Suppressive myeloid cells mediate resistance to immune checkpoint blockade. PI3Kγ inhibition can target suppressive macrophages, and enhance efficacy of immune checkpoint inhibitors. However, how PI3Kγ inhibitors function in different tumor microenvironments (TME) to activate specific immune cells is underexplored. The effect of the novel PI3Kγ inhibitor AZD3458 was assessed in preclinical models. AZD3458 enhanced antitumor activity of immune checkpoint inhibitors in 4T1, CT26, and MC38 syngeneic models, increasing CD8+ T-cell activation status. Immune and TME biomarker analysis of MC38 tumors revealed that AZD3458 monotherapy or combination treatment did not repolarize the phenotype of tumor-associated macrophage cells but induced gene signatures associated with LPS and type II INF activation. The activation biomarkers were present across tumor macrophages that appear phenotypically heterogenous. AZD3458 alone or in combination with PD-1–blocking antibodies promoted an increase in antigen-presenting (MHCII+) and cytotoxic (iNOS+)-activated macrophages, as well as dendritic cell activation. AZD3458 reduced IL-10 secretion and signaling in primary human macrophages and murine tumor-associated macrophages, but did not strongly regulate IL-12 as observed in other studies. Therefore, rather than polarizing tumor macrophages, PI3Kγ inhibition with AZD3458 promotes a cytotoxic switch of macrophages into antigen-presenting activated macrophages, resulting in CD8 T-cell–mediated antitumor activity with immune checkpoint inhibitors associated with tumor and peripheral immune activation.
Immune checkpoint blockade (ICB) has changed the cancer treatment paradigm (1). However, the tumor microenvironment (TME) contains multifactorial resistance mechanisms, including suppressive myeloid cells associated with resistance to ICB and poor survival (2–4). Developing strategies to target myeloid resistance mechanisms to enhance response to ICB therapy is important. Tumor myeloid cells are plastic and phenotypically heterogeneous with positive and negative influence on antitumor immunity. Several therapeutic approaches such as CSF1R, CCR2, CXCR2, CXCR4 and most recently PI3Kγ inhibitors (reviewed in ref. 4) deplete or reprogram tumor myeloid cells have yielded modest success. Gaining insight into the mechanisms increasing ICB efficacy is important to refine therapeutic strategies myeloid targeting agents and identify patients likely to benefit.
PI3Kα/β isoforms are ubiquitously expressed (5), whereas PI3Kδ/γ are expressed primarily in immune cells (6). The PI3Kγ isoform regulates immune cell proliferation, survival, migration and function in respiratory, inflammatory, metabolic disorders, and cancer. PI3Kγ inhibitors selectively inhibit the class I PI3K that is expressed in leukocytes and controls chemokine-dependent leukocyte chemotaxis and mast cell activation (7). In the context of cancer, PI3Kγ inhibition is reported to reprogram or repolarize immunosuppressive macrophages from a phenotype to elicit an antitumor immune response (8–10). This repolarization of tumor macrophages is suggested to promote specific pro-inflammatory “M1 macrophage” phenotypes, with activated IFNγ, lipopolysaccharide (LPS), or TNFα signaling with pro-inflammatory and microbicidal functions, and downregulate immunosuppressive “M2 macrophages” that express high levels of IL-10 and scavenger markers, for example, CD206 (11). Therefore, selective PI3Kγ inhibitors may modulate tumor macrophages, changing the immune status to enhance efficacy of checkpoint inhibitors by increasing recruitment or activation of cytotoxic T cells (9, 12). One selective PI3Kγ inhibitor (IPI-549) has been described to enhance ICB in preclinical models and is currently being tested in combination with the anti-PD1 checkpoint inhibitor nivolumab in solid tumors (13, 14).
How PI3Kγ inhibition facilitates improved antitumor T-cell responses has not been widely explored. We have therefore studied AZD3458 a novel selective PI3Kγ kinase inhibitor (15, 16) as a monotherapy and in combination with ICB in syngeneic mouse tumor models. We find that in contrast with studies in other preclinical models AZD3458 changes the activation status of specific cells in the TME, but importantly by modulating myeloid cell–mediated immunosuppression/activation signaling rather than global repolarization of tumor macrophage subtypes.
Materials and Methods
In vivo studies
All animal studies were performed according to UK Home Office and IACUC guidelines. Cell lines CT-26, 4T1, and MC-38 we purchased from the ATCC. CT-26 (5×10^6 cells/mouse) or MC-38 (5×10^6 cells/mouse) tumor cells were implanted subcutaneously in the flank of female Balb/c and C57/Bl6 mice, respectively. 4T1 (5×10^6 cells/mouse) tumor cells were implanted orthotopically in mammary fat pad (o.t.) of female Balb/c mice or subcutaneous. Four days (CT-26 or 4T1) or one day (MC-38) after implantation mice were randomized by body weight before dosing.
