STING activation promotes robust immune response and NK cell–mediated tumor regression in glioblastoma models

Although STING is considered a promising target in cancer, expression of STING pathway components in GBM has not been studied. Therefore, we first characterized the expression of the key components of the cGAS-STING-IRF3 pathway in GBM, in both patient GBM samples and GBM cell lines. Western blotting showed low levels of cGAS expression in patient GBM samples and various levels of STING, TBK1, and IRF-3 (Fig. 1A). All the GBM cell lines tested, comprising patient-derived neurosphere cells G9 and G30 (48), as well as the two murine models used in the present work (GL261 and CT-2A), express cGAS, STING, TBK1, and IRF-3 at the protein level (Fig. 1A). Analysis of STING and phospho-TBK1 (Ser172) immunohistochemical (IHC) staining in a tissue microarray (TMA) showed a range of expression that is highest in GBM and lowest in normal brain, with intermediate levels in anaplastic astrocytoma and oligodendroglioma specimens (Fig. 1B). In the normal human brain, STING is expressed in a subset of cells and is particularly prominent in the vasculature, and this picture holds true for the majority of GBM cases in our TMA (Fig. 1C and D). Staining was also observed in individual cells scattered throughout the tumor parenchyma. pTBK1 was detected in the vasculature of GBM samples, indicating that there may be some STING pathway activation in GBM vasculature, in contrast to normal brain vasculature, where pTBK1 staining is absent (Fig. 1C and D) (49). In mice, STING is readily detectable in both CT-2A and GL261 tumors in vivo (SI Appendix, Fig. S1). The cGAS-STING pathway shows some degree of baseline activation in GL261 tumors, pTBK1 being colocalized with markers of the vasculature such as CD31 and α-SMA (Fig. 1E). Tumors are also infiltrated and surrounded by STING-positive cells (Fig. 1F and G) that mainly comprise F4/80⁺ and IBA1⁺ cells, as members of the innate myeloid immune population and microglia (Fig. 1H and I, respectively).

CXCL10 is an important cytokine produced downstream of type I IFN after STING activation and is commonly used to measure STING activity in human cancer models (50, 51). Using a CXCL10 enzyme-linked immunosorbent assay (ELISA), we found that the STING pathway is nonfunctional in all of the tested human GBM cell lines (SI Appendix, Fig. S2). Similarly, the murine glioma cell lines CT-2A and mut3 were not responsive to STING agonists, with GL261 cells being a notable outlier that responded strongly to STING agonist treatment, as demonstrated by CXCL10 release (Fig. 2A). Human brain vascular pericytes (HBVPs) and the human brain microvascular endothelial cell line hCMEC/D3 were responsive to STING agonist treatment, supporting a role of STING in the vasculature (Fig. 2A). We then investigated the feasibility of activating STING in GBM, using patient GBM specimens that were cultured as explants in suspension, and CXCL10 release measured after treatment with the STING agonist ADU-S100 (49). These patient samples were responsive to treatment with the STING agonist, producing various levels of inflammatory cytokines, indicating the activation of the STING pathway within the tumor tissue (Fig. 2B and C), and establishing that STING can be activated in human GBM and is therefore a potential immunotherapeutic target. After having established that the STING pathway was present and functional in GBM tissue, we tested the activity of STING agonists in vitro through coculture immune-mediated cell killing assays (Fig. 2D and E). GBM neurospheres made from GFP-expressing G9pCDH patient-derived GBM cells (48) were incubated with human peripheral blood mononuclear cells (PBMCs) freshly extracted from healthy donors at different ratios and treated with ADU-S100 (Fig. 2E). This showed that ADU-S100 is not toxic to GBM neurospheres, even at the highest concentration (Fig. 2E). In contrast, in the presence of PBMCs we observed an ADU-S100 concentration-dependent immune cell killing of the GBM neurospheres (Fig. 2F) with immune-mediated cytotoxicity being efficient between 12.5 and 50 µM, while the effect is reduced at 100 µM probably due to T cell toxicity at high STING agonist concentrations, as evidenced previously (52). The effects of the STING agonist increased with PBMC concentration, indicating an immune-mediated killing effect (Fig. 2F).

