Dose-specific intratumoral GM-CSF modulates breast tumor oxygenation and anti-tumor immunity

Abstract

GM-CSF has been employed as an adjuvant to cancer immunotherapy with mixed results based on dosage. We previously showed GM-CSF regulated tumor angiogenesis by stimulating soluble VEGFR-1 from monocytes/macrophages in a dose-dependent manner that neutralized free VEGF, and intratumor injections of high-dose GM-CSF ablated blood vessels and worsened hypoxia in orthotopic PyMT triple-negative breast cancer (TNBC). In this study, we assessed both immune-regulatory and oxygen-regulatory components of low-dose vs. high-dose GM-CSF to compare effects on tumor oxygen, vasculature, and anti-tumor immunity. We performed intratumor (IT) injections of low-dose GM-CSF or saline controls for 3 weeks in FVB/N PyMT TNBC. Low-dose GM-CSF uniquely reduced tumor hypoxia and normalized tumor vasculature by increasing NG2⁺ pericyte coverage on CD31⁺ endothelial cells. Priming of “cold”, αPD1-resistant PyMT tumors with low-dose GM-CSF (hypoxia reduced) sensitized tumors to αPD1, while high-dose GM-CSF (hypoxia exacerbated) did not. Low-dose GM-CSF reduced hypoxic and inflammatory TAM transcriptional profiles, however no phenotypic modulation of TAM or TIL were observed by flow cytometry. On the other hand, high-dose GM-CSF priming increased infiltration of TAMs lacking MHCII^hi phenotype or immune stimulatory marker expression, indicating immune-suppressive phenotype under hypoxia. However, in αPD1-susceptible BALB/c 4T1 tumors (considered “hot” vs. PyMT), high-dose GM-CSF increased MHCII^hi TAM and immune stimulatory molecules, suggesting disparate effects of high-dose GM-CSF across PyMT vs. 4T1 TNBC models. Our data demonstrate a novel role for low-dose GM-CSF in reducing tumor hypoxia for synergy with αPD1 and highlight why dosage and setting of GM-CSF in cancer immunotherapy regimens require careful consideration.

INTRODUCTION

Triple negative breast cancer (TNBC) is the most aggressive type of breast cancer with limited treatment options and poor survival. A classic feature of the TNBC tumor microenvironment (TME) is hypoxia. Hypoxia is a hallmark of solid tumor progression and correlates with worse overall prognosis in breast cancer patients (1). The tumor attempts to meet increasing oxygen demand as it grows by excessive production of pro-angiogenic factors such as vascular endothelial growth factor (VEGF) to stimulate tumor angiogenesis. Ironically, tumor angiogenesis is highly dysregulated and fails to resolve, and even exacerbates, hypoxia by forming excessive, abnormal vessels that lack pericyte coverage and cannot perfuse blood and oxygen (2). Hypoxia drives resistance to a number of treatment modalities, including anti-programmed cell death protein 1 (αPD1) immune checkpoint blockade (ICB) recently approved to treat TNBC (3). However, TNBC response rates to αPD1 are an abysmal 19% (4). A strategy aimed to overcome tumor hypoxia and improve ICB efficacy is vessel normalization, in which excess tumor VEGF is neutralized but sufficient VEGF remains to promote vessel stability, regain or maintain pericyte coverage, and increase perfusion (2, 5). Strategies such as bevacizumab (VEGF-neutralizing antibody) or VEGF Trap (VEGFR binding domains fused to Fc receptors) have been employed at low doses for vessel normalization. However, one hurdle to these treatments is the reinstatement of dysfunctional angiogenesis by tumor myeloid cells to promote resistance (6); thus, modulating myeloid-driven angiogenesis may present a better option.

Tumor-associated macrophages (TAMs) are the dominant myeloid population in breast TME that can comprise up to 50% of the tumor mass (7). Along with hypoxia, increased numbers of TAMs predict poor outcomes in breast cancer (8). Macrophages recruited into the tumor are highly plastic, which the immune-suppressive and hypoxic TME uses to its advantage by producing tumor-derived factors for polarization into “tumor-educated” “M2-like” TAMs. Immunosuppressive TAMs subvert anti-tumor CD8⁺ responses. Furthermore, TAMs accumulate in hypoxic regions and are a substantial source of VEGF to drive the formation of dysfunctional tumor vessels that perpetuate hypoxia (9, 10). Thus, therapeutic re-shaping of TAM towards immunostimulatory, anti-angiogenic phenotype may reduce tumor hypoxia and immunosuppression for improved ICB efficacy.

Granulocyte-macrophage colony-stimulating factor (GM-CSF/CSF2) drives development, maturation, and differentiation of myeloid cells. In vitro, GM-CSF induces monocyte differentiation into immunostimulatory “M1” macrophages, characterized by MHCII, CD80/86 and iNOS. GM-CSF may re-direct immunosuppressive, “M2-like” pro-tumor TAMs into immunostimulatory, MHCII^hi “M1-like” anti-tumor phenotype (11). In this regard, GM-CSF has been explored as an adjuvant to cancer immunotherapies (12) in various modalities ranging from GM-CSF-expressing oncolytic viruses (T-VEC), GM-CSF-secreting cancer vaccines (GVAX), or co-administration of GM-CSF with peptide antigens/antigen-loaded DCs (13, 14). While GM-CSF has been primarily studied in melanoma, autologous GM-CSF-secreting tumor vaccines are currently being tested in breast cancer (15). However, conflicting results have been reported to-date, and clinical trials have been disappointing. While some studies showed that administration of GM-CSF increases TAM MHCII^hi “M1-like” phenotype (11), antigen-presenting capacity, and proinflammatory cytokines (16), other studies reported that GM-CSF induces expansion of immunosuppressive myeloid cells that suppress CD8⁺ TIL and promotes tumor progression (14, 17). While these outcome differences could be cancer type-specific(18), another possibility is the dosage variation of GM-CSF reported across studies. In vitro work showed dosage of GM-CSF critically determines its effects on macrophages (19). Another possibility is that GM-CSF effects on macrophages other than immunoregulatory properties has not been considered.

Our lab previously reported that GM-CSF modulates TAM-mediated angiogenesis within the murine breast TME. We showed that GM-CSF induced the production of the anti-angiogenic molecule soluble VEGF receptor-1 (sVEGFR-1/sFLT1), an alternatively spliced form of the membrane-bound VEGFR-1, from monocyte-derived macrophages/TAM. We found that GM-CSF induced VEGFR-1 production in a dose-dependent manner, where 100ng expressed the most abundant sVEGFR-1 levels (20). Intratumor (IT) injections of 100ng “high-dose” GM-CSF 3 times per week (3x/wk) in orthotopic PyMT murine breast tumors “overwhelmed” the tumor with high TAM sVEGFR-1 leading to vessel ablation and exacerbation of hypoxia (21). Although we now know that driving increased hypoxia is unsuccessful as a treatment in breast cancer (22), these data set a precedent for GM-CSF-mediated regulation of tumor vasculature and oxygen via TAM sVEGFR-1, as effects were lost when co-treated with a neutralizing antibody to sVEGFR-1 (21). Mixed benefits of GM-CSF may be due in-part to a lack of consideration of effects of GM-CSF on tumor angiogenesis, macrophage sVEGFR-1 production, and tumor oxygenation - important parameters that influence tumor progression and resistance to ICB. Further, because we observed substantially less sVEGFR-1 at lower doses (1–10ng) of GM-CSF vs. 100ng “high-dose” (20), we hypothesized that low-dose IT GM-CSF would normalize tumor vasculature and reduce tumor hypoxia, thereby promoting anti-tumor immunity.

Here, we report disparate effects of high-dose (100ng) vs. low-dose (5ng) GM-CSF in regulating tumor angiogenesis, intra-tumoral oxygen levels, and immune cell phenotypes within the TME, which correlate with their differential effects on sensitizing the immunologically “cold” PyMT tumors to αPD-1 therapy. In addition, we compared the effects of GM-CSF on the immune cell profile within the TME in “cold” PyMT to the 4T1 murine model of TNBC, which is considered “hot” relative to PyMT. Our findings suggest both immune-regulatory and oxygen-regulatory components of GM-CSF are critical to inform optimal dose and setting in cancer immunotherapy for TNBC and other solid tumors.

