HFB301001, an OX40-based immunotherapy, drives Treg clearance and CTL activation through optimized OX40 receptor clustering

Abstract

Background

OX40, a key co-stimulatory receptor that amplifies T cell-mediated anti-tumor immunity, is a promising immunotherapeutic target. Despite most reported OX40 agonists in clinical trials having high affinity, the relationship between affinity and agonistic activity remains complex, necessitating further clarification of affinity’s impact on OX40-based immunotherapy efficacy and its underlying mechanisms.

Methods

We generated the different affinity OX40 agonist antibodies were generated by phage display. Antibody-receptor interactions were modeled using AI-based prediction and validated by hydrogen-deuterium exchange. We assessed the receptor clustering, T cell activation, and regulatory T cell (Treg) depletion effect of OX40 antibodies with different affinities by confocal microscopy and reporter cell assays. We further evaluated the anti-tumor efficacy in multiple murine tumor models. The effects of HFB301001 treatment on tumor-infiltrating T cells, safety in cynomolgus monkeys, and immune activation in clinical samples were investigated using single-cell RNA sequencing (scRNA-seq), flow cytometry, ELISpot, and immunofluorescence.

Results

We identified the low-affinity OX40 agonist antibody HFB301001 and generated variants with different affinities via phage display. Compared with its high-affinity mutant, HFB301001 induced stronger receptor clustering, enhanced T cell activation, and mediated more potent natural killer-mediated antibody-dependent cell-mediated cytotoxicity for Treg depletion than its high-affinity mutant in vitro. Consistently, HFB301001 outperformed the high-affinity mutant by boosting intratumoral T cell infiltration/activation and Treg clearance in vivo. Additionally, HFB301001 exhibited favorable safety in cynomolgus monkeys and effectively activated tumor-infiltrating T cells in a clinically relevant tumor slice culture system.

Conclusions

The reduced-affinity strategy represents a promising framework for the clinical development of OX40-targeted cancer immunotherapies. Currently, HFB301001 is concluding in a phase I clinical study involving patients with advanced solid tumors (NCT05229601).

WHAT IS ALREADY KNOWN ON THIS TOPIC.

WHAT THIS STUDY ADDS

• The low-affinity OX40 agonist HFB301001 promotes stronger receptor clustering and enhances T-cell activation, furthermore facilitating more pronounced Treg depletion than high-affinity OX40 agonists both in vitro and in vivo.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This study introduces a reduced-affinity strategy for optimizing OX40-targeted immunotherapy. Unlike previous high-affinity OX40 agonists in clinical trials, our low-affinity antibody promotes stronger T-cell activation and Treg depletion, providing new insights into OX40 biology and accelerating the development of OX40-targeted cancer immunotherapies.

Introduction

Immune checkpoint blockade (ICB), including antibodies against programmed cell death protein 1 (PD-1)/programmed death-ligand 1 (PD-L1) and cytotoxic T-lymphocyte associated protein 4 (CTLA-4), has transformed cancer therapy by modulating the tumor microenvironment (TME).1,3 However, only a subset of patients benefits from this treatment, and many eventually experience a relapse.^{4 5} Direct stimulation of effector T cells therefore represents an alternative strategy that may help to overcome both primary and acquired resistance to ICB.^{6 7}

OX40 (CD134, TNFRSF4), a co-stimulatory receptor, is primarily expressed on activated CD4⁺ and CD8⁺ T cells. In the TME, OX40 is mainly found on CD4⁺ T cells, especially Tregs.^{8 9} Engagement by its ligand, OX40 ligand (OX40L), or by agonist antibodies delivers potent signals that enhance T cell proliferation, effector function, and memory formation.^{10 11} Agonist OX40 antibodies exert a dual mechanism of action by amplifying effector T cell responses and depleting intratumoral Tregs via Fc gamma receptor (FcγR)-mediated antibody-dependent cell-mediated cytotoxicity (ADCC),12,14 a dual mechanism supported by early clinical studies for solid tumors showing enhanced tumor-reactive T-cell activation and immune remodeling in patients treated with OX40 agonists.15,19

Unlike conventional inhibitory antibodies that block ligand-receptor interactions, agonistic immunomodulatory antibodies function through distinct mechanisms.^{20 21} In addition to target engagement, they must induce receptor clustering to activate downstream signaling pathways.^{22 23} Holay et al demonstrated that the hexavalent OX40 agonist INBRX-106 achieves superior activity compared with bivalent antibodies by promoting enhanced receptor clustering by multimeric binding.²⁴ Yu et al reported that lower-affinity agonist antibodies targeting CD40, 4–1BB, and PD-1 can paradoxically elicit stronger T cell activation, challenging the assumption that higher affinity ensures better efficacy.²⁵ However, the impact of antibody affinity on the anti-tumor activity of OX40 agonists remains incompletely understood and warrants further investigation. Moreover, although OX40 agonist antibodies can also deplete Tregs, whether higher affinity influences this effect remains to be elucidated.

Here, we describe an affinity-tuning strategy designed to optimize OX40-targeted immunotherapy. We developed HFB301001, a low-affinity, non-competitive antibody that binds OX40 without disrupting OX40L interaction. HFB301001 promotes enhanced receptor clustering, which translates into more robust T cell activation, Treg depletion, and antitumor activity compared with high-affinity counterparts. This study supports reduced-affinity engineering as a potential strategy to enhance OX40-targeted immunotherapy.

Materials and methods

Cell culture

The mouse colon cancer MC38 and E.G7 cell lines were purchased from Shanghai Model Organisms Center. The OX40 NFκB-Luc Jurkat cell line (Cat: CBP74061, Cobioer, China) was used to evaluate the agonist activity of OX40 antibodies. These cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM, Cat: PM150210, Pricella, China) or RPMI-1640 (Cat: PM150110, Pricella, China) medium, with 10% fetal bovine serum (Cat: 16421050, Pricella, China), 10 mM HEPES (Cat: PB180325, Pricella, China), and 1% penicillin/streptomycin (Cat: PB180120, Pricella, China). Peripheral blood mononuclear cells (PBMCs) were isolated from a healthy donor as described in the operation manual (Cat: P8610, Solarbio, China) and cultured in the natural killer (NK) cells serum-free culture kit (Cat: NC0102.F, Yocon, China).

Mice

The female C57BL/6J mice were obtained from Beijing Vital River Laboratory Animal Technology. The humanized OX40 (hOX40) mice (Cat: NM-HU-00041) and Fcgr3 KO mice (Cat: NM-KO-190189) were procured from the Shanghai Model Organisms Center. All experiments were conducted using mice aged between 6 and 8 weeks. Animals were housed under specific pathogen-free conditions at the Nankai University animal research facility, with an ambient temperature maintained at 21°C–22°C and a 12-hour light/dark cycle (lights on from 06:00 to 18:00).

Phage display screening

A phage display Fab library was constructed in order to randomly diversify two positions of the complementarity-determining region 3 (CDR3) region of both Variable region of the heavy chain (VH) and Variable region of the light chain (VL), as described previously.²⁶ The introduction of mutations to the CDR3 region was achieved through the utilization of PCR primers comprising degenerate codons MNN (where N=A/C/G/T, M=A/C) at specific positions. The amplified VH-CH1 and VL-CL genes were cloned into the pComb3XTT vector and introduced into E. coli XL1-Blue cells, and the size of the library was determined to be 10⁸.

