Characterization of AB598, a CD39 Enzymatic Inhibitory Antibody for the Treatment of Solid Tumors

AB598 (1) is a novel monoclonal antibody under clinical development as a solid tumor immunotherapy targeting the enzymatic activity of cluster of differentiation 39 (CD39; gene: ENTPD1, also called NTPDase1 or ecto-apyrase). CD39 is widely expressed on the surface of nontumor cells in the tumor microenvironment (TME), such as on the infiltrating immune cells [macrophages, dendritic cells, regulatory T cells (Treg), NK cells, and exhausted T cells], endothelial vasculature, and stroma (2, 3). CD39 is an ectonucleoside triphosphate diphosphohydrolase (ENTPD) that catalyzes the sequential removal of the terminal gamma and beta phosphates of adenosine triphosphate (ATP), resulting in one adenosine monophosphate (AMP) and two inorganic phosphates as products. The enzymatic activity is carried out by the extracellular (EC) domain of CD39, which exists as a loop anchored into the membrane by transmembrane helices at both ends (4). Inhibiting CD39 results in an increase in its substrate, EC ATP (eATP), an alarmin known for its immunostimulatory effects on immune cells, concurrently with a decrease in its product, EC AMP (eAMP), which is easily converted to adenosine, a molecule inhibitory to immune cells (3). Therefore, inhibiting CD39 has the power to affect both the cells expressing CD39 and the EC environment of the TME, providing relief of passive immunosuppression through the decreased formation of AMP, a precursor of adenosine, simultaneously with natural agonism of immune function through an increase in eATP.

Although the breakdown of ATP to AMP by CD39 is a primary contributor to eAMP, CD38, ectonucleotide pyrophosphatase/phosphodiesterase 1 (ENPP1), other ENTPD and ENPP isoforms, and alkaline phosphatases all contribute to AMP formation (5). For this reason, targeting the AMP pool and adenosine signaling can be more effectively achieved through targeting the downstream proteins CD73, an ectoenzyme that converts AMP into adenosine, and the cell surface adenosine A2a and A2b receptors (A_2aR and A_2bR), a strategy that is being evaluated clinically. For example, quemliclustat, a CD73 small molecule inhibitor, and etrumadenant, an A_2aR/A_2bR small molecule dual receptor antagonist, are currently being evaluated in clinical trials in solid tumors. Therefore, preclinical work for AB598 has focused on the contribution of increased levels of eATP to induce immune stimulation as a therapeutic hypothesis.

ATP drives immune cell stimulation through the P2X and P2Y purinergic receptors. There are seven P2X receptors, which are all trimeric ATP gated–cation channels, and eight P2Y G protein–coupled receptors (GPCRs), which bind several di- and triphosphate nucleotides including ATP (6, 7). P2X and P2Y receptors have varied affinity for ATP, with half-maximal effective concentrations (EC₅₀s) ranging from 0.1-µmol/L ATP for P2Y2 to >100 µmol/L for P2X7 (6). In healthy tissues, the eATP concentration is low, 10 to 100 nmol/L (3), rendering the most sensitive P2 receptors constitutively active whereas most receptors remain inactive. Stressful intratumoral conditions such as cell death, hypoxia, and treatment with chemotherapy can increase intratumoral concentrations of eATP into the hundreds of micromolar range (8, 9). Stabilization of these increased eATP concentrations by CD39 inhibition results in the activation of the P2 receptors with the poorest affinity for ATP. P2X7 and P2Y11 are two of the least sensitive P2 receptors, with EC₅₀ > 100 and 17 µmol/L, respectively (6). Demonstrated here is that both are essential for the full immune activation of myeloid cells, illustrating the need to maintain a high eATP reservoir to enhance immune stimulation.

The inhibition of CD39 has the potential to further elevate eATP levels resulting in P2X7-dependent activation of inflammasomes in macrophages along with the maturation and release of interleukin (IL)-1β and IL-18 (10). The immunostimulatory properties of IL-18, in combination with IL-12 or IL-2, can drive the production of interferon-gamma (IFNγ) in lymphoid cells, a key cytokine for immune cell activation (11). Moreover, IL-18 can expand the population of stem-like exhausted precursor T cells, which can be targeted by immune checkpoint therapy to produce a more functional T cell phenotype (12). ATP signaling through P2Y11 has been shown to increase the maturation of monocyte-derived dendritic cells (moDCs; ref. 13), which can lead to an increased presentation of antigen by antigen-presenting cells (APC) in the TME, increasing T cell engagement and ultimately tumor-targeted cytolytic activity.

CD39 inhibition is particularly relevant to the TME, an area with naturally elevated eATP due to cellular stress and treatment with chemotherapies capable of inducing ATP release. The utility of eATP to promote an immunostimulatory TME to prompt host immune control of the tumor supports the development of AB598, a potent and specific CD39 inhibitor for the treatment of solid tumors, and the study of its molecular mechanism for immune activation. AB598 is currently being evaluated in a Phase 1/1b clinical trial (NCT05891171) in combination with chemotherapy and immunotherapy.

Extended methods can be found in the supplemental information.

