KLRG1 Cell Depletion as a Novel Therapeutic Strategy in Patients with Mature T-Cell Lymphoma Subtypes

Mature T-cell and NK-cell neoplasms comprising 34 subtypes represent a group of rare and heterogeneous neoplasms (1). Despite the introduction of anthracycline-based chemotherapy protocols, with or without consolidative hematopoietic transplantation and a plethora of new agents, the progression-free survival (PFS) of patients with advanced peripheral T-cell lymphoma (PTCL) needs to be improved (2, 3). One approach to treating malignancies is the use of therapeutic monoclonal antibodies (mAb) that bind to surface molecules on neoplastic cells and mediate effector cell function, such as antibody-dependent cellular cytotoxicity (ADCC) and antibody-dependent cellular phagocytosis (ADCP). Killer-cell lectin-like receptor G1 (KLRG1) is an inhibitory cell-surface receptor expressed predominantly on late-differentiated effector and effector memory CD8⁺ T (Tem and TemRA) and NK cells (4, 5). Accordingly, its expression on mature T-cell and NK-cell neoplasms might be expected and a therapeutic antibody designed to deplete KLRG1⁺ cells might be effective for these disorders.

Here, we define the molecular epidemiology of surface KLRG1 across various subtypes of mature T-cell and NK-cell neoplasms and the potential for anti-KLRG1 mAbs to deplete lymphoma cells from in vitro cell lines, in vivo xenografts, and ex vivo primary patient samples. Our group has previously demonstrated that the PI3K-δ/γ inhibitor (PI3Ki), duvelisib can induce a shift in TAM from the immunosuppressive M2-like to the inflammatory M1-like phenotype (6). We now demonstrate the efficacy of a first-in-class humanized afucosylated anti-KLRG1 mAb alone and in combination with the duvelisib in enhancing macrophage-mediated clearance of lymphoma cells in TCL xenografts.

Jurkat, H9, HuT-78, MJ, HH, and Raji cells were obtained from ATCC. L-82, SR-786, KI-JK, SUP-M2, SU-DHL1, DERL-7, and DERL-2 cells from DSMZ, Karpas 299 (K-299) and Karpas 384 cells from Sigma-Aldrich, DL-40, MTA, and KHYG-1 cells from JCRB. FEPD, MAC2A, OCI-Ly13.2, NKL, HuT-102, SMZ1, SNK6, Myla, SeAx, and OCI-Ly12 cells were acquired as previously described (7). CHO wild-type (CHO-WT) and stably transfected with human KLRG1 construct (CHO-KLRG1) cells were provided by Abcuro, Inc. All cell lines were routinely tested for mycoplasma detection with the MycoSEQ Mycoplasma Detection Assay from Applied Biosystems prior to use and authenticity was validated by STR profiling. Cellular populations such as human CD4, CD8, and NK cells were isolated from healthy donors using EasySep isolation kits from STEMCELL Technologies. CD8⁺ Tn, CD8⁺ Tcm, CD8⁺ Tem, and CD8⁺ TemRA populations were isolated using FACS utilizing a cocktail of Zombie Aqua, CD8-FITC, CCR7-PE-Cy7, and CD45RA-BV421.

The human T-cell lymphoblast leukemia cell line, Jurkat and PTCL-NOS cell line, SMZ1 were chosen for the generation of stable T-cell lymphoma cell lines expressing full-length KLRG1 protein. Jurkat, SMZ1, and 293T cells were maintained at RPMI-1640 with 10% and 20% fetal bovine serum (FBS), respectively, and DMEM with 10% FBS, respectively. 293T cells were transfected with packaging plasmids psPAX2 and pMD2.G (Addgene) together with pLOC-KLRG1 plasmid (Horizon). After 48 hours, the lentiviral-contained supernatants were collected and filtered through a 0.45 μm filter for the following transduction. Jurkat and SMZ1 were washed with PBS and resuspended with 8 μg/mL polybrene containing lentiviral medium and incubated at 37 degrees for 3 days. Following the incubation, infected Jurkat and SMZ1 cells were maintained under Blasticidin selection and confirmed to have KLRG1 surface expression by flow cytometry using the 13F12F2 clone.

Unconjugated and PE-conjugated anti-human KLRG1 clone, 13F12F2 and its isotype control murine IgG2aκ were obtained from eBioscience. Primary conjugated antibodies against human CD3-AF488 (OKT3), CD4-PB (RPA-T4), CD8-APC-Cy7 (SK1), CD16-APC-Cy7 (3G8), CD56-PB (5.1H11), CD57-APC (HNK-1), CD94-PE/Cy7 (DX22), CD14-APC (63D3), murine CD11b-APC (M1/70), CCR7-PE-Cy7 (G043H7), CD45 RA-BV421 (HI100), I-A/I-E-PB (M5/114.15.2), CD206-PE (C068C2), and Zombie Aqua (ZA) viability stain were purchased from BioLegend. Functional grade anti-human CD47 mAb, MIAP410, and murine IgG1κ isotype control were obtained from Bio X Cell. Proprietary unconjugated and conjugated anti-human KLRG1 antibodies GA015 and GA015-AF488 and its human IgG1 isotype controls, mAb008 and its murine IgG2cκ isotype control, mAb208 and its murine afucosylated IgG2aκ isotype controls were provided by Abcuro, Inc. Mouse complements were obtained from Innovative Research. Human male AB plasma as a source of human complement was obtained from Sigma-Aldrich. Rituximab, the human anti-CD20 mAb, and staurosporine were purchased from Selleckchem, whereas duvelisib was purchased from MedChemExpress. Liposaosmal clodronate (Clodrosome) was purchased from Encapsula Nano Sciences. A table summarizing the pertinent properties of various anti-KLRG1mAbs used in the manuscript is included in the Supplementary Methods section. Different antibodies had to be used due to their inherent characteristics as some antibodies were better for flow cytometry, some better for IHC, some were functional with an afucosylated Fc and some were depleting versus others remained unknown or were neutralizing, some proprietary, and some commercial grade. Different assays required different antibodies with unique properties and hence different antibodies have been used in the manuscript.

Various CD8⁺ T-cell subsets, bulk CD8⁺ T cells, and CD4 and NK cells were isolated from healthy donors and submitted for whole-transcriptome sequencing. Human blood CD14⁺ monocytes were prepared from normal donor apheresis leukoreduction collars (Crimson Core, Brigham, and Women's Hospital) using negative selection kits (STEMCELL Technologies) and cryopreserved. Monocytes were differentiated to macrophages by culturing in RPMI supplemented with 10% FBS, 100 U/mL penicillin, 100 μg/mL streptomycin, and human M-CSF (100 ng/mL) for approximately 7 days. These human monocyte-derived macrophages (hMDM) were treated with duvelisib or DMSO-control media for 24, 48, and 72 hours. Supernatants were harvested at the above time points to assess human cytokine secretion by multiplexed analysis (Eve Technologies). Macrophages were harvested and subjected to RNA extraction and downstream bulk RNA-seq. Libraries were prepared using Roche Kapa mRNA HyperPrep strand-specific sample preparation kits from 200 ng of purified total RNA according to the manufacturer's protocol on a Beckman Coulter Biomek i7 (8). The finished dsDNA libraries were quantified by Qubit fluorometer and Agilent TapeStation 4200. Uniquely dual-indexed libraries were pooled in an equimolar ratio and shallowly sequenced on an Illumina MiSeq to further evaluate library quality and pool balance. The final pool was sequenced on an Illumina NovaSeq 6000 targeting 40 million 150 bp read pairs per library at the Dana-Farber Cancer Institute Molecular Biology Core Facilities. For RNA-seq analysis sequenced reads were aligned to the UCSC hg38 reference genome assembly and gene counts were quantified using STAR (v2.7.3a). Differential gene-expression testing was performed by DESeq2 (v1.22.1) (9). RNA-seq analysis was performed using the VIPER snakemake pipeline (10). To permit analysis of the aggregated human PBMC-derived T cells, TCL cell lines, and patient-derived xenograft (PDX) data, we first performed quantile normalization to adjust for library depth and platform-specific differences. Batch effects within PDX and cell lines were successfully removed using the ComBat approach from SVA 3.18.00.

