Blockade of Immune-Checkpoint B7-H4 and Lysine Demethylase 5B in Esophageal Squamous Cell Carcinoma Confers Protective Immunity against P. gingivalis Infection

Between 15% and 20% of newly diagnosed cancers are likely related to infection (1, 2). Although pathogens themselves might not be oncogenic, they compromise the overall health of the host and can result in an exacerbation of lymphocyte exhaustion and disruption in the immune surveillance required for clearing transformed cells within the site of infection. This cascade favors the development of oncogenic processes (3). Numerous studies have examined resistance to infection in patients and preclinical models. However, the mechanisms driving bacterial evasion of host defenses and infection-associated immunopathology remain unclear.

Porphyromonas gingivalis, an anaerobic gram-negative bacterium, is the primary causative agent of chronic periodontitis (4). This bacterium is frequently present in specimens of esophageal squamous cell carcinoma (ESCC) and esophageal dysplasia (5, 6). The persistence of infection suggests that clearance of this organism by the host immune response is inadequate. Therefore, understanding the interactions that occur between P. gingivalis and the host immune defense system is important (7, 8).

B7-H4 (also known as B7x and B7S1, encoded by the VTCN1 gene) is reported to be a B7 T-cell costimulator (9), but limited advances have been made on its involvement in the immune response to infection. An earlier investigation has shown that B7-H4^−/− mice have a relatively stronger T helper (Th)1 response and lower degree of parasite burden following challenge with Leishmania major (10), whereas another study demonstrated that B7-H4 knockout mice exhibit increased resistance to a lethal challenge with Listeria monocytogenes by mounting an augmented neutrophil response (11). A third study reported that B7-H4 knockout mice confer resistance to a lethal pulmonary infection with Streptococcus pneumonia (12). In terms of cancer, B7-H4 knockout mice with lung tumors show significantly enhanced survival and memory responses to a secondary challenge (13). Thus, B7-H4 targeting represents a promising strategy for reinvigorating immunity.

Due to the complexities associated with the immunoregulatory network and tumor–host heterogeneity, immunotherapeutic strategies remain clinically challenging, as not all cancers respond to immunotherapy, especially those exhibiting poor immunogenicity (14, 15). Therefore, researchers are seeking hosts capable of regulating endogenous immune priming, as well as T-cell trafficking into the tumor beds, as research subjects for augmenting therapeutics. In lung, as well as ovarian cancers, DNA methyltransferase inhibitors (DNMTi) and histone deacetylase inhibitors (HDACi) have been shown to enhance responses to immune-checkpoint blockade by removing the epigenetic markers that suppress chemokine expression (16, 17). Lysine demethylase 5B (KDM5B or JARID1B) belongs to the JumonjiC domain-containing (JmjC) demethylase family and functions as a repressor by removing three methyl groups at lysine 4 of histone 3 (H3K4me3), thereby affecting target genes (18). Selective inhibitors that have been identified (19, 20) may be utilized for the development of novel strategies in the treatment of poorly immunogenic cancers.

In this study, we explored the mechanisms used by P. gingivalis to establish a chronic infection and to avoid the immunogenic responses of the host. We also examined how the immune-checkpoint molecule B7-H4 and the histone demethylase KDM5B are involved in conferring a survival advantage, which can potentially be exploited therapeutically for ESCCs exposed to infection with P. gingivalis.

This multicenter study on esophageal cancer was performed at The First Affiliated Hospital of Henan University of Science and Technology (HUST; Luoyang, Henan, China) and Anyang Tumor Hospital (ATH; Anyang, Henan, China). Patients were enrolled from 2010 to 2014, primarily from the Taihang Mountains region of the Henan provinces, where ESCC is prevalent. Prior to sample collection, informed consent was obtained from all patients in accordance with the protocols approved by the institutional review boards of HUST and ATH. All procedures involving patients were performed in accordance with the Declaration of Helsinki ethical guidelines (21). The present study enrolled 120 ESCC patients undergoing surgery or diagnostic upper gastrointestinal endoscopy, among which 50 were P. gingivalis positive. No restrictions regarding age, sex, or disease stage were established. The exclusion criteria were (i) patients who had been treated with chemotherapy, radiotherapy, or immune therapy; (ii) patients who had received transfusions within 6 months prior to the study; and (iii) patients who had used antibiotics, proton pump inhibitors, and nonsteroidal anti-inflammatory drugs in the preceding 6 days. Each specimen was cut into pieces and treated differently for various uses, put in liquid nitrogen to extract RNA for RNA sequencing (RNA-seq), or fixed in 10% formaldehyde for making paraffin-embedded blocks. Additional fresh ESCC specimens were collected for explants cultivation for chemokines measurement or generation of patient-derived xenografts (PDX). CD8⁺ T cells were isolated and sorted from heparinized venous blood samples of healthy adult volunteers negative for P. gingivalis for coculture and chemotaxis functional studies, and those from ESCC donor patients positive for P. gingivalis were for adoptive T-cell transfer and development of humanized mouse model.

The human ESCC cell lines Kyse-410, Kyse-150, Kyse-180, and Yes-2 were kindly donated by Dr. Shimada at Kyoto University (Kyoto, Japan). 293T cells were maintained in our laboratory. Cell lines were cryopreserved in aliquots or maintained at a density of 5 × 10⁵ cells/mL in RPMI-1640 medium (GE Healthcare) supplemented with 10% fetal bovine serum (GE Healthcare) in an incubator at 37°C and 5% CO₂. To help prevent or control contamination with Mycoplasma, anti-Mycoplasma reagent (InvivoGen) was used prophylactically at a working concentration of 2.5 μg/mL. All cells were maintained or passaged for less than 3 months, prior to cryopreserving aliquots for further procedures. Short tandem repeat profiling was carried out by GENEWIZ to authenticate the cell lines.

NOD/SCID IL2γc^−/− (NSG) and C57BL/6 mice (Biocytogen) were housed under specific pathogen-free conditions at the animal care facility of HUST. All procedures involving animals were approved and monitored by the animal care and use committee of HUST.

P. gingivalis strain ATCC 33277 was kindly provided by Richard J. Lamont of the University of Louisville (Louisville, KY). The bacteria were grown on anaerobic blood agar plates (BD) in a chamber with 85% N₂, 5% H2, and 10% CO₂ for 5 to 7 days. It was then inoculated into a liquid broth of brain heart infusion (BHI) for 24 hours until the culture reached an optical density of 1.0 at 600 nm corresponding to 1 × 10⁹ colony-forming units (CFU) per/mL. Bacterial strains were maintained as frozen stocks and after being started, were passed three times through growth medium before being used. P. gingivalis strain 33277 was applied to studies of bacteria–host cross-talk in vitro, as well as animal infection studies. The P. gingivalis strain L2 was derived from a clinical isolate associated with ESCC lesions from the high incidence areas of the Taihang Mountains, as we described (6), and were used for inoculation of human ESCC explants. For each assay, bacterial numbers were established by the number of CFU of P. gingivalis serially diluted on blood agar plates. The viability of the bacterial cells was evaluated under a phase-contrast microscope, Nikon TE2000 (Nikon Corp.), before the cells were used for subsequent analysis.

Explant cultures were used for studies of bacteria–host cross-talk in vivo to mimic human infection. Surgical specimens were collected after informed consent was given from 12 ESCC donor patients negative for P. gingivalis (3 women, 9 men; mean age: 64 years) for explants culture.