Sprague Dawley rats were immunized with 30 to 200 mg of human CD39 antigen per animal, either protein or plasmid DNA. The adjuvant mixture included Adju-Phos, CpG-ODN, or Titer-Max. Animals were injected once every other week via footpad, subcutaneous, intraperitoneal, intramuscular, and intradermal routes. The serum titer was measured by ELISA or FACS. When the serum titer was sufficiently high, the animal was given a final boost with protein and cell membrane lysates in sterilized PBS without adjuvant. After 48 to 96 hours, the animals were euthanized, and lymph nodes and spleen B cells were used for cell fusion with Sp2/0 myeloma cells. The antibody was subsequently humanized to generate AB598, which has an IgG1 Fc-silent (FcS) heavy chain constant domain. The heavy chain and light chain sequences of AB598 were codon-optimized and synthesized. The sequence of AB598 is available in the pending patent application WO2023164872A1, in which AB598 is referenced as hu39.5_IGG1.AA, with SEQ ID NO: 66 for the mature heavy chain and SEQ ID NO: 67 for the mature light chain. The sequences can be found in the Sequence Listing Table in the pending patent application (1). The DNA fragments obtained were cloned into separate heavy chain and light chain expression vectors. After Sanger sequencing validation, the vectors were prepared for cell transfection, and CHO-K1 cells were transfected by electroporation. A transfected stable pool was selected using antibiotics and scaled up. After 2 weeks’ passage, 2 weeks of fed-batch protein production was performed at a 3 L scale. The purification protocol for AB598 consisted of harvested cell culture fluid clarification, followed by affinity capture using Mabselect SuRe resin (Cytiva) and cation exchange chromatography with POROS XS resin (Thermo Scientific). The protein was buffer exchanged into isotonic histidine buffer pH 6.0 and concentrated to 50 mg/mL via ultrafiltration and diafiltration, aliquoted and stored frozen at −80°C.

Further information on the generation (Examples 1 and 2) and characterization (Examples 3–7) of AB598 can be found in the patent application WO2023164872A1 (1).

CHO-K1 (ATCC, Cat. No. CCL-61, female, hamster), MOLP8 (DSMZ, Cat. No. ACC569, male, human), OAW42 (Sigma-Aldrich, Cat. No. 85073102, female, human), SK-MEL-5 (ATCC, Cat. No. HTB-70, female, human), A549 (ATCC, Cat. No. CCL-185, male, human), NCI-H526 (ATCC, Cat. No. CRL-5811, male, human), and SNU-1 (ATCC, Cat. No. CRL-5971, male, human) cell lines were grown for minimal passages (1–2 weeks) before banking and used before passage 20 upon defrost for experiments. All cell lines were tested for mycoplasma contamination using RT-PCR and all cell lines were authenticated with short tandem repeat, except CHO-K1 because short tandem repeat profiling for hamster cell lines was unavailable.

AB598 was evaluated by flow cytometry for binding to human CD39 expressed on the cell surface of peripheral blood mononuclear cells (PBMC). Previously purified, frozen PBMCs were defrosted, and treated with 0 or 400 µmol/L ATP for 30 minutes followed by AB598 or IgG1 FcS isotype control (clone MOPC-21, Absolute Antibody, Cat. No. Ab00178-10.3) for an additional 30 minutes. CD39 binding by AB598 was detected with 0.5 µg/mL of PE-labeled goat anti-human IgG. All incubation and staining steps were conducted at 4°C in Dulbecco’s PBS supplemented with 5% goat serum. The geometric mean of the fluorescence intensity of PE was obtained for the live single-cell population gated on either monocytes (CD14⁺) or B cells (CD19⁺).

PBMCs or B cells isolated from healthy blood donors, or MOLP8 multiple myeloma cells, were treated with a saturating concentration (67–100 nmol/L) of the treatment antibody. In the case of ATP addition, ATP was added 1 hour later. Assay plates were incubated for 0 to 72 hours, in the time course experiment, 36 hours in the ATP addition experiment, or 48 hours in the MOLP8 experiment before staining for cell surface CD39 with anti-CD39 APC clone A1 (Biolegend, Cat. No. 328210, RRID:AB_1953233), anti-CD39 AF647 clone A1 (BD, Cat. No. 567505, RRID:AB_2916629), anti-CD39 BV421 clone A1 (BD, Cat. No. 567524, RRID:AB_2916635), or anti-CD39 AF647-AB598 [conjugated with Alexa Fluor 647 Conjugation Kit (Fast)—Lightning-Link, Abcam, Cat. No. ab269823]. In the PBMC experiment, CD39 MFI was recorded on B cells gated on live, single, CD45⁺CD3⁻CD19⁺ cells. In the matched MOLP8 western blot experiment, CD39 was detected with primary anti-CD39 clone EPR20627 (Abcam, Cat. No. Ab223842, RRID:AB_2889212) added to a final concentration of 0.57 μg/mL and 1:1,000 secondary IRDye 800CW Goat anti-Rabbit IgG Secondary Antibody (LI-COR Biosciences, Cat. No. 926-32211, RRID:AB_621843) and imaged on a LI-COR Odyssey Fc Imager. Beta actin loading control DyLight 680 clone BA3R (Thermo Fisher Scientific, Cat. No. MA5-15739-D680, RRID:AB_2537665) was used for normalization. Analysis of band intensities was performed in Image Studio Lite Ver 5.2. The CD39 band intensity was normalized to the beta actin–band intensity and the AB598-treated samples were then compared with the matched isotype control–treated samples.

