Cells were maintained at 37°C in a 5% CO₂ incubator. Human MDA-MB-231 TNBC cells were purchased from ATCC (Manassas, VA, USA) and cultured in DMEM supplemented with 10% FBS and 1% penicillin/streptomycin. The E-cadherin^high MDA-MB-231 cell line variant was generated and maintained as described by Sulaiman et al. (30). Human SUM149-PT TNBC cells were purchased from Asterand (Detroit, MI, USA) and cultured in Ham’s F-12 supplemented with 5% FBS, 1% penicillin/streptomycin, 5 µg/mL insulin, 1 µg/mL hydrocortisone, and 10 mM HEPES. Mouse AT3 TNBC cells of C57BL/6 origin and transduced with retroviral vectors expressing chicken OVA cDNA (AT3^ova) were purchased from the Peter MacCallum Cancer Institute (East Melbourne, AU). AT3^ova cells were maintained in DMEM supplemented with 10% FBS and 1% penicillin/streptomycin. Cells were routinely tested for mycoplasma using PCR. All relevant reagents and their sources are listed in Supplementary Table 1.

The docking software Molecular Operating Environment (MOE) purchased from Chemical Computing Group Inc (Montreal, QC, CA). The CD73 protein structure (accession code: 4H2G) was downloaded from Protein Data Bank and optimized for hydrogens and lone pairs within MOE. Natural compound structures were downloaded as potential ligands from PubChem in the SDF file format. The docking score, protein-ligand interactions, and the respective energies released from the interaction were generated and recorded using MOE Align/Superpose functions.

Cells were seeded into 6-well plates (3.0 x 10⁵ cells/well). Total RNA was extracted 72 hours after treatment using the Qiagen RNeasy kit (Toronto, ON, CA). cDNA was obtained from mRNA using the iScript cDNA Synthesis Kit purchased from Bio-Rad (Hercules, CA, USA), as per manufacturer instructions. Gene expression levels were determined using the Bio-Rad MyiQ real-time PCR system in a reaction mixture consisting of 50% SyBr Green, 37.5% RNAse-free water, 5% of forward/reverse primers, and 2.5% cDNA. Specific forward and reverse gene primers are listed in Supplementary Table 2. Results were normalized to 18S or GAPDH housekeeping genes, and relative fold changes in gene expression was calculated using the ^2ΔΔCT method.

Cells were seeded into 6-well plates (3.0 x 10⁵ cells/well). Harvested cells were filtered through a 40 um strainer and suspended in PBS supplemented with 2% FBS and 2 mM EDTA 96 hours post-treatment. Non-specific binding was reduced with mouse anti-human IgG Fc from Thermo Fisher (Waltham, MA, USA) or rat anti-mouse CD16/CD32 from BD Biosciences (Franklin Lakes, NJ, USA). Cells were then incubated with fluorescently labeled antibodies (Supplementary Table 3). For human cells, apoptosis was determined 7-aminoactinomycin (7-AAD) purchased from eBioscience (San Diego, CA, USA). For mouse cells, apoptosis was determined prior to incubation with fluorescently labeled antibodies using the LIVE/DEAD fixable far red dead cell stain kit from Thermo Fisher (Waltham, MA, USA). ALDH activity was determined as per manufacturer instructions using the ALDEFLUOR kit from StemCell Technologies (Vancouver, BC, CA). Flow cytometry was performed on LSRFortessa (BD Biosciences) and data was analyzed using FloJo software (Ashland, OR, USA).

Cells were seeded into 96-well plates (5.0 x 10³ cells/well). Cells were washed twice with pre-warmed phosphate-free buffer (distilled water supplemented with 2 mM MgCl₂, 125 mM NaCl, 1 mM KCl, 10 mM glucose, and 10 mM HEPES up to pH 7.2) 48 hours post-treatment. Cells were then incubated at 37°C with 100 µM adenosine 5’-(α,β-methylene)diphosphate (APCP) or 250 µM AMP for 10 minutes. Resultant media phosphate concentrations were assessed as per manufacturer’s instructions using the Malachite Green Phosphate Detection Kit, purchased from R&D Systems (Minneapolis, MN, USA).

Cells were seeded into 24-well plates (1.5 x 10³ cells/well). A viability analysis was performed 120 hours post-treatment with a 3-hour incubation at 37°C using 10% 3-(4,5-Dimethylthiazol-2-Yl)-2,5-Diphenyltetrazolium Bromide (MTT, 1 mg/mL). The supernatant was aspirated, and formazan crystals were dissolved in DMSO. Resulting absorbance was measured at 570 nm.