For the biopanning procedure, phage particles were incubated with biotin-conjugated human OX40 protein (Cat: TN4-H82E4, ACRO Biosystems, USA) at room temperature for 2 hours. The resulting phage–antigen complexes were isolated using streptavidin-coated Dynabeads M280 (Cat: 11 205D, Invitrogen, USA). The release of bound phages was achieved through the utilization of an acid elution method, employing a solution of 0.1 M glycine-HCl (pH 2.2) for 10 min, followed by a neutralization step using 1 M Tris-HCl (pH 8.0) to achieve a final pH of approximately 7.5. Plasmid DNA from the enriched phage pool was extracted using a TIANprep mini plasmid kit (Cat: DP103-03, TIANGEN, China) for subsequent analysis.

Structure prediction of hOX40-HFB301001 complex

The structure prediction of the hOX40-HFB301001 complex was performed using EnsembleFold, a recently developed framework for biomolecular structure modeling built on the CASP15-winning algorithm DMFold.^{27 28} By integrating DeepMSA2 for high-quality multiple sequence alignment (MSA) generation with a modified AlphaFold3 prediction engine, EnsembleFold enables accurate and efficient modeling of biomolecular complexes.²⁹

The first step of EnsembleFold involves MSA construction via DeepMSA2, which performs homology searches against an extensive collection of genomic and metagenomic databases containing over 40 billion sequences. For monomeric targets, DeepMSA2 employs three parallel search pipelines (dMSA, qMSA, and mMSA), each performing iterative database searches to achieve sufficient alignment depth. Up to 10 MSAs are generated and then ranked using a simplified AlphaFold3 module, and the alignment yielding the highest predicted Local Distance Difference Test (pLDDT) score is selected as the final alignment. For multimeric MSA construction, DeepMSA2 first produces monomeric MSAs for each chain, followed by MSA pairing, sequence linking, and multimeric MSA selection. In the case of homomeric complexes, monomeric MSAs are concatenated directly. For heteromeric complexes with $N$ distinct component chains, the top $M$ MSAs for each chain are paired to form $M^N$ hybrid alignments, with $M$ adjusted to balance computational cost. Sequence linking is then guided by shared species information. Finally, the top 25 multimeric MSAs are selected for heteromeric complexes based on a composite score combining the depth of the MSAs and the pLDDT score of the monomer chains.

In the second step, structural modeling is carried out using a modified AlphaFold3 structural module, featuring key architecture enhancements for comprehensive and accurate modeling of biomolecular complexes. Specifically, the original Evoformer module is replaced with a simplified Pairformer, which enhances computational efficiency and mitigates reliance on extensive MSAs. Additionally, the conventional invariant point attention mechanism is substituted with a diffusion-based structure module, which predicts atomic coordinates through a denoising diffusion process. This design broadens the applicability of the model to chemically diverse complexes without imposing torsion-based parameterization or stereochemical limitations. Further refinements include a specialized training strategy aimed at minimizing structural hallucinations and a diffusion rollout protocol for estimating model confidence.

Within the EnsembleFold pipeline, the MSAs generated in the first step are fed directly into AlphaFold3 in place of its default sequence alignments. The AlphaFold3 modeling engine is further customized with several modifications: (1) optional use of structural templates, (2) tunable dropout rate, (3) increased diffusion sampling, (4) an optional early-stopping scheme, and (5) increasing the number of recycling iterations. Final model selection is based on a composite score combining the predicted TM-score (pTM), interface-predicted TM-score (ipTM), a disorder region reward, and a clash penalty.

$$\begin{array}{c}0.8\cdot \text{i}\text{p}\text{T}\text{M}+0.2\cdot \text{p}\text{T}\text{M}+0.5\cdot \text{d}\text{i}\text{s}\text{o}\text{r}\text{d}\text{e}\text{r}-100\cdot \text{h}\text{a}\text{s}\text{\_}\text{c}\text{l}\text{a}\text{s}\text{h}\#\left(1\right)\end{array}$$

In this study, a total of 2,000 hOX40–HFB301001 complex models were generated using EnsembleFold with sixteen random seeds across twenty-five MSAs. The top-ranked model was subsequently selected to define the epitope region based on changes in accessible surface area (ASA) on antibody binding. Specifically, the ASA of each residue was calculated using the program DSSP for both the bound and unbound states, and the difference ($∆ASA$) was obtained.³⁰

$$∆ASA=AS{A}_{unbound}-AS{A}_{bound}$$

A residue was then defined as an epitope residue if its $∆ASA$ is larger than 10${\mathrm{\AA }}^2.$

Hydrogen/deuterium exchange mass spectrometry

Hydrogen/deuterium exchange mass spectrometry (HDX-MS) was employed to map the epitope on the His-tagged human OX40 extracellular domain (Cat: 10481-H08H, Sino Biological, China), which is recognized by monoclonal antibody HFB301001. Purified hOX40 (2 mg/mL), either alone or in complex with an excess of HFB301001, was incubated in deuterated Phosphate-buffered saline (PBS, pH 7.4) for 300–3600 s at 20°C. The reaction was quenched with a prechilled buffer (final pH ~2.5) and subsequently digested online using a pepsin/protease XIII column at 0°C. Peptides were separated by ultra-performance liquid chromatography and analyzed using a Q Exactive Plus Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo Scientific). Data processing was performed with HDX WorkBench (v3.3), and differences in deuterium uptake of at least 5% with p<0.05 (unpaired t-test) were considered significant. Peptide coverage exceeded 95%, with a back-exchange rate of less than 30%, ensuring reliable epitope identification.

Construction of OX40 NF-κB-GFP Jurkat, cynomolgus monkey OX40 NF-κB-GFP Jurkat, and HEK293T-OX40-GFP

Jurkat or HEK293T cells were transfected with the pCMV-OX40-NF-κB-GFP, pCMV-cynomolgus monkey OX40-NF-κB-GFP, or pLV2-UBC-OX40-GFP plasmid, alongside the respective packaging plasmids psPAX2 (Cat: P0261, MiaoLingPlasmid, China) and pMD2.G (Cat: P0262, MiaoLingPlasmid, China) in a 6:3:1 ratio. Following a 48-hour co-transfection, a supernatant containing the lentivirus was collected and subsequently concentrated. The collected lentivirus was used to transduce Jurkat or HEK293T cells, and green fluorescent protein (GFP)-positive cells were isolated by flow cytometry using the BD FACS Aria Fusion instrument 24 hours after transduction.

Surface plasmon resonance assays

The binding affinity and kinetics of protein–protein interactions were determined by surface plasmon resonance (SPR) using a CM5 sensor chip (Cat: 10345041, Cytiva, USA) on a Biacore T200 instrument (Cytiva, USA). A mouse monoclonal anti-human IgG Fc antibody (Cat: ab99757, Abcam, USA) was immobilized on the CM5 biosensor chip. During each cycle, test antibodies were captured onto the chip surface at a flow rate of 10 µL/min. Subsequently, the recombinant human OX40 protein (Cat: OX0-H5224, ACRO Biosystems, USA) was injected over the chip surface at a flow rate of 30 µL/min for 120 s and dissociated for 300 s for measurements. Before the next cycle, the chip was regenerated using 3 M magnesium chloride (Cat: BR100839, Cytiva, USA) at a flow rate of 30 µL/min for 30 s. All data were processed using double referencing by subtracting signals from both a reference flow cell and buffer-only injections. Binding data were globally fitted to a 1:1 Langmuir binding model using Biacore T200 Evaluation Software to derive kinetic and affinity parameters.

Antibody activation assay

The OX40 NF-κB-Luc Jurkat cell line was seeded into 96-well plates. Cells were treated with different affinity OX40 agonist antibodies in the presence or absence of a goat anti-human IgG-Fc secondary antibody (2.5 µg/mL, Cat: SSA015, Sino Biological, China), and incubated at 37°C for 5.5 to 6 hours. D-Luciferin, sodium salt (Cat: 40 901ES03, Yeasen, China) was then added, and the plate was further incubated at room temperature. Luminescence was then measured using a microplate reader.