Whole blood from healthy human donors was analyzed to quantify changes in total versus unbound CD39 in both the EC and intracellular (IC) compartments after exposure to AB598. On the same day as collection, whole blood samples were treated with either AB598 or isotype control (100 nmol/L). Samples were maintained at 37°C for 30 minutes, 24 hours, or 48 hours then were placed on ice for subsequent staining steps. Negative control conditions for detecting EC CD39 were first blocked with excess unlabeled AB598 or A1 (Thermo Scientific, Cat. No. 14-0399-82, RRID:AB_837126) antibodies, then all samples were stained with a cocktail of antibodies to detect EC lineage markers including anti-CD3 PerCP-Cy5.5 clone SP34-2 (BD, Cat. No. 552852, RRID:AB_394493). Additional antibodies for CD39 were included in the EC staining cocktail for conditions intended for the detection of EC CD39, including negative control conditions: anti-CD39 PE clone A1 (Thermo Scientific, Cat. No. 12-0399-42, RRID:AB_1272091) and anti-CD39 AF647-AB598 (labeled using GlyCLICK Fluorophore conjugation kit, Genovis, Cat. No. L1-F03-200). After EC staining, samples were lysed, washed, permeabilized in 1× Fix/Perm Buffer (Thermo Scientific, Cat. No. 00-5523-00), and then blocked using 20% normal mouse serum diluted in permeabilization buffer (Thermo Scientific, Cat. No. 00-5523-00). For IC CD39 negative control conditions, excess unlabeled antibodies (AB598 or clone A1) were included in the blocking solution. All samples were then intracellularly stained with anti-FoxP3 eFluor450 clone PCH101 (Thermo Scientific, Cat. No. 48-4776-41, RRID:AB_1834365). Additional antibodies for CD39 were included in the IC staining cocktail for conditions intended for the detection of IC CD39, including negative control wells: anti-CD39 PE clone A1 (Thermo Scientific, Cat. No. 12-0399-42, RRID:AB_1272091) and anti-CD39 AF647 AB598. Samples were finally washed and resuspended for data acquisition. Data were collected using a FACSCanto flow cytometer.

Female 6- to 8-week-old CB17.SCID mice obtained from Charles River Laboratories were implanted with 5 million MOLP8 cells (human multiple myeloma) in PBS mixed 1:1 with Matrigel (Corning, Cat. No. 354234) subcutaneously on the right flank on day 0. For the pharmacodynamic studies, when tumors reached 300 to 500 mm³, mice were treated with a single 10 mg/kg i.v. dose of AB598.mIgG2a FcS, which contains identical complementarity-determining regions (CDR) to AB598 and a murine FcS constant domain or a matched FcS isotype control. Mice were sacrificed after 48 hours for the intratumoral enzymatic inhibition and intratumoral ATP assays and after 72 hours for the intratumoral flow cytometry assays. For the efficacy study, 10 mg/kg AB598.mIgG2a FcS or matched FcS isotype control was administered on day 0 and then twice per week. Mice were sacrificed when the tumor volume measurement exceeded 2,000 mm³, and the study concluded after several mice reached this endpoint. Experiments were performed at Arcus Biosciences in accordance with federal, state, and institutional guidelines and were approved by Arcus Biosciences’ Institutional Animal Care and Use Committee.

CD39 enzymatic inhibition was measured by ATP depletion with Kinase Glo Plus Assay (Promega, Cat. No. V3771) or AMP formation with AMP Glo Assay (Promega, Cat. No. V5012). The assays were run in the presence of 20 µmol/L ATP for low ATP conditions and 400 µmol/L ATP for high ATP conditions for the experiments on monocytes and cell lines. Intratumoral ATP was determined using supernatants from tumors digested in the presence of protease and phosphatase inhibitors and measured using the Kinase Glo Plus Assay. Extracellular ATP in the chemotherapy-treated cancer cell lines was measured using RealTime-Glo Extracellular ATP Assay (Promega, Cat. No. GA5010). Tumor enzymatic inhibition was determined using the Kinase Glo Plus Assay. To measure the amount of in vivo CD39 inhibition, tumors were treated ex vivo with both AB598 and isotype control. The percent of in vivo AB598.mIgG2a intratumoral ATPase inhibition is the fraction of residual ATPase activity in the ex vivo isotype control-treated sample compared with the ex vivo AB598-treated sample.

Male and female cynomolgus monkeys received weekly doses of 200 mg/kg AB598 i.v. for 3 weeks. Longitudinal blood samples were analyzed by flow cytometry to measure receptor occupancy (RO) and total levels of EC CD39. On peripheral immune cells, total CD39 was quantified by a noncompetitive CD39 antibody, anti-CD39 PE clone A1 (Thermo Fisher Scientific, Cat. No. 12-0399-042, RRID:AB_1272091), and unbound CD39 was quantified by fluorescently labeled AF647-AB598 [produced using Lightning-Link Alexa Fluor 647 (Fast) Conjugation kit, Abcam, Cat. No. ab269823]. Tissue samples were also collected at necropsy 24 hours after the third dose. In tissues, total CD39 protein was detected in FFPE-sections by immunohistochemistry (IHC) using a noncompetitive CD39 antibody, clone EPR20461 (Abcam, Cat. No. ab223828, RRID:AB_2889212); unbound CD39 was detected using AB598.mIgG2a FcS, which competes for the same epitope as AB598. Enzymatic activity was measured by enzyme histochemistry to detect the generation of free phosphate. Frozen tissue sections were treated with 0.25 mmol/L ATP substrate and 2.5 mmol/L levamisole to block nonspecific phosphatase activity. The addition of 2-mmol/L lead nitrate resulted in lead–phosphate deposition observed by brown coloring. Whole slide scanning was performed on the Pannoramic Midi II brightfield scanner (3D Histech). Positive signal was quantified using an area quantification algorithm on HALO Digital Imaging software (Indica Labs). All procedures involving the care and use of animals were performed in accordance with federal, state, and institutional guidelines and were approved by the Testing Facility’s Institutional Animal Care and Use Committee.