Cells were seeded into 24-well plates (1.0 x 10⁵ cells/well). Cells were transfected with 1000 ng of either Wnt M50 Super 8x TOPFlash-luciferase (Addgene Plasmid #12456) or YAP 8xGTIIC-luciferase (Addgene Plasmid #34615), in conjunction with 1000 ng of Renilla pRL-SV40P (Addgene Plasmid #27163) using Lipofectamine 3000 purchased from Invitrogen (Carlsbad, CA, USA). Cells were treated 24 hours post-transfection and lysed for Firefly and Renilla luciferase activity 48 hours post-treatment using the Dual-Luciferase Reporter Assay System, purchased from Promega (Madison, WI, USA).

Cells were resuspended in 1:1 DMEM:F-12 supplemented with 2% B27, 1% sodium pyruvate, 1% penicillin/streptomycin, 20 ng/mL basic fibroblast growth factor, and 20 ng/mL epidermal growth factor and seeded into 96-well ultra-low attachment plates (1.5 x 10³ cells/well). Tumorspheres were counted (>100 µm diameter) and representative pictures were captured 10 days post-treatment using the Zeiss Axiovert 40 CFL microscope purchased from Carl Zeiss AG (Feldback, Switzerland). A cell viability assay was also performed as previously described.

The TNBC patient-derived xenograft samples HCI-001, HCI-002, HCI-015, and HCI-016 were obtained from and characterized by the University of Utah (33, 34). Extracted tumors were sliced with a scalpel to obtain 2 mm x 2 mm tumor fragments. These fragments were cultured in 48-well plates with 1:1 DMEM:F-12 supplemented with 10% FBS, 1 ug/mL insulin, 0.5 ng/mL hydrocortisone, 3 ng/mL epidermal growth factor, and 1% penicillin/streptomycin. Tumor fragments were treated, and viability was assessed daily with 10% resazurin sodium salt. Media fluorescence intensity was measured at 560 nm excitation and 530 nm emission following a 3-hour incubation at 37°C.

C57BL/6 wild-type (5-6 week old, female) mice were purchased from the Charles River Laboratory. Tumors were established in the mammary fat pads using 1 x 10⁶ AT3^ova cells resuspended in Matrigel purchased from BD Biosciences (Mississauga, ON, CA). Established AT3^ova tumors (3 mm x 3 mm) were treated intraperitoneally with paclitaxel (P, 10 mg/kg), luteolin (L, 20 mg/kg), quercetin (Q, 5 mg/kg), or adenosine 5’-(α,β-methylene)diphosphate sodium (APCP, 20 mg/kg) where indicated. Tumors were measured every 2 days using calipers. Mice were humanely euthanized once tumors reached a mean volume of 950 mm³. Tumors were minced with scissors and enzymatically digested with collagenase/hyaluronidase from StemCell Technologies (Vancouver, BC, CA) and DNAse I from Sigma Aldrich (St Louis, MO, USA).

Data are expressed as mean ± standard deviation (SD) or standard error (SE), where indicated, without data transformation and outlier exclusion. For relative comparison, the data were normalized to control group and then compared as indicated in each figure. Data distributions were tested by one-way ANOVA. When appropriate, statistical differences between groups were assessed by unpaired Student’s t-test or Mann-Whitney test (comparison of 2 groups). Statistical tests were completed using GraphPad Prism 9 software (GraphPad, San Diego, CA). Unless stated otherwise, experiments have a minimum of three biological repeats.

Targeting enriched CSCs in TNBC post-chemotherapy remains challenging, owing to limited knowledge of immunotherapeutic targets. To identify novel targets, we adopted the METABRIC RNA-seq dataset to compare 165 TNBC samples post-chemotherapy with 2344 non-TNBC breast cancer samples post-treatment (chemotherapy/hormone therapy/radiotherapy). Components of the adenosinergic signaling pathway were identified as potential clinical targets, including CD73 (Figure 1A) and adenosine receptor ADORA2B (Figure 1B). Given the limitation of current CD73 inhibitors for clinical translation, we used quantitative structure-activity relationships (QSAR) from high throughput in silico molecular docking between CD73 and natural compounds for virtual screening. Using the CD73 inhibitor APCP as a positive control (Figure 1C), we found that the flavonoids quercetin (Figure 1D) and luteolin (Figure 1E) exhibited relatively high docking scores, indicating a notable specificity for CD73 (PDB# 4H2G), with quercetin yielding extensive molecular interactions (Supplementary Table 4). Of note, quercetin and luteolin have been safely used as over-the-counter medicine/supplements.