To understand the effects of STING agonists on the GBM TME, we characterized the immune response and immune infiltrates after ADU-S100 treatment of GL261 and CT-2A tumors in immunocompetent mice. Initial pilot studies using 2′,3′-cGAMP or ADU-S100 to activate STING indicated a strong innate immune response 3 d after treatment and an increase in myeloid populations, together with a survival benefit (SI Appendix, Fig. S3). We then analyzed in detail brain infiltrating leukocytes (BILs) by flow cytometry by using a panel of 13 immunological markers after an IT bolus of 50 µg of ADU-S100 (39, 53) at day 14 after GL261 intracranial implantation. Three days after treatment, we observed the suppression of microglia, and both CD4+ and CD8+ T cell populations, and infiltration of NK and CD11b⁺/Gr1⁺ inflammatory immune cells (Fig. 3B). At 7 d after treatment, T cell proportions recovered and were increased compared with baseline levels. PD-L1 expression was sharply increased shortly after treatment in CD45-negative cells (Fig. 3C, day 3). In this regard, granulocytic myeloid-derived suppressor cell (G-MDSC) populations also increase, although these may not be mature and immunosuppressive but rather inflammatory (Fig. 3D). These populations do not show increased expression of PD-L1 (SI Appendix, Fig. S4A).

To obtain a more global and precise picture of immune activation following IT ADU-S100 treatment in the brain, we analyzed the multidimensional data by t-distributed stochastic neighbor embedding (t-SNE) coupled to FlowSOM (54), allowing efficient clustering of populations and visualization of the global shift in immune populations for treated samples. Fig. 3E shows two-dimensional (2D) t-SNE density plots of the controls and ADU-S100–treated groups at day 3; the immunological landscape reveals extensive reorganization of the TME after treatment and activation of the STING pathway (Fig. 3E). The TME at day 7 also follows a deep remodeling (Fig. 3E). We proceeded to cluster these populations by using FlowSOM, and this revealed that the cell types making the bulk of the immune profile of the treated brain infiltrates were of two main types, which are mapped on the tSNE plot at day 3 in Fig. 3F: NK populations, defined as CD3⁻/CD11b⁺/CD11c^Int/Gr1⁻/CD49b⁺, and inflammatory myeloid cells comprising macrophages, dendritic cells (DCs), and neutrophils, collectively defined as CD3⁻/CD11b⁺/CD11c^Int/Gr1⁺/CD49b⁻. These could also comprise MDSC-type populations, although in this case they represent immature myeloid cells that would not yet convey suppressive phenotypes (55). These inflammatory populations are seen in other inflammatory states in the brain, such as in the response to traumatic brain injury (56). Fig. 3G shows the clustering of the FlowSOM populations at day 3 for GL261, and the main increased cell populations are highlighted. We can divide these into two major groups, one belonging to the NK type, with low Gr1 and mid- to high expression of CD49b, and the other group comprising Gr1⁺/CD49b⁻ inflammatory populations. Cytotoxic infiltrates of F4/80⁺ cells are further evidenced by the multiplexed immunofluorescence images 72 h after treatment of GL261 tumors (Fig. 3H).

We then performed a similar study in CT-2A tumors treated with a 50 µg ADU-S100 bolus (day 7 postimplantation), and the tumor-bearing hemispheres were analyzed for BILs (Fig. 4A). As in the case of GL261, we observed up-regulation of PD-L1 and changes in the MDSC populations; however, the latter did not show significant changes in PD-L1 expression (Fig. 4B and SI Appendix, Fig. S4B). 2D t-SNE once again revealed a complete remodeling of the TME and its immune profile, with density plots that are completely shifted (Fig. 4C), showing consistent and significant activation of the innate immune system at day 3, with an inflammatory TME comprised of NK populations and inflammatory immature innate cells of the myeloid lineage (Fig. 4D–F). A similar picture was obtained at day 7 with NK and innate immune cell infiltrates (SI Appendix, Fig. S5). Our results in the brain therefore support the critical participation of NK populations in the STING-induced antitumor effects (57, 58). Thus, IT delivery of STING agonists in murine GBM results in a strong innate immune response characterized by partially immature myeloid infiltrates and PD-L1 up-regulation.

