Formyl-methionyl-leucyl-phenylalanine or fMLF (F3506), DMSO, and Pertussis toxin (PT, P7208) were purchased from Sigma-Aldrich. Human recombinant CXCL8 (CHM349), CXCL1 or GROα (CHM329) and GM-CSF (CYT-838) were purchased from ProSpec.

We used a panel of 13 different cell lines of breast epithelial origin in the study. Cell lines of different receptor expression status and malignant potential were cultured using medium and serum described in Table 1 . All cell lines were maintained at 37°C with 5% CO₂ except SUM149 cells, which were maintained with 10% CO₂ (30). All cell lines tested negative for mycoplasma contamination using the Mycoalert detection kit (Lonza).

Blood was obtained from healthy human male and female subjects aged 19–65 years, through the Platelet Pharmacology and Physiology Core at the University of Michigan. The Core maintains a blanket IRB for basic science studies, where HIPAA information is not required. Therefore, while the IRB-approved Core enrolls healthy subjects that conform to the protection of human subject standards, we did not have access to this information. The samples that we received were fully de-identified. Neutrophils were purified using dextran-based sedimentation followed by histopaque-based density gradient centrifugation, adapted from the method described previously (31, 32). Endotoxin free (< 0.005 EU/mL) cell culture grade water (SH30529, Cytiva) was used to dilute any reagents used for neutrophil isolation from the blood. All tubes used were coated with 0.5% tissue culture grade BSA (A9576, Sigma-Aldrich) diluted in DPBS in order to prevent non-specific cell activation by adhesion to the tube wall. Isolated neutrophils were cleared of residual red blood cells by ACK lysis buffer (10-548E, Lonza) and resuspended in mHBSS buffer (20mM HEPES, pH 7.4, 150mM NaCl, 4 mMKCl, 1.2mM MgCl2, and 10 mg/ml glucose). Viable cells were counted using 0.1% trypan blue (SV30084.01, Cytiva). Cells were more than 99% viable immediately following isolation. In order to address donor-to-donor variability of neutrophil response, cells were routinely tested for minimum basal activity and a robust response to fMLF stimulation before proceeding with daily studies. A dose dependent conversion from basal circular to elongated polarized morphology of fMLF stimulated neutrophils were confirmed by bright-field imaging of two randomly selected fields using 40x objective lens on a Zeiss Axiovert microscope.

Tumor spheroids were generated using the hanging drop method (33). Briefly, 1000 cancer cells resuspended in full serum media were seeded in 20 µl drops on the inside surface of petri dish lids. Lids were then inverted and placed onto media filled dishes. Spheroids were grown for 4 days in the incubator at 37°C. Day-4 spheroids, visible as aggregates at the bottom of each hanging drop, were harvested and used either for imaging or to generate TCM.

Tumor spheroids were imaged using 2-photon microscopy. Briefly, day-4 spheroids were fixed with 4% PFA for 1 hour at room temperature. Spheroids were stained with DRAQ5 (108410, Abcam) (1:1000) overnight at 4°C and then mounted onto poly-L-lysine coated coverslips. To reduce scattered light and improve image quality, the spheroids were optically cleared using the ultrafast optical clearing method (FOCM) (34). Lastly, the coverslips were mounted onto glass slides using ImmuMount (9990402, Thermo Scientific) with small spacers between the slides and the coverslips to prevent the spheroids from being flattened. The spheroids were imaged at 1040 nm using a C-Plan Apochromat 40x/1.3x oil objective on a Zeiss LSM 780. Image stacks were first processed by Histogram Matching with nTracer-AlignMaster (35) to equate brightness differences among slices within the same stack. A median filter with a radius of 0.5px was then applied to suppress the background noise. Center slices of each image stacks were used to representatively demonstrate the spheroids’ structure from each cell type. Movies traversing the image stacks from top to bottom were also generated.

To generate conditioned media (CM or TCM), cell lines were seeded at a density of 0.15X10⁶/ml in 2 ml (6-well tissue culture plate), 16 ml (T75 tissue culture flask), or 32 ml (T150 tissue culture dish) volume of complete medium for 24 hrs, at which point they reached ~ 70% confluence. The culture medium was then removed and cells were gently washed twice with calcium and magnesium free sterile DPBS (Gibco) to remove left over serum containing media. Cells were then incubated with fresh medium supplemented with reduced serum (1%) ( Figure 1A ) or without serum ( Figure 2A ) and incubated for an additional 24 or 48 hrs, respectively. The media was then harvested and filtered through 0.22 µm membrane filter to remove dead cell debris. Aliquots were frozen at -30°C until analyzed.