For in vivo treatment, anti-mouse CD8 antibody (Bio X Cell) 25 mg/kg, intraperitoneal dosing was started on one day before MC38 cell being implanted for 2 days, then dosed every 5 days for 2 weeks; AZD3458 20 mg/kg, PO dosing, twice a day (8 hours apart) was started on either day 1 or 3 post MC38 cell being implanted; anti mouse PD-1 (Bio X Cell) antibody 10 mg/kg, and IgG2b isotype control antibody (Bio X Cell) 10 mg/kg, α-PD-L1 antibody 10 mg/kg (mouse IgG1, clone D265A; AstraZeneca) and α-CTLA-4 10 mg/kg (mouse IgG1, clone 9D9; AstraZeneca) or the respective isotype controls (αNIP; AstraZeneca). The tumor and mice bodyweight were measured twice a week. Tumor growth inhibition (TGI) = 100 × [(geometric mean (control RTV) − geometric mean (treated RTV)]/[geometric mean(control RTV)].
At the end of the study, tumor tissues were then transferred into the gentleMACS C Tube containing RPMI. Cells were liberated using a mouse tumor dissociation kit from Miltenyi Biotec and octodissociator (Miltenyi) according to the manufacturer's instructions.
The fluorophore-conjugated antibodies and gating strategy used in this study are listed in Supplementary Tables S1 and S2. All antibodies were purchased from BioLegend, eBioscience, BD Biosciences or Cell Signaling Technology. Cells were stained with a viability marker (Live/Dead Aqua, Thermo Fisher Scientific) according to the manufacturer's instructions, and stained for surface/intracellular markers as described previously (17).
Ex vivo CD11b blood assay
For in vitro compound treatment, non-activated whole blood was collected via vena cava into EDTA tubes from female Balb/C mice. The blood was pooled and used for dose response of AZD3458 for 15 minutes (DMSO<0.01%). Whole blood (100 μL) was then stimulated for 30 minutes at room temperature by 100 ng/mL MIP-2.
Following MIP-2 stimulation, red blood cells were lysed using ACK buffer (Lonza #10–548E). Leucocytes were stained with αCD11b-APC (M1/70), αLy6G/Ly6C(GR-1)-PE (RB6–8C5), and αCD3-BUV395 (17A2) with FC block (Thermo Fisher# 14–0161–86, 1in100) for 30 minutes. Following two PBS washes, the cells are fixed with Cytofix (BD Biosciences #554655) for 15 minutes at 4°C.
Acquisition of sample by flow cytometry and analysis.
Cells were analyzed on a BD Fortessa flow cytometer and analyzed using FlowJo software (V.10, Treestar). Tumor statistical analysis was performed by the Tukey's multiple comparison test in conjunction with ANOVA using in house FAST FACS data pipeline. High-dimensional visualization and clustering analysis (t-SNE) was analyzed using Cytobank (18).
Ex vivo tissue culture
Human peripheral blood mononuclear cells (PBMC) were enriched from Leukocyte cones by density centrifugation (Lymphoprep, STEMCELL Technologies), from which CD14+ monocytes are isolated by positive selection (STEMCELL #17858). Naïve monocytes are either used directly or matured for 5 days in full RPMI (Sigma R8758–500 mL, supplemented with 10% FBS), using M-CSF or GM-CSF (at 100 ng/mL; STEMCELL), as indicated, in a humidified CO2 (5%) incubator. For Western blot analyses, CD14+ cells were plated at 1 × 106 cells per mL in full RPMI and treated with AstraZeneca compounds, glycolysis inhibitors (2-Deoxy-D-Glucose and 3-Bromopyruvate from Sigma) and LPS (O26:B6, 100 ng/mL; Sigma) for the indicated times and lysed in RIPA buffer (Thermo Fisher Scientific; supplemented complete protease and phosphatase inhibitors; Sigma). Lysates were quantified for protein content and 10 μg separated by SDS-PAGE on 4%–12% BIS-TRIS gels (Invitrogen). Gels were transferred to nitrocellulose using iBlot2 (Thermo Fisher Scientific) and probed with antibodies from Cell Signaling Technology: HIF-1α (#36169), Hexokinase II (#2867), Phospho-Stat3 Tyr705 (#9131), Phospho-S6 Ribosomal Protein Ser235/236 (#2211), phospho-AKT Thr308 (#2965) and Ser473 (#9271), phospho-ERK p42/p44 Thr202/Tyr204 (#9101S), phospho-NFkB p65 Ser536 (#3033), phospho-mTOR Ser2448 (#2971), and phospho-C/EBPβ Thr235 (#3084). For RNA analyses, M-CSF macrophages were treated as above and lysed in TaqMan lysis buffer (Thermo Fisher Scientific). Expression of target genes was quantified on a Lightcycler 480 II (Roche) using probes for IL-10 (Hs00961622_m1), IL-12B (Hs01011518_m1), multiplexed with 18S RNA (4352930E). Secreted proteins were quantified using MesoScale Diagnostics (MSD) v-plex pro-inflammatory plates (K15049D) or IL-12 p40 plates (K151AQB), digitized using a SECTOR Imager 6000 (MSD).
