Sym024 interacts with a unique epitope on the CD73 homodimer, favoring effective bivalent binding to improve anti-PD1 therapy

Abstract

Purpose

Adenosine signaling may be a central immune suppressive mechanism in several cancers, and blockade of the rate-limiting CD73 AMP-to-adenosine enzyme has been demonstrated to improve clinical efficacy of PD(L)-1 immune therapy. However, deep inhibition of CD73 activity could prove difficult in tumor environments with a constant AMP supply and high CD73 levels. Here, we sought to identify, characterize, and benchmark a novel antagonistic anti-CD73 antibody, Sym024 (S95024), and to structurally decode its mode of action.

Experimental Design

Sym024, selected via functional antibody repertoire screening, was tested against benchmark anti-CD73 antibodies in primary cell, cell line in vitro binding, CD73 enzymatic activity, and T cell activation assays. Its in vivo tumor growth inhibition was examined in transplanted human or mouse tumors in immunocompetent or immunodeficient mice, and intra-tumoral enzymatic inhibition and immune cell recruitment were assessed. We investigated Sym024-CD73 interaction using surface plasmon resonance, cryo-electron microscopy, site-directed mutagenesis, and population level complex formation through size-exclusion-chromatography with light scatter mass detection. Preclinical safety and pharmacokinetics were assessed in monkeys.

Results

Sym024 effectively blocked CD73 across a large range of enzyme expression levels, comparing favorably to benchmark anti-CD73 antibodies; it improved the efficacy of PD-1 blockade in vitro as well as in vivo. Our structural data indicate that a unique one-to-one Sym024-CD73 interaction engenders this comprehensive inhibition. No pre-clinical safety flags were observed, and the pharmacokinetics profile of Sym024 supported a standard clinical dosing regimen.

Conclusions

The comprehensive CD73 inhibition exhibited by Sym024 may improve the efficacy of anti-PD(L)-1/anti-CD73 combination treatment.

Statement of Translational Relevance

Clinical proof-of-concept of combining CD73 inhibition with PD(L)-1 blockade has been demonstrated in non-small-cell lung cancer. The tumor environment, however, may contain very high CD73 levels and provides a steady supply of AMP substrate for CD73-mediated conversion to adenosine, suggesting that a comprehensive abrogation of this immune-suppressive signal pathway could be challenging, limiting treatment impact. Here, we show that Sym024 is a highly efficacious inhibitory antibody with a unique target binding mode of action. Sym024 inhibits adenosine signaling even in tumor environments with high CD73 activity and may thus considerably expand the patient population benefiting from anti-PD(L)-1/anti-CD73 combination treatment.

Introduction

Cancer cells are surrounded by a complex tumor microenvironment (TME) that influences each step of tumorigenesis (1). A key feature of this environment is a strong immune suppression exerted via various signaling pathways (1,2).

Cytotoxic T lymphocyte antigen 4 (CTLA-4) and programmed cell death protein 1 (PD-1) signaling pathways exert their effect at several levels to block T effector cell activation and impede the cancer immunity cycle (3). While anti-CTLA4, anti-PD-1, and anti-PD-1 ligand 1 (PD-L1) therapies can improve outcomes in several cancers (4), many patients remain unresponsive to these treatments, potentially due to additional TME immune suppressive mechanisms.

Adenosine has recently garnered interest as a key immune suppressive agent (5-9). Four distinct G-protein-coupled receptors, A1, A2A, A2B, and A3, are adenosine-specific. Immune cells predominantly express the relatively high-affinity A2A (K_D ~310 nM) and low-affinity A2B (K_D ~15 mM), with the latter generally limited to myeloid cells (5). Signaling through these receptors can dampen T, B, and NK cell activation, maturation, proliferation, and/or cytokine production, promote regulatory T cell differentiation, and induce an immune suppressive phenotype in myeloid cells, resulting in the secretion of an array of immunosuppressive cytokines (5,10).

In the TME, adenosine levels rely on constant replenishment, with hydrolysis of pro-inflammatory adenosine triphosphate (ATP) released from dying or damaged tumor cells constituting a major source of adenosine. The extracellular ectonucleotidases CD39 and CD73 catalyze the conversion of ATP to adenosine monophosphate (AMP), and AMP to adenosine, respectively. Together, they constitute a common immunosuppressive pathway, with CD73 considered the rate-limiting enzyme (5,9). Compatible with the notion of CD73 implication in cancerogenesis, over-expression of this enzyme correlates with poor prognosis in various cancers (5,8,11-13).

Several lines of evidence suggest that CD73 inhibition in combination with other immune-modulating agents may be an effective approach to immune-oncology therapy. Firstly, in CD73-deficient mice, tumor growth is restricted (14-17), and abrogation of adenosine signaling has been demonstrated to substantially augment PD(L)-1 blockade effects in numerous mouse cancer models (18-22). Secondly, two clinical trials recently showed that combination therapy with CD73 inhibition is superior to PD(L)-1 blockade alone for immune-oncology treatment of non-small-cell lung cancer (NSCLC) (23,24).

Several monoclonal antibodies (mAbs) that inhibit the enzymatic activity of CD73 are currently in clinical development, prominent among which are oleclumab (AstraZeneca), mupadolimab (Corvus Pharmaceuticals), IPH5301 (Innate Pharma) (biosimilar preparations of these are henceforth collectively referred to as anti-CD73 “benchmark antibodies” or simply “benchmarks”), and Sym024 (also known as S095024).

Here, we describe the identification and pre-clinical characterization of the fully human, Fc-attenuated (IgG1-LALA format) anti-CD73 antibody Sym024. Sym024 comprehensively inhibited CD73-mediated catalysis of AMP to adenosine in vitro and displayed better efficacy and/or inhibition potency than anti-CD73 benchmark antibodies across a range of enzyme expression levels. In the context of extended experimental incubation, emulating the TME, Sym024 displayed substantially stronger inhibitory efficacy than anti-CD73 benchmark antibodies. Sym024 inhibited CD73 enzyme activity in a dose-dependent manner and delayed tumor growth alone or in combination with PD-1 blockade in human and murine cancer transplant mouse models. Structural and functional data implicated a unique bivalent 1:1 binding between Sym024 and homo-dimeric CD73 as the molecular mechanism behind the effective CD73 inhibition. Sym024 displayed favorable preclinical safety and pharmacokinetic (PK) profiles in cynomolgus monkeys.

Materials and Methods

Cell lines

All cell lines were mycoplasma tested/authenticated using CellCheck 16 methodology (IDEXX BioAnalytics, Germany) or mycoplasma tested using MycoAlert PCR (Statens Serum Institut, Denmark). All cells were cultured per supplier recommendations, used at less than 15 passages from receival from supplier and continuously evaluated for visual changes. Cell lines were obtained from American Tissue Culture Collection (ATCC), USA (A-375/RRID:CVCL_0132, A-549/RRID:CVCL_0023, AsPC-1/RRID:CVCL_0152, Calu-3/RRID:CVCL_0609, Calu-6/RRID:CVCL_0236, Capan-2/RRID: CVCL_0026, Detroit 562/RRID:CVCL_1171, DU145/RRID:CVCL_0105, MDA-MB-231/RRID: CVCL_0062, MDA-MB-468/RRID:CVCL_0419, NCI-H1975/RRID:CVCL_1511, NCI-H292/RRID: CVCL_0455, NCI-H441/RRID:CVCL_1561, SK-BR-3/RRID:CVCL_0033); Leibniz-Institut DSMZ-Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH (DSMZ), Germany (BFTC-905/RRID:CVCL_1083, BT-474/RRID:CVCL_0179, KYSE-30/RRID:CVCL_1351); National Cancer Institute DTP, DCTD Tumor Repository (NCI), USA (HCT-116/RRID:CVCL_E7EB, MCF-7/RRID:CVCL_0031, T-47D/RRID:CVCL_0553); Japanese Collection of Research Bioresources Cell Bank (JCRB), Japan (HEC-251/RRID:CVCL_2927). Full information for all cell lines is summarized in Supplementary Table S1.

Antibodies used in functional studies

Details on the antibodies used for in vitro and in vivo functional and structural investigations can be found in Supplementary Table S2.

