains an unmet critical need. Herein, through multiomics analyses, we identify the chromatin assembly factor CHAF1A as a suppressor of the tumor-intrinsic innate immune response in ESCC. We found that CHAF1A was overexpressed in ESCC and negatively correlated with type I interferon production and CD8⁺ T-cell infiltration. Mechanistically, CHAF1A maintained heterochromatin silencing mediated by H3K9me3, thereby repressing endogenous retroviruses. This suppression prevented the accumulation of double-stranded RNA and the subsequent activation of the MAVS–IRF3 signaling pathway. Concurrently, CHAF1A preserved genomic stability, limiting the release of double-stranded DNA and activation of the cGAS–STING pathway. Loss of CHAF1A potentiated the response to immunotherapy through the coordinated activation of these dual pathways. We then performed a small-molecule compound screen and identified a CHAF1A inhibitor, baimaside, which enhanced the effect of anti–PD-1 therapy to augment antitumor immunity. Collectively, these data indicate that CHAF1A represents a potential therapeutic target for sensitizing ESCC to immunotherapy and provide a potential combination strategy for reversing immunotherapy resistance.

Introduction

Combining immune checkpoint inhibitors with chemoradiotherapy has become the standard of care for advanced esophageal squamous cell carcinoma (ESCC), but the objective response rate remains suboptimal, and the majority of patients develop either primary or acquired resistance (1–3). Chemoradiotherapy induces tumor cell DNA damage, promotes micronuclei formation, and triggers the release of damage-associated molecular patterns. These events lead to the activation of core innate immune signaling pathways, including the cGAS–STING and RIG-I/MDA5–MAVS pathways, which drive a type I interferon (IFN-I) response, reshape the tumor immune microenvironment, and ultimately enhance the efficacy of immunotherapy (4–9). However, tumor cells can actively suppress the activation of these innate immune pathways through diverse epigenetic and transcriptional regulatory mechanisms. This suppression results in insufficient IFN-I production, impaired antigen presentation, and a paucity of T-cell infiltration and represents a key tumor-intrinsic mechanism of immunotherapy resistance in ESCC (10–12). Consequently, identifying critical negative regulators of the tumor-intrinsic innate immune response represents an urgent clinical need to improve therapeutic outcomes (13, 14).

Activating the tumor-intrinsic innate immune response has emerged as a pivotal strategy to convert immunologically “cold” tumors into “hot” ones and thereby overcome immunotherapy resistance (13, 14). A core mechanism underlying this response lies in the sensing of aberrant cytosolic nucleic acids by tumor cells. Endogenous double-stranded RNA (dsRNA), derived primarily from retrotransposons and other repetitive genetic elements, is recognized by RIG-I or MDA5, which in turn activates IFN-I production through the mitochondrial antiviral signaling (MAVS) axis (15, 16). In contrast, cytosolic double-stranded DNA (dsDNA), which frequently arises from genomic instability, DNA damage, or mitochondrial stress, is sensed by cGAS, leading to activation of STING-dependent signaling and the subsequent production of IFN-I (16–19). Recent studies have demonstrated that targeting specific epigenetic or transcriptional regulators can effectively activate these innate immune pathways by relieving the suppression of cytosolic dsRNA or dsDNA accumulation (15, 20–22). Moreover, emerging evidence indicates that interventions directed at distinct molecular checkpoints can cooperatively amplify the tumor-intrinsic IFN-I response, thus overcoming the immunosuppressive barriers that limit the efficacy of current immunotherapies (17, 23–29).

However, the key determinants that initiate the tumor-intrinsic innate immune response in ESCC remain largely unknown. Through multiomics analysis, we identified the chromatin assembly factor CHAF1A as a critical suppressor of this pathway. We found that CHAF1A was overexpressed in ESCC, and its expression inversely correlated with intratumoral IFN-I signaling and CD8⁺ T-cell infiltration. Mechanistically, CHAF1A promoted immune resistance through a dual shutdown mechanism. Inhibition of CHAF1A in vivo substantially remodeled the tumor immune microenvironment and enhanced the effect of anti–PD-1 therapy, establishing CHAF1A as a promising therapeutic target for overcoming immunotherapy resistance in ESCC.

Materials and Methods

Patient samples and data acquisition

For patient sample and data acquisition, human ESCC tumor samples were obtained from Harbin Medical University Cancer Hospital. Written informed consent was obtained from all patients, and the studies were conducted in accordance with recognized ethical guidelines (e.g., Declaration of Helsinki, Council for International Organizations of Medical Sciences, the Belmont Report). Two independent cohorts were enrolled between January 2022 and December 2024. Cohort 1 comprised 40 treatment-naïve patients undergoing surgery during which paired tumor and adjacent normal tissues were collected. Cohort 2 included 17 patients with advanced ESCC who had received immunotherapy, with postimmunotherapy tumor tissues obtained via esophagoscopy. The study was approved by the Ethics Committee of Harbin Medical University (Ethics No.: JS2024-40). Clinical samples were either rapidly frozen in liquid nitrogen and stored at −80°C or fixed in formalin and embedded in paraffin.

Cell lines

The AKR cell line was purchased from Xiamen Immocell Biotechnology in 2025. The KYSE-150 (RRID:CVCL_1348), KYSE-30 (RRID:CVCL_1351), HEK293T (RRID:CVCL_0063), TE-1 (RRID:CVCL_1759), KYSE-510 (RRID:CVCL_1354), KYSE-410 (RRID:CVCL_1352), KYSE-180 (RRID:CVCL_1349), EC109 (RRID:CVCL_6898), AGS (RRID:CVCL_0139), HT29 (RRID:CVCL_0320), BT-549 (RRID:CVCL_1092), and NCI-H1299 (RRID:CVCL_0060) cell lines were purchased from Procell in 2025. The mEC25 cell line (CVCL_C6JH) was purchased from Wenhuashidai in 2025. The EC9706 cell line (RRID:CVCL_E307) was purchased from Shanghai Jingyuan Biotechnology in 2025. All cell lines were obtained directly from the indicated commercial suppliers and were used within 6 months of receipt. Cell line authentication was performed by the suppliers via short tandem repeat profiling; no further authentication was carried out in our laboratory. Mycoplasma contamination was monitored regularly throughout the study using PCR-based detection, with the latest negative screening completed before all in vitro experiments. The biological sex of each cell line is as follows: KYSE-150, female; KYSE-30, male; HEK293T, female; TE-1, male; KYSE-510, male; KYSE-410, male; KYSE-180, male; EC109, female; EC9706, female; AKR, unknown; mEC25, female; AGS, female; HT29, female; BT-549, female; and NCI-H1299, male. Cells were cultured in DMEM (Gibco, #11965092) or RPMI 1640 (Gibco, #C11875500BT) supplemented with 10% fetal bovine serum (Gibco, #A5669801) and 1% penicillin–streptomycin (Gibco, #15140122). Cultures were maintained at 37°C in a humidified atmosphere containing 5% CO₂. Cells were used between passages 5 and 15 and were passaged every 2 to 3 days to prevent overgrowth.

Lentivirus infection and CRISPR-Cas9 gene editing

To establish CHAF1A-knockdown cells, a lentiviral vector encoding a pool of two distinct short hairpin RNAs (shRNA) targeting CHAF1A was used (see Supplementary Table S1 for shRNA sequences). The target gene sequences of CHAF1A were retrieved from NCBI GenBank and subsequently cloned into the pLent-U6-shRNA-CMV-GFP-P2A-Puro vector (Synbio Technologies). The vector was packaged into lentiviral particles in HEK293T cells. The viral supernatant was then collected and used to infect mouse ESCC cells for 6 hours in the presence of jetPRIME transfection reagent (Polyplus, # 101000046). CHAF1A-knockdown mouse ESCC cells were selected using medium containing puromycin (Sigma, #P9620-10ML).

To generate CHAF1A-knockout (KO) cells, we employed a CRISPR/Cas9 system using plasmids provided by Synbio Technologies. Briefly, the CHAF1A-KO plasmids were designed based on predicted sequences obtained from the Eukaryotic Pathogen CRISPR guide RNA/DNA Design Tool (EuPaGDT) and inserted into the lentiCRISPR v2-GFP vector (Addgene, #52961). A total of 10 μg of CHAF1A-sgRNA plasmids was transfected into KYSE150 or KYSE30 cells plated on a 100 mm dish using 500 μL of jetPRIME buffer and 20 μL of jetPRIME transfection reagent (Polyplus, #101000046). Forty-eight hours later, the single-cell suspension was plated into 96-well plates for clonal expansion. When the clones reached confluence, they were passaged and expanded. Transfected human ESCC cells were sorted by fluorescence-activated cell sorting based on GFP fluorescence. Cell clones were enriched in 96-well plates. Reverse transcription quantitative PCR (qRT-PCR) and Western blot analyses were used to validate CHAF1A expression 14 to 21 days later. The sequences of the CHAF1A sgRNAs are listed in Supplementary Table S1.

For transient knockdown of MAVS and cGAS, small interfering RNAs (siRNA) targeting human MAVS or cGAS were chemically synthesized and purchased from Synbio Technologies and transfected into human ESCC cells using Lipofectamine 3000 (Invitrogen) according to the manufacturer’s instructions. Briefly, siRNA mediates sequence-specific degradation of complementary target mRNA via the RNA interference pathway, thereby silencing target gene expression at the posttranscriptional level. All sequences of siRNAs are listed in Supplementary Table S1.

