Targeting Adenosine with Adenosine Deaminase 2 to Inhibit Growth of Solid Tumors

The tumor microenvironment (TME) has several mechanisms that can suppress immune cell activity and cause resistance to therapy in solid cancers (1, 2); the accumulation of extracellular adenosine is potentially one such mechanism. Accumulation of extracellular adenosine may result from events associated with tumor growth, such as increased metabolic activity, hypoxia, inflammation, and cell death, and is a characteristic of the microenvironment of many human solid tumors (3, 4). Adenosine triphosphate released in the TME by apoptotic cells, or by cells undergoing high metabolic activity or stress, may be converted to adenosine by ectonucleotidases, resulting in the accumulation of extracellular adenosine (3, 4).

Extracellular adenosine can bind and activate four G-protein-coupled receptors: A_1R, A_2AR, A_2BR, and A_3R (5). Nonclinical and clinical evidence suggest that adenosine signaling in immune and tumor cells promotes tumor growth and metastasis (4). Deletion of A_2AR in natural killer (NK) cells, T cells, and antigen-presenting cells (APC), as well as systemic or local treatment with A_2AR antagonists, reduces growth and metastasis of syngeneic melanoma, lung, bladder, and breast tumors in in vivo murine models (6–10). A_2BR on APCs and tumor cells has been associated with the growth and metastasis of bladder and breast tumors (7, 9). In fact, A_2BR signaling in APCs intrinsically regulates tumor antigen-specific CD8⁺ T-cell responses and promotes growth of solid tumors (11). It has also been suggested that adenosine signaling via A_1R, A_3R, or intracellular metabolic pathways involving adenosine kinase and 5′ AMP-activated protein kinase (AMPK) activation may delay the growth of colon carcinoma, mesothelioma, and breast tumors (12–15). Also, recent evidence suggests the requirement of A2AR signaling in endogenous CD8⁺ T cells for their maintenance and responsiveness to immune-checkpoint blockade (6, 16–18). Therefore, the overall impact of adenosine accumulation on tumor growth, rather than on particular receptors, is poorly understood.

Adenosine deaminase (ADA) exists mainly as two isoforms in humans: ADA1, encoded by the ADA gene, a predominantly intracellular monomeric enzyme that catalyzes the deamination of both adenosine and deoxyadenosine; and ADA2, encoded by the CECR1 gene, a serum dimeric enzyme expressed by multiple cell types (19), which mainly catalyzes adenosine deamination (20). ADA2 has optimal activity at pH 6.5 (19), which is a pH level commonly found in many solid tumors. In addition to an active site, ADA2 has a heparin-binding region and a putative growth factor-like domain (19). The K_m value of ADA2 is higher than ADA1 (21), suggesting the affinity of ADA2 to adenosine is lower than that of ADA1. Notably, both serum and tumor levels of ADA2 are elevated in patients with breast cancer (22, 23). Overall, these qualities suggest a role for ADA2 in the pathologic conditions associated with elevated extracellular adenosine rather than steady-state conditions. However, it is not known whether ADA2 expression is prognostic in patients with cancer or whether treatment with ADA2 can influence tumor growth or metastasis.

Here, we show that ADA2 gene (CECR1) expression is associated with favorable outcomes in several solid tumor indications as well as changes in immune responses, suggesting that ADA2 may be a promising anticancer immunotherapy candidate. To develop ADA2 as an anticancer immunotherapy, we engineered ADA2 in two respects. First, we increased ADA2's circulating half-life through PEGylation and, second, we increased ADA2's catalytic activity by introducing two key mutations in the catalytic site. We performed reverse translational experiments in vitro using mouse and human cells and in vivo using syngeneic mouse models in preclinical settings. We show that treatment with engineered ADA2 inhibits tumor growth and metastasis by modulating adaptive immune system responses, and that combining ADA2 with immune checkpoint inhibitors more effectively blocks tumor growth in different tumor models.

An A375 melanoma cell line (CD73^high/CD26⁻) was obtained through the ATCC. Cell lines were sourced as follows: MB49 bladder carcinoma cell line from Dr. Timothy Ratliff of Purdue University, West Lafayette, Indiana. Metastatic breast adenocarcinoma cell line 4T1, expressing luciferase enzyme were from Tufts University (Medford, Massachusetts), EMT6 murine breast cancer and CT26 murine colon cancer cell lines [ATCC; authenticated by short tandem repeat (STR) DNA profiling], and MC38-ev murine colon cancer cell line (Mark J. Smyth Lab; QIMR Berghofer Medical Research Institute, Brisbane, Australia).

All cell lines were maintained in complete RPMI medium or DMEM, supplemented with 10% FBS, and were routinely tested using MycoAlert Mycoplasma Detection Kit (Lonza) and polymerase chain reaction (IDEXX). Only low-passage (<3) cell lines were used for the study.

C57BL6 and BALB/c female mice, 5 to 6 weeks of age, were purchased from Taconic and Charles River Laboratories, and handled in accordance with approved Institutional Animal Care and Use Committee protocols. Mice were subcutaneously inoculated with MB49, CT26, or MC38-ev cells into the right flanks, or orthotopically inoculated with EMT6 cells into the ninth mammary fat pad. When tumors reached 100 to 200 mm³, mice were staged to receive the intravenous vehicle control and PEGADA2 variants at multiple doses twice weekly for 2 to 3 weeks. For CD8⁺ T-cell depletion, mice bearing EMT6 tumors were treated by intraperitoneal injection with 200 μg anti-mouse CD8α antibodies (Clone L3; Bio X Cell) or Rat IgG2b isotype controls. PEGADA2^HCA treatment (3 mg/kg) was performed by intraperitoneal injection biweekly in the first week and three times a week in subsequent weeks. 4T1-Luc cells (10⁶) were injected intravenously via tail vein into Balb/c mice. PEGADA2^HCA treatment was performed twice weekly for the first week, then every other day until mice were analyzed for lung metastases. Intraperitoneal injection of 150 mg/kg D-luciferin was performed to measure luciferase activity by IVIS imaging (PerkinElmer) as a measure of lung dissemination.