The methods for the isolation and processing of RNA and the acquisition of raw expression data have been described previously (11, 12). Briefly, raw data were uploaded into BRB-ArrayTools (version 3.7.0) and MAS5.0 and quantile normalized and batch corrected using ComBat (CITE; ref. 11). Data were generated on the HG-U133A and HG-U133-plus 2.0 arrays (Affymetrix Inc.). All data were downloaded from previously published data sets and are accessible through GEO under accession numbers GSE57520, GSE5788, GSE10631, GSE19069, GSE19067, and GSE58445 or EMBL's European Bioinformatics Institute under E-TABM-702 and E-TABM-783. Data were generated on the HG-U133A and HG-U133-plus 2.0 arrays (Affymetrix Inc.).

A tissue microarray (TMA) was constructed by us as previously described using 68 cases of T-cell and NK-cell neoplasms and 8 tonsil controls seen at Brigham and Women's Hospital (BWH) (7). Based on preliminary staining results additional individual cases of ENKTCL, T-PLL, and G/D TCL were stained for human KLRG1 for a total of 182 specimens. B-cell non-Hodgkin lymphoma specimens representing various subtypes were used as negative controls (n = 10). IHC staining was performed with the anti-human KLRG1 murine antibody (mAb008, Abcuro Inc) at a dilution of 1:100 using the Leica Biosystems Refine Detection Kit with EDTA antigen retrieval on the Leica Bond RX automated staining platform. Modified H-score to assess KLRG1 expression was generated per BWH protocol as follows: stained slides were scored by R.W., A.L.J., A.C., and J.B.Z. for percentage of malignant cells with positive staining over background (0 = 0%–9%, 1 = 10%–29%, 2 = 30%–59% and 3 = ≥60%) and the intensity of staining (0 = no staining, 1 = weak staining above background, 2 = moderate staining, 3 = strong staining). Multiplying the percentage of malignant cells with positive staining and the intensity of positive staining generated a modified H-score ranging from 0 to 300. Scores generated by individual pathologists were cross-verified by other participating pathologists to reach a consensus score.

These studies were performed in accordance with the Declaration of Helsinki. Blood samples were obtained from normal healthy donors, patients with T-LGLL, NK-LGLL, and other mature T-cell neoplasms including T-PLL as per institutionally approved protocols at Massachusetts General Hospital and at the University of Virginia School of Medicine and in accordance with an assurance filed with and approved by the U.S. Department of Health and Human Services, wherever appropriate. Informed written consent was obtained from each subject under the local IRB protocol. T-LGL leukemia patients (CD8⁺ TCRαβ and TCRγδ; no CD4⁺) meeting diagnosis criteria by the World Health Organization (evidence of clonality, persistence of chronic expansion >6 months) were recruited for this study. Similarly, for NK-LGLL, all patients met one or more of the following criteria for a diagnosis: >2×10⁹/L atypical lymphocytes; evidence of NK-cell invasion of the marrow; aberrant NK populations detected by flow cytometry. Peripheral blood mononuclear cells (PBMC) were isolated from these samples by Ficoll centrifugation. For T-LGLL, all patients had CD3⁺/CD8⁺ LGL fractions between 73%–96% in peripheral blood as assessed by clinical flow cytometry results in the initial screen. One patient had evidence of T-LGLL based on a huge increase in T-LGLs but also had an expanded NK-cell population (>500 and aberrant phenotype as demonstrated by a lack of CD56 and positivity for CD94). So, this patient was included in both the T-LGLL and NK-LGLL flow-cytometric cohorts to assess KLRG1 expression on T-LGL and NK-cell populations. For patients with T-LGLL and NK-LGLL, various subsets of leukemic cells as generally deemed to be pathogenic were analyzed for surface KLRG1 expression by flow cytometry. At least two human anti-KLRG1 primary conjugated antibodies: 13F12F2-PE and GA015-AF488 were utilized. For healthy donors, surface KLRG1 expression was analyzed on gated CD4⁺ and CD8⁺ T cells. PBMCs were also isolated from a patient with T-PLL and surface KLRG1 expression was assessed with GA015-AF488 by gating on human CD2 and CD52 double positive population. Splenocytes were harvested from a PDX model of T-PLL and hepatosplenic T-cell lymphoma (HSTCL) and analyzed for surface KLRG1 expression with both GA015-AF488 and 13F12F2-PE by gating on human CD2⁺ and CD45⁺ double positive cells.

Apoptosis was determined using a fluorescein isothiocyanate Annexin V Apoptosis Detection Kit (BD Pharmingen) according to the manufacturer's instructions in 500 μL aliquots of 10⁶ cells per 4 mL treated as indicated. Experiments were performed in triplicates and repeated thrice for cell lines as target cells (SMZ1-WT and SMZ1-KLRG1). To enrich for B, NK-cell, CD4⁺, and CD8⁺ T-cell subsets whole blood and PBMCs isolated from 3 healthy donors were subjected to a human NK-cell enrichment cocktail, CD4⁺, CD8⁺, and CD19⁺ magnetic bead isolation kit from Stem Cell Technologies as per the manufacturer's instructions. These populations were incubated with antibodies at various concentrations up to 100 μg/mL for 2, 6, and 24 hours. CD8⁺ T cells and DFTL-81777 PDX-derived HSTCL cells were incubated with DMSO or duvelisib at 1 μmol/L for 24, 48, and 72 hours.

Mouse macrophages were differentiated from the bone marrow of 8–10-week-old C57BL/6 mice. Unfractionated bone-marrow cells were cultured in conditioned media (RPMI supplemented with 10% L929 supernatant as a source of colony-stimulating factor, 10% FBS, and 100 U/mL penicillin and 100 μg/mL streptomycin) for 7–9 days to differentiate into murine bone-marrow–derived macrophage (mBMDM). Target cells (tumor or CD8⁺ T cells) were labeled with 0.25 μmol/L CFSE using the CellTrace CFSE Cell Proliferation Kit. Following serum starvation for 2 hours, mBMDMs or human monocyte-derived macrophages (hMDM) were cultured with labeled tumor cells (target to macrophage ratio of 3–4:1) in the presence of indicated antibodies at 10 μg/mL for 2 hours. Cocultures were washed, adherent macrophages scraped, and prepared for analysis by flow cytometry. Murine macrophages were identified by staining with fluorophore-conjugated antibodies to CD11b (clone M1/70, BioLegend). Human macrophages were identified by staining with fluorophore-conjugated antibodies to CD14 (clone M5E2, BD Biosciences). Gating was performed on macrophages to determine engulfed tumor cells and phagocytosis was evaluated as the percentage of CFSE⁺ (of CD11b⁺ for murine and of CD14⁺ for human) macrophages. Each experiment was done in triplicate and repeated thrice.

CD8⁺ T cells were isolated using Miltenyi microbeads from PBMCs derived from three healthy donors and stained with 0.25 μmol/L CFSE dye as mentioned above. SMZ1-KLRG1 cells were stained with 0.5 μmol/L CellTrace Violet (CTV) fluorescent dye (Invitrogen). Both cell types were opsonized with various antibodies at concentrations up to 10 μg/mL and incubated with human macrophage cells as described above. Gating was performed on macrophages to determine engulfed tumor cells and phagocytosis was evaluated as the percentage of CFSE⁺ or CTV⁺ (of CD14⁺) macrophages. PBMCs isolated from three independent patients with T-LGLL were subject to FACS sorting utilizing a cocktail of primary conjugated mAbs. Sorted target cells, CD8⁺/CD57⁺ and CD8⁺/CD57⁻ cells were stained with CFSE and CTV, respectively, and subject to phagocytosis with human macrophages from a healthy donor as above.