Identification and in vitro functional assessment of anti-CD73 antibodies

Anti-CD73 antibody discovery via a humanized rodent platform (OmniRat^®)

To identify an antibody against CD73 with optimal inhibitory activity, a repertoire of anti-CD73 antibodies was generated by OmniRat^® (OmniAb) immunization using recombinant humanor cynomolgus CD73 purchased from Sino Biological and RD System as well as produced in-house (described in Supplementary Materials), and antibody-secreting cells isolated using markers for irrelevant cell deselection as well as B cell markers. Using Symplex RT-PCR technology, cognate pairs of variable heavy chain and variable light chain were cloned into an expression vector (25,26). Antibody clones were expressed by transient transfection of HEK293 (RRID:CVCL_0045) cells using a proprietary protocol. Antibody supernatants were diluted 250-fold and screened for binding to CHO (Freestyle CHO-S, Invitrogen #R800-07, RRID: CVCL_D604) cells transfected per manufacturer’s instructions (Thermo Fisher Scientific) with full length human, cynomolgus, or mouse CD73 using high-throughput flow cytometry (iQue^® Screener, IntelliCyt) and secondary antibody as described. Target-transfected cells were labelled with carboxyfluorescein succinimidyl ester in different intensities to discriminate populations seeded in the same well. Mock-transfected cells were included as negative control.

Anti-CD73 antibody binding to human, Cynomolgus and mouse CD73

Flow cytometry binding assays were performed using a secondary fluorophore-labeled antibody (AF647-conjugated anti-human IgG (H+L) ((Molecular Probes #A-21445, RRID:AB_2535862))) and either CHO-S cells carrying recombinant human, Cynomulgus or mouse CD73, Cynomulgus CD73-positive cell line Cynom-K1 or primary human CD19+ B cells or CD3+ T cells from purified buffy coats as described below. CD73-expressing cells were incubated with anti-CD73 antibodies for 30 minutes at 4°C, twice washed in wash buffer (Opti-MEM with 2% heat-inactivated FCS, both Gibco (Thermo Fisher Scientific) and 1 nM EDTA Invitrogen (Sigma-Aldrich)) and exposed to the AF647-conjugated secondary antibody for 20 minutes at 4°C. After an additional washing step, binding was detected via high-throughput flow cytometry (iQue Screener PLUS, Sartorius) measuring the geometric mean of the AF647 signal. A 12-point antibody titration curve was generated using triplicate measurements.

In-house generation, expression and purification of recombinant, soluble CD73

The human CD73 sequence from Uniprot DB (RRID:SCR_002380) P21589 was cloned into the NheI-XhoI of the pcDNA3.1+ (RRID:Addgene_51207) mammalian expression plasmid, together with a C-term six His-tag encoding linker sequence in the XhoI-XbaI site. This was transfected into Expi293F HEK293 cells (ThermoFisher #A14527, RRID:CVCL_D615) using ExpiFactamine (ThermoFisher #A14524, RRID:CVCL_D615) as described by the manufacturer. Supernatants were harvested and loaded directly unto 5 mL HisTrap Excel columns (MERCK #GE17-3718-06) for purification and eluded with 500 mM Imidazole. Subsequent dialysis was done three times against 1L PBS at pH 7.4

Anti-CD73 antibody functional evaluation; enzymatic assays

Enzymatic inhibition was examined for recombinant soluble CD73 (sCD73) or CD73 expressed on cancer cell lines or primary B (CD19+) and T cells (CD8+ or CD4+) isolated from healthy human donor buffy coats using Leucosep kits (Greiner-Bio-One) and MACS positive selection (Miltenyi Biotec), following manufacturer’s instructions. For sCD73 inhibition, antibodies were incubated with 200 ng/mL recombinant CD73 (hook effect test; produced in-house as above, with 2 hour incubation and 300 μM AMP, 100 μM ATP), or purchased CD73 (screens; 2000 ng/mL sCD73 SinoBio #10904-H08H, 30 min, 900 μM AMP, 100 μM ATP) in 20 μL Tris-buffer (25 mM Tris, 5 mM MgCl2, pH 7.5), and 20 μL CellTiter-Glo^® 2.0 reagent (Promega Corporation) added to measure CD73 enzymatic activity via AMP inhibition of ATP-dependent luciferase (27). For cell-surface CD73, cells were incubated with antibodies for 30 minutes at 37°C, whereupon 400 μM (cell lines) or 50 μM (primary cells) AMP substrate was added. The cells were incubated for 3, 6, or 24 hours (cell lines, 10,000 cells/well), 24 hours (B cells; 50,000 cells/well), or 48 hours (T cells; 200,000 cells/well) at 37°C in serum-free RPMI medium (Thermo Fisher Scientific). After incubation, cell supernatants and 50 μM (cell lines) or 25 μM ATP (primary cells) added and residual AMP measured using the CellTiter-Glo^® 2.0 kit.

Anti-CD73 antibody functional evaluation; T cell activation

Direct T cell activation was assessed in vitro using primary T-cells. CD4+ T or CD8+ cells were isolated as described above, activated for 1 hour at 37°C with interleukin-2 (RnD Systems) and anti-CD3/CD28 beads (Thermo Fisher Scientific), and subsequently incubated with anti-CD73 antibodies and 100 mM AMP for 48 hours in X-Vivo medium (Lonza) in 96-well plates. They were then incubated with 1 mCi 3H-thymidine (PerkinElmer) per well for an additional 24 hours at 37°C. T cell proliferation was measured as 3H-thymidine incorporation using a scintillation assay and luminometer (MicroBeta2, PerkinElmer).

T cell activation by allogeneic dendritic cell (DC) activation (mixed lymphocyte reaction (MLR)) was examined by mixing CD4+ T cells (50,000 cells/well), isolated as described above, with DCs (10,000 cells/well) differentiated from CD14+ monocytes by 7 days of culture with 20 ng/mL granulocyte-macrophage colony-stimulating factor (GM-CSF, RnD Systems) and 20 ng/mL interleukin-4 (RnD Systems), as described in (1). T cells and DCs were incubated in X-Vivo medium (Lonza) for 72 hours at 37°C with (or without) 50 mM AMP and antibodies as indicated. The anti-PD-1 antibody used was Sym021 (1). Interferon gamma (IFNγ) in supernatants was measured by ELISA (Thermo Fisher Scientific).

AMP competition assay, inhibition of soluble CD73

To assess antibody competition with AMP, anti-CD73 antibodies (2 nM) or APCP (800 nM, an AMP-competing CD73 small molecule inhibitor, Sigma-Aldrich #M3763) were incubated with recombinant sCD73 (2 nM, Sino Biological #10904-H08H) for 1 hour at 22°C followed by addition of the CD73 substrate AMP (Sigma-Aldrich #A1752) up to 2000 μM, and additional incubation for 15 minutes at 22°C. Malachite reagent A (RnD systems # DY996) was added, plates spun down to mix and incubated for 10 minutes at 22°C. Malachite reagent B was added, plates spun down to mix and incubated for 20 minutes at 22°C, and luminescence was measured as a readout of free phosphate (Envision luminometer, PerkinElmer). Assay established in (28).

Test of intrinsic tumor growth inhibition by CD73-blockade on MDA-MB-231 cells in vitro.

Antibodies were distributed to 384 well plates and cells were seeded at 1000 cells pr well and incubated for 2 hours at 37°C. Subsequently, AMP (Sigma-Aldrich #A1752) was added at a 300 μM final concentration and cells incubated for four days at 37°C in a total volume of 20 μL. On day five, 20 μL of WST-1 (Roche #11644807001) in a 1:5 dilution with RPMI medium was added and absorbance measured at wavelengths 450 and 620 nM after 90 min and 210 min to assess mitochondrial activity indicating live, metabolically active cells. Data were background (growth medium values) subtracted and normalized to no treatment (no antibody =100).

Assessment of CD73 protein levels in cell lines

Measurement of CD73 protein was performed on frozen cell pellets lysates exposed to standard RIPA buffer with protease inhibitors (Pierce) and sonication (Branson Sonifier 150, Thermo Fisher Scientific). Protein content was standardized using a BCA kit (Pierce) and submitted to quantification by capillary western (Simple Western, Sally Sue, Bio-Techne), using anti-CD73 (RnD Systems #AF5795, RRID: AB_1964487) and anti-pan-Actin (Cell Signaling #4968, RRID:AB_2313904). CD73 levels were normalized to pan-Actin levels, both calculated with ‘area under curve’ using automatic settings. Two independent experiments were performed on the Sally instrument (Supplementary Table S3, same batch of polyclonal anti-CD73 and anti-pan-Actin), while an addition independent experiment was performed on the newer Jess instrument (new batch of polyclonal anti-CD73 and anti-pan-Actin). Analyses were done with Compass for Simple Western Ver#6.2, Bio-Techne software.

In vivo assessment of anti-CD73 antibody activity

Analysis of in vivo anti-CD73 antibody enzymatic activity in human tumor xenograft and syngeneic mouse tumor models

To generate xenograft tumor models, human tumor cells (MDA-MB-231, A375 or Calu-6) were inoculated subcutaneously into the flanks of 8-10-week-old NODscid (RRID:IMSR_ARC:NODSCID) or NXG (RRID:IMSR_RJ:NXG) mice (Janvier Labs). To generate peripheral blood mononuclear cell (PBMC)-humanized mice with A375 or Calu-6 xenografts, the NXG mice were inoculated with tumor cells on day 0 and received human PBMC intraperitoneally on day 1 or 2.

