Paclitaxel induces micronucleation and activates pro-inflammatory cGAS-STING signaling in triple-negative breast cancer

Abstract

Taxanes remain one of the most effective medical treatments for breast cancer. Clinical trials have coupled taxanes with immune checkpoint inhibitors in triple-negative breast cancer (TNBC) patients with promising results. However, the mechanism linking taxanes to immune activation is unclear. To determine if paclitaxel could elicit an antitumoral immune response, we sampled tumor tissues from patients with TNBC receiving weekly paclitaxel (80 mg/m²) and found increased stromal tumor-infiltrating lymphocytes and micronucleation over baseline in three of six samples. At clinically relevant concentrations, paclitaxel can induce chromosome missegregation on multipolar spindles during mitosis. Consequently, post-mitotic cells are multinucleated and contain micronuclei, which often activate cyclic GMP-AMP synthase (cGAS) and may induce a type I interferon response reliant on the stimulator of interferon genes (STING) pathway. Other microtubule-targeting agents, eribulin and vinorelbine, recapitulate this cGAS/STING response and increased the expression of immune checkpoint molecule, PD-L1, in TNBC cell lines. To test the possibility that microtubule-targeting agents sensitize tumors that express cGAS to immune checkpoint inhibitors, we identified ten TNBC patients treated with PD-L1 or PD-1, seven of whom also received microtubule-targeting agents. Elevated baseline cGAS expression significantly correlated with treatment response in patients receiving microtubule-targeting agents in combination with immune checkpoint inhibitors. Our study identifies a mechanism by which microtubule-targeting agents can potentiate an immune response in TNBC. Further, baseline cGAS expression may predict patient treatment response to therapies combining microtubule-targeting agents and immune checkpoint inhibitors.

Introduction

Paclitaxel is one of the most widely used chemotherapy agents in breast cancer (1). Triple-negative breast cancer (TNBC) is an aggressive breast cancer subtype and paclitaxel-based regimens remain the standard of care. Paclitaxel, which stabilizes polymerized microtubules, has predominantly been thought to cause mitotic arrest and cell death in a subset of the arrested population (2). More recently, paclitaxel has been shown to reach concentrations of 1–9 μM in human tumors, inducing chromosome missegregation events and leading to cell death in a portion of daughter cells (3).

The IMpassion130 clinical trial demonstrated progression-free survival benefit in metastatic and unresectable locally advanced TNBC patients expressing program death ligand-1 (PD-L1), who received PD-L1 inhibitor, atezolizumab, with nab-paclitaxel, compared to patients receiving placebo with nab-paclitaxel (4) and changed the treatment paradigm in metastatic TNBC patients. A similar clinical trial, KEYNOTE-355 also showed progression-free survival benefit for TNBC patients treated with chemotherapy regimens that included paclitaxel with PD-1 inhibitor, pembrolizumab (5). Although, these clinical trials were not designed to distinguish additive or synergistic drug effects, it remains possible that taxanes can activate the immune system.

Cyclic GMP-AMP synthase (cGAS) is a key component of innate antiviral immunity and induces a pro-inflammatory type I interferon response (6). As a cytosolic DNA sensor, cGAS binds to double-stranded DNA (dsDNA) and catalyzes synthesis of 2’3’-cGAMP, a cyclic dinucleotide that further activates downstream stimulator of interferon genes (STING). STING activates transcription of type I interferons such as interferon-beta (IFN-β). Classically, these interferons can be secreted to alert the immune system to the presence of virally infected host cells for elimination. cGAS can also be activated by nonspecific dsDNA sequences including self-DNA. For example, when one or a few chromosomes are erroneously separated from bulk DNA during mitosis, a separate but fragile nuclear envelope forms around the isolated chromosome to create a structure known as a ‘micronucleus’. Such micronuclei are prone to rupture, exposing enclosed DNA to cytoplasm and eliciting cGAS-STING signaling, which can sometimes result in the production of type I interferons (7–9). These interferons can upregulate immune checkpoint molecules such as PD-L1 on cancer cells, providing a rationale for combining paclitaxel and potentially other microtubule-targeting agents (MTAs) with immune checkpoint inhibitors (ICIs) (10). In addition, both 2’3’-cGAMP and type I interferons can recruit M1 macrophages and Batf3 dendritic cells, which prime effector T-cells for antitumoral cytoxicity (11). Further, 2’3’-cGAMP can recruit and activate immune cells including CD8+ T-cells, which can form a component of stromal tumor-infiltrating lymphocytes (sTILs) and is also associated with improved survival in TNBC patients (12,13).

Here, we investigate a mechanism by which paclitaxel and other MTAs can potentiate ICIs and identify a predictive biomarker for these combination therapies in TNBC patients. Paclitaxel and other MTAs can trigger a pro-inflammatory immune response potentially via micronuclei generated from aberrant mitoses. Ruptured micronuclei can permit cGAS entry, 2’3’-cGAMP synthesis, STING signaling and production of type I interferons in TNBCs expressing cGAS. Elevated baseline cGAS expression correlates with treatment response in patients receiving MTAs in combination with ICIs. Our findings suggest that cGAS expression can predict patient populations that are likely to benefit from MTAs and ICIs.

Materials and Methods

Cell lines and culture conditions

All cell lines were either directly purchased from ATCC/NCI or were gifted from other research groups. Cell lines gifted from other research groups were validated using polymorphic short tandem repeat loci throughout 2019 through the Small Molecules Screening Facility at UW-Madison. Mycoplasma contamination was regularly monitored using the R&D Systems Mycoprobe Mycoplasma Detection Kit. All cell lines were maintained at 37°C and 5% CO2 in a humidified incubator in growth media supplemented with 10% fetal bovine serum, 100 units/mL penicillin-streptomycin and 1X plasmocin prophylactic. MDA-MB-231, BT-549 and MDA-MB-468 were obtained from NCI. MDA-MB-436 cells were obtained from ATCC. MDA-MB-453, Hs578T and HCC1806 were gifts from Ruth O’Regan. THP-1 was a gift from David Beebe. MDA-MB-231 stably expressing RFP-H2B and GFP-Tubulin was a gift from Beth Weaver. All triple negative breast cancer cell lines were propagated in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 4.0 mM L-Glutamine, 4500 mg/L glucose and 1X plasmocin prophylactic. THP-1 was propagated in RPMI 1640 medium with 2 mM L-glutamine, 1.5 g/L sodium bicarbonate, 4.5 g/L glucose, 10 mM HEPES, 1.0 mM sodium pyruvate, 1X glutamax and 1X plasmocin prophylactic.

Reagents and antibodies

Reagents used in this study were Paclitaxel/Taxol (Tocris 1097), Phorbol 12-myristate 13-acetate/PMA (Sigma P8139), digitonin (Sigma D141), Cyclic [G(2’,5’)pA(3’,5’)p]/2’3’-cGAMP (Invitrogen tlrl-nacga23) and recombinant human IFN-γ (BD Biosciences 554616). The following primary antibodies were obtained from Cell Signaling Technologies: cGAS D1D3G Rabbit mAb #15102 (immunoblot 1:1000 dilution, immunofluorescence 1:200 dilution); STING D2P2F Rabbit mAb # 13647 (immunoblot 1:1000 dilution, immunofluorescence 1:200 dilution); Phospho-IRF-3 (Ser396) (4D4G) Rabbit mAb #4947 (immunoblot 1:1000 dilution); Phospho-Stat1 (Tyr701) (58D6) Rabbit mAb #9167 (immunoblot 1:1000 dilution), Pan-keratin (C11) Mouse mAB #4545 (IHC 1:100 dilution), PD-L1 Extracellular Domain Specific D8T4X Rabbit mAb Alexa Fluor 647 (flow cytometry 1:100). Other primary antibodies used were α tubulin, clone YL1/2 rat monoclonal IgG2a from Millipore Sigma MAB1864 (immunofluorescence 1:2500 dilution); β actin ab6276 from Abcam (immunoblot 1:1000 dilution); human GAPDH 4650S from Developmental Studies Hybridoma Bank (immunoblot 1:1000).

Infrared 800CW and 680LT secondary antibodies were used at 1:10,000 dilution and were obtained from LI-COR. Rabbit and mouse HRP antibodies were used at 1:10,000 dilution and were obtained from Jackson ImmunoResearch. Detection was performed either on an Odyssey infrared imaging system (LI-COR) or on a film developer. For immunofluorescence microscopy, Alexa-488, Alexa-594, and Alexa-647 coupled secondary antibodies from Jackson ImmunoResearch were used.

