Adenosine A2a Receptor Antagonism Restores Additive Cytotoxicity by Cytotoxic T Cells in Metabolically Perturbed Tumors

Cytotoxic T lymphocytes (CTL) are antigen-specific effector cells with the ability to eradicate cancer cells in a contact-dependent manner. The presence of CTLs in solid tumors positively correlates with improved clinical outcomes across a broad range of tumor types (1). Consequently, strategies have been developed to augment T-cell infiltration into the tumor microenvironment, including adoptive T-cell transfer (ACT) and dendritic cell (DC) vaccination (2, 3). However, although these strategies favor accumulation of tumor-specific CTLs in the tumor, only a fraction of patients reach tumor rejection and long-term survival (4–6). The mechanisms underlying this clinical failure are diverse, including CTL paralysis induced by an immunosuppressive tumor microenvironment (7). Thus, overcoming immunosuppression in tumors represents an unmet need for improving CTL effector function and anticancer immunotherapy.

The efficacy of the CTL effector response critically depends on the multistep recognition and elimination of antigenic target cells. Activated CTLs in the tumor probe the local tissue microenvironment in search for antigenic target cells. Target cell recognition initiates a stop signal to slow down CTL migration, thereby facilitating prolonged CTL engagement and the formation of an immune synapse (8). The immune synapse mediates cell–cell interaction and T-cell signaling to trigger the release of perforins and granzymes toward the target cell (9). Cytotoxic contacts are terminated upon CTL detachment from the target cell, followed by interstitial CTL migration and subsequent engagements with neighboring target cells (10, 11). The cytototoxic attack of CTL against solid tumor cells is inefficient and rarely completed by a single CTL. Instead, multiple CTLs sequentially engage with the same target cell to deliver a sequence of sublethal hits that accumulate overtime to induce target cell death (termed additive cytotoxicity; ref. 12). This complex sequence of migration, adhesion, and signaling steps enables serial killing and CTL cooperation to eliminate tumor cells.

Immunosuppressive factors released in tumor lesions have been shown to interfere with the CTL effector response. The nucleoside adenosine constitutes a potent immunosuppressive metabolite in the tumor microenvironment. While present at nanomolar concentrations in healthy tissues, extracellular adenosine accumulates up to micromolar levels in solid tumors, with higher concentrations in the tumor core than in the periphery (13, 14). The majority of adenosine in the tumor interstitium comes from the sequential hydrolysis of extracellular ATP by the cell-surface enzymes CD39 and CD73. In preclinical animal models, targeted inhibition of these ectonucleotidases restores antitumor immunity and enhances the efficacy of cancer immunotherapies (15–19). In addition, CD39 and CD73 predict poor clinical outcomes in various human cancers (20). Extracellular adenosine exerts its biological functions through activation of G-protein–coupled adenosine receptors (A1, A2a, A2b, and A3), which are differentially expressed on tumor, stromal, and infiltrating immune cells. The adenosine A2a receptor (ADORA2a) is the predominant subtype expressed by CTLs (21). ADORA2a signaling reduces T-cell crawling on ICAM-1 substrates that mimic cell surfaces (22). Adenosine also activates the C-terminal Src kinase (Csk) in naïve CTLs, which exerts tonic inhibition of the Src family kinases Lck and Fyn, thereby dampening the strength of T-cell receptor (TCR) signaling (23, 24). The bulk suppressive effect of adenosine/ADORA2a signaling on CTL-mediated antitumor immune control has been well established (14, 25, 26), yet the impact on the individual substeps and associated contact kinetics of the CTL effector response remain obscure.

Understanding the cellular and molecular mechanisms by which adenosine compromises CTL effector function requires time-resolved analysis of the multistep CTL effector response at cellular and molecular levels. We used single-cell reporter strategies to detect perforin hit delivery toward antigenic B16F10 melanoma cells in 3D collagen coculture in vitro and progressing tumors in vivo. This revealed that adenosine destabilizes CTL–target cell contacts, which resulted in reduced perforin hit delivery and target cell killing. Pharmacologic inhibition of ADORA2a restored target cell killing by stabilizing CTL–tumor cell interactions and increasing repetitive hit delivery. This identifies the adenosine–ADORA2a signaling axis in perturbing sublethal multihit delivery, which can be restored by ADORA2a antagonism.

Mouse melanoma B16F10 cells were obtained from ATCC (CRL-6475). B16F10 cells expressing the ovalbumin-derived CTL epitope SIINFEKL were obtained by electroporation. Stable H2B-GFP- or H2B-mCherry–expressing B16F10/OVA cells were obtained by lentiviral transduction and blasticidin selection (10 mg/mL; Thermo Fisher Scientific, R2101). B16F10/OVA H2B-mCherry cells were lentivirally transduced to stably express the calcium sensor GCaMP6s followed by blasticidin selection (10 mg/mL; ref. 27). The cells were cultured in RPMI-1640 medium (Gibco, 21875-034) supplemented with 10% FCS (Sigma, F7524), 1% sodium pyruvate (Gibco, 13360-039), and 1% penicillin and streptomycin (PAA, P11-010). OT-I CTLs (discussed below) were cultured in T-cell medium (TCM), consisting of RPMI-1640 (Gibco, 21875-034) supplemented with 10% FCS (Sigma, F7524), 10 mmol/L HEPES (Gibco, 15630-056), 500 mmol/L 2-mercaptopethanol, 1% penicillin, and streptomycin (PAA, P11-010), 1% sodium pyruvate (GIBCO, 11360-039), and 0.1 mmol/L nonessential amino acids (Gibco, 11140-035). During 3D collagen interface coculture, CTLs were cultured in TCM containing heat-inactivated FCS (15 minutes at 80°C) to limit FCS-induced metabolic breakdown of adenosine. Cell culture was performed at 37°C in a humidified 5% CO₂ atmosphere. Identity of the B16F10 cells was verified by short tandem repeat (STR) DNA profiling (IDEXX BioResearch, last profiled in May 2018). No mammalian interspecies contamination was detected. Lack of contamination with Mycoplasma was routinely verified using the MycoAlert Mycoplasma Detection Kit (Lonza, LT07-318, last tested in August 2020). Cell lines were cultured up to a maximum of 30 passages.

