Extracellular cGAMP is a cancer cell-produced immunotransmitter involved in radiation-induced anti-cancer immunity

Abstract

2’3’-cyclic GMP-AMP (cGAMP) is an intracellular second messenger that is synthesized in response to cytosolic double-stranded DNA and activates the innate immune STING pathway. Our previous discovery of its extracellular hydrolase ENPP1 hinted at the existence of extracellular cGAMP. Here, we detected that cGAMP is continuously exported but then efficiently cleared by ENPP1, explaining why it has previously escaped detection. By developing potent, specific, and cell impermeable ENPP1 inhibitors, we found that cancer cells continuously export cGAMP in culture at steady state and at higher levels when treated with ionizing radiation (IR). In mouse tumors, depletion of extracellular cGAMP decreased tumor-associated immune cell infiltration and abolished the curative effect of IR. Boosting extracellular cGAMP with ENPP1 inhibitors synergized with IR to delay tumor growth. In conclusion, extracellular cGAMP is an anti-cancer immunotransmitter that could be harnessed to treat cancers with low immunogenicity.

The second messenger 2’3’-cyclic GMP-AMP (cGAMP)¹ plays pivotal roles in anti-cancer and anti-viral innate immunity. It is synthesized by the enzyme cyclic-GMP-AMP synthase (cGAS)² in response to double-stranded DNA (dsDNA) in the cytosol, which is a danger signal for damaged or cancerous cells and intracellular pathogens^3–7. cGAMP binds and activates its endoplasmic reticulum (ER) surface receptor Stimulator of Interferon Genes (STING)⁸ to activate production of Type 1 interferons (IFNs). These potent cytokines trigger downstream innate and adaptive immune responses to clear the threat.

In addition to activating STING within its cell of origin, cGAMP can spread to bystander cells through gap junctions in epithelial cells⁹. This cell-cell communication mechanism alerts adjacent cells of the damaged cell and also, unfortunately, accounts for the spreading of drug-induced liver toxicity^10,11 and brain metastases¹². In addition, cytosolic cGAMP can be transmitted to other cells via viral particles^13,14. In both transmission modes, cGAMP is not exposed to the extracellular space. Finally, tumor-derived cGAMP has been reported to activate STING in non-cancer cells through unknown mechanisms and eventually activate NK cell responses¹⁵.

cGAMP is synthesized in the cytosol and cannot passively cross the cell membrane, due to its two negative charges. However, two pieces of evidence hinted that cGAMP is exported to the extracellular space to signal other cells. First, we identified a cGAMP hydrolase, ectonucleotide pyrophosphatase phosphodiesterase 1 (ENPP1)¹⁶, which is the only detectable cGAMP hydrolase reported. Interestingly, ENPP1 is annotated as an extracellular enzyme, both as a membrane-bound form and as a soluble form in the serum¹⁶. Second, when added to cell media or injected into tumors, cGAMP and its analogs can cross the cell membrane to activate STING in most cell types^17,18. We and others subsequently identified a direct cGAMP importer, SLC19A1^19,20.

Here we report direct evidence for cGAMP export by cancer cells and the role of extracellular cGAMP in anti-cancer immune detection. We subsequently developed small molecule inhibitors of ENPP1 with nanomolar potency and used them to boost extracellular cGAMP concentration, immune infiltration, and tumor progression. Together, we characterize cGAMP as an immunotransmitter that can be harnessed to treat cancer.

cGAMP is exported from 293T cGAS ENPP1^−/− cells as a soluble factor

To test the hypothesis that cGAMP is present extracellularly, we first developed a liquid chromatography-tandem mass spectrometry (LC-MS/MS) method to quantify cGAMP down to 0.3 nM in cell media (extracellular) and cell extracts (intracellular) (Extended Data Fig. 1a–d)^21,22. We chose to use 293T cells, which express undetectable amounts of cGAS and STING proteins^2,8,23 (Extended Data Fig 1e). By stably expressing mouse cGAS and knocking out ENPP1, we created a 293T cGAS ENPP1^low cell line and then isolated a single clone to create a 293T cGAS ENPP1^−/− cell line. (Extended Data Fig. 1e). We also used serum-free media because serum contains a soluble form of ENPP1²⁴. Using this ENPP1-free cell culture system, we detected basal low micromolar intracellular cGAMP concentrations in the 293T cGAS ENPP1^−/− cells without any stimulation (Fig. 1a), but not in the parent 293T cells (Fig 1b). This was surprising at first, but can now be explained by multiple reports showing that cancer cells harbor cytosolic dsDNA, the activator of cGAMP synthesis, as a result of erroneous DNA segregation^25–27. After replenishing the cells with fresh media, we measured a linear increase of extracellular cGAMP up to 30 h (Fig. 1c). The amount of cGAMP exported is substantial given that after 30 h, the number of cGAMP molecules outside the cells was equal to the number inside (Fig. 1d). We detected negligible amount of cell death based on extracellular lactose dehydrogenate (LDH) activity, suggesting that cGAMP in the media is exported by live cells (Fig. 1e). We calculated the export rate (v_export) to be 220 molecules cell⁻¹ s⁻¹ (Fig. 1d). Finally, although human cGAS has been shown to be slower than its mouse counterpart²⁸, we measured both intracellular and extracellular cGAMP in human cGAS expressing 293T cells at steady state, which can be further induced with dsDNA transfection (Fig. 1f, g).

Figure 1 |. cGAMP is exported from 293T cGAS ENPP1^−/− cells.

a, Intracellular concentrations of cGAMP from 293T cGAS ENPP1^−/− cells without exogenous stimulation. At time 0, cells were replenished with serum-free media. Dashed line represents mean cGAMP concentration over all time points. Data are from 1 experiment, representative of 3 independent validations (Supplementary Fig. 1a, b); 2 cell culture replicates are plotted for each time point.

b, Intracellular concentrations of cGAMP from 293T WT and 293T cGAS ENPP1^−/− cells without exogenous stimulation at steady state. Data are from 1 experiment (supported by experiments in Fig. 2c and Supplementary Fig. 1c); 2 cell culture replicates are plotted.

c, Extracellular concentrations of cGAMP from experiment in a. Dashed line represents a linear regression. Data and regression are from 1 experiment, representative of 3 independent validations (Supplementary Fig. 1a, b); 2 cell culture replicates are plotted for each time point.

d, Linear regression of cGAMP exported per cell over time. Data reanalyzed from periment in a and c. Linear regression is representative of regressions from 3 independent validations (Supplementary Fig. 1a, b).

e, The fraction of extracellular/total cGAMP molecules (left y-axis) (data reanalyzed from experiment in a and c) compared to the fraction of extracellular/total lactate dehydrogenate (LDH) activity as a proxy for cell death (right y-axis). Data from lactate dehydrogenase assay are from 1 experiment, representative of 4 independent validations (Supplementary Fig. 1d–f); 2 cell culture replicates are plotted. Linear regressions were performed for data shown.

f, Expression of cGAS in 293T WT, 293T stably expressing mouse cGAS, and 293T stably expressing human cGAS assessed by western blot. Data are from 1 experiment (full scan of blot available in Source Data).

g, Intracellular and extracellular concentrations of cGAMP from 293T stably expressing mouse cGAS and 293T stably expressing human cGAS. Cells were transfected ± 0.5 μg/mL empty pcDNA6 vector. After 24 hours, cells were refreshed with serum-free media and incubated for another 24 hours before measuring cGAMP. Data are from 1 experiment, representative of 2 independent validations (Supplementary Fig. 1g); 2 cell culture replicates are plotted.

In a, b, c, d, e, and g, cGAMP is measured by LC-MS/MS and replicates are plotted individually with a bar representing the mean.

We have previously shown that there are multiple cGAMP importers, including the solute carrier SLC19A1^19,20. We postulate that there are also multiple cGAMP export mechanisms. To characterize the cGAMP export mechanism in 293T cells, we first determined whether cGAMP was enclosed in extracellular vesicles (as previously reported^13,14) or freely soluble. We filtered conditioned media from 293T cGAS ENPP1^−/−cells through a 10 kDa MWCO filter, which retains extracellular vesicles and proteins. cGAMP passed through the filter, suggesting that it is exported as a freely soluble molecule (Fig. 2a). To further confirm that extracellular cGAMP exported by 293T cells is predominantly in a soluble form, we used CD14⁺ human peripheral blood mononuclear cells (PBMCs) as a reporter. These cells have previously been shown to take up soluble cGAMP, which leads to IFN-β production^18,19. We observed that CD14⁺ PBMCs respond to submicromolar concentrations of soluble cGAMP by upregulating IFNB1 (Extended Data Fig. 2a, b). Conditioned media from DNA-transfected cGAS-expressing 293T cGAS ENPP1^low cells, but not DNA-transfected 293T cells, induced IFNB1 expression in CD14⁺ cells, suggesting that the activity is a result of extracellular cGAMP produced by 293T cells (Fig. 2b). Addition of purified soluble recombinant mouse ENPP1 (mENPP1) (Extended Data Fig. 2c) depleted detectable cGAMP in the conditioned media and also ablated this activity (Fig. 2c, d). Because soluble ENPP1 (MW = ~100 kDa) cannot permeate membranes and, thus, can only access soluble extracellular cGAMP, we conclude that 293T cells export soluble cGAMP.

Figure 2 |. Characterization of cGAMP export mechanism in 293T cGAS ENPP1^−/− cells.

a, Intracellular and extracellular cGAMP from 293T cGAS ENPP1^−/− cells measured ± filtering the media (10 kDa MWCO). Data are from 1 experiment, representative of 2 independent experiments (Supplementary Fig. 2a); 2 cell culture replicates are plotted.

b, Schematic of experiments in c and d. 293T or 293T cGAS ENPP1^low cells were transfected with empty pcDNA6 (0.5 μg/mL) and treated ± 20 nM recombinant mouse ENPP1 (mENPP1). Conditioned media was incubated with CD14⁺ PBMCs for 16 h.

c–d, Extracellular cGAMP and IFNB1 expression (mRNA levels were normalized to CD14 and fold induction calculated relative to untreated CD14⁺ cells). Data shown in each panel are from 1 experiment; 2 cell culture replicates are plotted. Data in c are representative of 3 independent validations (Supplementary Fig. 2b, c). Data in d are supported by data in Fig. 3b, e, h, j, and Fig. 4a.

e–f, The amount of total (e), intracellular (exponential fit), and extracellular (polynomial fit) (f) cGAMP over time produced by 293T cGAS ENPP1^−/− cells after transfection with empty pcDNA6 (1.5 μg/mL). Data and fits are from 1 experiment, representative of 3 independent validations (Supplementary Fig. 2d, e).

g, cGAMP export rates plotted against intracellular concentration of cGAMP, reanalyzed from data in f. Data are fit with allosteric sigmoidal model$v_{\text{export}}=V_{\max }{\left[substrate\right]}^n/{\left(K_{\text{half}}\right)}^n+{\left[substrate\right]}^n$, where V_max = maximal transporter velocity, n = Hill slope, and K_half = substrate concentration at half V_max. If we constrain V_max < 5000 molecules cell⁻¹ s⁻¹, then K_m > 60 μM. Analysis is from the same experiment shown in e–f, representative of 3 independent validations (Supplementary Fig. 2d, e).

h, Potential mechanisms of cGAMP export.

