STAT3 Antisense Oligonucleotide Remodels the Suppressive Tumor Microenvironment to Enhance Immune Activation in Combination with Anti–PD-L1

During the last decade, immune-checkpoint inhibitors have revolutionized the treatment of cancer. However, response rates are limited, often less than 20%, and the magnitude of benefit in terms of survival is likewise limited in most tumor types, likely due to additional immune-suppressive mechanisms within the tumor and its microenvironment and insufficient tumor antigen priming and presentation. Myeloid lineage cells, in particular, are implicated in driving immune suppression through the inhibition of T-cell and natural killer (NK) cell activity and the stimulation of suppressive regulatory T cells (Tregs), as well as driving resistance to checkpoint blockade (1). As an example, increased frequency of circulating myeloid cells is correlated with resistance to ipilimumab in patients with melanoma (2). Therefore, a therapeutic modality to relieve immunosuppression together with restoration of T-cell function may broaden responses and enhance the durability of checkpoint blockade.

STAT3 is a ubiquitously expressed transcription factor and master regulator of immune suppression (3–5). STAT3 also has a multifaceted role in promoting tumor cell survival and proliferation and is an attractive cancer drug target (3, 6). With respect to the role of STAT3 in driving immune suppression, STAT3 signaling has been linked to the promotion of a suppressive myeloid-derived suppressor cell (MDSC) phenotype (7) as well as protumor macrophage polarization (8). Consistent with these observations, CD14⁺ MDSC isolated from pancreatic cancer (9) and squamous cell carcinoma of the head and neck (HNSCC) patients (10) are enriched in a JAK/STAT3 signature and express higher levels of arginase 1, suggesting that activated STAT3 signaling is driving suppressive MDSC function clinically. Preclinical studies have demonstrated that hematopoietic cell-specific loss of STAT3 enhances antitumor immunity (11), and deletion of STAT3 from the myeloid compartment can improve the activity of adoptively transferred T cells (12). However, clinical validation of these preclinical observations has not been rigorously assessed due to the lack of selective STAT3 inhibitors.

Therapeutic inhibition of STAT3 has been attempted using numerous strategies, including inhibition of upstream JAK kinases, STAT3–SH2 domains, and STAT3–DNA binding, all of which have demonstrated clinical development challenges and are summarized elsewhere (13, 14). One way to selectively target STAT3 is with antisense oligonucleotides (ASO). Recent advances have led to a combination of oligonucleotide chemistry and base modifications that have enhanced the safety, selectivity, and potency of oligonucleotides such that target knockdown can be achieved in preclinical species and in humans (15). Previously, we have reported on the safety and activity of a selective STAT3 ASO, danvatirsen (AZD9150), in patients with lymphoma (16). Here we present analysis of pre- and posttreatment biopsies from danvatirsen-treated patients showing the selective uptake of danvatirsen into cells of the tumor microenvironment (TME; but not cancer cells) and increased proinflammatory gene expression. In vitro studies with cultured human macrophages and dendritic cells showed that danvatirsen treatment enhanced proinflammatory cytokine secretion and resistance to IL10 mediated suppression. Data from preclinical models revealed a unique pattern of STAT3 knockdown and ASO uptake in a subset of stromal/immune cells of the TME, specifically myeloid lineage cells, Tregs, endothelial cells, and cancer-associated fibroblasts. Knockdown of STAT3 in myeloid cells resulted in remodeling of infiltrating monocytes and reduction of myeloid-induced immune suppression in the TME, which enhanced T-cell functionality and antitumor activity when STAT3 ASO was administered in combination with anti–PD-L1. These alterations in infiltrating immune populations as a direct consequence of STAT3 knockdown demonstrate a novel role for STAT3 ASO in enhancing antitumor immunity.

Patient samples were collected in a phase I clinical study (NCT01563302) from patients treated with danvatirsen. The studies were conducted in accordance with the Declaration of Helsinki and were approved by institutional review boards. Written informed consent for sample acquisition and analysis for research purposes was obtained from the participants.

Five-micrometer sections of formalin-fixed paraffin-embedded (FFPE) tumor biopsies were stained with anti-STAT3 (CST; No. 4904) as described for mouse samples and evaluated by two pathologists and/or by Aperio image analysis color deconvolution software to quantify staining. Anti-ASO staining was done on a Labvision Autostainer with Proteinase K epitope retrieval, peroxidase, and serum-free protein blocking (Dako; X0909), followed by rabbit anti-ASO diluted 1:6,000 and Dako Envision+/HRP. Detection was done with Liquid DAB+ and counterstained in Carazzis hematoxylin.

RNA was isolated from FFPE tissue using the Qiagen RNAeasy FFPE kit (catalog No. 73504), six 5-μm sections per biopsy, as indicated in Supplementary Table S1. RNA was isolated from PBMCs using the Qiagen QIAamp RNA blood kit (catalog No. 52304). RNA was quantified by Qubit and analyzed using the NanoString nCounter Human Immunology v2 codeset. Differential expression analysis was performed from calibrated counts from patient tumor and whole blood as described (17), quantile normalized, Wilcoxon rank-sum or Wilcoxon signed-rank 2-tailed statistical tests between pre- and post-AZD9150 with/without baseline subtraction, and heat maps generated with unsupervised Euclidean clustering for the genes with raw P < 0.05, all within the R coding environment (18, 19). (Statistical results were generated using this online tool: https://www.socscistatistics.com/tests/.)

T-cell receptor (TCR) abundance was determined at Adaptive Biotechnologies using DNA isolated from frozen tumor biopsies that were collected in parallel to the FFPE biopsies.

Statistically significant gene signature changes: A collection of ∼90 gene signatures derived from the literature and internal efforts were analyzed using gene set variation analysis of the NanoString gene-expression data derived from diffuse large B-cell lymphoma (DLBCL) patient tumor biopsy samples and applying a significant P value of <0.05.