The tumor tissue samples (diameter, 2.5 cm) were transferred to 0.9% NaCl solution immediately after removal and sent to the laboratory at Anyang Cancer Hospital (Anyang, Henan, China). The underlying tissue layers were removed, and then three parts (diameter, 2 mm) were punched out from each epithelium. The presence of P. gingivalis in the tissue material was analyzed by 16S rDNA.

Explants were applied to each insert cup (Transwell; Corning) with the mucosal side facing up and with capillary contact with Iscove's medium (Biochrom KG), supplemented with 5% fetal calf serum (Invitrogen, Inc.) and 4 mL/L Skirrows supplement (Oxoid) below the grid in a culture dish (Falcon Plastics). The insert cup's membrane pore size (0.4 μm) prevented P. gingivalis from passing through the insert into the medium below.

P. gingivalis strain L2 was cultured on blood agar plates, harvested, washed, and resuspended in Iscove's medium. An inoculum of 10⁷ CFU organisms suspended in 200 μL of medium was added to cover the explants. After incubation for 1 hour at 37°C, the bacterial suspension was removed, and explants were washed 5 times with 600 μL of the medium. At 24 hours, 800 μL of the culture medium underneath the insert cups was collected and stored at –80°C; the same volume of fresh medium was added, and incubation continued for another 24 hours before all media were collected for subsequent chemokine quantification using human chemokine array kits (R&D Systems) according to the manufacturer's instructions.

PCR for amplification of 16S rDNA of P. gingivalis was performed as we previously described (5) in a total volume of 25 μL containing 2 μmol/L of primers and 10 ng of template DNA. 16S rDNA samples were amplified using P. gingivalis–specific and universal 16S rDNA primers. P. gingivalis 16S rDNA primer sequences were 5′AGGCAGCTTGCCATACTGCG3′ (forward) and 5′ ACTGTTAGCAACTACCGATGT 3′ (reverse), and the PCR product size was 404 base pairs (bp). The universal 16S rDNA primer sequences were 5′GATTAGATACCCTGGTAGTCCAC3′ (forward) and 5′CCCGGGAACGTATTCACCG3′ (reverse), and the PCR product size was 688 bp. The PCR cycling conditions were 30 cycles of denaturation at 94°C for 30 seconds, annealing at 65°C for 30 seconds with a decrease of 0.2°C per cycle, and extension at 72°C for 30 seconds.

The libraries were constructed by the Shanghai Biotechnology Corporation. We performed RNA-seq of Kyse-150 and Kyse-410 ESCC cells in the presence or absence of P. gingivalis isolates, followed by in vivo screening of P. gingivalis–positive subjects. Total RNA was isolated and purified from ESCC tissues and cells using the RNeasy Mini kit (Qiagen). All the coding RNAs were captured with the TruSeq Stranded Total RNA Sample Preparation kit (Illumina) in accordance with the manufacturer's instructions. Briefly, the purified mRNAs were randomly fragmented and transcribed into cDNAs. After 3′ adenylation, sequencing adaptors were added to the cDNAs, and the fragments were amplified by PCR. The PCR products were quantified using a Qubit 2.0 fluorometer (Life Technologies), and the quality of the products was evaluated using an Agilent 2100 bioanalyzer (Agilent Technologies). The RNA integrity numbers of the samples used were >7. The concentration of each library was normalized to 10 pmol/L, and then the libraries were sequenced on the Illumina HiSeq X Ten (Illumina).

The constructed libraries were sequenced using an Illumina HiSeq 2500 system (Illumina). Low-quality reads, short reads, rRNAs, and reads containing primer/adaptor contamination were removed. The remaining high-quality reads were mapped to a reference genome (mm10) with two mismatches using Tophat v2.0.9 fragments per kilobase of transcript per million mapped reads (FPKM). The genes in the libraries were produced using the Cufflinks v2.1.1 tool in accordance with the reference annotation set (mm10). Fold changes for the genes in different samples were estimated based on their FPKM values, and the significance threshold was determined according to the false discovery rate (FDR). In this study, a differentially expressed gene was defined as a gene whose expression changed more than 2-fold among the different samples with FDR ≤ 0.05. The RNA-seq results were deposited into the Sequence Read Archive database under accession number PRJNA516898. All the data from this study are available upon a reasonable request.

Total RNA was isolated using TRIzol reagent (Invitrogen) according to the manufacturer's instructions. Two micrograms of total RNA was reverse transcribed using the reaction ready, first-strand cDNA synthesis kit (SuperArray Bioscience). Real-time PCR was performed in triplicate using SYBR Green PCR Master Mix (Applied Biosystems) and the ABI Prism 7900 sequence detection system (Applied Biosystems). Quantitative PCR reactions were performed under conditions standardized for each primer. Standard curves were generated using 10-fold dilutions of standard plasmids. Analysis of quantitative real-time PCR (qRT-PCR) data was carried out using the comparative ΔΔC_T method. For each sample, the intensity of the amplimer was normalized against that of the internal control human β-actin, and then the data were each depicted as several fold changes (2^−ΔΔC_T) with respect to the transcription level observed in the control experiment. The KDM5B (QPH12457A) primers were purchased from SuperArray Bioscience. Primers used in qPCR are listed in Supplementary Table S1.

Human ESCC cells (Kyse-410, Kyse-150, Kyse-180, and Yes-2) incubated with or without P. gingivalis were lysed for extraction of total proteins using an extraction kit (Pierce). Briefly, 10 μg of extracts were separated on SDS-PAGE (12% polyacrylamide) and transferred onto a polyvinylidene difluoride (PVDF) membrane (Invitrogen). The membrane was blocked with skim milk and washed before incubating with specific antibodies. The antibodies used in immunoblotting included anti–B7-H4 (Abcam; cat. #ab209242) and anti-KDM5B (NOVUS Biologicals; cat. #NBP1-97310SS).

Before performing the surface staining for flow cytometry, 1 × 10⁶ cells were preincubated with human serum and horse serum in phosphate-buffered saline (PBS) for 30 minutes on ice. After washing twice with PBS, the expression markers on T cells and tumor cells were determined by fluorescence-activated cell sorting (FACS; Beckman Coulter) analyses after surface staining with the respective conjugated antibody (Ab; 1 μg) and also with the proper isotype control Ab (1 μg) for 45 minutes on ice. Anti–B7-H4 (cat. #12594942) antibody was from Invitrogen; CD4-FITC, human (cat. #130-113-775), CD8-PE, human (cat. #130-104-130), CD3-PE, human (cat. #130-091-374), CD3-FITC, human (cat. #130-113-128), CD4-FITC, mouse (cat. #130-120-750), CD8-PE, mouse (cat. #130-102-595), CD3-PE, mouse (cat. #130-109-879), and CD3-FITC, mouse (cat. #130-109-878) were all purchased from Miltenyi.