In a, c, and d, replicates are plotted individually with a bar representing the mean.

We then determined ATP-dependence of the dominant cGAMP exporter in 293T cells. We depleted ATP for an hour and cell viability was not affected (Extended Data Fig. 2d). In this time period, the intracellular cGAMP concentration remained constant, so did the cGAMP export activity (Extended Data Fig. 2e). Therefore, in this cell line, cGAMP is not exported by exocytosis or ATP-hydrolyzing pumps, but by an ATP-independent transporter or channel, likely driven by the electrochemical gradient across the cell membrane. To further characterize the kinetics of the dominant exporter, we varied intracellular cGAMP concentrations by dsDNA stimulation. Because there is no cGAMP degradation in this ENPP1^−/− cell line, the total amount of cGAMP synthesized is the sum of intracellular and extracellular cGAMP (Fig. 2e, f). The synthesis rate was linear for the first 12 h and then slowed slightly, possibly due to loss of the dsDNA/cGAS complex over time. By plotting export rates as a function of intracellular cGAMP concentrations, we observed that the vexport did not plateau in the concentration range we tested (Fig. 2g). These kinetics are characteristic of a channel (no measurable K_m) or an allosterically controlled transporter with a V_max > 5000 molecules cell⁻¹ s⁻¹ and a K_m > 60 μM, as the curve appears sigmoidal instead of hyperbolic. This characterization will aid in future identification of cGAMP exporter(s) (Fig. 2h).

Development of a cell impermeable ENPP1 inhibitor to enhance extracellular cGAMP activity

Having established the presence of extracellular cGAMP by carefully removing sources of ENPP1 from culture conditions, we then determined whether intracellular and/or extracellular cGAMP is degraded by ENPP1. Despite its extracellular annotation, it is possible that ENPP1 could flip orientation on the membrane, as reported for a related enzyme CD38²⁹, or it could be active when being synthesized in the ER lumen and cGAMP may cross the ER membrane. To investigate the localization of ENPP1 activity, we transfected 293T cGAS ENPP1^−/− cells with human ENPP1 expression plasmid and confirmed its activity in whole cell lysates (Fig. 3a). In intact cells, ENPP1 expression depletes extracellular cGAMP, but does not affect the intracellular cGAMP concentration (Fig. 3b). Therefore, only extracellular cGAMP is regulated by ENPP1 in these cells. However, we cannot exclude the possibility that ENPP1 has intracellular activity in other cell types or under certain stimulations (Fig. 3c).

Figure 3 |. Development of a cell impermeable ENPP1 inhibitor to enhance extracellular cGAMP activity.

a–b, 293T cGAS ENPP1^−/− cells were transfected with pcDNA6 (empty or containing human ENPP1). ENPP1 expression and activity (³²P-cGAMP hydrolysis by thin layer chromatography (TLC), pH 9.0) (a, full scan of blot available in Source Data) and cGAMP concentrations (b) are shown. Data are from 1 experiment, representative of 3 independent validations (Supplementary Fig. 3a–d); 2 cell culture replicates are plotted in b.

c, Possible cellular locations of ENPP1.

d, Structure of ENPP1 inhibitor STF-1084.

e, Inhibition by STF-1084. In vitro (³²P-cGAMP TLC assay, pH 7.5, purified mouse ENPP1: K_i,app = 110 nM. Three independent experiments are plotted. In cells (cGAMP export assay, human ENPP1 transfected into 293T cGAS ENPP1^−/− cells): IC₅₀ = 340 nM. Data are from 1 experiment; 2 cell culture replicates are plotted (supported by data in Fig. 3h, j).

f, Mean apparent permeability (P_app) for STF-1084 and controls. For PAMPA and MDCK, mean was calculated from 2 cell culture replicates, 1 experiment. For Caco-2, mean was calculated from 2 independent experiments.

g, PBMCs were incubated with STF-1084 for 16 h. Data are from one experiment, representative of 2 independent validations (Supplementary Fig. 3e); 3 cell culture replicates are plotted.

h, Intracellular and extracellular cGAMP for 293T cGAS ENPP1^−/− cells transfected with pcDNA6 (empty or containing human ENPP1) and treated ± 10 μM STF-1084 after 24 hours. Data are from 1 experiment, representative of 2 independent validations (Supplementary Fig. 3f); 3 cell culture replicates are plotted.

i, Schematic of experiment in j. 293T cGAS ENPP1^low cells were transfected with human ENPP1 and treated ± 10 μM STF-1084. Conditioned media was incubated with CD14⁺ PBMCs for 16 h.

j, Extracellular cGAMP and IFNB1 expression (mRNA levels were normalized to CD14 and fold induction calculated relative to untreated CD14⁺ cells). Data are from 1 experiment (supported by data in Fig. 2d, Fig. 3b, e, h, and Fig. 4a); 2 cell culture replicates are plotted.

k, PBMCs were electroporated ± 200 nM cGAMP and incubated ± 50 μM STF-1084 for 16 h. IFNB1 and CXCL10 mRNA levels were normalized to ACTB and fold induction calculated relative to untreated cells. Data are from 1 experiment, representative of 2 independent experiments (Supplementary Fig. 3g, h); 2 cell culture replicates are plotted.

In b, e, g, h, j, and k, replicates are plotted individually with a bar representing the mean. BQL = below quantification limit.

To study the physiological relevance of extracellular cGAMP, we sought to develop cell impermeable ENPP1 inhibitors that only affect extracellular ENPP1 activity. We first tested a nonspecific ENPP1 inhibitor QS1^30,31 (Extended Data Fig. 3a, b). QS1 can inhibit extracellular cGAMP degradation in cells overexpressing ENPP1. However, in ENPP1 null cells, QS1 also increased intracellular cGAMP and decreased extracellular cGAMP concentrations, suggesting that it blocks the cGAMP exporter(s) (Extended Data Fig. 3c). This export blockage activity excludes QS1 as a tool to study extracellular cGAMP. We therefore designed a phosphonate analog, STF-1084, to chelate Zn²⁺ at the ENPP1 catalytic site and to minimize cell permeability and avoid intracellular off-targets (Fig. 3d). STF-1084 is sixty-fold more potent than QS1 (K_i,app = 110 nM in an in vitro biochemical assay) (Fig. 3e, Extended Data Fig. 3b).

We confirmed that STF-1084 is cell impermeable by performing three independent permeability assays: the parallel artificial membrane permeability assay (PAMPA), the intestinal cells Caco-2 permeability assay, and the epithelial cells MDCK permeability assay (Fig. 3f, Extended Data Fig. 3d). Compared to control compounds with high cell permeability and low cell permeability, STF-1084 falls into the category of impermeable compounds in all three assays. In addition, it has low activity towards the closely related ectonucleotidases alkaline phosphatase (K_i,app > 100 μM) and ENPP2 (K_i,app = 5.5 μM) (Extended Data Fig. 3e). Although we do not expect STF-1084 to have intracellular off-targets due to its low cell permeability, we tested its binding against a panel of 468 kinases to further determine its specificity. Despite its structural similarity to AMP, STF-1084 binds weakly to only two kinases at 1 μM (Extended Data Fig. 3f). STF-1084 also shows high stability (t_1/2 > 159 min) in both human and mouse liver microsomes, and is non-toxic to primary human PBMCs at 100 μM (Fig. 3g). Together, we demonstrated that STF-1084 is a potent, cell impermeable, specific, stable, and non-toxic ENPP1 inhibitor.

Next, we measured the efficacy of STF-1084 in maintaining extracellular cGAMP concentrations of ENPP1 overexpressing 293T cGAS cells and obtained an IC₅₀ value of 340 nM, with 10 μM being sufficient to completely block extracellular cGAMP degradation (Fig. 3e). Unlike QS1, STF-1084 had no effect on intracellular cGAMP, demonstrating that it does not affect cGAMP export (Fig. 3h). Finally, we tested the efficacy of STF-1084 in boosting extracellular cGAMP signal detectable to CD14⁺ PBMCs. Conditioned media from ENPP1 overexpressing 293T cGAS cells failed to induce IFNB1 expression in CD14⁺ cells (Fig. 3i, j). The presence of STF-1084 rescued extracellular cGAMP levels in the media and induction of IFNB1 expression in CD14⁺ cells (Fig. 3j). STF-1084 had no effect on cytokine production when cGAMP was electroporated into primary human PBMCs (Fig. 3k), demonstrating that STF-1084 only boosts extracellular cGAMP signaling by preventing its degradation by ENPP1.

cGAMP export by cultured cancer cells is continuous at steady state and can be induced by ionizing radiation

Chromosomal instability of cancer cells has been reported to lead to micronuclei formation and rupture in the cytosol, and cGAS accumulates at these regions^25–27,32. With STF-1084 as a specific extracellular ENPP1 inhibitor, we were poised to test whether cancer cells produce cGAMP and export it. In unstimulated 4T1-luc cells (a mouse triple negative metastatic cancer cell line with a luciferase reporter), we were able to detect 350,000 molecules/cell (~150 nM) of intracellular cGAMP (Extended Data Fig. 4a). When we knocked down cGAS in these cells and reduced its protein level by approximately 3-fold, we also detected a reduction of cGAMP concentration by 3-fold (Extended Data Fig. 4a). In addition, we detected 350,000 molecules/cell extracellular cGAMP in the media after 48 hours (Fig. 4a). Incubating the media with recombinant ENPP1 abolished the cGAMP signal, demonstrating that extracellular cGAMP is in a free soluble form (Fig. 4a). Strikingly, when we used STF-1084 to inhibit cell surface and soluble ENPP1 in the cell culture media, we measured 3,000,000 molecules/cell of extracellular cGAMP after 48 hours (Fig. 4a). This is approximately 10-fold more than the amount of intracellular cGAMP, demonstrating that these cells export at least 90% of the cGAMP they synthesize every 48 hours. We detected similar levels of extracellular cGAMP in both mouse (4T1-luc, E0771, MC38) and human cancer cells lines (MDA-MB-231 and MCF-7), as well as the immortalized normal mouse mammary gland cells, NMuMG (Fig. 4b). On the contrary, we detected much lower extracellular cGAMP levels in 293T cells, which have very low cGAS expression, and in primary human PBMCs, which have high cGAS expression, but likely no cytosolic dsDNA to stimulate cGAS activity (Fig. 4b). We measured export over time in MC38 cells and it followed linear kinetics as observed in our model 293T cGAS ENPP1^−/− cells (Extended Data Fig. 4b). Besides the 4T1-luc cells, the cells we tested (E0771, NMuMG, and Panc02) had intracellular cGAMP levels below our limit of detection, corresponding to less than ~40 nM intracellular cGAMP (Extended Data Fig. 4c). Interestingly, out of the cell lines that we tested for intracellular cGAMP, only 4T1-luc cells have undetectable amounts of STING protein (Extended Data Fig. 4d). It is possible that cancer cells upregulate their cGAMP export mechanism(s) to clear intracellular cGAMP to avoid activation of their own STING and subsequent IFN-β production.