Human monocytes were cultured for 6 days in RPMI medium with 10% fetal bovine serum (FBS) and supplements containing 100 ng/mL human M-CSF (PeproTech). After 6 days of culture, the medium was replaced with fresh medium containing 100 ng/mL human M-CSF and 50 ng/mL human IL10 (PeproTech). A day later, cells were replated in 96-well flat-bottom plates at 100,000 cells/well in medium with M-CSF and IL10 and either left untreated or treated with 5 μmol/L control ASO or 5 μmol/L human Stat3 ASO (Ionis Pharmaceuticals). Cells were treated with 100 ng/mL lipopolysaccharide (LPS; Sigma) for 18 hours, after which culture supernatants were analyzed for cytokines by using MSD multiplex plates. Cells were surface stained for CD80 (BioLegend; No. 305220) and CD86 (BD; No. 562432) and analyzed on a flow cytometer (BD Fortessa).

Human monocytes were cultured for 7 days in RPMI medium with 10% FBS and supplements containing 100 ng/mL human GM-CSF and 100 ng/mL IL4 (PeproTech). After 7 days, cells were harvested and plated in 96-well flat-bottom plates at 100,000 cells/well. Cells were plated in medium with GM-CSF and IL4 and either left untreated or treated with 5 μmol/L control ASO or 5 μmol/L danvatirsen (human STAT3 ASO; Ionis Pharmaceuticals). Where indicated, human IL10 was added at a final concentration of 1 ng/mL. Cells were treated with 100 ng/mL LPS (Sigma) for 18 hours, and supernatants were analyzed for cytokines using the Mesoscale Discovery proinflammatory panel 1 (MSD; cat. No. K15049D). Cells were surface stained for CD86 (BD; No. 562432) and analyzed on a flow cytometer (BD Fortessa).

Primary human T cells were plated in 96-well flat-bottom plates (200,000 cells/well) in RPMI medium with 10% FBS and cultured for 3 days ± 5 μmol/L ASO (control or STAT3). After 3 days, 5 μL of Immunocult T-cell activator (STEMCELL Technologies) was added to the cells in each well. Twenty-four hours later, cell culture supernatants were analyzed for IFNγ, TNFα, and IL2 using Meso Scale Discovery multiplex plates. Cells were cultured for 4 days with addition of EdU during the last 18 hours of culture. Proliferation was analyzed by staining for incorporated EdU (Click-iT Plus EdU Alexa Fluor 647 Flow Cytometry Assay Kit) followed by flow analysis.

Human monocytes were cultured for 6 days in RPMI medium with 10% FBS and supplements containing 100 ng/mL human GM-CSF (PeproTech) and 50 ng/mL human IL6 at 37°C, and then harvested and treated with either 10 μmol/L control ASO or 10 μmol/L human STAT3 ASO. Three days after ASO treatment, culture supernatants were analyzed for cytokines using MSD multiplex plates. T-cell proliferation was evaluated by CFSE (Invitrogen), and proliferation was analyzed post anti-CD3/CD28 stimulation. Pellets were harvested for Western blot analysis.

Treated CT26 tumors were homogenized, and lysates were analyzed using the mouse 44-plex cytokine array from Eve Technologies. Cytokine signals from CT26 tumors were log₂ transformed. Sample-level quality control (QC) is assessed by the distribution of the overall cytokine signals and principal components analysis. The samples that failed QC were excluded in the downstream analysis. Differential expression analyses for cytokines were carried using ANOVA with a P value and FDR threshold of 0.05. Hierarchical clustering was used to identify sets of cytokines whose expression levels correlated among individuals within a population. A heat map was produced with R package heat map (version1.0.12) using default parameters.

Total RNA was extracted from two to three 5-μm-thick FFPE sections of CT26.WT tumors using the miRNeasy FFPE Kit (Qiagen). RNA integrity and quantity were assessed on the TapeStation 2200 using the RNA ScreenTape System (Agilent). Manufacturer's recommended protocols were followed. RNA was analyzed using the NanoString nCounter FLEX Analysis System and the 770-gene mouse PanCancer Immune Profiling Panel (NanoString) following the manufacturer's standard XT CodeSet Gene-Expression Assays protocol. Post-hybridization sample processing on the Prep Station using the high-sensitivity setting was followed by data collection on the Digital Analyzer using a scan setting of 555 fields of view. Preprocessing of the raw count data, which included background subtraction of the negative control probes, positive control normalization, and housekeeping gene normalization, was performed in the nSolver 4.0 (NanoString) software using the geometric means and default parameters. All samples included in the analysis fell within the default nSolver QC parameters. The background-subtracted, normalized count data were uploaded to ROSALIND v3.12.0.5 (OnRamp) for differential gene-expression and pathway analyses.

Cell proliferation was determined by cell number increase using the CellTiter-Glo luminescent cell viability assay kit (Promega). In brief, CT26 cells were grown in an RPMI plus 10% FBS. Cells were plated in 96-well plates at 100,000 cells/well and treated at the indicated compound concentrations. Cells were lysed with CellTiter-Glo luminescent cell viability assay reagent, and luminescence was read using the Envision plate reader. On day 5, supernatant was collected for cytokine analysis using Mesoscale Discovery according to the manufacturer's instructions (MSD; cat. No. K15255D-1).

Bone marrow cells were isolated from mouse femurs and tibiae and cultured for 6 days in culture medium containing 100 ng/mL mouse M-CSF (PeproTech; 315-02). On day 6, IL10 (PeproTech; 210-10) was added at a final concentration of 50 μg/mL. On day 7, cells were harvested and plated for treatment with mouse STAT3 antisense (No. 481549; Ionis Pharmaceuticals) or control antisense (No. 792169; Ionis Pharmaceuticals) for 3 days. Cells were lysed using PhosphoSafe Extraction Reagent (Millipore Sigma; cat. No. 71296). Protein extracts were separated by SDS-PAGE, transferred onto PVDF membranes, and probed with primary antibodies against p-STAT3 (CST; No, 9145), T-STAT3 (CST; No. 9139), or GAPDH (CST; No. 2118) followed by secondary HRP-conjugated antibody, and visualized with the Pierce ECL Western blotting substrate (Thermo Fisher Scientific; cat. No. 34076).