Tumor growth was tracked using an electronic caliper (Vernier Software & Technology, Beaverton). Tumor volume in mm³ was calculated using the formula: tumor volume = ½ [length × (width)²]. Body weight (BW) was measured as percent change as compared with baseline was calculated for each individual mouse by BW/Baseline BW (BW before the first dose) × 100. Terminal bleeding was performed to assess plasma levels of the alanine aminotransferase (ALT), creatinine, blood urea nitrogen (BUN), and glucose from plasma using LIASYS clinical chemistry analyzer. To assess the effects of PEGADA2 and anti-PD-L1 on tumor growth, mice with ∼100 mm³ MB49 or MC38-ev tumors were randomized into treatment groups: in the MB49 model, 3 μg/kg i.v. PEGADA2^HCA, and 0.3 mg/kg i.p. anti-PD-L1 mouse immunoglobulin G2b (Rat IgG2b, κ isotype; clone 10F.9G2; Bio X Cell) were administered alone or in combination; in the MC38-ev model, 3 μg/kg i.v. PEGADA2^HCA and 1 mg/kg i.p. anti-PD-L1 mouse IgG2b were administered alone or in combination (from one experiment, n = 8/group). Anti-mouse PD-L1 (clone 10F.9G2) and anti-mouse PD-1 (RMP1–14) were purchased from Bio X Cell and were diluted in phosphate-buffered saline (PBS) before administration.

ADA2 variants were designed on the basis of its structure–function relationship [Protein Data Bank (PDB) code: 3LGD; ref. 19]. Surface residue K374, residues R222, L224, E133, D416, S265, H267, and those in proximity to ADA2 active site were selected to introduce single or multiple mutations to either improve catalytic efficiency or to deactivate ADA2 enzymatic activity for mechanism of action studies. The Phyre 2 server was used to model the structures of ADA2 variants. Modeled structures were superimposed on the structure of ADA2 WT complexed with the adenosine transition state analog coformycin (PDB 3LGG) using the PyMOL Molecular Graphics System (https://www.pymol.org/sites/default/files/pymol_0.xml), Version 2.0 (Schrödinger, LLC).

Variants were generated using the QuickChange Lightning Multi Site-Directed Mutagenesis Kit (catalog no. 210518; Agilent Technologies) and were expressed in Chinese hamster ovary-suspension cells. FLAG-tagged ADA2 variants were purified using an anti-FLAG M2 affinity resin (catalog no. L00432; GenScript). The untagged ADA2 variants, NME8058 and NME8062, were purified by cation-exchange chromatography (CM-650M; Tosoh Bioscience LLC), followed by hydrophobic interaction chromatography (Butyl-650S; Tosoh Bioscience LLC). The butyl elution pool containing the ADA2 enzymes were exchanged into 1× PBS buffer (10 mmol/L PO₄³⁻, 137 mmol/L NaCl, and 2.7 mmol/L KCl, pH 7.4). Once purified enzymes were tested for enzymatic activity, purity by reversed-phase high-performance liquid chromatography (HPLC) and endotoxin level by Limulus amebocyte lysate (LAL) test cartridges (catalog no. PTS20F; Charles River Laboratories).

ADA2 was PEGylated using 20 kDa PEG (JenKem Technology USA, Inc.) in a 15:1 molar ratio of PEG to ADA2 at 4°C for 18 hours, and the reaction was terminated by 25 mmol/L glycine. PEGADA2 was purified using hydrophobic interaction chromatography (GE Phenyl Sepharose HP; GE Healthcare). After purification, the endotoxin level was retested for PEGADA2 as described. The elution pool containing PEGADA2 was buffer-exchanged into 1× PBS, and retested for enzymatic activity, protein concentration, and endotoxin level before the material was released for in vitro and in vivo studies. Completion of the PEGylation reaction was verified by SDS-PAGE analysis.

Enzymatic activities of 5 μg/mL PEGADA2 variants were tested using the ADA Assay Kit (Cat. BQ014EALD; GenWay Biotech). Table 1 lists the ADA2 variants used in this study and their key properties. Variants with similar enzymatic activity (Supplementary Figs. S1 and S2) were used interchangeably for some studies, due to the limitations of material availability.

A375 melanoma cells (5 × 10⁴/0.5 mL) were incubated (60 minutes, 37°C) in AIM V serum-free medium in the presence of 200 μmol/L AMP (Sigma-Aldrich), with or without ADA2 variants (1.0, 2.5, 5.0, and 10.0 μg/mL). After incubation, supernatants were immediately stored at −80°C or processed for HPLC analysis using a Waters Alliance 2965 Separation Module (Waters Corporation). Supernatants (100 μL) were precipitated with 400 μL of cold methanol (VWR Chemicals) and filtered with Phree phospholipid removal tubes (Phenomenex). Samples were dried overnight in a vacuum desiccator and reconstituted in 100 μL of ultrapure Milli-Q water (Merck Millipore) before injection (50 μL) into the HPLC system. Separation of AMP, adenosine, and inosine was performed using a method based on a binary mobile phase, which consists of 7 mmol/L ammonium acetate (Buffer A; Sigma-Aldrich), pH 3, and acetonitrile (Buffer B; VWR Chemicals) at a flow rate of 1 mL/minute and UV detection set at 260 nm. Peak identities were confirmed by using reference standard compounds, and μmol/L concentrations of adenosine were inferred by comparing the peak area of samples with the calibration curve for the peak areas of adenosine standard. Levels of AMP, inosine, and adenosine were analyzed using HPLC to assess whether PEGADA2 effectively depleted adenosine in a tumor-like environment in vitro.

