Adipocyte-derived LTB4 programs human NKG2A+ γδ T cells as cytotoxic sentinels at the adipose-tumor interface in breast cancer

Abstract

Background

The peritumoral microenvironment has emerged as a key role in affecting tumor invasion and immunotherapy responses. In adipose-enriched tumors, such as breast cancer (BC), peritumoral adipose tissue (PA) harbors unconventional immune populations, yet its immunological functions remain poorly understood. In particular, how adipocyte regulate innate-like lymphocytes, such as γδ T cells, remains unclear.

Methods

We performed single-cell RNA sequencing and spatial profiling of paired specimens from patients with BC. Integrated multi-omics analyses, immunofluorescence staining, human γδ T-cell expansion assays, functional assays, and in vivo models were used to define the immune cell states and evaluate the impact of lipid mediator leukotriene B4 (LTB4) on γδ T-cell activation and signaling. Clinical correlations were assessed using our cohort and patient datasets.

Results

We identified a previously unrecognized population of NKG2A⁺γδ T cells with predominant Vδ2 usage that preferentially accumulated in PA, particularly at the adipose-tumor interface. Multi-omics analyses revealed that they exhibited potent cytotoxic activity and extensive interactions with dendritic cells, coordinating a local immune surveillance network. Mechanistically, peritumoral adipocytes showed enhanced activation of the 5-lipoxygenase pathway and secreted the lipid mediator LTB4, which selectively combined to LTB4 receptors, thereby activating STAT1 signaling in γδ T cells and upregulating NKG2A expression. NKG2A⁺γδ T cells were preferentially enriched in ductal carcinoma in situ and early-stage BC. These cells were also associated with favorable clinical outcomes across multiple adipose-enriched tumors. Assays using human specimens confirmed that LTB4 potentiated γδ T cell-mediated antitumor responses. Importantly, LTB4-programmed γδ T cells displayed superior killing capacity using in vitro and in vivo assays.

Conclusions

These findings redefine PA as an active immunological niche that programs γδ T-cell immunity through adipocyte-derived LTB4 signaling. Our study identifies NKG2A⁺γδ T cells as cytotoxic sentinels at the adipose-tumor interface during early tumor development, providing a rationale for leveraging LTB4-mediated γδ T-cell programming in translational cancer immunotherapy.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Clinical observations indicate that fatty breasts are associated with better clinical outcomes and immunotherapy responses.

• Protective Vδ1⁺γδ T cells are reported in healthy women breasts.

WHAT THIS STUDY ADDS

• Using transcriptional and spatial profiling, we identified a distinct population of NKG2A⁺γδ T cells with predominant Vδ2 T-cell receptor usages in peritumoral adipose tissue, unlike Vδ1 T cells.

• NKG2A⁺γδ T cells exhibit potent cell-killing capacity, and interact with antigen-presenting cells to support local immune surveillance.

• Leukotriene B4, a lipid signal, is predominantly produced by peritumoral adipocytes and specifically activates γδ T cells via the STAT1 signal to promote their cytotoxicity.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• The abundance of NKG2A⁺γδ T cells at the adipose-tumor interface contributes to restrained tumor progression and favorable clinical outcomes.

• Applying leukotriene B4 to potentiate γδ T cell-mediated antitumor effects may benefit γδ T cell-based immunotherapy.

Introduction

The peritumoral adipose tissue (PA) is considered to be a key factor affecting the progression and outcome of adipose-enriched epithelial tumors, especially in breast cancer (BC).¹ Adipocytes represent 37% of breast stromal tissue and play a pivotal role in support effects, energy supply, metabolic regulation, and cytokine release due to their anatomical distribution.² Clinically, women with fatty breasts have a lower risk of developing BC compared with those with dense breasts.³ Adipocyte-derived leptin has been found to boost the efficacy of immunotherapy in patients with BC.⁴ However, the crosstalk between adipocytes and BC cells was also demonstrated to result in changes in adipocyte function and paracrine signaling, formating a microenvironment that promotes tumor growth.⁵ The controversy over the role of adipose tissue (AT) in tumor development, treatment efficacy, and prognosis of BC varies among different studies, which may be related to the fact that these studies failed to clearly define the specific anatomical distribution of adipocytes relative to the tumor.

As an important component in the peritumoral microenvironment (PME), AT harbors a diverse array of immune cell types.^{6 7} Among these cells, γδ T cells are demonstrated to expand significantly within murine AT under physiological conditions,⁸ functioning as a crucial link between innate and adaptive immunity.⁹ Subcutaneous adipocytes can directly activate γδ T cells via CD40/Tnfsf18 signaling during chronic infection in vivo.¹⁰ Adipocytes-derived adiponectin could prevent skin inflammation via suppressing γδT17 cells in psoriasis.¹¹ These findings indicate that adipocytes participated in the inflammation immune response via regulating γδT cell function in the infectious mouse model, while the interactions between γδT cells and adipocytes were unexplored in the tumor model.

Our previous work has revealed that human γδ T cells with dominant Vδ1 chains play immunomodulatory roles in the tumor microenvironment (TME).^12–15 However, it is still unclear whether AT can regulate the characteristics and functions of cells in patients with BC, and whether cells are regulated by the PA or by the distant adipose tissue (DA).

In this study, we aimed to investigate the abundance, functional states, and regulatory cues of γδ T cells in human AT and their roles in BC. We collected specimens from patients with BC, including paired tumor (T) and PA, DA, draining lymph nodes (LNs), and peripheral blood (PB). Our findings revealed that the γδ T cells in PA exhibited potent cytotoxic ability. Among them, the abundance of NKG2A⁺ γδT cells had significantly increased and unexpectedly mainly expressed the Vδ2 T-cell receptor, rather than the tissue-resident Vδ1 subsets. The PA infiltrating NKG2A⁺γδ T cells contributed to immune surveillance by targeting tumor cells and interacting with dendritic cells (DCs) that had escaped the primary site. Further analysis showed that peritumoral adipocytes upregulated arachidonic acid metabolism, leading to the production of leukotriene B4 (LTB4), which in turn programmed NKG2A⁺γδ T cells through the STAT1 signaling pathway, enhancing their cytotoxic function.

Collectively, our findings highlight the critical role of NKG2A⁺γδ T cells accumulated in PA and their rapid mobilization within the adipose-tumor interface, establishing a PA-dependent immune surveillance network. These insights highlight their potential applications in γδ T cell-based immunotherapy.

Materials and methods

Patient cohort and human samples

Fresh and paraffin-embedded human samples of patients with BC were pathologically confirmed and untreated before, collected from 2021 to 2026 from the Second Affiliated Hospital of Zhejiang University, School of Medicine (SAHZU). PA was collected from the adipose-tumor interface, less than 2 cm away. DA was collected from more than 5 cm away from T. Summarized and detailed patient information for in vivo analysis and patient-derived tumor fragments (PDTFs) was included in online supplemental table 1.

All samples were obtained with patient consent and approval of the internal review and ethics boards of SAHZU (IR2021001115, IR2021001330, and I20241341).

Preparation of samples and single-cell suspension

Tissues were dissected into small pieces with a diameter <1 mm and dissociated using 1 mg/mL Collagenase I, II, and IV (Sigma, Cat# #V900891, Cat# #V900892, Cat# #V900893) plus 0.25 mg/mL DNase (Sigma, Cat# DN25) in RPMI 1640 (Gibco, Cat# A1049101) with gentleMACS Dissociator for 30–90 min. After digestion, cells were washed with phosphate-buffered saline (PBS) three times (350 g and 5 min) and filtered through 100 µm cell strainers. Primary peripheral blood mononuclear cells (PBMCs) were isolated by density gradient centrifugation using Ficoll (Dekewe, Cat# 7111012) following the manufacturer’s recommendation.

Cell and patient-derived organoids culture

The human cell line MDA-MB-231, MCF-7, and HEK293T were obtained from the American Type Culture Collection and cultured according to standard protocols.

Patient-derived organoids (PDOs) were obtained as we previously described.¹⁰ Briefly, the pellet from tissue processing was resuspended in a cold Matrigel matrix (BD, Cat# 356231) and seeded on a 24-well plate at 37°C for 15 min. Then, 400 µL organoid culture medium (Advanced DMEM/F12 (Gibco, Cat# 12634–010) supplemented with GlutaMAX-I (Gibco, Cat# 35050–061), B27 (Gibco, Cat# 17504–044), SB202190 (Sigma-Aldrich, Cat.# S7067), EGF (PeproTech, Cat.# AF-100-15), Neuregulin 1 (PeproTech, Cat.# 100–03), Noggin (PeproTech, Cat.# 120–10C), R-spondin1 (PeproTech, Cat.# 120–38), A-83-01 (Tocris, Cat.# 2939), Y27632 (AbMole, Cat.# Y-27632), KGF (PeproTech, Cat.# 100–19), FGF10 (PeproTech, Cat.# AF-100-26), N-acetylcysteine (Sigma, Cat# A9165-5 g), nicotinamide (Sigma, Cat# N0636), β-estradiol (Sigma, Cat# E2758), primocin (Life Technologies) and penicillin/streptomycin (Life Technologies)) was added. Organoids were incubated in 37°C, 5% CO₂, humid incubators (Thermo Fisher). The medium was changed every 3 days. Organoids were passaged at ratios (1:1 to 1:4) for 1–4 weeks using TryPLE Express (Gibco, Cat# 12605–028), digested for 5 min, and centrifuged at 400×g for 10 min.