MATERIALS AND METHODS

Murine tumor models

PyMT tumors were generated via orthotopic implantation of PyMT MET-1 cells or transplantation of isolated primary tumor cells from 12⁺ wk Stage IV FVB/N MMTV-PyMT spontaneous tumors into the 4^th mammary fat pad of WT FVB/N female virgin mice (6-wk old, Jackson Laboratories, 001800) using an insulin syringe. PyMT MET-1 cells from MMTV-PyMT murine tumors of FVB/N background were a gift from Alexander Borowsky (23). MET-1 cells were cultured in DMEM (Gibco #11995–065) supplemented with 10% FBS (R&D Systems #S11550), 10 μg/mL recombinant human Insulin (MP Biomedicals #193900) and 5 ng/mL recombinant human epidermal growth factor (Gibco #PHG0311). 0.5×10⁶ MET-1 cells (in 50 μl DMEM) or 0.5×10⁶ PyMT tumor cells (in 50 μl Kaighn’s F12K) from female MMTV-PyMT Stage IV tumors were injected for implantation. Stage IV MMTV-PyMT-derived tumor cells were isolated by tumor homogenization using Miltenyi tumor homogenization kit and Octomacs instrument followed by co-depletion of CD31⁺ and CD45⁺ cells by Miltenyi magnetic column separation. For 4T1 model, 4T1 cells (ATCC 2539) were cultured in RPMI 1640 + 10% FBS. Five hundred thousand 4T1 cells were orthotopically implanted as described above in 50 μl RPMI into WT BALB/cJ female virgin mice (6–8 wk, Jackson Laboratories, 000651). Tumor growth was determined using calipers and the following equation: volume=0.5(shortest dimension² x longest dimension). All in vivo experiments were done in strict accordance with protocols approved by the Institutional Animal Care and Use Committees at West Virginia University. Mice with intraperitoneally invading tumors were sacrificed and excluded from analysis. Mice were assigned treatment groups completely at random. The experimenter assigned unique identifiers to each mouse based on each animal’s cage number, ear clipping, and tail stripe (I-V) applied with a marker for blinding of self and others. Mice with intraperitoneally invading tumors, heavily ulcerated tumors, or tumors that naturally disappeared shortly after presentation were sacrificed and excluded from analysis. Our study utilized only female mice because male breast cancer represents less than 1% of breast cancer cases.

Intratumor (IT) GM-CSF treatments

Recombinant murine GM-CSF (carrier free) was purchased from R&D Systems (415-ML-050/CF) and reconstituted in phosphate buffered saline. GM-CSF aliquots were periodically validated for functionality by assessing phospho-STAT5 on GM-CSF-treated bone marrow monocytes. Once mice formed established tumors (minimum of ~50mm³ volume to allow for intratumor injection), mice were randomly split into groups and subjected to intratumor (IT) injections of GM-CSF (dose-specified according to each experiment) in 50 μL isotonic saline, or isotonic saline alone as a control using an insulin syringe. Mice were blinded from the experimenter during treatments and analyses by assigning unique identifiers to each mouse based on each animal’s cage number, ear clipping, and tail stripe (added to experimental mice by marker) that could then be matched back to a specific treatment. Treatments were performed 3 times a week (3x/wk) for 3 weeks total (total of 9 treatments). After 3wks, mice were euthanized and subjected to various experimental procedures as outlined. For consistency, all analyses were performed the day immediately following the final IT treatment, except EPR described below which was performed on the final day.

Electron Paramagnetic Resonance (EPR)

In vivo tumor oxygen was measured on the final day of IT GM-CSF administration using L-Band EPR Spectroscopy (~1.2 GHz) (Magnettech, Germany) or Rapid Scan EPR Imaging (24). Mice were anesthetized using air-isoflurane mixture. Soluble oxygen-sensing trityl radicals Ox071 (2 mM solution in saline, pH=7.4) or HOPE (3 mM solution in saline, pH=7.4) were injected intratumorally with an insulin syringe. The injection volumes were adjusted to the tumor size to cover similar volumes of the tumors (20–200 μL for Ox071 and 20–50 μL for HOPE). Both Ox071 and HOPE spin probes were synthesized by our group as previously described (25, 26). Both probes are green which allow for visual confirmation of successful injection. HOPE and Ox071 oxygen-sensing soluble probes are green which allow for visual confirmation of successful injection. Injection volumes were adjusted to tumor size to cover similar volumes of the tumors (20–200 μL for Ox071 and 20–50 μL for HOPE). After probe injection into the tumor, a 15-minute equilibration period was utilized prior to data collection to account for environmental oxygen (pO₂) delivered with the probe. Anesthetized mice were then placed on the warmed platform and surface coil resonator placed directly onto the mammary tumor at the site of injection for EPR readings. Three EPR spectra were recorded using the following acquisition parameters. For Ox071; sweep width: 0.96G, sweep time: 30 s, modulation amplitude: 30 mG, modulation frequency: 100 kHz, non-saturating power. For HOPE; sweep width: 0.96G, sweep time: 60 s, modulation amplitude: 40 mG, modulation frequency: 100 kHz, non-saturating power. The in vivo spectra of Ox071 were simulated using EasySpin (27). The oxygen-sensitive Lorentz component of the linewidth was converted to pO₂ using a calibration curve. The in vivo spectra of HOPE were fitted using a MATLAB algorithm/application reported previously to extract pO₂ (26). For RS-EPRI, the anesthetized mice were placed within the surface coil resonator at the site of injection. Data acquisition was performed using a locally built RS-EPR imaging system (24, 28–30). For image reconstruction, a previously developed algorithm (31) was integrated into the locally developed software. Images were constructed from the total number of 3276 projections acquired within 10–15 mins. A recently developed automatic digital control for the resonator tuning and coupling ensured consistent critical coupling conditions during in vivo measurements (28). A microcontroller Teensy 4.1 was programmed to adjust tuning and coupling in near real-time. The firmware uploaded into the microcontroller was developed using a compatible Arduino Integrated Development Environment software. Control over the Teensy was achieved via serial port communication from a personal (Windows) computer. A Graphic User Interface (GUI) was written in MATLAB’s appdesigner tool. Beside the Teensy, GUI controls other external hardware units, including the magnet, gradient powers supplies, function generator, arbitrary waveform generator, and an audio amplifier.

Tumor homogenization and flow cytometry

Excised tumors were injected with 1 mg/mL collagenase D and 200 mg/mL DNase I in RPMI 1640 using an insulin syringe and minced with a razor blade/scissors. After incubation in 37°C water bath for 30 min, homogenates were pushed through a 70 μM cell strainer with a plunger, diluted 1:1 in RPMI and centrifuged at 1500 RPM for 5 min at 4°C to achieve a final single cell tumor suspension in 1x PBS, 2% FBS and 1 mM EDTA. Fc receptors were blocked with anti-CD16/CD32 (2.4G2, BD Biosciences) for 10 minutes on ice prior to enrichment of tumor leukocytes using CD45 Mouse TIL Isolation Kit (StemCell Technologies, Cat #100–0350) together with StemCell Technologies EasySep^™ magnets according to the manufacturer’s protocol. Purified leukocytes were plated into 96-well V-bottom plates and again subject to brief incubation with anti-CD16/CD32 prior to subsequent immunostaining.

Surface and Live/Dead staining were performed simultaneously in 100 μL FACs Buffer (PBS + 1% FBS) for 30 min at 4°C. For surface marker staining, cells were then fixed in 2% paraformaldehyde (PFA, Thermofisher) for 20 min at 4°C prior to flow cytometric analyses, unless otherwise indicated. For staining of intracellular or nuclear proteins, cells were fixed with eBioscience^™ Foxp3 Fixation/Permeabilization buffer (Fisher) for 30 min at 4°C after surface staining, followed by washing and intracellular staining in 1x eBioscience^™ Perm/Wash buffer for 30 min at 4°C. Panels with TIM3 were first fixed in 2% PFA for 20 min after regular surface staining (to preserve TIM3 staining) followed by fixation in eBioscience^™ Foxp3 Fixation/Permeabilization buffer (Fisher) for 20 min. For intracellular cytokine staining, cells were stimulated with 50 ng/mL PMA (MilliporeSigma P1585) and 1 μM ionomycin (MilliporeSigma I3909) in the presence of GolgiPlug (1:2000, BD Biosciences 555029) and GolgiStop (1:3000, BD Biosciences 554724) at 37°C for 4 hr. Cells were then stained with surface markers, fixed in 2% PFA, and incubated overnight in FACs buffer at 4°C. The next morning, cells were permeabilized using 1X BD Perm/Wash Buffer and subjected to intracellular cytokine staining in Perm/Wash for 30 min at 4°C.

All antibodies and dilutions used in the study are listed in Supplementary Data Table 1. Samples were stored in FACs buffer at 4°C until flow cytometric analyses were performed the next day if fixed, or later that same day if not fixed (specified in each figure legend). Cells were gated on singlets, then live CD45⁺ cells, and the corresponding gating strategy for each population. Data were run on a BD LSR Fortessa. Compensation was performed using FACs DIVA software with the use of unstained tumor controls, single stained tumor controls, and FMO controls. Data were analyzed with FlowJo V.9 software and graphed using GraphPad Prism.