We analyzed the transcriptome of GL261 tumors after ADU-S100 treatment and isolation of the BILs from the tumor-bearing hemisphere at 72 h. We can clearly distinguish the two conditions and the effect of the STING therapy, as seen from the principal component analysis of the individual samples (SI Appendix, Fig. S6A). This showed an acute specific activation of innate defense mechanisms (SI Appendix, Fig. S6 B and D). A volcano plot highlighted 20 of the most differentially expressed genes, showing high induction of the Ifit1 gene (IFN-induced protein with tetratricopeptide repeats 1B-like 1), in accordance with the activation of the type I IFN pathway. IFIT proteins are important for viral defense and are known to be widely expressed in the mouse brain. Activation of the ifit1 gene downstream of DNA sensing has been established (59). Other highly overexpressed genes included Ifi47, Tnf, and Mx2, all of which have been implicated in IFN signaling (SI Appendix, Fig. S6B) (60). It is worth noting that the tumor necrosis factor (TNF) alpha response is also seen in our human Luminex panel on the fresh GBM samples (Fig. 2C). Gene enrichment analysis also revealed terms related to the type I IFN signaling pathway being highly enriched in the ADU-S100 treated samples, in addition to terms related to NF-κB activation and translocation, cytokine production, and inflammatory response, while the lipoxygenase pathway was down-regulated (SI Appendix, Fig. S6D) (61), further supporting immune activation. We also compared gene expression signatures between ADU-S100–treated and PBS controls for genes known to be responsive to IFNγ (type II IFN), IFNβ (type 1 IFN), and IFNα (type 1 IFN) treatment (SI Appendix, Fig. S6C) (62). Treated samples had higher average gene expression signatures for all three IFN responsive gene sets that were investigated, with IFNα and IFNβ showing statistically significant differences between the two conditions (false discovery rate [FDR]–adjusted P values of 0.0062 and 0.0032, respectively). Transcriptomic analyses therefore showed that the GL261 tumors were highly responsive to ADU-S100 treatment, characterized by potent IFN signaling and innate immune activation. The involvement of NK cells in the response to the STING agonist was highlighted by increased NK cell gene signature shown in the violin plot (SI Appendix, Fig. S6C). NCR1, KLRD1, CD247, PRF1, and TNF are found among the highly up-regulated genes after ADU-S100 treatment. NCR1 (NKp46) is the most highly up-regulated and is one of the main activators of NK cells (63, 64). KLRD1, also known as CD94, is widely expressed in NK cells and some T cells and functions together with NKG2A/C as an immune checkpoint (65). CD247, expressed in NK and T cells, associates with NCR1, NCR3, and CD16. It is involved in the responsiveness of NK cells, such as degranulation (65). PRF1 (perforin 1) is a major component of cytolytic granules and is involved in the cytotoxic activity of NK and T cells (65). A volcano plot for this specific gene set (i.e., NK-mediated cytotoxicity) can be found in the SI Appendix, Fig. S6E.