Tumor-draining lymph nodes (TdLNs) derived from lung cancer patients were minced into small fragments and enzymatically digested into single-cell suspensions. The cells were stimulated for 48 hours with a primary antibody at 10 µg/mL, a goat anti-human IgG-Fc secondary antibody (Cat: SSA015, Sino Biological, China), and OKT3 (anti-CD3) (Cat: CDE-M120a, ACRO Biosystems, China) at 1 µg/mL. After stimulation, the single-cell suspensions were prepared and subjected to flow cytometry analysis.

Antibody preparation

Antibodies were produced by transient transfection of HEK293F suspension cells. Briefly, cells were co-transfected with equivalent amounts of plasmids encoding the heavy and light chains of the antibody. Following transfection, cells were cultured for 5 days in a humidified incubator at 37°C with 5% CO₂ and 125 rpm agitation. The culture supernatant was collected and filtered through a 0.45 µm membrane to remove cell debris. Antibodies were subsequently purified using a HiTrap Protein A affinity column (Cat: 17040301, Cytiva, USA) on an ÄKTA purifier system.

Co-culture assay

A Jurkat cell line that had been genetically modified to express the high-affinity FcγRIIIa V158 variant and an NFAT-driven luciferase reporter (FcγRIIIa NFAT-luc Jurkat reporter cell line) was used as effector cell. Spleen-derived Treg cells were isolated from hOX40 mice using a mouse CD4⁺ CD25⁺ Regulatory T Cell Isolation Kit (Cat:130-091-041, Miltenyi Biotec, Germany) according to the manufacturer’s protocol and used as target cells. The effector and target (E:T) cells were cocultured at an E:T ratio of 1:10 in a 96-well plate. Following a 24-hour incubation with HFB301001 at 37°C in a humidified CO₂ incubator, luciferase activity was measured using the Luciferase Assay System (Cat: G720A, Promega Corporation, USA) in accordance with the manufacturer’s instructions.

To assess the cytotoxicity of different antibodies against Treg cells, human NK cells that had been differentiated from PBMCs were used as effector cells and co-cultured with Carboxyfluorescein succinimidyl ester (CFSE)-labeled (5 mM; Cat. 423801, BioLegend, USA) Treg cells isolated from the spleens of hOX40 mice as target cells. The E:T ratio was set at 3:1, and the co-culture was maintained for 48 hours. The proportion of 7-Aminoactinomycin D (7-AAD, Cat: 559763, BD Pharmingen, USA) positive Treg cells was then determined by flow cytometry.

Mouse experiments

Female humanized OX40 mice (aged 6–8 weeks) were subcutaneously inoculated with MC38 (1×10⁶) or E.G7 (1×10⁶) cells. Once the tumors had reached 50–100 mm³, mice were treated intraperitoneally (i.p) with HFB301001 (10 mg/kg), its high-affinity variant HFB301001^hi (10 mg/kg), different Fc variants of HFB301001, or an isotype control (hIgG1). Treatments were administered every 3 days for a total of three doses.

For pharmacodynamic evaluation, MC38 tumor-bearing mice were randomized to receive HFB301001 at 0.1, 1, or 10 mg/kg, or an isotype control at 10 mg/kg, on days 6, 9, and 12. Tumor volumes were measured every 3 days, and survival was monitored throughout the study.

For tumor rechallenge, mice that had completely rejected primary MC38 tumors following HFB301001 therapy were re-inoculated with 1×10⁶ MC38 cells 2 months after initial tumor clearance. Tumor growth and survival were then assessed every 3 days.

For the Treg cell transfer experiment, wild-type (WT) and FcγRIIIa knockout (KO) mice bearing MC38 tumors were intravenously transferred with 1×10⁵ Treg cells, isolated from the spleens of humanized OX40 mice using a Treg cell sorting kit (Cat: 18783, STEMCELL Technologies, Canada), on days 6 and 12. Simultaneously, on days 6, 9, 12, and 15, the mice were i.p. injected with 5 mg/kg of either HFB301001 or HFB301001^hi antibody. Tumor volumes were measured every 3 days.

For immune cell depletion, the mice were treated with anti-CD4 (300 µg/dose, Cat: A2101, Selleck, USA, selleckchem.com) and anti-CD8α (300 µg/dose, Cat: BE0004-1, Bio X Cell, USA) antibodies on the day prior to HFB301001 administration, with a total of three doses. Tumor size was measured every 3 days and calculated using the formula: (width²×length)/2.

Flow cytometry

The excised tumors and TdLNs were minced into small fragments and then enzymatically digested in a solution containing 1 mg/mL Collagenase IV (Cat: C8160, Solarbio, China), 1 mg/mL Hyaluronidase (Cat: H8030, Solarbio, China), and 25 µg/mL DNase I (Cat: D8071, Solarbio, China). Red blood cells were lysed using Erythrocyte Lysis Buffer (Cat: R1010, Solarbio, China). Following digestion, single-cell suspensions were prepared and subjected to flow cytometry analysis. The viability of cells was identified using a fixable viability dye (Cat: 423106, BioLegend, USA), and Fc receptors were blocked using an anti-mouse CD16/CD32 antibody (Cat: 101320, BioLegend, USA). The cells were then stained with fluorescently conjugated antibodies that target surface and intracellular markers relevant to immune cell phenotyping and functional assessment. Antibody panels used for each experiment are detailed in online supplemental table S1 and online supplemental figure S1. Data were acquired using a BD LSRFortessa X-20 flow cytometer and analyzed using FlowJo software (V.10).

scRNA-seq and analysis

Tumor tissues from mice were dissociated into single-cell suspensions. Single-cell libraries were prepared using the Chromium Next GEM Single Cell 5ʹ Reagent Kit v2 (Cat: CG000330, 10x Genomics, USA) according to the manufacturer’s protocol, and sequenced using an Illumina NovaSeq 6000 platform. Raw UMI counts were then normalized and scaled in Seurat (V.5.0.1), with the mitochondrial content regressed out. Following quality control and initial clustering, T cells were identified based on canonical markers and subsequently re-clustered to resolve finer subpopulations. Principal component analysis was performed on the top 2000 variable genes, and the first 18 principal components were used for Uniform Manifold Approximation and Projection (UMAP) visualization and graph-based clustering. Differentially expressed genes (DEGs) were identified using Seurat’s FindMarkers functions with the Wilcoxon rank-sum test, applying an adjusted p<0.05 as the significance threshold. Functional enrichment analyses of DEGs were conducted using the R package clusterProfiler (V.4.16.0) across Gene Ontology (GO) and KEGG. Overrepresentation significance was evaluated by Fisher’s exact test and adjusted for multiple testing using the false discovery rate.

ELISpot

The spleens of E.G7-OVA tumor-bearing mice that had received an i.p. injection of HFB301001 were harvested. T cells were then isolated and seeded into ELISpot plates that had been pre-coated with an anti-interferon γ (IFN-γ) capture antibody (Cat: 2210002, Dakewe, China), followed by co-culture with E.G7-OVA tumor cells for 36 hours at 37°C. After the incubation period, the cells were removed, and the plates were sequentially incubated with a biotin-labeled anti-IFN-γ detection antibody, streptavidin-HRP, and an AEC substrate in order to visualize the cytokine-producing cells. Spot-forming units were counted using an automated ELISpot analyzer.