Figure 4 |. Cancer cells continuously export cGAMP in culture at steady state and ionizing radiation further increases export.

a, Extracellular cGAMP produced by 4T1-luc cells after 48 hours in serum-containing media in the absence or presence of 50 μM STF-1084. Conditioned media without STF-1084 was also incubated with 10 nM of recombinant mENPP1 overnight. Data are from 1 experiment (supported by data in Fig. 3b, e, h, and j; 3 cell culture replicates are plotted.

b, Extracellular cGAMP produced by 4T1-luc (5 cell culture replicates from 2 independent experiments), E0771 (2 cell culture replicates), MC38 (2 cell culture replicates), NMuMG (3 cell culture replicates), MDA-MB-231 (2 cell culture replicates), MCF-7 (2 cell culture replicates), 293T (3 cell culture replicates), and primary human PBMCs (6 cell culture replicates from 2 donors) measured after 48 h in the presence of 50 μM STF-1084. BQL = below quantification limit. Data are from 1 experiment except where indicated for 4T1-luc.

c, Extracellular cGAMP produced by cancer cell lines 4T1-luc, E0771, Panc02, Neuro-2a, MDA-MB-231, and HeLa after 24 and 48 hours. At time 0, cells were left untreated or treated with IR (8 Gy or 20 Gy) and refreshed with media supplemented with 50 μM STF-1084. Data for each cell line are from 1 experiment. For 4T1-luc and E0771, data are representative of 2 independent validations (Supplementary Fig. 4); 2 cell culture replicates.

In all panels, cGAMP is measured by LC-MS/MS and replicates are plotted individually with a bar representing the mean. BQL = below quantification limit.

Ionizing radiation (IR), as a standard cancer treatment, has been shown to increase erroneous chromosomal segregation and cytosolic DNA^{25–27,33,34}. Indeed, IR increased extracellular cGAMP production in all the cancer cell lines we tested (4T1-luc, E0771, Panc02, Neuro-2a, MDA-MB-231, and HeLa) (Fig. 4c) while causing negligible amounts of cell death (Extended Data Fig. 4e). Interestingly, IR induced more than 10-fold higher extracellular cGAMP in E0771 cells than in other cells lines, despite their similar levels of basal extracellular cGAMP. Together, our data demonstrate that both mouse and human cancer cells, regardless of their tissue of origin, constantly produce and efficiently export cGAMP and can be stimulated with IR to produce more extracellular cGAMP.

Extracellular cGAMP produced by cancer cells and sensed by host STING is responsible for the curative effect of ionizing radiation

We next directly probed the physiological function of extracellular cGAMP in mouse models. First, to determine the importance of cancer versus host cGAMP, we knocked out Cgas in cancer cells (Extended Data Fig. 5a) and utilized Cgas^−/− and Sting^−/− mice in the C57BL/6 background (Fig. 5a). We also developed a neutralizing protein agent, which should not be cell permeable, as a tool to specifically sequester extracellular cGAMP. We took advantage of the soluble cytosolic domain of STING (Fig. 5b), which binds cGAMP with a K_d of 73 ± 14 nM (Fig. 5c). We also generated an R237A¹⁷ mutant STING as a non-binding STING control (Fig. 5b–d). In cGAMP-treated human CD14+ PBMCs, wild type (WT) STING (neutralizing STING) was able to neutralize extracellular cGAMP with the predicted 2:1 stoichiometry, while the non-binding STING had no effect (Fig. 5e, Extended Data Fig. 5b). We observed similar results in primary mouse bone marrow cells (Extended Data Fig. 5c, d), validating in vitro the use of the STING proteins as tools to probe extracellular cGAMP.

Figure 5 |. Extracellular cGAMP is produced by cancer cells and sensed by host STING.

a, Experimental setup to assess the role of extracellular cGAMP in vivo.

b, Coomassie gel of recombinant cytosolic mouse WT and R237A STING. Data are representative of 5 independent experiments.

c, Binding curves for neutralizing STING (WT) and non-binding STING (R237A) determined by a membrane binding competition assay with ³⁵S-cGAMP. Data from 2 independent experiments are plotted.

d, Structure of WT STING with cGAMP with R237 highlighted in pink (PDB ID 4LOJ).

e, IFNB1 mRNA fold induction in CD14⁺ PBMCs treated with 2 μM cGAMP in the presence of neutralizing or non-binding STING (2 μM to 100 μM, 2.5-fold dilutions). Data are from 1 experiment (supported by data in Extended Data Fig. 5 b–d); 2 qPCR replicates are plotted with a bar representing the mean.

f–i, The indicated cells (1x10⁶ of E0771 or 1x10⁶ of 4T1-luc) of the indicated genetic background (WT or Cgas^−/−) were orthotopically injected into mice (C57BL/6 for E0771 or BALB/cJ for 4T1-luc) of the indicated genetic background (WT, Cgas^−/−, or Sting^−/−) on day 0. Non-binding STING, neutralizing STING, PBS, or recombinant mouse ENPP1 (mENPP1) were intratumorally injected on day 2. Tumors were harvested and analyzed by FACS on day 3. Sample sizes of n mice, from left to right (non-binding STING, neutralizing STING): (f) n = (5, 5); (4, 5); (5, 5); (5, 4). (g) n = (5, 5); (5, 5); (5, 4). (h) n = (2, 3); (5, 5). (i) n = (5, 6). Mean ± SD, unpaired two-tailed t test with Welch’s correction. j, E0771 cells (5x10⁴) were orthotopically injected into WT (n = 10 mice) or Enpp1^−/− (n = 6 mice) C57BL/6J mice. P value for tumor volume determined by pairwise comparisons using post hoc tests with a Tukey adjustment and for Kaplan Meier curve determined using the Log-rank Mantel-Cox test.

We established E0771 orthotopic tumors in mice, followed by intratumoral injection of neutralizing STING to deplete extracellular cGAMP, and excision of the tumors to stain for tumor-associated leukocytes. In WT E0771 tumors, neutralizing STING significantly decreased the CD11c⁺ (dendritic cell, DC) and the CD103⁺ CD11c⁺ (conventional type I dendritic cell, cDC1) populations³⁵ in the CD45⁺ MHC II⁺ (tumor-associated antigen presenting cell, APC) population (Fig. 5f, g and Extended Data Fig. 6a). This suggests that extracellular cGAMP can be detected by the immune system to activate DCs that are important for the anti-cancer response³⁵. Extracellular cGAMP depletion also diminished the CD11c⁺ population when tumors are grown in Cgas^−/− mice, suggesting that host cells do not contribute significantly to extracellular cGAMP production (Fig. 5f). In contrast, extracellular cGAMP depletion did not affect the CD11c⁺ or CD103⁺ CD11c⁺ population when Cgas^−/− E0771 cells or Sting^−/− mice were used (Fig. 5f, g). Depleting extracellular cGAMP did not affect the F4/80⁺ (macrophage) population in the APC population in any of the experiments (Extended Data Fig. 6b). Together, our data demonstrate that cancer cells, but not host cells, are the dominant producers of extracellular cGAMP, which is then sensed by host STING and leads to infiltration of DCs, in particular cross-presenting cDC1s.

We also tested the orthotopic 4T1-luc tumor model. Although Cgas and Sting knockout strains have not been established in the BALB/c background, we knocked out Cgas in the 4T1-luc cells. Intratumoral injection of neutralizing STING into the WT 4T1-luc tumors significantly decreased the tumor-associated CD11c⁺ population in the CD45⁺ MHC II⁺ population (Fig. 5h). In contrast, extracellular cGAMP depletion had no effect in Cgas^−/−4T1-luc tumors (Fig. 5h). As an orthogonal approach, we depleted extracellular cGAMP by intratumoral injection of mouse ENPP1 protein (Extended Data Fig. 2c) and again observed diminished CD11c⁺ cells in the CD45⁺ MHC II⁺ population (Fig. 5i). Together, our results demonstrate that extracellular cGAMP is produced by cancer cGAS and sensed by host STING, which leads to increased immune cell infiltration.

We next tested whether this basal level of extracellular cGAMP can also limit tumor growth. Long term administration of neutralizing STING compared to the non-binding STING control did not alter the course of tumor progression in the E0771 model (Extended Data Fig. 6c), suggesting that endogenous ENPP1-mediated degradation was sufficient to abolish the anti-cancer effect of cancer-derived extracellular cGAMP. E0771 cells do not have particularly high ENPP1 activity (Extended Data Fig. 6d), but ENPP1 is also expressed on host cells and present in the serum as a soluble form^16,24,36. We therefore implanted WT E0771 cells into Enpp1^−/− mice and, indeed, observed slower tumor growth, suggesting that host ENPP1 promotes tumor growth in this model. (Fig. 5j).

We next tested the physiological role of endogenous extracellular cGAMP when stimulated by IR, without ENPP1 inhibition. It was previously reported that IR exerts tumor-shrinkage effects in a host STING-dependent manner³⁷ and activates cGAS-dependent IFN-β production in cancer cells^25,26,38. Indeed, IR treatment induced cytokine production in both 4T1-luc and E0771 cells (Extended Data Fig. 7). Since E0771 cells export high levels of cGAMP upon IR treatment (Fig. 4c), we investigated the role of extracellular cGAMP in the tumor shrinkage effect of IR in the E0771 breast tumor model. We did not observe a significant increase of dead cells or cleaved caspases in both CD45⁻and CD45⁺ cells in response to treatment with IR (8 Gy) at 24h, suggesting that IR does not directly kill cancer or immune cells (Extended Data Fig. 8a). Treating established E0771 tumors with 8 Gy of IR resulted in tumor-free survival in 4 out of 17 mice, demonstrating the efficacy of IR in this model (Fig. 6a). No tumor-free survival was observed in Sting^−/− mice, confirming that the curative effect of IR depends on host STING in this model, but probably not cytokines produced by cancer cells (Fig. 6a). We sequestered extracellular cGAMP by injecting neutralizing STING for the duration of the experiment with non-binding STING as a control. Remarkably, depletion of extracellular cGAMP completely abolished the curative effect of IR (Fig. 6b), suggesting that extracellular cGAMP-induced host STING activation accounts for the curative effect of IR in this model.

Figure 6 |. Extracellular cGAMP is responsible for the efficacy of ionizing radiation.

a, E0771 cells (5x10⁴) were orthotopically injected into WT (n = 17 mice) or Sting^−/− (n = 9 mice) C57BL/6J mice. The tumors were treated with IR (8 Gy) when they reached 100 ± 20 mm³. Tumor volumes for individual mice are plotted.

b, E0771 cells (5x10⁴) were orthotopically injected into WT C57BL/6J mice. The tumors were treated with IR (8 Gy) when they reached 100 ± 20 mm³ and injected with non-binding (n = 8 mice) or neutralizing STING (n = 10 mice) every other day for the duration of the experiment. Tumor volumes for individual mice are plotted.

c–d, E0771 cells (5x10⁴) were orthotopically injected into WT C57BL/6J mice. The tumors were treated with IR (8 Gy) when they reached 100 ± 20 mm³ and injected with non-binding (n = 5 mice) or neutralizing STING (n = 5 mice) on days 2 and 4 after IR. Tumors were harvested and analyzed by FACS on day 5. Mean ± SD, unpaired two-tailed t test.