Human T-cell isolation

Primary human lymphocytes or T cells were incubated with APC/Cy-7 anti-human CD45 (BioLegend, Cat# 304014), FITC anti-human CD3 (BioLegend, Cat# 300406), and 7-AAD Viability Staining Solution (BioLegend, Cat# 420404) for 30 min at 4°C protected from light and sorted by fluorescence-activated cell sorting (FACS) using a flow cytometer (BD biosciences, Melody) for single-cell RNA sequencing (scRNA-seq). For in vitro experiments, primary human T cells were purified by CD3 Microbeads (Miltenyi Biotec, Cat# 130–050-101). Live γδ T cells and CD8⁺T cells were then incubated with APC/Cy-7 anti-human CD45 (BioLegend, Cat# 304014), PE/Cy-7 anti-human TCR γ/δ (BioLegend, Cat# 331222), APC anti-human CD159a/NKG2A (BioLegend, Cat# 375108), FITC anti-human CD8a (BioLegend, Cat# 301050), and 7-AAD Viability Staining Solution (BioLegend, Cat# 420404) for 30 min at 4°C protected from light and sorted by FACS (BD biosciences, Melody). Viability and cell purity were confirmed >90% in all samples.

Performance scRNA-seq samples

For sample patient SC1 (PB, LN, DA, PA, and T), scRNA-seq was performed on live CD45⁺ cells sorted from human samples by FACS. Single-cell suspensions were labeled with distinctive sample tags following the manual of BD Human Single-cell Multiplexing Kit, and pooled together at an equal ratio of all samples to obtain 4×10⁴ live cells. Cell Load Scan was performed with BD Rhapsody Scanner. The construction of scRNA-seq libraries was carried out using the Single-cell 3’ whole transcriptome amplification with sample multiplexing kit protocol following the manufacturer’s guidelines.

For sample patient SC2-3 (PB, LN, DA, PA, and T), scRNA-seq was conducted on viable CD45⁺CD3⁺ cells isolated from human samples using FACS. Single-cell suspensions (1×10³ cells/µL) were enclosed in emulsion droplets employing a Chromium Controller (10x Genomics). The construction of scRNA-seq libraries was carried out using the Chromium Next GEM Single Cell 5’ Reagent Kits v2 (10x Genomics) following the manufacturer’s guidelines. Quality assessment of the libraries involved quantification using a Qubit V.4.0 Fluorometer and pattern analysis using a 2100 Bioanalyzer. Subsequently, the libraries underwent sequencing on an Illumina NovaSeq 6000 system.

Processing scRNA-seq data

For samples from SC1, the FASTQ files were processed using the BD Rhapsody Targeted analysis pipeline V.1.9.1 on the Seven Bridges platform (https://www.sevenbridges.com) following the guidance.

For samples from SC2-3, scRNA-seq data were processed using the 10x Genomics CellRanger (V.3.1.0) pipeline. All the reads in the fastq files were aligned against GRCh38 genomes. scRNA-seq using the BD Rhapsody platform was performed by Genozhou Biotechnology (Shanghai, China). scRNA-seq using the 10x Genomics platform was performed by Majorbio Bio-Pharm Technology (Shanghai, China).

Analysis of scRNA-seq data

Quality control and downstream analysis were performed using the Seurat R package (V.4.1.0). Cells were filtered out if they had fewer than 500 unique molecular identifiers (UMIs), expressed less than 300 genes, or had a mitochondrial count ratio exceeding 0.2. Cells with a log10GenesPerUMI metric below 0.84 were also excluded, along with genes expressed by fewer than 10 cells. Doublets were identified and removed using the DoubletFinder R package (V.2.0).

To address technical and biological variations among samples, normalization, and removal of irrelevant sources of variation, including mitochondrial expression, were performed using the NormalizeData, CellCycleScoring, and SCTransform functions before integration. The RunHarmony function from the Harmony R package (V.1.0) was applied. Integration was performed using “patient” as the batch variable in the group.by.vars parameter. This approach was selected to ensure that both interpatient biological variation and platform-specific technical noise were addressed while preserving the shared biological signals across the five distinct sampling sites per patient. The project.dim parameter was set to FALSE, the theta parameter was set to 2, the sigma parameter was set to 0.1, and the lambda parameter was set to 1. Numerical convergence was defined by the stabilization of the objective function (tolerance <10⁻⁴). The integration quality was validated via Uniform Manifold Approximation and Projection (UMAP) visualization colored by platform and patient, which confirmed a high degree of intersample mixing.

A UMAP plot was generated using 40 principal components (PCs) for cluster identification and dimensionality reduction. The number of PCs was determined by cumulative variance calculation, ensuring that the selected dimensions captured over 90% of the total biological variance while minimizing technical noise. CD3⁺ cells were clustered using the FindNeighbors and FindClusters functions. Subsequently, γδ T cells were annotated, isolated, and reclustered. Specifically, a shared nearest neighbor graph was constructed based on the top 40 PCs using the Euclidean distance metric via FindNeighbors. Subsequently, the Louvain algorithm was applied to identify distinct cell clusters at a resolution of 0.1 using FindClusters, with all other parameters set to their default values. The clustering resolution was determined based on the optimization of cluster cohesion and separation via silhouette analysis to ensure cluster stability. Before presenting expression values and further analysis, the raw expression matrix underwent processing with the NormalizeData function. Differential expression analysis was performed using the FindAllMarkers function in Seurat to identify cluster-specific marker genes. Statistical significance was determined using the Wilcoxon rank-sum test, with markers defined as genes exhibiting adjusted p value<0.05 and log₂ fold change (FC) >0.25. To ensure the reliability of these markers, a minimum detection threshold of 10% was required in the clusters being compared.

The singscore levels of gene sets related to cytotoxicity, innate immune response, and inflammation response were performed using the irGESA R package (V.2.1.5). All gene sets were included in online supplemental table 2. Gene Set Enrichment Analysis (GSEA) was performed using the GSEABase R package (V.1.52.1) and gene sets from the MSigDB repository (V.2023.2).

Flow cytometry analysis

Dead cell exclusion was performed using the fixable viability stain Zombie Red (BioLegend, Cat# 423109). To stain cell surface antigens, single-cell suspensions were stained in FACS buffer, PBS with 0.5% fetal bovine serum (FBS), for 30 min at 4°C in the dark with fluorescence-conjugated antibodies. Before intracellular cytokine staining, cells were stimulated with stimulation cocktail plus protein transport inhibitor (eBioscience, Cat# 00–4975-03) for 4–6 hours at 37°C. Intracellular staining was then performed using fixation with or without permeabilization buffer (BD, Cat# 554714; eBioscience, Cat# 00–5523-00). Cells were fixed for 30 min at room temperature, permeabilized and intracellular antibodies were incubated for 30–60 min at 4°C protected from light. Flow cytometry was performed by BD Fortessa. Data were analyzed with FlowJo V.10.

The monoclonal antibodies were included in online supplemental table 3.

Spatial transcriptomic and proteomic profiling using Stereo-seq

Spatially resolved transcriptomic and proteomic analyses were performed using the Stereo-seq platform (BGI, China) following the manufacturer’s instructions.¹⁶ The optimal cutting temperature (OCT) embedded tissue was cut in the cryostat at a thickness of 10 µm and transferred onto a Stereo-seq chip. Next, the Stereo-seq chips with the tissue section were immediately warmed to 37°C for 5 min on a thermocycler adaptor. The tissue sections were fixed with pre-chilled methanol at −20°C for 5 min. Next, the chips were blocked for 20 min with the Blocking Buffer and incubated with the TotalSeq-A antibody mix (online supplemental table 4) in Blocking Buffer for 64 min at room temperature. After 4’,6-diamidino-2-phenylindole (DAPI) staining, fluorescence images were captured using a fluorescence microscope. Next, sections were permeabilized using permeabilization mix for an appropriate time determined by Stereo-seq Permeabilization Kit. After reverse transcription, complementary DNA (cDNA) and antibody derived tag (ADT) products were released from the chips by Release Enzyme Mix and collected, followed by pre-amplification. The cDNA fraction was separated from pre-amplified product by 0.8X beads selection, and was subjected to cDNA sequencing library preparation according to Stereo-seq Transcriptomics T Kit. The supernatant ADT product was cleaned up by 2.0X beads selection and amplified using Protein PCR primer Mix. ADT and cDNA libraries were sequenced with MGI DNBSEQ-T7 Sequencer. Stereo-CITE-seq library preparation and sequencing were performed by Shanghai Majorbio Bio-pharm Technology (Majorbio).

Stereo-seq data processing and analysis

Gene expression and ADT Fastq files were generated by MGI DNBSEQ-T7 or DNBSEQ-G400 Sequencer. Stereo-seq transcriptomic raw data were processed as described previously.¹⁷ The pipeline was publicly available at https://github.com/BGIResearch/SAW. Stereo-seq ADT Fastq files were generated by MGI DNBSEQ-T7 Sequencer. CID and MID are contained in read 1 (CID: 1–25 bp, MID: 26–35 bp) while read 2 consists of 21 bp PCR handle and 15 bp antibody barcodes. CID sequences on read 1 were first mapped to the designed coordinates of the in situ captured chip achieved from the sequencing, allowing 1 base mismatch. Reads with MID containing either N bases or more than 2 bases with quality score lower than 10 were filtered out. We used the in-house tag quantification pipeline to map antibody tags to their respective antibody barcodes. Finally, this information was used to generate a CID-containing ADT profile matrix.

The expression profile matrix was processed using the CellBin pipeline (BGI), which performs cell-level segmentation based on in situ spatial transcriptomic signals and nuclear morphology. Compared with fixed square-bin aggregation, CellBin enables cell-resolved transcriptomic profiling and reduces bias introduced by arbitrary bin size selection. Transcriptomic and proteomic expression matrices were clustered using SCANPY V1.9.1 33. The transcriptomic data were preprocessed by adjusting gene expression, selecting highly variable genes, and performing principal component analysis (PCA). Clustering was then performed using the Leiden algorithm and visualized with UMAP. For the proteomic data, a nearest-neighbor graph was constructed, followed by Leiden clustering and UMAP dimensionality reduction. Spatial visualization was performed with color-coded clusters. The joint analysis of transcriptome and proteome was performed using totalVI,¹⁸ and the resulting multi-omics model was clustered and visualized with UMAP and cluster spatial distribution.