Splenocytes were derived from spleens of OT-I TCR transgenic mice (Jackson Laboratories, stock number: 003831) or from double-transgenic dsRed/OT-I mice obtained by crossing dsRed.T3 mice (Jackson Laboratories, stock number: 006051) with OT-I TCR transgenic mice (Jackson Laboratories, stock number: 003831). Splenocytes and OT-I T lymphocytes were cultured in TCM (see above). Splenocytes were harvested by mashing the spleen tissue through a 100-μm nylon cell strainer (BD, 352360). Erythrocytes were depleted by incubating the splenocytes in ammonium chloride (0.83% NH₄Cl, 0.1% KHCO₃, 0.37% Na₂EDTA; 5 minutes at room temperature).

For the expansion of antigen-specific CTLs, splenocytes were cultured in TCM in 24-well plates (5 × 10⁵ cells/well) in the presence of SIINFEKL peptide (0.5 μg/mL; Sigma-Aldrich, S7951). After 3 days of culture, the cells were resuspended in a mixture of 50% CTL expansion culture supernatant with 50% fresh TCM and cultured in 24-well plates (4 × 10⁵ cells/well) in the presence of IL2 (100 units/mL; ABD Serotec, PMP38). After 48 hours of culture, CTLs were isolated by Ficoll gradient centrifugation (Axis-Shield, 1114544). Purity typically exceeded 96% Vα2⁺CD8⁺CD62L^lowCD44^hi cells, as determined by flow cytometry (BD FACSCalibur or BD FACS Lyric) using PerCP-Cy5.5–conjugated rat antimouse CD8a (BD Biosciences, 551162), APC-conjugated rat antimouse Va2 TCR (BD Biosciences, 560622), FITC-conjugated rat antimouse CD62 L (BD Biosciences, 561917), and PE-conjugated rat antimouse CD44 (BD Biosciences, 553134).

Unless stated otherwise, B16F10/OVA target cells were stimulated with interferon-gamma (IFNγ, 200 units/mL; PeproTech, 315-05) for 24 hours prior to seeding to enhance SIINFEKL antigen presentation (28). Target cells were seeded in a 96-well flat bottom plate (7,000 B16F10/OVA cells/well with 200 units/mL IFNγ) and allowed to adhere and spread overnight to form a subconfluent target cell layer. Preactivated OT-I CTLs were resuspended in bovine collagen solution (1.7 mg/mL; Advanced BioMatrix, 5005) and overlaid onto the target cell monolayer at CTL-target cell ratios ranging from 4:1 to 1:16. After collagen polymerization (30 minutes at 37°C), TCM was added to each well and the coculture was incubated for 30 hours. For proliferation analysis, preactivated OT-I CTLs were stained with 400 nmol/L CFSE (10 minutes at room temperature; Thermo Fisher, C1157) prior to embedding in the collagen matrix. The 3D cytotoxicity cocultures were treated with adenosine (25 μmol/L or 50 μmol/L; Sigma-Aldrich, A4036) and ADORA2a antagonist ZM-241385 (0.5 μmol/L; Tocris, 1036). CTL proliferation, viability, and cell-surface CD107a expression were assessed via flow cytometry as described below.

Cells were harvested from the 3D interface coculture by enzymatic digestion of the collagen gel using collagenase I (190 U/well, 30 minutes at 37°C; Sigma-Aldrich, C0130). The resulting cell suspension was collected, and the remaining adherent cells were detached using trypsin (0.075%; Gibco, 15090046) and EDTA (1 mmol/L; Invitrogen, 15575020). The cell suspension after trypsin/EDTA treatment was combined with the cell suspension after collagen digestion (without centrifugation), and the whole volume was analyzed by flow cytometry using a BD FACSCalibur and BD FACS Lyric. For measuring the effect of adenosine on CTL viability, OT-I CTLs (1.05 × 10⁵ cells) were maintained in 2D culture in TCM for 6 hours without target cells in the absence (milliQ) or presence of adenosine (50 μmol/L) and analyzed by flow cytometry. To determine cell-surface expression of CD107a, cells were stained with Alexa Fluor 488–conjugated anti-mouse CD107a (20 μg/mL; BioLegend, 121608) or the respective isotype control (20 μg/mL; BioLegend, 400525) in ice-cold PBS for 1 hour. After incubation, cells were stained with Alexa Fluor 488-conjugated donkey anti-rat antibody (5 μg/mL; Thermo Fisher, A-21208) in ice-cold PBS for 1 hour. Cells were gated on intact morphology, viability by propidium iodide (Sigma-Aldrich, P4170) or Sytox Blue (Thermo Fisher Scientific, S11348) exclusion, and dsRed expression using FlowJo software (Tree Star, version 10; Supplementary Fig. S1a). For CTL proliferation analysis, the CFSE signal intensity was determined after gating for viable dsRed-labeled CTLs.

B16F10/OVA target cells were stimulated with IFNγ (200 units/mL; PeproTech, 315-05) for 24 hours prior to seeding to enhance SIINFEKL antigen presentation (28). Target cells were seeded in a 96-well flat bottom microplate (7000 B16F10/OVA cells/well with 200 units/mL IFNγ; Greiner Bio-One, 665090) and allowed to adhere and spread overnight to form a subconfluent target cell layer. Preactivated OT-I CTLs were resuspended in bovine collagen solution (1.7 mg/mL; Advanced BioMatrix, 5005) and overlaid onto the target cell monolayer at a CTL–target cell ratio of 1:1. After collagen polymerization (30 minutes at 37°C), TCM was added, and the coculture was monitored by brightfield time-lapse microscopy using a BD Pathway 855 High-Content Bioimager with a plan -NEOFLUAR 10x/0.3 air objective and Modicam Olympus UApo/340 20×/0.75 or Zeiss Axiovert 200M with Moticam-pro 2850 CCD Camera in an OKOlab stage-top CO₂ incubator. Images were taken with a 30- to 98-second frame interval for up to 30 hours.

