An Antibody Targeting ICOS Increases Intratumoral Cytotoxic to Regulatory T-cell Ratio and Induces Tumor Regression

The last decade has seen a paradigm shift in cancer therapies with the approval of antibodies targeting immune checkpoints. These immune checkpoint inhibitors (ICI) trigger a durable response in malignancies, including metastatic melanoma, non–small cell lung cancer (NSCLC), head and neck cancer, renal, and bladder cancer (1). However, there is still a high proportion of patients exhibiting resistance to ICIs that may benefit from novel combinatory approaches.

Multiple molecular and cellular mechanisms associate with resistance to ICIs (2). For example, low incidence of cytotoxic T cells and the presence of immunosuppressive cells prevent an antitumor response. One class of immunosuppressive cells are regulatory T cells (Treg), which block the cytotoxic potential of effector T cells (T_Eff) through various mechanisms (3, 4). Thus, high numbers of intratumoral Tregs negatively correlates with survival and response to treatments (5, 6). In fact, Treg depletion modifies the tumor microenvironment (TME) and favors an antitumor response in preclinical models (7–9). For these reasons, Treg cells have been investigated as a prognostic cell type and as therapeutic targets.

The use of therapeutic antibodies for the preferential depletion of intratumoral Treg cells relies on the identification of a marker preferably expressed on these cells. One such potential target is ICOS, which belongs to the CD28/CTLA-4 family (10). Unlike CD28, ICOS is not expressed on naïve T_Eff cells but is induced upon T-cell receptor (TCR) engagement (11, 12). Following activation, ICOS is expressed at different levels on different T-cell subtypes where it can engage with its ligand (ICOS-LG, CD275) expressed on antigen-presenting cells. ICOS/ICOS-LG interaction initiates a costimulatory signal that results in production of either pro- or anti-inflammatory cytokines (IFNγ and TNFα by T_Eff cells; IL10 expression by Treg; ref. 13), thus regulating the immune cell homeostasis (14, 15). Of relevance, the accumulation of ICOS⁺ Treg cells in the TME is associated with disease progression (16, 17). In marked contrast, the upregulation of ICOS on CD4 T_Eff cells associates with better prognosis in patients treated with anti–CTLA-4 (18–21). The relative expression of ICOS varies between T-cell subtypes, with intratumoral Treg exhibiting higher ICOS expression followed by CD4⁺ and CD8⁺ T_Eff cells (12, 17, 19, 22). This differential expression suggests that ICOS represents a relevant target for a Treg depletion strategy. In fact, ICOS antibodies with depleting capability reduce the numbers of ICOS⁺ Treg cells and induce an antitumor response when combined with a vaccine strategy (7).

In this study, using a transgenic mouse platform (23), we identified a fully human ICOS IgG1 antibody called KY1044 that bound to mouse and human ICOS. In vitro, we established that KY1044 has costimulatory effect on ICOS^Low T_Eff cells and a depleting effect on ICOS^High cells. In vivo, KY1044 induced preferential depletion of ICOS^High cells, improved the T_Eff to Treg ratio in the TME and increased secretion of proinflammatory cytokines, resulting in antitumor efficacy.

All cell lines (except MC38) were obtained from LGC standards ATCC. CT26.WT (mouse colon carcinoma, ATCC catalog no. CRL-2638), B16-F10 (mouse melanoma, ATCC catalog no. CRL-6475) and MJ (G11; Human T-cell lymphoma, ATCC catalog no. CRL-8294) cells were acquired between July and November 2015, J558 (mouse plasmacytoma, ATCC catalog no. TIB-6) cell line was obtained in May 2016. A20 (A-20; mouse B-cell lymphoma, ATCC catalog no. TIB-208), and EL4 (mouse lymphoma, ATCC catalog no. TIB-39) cells were obtained in May 2017. MC38 cells (mouse colon adenocarcinoma) were obtained in August 2017 from NCI under a license agreement. The specific pathogen-free status of these cells was confirmed by PCR screening for mouse/rat comprehensive panel (Charles River). MC38, J558, and B16-F10 cells were cultured in antibiotic-free DMEM (Gibco, catalog no. 41966-029) + 10% FBS (Gibco, catalog no. 10270) complete cell culture media. CT26.WT cells were cultured in antibiotic-free RPMI (Gibco, catalog no. 2187) + 10% FBS complete cell culture media. A20 cells were cultured in antibiotic-free RPMI + 10% FBS + 0.05 mmol/L 2-mercaptoethanol (Gibco, catalog no. 21985-023) complete cell culture media. MJ cells were culture in IMDM (Gibco, catalog no. 12440053) + 20% FBS (I20 media). The passage number of cells were kept below 10 generations.

The RNA-sequencing data from The Cancer Genome Atlas (TCGA) consortium was downloaded from the UCSC Xena platform (TCGA Pan-Cancer, 10,460 samples in total; ref. 24). Samples classified as nontumor tissue (727 samples) were excluded as were leukemias, lymphomas, and thymomas (combined 341 samples). Single-sample gene-set enrichment analysis (ssGSEA; refs. 25, 26) was performed for ICOS and FOXP3. Samples were grouped by primary disease and the ssGSEA scores for each group were compared across the primary disease groups.

Peripheral blood mononuclear cell (PBMC) and tumors from 5 NSCLC donors were processed using the BD Rhapsody system. Sequencing libraries were generated using the Immune Response Human targeted panel and sequenced using a 2 × 75 bp paired-end run on the Illumina HiSeq 4000 System. Reads were processed by applying the BD Rhapsody processing pipeline to generate cell count matrices. The counts were filtered, normalized, and visualized using R and Bioconductor packages for scRNA-seq data (27–30). Cell type–specific gene sets were constructed by performing a literature search and cells were classified into one of 27 cell types (Supplementary Table S1) using the R package AUCell (31). The sequences have been submitted to ArrayExpress (accession E-MTAB-9451).