Animal models

We purchased 6-week-old female C57BL/6 and BALB/c nude mice from Changsheng Biotechnology. All animal experiments were approved by the Institutional Animal Care and Use Committee of Harbin Medical University (Approval No. KY2022-37).

Subcutaneous tumor models

AKR or mEC25 cells (5 × 10⁵) in 100 μL PBS were subcutaneously injected into the right flank of the mice. Mice were divided into three groups: wild-type (WT), shCHAF1A#1, and shCHAF1A#2, with six mice randomly assigned to each group. Tumor size was continuously monitored every other day using calipers. Tumor volume was calculated using the following formula: V = [length × (width)²]/2. Mice were euthanized on days 21 to 24.

Anti–PD-1 therapy

Mice were divided into four groups: WT + isotype IgG, WT + anti–PD-1, Sh + isotype IgG, and Sh + anti–PD-1. Each mouse was subcutaneously injected with 5 × 10⁵ cells. Starting from day 7, anti-mouse PD-1 (CD279, Selleck Chemicals, #A2122) treatment or rat IgG2a isotype control (Selleck Chemicals, #A2123) was administered every other day (0.2 mg per mouse, twice weekly, intraperitoneal injection). Mice were euthanized on days 21 to 24.

CD8⁺ T-cell depletion model

Mice were randomly divided into four groups: WT combined with isotype IgG (rat IgG2a isotype control, Selleck Chemicals, #A2123), WT combined with anti-CD8 (anti-mouse CD8α, Selleck Chemicals, #A2102), CHAF1A knockdown combined with isotype IgG, and CHAF1A knockdown combined with anti-CD8. CD8⁺ T-cell depletion was performed as described previously (30). On day −1 (1 day before tumor cell inoculation), each mouse received an intraperitoneal injection of 200 μg neutralizing anti-mouse CD8 or an equivalent dose of matched isotype control IgG. On day 0, all mice were subcutaneously injected with 5 × 10⁵ ESCC tumor cells. A second dose of 200 μg of the corresponding antibody was intraperitoneally administered on day 1 after tumor inoculation, and no subsequent supplementary injection was performed. All mice were humanely euthanized on day 21. Subcutaneous tumors were isolated, photographed, and measured to evaluate tumor growth characteristics.

Combination therapy with baimaside and anti–PD-1

AKR or mEC25 cells (5 × 10⁵) in 100 μL PBS were subcutaneously injected into the right flank of the mice. When tumor volumes reached approximately 100 mm³, mice were randomly assigned to the following treatment groups (n = 6 per group): control group: i.p. injection of PBS; anti–PD-1 group: i.p. injection of anti-mouse PD-1 (200 µg, every 3 days); baimaside group: oral gavage of baimaside, which is a previously reported compound (31) identified in this study as a CHAF1A-specific inhibitor, at 50 mg/kg daily (MedChemExpress, # HY-N2183); and combination group: administration of anti-mouse PD-1 and baimaside at the same doses and schedules as the monotherapy groups.

Virtual screen to identify CHAF1A inhibitors

To identify potential CHAF1A inhibitors, we conducted a structure-based virtual screen using Schrödinger software. The three-dimensional structure of human CHAF1A was obtained from the AlphaFold database (AF-Q13111-F1) and processed using the Protein Preparation Wizard module for hydrogen addition and energy optimization. Compound libraries, namely the MCE Bioactive Compound Library (HY-L001) and the lead-like Diverse Library Plus (HY-L901P), totaling 100,000 compounds, were prepared through the LigPrep module for energy optimization and 3D structure generation. Subsequently, a tiered molecular docking screen was performed using the Glide module: All compounds were initially screened in high-throughput virtual screening mode, with the top 30% of scored compounds from the HY-L001 library and the top 15% from the HY-L901P library advancing to standard precision docking. From these, the top 30% and top 15% of compounds, respectively, were further subjected to extra precision (XP) docking. Finally, candidate molecules were selected based on their docking scores for subsequent experimental validation.

Molecular dynamics simulation of the CHAF1A–baimaside complex

Molecular dynamics (MD) simulations were performed to evaluate the binding stability and dynamic conformational behavior of the human CHAF1A–baimaside complex. The protein structure was preprocessed by adding missing hydrogen atoms and correcting protonation states. Baimaside was parameterized using the Antechamber program in the AMBER toolkit, with atomic charges calculated using the AM1-BCC method. A mol2 file containing ligand bonding information was generated, and missing force-field parameters were supplemented using parmchk2 to generate the corresponding frcmod file. The General Amber Force Field (GAFF) was loaded in tleap, and the ligand topology and coordinate files were integrated with the protein structure to construct the initial CHAF1A–baimaside complex. The AMBER ff99SB force field was applied to the protein and GAFF to baimaside. The complex was solvated in a cubic box of OPC water, and K⁺ and Cl⁻ ions were added to neutralize the system and achieve a salt concentration of 0.15 mol/L. Following energy minimization, the system was heated under the NVT ensemble from 0 to 100 K for 100 picoseconds and subsequently from 100 to 300 K for 500 picoseconds. The system was then equilibrated for 1 nanosecond under the isobaric-isothermal (i.e., the NPT) ensemble at 300 K and 1 bar, followed by a 100-nanosecond unrestrained production MD simulation with a time step of 2 femtoseconds. The root-mean-square deviations (RMSD) of the protein backbone and baimaside were calculated throughout the trajectory to assess structural stability. In addition, the distance between the O13 atom of baimaside and an oxygen atom of Arg6 (R6) in CHAF1A was monitored to assess the stability of this interaction. Trajectory visualization and analysis of dynamic conformational behavior and interaction patterns were performed using PyMOL and VMD.

Immunohistochemical staining

Human ESCC tissue sections were obtained from the two patient cohorts described above at Harbin Medical University Cancer Hospital (Ethics No.: JS2024-40). For immunohistochemical (IHC) analysis, tumor-bearing mouse dorsal skin samples were fixed in formaldehyde and embedded in paraffin. Subsequently, the sections were deparaffinized, rehydrated, and subjected to antigen retrieval. After blocking with hydrogen peroxide, the sections were incubated with 5% BSA. Primary antibodies targeting the proteins of interest were applied and incubated overnight at 4°C. The following antibodies were used: anti-CHAF1A (1:300), anti–γ-H2AX (1:500), anti-H3K9me3 (1:200), anti-IFNA (1:200), anti-IFNB (1:200), and anti-CD8 (1:500; Supplementary Table S2). The following day, the sections were thoroughly washed with PBS and incubated with horseradish peroxidase (HRP)–conjugated secondary antibodies (goat anti-rabbit IgG polymer, ZSGB-BIO, #PV-6001; goat anti-mouse IgG polymer, ZSGB-BIO, #PV-6002) at room temperature for 1 hour. Diaminobenzidine (ZSGB-BIO, #ZLI-9019) was used for color development, with the reaction time controlled to ensure optimal staining. The sections were then lightly counterstained with hematoxylin, dehydrated in graded ethanol, cleared in xylene, and mounted with neutral gum. The slides were examined under a Zeiss Axioplan 2 microscope, and images were captured using a Zeiss Axiocam digital camera. Quantitative analysis of the positive signal area and intensity was performed using ImageJ software (version 2.14.0).

Western blotting

We performed Western blot analysis to detect the expression levels of specific proteins. Briefly, total proteins were extracted from tissue samples using RIPA lysis buffer (Thermo Fisher Scientific, #89901) supplemented with protease inhibitors (Beyotime Biotechnology, #P1005). The protein concentration was determined using the BCA assay kit (Beyotime Biotechnology, #P0009). Equal amounts of protein were separated by SDS-PAGE on 10% polyacrylamide gels and then transferred to polyvinylidene difluoride (PVDF) membranes. The membranes were blocked with 5% nonfat milk in Tris-buffered saline with Tween 20 (TBST) for 1 hour at room temperature. Primary antibodies against target proteins were applied and incubated overnight at 4°C. The following antibodies were used for Western blotting: anti–β-Tubulin (1:3,000), anti-GAPDH (1:3,000), anti-CHAF1A (1:1,000), anti-MAVS (1:1,000), anti–P-IRF3 (1:1,000), anti-IRF3 (1:2,000), anti–P-TBK1 (1:1,000), anti-TBK1 (1:1,000), anti-cGAS (1:1,000), anti–P-STING (1:1,000), anti-STING (1:1,000), anti–γ-H2AX (1:1,000), anti-H3K9me3 (1:1,000), anti-H3K27ac (1:1,000), anti-H3K4me3 (1:1,000), anti-H3K27me3 (1:1,000), anti-H3 (1:2,000), J2 (1:1,000), anti-Chk1 (1:1,000), anti–P-Chk1 (1:1,000), anti-RPA32 (1:1,000), anti–P-RPA32 (1:1,000), anti-HP1α (1:1,000), anti-KAP1 (1:1,000), anti-SETDB1 (1:1,000), anti-MDA5 (1:1,000), anti-SUV39H1 (1:1,000), and anti–Rig-I (1:1,000; Supplementary Table S2). The following day, the membranes were washed with TBST and incubated with HRP-conjugated secondary antibodies (goat anti-mouse IgG HRP, Abmart, #M21004S; goat anti-rabbit IgG HRP, Abmart, #M21007S) for 1 hour at room temperature. The protein bands were visualized using an enhanced chemiluminescence detection system (Beyotime Biotechnology, #P0018S). The density of the bands was quantified using ImageJ software (version 2.14.0).