Syngeneic tumor models were conducted at CrownBioscience (China) (MC38-huCD73) and GemPharmatech (China) (CT26-hCD73/hPDL-1) using subcutaneous inoculation into the flanks of 7-9 week old mice. MC38-huCD73 cells (MC38 with exogenous expression of human CD73) were inoculated in CD73 HuGEMM (C57BL/6, RRID:MGI:2159769 parental) mice, while CT26-hCD73/hPDL-1 (CT26 with exogenous expression of human CD73 and human PDL-1) were inoculated in BALB/c-hPD-1/hPD-L1/hCD73 mice (RRID:IMSR_CRL:028 parental).

Flow cytometric analysis of tumor-infiltrating lymphocytes

For flow cytometric analysis of tumor-infiltrating lymphocytes, tumors were harvested and processed into a single cell suspension using a tumor cell isolation kit (Miltenyi Biotec) and the GentleMACS Octo Dissociator (Miltenyi Biotec) according to manufacturers’ instructions. 1x10⁶ cells were added to a 96-well plate, pre-incubated with Fc block (Miltenyi Biotec), washed with staining buffer (phosphate-buffered saline (PBS), 2% fetal calf serum (FCS), and 0.03% NaN₃), and stained for cell surface markers at 4 °C, in the dark for 15 min. The following antibodies were used: CD45-BV510 5 (BD Biosciences #563891, RRID:AB_2734134), CD4 BV786 (Biolegend #100453, RRID:AB_2565843), CD8 PerCP-Cy5.5 (BD Biosciences #551162, RRID:AB_394081), CD3 FICH (BD Biosciences #555274, RRID:AB_395698), and Granzyme B PE-cy7 (Biolegend #396410, RRID:AB_2801079) . After a second wash with staining buffer, cells were acquired on a FACSVerse flow cytometer (BD Biosciences) with a three-laser, eight-color (4-2-2) configuration using the FACS Suite software, and analyzed with FlowJo Software (RRID:SCR_008520, TreeStar Inc., USA).

CD73 enzyme activity on ex-vivo tumor cell suspension and tissue sections

To evaluate anti-CD73 antibodies for their ability to inhibit the enzymatic activity of CD73 expressed in the tumor mass, dissociated tumor cell suspensions were incubated with CD73 substrate AMP for 3 hours at 37°C. CD73 activity was investigated via CellTiter-Glo^® 2.0 kit (Promega Corporation) as described in the in vitro methods (27). CD73 enzyme activity on tissue sections was determined at MicroMorph (Sweden). Briefly, 8 μm cryosections from tumors were fixed in 1:1 cold phosphate-buffered formalin and acetone and were pre-incubated with 50mM Tris-maleate buffer pH 7.4 containing 2 mM CaCl2 and 0.25 M sucrose. The buffer was then replaced with the same buffer supplemented with 5 mM MnCl2, 2 mM Pb(NO3)2, 2.5% Dextran T200, 2.5mM levamisole, and 1 mM AMP. The enzymatic reaction was carried out for 1 hr at 37°C and then stopped by 1% (NH4)2S. The cryosections were then counterstained in hematoxyline, dehydrated, and mounted.

In all in vivo experiments, mice were monitored daily, and tumors were measured two or three times weekly by caliper. Tumor volume was calculated using the formula: 0.5 x length x (width)². At a predetermined tumor size, mice were randomized into groups and treatment initiated. For PBMC-humanized NOG mice, treatment was initiated the same day as PBMC inoculation. Antibodies were administered two-three times weekly via intraperitoneal injection.

Structural studies of Sym024-CD73 interaction

Anti-CD73 antibody binding by surface plasmon resonance (SPR)

Kinetic binding analyses were performed using an SPR imaging system with a Continuous Flow Microspotter (CFM, Wasatch Microfluidics) and an Instrument for Biomolecular Interaction Sensing MultipleX 96 (IBIS-MX96, IBIS Technologies). Proteins were expressed as either C-terminal 6x histidine (His) tag or as human IgG1 Fc fusion proteins using the ExpiCHO^™ (RRID:CVCL_5J31) expression system and purified by standard nickel-nitrilotriacetic acid (Ni-NTA) chromatography or Protein A purification. Anti-CD73 Fab fragments were generated by digesting IgG1 antibodies with the GingisKHAN enzyme (Genovis).

Antibodies were captured on a G-a-hu-IgG Fc SensEye^® (Ssens BV) for 15 minutes by the CFM. After spotting, the SensEye was docked in the IBIS MX96, and spotted antibodies were fixed via SensEye FixIt kit (Ssens BV). Kinetic analysis involved injecting His-tagged antigens at increasing concentrations (0.16 to 10 nM) and regenerating the surface with 100 mM H₃PO₄ at pH 3. For Fab fragments, Fc-fusion CD73 were captured on a G-a-hu-IgG Fc SensEye^® and fixed by fix-it kit (Ssens BV). Kinetic analysis was conducted by injecting Fab fragments at concentrations from 0.8 to 300 nM, with surface regeneration using 10 mM glycine at pH 3 and 10% glycerol. Binding responses were fitted to a Langmuir 1:1 binding model using Scrubber 2.0 software to calculate the on-rate (K_on or K_a), off-rate (K_off or K_d), and affinity (K_D) constants.

To validate cryogenic electron microscopy (cryo-EM) data, binding experiments were performed with mutated human CD73 antigen. Human and rat (Rattus norvegicus) CD73 protein sequences were obtained from Uniprot (RRID:SCR_002380, accession P21589 and P21588, respectively). The sequences were aligned, and selected surface-exposed residues differing between human and rat CD73 were mutated to alanine. Constructs were transiently expressed in ExpiCHO^™ cells as Fc-fusion proteins, and the resulting supernatants containing CD73 fusion proteins were assessed for their binding affinity to anti-CD73 Fabs using SPR. The CD73 fusion proteins were captured and analyzed following the previously described protocol.

Stoichiometry of antibody-CD73 complexes formed in solution via size-exclusion chromatography with multi-angle light scattering (SEC-MALS)

Anti-CD73 antibodies and His-tagged CD73 were analyzed, both individually and in various ratios, to determine the size of complexes formed between CD73 homodimers and anti-CD73 antibodies by SEC-MALS. Samples were prepared by mixing 900 μmol of CD73-His with 900, 450, 90, or 0 μmol of antibody, diluted in PBS at pH 7.4. The mixtures were incubated for 30 minutes at room temperature, then separated using an ultra-high performance liquid chromatography (UHPLC) UltiMate 3000 system (Thermo Fisher Scientific) with an SEC X-Bridge column (Waters) at a flow rate of 1.2 mL/min. The running buffer was 0.01 M citrate, 250 mM L-arginine, hydrochloric acid, pH 6.0. After HPLC separation, all samples were analyzed using a MiniDAWN TREOS MALS detector (Wyatt) and Optilab T-rEX refractive index detector (Wyatt). Data plots were generated using GraphPad Prism.

Cryogenic electron microscopy sample preparation, data collection, and image processing

3 μL of the solution containing the CD73-Sym024 complex were applied either to the foil side of a plasma-cleaned UltraAuFoil holey gold grid (Quantifoil R 0.6/1.0, Au 300) or to the grid bar side of an UltraAuFoil R 0.6/1 with a single layer of graphene, prepared as previously described (29). Using a Vitrobot Mark IV (Thermo Fisher Scientific), the grids were blotted at 4°C, 100% humidity, 1.5-second blot time, blot force 10, and were plunge-frozen in liquid ethane pre-cooled by liquid nitrogen (30). For graphene grids, a 30-second wait was added to the grid freezing process. For both sets of grids, micrographs were recorded using a Titan Halo (Thermo Fisher Scientific) electron microscope operating at 300 kV and equipped with a Gatan K3 camera with a nominal magnification of 37,000x and calibrated with a pixel size of 0.8465 Å/pixel. Of 5534 movies collected for the UltraAu foil grid, 4366 were kept, 3195 were collected, and 2439 were kept for the graphene grid. Movies were collected with a total dose of 60 e⁻/Ų fractioned into 70 frames for each movie using SerialEM (31).