To probe the cellular mechanism of extracellular cGAMP, we excised established tumors five days after treating them with IR followed by extracellular cGAMP depletion and analyzed the immune cell populations (Extended Data Fig. 8b). The amount of tumor infiltrating CD11c⁺ and F4/80⁺ cells were not significantly altered when depleting extracellular cGAMP with neutralizing STING compared to the non-binding STING control (Fig. 6c). However, the CD103⁺ CD11c⁺ subpopulation decreased, indicating weakened cross-presentation when extracellular cGAMP was depleted (Fig. 6c). Although the numbers of CD8⁺ T cells were not altered significantly (Fig. 6d), expression levels of the activation markers CD62L, CD25, and Granzyme B were dampened when extracellular cGAMP was depleted (Fig. 6d). Together, our results suggest that IR leads to complete tumor regression by increasing the extracellular cGAMP level, which then directly or indirectly increases tumor infiltrating cross-presenting cDC1s, leading to activation of CD8⁺ cytotoxic T cells.

ENPP1 inhibitors synergize with IR to shrink tumors

ENPP1 is highly expressed in some breast cancers and its level has been correlated with poor prognosis^39–41 (Fig. 7a). High ENPP1 expression may be a mechanism that breast cancers utilize to deplete extracellular cGAMP and dampen immune detection. We measured ENPP1 activities in three triple negative breast cancer cells 4T1-luc, E0771, and MDA-MB-231, with MDA-MB-231 and 4T1-luc exhibiting high ENPP1 activities (Extended Data Fig. 6d). We, therefore, chose the 4T1-luc mouse model to probe the effect of ENPP1 on tumor immune detection, growth, and responses to treatment. We first tested the effect of ENPP1 on tumor infiltrating dendritic cells. Three days after implanting WT and Enpp1^−/− tumors (Extended Data Fig. 9a) orthotopically in mice, we excised the tumors and analyzed their tumor-associated leukocyte compositions. Enpp1^−/− tumors have a larger tumor-associated CD11c⁺ population than WT tumors when left untreated or treated with IR (20 Gy) (Fig. 7b). To further test that this effect was due to increased extracellular cGAMP, but not potential membrane scaffolding effects of ENPP1, any unidentified intracellular activity, or because these 4T1-luc cells express Cas9 due to the CRISPR knock out procedure, we used our cell impermeable ENPP1 inhibitor. We intratumorally injected STF-1084 immediately after IR treatment and tested its effect after 24 h. Indeed, compared to vehicle control, STF-1084 mirrored the effect of Enpp1^−/− on increasing the tumor-associated CD11c⁺ population (Fig. 7c).

Figure 7 |. Genetic and pharmacological inhibition of ENPP1 increases immune detection of cancer and synergizes with ionizing radiation.

a, ENPP1 expression in human cancers. Data are represented as box plots showing range of mRNA expression levels. RNA sequencing data from the TCGA Research Network (https://www.cancer.gov/tcga) PanCanAtlas and visualized with cBioPortal.

b, 4T1-luc WT or Enpp1^−/− cells (1x10⁶) were orthotopically injected into WT BALB/cJ mice on day 0. Tumors were left untreated or treated with IR (20 Gy) on day 2. Tumors were harvested and analyzed by FACS on day 3. n = 5 mice for all groups. Mean ± SD, unpaired two-tailed t test with Welch’s correction.

c, 4T1-luc cells (1x10⁶) were orthotopically injected into WT BALB/cJ mice on day 0. Tumors were treated with IR (20 Gy) and intratumorally injected with PBS (n = 4 mice) or STF-1084 (n = 5 mice) on day 2. Tumors were harvested and analyzed by FACS on day 3. Mean ± SD, unpaired two-tailed t test with Welch’s correction.

d, Established 4T1-luc WT (harboring scrambled sgRNA) or Enpp1^−/− tumors (100 ± 20 mm³) were treated once with IR (20 Gy) followed by three intratumoral injections of 10 μg cGAMP on day 2, 4, and 7 after IR (n = 10 mice for WT, n = 11 mice for Enpp1^−/−). Tumor volumes for individual mice are shown. P value determined by pairwise comparisons using post hoc tests with a Tukey adjustment at day 20. In the Enpp1^−/−4T1-luc + IR (20 Gy) + cGAMP treatment group, 3/11 mice were tumor free verified by bioluminescent imaging (tumor area outlined in red).

d, Established 4T1-Luc tumors (100 ± 20 mm³) were treated once with IR (20 Gy) followed by three intratumoral injections of 10 μg cGAMP alone (n = 9 mice) or 10 μg cGAMP + 100 μL of 1 mM STF-1084 (n = 10 mice) on day 2, 4, and 7 after IR. Tumor volumes for individual mice are shown. P value determined by pairwise comparisons using post hoc tests with a Tukey adjustment at day 40.

We then tested the effect of ENPP1 expressed by 4T1-luc on tumor rejection. We did not observe significant growth delay of 4T1-luc Enpp1^−/− tumors compared to WT tumors harboring a scrambled sgRNA and Cas9 when they were left untreated, or treated with IR or intratumoral cGAMP injections individually (Extended Data Fig. 9b–d). Strikingly, 3/11 mice inoculated with Enpp1^−/− tumors achieved tumor free survival when treated with a combination of IR and cGAMP injections, whereas no mice inoculated with WT tumors survived (Fig. 7d). This demonstrates that ENPP1 expressed on the surface of 4T1 cells was sufficient to abolish the tumor shrinkage effect of combination therapy. Together, we have demonstrated that ENPP1 expressed on both cancer cells (using the 4T1 model) and by the host (using the E0771 model) play a role in clearing extracellular cGAMP and promote tumor growth.

Although it is still an open question whether cancer or host ENPP1 plays a bigger role, small molecule inhibitors should, in principle, inhibit both. We tested the ENPP1 inhibitor STF-1084 in this combination therapy. STF-1084 has fast pharmacokinetics. Without optimizing its route of administration and pharmacokinetic properties, we intratumorally injected it into established orthotopic 4T1-luc tumors. STF-1084 synergized with IR and cGAMP to significantly delay tumor progression, and resulted in tumor free survival of 1/10 mice. (Fig. 7e).

Other than breast cancers, pancreatic cancers also express high levels of ENPP1 (Fig. 7a)^42,43. To be able to access these tumors that are not easily accessible through intratumoral injections, we sought to develop an analog of STF-1084 that can be administered systemically. We developed STF-1623 (Extended Data Fig. 10a), with improved K_i,app of 16 nM (Extended Data Fig. 10b). We performed a similar suite of assays on STF-1623 as we did on STF-1084. We confirmed that it is also cell impermeable (Extended Data Fig. 10c, d), is not toxic to primary human PBMCs (Extended Data Fig. 10e) does not target kinases (Extended Data Fig. 10f), and is stable to human and mouse microsomes (t_1/2 > 159 minutes). In addition, STF-1623 affected only extracellular cGAMP concentrations (Extended Data Fig. 10g) and had no effect on cytokine production when cGAMP was electroporated into primary human PBMCs (Extended Data Fig. 10h). Importantly, STF-1623 shows an improved pharmacokinetic profile compared to STF-1084 for systemic dosing. With subcutaneous injections, we can achieve >100 nM plasma concentrations after 24 hours (Extended Data Fig. 10i).

We then tested STF-1623 in the Panc02, syngeneic, subcutaneous pancreatic tumor model. STF-1623 delayed tumor growth as a single agent and synergized with IR to delay tumor growth, as well as achieve stable disease and tumor regression in some mice. (Extended Data Fig. 10j). Since this Panc02 model is not metastatic, we predict that tumor regression would lead to increased survival. Future studies are needed to determine if treatment with STF-1623 leads to a survival advantage in metastatic models. Together, our results demonstrate that the anti-tumor effect of extracellular cGAMP can be enhanced by inhibiting its degradation enzyme ENPP1. STF-1623 serves as a starting point for new classes of anti-cancer agents that can synergize with the endogenous extracellular cGAMP exported by cancer cells and induced by IR.

Extracellular cGAMP in cancer

Here, we provide in vitro and in vivo evidence that cGAMP can signal through the extracellular space. In all the cell types we have tested, cGAMP can be exported at various levels, suggesting that cGAMP export is not a cancer specific phenomenon and cGAMP exporters are likely expressed in most cell types. Since chromosomal instability and aberrant cytosolic dsDNA are cancer-intrinsic properties^44,45 and cancer cells rarely inactivate cGAS²⁷, we reason that cGAMP overproduction and increased export may also be properties intrinsic to cancer cells. Since no cytosolic cGAMP hydrolase has been identified and ENPP1 cannot degrade intracellular cGAMP, export is currently the only known mechanism by which cGAMP is removed from the cytosol, and represents another way to turn off intracellular STING signaling in addition to ubiquitin-mediated STING degradation⁴⁶. This clearance mechanism, however, exposes cancer cells to immune detection.

Indeed, our results demonstrate that cGAMP exported by cancer cells is a danger signal detected by the immune system. It is well known that neoantigens from cancer cells are presented by APCs to cross prime cytotoxic CD8⁺ T cells that eventually perform cancer-specific killing^7,18. However, it is less understood how APCs initially detect cancer cells. It has been shown that immunogenic tumors release dsDNA as a danger signal to CD11c⁺ dendritic cells^7,47. The evidence for IFN as a danger signal is mixed; one study showed that cancer cells respond to their own cytosolic dsDNA induced by IR and produce IFNs as a danger signal⁴⁸, whereas other studies have reported that cancer cells can lose their ability to make IFN via the STING pathway, or even repurpose the STING pathway to aid in metastasis^27,49. A recent study showed that the catalytic activity of cancer cGAS correlates with anti-cancer immunity in the B16 melanoma model in a host STING dependent manner¹⁵, but the mechanism of this suggested transfer of cGAMP from cancer to host cells was unknown. Here, we provide direct evidence that cancer cells produce soluble extracellular cGAMP as a danger signal, which leads to increased numbers of dendritic cells, specifically cross-presenting type I conventional dendritic cells, and cytotoxic T cell activation in the tumor microenvironment. cGAMP export is an important mode of cGAMP communication among cells that are not physically connected but are in close proximity. Unlike cytokines, it is unlikely that extracellular cGAMP can travel long distances in the extracellular space without being degraded and/or diluted to below its effective concentrations. We call cGAMP an immunotransmitter, due to these shared properties with neurotransmitters and its immune signaling functions. Extracellular cGAMP should be studied both for its basic biology as an immunotransmitter and for its therapeutic potential in cancer.

Methods:

Reagents and antibodies

[α−³²P]ATP (800 Ci/mmol, 10 mCi/mL, 250 μCi) and [³⁵S]ATPαS (1250 Ci/mmol, 12.5 mCi/mL, 250 μCi) were purchased from Perkin Elmer. Adenosine triphosphate, guanosine triphosphate, adenosine-¹³C₁₀,¹⁵N₅, 5’-triphosphate, 4-nitrophenyl phosphate, and bis(4-nitrophenyl) phosphate were purchased from Sigma-Aldrich. 2’3’-cGAMP and isotope labeled cGAMP were synthesized as described previously¹⁹. Caco-2 assay was purchased from Cyprotex. Kinome screens were conducted by Eurofins (data visualized using TREEspot™ Software Tool and reprinted with permission from KINOMEscan®, a division of DiscoveRx Corporation. ©DiscoveRX Corporation 2010). PAMPA and MDCK permeability assays were conducted by Quintara Discovery. Total protein content was quantified using the BCA assay (Thermo Fisher). Cell viability was quantified using the CellTiterGlo assay (Promega) or the lactate dehydrogenase assay (Pierce, Thermo Fisher). Mouse CXCL10 production was quantified with the mouse CXCL10/IP-10/CRG-2 DuoSet ELISA (R&D Systems) and the TMB substrate reagent set (BD Bioscience). Full length human ENPP1 was cloned into pcDNA6 vector. QS1 was synthesized as previously described³¹. The following monoclonal antibodies were used for western blotting: rabbit anti-cGAS (D1D3G Cell Signaling, 1:1,000) rabbit anti-mouse cGAS (D2O8O Cell Signaling, 1:1,000), mouse anti-tubulin (DM1A Cell Signaling, 1:2,000), rabbit anti-STING (D2P2F Cell Signaling, 1:1,000), rabbit anti-phospho-IRF3 (4D4G Cell Signaling, 1:1,000), rabbit anti-IRF3 (D83B9 Cell Signaling, 1:1,000), IRDye 800CW goat anti-rabbit (LI-COR, 1:15,000), and IRDye 680RD goat anti-mouse (LI-COR, 1:15,000).