Gene profiling and GSVA score analysis
Total RNA was isolated from snap-frozen tissue and cells using Qiashredder and Qiazol Lysis Buffer on Qiacube-HT following the RNeasy 96 QIAcube HT total RNA cell with DNase protocol according to the manufacturer's instructions (Qiagen). Reverse transcription was performed from 50 ng of total RNA (Thermo Fisher Scientific #4374967) and genes of interest were preamplified (Thermo Fisher Scientific #4488593; 14 cycles) using a pool of TaqMan primers (listed in Supplementary Table S1), following the manufacturer's instructions (Thermo Fisher Scientific), and further run on a 96.96 Fluidigm Dynamic array on the Biomark according to the manufacturer's instructions (Fluidigm). Data were collected and analyzed using Fluidigm Real-Time PCR Analysis 2.1.1 providing Ct values. All gene expression calculations were performed in Jmp13.0.1, and data represented in TIBCO Spotfire 6.5.2 or GraphPrism. Ct values were normalized to the average of housekeeping genes (dCt), and all treatment group compared with the average control group (−ddCt) and fold Change was calculated by taking 2^−ddCt. Statistical analysis of gene expression data (−ddCt) was performed in Jmp13.0.1, using a pairwise Student t test, which identify genes significantly modulated compared with control. GSVA scoring (19) was performed using genes defined in Rooney and colleagues (20).
For RNA sequencing, total RNA was extracted using the RNeasy 96 Qiacube HT Kit (Qiagen), quality validated using nanodrop and Quantit RNA Assay Kit (Thermo Fisher Scientific). For tumor and lymph-node libraries were made using TruSeq stranded mRNA kit and sequenced on the NovaSeq system (100bp, single end). For peripheral blood, libraries were made using TruSeq total RNA stranded with globin depletion and pooled libraries were run on 2 lanes of HiSeq4000 (50bp, single end). For sorted macrophages, libraries were made using clontech ultra low input mRNA kit and pooled libraries were run on 2 lanes of HiSeq4000 (50bp, single end). The python toolkit bcbio 1.0.8 (https://github.com/bcbio/bcbio-nextgen) was used for quality control read processing and quality control and analyze the sequencing data. Within bcbio, the sequencing reads were aligned using hisat2 2.1.0 for quality control purposes and a QC report was generated using multiqc Quantification of expression Estimation of the transcripts (tpm values) abundance was performed directly against the mouse mm10 Ensembl transcriptome using the Salmon tool 0.9.1 (21) without alignment, or adapter trimming (21, 22). Next, genes with an average count of less than 1 per samples were removed. Lowly abundant transcripts were removed and subsequently, the DESeq2 R package (version 1.16.1) was used to normalize for library size and perform differential expression analysis (23). Data were analyzed through the use of QIAGEN's Ingenuity Pathway Analysis (IPA, QIAGEN, www.qiagen.com/ingenuity). Pathway and Upstream regulator analysis (Supplementary Methods and Supplementary Table S1) were performed with IPA QIAGEN Inc. (24) using >0.5-log2FC fold changes and >0.1 P values obtained by DESeq2. Data accession number E-MTAB-10057.
Error bars relate to SEM unless indicated in figure legends. Appropriate statistical testing was performed using GraphPad Prism (V7) and indicated in the legend. Statistical significance is indicated as follows: *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001.