Tumors and spleens were harvested from CT26.WT tumor–bearing mice. Tumors were dissociated into single-cell suspensions using Miltenyi Tumor Dissociation Kit (catalog no. 130-096-730). Spleens were placed into C-tubes (Miltenyi Biotec, catalog no. 130-096-334) dissociated into single-cell suspensions using gentle MACS dissociator. Tumor samples were filtered through 70-μm nylon filters. Spleens samples were filtered through 40-μm nylon filters. Red blood cells were lysed using RBC lysis buffer (Sigma, catalog no. R7757). For flow cytometry profiling, 2 × 10⁶ tumor samples or 1 × 10⁶ spleen samples were plated in 96-well plate (Sigma, catalog no. CLS3957). Cell suspensions were preincubated with Live/Dead fixable Yellow Dead Cell Stain Kit (Invitrogen, catalog no. L34959). Prior to antibody labeling, cells were incubated with Fc receptor blocking solution (anti-CD16/CD32 BioLegend, catalog no. 101320) at 4°C for 10 minutes. The staining was performed at 4°C for 30 minutes using fluorochrome-conjugated anti-mouse antibodies. Intracellular and intranuclear staining was performed using Foxp3 staining buffers (Thermo Fisher Scientific, catalog no. 00-5523-00). All flow cytometry antibodies or isotype controls were purchased from Thermo Fisher Scientific. Antibodies used include: anti-CD45 (30-F11), anti-CD3 (17A2), anti-CD4 (RM4-5), anti-CD8 (53-6.7), anti-CD25 (PC61.5), anti-ICOS (7E.17G9), anti-Foxp3 (FJK-16s), anti-CD69 (H1.2F3). For the T-cell cytokine staining, single-cell suspensions were plated at 1 × 10⁶ cells per well in RPMI + 10% FBS cell culture media with 1× Brefeldin A solution (eBioscience, catalog no. 00-4506-51) for 4 hours at 37°C 5% CO₂. Cells were surface stained as above and subsequently fixed/permeabilized for intracellular staining with anti-IFNγ (XMG1.2) and anti-TNFα (MP6-XT22). All flow cytometry data was acquired using Attune FxT flow cytometer and data were analyzed using FlowJo software V10.

Blood samples (0.1 mL) were collected into sodium EDTA or lithium heparin tubes, mixed, and red blood cells (RBC) were lysed. The leukocytes were washed with FACS buffer (PBS with 2% FBS), stained with antibody cocktail by incubation in the dark for 45 minutes at room temperature. Tissue samples (lymph node), taken from animals at termination were mechanically disrupted (Medimachine, Becton Dickinson GmbH), single-cell filtered, and stained as for blood. Intracellular FoxP3 staining was carried out by permeabilization with 10× FACS lysing solution (1:5 with Aqua dest. +0.1% Tween-20) prior to incubation in the dark, 45 minutes at room temperature. All antibodies or isotype controls were purchased from BD Biosciences, eBioscience (via Fisher Scientific GmbH) or BioLegend. Antibodies used included: anti-CD3 (SP34), anti-CD20 (L27 or 2H7), anti-CD14 (M5E2), anti-CD4 (M-T477), anti-CD8 (SK1), anti-CD28 (CD28.2), anti-CD25 (M-A251), anti-Foxp3 (PCH101), anti-CD95 (DX2), and anti-CD185/CXCR5 (MU5UBEE). After staining, cells were washed and fixed in 1× BD Stabilizing Fixative. Lymphocytes were gated by forward scatter (FSC), sideward scatter (SSC), and CD45. Multicolour flow cytometric analysis was performed using the following leukocyte phenotypic characteristics: CD4⁺ Th cells: CD3⁺/CD4⁺/CD8⁻/CD14⁻/CD20⁻; CD8⁺ cytotoxic T cells: CD3⁺/CD4⁻/CD8⁺/CD14⁻/CD20⁻; Total memory CD4 T cells: CD28^+/DIM/− CD95^+/high; Follicular Th cells: (CD4⁺/CD185⁺), and Regulatory CD4 T cells: CD25⁺/FoxP3⁺. Acquisition of flow data was performed on a FACSVerse flow cytometer (BD Biosciences) and relative percentages of each of these subpopulations were determined using FlowJo software. Fifty thousand events were counted for all analyses.

In accordance to the declaration of Helsinki and upon approval by the Hamburg Medical Association's ethic commission (PV5035), NSCLC tissue and whole blood from the same patients were obtained from consented subjects, with ethical approval for analysis of protein, RNA, and DNA content. The NSCLC tumor samples were dissociated into single-cell suspensions using enzymatic digestion, while whole blood was processed into PBMCs using density centrifugation, all specimens were cryopreserved in Gibco Recovery cell culture freezing medium before used. Alongside NSCLC specimens, PBMCs from 5 healthy individuals were used. Flow cytometry was performed after cells were thawed and incubated in RPMI1640 + 10% FBS supplemented with 100 U of RNase free DNase at 37°C 5% CO₂ for 30 minutes. Cell suspensions were pre-incubated with Live/Dead fixable Yellow Dead Cell Stain Kit (Invitrogen, catalog no. L34959) and Fc receptor blocking solution (BioLegend, catalog no. 422302), before staining with fluorochrome-conjugated anti-human antibodies specific to anti-CD3 (UCHT1)/anti-CD45 (2D1), anti-CD4 (2A3), anti-CD8 (RPA-T8), anti-CD25 (MA251), anti-CD45RA (H100), anti-ICOS (C398.4A). Cells were incubated for 1 hour at 4°C, washed, and fixed overnight at 4°C with eBioscience intracellular fixation and permeabilization buffer (catalog no. 00-5523-00). Anti-FoxP3 (236A-E7) staining was then performed in permeabilization buffer for 1 hour at 4°C before washing and resuspending cells in DPBS (Gibco). All flow cytometry data was acquired using Attune NxT flow cytometer and data was analyzed using FlowJo software V10.

An IHC protocol using the Ventana Discovery Ultra platform (Roche) was developed for costaining of ICOS and FoxP3. The method was optimized in FFPE tonsil tissue, using clone D1K2T for ICOS (Cell Signaling Technology) and clone 236A/E7 (Abcam) for FoxP3. Briefly, the FFPE tissue sections underwent deparaffinization and a pH 9-based antigen retrieval step for 64 minutes. The slides were incubated with anti-FOXP3 antibody or the isotype control at 40 μg/mL for 60 minutes. An anti-mouse HRP (Roche catalog no. 760-4310) labeled detection antibody was applied for 16 minutes and an automated DAB detection (Roche catalog no. 760-159) was carried out as per manufacturer's recommendations. The slides were then incubated with the anti-ICOS antibody or the isotype control at 1 μg/mL for 60 minutes. An anti-rabbit HRP (Roche, catalog no. 760-4311) labeled detection antibody was applied for 28 minutes and an automated purple detection was carried out as per manufacturer's recommendations. Finally, the slides were counterstained with hematoxylin for 12 minutes and bluing reagent for 8 minutes. Methods were applied for the staining of tumor microarrays (TMA) from patients with cervical (CR2088), esophageal (ES2082), lung (LUC2281), head and neck (HN802a), gastric (STC2281), and bladder (BL2081a) cancer (all US Biomax, Inc). The slides were scanned (20× equivalent magnification) and quantified using the Indica Labs Halo platform to determine the number of ICOS/FoxP3 single and double positive cells per mm².