Clinical samples

All studies were conducted in accordance with the Declaration of Helsinki ethical guidelines and were performed after approval by the University of Wisconsin-Madison Institutional Review Board. Authors obtained written informed consent from all patients.

Histology and immunohistochemistry

Tissue cutting and H&E staining were completed by the Translational Research Initiatives in Pathology (TRIP) and Experimental Pathology laboratories at UW-Madison. Optimization of IHC staining for cGAS D1D3G Rabbit mAb #15102 was completed by the TRIP lab and used at 1:100 dilution. Cancer cells were visualized by Pan-keratin (C11) Mouse mAB #4545 and used at 1:100 dilution. DNA was visualized using DAPI at 1μg/mL dilution. Staining for both cGAS and pan-keratin used the immunofluorescence IHC protocol for pan-keratin, which was obtained from Cell Signaling Technology with the modification that antigen retrieval was performed in a pressure cooker at 250 Fahrenheit for 5 minutes in pH 6 citric acid buffer with 0.1% Triton-X.

Quantification criteria for tumor-infiltrating lymphocytes in histological samples

Tumor-infiltrating lymphocytes were quantified by based on the International TILs Working Group (14). Briefly, TILs are reported for the stromal compartment, where the numerator and denominator denote areas and not individual cells. The denominator represents the area of stroma and numerator represent the percentage of the stromal area occupied by lymphocytes. TILs are evaluated within the borders of the invasive tumor and are excluded outside of the tumor border and around DCIS and normal lobules. Tumor zones with crush artifacts, necrosis, and regressive hyalinization as well as in the previous core biopsy site are also excluded from the analysis. All mononuclear host immune cells (including lymphocytes and plasma cells) are scored with exclusion of polymorphonuclear leukocytes. At least one section (4–5 μm, magnification ×200–400) per patient is assessed. Biopsies are not preferred but can be used if full sections are not available as in this study. These criteria are designed for TILs assessment in the pre-therapeutic neoadjuvant setting but we have also used them in the neoadjuvant setting since no other criteria exist for evaluation in the neoadjuvant setting. TILs hot spots are excluded with a full assessment of average TILs in the tumor area by the pathologist as the final number being used in this study.

Quantification criteria for cGAS in immunohistochemistry samples

Quantification of cGAS expression is determined by average (mean) cGAS signal intensity of tumor cell clusters as visualized by the presence of pan-keratins. Each tumor cluster was assessed for cGAS by using the NIS elements ROI auto-detect tool. Tumor clusters from at least 10 random fields were assessed for each patient treatment condition at 600x magnification. Background subtraction was performed using the minus primary antibody negative control.

RNA interference

shcGAS and shSTING plasmids were purchased from Vector Builder. Predesigned MISSION siRNA against cGAS and STING were purchased from Sigma Aldrich. Transfection of siRNA was completed using Lipofectamine 3000 following manufacturer’s instructions.

Time-lapse video microscopy

For live cell imaging, MDA-MB-231 cells stably expressing H2B-RFP and tubulin-GFP were seeded onto glass-bottomed plates and maintained at 37 °C in DMEM HG media. Prior to imaging, MDA-MB-231 cells were treated with 10 nM paclitaxel. Image acquisition of MDA-MB-231 cells undergoing mitosis in either DMSO or paclitaxel was performed every 4 minutes at 20x magnification on a Nikon Eclipse Ti inverted microscope equipped with a 100x/1.4NA (Plan Apo) DIC oil immersion objective, motorized stage (Prior Scientific) and ORCA Flash4.0 V2+ digital sCMOS camera (Hammamatsu). Representative live cell images were captured at 60X objective using a Nikon Spinning Disk Confocal Microscope equipped with Yokogawa CSU-W1, 2 Hamamatsu Orca Flash 4 cameras, motorized stage, generation 4 Perfect Focus System and Tokei Hit. Video montages of single and combined wavelength channels of confocal microscope images were produced using FIJI.

Immunofluorescence microscopy

Cells were seeded on glass coverslips at low density in 24-well plates and allowed to grow until 50% confluence. Coverslips were washed twice in PBS before being fixed in 4% paraformaldehyde in PHEM buffer for 15 min at room temperature (RT), washed 3 times in PBS, and then blocked for 30 min at RT in 3% bovine serum albumin (BSA) and 0.1% Triton X-100 in PBS (PBSTx+BSA). Primary antibodies were pooled and diluted in PBSTx+BSA. Coverslips were incubated in primary antibodies for 1 h at RT and washed 3 times in PBSTx. Alexa Fluor (Invitrogen) secondary antibodies were pooled and diluted at 1:350 in PBSTx+BSA. Coverslips were incubated in secondary antibodies for 45 min at RT and then washed twice with PBSTx. Coverslips were counterstained with DAPI and mounted on glass slides with Prolong Diamond anti-fade medium (Invitrogen) and allowed to cure overnight. Image acquisition was performed on a Nikon Eclipse Ti inverted microscope equipped with a 100x/1.4NA (Plan Apo) DIC oil immersion objective, motorized stage (Prior Scientific), and CoolSNAP HQ2 CCD camera (Photometrics). Optical sections were taken at 0.2 μm intervals and, except for extraction experiments, deconvolved using the AQI 3D Deconvolution module in Nikon Elements.

For quantification of cGAS puncta, observer blinding was performed by slide label concealment. cGAS puncta were quantified as positive if immunofluorescence in puncta is above the background signal and there is also an overlaying DAPI signal. At least 250 cells were quantified for each replicate for a total of 3 replicates per treatment condition. Phenotypes observed upon paclitaxel challenge were also quantified. A classification scheme of 9 categories of DNA organization was employed for this purpose explained in more detail in the micronuclei quantification criteria section. Diffuse and punctate staining of cGAS positivity was also quantified. CREST signals were also quantified in cGAS puncta.

Quantification of cytoplasmic fluorescence intensity was performed using Nikon Elements. Cells were also imaged at 60x magnification with oil immersion and cell diameters were measured using the Nikon NIS Elements software. Images of at least 100 cells were acquired for each condition for paclitaxel. Threshold levels were equally applied to all images to exclude background intensity. cGAS immunofluorescence intensity was qualitatively determined by examining surrounding cell intensities. Sample size was selected for cell biology experiments based on prior experience and biologically significant effect size. For immunofluorescence, the sample size was typically ~100 cells. Three biological replicates were performed each with this sample size.

Data analysis was performed using Prism 8 (GraphPad). Statistical significance was determined using an ordinary one-way ANOVA with Dunnett’s multiple comparisons test with a single pooled variance when comparing multiple cell lines against the vector control or a two-tailed, unpaired t-test with Welch’s correction when comparing a single cell line with different chemical treatments.

Micronuclei quantification criteria

Micronuclei were qualitatively assessed based on nuclear morphologies present in untreated MDA-MB-231 cells. Major (M) nuclei have one nucleus and represent the most abundant phenotype seen in the cell population. The bi-nucleated+ category includes all nuclear morphologies that have two distinct nuclei categorized as follows. Major Minor (MR) have one large and one small nuclei (equivalent to nuclear structure with 3+ CREST puncta). Major Mini (MI) have one large and a qualitatively smaller nuclei (equivalent to nuclear structure with 2 CREST puncta). Major Micro (MC+) have the smallest nuclei (equivalent to nuclear structure with 1 or 0 CREST puncta). Major Micro+ (MC+) have a major, minor and micronucleus. Major Major+ (MM+) have two Major nuclei with or without micronucleus attached. The multinucleated category includes nuclear categories that typically only show up after paclitaxel treatment. These cells have 3 or more nuclei of similar sizes similar to a bunch of grapes. We divided the multinucleated cells into 3–5 MultiN (MN3+) with three to five nuclei and 6+ MultiN (MN6+) with six or more nuclei. This separation was made because paclitaxel at micromolar concentrations is known to arrest cells in mitosis, which can cause slippage, resulting in the formation of many smaller nuclei, typically greater than six. Cells that are delayed by mitosis but still progress through it typically have fewer nuclei being formed even though slippage can still occur.

Nuclear blebbing is defined as protrusions from the nuclear surface in interphase cells with a phenotype that suggests these structures to be connected to the main and we employed several stringent criteria to define micronuclei to exclude blebbing from our analysis. Micronuclei were defined as structures stained by DAPI that are separate from the main nucleus (if one exists). Micronuclei were also only quantified if they are oval or circular in shape. Z-stacks were employed over the entire height of the cell to ensure that there is no connection between the micronucleus and potentially other surrounding nuclei. For micronuclei overlapping with larger nuclei, changes in chromatin patterns between overlapping structures should exist or nuclear borders must be seen to clearly demarcate the borders of overlapping structures into either an oval or circular shape.