CD14⁺ cells were isolated from healthy human peripheral blood by positive selection, except in the case of CRISPR experiments, in which cells were isolated by negative selection. After isolation, cells were differentiated into moDCs or M0 macrophages with 100 ng/mL recombinant human (rh)GM-CSF (R&D Systems, Cat. No. 215-GM) and 100 ng/mL rhIL4 (Peprotech, Cat. No. 200-04) or 50 ng/mL rhM-CSF (R&D Systems, Cat. No. 216-GM) for 6 days with refreshment of medium and growth factors on day 3. The CRISPR knockout (KO) protocol used in this study (Lonza buffer P3, program DK-100) was modified from earlier publications (14, 15) and two guide conditions containing single or combinations of 2 to 3 guides were tested per gene (Supplementary Table S1). In macrophage assays, on day 6, the cells were treated with 1 ng/mL lipopolysaccharide and 100 nmol/L of AB598 or isotype control antibodies and incubated for 3 hours. The cells were then treated with 500 µmol/L ATP and incubated for an additional 4 hours. Following incubation, supernatants were tested for IL-1β by ELISA (R&D Systems, Cat. No. QK201) or CBA (BD, Cat. No. 561509) and IL-18 by ELISA (Fisher Scientific, Cat. No. DL180). In moDC assays, cells were treated with AB598 or isotype control antibodies for 1 hour, followed by 300 µmol/L ATP. After 16 to 18 hours, CD83 and CD86 were measured by flow cytometry.

Statistical significance is indicated as ^∗, P ≤ 0.05; ^∗∗, P ≤ 0.01; ^∗∗∗, P ≤ 0.001; ^∗∗∗∗, P ≤ 0.0001; ns, nonsignificant. Unless otherwise noted, the bar heights or points represent the mean, and the error bars represent the standard deviation. Flow cytometry results were analyzed using FlowJo Software (BD). Curve fits and statistics were calculated using GraphPad Prism.

The data generated in this study are available within the article and its supplementary file. Raw data depicted in plots are available from the corresponding author upon reasonable request.

AB598 is a humanized FcS anti-CD39 antibody generated from a rat immunization and screening campaign. Humanization of selected rat antibodies was performed using the CDR grafting technique (16) where variable heavy and variable light chain sequences of the rat antibodies were used to identify the closest human germlines for each chain to receive the rat CDR peptides, with a minimum of rat sequence back mutations. Human acceptors for the variable heavy and variable heavy frameworks were searched for within the GenBank database (17). Humanized antibodies were produced and tested, and the AB598 variable domain sequences were selected based on their ability to bind and inhibit human CD39 specifically and potently, and not any other human CD39 family members, CD39L1, CD39L2, CD39L3, or CD39L4 (Supplementary Fig. S1). Species profiling of AB598 showed that AB598 binds human and cynomolgus monkey CD39 but not mouse or rat CD39 (Fig. 1A).

Surface plasmon resonance showed that AB598 binds potently to rhCD39, with an equilibrium dissociation constant, K_D = 2.4 (±0.3) × 10⁻¹⁰ mol/L, with a fast on-rate, k_a = 3.0 (±0.3) × 10⁵ M⁻¹ s⁻¹ and a slow off-rate, k_d = 7.2 (±0.4) × 10⁻⁵ s⁻¹ resulting in a long residence time and high subnanomolar affinity to CD39 (Supplementary Fig. S2; Supplementary Table S2). AB598 potently binds cell surface CD39, including CD39 on primary human monocytes and B cells, with a mean EC₅₀ across 6 to 8 donors of 0.2 to 0.3 nmol/L (Fig. 1B and C; Supplementary Table S3). Cell surface binding was unaffected in the presence of 400 µmol/L ATP. Similarly, AB598 potently inhibited cell surface CD39 enzymatic activity with no loss of potency due to high ATP concentration across six human cell donors, with IC₅₀ = 0.5 ± 0.1 nmol/L at 20-µmol/L ATP and 0.2 ± 0.1 nmol/L at 400 µmol/L ATP (Fig. 1D; Supplementary Table S4), indicating that AB598 treatment is suitable for the high eATP conditions expected in the TME. In the monocyte enzymatic inhibition assay (Fig. 1D), which measures AMP formation after substrate ATP addition, AB598 inhibited approximately 75% of AMP formation compared with untreated cells. AB598-mediated CD39 enzymatic inhibition was also measured by ATP depletion on CD39-expressing cancer cell lines (Fig. 1E). AB598 fully inhibited the EC ATPase activity of MOLP8 and OAW42 cells; however, in SK-MEL-5, approximately 25% of the EC ATPase activity remained after adding saturating amounts of AB598. Given that the amount of ATPase activity post-AB598 addition fluctuated with cell type, it was suspected that the uninhibited ATPase activity could be due to the expression of additional ATP-degrading or AMP-generating enzymes, and the CD39-specific ATPase activity was fully inhibited. In SK-MEL-5 cells, AB598 inhibited ATPase activity to levels of ATPase activity seen in CD39 CRISPR KO SK-MEL-5 cells (Fig. 1F; Supplementary Fig. S3), indicating that the residual AMP generation in the monocyte and SK-MEL-5 cell enzymatic activity assays (Fig. 1D–F) is not due to CD39 and that AB598 completely inhibits CD39 enzymatic activity.