To validate our in silico findings in vitro, we tested clinically relevant doses of quercetin (35) and luteolin (36) with the chemotherapeutic, paclitaxel, on CD73 activity. By measuring media phosphate concentrations normalized to cell count in the presence of AMP substrate, we found that 10 µM quercetin potently inhibited paclitaxel-upregulated CD73 activity, comparable to paclitaxel combined with 100 µM APCP in human TNBC MDA-MB-231 (Figure 1F) and SUM149-PT (Figure 1G) cell lines. Strikingly, quercetin, but not luteolin, antagonized the paclitaxel-mediated upregulation of CD73 activity (Figures 1F, G), highlighting the efficacy of quercetin as a clinically translatable CD73 inhibitor.

Next, we employed flow cytometric analysis to confirm that the observed reduction in CD73 activity by quercetin reduced paclitaxel-enriched CD73 protein expression in TNBC cells. The CD73 inhibitor APCP in combination with paclitaxel was used as a positive control. We found that quercetin did not lower CD73 expression but reduced the frequency of living CD73^high cells alone and antagonized the paclitaxel-mediated enrichment of living CD73^high cells in combination (Figures 1H, I; Supplementary Figures S1A, B). Similar results were obtained from the chemo-resistant mouse AT3^ova TNBC cells (Supplementary Figure S1C) and TNBC PDX organotypic cultures (Figure 1J), implying that clinically achievable doses of quercetin may effectively reduce CD73^high tumor cells post-chemotherapy.

Given that luteolin did not inhibit CD73 activity, we then tested its inhibitory efficacy on TNBC cell viability to validate the three-drug combination. Paclitaxel used in the clinical treatment of TNBC, and paclitaxel combined with CD73 inhibitor APCP were taken as positive controls. The combination of quercetin, luteolin, and paclitaxel was most effective at killing human MDA-MB-231 (Figure 2A) and SUM149-PT (Figure 2B), and chemo-resistant mouse AT3^ova (Figure 2C) TNBC cell lines in MTT viability assays. The triple-drug combination also reduced cell viability more significantly than any other dual-combinations (Supplementary Figures S2A–C). In contrast, the inclusion of quercetin and luteolin in the triple-drug combination did not lead to a reduction in the viability of MCF10a cells (a non-tumorigenic mammary cell line) beyond that observed with paclitaxel alone (Supplementary Figure S2D). Interestingly, the chemo-resistant mouse AT3^ova cells were sensitized to treatment, mainly by luteolin. Using the Chou-Talalay method, synergism was also observed in the triple-drug combination using the clinically relevant doses of each drug (Supplementary Figures S2E, F).

Since MTT cell viability assays are metabolic activity-dependent, we performed flow cytometry and confirmed that the triple-drug combination was the most effective in increasing frequencies of late apoptotic and dead TNBC cells (Figures 2D–F). Furthermore, we found that clinically relevant doses of the triple-drug combination significantly reduced ex vivo viability of four distinct TNBC PDX organotypic cultures compared to control or paclitaxel alone (Figures 2G–J). Of note, chemo-resistant PDX samples were sensitized to triple-drug treatment and exhibited significant reductions in viability (Figures 2G, H). Together, these data suggest that luteolin played an important role in the triple-drug combination by promoting TNBC cell killing via other mechanisms associated with treatment resistance post-chemotherapy.

Since CSCs are crucial for tumor relapse, we then asked whether triple-drug combination could suppress two key CSC-promoting pathways, YAP and Wnt, that have been closely linked to the maintenance and enrichment of mesenchymal-like and epithelial-like CSCs, respectively (28, 37–39). Gene expression analysis showed that luteolin significantly reduced expression of YAP target genes Ctgf, Cyr61 and Ankrd1 in MDA-MB-231 human TNBC cells, comparable to simvastatin, a known YAP inhibitor (Figure 3A). Similarly, luteolin significantly reduced expression of Wnt target genes Tcf4, Lef1, and Axin2 in SUM149-PT human TNBC cells, comparable to PRI-724, a known Wnt inhibitor (Figure 3B). In both instances, luteolin but not quercetin antagonized the paclitaxel-mediated upregulation of those target genes (Supplementary Figures S3A, B). To consolidate these findings, we performed dual-luciferase reporter assays to assess transcriptional activity of YAP and Wnt. We found significant reductions in the transcriptional activity of both CSC-promoting pathways by the triple-drug combination in human MDA-MB-231 (Figure 3C) and SUM149-PT (Figure 3D) TNBC cells, respectively. These results imply that triple-drug combination may not only suppress CD73 but also mesenchymal-like and epithelial-like CSCs.