With evidence of immune-dependent tumor cell killing, remodeling of the TME, and elevated innate immune response after IT injection of ADU-S100, we performed efficacy studies in murine GBM models. This was performed by using ADU-S100 loaded IT implants, made of cross-linked alginate chains and designed to release the STING agonist over a period of a few days (Fig. 5A and B) (66). This approach was chosen as it may be more clinically applicable than the bolus injection and can also be adapted and combined with other therapies such as immune checkpoint blockade. The gel matrix used here was composed to perform as a rapid release system for small molecules and a slow release delivery for larger molecules. Therefore, the small molecule STING agonist is released quickly, to mimic the bolus injection and trigger the immune reaction, while the aPD-1 antibody slowly comes into effect to oppose the chronic immunosuppressive effects of STING activation over time. To test the hydrogel approach, GL261 tumors were treated with 50 µg of ADU-S100 in biodegradable hydrogels 2 wk after implantation, and the tumors were monitored by bioluminescence and MRI starting 4 d after treatment (Fig. 5C–E). A striking effect of the therapy was seen shortly after treatment, with the IVIS signal increasing sharply in the control group (Fig. 5E), together with the tumor size as seen from MRI (Fig. 5D; for full MRI data, see SI Appendix, Fig. S7). The tumors completely regressed 2 wk after treatment in two-thirds of the treatment group, while the mock treated mice were already beyond their survival endpoints. The observed survival rate of STING monotherapy is thus comparable to that of the reported combined aPD-1 and aCTLA-4 therapy in the same model (18). Two-thirds of the treated mice were long-term survivors and were rechallenged by injection of GL261 cells into the contralateral hemisphere at day 150 and did not develop tumors, showing the establishment of long-term adaptive immunity after STING agonist treatment of GL261 tumor-bearing animals (Fig. 5C). Analysis of BILs 17 d after treatment with our ADU-S100 implants showed a sustained myeloid infiltration and a low PD-L1 expression on CD45-negative cells, in contrast with the acute up-regulation shortly after treatment (SI Appendix, Fig. S8 C and D). The treated mice were tumor free at that time point and went on to become long-term survivors.

We then turned to the aggressive and notably colder, nonimmunogenic CT-2A model as a more challenging syngeneic GBM mouse model (67, 68). The high immunogenicity of GL261 tumors may not correlate well with the clinical reality of GBM (69, 70). A similar efficacy study using cross-linked hydrogels to deliver ADU-S100 was conducted with the CT-2A model. This showed a significant increase in the median survival from 18 to 29 d (Fig. 5F). Representative MRI pictures are shown in Fig. 5G, and the full set of images can be found in SI Appendix, Fig. S9. No long-term survivors were observed in the STING monotherapy group, supporting previous observations that the CT-2A model is more resistant to immunotherapy than GL261 (67). We then performed an initial pilot study of the combination of ADU-S100 and aPD-1 loaded into hydrogels. This study led to the emergence of long-term survivors (Fig. 5H), supporting further detailed studies with this approach.

Finally, to evidence the critical involvement of NK populations in the tumor cell killing and the preclinical efficacy of ADU-S100 treatment in our models, we performed an NK depletion study on GL261-bearing mice. NK cell depletion was performed before treatment (Fig. 5I) and throughout follow-up, showing a complete loss of efficacy for the ADU-S100 treatment, while the isotype group treated with a 50 µg ADU-S100 bolus consistently showed a high percentage of tumor-free survivors (about two-thirds; Fig. 5J). This finding highlights the importance of alterations in tumor infiltrating immune cells after STING agonist treatment (Fig. 3E–G) and the critical reshaping of the TME with profound infiltration by myeloid and NK populations, the latter being responsible for tumor rejection.

GL261Luc2 murine glioma cells were purchased from PerkinElmer. CT-2A murine glioma cells were a gift from Thomas Seyfried (Boston College, Boston, MA). These two murine glioma cell lines were cultured in Dulbecco’s modified Eagle’s medium (Life Technologies), containing 10% heat-inactivated fetal bovine serum (FBS) and 1% penicillin/streptomycin. HBMEC/D3 and HBVP cells were purchased from ScienCell Research Laboratories and grown according to the manufacturer’s recommendations. Primary human glioblastoma stem cells (GSCs) (G9) were obtained by dissociation of gross tumor samples and cultivated in neurosphere media (briefly, neurobasal medium containing vitamin A–depleted B27, 1% GlutaMAX, 20 ng/mL epidermal growth factor, 20 ng/mL fibroblast growth factor, Primocin, and Plasmocin), as previously described (87). All cells were cultured at 37 °C with 5% CO₂, and mycoplasma testing was routinely done by PCR. The STING agonists ADU-S100 and 2′,3′-cGAMP were purchased from ChemieTek as the disodium salts (catalog nos. CT-ADUS100 and CT-CGAMP).