In vivo adenosine measurement studies were performed by Charles River Laboratories. Briefly, in mice, a microdialysis tumor probe with a 4 mm polyacrylonitril membrane (NO-PAN; Charles River Laboratories) was positioned in CT26 tumors when tumor size reached 250 mm³; 1 day after surgery, baseline samples were collected. After the treatment of 3 mg/kg PEGADA2^HCA, samples were collected in 30-minute intervals for 5 hours and also collected after 24 hours. HPLC with MS-MS detection using 13C5-adenosine as internal standard (IS) to assess adenosine concentrations.

Human CD8⁺ T cells were isolated from fresh peripheral blood mononuclear cells (PBMC; from the San Diego Blood Bank) using the Human CD8⁺ T Cell Enrichment Kit (StemCell Technologies) and resuspended in 10 ml PBS containing 2 μmol/L CellTrace Violet proliferation dye (Thermo Fisher Scientific). They were then incubated for 15 minutes at 37°C and 5% CO₂ in the dark, quenched for 5 minutes with 5 mL FBS, and washed and seeded at 3 × 10⁵ cells/well in RPMI medium with 10% FBS supplemented with nonessential amino acids, 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid, and sodium pyruvate. After addition of PEGADA2, cells were treated with 500 μmol/L adenosine and with Human T-Activator Dynabeads (Life Technologies) coated with anti-CD3/anti-CD28 (1:2 bead:cell ratio) for 72 hours. Adenosine (500 μmol/L final) was added every 24 hours. At the end of the incubation proliferation was analyzed using the Novocyte Flow Cytometer (ACEA Biosciences).

Tumors and spleens were kept in PBS on ice during processing. Tumors were minced, placed in 2.5 mL RPMI medium containing 0.2 Wünsch units/mL Liberase DH and 100 Units/mL DNaseI (Roche), and dissociated using the GentleMACs Octo Dissociator (Miltenyi Biotec). Samples were dissociated using the pre-installed program 37C_m_TDK_1, then filtered through a 70 μmol/L cell strainer to remove any remaining large tissue fragments and washed with 10 mL RPMI medium. Cell suspension was then centrifuged at 300 × g for 7 minutes and the cell pellet was resuspended in 2 mL of staining buffer (2% FBS in PBS with 0.05% sodium azide). Spleens were dissociated with a syringe plunger, filtered through a 70 μmol/L cell strainer with 2% FBS in PBS, washed, and resuspended in 5 ml of staining buffer.

For each staining panel, tumor (200 μL) and spleen (50 μL) samples were transferred to 96-well round bottom plates, centrifuged at 500 × g for 3 minutes, and resuspended in 50 μL of antibody cocktail (see Supplementary Table S1) in staining buffer containing 0.5 μL/sample purified mouse Fc-blocking antibody (eBioscience) and 3 μL/ml Near InfraRed Fixable viability dye (Thermo Fisher Scientific). Plates were incubated on ice in the dark for 20 minutes, then centrifuged. Cell pellets were resuspended in 100 μL of fixation/permeabilization buffer (eBioscience) and incubated at room temperature for 10 minutes. Samples were then resuspended in staining buffer [1× permeabilization buffer for the lymphoid panel (eBioscience)] and intracellular staining was performed.

Flow cytometry was conducted using a Novocyte 3-laser flow cytometer and analyzed using NovoExpress software (ACEA Biosciences). The gating strategy used for the flow cytometry analysis is shown in Supplementary Fig. S3.

C57BL/6 mice were injected with 4% thioglycolate medium, and peritoneal exudates containing macrophages were collected 3 days later. Adherent macrophages cultured for 3 hours in tissue culture plates with RPMI with 10% FBS were detached and seeded at a density of 10⁵ cells/well in 96-well flat-bottom plates. Cells were stimulated with 0.1 μg/mL ultra-purified LPS-EB (InvivoGen) with or without 50 μmol/L adenosine, in the presence or absence of varying concentrations of PEGADA2 variants for 24 hours. LPS alone and no-add groups were used as positive and negative controls, respectively. In some experiments, CPI-444 (Chemgood LLC), an adenosine A2 receptor antagonist was used. After LPS stimulation, cell culture supernatants were collected, and concentrations of IL12p70 and IL10 were determined using the U-PLEX platform (MesoScale Discovery).

MDA-MB-231 breast cancer cell line (ATCC) was maintained in tissue culture as an adherent monolayer in RPMI with L-glutamine and 10% FBS at 37°C and 5% CO₂. On the day prior to the experimental procedure, cells were detached with trypsin and ethylenediaminetetraacetic acid, counted using a Nexcelom Cellometer Auto2000 (Nexcelom Bioscience), diluted to 1 × 10⁶ cells/mL, and seeded at 1 × 10⁵ cells/well in flat-bottom 96-well plates. Total human T cells were isolated from frozen PBMCs (from San Diego Blood Bank) using T cell enrichment kits (StemCell), counted, seeded at 5 × 10⁵ cells/well in flat-bottom 96-well plates in the presence of 2.5 × 10⁵ Dynabeads Human T-Activator CD3/CD28, and incubated for 24 hours at 37°C and 5% CO₂. For both cell types, culture medium and activation beads were removed from the wells. Cells were rinsed with PBS and resuspended in 50 μmol/L rolipram or PBS (test and control, respectively). Samples were incubated for 30 minutes (37°C, 5% CO₂), then PEGADA2 100 to 0.001 μg/mL was added followed by 50 μmol/L adenosine in PBS. Cells were incubated again, washed with PBS, and resuspended in lysis buffer provided with the Cyclic AMP XP Assay Kit (Cell Signaling Technology), and assayed for cAMP content with an ELISA. For MDA-MB-231, 1 μmol/L of adenosine receptor agonists 2′MeCCPA (A_1R), CGS21680 (A_2AR), BAY 60–6583 (A_2BR), and 2-Cl-IB-MECA (A_3R), and adenosine analog N-ethylcarboxamidoadenosine (control) were added to respective wells and incubated for 15 minutes at 37°C and 5% CO₂, followed by cAMP ELISA as described. Adenosine receptor antagonists (1 μmol/L), PSB36 (A_1R), SCH58621 (A_2AR), PSB603 (A_2BR), MRS1523 (A_3R), and XAC (nonspecific adenosine receptor antagonist control) were added to respective wells in the presence of 50 μmol/L adenosine and incubated for 15 minutes at 37°C and 5% CO₂, followed by cAMP ELISA. Rolipram (50 μmol/L) was added to the cells before adenosine receptor activation to allow cAMP accumulation.