All samples for IHC were FFPE into blocks. FFPE blocks were sectioned at 4 μm. IHC was run on a Ventana Discovery Ultra. Primary antibodies used are listed in Supplementary Methods. The Chromomap DAB detection kit (Ventana; 760-159) and OmniMap anti-Rabbit HRP (Ventana; 760-4311) were used for rabbit primary antibodies. DABMap detection kit (Ventana; No. 760-124) and biotinylated anti-rat secondary antibody–mouse adsorbed (Vector Labs; No. BA9401)—were used for rat primary antibodies. All chromogenic staining was done using DAB as a substrate and counterstained using Hematoxylin II (Ventana; No. 790-2208) and Bluing Reagent (Ventana; No. 760-2037). For fluorescent multiplexing, Ki67, granzyme B, and CD8 primary antibodies were sequentially stained, with a CC2 denaturation step in between. The following substrates were used for a fluorescence Red 610 kit (Ventana; No. 760-245), DCC kit (Ventana; No. 760-240), and FAM (Ventana; No. 760-243), and the counterstain used was QD DAPI (Ventana; No. 760-4196). Chromogenic slides were scanned using an Aperio AT2 (Leica), and fluorescent slides were scanned with an Aperio Versa (Leica). Image analysis was performed on the HALO platform (Indica Labs).

CT26 cells [clone CT26.WT (ATCC) and clone CT26.AZ (obtained from Ian Hart)], A20 cells (ATCC TIB-208), 4T1 cells (ATCC), and MC-38 cells (NCI) were grown, under standard conditions, from frozen stocks maintained in the AstraZeneca cell bank repository. All cell lines were authenticated by STR fingerprinting and mycoplasma tested between 2016 and 2017 prior to use (RADIL). CT26 (5 × 10⁵), A20 (2 × 10⁵), and 4T1 (5 × 10⁴) were maintained between passages 4 and 10 and implanted into Balb/c mice (Envigo), subcutaneously. MC-38 (5 × 10⁵) were implanted into C57/Bl6 (Charles River) subcutaneously. Models were performed at AstraZeneca according to AstraZeneca Institutional Animal Care and Use Committee guidelines. Mice were maintained in a controlled, specific pathogen-free environment at 20°C to 25°C, 40%–70% humidity and 12 hours light-to-dark cycle. Tumor length and width were measured by caliper, and tumor volume was calculated using the formula volume = (length × width²)*π/6. mSTAT3 ASO (No. 481549) and control ASO (No. 792169) were obtained from Ionis Pharmaceuticals. Mouse anti–PD-L1 was generated at AstraZeneca. Anti-CD8a used for depletion (Bio X Cell No. BP0061) was administered on days −1, 0, 1, 5, 10, and 15 with respect to tumor implant. AZD4205 was administered by oral gavage at 50 mg/kg daily.

Harvested tumors were dissociated using the gentleMACS and Miltenyi mouse tumor dissociation kit. Cells were filtered with 70-μm cell strainer, and 50 μL of 2 × 10⁶ cells was blocked with Fc block (eBioscience). Cells were incubated for 15 minutes with live/dead fixable zombie UV stain dye. Samples were incubated with master mix of flow cytometry antibodies and washed twice with flow wash buffer. T cells' ex vivo stimulation was done in the presence of cell stimulation cocktail (eBioscience). Four hours after stimulation, cells were incubated with T-cell functional marker antibodies. Samples were acquired by BD LSRFortessa flow cytometer, and analysis of data was done using FlowJo 10.6. Antibodies used for flow cytometry studies are list in Supplementary Table S4. We have performed one-way ANOVA and have highlighted significant value changes in the graphs.

Tumor tissues were chopped and transferred into a tube containing RPMI (Thermo Fisher, Gibco 61870-044). Single-cell suspension was prepared by treating tumors with a mix of enzymes using mouse tumor dissociation kit (Miltenyi Biotec; No. 130-096-730) for 40 minutes at 37°C with agitation. Cells were counted (Vi-Cell XR; Beckman Coulter), and 3 × 10⁶ cells from each sample were arranged into groups of 20 tumors. Samples were stained with 5 μmol/L Cell-ID Cisplatin (Fluidigm, No. 201064) for 1 minute at room temperature in MaxPar PBS (Fluidigm; No. 201058) and were then barcoded using the Cell-ID 20-Plex Pd Barcoding Kit (Fluidigm; No. 201060) according to the manufacturer's instructions. Cells were washed, and each group of 20 samples was pooled into a single tube and fixed using the Fix/Perm (Thermo Fisher; 00-5123-43 and 00-5223-56) for 15 minutes at 4°C. Cells were blocked in anti-CD16/CD32 antibody (Thermo Fisher; 14-0161-86) and stained overnight with metal- or fluorophore-conjugated antibodies (Supplementary Table S5) in Perm Buffer (Thermo Fisher; 00-8333-56). Samples were washed twice and stained using metal-conjugated secondary antibodies incubated for 30 minutes at 4°C, washed twice, and stored in MaxPar Fix and Perm Buffer (Fluidigm; No. 201067) with Cell-ID Intercalator-Ir (Fluidigm; 201192A) at 125 nmol/L until acquisition.

Compensation of spectral overlap between antibodies was performed by first staining anti-mouse or anti-rat/hamster beads (BD Biosciences; No. 552843 and No. 552845) individually with each of the metal-conjugated antibodies. For antibodies raised in rabbit or goat, anti-mouse beads were first coated with mouse anti-rabbit IgG or mouse anti-goat IgG (Thermo Fisher; No. 31213 or No. 31107), washed, and then incubated with rabbit or goat metal–conjugated antibody. Stained beads were then pooled for acquisition.