To generate TCM from cells grown in 3D spheroids, approximately 30 day-4 spheroids were placed in each well of poly-HEMA-coated, 96-well, V-bottom plates. Once settled to the bottom, spheroids were gently rinsed with DPBS and incubated in 200 µl serum-free media for 48 hrs. The media was then harvested and filtered through a 0.22 µm membrane filter. Aliquots were frozen at -30°C until analyzed.

For the analysis of actin polymerization and polarized morphology, neutrophils were seeded at 1-2x10⁶ cells/ml in 1 ml volume in a 35 mm dish with 20 mm #1.5 coverslip (MatTek). The coverslips were pre-coated with 25 µg/ml fibrinogen (F4883, Sigma-Aldrich) in DPBS for 1 hr at 37°C. Neutrophils were incubated 5-10 min in the incubator at 37°C to allow floating cells to settle down on the coverslip. Cells were then stimulated with an equal volume of TCM, media control, positive control fMLF or vehicle control DMSO and were further incubated for 15-20 min at 37°C. Following stimulation, the entire liquid medium was gently discarded from each dish and the adherent cells were immediately fixed with 4% PFA (Electron Microscopy Sciences), 0.1% glutaraldehyde (G5882, Sigma-Aldrich), 5% sucrose (BP220, Fisher Scientific), and 0.1M cacodylate (11652, Electron Microscopy Sciences), for 15 min at room temperature (RT). Once fixed, cells were washed with 0.1M glycine (G8898, Sigma Aldrich) in DPBS to quench the autofluorescence of glutaraldehyde. Washed cells were next permeabilized with 0.05% saponin (84510, Sigma Aldrich) in 3% BSA-containing DPBS for 15 min at RT and stained with phalloidin-TRITC (P1951, Sigma Aldrich) and DAPI (D9542, Sigma Aldrich). Cells were imaged using a 63x objective lens on a Zeiss LSM880 confocal microscope. Cells in three to five different fields across the coverslip were captured randomly per condition in each experiment. F-actin polymerization and cell circularity were measured using the ImageJ image analysis software. Briefly, integrated density, which is a product of cell area and mean fluorescence intensity, and circularity were obtained from ImageJ after manually identifying the boundary of each cell in a given field. Polarization index was calculated by normalizing the integrated density of each cells with respect to the positive control (fMLF) in each experiment.

To analyze cell circularity in response to TCM fractions or TCM in the presence of inhibitors, 0.15x10⁶ neutrophils were seeded in each well of a fibrinogen coated 8 well chambered cover glass with #1.5 cover glass (Cellvis). Cells were allowed to adhere to the cover glass by incubating at 37°C/5% CO₂ for 5 min followed by stimulation with an equal volume of TCM fractions, controls or TCM with or without the inhibitor for an additional 15 min at 37°C/5% CO₂. Bright-field images of the cells in two to three randomly selected fields per condition were captured using a 40x objective lens on a Zeiss Colibri microscope. Circularity of each cell was measured by identifying cell boundary manually using ImageJ software.

To quantify the ability of TCM to induce neutrophil migration, transwell migration chambers with membrane inserts of 3 µm diameter pore size (Corning and Greiner Bio-One) were used. Both the inserts and the bottom wells were coated with 2% tissue culture grade BSA for 1 hr at 37°C to prevent strong neutrophil adhesion. Coated inserts and wells were rinsed with DPBS twice to remove residual BSA. Freshly isolated neutrophils at a density of 4X10⁶/ml were seeded onto the inserts (100 or 200 µl), and placed in 24-well or 12-well plates, respectively. Control chemoattractants or TCM were gently added in 600 or 1200 µl volume to the bottom wells of 24-well or 12-well plates, respectively, avoiding any bubble formation between the insert and the liquid in the well. Migration was allowed to take place in 37°C/5% CO₂ for 2 hrs. The percentage of neutrophils migrated to the bottom chamber was calculated from the cell counts obtained using a hemocytometer.

Under agarose migration assay was performed following the protocol described elsewhere (36). Briefly, 0.5% SeaKem ME agarose (Lonza) in 50% each of DPBS and mHBSS was poured and allowed to solidify in 1% BSA-coated 35 mm glass bottom dish with 20 mm micro-well #1.5 cover glass (Cellvis). Using a metallic hole punching tool, three 1 mm diameter wells were carved out at a 2 mm distance from each other. Neutrophils were stained with 0.5 µM CellTracker Red CMPTX dye (Invitrogen) for 15 min at 37°C in rotation and re-suspended in mHBSS. In some experiments, cells were pre-treated with pertussis toxin (PT) for 2 hrs followed by staining in PT supplemented buffer. 7 μl (5×10⁴ cells) of stained cells were added to the wells at the edges and cells were allowed to migrate towards the center well containing 7 μl of either fMLF (100 nM) or 50 μg/ml of GM-CSF for 2 hrs at 37°C/5% CO₂. End point images of cells migrating towards the chemoattractant were acquired using a 10x objective lens (Zeiss Axiovert microscope).