Survival analysis of high versus low ADA2 expressors for different solid tumor indications was performed using OncoLnc tool (from http://www.oncolnc.org; ref. 24) and bioinformatics analysis for differential gene expression in lung and breast cancer was performed by Rancho BioSciences using The Cancer Genome Atlas (TCGA) database. Briefly, TCGA RNA-sequencing (RNA-seq) data for breast cancer and lung adenocarcinoma were downloaded from the Genomic Data Commons Data Portal (https://portal.gdc.cancer.gov/repository) using the ‘gdc-client’ (RNAseq measurements) and ‘gdc-tsv-tool.py’ (clinical data) command-line tools. Datasets were assembled by simplifying clinical sample annotation and labeling samples (e.g., BRCA_F_WH_IIA_C50.9_1, cancer_gender_race_cancer stage_tissue code_number for individual samples), filtering samples not fully annotated for these attributes, removing normal samples and genes with low expression, and keeping genes with count >5 in ≥260 (lung) or 550 (breast) samples. The assembled datasets were used in subsequent analyses.

To identify differential gene expression patterns for patients with ADA2^high versus ADA2^low, RNA-seq data from the tumors of patients with breast and lung cancer were downloaded from TCGA database. The top and bottom 30% CECR1 gene (the gene coding for ADA2) expressers, based on RNA counts (ADA2^high vs. ADA2^low, respectively), were compared using Morpheus software (Morpheus, http://software.broadinstitute.org/morpheus). Gene expression patterns for each individual patient were adjusted using z-score (minimum value of −2, maximum value of 2) and grouped as CECR1^high versus CECR1^low patient groups. Differential gene expression analysis was performed between these two groups. Pathway analysis was performed on the lists of genes showing significant difference [P < 0.05 and Log2FC>1 (i.e., an average change of two-fold or more)] between the two conditions, using the hypergeometric test in R's clusterProfiler-package and the Gene Ontology-Biological Pathway (GO-BP; refs. 25, 26).

The correlation of ADA2 expression with T-cell infiltration into tumors and with T-cell activation was analyzed using GEO dataset GSE58644 (27, 28). T-cell infiltration and costimulation scores for each patient were calculated using published gene signatures (29, 30), as the sum of z scores of all the genes in the signature.

Survival analyses for lung and breast cancer, including CD73/CD39 stratification, and breast and colon cancer subtypes based on the expression of ADA2 were performed using the KMPlotter database (27, 31) and the GEO dataset, GSE39582 (32), respectively. Significance was calculated using Log-rank test.

We analyzed the association of survival with CECR1 expression in patients with breast cancer, lung adenocarcinoma, sarcoma, melanoma, pancreatic cancer, or kidney cancer, using data from TCGA database, and observed that those with high CECR1 mRNA counts (designated ADA2^high; i.e., the top 30%) had significantly longer survival than patients whose tumors exhibited low ADA2 mRNA counts (ADA2^low; i.e., the bottom 30%; Fig. 1A). In an analysis of the Gene Expression Omnibus (GEO; ref. 33) dataset GSE39582, which has gene expression-profiling data for five different subtypes (32), higher ADA2 expression was found to predict better survival in the C2 and C5 molecular subtypes. Although the C5 tumors were characterized by high chromosomal instability, frequent mutations in KRAS and TP53, and activated Wnt signaling, the C2 subtype was characterized by high microsatellite instability and increased immune system activation (Fig. 1A). We also observed increased survival in patients with breast and lung cancer based on higher ADA2 expression, as assessed using microarray-based datasets (Supplementary Fig. S4; ref. 31). Association between elevated ADA2 expression and increased survival was particularly substantial among patients with triple-negative (basal) and HER2+ breast cancer subtypes (Supplementary Fig. S4), two of the most aggressive subtypes of breast cancer.

The K_m value and catalytic activity of ADA2 is low (21). Therefore, we introduced single-point mutations in the ADA2 catalytic site to increase its catalytic activity. Engineered ADA2 variants with FLAG tags were expressed in CHO-S cells, purified by anti-FLAG resin, screened by activity assay, and ranked on the basis of their activity. Among tested mutants, the single mutants, R222Q and S265N, showed significant improvement in kinetic parameters compared with the wild-type enzyme (ADA2 WT; Fig. 1B). After combining these mutations (R222Q/S265N), the catalytic efficiency increased 16-fold in comparison with ADA2 WT (Table 1). Therefore, the double mutant (NME62) was selected for further biological characterization. The triple mutant, R222Q/S265N/K374D (NME58), with an additional mutation in its heparin binding site had similar activity to NME62 and was used in some experiments. A variant with double mutation, E333A/D416V (NME133), had no detectable activity in a screening assay and was used as a negative control in biological mechanism-of-action studies (Supplementary Table S2).