ADCC with murine FcyRIII expressing cells was determined using Promega's Murine FcyRIII ADCC Reporter Bioassay (CS1779B08) as per the manufacturer's instructions. Experiments were performed in triplicate with an effector–target ratio of 25:1. Antibodies were incubated at 10 μg/mL for 6 hours, and fold induction was calculated per vendor manual. For ADCC with human effectors, human PBMCs were isolated from three healthy donors and used as effector cells. The PBMCs were incubated overnight in IL2 containing media. On the day of the experiment, target cells (tumor cells) were labeled with 0.25 μmol/L CFSE dye, opsonized with antibodies at concentrations up to 10 μg/mL, and incubated with effector cells at a 1:25 ratio for 4 hours. Propidium iodide (PI) was added to identify dead cells, and flow cytometry was performed by gating on CFSE⁺ target cells to determine the percentage of PI-positive cells. To enrich for B, NK-cell, CD4⁺, and CD8⁺ T-cell subsets as target cells for ADCC assay, whole blood and PBMCs isolated from three healthy donors were subjected to human NK-cell enrichment cocktail, CD4, CD8, and CD19 magnetic bead isolation from Stem Cell Technologies as per manufacturer's instructions. Healthy target cells (B, NK, CD4⁺, and CD8⁺ cells) were CFSE labeled as above and incubated with autologous human PBMCs as effector cells in the presence of various antibodies at a concentration of 10 μg/mL at 1:25 ratio for 4 hours as above and percentage of PI-positive cells by gating on CFSE⁺ cells was determined. Additionally, PBMCs isolated from 3 other healthy donors were subject to FACS sorting utilizing a cocktail of primary conjugated mAbs. Various target cell populations (CD8⁺ T-cell subsets) were sorted as follows: central naïve (Tn, CCR7⁺/CD45RA⁺/ZA⁻), central memory (Tcm, CCR7⁺/CD45RA⁻/ZA), effector memory cells (Tem, CCR7⁻/CD45RA⁻/ZA⁻), terminally differentiated effector memory cells reexpressing CD45RA (TemRA, CCR7⁻/CD45RA⁺/ZA⁻). Target cells were CFSE labeled as above and incubated with autologous sorted effector cells (CD4⁻/CD8⁻/CD57⁻/CD56⁻) in the presence of various antibodies as above, and the percentage of PI-positive cells by gating on CFSE⁺ cells was determined.

PBMCs isolated from four patients with T-LGLL were subject to FACS sorting utilizing a cocktail of primary conjugated mAbs. Sorted target cells (CD8⁺/CD57⁺/ZA⁻) were CFSE labeled as above and incubated with autologous sorted effector cells (CD4⁻/CD8⁻/CD57⁻/CD56⁻) in the presence of various antibodies at a concentration of 10 μg/mL at 1:25 ratio for 4 hours as above and percentage of PI-positive cells by gating on CFSE⁺ cells was determined.

Treatment-naïve patients with T-PLL (n = 1), AITL (n = 1), and PTCL-NOS (n = 1) were identified. Patients with T-PLL had >90% circulating CD2⁺/CD52⁺tumor cells (of the total lymphocyte count) with surface KLRG1 expression in 40% of the CD2⁺/CD52. The AITL patient had >60% circulating CD4⁺ population (of the total lymphocyte count) with 24% of the population positive for surface KLRG1. We made a single-cell suspension from an excisional lymph node biopsy derived from a patient with PTCL-NOS with 90% tumor cells based on the patient's clinical IHC evaluation. Up to 30% of the tumor cells were positive for surface KLRG1 expression in this sample. As above, these specimens were CFSE labeled and incubated with allogeneic effector cells in the presence of various antibodies at a concentration of 10 μg/mL at a 1:25 ratio for 4 hours as above and percentage of PI-positive cells by gating on CFSE⁺ cells was determined.

CD8⁺ T cells were isolated using Miltenyi microbeads from PBMCs derived from three healthy donors and stained with 0.25 μmol/L CFSE dye, as mentioned above. SMZ1-KLRG1 cells were stained with 0.5 μmol/L CTV. Both cell types were opsonized with various antibodies at concentrations up to 10 μg/mL and incubated with effector cells at a 1:25 ratio for 4 hours as above. PI was added and flow cytometry was performed by gating on CFSE⁺ or CTV⁺ target cells to determine the percentage of PI-positive cells as mentioned above. CD8⁺/CD57⁺ and CD8⁺/CD57⁻ underwent FACS from T-LGLL patient-derived PBMCs. They were stained with CFSE and CTV, respectively, and incubated with autologous sorted effector cells (CD4⁻/CD8⁻/CD57⁻/CD56⁻) in the presence of various antibodies at a concentration of 10 μg/mL at 1:25 ratio for 4 hours as above and percentage of PI-positive cells by gating on CFSE⁺ or CTV⁺ cells was determined as above.

CDC with murine and human complements was determined as previously described (http://cshprotocols.cshlp.org/content/2018/2/pdb.prot093799; ref. 7). Experiments were performed in triplicates. Antibodies were incubated at 10 μg/mL for 4 hours.

Quantum Simply Cellular microspheres were used in the quantitative analysis of cellular KLRG1 expression. Cells with varying degrees of KLRG1 expression were stained with 13F12F2-PE–conjugated primary antibody or isotype control as per the manufacturer's instructions to permit the determination of the antibody binding capacity (ABC) of the cells. Geo mean for the populations of interest was recorded and inserted into the Lot-specific QuickCal analysis template to generate the ABC which serves as a surrogate for a number of surface receptors.

HSTCL PDX-derived TCL cells were incubated with increasing concentrations (0.001 μg/mL to 100 μg/mL) of either isotype control or unconjugated human anti-KLRG1 mAbs (GA015 or 13F12F2) for 2 hours. Cells exposed to 13F12F2 were fixed, washed, and then stained with GA015-AF488 conjugated primary mAb or isotype control at the concentration of 10 μg/mL, whereas cells exposed to GA015 were stained with 13F12F2-PE mAb or isotype control at the concentration of 10 μg/mL. hCD2⁺/hCD45⁺ HSTCL cells were gated to determine the expression of hKLRG1 by flow cytometry.

CD8⁺ T cells isolated from a healthy donor were incubated with increasing concentrations (0.001 μg/mL to 100 μg/mL) of either isotype control or unconjugated human anti-KLRG1 mAb, mAb208 for 2 hours. Cells were washed, incubated with GA015 (which has overlapping but distinct epitopes along with a distinct human IgG1 Fc from mAb208), and then stained with anti-human Fc secondary mAb or isotype control at the concentration of 10 μg/mL. Human CD8⁺ T target cells were gated to determine the expression of KLRG1 by flow cytometry.

The thymic single-cell data used in this study have previously been published (13–16) and the analysis has been described in detail (16). In brief, fastq files were mapped to the human reference genome GRCh38 using CellRanger version 6.0.1 and the obtained h5 files were analyzed in R using the Seurat package (17). Cells with over 10% mitochondrial reads, fewer than 700 reads or expressing fewer than 250 genes were considered to be of low quality and removed from the data set. Doublets were identified with the scDblFinder package (18). In addition, cells with a gene count over 2,500–6,000 genes depending on the sequencing depth of the respective library were also removed as putative homotypic doublets. Finally, genes expressed in fewer than 10 cells across the full data set were removed. Gene expression was log-normalized, and the 2,000 most variable genes (HVG) were identified. Gene expression was scaled and cell-cycle scoring on the G₂–M and S phase marker genes provided in the Seurat package was performed. Differences in sequencing depth and the G2M-S score difference were regressed out. PCA was performed on the scaled HVGs, and batch correction was applied to the PCA matrix via the reduced MNN function from the Batchelor package (19). Louvain clustering with a resolution of 0.1–0.8 was performed and UMAP was used to visualize the results. Known marker genes for distinct stages of thymocyte development were used to annotate the clusters. Clusters consisting of non-T lineage cells were removed from the data set prior to downstream analyses. Markov affinity-based graph imputation of cells (MAGIC; ref. 20) was used to denoise the data and impute dropout values for improved visualization.