Pharmacological PI3Kγ inhibition with AZD3458 suppresses myeloid cell activation ex vivo and in vivo
AZD3458 is a potent selective PI3Kγ inhibitor developed for combination with immune checkpoint inhibitors (Fig. 1A). In cell assays, AZD3458 is selective for PI3Kγ with an IC50 value of 7.9 nmol/L (RAW264), 100-fold selective versus PI3Kδ (Jeko), and 1,000-fold selective versus PI3Kα (BT474c) and PI3Kβ (MDA-MB-468; Fig. 1B). In an ex vivo whole-blood mouse primary cell pharmacodynamic assay, AZD3458 inhibited MIP-2–induced CD11b expression with an average IC50 value of 429.6 nmol/L total drug (58.4 nmol/L free drug corrected for protein binding; Fig. 1C and D). AZD3458 exhibits good murine oral pharmacokinetics (Supplementary Fig. S1), and when dosed at 20 mg/kg reduced MIP-2–mediated neutrophil activation as effectively as the CXCR2 inhibitor AZD5069 (ref. 25; Supplementary Fig. S2).
PI3Kγ inhibition is proposed to modulate the TME by increasing macrophage IL12 secretion and reducing IL10 through parallel metabolic and transcriptional regulation pathways (9). To assess regulation of cytokine secretion and signaling by AZD3458 in human macrophages, hPBMC-derived CD14+ monocytes were differentiated with M-CSF and stimulated with LPS (9, 26). AZD3458 inhibited LPS-stimulated pAKT with an IC50 value of 230 nmol/L (Fig. 1E) and decreased LPS-mediated IL-10 mRNA production (Fig. 1F) and IL-10 protein secretion (Fig. 1G). mTORC1/2 inhibition with AZD2014 and inhibition of glycolysis with 2-deoxyglucose (2-DG) also reduced IL-10 secretion. Surprisingly, AZD3458 had little effect on IL-12 mRNA expression and protein secretion under these conditions (Fig. 1H and I). In contrast with AZD2014 and to some extent 2-DG induced expression of IL-12 mRNA and protein secretion (Fig. 1H and E). Although the net effect was a shift in the ratio of IL-12 to IL-10 secretion (Fig. 1J), AZD3458 treatment primarily regulated IL-10.
Inhibition of PI3Kγ in macrophages is reported to modulate LPS-driven regulation of NFκB, mTORC1/2, and CEBPβ pathways (9, 26), as well as HIF1 and STAT3 that control IL-10 and IL-12 expression (27, 28). LPS stimulated the mTORC1/2 (p-S6/p-NDRG1), NFκB (p-P65), and C/EBP (p-C/EBPβ) pathways (Supplementary Fig. S3A). Although AZD3458 inhibited the LPS-induced transient induction of pAKT activity, it did not reduce mTOR pathway biomarkers, pS6 or pNDRG1 (Supplementary Fig. S3A). In contrast, AZD2014 (300 nmol/L) effectively inhibited each of these biomarkers. AZD3458 had differential effects on key transcriptional pathways, suppressing the induction of HIF-1α but not pSTAT3, whereas AZD2014 robustly inhibited both HIF-1α and pSTAT3 induction. Inhibiting glycolysis with 2-DG (Fig. 3B) recapitulated the modulation of HIF1α and pSTAT3 observed with AZD2014, without suppressing pAKT. Therefore, while modulating glycolysis (2DG) and mTORC1/2 (AZD2014) overrides the activation of LPS-induced transcriptional programs associated with IL-10 and IL-12 regulation, PI3Kγ inhibition with AZD3458 specifically downregulated IL-10 expression.
PI3Kγ inhibition modifies the immune TME improving checkpoint blockade response in T-cell infiltrated and excluded tumor models
Given the differential transcriptional regulation observed in human macrophages, AZD3458 modulation of the TME in the ICB-sensitive CT-26 mouse syngeneic tumor model (29, 30) was assessed (Fig. 2A). AZD3458 had limited monotherapy efficacy when dosed orally up to 20 mg/kg QD or BID, which achieves drug concentrations sufficient to fully inhibit PI3Kγ (Supplementary Fig. S1). Higher dose levels (100 mg/kg QD) gave significant TGI of 34% compared with control (P = 0.0295) and a 44.4% response rate (Fig. 2A). At 100 mg/kg, AZD3458 achieves concentrations that may give transient inhibition of PI3Kδ affecting regulatory T cell (Treg) function and CD8+ T-cell activation (31). Analysis of tumor TME by GSVA gene expression analysis (19, 20) revealed a modest decrease in macrophage gene signatures (Fig. 2B) at lower doses that was more striking at 100 mg/kg. However, although increased CD8+ T cells (Fig. 2C) were observed, there was no significant impact on Tregs (Fig. 2D). Greatest monotherapy activity was observed at the 100 mg/kg dose, suggesting that blood concentrations delivering many multiples of cover over PI3Kγ may be required for monotherapy activity. Gene expression analysis of monotherapy-treated tumors revealed increased T-cell activation and type-II IFN γ signaling (Fig. 2E), with the most significant effects at the 100 mg/kg dose. Collectively, these data suggest that AZD3458 modulates the activation status of the TME, increasing IFNγ and T-cell activation, but has limited monotherapy activity in CT26. 20 mg/kg was selected for further efficacy testing as this was sufficient to achieve full cover over PI3Kγ without targeting PI3Kδ, which may deplete Treg cells or enhance T effector cell activation through PI3Kγ-independent mechanisms (31, 32).