Confocal microscopy

To assess the effects of OX40 agonists on receptor clustering, 3×10⁵ HEK293T-OX40-GFP cells were plated onto confocal dishes (Cat: 801001, NEST, China). The cells were then incubated with HFB301001 or HFB301001^hi (2 nM and 10 nM), respectively, in the presence of a goat anti-human IgG-Fc secondary antibody for 12 hours at 37 °C. Following Hoechst nucleus staining (Cat: C0031, Solarbio, China), images were acquired using a LEICA STELLARIS confocal microscope and analyzed using LAS X analysis software.

Multiplex immunofluorescence

Following fixation, tissue sections underwent two washes with PBS and were permeabilized with 0.1%–0.5% Triton X-100 in PBS for 10 min at room temperature. After two additional PBS washes, non-specific binding sites were blocked by incubating the sections in 10% normal goat serum in PBS for 30 min at room temperature. Primary antibodies, including those directed against CD4 (Cat: 25 229T, Cell Signaling Technology, USA), CD8 (Cat: 98 941T, Cell Signaling Technology, USA), and Foxp3 (Cat: 12 653T, Cell Signaling Technology, USA) were applied according to the manufacturers’ instructions. Fluorophore-conjugated secondary antibodies were then added for detection. Nuclear staining was performed using 1mg/mL DAPI (Cat: D9542, Sigma-Aldrich, Germany), followed by PBS washes. The slides were mounted with an anti-fade medium (KGF028) and scanned at 20×using a VS200 system.

Flow cytometry-based binding assay of HFB301001 to cynomolgus monkey OX40

The apparent binding affinity of HFB301001 to cynomolgus monkey OX40 was determined using flow cytometry-based saturation binding assays. Jurkat cells stably expressing cynomolgus OX40 were incubated with serial dilutions of HFB301001 at 4°C for 90–120 min to allow equilibrium binding. Receptor internalization was minimized, and binding was analyzed by flow cytometry using Alexa Fluor 488 goat anti-human IgG (H+L) (Cat: A-11013, Invitrogen, USA). Binding curves were generated by plotting specific mean fluorescence intensity (MFI) and fitted by nonlinear regression using a one-site specific binding model in GraphPad Prism (V.10.6.1).

Toxicity study in non-human primates

In the single dose toxicity study, HFB301001 was administered to eight cynomolgus monkeys (one/sex/group) as a 1-hour intravenous infusion at a rate of 10 mL/kg/hour, at the dose levels of 1, 10, or 100 mg/kg. The assessment of toxicity was based on survival, clinical observations, local reactions, and body weight. Furthermore, the study involved the collection of blood samples for the purpose of conducting toxicokinetic and immunophenotyping evaluations at regular intervals. The toxicity study in non-human primates was conducted by Charles River Laboratories Evreux.

Heterogeneity of OX40 expression in non-small cell lung cancer patients

PBMCs were isolated from peripheral blood, and single-cell suspensions were dissociated from lymph nodes and tumors of six non-small cell lung cancer (NSCLC) patients. These samples were then analyzed by flow cytometry to assess OX40 expression on Treg, CD4⁺ T, and CD8⁺ T cells.

Patient-derived tumor tissue slices

Tumor tissues derived from NSCLC patients were immersed in low-melting agarose (Cat: A8350, Solarbio, China) and processed into 200 µm slices with a vibratome (Leica VT1200S, USA) at 0.22 mm/s with an amplitude of 1 mm. The slices were then transferred to a tissue culture chamber and submerged in RPMI-1640 supplemented with 10% FBS (Cat: A5669701, Gibco, USA) and 1% P/S. Thereafter, the slices were stimulated with HFB301001 (10 mg/mL), goat anti-human IgG-Fc secondary antibody, and OKT3. The tumor slice was analyzed by flow cytometry.

Statistical analysis

The data are presented as the mean±SD. For the purpose of conducting multiple group comparisons, the one-way or two-way analysis of variance was applied, while for the assessment of differences between two groups, a two-tailed t-test was used. For two-group comparisons, a t-test was applied. Survival analyses were performed with the Kaplan-Meier method and evaluated by the log-rank test. Statistical analyses were conducted in GraphPad Prism version 10.3.1, and p<0.05 was considered to indicate statistical significance.

Results

AI-guided prediction of HFB301001-OX40 structure and experimental validation

We performed the high-throughput functional screening based on droplet-based microfluidics for functional screening of OX40 agonist antibodies (figure 1A).³¹ The activity of OX40 antibodies was examined with and without the secondary antibody for crosslinking. Among all the agonist antibodies, Ab-1 (hereafter referred to as HFB301001) and Ab-2 are dependent on Fc-mediated crosslinking (figure 1B and online supplemental figure S2A). When cocultured with HEK293 cells expressing FcγRIIa, HEK293 cells expressing FcγRIIb, or Raji cells, HFB301001 and Ab-2 induced robust reporter activation (figure 1C–F and online supplemental figures S2B–D). Among all the OX40 antibodies evaluated, treatment with HFB301001 resulted in the most significant delay in tumor growth in humanized OX40 mice bearing MC38 tumors (online supplemental figure S2E and S2F). Consequently, HFB301001 was selected for further investigation and development.

Figure 1. AI-guided prediction of HFB301001-OX40 structure and experimental validation. (A) Schematic illustration of the AI-guided prediction of the HFB301001-OX40 structure and its experimental validation (created with BioRender.com). High-throughput functional screening based on droplet-based microfluidics for preliminary functional screening of OX40 agonist antibodies, and subsequently combined AI-guided structural prediction with experimental validation to elucidate their interaction interfaces. (B) The activation function of Ab-1 (HFB301001). hOX40 NF-κB-GFP Jurkat reporter cells were stimulated with HFB301001 in the presence or absence of a secondary cross-linking antibody, and GFP expression was measured by flow cytometry. (C–F) Schematic illustration depicting the binding profile of HFB301001. hOX40 NF-κB-GFP Jurkat reporter cells were co-cultured with (D) FcγRIIa-positive HEK293FT cells, (E) FcγRIIb-positive HEK293FT cells, and (F) FcγRIIb-positive Raji cells in the presence of different concentrations of HFB301001. GFP expression was analyzed by flow cytometry. (G, H) The epitope regions defined by (G) AI-based structural prediction and (H) HDX-MS analysis were mapped onto the X-ray structure model of the OX40- and OX40L in complex. HFB301001 binding epitope on OX40 as determined by peptide mapping. The epitope region ‘CSRSQNTVCRPCGPGFYND’ was mapped onto a 3D model adapted from PDB 2HEV. (I) Schematic illustration of competitive binding between OX40 antibodies and OX40L. OX40 NFκB-Luc Jurkat reporter cells were treated with recombinant human OX40L with serial dilutions of HFB301001 or BMS 986178. After 24 hours of incubation, luminescence was measured. AI, Artificial Intelligence; HDX-MS, hydrogen/deuterium exchange mass spectrometry.

To elucidate the binding epitope of HFB301001 in silico, we employed EnsembleFold (online supplemental figure S3A), a recently developed framework for biomolecular structure modeling built on the CASP15-winning algorithm DMFold.^{27 28} Using EnsembleFold, 2000 structural models of the hOX40-HFB301001 complex were generated, and the top-ranked model was analyzed to identify the epitope region based on the changes in ASA (see Methods for details). The predicted epitope was primarily localized to the CRD1 of hOX40 (figure 1G and online supplemental figure S3B), indicating that HFB301001 does not affect the binding of OX40L to OX40.