Antibody-dependent cellular cytotoxicity (ADCC) was first tested in vitro using a reporter bioassay (Promega, catalog no. G7102). In brief, the assay uses a Jurkat reporter cell line expressing human FcγRIIIa V158 and NFAT-induced luciferase. Following engagement with the Fc region of a relevant antibody bound to a target cell, the FcγRIIIa receptors can activate an intracellular signals resulting in NFAT-mediated luciferase activity that can be quantified via a luminescence readout. CHO cells expressing either human, mouse, rat, or cynomolgus ICOS were used as target cells and coincubated with Jurkat reporter cells at a 5:1 ratio. Serial dilutions of KY1044 or isotype IgG1 control were added to the culture plates, incubated at 37°C overnight and the luciferase activity was measured using the Bio-Glo Luciferase Assay System (Promega, catalog no. G7940) on the EnVision Multilabel Plate Reader (Perkin Elmer). Graph data were normalized to background and plotted versus log10 antibody concentration.

For the primary ADCC cell assay, human natural killer (NK) cells were purified using the NK Cell Isolation Kit (StemCell Technologies, catalog no.17955). ICOS-transfected CCRF-CEM (ICOS CEM, target cells) were preloaded with the fluorescence-enhancing ligand (BATDA) for 30 minutes in the dark at 37°C. The cells were loaded and washed in the presence of an inhibitor of organic anion transporters (1 mmol/L Probenecid) to avoid spontaneous dye release from cells. KY1044 was diluted (1:4 dilutions, 10 points, starting from 33.3 nmol/L) in assay buffer. The target ICOS CEM cells (50 μL/well), effector cell (50 μL/well), and reagent dilutions (50 μL/well) were cocultured with 50 μL of the diluted antibody at 37°C, 5% CO₂ for 2–4 hours (NK cells to target ratio was 5:1). A digitonin-based lysis buffer (Perkin Elmer) was used to determine complete target cell lysis (100%).

For the MJ cell assays, KY1044 human IgG1 was presented in three different formats: plate-bound, soluble, or soluble with or without F(ab')₂ Fragments (Fc linker, catalog no. 109-006-170, Jackson Immunoresearch). Plate-bound KY1044 and the hIgG1 isotype control were diluted 1:2 in PBS (concentrations ranging from 10 μg/mL to 40 ng/mL, 10 points). One-hundred microliters of diluted antibodies were coated in triplicate into a 96-well, flat-bottom plate (Corning EIA/RIA plate) overnight at 4°C and then washed. MJ cells (15,000 cells/well) were added to precoated wells. For the soluble/cross-linked experiment, KY1044 and the isotype control were serially diluted 1:2 in I20 media (soluble Ab) or in I20 media containing 30 μg/mL of F(ab')₂ Fragments (cross-linked Ab) to give an 2× Ab stock concentrations ranging from 20 μg/mL to 80 ng/mL (10 points). Fifty microliters of diluted antibodies were added to 96-well with 50 μL of MJ cells (3 × 10⁵/mL). For the assays the cells were cultured for 72 hours at 37°C/5% CO₂ and cell-free supernatants were then collected and used to perform IFNγ ELISA using the Human IFNγ DuoSet assay (R&D Systems, DY285).

For the primary T-cell assays, leukocyte cones were obtained (HTA IRAS project number 100345). PBMCs were isolated by density-gradient centrifugation. T lymphocytes were them purified using the Stemcell EasySep Isolation kit (catalog no. 17951). For ICOS induction, isolated T cells were cultured at 2 × 10⁶/mL in R10 media (RPMI 10% heat-inactivated FBS) in the presence of 20 μL/ml of Dynabeads Human T-Activator CD3/CD28 (Life Technologies, catalog no. 111.31D). These activated primary T cells were tested as for the MJ cell assays in three different formats: plate-bound (5 μg/mL), soluble (15 μg/mL), or soluble plus F(ab')₂ Fragments cross-linker. T-cell suspension were added to antibody-containing plates to give a final cell concentration of 1 × 10⁶ cells/ml and cultured for 72 hours at 37°C and 5% CO₂ until IFNγ ELISA quantification. For the three-step culture (stimulation-rest-costimulation assay), the T cells were prestimulated by Dynabeads for 3 days to induce ICOS before being rested for 3 days to reduce their activation levels. These stimulated/rested T cells were then cultured with KY1044 in the presence or absence of an anti-CD3 antibody (clone UCHT1, eBioscience) to assess the requirement of TCR engagement. The effect of ICOS costimulation was assessed after 72 hours by measuring the levels of IFNγ and TNFα present in the culture (MSD multiplex assay).

For the gene expression analysis, T cells were harvested from the 3-step culture (stimulation–rest–costimulation assay) after 6-hour plate-bound antibody stimulation. Total RNA was extracted from the cell pellets with the RNeasy Micro Kit (Qiagen), quality controlled on the Agilent 2100 Bioanalyzer (Agilent Technologies) and subjected to SE50 sequencing following mRNA enrichment (BGI). The sequence reads were aligned using kallisto (32) and further processed using limma (33) and metascape (34) and GSVA (26). The sequencing data has been deposited into ArrayExpress (E-MTAB-9500).

Jurkat-Lucia NFAT cells (Invivogen, catalog no. jktl-nfat) stably expressing luciferase under the control of NFAT response elements were further transfected with either the human Icos gene sequence or a chimeric construct of Icos fused with CD247 (CD3ζ). Following selection of stable transgene integration, a luciferase activity bioassay was performed. High binding 96-well assay plates were coated overnight with either anti-CD3 (UCHT-1, eBioscience, catalog no. 16-0038-85; 10 μg/mL), anti-ICOS (C398.4A, BioLegend catalog no. 313512; 10 μg/mL), isotype control (HTK888, BioLegend, catalog no. 400902; 10 μg/mL), anti-CD3 + anti-ICOS (5 μg/mL each) or anti-CD3 + isotype control (5 μg/mL each). Transgene expressing cells or control cells (untransfected Jurkat-Lucia NFAT cells) were seeded at 50,000 cells/well. Following overnight incubation luciferase activity was measured by adding BioGlo reagent (Promega, catalog no. G7940) and reading luminescence on a plate reader. Luciferase activity was normalized and scaled to 100% by comparing to cells stimulated using Cell Stimulation Cocktail (eBioscience, catalog no. 00-4970-93).

The thaw-and-use format of the PD-1/PD-L1 Blockade Bioassay (Promega, catalog no. CS187111) was performed in a 96-well plate format according to the manufacturer's instructions. Briefly, PD-L1–expressing APC/CHO-K1 cells were plated on day –1 in cell recovery medium and incubated overnight at 37°C. The following day, assay media was replaced with 10-point serial dilutions of AbW IgG1 or isotype control, prior to addition of PD-1 expressing NFAT Jurkat effector cells. Cells were incubated at 37°C for 6 hours, followed by the addition of the BioGlo reagent and luminescence signal quantified on a plate reader.