Next, AZD3458 was tested in combination with ICB in CT26 (responsive) and 4T1 (ICB resistant) syngeneic tumor models (29, 30). In CT-26, AZD3458 (20 mg/kg BID) and anti–PD-L1 resulted in enhanced response rate compared with PD-L1 monotherapy-driven efficacy (Supplementary Fig. S4A), increase in CD8+ T cells (Supplementary Fig. S4B), T-cell activation markers IL-2Ra (CD25; Supplementary Fig. S4C), and GZMB (Supplementary Fig. S4D) gene expression. No impact on Tregs was observed (Supplementary Fig. S4E). In the ICB refractory macrophage cell enriched 4T1 orthotopic model, monotherapy AZD3458 decreased total macrophages by 2-fold (Supplementary Fig. S5A) and reduced expression of immunosuppressive markers CD206 and PD-L1 measured by flow cytometry (Supplementary Fig. S5B and S5C, respectively). AZD3458 increased GzmB+CD25+ CD8+ cytotoxic T cells (Supplementary Fig. S5D), CD25 (IL2Ra) expression by flow cytometry (Supplementary Fig. S5E) and expression of CD25 and IFN-inducible genes (Supplementary Fig. S5F). Combining AZD3458 with anti–PD-1 checkpoint blockade promoted 66% TGI (P = 0.045) at day 14 after first dose (Supplementary Fig. S5G), comparable with IPI-549 (8), demonstrating that AZD3458 is immune modulatory in vivo in CPI-sensitive and -insensitive models.
AZD3458 and ICB regulate immunomodulatory gene signatures in MC-38 tumors
The MC38 model was selected to explore broader CPI combinations as it shows moderate response to CPI and a TME that is not dominated by any individual cell type (29, 30, 33). AZD3458 enhanced activity of anti–CTLA-4, anti–PD-L1, and anti–PD-1 (Fig. 3B–E). Compared with single-agent arms combination of AZD3458 enhanced the CPI monotherapy response rate of anti–CTLA-4 from 0% to 50% (P = 0.0172), anti–PD-L1 from 24% to 36% (P = 0.1040), and anti–PD-1 from 10% to 81% (P = 0.0007) relative to vehicle control arms in this model (Fig. 3F). The greatest efficacy enhancement occurred with anti-PD1 resulting in increased tumor responses (Fig. 3F).
RNAseq analysis of AZD3458 monotherapy and combination with anti–PD-1 of whole tumors, isolated tumor macrophages (Mac; CD45+CD11b+F4/80+), whole blood (peripheral blood) and tumor-draining lymph nodes (TDLN) was performed (Fig. 4A) to explore mechanisms associated with combination activity. The dominant pathways modulated following monotherapy or combination treatment were related to immune cell function (Supplementary Fig. S6). The pathways predicted by IPA analysis (activation prediction z-scores, adjusted P values and relevant GSVA scoring; ref. 19) are shown (Fig. 4, Supplementary Table S1). In tumor samples, anti–PD-1 monotherapy induced T-cell and IFNγ activation signatures, including IL-2 (+2.2-fold) and IFNγ (+3.3-fold; Fig. 4B). Notably, anti-PD-1 induced no changes in peripheral tissues. Monotherapy AZD3458 modified macrophage-associated signatures in whole-tumor tissue. More marked effects were seen in tumor-derived macrophages (Fig. 4C). In whole tumors, upregulation of TNF, STAT6, IL4, IL13 signaling was dominant, whereas in tumor-derived macrophages activation of TNF, STAT, LPS, IFNγ, NFKβ signatures were evident. Induction of LPS signaling in macrophages is consistent with PI3Kγ activation limiting TLR4 signaling and macrophage reprogramming by biasing inflammatory cytokine outputs (26). The modulation of IL-4 signaling implies multiple inputs influencing the phenotype of tumor-derived macrophages, as IL-4 signaling suppresses LPS-induced production of pro-inflammatory cytokines TNFα and IL-1β impairing phagocytosis but potentiating microbial-induced signaling and cytokine secretion (34). Across the whole analysis in all tissues and cell isolates changes in transcripts associated with both pro- and anti-inflammatory macrophage profiles was evident indicating heterogeneity in the AZD3458 response rather than inducing polarized phenotypes.