Immunofluorescence

Cells were seeded on slides and cultured overnight, then fixed with 4% paraformaldehyde and permeabilized with 0.3% Triton X-100. After blocking, the cells were incubated with primary antibodies at 4°C overnight, followed by staining with fluorescence-conjugated secondary antibodies for 1 hour. The antibodies used for immunofluorescence (IF) staining were as follows: anti-cGAS (1:150), anti–γ-H2AX (1:150), anti-J2 (1:150), and anti-MAVS (1:150; Supplementary Table S2). Nuclei were counterstained with Antifade Mounting Medium containing 4′,6-diamidino-2-phenylindole (DAPI; Beyotime Biotechnology, #P0131). Images were acquired using a ZEISS LSM 900 confocal microscope and processed with ZEISS software (version 3.6).

RNase III and cGAS inhibitor treatment

To degrade endogenous cytoplasmic dsRNA, recombinant RNase III (Beyotime Biotechnology, #R7086M) was delivered into ESCC cells using the Protein Transfection Reagent (MedChemExpress, # HY-K2016) according to the manufacturer's instructions. Briefly, 10 μg RNase III was mixed with 8 μL Protein Transfection Reagent as per the manufacturer’s six-well protocol, incubated for 3 minutes at room temperature, diluted in 2 mL serum-free medium, and added to cells. After incubation at 37°C for 3 hours, the transfection medium was replaced with complete medium, and cells were cultured for an additional 21 hours before protein extraction for Western blotting and qRT-PCR. For IF, cells were fixed, permeabilized with Triton X-100, and incubated with 20 U/mL RNase III at 37°C for 30 minutes before J2 immunostaining to verify the specificity of dsRNA staining. For cGAS inhibition experiments, cells were treated with RU.521 (MedChemExpress, # HY-114180) at 10 μmol/L for 24 hours before downstream analyses.

Semi-denaturing detergent agarose gel electrophoresis

MAVS oligomerization was assessed by semi-denaturing detergent agarose gel electrophoresis. Cells were harvested and lysed in ice-cold, nonreducing lysis buffer (1% Triton X-100) to preserve protein polymers (32). Cell lysates were treated with 2% nonreducing loading buffer (2% SDS) at room temperature and separated on a 1.5% agarose gel supplemented with 0.1% SDS at 40 to 50 V in an ice bath. The proteins were then transferred onto a PVDF membrane by capillary transfer, followed by methanol fixation and immunoblotting with anti-MAVS antibody (1:150; Supplementary Table S2).

Flow cytometry

ESCC cells were isolated from subcutaneous tumors in mice. After euthanizing the mice, the excised subcutaneous xenografts were digested with collagenase IV (Gibco, #17104019) for 1 hour, filtered through a 45 µm mesh sieve, lysed of red blood cells, and then transferred into 1.5 mL EP tubes. Brefeldin A (eBioscience, #00-4506-51) was added at a concentration of 1,000× to block cytokine secretion, and the samples were further incubated for 4 hours. After induction for 4 hours, the cells were stained with near-infrared (NIR; BioLegend, #423105) for 30 minutes, followed by washing with PBS and centrifugation. Next, surface staining was performed with anti-CD3 (BioLegend, #100203), anti-CD45 (BioLegend, #157207), and anti-CD8 (BioLegend, #100711) for 30 minutes, after which the cells were washed again with PBS. Finally, after surface staining, the cells were fixed with 4% paraformaldehyde for 30 minutes and permeabilized for 40 minutes, followed by staining with intracellular anti-IFNγ (BioLegend, #631053). The cells were then filtered through a 45 µm mesh sieve before data acquisition on a BD FACSMelody (BD Biosciences). Data were analyzed using FlowJo software (version 10.10.0).

ELISA

Cell lysates from cultured human ESCC cell lines and cultured mouse ESCC cell lines and subcutaneous tumor tissues collected from tumor-bearing mice were analyzed by ELISA. For cultured cells, one 6 cm dish of logarithmically growing cells was washed with PBS and detached using a cell scraper. The cells were resuspended in 1 mL PBS and transferred to a 1.5 mL microcentrifuge tube. After centrifugation at 12,000 rpm for 5 minutes, the cell pellets were lysed in PBS supplemented with a protease inhibitor cocktail (MedChemExpress, #HY-17541A) on ice for 30 minutes. The lysates were then centrifuged at 12,000 rpm for 10 minutes, and the supernatants were collected for ELISA analysis. For mouse tumor samples, subcutaneous tumor tissues were homogenized in PBS supplemented with a protease inhibitor cocktail (MedChemExpress, # HY-17541A) using a tissue grinder. The homogenates were centrifuged at 12,000 rpm for 30 minutes, and the supernatants were collected into new 1.5 mL microcentrifuge tubes. ELISA assays were performed using species-specific human or mouse IFNA and IFNB ELISA kits (Meimian; Supplementary Table S2) according to the manufacturer’s instructions. Absorbance was measured at 450 nm with a Spark multimode microplate reader (Tecan).

Cell Counting Kit-8 assay

Cell viability was assessed using the Cell Counting Kit-8 (CCK-8; GlpBio, # GK10001). Cells were seeded into 96-well plates at a density of 1,500 cells per well in 100 μL of culture medium. For comparison of genetically modified cells with their corresponding control cells, CCK-8 measurements were performed at 0, 24, 48, 72, and 96 hours after seeding. For drug treatment experiments, cells were treated with DMSO or baimaside at 50 or 100 μmol/L, and CCK-8 measurements were performed at 0, 24, 48, and 72 hours. At each indicated time point, 10 μL of CCK-8 reagent was added to each well, followed by incubation at 37°C for 1 hour. Blank wells containing culture medium and CCK-8 reagent without cells were included to correct for background absorbance. Before measurement, the plates were gently shaken on an orbital shaker for 10 minutes to ensure uniform color development. Absorbance at 450 nm was measured using a Spark multimode microplate reader (Tecan), and the blank absorbance value was subtracted from the corresponding sample values. Each experimental condition was assessed in three replicate wells.

Comet assay

To evaluate DNA damage, we used a comet assay kit (Beyotime Biotechnology, #C2041M). To initiate the experiment, samples were prepared as follows: For adherent cells, the supernatant was removed, and the cells were digested with trypsin (Gibco, #25300054), washed, and resuspended to a density of 1 × 10⁶ cells/mL. For slide preparation, a two-layer agarose gel system was employed. The first layer consisted of 1% normal melting point agarose, which was spread on a slide and allowed to solidify at 4°C. The second layer was prepared by mixing the cell suspension with 0.7% low melting point agarose and overlaying it onto the first layer, followed by solidification at 4°C. Subsequently, the samples were subjected to lysis using a solution composed of Lysis Buffer and DMSO in a 9:1 ratio, incubated at 4°C for 1 to 2 hours, and then washed with PBS. On the following day, DNA unwinding was performed by placing the slides in an electrophoresis tank and allowing them to stand at room temperature for 20 to 60 minutes. Electrophoresis was conducted at a low voltage of 25 V for 20 to 30 minutes under low-temperature and light-protected conditions. After electrophoresis, the slides were neutralized with neutral buffer at 4°C for 1 to 3 cycles, each lasting 5 to 10 minutes, and then stained with propidium iodide for 10 to 20 minutes in the dark. Finally, the slides were washed with ultrapure water and mounted with a coverslip. Analysis was conducted using a fluorescence microscope, with at least 80 cells selected per sample and tail moment scores calculated using CaspLab (version 1.2.3_b2).

qRT-PCR

Total RNA was isolated from cells using TRIzon Pal (Cwbio, #CW3166M) and TRIzon Reagent (Cwbio, #CW0580S) according to the manufacturer’s instructions. A total of 1 μg of purified RNA was reverse transcribed into cDNA using the PrimeScript Reverse Transcriptase kit (Takara, #D2680A) on a T100 Thermal Cycler (Bio-Rad, #621BR65372). qRT-PCR was performed using the Talent qPCR PreMix kit (TIANGEN, #KR123) on an ABI Real-Time PCR System (Thermo Fisher Scientific, #7500). Each 10 μL reaction contained 50 ng of cDNA template. The amplification protocol consisted of an initial incubation at 50°C for 2 minutes and 95°C for 3 minutes, followed by 40 cycles of 95°C for 5 seconds and 60°C for 32 seconds. A melt curve analysis was performed after amplification to verify the specificity of the PCR products. The primers used for qRT-PCR are listed in Supplementary Table S3. Relative gene expression levels were calculated using the 2^−ΔΔCt method with human ACTB or mouse Actb as the internal reference gene. Each sample was analyzed in triplicate.