Regions of interest (ROIs) were defined based primarily on the single-stranded DNA (ssDNA) image from the Stereo-seq section together with spatial transcriptomic features. Spatial gene expression patterns, CellBin-based cell segmentation results, and unsupervised clustering outputs were visualized using StereoMap (BGI) to guide ROI selection. Adjacent H&E-stained sections were reviewed as pathological references for tissue annotation. ROIs were selected to represent biologically relevant regions, including adipose tissues adjacent to tumor regions. Two ROIs were independently analyzed and subsequently combined for downstream statistical analyses.

To explore intercellular communication patterns, CellChat (V.2.1.2) was applied to infer ligand-receptor interactions based on the annotated clusters. Communication probability and pathway centrality were visualized using the built-in CellChat functions. All statistical analyses and figures generation were performed in R (V.4.3.2).

Isolation of adipocytes

We harvested human adipocytes according to a previously published protocol.¹⁹ Briefly, upper layers were collected from adipose after digestion and spun for 30 min at room temperature. Drastic temperature changes needed to be avoided. After centrifugation, cells were packed with three layers (from upper to bottom): free lipids, mature adipocytes, and wash buffer. An 18-gage needle and syringe were used to remove the free lipid and wash buffer. We collected mature adipocytes for further experiments.

Total RNA extraction and RT-qPCR

Total RNA extraction was prepared using TRIzol reagent (Invitrogen, Cat# 15596026CN) and the cDNAs were generated by reverse transcription (TAKARA, Cat# RR047A). Reverse transcription quantitative polymerase chaine reaction (RT-qPCR) was performed for all of the genes with SYBR green mix (AG, Cat# AG11701) using a Bio-Rad CFX Connect (Bio-Rad). The expression of messenger RNA (mRNA) for genes of interest was normalized to ACTB. A list of the primers used for RT-qPCR was provided in online supplemental table 5.

RNA-seq and data analysis

Library was prepared and sequencing was performed on an Illumina NovaSeq 6000 (BGI Genomics, Hangzhou, China). After removing any raw data with low-quality, adapter pollution or high levels of N bases, clean data were obtained and then aligned against the NCBI human reference genome via the HISAT2 software (V.2.0.4). Differential gene expression was analyzed using the DESeq2 R package (V.1.34.0). GSEA was performed using the GSEABase R package (V.1.52.1) and gene sets from the MSigDB repository (V.2023.2).

Detection of LTB4 production

Mature adipocytes were incubated in vitro with RPMI 1640 medium and 10% FBS for 96 hours. Cells were removed and supernatants were collected by centrifugation at 1,000 g for 10 min at 4°C. Supernatants were frozen at −80°C until thawed for LTB4 measurements using a detection assay (Elabscience, Cat# E-EL-0061).

Isolation and culture of human peripheral γδ T cells

Human peripheral γδ T cells were collected as in the previous report with adjustment.^{20 21} Briefly, PBMCs were resuspended in CTS AIM V SFM (Gibco, Cat# 0870112DK) with 1% HEPES (Gibco, Cat# 15630130), 1% L-glutamine (Gibco, Cat# 25030081), 0.1% 2-mercaptoethanol (Gibco, Cat# 21985023) at 1–2 million/mL. Cells were seeded onto 48-well plates precoated with 0.3 μg/cm² pan TCR γ/δ antibody (Beckman Coulter, Cat# IM1571U) and cultured for 3 days. Subsequently, cells were transferred to uncoated 48-well plates with rhIL-2 (PeproTech, Cat# 200–02-50UG) 10 ng/mL, rhIL-15 (PeproTech, Cat# 200–15-10UG) 10 ng/mL, and rhIL-21 (PeproTech, Cat# 200–21-10UG) 10 ng/mL. Cells were passaged every 3–5 days.

Expanded cells were assessed by flow cytometry before downstream experiments, and the majority of expanded γδ T cells exhibited a Vδ2 phenotype. Therefore, generally live CD3⁺Vδ2⁺cells were sorted for further experiments. Cell viability and purity were confirmed >90%. For LTB4 treatment, on day 9, 100 nM LTB4 with or without 1 µM etalocib was added to γδ T cells in the absence of exogenous cytokines for an additional 3 days.

Cytotoxicity and infiltration assay in vitro

For cell line cytotoxicity assays, target cells (cell lines or PDOs) and sorted γδ T or CD8⁺T cells were incubated (effector cells: target cells=5:1) for 6 hours at 37°C. Cytotoxicity was assessed using 7-aminoactinomycin (7-AAD) staining. CD45⁻7-AAD⁺ cells were measured as a percentage of cell death rate in target cells.

For PDOs, they were labeled by DiI (Beyotime, Cat# C1036) at 37°C for 20 min and washed three times with RPMI 1640 medium supplemented with 10% FBS. Immune cells were labeled by carboxyfluorescein succinimidyl ester (CFSE, eBioscience, Cat# 65–0850-84) following the same protocol. Labeled immune cells and PDOs were incubated (E:T=5:1) for 12–24 hours at 37°C for immune infiltration in 50% matrix (diluted by PBS), which was imaged once an hour using confocal (Olympus, FV3000). To analyze cell death rate, they were co-cultured for 6 hours at 37°C and stained by 7-AAD. DiI⁺7-AAD⁺ cells were measured as a percentage of cell death rate in target cells.

For NKG2A blockade assays, 20 µg/mL Monalizumab (MedChemExpress, Cat # HY-P99032) or isotype (MedChemExpress, Cat # HY-P99003) were incubated with NKG2A⁺γδ T cells before co-culturing with cell lines.

H&E staining and immunofluorescence staining

Tissues collected from patients with BC were obtained and fixed with 4% paraformaldehyde for 24 hours, embedded in paraffin and sectioned at 4 µm thickness for sections. The slices were deparaffinized and rehydrated with dimethylbenzene, ethanol, and distilled water in order according to the standard protocol and were stained with H&E, respectively.

For immunofluorescence staining, after being deparaffinized and rehydrated, slices were subjected to antigen retrieval using Tris-EDTA (pH=9) buffer in a thermostatted container at 98°C for 30 min. Sections were blocked with endogenous peroxidase blocking solution (Boster, Cat# AR1108) for 20 min and goat serum (Boster, Cat# AR0009) for 30 min at room temperature. Slices were stained with the following primary antibodies overnight at 4°C: T-cell receptor (TCR) γ/δ monoclonal antibody (Invitrogen, Cat# TCR1061; 1:200) and anti-NKG2A antibody (Abcam, Cat# ab260035, 1:200). After three washes with PBS containing Tween-20 (PBST), they were then incubated with Alexa Fluor-conjugated secondary antibodies (Abcam, Cat# ab150113, Cat# ab150078, 1:200) at room temperature for 2 hours. Both H&E and immunofluorescence section staining were scanned by 3D Histech for further analysis.

Isolation of myeloid cells

Tissues were digested into single-cell suspensions. CD11b⁺myeloid cells were isolated using anti-CD11b Microbeads (Miltenyi Biotec, Cat# 130–097-142) according to the manufacturer’s instructions. Briefly, cells were incubated with anti-CD11b Microbeads for 15 min at 4°C and subsequently applied to magnetic-activated cell sorting (MACS) columns (Miltenyi Biotec, Cat# 130–042-401) placed in a magnetic field. After washing, magnetically labeled CD11b⁺cells were retained in the column. Purified myeloid cells (purity >90%) were used for experiments.

Protein extraction and western blotting

Adipocytes and myeloid cells were lysed in radioimmunoprecipitation assay (RIPA) buffer (Beyotime, Cat# P0013B) supplemented with protease and phosphatase inhibitor cocktails (Yeasen Biotechnology, Cat# GRF101) on ice for 30 min. Lysates were centrifuged at 12,000×g for 15 min at 4°C, and the supernatants were collected.

Equal amounts of protein (20–40 μg) were mixed with 5×sodium dodecyl sulfate (SDS) loading buffer (Yeasen Biotechnology, Cat# LT101S), boiled at 95°C for 10 min, and separated by FuturePAGE 12% 15 Wells (ACE Biotechnology, Cat# ET15012Gel). Proteins were transferred onto NC membranes (Cytiva, Cat# GE10600002) using a wet transfer system, followed by blocking with 5% non-fat milk in TBS containing Tween-20 (TBST) for 1 hour at room temperature. Membranes were incubated overnight at 4°C with primary antibodies (HUABIO, Cat# ET1601-04; Abcam, Cat# ab169755), washed three times with TBST, and then incubated with HRP-conjugated secondary antibodies (HUABIO, Cat# HA1001) for 1 hour at room temperature.

STAT1 phosphorylation expression by flow cytometry

Human peripheral γδ T cells were stimulated with LTB4 (100 nM) for 1 hour. Cells were then immediately fixed with pre-cooling 4% paraformaldehyde for 15 min on ice, and permeabilized with methanol for 60 min at room temperature. Cells were stained with anti-STAT1 (phospho Y701) antibody (Abcam, Cat# 278713, 1:2000) and APC/Cy-7 anti-human CD45 (BioLegend, Cat# 304014). They were then incubated with Alexa Fluor 488-conjugated secondary antibody (Abcam, Cat# 150077) for 1 hour at room temperature protected from light. Flow cytometry was performed by BD Fortessa.