Migration in 3D matrix of type I collagen was performed as previously described (37). Briefly, neutrophils were fluorescently labeled by staining with 0.5 µM CellTracker Red CMTPX dye (Invitrogen) for 15 min at 37°C in rotating condition. Collagen mix was prepared on ice by adding sodium bicarbonate (S8761, Sigma Aldrich), MEM (10X, M3024, Sigma Aldrich), and Purecol^® (3 mg/mL, 5005, Advanced Biomatrix) in a 1:2:15 ratio. Labeled cells were washed and resuspended in 1x mHBSS. 0.5x10⁶ labeled cells were gently added to the collagen mix such that the final concentration of collagen is 0.2%. The collagen-cell mix was then immediately added to fill two-thirds of a migration chamber made of coverslip and glass slide. The slide with the chamber was then placed in upright position in the incubator at 37°C/5% CO₂ for at least 20 min to allow for the gel to polymerize. The upper one-third empty space of the chamber was filled with chemoattractant, either TCM or controls such as media, CXCL8 or buffer. The chamber was immediately sealed with paraffin mix. Time-lapse video was captured for 40 mins at different positions (see Figure 3A ) across the boundary of the chemoattractant and collagen-cell mix using an environmental controlled Zeiss LSM 880 at 10x magnification with the pinhole open. The Manual Tracking plugin in ImageJ software was used to analyze the migration data, where about 70-100 cells were manually tracked over the 40 min time course in each of three randomly selected positions per condition in each experiment. For CXCL8 and buffer control, cells located 1 to 1.5 mm below the matrix boundary were tracked as cells near the boundary mostly exhibited random slow migration, possibly because high concentration of CXCL8 in the boundary saturates the receptors (CXCR1/2). In contrast, cells in the immediate vicinity of the boundary (~500 µm) showed maximum directional movement for TCM and were tracked for both TCM and media control. The software quantified the velocity and directionality of migration, where the directionality was represented by X-FMI, which reflects the efficiency of forward migration of the cells parallel to the gradient (38).

To evaluate neutrophil recruitment toward cancer cells in real time, a co-culture model system was developed ( Figure 5A ). Wells of a 24 well glass-bottom plate (P24-1.5H-N, Cellvis) were coated with 200 µg/ml fibronectin (F1141, Sigma-Aldrich) for at least 1 hr at 37°C. Approximately 20 spheroids resuspended in 250 µl fresh media ( Table 1 ) supplemented with reduced serum (1%) were plated in each well. Spheroids were allowed to attach to the plate for about 4 hrs in the incubator at 37°C/5% CO₂. Towards the end of incubation, neutrophils were primed with 5 ng/ml GM-CSF for 30 mins. 5x10⁴ or 1x10⁵ primed neutrophils were seeded to each well in 250 µl of 1% serum media. Multiple spheroids in each well were imaged for 24 hrs using an environmental controlled Zeiss Colibri microscope at 10x magnification. To analyze the migration of the neutrophils, the paths of at least 40 neutrophils per video, spaced out both spatially and temporally, were manually tracked using the Manual Tracking plugin in ImageJ software. The position data acquired from the tracking were used in a custom MATLAB script to quantify X-FMI.

Neutrophils were treated with pertussis toxin as previously described (39). Briefly, 4x10⁶ neutrophils/ml were pre-incubated in mHBSS with 10% heat inactivated fetal bovine serum without or with 500 ng/ml pertussis toxin for 2 hours at 37°C with gentle rotation. After 2 hrs, the cells were centrifuged and the pellet was resuspended in mHBSS with 0.5% BSA with or without pertussis toxin. The cells were then loaded into transwell inserts and their migration was assessed as described above.

30 ml serum free TCM was collected from M4 and MDA-MB-231 cell lines as described above. TCM was concentrated 100-120 fold using Amicon Ultra-15 centrifugal tubes (Millipore Sigma) of 3 kDa molecular weight cut-off. Briefly, 15 ml TCM was centrifuged at 4000 rpm for 40 min at RT. Filtrates were discarded and an additional 15 ml TCM was added to the resulting concentrate, mixed well and centrifuged as before. The final concentrate was buffer exchanged twice with sterile DPBS to get rid of the residual media components and frozen at -30°C until use. Concentrated TCM was applied to a 30 cm x 10 mm Superdex 75 10/300 column (GE Life Sciences) equilibrated with 30 ml DPBS. The column was calibrated using four standards: aprotinin MW 6.5 kDa (A3886, Sigma-Aldrich), cytochrome c MW 12.4 kDa (C7150, Sigma Aldrich), carbonic anhydrase MW 29 kDa (C7025, Sigma-Aldrich), and bovine serum albumin MW 66 kDa (A8531, Sigma Aldrich). Sample was run at a rate of 0.2 ml/min and fractions of 150 µl were collected using DPBS as an eluent. Fractions were stored at 4°C until use.