Chromatin immunoprecipitation and chromatin immunoprecipitation–qPCR

Cells were cross-linked with 1% formaldehyde for 10 minutes and quenched with 1.25 mL of 1 mol/L glycine. After washing with PBS, chromatin was sonicated in RIPA lysis buffer (Thermo Fisher Scientific, #89901). For each immunoprecipitation, 8 μg of anti-H3K9me3 antibody (rabbit monoclonal, PTM BIO, # PTM-616RM) or normal rabbit IgG (Abcam, # ab171870) as a negative control was added to 1 mL of chromatin lysate and incubated overnight at 4°C. Antibody–chromatin complexes were captured using pretreated Protein A/G beads (Thermo Fisher Scientific, #20423) and washed sequentially with low-salt, high-salt, LiCl, and tris‑EDTA buffer (TE) buffers. Complexes were eluted with a solution containing 1% SDS. Both eluates and input samples were de-cross-linked at 65°C overnight. Following treatment with RNase A (Thermo Fisher Scientific, #EN0531) and proteinase K (Thermo Fisher Scientific, #EO0491), DNA was recovered using phenol–chloroform extraction and isopropanol precipitation. The enrichment of specific DNA fragments was analyzed by qPCR. The primers used for chromatin immunoprecipitation (ChIP)-qPCR are listed in Supplementary Table S3.

RNA sequencing

Bulk RNA sequencing (RNA-seq) was performed by Novogene. Total RNA was extracted from KYSE150 wild-type (KYSE150-WT) and CHAF1A-KO (KYSE150-KO) cells using TRNzol Universal Reagent (TIANGEN, # DP424) according to the manufacturer’s instructions. Three independent biological replicates were analyzed for each group. RNA integrity and quantity were assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies). Sequencing libraries were prepared using the Fast RNA-seq Lib Prep Kit version 2 (ABclonal, # RK20306). Briefly, mRNA was enriched from total RNA using oligo (dT) magnetic beads and fragmented. First-strand cDNA was synthesized using random primers, followed by second-strand cDNA synthesis, end repair, dA-tailing, adapter ligation, size selection, library amplification, and purification. The adapter sequences were as follows: P5 adapter, 5′-AATGATACGGCGACCACCGAGATCTACAC[i5]ACACTCTTTCCCTACACGACGCTCTTCCGATCT-3′; and P7 adapter, 5′-GATCGGAAGAGCACACGTCTGAACTCCAGTCAC[i7]ATCTCGTATGCCGTCTTCTGCTTGCAAGC-3′. Libraries were quantified using a Qubit 2.0 Fluorometer, assessed for fragment size distribution using an Agilent 2100 Bioanalyzer, and further quantified by RT-qPCR. Qualified libraries were sequenced on an Illumina NovaSeq X Plus platform (Illumina) using a NovaSeq X Series 25B Reagent Kit (300 cycles; # 20104706) with a paired-end 150-bp (PE150) sequencing strategy. Raw reads in FASTQ format were processed using fastp to remove adapter-containing reads, reads containing undetermined bases (N), and low-quality reads in which bases with Q_Phred ≤5 accounted for ≥50% of the read length. Q20, Q30, GC content, and overall sequencing error rates were calculated for the clean reads. Sequencing quality criteria included an overall error rate <1%, Q20 > 90%, and Q30 > 85%. All downstream analyses were performed using the resulting high-quality clean reads. Paired-end clean reads were aligned to the human reference genome (Homo sapiens GRCh38/hg38) using HISAT2 version 2.0.5. Gene-level read counts were generated using featureCounts version 1.5.0-p3, and gene expression levels were calculated as fragments per kilobase of transcript per million mapped reads. Differential expression analysis between KYSE150-WT and KYSE150-KO cells was performed using the DESeq2 R package version 1.20.0. P values were adjusted using the Benjamini–Hochberg method to control the false discovery rate (FDR). Genes with an adjusted P value ≤0.05 and |log₂(fold change)| ≥ 1 were considered differentially expressed.

Assay for transposase-accessible chromatin with high-throughput sequencing

Assay for transposase-accessible chromatin sequencing (ATAC-seq) was performed by Novogene using KYSE150-WT and CHAF1A-KO (KYSE150-KO) cells. One sample per group was analyzed, with no biological or technical replicates. Briefly, 5 × 10⁴ viable cells per sample were collected and centrifuged at 500 × g for 5 minutes at 4°C. Cell pellets were resuspended in 50 μL lysis buffer and incubated on ice for 5 minutes. After the addition of 1 mL wash buffer, nuclei were collected by centrifugation at 500 × g for 5 minutes. The nuclear pellet was resuspended in a Tn5 transposase reaction mixture, and tagmentation was performed at 37°C for 30 minutes. Equimolar adapter 1 and adapter 2 were added following transposition, and libraries were amplified by PCR using the ATAC-seq_Novogene Kit. PCR-amplified libraries were purified using AMPure beads, and library quality was assessed using a Qubit fluorometer. Index-coded libraries were clustered using a cBot Cluster Generation System with the TruSeq PE Cluster Kit version 3-cBot-HS (Illumina) according to the manufacturer’s instructions and sequenced on an Illumina NovaSeq platform to generate 150-bp paired-end reads. Raw sequencing reads in FASTQ format were processed using fastp version 0.20.0. Adapter-containing reads, reads containing poly-N sequences, and low-quality reads in which bases with quality scores <15 accounted for >40% of the read length were removed; trimmed reads shorter than 18 bp were also excluded. Q20, Q30, and GC content were calculated for the resulting clean reads. Clean reads were aligned to the human reference genome (GRCh38/hg38) using BWA-MEM version 0.7.12. Mitochondrial reads, improperly paired reads, PCR duplicates, and reads with mapping quality scores <13 were removed. Only uniquely mapped, properly paired, and deduplicated reads with Mapping Quality Score (MAPQ) ≥13 were retained for downstream analysis. Accessible chromatin peaks were identified using MACS2 version 2.1.0 with the parameters -q 0.05 –call-summits –nomodel –shift -100 –extsize 200 –keep-dup all and peaks with q values ≤0.05 were retained. Peaks from the two groups were merged using bedtools merge, and reads per million mapped reads were calculated for the merged peak regions. Regions with |log₂(fold change)| > 1 between KYSE150-WT and KYSE150-KO cells were considered differentially accessible. Peak-associated genes and genomic regions were annotated using ChIPseeker. ATAC-seq profiles were visualized using the Integrative Genomics Viewer (IGV, version 2.19.5).

Cleavage under targets and tagmentation

Cleavage under targets and tagmentation (CUT&Tag) libraries were prepared by Kaitaibio using the Hyperactive Universal CUT&Tag Assay Kit for Illumina Pro (Vazyme, # TD904). Briefly, intact nuclei were isolated from approximately 5 × 10⁴ cells using a nonionic detergent-based lysis buffer and immobilized on ConA-coated magnetic beads. Nuclei were incubated overnight at 4°C with an anti-H3K9me3 antibody (PTM BIO, # PTM-616RM), followed by incubation with a secondary antibody and pA/G-Tn5 transposase preloaded with sequencing adapters. The pA/G-Tn5–bound nuclei were then subjected to tagmentation at 37°C for 60 minutes, allowing simultaneous DNA fragmentation and adapter insertion. DNA fragments were purified using magnetic beads and amplified by limited-cycle PCR (9–20 cycles depending on input amount). Libraries were purified, quantified using a Qubit dsDNA HS Assay, and assessed for fragment size distribution before sequencing. Libraries with fragment sizes of approximately 100 to 1,000 bp were sequenced on an Illumina NovaSeq 6000 platform using paired-end 150-bp sequencing, with a target depth of 10 to 20 million paired-end reads per sample. Raw sequencing reads were processed using fastp (version 0.23.4) to remove adapter-contaminated reads, low-quality reads (>30% of bases with a Phred score <15), and reads containing >5% ambiguous bases. Clean reads were aligned to the H. sapiens reference genome (GRCh38) using Bowtie2 (version 2.5.3). Reads with mapping quality (MAPQ <30), mitochondrial reads, and PCR duplicates identified using Picard MarkDuplicates (version 3.1.1) were removed before downstream analyses. Peak calling was performed using MACS3 (version 3.0.1). As CUT&Tag libraries were generated from one biological sample per group, differential peak analysis was performed using the pseudoreplicate strategy following the ENCODE guidelines for CUT&Tag datasets without biological replicates, and differential peaks were identified using DiffBind (version 3.12.0) together with edgeR (version 4.0.16). Differential peaks were subsequently annotated using ChIPseeker, functional enrichment analyses were performed using clusterProfiler, and genome browser visualization was carried out using IGV (version 2.19.4).

Gene set enrichment analysis

Gene set enrichment analysis (GSEA) was performed using the clusterProfiler package in R (version 4.4.2). Differentially expressed genes identified by DESeq2 were ranked based on adjusted P value (P_adj), and log₂ fold changes were used to generate a preranked gene list. The Gene Ontology gene sets (c5.all.version 7.0.symbols.gmt) from the MSigDB database (version 7.0) were employed as the reference. A random seed was set (set.seed = 1) to ensure reproducibility. Results were visualized using ggplot2, with normalized enrichment scores (NES) displayed in directional bar plots indicating upregulated and downregulated gene sets. The significance threshold was set at FDR <0.25 (default GSEA standard).