Confocal microscopy

Human peripheral γδ T cells were seeded in poly-L-lysine-coated coverslips (LABSELECT, Cat# 18122). Cells were stimulated with LTB4 (100 nM) for 1 hour, then fixed in 4% paraformaldehyde for 20 min at room temperature. Cells were permeabilized by 0.25% Triton X-100 in double-distilled water for 30 min. Samples were blocked with goat serum (Boster, Cat# AR0009) for 1 hour. Cells were incubated with the primary antibody, anti-pSTAT1 (Cell Signaling Technology, Cat# 9167, 1:50) overnight at 4°C, and with Alexa Fluor 647-conjugated secondary antibody (Abcam, Cat# ab150083, 1:200) for 1 hour at room temperature. Cell nuclei were stained with DAPI (Beyotime, Cat# C1005, 1:200). Coverslips were mounted onto microscope slides. Images were acquired using a confocal microscope (Olympus, FV3000).

Luciferase reporter assay

The human KLRC1 promoter region (−2 kb of Transcription Start Site) was synthesized and cloned into the pGL3-Basic luciferase reporter vector upstream of the firefly luciferase gene. The pGL3-control plasmid was transfected into the cells as a positive control. Luciferase reporter assay was performed using the Dual Luciferase Reporter Assay Kit (Vazyme, Cat# DL101) following the manufacturer’s protocol. Briefly, HEK293T cells were seeded into 24-well plates and co-transfected with the pGL3 reporter plasmid and pRL-TK vector using Lipofectamine 3000 Reagent (Invitrogen, Cat# L3000015). 24 hours after transfections, the cells were stimulated with rhIFN-γ (50 ng/mL) or LTB4 (100 nM) for 8 hours. Cells were lysed to detect the firefly and renilla luciferase activity. Firefly luciferase activity was normalized to renilla luciferase activity to reflect the relative promoter activity.

CUT&Tag-seq

The cleavage under targets and tagmentation sequencing (CUT&Tag-seq) was performed using the Hyperactive In-Situ ChIP Library Prep Kit (Vazyme, Cat# TD904-C2) according to the manufacturer’s instructions with modifications. Briefly, 2×10⁵ expanded human peripheral γδ T cells were collected with or without LTB4 (100 nM) for 2 hours. Cells were bound to concanavalin A-coated magnetic beads and incubated with anti-STAT1 antibody (1:50, Cell Signaling Technology, Cat# 14994T) at 4°C overnight, followed by incubation with a secondary antibody. Subsequently, pA-Tn5 transposase was added to bind antibody-chromatin complexes. Tagmentation was activated by Mg²⁺ at 37°C, and DNA fragments were then PCR-amplified with indexed primers, purified using magnetic beads, and sequenced on an Illumina platform provided by Hangzhou Lianchuan Biotechnology.

Patient-derived tumor fragment

PDTFs were established as previously described.²² Briefly, tumors were cut into 1–2 mm³ pieces. Tumor fragments from adipose-tumor interface regions were mixed to reduce heterogeneity. Samples from nine patients were then mixed with ice-cold matrigel (BD, Cat# 356231) and transferred to a 96-well flat-bottom plate under identical conditions (DMEM supplemented with 10% FBS, 1% penicillin-streptomycin, 1% GlutaMAX, 1% sodium pyruvate, and 1% non-essential amino acid). Each patient sample was processed and analyzed as an independent biological replicate.

For LTB4 treatment, 100 nM LTB4 was added for 3 days. Flow cytometry of T cells was then performed. The culture supernatant was collected for detecting cytotoxic cytokines (Multi Science, Cat# EK180, Cat# EK158, and Cat# EK182).

Mice

All animal experiments were performed in accordance with the institutional guidelines and were approved by the Animal Care and Use Committee of the SAHZU (2024–352). 4-week-old female NSG-rhIL15 mice (Nanfang Model Animals) were used for xenograft tumor models. MDA-MB-231 cells were resuspended in a 1:1 mixture of serum-free DMEM and matrigel (Yeasen, Cat# 40183ES08) at a final concentration of 2×10⁶ cells per 100 µL. The cell suspensions were subcutaneously injected. Tumor growth was monitored every 3–4 days, and tumor volume was calculated using the formula: (length×width²)/2.

On day 17 after tumor injection, mice received an intravenous injection via the tail vein of 10⁶ cells suspended in 100 µL PBS or 100 µL PBS. On day 30, mice were sacrificed, and blood, spleens, lungs, and tumors were harvested for further analysis.

Data and statistics

In analysis of RNA-sequencing, genes with an adjusted p<0.05 and logFC >0.5 were considered as differentially expressed genes. Gene set enrichments were analyzed using preranked GSEA, and results with p<0.05 and Q<0.05 were considered significant.

Statistical analyses were performed with Prism V.9 (GraphPad Software). Results are presented as means±SEM. Unpaired or paired, according to samples, two-tailed Student’s t-tests were used for two groups. For multiple comparisons, unpaired or paired, according to samples, one-way analysis of variances (ANOVAs) were used. Statistical significance was defined as p≤0.05.

Results

Enriched NKG2A⁺γδ T cells are identified at the adipose-tumor interface with Vδ2 usage

We first collected AT samples from women and observed that γδ T cells were also highly enriched in human breast AT compared with PB and LNs (online supplemental figure S1), consistent with the phenomenon observed in murine adipose depots.⁸ To further characterize γδ T cells in patients with BC, we sorted live CD45⁺cells or CD45⁺CD3⁺cells from paired T, PA, DA, PB, and LNs specimens of three patients with BC who received breast-conserving surgery for scRNA-seq following confirmation of cell purity (online supplemental figure S2A,B).

We first analyzed gene sets associated with cytotoxic activity across various tissues and found that γδ T cells from PA exhibited the highest cytotoxic potential (figure 1A). Their innate immune responses and inflammatory responses were comparable to those observed in tumors, whereas peripheral γδ T cells displayed lower expression levels of these features (online supplemental figure S2C,D). We further analyzed TCR repertoires in PA and DA. Notably, Vδ1 chains dominated in DA, whereas Vδ2 chains were significantly enriched in PA, which were confirmed by flow cytometry (figure 1B,C, and online supplemental figure S2E).

Figure 1. NKG2A⁺γδ T cells are enriched at the adipose-tumor interface of patients with breast cancer. (A) Box plots showing mean singscore of gene sets related to cytotoxicity across all γδ T cells within each organ. Gene sets were included in online supplemental table 2. (B) Each pie represented matched γ and δ chain genes detected by scRNA-seq in different tissues. Expanded clones were fused. Colored by δ usages. The light purple pie was expanded Vδ2Vγ9 cells. (C) Representative flow plots of δ chain usages in T cells (left). Summarized percentages of NKG2A expression are also shown (right). N=8. (D) UMAP representation of scRNA-seq data from patients with BC. Colored by γδ T cell clusters. (E) Percentages of different γδT cell clusters. Colored by γδ T-cell clusters. Increased proportion in PA were labeled by red color. (F) Spider plot showing the cytotoxic gene score of Cluster 2, Cluster 3, and Cluster 5 in γδ T cells across organs in patients with BC. The cytotoxic gene score was calculated as the average singscore value of cytotoxicity-associated gene sets (online supplemental table 2). PA-derived Cluster 3 was labeled by red color. (G) Heatmap analysis displaying the average expression levels of cytotoxic genes in Cluster 3 and other clusters of γδ T cells from varied organs. Cytotoxic genes were enriched in online supplemental table 2. Cluster 3 was labeled by red color. (H) Representative flow plots (left) and summarized percentages (right) of NKG2A expressions in γδ T cells from DA, PA, and T of patients with BC. N_(DA)=47; N_(PA)=47; N_(T)=15. (I) The absolute numbers of NKG2A⁺γδ T cells per million live cells in DA and PA assessed by flow cytometry. (H) Summarized percentages of δ chain usages in NKG2A⁺ and NKG2A^negγδ T cells by flow cytometry. N=6. (K) Locations of NKG2A⁺γδ T cells in patients with BC. H&E staining showed histological differences of DA and PA (left panels). Immunofluorescence staining of NKG2A⁺γδ T cells (middle and right panels). Yellow arrows indicate the regions magnified in the right panels. Red: NKG2A; green: TCR δ; blue: DAPI for nucleus. (L and M) The quantification of γδ T cells (K) and NKG2A⁺γδ T cells (l) per field localizing at DA and PA. N=5. Data shown as mean±SEM in H–J, L, and M. T-tests were used in C, I, L, and M. In H and J, one-way ANOVA test was used. P value≤0.05 was considered significant. ANOVA, analysis of variance; BC, breast cancer; DA, distant adipose; LN, lymph node; PA, peritumoral adipose; scRNA-seq, single-cell RNA sequencing; T, tumor; UMAP, Uniform Manifold Approximation and Projection.

By employing UMAP, we identified eight distinct clusters of γδ T cells (figure 1D) to categorize their functional subsets. Clusters 2, 3, and 5 were more prevalent in PA (figure 1E). Among them, Cluster 3 displayed the strongest killing signature in PA (figure 1F). Cluster 3 was enriched with multiple innate-like (KLRC1, KLRD1, NCR3, and CD160) and cytotoxic (GZMA, GZMB, GZMM, PRF1, TNF, and FASLG) genes compared with other clusters, particularly in PA (figure 1G). Given that KLRC1 (encoding NKG2A) is a key functional rheostat for immune cells in the TME,^23–25 we therefore defined Cluster 3 as NKG2A⁺γδ T cells, characterized by KLRC1 expression (online supplemental figure S2F,G). Flow cytometry analysis confirmed the selective enrichment of novel NKG2A⁺γδ T cells in PA (figure 1H,I, and online supplemental figure S2H). Our findings showed that NKG2A was highly expressed on cytotoxic γδ T-cell subset with Vδ2 usage, while programmed cell death protein 1 (PD-1) was mainly expressed on Vδ1 cells in AT (figure 1J, and online supplemental figure S2I).

Given the primary focus on the PME, direct quantitative comparisons with intratumoral regions were not systematically performed. Histological and immunofluorescence staining assays verified that γδ T cells, especially NKG2A⁺γδ T cells were preferentially enriched at the adipose-tumor interface (figure 1K–M). Together, these data capture a previously unrecognized γδ T subsets, NKG2A⁺γδ T cells, preferentially infiltrating into PA, highlighting preferential localization at the adipose-tumor interface.