Full-length human PD-L1 sequence (Uniprot sequence ID: Q9NZQ7-1) was codon optimized for mammalian expression and cloned into an expression vector under the CMV promoter flanked by 3′ and 5′ piggyBac-specific terminal repeat sequences, facilitating stable integration (puromycin selection) into the genome of transfected CHO cells (35). Stable PD-L1–expressing CHO cells were coincubated with a 30 nmol/L concentration of biotinylated ligand (either PD-1 or CD80) and serial dilutions of antibody (150 nmol/L to 0.0076 nmol/L of either AbW hIgG1 or control hIgG1) for 1 hour at 4°C. The cells were then washed, ligand binding was detected using Alexa Fluor 647–labeled streptavidin and the samples were acquired on a BD FACSArray Bioanalyzer.

A T-cell activation assay using the OVA-specific DO11.10 mouse T-cell hybridoma was used as described previously (36). In brief, human PD-L1–transfected LK35.2 cells (antigen-presenting cells) were incubated with mouse PD-1–expressing DO11.10 cells in the presence of OVA peptide. Serial dilutions of AbW IgG1 or isotype control were added and incubated at 37°C overnight. The following day mouse IL2 release was quantified using the Mouse IL2 DuoSet ELISA (R&D Systems, catalog no. DY-402).

The ability of KY1044 to be internalized was assessed by capturing time course images on the IncuCyte ZOOM Live Cell Imaging Platform (Sartorius). Briefly, MJ cells endogenously expressing ICOS were first labeled with the Incucyte CytoLight Rapid Green Reagent (Sartorius) to enable detection of the cytoplasm. Serial dilutions of either KY1044, a CD71 antibody (positive control; Sigma, catalog no. SAB4700520-100UG) or a human IgG1 isotype control (negative control) were preincubated with secondary anti-human or anti-mouse antibodies covalently labeled with a pH-sensitive dye (pHrodo Red; Thermo Fisher Scientific, catalog no. P36600) and then added to the target cells. Antibody internalization was tracked for up to 48 hours and the areas under the time-course curves plotted as a function of antibody concentration.

All mice in vivo work was performed in the UK under Home Office licence (70/08759). All procedures were conducted in accordance with the United Kingdom Animal (Scientific Procedures) Act 1986 and associated guidelines, approved by institutional ethical review committees (Babraham Institute AWERB). Eight- to 10-week-old wild-type female Balb/C or C57BL/6J mice were sourced from Charles River UK Ltd and housed in transparent plastic cages with wire covers (391 W × 199 L × 160 H mm, floor area: 500 cm²) containing Grade 6 Wood Chip that can be replaced with Lignocell (IPS Product Supplies Ltd, BCM IPS Ltd.; WC1N 3XX) and bale shredded nesting material (IPS Product Supplies Ltd, BCM IPS Ltd.; WC1N 3XX). Four to five mice were housed per cage in a room with a constant temperature (19°C–23°C) and humidity (40%–70%) and a 12-hour light–dark cycle (lighting from 7 am to 7 pm). Mice were provided with pellet food (CRM(P), Specialist Diet Services) and RO water ad libitum using an automatic watering system.

Early passage (below P10) MC38 (3 × 10⁶ cells), J558 (1 × 10⁶ cells), CT26.WT (1 × 10⁵ cells), A20 (5 × 10⁶ cells), EL4 (1 × 10⁴ cells), and B16-F10 (1 × 10⁵ cells) tumor cells were prepared in either PBS or Matrigel (Corning, 354230) and the resulting cell suspension were injected subcutaneously into the flank of the mice (study day 0). Prior to tumor cell implantation, mice were anesthetised with isoflurane and the right flank of the mice was shaved. For the implantation, 100 μL of the cell suspension were injected using 25 G needles (BD Microlance TM 3. VWR 613-4952). Cell numbers and viability (required to be above 90%) were determined preimplantation by the trypan blue assay. Tumor growth was measured using digital calipers three times a week until end of the study. The tumor volumes (mm³) was estimated using a standard formula: (L × W²)/2 (with L being the larger diameter, and W the smaller diameter of the tumor). All data were plotted using GraphPad Prism V10.

Both KY1044 mIgG2/mIgG1 and anti–PD-L1 (AbW) mIgG2a/mIgG1 were produced by Kymab Ltd. The PD-1 (clone RMP1-14) mAb was purchased from BioXcell. Antibodies were dosed between 0.1 and 10 mg/kg or by flat dose of 20 to 200 μg/dose via the intraperitoneal route. For the efficacy and pharmacodynamic studies, the treatment groups were not blinded. The in vivo depletion of CD8⁺ and CD4⁺ T cells were conducted using a flat dose 200 μg of anti-CD8a (53-6.7) and/or anti-CD4 (GK1.5). Dosing was performed twice a week for 3 weeks starting from day 3 following tumor cell implantation. The CD8⁺ and CD4⁺ T-cell depletion was determined by flow cytometry of tumor, spleen, tumor draining lymph node (inguinal lymph node), and blood cells using an anti-CD3 antibody (17A2).

The effects of KY1044 in nonhuman primates (NHP) were studied as part of a repeat dose toxicity study. Naïve male cynomolgus monkeys (Macaca fascicularis) were obtained from a certified supplier, group housed, allowed access to water ad libitum and fed on a pelleted diet for monkeys supplemented with fresh fruit and biscuits. Animals ranged from 4 to 7 years old and weighed 3–6 kg at time of dosing. Four groups of three cynomolgus monkeys received weekly intravenous doses of KY1044 (slow bolus over approximately 1 minute) for a month (5 doses in total) at doses of 0 (vehicle control; PBS pH 7.4), 10, 30, or 100 mg/kg at a dose volume of 2 mL/kg. Blood samples were taken at 1 or 2 timepoints prior to dosing and at multiple timepoints up to 29 days after the first dose for measurement of serum KY1044 using a qualified ELISA assay, ICOS occupancy on CD4⁺ cells in blood determined using a validated flow cytometry method and/or immunophenotyping of whole blood (described above). For the receptor occupancy, free ICOS was measured by using a competing anti-ICOS that does not displace KY1044 while an anti-human IgG was used to detect KY1044 directly or after saturating the cells with an excess of KY1044 to quantify bound and total ICOS, respectively. Scheduled necropsies were conducted 1 day after the final dose (day 30) and spleen and mesenteric lymph nodes were taken for immunophenotyping.