Next, AB598 was analyzed for its activity in additional cell-based systems. A decrease in cell surface CD39 was observed when incubating B cells with AB598 for extended periods of time, reaching a maximal effect in vitro at 48 hours (Fig. 2A). The same pattern of CD39 cell surface decrease was observed using B cells isolated from PBMCs with or without ATP and in the presence of a commercially available, noninhibitory CD39 antibody, clone A1 (Fig. 2B; Supplementary Figs. S4 and S5). From these experiments, it was clear that the observed cell surface decrease of CD39 was not AB598-specific, nor was it a result of CD39 enzymatic inhibition or elevated eATP. The lack of detection of CD39 was not due to the occlusion of a specific epitope by CD39 clustering because AF647-A1 was used to detect CD39 in cells treated with AB598, and AF647-AB598 was used to detect CD39 in cells treated with A1. The relatively long timeline to observe the loss of cell surface CD39 suggested that ligand-induced receptor internalization may not be fully responsible for this effect. MOLP8 cells treated with AB598 or isotype control for 48 hours showed that cell surface and total CD39 were both decreased with AB598 treatment, but no change in ENTPD1 mRNA level was observed with AB598 exposure compared with isotype control treatment (Fig. 2C).

To better understand CD39 dynamics beyond the cell surface, whole blood from three healthy human donors was treated ex vivo with AB598 or isotype control, then simultaneously assessed for total and unbound levels of EC and IC CD39 after antibody exposure between 30 minutes and 48 hours at 37°C (Fig. 2D). Total CD39 was measured by flow cytometry using the commercial noncompetitive anti-CD39 antibody (PE-A1) and unbound CD39 was measured using labeled AF647-AB598 as a competitive reagent to AB598.

Analysis of viable immune cell subsets within ex vivo–dosed whole blood revealed cell type–specific differences in EC and IC CD39. Among these populations, bulk T cells did not express appreciable levels of either EC or IC CD39, and no changes were observed after AB598 exposure (Supplementary Fig. S6). As with isolated B cells and B cells within PBMC samples (Fig. 2A and B), progressive loss of EC CD39 was observed on whole blood B cells (CD19⁺), which occurred between 30 minutes and 48 hours post-AB598 exposure. Complete EC target engagement was maintained throughout this timeframe as evidenced by a lack of signal from competitive AF647-AB598 on the AB598-treated cells (Supplementary Fig. S6). CD3⁺FoxP3⁺ cells (Tregs) expressed high levels of EC CD39 which decreased significantly following AB598 exposure in correspondence with EC target engagement; however, unlike in B cells, Treg expressed high baseline levels of IC CD39, which enabled observation of changes in IC target engagement. Although internalization of EC CD39 is not apparent based on increases in IC CD39 following AB598 exposure within any cell type, complete loss of unbound IC CD39 was observed in AB598-treated Treg by 24 hours with complete IC target engagement maintained at 48 hours (Fig. 2D; Supplementary Fig. S6). These results indicate that AB598 does enter these cells, presumably by binding and internalization of the EC CD39:AB598 complex. However, this experiment does not allow for the comparison of the rate of CD39:AB598 internalization to the basal rate of CD39 internalization. Given that little or no increases in total IC CD39 are observed during the timeframe when EC CD39 is decreasing, it is likely that internalized CD39 either recycles to the cell surface and is lost by some other mechanism or is degraded intracellularly. To test whether CD39 was being internalized and then degraded, the MOLP8 western blot experiment in Fig. 2C was repeated in the presence of chloroquine, which prevents lysosome acidification, and MLN4924, which inhibits proteasomal degradation by preventing neddylation. Neither of these inhibitors resulted in an accumulation of CD39 suggesting degradation is not the main pathway for CD39 loss with AB598 treatment (Supplementary Fig. S7).

To explore if these cell surface CD39 changes were also observed in vivo, AB598 was studied in a xenograft model. AB598 is highly specific for human and cynomolgus monkey CD39 but does not bind mouse or rat CD39 (Fig. 1A). Therefore, immunocompromised SCID mice were inoculated with human multiple myeloma MOLP8 cancer cells which endogenously express human CD39. Treatment with a murinized version of AB598, AB598.mIGg2a FcS, resulted in full enzymatic inhibition of intratumoral CD39 (Fig. 3A) and an increase in intratumoral ATP (Fig. 3B) by 48 hours, indicating that by inhibiting the breakdown of ATP, AB598 can increase ATP levels in the TME. The mixed species model, which has human MOLP8 cells and mouse immune cells, provided further confirmation that the decrease in cell surface CD39 was AB598-mediated rather than due to other cellular changes resulting from CD39 inhibition such as increased ATP or decreased AMP. Cell surface CD39 was decreased on the human MOLP8 cells of AB598-treated mice, to which AB598 could bind (Fig. 3C), but not on the murine immune cells, to which AB598 could not bind (Fig. 3D), indicating that the direct interaction of AB598 with CD39 is the cause of the CD39 decrease on the cell surface.

CD39 inhibition is designed to promote tumor control through the stabilization of ATP to stimulate immune cells. There is no direct cytotoxic effect of ATP on the CD39-expressing MOLP8 multiple myeloma cell line used for xenograft studies (Supplementary Fig. S8). The SCID mouse model retains some of the critical components of the tumor immunity cycle (phagocytes and NK cells) but lacks others (mature B and T cells). Therefore, a prophylactic efficacy study was performed as these mice lack a fully functional immune system. Mice were dosed with 10 mg/kg AB598.mIgG2a FcS starting on the same day as tumor inoculation, day 0, and then biweekly starting at day 7. A drastic reduction in tumor growth in mice treated with AB598.mIgG2a FcS was observed, further supporting the model that AB598.mIgG2a FcS inhibition of intratumoral CD39 results in increased intratumoral eATP and enhanced tumor control.