Public dataset acquisition and bioinformatics analysis

RNA-seq raw count data and transcripts per million–normalized expression data for The Cancer Genome Atlas (TCGA) esophageal carcinoma cohort (n = 183) were obtained from the Genomic Data Commons using the R package TCGAbiolinks, and the corresponding clinical annotations were obtained from UCSC Xena. Gene expression matrices and sample annotation data for GSE70409 (n = 34), GSE77861 (n = 14), GSE17351 (n = 10), GSE47404 (n = 71), GSE53624 (n = 283), and GSE53625 (n = 358) were obtained from the Gene Expression Omnibus (GEO) using the R package GEOquery. GSE53624 and GSE53625, which contained both gene expression and survival data, were included in the survival analyses. Differential expression analysis was performed using DESeq2 for TCGA RNA-seq count data and limma for GEO expression data. Genes with an absolute log₂ fold change >1 and an FDR < 0.05 were considered differentially expressed. Based on predefined immune-related gene sets, immune activation and IFN production scores were calculated for each sample using gene set variation analysis (GSVA).

Statistical analysis

Statistical analyses were performed utilizing GraphPad Prism software (version 10.1.1) and R software (version 4.4.2). Quantitative data are presented as mean ± standard error of the mean. Fisher exact test was used for comparing categorical data, one-way analysis of variance (ANOVA) for comparisons among multiple groups and Student t test for comparing means between two groups. Pearson correlation analysis was performed to assess the relationship between two variables. To evaluate the association between epigenetic regulator expression and immune microenvironment activity, we first derived an immunome activation score for each sample using the GSVA algorithm based on a predefined set of immune-related genes. Spearman rank correlation analysis was then performed between the expression of each epigenetic regulator and the immunome activation score in tumor samples. Overall survival (OS) curves were generated using the Kaplan–Meier method, and the log-rank test was used for comparison. Differential expression analysis was performed using the limma algorithm with an FDR threshold of <0.05. Statistical significance is denoted as *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Results

CHAF1A impairs IFN-I production in ESCC

We identified candidate genes closely linked to IFN production in ESCC development. Integrating transcriptomic data from TCGA and GEO databases (including GSE70409, GSE77861, GSE17351, and GSE47404), we performed a multicohort integrative analysis (Fig. 1A). In each independent cohort, we screened for genes meeting at least two criteria simultaneously: overexpression in ESCC tissues, negative correlation with innate immune signatures (e.g., immune activation score and IFN production), and association with poor patient prognosis. Employing this approach, CHAF1A recurred consistently across multiple cohorts (TCGA, GSE70409, and GSE77861; Supplementary Fig. S1A–S1E). In TCGA data, CHAF1A expression was elevated in ESCC (Fig. 1B). To validate TCGA findings, we analyzed paired tumor and adjacent normal esophageal tissues from a cohort of 40 treatment-naïve patients with ESCC who underwent surgery at Harbin Medical University Cancer Hospital between January 2022 and December 2024. qRT-PCR analysis of 20 paired samples confirmed significantly increased CHAF1A mRNA expression in ESCC tissues (Fig. 1C). Western blot analysis of an independent set of six randomly selected paired samples from the same cohort further confirmed elevated CHAF1A protein expression in ESCC tissues (Fig. 1D). IHC analysis of all 40 paired samples further revealed strong nuclear CHAF1A staining in ESCC tissues compared with adjacent normal tissues (Fig. 1E). Across ESCC cell lines, CHAF1A protein levels inversely correlated with IFNB mRNA abundance (Fig. 1F and G). In parallel, high CHAF1A corresponded to reduced IFNA/B in IHC analysis of paired ESCC samples (Fig. 1H and I). Validating the role of CHAF1A in intrinsic IFN production, CHAF1A loss boosted IFNA/B secretion by KYSE150 and AKR cells (Fig. 1J). Furthermore, elevated CHAF1A expression predicted unfavorable OS in ESCC (GSE53624 and GSE53625), with a consistent clinical trend observed in esophageal cancer and esophageal adenocarcinoma (Supplementary Fig. S1F–S1I). In an independent cohort of 17 patients with advanced ESCC who received immune checkpoint blockade (ICB), CHAF1A expression in posttreatment tumor tissues obtained by esophagoscopy was significantly higher in nonresponders than in responders (P = 0.015; Supplementary Fig. S1J). Consistently, IHC analysis of samples from this cohort revealed that CHAF1A-low tumors exhibited stronger PD-L1 intensity and higher CD8⁺ T-cell infiltration (Supplementary Fig. S1K). These results revealed that CHAF1A overexpression in ESCC suppresses intrinsic IFN-I production, thereby promoting ESCC progression.

Figure 1.

CHAF1A reduces intrinsic IFN production in ESCC tumor cells. A, Schematic workflow of the multicohort integrative screening strategy combining TCGA and GEO datasets (GSE70409, GSE77861, GSE17351, and GSE47404) to identify genes upregulated in tumors but negatively correlated with innate immune signaling. CHAF1A, together with DNMT3B, SMYD5, TET1, and TRIM28, was identified as a top candidate suppressing the cGAS–STING–IRF3–IFN axis. B, CHAF1A mRNA expression levels in esophageal cancer (ESCA) and normal esophageal tissues analyzed using the TCGA dataset. C, qRT-PCR analysis of CHAF1A mRNA expression in 20 paired ESCC tumor and adjacent normal esophageal tissues (n = 20 pairs). D, Immunoblotting analysis of CHAF1A protein expression in six paired (P) ESCC and adjacent normal tissue samples. E, Representative IHC staining of CHAF1A in human ESCC and adjacent normal tissues and quantitative comparison of IHC scores (n = 40). Scale bars, 60 μm. F, Immunoblotting analysis of CHAF1A protein expression across the indicated normal esophageal epithelial and murine and human ESCC cell lines, with β-tubulin as the loading control. G, qRT-PCR analysis of IFNA and IFNB mRNA expression across the indicated human esophageal cell lines (left) and Pearson correlation analysis of IFNA or IFNB mRNA levels with relative CHAF1A protein abundance determined by immunoblotting densitometry. H, Representative IHC images of CHAF1A, IFNA, and IFNB expression in human ESCC tissues with high or low CHAF1A expression. Scale bars, 100 μm. I, Correlation analysis between CHAF1A expression and IFNA or IFNB expression based on IHC staining categories (low, medium, and high) in human ESCC tissues (n = 40). The statistical significance was determined by Fisher exact test for each 3-by-3 contingency table. J, ELISA analysis of IFNA and IFNB secretion in KYSE150 and AKR cells following CHAF1A knockdown or KO compared with WT controls. Data are presented as mean ± SD; statistical significance was determined using Student t test analysis as appropriate. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. FC, fold change; Hi, high; Lo, low; Med, medium; TPM; transcripts per million.

CHAF1A depletion suppresses ESCC tumor growth via IFN production

Based on GSEA of TCGA data, samples with low CHAF1A expression were enriched for immune pathways, including innate immune response, JAK-STAT signaling, and viral response (Supplementary Fig. S2A). Integrated RNA-seq/ATAC-seq analysis also showed sustained enrichment of these pathways upon CHAF1A KO, with IFN-I production being a top-ranked signature (NES = 2, P < 0.001; Supplementary Fig. S2B–S2E). To investigate the functional impact on ESCC progression, we implanted CHAF1A-KO or WT ESCC mouse cell lines into immunocompetent C57BL/6 mice (Fig. 2A). CHAF1A-knockdown tumors exhibited impaired growth (Fig. 2B–D; Supplementary Fig. S2F–S2H). Using IHC and ELISA, we observed that CHAF1A knockdown promoted intrinsic IFNA/β production, CD8⁺ T-cell infiltration, and γH2AX expression (Fig. 2E–H; Supplementary Fig. S2I–S2Q). Concurrently, CHAF1A knockdown increased infiltrating CD8⁺ T-cell proportions (Fig. 2I and J). However, in immunodeficient BALB/c nude mice, the tumor-suppressive effect of CHAF1A knockdown was abolished (Fig. 2K–N). Consistent with this, CD8⁺ T-cell depletion in immunocompetent mice abolished the tumor suppression induced by CHAF1A knockdown (Fig. 2O–R; Supplementary Fig. S2R–S2T), with flow cytometry confirming a decrease in CD8⁺ T-cell proportions to below 1% (Fig. 2S and T). These results indicate that the antitumor effect of CHAF1A knockdown is mediated by CD8⁺ T cells, and such tumor suppression relies on an intact host immune system.

Figure 2.