The cynomolgus monkey study was conducted at Covance preclinical Services GmbH in strict accordance with a study plan reviewed and approved by the local Institutional Animal Care and Use Committee (IACUC) of the testing facility and the German Animal Welfare Act. The study was performed according to DIRECTIVE 2010/63/EU OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL of September 22, 2010 on the protection of animals used for scientific purposes and the Commission Recommendation 2007/526/EC on guidelines for the accommodation and care of animals used for experimental and other scientific purposes (Appendix A of Convention ETS 123). The study was in compliance with the Kymab Working Practice on Experiments involving Animals and was approved by the Kymab Ethics Committee.

Unless stated otherwise in the figure legends, the efficacies observed for the different treatment groups were compared using either t test or one-way ANOVA and Tukey multiple comparison post hoc test. Differences between groups were significant at a P < 0.05. On the graphs, *, P < 0.05; **, P < 0.01; ***, P ≤0.001; ****, P ≤ 0.0001. Statistical analyses were performed with GraphPad Prism 10.0 (GraphPad Software, Inc.).

Ovarian, gastric, and liver cancers, are highly infiltrated with ICOS⁺ Treg cells (16, 17, 37) and this results in poor prognosis (17, 38). To assess ICOS expression on intratumoral T cells and to identify additional tumor types containing high ICOS⁺ Treg cells, we analyzed the content of T cells in tumors using different approaches.

Using the TCGA datasets, we identified tumors originating from the head and neck, stomach, cervix, thymus, testis, skin, and the lungs as indications with high mRNA expression of ICOS and FOXP3 (a marker of Treg; Fig. 1A). To establish which types of cells expressed ICOS and/or FOXP3, we assessed these markers at the mRNA and protein levels by single-cell transcriptomics and FACS analysis, respectively. We generated paired PBMCs and tumor scRNA-seq data for 79,544 cells from 5 patients with NSCLC (Supplementary Table S2), which we subsequently stratified into 27 cell subtypes based on published immune gene expression signatures. As expected, genes such as foxp3, ccr8, il2ra, tnfrsf4 were highly expressed in intratumoral Tregs (Fig. 1B). ICOS mRNA expression was higher in Tregs than in other T-cell subsets. Finally, ICOS expression was higher in the TME than in the periphery, whereas the expression pattern in both compartments followed the general trend of: Treg > CD4_non-Treg > CD8 > Other (Fig. 1B). By flow cytometry analysis (Fig. 1C; Supplementary Fig. S1A), ICOS expression was higher on Treg (CD4⁺/FOXP3⁺) than on the other T-cell subsets (CD8⁺ and CD4⁺/FOXP3⁻). ICOS protein was also more expressed in the TME than on PBMCs (P < 0.001 for Tregs). No differences in ICOS expression between PBMCs from healthy donors and patients with NSCLC were observed. Finally, T-cell immunophenotyping of CT26.WT tumor–bearing mice revealed that ICOS was also more expressed in the TME than in the spleen. This analysis also confirmed higher ICOS expression on Treg than on CD4/CD8 T_Eff cells (Supplementary Fig. S1B–S1D).

ICOS immunostaining on tumor tissue microarrays (from esophageal, head and neck, gastric, lung, bladder, and cervical cancer biopsies) confirmed the upregulation of ICOS in the TME (Fig. 1D). The costaining for ICOS and FOXP3 (Fig. 1E), also highlighted a high number of ICOS⁺ Treg cells for the six selected indications, with only bladder cancer demonstrating a lower density of ICOS/FOXP3-positive cells (Fig. 1F).

Altogether, our data confirmed that ICOS was not homogenously expressed on the T-cell subsets and was induced in the TME, especially on the surface of Treg. In addition, we identified indications with high levels of intratumoral ICOS⁺ Treg cells including esophageal, head and neck, gastric, lung, and cervical cancers.

The strong expression of ICOS on intratumoral Tregs suggests that the targeting of these cells with an effector enabled ICOS antibody could lead to their preferential depletion (e.g., via ADCC and/or ADCP). The level of target expression and the amount of antibody bound to a target are key parameters influencing cell killing by ADCC (39). Therefore, it is expected that ICOS^Low cells (e.g., CD8⁺ T_Eff cells) will be less sensitive to depletion than ICOS^High Treg, but could be sensitive to costimulation if the antibody also harbors agonistic properties (via target clustering through FcγR receptors; refs. 40, 41). With this in mind, we identified a fully human monoclonal IgG1 antibody called KY1044. KY1044 was generated in a transgenic mouse line for the human immunoglobulin genes in which the endogenous Icos gene was knocked-out. Lack of endogenous ICOS expression was aimed to facilitate the generation of a mouse cross reactive antibodies. KY1044 binds ICOS from different species (human, Cynomolgus monkey, rat, and mouse) with similar affinity (a K_d below 3 nmol/L for the Fab, Supplementary Fig. S2A).

KY1044-dependent ADCC was assessed using a luciferase reporter assay. As shown in Fig. 2A, KY1044 significantly induced the luciferase signal (EC₅₀ of 0.15 nmol/L, n = 3). A similar EC₅₀ was obtained against mouse ICOS (0.53 nmol/L, n = 3), rat ICOS (0.48 nmol/L, n = 3), or cynomolgus monkey ICOS (0.22 nmol/L, n = 3). We validated the data using human NK cells as effector cells. When incubated with KY1044 and CEM cells expressing hICOS (5:1 effector:target cell ratio), these NK cells induced potent ADCC-mediated killing (EC₅₀ of 5.6 pmol/L, n = 8, Fig. 2B). Altogether, these data confirmed that KY1044 triggers ADCC of ICOS^High-expressing cells.

The agonistic potential of KY1044 was assessed using the MJ [G11] CD4⁺ cells that we demonstrated express ICOS and do not require a primary stimulatory signal (e.g., TCR engagement) for cytokine production. KY1044 was either precoated on culture plates (mimicking cross-presentation/clustering) or used in solution with or without addition of a secondary cross-linking antibody. While soluble KY1044 did not induce IFNγ production, plate-bound, and cross-linked KY1044 effectively induced IFNγ secretion (EC₅₀ of 10.5 nmol/L ± 2.7 nmol/L; n = 2 and 0.50 nmol/L ± 0.18 nmol/L, n = 3, respectively; Fig. 2C and D). This agonism potential was confirmed using primary T cells preactivated with anti-CD3/CD28 (to induce ICOS expression; Supplementary Fig. S2B) and cultured with KY1044. Plate-bound (Fig. 2E) and cross-linked (Fig. 2F) KY1044 induced IFNγ production in primary cells. KY1044-dependent IFNγ secretion was significantly higher than for the isotype-treated cells in both plate-bound (2.4 ± 0.5-fold at 5 μg/mL mAb, P < 0.01) and cross-linked assays (2.5 ± 0.2-fold at 15 μg/mL mAb, P < 0.01). Importantly, a three-step stimulation–resting–costimulation experiment confirmed that TCR engagement was required for KY1044-dependent upregulation of cytokines (such as TNFα; Fig. 2G).