AB598 binds to cynomolgus monkey CD39 (Fig. 1A), which provided an opportunity to evaluate peripheral and tissue RO, changes in cell surface CD39, and the enzymatic inhibition of CD39 in tissues in a higher species. AB598 was well tolerated in monkeys with no overt toxicities. Monkeys dosed with AB598 achieved complete RO on circulating monocytes within 48 hours of the first dose which was maintained for the duration of the study with weekly AB598 dosing (Fig. 4A and B). Cell surface CD39 was decreased across multiple immune cell types by 48 hours and reached the maximum observed decrease by day 8. Loss of EC CD39 was particularly prominent on the surface of B cells (CD19⁺CD14⁻ cells) and other CD39-expressing lineage negative lymphocytes (CD19⁻CD14⁻ cells, likely representing T cells and NK cells), and occurred to a lesser extent on myeloid and granulocytic cell subsets.

CD39 expression in the spleens of both vehicle and AB598-dosed animals was localized to the red pulp and to a lesser degree the white pulp. IHC on esophagi showed CD39 staining in the smooth muscle of the muscularis mucosa and externa, immune cells and vasculature in the submucosa and lamina propria, and mucosal intraepithelial immune cells (Fig. 4D). In the spleen and esophagus of AB598-dosed animals, CD39 expression was decreased on the immune cells including the lamina propria of the esophagus but maintained on the vascular endothelium in both the spleen and esophagus and the muscularis mucosa of the esophagus. IHC to detect unbound CD39 showed a staining pattern consistent with the total CD39 IHC in vehicle-dosed tissues and no staining in AB598-dosed tissues, indicating full target engagement in the AB598-dosed tissues (Fig. 4D). Using enzyme histochemistry to measure ATPase activity, tissues from vehicle-dosed animals showed high levels of lead–phosphate deposition matching the expression pattern of CD39, whereas AB598-dosed animals showed lead–phosphate signal reduced to non-CD39 dependent levels, determined by high ex vivo dosing of AB598 (Fig. 4D; Supplementary Fig. S9), indicating full target inhibition. Notably, despite the preservation of CD39 on vascular endothelium compared with the immune cells after AB598 treatment, the CD39 enzymatic activity was still blocked on the vascular endothelium, highlighting the importance of using a CD39 inhibitory antibody in a therapeutic context rather than a CD39 binding antibody which only decreases cell surface CD39 on specific cell populations.

Next, the immunostimulatory effect of ATP on primary human myeloid cells was characterized using in vitro-derived moDCs and macrophages treated with exogenously added ATP. CD83, a maturation marker, and CD86, a costimulatory molecule, were both increased on moDCs treated with 300 µmol/L ATP for 18 hours (Fig. 5A). The increase was more substantial in the presence of AB598 (Fig. 5A), indicating the ability of AB598 to preserve ATP to alter the moDC phenotype toward a more mature and immunostimulatory profile. In vitro-derived macrophages were treated with AB598 or isotype control for 1 hour, followed by 1 ng/mL lipopolysaccharide for 3 hours, and then 500 µmol/L ATP for 4 hours, before secreted IL-1β and IL-18 were measured in the supernatant (Fig. 5B). IL-1β and IL-18 were not detected in the supernatants of cells treated with isotype control or AB598, but no ATP. ATP-driven IL-1β and IL-18 secretion is suggestive of inflammasome activation, which was confirmed by monitoring caspase-1 activation 30 minutes after ATP addition (Supplementary Fig. S10). Activation of the inflammasome, and production of IL-18 in particular, has been shown to be advantageous to producing a robust antitumor response (12, 18).

In addition to preserving immunostimulatory ATP, inhibition of CD39 can reduce the generation of immunosuppressive adenosine by decreasing the availability of AMP to be converted into adenosine by CD73. To specifically understand the contribution of increased ATP, relative to decreased adenosine in the immunostimulatory phenotype observed in ATP + AB598-treated samples (Fig. 5A and B) the dependency of AB598 treatment on ATP signaling mediated through purinergic receptors was explored. P2X7, P2Y2, and P2Y11 purinergic receptors were selected to profile for ATP/AB598-driven responses due to (i) their expression on these cell types, (ii) a validated response to ATP, and (iii) literature supporting the activation of macrophages or DCs through these receptors (6, 13). Additionally, receptors with relatively poor affinity for ATP were prioritized, as CD39 inhibition would have the greatest impact in activating these receptors. P2X7 has an ATP EC₅₀ > 100 µmol/L and P2Y11 an ATP EC₅₀ = 17 µmol/L (6). The P2Y2 ATP EC₅₀ is relatively potent at 0.1 µmol/L, but P2Y2 was included for its known role in DC chemotaxis. On both moDCs and macrophages, the exogenous addition of ATP, preserved by AB598, increased the activation of these cells (Fig. 5A and B). In moDCs, the increases in CD83 and CD86 were reduced to baseline with KO of P2Y11 using multiple CRISPR guides (Fig. 5C; Supplementary Fig. S11), and there was no effect of AB598 on cells without ATP treatment (Fig. 5A), indicating that AB598 is inducing moDC activation through ATP-mediated signaling via the P2Y11 receptor. In in vitro-derived macrophages, the increases in IL-1β and IL-18 when AB598 was combined with ATP was substantially reduced with KO of P2X7 using multiple CRISPR guides (Fig. 5D; Supplementary Fig. S11). IL-1β and IL-18 levels were not measurable in the absence of ATP addition. Together, these observations indicate that AB598 is inducing macrophage inflammasome activation through ATP-mediated signaling via the P2X7 receptor.