To validate the AI-predicted structure, we performed HDX-MS analysis of HFB301001. HDX-MS analysis comparing the deuterium content of free and HFB301001-bound human recombinant OX40 demonstrated that the peptides on OX40 protein corresponding to the amino acids CSRSQNTVCRPCGPGFYND (56-74), which were significantly protected by HFB301001 and hence defined as the epitope. The epitope region was mapped onto the X-ray structure model of the OX40- and OX40L in complex (PDB number: 2HEV). Notably, the HFB301001 epitope does not overlap with the OX40L ligand binding interface (figure 1H), corroborating the AI-predicted outcomes. To determine whether HFB301001 competes with OX40L for binding to OX40, we evaluated its ability to reduce the activation of OX40 reporter cells induced by OX40L. As shown in figure 1I, HFB301001 failed to interfere with OX40L-mediated signaling. In contrast, the OX40L-competitive OX40 antibody BMS 986178 exerted a dose-dependent inhibitory effect on OX40L-mediated signaling.

Affinity maturation of OX40 antibody

Next, to investigate whether enhanced affinity translates into improved antitumor efficacy, HFB301001 was affinity-matured to strengthen its binding to OX40. The CDR3 regions of the heavy chain and light chain of HFB301001 were subjected to random mutagenesis at two amino acid positions in order to construct a phage display library for affinity maturation (figure 2A).²⁶ The CDR3-targeted mutagenesis strategy, when coupled with increasingly stringent wash conditions, has been shown to lead to the generation of higher-affinity antibody mutants. After three rounds of phage display screening, 15 monoclonal clones were randomly selected for DNA sequencing. The clone HFB301001-mut-8 was the most frequently observed sequence, followed by the clones HFB301001-mut-10 and HFB301001-mut-11 (figure 2B). These three representative clones were selected and engineered into full-length antibodies containing a wild-type Fc region for affinity analysis using SPR. Clone HFB301001-mut-8 bound to OX40 with a K_D of 7.73×10^–9 M and was designated as the high-affinity variant (HFB301001^hi), while clone HFB301001-mut-10 (HFB301001^med) exhibited a medium affinity (K_D=1.91×10^–8 M). The original HFB301001 showed the lowest affinity (K_D=2.86×10^–7 M) (figure 2C, online supplemental figure S4A, and online supplemental table S2). Together, based on HFB301001, we employed phage display to generate higher-affinity OX40 agonist antibodies.

Figure 2. The high-affinity mutant of HFB301001 displays compromised functional activation and induces weaker clustering of the OX40 receptor in vitro. (A) Schematic illustration of HFB301001 affinity maturation using phage display. Two amino acid positions in the CDR3 regions of HFB301001 heavy and light chains were randomly mutated to generate a phage library. The phages were incubated with biotin-conjugated human OX40, washed, eluted, and subjected to iterative rounds of amplification and selection. The enriched phages were then sequenced. (B) Amino acid sequences of enriched OX40 agonist antibodies. (C) SPR analysis of antibody binding to OX40. Anti-OX40 antibodies were captured on an anti-human Fc-coated CM5 sensor chip, and OX40 protein was flowed over the surface. The raw data and the corresponding fitted curves were analyzed by SPR. (D) Activation potency of OX40 agonists in OX40 NF-κB-Luc Jurkat cells. OX40 NF-κB-Luc Jurkat cells were treated with HFB301001 or HFB301001^hi in the presence or absence of a secondary cross-linking antibody (crosslinked vs soluble format) for 6 hours, after which luciferase expression was measured. The background signal without antibody is subtracted from that of the antibody treatment. (E) Activation potency of OX40 agonists in cancer patient-derived TdLNs Cell suspensions were prepared from lymph nodes of cancer patients and stimulated with HFB301001 or HFB301001^hi in the presence of a secondary cross-linking antibody and OKT3 for 48 hours. The proportions of CD25⁺ and CD69⁺ T cells, as well as the expression of IFN-γ and GZMB, were analyzed by flow cytometry. (F) OX40 receptor clustering potency of OX40 agonists in OX40-GFP HEK293 cells. OX40-GFP HEK293 cells were treated with HFB301001 or HFB301001^hi (2 nM and 10 nM) in the presence of a secondary cross-linking antibody for 12 hours, and GFP clustering was visualized by confocal microscopy. OX40-GFP was shown in green, nuclei in blue. (G) Schematic illustration showing the co-culture of FcγRIIIa NFAT-Luc Jurkat cells and Treg. The cells were co-cultured at a ratio of 10:1 in the presence of a secondary cross-linking antibody for 24 hours, after which luciferase expression was measured. (H) Quantification of 7-AAD⁺ CFSE⁺ Treg cells following a 48-hour in vitro co-culture of human NK cells (effector cell) with Treg cells (target cell) isolated from hOX40 mice at an effector-to-target ratio of 1:3 across a range of concentrations. Data in (D), (E), and (G) are presented as mean±SEM. Statistical significance was determined by one-way ANOVA with multiple comparisons. ns, not significant; *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. ANOVA, analysis of variance; CFSE, Carboxyfluorescein succinimidyl ester; NH, natural killer; RLU, Relative Light Units; SPR, surface plasmon resonance; TdLN, tumor-draining lymph node; 7-AAD, 7-Aminoactinomycin D.

High-affinity mutant of HFB301001 displays compromised activation and lower clustering of the OX40 receptor compared with the low-affinity parental antibody in vitro

To assess the functional activity of OX40 agonists with varying affinities, OX40 NFκB-Luc Jurkat reporter cells were stimulated in the presence or absence of secondary antibody-mediated cross-linking. Interestingly, HFB301001, despite its lower binding affinity for OX40, induced stronger reporter cell activation than the high-affinity variants (figure 2D and online supplemental figure S4B). To further evaluate the activation function of HFB301001 and HFB301001^hi in clinically relevant samples, T cells isolated from lymph nodes from lung cancer patients were stimulated with both antibodies. Compared with HFB301001^hi, treatment with HFB301001 resulted in an augmentation of the frequencies of CD25⁺ CD69⁺, IFN-γ⁺, and GZMB⁺ CD3⁺ T cells in lymph node samples derived from three distinct lung cancer patients, indicating that HFB301001 promoted a higher proportion of activated T cells (figure 2E). We hypothesized that low-affinity antibody HFB301001 would enhance receptor clustering compared with its high-affinity mutant. Using Jurkat cells overexpressing hOX40-GFP fusion protein, we found that HFB301001 induced stronger OX40 clustering than HFB301001^hi at both high and low doses, which was associated with enhanced downstream signaling (figure 2F).

To monitor the activity of ADCC of antibodies with different affinities, a FcγRIIIa NFAT-luc Jurkat reporter cell line was constructed and co-cultured with or without OX40 humanized mice-derived Treg cells in the presence of HFB301001 or HFB301001^hi. HFB301001 treatment exhibited more activation of FcγRIIIa NFAT-luc Jurkat reporter cells than HFB301001^hi treatment, suggesting that HFB301001 induced enhanced ADCC (figure 2G). Moreover, in co-cultures of human NK cells with Treg cells isolated from hOX40 mice, HFB301001 induced markedly more potent cytotoxicity against Treg cells than HFB301001^hi, across a range of concentrations (figure 2H).

Collectively, these in vitro findings indicate that the high-affinity mutant of HFB301001 induced weaker OX40 receptor clustering and exhibited compromised functional activity.

High-affinity variants of HFB301001 exhibited inferior anti-tumor efficacy and immune response to HFB301001 in vivo

Given the importance of OX40 in T cell activation with distinct affinities, we next compared the anti-tumor efficacy of HFB301001 and HFB301001^hi. MC38 and E.G7 tumor-bearing mice were treated with either antibody (figure 3A). HFB301001^hi exhibited markedly reduced anti-tumor effect relative to HFB301001, resulting in tumor control and minimal survival benefit in both models. This finding indicates the disadvantage of high-affinity receptor engagement (figure 3B–E). In addition, no significant body weight reduction was observed in either tumor model, suggesting the treatment was well tolerated (figure 3F and G).