CRISPR-cas9 knockout of cGAS

A polyclonal pool of cGAS knockout MDA-MB-231 cells was produced by transfection of wild type MDA-MB-231 with Cas9-GFP Protein (Item No. CAS9GFPPRO) from Sigma Aldrich with predesigned Alt-R® CRISPR-Cas9 guide RNA against cGAS from Integrated DNA Technologies according to manufacturer’s instructions.

ELISAs for 2’3’-cGAMP and IFN-β

2’3’-cGAMP ELISA Kit (Item No. 501700) was purchased from Cayman Chemicals and used according to manufacturer’s instructions. Lysates for the 2’3’-cGAMP ELISA Kit were prepared by mixing 100uL of lysis buffer with previously frozen cell pellets consisting of 1 million cells flash frozen in liquid nitrogen and stored at −80°C.

For MDA-MB-231, Human IFN Beta Construction Kit (Item No. RHF842CK) was used to measure for supernatant IFN-β according to manufacturer’s instructions. For other breast cancer cell lines, Human IFN-beta DuoSet ELISA (Item No. DY814–05) was used with DuoSet ELISA Ancillary Reagent Kit 2 (Item No. DY008) to measure supernatant IFN-β according to manufacturer’s instructions. For each condition in measuring IFN-β, 50uL of supernatant was harvested from 2mL of total media in each well harboring 500 thousand cells of 6-well plates.

Human macrophage differentiation

Frozen PBMCs from individual donors without cancer were obtained from UW-Carbone Cancer Center Biobank. PBMCs were differentiated into M0 and M1 macrophages using Macrophage Base Medium XF (Item No. C-28057) and M1-Macrophage Generation Medium XF (Item No. C-28055) from Promo Cell according to manufacturer’s instructions. Macrophage CD80 was detected using CD80/B7–1 Monoclonal Antibody (Item No. 66406–1-IG) from Thermo Fisher conjugated to Alexa Fluor 488 Dye (immunofluorescence 1:100).

THP-1 differentiation

THP-1 cells between passage numbers 5 and 10 were grown in RPMI 1640 media to a density of 1–10 × 105. For differentiation to an M0 macrophage phenotype, THP-1 cells incubated with 320nM of PMA diluted in media in 6 well tissue culture plates for 24 hours. Subsequently, cells were washed three times with Hank’s Balanced Salt Solution (HBSS) before replacing with 10% FBS supplemented DMEM-High Glucose media. Cells were rested unperturbed for another 24 hours prior to experimentation.

Digitonin permeabilization

Digitonin permeabilization was carried out in a buffer containing 50 mM HEPES (pH 7), 100 mM KCl, 3 mM MgCl2, 0.1 mM DTT, 85 mM sucrose, 1 mM ATP, 0.1 mM GTP and 0.2% (v/v) FBS. 5ug/mL of digitonin was added either in isolation or with 10 μg/mL of 2′3′ cGAMP in permeabilization buffer for 10 min at 37 °C before replacing with fresh DMEM HG media.

Real-time quantitative PCR

cDNA preparation and quantitative PCR. cDNA was prepared as follows. ~ 3 × 106 cells were harvested and resuspended in 1 ml TRIzol (Life Technologies, # 15596018). For RNA extraction, 0.2 ml chloroform was added after incubation on ice for 10 minutes, and samples were thoroughly mixed. Samples were allowed to stand at room temperature for 10 min before centrifugation for 10 min (10,000 g; 4 °C). The aqueous phase was removed and mixed with an equal volume of isopropanol and centrifuged. The supernatant was removed and the remaining pellet was washed with 75% ethanol and allowed to dry at room temperature. Samples were eluted using 50 μl of nuclease-free water. ~ 300 ng RNA of each sample were used to generate cDNA using the Transcriptor First Strand cDNA Synthesis kit (Roche, # 04 379 012 001) according to the manufacturer’s instructions (both Anchored-oligo (dT)18 as well as random hexamer primers were used). 2 μl of each cDNA reaction were used for quantitative PCR with the LiqhtCycler 480 SYBR Green I Master mix (Roche, # 04 707 516 001), carried out on an iCycler (Bio-Rad) equipped with the IQ 5 detection system. The following temperature program was used: 5 min 95 °C; 50 × (10 sec 95 °C; 20 sec “annealing temperature”; 20 sec 72 °C).

The following primer sequences comprise the Type I interferon response panel with an annealing temperature of 55°C: STAT1 forward: 5’-CAGCTTGACTCAAAATTCCTGGA-3’, reverse: 5’-TGAAGATTACGCTTGCTTTTCCT-3’; STAT2 forward: 5’-GAGCCAGCAACATGAGATTGA-3’, reverse: 5’-GCCTGGATCTTATATCGGAAGCA-3’; IRF3 forward: 5’-AGAGGCTCGTGATGGTCAAG-3’, reverse: 5’-AGGTCCACAGTATTCTCCAGG-3’; IRF9 forward: 5’-GCCCTACAAGGTGTATCAGTTG-3’, reverse: 5’-TGCTGTCGCTTTGATGGTACT-3’; OAS1 forward: 5’-TGTCCAAGGTGGTAAAGGGTG-3’, reverse: 5’-CCGGCGATTTAACTGATCCTG-3’; IFNB1 forward: 5’-ATGACCAACAAGTGTCTCCTCC-3’, reverse: 5’-GGAATCCAAGCAAGTTGTAGCTC-3’.

The following primer sequences comprise the Type I interferon response panel with an annealing temperature of 57°C: IRF7 forward: 5’-CCCAGCAGGTAGCATTCCC-3’ reverse: 5’-GCAGCAGTTCCTCCGTGTAG-3’; OAS2 forward: 5’-ACGTGACATCCTCGATAAAACTG-3’, reverse: 5’-GAACCCATCAAGGGACTTCTG-3’; NLRC5 forward: 5’-ACAGCATCCTTAGACACTCCG-3’, reverse: 5’-CCTTCCCCAAAAGCACGGT-3’; IFI16 forward: 5’-GTTTGCCGCAATGGGTTCC-3’, reverse: 5’-ATCTCCATGTTTCGGTCAGCA-3’; IFI27 forward: 5’-TCTCTGCCCGGTGTTTTTGT-3’, reverse: 5’-TTCCGTGGCATTCCAGAGTC-3’

The following primer sequences comprise the THP-1 macrophage differentiation panel with annealing temperature of 56°C: RPLP0 forward: 5’-GCAGCATCTACAACCCTGAAG-3’, reverse: 5’-CACTGGCAACATTGCGGAC-3’; CCL18 forward: 5’-AAAATTGGCCAGGTGCAGTG-3’, reverse: 5’-TGAGGTTTCACCATGTTGGC-3’; CXCL10 forward: 5’-GTGGCATTCAAGGAGTACCTC-3’, reverse: 5’-TGATGGCCTTCGATTCTGGATT-3’; CXCL11 forward: 5’-AAGCAGGAAAGGTGCATGAC-3’, reverse: 5’-AGCTTTGCTGCTCTTCTTGG-3’; MMP9 forward: 5’-TGTACCGCTATGGTTACACTCG-3’, reverse: 5’-GGCAGGGACAGTTGCTTCT-3’; IL10 forward: 5’-GACTTTAAGGGTTACCTGGGTTG-3’, reverse: 5’-TCACATGCGCCTTGATGTCTG-3’. Readings were normalized to RPLP0.