Finally, the ability of CD39 inhibition to elevate eATP levels resulting from chemotherapy-induced cell death to enhance myeloid cell activation, supporting the rationale behind pursuing CD39 inhibition in combination with chemotherapy for the treatment of solid tumors, was tested. The intratumoral environment naturally produces larger amounts of eATP compared with healthy tissues given the various cellular stresses of the TME, and CD39 inhibition resulted in increased ATP in MOLP8 tumors (Fig. 3B). Intratumoral eATP levels can be further enhanced by stimulating ATP release upon chemotherapy-induced cell death. Treatment of A549 and H526 lung cancer cell lines with carboplatin or the SNU1 stomach cancer cell line with oxaliplatin resulted in ATP release from the tumor cells (Fig. 6A–C). Furthermore, the combination of chemotherapy and AB598 on CD39-expressing SK-MEL-5 cells demonstrated that CD39 inhibition can prevent the breakdown of eATP, leading to overall increased eATP levels (Fig. 6D).

A co-culture model was established to determine if chemotherapy can induce ATP release from tumor cells to enhance APC activation (Fig. 6E–H). In this model, no exogenous ATP is added, instead, eATP is generated upon docetaxel-induced cell death of A549 cells (Fig. 6F). AB598 has minimal or no effect in conditions lacking either docetaxel or A549 cells (Fig. 6G and H), because these conditions do not have appreciable amounts of eATP present. When moDCs are combined with both A549 cancer cells and docetaxel treatment, AB598 preserves eATP to amplify moDC activation as seen by increases in CD83 and CD86 (Fig. 6G and H). This in vitro co-culture system demonstrates the therapeutic hypothesis of CD39 inhibition in both a human system and a treatment-relevant scenario in which chemotherapy-induced ATP release from cancer cells, stabilized by CD39 inhibition, results in increased moDC activation.

CD39 is an emerging immuno-oncology target, contributing to immunosuppression in the TME by depleting the system of immunostimulatory eATP and generating eAMP to fuel the immunoinhibitory adenosine pool. A comprehensive characterization of the CD39 inhibitory antibody AB598 revealed that AB598 is an exquisitely potent antibody with subnanomolar affinity for recombinant and plasma membrane CD39. At 2.4 (±0.3) × 10⁻¹⁰ mol/L, AB598 has the most potent K_D of all published clinical anti-CD39 antibodies (19–21). Direct comparison of maximal CD39 enzymatic inhibition in vitro, using cell-based systems, and in vivo across the CD39-inhibitory clinical molecules is difficult due to the different model systems used by different groups and the varied expression levels of additional ATP-degrading and AMP-generative enzymes such as TNAP, CD38, and ENPP1 on different cell types. However, an isogenic CD39 wild-type/KO pair demonstrated that AB598 can inhibit 100% of CD39 enzymatic activity in a cell-based system. When delivered systemically, the murinized version of AB598 inhibits 100% of the CD39 enzymatic activity in MOLP8 xenograft tumors and this study is the first to show an increase in intratumoral ATP levels in mice treated systemically with a CD39 inhibitory antibody. AB598 delivered systemically in cynomolgus monkeys shows complete inhibition of CD39 in the tissues evaluated and complete RO on peripheral monocytes, an immune cell type with high CD39 expression in both human and monkey whole blood. Full RO on monocytes is achieved with the first dose of AB598 and sustained for the duration of the study (Fig. 4B).

Most of the clinical-stage anti-CD39 antibodies inhibit the enzymatic activity of CD39 and are designed to lack or attenuate Fc effector function, highlighting the emphasis on enzymatic inhibition, as opposed to depletion, as a key requirement for antitumor immunity. This study, and others, have reported a decrease in cell surface CD39 both in vitro and in vivo, with anti-CD39 antibody treatment. However, there is little information in the public domain addressing this observation. Warren and colleagues reported a dose-dependent decrease of cell surface CD39 in peripheral B cells from patients in their SRF617 clinical trial, in PBMCs and splenocytes from SRF617-dosed human CD39 knock-in mice, and on PBMCs from human whole blood treated ex vivo with SRF617 (19). In a xenograft mouse model using autologous human EBV-specific T cells against lymphoblastoid cell lines inoculated into NRG mice, treatment with another clinical-stage anti-CD39 antibody, TTX-030, resulted in a decrease of cell surface CD39 on the tumor-infiltrating human CD8⁺ T cells (22), with a separate study showing that TTX-030 is not internalized on SK-MEL-28 cells (21).

For the first time, this study demonstrates that the binding of CD39 by both AB598 and a noninhibitory antibody can trigger the loss of CD39 from the surface of immune cells, indicating that this phenomenon is not a feature specific to AB598 binding or CD39 inhibition. The timeframe to observe the effect seems comparable both in vitro and in vivo with a similar degree of CD39 lost within the first 48 hours after exposure to anti-CD39 antibody, although the full effect is typically seen at timeframes greater than those monitored using in vitro assays (roughly 168 hours in vivo in cynomolgus monkeys). One critical observation was that the cell surface decrease seen on immune cells is greater than the effect observed on nonimmune cells, as CD39 loss was not observed on the vasculature of the AB598-dosed cynomolgus monkey tissues (Fig. 4D). Future studies in patients dosed with AB598 will be needed to confirm these findings in a clinical setting, yet these observations reinforce the need for the development of a CD39 enzymatic inhibitory antibody, as opposed to a noninhibitory antibody capable of inducing a decrease in cell surface CD39. Inhibiting CD39 widely in the TME, and not solely on the immune cells, is important for maintaining a high intratumoral eATP concentration capable of stimulating intratumoral myeloid cells.