NKG2A⁺γδ T cells are memory cells with stronger antitumor cytotoxicity

We then assessed the differentiation status and phenotypes of PA-infiltrating γδ T cells. Our result indicated that NKG2A⁺γδ T cells in PA tissue were primarily CD45RA⁻CD27⁺ central memory T cells and CD45RA⁻CD27⁻ effector memory T cells in AT (online supplemental figure S3A). Both scRNA-seq and flow cytometry showed increased expression of T-cell homing markers in NKG2A⁺γδ T cells, including SELL (encoding CD62L) and IL-7R (online supplemental figure S3B, figure 2A–C). Elevated TCF-1 expression further indicated a non-exhausted phenotype (figure 2A,D). In contrast, γδ T cells from DA showed no significant differences in IL-7R or TCF-1 expression (online supplemental figure S3C,D).

Figure 2. Features and potent cytotoxicity capacity of NKG2A⁺γδ T cells in peritumoral adipose. (A) Violin plots showing the expression levels of SELL, IL7R, and TCF7 from scRNA-seq data. (B–D) Representative flow plots and summarized percentage of CD62L (B), IL-7R (C), and TCF-1 (D) in NKG2A⁺γδ T cells and NKG2A^negγδ T cells from PA. N_(CD62L)=6. N_(IL-7R)=5. N_(TCF-1)=6. (E) GSEA displayed upregulated and downregulated gene sets of NKG2A⁺γδ T cells, compared with NKG2A^negγδ T cells from scRNA-seq data. Red bars: upregulated gene sets. Blue bars: downregulated gene sets. (F–H) Representative flow plots and summarized percentage of CD39 (F), IFN-γ (G), and granzyme B (H) in NKG2A⁺γδ T cells and NKG2A^negγδ T cells from PA. N_(CD38)=5; N_(IFN-γ)=5; N_{(granzyme B)}=5. (I) In vitro cytotoxicity assays were performed to detect the killing ability of γδ T cells against MDA-MB-231 after 6-hour co-culture. Cell death rates were assessed by 7-AAD staining. Black: MDA-MB-231 cells alone. Gray: cancer cells co-cultured with NKG2A^negγδ T cells. Red: cancer cells co-cultured with NKG2A⁺γδ T cells. (J) Confocal imaging of co-culture assays between γδ T cells and PDOs. Red: PDOs; Green: γδ T cells. Scaled bar: 100 µm. (K) The cytotoxic ability of γδ T cells against PDOs was performed by in vitro cytotoxicity assays. Tumor cell death rates were assessed by 7-AAD staining. Gray: PDOs co-cultured with NKG2A^negγδ T cells. Red: PDOs co-cultured with NKG2A⁺γδ T cells. (L) In vitro cytotoxicity assays were performed to detect the killing ability of γδ T cells against MCF-7 after 6-hour co-culture. Cell death rates were assessed by 7-AAD staining. Black: MCF-7 cells alone. Gray: cancer cells co-cultured with NKG2A^negγδ T cells and isotype (20 µg/mL). Red: cancer cells co-cultured with NKG2A⁺γδ T cells and isotype (20 µg/mL). Purple: cancer cells co-cultured with NKG2A^negγδ T cells and Monalizumab (20 µg/mL). Data were analyzed using t tests in B–D, F–H, and K. One-way ANOVA tests were performed in I and L. P value≤0.05 was considered significant. IFN, interferon; PA, peritumoral adipose; PDO, patient-derived organoid; scRNA-seq, single-cell RNA sequencing;7-AAD, 7-aminactinomycin D.

The GSEA of scRNA-seq data revealed that NKG2A⁺γδ T cells exhibited a natural killer (NK)-like phenotype and upregulated γδ T-cell activation signalings (figure 2E). NKG2A⁺γδ T cells showed higher expression of CD38 and PD-1 compared with NKG2A^negγδ T cells (figure 2F and online supplemental figure S4A). Moreover, PA-derived NKG2A⁺γδ T cells secreted significantly more interferon (IFN)-γ and granzyme B than both NKG2A^negγδ T cells and DA-derived γδ T cells (figure 2G,H, and online supplemental figure S4B). These findings suggest that PA-derived NKG2A⁺γδ T cells were activated and cytotoxic.

To functionally evaluate their cytotoxic activity, we sorted NKG2A⁺γδ T cells and NKG2A^negγδ T cells from PA tissues and co-cultured with the BC cell line and PDOs, respectively. Because tumor cells express HLA-E, the ligand of NKG2A, we first examined HLA-E expression on cell lines and PDOs using flow cytometry. We found that the MDA-MB-231 cell line and PDOs used in this study expressed low levels of HLA-E, whereas the MCF-7 cell line exhibited high HLA-E expression (online supplemental figure S4C). When co-cultured with MDA-MB-231 cells, NKG2A⁺γδ T cells exhibited potent cytotoxic activity against BC cells (figure 2I). In addition, after co-culture with PDOs, NKG2A⁺ γδ T cells exhibited enhanced infiltration into the organoids (figure 2J and online supplemental figure S4D). Strong cytotoxic effects were validated by a reduction in the organoid size and subsequent cell death (online supplemental figure S4E, figure 2K, and online supplemental video 1). PA-derived NKG2A⁺CD8⁺ T cells or DA-derived NKG2A⁺γδ T cells were less effective than PA-derived NKG2A⁺γδ T cells (online supplemental figure S4F–H). In contrast, after co-cultured with MCF-7, NKG2A⁺γδ T cells showed weaker cytotoxicity. Notably, NKG2A blockade, Monalizumab, restored their cell-killing capacity (figure 2I).

Overall, our findings demonstrate that NKG2A⁺γδ T cells in PA tissue are highly cytotoxic, possess robust activation potential, and are capable of infiltrating into tumor tissues to exert strong antitumor effects. Their cytotoxicity activity may be conditioned by HLA-E expression on tumor cells, and Monalizumab could restore their cell-killing capacity.

Spatial profiling reveals preferential localization and enhanced immune crosstalk of NKG2A⁺ γδ T cells at the adipose-tumor interface

To further validate the preference of NKG2A⁺γδ T cells at the adipose-tumor interface, we performed the spatial transcriptomic and protein profiling. Two ROIs located at the tumor-adipose interface were manually selected based on H&E staining and pathological annotation (figure 3A). We identified several cellular components (figure 3A,B, and online supplemental figure S5) via integrating proteomic and transcriptomic analysis. Spatial mapping of ROIs confirmed the localization of both NKG2A^neg and NKG2A⁺γδ T cells (figure 3C). Quantitative analysis suggested that NKG2A⁺γδ T cells tended to localize closer to cancer cells, although statistical significance was not reached (figure 3D). Transcriptome profiling further revealed that NKG2A⁺γδ T cells also exhibited enhanced cytotoxicity and reduced exhaustion markers (figure 3E). Gene Ontology enrichment analysis of spatial profiling also suggested activation of innate immunity in NKG2A⁺γδ T cells compared with that in NKG2A^negγδ T cells (online supplemental figure S6A).

Figure 3. NKG2A⁺γδ T cells spatially localize in peritumoral adipose and interact with antigen-presenting cells. (A) H&E staining (left panel) and mapped cell clusters (right panel, colored by cell clusters) of the spatial profiling sample. Two ROIs were labeled. Scaled bar: 1000 µm. (B) UMAP of cell composition of spatial profiling conducting transcriptional and protein expressions. Upper: UMAP plot. Lower: colored by cell clusters. (C) Spatial mapping of NKG2A⁺, NKG2A^negγδ T cells, and cancer cells in ROIs. Left: H&E staining. Right: Spatial mapping. Upper: ROI_1. Lower: ROI_2. Scaled bar: 500 µm. Red dots: cancer cells. Blue dots: NKG2A^negγδ T cells. Green dots: NKG2A⁺γδ T cells. Gray: background (other cells). (D) Box plot showing the proximity of NKG2A⁺ or NKG2A^neg γδ T cells to cancer cells in two ROIs. (E) Transcriptomic analysis of activation and exhaustion features in NKG2A⁺ and NKG2A^neg γδ T cells from spatial profiling data. (F) CellChat analysis showing cellular interaction of γδ T cells sending to other cell types. Thickness of the connecting lines was proportional to the number of interactions between two cell-nodes. Each cell type was labeled by different color. (G) Dot plot indicating APP-CD74 signaling of γδ T cells and other cell types. Sizes of dots were communication probability. Wilcoxon rank-sum test was used in D. DC, dendritic cell; NK, natural killer; CAF, cancer-associated fibroblast; ILC, innate lymphoid cells; ROI, regions of interest; UMAP, Uniform Manifold Approximation and Projection.

To assess the cellular crosstalk of γδ T cells within the TME, we applied CellChat analysis. Compared with NKG2A^negγδ T cells, NKG2A⁺γδ T cells exhibited more extensive communication networks, including cancer cells, B cells, and DCs (figure 3F). Intersection analysis highlighted enhanced APP-CD74 signaling between NKG2A⁺γδ T cells and other cells (online supplemental figure S6B), predominantly antigen-presenting cells (APCs), potentially indicating improved antigen presentation capacity. Among these, DCs were the primary signal recipients (online supplemental figure S6C). The strongest APP-CD74 interactions between DCs and NKG2A⁺γδ T cells also presented (figure 3G). Taken together, these results suggest that NKG2A⁺γδ T cells are spatially localized in PA, linking with cancer cells and APC-mediated antigen presentation at the adipose-tumor interface, which generate a potential local immune surveillance network at the adipose-tumor interface.