Mammalian cell lines and primary cells:

293T, NMuMG, MDA-MB-231, HeLa, MCF-7, and Neuro-2a were procured from ATCC, Panc02 was procured from the DTP/DCTD/NCI Tumor Repository, E0771 was procured from CH3 BioSystems, 4T1-luciferase (4T1-luc) was a gift from Dr. Christopher Contag (Stanford University, Stanford CA, USA)⁵⁰, and HEK293S GnT1⁻ expressing secreted mENPP1 was a gift from Dr. Osamu Nureki (University of Tokyo, Tokyo, Japan)⁵¹. All cell lines were maintained in a 5% CO₂ incubator at 37°C. 293T, Neuro-2a, MC38, MDA-MB-231, HeLa, and L929 were maintained in DMEM (Corning Cellgro) supplemented with 10% FBS (Atlanta Biologics) (v/v) and 100 U/mL penicillin/streptomycin (P/S) (ThermoFisher). NMuMG and MCF-7 were maintained in DMEM supplemented with 10% FBS, 100 U/mL P/S, and 10 μg/mL bovine insulin (Millipore Sigma). 4T1-luc, and Panc02 were maintained in RPMI (Corning Cellgro) supplemented with 10% FBS and 100 U/mL P/S. E0771 cells were maintained in RPMI supplemented with 10% FBS, 100U/ml P/S, and 10 mM HEPES. Primary human peripheral blood mononuclear cells (PBMCs) were isolated by subjecting enriched buffy coat from whole blood (Stanford Blood Center) to a Percoll density gradient. CD14⁺ PBMCs were isolated using CD14⁺ MicroBeads (Miltenyi). PBMCs were cultured in RPMI supplemented with 2% human serum and 100 U/mL P/S. Primary mouse bone marrow cells were isolated by opening the ends of the femur and tibia removing cells by centrifugation⁵² and were cultured in RPMI supplemented with 10% FBS and 100 U/mL P/S. Bone marrow cells were differentiated into bone-marrow derived macrophages (BMDMs) by culturing as above, plus 10% conditioned media from L929 cells.

Making cell lines:

293T cells were virally transfected to stably express mouse or human cGAS. 293T cGAS ENPP1^low cells were created by viral transfection of CRISPR sgRNA targeting human ENPP1 (5’-CACCGCTGGTTCTATGCACGTCTCC-3’), and 293T mcGAS ENPP1^−/− cells were selected after single cell cloning from this pool. 4T1 and E0771 Cgas^−/− cells were created by viral transfection of CRISPR sgRNA (using lentiCRISPRv2-blast, Addgene plasmid #83480⁵³) targeting mouse Cgas (5’-CACCGGAAGGGGCGCGCGCTCCACC-3’). Cells were single cell cloned, and multiple knockouts were pooled after verification by western blotting. 4T1-luc Enpp1^−/− and scrambled cells were created by viral transfection of CRISPR sgRNAs (using lentiCRISPRv2-blast) targeting mouse Enpp1 (5’- GCTCGCGCCCATGGACCT-3’ and 5’- ATATGACTGTACCCTACGGG −3’) or a scrambled sequence. Cells were selected with 0.5 – 2 μg/mL blasticidin, single cell cloned, and multiple knockouts were pooled after verification by activity assay. Alternatively, 4T1-luc Enpp1^−/− and scrambled cells were created by transient transfection with Lipofectamine 3000 of the same CRISPR sgRNAs as above or a scrambled sequence (using PX458, Addgene plasmid # 48138), followed by single cell cloning of GFP positive cells. Multiple clean knockouts were pooled after verification by activity assay (commercial antibodies are not sensitive enough for verification of protein expression). 4T1-luc shcGAS cells were created by viral transfection of shRNA (5’-CAGGATTGAGCTACAAGAATAT-3’)⁴⁸ using the plasmid pGH188. Cells harboring the shRNA were selected with 0.5 – 2 μg/mL blasticidin, sorted for GFP expression, and were used as a pool.

Expression and purification of recombinant proteins

sscGAS was produced as described previously¹⁹. mENPP1 was produced as described previously^51,54.

Mouse STING (residues 139–378) was inserted into the pTB146 His-SUMO vector (a generous gift from T. Bernhard, Harvard Medical School) and expressed in Rosetta cells. Cells were grown in 2xYT medium with 100 μg/mL ampicillin and induced when the OD₆₀₀ reached 1 with 0.75 mM IPTG at 16 °C overnight. All subsequent procedures using proteins and cell lysates were performed at 4 °C. Cells were pelleted and lysed in 50 mM Tris pH 7.5, 400 mM NaCl, 10 mM imidazole, 2 mM DTT, and protease inhibitors (cOmplete, EDTA-free protease inhibitor cocktail Roche). Cells were lysed by sonication and the lysate was cleared by ultracentrifugation at 50,000 rcf for 1 hour. The cleared supernatant was incubated with HisPur cobalt resin (ThermoFisher Scientific; 1 mL resin per 1 L bacterial culture) for 30 minutes. The resin-bound protein was washed with 50 column volumes of 50 mM Tris pH 7.5, 150 mM NaCl, 2% triton X-114, 50 CV of 50 mM Tris pH 7.5, 1 M NaCl (each wash was set to a drip rate of 1 drop/2–3 seconds and took 2–3 hours), and 20 column volumes of 50 mM Tris pH 7.5, 150 mM NaCl. Protein was eluted from resin with 600 mM imidazole in 50 mM Tris pH 7.5, 150 mM NaCl. Fractions containing His-SUMO-STING were pooled, concentrated, and dialyzed against 50 mM Tris pH 7.5, 150 mM NaCl while incubating with the SUMOlase enzyme His-ULP1 to remove the His-SUMO tag overnight. The solution was incubated with the HisPur cobalt resin again to remove the His-SUMO tag, and STING was collected from the flowthrough. Protein was dialyzed against 20 mM Tris pH 7.5, loaded onto a HitrapQ anion exchange column (GE Healthcare) using an Äkta FPLC (GE Healthcare), and eluted with a NaCl gradient. Fractions containing STING were pooled and buffer exchanged into PBS.

Liquid chromatography-tandem mass spectrometry

Measurement of cGAMP: Cyclic GMP-¹³C₁₀,¹⁵N₅-AMP was used as an internal standard at 0.5–1 μM. Samples were analyzed for cGAMP, ATP, and GTP content on a Shimadzu HPLC (San Francisco, CA) with an autosampler set at 4°C and connected to an AB Sciex 4000 QTRAP (Foster City, CA). A volume of 10 μL was injected onto a Biobasic AX LC column, 5 μm, 50 x 3 mm (Thermo Scientific). The mobile phase consisted of 100 mM ammonium carbonate (A) and 0.1% formic acid in acetonitrile (B). Initial condition was 90% B, maintained for 0.5 min. The mobile phase was ramped to 30% A from 0.5 min to 2.0 min, maintained at 30% A from 2.0 min to 3.5 min, ramped to 90% B from 3.5 min to 3.6 min, and maintained at 90% B from 3.6 min to 5 min. The flow rate was set to 0.6 mL/min. The mass spectrometer was operated in electrode spray positive ion mode with the source temperature set at 500°C. Declustering and collision-induced dissociation were achieved using nitrogen gas. For each molecule, the MRM transition(s) (m/z), DP (V), and CE (V) are as follows: ATP (508 > 136, 341, 55), GTP (524 > 152, 236, 43), cGAMP (675 > 136, 121, 97; 675 > 312, 121, 59; 675 > 152, 121, 73), internal standard cyclic GMP-¹³C₁₀,¹⁵N₅-AMP (690 > 146, 111, 101; 690 > 152, 111, 45; 690 > 327, 111, 47).

Measurement of STF-1084 and STF-1623:

Measurements were performed using a Q-Exactive FT-mass spectrometer (Thermo) equipped with Vanquish uHPLC. Samples were diluted in water with 0.1% formic acid and injected onto a Phenomenex Synergi Hydro-RP column (4 μm particle size 2 mm ID, 30 mm length). The column compartment was at ambient temperature. The flow rate was 0.5 mL/min. Mobile phase A was water with 0.1% formic acid; mobile phase B was acetonitrile with 0.1% formic acid. Each run was five minutes; the gradient was as follows: 0–0.5 min 0% B, 0.5 to 2 min linear from 0 to 95% B, 2 to 3.5 min hold at 95% B, 3.5 to 3.6 min from 95% B to 0% B and 3.6 to 5 min at 0% B to re-equilibrate the column. Minutes 1 through 4.8 were sent to the mass spectrometer for analysis. Detection on the Q-Exactive was performed in positive mode between 100–1000 m/z, using an acquisition target of 1E6 with maximum IT of 100 ms at a resolution of 70,000. Quantification was done using TraceFinder 4.1 software (ThermoFisher).

Export assay in 293T cGAS ENPP1^−/− cells

293T cGAS ENPP1^−/− cells were plated in plates coated with PurCol (Advanced BioMatrix). In some experiments, the cells were transfected with indicated plasmids complexed with Fugene 6 (Promega) 24 hours prior to the export experiment. At the start of the experiment, the media was replaced with serum-free DMEM supplemented with 1% insulin-transferrin-selenium-sodium pyruvate (ThermoFisher) and 100 U/mL P/S. At indicated times, the media and cells were removed and centrifuged at 1000 rcf for 10 minutes at 4 °C. Cells were lysed in 30 to 100 μL of 50:50 acetonitrile:water supplemented with 500 nM internal standard and centrifuged at 15,000 rcf for 20 minutes at 4 °C. Media was supplemented with internal standard at 500 nM and 20% formic acid. If media cGAMP enrichment was necessary, the media was acidified with 0.5% acetic acid, supplemented with internal standard, and applied to HyperSep Aminopropyl SPE columns (ThermoFisher Scientific) as described previously²². Eluents were evaporated to dryness and reconstituted in 50:50 acetonitrile:water. The media and cell extract were submitted for mass spectrometry quantification of cGAMP, ATP, and GTP.

Conditioned media transfer

293T cGAS ENPP1^low cells were plated and transfected with plasmid DNA as described above. 24 hours following transfection, media was changed to RPMI + 2% human serum + 1% penicillin-streptomycin, ± 2 μM cGAMP, ± 20 nM recombinant mENPP1, or ± 50 uM STF-1084. 24 hours following media change, the conditioned media was removed from the 293T cGAS ENPP1^low cells and incubated with freshly isolated CD14⁺ PBMCs. Gene expression of CD14⁺ PBMCs was analyzed 14–16 h later.