CHAF1A promotes ESCC tumor growth in vivo by suppressing tumor-intrinsic IFN responses. A, AKR cell lines were transduced with WT controls or two independent CHAF1A shRNAs (shCHAF1A#1 and shCHAF1A#2) and subcutaneously injected into immunocompetent mice (C57BL/6). B, Representative images of tumors excised from mice bearing AKR. C and D, Final tumor weights (left) of AKR tumors (n = 6 mice per group) and tumor growth curves (right) of AKR tumors (n = 6 mice per group). Final tumor weights and tumor volume values at the final time point from growth curve measurements were analyzed using a two-tailed Student t test. All data are presented as mean ± SD. E, Tumor concentrations of IFNA (left) and IFNB (right) in mice bearing WT or CHAF1A-knockdown tumors, measured by ELISA (n = 3). F, Representative IHC picture of CHAF1A, CD8, and IFNA/B expression in tumor tissues derived from AKR cells (WT vs. shCHAF1A#1 and shCHAF1A#2). The infiltration of CD8⁺ T cells was evaluated separately in the tumor center (Core-CD8) and the invasive margin (Margin-CD8). Scale bars, 100 μm. G, Quantification of CD8⁺ T-cell infiltration (%) within the tumor core and margin areas across the indicated groups. H, Percentage of IFNA- and IFNB-positive staining areas relative to the total tissue area. Statistical significance was determined by a two-tailed Student t test. I, Representative flow cytometry gating plots showing the frequency of CD8⁺ T cells (gated on CD45⁺CD3⁺ population) in WT, shCHAF1A#1, and shCHAF1A#2 groups. J, Statistical quantification of the percentage of CD8⁺ T cells among total CD45⁺CD3⁺ cells (n = 3). K, AKR cells were transduced with WT controls or two independent CHAF1A shRNAs (shCHAF1A#1 and shCHAF1A#2) and subcutaneously injected into BALB/c nude mice (n = 6 mice per group). L, Representative images of tumors excised from mice bearing AKR. M and N, Final tumor weights of nude mice (M) and tumor growth curves (N). Final tumor weights and tumor volume values at the final time point from growth curve measurements were analyzed using a two-tailed Student t test. All data are presented as mean ± SD. O, Schematic timeline of the experimental design. Mice were injected with tumor cells on day 0, treated i.p. with anti-CD8 neutralizing antibody or isotype IgG control starting on day 1, and sacrificed on day 21. P, representative images of harvested tumors from the indicated treatment groups (n = 5 per group) at the endpoint. Q and R, Final tumor weights of nude mice (Q) and tumor growth curves (R). Final tumor weights and tumor volume values at the final time point of mice. Data are presented as mean ± SD. A two-way ANOVA followed by Bonferroni’s post hoc multiple comparisons test was performed. Intermediate time point values of tumor growth curves were plotted without statistical analysis. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. S, Flow cytometry plots showing the frequency of CD8⁺ T cells (pregated on CD45⁺CD3⁺ cells) to verify antibody-mediated depletion. T, percentage of CD8⁺ T cells among total CD3⁺ T cells across the indicated treatment groups. Data represent mean ± SD. *, P < 0.05; **, P < 0.01; ns, not significant (P ≥ 0.05).

CHAF1A loss activates an endogenous retrovirus–MAVS axis in ESCC

Endogenous retroviruses (ERV) produce dsRNA through bidirectional transcription and dsDNA via reverse transcription. Cytosolic RNA sensors recognize this dsRNA, triggering MAVS to propagate the signal (Fig. 3A). MAVS oligomers then recruit TBK1 to phosphorylate IRF3. The phosphorylated IRF3 ultimately initiates IFN-I gene expression (Fig. 3A). Western blot analysis showed that CHAF1A depletion upregulated RIG-I and MDA5 and induced MAVS accumulation, thereby promoting the phosphorylation of TBK1, IRF3, and STAT1 in ESCC cells (Fig. 3B; Supplementary Fig. S3A). Furthermore, CHAF1A KO triggered MAVS oligomerization and enhanced its mitochondrial colocalization, effectively mimicking polyinosinic-polycytidylic acid-induced activation (Supplementary Fig. S3B and S3D). To determine whether CHAF1A loss caused endogenous dsRNA accumulation, we used the dsRNA-specific J2 antibody for dot blot and IF analyses. These assays detected enhanced dsRNA levels in CHAF1A-deficient ESCC cells (Fig. 3C–E). As ERVs yield abundant intracellular dsRNA, we postulated that CHAF1A loss reactivated ERV transcription. RNA-seq analysis identified upregulated expression of numerous ERV families in CHAF1A-deficient cells, including >20 overexpressed LTR-containing molecules such as HERVH, HERV9N, and LTR7 (Fig. 3F). qRT-PCR confirmed these findings for specific ERVs (Fig. 3G). In contrast, non-ERV repetitive elements (e.g., ALU, L1ORF1) showed no changes in expression (Fig. 3H), indicating CHAF1A’s specific regulation of ERVs. We next asked whether this accumulated ERV RNA was biologically active and could provoke innate immune signaling. By RT-qPCR analysis, IFN-stimulated genes (ISG; e.g., CXCL9 and CXCL10) were broadly induced upon CHAF1A loss (Fig. 3I). Using RNase III treatment, we eliminated cytosolic dsRNA in CHAF1A-KO ESCC cells. The dsRNA degradation partially reduced ISG mRNA levels (Fig. 3I). Meanwhile, phosphorylated IRF3 and STAT1 levels were similarly halved upon RNase III treatment in CHAF1A-KO ESCC cells (Fig. 3J). These results suggest the existence of parallel mechanisms beyond the dsRNA axis. Collectively, CHAF1A loss derepresses ERV and activates the TBK1/IRF3 pathway in ESCC.

Figure 3.

Loss of CHAF1A enhances dsRNA accumulation through ERV activation in ESCC cells. A, Schematic diagram illustrating the cytosolic nucleic acid sensing pathways. B, Western blot image of cGAS–STING pathway components in CHAF1A-KO ESCC cells (KYSE150 and mEC25) or WT controls. C, J2 dot blot analysis with total RNAs extracted from WT control or CHAF1A-deficient KYSE150 cells. The graphs below show relative dot intensity. D and E, IF staining of dsRNA using the J2 antibody in KYSE150, mEC25, AKR, and KYSE30 cells with or without CHAF1A depletion. Representative images and quantification of fluorescence intensity are shown. Scale bars, 20 μm. F, Heatmap showing expression changes of ERV families and repetitive elements in CHAF1A-KO KYSE150 cells compared with WT controls, as determined by RNA-seq. G and H, qRT-PCR analysis of ERVs (G), short interspersed nuclear element (SINE), and long interspersed nuclear element (LINE) (H) in CHAF1A-KO KYSE150 cells and WT controls. I, qRT-PCR analysis of ISGs in CHAF1A-KO KYSE150 cells treated with or without RNase III and WT controls. Data represent mean ± SD of three independent experiments. Statistical significance was determined by one-way ANOVA followed by Tukey multiple comparisons test. *, P < 0.05, **, P < 0.01, ***, P < 0.001, ****, P < 0.0001. J, Western blot image of cGAS–STING pathway components in CHAF1A KO treated with or without RNase III, and WT KYSE150 cells.

CHAF1A silences ERVs via H3K9me3 in ESCC

Given that CHAF1A deficiency reactivates ERVs, we explored the underlying epigenetic mechanism. ATAC-seq analysis revealed increased chromatin accessibility at ERV loci upon CHAF1A loss (Fig. 4A). Western blotting further demonstrated that CHAF1A depletion selectively diminished H3K9me3, but not other ERV-related histone marks (H3K27ac, H3K4me3, H3K27me3; Fig. 4B). Mechanistically, immunoblotting revealed that CHAF1A KO reduced the protein levels of SETDB1, HP1α, and SUV39H1 (Fig. 4C). As the CAF-1 complex recruits SETDB1 and HP1α to generate the H3K9me1 substrate required for SUV39H1-mediated methylation, these results indicate that CHAF1A loss destabilizes the heterochromatin-silencing machinery, thereby disrupting H3K9me3 deposition in ESCC cells. Our ChIP assay showed clear enrichment of H3K9me3 at the MER4D-LTR1 locus, which was absent in the IgG control (Fig. 4D). We hypothesized that CHAF1A enforces ERV silencing by maintaining H3K9me3, a central repressive mark for genomic repeat elements. Consistently, ChIP-qPCR confirmed reduced H3K9me3 enrichment at specific ERV loci (e.g., MER21C LTR2, MTL2B4 LTR1, MLT1N2, MER4D LTR1) in CHAF1A-deficient cells (Fig. 4E and F). H3K9me3 CUT&Tag sequencing confirmed that CHAF1A KO reduced H3K9me3 enrichment at ERV genomic regions (Fig. 4G). Integrated CUT&Tag, ATAC-seq, and RNA-seq analyses further revealed that CHAF1A deletion decreased H3K9me3 occupancy, increased chromatin accessibility, and elevated ERV transcription (Fig. 4H). These data demonstrate that CHAF1A maintains H3K9me3-dependent heterochromatin silencing at ERV loci. ATAC- and RNA-seq analyses showed increased chromatin accessibility and elevated transcription of ISGs (e.g., BST2, ISG15), directly linking ERV derepression to downstream innate immune activation (Fig. 4I). In summary, we establish that CHAF1A mediates ERV silencing in ESCC by maintaining H3K9me3-marked heterochromatin.

Figure 4.