Image processing was identical for both datasets unless otherwise stated. For the single layer graphene grids, a BoxNet model was trained in Warp (32) to optimize the picking of particles with low signal to noise ratio from the dataset collected, in order to maximize particle count and capture rare views. Motion correction, contrast transfer function estimation and particle extraction using a box size of 320 pixels were done in Warp (32) (Supplementary Figure S1). The UltraAu foil dataset yielded a particle stack containing 477,028 particles, and the graphene dataset 773,442 particles. Further image processing, including 2D and 3D classification and refinements of cryo-EM maps, was done in cryoSPARC v3.0 (RRID:SCR_016501) (33). Particles were first down-sampled from 0.8465 Å/pixel to 3.4 Å/pixel, and down-sampled particles were used during 2D classification, resulting in 427,052 particles for the UltraAU dataset and 403,496 particles for the graphene dataset. The selected particles from the 2D classification underwent ab initio 3D reconstruction to generate initial 3D maps resembling the topology of PDB (RRID:SCR_012820)7BBJ (CD73-mAb 19 complex) (34). A further round of particle cleanup using heterogenous refinement with ab initio selected maps and decoy maps as input volumes, resulted in a final set of 226,436 particles for the UltraAu dataset and 235,351 particles for the graphene dataset. Particles which did not get sorted to decoy maps were further 3D-classified using heterogenous refinement; particles from the selected 3D classes were restored to resolution of collected data (0.8465 Å/pix) and used in non-uniform refinement (35) of selected 3D volumes to improve resolution. Particles from the high-resolution, non-uniform refined 3D volume were exported to Relion3.1 format (36) using Pyem (37) with home-made tools (https://github.com/EArmbruster/motion-star-editor) for Bayesian polishing in Relion3.1 (38). Bayesian-polished particles were imported back to cryoSPARC and the two datasets merged into a final dataset of 286,779 particles. This final set was subjected to a last round of non-uniform refinement, yielding a final map with a global resolution of 3.1 Å (Supplementary Figure S1). Local refinements focused on different regions of the complex were then performed using 10° rotation search and 5 Å shift search to improve local resolution and resolvability in dynamic map regions. Briefly, masks used in local refinement were made by filtering volume to 20 Å resolution, manually segmented to contain various regions of the volume, then dilating 2 Å with 12 Å of soft padding. The maps from local refinements reached resolutions between 2.7 and 2.9 Å based on gold-standard Fourier shell correlation of 0.143 criterion. They were combined in UCSF Chimera (RRID:SCR_004097) (39) to form a composite map used for model building and analysis of CD73-Sym024 complex.

Cryogenic electron microscopy structure modeling, refinement, and analysis

Initial coordinates to model CD73-Sym024 complex were taken from PDBs 7PBY (40) for CD73 and from 7BBJ (34) for the Fab region of Sym024. They were rigid-body fitted into the cryo-EM density map using UCSF Chimera (39). To account for different elbow angles between antibody domains, they were first separated and then rigid-body fitted into the cryo-EM map. Rigid-body fitted antibody domains were mutated in COOT (RRID:SCR_014222) (41) to match the Sym024 sequence. The rigid-body fitted CD73-Sym024 complex was then flexibly fitted to the cryo-EM map using Namdinator (42), which uses molecular dynamics to achieve flexible fitting. This was followed by iterative rounds of real-space refinement and model building in PHENIX (RRID:SCR_014224) (43), COOT (41), and ISOLDE (RRID:SCR_025577) (44). The final model was validated using MolProbity (45). The validated model was used for molecular analysis of the CD73-Sym024 complex. Figures were generated with PyMol (RRID:SCR_000305) (46) and UCSF ChimeraX (47).

Epitope mapping of anti-CD73 antibodies by CD73 mutagenesis and surface plasmon resonance (SPR)

Epitope mapping of Sym024 and benchmarking antibodies were performed by combining mutagenesis and Surface Plasmon Resonance (SPR). Protein sequences for human and rat (Rattus norvegicus) CD73 were obtained from Uniprot (Accession numbers P21589 and P21588, respectively). Full-length protein sequences for chicken (Gallus gallus) CD73 were downloaded from NCBI (XP_005552488.1 and XP_004940453.1, respectively) and aligned. Publicly available CD73 structures (PDB 4H2F, 4H2G, and 4H2I) were used to map surface-exposed amino acid residues. Residue positions that differed between human and rat CD73 were mutated to alanine. To map linear antibody epitopes, CD73 chimeric proteins were created by sequentially exchanging segments of 10 amino acids in the human CD73 extracellular domain (ECD) sequence with chicken sequence, overlapping by 5 amino acids. The cDNA coding for the extracellular domain of human CD73 was synthesized and cloned into a vector with a CMV promoter and human Ig Fc sequence (residues P101-K330), resulting in a fusion of Ig Fc to the C-terminal of the cloned CD73 ECD. Wild type and mutated human CD73 Fc fusion constructs were generated using standard gene synthesis techniques and expressed transiently in 2 ml cultures with an ExpiCHO^™ expression system. After harvesting, supernatants were tested for binding to anti-CD73 Fabs by SPR as described above.

Preclinical pharmacokinetics and toxicology in cynomolgus monkeys

Toxicity studies in cynomolgus monkeys (Macaca fascicularis) were performed by Labcorp (UK) and included standard toxicology, immunotoxicity, and toxicokinetic endpoints. The cynomolgus monkey was selected as the most appropriate species for prediction of human PK and potential human toxicity of Sym024 based on CD73 amino acid sequence homology and similar target binding strength between human and cynomolgus CD73.

In a pilot study, one male and one female cynomolgus monkey were dosed once-weekly for three weeks with escalating doses of Sym024 (5, 30, and 100 mg/kg/occasion) via 30-minute intravenous infusion. In a subsequent study, cynomolgus monkeys (three males and three females/group) were dosed five times, once weekly with 0 (vehicle), 10, 30 or 100 mg/kg/week Sym024 via 30-minute intravenous infusion. Three recovery groups (two males and two females/group) were dosed with 0 (vehicle), 10, or 100 mg/kg/week Sym024 for four weeks, followed by an eight-week dosing-free period.

An additional PK group (two male cynomolgus monkeys for GLP) were dosed with a single dose of 10 mg/kg Sym024 to obtain a full 3-week Sym024 elimination profile. Sym024 serum levels were measured using a validated method employing a generic anti-human IgG Gyrolab kit on the Gyrolab xP platform (Gyros AB). Anti-drug antibodies (ADA) were measured via electrochemiluminescence bridging immunoassay on the Meso Scale Discovery (MSD) platform (Meso Scale Diagnostics, Llc).

Statistical analyses

All statistical analyses were performed using GraphPad Prism (RRID:SCR_002798) Version 9 or 10 software (GraphPad Software, USA), and values were computed using a standard two-way ANOVA, unless otherwise stated. P-values equal to or below 0.05 (p ≤ 0.05) were considered statistically significant.

Data and material availability

Cryo-Electron Microscopy (Cryo-EM) maps (image in Supplementary Figure S1) and models are deposited in the EMDB (EMD-71128) and PDB (PDB 9P1M) databases. Additional accession codes are EMD-71125, EMD-71123, EMD-71126 and EMD-71127. Statistics regarding cryo-EM are given in Supplementary Table S4. Further data are available upon request from the corresponding author.

Ethics approval

All in vivo studies carried out in Denmark (at Symphogen Servier A/S) were approved by The Animal Experiments Inspectorate under the Danish Ministry of Food, Agriculture and Fisheries and performed in accordance with applicable laws and regulations relating to the care and use of laboratory animals. Mouse studies carried out in China were done at Crown BioSciences and GemPharmatech, which are accredited by the International Laboratory Animal Evaluation and Accreditation Management Committee (AAALAC International) and all studies reviewed and approved by the local Institutional Animal Care and Use Committee (IACUC), as well as at the Experimental Animals Administration Regulations, 2^nd Detective, Chinese Scientific-Technology Committee, or the Animal Management Committee of the Jiangsu Provincial Department of Science and Technology. Monkey studies were performed by Labcorp in the UK, approved by the Labcorp Animal Welfare and Ethical Review Body (AWERB) in compliance with establishment, project and personal licences granted by the UK Home Office board, in accordance with the Animal (Scientific Procedures) Act 1986 (ASPA) regulation.

Results

Identification of Sym024, an antibody with comprehensive inhibition of both cell-surface and soluble CD73, via deep functional screening

A repertoire of 8832 anti-CD73 antibodies was generated by OmniRat^® immunization. Of these, 1537 were shown to bind to both human and cynomolgus CD73. 260 anti-CD73 antibodies were selected for further evaluation and subjected to functional testing. We assessed their ability to inhibit enzymatic activity of recombinant sCD73 and CD73 endogenously expressed on two cell lines (Figure 1A). Based on functional ranking, binding properties, sequence diversity and developability data, 27 sequence-corrected anti-CD73 antibodies were selected (Figure 1B), produced in IgG1-LALA isotype and screened in CD73 activity assays. The antibody with the strongest and broadest inhibitory effect, Sym024 (and a close sequence relative) almost completely inhibited sCD73 (with 4% ± 0.28% SD residual activity).