PBMC electroporation of cGAMP and treatment with ENPP1 inhibitor

PBMCs (2 x 10⁶) were resuspended in electroporation buffer (90 mM Na₂HPO₄, 90 mM NaH₂PO₄, 5 mM KCl, 10 mM MgCl₂, 10 mM sodium succinate) with or without 200 nM cGAMP. Cells were electroporated in a cuvette with a 0.2 cm electrode gap (Bio-Rad) using program U-013 on a Nucleofector II device (Lonza) and immediately transferred to fresh media with or without ENPP1 inhibitor.

RT-PCR analysis

Total RNA was extracted using Trizol (Thermo Fisher Scientific) and reverse transcribed with Maxima H Minus Reverse Transcriptase (Thermo Fisher Scientific). Real-time RT-PCR was performed in duplicate with AccuPower 2X Greenstar qPCR Master Mix (Bioneer) on a 7900HT Fast Real-Time PCR System (Applied Biosystems). Data were normalized to CD14, ACTB, or GAPDH expression (human) and Actb expression (mouse) for each sample. Fold induction was calculated using ΔΔCt. Primers for human IFNB1: fwd (5’-AAACTCATGAGCAGTCTGCA-3’), rev (5’-AGGAGATCTTCAGTTTCGGAGG-3’); human CXCL10⁹: fwd (5’-TCTGAATCCAGAATCGAAGG-3’), rev (5’- CTCTGTGTGGTCCATCCTTG-3’) human CD14: fwd (5’-GCCTTCCGTGTCCCCACTGC-3’), rev (5’-TGAGGGGGCCCTCGACG-3’); human ACTB: fwd (5’-GGCATCCTCACCCTGAAGTA-3’), rev (5’-AGAGGCGTACAGGGATAGCA-3’); human GAPDH: fwd (5’-CCAAGGTCATCCATGACAAC-3’); rev (5’-CAGTGAGCTTCCCGTTCAG-3’); mouse Cxcl10⁹: fwd (5’- CTCTGTGTGGTCCATCCTTG-3’), rev (5’-GTGGCAATGATCTCAACACG-3’), mouse Actb⁹: fwd (5’-AGCCATGTACGTAGCCATCC-3’), rev (5’- CTCTCAGCTGTGGTGGTGAA-3’)

³²P-cGAMP degradation TLC assay

Radiolabeled ³²P cGAMP was synthesized as previously described¹⁶. Cell lysates were generated by scraping and lysing 1x10⁶ cells (293T) or 10x10⁶ cells (4T1-luc, E0771, and MDA-MB-231) in 100 μL of 10 mM Tris, 150 mM NaCl, 1.5 mM MgCl₂, 1% NP-40, pH 9.0. Samples were normalized to the amount of protein in each lysate reaction. The probe ³²P-cGAMP (5 μM) was incubated with mENPP1 (20 nM) or whole cell lysates in 100 mM Tris, 150 mM NaCl, 2 mM CaCl₂, 200 μM ZnCl₂, pH 7.5 or pH 9.0 for the indicated amount of time. To generate inhibition curves, 5-fold dilutions of ENPP1 inhibitor was included in the reaction. Degradation was evaluated by TLC as previously described¹⁶. Plates were exposed on a phosphor screen (Molecular Dynamics) and imaged on a Typhoon 9400 and the ³²P signal was quantified using ImageJ. Inhibition curves were fit to obtain IC₅₀ values with Graphpad Prism 7.03. IC₅₀ values were converted to Ki,app values using the Cheng-Prusoff equation$K_{\text{i},\text{app}}={\text{IC}}_{50}/\left(1+\left[\text{S}\right]/K_{\text{m}}\right)$.

ALPL and ENPP2 inhibition assays

Inhibition assays were performed by monitoring production of 4-nitrophenolate by absorbance at 400 nm. ALPL conditions: 0.1 nM ALPL, 2 μM 4-nitrophenyl phosphate, and inhibitor in buffer pH 9.0 containing 50 mM Tris, 20 μM ZnCl₂, 1 mM MgCl₂ at room temperature. ENPP2: 2 nM ENPP2, 500 μM bis(4-nitrophenyl) phosphate, and inhibitor in buffer pH 9.0 containing 100 mM Tris, 150 mM NaCl, 200 μM ZnCl₂, 2 mM CaCl₂.

Intracellular and extracellular cGAMP measurement in cancer cell lines

Cells were refreshed at time 0 with media supplemented with 50 μM STF-1084 (for IR experiments, cancer cell lines were also exposed to 8 Gy or 20 Gy of γ-radiation using a cesium source). At indicated times, media and cells were collected and centrifuged at 1,000 rcf. Cells were counted, lysed with 80% water + 20% methanol + 2% acetic acid and centrifuged at 15,000 rcf. cGAMP was enriched from the media and cell extract as described above.

Mouse models (4T1-luc, E0771, Panc02)

Five- to nine-week-old female mice were used for all experiments, and all tumor inoculations were performed using PBS as the vehicle. BALB/c mice (Jackson Laboratories) were inoculated with 5 x 10⁴ or 5 x 10⁵ 4T1-luc or 4T1-luc ENPP1^−/− cells suspended in 50 μL into the fifth mammary fat pad. C57B6/J WT, STING^gt/gt (referred to as Sting^−/−), or Enpp1^−/− mice (Jackson Laboratories) were inoculated with 5 x 10⁴ E0771 cells suspended in 50 μL into the fifth mammary fat pad. C57B6/J mice were inoculated with 3 x 10⁶ Panc02 cells suspended in 100 μL subcutaneously into the right hind flank. When tumor volume (determined by length² x width / 2) reached 100 ± 20 mm³, tumors were irradiated with 20 Gy (4T1-luc and Panc02) or 8 Gy (E0771) using a 225 kVp cabinet X-ray irradiator filtered with 0.5 mm Cu (IC-250, Kimtron Inc., CT). Anaesthetized animals were shielded with a 3.2 mm lead shield with a 15 x 20 mm aperture where the tumor was placed. For Panc02, the mice were implanted subcutaneously between the scapulae with an osmotic pump (Alzet 1002) containing a solution of 200 mg/mL STF-1623 in PBS or PBS alone one day prior to IR. Pumps were removed 8 days after implantation. Treatments after irradiation were administered as specified. Tumor volumes were recorded and analyzed in a generalized estimation equation in order to account for the within mouse correlation. Pair-wise comparisons of the treatment groups at each time point were done using post hoc tests with a Tukey adjustment for multiple comparisons. Animal death was plotted in a Kaplan Meier curve using Graphpad Prism 7.03 and statistical significance was assessed using the Log-rank Mantel-Cox test. Mice were maintained at Stanford University in compliance with the Stanford University Institutional Animal Care and Use Committee regulations and procedures were approved by the Stanford University administrate panel on laboratory animal care, or mice were maintained by Crown Biosciences in accordance with their regulations on animal laboratory care.

FACS analysis of tumors

BALB/c WT (4T1-luc tumors) or C57BL/6 (E0771 tumors) WT, Cgas^−/−, or Sting^gt/gt (referred to as Sting^−/−) mice were inoculated with 1 x 10⁶ tumor cells suspended in 50 μ into the fifth mammary fat pad. Two days after injection, tumors were intratumorally injected with 100 μL of 1 mM STF-1084 in PBS or with PBS alone. For experiments using STING and mENPP1, 100 μL of 100 μM neutralizing STING or non-binding STING (R237A) or 700 nM mENPP1 or PBS were injected intratumorally.

Alternatively, C57BL/6 mice were inoculated with 5 x 10⁴ E0771 tumor cells suspended in 50 μL into the fifth mammary fat pad. After reaching 100 ± 20 mm³, tumors were irradiated with 8 Gy. Two and four days after IR, tumors were intratumorally injected with 100 μL of 100 μM neutralizing STING or non-binding STING (R237A).

On the next day, the tumor was extracted and incubated in RPMI + 10% FBS with 20 μg/mL DNase I type IV (Sigma-Aldrich) and 1 mg/mL Collagenase from Clostridium histolyticum (Sigma-Aldrich) at 37 °C for 30 min. Tumors were passed through a 100 μm cell strainer (Sigma-Aldrich) and red blood cells were lysed using red blood cell lysis buffer (155 mM NH₄Cl, 12 mM NaHCO₃, 0.1 mM EDTA) for 5 min at room temperature. Cells were stained with Live/Dead fixable near-IR dead cell staining kit (Thermo Fisher Scientific, 1:1000 in accordance with the manufacturer’s description), Fc-blocked for 10 min using TruStain fcX (101320 BioLegend, clone 93, 1:100) and subsequently antibody-stained with CD8α-AF594 (100758 BioLegend, clone 53–6.7, 1:200), CD11c-PE (117308 BioLegend, clone N418, 1:200), CD45-AF700 (103128 BioLegend, clone 30-F11, 1:800) or BV650 (103151 BioLegend, clone 30-F11, 1:100), CD62L-BV785 (104440 BioLegend, clone MEL-14, 1:200), F4/80-APC (123116 BioLegend, clone BM8, 1:200), Granzyme B-AF647 (515406 BioLegend, clone GB11, 1:100), I-A/I-E-FITC (107606 Biolegend, clone M5/114.15.2, 1:800), CD3ε-PerCP-eF710 (46–0033-82 eBioscience, clone eBio500A2, 1:200), CD25-eF450 (48–0251-82 eBioscience, PC61.5, 1:200), and CD103-BUV395 (740238 BD Biosciences, clone M290, 1:400). Caspase activity was detected after red blood cell lysis using the FAM-FLICA Poly Caspase Assay Kit (ImmunoChemistry Technologies) according to the manufacturer’s description.

Cells were analyzed using an SH800S cell sorter (Sony), an LSR II (BD Biosciences), or an Aurora (Cytek). Data was analyzed using FlowJo V10 software (Treestar) and Prism 7.04 software (Graphpad) for statistical analysis and statistical significance was assessed using the unpaired two-tailed t test with Welch’s correction.

In vivo imaging

Mice were injected ip with 3mg XenoLight D-luciferin (Perkin-Elmer) in 200 μl water and imaged using a Lago X in vivo imaging system (Spectral Instruments Imaging). Object height was set to 1.5cm, binning to 4, FStop to 1.2, and the exposure time was 120s. Images were analyzed using aura 2.0.1 software (Spectral Instruments Imaging).

Synthesis of STF-1084

Preparation of dimethyl (E)-(2-(1-benzylpiperidin-4-yl)vinyl)phosphonate

Sodium hydride (2.16 g, 54.11 mmol) was carefully added to a stirred solution of bis(dimethoxyphosphoryl)methane (11.42 g, 49.19 mmol) in toluene (100 mL) at room temperature. The reaction mixture was then placed under an atmosphere of nitrogen and a solution of 1-benzylpiperidine-4-carbaldehyde (10 g, 49.19 mmol) in toluene (50 mL) was slowly added keeping the temperature below 40 °C. The resulting mixture was left to stir at room temperature for 16 h and then quenched by the addition of aqueous saturated ammonium chloride solution. The organic phase was separated, washed with brine, dried (MgSO₄) and evaporated to dryness. Chromatography (120 g SiO₂; 5 to 100% gradient of EtOAc in hexanes) provided dimethyl (E)-(2-(1-benzylpiperidin-4-yl)vinyl)phosphonate (6.2 g, 16%) as a colorless oil.