The duration, kinetics, and killing outcome of individual CTL–target cell interactions were quantified by manual analysis. CTLs migrating in an out-of-focus plane above the target cell in the 3D collagen matrix were not included for analysis. Direct CTL–target cell contacts were identified based on the following morphologic and kinetic criteria: polarization/flattening of the CTL on the target cell-surface, deviation of the CTL migration path along the target cell-surface, and slowed CTL migration speed or migration arrest. For analyzing the effect of adenosine on CTL migration speed, CTL migration in 3D interface cocultures was quantified during periods without conjugation to B16F10/OVA cells using computer-assisted cell tracking (AutoZell 1.0 Software; Center for Computing and Communication Technologies, University of Bremen, Bremen, Germany).

Preactivated OT-I CTLs (3.6 × 10⁶ cells/mL) were incubated with Fura-2 AM (2 μmol/L, 30 minutes, 37°C, in the dark; Thermo Fisher, F1221), washed twice with TCM and embedded in the 3D interface coculture with unlabeled B16F10/OVA target cells as described above. Ca²⁺ signals in the OT-I CTLs were monitored by spinning-disk confocal microscopy (BD Pathway 855 High-Content Bioimager; Olympus UApo/340 20×/0.75) using a frame interval of 104 to 112 seconds. To account for phototoxicity and bleaching, imaging periods were limited to 1 hour. CTL calcium fluxes were analyzed by manual region-of-interest (ROI) selection using the Fura-2 340/380 ratiometric images.

B16F10/OVA target cells were engineered to stably express the calcium sensor GCaMP6s (27). 3D cocultures of target cells and CTLs were set up as described above. CTL–target cell conjugation and associated intracellular Ca²⁺ events were obtained by long-term confocal microscopy of GCaMP6s and OT-I CTLs at frame intervals of 8 to 15 seconds for up to 12 hours (Leica SP8 SMD; Leica HC PL APO CS 40×/0.85) using a fixed focus, 600 μm optical section, and an excitation power of 0.05 mW for each 488 and 561 nm laser line (3.16 μsec pixel dwell time, 0.32 μm² pixel size). Lack of phototoxicity was verified based on unperturbed migration (CTL) and morphologic integrity (CTL, target cells). OT-I CTL interaction kinetics were coregistered with Ca²⁺ events and B16F10/OVA death induction. Target cell death was identified by increased GCaMP6s sensor brightness, followed by cell shrinkage. CTL-associated Ca²⁺ events were identified by manual frame-by-frame image intensity analysis and displayed for cell populations of multiple independent experiments.

B16F10/OVA target cells were stimulated with IFNγ (200 units/mL; PeproTech, 315-05) for 24 hours. B16F10/OVA target cells (5 × 10⁵ cells) or preactivated OT-I T cells (2 × 10⁶) were lysed using Buffer RLT (Qiagen, 79216) containing 1% β-mercaptoethanol. After mechanical homogenization of the lysate using a cell scraper, RNA was isolated using the RNeasy Micro Kit (Qiagen, 74004), and total RNA was reverse transcribed using the iScript cDNA Synthesis Kit (Bio-Rad, 1708890). Primers were designed using HomoloGene (Table 1; http://www.ncbi.nlm.nih.gov). Primers were validated in silico using Oligoanalyzer software (http://www.idtdna.com/analyzer/Applications/OligoAnalyzer/). Quantitative PCR was performed on a CFX96 Real-Time PCR Detection System with C1000 Thermocycler (Bio-Rad) and analyzed using the Bio-Rad CFX Maestro software (version 4.1.2433.1219). RNA expression for the genes of interest was normalized to the pooled levels of two housekeeping genes (Gapdh and Hprt1).

A 96-well flat-bottom microplate (Greiner Bio-One, 665090) was coated overnight with anti-mouse CD3ε (10 μg/mL; BioLegend, 100339). Preactivated CTLs (3×10⁶ cells/mL) were incubated in the absence (DMSO) or presence of ZM-241385 (0.5 μmol/L, 15 minutes at 37°C; Tocris, 1036), followed by the addition of milliQ water or adenosine (50 μmol/L, 30 minutes at 37°C; Sigma-Aldrich, A4036). The CTL suspension was added to the anti-CD3–coated microplate (150,000 CTLs/well) and incubated (10 minutes at 37°C), followed by fixation with Bouin solution (15 minutes, room temperature; Klinipath, 64096). After permeabilization (15 minutes, room temperature); 1× PBS, 0.1% Triton X-100 (Sigma-Aldrich, T8787), 10% normal goat serum (Thermo Fisher Scientific, PCN5000), 1% BSA (Sigma-Aldrich, A9647), and blocking of nonspecific binding sites (1 hour, room temperature); 1× PBS, 0.05 Tween-20 (Sigma-Aldrich, P7949), 10% normal goat serum, 1% BSA), the cells were stained with anti–phospho-Src (Tyr-416) family kinase (2.5 μg/mL; Cell Signaling Technology, 2101) and Alexa Fluor 647–conjugated anti-CD45 (2.5 μg/mL; BioLegend, 103124) overnight at 4°C in blocking solution. Background staining was detected using rabbit IgG isotype control (R&D Systems, AB-105-C). Primary antibodies were detected by incubating the cells with Alexa Fluor 568–conjugated goat anti-rabbit antibody (2.5 μg/mL; Thermo Fisher, A11036) overnight at 4°C in a blocking solution. Cells were imaged using a Zeiss LSM880 (63×1.4 NA oil immersion objective, Carl Zeiss). For phospho-Src analysis, confocal microscopy was performed with sequential 561 nm and 633 nm excitation. Emission light was collected using 568 to 642 nm and 646 to 756 nm filters for Alexa Fluor 568 and Alexa Fluor 647, respectively. All images were processed using Fiji/ImageJ, and the number of phosho-Src⁺ microclusters per CTL was analyzed using the Find Maxima plugin.