The preprocessed Seurat objects were downloaded from https://zenodo.org/record/4739231#.ZFOY0OxByAk, and the processing information can be seen in the original publication, which briefly includes individual quality control per sample, batch correction with the variational autoencoder tool scVI (ver 0.5.0) and clustering with optimal parameters defined for individual objects based on visual inspection (21). The downloaded Seurat objects were used to quantify the expression of KLRG1 in preidentified clusters, and the differential expression analysis was done with the Bonferroni corrected t test. In the scRNA-seq data, differential expressed genes were identified with Bonferroni corrected two-sided t test. Enriched pathways were identified from significantly differentially expressed genes (P_adj < 0.05) with Benjamini–Hochberg corrected one-sided hypergeometric testing implemented in ClusterProfiler (3.16.0) with KEGG, REACTOME, and HALLMARK categories gathered from MSigDB, where a significance threshold of P_adj  < 0.05 was used. Differentially expressed regulons were identified with the SCENIC workflow (ver 1.2.4), where the vignette was followed with the default parameters.

hMDM from healthy donor differentiated as above were treated with DMSO or duvelisib at 1 μmol/L. At 24, 48, and 72 hours, macrophages were harvested, washed, stained with a panel of surface antibodies for flow cytometry and analyzed. Antibodies included CD45-FITC, CD14-BV510^TM, CD86-PE-Cy7, HLA-ABC-AF^TM700, HLA-DR-BV605^TM, CD16-APC-Cy7, SIGLEC10-APC, CD276-PerCP, PD-L1-BV711^TM, LRP1-PE, TLR4-BV421^TM, and CD64-BV785^TM.

In vivo studies were performed as previously described (7). For the SMZ1-KLRG1 xenograft experiment, 1 million SMZ1-WT or SMZ1-KLRG1 cells were injected subcutaneously into the flank of 5–7-week-old NSG (Jackson laboratories) mice. For the SMZ1-KLRG1 xenograft, when the tumor volume reached 50–250 mm³, mice were randomized into one of the following four treatment arms: murine IgG2a isotype control mAb, mAb208 (10 mg/kg intraperitoneally (i.p.) every 48 hours for both antibodies), duvelisib (50 mg/kg by oral gavage b.i.d. daily), or the combination (administered at single-agent doses). Tumor volumes were monitored by calipers and treatment was continued till tumor volumes reached 2,000 mm³, at which point the mice were euthanized as per IACUC guidelines.

HSTCL or T-PLL PDX cells (DFTL-81777 or DFTL-28776, respectively) were injected into 5–7-week-old NOD.Cg-B2m^tm1Unc Prkdc^scid Il2rg^tm1Wjl/SzJ mice (Jackson laboratories) with 1–1.5 ×10⁶ cells per mouse via tail-vein injection. Upon engraftment with ≥5%–10% hCD2⁺/hCD45⁺ cells as determined by flow cytometry in the liver or spleen (hCD2 clone-RPA-2.10, APC, and hCD45 clone-HI30, FITC; both from BioLegend), mice were randomized, and treatment was initiated with either mAb208 (10 mg/kg i.p. every 48 hours diluted in PBS) for 6 or 12 doses for DFTL-81777 or 9 doses for DFTL-28776, murine IgG2a isotype control mAb, duvelisib (50 mg/kg by oral gavage b.i.d. daily for 11 days or 4 weeks for DFTL-81777 or 3 weeks for DFTL-28776), anti-CD47mAb, MIAP410 (10 mg/kg i.p. every 48 hours diluted in PBS) for 6 doses or the combination of duvelisib with anti-CD47mAb or duvelelib with mAb208 (administered at single-agent doses). Three to six mice per arm were euthanized 24 hours after the 6th or 12th dose for DFTL-81777 and after the 9th dose for DFTL-28776, and the splenocytes and the bone-marrow cells flushed from the femur from each mouse were harvested. Flow cytometry was performed on live cells to determine tumor burden (human CD45 and CD2 double-positive cells), and macrophage polarization was determined using a cocktail of primary conjugated antibodies that include murine F4/80-APC, Ly6G-PE, CD11b-PE-Cy7, CD206-BV 421, CD45-APC-Cy7, PD-L1-BV605, CD86-PerCP, FcGRIV-BV785, ZA, human CD45-FITC (all from BioLegend), and iNOS-AF700 (Invitrogen). Mouse plasma was also analyzed for murine and human cytokines by Eve Technologies. Tumor cells were harvested from the DFTL-81777 model for ex vivo flow cytometry assays including determination of human KLRG1 expression. In another independent in vivo study, mice engrafted with DFTL-81777 cells as above were treated continuously on one of the four arms as above till they were moribund with disease for survival analyses.

For the macrophage depletion xenograft experiment, for the PDX model DFTL-81777, after confirmation of engraftment and randomization of mice, treatment was initiated with either murine IgG2a isotype control, mAb208, duvelisib, or combination at the above doses. All mice received PBS containing liposomes (100 μL i.v on days 1, 3, and 5 of the above treatments. Four mice per arm were euthanized 24 hours after the last dose of clodronate. Cells were harvested from liver, spleen, and bone marrow and analyzed for tumor burden as above.

All animal work was performed according to MGH IACUC protocol # 2022N000132.

Continuous variables were summarized by the median and interquartile range and comparisons between groups were performed using the Wilcoxon rank-sum test and Kruskal–Wallis test. Categorical and ordinal variables were summarized by frequencies and percentages and comparisons between groups were performed using Pearson Chi-squared test and Fisher exact test. A continuous time-to-event approach was used (from the time of diagnosis to the time of death or time of progression), and a comparison between overall and PFS curves was performed using a log-rank test. Analyses were performed using R version 4.2.2 or greater (The R Foundation for Statistical Computing-http://www.r-project.org/) and a P < 0.05 determined statistical significance. Overall survival for PDX experiments was calculated from the time of treatment initiation to the time of death and compared using a log-rank test.

All primary data from RNA-seq as previously reported have been deposited in NCBI's Gene-Expression Omnibus and are accessible through the GEO series accession number GSE114085 (7). Further information and access to raw NGS files are freely available from the authors upon request.

Previous studies based on gene expression and scRNA-seq data set publications have reported higher KLRG1 expression at the transcript level in the Tem and TemRA cells (22). Bulk RNA sequencing studies of human TCL cell lines, PDX cells, and primary T cells demonstrated heterogeneity in transcript levels of human KLRG1 with higher expression in PDX-derived HSTCL (DFTL-81777) and ATLL (DFTL-69579) cells relative to other PDX's, G/D (Karpas 384) and NKL (KHYG-1) cell lines relative to other PTCL cell lines, and in CD8⁺Tn, Tfh and Teff cells derived from healthy donors (Fig. 1A). We analyzed gene-expression data sets generated from patients with PTCL (11, 12). We noted a higher expression of KLRG1 in subsets of patients with HSTCL, ENKTCL, NKT-GD (NK/TCL-gamma delta), PTCL-NOS, angioimmunoblastic TCL (AITL), T-prolymphocytic leukemia (T-PLL), stimulated CD8⁺ T cells relative to stimulated CD4⁺ T cells and the anaplastic large cell lymphoma (ALCL) subtypes (Fig. 1B).

Studies of mRNA expression may not necessarily reflect surface protein. Thus, we examined KLRG1 protein levels with multiple anti-KLRG1 antibodies (13F12F2 and GA015 are clones) on the surface of human TCL lines including H9, SeAx, MyLa, HuT-102, SR-786, HuT-78, MJ, HH, L-82, KI-JK, SUP-M2, SU-DHL1, DERL-2, K-299, Karpas 384, DL-40, KHYG-1, FEPD, MAC2A, OCI-Ly13.2, OCI-Ly12, SMZ1, NKL, MTA, MOTN-1, and DERL-7 in comparison with CD4⁺ and CD8⁺ T cells from healthy donors. None of the above lines representing several PTCL subtypes expressed surface KLRG1. In contrast, greater expression of KLRG1 protein was detected on various subsets of CD8⁺ T and NK cells compared with CD4⁺ T cells utilizing different antibodies as previously reported (Fig. 1C; ref. 23). To define concordance between transcript and protein levels, we performed whole-transcriptome sequencing (bulk RNA-seq) and flow cytometry on the same healthy donor-derived cellular subsets (Supplementary Fig. S1A and S1B). Overall, specimens with higher human KLRG1 transcript levels also exhibited wider expression of surface KLRG1. To determine the expression of the human KLRG1 gene across stages of human T-cell development, we analyzed publicly available scRNA-seq data sets (13–20). We observed a higher expression of KLRG1 in G/D T cells and Tregs (Supplementary Fig. S1C and S1D). We hypothesized that measurement of KLRG1 surface density would enable us to better understand differential responses to targeted antibodies. The surface density was highest on CHO and SMZ1 cells overexpressing KLRG1 (CHO-KLRG1 used as positive control) and negligible on wild-type CHO and SMZ1 cells (CHO-WT, SMZ1-WT) and mantle cell lymphoma (MCL, DFBL-39435) PDX cells (negative control; Fig. 1D). KLRG1 surface density on PDX-derived HSTCL cells was higher in comparison with bulk and specific CD8⁺ T cells and its subsets and PDX-derived T-PLL cells, highlighting heterogeneity.