Cell sorting and RNA Isolation

Purified tumor leukocytes were isolated as described above, stained with surface markers, and sorted for TAMs (Live CD45^hi CD11b⁺ F4/80⁺) using a BD FACs Aria III Sorter. Cells were sorted into phenol red-free DMEM supplemented with 10% low endotoxin (certified <0.06 EU/mL) FBS. Purity of >95% was confirmed by post-sorts on samples where cell numbers were sufficient. RNA from pre-sorted TAMs was isolated using RNeasy Plus Mini Kit (Qiagen 74136) and concentrated using RNeasy MinElute Cleanup Kit (Qiagen 74204).

RNA-Seq Library preparation and analyses

Library preparation and sequencing were performed at the Genomics Core Facilities of West Virginia University and Marshall University, respectively. RNA-Seq libraries were built using the NEBNext Single Cell/Low Input RNA Library Kit (New England Biolabs E6420S). We used the maximum volume of total RNA (7μl) for this protocol and generated cDNA using 21 PCR cycles during the initial cDNA Synthesis/Amplification step. Upon completion of the cDNA synthesis, all samples were quantified via Qubit Fluorometer (ThermoFisher) measurement. A total of 10 ng cDNA was used for the Fragmentation/End Prep portion of the protocol. NEBNext Multiplex Oligos for Illumina (Illumina specific UDI’s E6440S) were ligated onto the fragments and amplified via 8 cycles of PCR. The completed libraries were quantified using Qubit Fluorometer (ThermoFisher), and the quality of the libraries was determined by Bioanalyzer High Sensitivity DNA Analysis (Agilent). The libraries were sequenced in the Marshall University Genomics Core using the NextSeq2500 system to generate 50 bp paired-end reads (Illumina). We followed our previous work for RNA-Seq data analysis (32, 33) : paired-end short reads aligned against the mouse reference genome (mm10) by subread v2.0.1 (34), number of reads to RefSeq gene annotations summarized by featureCounts (35), gene expression level quantified by RPKM (reads per kilobase of exon model per million mapped). Differentially expressed genes were identified by EdgeR (36) with the following criteria FDR < 0.05; fold change > 1.5, and expression in at least two samples from the compared conditions. Gene Set Enrichment Analysis against selected MSigDB gene sets was conducted by GSEA (v4.1.0) (37, 38) using fold changes of all expressed genes as input. RNA-Seq sequencing data is deposited to GEO with accession # GSE228874 (secure token to access the data in GEO for reviewer purpose on https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE228874uvcxewkgztmbfaj).

Immunohistochemistry (IHC)

To observe lectin perfusion, tumor-bearing mice were retro-orbitally injected with 1 mg/mL Dylight594-labeled tomato lectin (Vector Laboratories) 5 minutes prior to euthanasia and then perfused intracardially with 1x PBS and heparin solution followed by 4% paraformaldehyde (Formalin). Tumors were removed, cut in half, and fixed in 4% paraformaldehyde overnight at 4°C, then cryoprotected with 20% sucrose, frozen in OCT medium, and sectioned at 10 μM. Tissue sections from each tumor half were stained with anti-CD31 antibody (1:50, FITC rat anti-mouse, BD Pharmingen 561813) and anti-NG2 antibody (1:100, rabbit anti-mouse, EMD Millipore AB5320). Detection of NG2 was accomplished using Alexa Fluor 647 secondary antibodies (1:5000, Invitrogen A21245). Sections were DAPI-stained and mounted with Prolong Antifade Mountant (Invitrogen). Images were captured using Nikon A1R Confocal system with Plan Fluor 40X Oil objective lens. Three images were collected per section at random in CD31⁺ regions with Hamamatsu Orca Flash 4.0 monochrome cMOS camera and processed using ImageJ by thresholding using MaxEntropy function for analysis. For HIF1α IHC, tumors were removed and fixed in 4% PFA for at least 24 hr at 4°C and then placed in PBS prior to paraffin embedding and sectioning at 4 μM. After deparaffinization, antigen retrieval, and blocking with 2% horse serum, sections were stained with anti-HIF1α antibody (5 μg/mL Cell Signaling 480485), horse anti-rabbit secondary antibody (at concentration supplied, Vector Laboratories MP-7401), and DAB substrate (per manufacturer’s recommendations, Dako K3468), and counterstained with Harris Hematoxylin (Leica 3801562). Slides were imaged using a SlideScanner with a 10X objective lens and quantified for HIF-1α positivity by the masking setting in ImageJ (39).

In vivo immunotherapy and chemotherapy studies

After PyMT tumors reached a volume of 50mm³, mice were randomized into treatment groups. Mice were randomized as they hit the required size criteria and cohorts of treatments were mixed throughout cages. First, mice receiving IT GM-CSF were administered IT GM-CSF or IT saline 3x/wk for a 2wk “priming” phase. Next, mice received immunotherapy or chemotherapy for 3wks, alongside continued IT GM-CSF or saline as a “combination phase,” for a total of 5wks. αPD1 or IgG antibodies were administered intraperitoneally 3x/wk at 10 mg/kg (BioXCell InVivoMAbs: αPD1 - clone J43 in InVivoPure pH=6.5 Dilution Buffer; IgG - polyclonal Armenian hamster, diluted in InVivoPure pH=7.0 Dilution Buffer), or Doxorubicin administered intraperitoneally at 1x/wk at 5mg/kg (Tocris, diluted in 1x PBS). All mice received a total of 15–16 IT treatments (3x/wk for 5 wk), 9 Ab treatments (3x/wk for 3wk), or 3 DOX treatments (1x/wk for 3wk). Mice were weighed and scored weekly for morbidity and monitored for survival. Tumor growth was monitored by measuring the largest dimension and perpendicular dimensions with calipers. Excised tumors were weighed for burden. Removal criteria for survival studies were met when tumors reached 20mm in largest diameter or other health-related criteria within our approved animal protocol.

Statistical Analyses

Analysis of Variance (ANOVA) with Tukey’s post-doc test was utilized for experiments with more than 2 groups without repeated measures. Mann-Whitney U-Test was utilized for experiments with 2 groups without repeated measures. Statistical outliers were determined by Grubb’s Outlier Test and excluded from analyses. Tumor growth data were analyzed using mixed-effects modeling, accounting for observational dependencies for each tumor. Normality assumptions were confirmed by residual plots generated by model. Data were analyzed using SAS 9.4 software (SAS Institute, Cary, NC). A threshold of p<0.05 was used to denote statistical significance.

RESULTS

Low-dose intratumor (IT) GM-CSF reduces tumor hypoxia

In our previous studies, we reported that GM-CSF induced the alternatively spliced soluble form of VEGF receptor-1 (sVEGFR-1/sFLT1) from monocyte-derived macrophages to sequester VEGF from bioactivity on endothelial cells (20, 21). We showed that macrophage sVEGFR-1 production occurred in a dose-dependent manner across 1–100ng, where 100ng GM-CSF led to maximum sVEGFR-1 production. In those works, we utilized 100ng “high-dose” intratumor (IT) injections of GM-CSF 3x/wk for 3wk, which ablated vessels and induced near anoxia, in vivo. However, the detriments of tumor hypoxia have since been elucidated, and studies now focus on vessel normalization to reduce hypoxia (2). We hypothesized that a lower treatment dose of GM-CSF could produce sVEGFR-1 sufficient to moderate instead of deplete tumor VEGF and potentially reduce tumor hypoxia, in vivo. We performed IT injections of low treatment doses of 5ng or 15ng GM-CSF -- 20x and 6.6x lower than 100ng high treatment dose GM-CSF from our previous studies, respectively, -- or isotonic saline controls 3x/wk for 3wk in murine orthotopic PyMT Stage IV TNBC. After 3 wks, average tumor oxygen was measured in vivo via L-Band Electron Paramagnetic Resonance Spectroscopy (EPR) (40, 41). Overall IT GM-CSF regimen and EPR methodology is outlined in Fig 1A. We observed ~30% increase in average tumor oxygen levels from both 5 and 15ng GM-CSF relative to saline (Fig 1B) without augmenting tumor burden (Fig 1C). We reasoned that 5ng was our optimal low treatment dose for hypoxia reduction as it required less protein for similar effects. Thus, from this point forward, 5ng GM-CSF is referred to as low-dose GM-CSF. In an expanded study comparing low- dose GM-CSF vs. saline, we found that while larger tumor volume significantly correlated with increased hypoxia in saline-treated tumors (r²=0.4249), this correlation was abolished in low-dose GM-CSF-treated tumors (r²=0.0002) (Fig 1D). Overall average tumor oxygen was moderately, although not significantly, increased to levels near normal mammary controls (Fig S1A). Given heterogeneity of oxygen throughout a solid tumor, we wondered how GM-CSF modulated changes in total tumor oxygen distribution, especially in large tumors at the end of a 3wk study. We utilized Rapid Scan EPR Imaging (RS-EPRI) which provides quantitative 3D spatial distribution of oxygen at voxel-to-voxel sensitivity via imaging throughout the tumor microenvironment (30) to compare tumor oxygen distribution between groups. Representative RS EPRI images, corresponding histograms of oxygen distribution of tumor oxygen levels, and quantification reveals that low-dose GM-CSF significantly increased overall tumor oxygen distribution to levels similar to normal mammary controls (Fig 1E). Tumor oxygen rescue after low-dose GM-CSF is evident in all spatial dimensions (Movie S1 and S2). We also observed a significant reduction of HIF-1α expression in low-dose IT GM-CSF-treated tumor sections vs. saline, supporting oxygen-dependent degradation of HIF-1α in those samples and further validating reduced hypoxia (Fig 1F). Lastly, to pinpoint that hypoxia reduction is unique to low-dose GM-CSF, we performed IT injections of high-dose (100ng) GM-CSF for 3wk and observed oxygen distribution similar to or lower than saline controls (Fig 1G), in alignment with our previous data (21). This demonstrates that high-dose GM-CSF maintains or even exacerbates hypoxia, whereas low-dose GM-CSF reduces hypoxia. Together, these data suggest that GM-CSF differentially regulates tumor oxygen based on dose, and that low-dose IT GM-CSF uniquely reduces tumor hypoxia.