Figure 3. High-affinity mutants of HFB301001 exhibited inferior antitumor efficacy than HFB301001 in vivo. (A) Schematic representation of HFB301001 and HFB301001^hi antitumor efficacy evaluation. MC38 or E.G7 tumor-bearing mice were treated with HFB301001 or HFB301001^hi for four doses, and tumor volumes were measured every 3 days. (B, E) Individual tumor growth curves, (C, D) survival curves, and (F, G) body weights were monitored (n=10). Statistical significance of survival was determined using the log-rank test. Data in (C) and (D) are presented as mean±SEM from one representative experiment of two independent replicates. ***p<0.001, ****p<0.0001.

To explore the immunological effects of HFB301001 and HFB301001^hi, tumors and TdLNs from MC38-bearing mice were analyzed by flow cytometry after three doses. As shown in figure 4A, treatment with HFB301001 significantly inhibited tumor growth. Treatment with HFB301001^hi resulted in a significant decrease in the infiltration of CD45⁺ and total T cells into the TME when compared with HFB301001 treatment (figure 4B and C). The present study found that both HFB301001^hi and HFB301001 treatment led to a significant increase in the proportion of CD44⁺ CD62L⁻ effector T cells compared with the control group. In contrast, the frequency of CD44^- CD62L⁺ naïve T cells was markedly decreased following both HFB301001^hi and HFB301001 treatment, suggesting that OX40 agonist treatment, regardless of antibody affinity, vigorously promotes the differentiation of naïve T cells into effector T cells (figure 4D and E).

Figure 4. High-affinity mutants induced limited CD8⁺ T-cell activation and inefficient Treg clearance in tumor and TdLNs. (A) Representative images of tumor tissues treated with HFB301001 and HFB301001^hi. MC38 tumor-bearing mice were treated with HFB301001 or HFB301001^hi for three doses. Tumor tissues were collected for imaging. (B–Q) Frequency of CD45⁺ cells and subsets of T cells following HFB301001 and HFB301001^hi treatment in TME and TdLNs. Both tumor tissues and TdLNs were subsequently digested into single-cell suspensions for flow cytometry analysis following HFB301001 or HFB301001^hi treatment (n=5). The quantitative flow cytometry data of the frequencies of CD45⁺, CD3⁺ T, naïve CD3⁺, effector CD3⁺, exhausted CD8⁺ T, Ki67⁺ CD8⁺ within tumors, and CD8⁺, activated CD8⁺, CD4⁺, Treg, and CD8⁺ T/Treg ratio within (B–L) tumors and (M–Q) TdLNs. Data are presented as means±SD. Statistical significance was determined by one-way ANOVA with multiple comparisons. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. ANOVA, analysis of variance; ns, not significant; TdLNs, tumor-draining lymph nodes.

Compared with HFB301001, HFB301001^hi resulted in a restricted activation (CD25⁺ CD8⁺) and proliferation (Ki67⁺ CD8⁺) of CD8⁺ T cells (figure 4F–I). The T cell exhaustion (PD-1⁺ CD8⁺) was not significantly alleviated by HFB301001^hi treatment. Additionally, CD4⁺ T and Treg were reduced to a greater extent in the HFB301001 group compared with the HFB301001^hi group. The intratumoral CD8⁺/Treg ratio was significantly lower in the HFB301001^hi group than in the HFB301001 treated group, suggesting a more immunosuppressive TME associated with HFB301001^hi treatment (figure 4J–L). Next, we examined T cell functional status in TdLNs. Analogous to its effects in the TME, the HFB301001^hi treatment led to a significant reduction in CD8⁺ T cell activation and the CD8⁺/Treg ratio compared with the HFB301001 group (figure 4M–Q). These analyses indicate that high-affinity antibody treatment impairs T cell activation and Treg clearance in both the TME and TdLNs, accounting for its suboptimal antitumor activity.

HFB301001 promotes robust CD8⁺ T-cell activation, facilitates Treg clearance, and establishes durable immunological memory

To evaluate the dose-dependent efficacy and pharmacodynamics of HFB301001, MC38 tumor-bearing hOX40 mice received i.p. injections of HFB301001 at 0.1, 1, or 10 mg/kg. Twice-weekly administration resulted in dose-dependent tumor growth inhibition, with the most significant tumor volume reduction observed at 10 mg/kg. Treatment with 10 mg/kg also conferred a significant survival advantage compared with the control group (figure 5A). Importantly, MC38-bearing mice cured by HFB301001 completely rejected tumor rechallenge 2 months after initial clearance, whereas all age-matched naïve controls developed tumors, demonstrating that HFB301001 induced durable antitumor immune memory (figure 5B). Moreover, CD8⁺ T-cell depletion markedly abolished the antitumor efficacy of HFB301001, whereas CD4⁺ T-cell depletion exerted no significant impact, indicating that CD8⁺ T cells are critical mediators of its antitumor activity (figure 5C).

Figure 5. HFB301001 promotes robust CD8⁺ T cell activation and establishes immunological memory. (A) The antitumor efficacy of HFB301001 at different doses. MC38 tumor-bearing hOX40 mice were treated with HFB301001 at various doses (n=10). Survival curve was plotted. Statistical significance of survival was determined using the log-rank test. (B) Schematic illustration of the rechallenge assay. MC38 tumor-bearing mice cured by HFB301001 and age-matched naïve controls were rechallenged with MC38 cells 2 months after the initial tumor clearance. Survival curve was subsequently monitored. The rechallenged mice (n=6) represent animals accumulated from multiple independent experimental cohorts, rather than a single experiment. (C) CD4⁺ and CD8⁺ T cell depletion assay. MC38 tumor-bearing mice were treated with HFB301001 or HFB301001^hi for three doses. Depleting antibodies against CD4⁺ or CD8⁺ T cells were administered intraperitoneally (i.p.) 1 day before treatment. Tumor volume was monitored. (D) UMAP plot of lymphoid cells. MC38 tumor-bearing mice were treated with PBS or HFB301001 for three doses. Tumor-infiltrating immune cells were analyzed by scRNA-seq and colored by T-cell subclusters. (E) Proportions of T-cell subclusters in control (PBS) and HFB301001-treated groups. (F) Bubble diagram depicting the expression patterns of activation and exhaustion across distinct T cell clusters. (G) GO-enriched functional pathways of tumor-infiltrating Treg, CD4⁺ T, and CD8⁺ T cells in PBS (Control) vs HFB301001-treated mice. Colors: Treg (purple), CD4⁺ (green), CD8⁺ (orange). (H–J) Schematic illustration of immunofluorescence analysis of mouse tumor tissues. MC38 tumor-bearing mice were treated with HFB301001 or HFB301001^hi for three doses. Tumor tissues were harvested and subjected to multicolor immunofluorescence staining. Blue indicates nuclei, red indicates CD4 or CD8, and green indicates Foxp3. (K) Schematic illustration of T cell-specific activation. E.G7-OVA tumor-bearing mice were treated with HFB301001 for three doses. Splenocytes were harvested and incubated with E.G7-OVA cells for 36 hours, after which IFN-γ production was measured. (L) The representative wells (left) and the number of IFN-γ⁺ spots per 2×10⁴ cells (right) are shown. Data are presented as means±SD. n=3 for (L), n=6–10 for A–C), p values were calculated using the t-test in (L) and two-way ANOVA in (C). *p<0.05, ****p<0.0001. ANOVA, analysis of variance; ns, not significant; PBS, Phosphate-buffered saline; scRNA-seq, single-cell RNA sequencing; UMAP, Uniform Manifold Approximation and Projection.