A structural analysis approach was used to help determine the cause of the significant improvement in the catalytic efficiency of NME62 and NME58. We observed two major differences when a structural model of NME62, predicted by the modelling server Phyre 2 (34), was compared with the structure of ADA2 WT complexed with the substrate, the adenosine transition state analog coformycin. The first difference observed was in the secondary structure changes in the active site region containing residues 81 to 89. D89 is a key residue that contributes to ADA2 catalysis, and H86 and H88 are the zinc-coordinating residues of ADA2. This region forms a β strand in the ADA2 WT-coformycin complex structure, whereas the same region was predicted to be a loop by Phyre 2 in NME62 (Fig. 1C). This indicates a possible increase in flexibility of the NME62 active site with a subsequent impact on substrate binding and turnover rate. The second difference was that two newly formed H-bonds were observed between the Q222 residue and the substrate coformycin in the structure model of NME62; in comparison, H-bonds were not observed between the R222 residue and the coformycin in the ADA2 WT. This may also contribute to changes in substrate binding (Fig. 1C).

Because all ADA2 variants R222Q/S265N and R222Q/S265N/K374D, with and without FLAG tags, had comparable enzymatic activity, we named them collectively “high catalytic activity” variants to simplify nomenclature. High catalytic activity variants are abbreviated as ADA2^HCA or PEGADA2^HCA for the PEGylated version (Supplementary Table S2).

The results of the pharmacokinetic study showed that both half-life and AUC of PEGADA2^HCA were significantly increased compared with non-PEGylated ADA2 WT (Supplementary Fig. S5), indicating successful pharmacokinetic (PK) improvement with PEGylation.

CD73 is expressed in the TME on the cell surface and converts adenosine monophosphate (AMP) to adenosine (35). We treated a CD73⁺ A375 melanoma cell line with AMP to simulate an adenosine-generating TME, and compared PEGADA2 and PEGADA2^HCA with respect to the depletion of tumor generated adenosine. PEGADA2 and PEGADA2^HCA each prevented extracellular adenosine accumulation in supernatants of CD73⁺ A375 melanoma cells treated with AMP. Notably, at least 10-fold less PEGADA2^HCA than PEGADA2 was required to prevent adenosine accumulation (Fig. 1D).

Previous studies have indicated that adenosine plays a role in the aggressiveness of breast tumors by directly stimulating adenosine receptors (9). In a test of PEGADA2's effect in human breast tumors, we confirmed that MDA-MB-231 human triple-negative breast cancer cell line (TNBC) accumulated cyclic AMP (cAMP) in response to adenosine signaling (Supplementary Fig. S6), and established that PEGADA2^HCA more effectively reversed cAMP accumulation than PEGADA2 in MDA-MB-231 cells stimulated with adenosine (Fig. 1E). These results suggest that engineered PEGADA2^HCA effectively reduces tumor-generated adenosine, which can translate into a decrease in intracellular adenosine signaling.

We evaluated whether PEGADA2 variants prevent the adenosine-mediated polarization of macrophages into an anti-inflammatory phenotype by incubating peritoneal macrophages from C57BL/6 mice with PEGADA2 or PEGADA2^HCA in the presence of lipopolysaccharide (LPS) and adenosine. Adenosine strongly suppressed the release of proinflammatory and immuno-stimulatory cytokine IL12, while increasing anti-inflammatory IL10 production by peritoneal macrophages from C57BL/6 mice in the presence of LPS (Fig. 2A). These adenosine-mediated polarization effects were reversed by both PEGADA2 and PEGADA2^HCA, with PEGADA2^HCA having greater potency (Fig. 2A). PEGADA2 variants demonstrated similar functional activity to the non-PEGylated variants for reversing the effect of adenosine on IL12 and IL10 release from macrophages (Fig. 2B), suggesting that the engineered ADA2 variants retained functional activity after PEGylation. PEGADA2^INACT, an engineered inactive variant with intact growth factor and heparin-binding domains, but no ability to convert adenosine to inosine based on enzymatic activity (Supplementary Table S2), was used as a control for possible nonenzymatic effects. This variant failed to reverse the effect of adenosine on macrophage polarization into an anti-inflammatory phenotype (Fig. 2C). Figure 2D shows that, at the same concentrations, PEGADA2^HCA but not the small molecule adenosine receptor blockade CPI-444, completely reversed adenosine inhibition of IL12 release. Collectively, these results suggest that PEGADA2 effectively reverses adenosine suppression of macrophages in an enzyme-activity dependent manner and this effect is not influenced by PEGylation.

Previous studies have shown that the MC38 and CT26 murine colon cancer models are responsive to checkpoint inhibitors (36). These tumors have very low levels of adenosine converting enzyme CD73; however, they have been shown to be sensitive to CD73-generated adenosine from nontumor resources such as regulatory T cells and stromal cells (37, 38). Nonmetastatic EMT6 and highly metastatic 4T1 orthotopic breast tumor models represents CD73⁺ TNBC subtypes (39–41) and found to be responsive to adenosine receptor/CD73 blockade (42). Therefore, we have tested the effect of PEGADA2^HCA on tumor growth on MC38, CT26, and EMT6 models, and tested lung dissemination using 4T1 model. Treatment of MC38 tumor-bearing mice with PEGADA2^HCA significantly reduced tumor growth in a dose-dependent manner. An antibody against programmed cell death ligand-1 (PD-L1) was used as a positive control (Fig. 3A). PEGADA2^HCA also reduced the growth of EMT6 breast adenocarcinoma (Fig. 3B) and CT26 colon carcinoma tumors (Fig. 3C). PEGADA2^HCA significantly reduced adenosine accumulation in the tumor (Fig. 3D). Inhibition of tumor growth was absent in the group receiving the enzyme-inactive variant (Fig. 3E). PEGADA2^HCA treatment significantly reduced the lung colonization of 4T1 tumors (Fig. 3F and G) versus vehicle. These data suggest that PEGADA2^HCA treatment effectively decreases tumor growth by targeting adenosine and reducing metastatic colonization.