Immediately before acquisition, beads or tumor samples were washed twice using Cell Staining Buffer (Fluidigm; No. 201068) and then twice with MaxPar Water (Fluidigm; No. 201069). Pellets were resuspended in MaxPar Water and filtered twice through 70-μm strainer (Greiner No. 542070) and acquired on Helios CyTOF System (Fluidigm). Sample de-barcoding was performed using CyTOF Software (Fluidigm) or manual gating (Supplementary Fig. S1), and data were analyzed using FlowJo software (V.10, Treestar) or Cytobank. Data compensation was performed using CATALYST package (20). Gating strategies are shown in Supplementary Fig. S2.

Antibodies that were not available as metal conjugates were labeled using Maxpar X8 Antibody Labeling Kits (Fluidigm; 201141A–201156A and 201158A–201176A; Sigma, No. 203440) according to the manufacturer's instructions or as published elsewhere (21).

Danvatirsen monotherapy was tested in two phase I clinical studies (NCT01563302 and NCT01839604) in which several durable clinical responses were observed (complete and partial responses up to 2 years), suggestive of immune-mediated antitumor responses (16). Danvatirsen was dosed systemically via intravenous injection, with three doses in the first week followed by weekly dosing thereafter. Initial evidence suggesting an immune versus tumor cell–mediated mechanism of action for danvatirsen was obtained by examining STAT3 knockdown in 12 sets of paired tumor biopsies (Supplementary Table S1) including nine DLBCL, two follicular lymphoma (FL), and one non–small cell lung cancer (NSCLC) taken from patients enrolled in clinical trial NCT01563302 at baseline and after 4 weeks of treatment when steady-state drug levels were achieved (15, 22). Seven biopsy pairs were stained with an anti-ASO antibody to identify which cell types take up danvatirsen. We observed anti-ASO staining in the stromal, but not tumor cells, with clear staining of endothelial cells as well as spindle cells and inflammatory cells (representative staining; Fig. 1A). Normal cells surrounding tumor cells in an NSCLC tumor that metastasized to liver were positive for anti-ASO (Fig. 1B), consistent with the observation that ASOs accumulate in the liver (23, 24), but interestingly, only weak or patchy anti-ASO staining was observed in tumor cells (Fig. 1A and B). Staining for STAT3 quantified by image analysis revealed little or no target knockdown in tumor cells (Fig. 1C and D) in all except one FL tumor, which exhibited a 30% decrease (Supplementary Table S1). In contrast, STAT3 protein knockdown was observed in endothelial cells of on-treatment biopsies from three patients (Fig. 1D; Supplementary Table S1), and in normal cells adjacent to the NSCLC liver metastasis (Fig. 1C). The patterns of danvatirsen uptake and STAT3 knockdown indicate that there was distribution of the ASO to tumor tissue in patients, but preferential uptake within the tumor by non-tumor (stromal) cells at clinically tolerated doses. Anti-ASO staining in on-treatment bone marrow biopsies demonstrated ASO uptake in myeloid and erythroid lineage cells (Supplementary Fig. S1A), and RNA isolated from baseline and on-treatment peripheral blood mononuclear cells from 11 DLBCL and 4 FL patients showed a statistically significant (P = 0.0024) median log₂ decrease of 0.49 (29%) in STAT3 expression 4 weeks after treatment (Fig. 1E). In order to evaluate the selectivity of STAT3 knockdown within the STAT family, STAT1 RNA expression was assessed in the same blood samples from the 11 DLBCL and four FL patients. STAT1 RNA expression was not reduced in these blood samples but was increased (median of 60%), which is likely secondary to the STAT3 knockdown (Supplementary Fig. S1B) and is consistent with the selectivity of danvatirsen to reduce STAT3, but not STAT1 (15). The danvatirsen uptake into stromal cells of the tumor and cells of the bone marrow coupled with STAT3 knockdown in immune cells in peripheral blood suggested that danvatirsen would elicit immune changes in the patients' tumors. To assess these changes, we analyzed the expression of 579 immune genes by NanoString technology using RNA isolated from the same paired lymphoma tumor biopsies used for IHC analysis. Unsupervised clustering demonstrated the reproducibility of biological replicates (Supplementary Fig. S2). Analysis of bulk tumor RNA did not show significant reduction in STAT3 RNA (Supplementary Fig. S1C). In contrast, consistent changes in genes associated with immune cells and function were observed in on-treatment samples. Notably, STAT1 increased on-treatment (median 49% increase; P < 0.05, Wilcoxon signed-rank test; Supplementary Fig. S1C). A heat map of the greatest magnitude and most highly significant changes in the on-treatment DLBCL samples showed that genes associated with increased IFNγ signaling, including five genes (STAT1, IFNγ, CXCL9, CXCL10, and IDO1) of a six-gene signature associated with response to PD-L1 axis blockade (25), were upregulated after danvatirsen treatment (Fig. 1F; Supplementary Table S2). Type I IFN and IFNγ signatures were among the most statistically significantly (P < 0.05) modulated in evaluation of a collection of immune signatures (Fig. 1G; Supplementary Table S3). The number of T cells, as indicated by TCR sequences present in paired pre- and on-treatment tumor biopsies, shows a trend of increasing with an average increase of 2.2-fold (Fig. 1H), although not reaching statistical significance. There was no consistent change in CD8A gene expression (data not shown).