To explore the in-depth antitumor mechanism of HFB301001, we performed scRNA-seq of tumor-infiltrating T cells from MC38 tumors treated with HFB301001 or PBS. Unsupervised clustering based on established markers identified seven T-cell subclusters: Treg, CD4_Tcm, CD8_Naive, CD8_Teff, CD8_Tex, CD8_Cycling, and gdT (figure 5D and online supplemental figure S5). There was a decrease in Treg cells, while a marked expansion of CD8_Teff following HFB301001 treatment relative to the control group (figure 5E). Moreover, HFB301001 treatment increased the proportion of CD8_Teff cells with elevated expression of effector markers, including Gzmb and Ifng, while reducing Pdcd1 expression (figure 5F). The GO pathway analysis revealed substantial enrichment of CD8⁺ T cell-related signaling pathways in the HFB301001 treatment, including T cell differentiation, T cell chemotaxis, IL-2 production, and T cell activation (figure 5G). Furthermore, multiplex immunofluorescence analyses revealed a pronounced increase in total CD8⁺ T cell infiltration and a reduction in Treg cells within tumors from treated mice (figure 5H–J). Consistent with these findings, splenocytes were co-cultured with E.G7-OVA tumor cells, and we observed a greater number of tumor-specific IFN-γ-secreting splenocytes derived from HFB301001-treated mice compared with those from vehicle-treated mice (figure 5K and L).

To determine whether Fc-mediated effector functions and Fc crosslinking-dependent agonism are critical for the antitumor efficacy of HFB301001, we generated multiple IgG isotypes of HFB301001 with distinct FcγR binding properties. Among the murine IgG subclasses, mIgG2a, which is functionally analogous to human IgG1, displayed the strongest FcγR engagement and ADCC potential, whereas mIgG1 and mIgG2a-SELF variants preferentially bound FcγRIIb with weaker FcγRIIa interaction. In contrast, the D265A/N297A (DANA) mutation abolished FcγR binding and consequently eliminated Fc-mediated crosslinking (online supplemental table S3). Functionally, HFB301001-mIgG2a exhibited the most pronounced tumor growth inhibition in the MC38 model, while DANA mutants were largely inactive. HFB301001-mIgG1 and -SELF variants showed intermediate efficacy (figure 6A–C).

Figure 6. HFB301001 suppresses tumor progression by mediating ADCC-dependent depletion of intratumoral Tregs. (A) Schematic representation of HFB301001 Fc variant efficacy evaluation. MC38 tumor-bearing mice were treated with different HFB301001 Fc variants for three doses, and tumor volumes were measured every 3 days. (B) Survival curves and (C) individual tumor growth were monitored (n=5). (D) Schematic of the Treg cell transfer experiment to evaluate the effects of Treg depletion in MC38 tumor-bearing mice. WT and Fcgr3 KO mice were intravenously transferred with Treg cells isolated from the spleens of humanized OX40 mice on days 6 and 12, and the mice were intraperitoneally injected with either HFB301001 or HFB301001^hi on days 6, 9, 12, and 15. (E) Survival curves and (F) Individual tumor growth were monitored (n=6). Survival was assessed using the log-rank test. Data are presented as mean±SEM. ns, not significant; *p<0.05, **p<0.01. ADCC, antibody-dependent cell-mediated cytotoxicity; WT, wild-type.

We further investigated ADCC using Fcgr3 KO mice by implanting MC38 tumors and transferring Treg cells derived from the spleens of OX40 humanized mice. In this model, endogenous T cells express murine OX40, while adoptively transferred Tregs express human OX40. Consequently, OX40 antibodies (HFB301001 or HFB301001^hi) exert their effects exclusively on Tregs. Treatment with HFB301001 effectively inhibited tumor growth in WT mice, whereas its efficacy was weaker in Fcgr3 KO mice. In contrast, HFB301001^hi demonstrated abrogated tumor inhibition in both mouse groups (figure 6D–F). These findings demonstrate that HFB301001 not only enhances T cell activation but also limits tumor progression through ADCC-mediated depletion of intratumoral Tregs.

Collectively, these data suggest that HFB301001 significantly suppresses the growth of MC38 tumor by eliciting tumor-specific T cells, facilitating Treg clearance, and establishing anti-tumor immunological memory.

HFB301001 demonstrates safety in the cynomolgus monkey

To support future clinical translation, we first assessed the cross-species binding of HFB301001. Flow cytometry assays demonstrated that HFB301001 binds to Jurkat cells overexpressing cynomolgus monkey OX40 with an affinity comparable to that observed for human OX40, thereby supporting the utility of cynomolgus monkeys for in vivo evaluations (online supplemental figure S6A). Based on these findings, an acute toxicity study was conducted in cynomolgus monkeys. Animals received a single 1 hour intravenous infusion of HFB301001 and were monitored for 15 days, during which serum drug exposure, hematological parameters, and immunophenotyping were assessed (online supplemental figure S6B).

No sex-related differences were observed in the pharmacokinetic parameters across the dose groups. Both Area Under the Curve (AUC)_0-t and C_end values increased proportionally with the dose, ranging from 1 to 100 mg/kg. The terminal half-lives for all animals were estimated to be between 225 and 356 hours, consistent with the typical behavior of monoclonal antibodies (online supplemental figure S6C and online supplemental table S4). Moreover, no HFB301001-related clinical signs were observed at any of the dose levels.

HFB301001 was well tolerated at all administered dose levels in this study, including a single dose of up to 100 mg/kg, which was the highest dose tested. No significant body weight loss was observed at any dose level (online supplemental figure S6D). Under the experimental conditions, the 100 mg/kg dose was determined to be below the Maximum Tolerated Dose. Administration of HFB301001 resulted in a minimal increase in lymphoid cellularity in the spleen at 100 mg/kg, which correlated with increased spleen weights (online supplemental table S5). No macroscopic findings were observed that were considered related to HFB301001 administration. Hematological assessments showed no treatment-related alterations in red blood cell-associated parameters, including erythrocytes, hemoglobin, packed cell volume, and mean cell volume, and platelet counts also remained stable (online supplemental figure S6E and online supplemental table S6). Differential leukocyte analysis showed no consistent treatment-related changes in major leukocyte subsets. However, total leukocyte counts decreased from day −3 to day 15 in both sexes across all dose groups, without alteration in lymphocyte proportions (online supplemental figure S6F and online supplemental table S6).

To evaluate functional changes in T cells following HFB301001 treatment, peripheral blood was analyzed by flow cytometry. The proportion of cytotoxic T (CD3⁺ CD8⁺ GZMB⁺) cells increased in both male and female monkeys on day 15 following treatment with HFB301001 at 1 mg/kg and 10 mg/kg. At a dose of 100 mg/kg, the proportion of cytotoxic T cells further increased in males while decreasing in females (online supplemental figure S6G). While there was no significant decrease in Treg cells in all doses. Activated NK cells showed a tendency to increase, whereas total NK cells, CD3⁺ T cells, and helper T (CD4⁺) cells displayed modest, non-dose-dependent fluctuations (online supplemental table S7). Together, these data indicate that HFB301001 is well-tolerated in cynomolgus monkeys.