Finally, a transcriptomic analysis of CD3⁺ T cells was performed following 6 hours of combined anti-CD3 and KY1044 costimulation. In line with previous reports (14), we observed cytokine upregulation, most notably IFNG, IL10, and IL4 (Fig. 2H). Gene set enrichment analysis confirmed that KY1044 induced genes involved in cytokine–cytokine receptor interactions and and lymphocyte activation (Fig. 2H). Of relevance, no significant change in genes associated with proliferation and cell-cycle progression were observed. However, in agreement with a previous report (42), the analysis showed an enrichment of genes involved in cell locomotion, adhesion, and differentiation. This finding also reflected our observations of MJ [G11] CD4⁺ morphology changes in response to anti-ICOS (C398.4A or KY1044; Supplementary Fig. S2C; Supplementary Movies S1 and S2). Even though ICOS signaling depends on the PI3K/AKT/mTOR axis (43), following KY1044 agonism, we noticed an overrepresentation of genes downstream of the nuclear factor of activated T cells (NFAT; Fig. 2H), as seen previously (44). NFAT-dependent ICOS agonism was confirmed using reporter cell lines (Supplementary Fig. S2D). Target-mediated internalization of KY1044 was limited (Supplementary Fig. S2E).

Altogether the data presented here demonstrated that KY1044 has a dual mechanism of action with both depleting and costimulatory properties.

KY1044 cross-reactivity to mouse ICOS facilitated the in vivo mouse pharmacology work for which the antibody was reformatted as a mouse IgG2a (the effector enabled format in mouse). Within the TCGA datasets, ICOS expression was high in diffuse large B-cell lymphoma (Supplementary Fig. S3). On the basis of the role of ICOS/ICOS-LG signaling in the generation and maintenance of germinal centres (10, 45) and on the antitumor efficacy of ICOS antibodies in lymphoma (46), we assessed the effect of KY1044 mIgG2a in the ICOS-LG⁺ A20 tumor lymphoblast B-cell model (47). Mice were dosed (2qW for 3 weeks at 10 mg/kg) starting from 6 days following tumor cell implantation with saline and KY1044 mIgG2a. KY1044 mIgG2a treatment triggered an antitumor response, with more than 90% of the mice being free from measurable disease at the end of the study (day 42; Fig. 3A). This efficacy was confirmed in another B-cell–derived syngeneic model, the J558 plasmacytoma model in which around 70% of the KY1044 mIgG2a–treated mice were tumor free at the end of the study (Fig. 3B).

Monotherapy efficacy was also assessed in models of hematologic malignancies (T-cell lymphoma EL4) and models of solid tumors (CT26.WT, MC-38, and B16.F10). Overall, the monotherapy response was absent or low in all these models. Mice harboring CT26.WT or MC-38 tumors showed only tumor growth delay (Supplementary Fig. S4).

Altogether, our in vivo studies demonstrated that KY1044 mIgG2a was highly effective as a monotherapy in B-cell–derived tumor models.

CT26.WT tumors respond poorly to anti–PD-L1 (48). Because anti-ICOS and the anti–PD-L1 act on complementary immune pathways, we combined KY1044 mIgG2a with anti–PD-L1 (AbW). AbW binds to mouse PD-L1 with high affinity (K_d = ∼2 nmol/L) and blocks PD-L1/PD-1 and PD-L1/CD80 interactions (Supplementary Fig. S5A). These studies (Fig. 4A) demonstrated a complete antitumor response in the majority (ranging from 50% to 90%) of mice treated with the KY1044 mIgG2a (equivalent of 3 mg/kg) and anti–PD-L1 (equivalent of 10 mg/kg) combination. Of relevance, we observed a trend (albeit not significant) showing higher antitumor efficacy when anti–PD-L1 was used as mIgG2a than as mIgG1 (Supplementary Fig. S6A). Finally, when KY1044 was reformatted as a mouse IgG1 (low depleting potential in mouse) and combined with anti–PD-L1, the resulting antitumor efficacy was weaker than the one observed with the corresponding KY1044 mIgG2a/anti–PD-L1 combination, thus arguing for the contribution of ICOS-mediated depletion to the stronger antitumor efficacy seen in the CT26.WT model (Supplementary Fig. S6B).

Mice that survived the first CT26.WT challenge in response to the KY1044 mIgG2a/anti–PD-L1 treatment were resistant to a CT26.WT rechallenge but sensitive to the growth of EMT-6 cells (Fig. 4A) suggesting a tumor antigen–specific memory response. Finally, we demonstrated that the response to the combination was primarily dependent on CD8⁺ T cells, with the improved survival fully abrogated when CD8⁺ T cells were depleted (Fig. 4B; Supplementary Fig. S5). Conversely, addition of a CD4-depleting antibody (which depletes both CD4 Tregs and CD4 non-Tregs) to the combination was still associated with tumor growth delay but CD4 depletion eliminated any long-term survival benefit (Fig. 4B). Altogether, the coadministration of KY1044 mIgG2a and anti–PD-L1 triggered a durable antitumor immune response in the CT26.WT tumor model. The combination of anti–PD-1 (clone RMP1-14) with KY1044 was poorly effective in the CT26.WT model (Supplementary Fig. S6C), whereas it was effective in another model (MC38; Supplementary Fig. S6D).

The KY1044 mIgG2a and anti–PD-L1 combination was also tested in other models (Supplementary Fig. S7A). Although the J558 model already responded well to either the anti-ICOS or the anti–PD-L1 monotherapy, a 100% response was achieved with the combination in this model. Similarly, the combination was effective in the MC38 model (Supplementary Fig. S7B). The B16F10, 4T1, and EL4 tumor models did not respond to the combination (Supplementary Fig. S7A).

Collectively, our efficacy studies demonstrated that the cotargeting of ICOS and PD-L1 resulted in a strong combinatorial effect in selected tumor models, including those in which both monotherapies were poorly effective.