We performed differential gene expression analysis between ADA2^high and ADA2^low breast cancer and lung adenocarcinoma using data from the TCGA database. We identified more than 1,000 genes differentially expressed in ADA2^high versus ADA2^low tumors using the criteria of P < 0.05 and log₂FC > 1 (Fig. 4A). More than 50% of the genes and pathways that are significantly higher in the ADA2^high group are common to both lung adenocarcinoma and breast cancer (Fig. 4B–E). Pathway analysis indicated that high ADA2 expression is associated with enrichment of immune-related pathways (Fig. 4C and D). The most significantly enriched pathway is the adaptive immune response pathway for both cancer types in ADA2^high tumors (Fig. 4E). ADA2^high tumors also show high expression of IFNγ gene signatures (Fig. 4B), which predict increased response to checkpoint blockade (43) and low expression of genes associated with tumor progression and metastasis such as TGFBR3 L (44) and STC2 (Supplementary Fig. S7; ref. 45). All the differentially expressed genes and pathways including the common genes and pathways between ADA2high expressors of patients with lung and breast cancer were indicated in Supplementary Table S3.

We tested the effect of PEGADA2^HCA on tumor infiltration of immune cells using syngeneic MC38 and CT26 tumor models 1 day after the third dose, when the sizes of tumors treated with PEGADA2HCA tends to diverge as compared with the vehicle control groups (Supplementary Fig. S8). PEGADA2^HCA significantly increased the density of the CD8⁺ tumor-infiltrating lymphocyte (TIL) population, CD4⁺ TIL, NK cells, and CD103⁺MHCII⁺CD11b^dim type I dendritic cells (cDC1) within MC38-ev tumors, but not in the spleen (Fig. 4F; Supplementary Fig. S9). PEGADA2^HCA did not change the density of tumor- or spleen-associated CD103⁻MHCII⁺CD11b⁺CD11c⁺ myeloid dendritic cell populations (MyDC; cDC2), macrophages, or Ly6G⁺ granulocytic myeloid-derived suppressor cell populations (GrMDSC; Fig. 4F; Supplementary Fig. S9). PEGADA2^HCA also significantly increased both the density and proportion of CD8⁺ TIL, cDC2, and cDC1 populations in a CT26 model (Fig. 4G).

Clinically, high tumor expression of ADA2 in patients with breast cancer was positively associated with the density of TIL and T-cell costimulation scores (Fig. 5A and B, respectively). In a test of the effect of PEGADA2 on direct adenosine signaling in primary human T cells, incubation of prestimulated T cells with PEGADA2^HCA prevented cAMP accumulation after adenosine administration (Fig. 5C). PEGADA2^HCA completely reversed adenosine suppression of T-cell proliferation in a dose-dependent manner (Fig. 5D). Depletion of CD8⁺ T cells from tumor-bearing animals completely reversed tumor growth inhibition (TGI) by PEGADA2^HCA (Fig. 5E). Collectively, these results suggest that tumor enrichment of ADA2 or PEGADA2 treatment can promote adaptive immune responses to inhibit tumor growth.

In both preclinical and clinical setting targeting both adenosine A2 receptors and PD-1 causes combinatorial effect in control of tumor growth (46). Our bioinformatics analysis also indicates that increased ADA2 expression was associated with increased expression of the immunotherapy response genes/signatures (Fig. 4B). Therefore, we tested the effect of PEGADA2HCA alone or in combination with anti-PD-L1 antibody on growth of MC38 and MB49 tumors. Although monotherapy with PEGADA2^HCA or anti-PD-L1 resulted in 39% and 35% TGI, respectively, combining these two agents resulted in 62% TGI (Fig. 5F). Likewise, we evaluated combining PD-L1 blockade with PEGADA2^HCA in MB49 tumors, which are known to be sensitive to adenosine receptor blockade (7). Combining PD-L1 blockade with PEGADA2^HCA resulted in significantly greater TGI, compared with the control group or treatment with single agents alone (Fig. 5G). Overall, these results suggest that enzymatic targeting of adenosine in the TME shows promise as a novel cancer immunotherapy, and this approach can potentially be further improved by increasing the catalytical activity and/or half-life of the enzyme.

When gene expression profiles of microdissected tumor epithelium and stroma from 38 TNBC tumors were analyzed, both ADA1 and ADA2 expression was found to be higher in stroma as compared with tumor cells (Fig. 6A). However, ADA2 expression was higher than ADA1 expression in both stroma and tumor (Fig. 6B). Overall survival of chemotherapy-treated patients with TNBC, based on ADA2 expression levels, was evaluated after stratifying patients as expressing high versus low levels of CD73 and CD39. Higher expression of ADA2 was significantly associated with better survival in only those patients with high levels of CD73 and CD39, but not those with low levels of CD73 and CD39 (Fig. 6C). Similarly, when the chemotherapy-treated patients with non–small cell lung cancer (NSCLC) were stratified on the basis of CD73 and CD39 expression, higher ADA2 levels were significantly associated with better disease-specific survival only in those expressing high levels of CD73 and CD39 (Fig. 6D). Increased ADA1 expression is not correlated with survival in CD73/CD39high or CD73/CD39low subgroups among patients with TNBC or NSCLC (Fig. 6E and F). These results underlie an important role for adenosine and expression of ADA2 in TME.