We sought to model the activity seen in patients to further understand the biology of immune modulation mediated by danvatirsen, using preclinical models. To that end, we investigated the antitumor effects of a mouse surrogate STAT3 ASO (herein mSTAT3 ASO) in several syngeneic mouse tumor models including a lymphoma model and several epithelial tumor models. Mice were treated with mSTAT3 ASO via subcutaneous injection at 50 mg/kg on a schedule of 5 days on 2 days off. In A20, MC-38, and CT26 tumor–bearing mice, treatment with mSTAT3 ASO resulted in partial tumor growth inhibition as monotherapy (Supplementary Fig. S3A). The antitumor effects of mSTAT3 ASO were unlikely to be driven through tumor-intrinsic inhibition of STAT3 signaling because in vitro treatment of CT26 cells with mSTAT3 ASO did not reduce the proliferation rate, and antitumor effects were lost when CT26 cells were implanted in immune-deficient NOD/SCID gamma (NSG) hosts (Supplementary Fig. S3B and S3C). Given these results, we focused the rest of our analysis on effects of mSTAT3 ASO within the immune compartment and tested whether combination with checkpoint inhibitors such as anti–PD-L1 could enhance the antitumor effects. Indeed, the combination of mSTAT3 ASO with anti–PD-L1 resulted in more significant tumor growth inhibition than either monotherapy alone, in three models sensitive to anti–PD-L1 monotherapy, MC38, CT26, and A20, but not in the checkpoint blockade refractory model 4T1 (Fig. 2A; Supplementary Fig. S4). We next evaluated whether inhibition of JAK1, an upstream activator of STAT3 as well as STAT1, would be as effective as STAT3 inhibition in CT26 and found that selective JAK1 inhibition using the small-molecule inhibitor AZD4205 (26) had no monotherapy activity, and actually antagonized the activity of anti–PD-L1 (Fig. 2B). Furthermore, CD8 T cells were necessary for the antitumor effects of mSTAT3 ASO because both monotherapy mSTAT3 ASO and combination antitumor effects were mitigated when CD8 T cells were depleted (Fig. 2C). Collectively, these findings suggested that mSTAT3 ASO has immunotherapeutic activity.

Based on the potential of mSTAT3 ASO treatment to improve clinical responses of epithelial tumors to PD-(L)1 blockade, we chose to exploit the immunogenic CT26 model to elucidate the mechanism by which mSTAT3 ASO modifies the TME to enhance antitumor immune responses. Effects were evaluated at two dose levels, 50 mg/kg (5 days on, 2 days off) and 25 mg/kg (5 days on, 2 days off), alone and in combination with anti–PD-L1. As observed in Fig. 2, 50 mg/kg (5 on 2 off) had modest monotherapy activity, but significantly enhanced activity in combination with anti–PD-L1. Similar results were observed at the lower dose of 25 mg/kg (5 on 2 off; Supplementary Fig. S5A), providing confidence to proceed with mechanistic studies using the two dose levels.

We performed mass cytometry (CyTOF) analysis at these two dose levels (25 mg/kg and 50 mg/kg) to link target engagement (STAT3 protein knockdown), and drug uptake (anti-ASO levels) to pharmacodynamic effects mediated by STAT3 reduction. t-distributed stochastic neighbor embedding (t-SNE) analysis showed that STAT3 knockdown and anti-ASO uptake were highly correlated and observed mainly in cells in the monocyte/macrophage cluster (Fig. 3A). Quantification of STAT3 knockdown in individual immune cell lineages revealed a dose-dependent reduction in the STAT3 protein in myeloid lineage cell types, including several macrophage subsets and dendritic cells, as well as CD4⁺ FOXP3⁺ Tregs, cancer-associated fibroblasts (CAF), and CD31⁺ endothelial cells. At the lower dose of 25 mg/kg, there was 38.7% reduction of STAT3 in the tumor compartment, whereas at the 50 mg/kg dose, STAT3 protein was reduced by 77.3%. Given the clinical observations of limited STAT3 knockdown in tumor cells in danvatirsen-treated human biopsies (Supplementary Table S1), we chose 25 mg/kg as the clinically relevant preclinical dose for future pharmacodynamic assessments. Minimal STAT3 reduction was observed in CD8 T cells (5.7% reduction), NK cells (19.3% reduction), Ly6C⁺ (36.2% reduction), and Ly6G⁺ (23.7% reduction) neutrophil-like cells (Fig. 3B; Supplementary Fig. S5B). By contrast, the greatest STAT3 reduction was observed in myeloid lineage cells, including ∼78% reduction in F4/80⁺ macrophages, 54% reduction in cDC1, and 50.9% reduction in cDC2. Interestingly, we found higher STAT3 reduction in CD4⁺ FOXP3⁻ T cells at the lower dose of 25 mg/kg (14.9%) versus 50 mg/kg (4.6%). The implication of this finding as it relates to CD4 T-cell biology is intriguing, and we are conducting further studies to verify and explain the observation.

To link the pharmacodynamic effects of STAT3 reduction with antitumor effects, we used multicolor flow cytometry to explore the global immune changes mediated by STAT3 ASO monotherapy and in combination with anti–PD-L1, after 14 days of treatment in the CT26 tumor model. Consistent with the observation that cells of the myeloid lineage had the most significant STAT3 reduction (Fig. 3B–D; Supplementary Fig. S5B), t-SNE analysis demonstrated significant remodeling within the myeloid compartment in response to mSTAT3 ASO and combination treatments (Fig. 4A). Specifically, we observed substantial reduction in suppressive tumor–associated macrophages (CD11b⁺ and CD206⁺) in mice treated with monotherapy mSTAT3 ASO (threefold decrease) and in combination with anti–PD-L1 (sixfold decrease). This was accompanied by an increase in presumed CD11b⁺ MHCII⁺ antitumor macrophages (1.7-fold increase with mSTAT3 monotherapy and 2.2-fold increase in combination-treated mice; Fig. 4B and C).

To further explore the tumoral myeloid compartment remodeling mediated by mSTAT3 ASO and combination therapies in CT26, we repeated the flow cytometry analysis on day 18. The flow panel included markers of proinflammatory and anti-inflammatory myeloid cells and markers of infiltrating monocytes, in particular CCR2 and CX3CR1 (27). t-SNE and unsupervised clustering analysis by FlowSOM revealed several cellular subtypes within the CD45⁺CD11b⁺CD11c^low immune population (Fig. 5A and B). The immunophenotypic markers iNOS, MerTK, CX3CR1, CCR2, CD124 (IL4R), and Arg1 were used to classify these subclusters as protumor and antitumor (28). The combination treatment group showed a fourfold increase of MerTK⁻CCR2^hiiNOS^hi antitumor immune cells (Fig. 5C). The protumor mega cluster, exemplified by MerTK⁺CD206⁺CCR2^lo, was decreased twofold in the combination-treated mice (Fig. 5C).