Brain tumor TMAs (GL2082) were purchased from US Biomax, Inc., and IHC staining for STING and phospho-TBK1 was performed on the Leica Bond III automated staining platform as previously published (53). The antibody for phospho-TBK1 (Cell Signaling Technology #5483, clone D52C2) was run at 1:50 dilution with the Leica Biosystems Refine Detection Kit with ethylenediaminetetraacetic acid (EDTA) antigen retrieval. The antibody for STING (Cell Signaling Technology #13647, clone D2P2F) was run at 1:50 dilution with the Leica Biosystems Refine Detection Kit with citrate antigen retrieval. IHC staining was quantified in QuPath software (0.2.0-m4). The positive pixel detection analysis was used with default settings for 3,3′-diaminobenzidine staining to detect and quantify positive pixels in each of three individual, randomly selected fields per tumor, which were then averaged.

CXCL10 ELISA (no. DIP100; R&D Systems) was used on media collected from cell culture according to the manufacturer’s instructions. Multiplex cytokine arrays were performed as previously described (53) via the bead-based immunoassay approach Bio-Plex Pro Human Cytokine 40-Plex Assay (catalog no. 171AK99MR2) on a Bio-Plex 200 system (catalog no. 171000201) (Bio-Rad Laboratories, Hercules, CA) and the Human Cytokine/Chemokine Magnetic Bead Panel (catalog no. HCYTMAG-60K-PX30) on a Luminex MAGPIX system (Merck Millipore, Billerica, MA). Conditioned media concentrations (pg/mL) of each protein were derived from five-parameter curve fitting models and plotted as log₂ fold changes. Lower and upper limits of quantitation were imputed from standard curves for cytokines above or below detection. Above assay readouts are marked with asterisks.

PBMCs were obtained from healthy human donors as approved by the Institutional Review Board (IRB) at Brigham and Women’s Hospital (all samples were de-identified) and isolated with Ficoll-Paque PLUS density gradient medium (GE Healthcare Life Sciences) according to the manufacturer’s instructions. A single-cell suspension of GBM cells was seeded at 2,000 cells/well (G9^pCDH) in ultra-low-attachment 96-well plates (Corning) and incubated for 24 h to allow sphere formation. PBMCs were then added with the different ADU-S100 concentrations, and treatment was repeated 3 d later. Cells were then incubated for another 3 d (total = 6 d of coculture). Microscope images of the spheres were taken daily (Nikon TI, 4× magnification), and the spheres’ fluorescence was measured in ImageJ with a dedicated macro.

The two polymeric components were prepared according to our previously published procedures (88). ADU-S100 was loaded by using an appropriate PBS solution of the drug to dissolve the norbornene–alginate component. Both components were mixed right before intracranial injection.

We purchased 7- to 8-wk-old female C57/BL6 mice from Envigo and injected a suspension of 50,000 cells (GL261Luc2, CT-2A) in 2 μL of PBS intracranially to establish mouse brain tumors (2 mm right lateral, 1 mm frontal to the bregma, and 3 mm deep). Successful tumor implantation was verified by bioluminescence imaging with the Perkin-Elmer IVIS Lumina 3 system. The study end point was considered a weight loss of 20%, onset of neurological symptoms, or signs of pain and distress. All animal experiments and procedures described in this study were approved by Brigham and Women’s Institutional Animal Care and Use Committee.

The tumor-bearing cerebral hemisphere was collected from each mouse at the indicated days after tumor implantation and therapy. Single mouse tumor cell suspensions were obtained with a mouse Tumor Disassociation Kit from Miltenyi Biotec (catalog no. 130-096-730). After leukocyte extraction with density gradient medium, cell suspensions were stained and analyzed by flow cytometry.