To test if PEGADA2^HCA treatment can potentially be well tolerated as a potential new therapy we have tested changes in weights and several clinical chemistry parameters in mice, indicative of renal or hepatotoxicity. Treatment of mice with different doses of PEGADA^HCA caused no weight reduction and lead to weight changes similar to that of vehicle group both in MC38-bearing C57BL/6 mice and CT26-bearing Balb/c mice (Supplementary Fig. S10). When terminal bleeding samples from the same mice were analyzed for basic markers of hepatotoxicity or renal toxicity from plasma, we did not observe any significant and/or dose-dependent changes after PEGADA2^HCA treatment as compared with vehicle group (Supplementary Fig. S11).

Adenosine has been associated with both protumoral (35) and antitumoral effects (13–15). Much of the in vivo evidence for pro-tumoral effects of adenosine relies heavily on targeting one or two receptor subtypes, adenosine generation mechanisms, and on evaluating the effects of particular adenosine receptors in knockout animals. Our study provides direct evidence that the extracellular adenosine molecule acts as a tumor promoter. We also introduce engineered ADA2 as a novel cancer immunotherapy, which can inhibit tumor growth by reducing adenosine in an enzyme activity-dependent manner and by modulating immune responses. Reducing adenosine through enzymatic degradation may offer advantages over other single-agent therapies that target the adenosine pathway, such as small molecule blockade of individual adenosine receptors or CD39/CD73 blockade. These potential advantages may arise for three reasons. First, more than one receptor promotes tumor growth through adenosine. Second, blockade of adenosine receptors promotes generation of more adenosine, and finally, there are other pathways that promote adenosine generation along with the CD39/CD73 pathway in the TME (35, 41).

ADA2 catalyzes deamination of adenosine into inosine, a more stable nucleoside than adenosine. Recent studies have indicated that inosine can be an additional carbon source for CD8 T cells in TME (47). Also, inosine generated by microbiome can weakly stimulate adenosine A2A receptors (17), which can provide basal level A2AR signaling required for T-cell maintenance and growth factor responsiveness of CD8⁺ T cells (6, 16, 17). Therefore, ADA2 may have a dual mechanism of action: inhibiting immunosuppressive adenosine signaling by reducing adenosine and increasing CD8+ T cell fitness by generating inosine. These observations provide a rationale to test PEGADA2^HCA in subset of patients with low levels of inosine-generating microbiomes and increased expression of recently discovered adenosine gene signature (46), which indicates activation of immunosuppressive myeloid cells through adenosine signaling.

ADA2 expression among patients with breast, lung, pancreas, melanoma, and sarcoma tumor types; and increased ADA2 expression in a subgroup of patients with colon and kidney carcinoma is associated with improved prognosis. A hallmark of some of these tumors is elevated adenosine levels. Some are known as inflamed tumors, which are associated with increased tumor mutational burden and immune cell infiltration (i.e., lung cancer and melanoma), and some are known to be tumors with limited immune cell infiltration (48, 49). We observed an association between ADA2 expression and survival in all these tumor types. Immune cells, tumor cells and cells adjacent to tumor cells can be sources of adenosine through high expression of CD39 and/or CD73 (50–52). In immunologically active tumors, adenosine generation can be achieved by infiltrating immune cells, causing autocrine/paracrine signaling for suppression, whereas in immune-excluded tumors, adenosine can be generated by tumor cells to suppress the immune responses even at early phases of tumor growth, causing these tumors to stay immunologically dormant. In addition, it is important to note that high ADA2 expression was not associated with prolonged survival in all cancer indications, suggesting that adenosine could couple with other suppressive mechanisms in these indications or play a minor role for tumor growth. Further research is required to contextualize how ADA2 and other sources of adenosine generation, such as nucleoside transporters, may fit in this picture.

Higher expression of CD73 (which converts AMP to adenosine, contributing to adenosine accumulation) has been associated with poor prognosis (51) or decreased survival (53) in patients with some cancers such as TNBC. Our data indicate that increased expression of the adenosine-lowering ADA2 enzyme is associated with a favorable prognosis, providing further evidence that activation of pathways leading to adenosine accumulation (low ADA2, high CD73) is immunosuppressive. Importantly, our analyses further showed that higher expression of ADA2 is significantly associated with better survival in both patients with TNBC and NSCLC with only high CD39/CD73 levels, suggesting a critical role for ADA2 in controlling adenosine levels in the tumor microenvironment. Current approaches targeting adenosine signaling utilize CD39 and CD73 blocking antibodies, and/or small-molecule inhibitors of adenosine receptors (54). Deletion of A_2AR significantly increases CD73 expression, suggesting a potential auto-regulatory loop exploited by tumors for generation of more adenosine (41). There are also overlapping and nonoverlapping tumor-promoting effects of A_2AR and A_2BR (7, 8, 10, 11, 55, 56). Therefore, evidence to date indicates that targeting multiple receptors, or targeting adenosine receptors with CD73 blockade, may result in a stronger effect on tumor growth than targeting single receptors or ectonucleotidases (41). Enzymatic depletion of tumor-associated adenosine with PEGADA2^HCA can be an alternative and more direct approach compared with targeting multiple components of adenosine signaling separately. Importantly, we found that suppression of macrophage activation by adenosine was completely reversed by PEGADA2^HCA but not by small molecule adenosine receptor blockade, and that PEGADA2^HCA alone caused monotherapy effects in a wide dose range and in models refractory to adenosine receptor blockade as a monotherapy, such as CT26 (57). These observations support the potential benefits of targeting the adenosine pathway by PEGADA2 as compared with other therapies. Co-blockade of A2AR and CD73 causes a strong combinatorial effect in control of tumor growth (41). It remains tested whether combinatorial targeting of purinergic signaling with adenosine receptor blockade and/or CD39/CD73 blockade and PEGADA2 will cause a stronger TGI as compared with single approaches alone.