Immunoblot

For all experiments, cells were challenged with 10nM paclitaxel for 2 or 4 days and cells were lysed and frozen down and stored at −80°C prior to use. Cell pellets were lysed in lysis buffer (50 mM HEPES pH 7.5, 100 mM NaCl, 0.5% NP-40, 10% glycerol) containing phosphatase inhibitors (10 mM sodium pyrophosphate, 5 mM β-glycerophosphate, 50 mM NaF, 0.3 mM Na3VO4), 1mM PMSF, 1x protease inhibitor cocktail (Thermo-Scientific) and 1 mM dithiothreitol. A 25-gauge syringe was used to provide additional mechanical lysis to the cell membrane before incubating lysate on ice for 30 minutes and centrifuging at 15,000 × g speed at 4°C to remove insoluble pellets. Protein concentration was measured using the Bradford assay. Proteins were separated by SDS-PAGE, transferred to Immobilon PVDF membrane (Millipore), and blocked for 30 min in 0.1% Tween-20 Tris buffered saline ph 7.4 supplemented with either 5% BSA (STING, phospho-IRF3) or 5% milk (cGAS, phospho-STAT1). Membranes were incubated with gentle agitation for 24 hours at 4°C with primary antibodies diluted in either TBST supplemented with 5% BSA or 5% milk, washed 3x with TBST, incubated for 1 h at room temperature in secondary antibodies conjugated to horseradish peroxidase diluted 1:10,000 in TBST supplemented with 5% milk and subsequently in LI-COR secondary antibodies after developing on film. Membranes were washed and developed with luminol/peroxide (Millipore) and visualized with film. Housekeeping proteins GAPDH and actin were developed

All results were obtained from single gels. To simultaneously probe for the protein of interest and the loading marker, the membrane was divided in two after transfer and incubated in separate antibody solutions. When identical-sized proteins prevented membrane division, the membrane was first probed for the protein of interest, stripped in an acidic glycine wash (100 mM glycine pH 2, 500 mM NaCl, 2% SDS), rinsed in deionized H2O, and then reprobed for the loading marker.

Human macrophage differentiation

Frozen PBMCs from individual donors without cancer were obtained from UW-Carbone Cancer Center Biobank. PBMCs were differentiated into M0 and M1 macrophages using Macrophage Base Medium XF (Item No. C-28057) and M1-Macrophage Generation Medium XF (Item No. C-28055) from Promo Cell according to manufacturer’s instructions in NUNC 6-well cell culture plates (Item No. 140675). Macrophage CD80 was detected using CD80/B7–1 Monoclonal Antibody (Item No. 66406–1-IG) from Thermo Fisher conjugated to Alexa Fluor 488 Dye (immunofluorescence 1:100).

Flow cytometry

Cells were harvested and labeled according to Cell Signaling Technology manufacturer’s flow cytometry protocol using PD-L1 Extracellular Domain Specific D8T4X Rabbit mAb Alexa Fluor 647 (flow cytometry 1:100). DAPI at 1ug/mL was used for live-dead staining. Cells were kept on ice before analysis on a ThermoFisher Attune N×T Flow Cytometer. FCS Express 7 Research Edition was used to make flow plots. Data analysis was performed using Prism 8 (GraphPad).

METABRIC and TCGA patient data analysis

We extracted altered mRNA sequencing gene expression Z-scores (<2 or >2) and survival data of cGAS and STING from patients of two public datasets, METABRIC and TCGA (breast cancer), using cBioPortal for cancer genomics (15,16). To compare cGAS expression levels for breast cancer subtypes, we extracted gene expression Z-scores from the METABRIC cohort for each patient sample relative to breast cancer subtype of hormone receptor presence (HR+) or absence (HR−) and HER2 presence or absence. Z-scores indicate the number of standard deviations that the cGAS expression of individual samples is away from the mean expression of all samples combined. Similarly, STING expression levels were compared for breast cancer subtypes using the same method. For Kaplan-Meier plots, cGAS and STING mRNA gene expression scores and recurrence free survival (RFS) data were extracted from the Breast KM plotter interface for patients who have undergone systemic treatment with chemotherapy but not endocrine therapy (17). Patient data were stratified by presence or absence of estrogen receptor. Patients were further split into expressing high (top 50%) or low (bottom 50%) levels of cGAS or STING by the median gene expression score. Kaplan-Meier plots were drawn using Prism 8 (GraphPad) based on high or low gene expression for cGAS and STING. Patients derived from some of these studies are managed according to the NIH Consensus guidelines for Adjuvant Therapy for Breast Cancer, which does not recommend chemotherapy for early-stage invasive disease, similarly to the NCCN guidelines.

Results

Some but not all triple-negative breast cancer patients develop tumor-infiltrating lymphocytes following neoadjuvant paclitaxel.

In TNBC, stromal tumor-infiltrating lymphocytes (sTILs) are associated with immune activity and better survival outcomes (18). To determine if paclitaxel can be associated with an immune response, we assessed sTILs in six primary breast tumors from treatment-naïve TNBC patients before neoadjuvant paclitaxel and at two time points following initiation of therapy (Figure 1A, B). Day 2 samples were collected 20 hours after the first dose of paclitaxel, when mitotic defects have been observed in our previous study (3). Day 15 samples were collected 20 hours after the third dose of paclitaxel, which is anticipated to be sufficient for accumulating post-mitotic cellular defects, such as micronuclei. We measured lymphocyte infiltration at each time point. Histologically, robust recruitment of sTILs was observed at Day 15 but not at Day 2 in two patient samples, P102 and P103. Similarly, patient sample P113 showed increased sTILs at Day 15, though these increases were modest (Figure 1B). In each case, sTILs measurements from separate cores were identical, suggesting little to no variation due to regional tumor sampling (Table S1). These data suggest that paclitaxel can induce recruitment of TILs in some, but not all TNBC samples, consistent with other investigations (19). Therefore, we sought to investigate the mechanism of sTILs recruitment.

Fig. 1. Some but not all triple-negative breast cancer patients develop tumor-infiltrating lymphocytes following neoadjuvant paclitaxel.

(A) (Above) Dose scheduling of patients with biopsies (P102, P103, P113, P114, P118, P119) undergoing neoadjuvant paclitaxel treatment. Treatment regimen comprises biweekly 80mg/m² neoadjuvant paclitaxel for four doses. (Below) Representative hematoxylin and eosin (H&E) images showing increased stromal tumor-infiltrating lymphocytes (sTILs) in one patient P102 after the third dose of 80mg/m² neoadjuvant paclitaxel 15 days after the start of therapy. Inset contains magnifications of representative patient images. Tumor is outlined in black and sTILs are represented by small cells with dark, purple nuclei in “Day 15” sample. Original images are at 400x magnification.

(B) Quantification of sTILs of patient H&E images by staff pathologist using International TILs Working Group criteria. See supporting information for more details. Qualitative TILs categories are A=low (0–20%), B=medium (20–50%), C=high (50–100%). Red shading represents patients with TILs increased to a higher qualitative category at “Day 15” compared to baseline. Blue shading represents patients with TILs that have either not increased or have not increased to a higher qualitative category.

(C) Representative images of tumor cells from patient biopsies of nuclear phenotypes after paclitaxel showing DNA (DAPI), cGAS and pan-cytokeratin (PCK). Micronucleus is represented by yellow arrow and magnified in the inset panels on the bottom-right corner. Scale bar, 50μm.

(D) Quantification of intratumoral cGAS expression by average (mean) cGAS signal intensity of tumor cell clusters. Red shading represents patients with TILs increased to a higher qualitative category at “Day 15” compared to baseline. Blue shading represents patients with TILs that have either not increased or have not increased to a higher qualitative category. AU=arbitrary units

(E) Spearman correlation of six patients treated with neoadjuvant paclitaxel from (A-D) plotted by change in TILs between “Day 15” and “Baseline”. Red squares represent patients with TILs increased to a higher qualitative category at “Day 15” compared to baseline. Blue squares represent patients with TILs that have either not increased or have not increased to a higher qualitative category. AU=arbitrary units

Paclitaxel is known to induce mitotic aberrations, which could generate micronuclei. Ruptured micronuclei activate innate immune signaling in certain contexts (7,8). To determine if we can observe increased micronuclei after paclitaxel in human TNBC, we quantified interphase DAPI fragments representing micronuclei (Figure 1C). We excluded mitotic cells as identified by condensed chromosomes and NUMA localization at spindle poles (Figure S1A). The percentage of cells with micronuclei increased at Day 15 compared with the percentage at baseline in three of four samples (Figure S1B). Further, we observed a positive trend between increased micronuclei and sTILs with two patient samples having both increased micronuclei and sTILs at Day 15 (Figure 1B, Figure S1C).

Previous studies show that radiation can recruit cGAS-positive micronuclei, which are associated with activation of downstream STING signaling and immune activation (7,8,20,21). Therefore, we evaluated cGAS recruitment to micronuclei in histological samples from patients on Day 15 (Figure 1C). Four of six patients had increased cGAS-positive micronuclei at Day 15 (Figure S1D). We also observed a positive trend between cGAS-positive micronuclei and sTILs, but this result was not statistically significant, suggesting that other modifying factors are at play (Figure S1E). When we quantified tumor cGAS expression, we found cGAS expression levels to positively correlate with sTIL recruitment (Figure 1E). These results suggest that paclitaxel can recruit cGAS signaling in tumors to elicit immune activation although the process may be more complex and influenced by tumor cGAS expression levels. Based on our findings, we hypothesized that tumor cGAS expression and micronucleation may play a role in downstream STING signaling after paclitaxel to promote an immune response. Therefore, we set out to interrogate this hypothesis mechanistically.