TCM concentrates were sent to RayBiotech, Inc. (GA) to screen for a panel of potential neutrophil chemoattracting factors encompassing CXCL chemokine family members CXCL1/2/3 aka GRO, CXCL5, CXCL6, CXCL8, CXCL12, CCL family members CCL2, CCL3, CCL5, growth factors G-CSF, GM-CSF, and cytokine TGF-β. The presence of the analytes was detected and quantified using a high throughput custom Quantibody multiplex enzyme-linked immunosorbent assay (ELISA) (RayBiotech) and the data were analyzed by RayBiotech, Inc. Briefly, concentrates were prepared as described above using 60 ml serum free TCM harvested from the respected cell lines such that the protein concentration approximated 1-2 mg/ml, as estimated using the Pierce BCA protein assay kit (Thermo Fisher). Concentrated samples were tested without dilution in quadruplicate, as per the manufacturer instructions. The protein concentration for each target analyte was quantified by comparing the fluorescence signal from the sample with the individual standard curve generated by known concentrations of each protein.

ERK phosphorylation was assessed using western blotting analysis. In brief, neutrophils were pretreated with 2 mM pefabloc protease inhibitor (Sigma Aldrich) for 15 min at 37°C with gentle rotation. 4x10⁶ neutrophils/ml were seeded in 500 µl in each well of a 12 well tissue-culture plate pre-coated with 25 ug/ml fibrinogen. Cells were allowed to settle down in the well by incubating the plate for 5 min at 37°C. Cells were then exposed to an equal volume of TCM, media control, positive controls including fMLF and CXCL1 or the corresponding vehicle controls. After 5 min stimulation at 37°C/5% CO₂, the plate was placed on ice and the cells were lysed in 1X Laemmli buffer (Biorad) containing 5% β-mercaptoethanol (VWR), phoSTOP (Millipore Sigma) and Halt™ protease and phosphatase inhibitor cocktails (Thermo Scientific) at 95°C for 10 min. An equal volume of each was resolved by 10% SDS–PAGE, transferred onto PVDF membranes (Millipore), blocked with 5% nonfat dry milk for 1 hr and probed with primary Abs: phospho-p44/42 MAPK (ERK1/2) (Thr202/Tyr204, dilution 1:2000, 4370, Cell Signaling Technology) and alpha tubulin (dilution 1:2000, 66031-1-Ig, Proteintech). Bands were visualized using HRP-conjugated secondary antibodies (dilution 1:8000, Jackson ImmunoResearch), SuperSignal™ West Pico PLUS Chemiluminescent Substrate, and a C600 digital imaging system (Azure). Integrated density for each band was measured using ImageJ software.

Neutrophils were pre-treated with the CXCR2 receptor antagonist AZD5069 (Cayman Chemical), the TGF-β1 receptor ALK5 antagonist SB431542 (Cayman Chemical), or the SMAD3 phosphorylation inhibitor SIS3 (Cayman Chemical) at 1 µM, 1 µM and 3 µM, respectively, or with vehicle control DMSO for 30 min at 37°C with gentle rotation before they were added to the polarization or transwell migration assay.

Cancer cell membrane integrity was analyzed for the detection of cell death using the LDH cytotoxicity assay kit (Thermo Scientific) based on the manufacturer instructions. Briefly, cancer cells were seeded at 0.15x10⁶ cells/ml in 100 µl volume in a 96-well tissue culture dish in two sets of triplicate wells. The control set was for maximum LDH release by subjecting cells to undergo complete lysis, and the test set was for spontaneous LDH release. Cells were subjected to incubation and subsequent media change as described for generating serum free TCM. Three days after initial seeding, 10 µl of 10x lysis buffer and 10 µl of sterile water were added to the control and test set, respectively. After incubating the plate for 45 min in the incubator at 37°C/5% CO₂, 50 µl from each well was transferred to a fresh 96 well plate, mixed with 50 µl LDH reaction mix, and incubated for additional 30 min at RT protected from light. Finally, the reaction was stopped by adding the stop solution and absorbance was measured at 490nm and 680nm by a SpectraMax M5 Multi-Mode Microplate Reader (Molecular Devices). Absorbance was corrected by subtracting background signal of the instrument followed by correcting for the absorbance values of corresponding media. The percentage of maximum LDH release was calculated with respect to corresponding control set.