Target cells were seeded in the presence of IFNγ (200 units/mL) in a 96-well flat bottom microplate (7,000 and 10,000 B16F10/OVA cells/well for granule polarization and NFAT1 analysis, respectively; Greiner Bio-One, 665090) and allowed to adhere and spread overnight to form a subconfluent target cell layer. Preactivated OT-1 CTLs (1.6 × 10⁶ cells/mL) were incubated in the absence (DMSO) or presence of ZM-241385 (0.5 μmol/L, 15 minutes at 37°C; Tocris, 1036), followed by the addition of milliQ water or adenosine (50 μmol/L, 30 minutes at 37°C; Sigma-Aldrich, A4036). For granule polarization analysis, CTLs were seeded onto the subconfluent target cell layer (35,000 CTLs/well) in the presence of DMSO/ZM-241385 (0.5 μmol/L) and milliQ/adenosine (50 μmol/L) and incubated for 4 hours, followed by fixation with Bouin solution (15 minutes, room temperature; Klinipath, 64096). For NFAT1 analysis, CTLs were seeded onto the subconfluent target cell layer (80,000 CTLs/well) in the presence of DMSO/ZM-241385 (0.5 μmol/L) and milliQ/adenosine (50 μmol/L) and incubated for 4 hours, followed by fixation with 2% paraformaldehyde (PFA; 15 minutes, room temperature; Merck, 104005). After permeabilization (15 minutes, room temperature; 1× PBS, 0.1% Triton X-100, 10% normal goat serum, 1% BSA) and blocking of nonspecific binding sites (1 hour, room temperature; 1× PBS, 0.05 Tween-20, 10% normal goat serum, 1% BSA), the cells were stained with anti–phospho-Src (Tyr-416) family kinase (2.5 μg/mL; Cell Signaling Technology, 2101), anti-NFAT1 (2 μg/mL; Cell Signaling Technology, 5861), Alexa Fluor 647–conjugated anti-CD45 (5 μg/mL; BioLegend, 103124), Alexa Fluor 488–conjugated anti-granzyme B (2.5 μg/mL; Thermo Fisher, 11-8898-82) or Alexa Fluor 568 Phalloidin (200 U/mL; Thermo Fisher, A12380) overnight in blocking solution at 4°C. Background staining was detected using rabbit IgG isotype control (R&D Systems, AB-105-C). Primary antibodies were detected by incubating the cells with Alexa Fluor 647–conjugated goat anti-rabbit antibody (2.5 μg/mL; Thermo Fisher, A21245), Alexa Fluor 568–conjugated goat anti-rabbit antibody (2.5 μg/mL; Thermo Fisher, A11036) and Alexa Fluor 488–conjugated goat anti-rat antibody (2.5 μg/mL; Thermo Fisher, A11006) overnight at 4°C in blocking solution. Nuclei were stained with DAPI (1 μg/mL, 15 minutes at room temperature; Sigma-Aldrich, D9542). Imaging was performed using laser scanning microscopy (LSM880; 63 × 1.4 NA oil immersion objective; Carl Zeiss). For CTL granule polarization analysis, confocal microscopy was performed with 488 nm and 647 nm excitation, followed by 561 nm excitation. Emission light was collected using 499 to 575 nm, 578 to 644 nm, and 645 to 701 nm filters for Alexa Fluor 488, Alexa Fluor 568, and Alexa Fluor 647, respectively. For NFAT1 analysis, Airyscan imaging was performed with sequential 405, 488, 561, and 633 nm excitation. Emission light was collected using the following emission filters: BP420–480/BP495–620 for DAPI, 3P 495–550/LP 570 for Alexa Fluor 488, 3P 570–620/LP645 for Alexa Fluor 568, and 3P 570–620/LP645 for Alexa Fluor 647. Raw images were reconstructed using the Zeiss Zen 2.1 Sp1, and images were processed using Fiji/ImageJ.

All experiments with mice were approved by the Ethical Committee on Animal Experiments and were performed in the Central Animal Laboratory of the Radboud University, Nijmegen (AVD10300 2019 7964), in accordance with the Dutch Animal Experimentation Act and the European FELASA protocol (https://felasa.eu/). Tumor size was continuously monitored by fluorescence microscopy, with a humane endpoint for tumors exceeding 2 cm³.

Histone-2B/mCherry expressing B16F10/OVA cells (5 × 10⁴ to 2 × 10⁵) were injected into the deep dermis of C57/B16J mice (Charles River Laboratories, strain code 632) carrying a dorsal skin-fold chamber, as described (29). Three and six days after tumor implantation, in vitro–activated dsRed OT-I CTLs (2× 10⁶) were administered retro-orbitally. Tumor-bearing mice were daily monitored for up to 13 days. Intravital multiphoton microscopy was performed on anesthetized mice (1%–3% isoflurane in oxygen; Pharmachemie, 45112106). During imaging, mice were maintained on a temperature-controlled stage. Tumor volume was obtained at consecutive time points from epifluorescence overview images and calculated as (tumor width)² × (tumor length) × π/6.