Combining AZD3458 and anti–PD-1 resulted in a more dramatic shift in the gene signature profiles in whole-tumor and tumor-derived macrophages. In tumor, the combination induced immune activation signatures associated with increased metabolic signaling, LPS signaling, IL-4, IL-1B, IL-13 and decreased IL-10RA signaling (Fig. 4D). Changes in the peripheral blood were also consistent with pro-inflammatory gene activation, including type I and II IFNs, TNFα, LPS, and IL1-β (Fig. 4D). Changes in the draining lymph node were modest. The data suggest that PI3Kγ inhibition induces an antitumor immune phenotype that is further enhanced when combined with ICB.
IFNγ pathway modulation (GSVA scoring of IFNγ gene signatures IFNG.GS.IO) occurred in all treatment groups across most tissues (Fig. 4E), particularly following combination treatment. Pattern of responses varied, for example, the IFNγ signature was highly upregulated in the tumor following anti-PD1 treatment, whereas AZD3458 drove changes in peripheral tissues. However, combining the two agents enhances multiple immune cell–associated pathways achieving robust elevation of IFNγ gene signature activation across both tumor and peripheral tissues.
PI3Kγ inhibition changes macrophage activation but not polarization in MC38 tumors
Tumor macrophages from AZD3458-treated MC38 tumors decreased IL-10/IL-10RA signaling and increased LPS-mediated activation as key activation signatures (Fig. 5A), mirroring the suppression of LPS-induced IL-10 expression observed in primary human macrophages (Fig. 1E). Analysis of specific macrophage transcript biomarkers shows that AZD3458 induces pro-inflammatory genes Nos2, IL-1α, IL-6 as well as the differentiation/immunosuppressive marker Arg1, which increased further following anti–PD-1. This is consistent with reports describing mixed macrophage phenotypes associated with antitumor responses (12, 26). Consistent with the human macrophages, a reduction in IL-10 expression was seen in macrophages from AZD3458 and PD-1+AZD3458–treated tumors (Fig. 5B). At a single gene level, changes were more pronounced in isolated macrophages (Fig. 5B) than in bulk tumor (Fig. 5C) likely due to dilution of the macrophage gene expression signal. Importantly, in contrast with other studies, here there was no clear change in expression patterns associated with conventional macrophage “M1/M2” polarization markers (3, 9, 26), with IL-12 transcript levels nearly undetectable in these samples. This suggests that antitumor activity is associated with an immune-activated macrophage phenotype across a more complex macrophage population than seen in other studies following PI3Kγ and PD-1 inhibition.
Enhancement of tumor macrophage and dendritic cell activation following PI3Kγ inhibition in MC38 tumors
The effect of treatment on macrophage, dendritic cell (DC) and T-cell activation status was next assessed by flow cytometry (Fig. 6). The proportion of CD11b+F4/80+ macrophages increased over time as tumors expanded, but neither AZD3458 alone or combination treatment (day 4 or 10) affected total macrophage infiltration (Fig. 6A). However, treatment resulted in significant increases in CD11b+F4/80+ macrophages expressing iNOS, a mediator of cytotoxicity by activated macrophages (Fig. 6B and C; refs. 35, 36). In the anti-PD1 group, the expression of iNOS was observed primarily in antigen-presenting MHCIIhi macrophages. Mouse and human TAMs can express PD-1–affecting phagocytic function (37), whereas PD-1 ablation on myeloid cells promotes macrophage function and antitumor immunity (37, 38). However, AZD3458 treatment increased iNOS expression in both MHCIIlo and MHCIIhi macrophages (Fig. 6D and E), further enhanced in the combination group suggesting an impact across the entire macrophage population. Collectively, the data suggest that AZD3458 promotes upregulation of cytotoxic potential across the macrophage subpopulations, and increased antigen-presenting macrophages in the TME. To confirm whether the activation was specific to a subset of macrophages or myeloid cells, t-SNE analysis was performed to visualize the distribution of pro- and anti-inflammatory markers in the macrophage population. Cells were clustered by lineage marker and functional marker distribution are depicted (Fig. 6G). Cells expressing MHC-II with pro-inflammatory marker iNOS had a partial overlap with the immunosuppressive macrophage marker CD206 (Fig. 6H). These observations are consistent with the gene expression analysis with increased expression of MHCII, iNOS, and CD206 in subpopulations of macrophages were important components of the mode of action of PI3Kγ alone and in combination with ICB in MC38 tumors.