Figure 1. Sym024 was identified by orthogonal functional screening with both cell-surface and soluble CD73 (sCD73) and displays favorable binding and inhibitory characteristics compared to benchmark antibodies.

A. Orthogonal functional screening of CD73 enzymatic inhibition of cell surface (H292 and Calu-6 cell lines) and recombinant sCD73. Antibodies with >20% residual sCD73 activity are shown in green; those with <20% residual sCD73 activity are shown in red (two). The unspecific control antibody is represented by a black square. Data is shown without error bars for clarity. B. Functional assessment of 27 selected antibodies. Residual enzyme (CD73) activity is shown in green (H292 surface enzymes), blue (Calu-6 surface enzymes), and red (sCD73). Sym024 was identified as the best inhibitor. 400 μM AMP, 50 μM ATP was used for cell lines and 900 μM AMP, 100 μM ATP for sCD73 assays. C. Flow cytometric measurement of anti-CD73 antibody binding expressed as mean fluorescence intensity (MFI) to CD73-transfected Chinese Hamster Ovary Suspension (CHO-S cells) (left panel), isolated primary T cells (CD3+; middle panel), or B cells (CD19+; right panel). Four different anti-CD73 antibodies were tested. D. Dose-response curves for enzyme (CD73) inhibition in isolated primary CD19+ B cells (left panel), CD8+ T cells (middle panel), or CD4+ T cells (right panel). 50 μM of AMP and 6.25 μM ATP was used. All data represent mean values of three replicates ± SEM. ****, P < 0.0001; standard two-way ANOVA. Activity values (AMP removal allowing ATP-dependent luciferase luminescence) were normalized to enzyme (CD73) activity with no antibody, and background values (no cells) subtracted.

Sym024 binds CD73 with high avidity and displays deep inhibition of CD73 on primary human leukocytes

Flow cytometry with CD73-transfected CHO-S cells or primary immune cells from healthy donors showed that the apparent affinity (avidity) of Sym024 was greater than that of a panel of biosimilar analogs of antibodies in recent clinical development (these are henceforth referred to as oleclumab, mupadolimab, and IPH5301, as a whole, “benchmark antibodies”); EC50 values for Sym024 versus oleclumab were approximately 5 versus 65 (in CHO-S cells), 8 versus 20 (in B cells) and 1 versus 18 ng/mL (in T cells) (Figure 1C and Supplementary Table S5). The maximum mean fluorescence intensity (MFI), as a proxy for CD73 cell surface density, was substantially higher in CHO-S cells (Sym024 MFI: 170,000) than primary cells (Sym024 MFI: 27,000 in B cells and 5,100 in T cells).

CD73 inhibition on primary CD4+, CD8+ T cells, and CD19+ B cells differed between Sym024 and the benchmark antibodies (Figure 1D). Higher concentrations of Sym024 resulted in almost complete inhibition of CD73 activity in all primary cell types. Meanwhile, oleclumab showed limited inhibition of CD73 activity in CD19+ B cells and CD8+ T cells and only partial inhibition in CD4+ T cells, correlating with known CD73 expression levels (48). IPH5301 was able to inhibit CD73 with potencies and efficacies similar to Sym024 in CD4+ and CD8+ T cells but performed worse in CD19+ B cells, with approximate IC50 values of 1 versus 2 μg/mL and residual activity of 9 versus 21% for Sym024 and IPH5301, respectively. Mupadolimab displayed markedly weaker potency and efficacy in all primary cells tested.

Sym024 was further found to bind and inhibit cynomolgus (Macaca fascicularis) CD73, but not mouse (Mus musculus) CD73 (Supplementary Figure S2).

Deep inhibition of CD73 fully relieves adenosine-mediated repression of CD4 and CD8 T cells and synergizes with anti-PD-1 for T cell activation

Both CD4+ and CD8+ T cells were shown to proliferate when CD73 was inhibited, abrogating the conversion of added AMP to suppressive adenosine (Figure 2A). All anti-CD73 antibodies counteracted adenosine-mediated suppression and stimulated CD4+ and CD8+ T-cell proliferation, albeit with differing potencies and efficacies. While Sym024, mupadolimab, and IPH5301 fully restored T-cell proliferation to pre-AMP levels (100%), Sym024 did so with the highest potency, with approximate EC50s of 40 (CD4+ cells) and 190 ng/mL (CD8+ cells). Oleclumab only partially restored proliferation up to a maximum of 28% for both T cell subtypes.

Figure 2. Sym024 effectively alleviates adenosine-mediated repression of T cell proliferation and activation and synergizes with PD-1 blockade.

A. Anti-CD3/anti-CD28 activated T cells proliferate in the presence of AMP when CD73-mediated AMP-to-adenosine conversion is inhibited. CD4+ T cells (left panel) or CD8+ T cells (right panel). Values are normalized to T cell proliferation without AMP, and background values without T cell activation are subtracted. B. Activation of T cells assessed by interferon gamma (IFNγ) release (mixed lymphocyte reaction (MLR)) upon inhibition of CD73 and/or PD-1 blockade. No AMP (left panel) or with added AMP (right panel, 50 μM). The No AMP/with AMP data are from the same donor and representative of three donor pairs. All data represent mean values of three replicates ± SEM. *, P < 0.05; ***, P < 0.001 (anti-PD-1 versus control antibody); ****, P < 0.0001 (anti-PD-1+Sym024 versus anti-PD-1 and Sym024 versus control antibody); standard two-way ANOVA.

To investigate PD-1 blockade and Sym024 combination treatment, we explored the functionality of an anti-PD-1 mAb (Sym021 (25)) in vitro using a one-way MLR with interferon gamma (IFNγ) as a read-out of T-cell activation. PD-1 blockade, either alone or in combination with Sym024, relieved repression of the one-way MLR reaction, likely exerted by PD-1 ligands (Figure 2B). When anti-PD-1 mediated activation was suppressed by addition of AMP, the one-way MLR reaction was not stimulated by blocking PD-1 alone. In contrast, Sym024 alone slightly increased IFNγ levels and the combination of the anti-PD-1 mAb and Sym024 strongly activated the MLR reaction (Figure 2B).

Extended incubation highlights the importance of comprehensive enzymatic inhibition as exerted by Sym024

In the TME, cancer cells often express highly elevated levels of CD73 (5). Further, even residual CD73 activity due to incomplete inhibition may yield impactful levels of adenosine over time. Thus, we examined the ability of Sym024 and the benchmark antibodies to inhibit AMP catalysis using CD73-high lung cancer cells (NCI-H292) over short, intermediate, and long incubation periods.

Following 3-hour incubation, there were no clear differences in inhibition efficacy among the tested antibodies, with the exception of oleclumab which, in accordance with data from primary cells (Figures 1D and 2A), only partially abolished CD73 activity (Figure 3A). Following 6-hour incubation, CD73 catalytic activity was not inhibited by oleclumab (100% residual CD73 activity), substantially inhibited by mupadolimab (16% residual CD73 activity) and IHP5301 (14% residual CD73 activity), and near-fully inhibited by Sym024 (2% residual CD73 activity) (Figure 3B). Following 24-hour incubation, Sym024 displayed substantial CD73 inhibition (22% residual CD73 activity) and the benchmark antibodies displayed no inhibition (100% residual CD73 activity).

Figure 3. Sym024 displays strong inhibitory efficacy at both long incubation periods and high antibody-to-soluble CD73 ratios, suggesting a unique mode of action.

A. Dose-response inhibition of cell surface enzymatic activity of the highly CD73-expressing cell line NCI-H292 with a 3-hour incubation period. Background values (no cells) subtracted. B. Single-concentration (25 μg/mL) inhibition of H292 enzymatic activity following 3-, 6- or, 24-hour incubations. C. and D. Single-concentration (25 μg/mL) enzymatic inhibition with a 20-cell-line panel, triplet groups of 3- (left), 6- (center), or 24-hour (right) incubations (as in panel B, illustrated by insert in C, cell lines indicated on X axis. Cell lines ordered according to residual activity for Sym024 at 24-hours. Background values (AMP removal with no cells) were not subtracted but set to 100%. C. Negative control antibody, showing the AMP removal activity with no inhibition of CD73. D. Sym024 (upper left panel; green), oleclumab (upper right panel; red), mupadolimab (lower left panel; yellow), and IPH5301 (lower right panel; blue). E. Dose-response inhibition of soluble CD73, 2-hour incubation. The vertical dotted line signifies equimolar amounts of antibody and dimeric enzyme. F. Dose-response inhibition of soluble CD73, 2-hour incubation. Comparison of the full length Sym024 antibody (bivalent) with F(ab’)2 (bivalent) and Fab (monovalent) fragments. Concentrations in nM for comparison. 400 μM AMP, 50 μM ATP was used for cell lines and 300 μM AMP, 100 μM ATP for sCD73 assays. All data represent mean values of three replicates ± SEM.