To assess the mechanism of action of KY1044 in vivo, we first conducted pharmacodynamic studies in the CT26.WT model. Tumor-bearing mice were dosed with KY1044 mIgG2a (ranging from 0.3 to 10 mg/kg) on day 13 and day 15 following tumor cell implantation (Fig. 5A). The T-cell content in the tumors and the spleens were analysed on day 16. Although KY1044 monotherapy did not induce much antitumor efficacy in this model (Fig. 4A), KY1044 mIgG2a was associated with intratumoral Treg cell depletion, even at doses as low as 0.3 mg/kg (Fig. 5B). The antibody format was critical to decrease Tregs as shown with partial and nonsignificant Treg depletion when using a mIgG1 (poor depleter; Supplementary Fig. S8A). In addition, an increase in the CD8⁺ T_Eff to Treg cell ratio (known to be associated with improved response to ICIs) in the TME was observed at all doses tested (Fig. 5B). However, the highest dose of 10 mg/kg showed a lower CD8⁺ T_Eff to Treg cell ratio than the one resulting from the 1 and 3 mg/kg doses. This decrease in the ratio at the highest dose may have been caused by the depletory effect of KY1044 (albeit not significant) on CD8⁺ T cells (Supplementary Fig. S8B). These effects were not observed in the spleen (Fig. 5C; Supplementary Fig. S8C) potentially due to lower ICOS expression in this tissue (Supplementary Fig. S1). We repeated similar experiment by dosing mice once with KY1044 mIgG2a. The tumors were then immunophenotyped up to 7 days posttreatment (Supplementary Fig. S8D). Treg depletion was observed after a single dose of 0.3 mg/kg; however, at this dose, the incidence of Tregs in the TME recovered quickly back to the level observed in the control group. A long-lasting Treg depletion and an increase of the CD8 to Treg ratio were observed at higher dose (3 and 10 mg/kg; Supplementary Fig. S8D). Altogether, these data suggest that KY1044 mIgG2a preferentially depleted ICOS^high Treg cells and improved CD8⁺ T_Eff to Treg cell ratio in the TME.

ICOS was expressed on cynomolgus monkey CD4⁺ memory T cells (T_M: CD3⁺/CD4⁺/CD95^high/CD28^−/Dim/+), CD4⁺ follicular Th cells (T_FH: CD3⁺/CD4⁺/CD185⁺) and on Treg cells (CD3⁺/CD4⁺/CD25⁺FoxP3⁺; Fig. 5D). Similarly, circulating NHP CD8⁺ T cells showed the lowest ICOS expression. We assessed the pharmacodynamic effects of KY1044 hIgG1 in NHP after repeated weekly intravenous administration at doses of 0, 10, 30, and 100 mg/kg for 4 weeks (5 doses). Exposure and full occupancy of ICOS on circulating CD4⁺ T cells was maintained at all doses (Supplementary Fig. S9A and S9B). Immunophenotyping indicated a decrease in T_M and T_FH cells in peripheral blood (Fig. 5E and F). KY1044 did not affect circulating Treg (Supplementary Fig. S9C), which represented less than 3% of circulating CD4⁺ cells in monkeys (Fig. 5D). KY1044 did not alter CD8⁺ T cells in blood (Supplementary Fig. S9D). Finally, we did not observe a significant decrease in ICOS^high cells such as T_M or Treg or in ICOS^low CD8⁺ T cells in the in the lymph node (or spleen) of treated monkeys (Supplementary Fig. S9E–S9G). In summary, these data demonstrated that high dose of KY1044 elicited some depletion of ICOS^high cells in peripheral blood but not in lymphoid tissues of NHPs.

To demonstrate that KY1044 activates and increases proinflammatory cytokine production in intratumoral T_Eff cells, we assessed the expression of the activation markers CD69 and CD44 on CD8⁺ T cells. T cells from CT26.WT tumors were analyzed 24 hours after a second dose of KY1044 mIgG2a. KY1044 mIgG2a induced a significant increase in CD69 expression (Fig. 6A; P < 0.05 vs. control) at doses of 1 and 3 mg/kg. Similarly, the analysis of CD69/CD44 double positive cells confirmed a significant increase (P < 0.05) of CD8 activation in response to KY1044 at a dose of 1 mg/kg (Fig. 6A). In a separate experiment, we examined KY1044-dependent induction of IFNγ and TNFα by CD4⁺ and CD8⁺ T cells. Using an intracellular staining approach on intratumoral CD4⁺ and CD8⁺ T cells collected 7 days posttreatment (3 mg/kg of KY1044 mIgG2a), we demonstrated a significant increase in cytokine production (Fig. 6B). Although, one cannot differentiate between direct activation (through direct ICOS costimulation on T_Eff cells) or indirect activation (via Treg depletion), this analysis of intratumoral T cells revealed that KY1044 was associated with some activation of both ICOS^Low CD8⁺ and CD4⁺ T_Eff cells.

The pharmacodynamic studies confirmed that KY1044 mIgG2a effectively depleted ICOS^high Treg and resulted in activation of ICOS^Low T_Eff cells (Fig. 4A). However, we noticed that when used at high doses, KY1044 mIgG2a also affected the numbers of intratumoral CD8⁺ T cells (Supplementary Fig. S8), resulting in a lower CD8⁺ T_Eff to Treg cell ratio. Because a higher baseline CD8⁺ T_Eff to Treg cell ratio positively correlates with a response to ICIs such as atezolizumab (49), we aimed to assess whether the antitumor efficacy would be superior at an intermediate dose. For this, we repeated the CT26.WT efficacy study using a range of different doses of KY1044 mIgG2a (20, 60, and 200 μg/dose) combined with a fixed dose of anti–PD-L1 (200 μg/dose). As shown in Fig. 7, all the combination treatments were associated with an improved response (vs. control or the anti–PD-L1 group). However, an improved response at 60 μg/dose of KY1044 mIgG2a was observed, with 90% of the mice being tumor free on day 54.

The approval of ICIs provides a novel approach that adds to existing cancer therapies. However, responses to ICIs are not universal (2). An immunosuppressive TME can underlie the lack of response. Notably, intratumoral Tregs maintain an immunosuppressive environment. Strategies reducing Treg numbers and improving the ratio of effector T cells to Treg cells have thus emerged. Blocking the recruitment, function, expansion, and/or the survival of Treg, may achieve this goal and a number of molecules, targeting CTLA-4, CD25, GITR, OX40, CCR8, CD137 and CCR2, are under investigation (50). However, the differential expression of these targets (i.e., within immune and other cell subtypes) is crucial to preferentially target T_Reg (49). Here, we demonstrated that ICOS expression differs between T-cell subsets, with the highest expression observed on Treg and confirmed that ICOS expression is higher in the TME than in the periphery (blood or spleen). In addition, we identified head and neck, gastric, esophageal, lung, bladder, skin, and cervical cancers as tumors with a high content of ICOS⁺ cells. However, costaining of ICOS with FOXP3 showed that these tumors differ by their incidence of ICOS⁺ Treg with cervical, esophageal, and head and neck cancers being more infiltrated by ICOS⁺ Treg. Because the ICOS expression on different cell subsets should affect the response to anti-ICOS therapies, this work suggests that the specific expression of ICOS on Treg may support a patient selection strategy.