Peritumoral adipocytes selectively programs NKG2A⁺γδ T cells via the LTB4

We next investigated the potent factors driving NKG2A⁺γδ T cells accumulation. Given the nature of the breasts AT and NKG2A⁺γδ T cells localization, we hypothesized that adipocytes may program γδ T cells. To test this, we cultured DA-derived T cells with adipocytes isolated from different anatomical sites (figure 4A), using a previously established protocol.¹⁹ Interestingly, only γδ T cells upregulated NKG2A expression after co-culture with peritumoral adipocytes, while PD-1 expression remained unchanged. No significant changes were observed in CD4⁺ or CD8⁺ T (figure 4B and online supplemental figure S7). These findings suggest that NKG2A⁺γδ T cells are specifically induced by peritumoral adipocytes.

Figure 4. Peritumoral adipocyte-secreted LTB4 programs NKG2A⁺γδ T cells. (A) Diagram of isolation and culture distal or peritumoral adipocytes with DA-derived T cells. (B) Representative flow plots showing NKG2A expression on CD4⁺T, CD8⁺T, and γδ T cells (left). Summarized percentages of NKG2A expression are also shown (right). (C) GSEA shows upregulated gene sets in peritumoral adipocytes, compared with distal adipocytes from RNA-seq. (D) GSEA of arachidonic acid metabolism in peritumoral adipocytes, compared with distal adipocytes. (E and F) LTB4 concentrations in adipocytes and myeloid cells from DA and PA of BC patients, measured per 100,000 cells (E) or per gram of the tissue (F). N=9. (G) Representative flow plots and summarized percentages of NKG2A expression on γδ T cells in different conditions are shown. Demethylnobiletin: 5-LOX inhibitor, 2 µM. LTB4: 100 nM. (H and I) Representative histograms and summarized percentages of LTB4R expressions of T cells (H), NKG2A^negγδ T cells, and NKG2A⁺γδ T cells (I) in PA. N=6. (J) Flow plots and percentages of NKG2A expressions in γδ T cells. LTB4: 100 nM. Etalocib: LTB4R antagonist, 1 µM. (K) Flow plots and percentages of NKG2A expressions on γδ T cells were detected after LTB4 stimulation for 3 days with or without inhibitors. LTB4, 100 nM. Fludarabine: STAT1 inhibitor (iSTAT1), 100 nM. FLLL32: STAT3 inhibitor (iSTAT3), 50 nM. Alpelisib: PI3K inhibitor (iPI3K), 500 nM. (L) Phospho-STAT1 (pSTAT1) expressions of NKG2A^negγδ T cells and NKG2A⁺γδ T cells in PA. N=6. (M) pSTAT1 of γδ T cells was detected without (blue) or with LTB4 stimulation in vitro after 60 min (red) by flow cytometry. (N) Confocal imaging of pSTAT1 expressions without or with LTB4 stimulation in vitro in γδ T cells after 60 min. Pink: pSTAT1; Blue: DAPI for nucleus. Data in B and E–K shown as mean±SEM. Data were analyzed using t-tests in B, I, L, and M. One-way ANOVA tests were performed in E–H, J, and K. P value≤0.05 was considered significant. 5-LOX, 5-lipoxygenase; ANOVA, analysis of variance; APC, antigen-presenting cell; DA, distant adipose; GSEA, Gene Set Enrichment Analysis; LTB4, leukotriene B4; MFI, mean fluorescence intensity; NES, normalized enrichment score; PA, peritumoral adipose; RNA-seq, RNA sequencing.

To further dissect the underlying mechanism, we performed RNA-sequencing on adipocytes isolated from paired DA and PA samples and discovered distinct transcriptional profiles between the two adipocyte populations (online supplemental figure S8A–C). Histologically, peritumoral adipocytes appeared smaller than those from DA (online supplemental figure S8D), consistent with previous studies.²⁶ Genes associated with lipid metabolism also indicated increased lipolysis in PA adipocytes (figure 4C and online supplemental figure S8E). Given the upregulation of arachidonic acid metabolism (figure 4C,D), we next examined the three classical arachidonic acid metabolic pathways (online supplemental figure S8F). The mRNA expression levels of key enzymes, including PLA2G2A, ALOX5, and GPX3, were significantly increased in PA-derived adipocytes compared with those from DA, as assessed by RT-qPCR (online supplemental figure S8G). These enzymes are critical for the production of LTB4. Furthermore, the 5-lipoxygenase (5-LOX) pathway was specifically elevated in adipocytes at the protein levels (online supplemental figure S8H).

Although myeloid cells are a known important source of LTB4, the mRNA expression of these enzymes was comparable between PA and DA-derived myeloid cells (online supplemental figure S8G). Moreover, adipocytes vastly outnumbered myeloid cells in AT. Quantification confirmed elevated LTB4 levels in PA-derived adipocytes, and these increases were primarily attributable to adipocytes rather than myeloid cells in AT (figure 4E,F). Importantly, treatment with the 5-LOX inhibitor demethylnobiletin significantly reduced NKG2A expression on γδ T cells, without affecting CD4⁺ or CD8⁺ T cells. This effect was reversed on supplementation with exogenous LTB4 (figure 4G and online supplemental figure S9), confirming that LTB4 derived from PA adipocytes promotes NKG2A expression on γδ T cells. These findings suggest that lipolysis via the arachidonic acid pathway enhances LTB4 production by peritumoral adipocytes, which in turn serves as a key inducer of NKG2A expression on γδ T cells.

LTB4/LTB4R axis activates NKG2A⁺γδ T cells through the STAT1 signaling pathway

To explore the molecular mechanism of NKG2A⁺γδ T cell activation by LTB4, we first detected the LTB4 receptor (LTB4R) expression by flow cytometry. The result revealed higher LTB4R expression on γδ T cells compared with αβ T cells, which may explain the subset-specific polarization (figure 4H). Additionally, NKG2A⁺γδ T cells expressed higher levels of LTB4R than NKG2A^neg γδ T cells, which was verified by scRNA-seq, spatial profiling and flow cytometry assays (online supplemental figure S2F, figures 3E and 4I).

Next, to investigate the direct activation role of LTB4, γδ T cells were isolated from PB and stimulated with LTB4 in vitro. NKG2A expression was upregulated on γδ T cells, and the expression level decreased with the treatment of LTB4R antagonist etalocib (figure 4J). Canonical pathways downstream of LTB4/LTB4R signaling have been previously reported, including STAT1, STAT3, and PI3K.^{27 28} Accordingly, we conducted in vitro cultures with inhibitors targeting these pathways following LTB4 stimulation. The results showed that inhibitors of STAT1 (fludarabine, iSTAT1) reduced NKG2A expression with the most pronounced effect (figure 4K). In contrast, STAT3 and PI3K inhibition had no noticeable effect. These data suggest that STAT1 signaling mediates NKG2A upregulation in response to LTB4 stimulation.

Flow cytometry also revealed that NKG2A⁺γδ T cells in PA exhibited significantly higher phospho-STAT1 (pSTAT1) than NKG2A^negγδ T cells (figure 4L). We next isolated γδ T cells to determine whether LTB4 directly induces STAT1 activation in vitro. As expected, pSTAT1 levels increased following LTB4 stimulation (figure 4M). Confocal imaging further confirmed that LTB4 promoted the nuclear translocation of pSTAT1 (figure 4N).

To explore how STAT1 activation enhances NKG2A expression, we cloned the KLRC1 promoter region into a luciferase reporter construct and demonstrated that IFN-γ induced STAT1 activation enhanced KLRC1 transcription. A similar effect was also observed following LTB4 treatment (online supplemental figure S10A). We then used the JASPAR database to screen their potential STAT1 binding sites within the KLRC1 promoter region, and the top three predicted binding sites were shown (online supplemental figure S10B,C). These findings suggest that activated STAT1 may directly bind to the KLRC1 promoter, thereby promoting KLRC1 transcription. To confirm the transcription factor-DNA binding, we further analyzed public datasets. In CD4⁺T cells, activated STAT1 was enriched at the Klrc1 promoter locus near its transcription start site (<2 kb) (online supplemental figure S10D,E). We next performed CUT&Tag-seq in human γδ T cells and found that LTB4-treated cells exhibited increased STAT1 binding peaks in the KLRC1 promoter locus (online supplemental figure S11F). Together, these results demonstrate that LTB4/LTB4R signaling promotes STAT1 activation in γδ T cells, which directly binds to the KLRC1 promoter region, thereby driving the polarization and cytotoxic function of NKG2A⁺γδ T cells.

Abundance of NKG2A⁺γδ T cells correlated with favorable clinical outcomes

We then examined the association of NKG2A⁺γδ T cells with clinical features and patient survival. We total detected samples from patients with BC (N=47) and analyzed the correlation between the number of NKG2A⁺γδ T cells in AT with clinical pathological features. The highest abundance of NKG2A⁺γδ T cells was observed in ductal carcinoma in situ (DCIS), while the number of NKG2A⁺γδ T cells in PA was gradually reduced with the tumor progression from DCIS to invasive tumors and further decreased from T1 stage to T2 stage (figure 5A). Clinically, the abundance of NKG2A⁺γδ T cells markedly differed across different molecular subtypes of invasive BC and was notably higher in non-luminal subtypes (figure 5B). However, their levels were significantly reduced in invasive patients with BC with a body mass index (BMI) over 24 kg/m² (figure 5C). Additionally, breast composition also influenced their distribution. PA from fatty breasts exhibited a greater abundance of NKG2A⁺γδ T cells compared with that from dense breasts (figure 5D). Notably, the abundance of NKG2A⁺γδ T cells was not associated with axillary lymph node metastasis or patient ages (online supplemental figure S11A,B). Importantly, a higher abundance of NKG2A⁺γδ T cells was positively correlated with better overall survival (OS) and recurrence-free survival (RFS) in patients with BC (figure 5E,F). LTB4R expressions of γδ T cells greatly contributed to favorable OS and RFS (figure 5G,H).