Upon extension of the short-, intermediate-, and long-duration enzymatic inhibition assays to a panel of 20 cell lines representing very high to low or no CD73 activity (see Figure 3C and Supplementary Table S3), we observed Sym024 to inhibit CD73 activity comprehensively even at the 24-hour time point (Figure 3D); CD73 residual activity was strongest in HEC-251, Capan-2, and Calu-6 cells. While the HEC-251 cell line harbors low levels of CD73 expression at the protein level, it nevertheless effectively catalyzes the conversion of AMP, possibly due to expression of alkaline phosphatase, as indicated by RNAseq DepMap database data (Supplementary Table S3). In Capan-2 and Calu-6 cell lines, which harbor the highest CD73 protein levels of all cell lines tested, incomplete inhibition may represent a scenario of partial antibody pool saturation.

The benchmark antibodies inhibited CD73 less comprehensively, especially after 24 hours of incubation, with oleclumab displaying the least inhibition. Two additional anti-CD73 clinical candidates, uliledlimab and dresbuxelimab, exhibited similarly limited CD73 inhibition after 24 hours (Supplementary Figure S3A).

To confirm antibody stability during assays, we pre-exposed all antibodies to 37°C for 24 hours prior to a 3-hour incubation activity test and found little change in inhibitory capacity (Supplementary Figure S3B and C).

Sym024 and all benchmark antibodies maintained their ability to inhibit sCD73 in the presence of increasing levels of its substrate AMP up to 2000 μM, indicating a non-competitive mechanism and retained inhibition at high substrate levels. In contrast, the small molecule inhibitor APCP lost its inhibitory effect at high AMP levels, confirming its AMP-competitive nature (Supplementary Figure S4).

Titrating the antibodies using a constant amount of sCD73 confirmed a ‘hook’-effect for oleclumab (28), whereby inhibition was lost at high antibody-to-enzyme ratios (Figure 3E). A similar effect was observed for mupadolimab and, to a lesser extent, for IPH5301. Sym024 displayed no loss of inhibition.

To investigate whether bivalent bridging of enzyme dimers is required for Sym024-mediated inhibition (similarly to oleclumab-mediated inhibition (28)), we compared full-length Sym024 IgG1-LALA with monovalent Fab and bivalent F(ab’)2 fragments (Figure 3F). We demonstrated a higher potency for the bivalent full-length antibody and F(ab’)2 (IC50 values of 2.67 and 1.97 nM, respectively), than for the Fab fragment. Even at the highest concentration of 666.67 nM (corresponding to 100 μg/mL for an IgG1 antibody), the Fab fragment did not fully inhibit CD73.

Sym024 interacts with a unique epitope on the CD73 homodimer favoring effective bivalent binding to the open conformation

To understand the molecular mechanism underlying the highly efficacious, bivalency-dependent Sym024 inhibition of CD73 enzymatic activity, we characterized complex formation by surface plasmon resonance (SPR), size-exclusion chromatography with multi-angle light scattering (SEC-MALS), cryo-EM, and alanine mutagenesis.

The monovalent binding affinity (K_D) of the Sym024 Fab fragment to the CD73 homodimer was 55±4 nM, with fast on- and off-rates (Figure 4A). In contrast, the bivalent binding (avidity) of full-length Sym024 showed an apparent affinity (K_D,app) of 0.22±0.03 nM with a slow off-rate (Supplementary Table S6 and Figure 4B). IPH5301 demonstrated a similar difference between its monovalent binding affinity (K_D = 10±1 nM) and bivalent binding (K_D,app = 0.10±0.04 nM). Conversely, oleclumab exhibited a high CD73 affinity for both monovalent (Fab) (K_D = 0.06±0.01 nM) and bivalent (mAb) binding (K_D,app = 0.02±0.01 nM) (Figure 4A and Supplementary Table S6). SPR binding data indicated varying dependencies on bivalent interactions among the antibodies, with Sym024 showing the highest dependence, followed by IPH530, and oleclumab showing almost no dependence.

Figure 4. Investigation of Sym024 binding to Human CD73 extracellular domain indicates a unique 1:1 interaction.

A. Representative SPR sensograms of Fabs and mAbs binding to recombinant CD73 heterodimer. Colored lines represent SPR data, while black lines indicate global fits to a 1:1 binding model for Sym024 (green), IPH5301 (blue), and oleclumab (red). Average dissociation constants (K_Ds) with standard deviations are presented, along with the highest antigen concentration used (n=6). CD73 concentrations indicated in grey. B. Size-exclusion chromatography (SEC) profiles of antibodies Sym024 (green), IPH5301 (blue), and oleclumab (red), mixed with CD73 in a 1:1 molar ratio (n:n) of mAb:CD73 (6 μM of each, corresponding to ~900 μg/mL of antibody). SEC profiles of antibodies alone (dotted lines) and CD73 alone (cyan) are also shown. The multi-angle light scattering (MALS)-calculated sizes of each peak are indicated and detailed in Supplementary Table S7. Experiment was run twice for Sym024 and once for benchmarks. C. Cryo-EM density map of Sym024_CD73 complex viewed from two different perspectives. Sym024 is shown in green, and CD73 in cyan. D. Ribbon diagram with surface representation in gray of the Sym024:CD73 structure. Sym024 is shown in green, and the CD73 heterodimer in cyan. Distance between each paratope is indicated. E. The paratope and epitope are highlighted in the square (dotted lines) and in zoom. Molecular interactions are detailed in Supplementary Tables S8A and S8B.

SEC-MALS showed a highly homogeneous complex of Sym024:CD73 extracellular domain with one main peak with an average size of 269.1 kDa, corresponding to a stoichiometry of 1:1 Sym024:CD73 (Figure 4B, and Supplementary Table S7). A second smaller peak corresponded either to excess Sym024 (149.7 kDa) or CD73 dimers (124.2 kDa) (Figure 4B and Supplementary Table S7). Notably, IPH5301 exhibited a mixed population of complexes, including a peak corresponding to one antibody per CD73 homodimer and a peak reflecting higher-order oligomer complexes (510.8 kDa). Oleclumab formed large, CD73 dimer-cross-linking complexes (Figure 4B and Supplementary Table S7), in line with previous observations (28). For the other anti-CD73 clinical candidates, mupadolimab, uliledlimab and dresbuxelimab, similar higher-order complex formations were observed (Supplementary Figure S5A and Supplementary Table S7).

A 3.2Å cryo-EM structure of full-length Sym024 binding a CD73 homodimer (Figures 4C and 4D) revealed a 1:1 stoichiometry between the CD73 homodimer and Sym024, with each CD73 homodimer bound by one arm (Fab) of Sym024 at the N-terminal domains of CD73, perpendicular to the dimerization interface of the CD73 homodimer and on the opposite side of the catalytic center (Figures 4C and 4D).

The flexible Fc region of Sym024 was not resolved via cryo-EM but instead modelled for illustration (Figure 4D). Per our model, the epitope consists of residues Q69, R73, A74, P76, N77, R109, P165, G167, and D168. The recognition of CD73 by the variable fragment of Sym024 is achieved through a combination of hydrogen bonding, and electrostatic and hydrophobic interactions by the complementarity-determining regions (CDRs) H1, H3, and L1, which form the paratope of the antibody (Figure 4E (zoom) and Supplementary Table S8). Alanine substitutions of selected CD73 residues showed that residues R73, R109, and D168 are particularly critical for binding, as their substitution led to a complete loss of Sym024 Fab binding (Supplementary Table S9), validating the structure of Sym024:CD73 complex resolved by cryo-EM. The CD73 epitope for Sym024 differs markedly from benchmark epitopes, as illustrated in Supplementary Figure S5B.

In the resolved structure of the complex, the CD73 homodimer is, in its open conformation, comparable to an earlier reported structure (PDB 4H2G (49)). The molecular distance of approximately 9-10 nanometers between the epitopes on each CD73 homodimer (Figure 4D and Supplementary Figure S5B) falls within the optimal range for maximal avidity enhancement of human IgG1 (50-52). This plausibly explains why bivalent binding of Sym024 to the CD73 open conformation is favored, resulting in effective locking of the conformation and thus restriction of movement. In contrast, IPH5301 binds to epitopes in closer proximity, thereby stabilizing an intermediate conformation of CD73 as previously suggested (53) and illustrated in Supplementary Figure S5B.