LC-MS: m/z = 309.8 [M+H]⁺

¹H NMR (500 MHz, Chloroform-d) – δ 7.47–7.21 (m, 5H), 6.86–6.73 (m, 1H), 5.65–5.53 (m, 1H), 3.75 (s, 3H), 3.70 (s, 3H), 3.52 (s, 2H), 2.98–2.87 (m, 2H), 2.22–2.11 (m, 1H), 2.09–1.99 (m, 2H), 1.79–1.70 (m, 2H) and 1.54–1.44 (m, 2H).

Preparation of dimethyl (2-(piperidin-4-yl)ethyl)phosphonate

To a mixture of dimethyl (E)-(2-(1-benzylpiperidin-4-yl)vinyl)phosphonate (6.2 g, 20.0 mmol) in ethanol (80 mL) was added Pd(OH)₂/C (0.5 g). The mixture was exchanged with hydrogen gas for three times and stirred under hydrogen balloon at room temperature for 12 h. The mixture was filtered through Celite® and evaporated to dryness to give 4.4 g (100% yield in ~90% purity) of dimethyl (2-(piperidin-4-yl)ethyl)phosphonate as colorless oil.

¹H NMR (500 MHz, Chloroform-d) δ 3.75 (s, 3H), 3.73 (s, 3H), 3.08 (dt, J = 12.5, 3.2 Hz, 2H), 2.58 (td, J = 12.2, 2.5 Hz, 2H), 1.80–1.60 (m, 4H), 1.60–1.50 (m, 2H), 1.38 (m, 1H), 1.11 (qd, J = 12.1, 4.0 Hz, 2H).

Preparation of dimethyl (2-(1-(6,7-dimethoxyquinazolin-4-yl)piperidin-4-yl)ethyl)phosphonate

Diisopropylethylamine (0.6 g, 8.9 mmol) was added to a mixture of dimethyl (2-(piperidin-4-yl)ethyl)phosphonate (1.1 g, 4.9 mmol) and 4-chloro-6,7-dimethoxy-quinazoline (1.0 g, 4.5 mmol) in isopropyl alcohol (20 mL). After stirring at 90 °C for 3 h, the reaction mixture was cooled and evaporated to dryness. Purification of silica gel (5% MeOH in dichloromethane) provided dimethyl (2-(1-(6,7-dimethoxyquinazolin-4-yl)piperidin-4-yl)ethyl)phosphonate (755 mg, 37%) as oil.

LC-MS: m/z = 410.25 [M+H]⁺

¹H NMR (500 MHz, CDCl₃) δ 8.65 (s, 1H), 7.23 (s, 1H), 7.09 (s, 1H), 4.19 (dq, J = 14.0, 2.9, 2.4 Hz, 2H), 4.02 (s, 3H), 3.99 (s, 3H), 3.77 (s, 3H), 3.75 (s, 3H), 3.05 (td, J = 12.8, 2.3 Hz, 2H), 1.93 – 1.77 (m, 4H), 1.67 (ddd, J = 14.1, 9.5, 5.9 Hz, 3H), 1.46 (qd, J = 12.2, 3.7 Hz, 2H).

Preparation of dimethyl (2-(1-(6,7-dimethoxyquinazolin-4-yl)piperidin-4-yl)ethyl)phosphonic acid hydrogen bromide salt (STF-1084).

Bromotrimethylsilane (3.67 g, 24 mmol) was added to a cooled solution of dimethyl (2-(1-(6,7-dimethoxyquinazolin-4-yl)piperidin-4-yl)ethyl)phosphonate (3.25 g, 7.94 mmol) in chloroform (60 mL) that was cooled by an ice bath. The reaction mixture was allowed to warm to room temperature and after 90 minutes was quenched by the addition of methanol (20 mL). The mixture was evaporated to dryness under reduced pressure and then solvated in methanol (100 mL). The reaction mixture was concentrated to half volume, filtered to remove precipitate, and then evaporated to dryness. The residue was crystalized with dichloromethane, filtered and dried under vacuum to give dimethyl (2-(1-(6,7-dimethoxyquinazolin-4-yl)piperidin-4-yl)ethyl)phosphonic acid (2.1 g, 69%).

LC-MS: m/z = 381.8 [M+H]⁺

¹H NMR (500 MHz, DMSO-d₆) δ 8.77 (s, 1H), 7.34 (s, 1H), 7.23 (s, 1H), 4.71 (d, J = 13.1 Hz, 2H), 3.99 (s, 3H), 3.97 (s, 3H), 3.48 (t, J = 12.7 Hz, 2H), 3.18 (s, 1H), 1.97–1.90 (m, 2H), 1.62–1.43 (m, 4H), 1.40–1.27 (m, 2H).

¹³C NMR (126 MHz, CD₃OD) δ 162.70, 158.44, 150.72, 147.43, 138.54, 107.54, 107.34, 100.15, 57.32, 56.98, 49.54, 37.44, 37.32, 33.20, 30.20 (d, ¹J_C-P = 3.78 Hz), 25.85, 24.74.

Synthesis of STF-1623

(2-(1-(8-methoxyquinazolin-4-yl)piperidin-4-yl)ethyl)phosphonic acid (STF-1623) was prepared according to the same synthetic procedure as STF-1084, using the 4-chloro-8-methoxyquinazoline instead of the 4-chloro-6,7-dimethoxyquinazoline in the preparation of dimethyl (2-(1-(8-dimethoxyquinazolin-4-yl)piperidin-4-yl)ethyl)phosphonate.

LC-MS: m/z = 352.1 [M+H]⁺

¹H NMR (400 MHz, DMSO-d₆): δ 8.62 (s, 1H), 7.63–7.53 (m, 3H), 6.65 (d, J = 12.4 Hz, 2H), 4.02 (s, 3H), 3.39–3.36 (m, 2H), 1.92 (m, 2H), 1.58–1.56 (m, 1H), 1.54–1.45 (m, 4H) and 1.35–1.30 (m, 2H).

¹³C NMR (126 MHz, D₂O) δ 162.77, 152.41, 150.66, 139.37, 125.78, 116.72, 115.40, 111.97, 55.61, 49.96, 36.26, 36.13, 31.35, 29.77 (d, ¹J_C-P = 3.78 Hz), 25.85, 24.79.

Statistics and Reproducibility

Treatment of established tumors in mice: power calculations were performed to estimate that cohorts of 9 mice were needed. Calculations were based on a pilot study using a 0.05, 2-sided significance level and power of 0.8, determined by Mann-Whitney U-test. The effect size of 1.9 and power calculations were performed using G*Power 3.1. Mice from different treatment groups were randomly co-housed in each cage to eliminate cage effects. The experimenter was blinded to group allocation and analysis. No data were excluded from the analyses.

For all other experiments, no statistical method was used to predetermine sample sizes. For FACS analysis of tumors, sample sizes were chosen to be 2–6 mice. For in vitro and cell culture experiments, sample sizes were chosen to be 2–3 biological replicates except in Fig. 2e–g, Fig. 5e, Extended Data Fig. 2b, Extended Data Fig. 3d, and Extended Data Fig. 10d, where titrations or time courses were performed and the sample size was chosen to be 1, with 2 technical replicates. The experiments were not randomized. The experimenter was not blinded to group allocation. No data were excluded from the analyses.

Either independent experiments (experiments repeated with identical assay conditions) or independent validations (experiments repeated with similar, but not identical conditions that validate results overall, but not precisely) were performed as indicated in the figure legends. Data from independent validations is shown in the Supplementary Figures as indicated. For data shown in Fig. 1b, f; Fig. 2d; Fig. 3e (in cells assay), f, j; Fig. 4a, b, c (Neuro-2a, MDA-MB-231, HeLa); Extended Data Fig. 3f, Extended Data Fig. 4a, e; and Extended Data Fig. 10b (in cells assay), d–f, the experiment was performed once. The result was supported by orthogonal experiments in other panels, as indicated in figure legends.

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data Availability

The ENPP1 mRNA expression data were derived from the TCGA Research Network: http://cancergenome.nih.gov/. Source Data (numerical and uncropped western blots) and Supplementary Data (independent validations) for Fig. 1–5, and Extended Data Fig. 1–7, 9–10 are provided. All other data that support the findings of this study are available from the corresponding author upon request.

Extended Data

Extended Data Figure 1 |. Measuring cGAMP in 293T cGAS cell lines by LC-MS/MS.

a, Chemical structures of cGAMP (top) and single isotopically-labeled cGAMP (bottom) used as an internal standard at a concentration of 0.5–1 μM.

b-c, Full (7–8 point) standard curves and LC traces of lowest cGAMP standards in 50/50 acetonitrile/water spiked in directly (LOQ = 4 nM) (b) or after concentrating and extracting 12.5x from complete cell culture media (LOQ = 0.3 nM from original sample, 4 nM in concentrated sample) (c). IS = internal standard. R² = coffecient of determination, determined after linear least squares regression with 1/Y² weighting. Data are from 1 experiment, representative of (b) 64 independent experiments and (c) 13 independent experiments.

d, Calibration of cell number to ATP concentration measured by LC-MS/MS. Data are from one experiment, representative of two independent experiments, 8 individual cell culture replicates are plotted.

e, cGAS expression of 293T, 293T cGAS ENPP1^−/−, and 293T cGAS ENPP1^low cell lines analyzed by western blot (left; full scan of blot available in Source Data). ENPP1 hydrolysis activity of ³²P-cGAMP in whole cell lysates from 1 million each of 293T cGAS, 293T cGAS ENPP1^−/−, and 293T cGAS ENPP1^low cells, measured by TLC and autoradiography (right). Data are from 1 experiment, representative of 5 independent experiments.

Extended Data Figure 2 |. Development of assays to investigate the mechanism of cGAMP export.

a, Schematic of experiment for b. CD14⁺ PMBCs were stimulated with increasing concentrations of extracellular cGAMP for 16 h.

b, IFNB1 mRNA levels were normalized to indicated gene and fold induction was calculated relative to untreated CD14⁺ cells. Each donor was performed as a one independent experiment; 2 qPCR replicates are plotted with a bar representing the mean.

c, Coomassie gel of recombinant mouse ENPP1 purified from media; elution fractions were pooled before use (left). ³²P-cGAMP degradation by mouse ENPP1 analyzed by TLC (right). Data are representative of 10 independent experiments.

d–e, 293T cGAS ENPP1^low cells were incubated with serum-free ATP depletion media (no glucose, 6 mM 2-deoxy-D-glucose, 5 mM NaN₃) or serum-free complete media for 1 hour. Levels of analytes were measured: (d) ATP by LC-MS/MS, total protein by BCA, cell death by extracellular lactate dehydrogenase activity, and (e) cGAMP by LC-MS/MS. BQL = below quantification limit. Data are from 1 experiment; 3 cell culture replicates are plotted (except for extracellular cGAMP, where 2 cell culture replicates are plotted). ATP data are representative of 3 independent validations (shown in Supplementary Fig. 5).