Figure 1. Low-dose intratumor (IT) GM-CSF reduces tumor hypoxia.

(A) Schematic of GM-CSF IT treatment regimen and methodology for measurement of in vivo tumor oxygen with oxygen-sensing trityl radicals by Electron Paramagnetic Resonance (EPR) Oximetry. PyMT cells were orthotopically implanted into WT FVB/NJ female mice and allowed to grow to a minimum of 50 mm³ volume. 15ng GM-CSF, 5ng GM-CSF, or saline was administered IT by an insulin syringe 3x per week for 3 weeks total. After 3 weeks of IT GM-CSF therapy, mice were anesthetized, injected intratumorally with oxygen-sensing soluble trityl radical, and subjected to L-Band Electron Paramagnetic Resonance (EPR) Spectroscopy to measure in vivo tumor oxygen. (B) Average tumor oxygenation in PyMT tumor-bearing mice measured by L-Band EPR after 3ks of 5ng or 15ng IT GM-CSF. Saline n=6, 15ng GM-CSF n=8, 5ng GM-CSF n=6. Boxes represent the median (line) of tumor oxygen in mmHg ± SD. (C) Tumor burden of mice in B after sacrifice. Error bars = ± SEM. (D) Correlations between average tumor oxygen (measured by L-Band EPR) and tumor volumes (measured by caliper) in mice after 3ks of 5ng (low-dose) IT GM-CSF or IT saline. Spearman’s correlation test was utilized to generate R² and p-values. Saline n=11, 5 g (low-dose) GM-CSF n=10. (E) Representative Rapid-Scan EPR Imaging (RS-EPRI) of tumors after 3ks of low-dose GM-CSF or saline alongside a normal mammary control (left) and histogram distributions of whole tumor voxel oxygen values detected (middle). RS-EPRI allows for 3D spatial mapping of tumor oxygen distribution throughout the tumor tissue. Each graph has a corresponding key for oxygen partial pressure (mmHg) indicated by colors –images depict a shift in coloration from relatively dark blue (less oxygen) in saline to light blue/green (more oxygen) in low-dose GM-CSF. Dotted lines indicate tumor volume measured by caliper. Quantification of RS-EPRI data (right). Saline n= 8, low-dose GM-CSF n=8, normal mammary n=4. Error bars: ± SEM. *p<0.05, **p<0.01 by One-Way ANOVA and Tukey’s post-doc correction. (F) Representative images of HIF-1α staining of tumors after 3wk of low-dose GM-CSF or saline by IHC (left) and image staining quantification (right). Saline n=7, low-dose GM-CSF n=5. Error bars: ± SEM. *p<0.05 by Mann-Whitney U-Test. (G) Representative RS-EPRI of PyMT tumors treated with high-dose (100ng) GM-CSF vs. saline after 3wk (left) and histogram (middle). Each graph has a corresponding key for oxygen partial pressure indicated by relative colors. Dotted lines indicate tumor volume measured by caliper. Quantification of RS-EPRI data (right). Saline n=4, high-dose (100ng) GM-CSF n=6. ns= not significant by student’s unpaired t-test. NOTE: For E and G, x axis scales differ across conditions, and colors correspond to relative oxygen levels in the scale bar specific to each individual image.

Low-dose GM-CSF normalizes tumor vasculature by increasing NG2⁺ pericyte coverage on CD31⁺ tumor vessels

Next, we studied how low-dose GM-CSF influences tumor vasculature. Previous studies showed that high-dose GM-CSF ablated tumor blood vessels and worsened hypoxia by sVEGFR-1 sequestration of free tumor VEGF (21). We hypothesized that low-dose GM-CSF normalized tumor vasculature for better oxygen delivery. While VEGF binding to VEGFR-1 on NG2⁺ pericytes lead to their migration away from stable vessels to augment dysfunctional angiogenesis (42), adequate doses of anti-VEGF therapy may promote vessel normalization by improving vessel pericyte coverage for improved stability and function (5). Therefore, after 3wks of low-dose GM-CSF therapy in PyMT tumors - the same timepoint in which we observed significant increases in oxygen above - we utilized FL-IHC on tumor sections to analyze tumor blood vessel density, perfusion, and pericyte coverage. We found that low-dose GM-CSF resulted in increased NG2⁺ pericyte co-localization with CD31⁺ tumor vessels without changing pericyte abundance, indicating improved pericyte coverage around some portions of tumor vasculature (Fig 2A). We also observed NG2⁺ pericytes had moderately increased co-staining with fluorescently labeled tomato lectin after low-dose GM-CSF treatment (Fig 2B), and NG2⁺ CD31⁺ dual-stained regions exhibiting lectin perfusion were also modestly increased (Fig S1B), suggesting that blood perfusion is improved specifically in pericyte-covered tumor vasculature. However, we did not observe significant changes in CD31⁺ density or CD31⁺ perfusion (Fig S1C–D), suggesting that low-dose GM-CSF induced stabilization of existing vasculature via modulation of pericytes rather than a reduction of overall endothelial cells within the TME. Overall, these data imply that low-dose GM-CSF induces vascular normalization via improved pericyte coverage around existing tumor blood vessels to improve vessel health independent of CD31 modulation.

Figure 2. Low-dose IT GM-CSF normalizes tumor vasculature by increasing NG2⁺ pericyte coverage on CD31⁺ tumor vessels.

(A) PyMT cells were orthotopically implanted into WT FVB/NJ female mice and allowed to grow to a minimum of 50 mm³ volume. After 3wk of low-dose IT GM-CSF or IT saline administration, tumors were harvested, sectioned, and immunostained with anti-CD31 antibody (green) and anti-NG2 antibody (magenta) to stain endothelial cells and pericytes, respectively. Pericyte-covered vessels are indicated by CD31⁺ NG2⁺ overlay (white on far-right images). Low-dose GM-CSF treated-tumors have significantly increased co-localized CD31⁺ NG2⁺ cells (*p<0.05) while overall NG2⁺ pericyte abundance is unchanged. Images were taken non-serially at 3 images per half of tumor, and 6 total images were analyzed per n for n=11 saline, and n=9 low-dose GM-CSF for CD31⁺NG2⁺ ; n=10 for NG2⁺ only. Results represent mean pixel count ± SEM and were analyzed by Mann-Whitney U-Test. Scale bar = 50 μm. (B) PyMT cells were orthotopically implanted into WT FVB/NJ female mice and allowed to grow to a minimum of 50 mm³ volume. After 3wk of low-dose IT GM-CSF or IT saline administration, mice were given retroorbital injection of fluorescently labeled tomato lectin (green), which labels perfused blood vessels. Tumors were harvested, sectioned, and immunostained with anti-NG2 antibody (magenta) for pericytes to indicate perfusion of pericyte-containing regions (Lectin⁺ CD31⁺ overlay in white on far-right images). Low-dose GM-CSF treated-tumors have moderately increased co-localized Lectin⁺ NG2⁺ regions (p=0.1145). Two images were taken non-serially per one half of tumor, and 2 total images were analyzed per n for n=12 saline, and n=7 low-dose GM-CSF. Results represent mean pixel count ± SEM and were analyzed by Mann-Whitney U-Test. Scale bar = 50 μm. HPF= high powered field.