Paclitaxel causes cGAS-positive micronuclei to form in interphase following delayed mitosis on multipolar spindles.

To determine how paclitaxel causes micronuclei to form in cells, we challenged fluorescently tagged RFP-H2B, GFP-Tubulin MDA-MB-231 cells with paclitaxel and tracked them through mitosis into the subsequent interphase by live-cell imaging (Figure 2A, Videos 1, 2). In many cases, cells exposed to clinically relevant doses of paclitaxel are known to form multipolar spindles and divide, sometimes after a delayed mitosis (3). In accord with this, we found most cells to complete aberrant cell division on multipolar spindles (Figure 2B). In the subsequent interphase, cells became multinucleated and contained a subset of irregularly sized nuclear fragments, some resembling micronuclei. These findings suggest that mitotic transit on multipolar spindles is sufficient to generate cells with nuclear fragments.

Fig. 2. Paclitaxel causes cGAS-positive micronuclei to form in interphase following delayed mitosis on multipolar spindles.

(A) Representative images of MDA-MB-231 cells expressing H2B-RFP and tubulin-GFP filmed with 60X objective from mitosis to subsequent interphase with no treatment (UNT) at 5-min intervals or 10 nM paclitaxel (PAX) at 4-min intervals.

(B) Quantification of live cell images from (A) for mitotic duration and chromosome segregation defects from Nuclear Envelope Breakdown (NEBD) to anaphase onset and daughter cell interphase nucleus morphology. N=50 cells for UNT and N=150 cells for 10 nM PAX.

(C) Representative images of multinucleated MDA-MB-231 after 10 nM paclitaxel with and without cGAS puncta. Scale bar, 10 μm.

(D) Quantification of nuclei phenotypes (sum of blue and green bars) and cGAS-positivity (green bar) in MDA-MB-231 interphase cells after 0, 24, 48 and 72 hours of 10 nM paclitaxel. N=3 independent experiments comprising ≥250 asynchronous cells per condition. Mean and SEM are plotted. P values are calculated by one-way ANOVA with Tukey-Kramer test (*P < 0.05, **P < 0.01, ***P < 0.001, ns=not significant). See Figure S2A for representative images of nuclei phenotypes.

(E) Quantification of micronuclei (sum of blue and green bars) of multinucleated cells of multiple TNBC cell lines after 0 and 48 hours of 10 nM paclitaxel N=3 independent experiments comprising ≥250 asynchronous cells per condition. Results for MDA-MB-231 from (D). Mean and SEM are plotted. P-values were calculated by two-tailed, paired Student’s T Test (*P < 0.05, **P < 0.01, ***P < 0.001, ns=not significant). TNBC subtypes are M=mesenchymal, MSL=mesenchymal stem-like (retired classification), BL1=basal-like 1, BL2=basal-like 2, LAR=luminal androgen receptor.

(F) Representative images of multinucleated MDA-MB-231 after 10 nM paclitaxel stained for cGAS and CREST. Yellow arrows indicate a cGAS-positive micronucleus with one CREST foci, which is magnified in the top-left inset. Scale bar, 10 μm.

(G) Quantification of MDA-MB-231 CREST foci with no treatment (UNT) or after two days of 10 nM paclitaxel (10 nM PAX) categorized by number of cGAS-positive or cGAS-negative micronuclei. N=3 independent experiments comprising ≥50 asynchronous cells (≥ total 100 micronuclei) per condition. Mean and SEM are plotted. P values were calculated by one-way ANOVA with Tukey-Kramer test (*P < 0.05, **P < 0.01, ***P < 0.001, ns=not significant).

We extended our findings to fixed cell immunofluorescence analysis of wild type MDA-MB-231 cells exposed to clinically relevant, 10 nM paclitaxel. Multinucleated cells were frequently observed (68% +/− 6.4) 48 hours after paclitaxel challenge, when compared to controls (1.1% +/− 0.6) (Figure 2C, D). Moreover, a proportion of multinucleated cells had nuclear fragments that expressed high levels of cGAS, which far exceeded untreated controls (Figure 2D). We reasoned that micronuclei appeared sooner in cell culture compared to patient samples due to faster growth rates in vitro (3).

To determine if this effect can be generalized, we quantified the frequency of cGAS-positive multinucleated cells in other cell lines corresponding to distinct molecular subtypes of TNBCs as determined from previous studies (22). Indeed, paclitaxel induced cGAS-positive micronuclei in other TNBC cell lines, though, importantly, it was absent in MDA-MB-468 and MDA-MB-453 cells, which are two lines that do not express detectable cGAS by immunoblot (Figure 2E, S2C).

Since cGAS recruitment was commonly found on small nuclear fragments, we reasoned micronuclei with few chromosomes are most prone to rupture. To test this, we co-stained cells with CREST antibody, which labels centromeres, to mark individual chromosomes (Figure 2F). In untreated cells, we sometimes found cGAS recruitment, most prominently to fragments that lacked CREST foci, suggesting acentric chromosome fragments. However, after paclitaxel, most cGAS-positive micronuclei contained one or two CREST foci, consistent with paclitaxel inducing whole chromosome missegregation. By contrast, we did not find a significant number of cGAS-positive fragments in micronuclei with three or more CREST foci, suggesting that these larger nuclear fragments are less likely to be exposed to cytosolic sensors. This may be due to less lamin B1 being incorporated into small compared to large micronuclei, which would increase nuclear envelope fragility and promote rupture events (23).

To further characterize the nuclear fragments that contribute to increased cGAS positivity, we categorized nuclear morphologies observed after paclitaxel treatment of TNBC cell lines, resulting in eight distinct phenotypes (Figure S2A). Next, we quantified the nuclear morphologies after paclitaxel exposure in a range of TNBC cell lines with different levels of cGAS expression (Figure S2B–C). Cells with high (MDA-MB-231), intermediate (MDA-MB-436, BT-549, HCC-1806) and low cGAS (MDA-MB-468, MDA-MB-453) expression levels were examined. All cell lines exhibited similar patterns of nuclear phenotypes, with the majority of single-nucleated cells converting to multinucleated cells after paclitaxel. In cell lines expressing cGAS, many multinucleated cells also contained cGAS-positive micronuclei (Figure S2C, D). These results suggest that the formation of cGAS-positive micronuclei is a general effect elicited by treatment with paclitaxel. Similarly, nanomolar concentrations of other MTAs, vinorelbine and eribulin, also induced cGAS-positive micronucleation in MDA-MB-231 cells (Figure S2E, F). Given the strikingly similar nuclear phenotype, these results support the conclusion that MTAs in general can cause cGAS-positive micronuclei formation.

Although these findings suggested a clinically important mechanism of immune activation, we considered alternative explanations. We found cGAS puncta are not simply a consequence of generally increased expression, as judged by unchanged cGAS levels after 4-days exposure to paclitaxel (Figure S3A). We ruled out nonspecific staining with two additional antibodies targeting distinct regions of cGAS, which also stained micronuclei (Figure S3B). Further, we generated biallelic knockout mutants of cGAS in MDA-MB-231 cells using CRISPR-Cas9 and validated clones by immunoblot, TA cloning and 2’3’-cGAMP production after radiation-induced stimulation (Figure S3C–E). As expected, cGAS-positive puncta were abrogated in cGAS-knockout MDA-MB-231 cells before and after paclitaxel (Figure S3F), suggesting that observed signals are specific to cGAS. To ensure that the mechanism is consistent with previous studies, we assessed the integrity of the nuclear envelope of micronuclei for rupture, which is known to precede cGAS recruitment to micronuclei in other contexts (7–9). As expected, paclitaxel-induced cGAS-positive micronuclei have reduced levels of nuclear envelope and intranuclear proteins (Figure S3G, H) (24). Taken together, our findings demonstrate that paclitaxel and potentially other MTAs cause multipolar cell divisions in TNBC cell lines, producing daughter cells with micronuclei containing defective nuclear membranes, which can rupture and recruit cGAS.