Sym024 inhibits tumor growth in murine and human tumor models

The capability of Sym024 to inhibit tumor growth was explored in various human and murine tumor models.

Inhibiting CD73 activity of MDA-MB-231 limits tumor growth of this triple negative breast cancer cell line in immunocompromised mice (54,55). Thus, to test the direct effect of inhibiting CD73 activity of in vivo tumor growth in the absence of immune cells, Sym024, oleclumab, and mupadolimab were evaluated in MDA-MB-231-grafted immunodeficient mice (Figure 5A). Sym024, oleclumab and mupadolimab effectively controlled tumor growth, with no tumor size increase during treatment. At the end of treatment (on day 40), a significant tumor growth inhibition (P<0.005) was observed for Sym024, oleclumab, and mupadolimab. Sym024 and mupadolimab responses were sustained until 50 days post-treatment, in contrast to 20 days for oleclumab. At 28 days post-treatment (day 68), oleclumab-treated tumors showed levels of CD73 activity similar to vehicle-treated tumors, with significantly suppressed activity by Sym024 (Supplementary Figure S6). At termination (day 123), tumor regrowth was seen in all treatment groups; however, tumor growth was significantly reduced in Sym024 compared to oleclumab treated mice (P<0.05) (Figure 5A).

Figure 5. Sym024 inhibits CD73 and tumor growth in vivo.

A. Tumor growth in NODscid mice subcutaneously engrafted with MDA-MB-231 cells. The mice were treated twice weekly with vehicle or 10 mg/kg Sym024, oleclumab, or mupadolimab (n=10 mice/group). The grey area denotes the treatment period. Data are presented as means ± SEM. At day 45, a two-way ANOVA with Bonferroni’s multiple comparisons test was applied to compare tumor volumes at each time-point between treatment groups; at the end of treatment, an unpaired T test was applied. B and C. Peripheral blood mononuclear cell (PBMC)-humanized mice engrafted with A375 tumor cells received vehicle buffer, 5 mg/kg, 20 mg/kg, or 50 mg/kg of Sym024 three times weekly for two weeks (n=5/group). Tumors were harvested one day post last dose and analyzed for CD73 activity. B. Enzyme activity (AMP removal) analyzed in dissociated cells from the tumor (including tumor, immune, and stroma cells). An unpaired T test was applied. C. Phosphatase (CD73 enzyme) activity analyzed by lead(II) phosphate disposition on tumor sections from mice receiving vehicle or 50 mg/kg Sym024. T and S indicate human tumor and mouse stroma cells, respectively. D. Tumor growth inhibition of Sym024 treatment analyzed in 11 different studies using 11 PBMC donors engrafted with human melanoma A375 cells or human lung carcinoma Calu-6 cells. Mice were treated with vehicle buffer or 50 mg/kg of Sym024 three times weekly for two weeks (n=8-10 mice/group). Percent Sym024 tumor growth inhibition compared to control was calculated from end of study ((1-(median Vmax Sym024- medianVstart)/(medianVmax Vehicle-medianVstart)) x 100). A two-way ANOVA with Bonferroni’s multiple comparisons test was applied to compare tumor volumes at each time-point between treatment groups (n=8-10 mice/group). See Supplementary Figure S8B for individual donor data. E. PBMC-humanized mice (donor 11) engrafted with A375, treated as in D (n=10 mice/group). The grey area denotes the treatment period. F. CD73 HuGEMM mice were subcutaneously engrafted with MC38-hCD7. On day six after cell inoculation and at average tumor volume of 60 mm³, mice were randomized and treated twice weekly for three weeks with vehicle buffer, Sym024 (20 mg/kg), anti-PD-1 (Sym021) (1 mg/kg) or Sym024+anti-PD-1. Each graph shows tumor growth of individual mice (n=10 mice/group). G. BALB/c-hPD-1/hPD-L1/hCD73 mice were subcutaneously engrafted with CT26-hCD73/hPDL-1. On day ten after cell inoculation and at average tumor volume of 50 mm³, mice were randomized and and treated twice weekly for three weeks with vehicle buffer, Sym024 (50 mg/kg), anti-PD1 (pembrolizumab) (5 mg/kg), or Sym024+anti-PD-1. Each graph shows tumor growth of individual mice (n=10 mice/group). H. Tumor infiltrating lymphocytes of tumors from G was analyzed by flow cytometry for CD45+, CD3+, CD4+, CD8+ and Granzyme B+ cells. One-way ANOVA was applied. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

To explore in vivo effects of Sym024 in the context of an immune system, we selected two cell line-derived human xenograft models with mid (A375) or high (Calu-6) CD73 expression (Supplementary Table S3) for which treatment with anti-CD73 antibodies in the absence of human immune cells had no or limited effect on the tumor growth, in contrast to MDA-MB-231 (Supplementary Figure S7). Both cell lines express substantial CD73 upon engraftment (Supplementary Figure S8A). In dissociated tumor cell suspensions from mice reconstituted with human PBMC and engrafted with A375 cells, Sym024 demonstrated dose-dependent inhibition of CD73 activity after repeat dosing. 50 mg/kg Sym024 inhibited CD73 activity by 87%, 20 mg/kg by 75%, and 5 mg/kg by 68% (Figure 5B). Sym024 is not cross-reactive to mouse CD73, and Sym024-mediated CD73 inhibition was thus specific to human cells, leaving murine stroma cells to hydrolyze AMP to adenosine (Figure 5C).

The anti-tumor effect of Sym024 was analyzed in 11 PBMC-humanized xenograft mouse tumor models using PBMC isolated from 11 different healthy donors (one donor/study). Six of the mouse models displayed tumor growth inhibition above 10% after treatment with Sym024 (Figure 5D). Anti-tumor effects were significant in Sym024-treated mice compared to vehicle-treated mice for three human PBMC donors (Donor, 9, 10 and 11); donor 11 is shown in Figure 5E, with all individual data in Supplementary Figure S8B. For donor 11, flow cytometry analyses show a tendency to increased CD3+, CD8+ and CD4+ lymphocyte infiltration in tumors of Sym024-treated mice compared to those of vehicle-treated mice (Supplementary Figure S8C).

Sym024 was investigated for its ability to enhance anti-PD-1 therapy in two syngeneic tumor models with exogenous expression of human CD73 (huCD73) engrafted on huCD73 knock-in (KI) mice. While limited or no effect was observed for Sym024 as single treatment, combining Sym024 with anti-PD-1 treatment enhanced anti-PD-1 effects (Figures 5F and 5G and Supplementary Figure S9). Treatment with anti-PD-1 antibody Sym021, which is cross-reactive to mouse PD-1 (25), led to tumor growth regression in eight of ten MC38-huCD73 tumors engrafted on huCD73 KI mice, and complete response (CR) in four of ten tumors. Combined Sym021 and Sym024 treatment led to tumor regression in all mice and CR in eight of ten huCD73 KI mice (Figure 5F).

Treatment with pembrolizumab led to tumor regression in two of ten and CR in one of ten huPD-1/huPDL-1/huCD73 mice engrafted with CT26-hCD73/hPDL-1. Combined pembrolizumab and Sym024 treatment led to tumor regression in six of ten mice and CR in two of ten mice and significant CD8+ and CD8+/Granzyme B+ T cell infiltration (Figure 5G).

Sym024 has favorable PK properties and is well-tolerated in cynomolgus monkeys

Safety and pharmacokinetics of Sym024 were assessed in cynomolgus monkeys. Anti-drug antibody (ADA) responses in monkeys receiving Sym024 emerged after the second dose, and ADA serum titers generally increased with increasing number of doses with an inverse dose-relationship. In the 10 mg/kg dose group ADA responses were detected in 11 of 12 animals, while only 3/6 were ADA positive in the 30 mg/kg group. In the 100 mg/kg dose group no ADA was detectable in any of the 10 animals during the dosing phase but became detectable in all four recovery animals at end of the 8-week treatment-free period. ADA emergence generally resulted in decreased exposure and increased clearance of Sym024 in monkeys.

Pharmacokinetic (PK) elimination curves following first or single infusion of Sym024 showed dose-proportional PK profiles consistent with elimination of IgG1 antibodies (Supplementary Figure S10A). Following repeat dosing, the mean accumulation ratio in ADA-negative animals ranged from 1.7 to 2.4 for AUC_0-168h (Supplementary Figure S10B). The Sym024 serum half-life during the recovery phase was 6 days for the 10 mg/kg dose (n= 1) and 8 days for the 100 mg/kg dose (n= 4; range: 6 to 10 days), likely reflecting the half-life at saturation, with only minor impact of target-mediated clearance. Using one-species allometric scaling, cynomolgus monkey PK parameters were used to predict human PK; the human PK has since been consolidated in a first clinical study with Sym024 (NCT04672434).