Pre-treatment with low-dose GM-CSF sensitizes “cold” PyMT tumors to αPD1 immunotherapy

αPD1 immune checkpoint blockade (ICB) responses in TNBC are a mere 19% (4). PyMT tumors are reported to be notoriously immunologically “cold” (immune deserts) and resistant to αPD1, with characteristics very similar to human TNBC (43, 44). Because tumor hypoxia is recognized as a significant driver of immune-suppression and resistance to αPD1 (45), we hypothesized that pre-treatment of PyMT tumors with low-dose IT GM-CSF would improve responses to αPD1 ICB. Our model allows an approximate 5wk treatment window prior to reaching ethical endpoints in tumor size (46). Therefore, we performed a 2wk “priming phase” of low-dose GM-CSF 3x/wk (or saline controls) to first modulate TME blood vessels and oxygen as the tumors grew, followed by αPD1 for 3wks together with low-dose GM-CSF (both given 3x/wk) in a “combination phase” for 5wks total. We continued GM-CSF with αPD1 as studies report vessel modulation may be readily reversible upon discontinuation of anti-angiogenic therapy (47). Schema of study design is depicted in Fig 3A. First, we tested αPD1 resistance in the PyMT model. We observed no significant differences in tumor growth, burden, or survival in mice treated with αPD1 alone compared to IgG control to indicate a resistant phenotype, in line with other reports (44). Low-dose GM-CSF treatments alone also had no significant differences in these parameters over the extended 5wk period for tumor growth, final burden, or survival (Fig 3B). Next, we compared GM-CSF at low-dose and high-dose in combination with αPD1. While tumor volume remained unchanged throughout low-dose and high-dose priming, low-dose GM-CSF with αPD1 therapy resulted in significant reduction of PyMT tumor volume and burden compared to saline with αPD1 control (Fig 3C). However, high-dose GM-CSF with αPD1 tumors had no effect, suggesting that GM-CSF dosing matters for improved sensitization to ICB. We also tested whether low-dose GM-CSF could improve PyMT response to doxorubicin chemotherapy (DOX), a standard-of-care treatment in TNBC which works independently of immune activation. While tumors were sensitized to DOX alone, low-dose GM-CSF had no added benefit. All mice survived to endpoint with no survival differences between groups. These data show that priming with low-dose GM-CSF, but not high-dose, sensitizes “cold” PyMT tumors to ICB, which correlates with their effects in reducing and exacerbating hypoxia, respectively. Furthermore, we demonstrated that low-dose GM-CSF promotes the efficacy of αPD1 but not doxorubicin on PyMT tumors, suggesting that low-dose GM-CSF possibly improves immunotherapy via both elevation of tumor oxygen and its actions on immune cells.

Figure 3. Pre-treatment with low-dose IT GM-CSF sensitizes “cold” PyMT tumors to αPD1 immunotherapy.

(A) Schema of “priming” and “combination” study design. (B) Orthotopic PyMT tumors were allowed to reach 50mm³ in volume. Then, as indicated as in schema (A), mice received low-dose IT GM-CSF or IT saline 3x/wk for a 2-wk priming phase. After 2wk of priming (measure day 5), mice received αPD1 or IgG antibodies intraperitoneally (IP) 3x/wk together with continued IT administration of low-dose GM-CSF or saline 3x/wk for 3 wk total of combination. Dotted lines indicate start of αPD1 or doxorubicin (DOX). Tumor volumes were measured 2–3x per wk for a total of 12 measure days over 5 wks of study. Mice were monitored for morbidity/survival and tumors weighed for burden after sacrifice. Tumor growth was analyzed by mixed-effects modeling and were not significant. In this study, some mice reached early removal criteria (2 in Saline+IgG with last measure days 8 and 9, 1 in saline+ αPD1 with last measure day 11, and 2 in low-dose GM-CSF+IgG with last measure day 11 and 8). Growth of early removal tumors was mathematically predicted by using growth slopes, and these values are depicted in the tumor growth curve for visual representation to prevent an artificial drop in the growth curve. For statistical analyses (mixed-effects modeling), these predicted values were not included. Mice that died a natural death during the study (2 in Saline+IgG, 1 in low-dose GM-CSF+IgG) were pre-determined to be excluded from survival analyses but not tumor volume data. Tumor burdens were analyzed by one-way ANOVA and Tukey’s post-hoc correction, tumor growth was analyzed by mixed-effects modeling, and survival was analyzed by Kaplan-Meyer test. No statistical differences were found. Saline+IgG n=8, Saline+αPD1 n=9, low-dose GM-CSF+IgG n=9. (C) Orthotopic PyMT tumors were allowed to reach 50mm³ in volume. After, mice received dose-specific IT GM-CSF or IT saline 3x/wk for a 2-wk priming phase. After 2wk of priming (measure day 5), mice received αPD1 IP 3x/wk, or DOX (doxorubicin chemotherapy) 1x/wk, together with continued IT administration of low-dose GM-CSF or saline 3x/wk for 3 wk total of combination. Dotted lines indicate start of αPD1 or DOX. Tumor volumes were measured 2–3x per wk for a total of 13 measurement days over 5 wks of study. Mice were monitored for morbidity/survival and tumors were weighed for burden after sacrifice. For this study, all mice survived to the pre-determined endpoint of 5 study wks, so there were no survival differences. Low-dose GM-CSF+DOX n=7, Saline+DOX n=6, low-dose GM-CSF+αPD1 n=8, Saline+αPD1 n=7, high-dose GM-CSF+αPD1 n=8. Mice that were found dead naturally (3 in low-dose GM-CSF+DOX, 1 in Saline+DOX) were pre-determined to be excluded from survival study but included in tumor volume data. For tumor volumes, each point = average tumor volume ± SEM. For tumor burdens, error bars = ± SEM. Tumor growth was analyzed by mixed-effects modeling. Tumor burdens were analyzed by one-way ANOVA and Tukey’s post-hoc correction. *p<0.05; **p=<0.01.

TAM have reduced hypoxic, inflammatory transcriptional programs after low-dose GM-CSF therapy

We next sought to better understand how low-dose GM-CSF increases αPD1 efficacy in PyMT tumors. First, we determined cell types within TME that would directly respond to GM-CSF. While myeloid cells are recognized as significant responders to GM-CSF, some studies have reported aberrant expression of GM-CSF receptor on cells within TME such as endothelial cells and tumor cells, in which GM-CSF has been shown to stimulate their migration and proliferation (14, 48). We confirmed that myeloid cells highly express GM-CSF Receptor α, the essential component of the GM-CSF receptor α/β complex, whereas the expression level in tumor cells, endothelial cells, and T cells was very low, indicating that our observed effects of GM-CSF are most likely mediated through myeloid cells (Fig 4A).

Figure 4. TAM have reduced hypoxic, inflammatory transcriptional programs after low-dose IT GM-CSF priming.

(A) Untreated PyMT orthotopic tumors were grown for approximately 4 wks to allow for sufficient cell numbers for analysis and were then harvested, enzymatically digested, stained for CD45, TCRβ, CD31, and GM-CSFRα, fixed in 2% PFA, and analyzed by flow cytometry the following day. GM-CSFRα was predominantly expressed by general leukocytes (CD45⁺), particularly myeloid cells (CD45⁺ TCRβ⁻), and not in T cells (CD45⁺ TCRβ⁺), tumor cells (CD45⁻ TCRβ⁻ CD31⁻ ), or endothelial cells (CD45⁻ TRCβ⁻ CD31⁺). Data is representative of n=4 tumors. (B) Pre-sorting gating strategy and representative post-sort purity assessment of 98% for CD45⁺ CD11b⁺ F4/80⁺ TAMs isolated from PyMT tumors for RNA-Seq analyses after 3wk of low-dose IT GM-CSF or IT saline administration. (C) Gene Set Enrichment Analyses (GSEA) of primary isolated TAM expressed genes sorted by fold change of expression (Low-dose GM-CSF/Saline, n=3 per treatment group) from high (red) to low (blue) against selected MSigDB hallmark gene sets (vertical bars): Hypoxia, Inflammatory response, IFN-γ response, IL2 STAT5 Signaling, IL6 JAK STAT3 signaling, TNF-α signaling, and OXPHOS. NES: normalized enrichment score.

GM-CSF has been explored as an adjuvant to immunotherapies and cancer vaccines due to stimulation of antigen presenting cells such as macrophages, and GM-CSF has been reported to drive “M1-like” macrophages in vitro (19). Understanding GM-CSF response on TAM is of particular importance in the breast cancer setting, where TAMs comprise up to 50% of the tumor mass (7). We hypothesized that low-dose GM-CSF, in addition to normalizing blood vessels to reduce hypoxia (data above) through modulation of TAM sVEGFR-1 (20), might also drive an immunostimulatory macrophage phenotype for a unique “double punch effect” on TAM for improved αPD1 efficacy. To test this, we performed RNA sequencing on CD45⁺ CD11b⁺ F4/80⁺ TAMs isolated from PyMT tumors after 3wk of low-dose GM-CSF or saline controls and assessed transcriptional profiles. Gating strategies and post-sort purity analyses are depicted in Fig 4B. Gene Set Enrichment Analyses revealed that TAMs from the low-dose GM-CSF cohort were significantly less enriched in genes for hypoxia coupled with higher oxidative metabolism, validating reduction of tumor hypoxia we observed by EPR and HIF-1α staining (Fig 4C). However, no differentially expressed genes were identified in low-dose GM-CSF vs saline, indicating that low-dose GM-CSF may not perturb overall TAM phenotype (Fig S2A). Intriguingly, low-dose GM-CSF TAMs had significantly less enrichment for gene sets related to inflammatory response and inflammatory signaling pathways, such as IFNγ, IL-2/STAT-5, IL-6/JAK/STAT3, and TNFα responses, suggesting that low-dose GM-CSF did not have inflammatory effects on TAM but rather quieted aberrant inflammatory TAM signaling. These data imply that TAMs from low-dose GM-CSF-treated tumors experienced less hypoxia and aberrant inflammatory signaling within the TME.