Paclitaxel activates cGAS signaling in triple-negative breast cancer cell lines.

cGAS preferentially binds to dsDNA to catalyze synthesis of STING-substrate, 2’3’-cGAMP (6,25). To interrogate this pathway, we evaluated if the appearance of cGAS-positive micronuclei corresponded with cGAS activity after paclitaxel exposure. To test this, wild type and cGAS knockout MDA-MB-231 cells were exposed to 10 nM paclitaxel over four days and 2’3’-cGAMP levels were assessed. Consistent with cGAS activation, paclitaxel increased 2’3’-cGAMP, which peaked after two days in wild type cells, but remained undetectable in cGAS-knockout cells, as expected (Figure 3A). Further, STING activation, as measured by phosphorylation of downstream proteins TBK1, IRF3, p65, and STAT1, increased after paclitaxel exposure in wild type MDA-MB-231 cells (Figure 3B, S4A). Importantly, paclitaxel failed to induce STING activation in both cGAS knockout MDA-MB-231 and cGAS knockdown MDA-MB-436 cells, suggesting that STING activation is dependent on cGAS and that this effect is not cell line dependent (Figure 3C, D). While si-RNA mediated cGAS knockdown showed increased pIRF3 upregulation at baseline potentially indicating off target effects, sh-RNA mediated cGAS and STING knockdown showed decreased expression of interferon-stimulated genes after paclitaxel in MDA-MB-231 cells compared to wild type (Figure S4B, C). These results demonstrate that paclitaxel activates cGAS-STING signaling in TNBC cells and that this is associated with the appearance of cGAS-positive micronuclei.

Fig. 3. Paclitaxel activates cGAS signaling in triple-negative breast cancer cell lines.

(A) ELISA analysis of 2’3’-cGAMP production in wild type and cGAS knockout MDA-MB-231 four days after exposure to 10 nM paclitaxel. N=3 independent experiments. Mean and SEM are plotted. P values were calculated by one-way ANOVA with Dunnett’s test with control as 0 day (*P < 0.05, **P < 0.01, ***P < 0.001, ns=not significant).

(B) Representative immunoblot of MDA-MB-231 wild type cells exposed to 10 nM paclitaxel after 0, 0.5, 1, 2 and 4 days probed with phosphoproteins downstream of the STING pathway that increase with activation of the pathway comprising pTBK1, pSTING, pIRF3 and pSTAT1.

(C) Representative immunoblot of cGAS, STING and pIRF3 from MDA-MB-231 wild type (WT) and cGAS knockout (KO) cells exposed to 10 nM paclitaxel for four days. Alpha tubulin serves as the loading control.

(D) Representative immunoblot of cGAS and pIRF3 from MDA-MB-436 wild type (WT) and siRNA-mediated cGAS knockdown (KD) cells exposed to 10 nM paclitaxel for four days. Alpha tubulin serves as the loading control.

(E) Interferon-beta measurements of wild type (WT) and cGAS knockout (KO) MDA-MB-231 exposed to no treatment (UNT) or 10 nM paclitaxel (PAX) over five days. N=2 independent experiments. Mean and SEM are plotted. P values were calculated by one-way ANOVA with Tukey-Kramer test (*P < 0.05, **P < 0.01, ***P < 0.001, ns=not significant).

(F) Interferon-beta measurements of MDA-MB-436 wild type (WT) and siRNA-mediated cGAS knockdown (KD) exposed to no treatment (UNT) or 10 nM paclitaxel (PAX) over five days. N=2 independent experiments. Mean and SEM are plotted. P values were calculated by one-way ANOVA with Tukey-Kramer test (*P < 0.05, **P < 0.01, ***P < 0.001, ns=not significant).

Although MDA-MB-231 cells express abundant cGAS, they are inconsistent in their ability to synthesize type I interferons (26,27). In our study, we did not detect IFN-β secretion by MDA-MB-231 after paclitaxel despite activation of the cGAS-STING pathway (Figure 3E). By contrast, MDA-MB-436 cells have an intact type I interferon pathway (28). Therefore, we also measured IFN-β protein in MDA-MB-436 cells, which readily form cGAS-positive multinucleated cells after paclitaxel (Figure S2D). After five days of 10 nM paclitaxel, we detected increased IFN-β protein in the supernatant (Figure 3F). Importantly, this effect was dependent on cGAS as we observed an attenuation of IFN-β production with siRNA-mediated cGAS knockdown. Collectively, our data show that paclitaxel activates the STING pathway, and induces synthesis of 2’3’-cGAMP and IFN-β, in a cGAS-dependent manner.

Paclitaxel induces secretion of cGAS-dependent soluble factors in breast cancer cells that polarize macrophages to an M1-like phenotype.

Following cGAS activation in the setting of self-DNA exposure, tumor cells can initiate a signaling cascade that results in the production of 2’3’-cGAMP, inflammatory cytokines, chemokines and IFN-β (6). These soluble factors are secreted and maintained through autocrine and paracrine loops (27,29). Among immune cells, macrophages have functional plasticity and can be reprogrammed to either an M1 or M2 phenotype in response to environmental cues by interferon-derived factors (30).

To test the hypothesis that soluble factors derived from tumor cells can promote macrophage M1 polarization, we differentiated THP-1 cells into macrophages and incubated them in either paclitaxel or paclitaxel-conditioned media from MDA-MB-231 cells and assessed for polarization (Figure S5A). Direct exposure of macrophages to 10 nM paclitaxel for four days only minimally upregulated pro-inflammatory chemokines associated with M1 polarization (Figure S5B). By contrast, paclitaxel-conditioned media from MDA-MB-231 cells induced M1 polarization of THP-1 cells, but M1 polarization was attenuated with conditioned media from cGAS knockout MDA-MB-231 cells (Figure S5C, D). As a complementary approach, we performed 2D co-culture of wild type or cGAS knockout MDA-MB-231 cells with THP-1 in the presence and absence of paclitaxel (Figure S5E). We observed increased expression of M1-associated genes, CXCL10 and CXCL11, despite concurrent upregulation of M2-associated genes in THP-1 cells co-cultured with MDA-MB-231 cells following paclitaxel exposure. Importantly, these two M1-associated genes were downregulated after cGAS knockout. Our results suggest that THP-1 macrophages in co-culture with MDA-MB-231 and in conditioned media from MDA-MB-231 both upregulated expression of M1-associated genes to a greater extent than in direct exposure to paclitaxel. These data suggest that paclitaxel induces MDA-MB-231 cells to secrete soluble factors, in a cGAS-dependent manner, that exhibit paracrine effects on THP-1 macrophages to elicit a phenotype similar to M1. Given that some M2 markers were upregulated as a result of paclitaxel, it is likely that the M1-like macrophage phenotype we observed has characteristics of M1 polarization but is not identical, suggesting a complex phenotype. Our results also suggest that other factors such as 2’3’-cGAMP and NF-κB stimulated genes may be involved in mediating the polarization of THP-1 macrophages as MDA-MB-231 do not secrete noticeable IFN-β (Figure 3E).

THP-1 cells have differences in gene expression, protein secretion and cellular morphology compared to macrophages derived from peripheral blood mononuclear cells (PBMCs) at baseline and after polarization (31). Therefore, to more closely mimic physiological conditions, we incubated macrophages derived from PBMCs in conditioned media from MDA-MB-231 previously treated with and without paclitaxel to assess for polarization. MDA-MB-231 cells are known to polarize macrophages derived from PBMCs to an M2-like phenotype with an elongated morphology and changes in surface marker expression (32). Given this, we measured expression of the M1 marker, CD80, and length of macrophages after exposure to conditioned media from MDA-MB-231 cells. PBMCs from three healthy volunteers were differentiated into M0 macrophages and separately incubated with conditioned media from MDA-MB-231 parental and cGAS knockout cell lines. Conditioned media were either obtained from MDA-MB-231 previously untreated or treated with two days of 10 nM paclitaxel before washing off and replacing with fresh media (Figure 4A). We also polarized M0 macrophages to an M1 phenotype to assess the validity of our CD80 marker (Figure S5H). For all conditions, following three days incubation with conditioned media, blood-derived human macrophages were assessed for M1 polarization. We quantified immunofluorescence images for maximum length of macrophages on bright field images and total CD80 immunofluorescence (Figure 4B). Exposure to conditioned media from MDA-MB-231 cells polarized macrophages to a pro-tumoral M2-like phenotype as characterized by an elongated shape and decreased CD80 expression (Figure 4C, D). By contrast, exposure to conditioned media from paclitaxel-treated MDA-MB-231 allowed macrophages to maintain round morphology and increased CD80 expression. Importantly, CD80 levels were increased from baseline M0 macrophages, suggesting paclitaxel-treated MDA-MB-231 can polarize macrophages to an M1 phenotype.