Peripheral blood mononuclear cells (PBMC) were purified from buffy coats from venous blood from 3 healthy adult donors negative for P. gingivalis using Ficoll-Hypaque gradient centrifugation. Bulk CD8⁺ T cells were isolated by negative selection with microbeads (Miltenyi Biotech) and were used in the cocultures described below. T cells were stimulated with a predetermined optimal concentration of activation beads (Miltenyi) coupled with anti-CD3 and anti-CD28 antibodies at a 2:1 beads:T-cell ratio. To determine the role of B7-H4 on ESCC cells for T-cell proliferation, Kyse-150 and Kyse-410 were exposed to P. gingivalis at a 50:1 bacterium/cell ratio for 6 hours. The cells were then washed twice with PBS and plated at 1 × 10⁵ cells per well in a 24-well plate for 24 hours before the addition of T cells. For the coculture study, CD8⁺ effector cells were added to target ESCC cells over an effector:target (E:T) ratio of 1:10 and incubated at 37°C with 5% CO₂ for 3 days. For blocking B7-H4, anti–B7-H4 (5 μg/mL; functional grade from eBioscience, cat. #16-5972-81) or an isotype control, human immunoglobulin G1 antibody (5 μg/mL; functional grade from eBioscience), was added to ESCC cells for 60 minutes before T-cell addition. The proliferation of T cells was determined by carboxyfluorescein succinimidyl ester (CFSE; Invitrogen) staining during the final 16 hours of coculture. After being washed two more times, the labeled cells were analyzed by FACS (Beckman Coulter), gating on the CFSE-labeled cells. Supernatants were collected from coculture wells after 7 days and used to quantitate the production of IL2 by enzyme-linked immunosorbent assays (ELISA; R&D Systems) according to the manufacturer's instructions.

CD8⁺ lymphocytes were selected from PBMCs by negative selection using magnetic beads and cultured with anti-CD3/anti-CD28–coated beads for 7 days to generate CD8⁺ effector cells (Miltenyi). These cells were loaded into the upper chambers of transwell inserts (5.0-μm pore size, Costar). In the bottom well, medium containing different amounts of neutralization antibodies to chemokines, or culture supernatant from P. gingivalis–infected cell lines (Kyse-410 and Kyse-150), was added. The KDM5B inhibitor CPI-455 (cat. #S8287) was from Selleck. For antibody blocking assays, neutralization anti-CXCL9 (MAB392), anti-CXCL10 (MAB266), and anti-CXCL11 (MAB672; R&D Systems) were added into culture supernatants and incubated at 37°C for 30 minutes before adding into T cells. The contents of the lower chamber were collected, and the percentage of CD8⁺ cells was determined by FACS.

A plasmid encoding KDM5B was inserted into the pcDNA 3.1 (−)/Myc-His A (Invitrogen) vector. pcDNA3.1-KDM5B-Myc-His (H499Y) and the deletions of the PHD2/PHD3 (Δ752–1544 aa) were generated by PCR as previously described (18). RNAi constructs were made by synthesizing oligonucleotides encoding 19 bp short-hairpin RNA that target human KDM5B (RNAi1: 5′-GGAGATGCACTTCGATATA-3′, RNAi2: 5′-CACTGGAGCTATTCAATTA-3′) and cloned into pHTPsiRNA vector.

For chromatin immunoprecipitation (ChIP) assays, pcDNA3.1-KDM5B-Myc-His (H499Y) or mutant vector were transfected into 293T cells with FuGene6. Analysis of H3K4 methylation at KDM5B-binding regions was performed in Kyse-450 cells (Mock) and Kyse-450 cells with KDM5B knockdown (5BKD). After 48 hours of transfection, cells were treated with DMEM containing 1% formaldehyde for 10 minutes. Cross-linking was stopped by the addition of 0.125 mol/L glycine for 5 minutes. After washing twice with PBS, the cells were resuspended in 1 mL of cell lysis buffer [10 mmol/L HEPES (pH 7.9), 0.5% NP-40, 1.5 mmol/L MgCl2, 10 mmol/L KCl, and 0.5 mmol/L DTT] and kept on ice for 10 minutes. After centrifugation at 4,000 rpm for 5 minutes, the cell pellets were resuspended in nuclear lysis buffer [20 mmol/L HEPES (pH 7.9), 25% glycerol, 0.5% NP-40, 0.42 M NaCl, 1.5 mmol/L, and 0.2 mmol/L EDTA] containing protease inhibitors to extract nuclear proteins at 4°C for 20 minutes, and then the chromatin was sonicated into fragments with an average length of 1 kb. After centrifugation at 13,000 rpm for 10 minutes, the supernatants were diluted in an equal volume of dilution buffer containing 1% Triton X-100, 2 mmol/L EDTA, 20 mmol/L Tris-HCl (pH 7.9), 50 mmol/L NaCl, and protease inhibitors. Primer pairs used in the ChIP assay are listed in Supplementary Table S2. Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as the housekeeping gene. The sources of the antibodies used are as follows: anti-KDM5B (1:2000, Cell Signaling Technology, 15327S), H3K4me1 (Invitrogen, PA5-40087), H3K4me2 (Invitrogen, PA5-31912), H3K4me3 (Invitrogen, 49-1005), and anti-Histone H3 (1:2000, Cell Signaling Technology, 9715).

Six-week-old specific pathogen-free C57BL/6 mice were housed in polycarbonate cages in isolators and fed a commercial pellet diet with water ad libitum. Briefly, 3 × 10⁸ CFU of low-passage P. gingivalis was diluted in 3% (w/v) NaHCO₃ in PBS in a total volume of 0.5 mL, which was then administered orally into mice (n = 70) using a blunt feeding needle (Popper & Sons) four times at weekly intervals. After 2, 8, or 16 weeks, the mice were euthanized. To evaluate the protective effects of the antibodies, mice (n = 40) were rechallenged with 5 × 10⁷ CFU of P. gingivalis 2 weeks after the last infection. After 2, 8, or 18 weeks, the mice were euthanized, and sera specimens were collected and stored at −80°C until tested. The esophageal mucosa was processed for quantitative culture (placed in BHI broth, 400 μL), histopathology, and IHC (placed in formalin, 2 mL) as described below.

For histopathology, stored esophageal mucosa fixed in 4% neutral-buffered formaldehyde was embedded in paraffin, cut in longitudinal sections (4 μm), and stained by hematoxylin and eosin (H&E). Examination of the tissue sections for the histopathologic lesions, including the occurrence of infiltrates of polymorphonuclear and mononuclear cells, erosive lesions, and edema, was performed blind (by Jianqiang Mi and Hongbo Han, HUST). The scoring grades were defined as follows: 0, none; 1: mild, scant number of leukocytes was observed in the mucosa; 2: moderate, a moderate number of leukocytes was observed in the deep to mid mucosa and occasional neutrophils; 3: heavy, numerous infiltrates in the deep to mid mucosa, and one or two lymphoid aggregates; and 4: severe, the lamina propria and submucosa were completely populated by diffuse infiltrates.

Serial sections of 4-mm thickness were prepared and deparaffinized by submersion in ethanol and rinsing continuously in distilled water for 5 minutes. Antigen retrieval was performed in antigen retrieval Citra plus solution (BioGenex), and slides were blocked with 1.5% normal goat serum (Vector Laboratories) for 30 minutes. Preimmune rabbit IgG was used as a negative control. Polyclonal rabbit anti–P. gingivalis 33277 (5, 6) was incubated for detection of P. gingivalis (1:1,000 dilution) for 12 hours at 4°C, followed by biotin-conjugated secondary antibody for 1 hour at room temperature, streptavidin-peroxidase for 30 minutes at room temperature, and enzyme substrate (Dako). Sections were counterstained with hematoxylin and evaluated by two senior pathologists (Jianqiang Mi and Hongbo Han). Staining intensity was classified using a numerical scale: grade 0 (none, 0%–10% staining), grade 1 (weak, 10%–30%); grade 2 (moderate, 30%–60%), and grade 3 (strong, over 60%), with a score of ≥2 considered positive of staining. Serial sections were also labeled with anti–B7-H4 (Abcam, cat. #ab209242) and anti-KDM5B (NOVUS, cat. #NBP1-97310SS), CD4, human (Abcam, cat. #ab133616), CD4, mice (Abcam, cat. #ab183685), CD8, human (Abcam, cat. #ab4055), and CD8, mice (Abcam, cat. #ab22378). The number of immunostained cells in four fields per specimen was counted using a light microscope (Eclipse 80i, Nikon) with 400 × magnification. A quantitative description of each score is defined as follows: 1: 100 ± 20 cells/mm²; 2: 200 ± 40 cells/mm²; 3: 400 ± 80 cells/mm²; and 4: 800 ± 160 cells/mm².