Recent work in the field of CD39 inhibition for the application of immuno-oncology implicated ATP-driven myeloid cell activation, and particularly inflammasome activation resulting in IL-18 production, as a key driver of antitumor immunity (22, 23). Li and colleagues and Yan and colleagues utilized genetic in vivo murine models to reach this conclusion. The current study complements and extends the understanding of CD39 inhibition as a driver of antitumor immunity via the stabilization of eATP, instead of as a reversal of adenosine-mediated immunosuppression. To understand the clinical relevance of the previous findings more fully, the current study applies these principles to primary human cell systems. Using genetic manipulation of primary human in vitro-differentiated macrophages, it is demonstrated that CD39 inhibition in the presence of ATP works through the ATP gated–P2X7 cation channel to induce inflammasome activation resulting in IL-1β and IL-18 secretion, confirming the previous murine findings in a primary human cell system. In moDCs, ATP activation, enhanced by the stabilization of ATP with AB598, results from signaling through the P2Y11 receptor. The KO studies demonstrate that the full phenotype of AB598 and ATP-induced myeloid cell activation can be reversed with interruption of signaling through these purinergic receptors, confirming for the first time in a primary human genetic P2 receptor KO system that increased ATP, not decreased adenosine, is directly responsible for the increased myeloid cell activation phenotype observed with CD39 inhibition in the presence of ATP.

Given that myeloid cells can be more fully activated through the addition of ATP, this study makes a compelling case for combining AB598 with chemotherapy in the clinic. Common chemotherapeutic agents such as carboplatin, oxaliplatin, and docetaxel can induce ATP release, and CD39 inhibition by AB598 can stabilize high levels of eATP. Although several previous studies and reviews have suggested that the combination of chemotherapy and CD39 inhibition bolsters eATP pools, this study is the first to directly demonstrate that tumor cells, treated with a CD39 inhibitory antibody and chemotherapy accumulate more eATP in the medium than cells treated with chemotherapy alone. In a co-culture system in which docetaxel induces ATP release from A549 lung cancer cells, eATP is stabilized by AB598 and induces an immunostimulatory phenotype on moDCs, demonstrating that the therapeutic hypothesis is feasible in a human setting. Taken in the context of the TME, increased APC activation can lead to increased APC:T cell interaction and increased tumor control resulting from increased T cell cytolytic activity and an increased capacity for an adaptive anti-tumor immune response.

Taken together, these findings support the development of AB598 for the treatment of solid tumors in combination with chemotherapy in the clinic. A Phase 1/1b clinical trial of AB598 in combination with chemotherapy and anti-PD-1 (NCT05891171) is currently underway.

All authors are current or former employees of Arcus Biosciences with stock equity. C.E. Bowman, A. Chen, X. Zhao, E. Fernandez-Salas, and N. Walker report a patent for WO2023164872A1 pending to Arcus Biosciences.

A.E. Anderson: Conceptualization, data curation, formal analysis, supervision, investigation, visualization, methodology, writing—original draft, writing—review and editing. K. Parashar: Formal analysis, investigation, methodology, writing—original draft. K. Jin: Conceptualization, formal analysis, investigation, methodology, writing—original draft. J. Clor: Data curation, formal analysis, validation, visualization, methodology, writing—original draft. C.E. Stagnaro: Investigation, visualization, methodology. U. Vani: Investigation, methodology. J. Singh: Conceptualization, formal analysis, supervision, investigation, methodology. A. Chen: Conceptualization, formal analysis, investigation, methodology. Y. Guan: Conceptualization, formal analysis, validation, investigation, methodology, writing—original draft. P. Talukdar: Formal analysis, investigation, methodology, writing—original draft. P. Sathishkumar: Investigation, methodology. D.J. Juat: Investigation, methodology. H. Singh: Formal analysis, investigation. R. Kushwaha: Conceptualization, formal analysis, supervision, validation, investigation, writing—original draft. X. Zhao: Conceptualization, formal analysis, supervision, methodology. A. Kaplan: Conceptualization, formal analysis, supervision, validation, investigation, visualization, methodology, writing—original draft. L. Seitz: Conceptualization, data curation, supervision, writing—review and editing. M.J. Walters: Conceptualization, data curation, supervision, writing—review and editing. E. Fernandez-Salas: Conceptualization, data curation, supervision, writing—review and editing. N.P.C. Walker: Conceptualization, data curation, supervision, writing—original draft, writing—review and editing. C.E. Bowman: Conceptualization, data curation, formal analysis, supervision, investigation, visualization, methodology, writing—original draft, writing—review and editing.

We would like to thank Gonzalo Barajas for conducting the MOLP8 xenograft studies and Janine Kline for her oversight and enablement of the MOLP8 studies. We would also like to thank Haley Yingling for assisting with the flow cytometry confirmation of the SK-MEL-5 CD39 knockout cells and Emily Zhou for assisting with the species selectivity experiment. We would like to thank the scientists at WuXi Biologics for their contributions reflected in WO2023164872A1. The research described in this manuscript was fully funded by Arcus Biosciences, Inc., a publicly traded biotechnology company.