We used IHC to further assess the changes in proinflammatory and protumorigenic macrophages. CD163 positivity, a commonly used marker of M2-like suppressive macrophages (and target of STAT3), decreased twofold, and iNOS-positive proinflammatory macrophages on the periphery of the tumor increased twofold in mice treated with monotherapy mSTAT3 ASO or in combination (Fig. 5D and E). These changes suggest that STAT3 knockdown produced functional changes in addition to macrophage reduction. Consistent with this hypothesis, treatment with a CSF1R inhibitor (29) to deplete suppressive macrophages showed no antitumor activity in CT26, as monotherapy or in combination with anti–PD-L1 (Supplementary Fig. S6). Lastly, we validated changes at the RNA level using NanoString gene-expression analysis of FFPE sections from treated and untreated CT26 tumors. Consistent with the flow cytometry and IHC data, we observed a decrease in CD163 RNA expression and an increase in Nos2 RNA expression in the tumors from mice treated with mSTAT3 ASO alone or in combination with anti–PD-L1 (Supplementary Fig. S7). Furthermore, we saw an increase in STAT1 RNA whereas STAT3 RNA expression was reduced (Supplementary Fig. S7A). Other markers associated with alternatively activated macrophages including Lyve-1 and F13a1 (30, 31) also showed decreased gene expression with monotherapy mSTAT3 ASO, whereas a key T-cell recruiting cytokine Cxcl9 showed increased expression (Supplementary Fig. S7B).

The immunophenotypic changes observed in myeloid cells of CT26 tumors from mSTAT3 ASO-treated mice suggested that danvatirsen could be capable of reversing immunosuppressive effects of STAT3 activation in various myeloid cell types. To test this, human monocytes were differentiated into suppressive human macrophages and cultured with danvatirsen or control ASO for 3 days. Danvatirsen treatment resulted in a proinflammatory phenotype in macrophages, enhancing cytokine secretion and costimulatory molecule expression upon LPS stimulation (Fig. 5F). Dose-response analysis of mouse macrophages and human macrophages treated with the mouse or human selective STAT3 ASO, respectively, revealed similar potency (Supplementary Fig. S8), providing further confidence regarding the translatability of our findings across species. Similarly, cultured human monocyte-derived dendritic cells were treated with danvatirsen for 3 days and then subsequently treated with IL10, to mimic an immunosuppressive TME. Upon LPS stimulation, we observed an increase in the cell-surface expression of CD86, and increased secretion of TNFα, IFNγ, and IL12 with danvatirsen, but not control ASO, demonstrating that STAT3 inhibition can rescue dendritic cells from IL10-mediated suppression (Supplementary Fig. S9A).

STAT3 signaling promotes the expansion and suppressive functionality of MDSCs (32). To understand functional consequences of STAT3 knockdown in MDSCs, human monocyte-derived MDSCs were treated with danvatirsen. Treatment promoted IFNγ and TNFα production in human MDSCs. In concert with reduction of STAT3 protein, and increased cytokine production, we also observed an increase in pSTAT1. In an MDSC/T-cell coculture assay, danvatirsen treatment of MDSCs led to increased T-cell proliferation to a similar extent of CD3/CD28-stimulated T cells (Supplementary Fig. S9B).

Abundance and functional fitness of intratumoral CD8⁺ T cells predict response to immunotherapy (33). In contrast with myeloid cells, treatment of isolated human CD3 T cells with danvatirsen did not significantly alter Th1 cytokine secretion (Supplementary Fig. S9C). We observed a minor reduction in CD3 viability following 3 days of treatment with 5 μmol/L STAT3 ASO (70.1% viable in vehicle treated and 66.3% viable in STAT3 ASO treated, donor 13216). In the CT26 model, at the clinically relevant dose of 25 mg/kg, we observed minimal STAT3 knockdown (14.9% reduction in CD4 and 5.7% reduction in CD8 T cells; Supplementary Fig. S5B), and an increase in Ki67⁺ CD8⁺ T cells (Fig. 6B), suggesting that the minor reduction in T-cell viability observed in ex vivo–treated human T cells does not translate in vivo. Collectively, these data highlight the ability for danvatirsen to remove myeloid-driven immune suppression to help restore T-cell proliferation and function.

To further explore the global changes associated with myeloid compartment remodeling, we analyzed whole tumor homogenates from CT26 tumors after 2 weeks of treatment. Using FDR adjustment, we identified statistically significant changes in cytokines (labeled P_adjusted). The only statistically significant changes (P_adj < 0.05) with either monotherapy treatment versus control ASO were decreases in IL3 and GCSF with anti–PD-L1 treatment, although a number of less significant (P < 0.05) changes were seen with both monotherapy treatments (Supplementary Fig. S10A). However, with the combination treatment, statistically significant (P_adj < 0.05) increases in cytokines associated with lymphocyte and macrophage recruitment [MDC (CCL22) and MIP3a (CCL20)], as well as a decrease in the immunosuppressive cytokine IL10, were observed (Fig. 6), and the IL3 and GCSF changes were maintained. These cytokine changes are consistent with an increase in activated inflammatory monocyte/myeloid populations and greater tumor growth inhibition with the combination than with either monotherapy.