To assess lymphoma cell–specific KLRG1 protein expression in primary samples, we first evaluated a 68-patient TMA, extensively annotated, and previously characterized by us, that includes multiple TCL subtypes by IHC staining (7). Multiple anti-human KLRG1 antibodies including MAB70293 and mAb008 were investigated for specific staining. mAb008 exhibited marked specificity for human KLRG1 protein based on staining of CHO-WT, CHO-KLRG1, SMZ1-WT, and SMZ1-KLRG1 cells and hence was selected for staining of our TMA (Fig. 1E). We generated an H-score (defined as % of lymphoma cells that are KLRG1 positive |$ \times $| intensity of KLRG1 staining from 0 to 3). We utilized the maximum H-score among the multiple cores from each biopsy included in the TMA. H-score for KLRG1 expression was highly heterogeneous both within and across the common TCL subtypes (Fig. 1F). The maximum percentage of lymphoma cells positive for KLRG1 membrane staining independent of staining intensity was also highly heterogeneous within and across the common subtypes (Fig. 1G; Supplementary Fig. S2A). Based on higher lymphoma-specific H scores observed in a subset of patients with ENKTCL, T-PLL, and HSTCL, we enriched and stained additional cases of these rare subtypes including G/D TCL with mAb008 and confirmed similar expression patterns. Additional pictures of staining are included in Supplementary Fig. S2B to demonstrate heterogeneity. To further validate our IHC results, we stained 10 cases of various B-cell NHL such as CLL/SLL, marginal zone lymphoma, MCL, and diffuse large B-cell lymphoma (DLBCL) of the GCB and ABC types. Lymphoma cells in these specimens did not exhibit any KLRG1 (Supplementary Figs. S2C and S2D), highlighting the specificity of the antibody. Various intensities of KLRG1 staining from absent to strong are depicted in Supplementary Fig. S2E. We further confirmed the lymphoma cell–specific expression with both IHC and flow (by gating on hCD2⁺/hCD52⁺ TCL cells) isolated from a treatment-naïve patient with T-PLL, the T-PLL PDX model (DFTL-28776) and by gating on hCD2⁺/hCD45⁺ TCL cells isolated from DFTL-81777 (Supplementary Fig. S2F and S2G). A PDX model of DLBCL (DFBL-20954) was used as a negative control. We attempted to immunophenotypically define KLRG1⁺ versus KLRG1⁻ neoplasms within the subtypes with greater KLRG1⁺ tumor cells, using clinical IHC and flow data (when available) but could not identify any specific pattern differentiating these cases (Supplementary Tables 1–5). Clinical characteristics for the 68 cases represented in the TMA have been previously described (7). Available characteristics for the remaining cases are described in Supplementary Tables S6–S8.

We attempted to divide patients represented within and outside the TMA into quartiles based on a % of KLRG1⁺ cells and compared their progression-PFS and overall survival (OS) from the time of their diagnosis. There were no significant meaningful differences in PFS or OS across quartiles based on a % of KLRG1⁺ lymphoma cells (Supplementary Figs. S3A–S4E). We applied a recursive partitioning approach but were unable to identify an optimal split based on the KLRG1 H-score or percentage of KLRG1⁺ lymphoma cells that resulted in a consistent and significant difference in OS and PFS between groups. There were no significant associations detected between the percentage of KLRG1⁺ lymphoma cells (divided into two cohorts based on above or below the median and mean) and treatment (Supplementary Table S9). Thus, although KLRG1 surface protein expression appears heterogeneous across subtypes of T-cell and NK-cell neoplasms, there are subsets that show a greater percentage of KLRG1⁺ tumor cells than others.

A population of KLRG1⁺ Tem and TemRA are expanded in both the CD4⁺ and CD8⁺ T-cell subpopulations in patients with the autoimmune muscle disease inclusion body myositis (IBM; ref. 23). IBM may be associated with T-LGLL, and 58% of patients with IBM have blood CD8⁺CD57⁺ T-cell expansions similar to patients with T-LGLL (23). This led us to determine surface KLRG1 expression in patients with T-LGLL and NK-LGLL, subtypes of mature T-cell and NK-cell neoplasms that have leukemic presentations. We systematically investigated several CD3⁺ and CD3⁻/CD56⁺ subsets in PBMCs derived from patients with T-LGLL and NK-LGLL, respectively. We observed expression of surface KLRG1 in the majority of the various CD3⁺ T-cell subsets in patients with T-LGLL, including CD3⁺/CD8⁺ and on CD3⁺/CD8⁺/CD57⁺ cells which are widely accepted as the pathogenic cells in T-LGLL. This was observed with anti-KLRG1 antibodies GA015 and 13F12F2 (Fig. 2A; Supplementary Fig. S5A and S5B).

Similarly, widespread expression of KLRG1 was observed in CD3⁻/CD56⁺ and its subsets, such as CD3⁻/CD56⁺/CD16⁺, CD3⁻/CD56⁺/CD16⁺/CD57⁺, and CD3⁻/CD56⁺/CD16⁺/CD57⁺/CD94⁺ cells in patients with NK-LGLL (Fig. 2B; Supplementary Fig. S6A and S6B). Staining was heterogeneous within the patients with both neoplasms but comparable in intensity across the two anti-KLRG1 mAbs (Supplementary Fig. S7A and S7B). Molecular and clinical details for the majority of these patients have been previously described (21, 24, 25). Molecular and clinical characteristics available for 10 out of the 12 with T-LGLL and NK-LGLL specifically included in our study are summarized in Supplementary Table S10. To define KLRG1 expression at the single-cell level in T-LGLL, which demonstrated ubiquitous surface expression, we utilized a recently published scRNA + TCRab-seq data set derived from sorted CD45⁺ cells from CD8⁺ T-LGLL (n = 9, 11 samples) and healthy (n = 6) samples (21). Within the CD45⁺ cells, the highest expression of KLRG1 was found in the clusters with CD8⁺ TemRA phenotype, with also some expression in NK cells (Supplementary Fig. S8A and S8B). Expression of KLRG1 was higher in CD8⁺ TemRA from patients with T-LGLL than in healthy (Fig. 2C; P < 0.0001, Bonferroni corrected t test). However, to answer whether KLRG1 is expressed similarly in reactive cytotoxic clones within the CD8⁺ T-LGLL clones, we focused on “hyperexpanded” T-cell clones, i.e., clones with at least 10 cells sharing the same T-cell receptor (TCR) from patients with T-LGLL and healthy individuals (Supplementary Fig. S8C). In this analysis, we found that KLRG1 is more highly expressed in CD8⁺ T-LGLL cells than in large oligoclonal expansions from healthy controls (Supplementary Fig. S8D). When focusing on CD8⁺ T-LGLL cells, we found that 6/9 T-LGLL patients expressed at least 10% KLRG1⁺ cells in their clusters (Fig. 2D). Intrapatient and interpatient clonotypic heterogeneity in KLRG1 expression was observed. At a phenotype level, 3/7 of the preidentified clusters expressed substantial KLRG1, including the STAT3 mutated and STAT3 wild-type clusters. No obvious differences with respect to KLRG1 expression were observed between the mutated versus wild-type clusters (Fig. 2E and F). Next, we sought to investigate whether anti-KLRG1 antibodies could eliminate leukemic cells from the peripheral blood of patients with T-LGLL. Incubation of CFSE-labeled CD8⁺/CD57⁺ live leukemic cells isolated from 4 patients with T-LGLL with autologous KLRG1⁻ sorted effector cells (CD4⁻/CD8⁻/CD57⁻/CD56⁻) resulted in cytotoxicity of these cells in the presence of anti-KLRG1 mAb208 (another clone) but not 13F12F2 (Fig. 2G). Thus, KLRG1 is a potential actionable target for patients with T-LGLL and NK-LGLL. Further, we performed pathway analyses using KEGG, REACTOME, and HALLMARK pathways between clusters KLRG1⁺ (0,2,3) and KLRG1⁻ (1,4, 6) clusters. The most upregulated pathways in cluster 0 included those involved in splicing, TCR signaling, and interferon signaling whereas in cluster 2 these included oxidative phosphorylation and IL12 and in cluster 3 TNFa, TCR and IFN signaling. Thus, despite heterogeneity, common pathways underlie clusters with high KLRG1 expression. When comparing pooled analysis of KLRG1⁺ versus KLRG1⁻ clusters unregulated pathways included TCR signaling, interferon signaling, and splicing, suggesting an active immunologic signaling present in these clusters. Downregulated pathways included those related to oxidative ribosomal RNA processing and oxidative phosphorylation suggesting a lack of common pathways underlying KLRG1⁻ clusters. Additionally, we also performed transcription factor usage analyses and observed that KLRG1⁺ clusters had homogenous transcription factor usage including dependency on Fos, Jun, IRF1, and JunB on the transcription factor regulatory network analysis SCENIC (Supplementary Fig. S8F).