Esophageal tissue was homogenized on ice for 80 seconds in 1 mL of PBS containing 0.05% Triton X-100. The resulting supernatant was isolated after centrifugation (10,000 × g) and serially diluted in sterile saline and plated on blood agar plates at 37°C under aerobic conditions (85% N₂, 5% H2, and 10% CO₂). P. gingivalis colonies were enumerated by the technique described above.

ESCC tumor samples or the esophageal mucosa of C57BL/6 mice with P. gingivalis challenge were processed into 20 μm–thick sections and washed thoroughly with PBS before being mounted onto microscope glass slides for RNAscope in situ hybridization. The mounted sections were then stored at –80°C for future analysis or immediately dehydrated in 50%, 70%, 95%, and 100% ethanol for 5 minutes each. The dehydrated sections were treated with hydrogen peroxide for 10 minutes at room temperature, immersed and boiled in RNAscope Target retrieval reagent (ACDBio; cat. #322000) for 15 minutes, and incubated with five drops of RNAscope Protease (ACDBio; cat. #322330) to entirely cover each section at 40°C in a HybEz oven (ACDBio) for 30 minutes. The sections were rinsed with ultrapure water between the steps. For probe hybridization, probes targeting different organisms were incubated with the sections and allowed to hybridize at 40°C in a HybEz oven for 2 hours. Four to five drops of probes were added to cover the entire section. P. gingivalis probes (cat. #532131), mouse probes (cat. #313911), and human probes (cat. #313901, used as positive controls) were all purchased from ACDBio. After signal amplification and detection, and counterstaining with hematoxylin, the probe-hybridized sections were examined under a light microscope (Eclipse 80i, Nikon), and bacterial colonization was evaluated by an unbiased observer according to the following semiquantitative scoring system: 0: no bacteria; 1: mild colonization, scant number of bacterial cells was observed in the mucosa and submucosa; 2: moderate colonization, a moderate number of bacterial cells was observed in the submucosa; 3: heavy colonization, numerous bacterial cells were observed in the submucosa; and 4: severe colonization, the submucosa was completely populated by bacterial cells.

Flat-bottom, 96-well microtiter plates (Corning) were coated with 10 μg/mL sonicated whole-cell extracts of P. gingivalis in PBS. Wells were incubated with serial dilutions of 100-μL aliquots sera (1:100) or esophageal secretions (1:10) followed by blocking with PBS (0.1% BSA). The wells were then incubated with the dilutions of alkaline phosphatase (AP)–conjugated goat anti-mouse IgG1, IgA, and IgG2a (Abcam), followed by the phosphatase substrate (Sigma-Aldrich) in ethanolamine buffer. Optical density (OD) readings were done at 405 and 492 nm with an ELISA plate reader (PerkinElmer). The readings for uncoated wells were subtracted from those of the respective test samples. Cutoff values for each antibody class and each antibody sample type were determined from the mean OD values ± 2 SD for the corresponding uninfected samples. Samples with OD readings greater than these cutoff values were considered positive for P. gingivalis–specific antibodies.

CD4⁺ and CD8⁺ T cells were purified from the peripheral blood of patients with cancer from whom PDXs had been previously generated. These CD4⁺ and CD8⁺ T cells were expanded in vitro via culturing in TexMACS medium (Miltenyi; cat. #130-097-196) with hIL2 (50 U/mL; Miltenyi; cat. #130-097-742) and hIL-7-Fc (70 ng/mL; Miltenyi; cat. #130-095-361), and then stimulated using Dynabeads CD3/CD28/CD137 human T-Activator (Gibco; cat. #11163D). The Th1/Th2 phenotypes of the expanded cells were evaluated via flow cytometry and immune labeled for CD4 (Miltenyi; cat. #130-113-775), CD25 (Miltenyi; cat. #130-113-282), CD69 (Miltenyi; cat. #130-112-612), CD8 (Miltenyi; cat. #130-104-130), and CD107a (Miltenyi; cat. #130-095-510). Unlike the original T-cell proportions, the numbers of CD4⁺CD25⁺ and CD8⁺CD107a⁺ T cells were substantially increased among the cultured cells. NSG mice were then sublethally irradiated with 200 cGy. We isolated patient peripheral blood CD34⁺ hematopoietic stem cells (HSC) using Ficoll-Paque Plus (GE Healthcare) hypaque centrifugation and a positive selection kit (Stem Cell Technology; cat. #17856). The HSCs were then expanded in stem-cell serum-free medium (Stem Cell Technology; cat. #09655) and transplanted (at the density of 4 × 10⁶ cells/mL) into 40 irradiated 2- to 4-month-old female mice via tail-vein injection. Thirty days after bone marrow transplantation, the proportions of peripheral human versus mouse CD45⁺ cells were monitored by FACS to assess reconstitution. We calculated chimerism as follows: percentage of CD45⁺ human cells/total cells (human CD45⁺ cells plus mouse CD45⁺ cells). Forty-five days after HSC transplantation, we implanted a 1 × 2 mm fragment of a PDX that was obtained from a patient with ESCC. This PDX was previously grown in the fourth mammary fat pad of each of the 40 NSG mice (20 mice per patient tumor) that were engrafted with the patient's CD34⁺ HSCs. When the diameter of the PDX reached 5 × 5 mm in mice with hCD45⁺ cells constituting over 30% of peripheral blood cells, the mice were dosed according to the schedule outlined in “In vivo tumor studies.” The tumor-bearing mice received intravenous infusions of the patient's expanded peripheral CD4⁺ and CD8⁺ T cells (2 × 10⁶ cells/mL) for an additional 2 weeks, after which we euthanized the mice, removed the tumors for volume determination, and allocated them for histology and FACS analysis of infiltrating leukocytes as described below.

One- to 2-mm fragments of P. gingivalis–positive PDXs were implanted subcutaneously into the flank region of humanized mice. After injection, the mice were randomly divided into different groups (n = 10/group). Mice were treated with CPI-455 (50 mg/kg or 70 mg/kg, daily, intraperitoneal injection; Selleck; cat. #S8287) and anti–B7-H4 (188; 500 μg/mouse, weekly, intraperitoneal injection; eBioscience; cat. #16-5972-81), followed by the sequential administration of CPI-455 and anti–B7-H4 Ab started on days 6 and 20, respectively; phased combined treatment for 14 days with CPI-455 started on day 6 and combined treatment with anti–B7-H4 Ab started on day 13 and continuing for 3 weeks; or extended phased combined treatment with CPI-455 for 28 days. The animals dosed according to the appropriate schema (n = 10 mice/group) were monitored daily for up to 2 months, and the objective response rate and survival were recorded.