Extended Data Figure 3 |. Characterization of ENPP1 inhibitors QS1 and STF-1084.

a, Structure of QS1.

b, Inhibition by QS1 (compared to STF-1084, replotted from Fig. 3e). in vitro (³²P-cGAMP TLC assay, pH 7.5, purified mouse ENPP1): K_i,app = 6.4 μM, 2 independent experiments are plotted with a bar representing the mean.

c, Intracellular, extracellular, and total cGAMP for 293T cGAS ENPP1^−/− cells transfected with pcDNA6 (empty or containing human ENPP1) and treated ± QS1 after 24 hours. Data are from 1 experiment, representative of 2 independent validations; 2 cell culture replicates are plotted with a bar representing the mean.

d, Permeability of compounds in Caco-2 assay. PA = peak area, IS = internal standard. Compounds were incubated on the apical side of a Caco-2 monolayer for 2 hours. Compound concentration on the basolateral side was monitored by LC-MS/MS. Apparent permeability rates (P_app) were calculated from the slope. Data are representative of 2 independent experiments; single points are plotted.

e, Inhibitory activity of STF-1084 against alkaline phosphatase (ALPL) and ENPP2. Data are from 1 experiment, representative of 2 independent validations; 2 reaction replicates are plotted, except for ENPP2 100 uM and 32 nM, where 1 point is plotted.

f, Kinome interaction map (468 kinases tested) for STF-1084 depicting kinase inhibition as a percent of control.

Extended Data Figure 4 |. Intracellular and extracellular cGAMP from cancer cell lines.

a, cGAS expression of 4T1-luc WT and 4T1-luc shcGAS analyzed by western blot (full scan of blot available in Source Data). Intracellular cGAMP (chromatograms and quantification displayed) in 4T1-luc WT and shcGAS cells without exogenous stimulation. Concentration reported in units of molecules/cell and nM (estimated using cell volume = 4 pL). IS = internal standard. Data are from 1 experiment; 2 cell culture replicates are plotted.

b, Extracellular cGAMP produced by MC38 cells over 48 hours. At time 0, cells were refreshed with media supplemented with 50 μM STF-1084. Line depicts linear regression. Data are from one experiment representative of two independent validations; 2 cell culture replicates are plotted with a bar representing the mean.

c, Chromatograms for E0771, Panc02, and NMuMG cell lysate from LC-MS/MS. E0771 cell lysate was spiked with 40 nM cGAMP to determine limit of quantification. IS = internal standard. Data are representative of 2 independent validations.

d, cGAS and STING expression of 4T1-luc, MC38, E0771, Panc02, and Neuro-2a cell lines analyzed by western blot (full scan of blot available in Source Data). Data are representative of 3 independent validations.

e, Cell viability of cancer cell lines 4T1-luc and E0771 measured by lactate dehydrogenase extracellular activity compared to intracellular activity. At time 0, cells were left untreated or treated with IR (8 Gy or 20 Gy) and refreshed with media supplemented with 50 μM STF-1084. Data are from 1 experiment representative of two independent validations; 2 cell culture replicates are plotted with a bar representing the mean.

Extended Data Figure 5 |. Validation of Cgas^−/− cell lines and tools to neutralize extracellular Cgamp.

a, E0771 (left) and 4T1-luc (right) Cgas^−/− cells subcloned from CRISPR knockout pools (full scan of blot available in Source Data). Data are representative of two independent validations. Six E0771 Cgas^−/− subclones (1, 2, 4, 6, 8, and 9) were pooled before injection into mice. Two 4T1-luc Cgas^−/− subclones were pooled before injection into mice.

b, 293T cGAS ENPP1^low cells were transfected with empty pcDNA6 (0.5 μg/mL) and incubated for 24 hours. Conditioned media was treated with nonbinding or neutralizing STING (1 hour pretreatment) and then incubated with CD14⁺ PBMCs for 16 h. Extracellular cGAMP measured by LC-MS/MS and IFNB1 expression (mRNA levels were normalized to CD14 and fold induction calculated relative to untreated CD14⁺ cells). Data are from 1experiment (supported by data in Fig. 5e and Extended Data Fig. 5 c, d); 2 cell culture replicates are plotted.

c, Cxcl10 mRNA fold induction (normalized to Actb and untreated cells) in primary mouse bone marrow cells treated with 20 μM cGAMP in the presence of neutralizing or non-binding STING (100 μM) for 16 h. Data are from 1 experiment (supported by data in Fig. 5e and Extended Data Fig. 5 b, d); cell culture replicates plotted are (from left to right) 3, 2, 2, 3, 2, 2.

d, Mouse bone marrow-derived macrophages were incubated with 10 uM cGAMP and indicated concentrations of neutralizing STING protein for 2 hours. Levels of pIRF3 and IRF3 were analyzed by western blotting. Data are from 1 experiment (supported by data in Fig. 5e and Extended Data Fig. 5 b, c) (full scan of blot available in Source Data).

Extended Data Figure 6 |. FACS analysis and tumor growth following extracellular cGAMP depletion in untreated tumors.

a, FACS gating scheme for experiments in Fig. 5 f–i, Fig. 7b, c and Extended Data Fig. 6b.

b, WT or Cgas^−/− E0771 cells (1x10⁶) were orthotopically injected into WT, Cgas^−/− or Sting^−/− C57BL/6J mice on day 0. Neutralizing STING or non-binding STING was intratumorally injected on day 2. Sample sizes of n mice, from left to right (non-binding STING, neutralizing STING): n = (5, 5); (4, 5); (5, 5); (5, 4). Mean ± SD, unpaired two-tailed t test with Welch’s correction.

c, Established E0771 tumors (100 ± 20 mm³) were injected with non-binding (n = 8 mice) or neutralizing STING (n = 9 mice) every other day for the duration of the experiment. Tumor volumes for individual mice are shown. P value for tumor volume determined by pairwise comparisons using post hoc tests with a Tukey adjustment and for Kaplan Meier curve determined using the Log-rank Mantel-Cox test.

d, ENPP1 activity in 4T1-luc, E0771, and MDA-MB-231 cells using the ³²P-cGAMP degradation assay. Data from 1 experiment, representative of 3 independent validations.

Extended Data Figure 7 |. Cytokine production in cancer cell lines treated with ionizing radiation.

CXCL10 production by cancer cell lines 4T1-luc and E0771 measured by ELISA. At time 0, cells were left untreated or treated with IR (8 Gy or 20 Gy) and refreshed with media supplemented with 50 μM STF-1084. Data are from 1 experiment, representative of 2 independent validations; 2 cell culture replicates are plotted with a bar representing the mean.

Extended Data Figure 8 |. FACS analysis of tumors treated with ionizing radiation.

a, FACS gating scheme for live dead analysis in established tumors. E0771 cells (5x10⁴) were orthotopically injected into WT C57BL/6J mice. The tumors were treated with IR (8 Gy) when they reached 100 ± 20 mm³ and harvested and analyzed by FACS 24h after IR. For caspase activity, a single cell suspension was incubated for 1h with the FAM-FLICA Poly Caspase substrate before FACS stain and analysis. Mean ± SD, unpaired two-tailed t test with Welch’s correction.

b, FACS gating scheme for experiments in Fig. 6c, d.

Extended Data Figure 9 |. 4T1-luc Enpp1^−/− tumor growth.

a, Validating Enpp1^−/− 4T1-luc clones (11 clones were pooled) using the ³²P-cGAMP degradation assay (3 day incubation). Lysates were normalized by protein concentrations. Data are from 1 experiment, representative of three independent validations; 2 technical protein concentration replicates are plotted with a bar representing the mean.

b, Established 4T1-luc WT (harboring scrambled sgRNA) (n = 10 mice) or Enpp1^−/−tumors (n = 10 mice) (100 ± 20 mm³) were monitored without treatment. Tumor volumes for individual mice are shown.

c, Established 4T1-luc WT (harboring scrambled sgRNA) (n = 10 mice) or Enpp1^−/−tumors (n = 10 mice) (100 ± 20 mm³) were treated with IR (20 Gy) and monitored. Tumor volumes for individual mice are shown.

d, Established 4T1-luc WT (harboring scrambled sgRNA) (n = 9 mice) or Enpp1^−/−tumors (n = 9 mice) (100 ± 20 mm³) were treated with three intratumoral injections of 10 μg cGAMP on day 2, 4, and 7 and monitored. Tumor volumes for individual mice are shown.

Extended Data Figure 10 |. A systemic ENPP1 inhibitor delays Panc02 tumor growth as a single agent and synergizes with ionizing radiation.

a, Structure of ENPP1 inhibitor STF-1623.

b, Inhibition by STF-1623. In vitro (³²P-cGAMP TLC assay, pH 7.5, purified mouse ENPP1: K_i,app = 16 nM. 3 independent experiments are plotted. In cells (cGAMP export assay, human ENPP1 transfected into 293T cGAS ENPP1^−/− cells): IC₅₀ = 68 nM. Data are from 1 experiment; 2 cell culture replicates are plotted.

c, Mean apparent permeability (P_app) for STF-1623 and controls. For PAMPA and MDCK, mean was calculated from 2 cell culture replicates, 1 experiment. For Caco-2, mean was calculated from 2 independent experiments (data for atenolol and propranolol are reproduced from Fig. 3f for comparison).

d, Permeability of compounds in Caco-2 assay. PA = peak area, IS = internal standard. Compounds were incubated on the apical side of a Caco-2 monolayer for 2 hours. Compound concentration on the basolateral side was monitored by LC-MS/MS. Apparent permeability rates (P_app) were calculated from the slope. Data are representative of 2 independent experiments; single points are plotted.

e, PBMCs were incubated with STF-1623 for 16 h. Data are from one experiment; 2 cell culture replicates are plotted.

f, Kinome interaction map (468 kinases tested) for STF-1623 depicting kinase inhibition as a percent of control.

g, Intracellular and extracellular cGAMP concentrations for 293T cGAS ENPP1^−/− cells transfected with pcDNA6 (empty or containing human ENPP1) and treated ± 2 μM STF-1623 after 24 hours. Data are from 1 experiment; cell culture replicates plotted from left to right are (intracellular) 2, 3, 2, 5 and (extracellular) 2, 3, 2, 3.

h, PBMCs were electroporated ± 200 nM cGAMP and incubated ± 2 μM STF-1623 for 16 h. IFNB1 and CXCL10 mRNA levels were normalized to ACTB and fold induction calculated relative to untreated cells. Data are from 1 experiment; 2 cell culture replicates are plotted.

i, Mice were injected subcutaneously with 300 mg/kg STF-1623 at time 0. At indicated times, the mouse was sacrificed, blood was drawn by cardiac puncture, and serum isolated after clotting. STF-1623 concentrations were measure by LC-MS/MS. Data are from 1 experiment; 2 mice per time point are plotted except for 8 hours, where 3 mice are plotted.

j, Mice bearing established subcutaneous Panc02 tumors (100 ± 20 mm³) were implanted with a subcutaneous pump containing STF-1623 (50 mg/kg/day) on day 0 and left untreated or treated with IR (20 Gy) on day 1. No IR: n = 10 mice, no IR + STF-1623: n = 10 mice, IR (20 Gy): n = 10 mice, IR (20 Gy) + STF-1623: n = 15 mice. Pumps were removed on day 8. Tumor volumes for individual mice are shown. P value determined by pairwise comparisons using post hoc tests with a Tukey adjustment.

In b and g, cGAMP is measured by LC-MS/MS. BQL = below quantification limit. In b, e, g, h, and i, replicates are plotted individually with a bar representing the mean.