We hypothesized that mAbs that deplete surface KLRG1 could be deployed for the eradication of KLRG1⁺ lymphoma cells. Because none of the above cell lines expressed surface KLRG1, we generated stably transduced PTCL-NOS cell line, SMZ1, and T-ALL line Jurkat to express full-length protein. First, we investigated if the KLRG1-depleting antibody mAb208 eradicated cells by inducing apoptosis of KLRG1⁺ cells. Exposure of SMZ1-KLRG1 cells to mAb208 at concentrations as high as 100 μg/mL up to 24 hours did not induce apoptosis in comparison with staurosporine (positive control; Fig. 3A). No apoptosis of CD4⁺, CD8⁺ T cells, NK, or B cells was seen either despite surface expression of KLRG1 ranging from 1%–19.7% for NK cells, 8%–10.6% for CD4⁺, and 25.1%–36.6% for CD8⁺ T cells (Supplementary Fig. S9A).

Incubation of SMZ1-KLRG1 cells, but not SMZ1-WT, with multiple anti-KLRG1 mAbs including mAb208 and 13F12F2 in the presence of both human monocyte-derived macrophages (hMDM) and murine bone-marrow– derived macrophages (mBMDM) increased phagocytosis (Fig. 3B and C). The fold change in phagocytosis induced by mAb208 and 13F12F2 was at least as robust as the effect of the anti-CD47 mAb on Jurkat cells, a known phagocytic checkpoint (positive control).

To assess ADCC, we adopted two different approaches. In the first assay, we directly tested ADCC of SMZ1-KLRG1 cells by human PBMCs while using Raji cells incubated with rituximab (positive control). Neither mAb208 nor 13F12F2 induced ADCC of SMZ1-WT cells (Fig. 3D). mAb208 induced potent ADCC of SMZ1-KLRG1 cells in a dose-dependent fashion as opposed to 13F12F2, which had negligible effect (Fig. 3E). These findings underscore the selective activity of mAb208 and the dose-dependent activity of anti-KLRG1 mAbs in effecting ADCC. Similar results for ADCP and ADCC were observed with Jurkat-KLRG1 cells relative to Jurkat-WT (KLRG1⁻) cells highlighting that the effects are not specific for a particular cell line (Supplementary Figs. S9B and S9C). These results were further confirmed in three additional primary specimens derived from patients with T-PLL, AITL, and PTCL-NOS (Supplementary Fig. S9D).

Second, we utilized an established assay involving Jurkat cells that express murine FcγRIII, the primary FcR on murine NK cells involved in ADCC. Neither anti-KLRG1 mAbs were able to induce a fold change in reporter activity of the effector cells when incubated with SMZ1-WT cells in comparison with the effect of the anti-CD20 mAb on Raji cells (positive control; Fig. 3F). In contrast, both mAb208 and 13F12F2 induced ADCC of SMZ1-KLRG1 cells in a dose-dependent effect (Fig. 3G) with mAb208, demonstrating greater potency when compared with 13F12F2.

No induction of CDC of SMZ1-WT cells with either murine or human complements upon exposure to mAb208 (Fig. 3H) or 13F12F2 (Fig. 3J) was observed. Although incubation with human complements did not affect the cytotoxicity of SMZ1-KLRG1 cells (Fig. 3I), both antibodies induced cytotoxicity with murine complements in a dose-dependent manner (Fig. 3K) potentially due to isotype differences and their varied abilities to activate complements. Thus, antibodies such as mAb208 with the right isotype backbone and functional Fc can deplete KLRG1⁺ cells through a plethora of mechanisms including ADCC, ADCP, and CDC.

Next, we wanted to assess if the anti-KLRG1 mAbs capable of clearing tumor cells have different binding epitopes. We performed cross-blocking assays to discern this. After exposure to increasing concentrations of functional grade 13F12F2 to saturate the target, PDX-derived HSTCL cells with higher target density still demonstrated binding to GA015 (Fig. 4A). Reciprocal binding to 13F12F2 was observed despite incubation with saturating concentrations of GA015, thereby confirming differential binding sites (Fig. 4B).

We also wanted to determine if the above dose-dependent functional effects of mAb208 are due to increasing KLRG1 receptor saturation. Thus, after incubating the healthy donor-derived CD8⁺ T cells to their increasing concentrations, we further added GA015 to bind any unbound epitopes. We observed that concentrations as high as 10 μg/mL of mAb208 were needed for maximal binding to the KLRG1 receptors on healthy CD8⁺ T cells (Fig. 4C).

Next, we wanted to assess if anti-KLRG1 mAbs would eradicate KLRG1 expressing CD8⁺, CD4⁺, and NK cells relative to KLRG1⁻ B cells. We first confirmed surface KLRG1 expression on 10.1%, 17.5%, and 24.6% of CD8⁺ T cells, 4%, 11.3%, and 20.1% on CD4⁺ and 12.2%, 15.9%, and 21.6% of NK cells from three healthy donors as previously demonstrated in Fig. 1C. Exposure of flow cytometry-sorted CD8⁺ T cells from the healthy donors to autologous human PBMCs effectors without NK cells demonstrated dose-dependent ADCC mediated by mAb208 in contrast with 13F12F2, which lacks an afucosylated Fc region (Fig. 4D). A similar increase in ADCC of CD4⁺ T cells and NK cells was seen with mAb208 without any appreciable increase in ADCC of KLRG1⁻ B cells (Fig. 4E and F). No remarkable increase in phagocytosis of CD8⁺ T cells by autologous hMDM was observed upon incubation with anti-KLRG1 mAb (Fig. 4G). Thus, KLRG1-depleting antibodies specifically eliminate KLRG1-expressing T and NK cells.

KLRG1 expression in CD8⁺ T cells increases with antigen-driven T-cell differentiation with the lowest levels of expression in naïve T cells (Tn) and highest in highly differentiated TemRA and CD56⁺/CD8⁺ T cells (4). Accordingly, we assessed if the depleting antibodies were highly specific in eradicating CD8⁺ T-cell subsets with greater expression of KLRG1. Incubation of various CD8⁺ T-cell subsets including Tn, Tcm (central memory), Tem, and TemRA from 3 healthy donors with autologous KLRG1⁻ effector cells (CD4⁻/CD8⁻/CD57⁻/CD56⁻) in the presence of anti-KLRG1 mAbs 13F12F2 and mAb208 demonstrated cytotoxicity of Tem and TemRA subsets that exhibit greater KLRG1 expression than Tn, Tcm, B, and NK cells which have minimal to no expression (Fig. 5A). Further, mAb208 was more potent in exerting these cytotoxic effects relative to 13F12F2. These data highlight the potential of anti-KLRG1 mAb208 in inducing selective depletion of KLRG1⁺ CD8⁺ T cells. Next, we wanted to assess if anti-KLRG1 mAbs would preferentially eradicate KLRG1⁺ expressing TCL cells over healthy CD8⁺ T cells. Simultaneous exposure of SMZ1-KLRG1 and healthy CD8⁺ T cells to autologous effector cells in the presence of mAb208 demonstrated greater cytotoxicity of tumor cells relative to its isotype but similar levels of cytotoxicity of CD8⁺ T cells in comparison with its control (Fig. 5B). However, higher phagocytosis of tumor cells over CD8⁺ T cells was observed upon incubation with mAb208 (Fig. 5C). Similar results observed upon coincubation of CD8⁺/CD57⁺ leukemic cells and CD8⁺/CD57⁻ nonleukemic cells in T-LGLL patients (Fig. 5D and E). Coincubation of healthy donor-derived CD8⁺ TemRA cells and T-PLL cells from patients and PDX with comparable KLRG1 surface expression (20.5%–42.6%) with human PBMC-derived effector cells demonstrated higher cytotoxicity of the tumor cells with mAb208 relative to CD8⁺ T cells and in comparison, with their respective isotypes (Fig. 5F). Next, we hypothesized that duvelisib, which induces apoptosis of TCL cells and repolarizes macrophages to a proinflammatory phenotype, might be combined with anti-KLRG1 mAbs to enhance clearance of tumor cells (6). Indeed, both CD8⁺ T cells from a healthy donor and PDX HSTCL cells demonstrated increasing apoptosis upon incubation with duvelisib compared with control (Fig. 5G) resulting in increased ADCP with both anti-KLRG1 mAbs 13F12F2 and mAb208 at multiple timepoints (Fig. 5H and I). Thus, KLRG1-depleting antibodies are particularly able to remove KLRG1⁺ T cells and tumor cells.