IT GM-CSF priming at high-dose, but not low-dose, GM-CSF increases TAM in PyMT tumors

GM-CSF has been shown to exert immunostimulatory effects on myeloid cells in some instances, while others yield little to no benefit or even expansion of immunosuppressive myeloid populations (14, 17). We hypothesized that low-dose and high-dose GM-CSF treatments have differential modulating effects on TAMs. We determined the phenotypes of TAMs and other myeloid cells within PyMT tumors after 3wks of low-dose or high-dose IT GM-CSF administration. As expected, GM-CSF administration alone did not affect tumor burden (Fig S2B). We found that high-dose GM-CSF administration increased the percent of CD11b⁺ myeloid cells, including the CD11b⁺ F4/80⁺ TAM, within the TME (Fig 5A). This result was in line with our previous findings showing that high dose GM-CSF increased F4/80⁺ cells as determined by IHC (21). In contrast, administration of low-dose GM-CSF did not have the same effect. The increase of TAMs following high-dose GM-CSF administration could be due to increased infiltration of Ly6C⁺ CCR2⁺ monocytes, which further differentiate into CCR2⁺ CX3CR1⁺ then CCR2⁻ CX3CR1⁺ TAMs within TME, or the expansion of FLRβ⁺ mammary tissue-resident macrophages. FMO controls used for gating are shown in Fig S2C. We found an increase of Ly6C⁺ differentiating monocytes within the TME as indicated by expression of FCγR1⁺ (Fig 5B), which coincided with the reduction of CCR2⁺ CX3CR1⁺ and a corresponding increase of CCR2⁻ CX3CR1⁺ TAMs (Fig 5C). In contrast, percentages of FLRβ⁺ tissue-resident (Fig 5D) and Ki67⁺ proliferating macrophages (Fig 5E) were comparable with or without GM-CSF administration. Thus, our data demonstrate that high-dose GM-CSF stimulates the recruitment of CCR2⁺ monocytes from the periphery that transitioned into CX₃CR1⁺ tumor-residing macrophages rather than the expansion of tissue-resident macrophages.

Figure 5. IT GM-CSF priming at high-dose, but not low-dose, increases TAM in PyMT tumors.

PyMT cells were orthotopically implanted into WT FVB/NJ female mice and allowed to grow to 50mm³ volume. After 3wk of low-dose GM-CSF, high-dose GM-CSF, or saline treatment, tumors were harvested, homogenized, and enriched for CD45⁺ cells by magnetic separation. Cells were stained for TAM and various phenotypic markers, fixed and analyzed by flow cytometry the following day unless otherwise specified. (A) CD11b⁺ F4/80⁺ TAM and CD11b⁺ F4/80^neg myeloid cells, unfixed, representative gating strategy of CD11b/F4/80 on Live CD45⁺ cells (left) and quantification (right); (B) Ly6C⁺ FcγRI⁺ differentiating monocytes, representative gating strategy of Ly6C/FcγRI on Live CD45⁺ CD11b⁺ Ly6G^neg cells (left) and quantification (right); (C) CCR2⁺ CX3CR1⁺ TAM and CCR2^neg CX3CR1⁺ TAM, representative gating strategy and CCR2 and CX3CR1 in Live CD45⁺ CD11b⁺ F4/80⁺ TAM (left) and quantification (right); (D) FLRβ⁺ TAM, unfixed cells, representative gating strategy for FLRβ on Live CD45⁺ CD11b⁺ F4/80⁺ (left) and quantification (right); and (E) Ki67⁺ proliferating TAM, representative gating strategy of Ki67 in Live CD45⁺ CD11b⁺ F4/80⁺ TAM (left) and quantification (right). For (A, D): n=7 Saline, n=7 low-dose GM-CSF, n=5 high-dose GM-CSF. For (B, C, E): n=3 Saline, n=6 low-dose GM-CSF, n=4 high-dose GM-CSF. Due to limitations of the number of tumors able to be processed and immunostained in a single day, these data represent 1 independent experiment of 4–5 staggered cohorts of 3–5 tumors per day prepared and analyzed on different days combined together. Error bars = ± SEM. Significance was determined by One-Way ANOVA and Tukey’s post-hoc correction. *p<0.05, **p<0.01.

IT GM-CSF priming has disparate effects on TAM phenotype in “hot” 4T1 vs. “cold” PyMT tumors

Next, we asked whether GM-CSF administration alters the phenotype of TAMs. Given that only low-dose and not high-dose GM-CSF priming sensitized PyMT tumors to αPD1 therapy, we hypothesized that the recruited, monocyte-derived TAMs following high-dose GM-CSF administration were immunosuppressive. Previous reports showed that expression of several markers, including Arg1, PD1, CD206, correlates with the immunosuppressive phenotype of TAMs (49–51), however GM-CSF did not alter these markers on TAM (Fig S2D). Strikingly, we found that GM-CSF administration failed to increase MHC II, CD80, and CD86 expression on TAMs, even with 3wks of high-dose treatments (Fig 6A–B). In fact, high-dose GM-CSF TAMs even approached near statistical decreases in MHCII⁺ (p=0.056) and MHCII^hi phenotypes (p=0.085) vs. saline controls (Fig 6A). These results were unexpected because GM-CSF is well-known to promote the function of antigen-presenting cells through the upregulation of MHC II and/or costimulatory molecules which make it a potential immunotherapy adjuvant (11). Therefore, we further asked whether this phenotype is due to a general the lack of responsiveness of breast cancers to high-dose GM-CSF, or if it is PyMT “cold” tumor specific. An analyses of gene expression profiles across various tumor types classified 4T1 tumors as highly immunogenic due to up-regulation immune-activation genes (52) and they have been shown to mildly respond to αPD1 alone (53, 54). Thus, we characterize this breast cancer model as “hot” relative to the notorious “cold” nature of PyMT. To investigate this, we administered GM-CSF to 4T1 tumors and compared with those observed in PyMT tumors. As expected, 3wk IT GM-CSF alone did not alter 4T1 tumor burden (Fig S2E). Similar to the effect in PyMT tumors, administration of high-dose GM-CSF was trending toward increased percent of CD11b⁺ F4/80⁺ TAMs in 4T1 (p=0.228 due to 1 out of 7 outlier in treatment group) (Fig 6C). In contrast to what we found in PyMT tumors, the percent of FLRβ⁺ TAMs in 4T1 tumors were also increased, suggesting an expansion of tissue-resident macrophage population (Fig 6D). Importantly, high-dose GM-CSF administration significantly increased the percent of TAMs that express high level of MHC II (Fig 6E), and the expression of CD80 and CD86 were also moderately increased (p=0.109 vs. low-dose GM-CSF) (Fig 6F). However, similar to what we observed in PyMT tumors, low-dose GM-CSF had no effect. These data suggest that the “cold” PyMT and the “hot” 4T1 tumors responded to high-dose GM-CSF treatment differently, where in 4T1 tumors high-dose GM-CSF treatments drove TAMs toward M1-like phenotype.

Figure 6. IT GM-CSF priming has disparate effects on TAM phenotype in “hot” 4T1 vs. “cold” PyMT tumors.

PyMT or 4T1 cells were orthotopically implanted into WT FVB/NJ female mice and allowed to grow to 50 mm³ volume. After 3wk of low-dose GM-CSF, high-dose GM-CSF, or saline treatment, tumors were harvested, homogenized, and enriched for CD45⁺ cells by magnetic separation. Cells were stained for TAM and various phenotypic markers, fixed and analyzed by flow cytometry the following day unless otherwise specified. TAMs were identified as Live CD45⁺ CD11b⁺ F4/80⁺. (A) PyMT TAM MHCII^hi and MHCII⁺ unfixed, representative gating strategy for MHCII on Live CD45⁺ CD11b⁺ F4/80⁺ TAM (left) and quantification (right). (B) PyMT TAM CD80 and CD86, representative gating strategy for CD80⁺CD86⁺ on Live CD45⁺ CD11b⁺ F4/80⁺ TAM (left) and quantification (right). (C) 4T1 TAM and CD11b⁺ F4/80^neg myeloid cells, unfixed, representative gating strategy for CD11b/F4/80 on Live CD45⁺ CD11b⁺ F4/80⁺ TAM (left) and quantification (right). (D) 4T1 FLRβ TAM; unfixed, representative gating strategy for FLRβ on Live CD45⁺ CD11b⁺ F4/80⁺ (left) and quantification (right). (E) 4T1 TAM MHCII⁺ and MHCII^hi; unfixed, representative gating strategy for CD11b/F4/80 on Live CD45⁺ CD11b⁺ F4/80⁺ TAM (left) and quantification (right). (F) 4T1 TAM CD80 and CD86, representative gating strategy for CD80⁺CD86⁺ on Live CD45⁺ CD11b⁺ F4/80⁺ TAM (left) and quantification (right). For (A, B): Saline n=7, low-dose GM-CSF n=7, high-dose GM-CSF n=5. A: 1 statistical outlier in low-dose GM-CSF was removed. For (C-F): Saline n=5, low-dose GM-CSF n=5, high-dose GM-CSF n=7. E: 1 statistical outlier in saline was removed. Due to limitations of the number of tumors able to be processed and stained in a single day, these data represent 1 independent experiment of 4–5 staggered cohorts of 3–5 tumors per day prepared and analyzed on different days combined together. Error bars = ± SEM. Significance was determined by One-Way ANOVA and Tukey’s post-hoc correction. *p<0.05, **p<0.01.