In this study, the impact of AZD3458 in the TME was not restricted to macrophages. PI3Kγ activity can influence DC maturation or activation (39). Flow cytometry analysis revealed AZD3458 (but not anti-PD1) gave >3-fold increase in infiltrated CD11c+MHCIIhi antigen-presenting DCs at day 10 (Fig. 6G). Moreover, transcriptome-based signature analysis indicated the presence of enhanced DC maturation in blood and TDLN (Supplementary Fig. S6), suggesting increase peripheral antigen presentation.
Activated macrophages and DCs could both enhance T-cell function. PD-1 treatment increased frequency of CD8+ T cells; however, CD8+ T-cell increases were lower following AZD3458 monotherapy and combination treatment (Fig. 6H). Despite this, following treatment significant increases in activated CD8+ effector (Teff) cell phenotypes, measured by expression of GzmB, CD25 and ICOS (Fig. 6H) were observed. The induction of cytotoxic markers in the macrophages following AZD3458 in combination to PD-1–promoted CD8+ T-cell function, and depleting anti-CD8 antibodies in the MC-38 tumor model abrogated the antitumor response (Fig. 6I and J).
In summary, at concentrations delivering selective PI3Kγ inhibition AZD3458 efficacy in the MC38 model is achieved without depleting macrophages, instead it influences the activation status across the different subsets of tumor-associated macrophages consistent with a cytotoxic/killing MHCIIhi/iNOS+ profile and also increases infiltrating antigen-presenting DCs. The changes in the antigen-presenting myeloid cells are followed by activation of CD8+ T-cell phenotype resulting in immune-mediated antitumor response (Supplementary Fig. S7).
Macrophage-like or neutrophil-like myeloid cells are associated with limiting the response to ICB (40) and drive acute and chronic inflammation (41). Myeloid cells depend on PI3Kγ activity to promote tumor immunosuppression (10), AZD3458 a novel selective PI3Kγ inhibitor enhanced ICB activity, increasing activated tumor T cells. Transcript analysis of CT26 tumors revealed induction of pro-inflammatory gene signatures similar to those published for IPI-549, where PI3Kγ inhibition is associated with reducing “M2” or suppressive macrophage like cells (9, 12). However, biomarker analysis across three syngeneic tumor models (CT26, MC38, and 4T1) revealed that AZD3458 had variable effects on the total levels of tumor macrophages, whereas gene expression analysis revealed heterogeneous macrophage profiles across the population, as well as increasing DC activation. Increased efficacy was associated with activation of pro-inflammatory signaling nodes rather than switching macrophage polarization. These findings support preclinical and clinical observations, suggesting that M1-like and M2-like gene signatures are not mutually exclusive in tumors, with macrophages displaying both gene-expression profiles (42, 43). Hence, the MOA of PI3Kγ inhibitors is more complicated than a binary M1/M2 repolarization, and the biomarker response will vary according to TME context.
The potential heterogeneity is exemplified by the impact of AZD3458 on MC38 tumors. MC38 exhibit an immune-infiltrated TME with defined macrophage populations (44), and partial sensitivity to ICB (29, 30). Despite antitumor activity being similar to CT-26 and 4T1, AZ3458 did not deplete macrophages in MC38. Although tumor macrophages increased immune activation pathways a mixed profile of differentiation biomarkers was observed, suggesting enhanced efficacy without influencing the classical immunosuppressive polarization phenotype (M1/M2). AZD3458 treatment also increased DC infiltration and activation, which in conjunction with macrophages displaying antigen-presenting and -killing phenotype will help optimize antitumor response directly or by increasing effector T-cell activation. Therefore, although PI3Kγ inhibition may reduce tumor macrophages in some settings (9), depletion or repolarization is not essential as increased antigen presentation and cytotoxic potential within the TME can improve efficacy. AZD3458 also modulated neutrophil activation and recruitment, which also influence response to ICB (45). How PI3Kγ inhibitors influence tumor neutrophil function, and whether this contributes to efficacy remains to be established.