Intravital microscopy was performed on a customized near-infrared/infrared multiphoton microscope (TriMScope-II, LaVision BioTec a Milteny Company), equipped with two tunable Titanium:Sapphire lasers (Chameleon Ultra I and Ultra II; Coherent) and an optical parametric oscillator (MPX, APE) for excitation. Microscopy was performed using a 25 × 1.05 NA water-immersion objective lens (XLPLN25XWMP2, Olympus). Emission was filtered with 525/50 (GCaMP6s), 572/28 (dsRed), 620/60 (mCherry) and detected by alkali, GaAsP, or GaAs PMT detectors (H6780-20, H7422A-40 or H7422A-50, Hamamatsu). Time-lapse recordings of OT-I CTL effector dynamics in B16F10/OVA tumor tissue were acquired by scanning 3D tissue regions (360 μm × 360 μm × 100 μm, z-interval: 5 μm) with 1,090 nm at a laser power of 20 to 50 mW (time interval: 2 minutes, total imaging period: 60 minutes). CTL densities and B16F10/OVA mitosis/apoptosis frequencies were analyzed using 3D stacks (597 μm × 597 μm, imaging depth up to 273 μm, z-interval: 7 μm) with 1,090 nm at a laser power of 20 to 60 mW. For in vivo visualization of the calcium sensor GCaMP6s in B16F10/OVA cells, whole-tumor 3D stacks with an imaging depth of 308 μm (z-interval: 7 μm) were recorded with 910 nm (GCaMP6s) and 1,210 nm (mCherry) at a laser power of 15 to 35 mW. The setup was equipped with a warm plate (DC60 and THO 60-16, Linkam Scientific Instruments Ltd) and a custom-made objective heater (37°C), as described (30).

Images were processed using Fiji/ImageJ (version 2.0.0-rc-69/1.53f, 2.0.0-rc-43/1.53i and 2.0.0-rc-43/1.52h). Images were stitched to mosaics using the Stitch Grid/Collection plugin (31), and field drifts during time-lapse recording were corrected using the StackReg plugin (32) and the Correct 3D Drift plugin (33). For image quantifications, raw, unmodified images were used. For display purposes, images were scaled and adjusted for brightness and contrast to enhance visualization. Quantification of OT-I CTL density, B16F10/OVA mitosis/apoptosis frequencies, and OT-I CTL mitosis/death rates was performed on individual image sections from 3D stacks. To preclude repeated counting of the same cell, every third slice per stack was analyzed. Nuclei and CTLs were segmented by Gaussian Blur filtering, automated thresholding (Bernsen's algorithm), and Watershed separation of touching objects. Objects were filtered for size (Nuclei: ≥80 μm²) using the Analyze Particles command and counted per slice. To determine the B16F10/OVA apoptosis and mitosis frequencies from in vivo samples, apoptotic and mitotic nuclei were counted manually and calculated as the percentage of total nuclei per stack. Tumor subregions were defined manually, as follows: “edge” for cells located in the tumor mass within 300 μm from the tumor–stroma interface; and “center” for cells located >300 μm from the tumor margin. To quantify GCaMP6s intensity in B16F10/OVA tumors, background noise was reduced using a median filter (radius 2), followed by reconstruction of a maximum intensity Z-projection. The tumor boundary and ROI defining tumor subregions were manually segmented based on GCaMP6s expression in B16F10/OVA cells.

Parametric data sets were compared using an unpaired t test or one-way ANOVA. To correct for multiple comparisons, the one-way ANOVA was combined with Tukey or Sidak post-hoc testing. Nonparametric data sets were compared using the Mann–Whitney U or Kruskal–Wallis test. Dunn multiple comparisons correction was performed for Kruskal–Wallis testing. The extra sum-of-squares F test was used to compare the slopes of nonlinear regression curves on the relationship between OT-I CTL numbers and B16F10/OVA cell killing. Poisson regression was used to fit curves on the relationship between the CTL contact duration and the number of B16F10/OVA Ca²⁺ events per CTL contact. All data were analyzed using GraphPad Prism 9.

The data generated in this study are available within the article and its supplementary data files.

To identify the mechanisms by which extracellular adenosine compromised CTL effector functions against tumor cells, we used a 3D interface coculture of OT-I effector CTLs and B16F10/OVA melanoma cells (Fig. 1A). The 3D matrix environment of the 3D interface model depends on both active migration of CTLs to reach the lower tumor–matrix interface and recognition of the B16F10/OVA target cells (Supplementary Movie S1). Target cell killing increased with increasing CTL density, reaching up to 85% at a 4:1 CTL–target cell ratio (Fig. 1B; Supplementary Fig. S1a). The viability of B16F10 control cells, not expressing the OVA peptide, remained near 100%, irrespective of the E:T ratio (Fig. 1B). Low micromolar concentrations of adenosine, like those observed in tumors in vivo (34), reduced CTL-mediated tumor cell killing across a wide range of CTL densities and by up to 65% (Fig. 1C). A four-fold increase in CTL density, from E:T ratio of 1:2 to E:T ratio of 2:1, did not enhance B16F10/OVA cell killing under adenosine-rich conditions (Fig. 1C). The suppressive effect of adenosine on tumor cell killing by OT-I CTLs was caused by a dual effect: (i) lowering CTL numbers (by 45%) by reducing CTL proliferation without inducing direct CTL death (Supplementary Fig. S1b–S1d) and (ii) decreasing the killing rate per CTL, as indicated by a less steep slope of the regression curve between target cell killing and CTL density (Fig. 1D).

To investigate how adenosine affected CTL conjugation and target cell killing at the single-cell level, we performed time-lapse microscopy. Adenosine reduces T-cell crawling on ICAM-1–coated 2D substrates (22), and impaired CTL motility may delay the serial conjugation of CTLs with target cells. In 3D interface cocultures, adenosine did not affect the migration speed of OVA-stimulated OT-I CTLs during periods without target cell conjugation (Fig. 1E). However, adenosine shortened the CTL interaction time with individual tumor cells by 59%, from 27 minutes to 11 minutes (Fig. 1F and G; Supplementary Movie S1). The aggregated CTL contact duration with multiple tumor cells was reduced by 50%, from 8 to 4 hours (Fig. 1H). Thus, adenosine does not affect CTL migration speed in a 3D matrix environment but weakens CTL contact stability with target cells.