CHAF1A maintains H3K9me3-dependent heterochromatin silencing of ERVs. A, ATAC-seq analysis of chromatin accessibility at ERVs in WT and CHAF1A-KO KYSE150 cells. B, Western blot analysis showing the expression of histone modification markers in CHAF1A KO and WT ESCC cells (KYSE150 and mEC25). C, Western blot analysis of CHAF1A, SETDB1, SUV39H1, HP1α, and KAP1 protein levels with β-tubulin as the loading control. D, Analysis of DNA fragments enriched by ChIP using agarose gel electrophoresis. The specific PCR-amplified region is the MER4D LTR1 genomic locus. E and F, ChIP-qPCR validation of H3K9me3 binding at multiple ERV promoters in KYSE150 and mEC25 cells. Data represent mean ± SD of three biological replicates. Statistical analysis was performed using two-tailed unpaired Student t test. *, P < 0.05; **, P < 0.01. G, H3K9me3 CUT&Tag profiles across ERV genomic regions in WT and CHAF1A-KO KYSE150 cells. Heatmaps (left) and metaplots (right) show H3K9me3 enrichment centered on ERV loci within a 5 kb window; signal intensity is shown as RPKM. CUT&Tag was performed using one WT and one KO biological sample, and ENCODE-recommended pseudoreplicate analysis was used to assess signal reproducibility. H, Genomic tracks of ATAC-seq, RNA-seq, and H3K9me3 CUT&Tag. IGV snapshots show chromatin accessibility (pink), gene expression (blue), and H3K9me3 enrichment (orange) at specific ERV loci in WT and KO KYSE150 cells. Track scales are indicated. I, Genomic tracks of ATAC-seq and RNA-seq. IGV snapshots show chromatin accessibility and gene expression at specific ISG loci in WT and KO KYSE150 cells. Track scales are indicated.

CHAF1A deficiency induces DNA damage to activate dsDNA–cGAS–STING

Upon sensing cytosolic dsDNA, cGAS stimulates STING to recruit TBK1, leading to IRF3 phosphorylation and IFN gene transcription (Fig. 3A). IF identified elevated micronuclei formation in CHAF1A-deficient ESCC cells, indicating enhanced DNA damage (Fig. 5A and B). We found that CHAF1A loss or low expression correlated with diminished DNA repair capacity by GSEA (Fig. 5C). CHAF1A-deficient ESCC cells also displayed reduced mRNA expression of double-strand break repair-related genes (e.g., BANF1, RUVBL1, and XRCC6; Fig. 5D). We next found elevated γH2AX at the protein level in CHAF1A-deficient ESCC cells (Fig. 5E). Neutral comet assays quantified significantly higher relative olive tail moment upon CHAF1A loss in KYSE150 cells (Fig. 5F). The upregulation of ISGs induced by CHAF1A KO was substantially abrogated by simultaneous treatment with RNase III and the cGAS inhibitor RU.521 (Fig. 5G). Furthermore, IF revealed that CHAF1A deficiency led to an accumulation of γH2AX foci (Fig. 5H). We next asked if CHAF1A deficiency activated the dsDNA–cGAS–STING cascade. To delineate the specific pathways involved, immunoblotting showed that si-cGAS or RU.521 treatment suppressed the dsDNA axis (p-STING, p-IRF3, and p-STAT1), whereas si-MAVS inhibited the dsRNA axis (p-TBK1, p-IRF3, and p-STAT1; Supplementary Fig. S4A–S4C). Combined RNase III and cGAS inhibitor RU.52 treatment abolished the activation of all these downstream signaling molecules (Fig. 5I). qRT-PCR confirmed that CHAF1A KO upregulated cGAS mRNA levels in ESCC cells, consistent with its increased protein expression (Supplementary Fig. S4D and S4E). Mechanistically, CHAF1A depletion increased RPA32 and CHK1 phosphorylation (Supplementary Fig. S4F), showing elevated replication stress, although cell proliferation remained unaffected (Supplementary Fig. S4G and S4H). We further analyzed IκBα expression and observed no significant alterations between the WT and CHAF1A-knockdown groups (Supplementary Fig. S4I). Consistently, GSEA showed no notable enrichment in NF-κB, mTOR, or autophagy signaling pathways, suggesting that no compensatory bypass pathways were activated (Supplementary Fig. S4J). Furthermore, CHAF1A deficiency induced inconsistent activation levels of the cGAS–STING pathway across breast, gastric, colorectal, and lung cancer cells, with no uniform regulatory pattern observed (Supplementary Fig. S4K), underscoring an ESCC-specific vulnerability. Our findings reveal that in ESCC, CHAF1A loss induces genomic instability and cytosolic DNA accumulation, leading to cGAS–STING–IRF3 pathway activation.

Figure 5.

Genomic instability activates the cGAS–STING pathway and induces intrinsic immune signaling in CHAF1A-deficient ESCC cells. A and B, Representative IF images and quantification showing the accumulation of cytoplasmic DNA and micronuclei in CHAF1A-KO ESCC cells (KYSE150, mEC25, KYSE30) and WT controls. Nuclei were stained with DAPI. Scale bars, 20 μm (KYSE150), 5 μm (mEC25 and KYSE30). Data are mean ± SD; significance was determined by a two-tailed Student t test. C, GSEA of RNA-seq (left) in CHAF1A-KO KYSE150 cells compared with WT controls and GSEA of the TCGA dataset (right). D, Quantitative qRT-PCR validation showing the expression of representative DNA repair–associated genes between CHAF1A-KO and WT controls in KYSE150 and mEC25 cells. Data are presented as mean ± SD; P values were determined by a two-tailed Student t test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. E, Immunoblotting images of γH2AX in KYSE150 and mEC25 cells with or without CHAF1A KO. F, Representative images and quantification of the neutral comet assay showing increased tail moments in CHAF1A-KO ESCC cells. G, Quantitative qRT-PCR validation showing the expression of representative ISG genes in KYSE150 cells with CHAF1A KO, WT, and CHAF1A KO with cGAS inhibitor (RU.521). H, IF staining of γH2AX in KYSE150 and mEC25 cells with or without CHAF1A KO. Quantification of γH2AX foci per nucleus is shown on the right. Scale bars, 20 μm. I, Western blot image of cGAS-STING pathway components in CHAF1A KO treated with or without RNase III and RU.521, and WT KYSE150 cells. DSB, DNA double-strand break.

CHAF1A loss potentiates ICB therapy in ESCC

To evaluate whether CHAF1A deficiency potentiates the response to ICB, we generated syngeneic grafts using CHAF1A-knockdown and WT ESCC cells (Fig. 6A). Anti–PD-1 treatment markedly suppressed the growth of CHAF1A-deficient tumors (Fig. 6B), with corresponding reductions in tumor volume and weight (Fig. 6C and D). IHC showed stronger expression of CD8, IFNA/B, and γH2AX in CHAF1A-depleted tumors with anti–PD-1 therapy (Fig. 6E–G; Supplementary Fig. S5A and S5B). Additionally, flow cytometry confirmed increased infiltration of CD8⁺ T cells (Fig. 6H) and a higher proportion of IFNγ⁺CD8⁺ T cells (Fig. 6I). The levels of secreted IFNA/B were also elevated in CHAF1A-knockdown tumors following anti–PD-1 treatment (Fig. 6J). Collectively, these data indicate that CHAF1A depletion activates innate immunity and triggers CD8⁺ T-cell recruitment, thereby improving ICB efficacy.

Figure 6.

CHAF1A expression determines the responsiveness of ESCC tumors to anti–PD-1 immunotherapy. A, mEC25 and AKR cells with CHAF1A knockdown (shCHAF1A#1 or shCHAF1A#2) or WT control were subcutaneously injected into immunocompetent mice, followed by treatment with anti–PD-1 antibody or isotype IgG as indicated. Tumors were collected for analysis at the endpoint. B, Representative images of tumors excised from mice bearing mEC25 or AKR. C and D, Tumor growth curves and endpoint tumor weights of mEC25 (C) and AKR (D) tumors under the indicated treatments. Final tumor weights and tumor volume values at the final time point of mice. Data are presented as mean ± SD. A two-way ANOVA followed by Bonferroni post hoc multiple comparisons test was performed. Intermediate time point values of tumor growth curves were plotted without statistical analysis. E, Representative IHC images of tumor sections stained for CHAF1A, CD8, IFNA, IFNB, and γH2AX in mEC25 and AKR tumors. Scale bars, 50 μm. F and G, IHC analysis of CD8 and IFNA/B expression in CHAF1A-knockdown tumors of mEC25 (F) and AKR (G). H, Flow cytometric analysis of tumor-infiltrating CD8⁺ T cells. I, Quantification of CD8⁺ T-cell frequency and the proportion of IFNγ⁺ cells among CD8⁺ T cells. J, Tumor concentrations of IFNA and IFNB in tumor-bearing mice measured by ELISA. Data are presented as mean ± SD. Statistical analysis was performed by a two-way ANOVA followed by Bonferroni post hoc multiple comparisons test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Baimaside inhibits CHAF1A and potentiates ICB therapy in ESCC