We used flow cytometry and immunofluorescence to further explore changes associated with intratumoral CD8⁺ T cells in CT26 tumors after 2 weeks of treatment with mSTAT3 ASO. t-SNE analysis revealed an abundance of CD8-positive T-cell populations (Fig. 4A). We further interrogated the T-cell changes by performing functional marker analysis after PMA/ionomycin ex vivo stimulation of tumor-associated immune cells. Key effector cytokines, including IL2 and IFNγ, were upregulated (1.7-fold; anti–PD-L1 versus combination-treated animals; Fig. 6B). The tumor antigen–specific T cells expressing cytotoxic functional proteins granzyme B and Ki67 were more abundant in the combination-treated group (Supplementary Fig. S11). One-way ANOVA analysis showed significant differences between PD-L1 versus combination, suggesting STAT3 ASO treatment provides enhancement of functional responses beyond PD-L1 treatment alone. Combination treatment yielded significant increases in granzyme B–expressing CD8 T cells by flow cytometry analysis. Immunofluorescence studies further confirmed the doubling of granzyme B and Ki67-expressing CD8 T cells after combination treatment (Fig. 6C). Although we observed an increasing trend for Ki67 and IFNγ in control ASO-treated mice compared with vehicle, statistical analysis revealed nonsignificant changes for all the tested cytokine markers (Fig. 6B). Collectively, these data demonstrate that the myeloid immune compartment remodeling that occurs following STAT3 ASO treatment favors an increase in CD8 T-cell effector number and function, despite a lack of significant STAT3 knockdown directly in CD8 T cells.

STAT3 inhibition can lead to antitumor effects through tumor cell–intrinsic (STAT3 inhibition in tumor cells) as well as tumor cell–extrinsic (STAT3 inhibition in the TME, particularly immune cells) mechanisms (5). Previous preclinical work demonstrated that danvatirsen can inhibit tumor growth by blocking STAT3 activity in several human tumor models xenografted to immune-deficient mice (15). In those xenografts, danvatirsen could only work through a tumor-intrinsic mechanism because it does not hybridize to mouse STAT3. However, in patients, danvatirsen has the potential to work through extrinsic or intrinsic mechanisms or both. Because ASO drugs exhibit differential uptake depending upon cell type (24), it was important to evaluate which cell types showed drug uptake and/or target knockdown in patient biopsies in order to understand danvatirsen's potential mechanism of action. Although analysis of bulk tumor RNA did not show significant reduction of STAT3, this could be because immune and other stromal cells are minority cell types within the tumor, whereas tumor cells, which do not take up danvatirsen, represent the majority cell type in the bulk tumor RNA analysis. The results presented here demonstrate danvatirsen uptake and STAT3 knockdown in non-tumor cells in the stroma, and proinflammatory gene-expression changes in on-treatment patient biopsies, suggesting that the clinical activity of danvatirsen was mediated through a stromal/immune- rather than direct tumor-mediated mechanism at clinically tolerated doses. The increases in STAT1 and STAT2 observed in peripheral blood and bulk RNA from tumors are intriguing and may be secondary to the increase in IFN (34).

The maximum tolerated dose of danvatirsen is in part limited by thrombocytopenia (16), which is also observed for JAK inhibitors (e.g., ruxolitinib; ref. 35). This suggests a potential class effect of STAT3 inhibition and that approaches to chronically inhibit STAT3 in tumors will need to balance the antitumor effects with the broader effects of inhibiting STAT3 function in normal physiologic processes such as platelet formation (36). Small-molecule STAT3 inhibitors such as OPB-31121 and OPB-51602, which target STAT3 in tumor cells directly (37), have been evaluated in phase I clinical studies but did not progress into phase II due to lack of sufficient antitumor activity and drug-related toxicities such as peripheral neuropathies and pneumonitis (38, 39). Such drug-related toxicities were not observed for danvatirsen in this clinical study (16). Danvatirsen uptake in tumors was observed almost exclusively in cells of the TME, suggesting that the clinical benefit was not due to direct tumor cell modulation, but rather, through remodeling of the suppressive TME. Similarly, we were able to achieve robust antitumor effects in preclinical tumor models without significant direct inhibition within the tumor compartment, and unveiled a mechanism of myeloid immune compartment remodeling that favored proinflammatory macrophages and enhanced T-cell function. Collectively, these data suggest that direct tumor cell inhibition of STAT3 signaling will not be required to achieve clinical benefit.

STAT3 signaling is induced via phosphorylation by JAK kinases, which are activated by various cytokines including IL6, IL10, and growth factors that can be secreted by tumor cells as well as surrounding stroma (40). However, JAK1 and JAK2 kinases phosphorylate and activate other STAT family members as well, including STAT1, which enhances tumor cell killing through induction of IFNγ (41). Therefore, JAK1/JAK2 inhibitors approved for clinical use are not good candidates for immunotherapeutic approaches to address STAT3-mediated immune suppression in the TME (41, 42), in line with our data demonstrating that AZD4205, a JAK1 inhibitor, is actually antagonistic in combination with anti–PD-L1 (Fig. 2B). Furthermore, loss-of-function mutations in the IFNγ–JAK–STAT1 signaling pathway in tumor cells, analogous to JAK1 inhibition, cause resistance to immune-mediated therapy (43, 44). Danvatirsen's selectivity for STAT3, and its effects in patients' tumors with respect to enhancing the type I IFN response and IFNγ signaling, coupled with the preclinical findings demonstrating that mSTAT3 ASO relieves myeloid-driven immune suppression and enhances efficacy of anti–PD-L1 suggest that danvatirsen could enhance the effects of immune-checkpoint blockade in the clinic.

Our preclinical studies demonstrated preferential mSTAT3 ASO uptake and STAT3 knockdown in stromal/immune cells in syngeneic mouse tumors, as was seen in the patient tumors. The increase in type 1 interferon and IFNγ gene-expression signatures in clinical samples is consistent with the increase in IFNγ and CD8 T-cell activity and effects on cells of the myeloid lineage observed in preclinical models. Syngeneic tumor models treated with a mouse surrogate STAT3 ASO had a similar pattern of preferential immune and stromal STAT3 knockdown as was observed in clinical samples. We conducted detailed immunophenotypic analysis of mouse tumors and functional assays on human immune cells to further investigate potential mechanistic effects of STAT3 knockdown by danvatirsen in tumors. Our results suggest that the therapeutic effects of STAT3 ASO, specifically the increases in proinflammatory cytokines and gene expression, are elicited through direct activity on macrophages and dendritic cells that results in reduced immunosuppression and increased proinflammatory changes in the TME, rather than direct effects on NK cells and CD8 T cells.