An additional cohort of mice (n = 5/group) was included to conduct mechanistic studies. In this cohort, the mice were sacrificed on day 30 after tumor inoculation. Residual tumors were surgically removed before terminal escape (tumor with partial response, PR) or complete remission (tumor with complete response) and processed for IHC and flow cytometry analysis. IHC and flow cytometry results related to lymphocyte infiltration were determined, and a representative mouse from each treatment group [(i) animals receiving control therapy, (ii) animals receiving anti–B7-H4 Ab monotherapy, (iii) animals receiving 75 mg/kg CPI-455 monotherapy, and (iv) animals treated with extended phased therapy using 75 mg/kg dose CPI-455] in this separate cohort is shown, TV (mm³) = π/6 × length × width². Mice suffering from progressive disease or those used for subsequent analysis were euthanized when the TV was more than 2,500 mm³.

For studies on tumor rechallenging, 1- to 2-mm diameter pieces of tumor tissue taken from P. gingivalis–positive ESCC patients undergoing surgery or biopsy were implanted into the flank regions opposite of the primary implantation sites. Complete response was defined as complete response of the tumor without any recurrences. Primary and secondary challenge tumor growth was followed for up to 66 and 25 days, respectively.

ESCC PDXs were minced into small pieces followed by digestion with triple enzyme mixture containing collagenase type IV (Sigma-Aldrich), hyaronidase (Sigma), and deoxyribonuclease (Sigma) for 120 minutes at room temperature. After digestion, ESCC cells were washed twice in PBS and cultured in RPMI-1640 containing 10% human serum supplemented with L-glutamine (Sigma) and 2-mercaptethanol (Sigma) and IL2 (1,000 U/mL; Miltenyi; cat. #130-097-742) for the generation of T cells. Once T cells were released from tumor tissues, they were grown in high-dose IL2 medium for 1 week. These T cells were then transferred to a fresh well and grown in a low-dose IL2 (50 U/mL)-containing medium.

We used an adoptive T-cell transfer model, in which tumors from ESCC patients positive for P. gingivalis were orthotopically implanted into the mammary fat pad of NSG mice (n = 15/group) that were engrafted with the same patient's CD8⁺ T cells. Therapies, as indicated in the phased regimen, were started before T-cell transfusion. Carboxyfluorescein succimidyl ester (CFSE)–stained (10 μmol/L) CD8⁺ T cells were washed twice in PBS and then injected tail intravenously (7 × 10⁶/mouse in 200 μL of buffered saline) into tumor-bearing (tumor size about 0.5–1 cm) NSG mice. Mice were euthanized at 60 minutes 3, 5, 7, and 15 days after transfer, and tissues were harvested and digested into single-cell suspensions. Flow cytometry was carried out to examine the proportion of CFSE-positive cells.

Explants (10 mm × 10 mm) from the tumor of an ESCC donor patient were implanted subcutaneously into the flank of NSG mice at the start of the experiment. XenoLight DiR (Caliper Life Sciences) at a concentration of 320 μg/mL was added into a suspension of 7 × 10⁶ human CD8⁺ T cells from the same donor, and the mixture was incubated for 30 minutes and then washed, suspended in 0.2 mL of PBS, and injected intravenously via the tail vein into tumor-bearing NSG mice (n = 5–10/group). Therapies, as indicated in the phased regimen, were started before T-cell transfusion. Tumor volumes were measured every 3 days with calipers. Dorsal, right lateral, and ventral images of the mice were obtained at 1 hour and at 5, 7, and 15 days after the T-cell injection using an IVIS in vivo imaging system (Caliper). The fluorescent DiR signals were collected using a 710 nm (excitation)/760 nm (emission) filter set. The intensities of the signals were determined with Living Image 3.1 software (Xenogen) and expressed as photons/s/cm²/sr. Tumor tissues harvested from an additional cohort of mice (n = 5/group) at 7 and 15 days after transfer were used for mechanistic analyses. These tissues were formalin fixed, paraffin embedded and then processed for IHC staining for CD8 infiltration, which was calculated as CD8⁺ T-cell number/microscope field in the tumor tissues.

Statistical analyses were carried out using Prism 6.0 software (GraphPad Software) Macintosh version, and the data were expressed as mean ± standard error of the mean (SEM) unless noted otherwise. The differences in the median survival duration of the mice were determined by Kaplan–Meier survival plots and log-rank tests. Wilcoxon rank sum test was used for independent samples for analysis of significance. A P value of < 0.05 was considered statistically significant.

Our previous study demonstrated the presence of P. gingivalis in ESCC lesions (6). To clarify clinically relevant targets for host–bacterium interactions, we performed RNA-seq of ESCC cells in the presence or absence of P. gingivalis isolates, followed by in vivo screening of P. gingivalis–positive subjects. We identified 109 upregulated and 69 downregulated genes overlapping between Kyse-150 and Kyse-410 P. gingivalis–exposed cell lines, whereas the overlap with tumor profiles from P. gingivalis–positive subjects identified 40 upregulated and 21 downregulated genes exhibiting sustained expression differences in vivo (Supplementary Fig. S1A and S1B). These 61 differentially expressed genes formed our preclinical infection signature (Fig. 1A), from which the top-ranked genes were found to be involved in diverse immune processes (Fig. 1B). To identify the genes involved in the dynamic and heterogeneous immune responses that occur in response to infection in the ESCC microenvironment, we focused on the gene encoding B7-H4, an immune-checkpoint protein and drug target. We also identified epigenetic alterations in response to P. gingivalis infection, which encompassed several members of the JmjC family, including the H3K4 demethylase KDM5B, which encodes a targetable enzyme that functions as a repressor of target genes. P. gingivalis infection resulted in transcriptional silencing of a subset of genes, some of which encode chemokines and chemokine receptors. To validate the sequencing results, we assessed several key immune-related genes, including B7-H4 and KDM5B by immunoblots, qRT-PCR, and FACS and subsequently confirmed that P. gingivalis infection upregulated all 17 genes of interest (Fig. 1C and D; Supplementary Fig. S1C and S1D).

Cell line models might not always reveal relevant functions when cells are fully differentiated and in contact with the extracellular matrix. Therefore, we used cultured human ESCC explants to carry out a detailed study of chemokine responses during P. gingivalis infection using an array targeting 31 chemokines (Supplementary Table S3). Because using KDM5B inhibitors can stabilize bioactive chemokines in order to enhance adjuvant-based tumor immunity (17, 22), we used supernatants generated from ESCC explants in the presence or absence of P. gingivalis isolates, as well as P. gingivalis–exposed explants, in the presence of CPI-455, a neutralizing inhibitor of KDM5B for further analyses. As expected, explants, obtained from uninfected patients, were infected ex vivo with P. gingivalis. These explants showed reduction of a subset of overlapping chemokines, whose high concentrations were repeatedly observed in the respective explants pretreated with CPI-455 and infected with P. gingivalis. The array images of explants 1 (Fig. 1E), 2, and 3 (Supplementary Fig. S1E) are representative of results revealing similar patterns of expression. Array results related to chemokine concentrations in the representative explant 1 (Supplementary Fig. S2A–S2C), explant 2 (Supplementary Fig. S2D and S2E), and explant 3 (Supplementary Figs. S2F and S3A) are also shown, with CXCL11, CXCL9, and CXCL10 representing the top chemokines displaying rescued concentrations.