Flow cytometry on mouse tumors was performed as previously described (53). Briefly, after BIL isolation, cell suspensions were subjected to flow cytometry. After live/dead staining with the Zombie NIR Fixable Viability Kit (catalog no. 423106; Biolegend, San Diego, CA) according to the manufacturer’s instructions, single-cell suspensions were stained with fluorophore-conjugated primary antibodies (SI Appendix, Table S1), in PBS containing 2% FBS at 2 μg/mL. After washing, cells were resuspended in PBS containing 2% FBS and analyzed on an LSRFortessa flow cytometer (Becton Dickinson, Franklin Lakes, NJ). Levels were compared with isotype control antibodies. The data analyses were performed in FlowJo software (TreeStar). t-SNE was achieved with the embedded FlowJo t-SNE algorithm using default parameters. FlowSOM analysis was performed with the corresponding FlowJo plugin and R. Samples of the gating strategies can be found in the SI Appendix, Fig. S10.

The anti-NK1.1 depleting antibody was purchased from Bio X Cell (catalog no. BE0036), first injected 36 h before the intracranial treatment and biweekly after that. The mock groups used a mouse IgG2a from the same manufacturer (Bio X Cell, catalog no. BE0085). All antibodies were injected intraperitoneally (250 µg in 100 µL of InVivoPure pH 7.0 Dilution Buffer; Bio X Cell, catalog no. IP0070). Peripheral blood was harvested from the tail vein 32 h after the first injection of the depleting antibody and was analyzed by flow cytometry to determine the NK cell contents.

Mice were transcardially perfused with PBS (Thermo Fisher). Brains were removed and stored 24 h in 4% paraformaldehyde, then sucrose and slices were prepared with a vibratome (Campden Instruments) and immunostained for calretinin according to the following solutions and protocol: carrier solution, 1% normal horse serum (NHS; Vector Laboratories) with 0.5% Triton (Thermo Fisher) in PBS (Thermo Fisher); blocking solution, 10% NHS with 0.5% Triton in PBS. After several rounds of PBS washes, slices were blocked for 2 h at room temperature and incubated with primary antibody in carrier solution overnight at 4 °C. Slices were washed again in PBS and incubated for 2 h at room temperature in secondary antibody in carrier solution at a 1:1,000 dilution. After four final washes, slices were mounted on slides and coverslipped with 22 × 50 mm, 0.16- to 0.19-mm-thick cover glass (Fisher Scientific). Images were acquired with an LSM710 confocal microscope (Zeiss) and stitched with Zen 2.1 (black, Zeiss). Confocal images were post-processed in ImageJ (version 2.0).

Immunofluorescent multiplex staining was performed on the Leica Bond RX automated staining platform with a Leica Biosystems Refine Detection Kit. Antibodies were used as follows: pTBK1 (Cell Signaling Technologies; clone D52C2, catalog no. 5483) was run at 1:50 dilution with EDTA antigen retrieval; STING (Cell Signaling Technologies; clone D2P2F, catalog no. 13647S) was run at 1:50 dilution with citrate antigen retrieval; CD31 (Cell Signaling Technology; clone D8V9E, catalog no. 77699) was run at 1:100 dilution with citrate antigen retrieval; IBA1 (Wako; polyclonal, catalog no. 019-19741) was run at 1:500 dilution with citrate antigen retrieval. Imaging was performed on a Leica Versa 200 automated fluorescent/brightfield scanner at 20× magnification. Alexafluors 488, 555, 594, and 647 were used for each antibody (respectively).