STAT3 has long been suggested to promote immunosuppressive macrophage proliferation and survival in the TME (7). Clinically, tumor-infiltrating macrophage infiltration by CD163⁺ TAMs is associated with poor survival in human cancers (45). STAT3 knockdown by mSTAT3 ASO effectively orchestrates a fine-tuning of the TME with reversal of suppressive macrophage activity. The decrease in suppressive macrophages is accompanied by an increase of inflammatory macrophages and enhanced T-cell function. We also observed increased STAT1 expression and IFNγ signature in the TME of combination-treated groups, consistent with published reports showing reciprocal regulation and opposite roles of STAT3 and STAT1 in tumorigenesis (41). With reduced tumor-suppressive macrophages, we postulate that T-cell–derived IFNγ drives upregulation of the STAT1 signaling pathway and differentiation of newly recruited monocyte/macrophage cells toward the development of activated iNOS⁺ macrophages.

An interesting observation was the lack of significant STAT3 knockdown in CD8 T cells in CT26 tumors treated with mSTAT3 ASO, despite the fact that mSTAT3 ASO efficacy was lost in an immune-deficient NSG host or when CD8 T cells were depleted. Previous work has demonstrated that STAT3 signaling in CD8 T cells inhibits production of IFNγ, which limits CXCL10 production from myeloid cells, restricting recruitment of CD8 T cells to tumors (46). Furthermore, inhibition of STAT3 signaling in CD8 T cells has been used as a strategy to promote expansion of adoptively transferred T cells (47). Collectively, these data suggest that inhibition of STAT3 signaling in T cells would be beneficial for enhancing T-cell function. In the patient tumor biopsies, we saw an inconsistent increase in T cells with treatment, as evidenced by TCRs; however, we are still exploring gene signatures associated with T-cell functionality. In CT26 tumors and in vitro experiments, STAT3 ASO treatment did not lead to robust STAT3 knockdown in CD8 T cells, but we observed an increase in CD8 T-cell function and robust CD8 T-cell–dependent antitumor effects that are further enhanced with anti–PD-L1 treatment, likely as a consequence of remodeling the suppressive TME. It is interesting that an increase in CXCL9 and CXCL10 are some of the most statistically significant changes observed in on-treatment biopsies from danvatirsen-treated patients (Fig. 1F), and we see an average of ∼twofold increase in T cells in tumors from danvatirsen-treated patients' tumors (Fig. 1H). This suggests that the direct effects of STAT3 ASO on various myeloid populations indirectly affect CD8 T cells and further supports the rationale of combining with PD-L1 blockade.

We see uptake of ASO and knockdown of STAT3 in Tregs within tumors but have not observed a significant change in Treg frequency in the CT26 model following treatment with mSTAT3 ASO. Recent work in orthotopic head and neck mouse models has demonstrated that mSTAT3 ASO treatment was able to mitigate the radiation-induced increase in Tregs, as well as reduce suppressive macrophages and increase proinflammatory macrophages, which resulted in improved antitumor activity (48).

One of the potential weaknesses of the work described here is the use of only one preclinical tumor model to interrogate the cell types affected by STAT3 ASO treatment and the resulting pharmacodynamic changes. It would be interesting in the future to compare the knockdown profile across multiple tumor models, but more impactful would be broader analyses, including imaging approaches, to address changes elicited by danvatirsen in patient tumors. All pre- and on-treatment pairs of biopsies provided in this study were from patients lacking clinical responses. Greater molecular insight may have been possible had we also been able to evaluate samples from clinical responders. The functional consequences of STAT3 ASO on other cells of the TME, fibroblasts, and endothelial cells, where we observed significant STAT3 reduction in the CT26 model, are also of interest. Our findings suggest that STAT3 ASO is not merely depleting macrophages in the TME, but by shifting the balance of suppressive and proinflammatory macrophages, it favors an antitumor microenvironment with enhanced recruitment and function of T cells. Based on data presented here, the combination of danvatirsen and durvalumab, an anti–PD-L1 therapeutic antibody, is currently being tested in a phase I/II clinical study in HNSCC patients.

All authors are or were employees of AstraZeneca at the time this work was performed, with stock ownership and/or stock options or interests in the company. In addition, R. Woessner and P. McCoon report a patent pending for WO2016062772.

T. Proia: Conceptualization, data curation, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review, and editing. M. Singh: Conceptualization, formal analysis, investigation, visualization, methodology, and writing–original draft. R. Woessner: Conceptualization, investigation, methodology, and writing–original draft. L. Carnevalli: Investigation, visualization, and methodology. G. Bommakanti: Investigation, visualization, and methodology. L. Magiera: Investigation, visualization, and methodology. S. Srinivasan: Conceptualization, investigation, visualization, and methodology, writing–original draft. S. Grosskurth: Data curation, software, visualization, and methodology. M. Collins: Investigation and methodology. C. Womack: Formal analysis, investigation, and methodology. M. Griffin: Investigation and visualization. M. Ye: Investigation and methodology. S. Cantin: Investigation and methodology. D. Russell: Data curation, investigation, and methodology. M. Xie: Data curation, software, investigation, and visualization. A. Hughes: Investigation and visualization. N. Deng: Investigation. D.A. Mele: Investigation and methodology. S. Fawell: Conceptualization. S. Barry: Conceptualization. C. Reimer: Conceptualization. J.C. Barrett: Conceptualization. P. McCoon: Conceptualization, data curation, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review, and editing.

We are grateful to Maria Udriste for her expert assistance with analysis of clinical tumor biopsies and to Deborah Shuman for her help with figure preparation. We also thank Ionis Pharmaceuticals for providing the mSTAT3 ASO and for consulting on this work.

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