We then screened a library of JmjC histone demethylase inhibitors to assess their ability to increase the expression of chemokines associated with T cells following P. gingivalis exposure. We initially examined the effect of JIB-04, a pan-selective JmjC demethylase inhibitor, on chemokine expression in ESCC cells. As expected, treatment with JIB-04 led to higher CXCL11, CXCL9, and CXCL10 in Kyse-150, Kyse-410, and Kyse-180 cells (Supplementary Fig. S3B and S3C). Treatment with CPI-455 to selectively target H3K4-specific JmjC demethylases increased CXCL11, CXCL9, and CXCL10 following infection, with maximum levels observed 48 hours after infection. The concentrations of other chemokines, including CCL5 and CCL2, were unaffected (Supplementary Fig. S3B and S3C). These results identified KDM5B as a potential epigenetic regulator of the Th1-type chemokine response to P. gingivalis.

Next, we performed in vitro chemotaxis assays to identify the most relevant chemokines in the functional recruitment of effector T cells into tumors. Use of different chemokine-neutralizing antibodies revealed that those targeting CXCL11, CXCL9, and CXCL10 eliminated the chemotactic activity induced by supernatants from ESCC cells pretreated with CPI-455 and infected with P. gingivalis, with their combined use resulting in an additive effect (Supplementary Fig. S4A and S4B). These results suggested that secretion of these Th1-type chemokines was responsible for T-cell trafficking.

ChIP assays were then performed to identify the association between KDM5B and chemokine release. We used four conserved regions in each gene, which showed that KDM5B immunoprecipitated with all three genes analyzed (Supplementary Fig. S4C). ChIP assays performed following KDM5B inhibition resulted in recovery of a higher quantity of PCR products from the CXCL11, CXCL9, and CXCL10 promoter regions relative to that observed in controls. However, no detectable amplification of products was obtained from the CCL2 and CCL5 promoters (Fig. 2A).

To investigate the effects of KDM5B binding, we analyzed the histone H3K4 methylation status of the KDM5B-binding sites in control and KDM5B-neutralized cells. KDM5B neutralization reduced its binding and increased H3K4me3, confirming that the observed ChIP signals were specific to KDM5B activity. The methylation at H3K4me2 and H3K4me1 sites, which are located within the KDM5B-binding regions, was less affected. Therefore, we concluded that KDM5B neutralization only affected the H3K4 methylation of the KDM5B-binding site because H3K4me3 in genomic regions that do not contain the KDM5B-binding site was not affected by KDM5B neutralization (Fig. 2B).

To investigate whether KDM5B represses transcription via H3K4 demethylase activity, we overexpressed wild-type KDM5B and a KDM5B mutant that possesses defective demethylase activity in 293T cells, and determined the transcription of KDM5B targets using real-time reverse transcription PCR. We found that wild-type KDM5B did not significantly alter the expression of KDM5B target genes, whereas cells harboring a catalytically deficient KDM5B mutant displayed increased target gene expression (Supplementary Fig. S4D). To investigate whether the KDM5B mutant alleviates H3K4me3 demethylation, we analyzed the histone H3K4 methylation status at the KDM5B-binding site in the CXCL10 promoter in cells overexpressing wild-type or mutated KDM5B. The expression of the wild-type enzyme reduced H3K4me3, indicative of active KDM5B-mediated demethylation of H3K4me3, whereas overexpression of the KDM5B mutant increased H3K4me3 methylation (Fig. 2C). These results suggested that KDM5B-mediated transcriptional regulation antagonized H3K4 methylation and that KDM5B bound to the promoters of Th1-type chemokines to alter H3K4 methylation.

To investigate the significance of the presence of B7-H4 on CD8⁺ T cells in vitro upon P. gingivalis challenge, we investigated T-cell proliferation following coculture with P. gingivalis–infected ESCCs in the presence of a B7-H4–neutralizing antibody or isotype control. CD8⁺ T-cell proliferation decreased after coculture with P. gingivalis–infected Kyse-150 (Fig. 3A and B) and Kyse-410 (Fig. 3C and D) cells but was increased after B7-H4 neutralization. To correlate CD8⁺ T-cell proliferation with IL2 synthesis, coculture supernatants were harvested at 48 and 72 hours, and IL2 was measured via ELISA. CD8⁺ T cells cocultured with P. gingivalis–infected Kyse-150 cells produced 8.5- and 12-fold higher IL2 at 48 and 72 hours, respectively, after blocking B7-H4 and relative to levels observed in cocultures treated with isotype control (Fig. 3E), with similar results obtained from P. gingivalis–infected Kyse-410 cells (Fig. 3F). These results are in agreement with those obtained via the CFSE assay, suggesting that B7-H4 presentation in response to P. gingivalis infection may help suppress the proliferation of activated CD8⁺ effector T cells.

Six-week-old male C57BL/6 mice were chronically infected with P. gingivalis strain 33277 over a period of months. The mice mounted a robust inflammatory response, which, however, was not sufficient to eliminate the bacterium (Supplementary Fig. S5A and S5B). We subsequently analyzed variability in treatment responses according to a 14-day phased schedule and compared immunopathology and bacterial colonization at intervals following inoculation. Inflammation was histologically evaluated in the esophageal mucosa from mice in the sham-treated group, as well as those receiving individual treatment with either CPI-455 or B7-H4–neutralizing antibody or combinatorial treatment with both agents. Histopathology analysis revealed no inflammation in either group at 2 weeks in response to the primary infection. However, at 8 weeks after inoculation, mice receiving monotherapy exhibited mild inflammation, whereas the combined treatment presented with heavy to severe inflammation, which persisted at 12 and 16 weeks after challenge (Fig. 4A). Evaluation of T-cell density indicated low infiltration in the mucosa from mice that received the combined treatment at 2 weeks after infection. However, massive infiltration of T cells (mostly CD8⁺) was observed at 8 weeks after infection, whereas few T cells were observed in the mucosa of mice receiving monotherapy (Fig. 4B and C). Leukocyte infiltration was mostly localized to the lamina propria, deep mucosa, and submucosa (Fig. 4C). We also observed continued increases in CD8⁺ T cells in the combined treatment group relative to the group receiving monotherapy (Fig. 4D).

Subsequently, we explored the effects of enhanced inflammation on P. gingivalis colonization. At 2 weeks after inoculation, the degree of colonization was comparable across the different groups as assessed by RNAscope in situ hybridization (Fig. 4E) and quantitative culturing (Fig. 4F). However, at 8 weeks after infection, mice that received the combined treatment displayed protection against infection rechallenge and a significant reduction in colonizing bacteria compared with mice that received monotherapy, with similar protection observed in the combined treatment group at 12 and 16 weeks after challenge. Assessment for P. gingivalis mRNA using RNAScope revealed that positive staining is diffusely present in the cytoplasm, with a punctate or granular appearance, and positive signals were detected in both the mucosa and submucosa. Rechallenge of mice with twice the original bacterial load and subsequent verification of colonization-free status in the combined treatment group confirmed a memory response against P. gingivalis (Supplementary Fig. S5C).