Endpoint migration and invasion assays were performed using a transwell system in 24-well plates (353504, Falcon). FluoroBlock™ filter inserts (351152, Corning) with 8 μm pore size were used that block light transmission from 400-700 nm and allow fluorescence detection only from the cells that migrated/invaded to the bottom side of the inserts with a bottom-reading fluorescence plate reader. The inserts were uncoated for the migration assays, and coated for the invasion assays with 0.2 mg/ml type I bovine collagen PureCol ^® of which the pH was neutralized to 7.2 – 7.4 using sodium bicarbonate. Cancer cells that were serum starved for 24 hrs were seeded onto the inserts at 0.5x10⁵ cells/ml in 100 µl of serum free media. Cells were allowed to migrate toward the bottom chamber containing 500 µl of full-serum media as the chemoattractant or serum-free media as the negative control. After a 24 hr incubation at 37°C/5% CO₂, the inserts were transferred to a fresh 24-well plate with black walls containing 500 μl of 4 μM Calcein AM (Biotium) in mHBSS per well. Cells were incubated for 1 hr at 37°C, and the fluorescence reading of migrated/invaded cells was measured from the bottom at wavelengths of 495/515 nm (Excitation/Emission) by a SpectraMax M5 Multi-Mode Microplate Reader (Molecular Devices).

The transcript expression levels of different CXL chemokine family members in breast cancers were analyzed using datasets identified from the cancer microarray database Oncomine (http://www.oncomine.org). A total of 7 datasets from independent studies were selected from a list of studies obtained by setting the search thresholds as follows: Data type, mRNA; gene rank, top 10%; fold change, 2; P−value, 0.0001. Studies that used Affymetrix platforms other than hgu133plus2 and hgu133a were excluded from further analysis. For each of the studies analyzed, data were downloaded from GEO per instructions of the respective publications. Farmer (40), Minn (41), Wang (42), Ivshina (43) used the Affymetrix hgu133A array while Miyake (44), Schuetz (45), Kao (46) used the Affymetrix hgu133plus2.0 array. For the Schuetz dataset (45), we could not get the raw CEL files and subsequently started from the expression matrix provided with the GEO download. CEL files for all other projects were processed using the Robust Multi-array Average (RMA) technique to obtain normalized expression values (47). Estrogen receptor annotation was obtained from the data download or from Ivshina (43), Wang (42), Minn (41), Miyake (44), Schuetz (45), and Farmer (40) datasets. Estrogen receptor annotation was not provided for Kao (46). Instead, using simple histograms, the cutoff for ER status was set by visually selecting a point between the bimodal distribution of ER expression; that cutoff was 10. The Limma package of Bioconductor was used to fit linear models to each dataset comparing ER- samples to ER+ (48). For Schuetz (45), the model controlled for variation from ductal carcinoma in situ (DCIS) and invasive ductal carcinoma (IDC) status. No other study had a co-variate considered. Data for CXCL chemokines were extracted and the log2 fold change between ER- and ER+ were plotted in a heatmap along with the fold change for the datasets respective estrogen receptor 1 (ESR1) gene (probeset 205225_at). Samples (columns) were sorted by ESR1 expression. From the TCGA data portal, we used the search “Project id” IS “TCGA-BRCA” AND “Workflow Type” IS “HTSEQ-FPKM” AND “Data Category” IS “Transcriptome Profiling” to find 1,222 FPKM (fragments per kilobase of exon model per million reads mapped) data files. The log2 of the FPKM values were plotted for our chemokines of interest and samples were ordered by the FPKM of ESR1.

GraphPad Prism software was used for data plotting and conducting statistical analysis by tests that are described in the respective figure legends along with the size of the samples. Tests used included two-tailed paired t test, unpaired t test, 1-way ANOVA with Dunnett’s multiple comparisons test or 2-way ANOVA with Sidak’s multiple comparisons test.

Numerous studies have suggested that soluble factors, such as CXCL chemokine family members, promote neutrophil infiltration in a variety of tumors. For example, CXCL members CXCL8 (aka IL-8) and CXCL1/2 have been implicated in driving neutrophil recruitment in breast cancer (49–52), and overexpression of these chemokines is associated with breast tumor samples and breast cancer cell lines that have negative estrogen receptor expression status and higher aggressiveness (50, 53–55). We confirmed these observations by comparing the expression pattern of all known CXCL chemokines across breast cancer subtypes that differ in ER expression status by analyzing transcriptome-based expression data identified from Oncomine and TCGA databases ( Figure S1 ). Out of 12 CXCL chemokines, we found that the expression of CXCL8, CXCL1, and CXCL2 transcripts was significantly elevated in ER- breast cancer specimens compared to ER+ samples in seven, four and three independent studies, respectively ( Figure S1C ). Similarly, TCGA data analysis revealed elevated expression patterns of CXCL8, CXCL1, and CXCL5 in ER- breast cancer specimens ( Figure S1D ).