High ICOS expression on intratumoral Tregs highlights the potential of this target for a depletion strategy. Using a transgenic mouse platform (23), we selected a fully human ICOS antibody called KY1044 and demonstrated that KY1044 depletes ICOS^high cells via ADCC (through the engagement of FcgRIIIa) and act as a costimulatory molecule on cells expressing lower ICOS levels, such as CD8⁺ T_Eff cells (through FcgR-dependent clustering). KY1044 has monotherapy efficacy in lymphoma models and combination efficacy (with anti–PD-L1) in models of solid tumors that are resistant to PD-1/PD-L1 blockade. In the CT26.WT tumor model, we showed better efficacy when combining anti–PD-L1 with KY1044 mIgG2a (effector enable) than with a poorly depleting mIgG1 format. Similarly, in this model, we showed better efficacy when combining KY1044 mIgG2a with anti–PD-L1 than with anti–PD-1. Although it was not clear why such a difference was observed in this particular model, one could postulate that anti–PD-1, which blocks binding to both PD-L1 and PD-L2, may be associated with a different phenotype to anti–PD-L1, which affects PD-1 and CD80 biology (51). Speculatively, there is also the possibility that anti-ICOS and anti–PD-1 may interfere with each other's functions within the immunologic synapse, as both ICOS and PD-1 are often expressed on the same cells (52–54).

There is an ongoing debate regarding the value of mouse models when predicting the depletion potential of effector function enabled antibodies in human (8, 9, 22, 55, 56). Experiments in mouse demonstrated complete depletion of intratumoral CTLA-4⁺ Treg, whereas depletion of these cells in patients treated with ipilimumab remains disputed. The differences in FcgR expression and Fc/FcgR interaction between rodents and primates may contribute to these discrepancies (57). Here, we performed pharmacodynamic studies in both mice (using mIgG2a) and NHPs (using hIgG1) to confirm that KY1044 decreased the frequency of ICOS^high cells in vivo in both species. With the mouse work, we demonstrated that KY1044 reduced Treg and improved the CD8⁺ T_Eff to Treg cell ratio, which led the tumors to respond to anti–PD-L1. Combined with the decrease in ICOS^high cells observed in NHP, these data suggested that KY1044 could also deplete ICOS^high cells in human tumors.

While depletion of Treg is an attractive strategy, it is crucial not to deplete all Treg in all tissues to avoid autoimmune diseases (58, 59). KY1044 did not deplete Treg in the spleen and lymph node. Although the reasons behind this selective depletion is not fully understood, this response has been shown for other depleting antibodies and, in our case, could be explained by the lower expression of ICOS on the surface of T cells in lymphoid tissues (Supplementary Fig. S1) or to lower expression of FcgRIIIa in the periphery. In the tumor rechallenge experiment, a long-term immune memory response was observed after treatment with KY1044. Because some memory T cells expressed ICOS, these data suggested that KY1044 has the potential to deplete ICOS^high cells without removing all ICOS cells. In cynomolgus monkeys, no adverse toxic effects were associated with the depletion of some of the ICOS^high cells. Finally, we demonstrated that KY1044 has costimulatory properties, as shown by the activation and secretion of cytokines such as IFNγ and TNFα. Although no effect on T-cell proliferation was observed, we noted a strong effect on cell morphology in response to KY1044-dependent ICOS stimulation. This costimulatory agonistic property was shown to require target clustering and concomitant engagement of the TCR.

Finally, while assessing the effect of KY1044 on the CD8⁺ T_Eff to Treg intratumoral ratio, we noticed that although the treatment improved the ratio at all doses tested, a bell-shaped response pattern was observed. Although all combinations of anti–PD-L1 with different doses of KY1044 mIgG2a were shown to trigger an antitumor immune response, the intermediate dose of 60 μg of KY1044 resulted in the strongest response.

Altogether, this study demonstrated that KY1044 was pharmacologically active, modified the intratumoral immune contexture, and induced a strong and long-lasting antitumor response. These findings, therefore, warrant the further assessment of KY1044 as a monotherapy or a combination therapy with anti–PD-L1 as a potential treatment for solid tumors.

R.C.A. Sainson reports other from Kymab Ltd (full-time employee) during the conduct of the study, as well as a patent for US9957323 issued (author/coinventor on the patent) and a patent for US16/323980 pending (author/coinventor on the patent). M. Kosmac reports a patent for US9957323 issued and a patent for US16/323980 pending. G. Borhis reports a patent for US9957323 issued and a patent for US16/323980 pending. N. Parveen reports a patent for US9957323 issued and a patent for US16/323980 pending. C. Van Krinks reports a patent for US9957323 issued and a patent for US16/323980 pending. H. Ali reports a patent for US9957323 issued to Kymab Ltd and a patent for US16/323980 pending to Kymab Ltd. H. Craig reports other from Kymab Ltd (full-time employee) during the conduct of the study. V. Germaschewski reports Kymab Ltd employment and share options. S. Quaratino reports other from Kymab Ltd (full-time employee) during the conduct of the study. M. McCourt reports personal fees from Kymab Ltd (full-time employee) during the conduct of the study, as well as a patent for US9957323B2 issued to Kymab Ltd and a patent for US20190330345A1 pending to Kymab Ltd. No potential conflicts of interest were disclosed by the other authors.

R.C.A. Sainson: Conceptualization, data curation, formal analysis, supervision, investigation, writing–original draft, writing–review and editing. A.K. Thotakura: Resources, data curation, writing–original draft. M. Kosmac: Resources, data curation, writing–original draft, writing–review and editing. G. Borhis: Resources, data curation, writing–original draft, writing–review and editing. N. Parveen: Resources, data curation. R. Kimber: Resources, data curation. J. Carvalho: Resources, data curation. S.J. Henderson: Resources, data curation. K.L. Pryke: Resources. T. Okell: Resources. S. O'Leary: Resources, data curation. S. Ball: Resources, data curation. C. Van Krinks: Data curation. L. Gamand: Resources, data curation. E. Taggart: Resources. E.J. Pring: Resources. H. Ali: Resources, data curation. H. Craig: Resources, data curation. V.W.Y. Wong: Resources, data curation. Q. Liang: Resources, supervision. R.J. Rowlands: Resources, data curation. M. Lecointre: Resources, data curation. J. Campbell: Resources, data curation, supervision. I. Kirby: Resources, data curation. D. Melvin: Software. V. Germaschewski: Resources, supervision. E. Oelmann: Writing–review and editing. S. Quaratino: Supervision, writing–review and editing. M. McCourt: Conceptualization, supervision, writing–review and editing.

The authors are extremely grateful to Kymab's molecular biology and antibody expression team as well as the Kymab BSU team. The authors also thank Propath UK and OracleBio for assistance with IHC staining and digital image analysis as well as Stephanie Grote-Wessels (Covance Preclinical Services GmbH) for their support with this study. They also thank the NCI Division of Cancer Treatment and Diagnosis, Developmental Therapeutics Program, Biological Testing Branch, Tumor Repository for providing the MC38 syngeneic tumor cell line.

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