Next, we investigated the effect of adenosine on CTL calcium signaling and degranulation. During conjugation with B16F10/OVA cells, 60% of the OT-I CTLs underwent a phase of calcium signaling (Fig. 1I and J; Supplementary Movie S2). In the presence of adenosine, the CTL fraction exhibiting calcium flux was decreased by 20%. The reduction in CTL calcium signaling was associated with labile (<30 minutes) CTL–target cell interactions (Fig. 1K; Supplementary Movie S2). Consistent with decreased CTL reactivation, the surface expression of the CTL degranulation marker CD107a was also reduced under adenosine-rich conditions (Fig. 1L). Thus, adenosine labilizes CTL–target cell interactions, dampens calcium signaling and degranulation, and decreases CTL killing efficacy.

We next explored how single CTL responses under adenosine-rich conditions could be restored. Adenosine mediates its biological effects by triggering adenosine transmembrane receptors (35). We profiled the expression of the known adenosine receptors in both OT-I and B16F10/OVA cells. Whereas B16F10/OVA cells did not express adenosine receptors above background levels, preactivated OT-I effector CTLs expressed the high-affinity ADORA2a abundantly, the low-affinity adenosine A2b receptor (ADORA2b) moderately, and none of the other adenosine receptors (Fig. 2A). Therefore, we hypothesized that adenosine compromised CTL effector functions via ADORA2a signaling. Indeed, under adenosine-rich conditions, the selective ADORA2a antagonist ZM-241385 completely restored the CTL killing efficacy to baseline levels (Fig. 2B). This immunostimulatory effect associated with prolonged CTL–tumor cell contacts and normalized CTL degranulation, which suggests that ADORA2a antagonism rescues the killing efficacy at the single CTL level (Fig. 2C and D; Supplementary Fig. S2a and S2b).

CTL effector functions against melanoma target cells are inefficient with a high failure rate, and individual CTL–target cell contacts rarely induce apoptosis at first encounter (12). Solid tumor cells can thus survive individual sublethal cytotoxic hits. In vivo, effective target cell killing requires a cooperative process in which multiple CTLs engage sequentially with the same target cell to initiate apoptosis by a series of sublethal interactions (12). We tested whether microenvironmental perturbation by adenosine decreases the rate of sequential hits by CTLs and whether sequential hit delivery was restored upon ADORA2a antagonization. The calcium sensor GCaMP6s (27) expressed by B16F10/OVA was used to monitor transient Ca²⁺ influx through perforin pores during CTL engagement. The induction of Ca²⁺ influx, subsequent nuclear envelope rupture, and DNA damage are strictly dependent on perforin expression by OT-I CTLs, without effect of soluble cytokines (12). Identified by time-gated analysis, CTL-associated Ca²⁺ signals originated at the CTL-tumor cell contact region and can be reliably discriminated from unspecific intracellular Ca²⁺ fluctuations, based on their signal intensity and duration (12). In control cultures, CTLs induced a calcium event during 55% of the CTL–target cell contacts (Fig. 2E and F). The majority of these productive CTL contacts were highly efficient, as 69% induced 2 or more calcium events by the same CTL during a single contact (Fig. 2G; Supplementary Fig. S2c). The destabilizing effect of adenosine on CTL–tumor cell interactions associated with a 28% lower probability to induce a calcium event in target cells, with a 19% reduction in repetitive calcium induction by productive CTL contacts (Fig. 2E–G; Supplementary Fig. S2c; Supplementary Movie S3). Both contact duration and calcium induction in target cells were restored by ADORA2a antagonism under adenosine-rich conditions (Fig. 2E–G; Supplementary Fig. S2c; Supplementary Movie S3). The restored number of calcium events induced in the target cell was correlated with rescue in CTL contact duration (Fig. 2H). Thus, adenosine reduces contact efficacy by limiting contact duration, and this is reversed by ADORA2a antagonism, resulting in restored repetitive hit delivery and cytotoxicity.

We next investigated the mechanism by which pharmacologic inhibition of adenosine-induced ADORA2a, triggering restored CTL contact efficacy. CTL conjugation is a multistep process that is orchestrated by TCR signaling in response to antigen recognition. Bulk population analysis by western blot has revealed that ADORA2a triggering suppresses anti-CD3–mediated TCR signaling in naïve CTLs (24). We validated these results for activated OT-I CTLs at the single-cell level by measuring activated Src kinases in individual CTLs stimulated with immobilized anti-CD3. Under control conditions, CD3 ligation by anti-CD3 induced phospho–Src-positive microclusters at the contact site/plane, with high cell-to-cell variability (Fig. 3A and B; Supplementary Fig. S3a–S3c). Adenosine reduced the number of phospho-Src⁺ microclusters by 60%, and this effect was completely restored by ADORA2a antagonism (Fig. 3A and B). We then performed ratiometric calcium imaging of fura-2–loaded OT-I CTLs cocultured with B16F10/OVA target cells to address whether a defective cytolytic synapse preceded target cell killing. Upon adenosine exposure, the fraction of CTLs with calcium mobilization decreased by 20% (Fig. 3C), and the amplitude of calcium influx into CTLs decreased under adenosine-rich conditions (Fig. 3D; Supplementary Fig. S3d), indicating weakened TCR signaling (36). Likewise, nuclear localization of the calcium-regulated nuclear factor of activated T cells 1 (NFAT1), a downstream effector of TCR activation and calcium signaling, was reduced (by 57%) upon OT-I CTL coculture with B16F10/OVA cells in the presence of adenosine (Fig. 3E and F). Thus, the strength of the CTL calcium influx, as well as downstream NFAT signaling, was dampened under adenosine-rich conditions. Both effects were completely rescued by ADORA2a antagonism with ZM-241385 (Fig. 3C and F).