Effect of GM-CSF priming on “cold” PyMT tumors is independent of the modulation of tumor-infiltrating lymphocytes (TILs)

Higher CD8⁺ TIL infiltration is associated with improved ICB in breast and other cancers (55). Reduction of hypoxia via vessel normalization have also been shown to promote TIL infiltration and effector function (56). We aimed to determine whether GM-CSF priming induced alterations in CD8⁺ TIL infiltration or phenotype based on dose and/or tumor type. We utilized flow cytometry to compare CD8⁺ TIL after 3wks of low-dose or high-dose IT GM-CSF in PyMT and 4T1 tumors. We found that CD8⁺ TILs isolated from 3 out of 7 4T1 tumors responded to high-dose GM-CSF treatments (responders), with significantly heightened production of granzyme B (p=0.0056), IFNγ/TNFα (p= 0.0021), and IL-2 (p=0.0014) compared to tumors administered saline, while low-dose GM-CSF treatments did not have these effects (Fig 7A, significant responders indicated by white square symbols). Correspondingly, the high-dose GM-CSF responders (indicated by white square symbols) had increased PD1^hiTIM3⁺ antigen experienced/exhausted phenotype (p=0.0050) (Fig 7B). In contrast, neither high- or low-dose GM-CSF affected CD8⁺ TIL phenotypes in PyMT tumors (Fig 7C–D). Correlations of these observed populations with PyMT and 4T1 tumor burdens amongst each treatment group suggest that increased proportion of IFNγ⁺TNFα⁺, GzmB⁺, and PD1^hiTIM3⁺ CD8⁺ TIL populations correlated with decreased tumor burden after 3 wk of low-dose GM-CSF monotherapy uniquely in PyMT tumors (Fig S3A–B). However, these data overall further support our notion that improved αPD1 sensitivity in PyMT tumors was not through substantial GM-CSF-mediated immune stimulation. Further, neither PyMT nor 4T1 tumors experienced increased infiltration of overall CD45⁺ cells, CD45⁺TCRβ⁺ TIL, CD4⁺ TIL, CD8⁺ TIL, or CD4⁺Foxp3⁺ Tregs in response to GM-CSF dosages (Fig 7E, Fig S3C–D), indicating that GM-CSF was not altering the abundance of immune supportive vs. immune suppressive TIL populations. Together, these data suggest that high-dose GM-CSF drove partial increases in CD8⁺ T cell polyfunctionality in 4T1 tumors, but substantial effects on TIL infiltration/phenotype were not observed in response to GM-CSF priming.

Figure 7. Effect of IT GM-CSF priming on “cold” PyMT tumors is independent of modulation of tumor-infiltrating lymphocytes (TILs).

(A) From the tumor-bearing mice described in Figure 6, some 4T1 tumors were harvested, homogenized, and enriched for CD45⁺ cells by magnetic separation. Cells were stimulated ex vivo with PMA and ionomycin for 4hr prior to antibody staining and flow cytometry to assess CD8⁺ TIL intracellular IFNγ/TNFα and IL-2. Intracellular granzyme B was assessed in unstimulated 4T1 CD8 T cells. Representative gating strategies on Live CD45⁺ TCRβ⁺ CD8⁺ TIL and corresponding quantification are shown for each population. For high-dose GM-CSF, the white squares indicate matched tumor samples that significantly responded to ex vivo stimulation vs saline controls. (B) PD1 and TIM3 expression in 4T1 Live CD45⁺ TCRβ⁺ CD8⁺ TIL. Gating strategy (left) and quantification (right) of PD1^hiTIM3⁺ CD8⁺ TIL. For high-dose GM-CSF, the white squares indicate matched tumor samples (same as in A that responded to ex vivo stimulation) that also yielded significant increases in PD1^hiTIM3⁺. (C) Assessment of PyMT CD8⁺ TIL intracellular IFNγ/TNFα, IL-2, and granzyme B as in A. (D) PD1⁺TIM3^hi CD8 TIL in PyMT tumors as in B. Gating strategy (left) and quantification (right). (E) Assessment of CD4, CD8, and CD4⁺ Foxp3⁺ regulatory T cell proportions in 4T1 (top) and PyMT (bottom) tumors. Representative gating strategies and corresponding quantification are shown for each population. For 4T1 studies (A-B, E top): Saline n=5, low-dose GM-CSF n=5, high-dose GM-CSF n=7. One statistical outlier was removed from IFNγ⁺TNFα⁺. For PyMT studies (C-D, E bottom): Saline n=7, low-dose GM-CSF n=7, high-dose GM-CSF n=5. Due to limitations of the number of tumors able to be processed and stained in a single day, these data represent 1 independent experiment of 4–5 staggered cohorts of 3–5 tumors per day prepared and analyzed on different days combined together. Error bars = ± SEM. Significance was determined by One-Way ANOVA and Tukey’s post-hoc correction. *p<0.05, **p<0.01.

DISCUSSION

In summary, our data demonstrate a dose-specific effect of GM-CSF in sensitizing “cold” PyMT tumors to αPD1 therapy: low-dose GM-CSF had minimal effect on infiltration, expansion, and phenotypical change of TAMs or TILs, but alleviated tumor hypoxia through vessel normalization, thereby sensitizing PyMT tumors to αPD1 therapy. In contrast, even though high-dose GM-CSF increased infiltrating monocyte-derived TAMs, it failed to upregulate MHC II and co-stimulatory molecules in these cells. In addition, high-dose GM-CSF treatment did not reduce tumor hypoxia and failed to sensitize PyMT tumors to αPD1 therapy. Interestingly, in αPD1-responsive 4T1 tumors, high-dose GM-CSF did upregulate TAM MHC II and drove immunostimulatory phenotype which led to moderately improved CD8 TIL polyfunctionality not observed in PyMT tumors. However, GM-CSF alone did not substantially modulate TIL infiltration at either dose or tumor type overall, suggesting primary effects are mediated through myeloid cells. Thus, this study provides evidence that the dosage of, and setting for, GM-CSF as an immunotherapy adjuvant for cancer treatments must be carefully determined.

GM-CSF has been the subject of a high volume of clinical testing as an adjuvant to cancer vaccines or ICB. However, dosage appears to be a critical factor for whether GM-CSF exerts pro- or anti-tumor effects. Tumor cell-derived GM-CSF at pg/mL level induced suppressive myeloid cell populations that suppressed CD8⁺ T cells (57, 58). Other studies report that exogenous high-dose GM-CSF drove immunosuppressive effects due to recruitment of immunosuppressive myeloid populations, while repeated administration of low-dose GM-CSF initiated anti-tumor responses or had little effect (17). In breast cancer, a GM-CSF-secreting tumor cell vaccine induced antigen-specific CD8⁺ T cell responses and improved overall survival (59). While GM-CSF shows promise in some settings, inconsistencies and unknown mechanisms have remained a longstanding issue. Regardless, autologous GM-CSF-secreting tumor vaccines are currently being tested in breast cancer (15). In our study, we observed that high-dose GM-CSF had two key effects on PyMT tumors: 1) an increase in proportion of TAM which notably lacked M1-like polarization, and 2) exacerbated breast tumor hypoxia. The presence of TAMs (8) or hypoxia (1), each correlate with worse prognosis in breast cancer patients, and hypoxia supports TAM differentiation, retention, and pro-angiogenic activity (60). Thus, these data elucidate a detrimental “double-punch” effect of high-dose GM-CSF on the TME that may limit ICB as we observed in our study.