PI3Kγ modulates macrophage production of IL-10, IL-12, and IL-6 via NFKβ and CEBPβ (2, 9, 26). Following stimulation increased glycolytic flux and PKM2 activation activates STAT3 and stabilizes HIF-1α (27, 28, 46, 47) that can also directly influences IL-10 expression (9). AZD3458 primarily downregulated IL-10 expression in LPS-activated M-CSF–differentiated human macrophages. Interestingly, both PI3Kγ and mTOR inhibitors suppressed the metabolic burst, but although AZD3458 partially suppressed induction of HIF-1α, AZD2014 fully inhibited HIF1α expression, pSTAT3 induction and NFKβ, S6 and CEBPβ. AZD2014 increased the expression of IL-12, whereas both PI3Kγ and mTORC1/2 inhibition reduced IL-10. This suggests that regulation of IL-10 and IL-12 may be decoupled, associated with the degree of transcription factor modulation achieved with each agent, and that targeting the pathway at different levels induces different effects. This differential of IL-10 and IL-12 modulation was also evident in MC38 tumors following monotherapy and combination treatment, with isolated macrophages derived from both monotherapy and combination groups exhibiting reduced IL-10 signaling, but activated TLR and metabolic signaling pathways. Increased IL1β and IL-12 was not apparent, despite being observed in other studies (9).
In whole, MC-38 tumors LPS and IFNγ pro-inflammatory pathways were increased by AZD3458, whereas combination treatment further increased expression of cytotoxic effector molecules iNOS, IL-1α as well as IL-6 (46). Despite inflammatory pathway induction, the suppressive macrophage marker CD206 was also expressed. Therefore, full macrophage repolarization was not required to elicit impact of PI3Kγ inhibitors in the TME, further suggesting that targeting PI3Kγ may be heterogeneous and not dependent on M1/M2 phenotypes. Interestingly, following combination treatment, changes in peripheral blood macrophage biomarkers mirrored some changes observed in the tumor, with increase in IFNγ, IL1B, and LPS signaling and decrease in IL-10 signaling. There is evidence that PI3Kγ influences DC function in development, inflammation, and cancer immunity (39, 48). In the MC38 model inhibiting PI3Kγ increased DC cell infiltration and activation status. Further work will be needed to determine whether this is direct or indirect, but could be a feature of PI3Kγ therapies that differentiates from other myeloid-modulate approaches.
In summary, we show that the PI3Kγ inhibitor AZD3458 can enhance checkpoint efficacy associated with modulation of tumor macrophage production of anti-inflammatory cytokines, increase in pro-inflammatory signaling and activated bacterial-like macrophage killing profile. In parallel activation of DCs is evident. Collectively, this enhances CD8 Teff cell activation and immune-mediated efficacy. Importantly, these data suggest that the effects of inhibiting PI3Kγ in the TME will vary based on context. Rather than simply depleting or repolarizing immunosuppressive macrophages, the effects may be more heterogeneous, and moreover impact other programming events such as the metabolic features, induction of cytotoxic functions, and antigen presentation potential. This has important implications for both biomarker selection and interpretation, as well as for optimization of treatment regimens.
L.S. Carnevalli reports other from AstraZeneca outside the submitted work. A. Ramos-Montoya reports employment and is a shareholder of AstraZeneca. D. Carroll reports other from Astrazeneca during the conduct of the study, as well as other from Astrazeneca outside the submitted work. M. Moschetta reports other from AstraZeneca during the conduct of the study. P. Morentin Gutierrez reports personal fees from AstraZeneca, and personal fees from AstraZeneca outside the submitted work. S.E. Critchlow reports employment and is a shareholder of AstraZeneca. T. Klinowska reports other from AstraZeneca during the conduct of the study, as well as other from AstraZeneca outside the submitted work. S.E. Fawell reports employment and is a shareholder of AstraZeneca. S.T. Barry reports employment and is a shareholder of AstraZeneca. No disclosures were reported by the other authors.
L.S. Carnevalli: Conceptualization, supervision, investigation, writing–original draft, writing–review and editing. M.A. Taylor: Data curation, investigation. M. King: Investigation, writing–review and editing. A.M.L. Coenen-Stass: Data curation, investigation. A.M. Hughes: Formal analysis, investigation. S. Bell: Investigation. T.A. Proia: Investigation. Y. Wang: Investigation. A. Ramos-Montoya: Investigation. N. Wali: Methodology. D. Carroll: Investigation. M. Singh: Investigation. M. Moschetta: Investigation. P. Morentin Gutierrez: Investigation. C. Gardelli: Investigation. S.E. Critchlow: Resources. T. Klinowska: Resources, writing–review and editing. S.E. Fawell: Resources. S.T. Barry: Conceptualization, writing–original draft.
We would like to thank AstraZeneca UK and US in vivo and PD groups for support with tumors models, provision of primary tissues and flow cytometry support. Elizabeth Kuczynski for critical review of the article. Graeme Smith, Elizabeth Hardaker, and members of the UK I/O groups for critical discussion of the work described in article. Illustrations Created with BioRender.com.
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.