Our group has previously demonstrated that the PI3K-δ/γ inhibitor duvelisib, which is recommended for the treatment of patients with T-cell lymphomas, induces M1 (MHCII⁺CD206⁻) phenotype in T-cell lymphoma PDXs (6). To gain further mechanistic insight into the transcriptional programs and other proteomic modifications induced by duvelisib, we performed whole-transcriptome sequencing of hMDM exposed to DMSO versus duvelisib at several time points. Overall, 1,697 differentially expressed genes were identified between the two groups using a log2Fold change of ≥ 1.5 and ≤ −1.5 (Supplementary Fig. S10). Consistent with previous publications (26), we observed that duvelisib was associated with an immune-stimulatory transcriptional program in MDM. As an example, duvelisib induced the expression of genes encoding costimulatory molecules (CD86), antigen presentation machinery (CIITA, HLA-DM1, HLA-DPB1, and HL1-DRB5), and chemokines (CXCL2, CXCL3, CXCL8, CXCL12, CCL20, and CCL19; Supplementary Fig. S11A). Conversely, expression of immune-suppressive cytokines (IL10), immune checkpoints (CD274, CD276), signaling pathways including RIG-1, TGFβ, NF-kB, and histone acetylation (LINC02085, LINC01451, TBP1, GPRC5B, and RCOR2) were reduced in duvelisib-incubated hMDM. Of note, increase in expression of critical phagocytic factors, such as prophagocytic receptors (several TLRs, LRP1, CD14, FCGR2A, FCGR3A, MFGE8, GAS6), as well as genes involved with cargo degradation in phagolysosomes (CTSS and LYZ), inflammation (NLRP6, CRISPLD2, BTNL8), and generation of reactive oxygen species (NOS2). In concert with these transcriptional changes, changes in surface protein expression of several key effectors and cytokines were observed in duvelisib-exposed hMDM compared with DMSO (Supplementary Fig. S11B and S11C). Thus, duvelisib is capable of reprogramming the transcriptomic and proteomic machinery in macrophages.

The ability to eliminate KLRG1⁺ TCL cells in vitro led us to test the efficacy of anti-KLRG1 mAb208 in vivo. Our group has previously demonstrated that duvelisib is effective in reducing tumor burdens in TCL PDXs and is a recommended treatment for patients with PTCL (6). Thus, we hypothesized that the addition of anti-KLRG1 mAb to duvelisib in vivo would further diminish tumor burden (Fig. 6A). In the first in vivo study of SMZ1-KLRG1 subcutaneous xenografts, treatment with combination therapy resulted in significant reductions in tumor volumes and markedly improved survival over either monotherapy (Fig. 6B and C). Conversely, treatment of KLRG1⁻ SMZ1-WT xenografts did not demonstrate any tumor reduction with mAb208 (Supplementary Fig. S12A).

The notable responses above led us to test mAb208 in highly disseminated PDX models of TCL that bear high fidelity to human disease that our group has previously characterized, and which have a diverse expression of surface KLRG1 (7). We utilized the HSTCL and T-PLL PDXs with varying surface expression (Supplementary Fig. S2F and S2G). Treatment with mAb208 alone resulted in a striking reduction in lymphoma involvement in the spleen of HSTCL PDXs and prolonging their survival (Fig. 6D and E). Combination with duvelisib further enhanced this effect. The combination was more effective than single agents in reducing the lymphoma burden in the liver and BM compartments of the T-PLL xenografts (Fig. 6F and G). To address the individual contribution of tumor-infiltrating macrophages to mAb208 efficacy in vivo, we depleted macrophages in vivo with clodronate. Depletion of murine macrophages in NSG mice engrafted with the HSTCL PDX markedly reduced the activity of mAb208 across all compartments (Supplementary Fig. S12B). To assess if the enhanced antitumor efficacy with the combination is an antibody class effect, we performed an in vivo experiment, analyzing the comparative efficacy of combinatorial regimens such as anti-CD47 mAb with duvelisib. Indeed, greater reductions in tumor burden across multiple compartments were seen with both the antibody combinations with duvelisib (Supplementary Fig. S12C). A decrease in the percentage of CD11b⁺/F4/80⁺CD206⁺ and an increase in CD11b⁺/F4/80⁺ MHC-II⁺ were observed with the combination (Supplementary Fig. S12D). No increase in cleaved-PARP or caspase-3 was seen with combination therapy in the HSTCL model. No significant difference in murine and human cytokines was noted in plasma with treatment. Of interest, the persistent lymphoma cells harvested from the HSTCL PDX treated with mAb208 demonstrated surface KLRG1 on a subset suggesting a few potential mechanisms of resistance to KLRG1 depleting antibodies including antigen escape and ineffective binding of the antibody among others (Supplementary Fig. S13). Thus, clinical therapeutic antibodies such as anti-KLRG1 in combination with PI3K inhibitors present an attractive combination strategy for patients with T-cell neoplasms (Supplementary Fig. S14).

Despite the heterogeneous nature of mature PTCLs, most patients with advanced disease have a poor outcome relative to their B-cell counterparts. As expected, the surface KLRG1 expression is varied across and within T-cell lymphoma subtypes and could likely reflect the underlying percentage of more or less differentiated cytotoxic CD8⁺ T cells. Our results indicate that depletion of KLRG1⁺ tumor cells by mAbs that engage FcγRs on macrophages and NK cells and activate complements is a novel therapeutic option for at least a subset of mature T-cell and NK-cell neoplasms. Highly specific anti-KLRG1 mAbs can exert this antitumor efficacy while sparing KLRG1 low/negative B, NK, Tn Tcm, and a subset of CD8⁺ Tem and TemRA cells, thereby providing a potential therapeutic window of tolerability alongside activity. Our observations also suggest that the nature of the antitumor immune response effected by KLRG1 depletion might depend on the isotype backbone and N-linked glycosylation in the Fc region of the antibody. A mIgG2a afucosylated anti-KLRG1mAb could induce potent ADCC and ADCP of tumor cells with both human and murine effectors. This might be secondary to the increased binding affinity of the mAb Fc to FcγRs on effector cells resulting in their activation and enhanced function.

In several solid tumor mouse models such as the 4T1 breast cancer model, monotherapy with KLRG1 neutralizing antibodies inhibited metastasis in vivo. Its combination with anti–PD-1 antibodies had additional activity in abrogating tumor burden and extending survival in the MC38 colon cancer and B16F10 melanoma models (22, 27). Of interest, the combination resulted in increased frequency and activation of CD8⁺ T cells. It increased the frequency and maturation of NK cells in the tumor microenvironment, highlighting its role as a novel checkpoint. Given reports of PD-1 blockade-associated hyperprogression in patients with PTCL (28), we deliberately deployed a distinct strategy of KLRG1 depletion versus neutralization to directly reduce tumor burden and report for the first time its potential as a tumor cell–depleting agent in hematologic malignancies such as PTCL. We, however, believe that akin to differential clinical effects of PD-1 blockade in PTCL, further genetic and pharmacologic manipulation studies would need to be performed to elucidate the antitumor role of KLRG1 blockade in PTCL and this strategy could demonstrate substantial efficacy in certain subsets of PTCL.