The differential chemokine expression pattern detected among breast cancer subtypes prompted us to investigate the neutrophil recruiting ability of a panel of breast cancer cell lines that differ in ER expression status and malignant potential. TNBCs and ER+ breast cancer cell lines along with non-malignant M1 and metastatic derivative M4 cells from the MCF10A breast cancer progression model (56–59) ( Table 1 ) were cultured in vitro under identical seeding cell density. The higher malignant potential of the TNBC cell lines compared to the ER+ breast cancer cell lines was confirmed by assessing their migration and invasion abilities ( Figures S2A, B ). We harvested TCM after growing cells in serum reduced media for 24 hrs ( Figure 1A ), and tested the ability of the TCM to activate neutrophils. Neutrophils stimulated with fMLF, which is a formylated bacterial tripeptide and a potent neutrophil chemoattractant (32), exhibited an elongated morphology with strong F-actin accumulation in the protrusions, detected using fluorescently tagged phalloidin (32) ( Figure 1B ii; quantified in Figures 1C, D ). In contrast, the majority of neutrophils exposed to DMSO control maintained a circular shape similar to resting neutrophils, and a uniform distribution of cortical actin ( Figure 1B i). Similar to fMLF, TCM derived from metastatic M4 cells induced strong polarized morphology and F-actin depositions at the poles ( Figure 1B v; quantified in Figures 1C, D ), while most of the cells stimulated with conditioned media (CM) collected from M1 cells remained circular ( Figure 1B iv; quantified in Figures 1C, D ). In addition, we found a significant decrease in circularity and increase in F-actin intensity of neutrophils stimulated with TNBC BT549 cell-derived TCM, while TCM derived from ER+ BT474 cells failed to induce such changes ( Figures 1C, D ). Next, we assessed the ability of the TCM to induce neutrophil migration using a transwell setup ( Figure 1E ). Similar to the polarization response, we found that a significant proportion of neutrophils migrated toward TCM derived from metastatic M4 and TNBC cells, but not from ER+ cells ( Figure 1F ). Because each breast cancer cell line has a unique genetic background, we also included in our panel bone-, lung- and brain-specific highly metastatic MDA-MB-231 cells selected from the same parental MDA-MB-231 TGL cells (41, 60, 61). Interestingly, neutrophil migration towards TCM derived from lung- and bone-specific MDA-MB-231 cells was significantly higher than towards TCM from the relatively less metastatic MDA-MB-231 TGL parental line ( Figure 1F ).

To minimize neutrophil activation by the presence of serum in the TCM and to mimic the lack of nutrient availability as encountered by cancer cells in an in vivo tumor mass, we harvested TCM from cells cultured in 0% serum for 48 hrs ( Figure 2A ) – conditions that lead to minimal cell death ( Figure 2B ). We found that serum free TCM derived from M4 and MDA-MB-231 cells induced strong neutrophil migration in transwell assays compared to the TCM harvested from MCF-7 cells cultured under the same conditions ( Figure 2C ). Notably, a greater percentage of neutrophils migrated toward serum free TCM ( Figure 2C ) than to serum reduced TCM ( Figure 1F ) derived from either M4 (57% vs. 49%) or MDA-MB-231 (42% vs. 26%) cells. Importantly, a gradual decrease in neutrophil recruitment with serially diluted serum free TCM showed that TCM triggers a dose-dependent response ( Figure 2D ). Together, these findings establish that TCM derived from breast cancer cells with triple negative receptor expression status and aggressive malignant potential induce neutrophil polarization and robust migration by secreting specific chemotactic factors. For subsequent studies, we used TCM generated under serum starved conditions from all cell lines.

We next assessed the ability of TCM to induce neutrophil migration in a more physiologically relevant interstitial tissue-like microenvironment by monitoring neutrophil motility through 3D matrices of type I collagen and time-lapse imaging ( Figure 3A ). By tracking the movement of neutrophils, we generated trace plots ( Figure 3B ) and rose plots ( Figure 3C ) that depict the behavior of cells in response to M4 TCM or media control. CXCL8 was used as a positive control. We found that neutrophils migrated directionally towards M4 TCM compared to media control ( Figure 3B iii, 3Biv, 3Ciii, 3Civ; see also Movies S1 , S2 ). We also found that neutrophils migrated towards M4 TCM with a significantly elevated velocity and directionality, compared to media control ( Figures 3F, G ). These increases in cell velocity and directionality also occurred in response to CXCL8 ( Figure 3B i, 3Bii, 3Ci, 3Cii, 3D&3E; see also Movie S3 , S4 ). However, a lower directionality was measured in response to M4 TCM (0.1) compared to CXCL8 (0.45), which could reflect the complexity of TCM composition. Indeed, unlike a single chemoattractant like CXCL8, the TCM includes multiple active factors that can potentially modulate their effect and have different diffusibility through the collagen. Taken together, these results indicate that active factors secreted from TNBC cells induce directional migration of neutrophils.