Sym024 was well tolerated in cynomolgus monkeys. There were no significant toxicologic observations or changes in hematology, blood chemistry, cytokine profiles, or urinalysis during or after once-weekly administration of Sym024 at 10, 30, or 100 mg/kg for 4 weeks. No late-onset toxicity was observed during the 8-week dosing-free recovery period, and 100 mg/kg was established as the no observed adverse effect level.

Discussion

Many cancer patients remain unresponsive to treatment with immune checkpoint inhibitors, likely due to additional TME immune suppressive mechanisms such as adenosine signaling (5). CD73 inhibition in combination with check-point inhibitors may be an effective approach to immune-oncology therapy (48). However, enzymatic inhibition might require a more comprehensive block than check point inhibition, since even residual enzymatic activity could result in accumulation of suppressive adenosine product.

Here, we identified Sym024, a mAb capable of blocking both soluble and cell-bound CD73. CD73 is a homodimer and requires active swiveling through two states for activity (49); it may thus offer unique targeting opportunities for bivalent antibodies due to the double set of available identical epitopes.

Sym024 comprehensively inhibited CD73-mediated catalysis of AMP to adenosine at high antibody-to-enzyme ratios, effectively alleviated CD73-driven inhibition of T cell proliferation and displayed greater efficacy than a panel of clinically relevant anti-CD73 benchmark antibodies across a range of enzyme expression levels (20 cell lines of diverse origin). In the context of extended in vitro incubation, emulating the TME, the difference in efficacy between Sym024 and benchmarks was substantial.

Both Sym024’s efficacious inhibition of sCD73 at high antibody-to-enzyme ratios and its enduring cell surface CD73 inhibition, suggest a unique molecular mode of action likely depending on the specific structural interplay between Sym024 antibody and enzyme.

SPR, SEC-MALS, and cryo-EM, provided deeper insights into the mechanism underlying CD73 Sym024 inhibition. Sym024 was shown to interact one-to-one (1:1) with unique epitopes on the CD73 homodimer N-termini, favoring an effective bivalent binding to the open conformation. Structurally, Sym024 forms an intradimer bridge to effectively lock CD73 movement and catalytic activity.

While all Sym024-CD73 complexes were shown to be exclusively of 1:1 stoichiometry, the benchmark antibodies formed either higher-order oligomerization complexes (oleclumab, uliledlimab, drebuxelimab) or a mix of oligomers and 1:1 complexes (IPH5301 and mupadolimab). In line with our data, oligomerization has previously been demonstrated to inhibit soluble CD73 only at low antibody-to-enzyme ratios (28). This is likely due to a shift towards ineffective monovalent CD73 occupancy which does not fully hinder movement of the enzyme at higher ratios (Figure 6).

Figure 6. Bivalent one-to-one interaction with CD73 is an effective inhibitory mechanism across all antibody-to-enzyme ratios.

Hypothetical illustration of antibody-enzyme complexes showing different inhibitory modalities. At very low antibody concentrations (bottom), no antibodies bind or inhibit CD73. At near-equimolar ratios (center), all anti-CD73 antibodies tested, except for Sym024, display some propensity to form antibody-enzyme oligomers with resulting enzymatic inhibition. At excess antibody ratios (top), benchmark antibodies exert less efficacious inhibition, likely due to an ineffective monovalent binding modus. This is not observed for Sym024, which our data suggests retains the 1:1 interaction with CD73. The illustration is limited to key representative binding modi; the full population is comprised of several complexes, depending on epitopes and affinities. Benchmark antibodies are shown in maroon, Sym024 in green, surface model of CD73 in grey (PDB: 7PBY). Created in BioRender. JAKOBSEN, J. (2025) https://BioRender.com/6tqd5hb

Our data suggest that a 1:1 stoichiometry is also superior for inhibition of cell-surface CD73 for all antibody-to-enzyme ratios tested. This might be explained by sterical challenges in including all membrane-tethered enzymes in oligomerization complexes. Interestingly, 1:1 interaction was previously identified as a possible binding modus for IPH5301 (53). Here, however, we found this to be only one of the observed modalities for IPH5301, emphasizing the importance of interrogating the entire antibody-enzyme complex population. Additionally, IPH5301 bind very distinct epitopes from Sym024, likely mandating the locking of an intermediate state, not the open confirmation.

In mouse models of human cancer, Sym024 inhibited CD73 enzyme activity, induced tumor infiltration of immune cells, and inhibited tumor growth both alone and in combination with an anti-PD-1 check-point inhibitor. In vivo investigations to explore Sym024’s effect on human tumor cells were impeded by a lack of Sym024 cross-reactivity to mouse CD73; thus, in mice carrying human tumor cells, the murine tumor stroma retained CD73 activity, while CD73 activity in human tumor cells was inhibited with high levels of Sym024. In immunodeficient mice, however, Sym024 almost abolished MDA-MB-231 tumor growth, mirroring previous observations with mouse CD73-inhibiting oleclumab (55). Further, Sym024, and to some extent mupadolimab, which is also not mouse cross-reactive (56), proved superior in maintaining tumor growth control after treatment withdrawal. This suggests that in this model the (human) tumor CD73 component is predominant and that inhibiting CD73 has a direct, tumor-intrinsic effect (even if we cannot exclude some contribution from relief of adenosine suppression of mouse NK and myeloid cells); this may confound interpretation when interrogating the immune-stimulatory effect of CD73. We thus used models less prone to this intrinsic effect for in vivo immune-effect investigations of Sym024 as monotherapy (human A375 and Calu-6) or in combination with anti-PD-1 (mouse CT26 and MC38). Sym024 impeded tumor growth as monotherapy and enhanced PD-1 blockade effect, in line with previous reports (18,19,57).

By parallel screening for inhibition of both cell-surface CD73 and sCD73 at high antibody concentrations, and thus high antibody to enzyme ratios, we avoided selecting for antibodies with oligomerization as the primary mode of action. In contrast, oleclumab was identified by screening for CD73 internalization, likely explaining its predisposition towards oligomerization, which is known to induce cellular uptake (55). This predisposition, together with a strong monovalent affinity, may augment the “binding-site barrier” effect and impede core tumor exposure (58), as well as lead to increased internalization-driven, target-mediated drug disposition. Further, for full inhibition of adenosine signaling, oleclumab has been shown to require supplementation with small molecule inhibitors, reinforcing the concept of oligomerization as a less effective inhibitory mechanism (59,60). Interestingly, oleclumab and Sym024 binding epitopes partially overlap (at the N-terminal domain distant from the catalytic site, in line with their non-competitive mode of action) (28) (Supplementary Figure S5), supporting the notion that diverse modes of action may arise from minute differences in binding orientation.

Conclusion

In this study, we showed Sym024 to be an effective inhibitor of both sCD73 and cell-bound CD73 enzymatic activity, even at high CD73 expression levels and extended incubation periods. Since even residual CD73 activity may lead to adenosine accumulation, the comprehensive Sym024-mediated inhibition might be key to achieving sufficient blockade of this immune suppressive signaling pathway, especially in the many cancers displaying high CD73 expression.

Sym024 (S095024) was recently investigated for clinical safety as a monotherapy and in combination with anti-PD-1 (Sym021) (NCT04672434) and is currently under investigation in a phase 1b/2 trial in combination with the anti-PD-1 compound cemiplimab (Regeneron) (NCT06162572). Data from this study will indicate whether Sym024’s robust pre-clinical CD73 inhibition translates into clinically meaningful improvement of the anti-neoplastic immune response.

Abbreviations list

AMP

Adenosine monophosphate

ATP

Adenosine triphosphate

ADA

Anti-drug antibody

AUC

Area under the curve

CR

Complete response

CDRs

Complementarity-determining regions

Cryo-EM

Cryogenic electron microscopy

CTLA-4

Cytotoxic T lymphocyte antigen 4

K_D

Dissociation constant

K_D,app

Dissociation constant, apparent

ECD

Extracellular domain

Fc

Fraction crystallizable

GLP

Good laboratory practice

huCD73

Human CD73

hPDL-1

Human PD ligand-1

HuGEMM

Humanized genetically modified mouse models

IFNg

Interferon gamma

KI

Knock-in

MLR

Mixed lymphocyte reaction

mAb

Monoclonal antibody

SEC-MALS

Size-exclusion chromatography (SEC) with Multi-angle static light scattering (MALS)

PBMC

Peripheral blood mononuclear cell

PK

Pharmacokinetic

PD-1

Programmed cell death protein 1

sCD73

Soluble CD73

SPR

Surface plasmon resonance

TME

Tumor microenvironment