Although KLRG1-depleting mAbs seem to induce clearance of TCL cells in vitro effectively, our in vivo studies in PDX revealed that depletion of KLRG1 alone was unable to induce a durable response despite continuous treatment. This could be partly explained by the lack of T and NK cells in NSG mice and hence the inability to exert the ADCC with reliance on ADCP alone as the primary driver of antitumor efficacy. Thus, the PDX studies do not capture many aspects of the in-human response to KLRG1-depleting mAbs. Interrogation of PDXs resistant to KLRG1 depletion demonstrated persistent expression of surface KLRG1 on a subset of tumor cells, suggesting that both antigen loss and escape contribute to resistance. Further, the PDX-derived tumor cells were able to bind to additional anti-KLRG1 mAbs ex vivo with similar and different Fab regions. Thus, highlighting that further optimization of our mAb208 would be necessary to fully unleash its antitumor activity. PI3K inhibition was able to enhance the antitumor efficacy of KLRG1 depleting mAbs in vivo through reprogramming of macrophages to a proinflammatory phenotype but the responses were not durable. Future planned studies will elucidate factors besides antigen loss and escape that contribute to tumor intrinsic and extrinsic mechanisms of innate and adaptive resistance to depletion of KLRG1-expressing tumor cells.

Future studies performed in humanized mice could help address the role of NK-cell fratricide in the efficacy and safety of anti-KLRG1 mAbs in vivo. Several studies have highlighted the dominant immunosuppressive and tumor-promoting characteristics of KLRG1⁺ Tregs in several murine models of non–small cell lung cancer fibrosarcoma, B16 melanoma, RLmale1 (radiation leukemia), and HPV16 E7-induced premalignant skin conditions (29–31). Thus, studies in conditional mouse models and humanized mice could discern the antitumor effects of depleting KLRG1⁺ Tregs in PTCL.

Our data add to the emerging literature that the specific elimination of KLRG1⁺ tumor cells utilizing therapeutic antibodies has the potential to diminish tumor burden in mature T-cell and NK-cell neoplasm. These data are also consistent with an ongoing first-in-human, open-label single ascending dose trial of ABC008, a first-in-class humanized afucosylated anti-human KLRG1 mAb in patients with IBM (NCT04659031; ref. 32). In study participants with IBM, a single subcutaneous dose of 0.1 mg/kg of ABC008 resulted in the depletion of CD8⁺KLRG1⁺ cells with no apparent safety concerns and no alterations in hematologic parameters, including regulatory T cells. CD8⁺ LGLs, because they are mostly KLRG1⁺, were also depleted. In concert, a phase I/II dose-escalation multicenter trial (NCT05532722) evaluating the safety and tolerability, of ABC008 in patients with T-LGLL who suffer from anemia and/or neutropenia is accruing. It is conceivable that screening of patients with discernible levels of surface KLRG1 on tumor cells by flow cytometry or IHC could be used as a selection criterion for clinical trial enrollment. Our data suggest that combinations with afucosylated anti-human KLRG1 IgG1 mAbs are needed to maximize the efficacy of the innate effector cells in subtypes of mature T-cell and NK-cell neoplasm that exhibit a greater expression of KLRG1⁺ tumor cells such as ENKTCL, T-PLL, and G/D TCL.

S. Mustjoki reports grants from Cancer Foundation Finland, Sigrid Juselius Foundation, and Academy of Finland during the conduct of the study as well as grants and personal fees from Novartis and BMS, grants from Pfizer, and personal fees from Dren Bio outside the submitted work. D.J. Feith reports grants from AstraZeneca, Recludix Pharma, and Kymera Therapeutics and grants and personal fees from Dren Bio outside the submitted work. T.P. Loughran reports personal fees from Keystone Nano, Flagship Labs 86, Dren Bio, Recludix Pharma, Kymera Therapeutics, and Prime Genomics outside the submitted work. J. Choi reports other support from Moonlight Bio outside the submitted work. T. Taghon reports grants from the Chan Zuckerberg Initiative, Research Foundation–Flanders (FWO), and Stichting Tegen Kanker during the conduct of the study. P.C. Johnson reports personal fees from Bristol Myers Squibb, AstraZeneca, AbbVie, Seagen, and ADC Therapeutics; grants and personal fees from Incyte; and grants from Medically Home outside the submitted work. E.D. Jacobsen reports personal fees from Pharmacyclics, Celgene, and Daiichi outside the submitted work. S.A. Greenberg reports personal fees and other support from Abcuro outside the submitted work; in addition S.A. Greenberg is cofounder of and scientific advisor to Abcuro, Inc. D.M. Weinstock reports grants from Abcuro during the conduct of the study as well as other support from Merck outside the submitted work; in addition, D.M. Weinstock is an employee of Merck/MSD. S. Jain reports grants and other support from Abcuro, Inc. during the conduct of the study as well as grants and other support from Abcuro, Inc outside the submitted work. In addition, S. Jain receives research funding, honoraria for scientific advisory board participation, and consulting fees from Abcuro, Inc; a consulting honorarium from Secura Bio; and consultancy or advisory committee or board membership with Mersana Therapeutics, Myeloid Therapeutics, SecuraBio, SIRPant Immunotherapeutics, Daiichi Sankyo, Crispr Therapeutics, and Acrotech LLC. S. Jain is affiliated with Mass General Brigham, which owns equity in Abcuro, Inc., one of the sponsors for this study. No disclosures were reported by the other authors.

B. Assatova: Data curation, methodology. R. Willim: Data curation, formal analysis, validation, methodology, writing–review and editing. C. Trevisani: Data curation, validation, methodology. G. Haskett: Data curation, formal analysis, methodology. K.M. Kariya: Data curation, methodology. K. Chopra: Data curation, methodology. S.R. Park: Formal analysis. M.Y. Tolstorukov: Supervision. S.M. McCabe: Data curation, methodology. J. Duffy: Data curation, methodology. A. Louissaint: Supervision, methodology. J. Huuhtanen: Data curation, methodology, writing–review and editing. D. Bhattacharya: Data curation, methodology. S. Mustjoki: Supervision. Min Jung Koh: Data curation, methodology. F. Powers: Data curation, methodology. E.A. Morgan: Supervision. L. Yang: Data curation, methodology. B. Pinckney: Data curation, formal analysis, methodology. M.J. Cotton: Data curation, methodology. A. Crabbe: Data curation, formal analysis, methodology. J.B. Ziemba: Data curation, methodology. I. Brain: Data curation, methodology. T.B. Heavican-Foral: Methodology. J. Iqbal: Supervision. R. Nemec: Data curation, methodology. A.B. Rider: Methodology. J.G. Ford: Data curation, methodology. Min Ji Koh: Formal analysis, methodology. N. Scanlan: Methodology. D.J. Feith: Data curation. T.P. Loughran: Data curation. W.S. Kim: Data curation, methodology. J. Choi: Data curation. J. Roels: Formal analysis, methodology. L. Boehme: Formal analysis, methodology. T. Putteman: Data curation. T. Taghon: Methodology. J.A. Barnes: Data curation. P.C. Johnson: Data curation. E.D. Jacobsen: Data curation. S.A. Greenberg: Resources, data curation, supervision, investigation, writing–review and editing. D.M. Weinstock: Resources, supervision, writing–review and editing. S. Jain: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, writing–review and editing.

This work was supported by Abcuro, Inc and the Center for Lymphoma Research Funds. S. Jain is supported by the National Cancer Institute K08 Career Development Award (K08CA230498) and MGH Lymphoma Research Funds. D.M. Weinstock is supported by NCI R35 CA231958 and P01 CA248384 and a Specialized Center of Research 7026-21 from the Leukemia and Lymphoma Society. E.D. Jacobsen is supported by the Reid Family Fund for Lymphoma Research. The authors thank University of Virginia LGL Leukemia Registry personnel Bryna Shemo and Matt Schmachtenberg for their support of this study. The LGL Leukemia Registry is supported by the Bess Family Charitable Fund and a generous anonymous donor (to T.P. Loughran).