Our analysis of public data suggested ADA2 expression is associated with the enrichment of pathways leading to adaptive immune system activation, increased T-cell costimulation and tumor infiltration by T cells. We observed PEGADA2^HCA-mediated infiltration of CD8⁺ T cells and cDC1 in two different syngeneic tumor models with different genetic backgrounds and showed that CD8⁺ T-cell depletion reversed TGI caused by PEGADA2^HCA. These suggest that the association between ADA2 expression and the enrichment of adaptive immune response pathways in humans can be caused by functional effects of ADA2 expressed in the TME. Patients with ADA2 deficiency syndrome (DADA2) commonly have mutations in the regions coding for the catalytic site (58). Therefore, further epidemiologic evaluation of patients with ADA2 deficiency with respect to cancer incidence and progression, or testing PEGADA2^HCA in the clinic, may shed further light on this association.

PEGylated monomeric isoform of adenosine deaminase (PEGADA1) is used in clinic as an enzyme replacement therapy (ERT) for the rare genetic disease called adenosine deaminase deficiency, which causes severe combined immunodeficiency (ADA-SCID). Both ADA1 and ADA2 have advantages over one another as a potential novel cancer therapy. ADA1 has higher affinity towards adenosine and higher catalytic activity as compared with ADA2. Also, there is more clinical experience with ADA1 as it has been used for decades for ERT. Being an intracellular protein ADA1 lacks a secretion signal sequence and has more potential to have immunogenicity among immunocompetent patient population (unlike patients with ADA-SCID) especially if high doses are required and if combined with other immunotherapies. Unlike ADA2, ADA1 can equally target both adenosine and deoxyadenosine whereas ADA1 has more selective towards adenosine. However, whether dual targeting of adenosine and deoxyadenosine in cancer setting is an advantage or disadvantage should be further investigated. ADA2 has a lower catalytic activity but has several features that position this enzyme well as a novel therapy for cancer such as acidic pH optimum often seen in the tumor microenvironment, being a secreted protein (which promotes its developability as a new molecular entity and reduces risk of immunogenicity), and potential to reduce not steady state but pathological levels of adenosine.

To further develop PEGADA2^HCA as an anticancer immunotherapy, more research is needed to explore dosing regimens and potential therapeutic response to PEGADA2^HCA in human clinical trials. Because high expression of ADA2 itself is potentially prognostic for better survival, and indicative of increased expression of IFNγ-gene signatures in some indications, ADA2 expression could serve as a possible biomarker for patient selection for immunotherapies or therapies targeting adenosine signaling.

L. Wang reports a patent for US 9,969,998 B2 issued. M.J. LaBarre reports other support from Halozyme outside the submitted work. H. Shepard reports employment with Halozyme Therapeutics, Inc. C.D. Thanos reports a patent 9,969,998 issued. O. Sahin reports grants from Halozyme Therapeutics, Inc. during the conduct of the study, personal fees from OncoCube Therapeutics LLC outside the submitted work, and is a co-founder of OncoCube Therapeutics LLC. C. Cekic reports employment and ownership of shares with Halozyme Therapeutics, Inc. No disclosures were reported by the other authors.

Halozyme Therapeutics, Inc. follows policies established by the International Committee of Medical Journal Editors. The studies were conducted by Halozyme Therapeutics, Inc., and the data are held by the company. The availability of the biological material, PEGylated adenosine deaminase 2, will require Material Transfer Agreements through Halozyme Therapeutics, Inc. Additional information about the studies and/or datasets can be obtained by contacting Halozyme Therapeutics, Inc., 11388 Sorrento Valley Road, San Diego, CA 92121. Phone: 1-858-794-8889; Email: publications@halozyme.com.

L. Wang: Conceptualization, resources, investigation, project administration, writing–review and editing. L.M. Londono: Formal analysis, investigation, methodology, project administration, writing–review and editing. J. Cowell: Investigation, methodology, writing–review and editing. O. Saatci: Data curation, formal analysis, investigation, writing–review and editing. M. Aras: Investigation, methodology, writing–review and editing. P.G. Ersan: Investigation, methodology, writing–review and editing. S. Serra: Investigation, methodology, writing–review and editing. H. Pei: Investigation, methodology, writing–review and editing. R. Clift: Investigation, methodology, writing–review and editing. Q. Zhao: Resources, methodology. K.B. Phan: Investigation, methodology, writing–review and editing. L. Huang: Resources, investigation, methodology. M.J. LaBarre: Project administration, writing–review and editing. X. Li: Investigation, methodology, writing–review and editing. H.M. Shepard: Investigation, project administration, writing–review and editing. S. Deaglio: Resources, investigation, project administration, writing–review and editing. J. Linden: Resources, investigation, project administration, writing–review and editing. C.D. Thanos: Conceptualization, investigation, project administration, writing–review and editing. O. Sahin: Resources, investigation, methodology, project administration, writing–review and editing. C. Cekic: Conceptualization, data curation, supervision, investigation, methodology, writing–original draft, project administration, writing–review and editing.

Development of this manuscript was supported by Halozyme Therapeutics, Inc. The authors would like to thank Sanna Rosengren, Robert Connor, Curt Thompson, Daniel C. Maneval, Keri Cannon, Charvi Nanavati, Michael Ouellette, Celine Derunes, Raju Koduri, Darin Taverna, Barbara Blouw, and Benjamin Thompson for their support with this program and for valuable scientific discussions. They thank Ozlem S. Sahin for helping with in vivo experiments. The results published here are, in part, based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga. Minor editorial and medical writing support, under the direction of the authors, was provided by Rachel O'Meara, PhD, Cristina Tomás, PhD, Natalie Morton, MSc, and Michelle Seddon, Dip Psych, all of Paragon Medica, and was funded by Halozyme Therapeutics, Inc.

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