To determine how the phased or sequential regimen affected the efficacy of combined therapy, we implanted a PDX from a P. gingivalis–infected tumor derived from a surgical ESCC specimen into NSG mice engrafted with CD34⁺ HSCs from the corresponding patient, followed by subsequent infusion of the patient's CD4⁺ and CD8⁺ T cells. Concurrently, a cohort of mice was treated with CPI-455 and anti–B7-H4 using a phased or sequential regimen (Supplementary Fig. S6A). For these studies, mice were treated with CPI-455 for 14 days, and anti–B7-H4 was administered either after the CPI-455 treatment (sequential regimen) or 7 days after the start of CPI-455 treatment (phased regimen). Our results showed that monotherapy delayed tumor growth, though rarely induced complete remission in these mice (Supplementary Fig. S6B, a–b). However, increased response was observed using the phased schedule, with an increased number of mice exhibiting complete response (Supplementary Fig. S6B, c). Mice treated using the phased schedule showed a synergistic response and significantly increased median survival compared with that of mice receiving CPI-455 monotherapy (35 vs. 26 days, respectively; Supplementary Table S4). Delayed administration of anti–B7-H4 allowed tumors to develop resistance to treatment. However, in the groups treated using delayed administration of anti–B7-H4, complete response was observed in 2 mice that survived symptom free until the end of the study (30 days after receiving the final dose of anti–B7-H4; Supplementary Fig. S6B, c; Supplementary Table S4). We also observed an improved response in the sequential schedule, suggested by the increase in the median survival time from 21 days (the median survival time of mice treated with anti–B7-H4) to 27 days (Supplementary Fig. S6B, d; Supplementary Table S4).

In a third dosing schedule, animals were treated for 28 days with the two doses of CPI-455 (Supplementary Fig. S6C). The higher dose of CPI-455 monotherapy (Fig. 5A, b) was more efficacious than the loser dose (Fig. 5A, a) in this extended treatment schedule. All the mice were responsive to the treatment, with one complete response observed, whereas mice receiving 50 mg/kg CPI-445 alone showed only a moderate (grade 2) response (Fig. 5A, a and b). Anti-B7-H4 monotherapy also resulted in less significant (grade 2) response (Supplementary Fig. S6D). However, both doses of CPI-455 were demonstrated to be effective when used with anti–B7-H4 in a phased regimen (Fig. 5A, c–f). Although the phased combination treatment for 14 days with 75 mg/kg CPI-455 increased the median survival to 44 days (vs. 31 days for 75 mg/kg CPI-455 monotherapy), the combined treatment led to an increased ratio of complete response (ratio of complete response: complete response/total mice; 30% vs. 10%; Fig. 5A, d; Supplementary Table S5). Although the extended phased regimen with 50 mg/kg CPI-455 (Fig. 5A, e) exhibited a similar ratio of complete response with the 14-day phased therapy with 75 mg/kg CPI-455 (Fig. 5A, d), the extended phased regimen produced increased median survival compared with that obtained using the shorter phased therapy from 44 to 50 days (Supplementary Table S5). The phased combination of the anti–B7-H4 treatment and the higher dose of CPI-455 showed significantly higher antitumor activity, generating an approximately 2-fold increase in the median survival (31 vs. 63 days), and 50% of the animals achieved complete responses compared with 10% of the mice treated with the 75 mg/kg CPI-455 monotherapy (Fig. 5A, f; Supplementary Table S5). To test the longevity of the antitumor response, surviving animals displaying a complete response were rechallenged with tumor cells 1 month after the end of the 28-day phased treatment. Thirteen completely remitted mice were protected from a second tumor challenge, suggesting that these mice developed immunologic memory (Supplementary Table S5).

To address the mechanisms associated with monotherapy or combined treatment, we analyzed the human lymphocyte density in tumors harvested from the mice. FACS (Fig. 5B) of tumor-infiltrating cells revealed an increase in the CD4⁺ and CD8⁺ T-cell populations after the combined treatment compared with those following sham treatment or monotherapy. IHC analysis (Fig. 5C and D) revealed significant (grade 4) lymphocyte infiltration progressing inward from the outer layers to the central regions of the tumors, with more intense staining and clearer infiltration observed at the central regions of the lesions from mice that received combined treatment relative to that from mice that received monotherapy.

We further applied the phased regimen to a model involving the adoptive transfer of autologous tumor-infiltrating lymphocytes (TIL) to examine the effect of treatment on T cells. After tumor inoculation and drug administration, CFSE-labeled CD8⁺ T cells were transferred into the mice carrying xenograft tumors. At 1 hour and 5, 7, and 15 days after transfer, cells obtained from PDXs were harvest and gated by CFSE fluorescence for analysis of proliferation. We observed no significant division by the T cells at 1 hour. However, on days 5 and 7, T-cell expansion was observed following anti–B7-H4 monotherapy or combined treatment, as well as following CPI-455 monotherapy, albeit to a far lesser degree than that observed following anti–B7-H4 therapy. On day 15, the majority of the CD8⁺ T cells had undergone at least one cycle of division in response to anti–B7-H4 monotherapy and combined treatment, whereas some of the T cells remained inactivated after CPI-455 monotherapy, as well as those that were untreated (Fig. 6A). This suggested that targeting B7-H4 facilitated the proliferation of adoptively transferred CD8⁺ T lymphocytes and improved antitumor responses.

Because KDM5B neutralization promoted T-cell chemotaxis in vitro, we validated this observation in vivo via the adoptive-transfer model and the phased regimen. CD8⁺ T cells were stained with DiR and injected into mice. For the imaging of T-cell trafficking and distribution, the mice were dorsally, right laterally, and ventrally viewed at 1 hour and 5, 7, and 15 days after T-cell injection. Shortly after injection (less than 3 days), T cells were randomly distributed in the spleen, lymph nodes, and tumor beds following treatment with each regimen. After more than 7 days, the number of T cells in tumor tissues increased following CPI-455 and combined therapy, with this accumulation persisting throughout the observation period of 15 days. We found that fewer CD8⁺ T cells accumulated in tumor sites following anti–B7-H4 monotherapy, with the cell number further decreasing at later time points in some cases (Fig. 6B and C). Subsequently, we confirmed via IHC that a large amount of CD8⁺ T cells accumulated in tumor tissues from the groups receiving CPI-455 and combined therapy, but not for those receiving sham treatment or anti–B7-H4 monotherapy (Fig. 6D and E). These results suggested that KDM5B neutralization in vivo promoted CD8⁺ T-cell trafficking into tumor beds, thus improving the antitumor activity of adoptively transferred T lymphocytes.

Our results indicated that tumors in mice that were not subjected to adoptive T-cell transfer grew more rapidly than those that were treated with adoptively transferred T cells. Without adoptive transfer, in most mice, treatment with both agents led to tumor growth comparable to either single-agent treatment (Supplementary Fig. S7A and S7B).

Finally, we evaluated the relationship between the concurrent expression of B7-H4 and KDM5B, the load of P. gingivalis, and CD8⁺ T-cell count in ESCC tissues. Biopsies revealed that concurrent B7-H4 and KDM5B expression was detected more frequently in tissues from P. gingivalis–infected patients than that in tissues from uninfected patients (Fig. 7A; Supplementary Table S6). The simultaneous expression of B7-H4 and KDM5B in ESCCs correlated more significantly with the bacterial load than with their individual expression (Supplementary Fig. S8A). We also found a negative correlation between coexpression of the two molecules and CD8⁺ T-cell count in ESCCs (Fig. 7A; Supplementary Fig. S8B).