We investigated the neutrophil recruitment ability of 3D tumor spheroids generated with breast cancer cell lines. First, we generated tumor spheroids with diameters ranging from 200 to 350 µm using the hanging drop method. Using 2 photon microscopy, we assessed the morphology of the spheroids, which differ in appearance, ranging from irregular, loose aggregates (MDA-MB-231 and BT549) to tight, symmetrical spheres (BT474 and M4) ( Figure 4A , Movies S5 – S8 ). We also determined that under our growth conditions, the spheroids did not harbor a hollow or necrotic core at their center ( Figure 4A ), ruling out the possible release of damage-associated molecular patterns (DAMPs) from damaged or dying cells at the spheroid core that may impact neutrophil recruitment neutrophil recruitment (62) ( Figure 4A , Movies S5 – S8 ).

To evaluate whether TNBC spheroids secrete neutrophil recruiting factors similarly to TNBC cells grown in monolayers, we assessed neutrophil migration toward spheroid-derived TCM and found that spheroid-derived TCM from all three TNBC cell lines (M4, MDA-MB-231, and BT549) had neutrophil recruiting activity, while spheroid-derived TCM from the ER+ cell line (BT474) was inactive ( Figure 4B ). These results establish that TNBC cells secrete neutrophil-recruiting factors in either 2D or 3D environments.

Next, we developed a neutrophil-tumor spheroid co-culture system to monitor the migration of neutrophils towards tumor spheroids by time-lapse imaging ( Figure 5A ). During the early hours of imaging, when neutrophils are not activated and therefore not attached to the surface, we noted that the neutrophils followed the fluid flow and some of them non-specifically got trapped at the boundary of the tumor spheroids ( Movies S9 , S10 ). However, over time, activated neutrophils clearly attached to the surface, actively migrated towards the M4 spheroids, and accumulated around the spheroid boundary ( Figures 5B , Movie S9 ). In contrast, we found that neutrophils never strongly adhered at the boundary or directionally migrated towards BT474 spheroids. Instead, they followed the direction of the flow in the system ( Figure 5C , Movie S10 ). Indeed, quantification of the directionality of neutrophil migration towards the spheroid revealed a higher directionality towards M4 spheroids compared to BT474 spheroids ( Figure 5D ).Taken together, these findings further affirm that TNBC cell lines induce directional migration of neutrophils by actively secreting specific chemotactic factors.

One of the major signaling pathways triggered in neutrophils by a wide range of activating factors, including chemoattractants, is the extracellular signal–regulated kinase (ERK) pathway (63, 64). ERK pathway is also important for activating the migration machinery of neutrophils (65). We therefore compared the ability of the secreted factors from different breast cancer cell lines to activate ERK by monitoring its phosphorylation status. We found that TCM derived from either M4 or MDA-MB-231 cells induced ERK phosphorylation, with M4 showing a more dramatic response (fMLF and CXCL1 were used as positive controls) ( Figures S3 and 6A ). Importantly, we measured no stimulation with TCM derived from the ER+ MCF-7 cells ( Figures S3 and 6A ).

To identify the nature of the activating factors driving neutrophil migration, we first used multiplex ELISA to quantify the relative presence of potential neutrophil chemoattractants/activators associated with cancer, including a number of CXCL and CCL chemokines, cytokines, and growth factors (1, 66) in TCM derived from M4, MDA-MB-231, and ER+ MCF-7 cell lines ( Figure 6B ). Among the CXCL chemokines associated with breast cancer of ER- status (GRO, CXCL5, CXCL6, and CXCL8) and breast cancer metastasis (GRO, CXCL5, CXCL6, CXCL8, and CXCL12) (67, 68), we found a greater than 500-fold increase in GRO chemokines (CXCL1/2/3) present in the TCM harvested from TNBC cells, compared to TCM harvested from ER+ cells. This high amount of GRO secretion by TNBC cells is in agreement with the robust expression of CXCL1 mRNA in ER- breast tumor specimen compared to ER+ ones [ (55) and Figure S1 ]. We detected a greater amount of CXCL8 present in the TCM from TNBC cells compared to the TCM from ER+ cells, which is in accordance with previous reports (53, 54). However, compared with GRO chemokines, the fold change in CXCL8 expression was more modest, suggesting that CXCL8 is not the key factor involved in neutrophil activation. In addition, we detected very little to no changes for CXCL5 and CXCL6 in any of the TCM. We also measured > 9-fold increase in CXCL12 in the TCM derived from M4 and MDA-MB-231 cells, compared to MCF-7 TCM. In contrast, the CCL family members were in general less abundantly secreted by breast cancer cells. We detected a 5-fold greater abundance of CCL2 in TCM from TNBC cells than ER+ cells, while very little to no CCL3 in any TCM. In contrast, CCL5 was secreted in equivalent amounts by all breast cancer cells, irrespective of the receptor expression status or malignant potential. Other than chemokines, we detected G-CSF and GM-CSF more abundantly in TCM harvested from TNBC cells compared to ER+ cells. Finally, we found that TGF-β1 was 12-30 fold more abundant in TCM harvested from TNBC compared to ER+ cells. Taken together, these data show that compared to ER+ breast cancer cells, TNBC cells overexpress a number of potential neutrophil recruiting factors, the most abundant of which being the CXCR2 specific GRO chemokines and TGF-β1.