To our knowledge, this is the first study suggesting that GM-CSF can serve as a vessel normalizing agent for reducing tumor hypoxia. The premise of this work was built on previous publications from our group in which intratumor administration of high-dose (100ng) GM-CSF for 3wks ablated vasculature and worsened tumor hypoxia in PyMT breast cancer, in vivo (21). The effects were driven by GM-CSF induction of sVEGFR-1 from monocytes/TAMs which sequestered free VEGF in the TME, as GM-CSF + αsVEGFR mAb abrogated the effects. However, we also found that GM-CSF stimulated sVEGFR-1 production in a dose-dependent manner, in which 100ng stimulated monocytes/macrophages to produce abundant sVEGFR-1, while the 1–10 g dosage range produced substantially less (20). We previously chose 100ng GM-CSF for intratumor administration as the focus of the field at that time was to deprive the tumor of oxygen, which has since yielded ineffective results due to potentiation of hypoxia and therapeutic resistance. We now show that GM-CSF has a dose-specific effect on tumor oxygen, where GM-CSF maintains or exacerbates hypoxia at high dosages sustaining suppression of anti-tumor immunity while lower doses reduce tumor hypoxia for improved responses. This is supported by our current data demonstrating that high-dose GM-CSF had no effect on αPD1 response, but low-dose GM-CSF reduces tumor hypoxia, normalizes tumor vasculature, and sensitizes tumors to αPD1 treatments. Our observation may help explain the “dosing matters” puzzle of GM-CSF as adjuvant of immunotherapy.

To investigate whole tumor oxygen longitudinally and in real-time with minimal invasiveness, we employed highly sensitive Rapid Scan EPR imaging technology to directly measure tumor oxygen distribution, in vivo. Other methodologies used for assessing hypoxia include pimonidazole staining, although this is an indirect measure of oxygen which is less sensitive, as it only detects <10mmHg and below and unable to detect changes between 1% and 3% oxygen tension (61), requires animal euthanasia, and can only assess hypoxia at microns thick sections. RS-EPR imaging allows us to directly measure and quantify 3D tumor oxygen distribution, which is important given the heterogenous and dynamic nature of oxygen distribution in tumors, while the animal is alive and the tumor is still in-tact.

Our study is unique from other GM-CSF studies in combination with ICB due not only to the dose comparison and correlation of vasculature/oxygen with anti-tumor effects, but also due to our strategy of pre-treating established tumors (~50mm³+ in volume) with dose-optimized GM-CSF aimed to re-shape the tumor microenvironment and “set the stage” for effective αPD1. Despite reduced hypoxia and improved αPD1 after low-dose GM-CSF, we did not observe increased infiltration or stimulatory phenotype of myeloid or CD8⁺ T cells after GM-CSF alone. Other studies report vaccination with irradiated autologous GM-CSF-expressing melanoma cells resulted in increased TIL infiltration and cytotoxic responses (16, 62). It is possible that GM-CSF as a monotherapy is insufficient for driving immune stimulation at these dosages and timepoints. In one study using perioperative Sargramostim in colorectal cancer patients, elevated plasma sVEGFR-1 levels limited post-laparotomy VEGF, but no immune-mediated benefits occurred (63). Future studies are underway in our laboratory to determine optimal GM-CSF dosing for both reduced tumor hypoxia and immune stimulation. Alternatively, low-dose GM-CSF with synthetic CpG or STING agonists may drive immune stimulation in a more physioxic tumor for further benefit. We analyzed the TME after 3wk of GM-CSF because we previously performed longitudinal EPR at wks 1, 2, and 3 after high-dose GM-CSF to show PyMT oxygen levels were significantly different at 3wk only (21). However, how high-dose GM-CSF delivery may regulate TAM recruitment and phenotype over time is of interest, especially after high-dose GM-CSF where appears to have earlier infiltration of CCR2⁻CX3CR1⁺ recruited TAM that differentiate into CCR2⁺CX3CR1⁺ tumor-residing TAM. Although low-dose GM-CSF did not increase immune infiltration, hypoxia reduction/vessel repair may promote improved TAM-CD8⁺ interactions. Along these lines, we are currently investigating mechanistically how TAMs/TILs may be modulated after αPD1 in GM-CSF-primed tumors.

Our findings may bring to light oxygen-regulatory effects in finding optimal dosages and promote consistency across studies, as doses often vary substantially making results difficult to interpret. However, GM-CSF is often administered via oncolytic viruses or tumor vaccines where dosage is difficult to control. Furthermore, the benefit of the herpes simplex virus (HSV) transfected with CSF2 gene (T-VEC) vs. cytolytic effects of the virus itself is unclear as the clinical trial lacked an important intralesional control arm (64). Intratumor injections allow for controlled delivery of dose-specific GM-CSF yet are not always clinically feasible. Our group recently developed GM-CSF-loaded biodegradable polymeric nanoparticles we are currently implementing for non-invasive, tumor-targeted delivery of dose-specific GM-CSF (65). It remains unclear how our preclinical dosage scales up for humans in the clinical setting, especially given difficulties measuring tumor oxygen within human tumors in real time. Fortunately, clinical EPR systems are currently under development (66), and their implementation in informing future clinical trials with GM-CSF may serve as a prognostic indicator for effective dosing.

Immunologically “cold” tumors are characterized by poor TIL infiltration (immune desert) and lack of neoantigens which leads to limited response to ICB. “Hot” tumors have increased TIL infiltration and neoantigen diversity. Since GM-CSF did not drive immunostimulation in the “cold” PyMT, we were curious whether GM-CSF effects depended on the immunological nature of the tumor to respond. Although other studies characterize 4T1 as relatively ‘cool’ (67), 4T1 has been shown to mildly respond to αPD1 alone (53, 54), unlike PyMT (44), and was classified as highly immunogenic (52). Thus, we consider this model a “hot” TNBC model relative to PyMT. To support this notion, we observed ~40% CD4⁺ TIL in saline-treated PyMT tumors whereas saline-treated 4T1 tumors had ~70% CD4⁺ TIL. Interestingly, while we observed immunosuppressive effects of high-dose GM-CSF in PyMT tumors, 4T1 tumors experienced immunostimulatory effects including M1-like phenotypic markers on TAM and production of anti-tumor molecules IFNγ, TNFα, IL-2 by CD8 TIL from almost half of the tumors generated. These data support disparate responses to GM-CSF across cancer types (18). As TAMs are already a negative prognostic indicator and respond to GM-CSF, de novo TAM phenotype likely plays a crucial role in response. We note that while saline-treated PyMT tumors had ~80% MHCII^hi TAMs (tumor suppressive), saline-treated 4T1 tumors had only ~5% which may have implications into which direction GM-CSF signaling may function. To our knowledge, independent of cytokine administration, our study is the first to directly compare the immune profiles of both TAM and TIL in PyMT vs. 4T1 TNBC models. Although further studies may fully elucidate the reasonings for observed differences and inform prognostic indicators to predict whether high-dose GM-CSF will exert pro- or anti-tumor effects immunologically in these models, hypoxia reduction will not be achieved at such dosage based on our data. Thus, mixed immunological effects based on setting may be an added negative variable of high-dose GM-CSF. Similar to both PyMT tumors and human TNBC, 4T1 tumors are also characterized as hypoxic (68). Ongoing work in our laboratory is focused on assessing GM-CSF-induced modulation of hypoxia in the 4T1 model to determine whether differential immunological impacts of high-dose GM-CSF may be related in-part to the extent of hypoxia. Lastly, the treatment window for 4T1 tumors is significantly shorter (3–4 weeks) than PyMT tumors (5–6 weeks), which limited us to be able to do the same 5-week treatment frame in 4T1 as we performed in PyMT for the immunotherapy studies. We felt that a shorter combination study (e.g. only 1 week of GM-CSF priming before αPD1, or less αPD1 therapy timeframe) would not permit accurate model-to-model comparison in this study. Current studies in our laboratory are focused on shorter duration and timepoint studies to inform equal comparison of effects of GM-CSF with αPD1 PyMT with 4T1 in future work, as well as other tumor types.

Overall, our findings connect GM-CSF modulation of TAM and αPD1 therapy with a novel oxygen-regulatory component that, together, better informs optimal dosing when utilizing GM-CSF in future cancer immunotherapy regimens.

KEY POINTS.

• Low-dose intratumor GM-CSF reduces breast tumor hypoxia and normalizes vasculature

• Low-dose GM-CSF priming sensitizes “cold” PyMT tumors to αPD1 immunotherapy

• GM-CSF effects on TAM vary between “hot” and “cold” TNBC models

List of Abbreviations

GM-CSF

Granulocyte-macrophage colony-stimulating factor

VEGF

Vascular endothelial growth factor

sVEGFR-1

soluble Vascular endothelial growth factor-1

ICB

Immune checkpoint blockade

PD1

programmed cell death-1

IT

intratumor

TAM

tumor-associated macrophage

TIL

tumor-infiltrating lymphocyte

TNBC

Triple-negative breast cancer

DOX

doxorubicin

MMTV-PyMT

mouse mammary tumor virus-polyoma middle T antigen

TME

Tumor microenvironment

EPR

Electron Paramagnetic Resonance

RS-EPRI

Rapid-Scan EPR Imaging