Figure 5. The abundance of NKG2A⁺γδ T cells is associated with clinical outcomes. (A) Percentages of NKG2A⁺γδ T cells in DCIS and invasive patients with BC. Invasive_T1: tumor size of invasive BC no more than 2 cm. Invasive_T2: tumor size of invasive BC more than 2 cm. (B and C) The NKG2A expressions in γδ T cells of invasive patients with BC with different molecular subtypes (B) and BMI (C) assessed by flow cytometry. Luminal: hormone receptor-positive BC tumors; non-luminal: hormone receptor-negative BC tumors. BMI (kg/m²) was calculated by dividing weight by the square of height. (D) The absolute numbers of NKG2A⁺γδ T cells per 1 million live cells in DA and PA with breast density assessed by flow cytometry. Breast density was determined using the ACR classification of breast mammography, categorizing as fatty (A–B) or dense (C–D). (E and F) The Kaplan-Meier survival curves/log-rank tests were used to compare OS (related to E) and RFS (relating to F) in groups with high and low numbers of NKG2A⁺γδ T cells in patients with BC. Cut-off values were automatically determined using the minimum p value approach. (G and H) The Kaplan-Meier survival curves/log-rank tests were used to compare OS (G) and RFS (H) in groups with high and low numbers of LTB4R⁺γδ T cells in patients with BC. Cut-off values were automatically determined using the minimum p value approach. (I–L) The Kaplan-Meier survival curves/log-rank tests of NKG2A⁺γδ T cell enrichment and OS in soft tissue tumors (I), bladder cancers (J), colon cancers (K), liver cancers (L). Cut-off values were automatically determined using the minimum p value approach. Data in A–D shown as mean±SEM. Data were analyzed using ANOVA tests in A–D. P value≤0.05 was considered significant. ANOVA, analysis of variance; BC, breast cancer; BMI, body mass index; DA, distant adipose; DCIS, ductal carcinoma in situ; OS, overall survival; RFS, PA, peritumoral adipose; RFS, recurrence-free survival.

To verify whether the association between NKG2A⁺γδ T cells and AT extends to other adipose-rich tumors, we analyzed The Cancer Genome Atlas (TCGA) and Clinical Proteomic Tumor Analysis Consortium (CPTAC) public datasets. Soft tissue tumors showed a positive correlation between NKG2A⁺γδ T cells enrichment and favorable prognosis (figure 5I). Similarly, in other AT-rich tumors, such as bladder and colon cancers, we found that NKG2A⁺γδ T cells were also significantly positively correlated with improved clinical outcomes (figure 5J,K). Liver cancer, which is frequently accompanied by adipocyte accumulation, demonstrated a positive correlation (figure 5L). In contrast, no significant differences were observed in non-AT-rich tumors, including lung adenocarcinoma, brain tumors, and hematologic malignancies (online supplemental figure S11C–F). Collectively, these results suggest that the prognostic relevance of NKG2A⁺γδ T cells extends broadly across AT-enriched tumors.

LTB4 enhances γδ T cell-mediated immune responses in human samples

To assess the immunomodulatory effects of LTB4 in a clinically relevant context, we employed the PDTF strategy to more accurately recapitulate the human TME.²² Briefly, tumor tissues obtained from patients with BC were cut into small fragments and cultured in vitro with or without exogenous LTB4 (figure 6A). The culture supernatant was then collected, and tumor-infiltrating immune cells were assessed by flow cytometry. Consistent with our previous findings, γδ T cells exhibited upregulated expression of NKG2A, suggesting their active engagement and modulation within the TME (figure 6B). Given the cell-interaction within the TME by spatial profiling (figure 3F,G), we further analyzed antigen presentation cells. Interestingly, both DCs and macrophages upregulated HLA-DR and CD86 expression, suggesting they acquired enhanced antigen presentation capacity at the adipose-tumor interface (figure 6C–E).

Figure 6. LTB4 improves antitumor effects in the tumor microenvironment. (A) Schematic representation of PDTF collection and culture. (B) Representative flow plots and summarized percentages of NKG2A expression on γδ T cells in PDTF. N=9. (C–E) Representative histograms and summarized percentages of HLA-DR or CD86 expressions on DCs (C) and macrophages (D and E). N=7. (F) The quantification data of cytotoxic cytokines between LTB4 untreated and treated conditions. N=9. (G) The correlation between the proportion of NKG2A⁺γδ T cells and the concentration of cytotoxic cytokines, including IFN-γ, granzyme B, and TNF-α. Blue plot: data from control groups. Red plot: data from LTB4-treated groups. Paired t-tests were performed in B−F. G was measured by Pearson tests. P value≤0.05 was considered significant. DC, dendritic cell; IFN, interferon; LTB4, leukotriene B4; PDTF, patient-derived tumor fragment; TNF, tumor necrosis factor.

Cytotoxic cytokines in the PDTF culture supernatant were quantified using ELISA. Notably, LTB4 treatment significantly increased the levels of cytotoxic cytokines, including IFN-γ, granzyme B, and TNF-α, indicating a reprogrammed TME favoring antitumor immunity (figure 6F). Moreover, the concentrations of these cytotoxic cytokines positively correlated with the percentages of NKG2A⁺γδ T cells (figure 6G). These findings indicate that LTB4 not only promotes γδ T-cell activation but also enhances their functional capacity to mediate antitumor immune responses in the setting of PDTF. These findings demonstrate that LTB4 promotes antitumor effects in TME.

LTB4-treated γδ T cells control tumor progression

We next investigated whether LTB4 could strengthen the activity of γδ T cells. γδ T cells are reported species-specific differences of cell biology in humans and mice.²⁹ Therefore, we used human samples to conduct more clinical translational studies. Human γδ T cells were expanded in vitro according to the previously established system.²¹ At day 10, LTB4 was added with or without inhibitors (etalocib and iSTAT1) for an additional 3 days (figure 7A). Both etalocib and iSTAT1 significantly reduced the production of cytotoxic cytokines (figure 7B). Total γδ T cells and NKG2A^neg/+γδ T cells were subsequently sorted for cell-killing assays. The cytotoxicity capacity of total γδ T cells was significantly increased following LTB4 treatment, and this effect was ameliorated by etalocib. Notably, the enhancement was only observed in NKG2A⁺γδ T cells, rather than in NKG2A^negγδ T cells (figure 7C).

Figure 7. LTB4-treated γδ T cells exhibit enhanced antitumor activity in vitro and in vivo. (A) Schematic overview of the cytotoxicity assay using LTB4-treated expanded γδ T cells. (B) Representative flow plots and summarized percentages of IFN-γ⁺TNF-α⁺ expressions in γδ T cells after LTB4 stimulation for 3 days without or with etalocib and iSTAT1. N=5. (C) Representative flow cytometry plots (left) and summarized data of tumor cell killing activity (right) are shown. N=3. (D) Schematic presentation of the in vivo ACT experiment. N=5 mice per group. (E) Tumor growth curves of mice over time following PBS or ACT treatment. (F and G) Quantification of the tumor volume (F) and the tumor weight (G) at the experimental endpoint. (H) Representative flow plots of tumor-infiltrating γδ T cells and summarized percentages of NKG2A expressions on γδ T cells across indicated organs after ACT in vivo. (I) Representative flow plots and summarized percentages of IFN-γ⁺TNF-α⁺ expressions on γδ T cells isolated from tumors after ACT in vivo. Data in B, C, and F−I shown as mean±SEM. ANOVA tests were performed in B, C, and F−H. Data were analyzed using t-tests in I. P value≤0.05 was considered significant. ACT, adoptive cell transfer; ANOVA, analysis of variance; FACS, fluorescence-activated cell sorting; IFN, interferon; LTB4, leukotriene B4; PBS, phosphate-buffered saline; PBMC, peripheral blood mononuclear cell; TCR, T-cell receptor; TNF, tumor necrosis factor.

To further evaluate the antitumor efficacy of LTB4-treated γδ T cells against tumors in vivo, we established xenograft tumor models in NSG (rhIL-15) mice using the MDA-MB-231 cell line. After tumor establishment, 10⁶ expanded Vδ2 T cells with or without LTB4, were adoptively transferred into tumor-bearing mice (figure 7D, online supplemental figure S12A). After sorting, transferred γδ T cells were readily detected in murine blood, spleens, lungs, and tumors (online supplemental figure S12B–D). Mice receiving LTB4-treated γδ T cells exhibited significantly smaller tumor sizes and lower tumor weights (figure 7E–G), suggesting that LTB4-treated γδ T cells retained enhanced cytotoxicity in vivo. Furthermore, LTB4-treated γδ T cells maintained higher NKG2A expressions within tumors, whereas the expressions of inhibitory molecules, including CD39 and TIM-3, were reduced (figure 7H, online supplemental figure S12E,F). Remarkably, LTB4-treated γδ T cells also produced more cytotoxic cytokines in vivo (figure 7I).

Collectively, our results demonstrate that LTB4 is a promising target for adoptive immunotherapy and reprograms the tumor immune microenvironment through activation of NKG2A⁺γδ T cells.

Discussion

Adipose tissue is composed of diverse innate immune cells and is also increasingly recognized as an essential immune organ. However, its roles in AT-rich tumor pathogenesis and immune surveillance at the adipose-tumor interface remain insufficiently understood. In this study, we identify a novel subset of PA-infiltrating NKG2A⁺γδ T cells that contribute to antitumor defense in patients with BC. The phenotypes and functions of these cells are characterized by: (1) predominant Vδ2 usage and enrichment in PA tissues, particularly at the adipose-tumor interface; (2) potent antitumor potential, demonstrated through scRNA-seq and spatial profiling analysis; (3) activation by LTB4 secreted from peritumoral adipocyte via downstream STAT1 signaling; and (4) a strong association with favorable clinical outcomes in BC and other AT-rich tumors.