Fig. 4. Paclitaxel induces secretion of cGAS-dependent soluble factors in breast cancer cells that polarize macrophages to an M1-like phenotype.

(A) Schematic of mono-culture conditioned media experiments with macrophages derived from PBMCs.

(B) Representative images of human macrophages expressing CD80 incubated with conditioned media from MDA-MB-231 parental and cGAS knockout cell lines untreated or pre-treated with 10 nM paclitaxel.

(C) Measurements of CD80 expression on macrophages following conditioned media from parental and cGAS knockout MDA-MB-231 untreated or pre-treated with paclitaxel. N=3 independent experiments. Each dot represents one cell. Mean and SEM are plotted. P values were calculated by one-way ANOVA with Tukey-Kramer test (*P < 0.05, **P < 0.01, ***P < 0.001, ns=not significant).

(D) Measurements of greatest dimension of macrophages following conditioned media from parental and cGAS knockout MDA-MB-231 untreated or pre-treated with paclitaxel. N=3 independent experiments. Each dot represents one cell. Mean and SEM are plotted. P values were calculated by one-way ANOVA with Tukey-Kramer test (*P < 0.05, **P < 0.01, ***P < 0.001, ns=not significant).

As expected, we found that macrophages exposed to conditioned media from paclitaxel-treated cGAS knockout MDA-MB-231 did not exhibit polarization, illustrating that intact cGAS signaling in tumor cells is important for macrophage polarization in co-culture (Figure 4C, D). Macrophage expression of CD80 is not known to be affected by downstream products of cGAS activation in MDA-MB-231, so we were surprised to see increased expression of CD80 in macrophages co-cultured with cGAS knockout MDA-MB-231 in the setting of paclitaxel exposure. However, we do not think that increased macrophage CD80 expression is indicative of M1 polarization because the length of cGAS KO cells did not increase compared to parental MDA-MB-231. Potentially, CD80 is regulated by downstream products of cGAS activation in MDA-MB-231 through an unknown mechanism. Since cGAS-STING activation results in production of 2’3’-cGAMP and IFN-β in some cell lines, we tested if these compounds could induce M1 polarization when exogenously added to conditioned media from cGAS knockout MDA-MB-231 cells. Previous studies suggest that both 2’3’-cGAMP and IFN-β have the capacity to polarize macrophages to an M1 phenotype (33,34). While we did not see any changes of CD80 expression in macrophages co-cultured with cGAS KO cells that were exposed to untreated, 2’3’-cGAMP and IFN-β media, we did see a decrease in cell length between untreated and both 2’3’-cGAMP and IFN-β media, which is a more sensitive marker of polarization (Figure 4C, D). Therefore, our data suggest that both 2’3’-cGAMP and IFN-β can partly rescue M1 polarization in MDA-MB-231 cGAS knockout cells. Since MDA-MB-231 do not secrete noticeable IFN-β (Figure 3E), endogenously produced 2’3’-cGAMP may at least partly be responsible for polarizing primary macrophages to an M1 phenotype as a result of paclitaxel exposure.

Nanomolar paclitaxel and other microtubule-targeting agents can increase surface PD-L1 expression in a cGAS-dependent manner in MDA-MB-436.

PD-L1 is an immune checkpoint marker whose expression on the surface of tumor cells can prevent T-cell activation. Importantly, PD-L1 can be upregulated by type I interferons (10,35). PD-L1 blockade not only allows activated T-cells to kill tumor cells expressing PD-L1 but may also promote T-cell priming by dendritic cells, which is important for generating an adaptive immune response (36). Interestingly, T-cell priming may depend on cGAS expression (37). Therefore, we reasoned that cGAS activation is a possible intermediary in the efficacy of ICIs and the need for them (10,20). PD-L1 mRNA expression increases after 20 hours of 30 nM paclitaxel in murine 4T1 and CT26 cancer cell lines (38); however, this report did not examine protein expression and surface localization. Therefore, we tested if these findings extend to human cancer cell lines with prolonged paclitaxel exposure, and if cGAS were necessary for this effect. For this purpose, we selected two cell lines – MDA-MB-436, which has a robust type I interferon response, and MDA-MB-231, which does not. As expected, nanomolar paclitaxel upregulated total PD-L1 expression in MDA-MB-436 cells and this effect was dependent on cGAS, as siRNA-mediated depletion of cGAS blocked this effect (Figure 5A). Since only surface rather than intracellular PD-L1 binds to immune cell receptors, we also assessed surface PD-L1 expression on MDA-MB-436 cells (Figure 5B). As expected, paclitaxel significantly upregulated surface PD-L1 in a cGAS-dependent manner, recapitulating our immunoblot results (Figure 5C). By contrast, MDA-MB-231 cells did not exhibit increased total or surface PD-L1 after paclitaxel (Figure S6A–C), due to a lackluster type I interferon response or high baseline PD-L1 expression. Interestingly, other MTAs such as vinorelbine and eribulin increased PD-L1 to a greater degree than paclitaxel in MDA-MB-436 cells (Figure 5D). Our results illustrate upregulation of PD-L1 by MTAs in a cGAS-dependent manner, particularly in cell lines with intact IFN-β signaling. These findings are consistent with other studies that have associated exposure of cells to type I interferons including IFN-β with tumor cell PD-L1 upregulation (10,39).

Fig. 5. Nanomolar paclitaxel and other microtubule-targeting agents can increase surface PD-L1 expression in a cGAS-dependent manner in MDA-MB-436.

(A) Representative immunoblots of total PD-L1 expression in wild type and si-cGAS MDA-MB-436 cells before and after three days of 10 nM paclitaxel. Alpha tubulin serves as the loading control.

(B) Flow cytometry strategy for PD-L1 expression.

(C) Quantification of surface PD-L1 by percent of the live parent population of MDA-MB-436 cells, wild type and siRNA-mediated cGAS knockdown, before and after three days of 10 nM paclitaxel. Mean and SEM of n=3 independent biological experiments comprising ≥100,000 events per condition are shown. P values were calculated by one-way ANOVA with Dunnett’s test with control as 0 day (*P < 0.05, **P < 0.01, ***P < 0.001, ns=not significant).

(D) Quantification of surface PD-L1 by percent of the live parent population of wild type MDA-MB-436 cells before and after three days of 10 nM eribulin and vinorelbine. Mean and SEM of n=3 independent biological experiments comprising ≥100,000 events per condition are shown. P values were calculated by one-way ANOVA with Dunnett’s test with control as 0 day (*P < 0.05, **P < 0.01, ***P < 0.001, ns=not significant).

Some triple negative breast cancer patients with high levels of tumor cGAS have durable responses to combination therapy.

While multiple cancer types may express cGAS, these levels vary greatly (40). Therefore, we reasoned that there may be differences in cGAS expression among breast cancer subtypes. We extracted cGAS and STING expression patient data from METABRIC, which is a public dataset of patient tumor biopsies with available mRNA sequencing of samples. While these samples are mostly comprised of tumor, they may contain other cell types including immune cells. We found that average cGAS levels are higher in the hormone-receptor negative subtypes and reach the highest levels in a small fraction of TNBC tumors (Figure S7A). We reasoned that if the cGAS-STING pathway mediates immune activation after paclitaxel, patients with higher cGAS expression could mount the most robust immune response after treatment. Consistent with this idea, relapse-free survival (RFS) after adjuvant taxane-based chemotherapy, which also includes anthracycline and cyclophosphamide, was superior with high cGAS expression, though no such effect was found in patients who did not receive any chemotherapy. While not common, adjuvant therapy is not always given to patients with triple negative breast cancers (Figure S7B). Interestingly, in hormone receptor-positive breast cancers, there was also a significant positive correlation between cGAS and RFS (Figure S7C). STING is activated by 2’3’-cGAMP, the product of cGAS activity. Unlike cGAS, STING exhibited similar distribution across all breast cancer subtypes (Figure S7D) and did not correlate with RFS in chemotherapy-treated patients (Figure S7E–F). These findings support the idea that cGAS expression, but not STING, varies among TNBC subtypes and could mediate therapeutic response.