The strength of CTL calcium signaling critically regulates CTL effector responses, mediating T-cell proliferation and cytotoxic granule exocytosis (37, 38). In OT-I CTLs, weakened calcium influx under adenosine-rich conditions associated with decreased proliferation (Fig. 3G and H; Supplementary Movie S4) and an increased frequency of conjugated CTLs that failed to polarize their cytotoxic granules toward the phospho-Src⁺ immune synapse (Fig. 3I and J; Supplementary Fig. S3e and S3f). Both effects were restored by ADORA2a antagonism (Fig. 3I and J; Supplementary Fig. S3g). Thus, triggering of the adenosine/ADORA2a pathway counteracts membrane-proximal TCR activation, calcium signaling, CTL proliferation, and lytic granule polarization to the immune synapse.

To address the effect of ADORA2a antagonism on CTL effector function in progressing melanoma tumors in vivo, activated OT-I CTLs were adoptively transferred into C57BL/6J mice bearing intradermal B16F10/OVA tumors, and tumor growth was monitored by intravital microscopy through a skin window (Fig. 4A). Adoptive OT-I CTL transfer delayed the growth of B16F10/OVA tumors by 45%, and this effect was further enhanced (by 44%) when mice were treated with ZM-241385 (Fig. 4B; Supplementary Fig. S4a–S4c). The tumor growth–suppressing effect of ADORA2a antagonism was dependent on adoptive CTL transfer, as ZM-241385 treatment alone did not affect B16F10/OVA growth (Fig. 4B; Supplementary Fig. S4a and S4b). To directly address the effect of ADORA2a antagonism on CTL effector responses, we analyzed the expansion, effector dynamics, and function of tumor-infiltrating OT-I CTLs by long-term time-lapse microscopy. ADORA2a antagonism increased CTL density in tumors, both at the center (by 3.7-fold) and edge (by 2.6-fold; Fig. 4C), which resulted from enhanced CTL proliferation rates with unaffected CTL death (Fig. 4D–F; Supplementary Movie S5). The ADORA2a antagonist in combination with adoptive OT-1 CTL transfer affected tumor growth in both the tumor core and periphery, with reduced tumor cell mitosis rates and a 3.4-fold (center) and a 2.6-fold (edge) increase in tumor cell apoptosis rates (Fig. 4G–I). Besides increasing CTL density, ZM-241385 treatment reduced CTL migration speed (by 22%) and favored migration arrest of CTL subsets in contact with tumor cells (Fig. 4J and L; Supplementary Movie S5). When observed in conjunction, decreased migration speed and increased tumor cell death induction is indicative of more focused cognate antigen recognition and effector function (39). The in vivo interaction kinetics were consistent with the observed stabilization of CTL–target cell interactions by ADORA2a antagonism in adenosine-rich 3D interface cocultures (Fig. 2C). Visualization of GCaMP6s activity revealed that treatment with the ADORA2a antagonist induced a 3-fold increase in tumor areas positive for B16F10/OVA Ca²⁺ events (Fig. 4M and N). In contrast, in control tumors treated with vehicle or ZM-241385 without adoptive CTL transfer, the GCaMP6s⁺ tumor area remained unchanged (Fig. 4M and N; Supplementary Fig. S5a–S5b). Thus, ADORA2a antagonism does not induce damage in tumor cells directly, but its efficacy depends on the presence of adoptively transferred CTLs. In conclusion, combining adoptive CTL therapy with ADORA2a antagonism results in enhanced local CTL expansion and improved CTL effector function against tumor cells.

The time-resolved single-cell analyses used in this study reveal that micromolar concentrations of adenosine, which match adenosine concentrations reached in tumors in vivo (34), interfere with multiple steps of the cytotoxic immune effector phase by (i) destabilizing CTL contacts with tumor cells, (ii) dampening proximal TCR signaling and downstream calcium influx, (iii) perturbing lytic granule polarization and exocytosis, and (iv) compromising sequential hit delivery and additive cytotoxicity. As a further consequence of impaired antigenic (re)stimulation, CTLs failed to expand in tumor lesions, resulting in reduced CTL numbers and cooperative engagements with tumor cells. This multilevel deterioration of CTL function was reverted by pharmacologic ADORA2a antagonization. ADORA2a inhibition restored tumor cell killing by stabilizing CTL–target cell conjugation and increasing sequential sublethal hit delivery to target cells, both in vitro and in vivo. Because enhanced adenosine signaling is a characteristic of a metabolically perturbed tumor microenvironment across solid tumor types, pharmacologic ADORA2a targeting may emerge as a critical supportive component for a range of immunotherapies.

Adenosine signaling perturbs multiple steps of CTL activation and antitumor effector function. Activation of the adenosine/ADORA2a pathway in naïve CTLs inhibits Src family tyrosine kinases (23), and inhibition of Src compromises CTL conjugation with DCs and subsequent T-cell activation (40). Similar to conjugation of naïve CTLs with DCs, we here showed that ADORA2a triggering by adenosine destabilizes conjugation of activated effector CTLs with target cells and severely limits sublethal hit delivery during individual contacts. Besides impairing lytic granule release, ADORA2a triggering has been shown to interfere with the production of CTL effector cytokines, including IFNγ and TNFα (21, 26, 41). Compromised cytokine secretion can be a consequence of instable CTL-target cell interaction and reduced CTL restimulation, for example, caused by experimental LFA-1 blockade, impaired TCR binding, or the presence of galectin-3 (42, 43). Thus, adenosine-induced labilization of CTL–target interaction and compromised antigenic restimulation may affect cell-mediated cytotoxicity and cytokine release in concert. The time-dependent integration of sublethal damaging events in cancer cells does not only depend on single CTL efficacy but also relies on a cooperative process in which multiple CTLs sequentially engage with the same target cell to induce apoptosis (12). Hence, by lowering the local density of CTLs, adenosine also compromises the serial delivery of perforin hits by CTL swarming. Delays in individual or cooperative hit delivery allow target cells to experience a prolonged time span between hits and, hence, an enhanced recovery time and chance to survive. Thus, ADORA2a signaling compromises both individual contact efficacy and CTL cooperativity required for serial killing of target cells.