To narrow down our search for tumor-secreted active factors, we used a combination of biochemical and pharmacological approaches. We set out to fractionate the TCM based on the molecular weight of their components and identify the fraction with optimum activity. We subjected the concentrates (>3 kDa) of MDA-MB-231 TCM to size exclusion chromatography ( Figure 7A ) and tested the resulting fractions for their ability to induce polarized morphology in neutrophils. We found that the fractions carrying proteins corresponding to ~12 kDa molecular weight (fraction “g” in Figure 7A ) induced maximum polarization as indicated by the elongated shape of the cells ( Figure 7B iv). In contrast, most of the cells retained a circular shape when stimulated with fractions containing proteins that were either larger ( Figure 7B ii&iii) or smaller than ~12 kDa ( Figure 7B v&vi). Indeed, when we quantified the extent of polarization induced by a series of fractions, we found that circularity was significantly reduced when exposed to MDA-MB-231 TCM itself or TCM-derived fractions corresponding to 12 kDa sized proteins ( Figure 7C , fractions e,f,g,h). A similar pattern of polarized morphology was also induced by M4 TCM derived fractions containing ~ 12 kDa sized proteins ( Figure S5 ). Importantly, both GRO chemokines (69) and monomeric TGF-β (70) have a molecular weight of 12 kDa.

Chemokines activate G protein-coupled receptors (GPCR) by signaling through Gi heterotrimeric G proteins, which are sensitive to pertussis toxin (PT) inhibition (71). To determine the role of chemokines in TCM induced neutrophil migration, we therefore, targeted chemokine specific GPCR Gi signaling by pretreating neutrophils with PT and measured the ability of cells to migrate towards TCM or CXCL8 as a positive control. Using transwell assays, we found that significantly fewer neutrophils migrated towards TCM derived from either M4 or MDA-MB-231 cells in the presence of PT ( Figure 7D ), indicating a role for chemokines in TCM-induced neutrophil recruitment. PT treatment was not toxic, as it did not alter the migration of neutrophils towards the growth factor GM-CSF ( Figure S4 ). Taken together, these data suggest that chemokines actively contribute to TCM induced neutrophil migration.

In order to specifically determine the identity of the mediators, we studied TCM induced neutrophil migration in the presence of different inhibitors targeting either the GRO receptor CXCR2 [AZD 5069 (72)] or the TGF-β1 receptor ALK5 [SB431542 (73)], or both. The presence of both inhibitors did not alter neutrophil migration towards fMLF, showing that the inhibitors are receptor specific ( Figure 8A ). We found that neutrophil migration induced by M4 TCM remained intact when cells were pretreated with either inhibitor alone ( Figure 8A ). In contrast, we noted a significant decrease (~38%) in the percentage of migrating neutrophils towards M4 TCM when cells were pretreated with both inhibitors, indicating that CXCR2 ligands and TGF-β mediate the effects of the TCM ( Figure 8A ). Similarly, we found a significant inhibition of M4 TCM induced neutrophil polarization as reflected by the increase in circularity in the presence of both the inhibitors compared to TCM alone ( Figure 8B ). However, neutrophil migration induced by MDA-MB-231 derived TCM was not significantly decreased by the inhibitors ( Figure 8A ).

To further elucidate the role of TGF-β signaling in mediating the effect of TCM on neutrophils, we targeted SMAD3, a downstream signaling component of TGF-β signal transduction pathway (74). Neutrophils pretreated with SIS3 [an inhibitor that specifically inhibits SMAD3 phosphorylation (75)] showed intact migration in response to M4 TCM ( Figure 8C ). However, when cells were pretreated with inhibitors targeting CXCR2, TGF-β receptor, and SMAD3 phosphorylation, we observed a remarkably stronger inhibition (~60%) of neutrophil migration towards M4 TCM ( Figure 8C ) than the inhibition (~38%) measured with CXCR2 and TGF-β receptor inhibitors ( Figure 8A ). Importantly, we also found a significant inhibition (~40%) of neutrophil migration towards MDA-MB-231 TCM in the presence of all three inhibitors, compared to vehicle control ( Figure 8D ). Together, these results indicate that a concerted action of secreted TGF-β and GRO specific chemokines plays a key role in TCM induced neutrophil recruitment.