Based on the tumor-promoting functions of CHAF1A observed in our study, we conducted a high-throughput virtual screening against its full-length structure (AlphaFold ID AF-Q13111-F1) to identify candidate compounds (Fig. 7A; Supplementary Fig. S6A; ref. 33). From an initial library of 10,000 molecules, two rounds of screening yielded 10 top-scoring hits (Supplementary Table S4). Among these, baimaside (XP docking score: −8.987) was selected for subsequent in silico and experimental validation, as its predicted binding pose revealed a favorable interaction network, including seven hydrogen bonds with residues SER787, GLU788, GLN783, CYS761, ARG762, and SER767 within the CHAF1A binding pocket (Fig. 7B–D; Supplementary Fig. S6B). To computationally validate the binding stability of the CHAF1A–baimaside interaction, we performed 100-nanosecond MD simulations (Supplementary Fig. S6C and S6D). Although the ligand RMSD equilibrated within the initial 20 nanoseconds, both protein and ligand trajectories remained stable thereafter, with the ligand consistently binding to the R6 residue at a distance of approximately 8 Å throughout the simulation. In AKR and mEC25 subcutaneous tumor models, baimaside significantly suppressed tumor growth (Fig. 7E–G; Supplementary Fig. S6E–S6G), with no appreciable toxicity; body weights remained stable, and histopathologic evaluation showed no major organ damage (Supplementary Fig. S6H–S6K). Given the immunomodulatory role of CHAF1A, we further evaluated baimaside together with anti–PD-1. The combination inhibited tumor growth compared with either monotherapy in both AKR and mEC25 models (Fig. 7H–J; Supplementary Fig. S6L–S6N). Consistently, CCK-8 assays showed that baimaside treatment had no significant impact on the proliferation of AKR and KYSE150 cells, further confirming its low cytotoxicity (Supplementary Fig. S6O and S6P). IHC analysis revealed elevated infiltration of CD8⁺ and CD4⁺ T cells, together with increased IFNB, γH2AX, and CHAF1A expression in the combination group (Supplementary Fig. S6Q), indicating activation of the antitumor immune microenvironment. Consistently, baimaside administration significantly reduced CHAF1A and H3K9me3 levels in AKR tumors (Supplementary Fig. S7A and S7B). Collectively, our data characterize baimaside as a well-tolerated compound with CHAF1A-targeting activity, single-agent efficacy, and anti–PD-1 potentiating effects, supporting its combinatorial development for ESCC immunotherapy. This mechanism is illustrated in Fig. 7K, which delineates how baimaside-mediated CHAF1A inhibition triggers endogenous retroviral transcription and innate immune activation, ultimately sensitizing ESCC cells to ICB.

Figure 7.

Baimaside, a CHAF1A inhibitor, synergizes with anti–PD-1 therapy by activating antitumor immunity in ESCC. A, Flowchart of the high-throughput virtual screening for CHAF1A inhibitors. B, Chemical structure of baimaside. C, Two-dimensional representation of the binding mode between baimaside and human CHAF1A protein. D, Three-dimensional representation of the binding mode between baimaside and human CHAF1A protein. E–G, AKR syngeneic tumors in C57BL/6 mice were treated with PBS or baimaside. Representative tumor photographs (E), growth curves (F), and weights of excised tumors (G) are shown. H–J, AKR syngeneic tumors in C57BL/6 mice were treated with PBS, baimaside, anti–PD-1 antibody, or baimaside combined with anti–PD-1 antibody. Representative tumor photographs (H), growth curves (I), and weights of excised tumors (J) are shown. Data are presented as mean ± SD; statistical analysis was performed by a two-way ANOVA followed by Bonferroni post hoc multiple comparisons test. *, P < 0.05; ***, P < 0.001; ****, P < 0.0001.  K, Graphical illustration of the mechanisms by which CHAF1A depletion overcomes immunotherapy resistance in ESCC. Loss of CHAF1A activates both the cGAS–STING and MAVS–IRF3 pathways, which synergize to establish an IFN-rich tumor microenvironment and enhance CD8⁺ T cell–mediated antitumor immunity. HTVS, high-throughput virtual screening; SP, single precision; XP, extra precision.

Discussion

In our study, we identified CHAF1A as a negative regulator of the tumor-intrinsic IFN-I response in ESCC through multiomics analysis. We found that CHAF1A was upregulated in ESCC tissues and negatively correlated with IFN-I production and CD8⁺ T-cell infiltration. Mechanistically, CHAF1A suppressed both the MAVS and cGAS–STING pathways, and its genetic or pharmacologic inhibition potentiated the effects of anti–PD-1 therapy in vivo. These findings establish CHAF1A as a promising target to overcome immunotherapy resistance in ESCC.

Mechanistically, we showed that CHAF1A maintained heterochromatin silencing at ERV loci. As the core subunit of the CAF-1 chromatin assembly complex, CHAF1A recruits SETDB1 and HP1α to promote H3K9me1 generation, which serves as the substrate for SUV39H1/H2-mediated H3K9me3 (34, 35). Loss of CHAF1A impaired SETDB1 recruitment and reduced H3K9me3 occupancy at ERV loci, leading to ERV derepression, dsRNA accumulation, and activation of the RIG-I/MDA5–MAVS axis (36–39).

In addition to derepressing ERVs, CHAF1A deficiency has been reported to compromise chromatin assembly during DNA replication and repair, resulting in replication fork stalling and collapse (40–43). Consistent with these findings, we observed an increased replication stress response following CHAF1A deficiency, as evidenced by increased phosphorylation of RPA32 and CHK1. Our data further showed that CHAF1A-deficient cells accumulated cytosolic dsDNA and activated the cGAS–STING pathway, thereby contributing to enhanced IFN-I production.

Beyond its direct role in nucleic acid sensing, CHAF1A has been implicated in various biological processes across cancers. It acts as a histone chaperone promoting c-MYC expression in gastric cancer (44), activates ERVs in hepatocellular carcinoma (HCC; ref. 45), participates in DNA damage repair (36–38), and blocks differentiation in neuroblastoma (46). These reports highlight CHAF1A as a multifunctional integrator of epigenetic regulation, genomic stability, and tumor progression.

Our study identified CHAF1A as an ESCC-specific upstream modulator governing H3K9me3 deposition at ERV loci. Although H3K9me3-mediated ERV silencing is well documented, the upstream regulators of this repressive mark remain largely undefined. A recent study reported that CAF-1 inhibition triggers innate immunity in HCC through H3.3 deposition rather than H3K9me3 reduction (45). In contrast, our study demonstrated that CHAF1A loss in ESCC reduced H3K9me3, not H3.3 deposition. Furthermore, our analysis indicated that the CHAF1A-mediated control of innate immunity was not a universal mechanism. CHAF1A depletion failed to induce a consistent immune response across breast, gastric, colorectal, and lung cancer models, reinforcing the clinical relevance of targeting CHAF1A as an ESCC-specific vulnerability.

Nevertheless, our study has several limitations. Our single-center ESCC immunotherapy cohort requires expansion for reliable survival analysis. The direct genomic targets of CHAF1A remain undefined, and future studies are needed to determine whether CHAF1A directly silences ERVs or indirectly regulates chromatin organization. Additionally, the mechanisms by which CHAF1A deficiency remodels myeloid and dendritic cell functions in the tumor microenvironment warrant further investigation.

In summary, our study identifies CHAF1A as a critical hub that links epigenetic regulation, genomic stability, and innate immunity in ESCC. By suppressing both ERV-derived dsRNA and genomic instability-induced dsDNA, CHAF1A prevents the activation of tumor-intrinsic innate immunity. We identified baimaside as a small-molecule CHAF1A inhibitor and demonstrated its single-agent and combinatorial activity with anti–PD-1 therapy in vivo. These findings provide a mechanistic rationale for targeting CHAF1A to improve immunotherapy outcomes in ESCC.

Supplementary Material

Supplementary Fig. S1

High CHAF1A expression mediates interferon suppression, T-cell exclusion, and anti–PD-1 insensitivity in esophageal cancer.

Supplementary Fig. S2

CHAF1A knock-down broadly up-regulates antiviral/immune pathways.

Supplementary Fig. S3

CHAF1A depletion activates the MDA5/RIG-I-MAVS innate immune sensing pathway.

Supplementary Fig. S4

Effects of CHAF1A deficiency on cGAS-STING and MAVS signaling pathways.

Supplementary Fig. S5

CHAF1A depletion synergizes with anti–PD-1 therapy to enhance DNA damage and antitumor immunity.

Supplementary Fig. S6

Baimaside targets CHAF1A to suppress esophageal cancer progression and potentiate anti–PD-1 immunotherapy.

Supplementary Fig. S7

In vivo evaluation of anti–PD-1 response and H3K9me3 alterations under Baimaside treatment.

Supplementary Tables

Supplementary Tables 1-4.

Data Availability

The high-throughput functional genomics and sequencing data generated during this study have been deposited in the Genome Sequence Archive at the China National Center for Bioinformation (https://ngdc.cncb.ac.cn/). RNA- and ATAC-seq data can be accessed under the number HRA014969. Cut&Tag data can be accessed under the number HRA017953. Other remaining data supporting the findings of this study are available from the corresponding author upon reasonable request.

Authors’ Disclosures

No disclosures were reported.

Authors’ Contributions

Y. Ma: Data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. M. Shi: Conceptualization, supervision, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. M. Ma: Data curation, validation, investigation, visualization, project administration. S. Li: Data curation, software. N. Cui: Validation, investigation, visualization. Y. Li: Software, formal analysis, supervision, methodology, writing–original draft, writing–review and editing. T. Dang: Resources, data curation, software, formal analysis, supervision, methodology, writing–original draft. R. Yang: Resources, data curation, supervision, validation, investigation. Y. Bai: Formal analysis, supervision, funding acquisition, validation. B. Wang: Resources, data curation, software, validation, investigation, visualization, methodology. C. Zhang: Resources, data curation, supervision, funding acquisition. C. Liu: Conceptualization, resources, data curation, supervision, funding acquisition, validation, investigation, visualization, project administration. Y. Zhang: Resources, funding acquisition, project administration.