HFB301001 treatment promoted T cell activation in patient tumor sections and TdLNs

To extend observations from preclinical models to human samples, we used NSCLC patient-derived tumor slices and TdLNs to assess the T cell activation capacity of HFB301001. We first assessed OX40 expression on immune cells from six NSCLC patients. In peripheral blood, OX40 expression was generally low on Treg cells. In contrast, Treg cells in tumor tissues exhibited higher OX40 expression, while CD8⁺ T cells showed low OX40 expression (online supplemental figure S7). Next, surgical samples of TdLNs and tumors from NSCLC patients were processed to obtain single-cell suspensions and tumor slices, respectively, and treated with HFB301001 in the presence of OKT3 (figure 7A). Treatment with HFB301001 and OKT3 resulted in enhanced T cell activation in TdLNs across multiple patients compared with either treatment alone, as shown by increased CD25 and CD69 expression on CD3⁺ cells (figure 7B and C). Analogous to its effects in TdLNs, HFB301001 treatment in the presence of OKT3 exhibited significantly potentiated T cell activation in human NSCLC slices (figure 7D and E). Moreover, treatment with HFB301001 in human NSCLC tumor tissue resulted in a reduction in the number of Treg cells (figure 7F and G). These data demonstrate that HFB301001 treatment enhances T cell activation in human primary tumor tissues and TdLNs, thus supporting the translational development of HFB301001.

Figure 7. HFB301001 treatment promotes T cell activation in patient-derived tumor tissue sections and TdLNs. (A) Schematic of HFB301001-mediated activation of T cells from patient-derived tumor slices and TdLNs. Surgical samples of TdLNs and tumors from NSCLC patients were processed into single-cell suspensions and tumor slices, respectively, and treated with HFB301001 in the presence of OKT3. T-cell activation (CD25⁺ CD69⁺) was assessed by flow cytometry. (B, C) Percentage of T-cell activation in TdLNs from two NSCLC patients. (D, E) Percentage of T-cell activation in tumor tissue slices from two NSCLC patients. (F, G) Percentage of Treg cells in tumor tissues from NSCLC patients. Tumor specimens obtained from two NSCLC patients were processed into single-cell suspensions and treated with HFB301001. The frequency of Treg cells (CD25⁺ Foxp3⁺) was quantified by flow cytometry. Data are presented as means±SD. n=3 for B–G. P values were calculated using one-way ANOVA in B–E and the t-test in F, G. *p<0.05, **p<0.01, ****p<0.0001. ANOVA, analysis of variance; NSCLC, non-small cell lung cancer; PBS, Phosphate-buffered saline; TdLNs, tumor-draining lymph nodes.

Discussion

In this study, we identified reduced affinity as an effective strategy to improve the efficacy of OX40 agonist antibody, providing new insights for antibody engineering. Mechanistically, the low-affinity OX40 agonist HFB301001 promotes receptor clustering, enhances T cell activation, and depletes Treg, thereby eliciting a potent antitumor response. HFB301001 showed a favorable safety profile. In support of its translational potential, HFB301001 further elicited robust T cell activation in patient-derived tumor slices and TdLNs. In a phase I study of heavily pretreated advanced solid tumors (NCT05229601), HFB301001 was well tolerated, with no dose-limiting toxicities. Consistent with its mechanism, HFB301001 enhanced intratumoral CD8⁺ T-cell and NK-cell infiltration without reducing OX40 surface expression and achieved encouraging disease control in refractory renal cell carcinoma and hepatocellular carcinoma.^{32 33}

High affinity is generally prioritized in therapeutic antibody development, while our findings indicate that this paradigm does not uniformly apply to agonistic antibodies. For agonistic antibodies engineered to regulate receptor signaling, such as OX40 agonists, lower affinity may instead confer superior functional activity, representing a novel design principle for enhancing T cell activation and antitumor efficacy. In accordance with this concept, Yu et al recently reported that lower rather than higher affinity promoted immune cell activation by enhancing receptor clustering for CD40, 4–1BB, and PD-1.²⁵ Recently, Singhaviranon et al³⁴ demonstrated that CD8⁺ T cells with high-affinity TCR-MHC-I interactions tend to undergo exhaustion and contribute to immunosuppression. In contrast, low-affinity T cells mediate more effective antitumor responses.²⁸ Consistent with these findings, our study demonstrates that reducing antibody affinity enhances the agonistic activity of anti-OX40 monoclonal antibodies. HFB301001 exhibited a faster dissociation rate compared with its high-affinity counterpart. This kinetic property enabled more dynamic binding and release from the receptor, facilitating receptor clustering.³⁵ Consequently, HFB301001 induced more robust OX40 clustering, amplified downstream signaling, and elicited stronger T cell activation and effector responses than the high-affinity variant. The OX40 antibody in clinical trials is engineered to have high affinity; for instance, BMS 986178 (K_D=2.27×10⁻⁹ M) showed lower efficacy compared with HFB301001 (online supplemental figure S8).³⁶ Our findings suggest fine-tuning of binding affinity as a strategy to optimize agonist antibody design.

Additionally, we reported that antibodies with reduced affinity can more effectively eliminate Treg cells, a key immunosuppressive component within the TME, through NK cell-mediated ADCC, thereby alleviating immunosuppressive pressure. The OX40 clustering may lead to NK16a (FcγRIIIa) clustering on NK cells and the formation of an ADCC lytic synapse.^{37 38} Moreover, although FcγRIIIa 158V/F polymorphism influences IgG1-mediated ADCC, with higher IgG1 affinity associated with the V158 allele.39,41 Thus, the efficacy of OX40 antibody treatment may be affected by FcγRIIIa polymorphism in patients.

Furthermore, we have demonstrated that HFB301001 does not interfere with the binding and signaling of the OX40L, consistent with the hypothesis that the HFB301001 epitope does not overlap with OX40L binding interface. Therefore, HFB301001 can deplete Treg without blocking OX40L/OX40 interactions, which may contribute to its potent antitumor efficacy. Similarly, CD25 is a selective target for Treg depletion. However, the anti-tumor efficacy of CD25 antibody is limited by IL-2 receptor signaling blockade in effector T cells. Solomon et al developed an anti-CD25 antibodies optimized to deplete Tregs while preserving IL-2 signaling on effector T cells.⁴²

The mechanisms of action of OX40 monoclonal antibodies are modulated by their isotype and by interactions with FcγR. For example, mIgG1 predominantly engages the inhibitory FcγRIIb to promote OX40 signaling and T-cell activation, whereas mIgG2a exerts cytotoxic effects by depleting OX40-expressing cells, particularly regulatory T cells.^{43 44} Similarly, isotype dependence effects have been observed with anti-PD-1 antibodies.^{45 46} Notably, Willoughby et al demonstrated that the anti-hOX40 hIgG1 antibody GSK3174998 engages both activating and inhibitory FcγRs to drive antigen-specific T-cell expansion.⁴⁷ In this study, we showed that HFB301001-mIgG2a exhibited the most potent tumor suppression. Variants with preferential FcγRIIb binding displayed intermediate efficacy, suggesting that engineering HFB301001 to enhance effector functions may further improve the therapeutic activity.

We also observed pronounced tissue- and cell-type-specific heterogeneity in OX40 expression in patients with cancer. This heterogeneity is likely to modulate the biological activity of OX40-targeting antibodies and may contribute to interpatient variability in clinical responses, with the underlying functional implications remaining to be fully elucidated. Consistent with this interpretation, cynomolgus monkey safety studies showed minimal changes in circulating Treg cells, which may reflect the low OX40 expression on peripheral Treg cells. In addition, the safety of HFB301001 was monitored for 15 days, but the rebound effect was not assessed during this period.

In summary, we report that the low-avidity OX40 agonist HFB301001 induces enhanced OX40 clustering, resulting in more effective T cell activation and Treg depletion, thereby improving antitumor responses. These findings establish low-affinity optimization as a promising strategy for developing the next-generation immunotherapies.

Data availability statement

Data are available on reasonable request.