Proteins were isolated from cell lines and content measured by bicinchoninic acid assay (Pierce Biotechnology). Protein extracts were subjected to polyacrylamide gel electrophoresis with either Criterion or Mini-Protean TGX precast gels, transferred to nitrocellulose (Millipore) membranes, and immunoblotted with antibodies (Cell Signaling Technologies, Danvers, MA) that specifically recognize STING (clone D2P2F, catalog no. 13647S), cGAS (clone D1D3G, catalog no. 15102), TBK1 (clone E8I3G, catalog no. 38066), IRF-3 (clone D83B9, catalog no. 4302), human or mouse anti-GAPDH (ab9484-200; Abcam, Cambridge, MA), and peroxidase-conjugated secondary antibodies (Jackson Laboratories, Bar Harbor, ME). 5% bovine serum albumin blocking buffer was used to dilute primary and secondary antibodies. Imaging of blots and quantitation of bands was performed with the Biorad Gel Doc XR Imaging System.

MRI data were acquired with either a BioSpec 3T or a Bruker 7 Tesla. Animals were kept under isoflurane narcosis throughout the scan. Respiration and heart rate were monitored. T2-weighted images were acquired via the RARE pulse sequence with the following settings: TE (echo time), 47.73 ms; TR (repetition time), 4,993.715 ms; RARE factor, 8; averages, 3; slice thickness, 0.5 mm; slicer orientation, axial; field of view, 20 mm × 20 mm; resolution, 0.078 mm × 0.078 mm.

The brain tumor samples were collected under the institutional banking IRB approved protocol 10-417. The samples were distributed under tissue sub-usage protocol approval. All patients undergoing brain tumor surgery at the Brigham are open to this banking protocol at the time of surgery. The IRB is approved by the DF/HCC IRB, and signed consent was obtained from all patients. Freshly isolated tumor tissue was harvested and processed within a few hours of surgery.

Total RNA was isolated directly from triplicate samples (BILs 3 d after ADU-S100) with the Qiagen (Hilden, Germany) RNeasy isolation kit (catalog no. 8028) via on-column DNase digestion. Briefly, libraries were prepared with SMARTer Stranded Total RNAseq version 2 Pico Input Mammalian sample preparation kits from 500 pg of purified total RNA according to the manufacturer’s protocol. The finished dsDNA libraries were quantified by Qubit fluorometer and Agilent TapeStation 4200. Uniquely dual indexed libraries were pooled in an equimolar ratio and shallowly sequenced on an Illumina MiSeq to further evaluate library quality and pool balance. The final pool was sequenced on an Illumina NovaSeq 6000 targeting 80 million 100-bp read pairs per library at the Dana-Farber Cancer Institute Molecular Biology Core Facilities. Data quality controls and replicate correlation were evaluated in the VIPER software package (89).

Sequencing reads were aligned using STAR (90) with an average of 3E7 uniquely mapped reads per sample and a mismatch rate per base of <1%. Aligned reads were first filtered to remove genes with <10 counts across all samples. Differential expression analysis was performed with DESeq2 in R (91), with an FDR-adjusted P value threshold of ≤0.1. A volcano plot showing log₂ fold change and –log₁₀ (adjusted P value) was then generated from the differential expression analysis in ggplot2. Gene enrichment analysis was performed with Enrichr (92) to investigate enriched pathways. The –log₁₀ (adjusted P values) of the most significant Gene Ontology (Biological Process) pathways for each condition were then plotted as bar graphs, alongside the corresponding values for the other condition for comparison. To further compare IFN gene signatures between ADU-treated samples and PBS controls, gene sets known to be responsive to IFNα, IFNβ, and IFNγ treatment were curated from published literature (62). We estimated the distributions of gene expression for the various IFN gene signatures per treatment group by first applying a regularized log transformation the filtered gene counts, followed by z-score normalization by gene. Violin plots were then used to visualize the distribution of the normalized gene expression values by condition. Statistical comparisons were performed via Wilcox test, with FDR for P value adjustment.

All graphs depict mean ± SEM unless otherwise indicated. Tests for differences between two groups were performed via unpaired two-tailed Student’s t test. Multiple comparisons used one-way or two-way ANOVA, as specified in the figure legends. Log-rank test was used for patient and mouse survival analyses. GraphPad Prism 9 was used for statistical analysis of experiments, data processing, and presentation. Sample sizes for in vivo studies were determined empirically based on results from prior publications.