The competitive antagonist ZM-241385 was used to inhibit ADORA2a signaling in effector CTLs. ZM-241385 exhibits high affinity for ADORA2a and dampens ADORA2a-mediated cAMP induction with an IC₅₀ of 54 nmol/L (44), with 319-, 63-, and >10,000-fold selectivity over ADORA1, ADORA2b, and ADORA3, respectively (45). Therefore, at the concentration (0.5 μmol/L) used here in tumor-mimetic 3D cocultures, ZM-241385 was expected to selectively inhibit the adenosine/ADORA2a pathway in CTLs, while sparing ADORA2b signaling. ADORA2a antagonism in well-defined adenosine-rich cocultures enhanced CTL contact stability, improved lytic granule polarization and exocytosis, and increased the delivery of sublethal perforin hits per CTL contact. We also showed that ADORA2a antagonism restored the functionality of tumor-infiltrating CTLs in progressing melanoma lesions in vivo by inducing a local effector phenotype with prolonged dwell time and improved sublethal hit delivery. This rescue of CTL effector function likely underlies the ability of other small-molecule ADORA2a antagonists (25, 46, 47) and genetic ADORA2a targeting strategies (26, 48) to enhance the efficacy of ACT in solid tumors. Thus, ADORA2a antagonism reinvigorates the killing efficacy of tumor-infiltrating CTLs by normalizing their immune synapse function.

Besides increasing the CTL killing efficacy at the single-cell level, ADORA2a antagonism converted the moderately immune-infiltrated melanoma microenvironment, characterized by modest CTL density, into a more immune-inflamed microenvironment with dense CTL infiltrates. Consistent with enhanced antigenic restimulation, ADORA2a antagonism restored membrane-proximal TCR signaling, CTL calcium influx, and CTL proliferation in adenosine-rich cocultures. ADORA2a antagonism also increased secondary expansion and swarming of tumor-infiltrating effector CTLs in B16F10/OVA tumors, thereby facilitating CTL cooperativity by serial CTL–tumor cell encounters (12). This increase in intratumoral CTL density has also been observed in renal cancer patients under active therapy with ADORA2a antagonists, which is associated with prolonged disease control in these patients (49). Similarly, expression of the ectonucleotidases CD39 and CD73 in preclinical tumor models suppresses the activation and proliferation of antigen-specific CTLs (46, 50). Thus, by supporting local CTL expansion in the tumor microenvironment, ADORA2a antagonism stimulates the cooperation of cytotoxic immune effector cells.

ZM-241385 has been used for proof-of-concept studies. Next-generation antagonists with extended in vivo half-life and good oral bioavailability have now been translated into the clinic, demonstrating tolerability and durable antitumor activity of adenosine antagonization regimens (49). Prolonged inhibition of ADORA2a is well tolerated in cancer patients, both as monotherapy and in combination with PD(L)-1 antibodies (49, 51, 52). To effectively target the adenosine/ADORA2a axis, predictive biomarkers are required to identify tumors that are under active adenosine-mediated immunosuppression and identify patients more likely to benefit from ADORA2a antagonists. As an example, gene-expression profiling can reveal ADORA2a signaling activity in the tumor microenvironment and identify patients with poor survival outcomes despite the accumulation of CTLs in tumor lesions (53). Besides its effect on CTL effector functions, adenosine signaling is known to inhibit NK cell maturation and impair perforin-mediated NK cell cytotoxicity against tumor cells (54, 55). Consequently, ADORA2a antagonism may exert a dual stimulatory effect on cumulative damage induction by reinvigorating effector responses of both tumor-infiltrating CTLs and NK cells (54). Thus, ADORA2a targeting may enable cooperation between cytotoxic leukocytes with complementary target-recognition strategies. The efficacy of ADORA2a targeting could be enhanced by combining ADORA2a antagonists with complementary approaches directed at increasing the susceptibility of tumor cells to sublethal damage, including radiotherapy to lower the setpoint of tumor cell apoptosis induction or DNA-repair inhibitors to delay target cell recovery between serial hits. Thus, ADORA2a antagonists can be positioned as an adjunct approach to other therapeutic strategies, with the aim to reinvigorate the CTL effector phase by normalizing immune synapse failure and CTL cooperativity in metabolically perturbed tumors.

None of the authors reported any disclosures.

J. Slaats: Conceptualization, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft. E. Wagena: Formal analysis, validation, investigation, visualization, methodology. D. Smits: Formal analysis, investigation, visualization, methodology. A.A. Berends: Formal analysis, investigation, methodology. E. Peters: Software, formal analysis, investigation, methodology. G.-J. Bakker: Software, formal analysis, investigation, visualization, methodology. M. van Erp: Software, formal analysis, investigation, visualization, methodology. B. Weigelin: Conceptualization, formal analysis, supervision, validation, investigation, visualization, methodology. G.J. Adema: Conceptualization, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft. P. Friedl: Conceptualization, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft.

This work was supported by a personal RadboudUMC PhD fellowship to J. Slaats (RvB15.52139). This work was also supported by grants from the European Research Council (617430-DEEPINSIGHT) and the Cancer Genomics Center to P. Friedl. Time-lapse confocal microscopy was enabled by an NWO investment grant to P. Friedl (834.13.003).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.