γδ T cells play a critical role in immune surveillance. Wu et al identified innate-like, tissue-resident Vδ1⁺T cells in the breasts of healthy women, which are also correlated with improved survival in patients with BC.³⁰ Our previous studies demonstrated that CD73⁺γδ T cells infiltrated tumors of patients with BC, different from those in normal breast tissue^{12 13} and remodeled the immunosuppressive TME through adenosine production. In the present study, we analyzed γδ T cells from various paired tissues of patients with BC. In contrast to earlier findings in healthy breast tissue, we discovered that NKG2A⁺γδ T cells are specifically enriched in PA and are predominantly localized at the adipose-tumor interface. Notably, these NKG2A⁺γδ T cells in PA are characterized by predominant Vδ2 chain usage and display strong cytotoxic activity. The abundance of PA-infiltrating NKG2A⁺γδ T cells is positively correlated with better clinical outcomes in patients with AT-rich tumors, suggesting a crucial immune surveillance role at the adipose-tumor interface. These findings also underscore key differences between healthy and tumor-bearing breast tissues. Collectively, γδ T cells with distinct TCR chains collaboratively contribute to maintaining homeostasis in the human breasts under both physiological and pathological conditions such as BC.

NKG2A was initially considered as an inhibitory receptor on NK cells,³¹ with its expression induced by γ_c family cytokines such as interleukin (IL)-2 and IL-15.^{32 33} Its ligand, HLA-E, is an unconventional class I HLA molecule. Pan-cancer analyses have revealed a lack of HLA-E expression in most cancers,³⁴ which was also observed in our PDOs, where HLA-E expression remained relatively low. Recently, NKG2A expression has also been reported on CD8⁺T cells, which correlates with the therapeutic response to anti-Lag3 and anti-PD-1 immunotherapies.^{35 36} Notably, NKG2A expression appears after PD-1 upregulation on CD8⁺T cells and is associated with improved OS in bladder cancer.³⁷ Tumor-reactive NKG2A⁺CD8⁺T cells are enriched in the tumor intraepithelial compartment.³⁸ As PD-1⁺γδT cells were reported to show opposite functions to PD-1⁺CD8⁺ T cells,²¹ we also observed that NKG2A⁺γδ T cells exhibited contrary function to previously reported immunosuppressive NKG2A⁺CD8⁺ T cells with elevated levels of activation markers and cytotoxicity-related molecules, enabling them to potently kill cancer cells and PDOs in vitro. In addition, NKG2A⁺γδ T cells displayed the stronger killing signature in PA than that in PB (figure 1F). These findings indicate that the same phenotype expressed on different cell types, or the same cell subpopulation in different anatomical locations, have different functions. Interestingly, GSEA analysis revealed a downregulation of the Fc receptor mediated stimulatory signaling pathway in NKG2A⁺γδ T cells, suggesting that their cytotoxicity is more reliant on direct cell killing rather than antibody-dependent cellular cytotoxicity. Moreover, PA-derived NKG2A⁺γδ T cells displayed superior cancer cell-killing potentials compared with their peripheral counterparts, suggesting local reprogramming of these cells. Monalizumab, a monoclonal antibody that blocks NKG2A, was developed to enhance antitumor effects by targeting CD8⁺T cells and NK cells.²⁵ However, their functional efficacy has been reported to depend on TIGIT expression on CD8⁺T cells.³⁷ Current clinical studies based on Monalizumab show a poor trend in head and neck squamous cell carcinoma.³⁹ Considering that ligand expression is varied in many tumors, the functional heterogeneity of the NKG2A/HLA-E across tumor types and cellular contexts should also be taken into account. Previous studies have shown that CD4⁺effector T cells can interact with MHC-II⁺CD11c⁺APCs at the tumor-invasive margins, thereby enhancing antitumor immunity.⁴⁰ Similarly, CD56^brightCD16^lowNK cells have been reported to suppress cDC1 recruitment.⁴¹ Our findings suggest that PA-derived NKG2A⁺γδ T cells may regulate the TME beyond direct cell-killing capacity. Spatial transcriptomic analysis and PDTFs revealed that NKG2A⁺γδ T cells elevate MHC-II expression on both DCs and macrophages. This interaction with DCs could underlie the NK-like functional properties of γδ T cells⁴² and reach a broader immunoregulatory role. However, downstream mechanisms and functional consequences of this crosstalk remain to be elucidated.

Fatty breasts, as opposed to dense breasts, are associated with a reduced risk of developing BC.³ Autologous adipose transfer does not increase local or regional recurrence in patients with BC who have undergone mastectomy and reconstruction⁴³ breast-conserving surgery followed by radiotherapy showed improved 10-year overall and relative survival compared with mastectomy in early BC.⁴⁴ This evidence suggests a potential protective role of AT. In our study, we also observed the immunoprotective effect of NKG2A⁺γδ T cells within PA at the early-stage of the BC, and their abundance decreased from DCIS to invasive BC, with T-stage progression. In addition, we identified significant enrichment of the arachidonic acid metabolic pathway in peritumoral adipocytes, leading to the production of LTB4, which specifically upregulates NKG2A expressions on γδ T cells. While the immunoregulatory effects of arachidonic acid metabolism are well established, most studies have focused on the classical cyclooxygenase and cytochrome P450 pathways.^45–47 In contrast, LOX and its metabolite LTB4 are well-known lipid chemoattractants that play key roles in inflammation and infection.^{27 48 49}

Although previous studies have identified myeloid cells as the primary source of LTB4,⁴⁸ we found that peritumoral adipocytes, rather than myeloid cells, are the major cellular source of LTB4 in AT. This observation is consistent with a recent study showing that adipocyte-derived LTB4 promotes thermogenesis under cold-exposure conditions.⁵⁰ LTB4 activates its G-protein coupled receptors, including BLT1 (high-affinity) and BLT2 (low-affinity).⁵¹ The expression and prognostic relevance of LTB4R vary across different cancer types, with high expression levels associated with improved OS in patients with BC.⁵² We discovered that γδ T cells express higher levels of LTB4R compared with CD4⁺ and CD8⁺T cells, indicating greater responsiveness to LTB4. LTB4-mediated signaling activates multiple downstream pathways, including JAK/STAT, ERK, PI3K/AKT, and Ca2⁺ influx.^{27 28 49 53 54} Notably, we observed elevated pSTAT1 levels in NKG2A⁺γδ T cells. Previous studies have shown that effector γδ T cells exhibit a marked upregulation of STAT1, whereas regulatory γδ T cells preferentially express STAT5A.⁵⁵ STAT1 signaling is also critical for T-cell self-renewal and effector function, and it promotes the differentiation and activity of T helper 1 cells.^56–58 Using CUT&Tag-seq analysis, we further demonstrated that STAT1 directly binds to chromatin regions associated with the KLRC1 gene in immune cells. Using an STAT1 inhibitor, we confirmed that LTB4-mediated polarization of NKG2A⁺γδ T cells and their production of IFN-γ and granzyme B and cytotoxic activity are STAT1-dependent. γδ T cells have emerged as promising candidates for adoptive cellular immunotherapy, and multiple studies have focused on strategies to enhance their in vivo antitumor activity, and in vitro activation during expansion has been considered a potential strategy to improve their therapeutic efficacy.^{14 59} In the present study, we transiently stimulated γδ T cells with LTB4 during in vitro expansion and evaluated their antitumor activity following adoptive transfer. Notably, LTB4-treated γδ T cells maintained enhanced long-term cytotoxicity and antitumor effects in vivo.

Our results deepen the understanding of γδ T cells in antitumor immunity. We are the first to identify a novel subset of NKG2A⁺γδ T cells, predominantly composed of Vδ2 chains, which forms an immune defense barrier at the adipose-tumor interface, distinct from the Vδ1⁺cells typically found in healthy AT. We demonstrate that NKG2A⁺γδ T cells possess potent cytotoxic activity. Furthermore, we show that peritumoral adipocyte-derived LTB4 activates the STAT1 signaling pathway in γδ T cells, leading to upregulation of NKG2A expression and enhanced tumor-killing function. Using the PDTF approach, we also demonstrate that LTB4 reprograms the immunosuppressive TME by increasing NKG2A expression on γδ T cells, thereby promoting antitumor immunity. These findings highlight a promising strategy for expanding and polarizing γδ T cells in therapeutic applications. In addition, we observe a higher accumulation of NKG2A⁺γδ T cells in patients with non-luminal BC compared with those with luminal pattern. Notably, in patients with a BMI >24 kg/m², local NKG2A⁺γδ T-cell numbers are reduced, potentially due to impaired adipocyte lipolysis. In contrast, a higher abundance of NKG2A⁺γδ T cells is observed in fatty breasts and early-stage BC. Overall, the presence of local NKG2A⁺γδ T cells is associated with reduced risk and better clinical outcomes in BC. Based on our findings, we further investigated the impact of the interactions between AT and NKG2A⁺γδ T cells across multiple tumor types. Interestingly, soft tissue tumors, which share characteristics with AT, showed a strong positive correlation with NKG2A⁺γδ T-cell infiltration. Similarly, positive associations were also observed in tumors anatomically adjacent to or significantly influenced by AT. However, in tumors lacking contact with AT, including non-solid and brain tumors, NKG2A⁺γδ T cells exhibited no significant association with clinical prognosis. These observations suggest that the contribution of NKG2A⁺γδ T cells to antitumor immunity may be context-dependent, with their prognostic relevance largely restricted to AT-rich TME.

In conclusion, our study identifies a novel subset NKG2A⁺γδ T cells and elucidates their role in tumor immune surveillance. We show that peritumoral adipocyte-derived LTB4 enhances the infiltration and cytotoxicity of NKG2A⁺γδ T cells in the PA, thereby preventing tumor invasion. These findings have important implications for the development of γδ T cell-based immunotherapies in AT-rich tumor treatment.

Data availability statement

Data are available upon reasonable request.