Cancer-induced immunosuppression allows tumors to evade immune surveillance, resulting in resistance to immunotherapy (23). Successful human pathogens have evolved numerous mechanisms to tolerate adverse conditions in hosts and to evade immune responses elicited upon infection (7, 24). P. gingivalis is an anaerobic gram-negative, rod-shaped bacterium, and much of its pathogenicity is a result of overall immunosuppression of the host cells (4, 25). This pathogenic mechanism may stem from its abilities to manipulate complement–Toll-like receptor cross-talk (26), myeloid-derived suppressor cell expansion (27), Th17/regulatory T-cell (Treg) imbalance (28), macrophage responsiveness (29), microbiota dysbiosis (30), and IFNγ-inducible chemokines release (31) in a manner that promotes periodontitis and related systemic diseases such as atherosclerosis, insulin resistance (32), and Alzheimer disease.

We previously identified P. gingivalis in nearly half of the ESCC lesions and made the clinicopathologic association with this deadly disease (5, 6). Thus, a hypothesis indicating that P. gingivalis ensures reduced recognition, as well as its persistence, in the epithelium by manipulation of key immune-sensing molecules and related signaling pathways is logical. Such sophisticated immune-escape tactics enable pathogens to evade host responses, subsequently increasing immunosuppression and facilitating poor immunogenicity. This cascade renders the mucosa more vulnerable to malignant transformation when encountering hypermutations and allows the progression of the precursor lesions for invasive cancer.

Immunotherapies, such as checkpoint blockade, are rapidly changing the standard treatment and outcomes for patients with advanced malignancies (33). However, improvements occur in only a fraction of patients, as not all cancers respond to immunotherapy, especially those exhibiting poor immunogenicity (34). Microbes have become increasingly recognized as key modulators of host immunity (35), raising the possibility that they may account for the varied responses and failures often observed during immune-checkpoint blockade. In this study, we examined how P. gingivalis contributes to the suppression of systemic and mucosal immunity in the host. Our findings suggest that P. gingivalis may mediate the equilibrium between promoters and suppressors of anticancer immunity in ESCC-bearing individuals in ways that shape the response to cancer immunotherapy. Therefore, manipulating the gut microbiota to prevent tolerance to immune-checkpoint inhibitors might be feasible. The ability of B7-H4 neutralization to restore immunogenicity to P. gingivalis–infected, poorly immunogenic ESCCs remains an important research focus. Therefore, incorporating these therapies into more powerful combinations is the next logical step.

Chemokines regulate leukocyte trafficking, thereby making them attractive candidates for facilitating endogenous immune priming and T-cell infiltration for improved immunotherapeutic efficacy (36). HDACi and DNMTi promote the secretion of the epigenetically repressed chemokines and the subsequent trafficking of immune cells into tumor beds (16, 17). Thus, an epigenetic modulation that activates the immune system in a synergistic manner can improve immunotherapy outcomes. Unfortunately, the use of posttranslational modifications to modulate immunity has not been extensively studied (22). KDM5B functions as a repressor by catalyzing demethylation of H3K4me3 modifications in target gene regions. Although the demethylation activity of KDM5B has been well demonstrated clinically, its significance in the regulation of host immunity has rarely been investigated. Accordingly, KDM5 inhibitors are potentially efficacious in the context of cancer stem cells and drug tolerance heterogeneity (37, 38). Our present study showed that KDM5B participated in epigenetic modification of chemokine expression as a strategy for enhancing tumor immunogenicity.

We also investigated individual or combined blockade of immunomodulatory molecules B7-H4 and KDM5B for enhancing tumor immunogenicity. NSG mice permitted both the transfer of T cells and the grafting of human tumors, thereby allowing the rapid progression of subcutaneous tumors. However, the use of this model is debated due to limitations associated with robust alloreactivity. In our investigations, we used allogeneic T cells. Both anti–B7-H4 and CPI-455, as single agents, controlled allogeneic tumors, and their combined use did not consistently produce better results. This notion can be demonstrated by ESCC PDXs used in more sophisticated reconstitution models involving the repopulation of mice with mature lymphocytes or CD34⁺ cells that can differentiate into lymphocytes, followed by the tracking of xenograft fate under different treatment regimens. Within this study's limited scope, we found that concurrent or sequential combined administration of anti–B7-H4 and CPI-455 did not differ significantly from that with monotherapy in generating an antitumor response, which suggests that the efficacy of the combined treatment needs to be optimized. Although we observed modest responses to monotherapy, multiple doses of CPI-455 were beneficial when used with anti–B7-H4 in the phased regimen. Specifically, the extended phased regimen using CPI-455 at a higher dose increased complete response by approximately 2-fold, and these constituted long-lasting remissions, as the mice remained tumor free after total regression. These results suggest the importance of evaluating different treatment combinations and regimens to identify synergistic or antagonistic effects. Our demonstration of delayed tumor growth following individual or combined administration of antibodies supports their continued development in ongoing clinical trials (NCT02437136 and NCT02546986).

Whether neutralization of B7-H4 and KDM5B results in similar types of immune impairment or differential effects on T-cell function remains unknown. The respective roles of these proteins are nonredundant, and their pattern of coexpression results in reduced efficacy. This phenomenon was illustrated in this study by the synergistic enhancement in tumor regression and reduction in bacterial load following dual blockade relative to the results following monotherapies. In addition, the simultaneous neutralization of B7-H4 and KDM5B did not significantly improve T-cell proliferation or lymphocyte trafficking compared with those obtained via blocking of B7-H4 or KDM5B alone. Therefore, the neutralization of different immunoregulatory factors triggers a differential restoration of immunity, indicating that synergistic effects might be characteristic of only specific cytokines. We suggested that targeting of B7-H4 enhanced tumor immunogenicity by preferentially boosting T-cell expansion, division, and development. Therefore, anti–B7-H4 therapy worked at the cellular level to enhance effective CTL-mediated killing of infected cells. By contrast, KDM5B neutralization altered the tumor microenvironment by favoring the attraction and accumulation of lymphocytes in the tumor bed, thereby “waking up” the T cells in the tumor and activating and sensitizing the immune system to respond to immunotherapy (Fig. 7B). Our results highlight the biological significance of B7-H4– and KDM5B-related pathways involved at various levels in lymphocytes during P. gingivalis infection.

No potential conflicts of interest were disclosed.

Conception and design: X. Yuan, F. Zhou, S. Gao

Development of methodology: Y. Liu, G. Li, M. Ma, H. Li, J. Kong, J. Sun, X. Hou, Y. Ma, F. Ren

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Liu, J. Kong, J. Sun, X. Hou, F. Ren

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Li, H. Li, J. Kong, J. Sun, G. Hou, F. Ren

Writing, review, and/or revision of the manuscript: X. Yuan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Liu, G. Li, Z. Lan, M. Ma, J. Sun, X. Hou, Y. Ma

Study supervision: Y. Ma, F. Zhou, S. Gao

The authors gratefully thank Dr. Chen at the University of Pittsburgh (Pittsburgh, PA) and Dr. Cao at the University of California Los Angeles (Los Angeles, CA), for their help in experimental technologies and for critically editing a draft of this article. The P. gingivalis ATCC 33277 strain was a generous gift from Professor Huizhi Wang and Richard J. Lamont at the University of Louisville. The ESCC cell lines were generous gifts from Dr. Shimada at Kyoto University. This work was supported in part by grants from the National Natural Science Foundation of China (U1404817 and 81702820, to X. Yuan; 81472234, to S. Gao), Major Projects of the Science and Technology Department in Henan Province (161100311200; to S. Gao), Young Academic Leaders of HUST (to X. Yuan), China Postdoctoral Science Foundation (171234; to X. Yuan), and funds from the Henan Education Department (16A320038; to X. Hou).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.