Our study so far suggests that paclitaxel can directly activate a cGAS-mediated immune response. To test if cGAS expression could predict response to combined MTAs and ICIs, we obtained pre-treatment FFPE slides from ten metastatic TNBC patients who were subsequently treated with ICIs with or without MTAs. Quantitative immunofluorescence illustrated variable cGAS levels in tumor specimens (Figure 6A, B). Staining specificity was validated with paraffin-embedded cell pellets from MDA-MB-231 cells which express cGAS and HEK-293T cells, which do not (Figure 6A, C). We found highest cGAS expression in P109, P811, P888, and P169 (Figure 6B). Among these, three of six patients had progression free survival exceeding 20 months after initiation of therapy, a durable response that greatly exceeds the average for triple-negative breast cancer. By contrast, patients with tumors expressing lower levels of cGAS had disease progression several months after initiation of therapy (Figure 6D). To evaluate this quantitatively, we determined a significant Spearman correlation between cGAS intensity and progression-free survival (Figure 6E). Interestingly, P811 remained an outlier with high expression of cGAS and rapid progression, suggesting that additional variables are at play, such as the use of a PD-L1 inhibitor with this subject versus a PD-1 inhibitor with the others. Based on our mechanistic findings and clinical validation data in patient tumor specimens, MTAs may activate an immune response in cells and tumors expressing cGAS. Therefore, cGAS expression could help select patients for combined MTAs and ICIs with a supportive underlying mechanistic rationale.

Fig. 6. Patients expressing higher levels of cGAS have increased disease control after combined microtubule-targeting agents and PD-1/PD-L1 inhibitors.

(A) Representative images of patient tumors from diagnostic biopsies with DNA (DAPI), cGAS and pan-cytokeratin (PCK) staining shown. Representative images by order of appearance from top to bottom are: patient sample with a high level of cGAS staining (cGAS^high), patient sample low level of cGAS staining (cGAS^low), background control with only secondary antibody and no primary antibody staining (−1°Ab), positive control for cGAS derived from paraffin-embedded MDA-MB-231 cell pellet (MDA-231), and negative control for cGAS derived from HEK293T cell pellet (HEK-293T). Scale bar, 50 μm.

(B) Quantification of intratumoral cGAS expression by average (mean) cGAS signal intensity of tumor cell clusters. AU=arbitrary units

(C) Immunoblot of cGAS in MDA-MB-231 and HEK293T, which serve as positive and negative controls for cGAS staining.

(D) Swimmer’s plot showing progression free survival of patients in months after combination therapy of paclitaxel and atezolizumab (PD-L1) or eribulin and pembrolizumab (PD-1) stratified from top to bottom by high to low cGAS expression. Legend indicates treatment regimen (top) and response status (bottom). Top legend color coding also applies to (E) for patient treatments.

(E) Spearman correlation of seven patients treated with both microtubule and immune checkpoint inhibitors from (B-C) plotted by mean cGAS intensity against progression free survival in months. This analysis excludes three patients without microtubule inhibitor treatment. AU = Arbitrary Units

Discussion

In this study, we found that paclitaxel and other MTAs can activate the cGAS-STING pathway and then explored the possibility of cGAS as a potential biomarker for combining MTAs and ICIs in patient samples (Figure S8). We employed clinically relevant levels of paclitaxel in cell culture to assess micronuclei formation and subsequent cGAS recruitment and activation in TNBC cell lines. At these concentrations, some TNBC, such as MDA-MB-231, can complete mitosis on multipolar spindles (Figure 2A–B, Videos 1,2), an effect that is markedly distinct from the mitotic arrest elicited by exposure to micromolar concentrations of paclitaxel (3,41).

Tumor cells act as a sink that concentrates intracellular levels of paclitaxel 67- to over 1000-fold relative to the extracellular concentration (2,3,42,43). Intratumoral paclitaxel levels accumulate to only 1–9 μM in human breast cancer (3). Therefore, cell culture paclitaxel concentrations between 5 nM and 50 nM, depending on cell type, best recapitulate physiologic intratumoral levels, contrasting with a number of prior preclinical studies that use 10- to 100-times higher concentrations. We expect our mechanistic findings in cell culture using these pharmacologically relevant concentrations reflect the effects of paclitaxel in humans, and further use clinical samples as a touchstone to verify concordance.

We hypothesize paclitaxel treatment in the short term to result in acute STING activation, which can result in antitumoral immunity through interferon-mediated pathways. In contrast, chronic STING activation can upregulate an unfolded protein response through autophagy-ER stress programs, which can drive tumorigenesis and immunosuppression (26,44). Further, this antitumoral immunity can be boosted by ICIs (21,37). Compared with local delivery of radiation or injected STING agonists, paclitaxel has the advantage of systemic administration, allowing it to reach tumor cells at virtually any site, including all metastatic—and micrometastatic—sites of disease. Therefore, the ability of paclitaxel to activate cGAS is highly suitable for treating or preventing metastatic breast cancer in tumors with high cGAS expression.

Here, we focused solely on cGAS, which is the only indispensable DNA sensor capable of mediating an IFN-β response to self-DNA (45). Other DNA sensors are unlikely to contribute to paclitaxel-induced IFN-β synthesis. For instance, AIM2 is a dsDNA sensor that can indirectly activate type I interferons but is not found in breast cancer cell lines and requires co-stimulation with IFN-γ, which is not known to be upregulated by paclitaxel (46). DAI/ZBP1 and IFI16 can directly activate STING but are expressed at low levels in breast cancer cell lines and tumors (47). On the contrary, DDX41 is a DNA sensor expressed in many cancers and can activate STING in vitro; however, such activation through this sensor is questionable in vivo (48). LRRFIP1 activates IRF3 signaling through β-catenin phosphorylation, and while the protein is expressed in breast cancers, this pathway appears to induce very little IFN-β synthesis, suggesting that it is only contributes little, if at all, to paclitaxel-mediated IFN-β signaling (49). While these prior studies cannot rule out minor roles of other sensors, our cGAS knockout and knockdown data support the idea that cGAS is the major DNA sensor involved in paclitaxel-mediated effects on immune activation.

We recognize that cGAS expression alone may not be sufficient to predict immunotherapy response for TNBC after paclitaxel. Immunotherapy response is influenced by many factors that regulate the immune system such as the presence of detectable neoantigens, and its suppression by the presence of PD-1/PD-L1, regulatory T cells and myeloid-derived suppressor cells (MSDC) among others (50). Paclitaxel may modulate some of these other factors, including inhibiting MDSC and regulatory T-cell populations while increasing PD-1/PD-L1 expression (51). While we did not investigate the effects of PD-1/PD-L1 blockade on other immune cells, PD-1 blockade on tumor-associated M2 macrophages can directly reactivate phagocytosis and tumor immunity and may contribute to increased immunity rather than just reactivation of cytotoxic T-cells (52).

Paclitaxel might also directly activate cGAS/STING in tumor-extrinsic host cells. For example, paclitaxel has contributes to reprograming M2 macrophages to an M1 phenotype (53). In addition, there different components of the cGAS/STING pathway can polarize macrophages to an M1 phenotype. For example, 2’3’-cGAMP secreted by tumors may polarize M2 macrophages to an M1 phenotype (54). Thus, we recognize that paclitaxel could modulate immunity in myriad ways, in addition to activating cGAS, complicating its mechanistic interactions with immunotherapy. Further, TNBCs are a highly heterogeneous group of cancers and the mechanism delineated here may be limited to cancers with cGAS expression and other characteristics that promote potential paclitaxel-induced immunostimulatory factors.

Our study has some limitations. First, we intentionally selected cell lines of different TNBC subtypes, and evaluated the biology in a small number of patient samples. Nevertheless, it remains possible that our findings are limited to specific situations or that other modifying factors impact outcomes since triple-negative breast cancer is highly heterogeneous. Second, we conclude that the biology is principally mediated by the cGAS cytosolic DNA sensor based in part on knockout and knockdown experiments. Since the underlying cell types have genomic instability, these experiments risk having clone-specific effects. This risk is mitigated in that the knockout cells are polyclonal and in that we performed knockdown experiments in other cell lines. Finally, the M1 polarization of macrophages is an early step in immune response. It would be of value to delineate the conditions under which this step leads to recruitment of TILs and generate an anticancer response.

In conclusion, our experiments suggest that paclitaxel can generate micronuclei that parallels cGAS-STING activation in surviving post-mitotic TNBC cells. Paclitaxel can also induce polarization of macrophages to an M1 phenotype in a cGAS-dependent manner and may contribute to lymphocyte recruitment in some TNBC samples as well as better survival for patients on combination therapy. Further studies are needed to confirm our findings in vivo. Given the current lack of effective biomarkers in clinical studies, our findings demonstrate the potential value of cGAS expression in predicting response to standard-of-care and experimental treatments with combined MTAs and ICIs in TNBC.