Next, we hypothesized that quercetin and luteolin would antagonize the paclitaxel-mediated enrichment of CD44^highCD24^low mesenchymal-like and ALDH1^high epithelial-like CSC populations (4). Through flow cytometric analysis, we found that the triple-drug combination circumvented the paclitaxel-mediated enrichment of CD44^highCD24^low CSCs in mesenchymal-like MDA-MB-231 cells (Figure 4A) and PDX organotypic cultures (Figure 4B). The same trend was observed with ALDH^high CSCs in epithelial-like SUM149-PT cells (Figure 4C) and mouse AT3^ova TNBC cells (Supplementary Figure S3C). Interestingly, quercetin in combination with paclitaxel was unable to reduce paclitaxel-mediated CSC enrichment as effectively as APCP in combination with paclitaxel (Supplementary Figures S3D, E).

Since CD73 has been related to CSC survival, we gated on CD73^high and CD73^low MDA-MB-231 cells to further understand the association between CD73 and CSCs. Interestingly, significantly higher frequencies of CD44^highCD24^low CSCs were observed in CD73^high, but not in CD73^low populations, irrespective of the different treatments (Figure 4D). This suggested that CD73 itself may be a promoter of mesenchymal-like CSCs. Given very low expression of CD73 in epithelial-like TNBC cells, we tested the effect of quercetin, luteolin, and CD73 inhibitor APCP on epithelial-like tumorsphere formation (an in vitro functional assay to assess CSCs). Quercetin and luteolin were observed to reduce the paclitaxel-enhanced tumorsphere formation in human MDA-MB-231 over-expressing E-cadherin (Figures 4E, F) and SUM149-PT (Figure 4G) epithelial-like TNBC cells, implying inhibition of CSC functions in vitro. In contrast, CD73 inhibitor APCP did not antagonize paclitaxel-mediated enrichment in epithelial-like tumorspheres, thereby emphasizing the potential benefit of quercetin, luteolin, and paclitaxel in the inhibition of CSCs in vitro.

To investigate whether the triple-drug combination would maintain superior treatment potential within a heterogenous tumor microenvironment, we injected chicken ovalbumin (ova)-expressing (AT3^ova) mouse TNBC cells into the mammary fat pads of syngeneic immune-competent C57BL/6 mice, followed by treatments (Figure 5A). We found that the triple-drug treatment significantly inhibited the growth of chemo-resistant AT3^ova tumors (Figure 5B). We repeated the experiment to further investigate the effect of each drug alone on tumor growth, with APCP in combination with paclitaxel as a positive control. The triple-drug combination was much more effective than APCP combined with paclitaxel at reducing tumor growth (Figure 5C). Moreover, quercetin and luteolin in combination were more effective at reducing tumor growth than each drug alone; notably, none of the treatments impacted mouse body weight (Supplementary Figures S4A, B).

Next, we investigated the frequencies of tumor-infiltrating lymphocytes, given the observations that CD73 inhibits CD8⁺ T and NK cells while upregulating CD4⁺ regulatory T cell activities (16–18). The OVA-expressing tumors allowed for the assessment of tumor-specific CD8⁺ T cell responses using the ovalbumin peptide SIINFEKL loaded tetramers. We found that the triple-drug combination markedly increased the frequency of tumor-specific activated CD8⁺ T cells (Figure 5D) and mature activated NK cells that were suppressed following paclitaxel treatment (Figure 5E). Of note, luteolin alone was the significant contributor to a higher frequency of tumor-specific activated CD8⁺ T cells, whereas quercetin alone was the significant contributor to a higher frequency of activated mature NK cells. We also found that the triple-drug combination reduced the paclitaxel-mediated increase in the frequencies of immune-suppressing CD4⁺ regulatory T cells (Figure 5F). CD73 expression on activated regulatory T cells was also reduced significantly by quercetin and its upregulation by paclitaxel was antagonized by the triple-drug combination (Supplementary Figure S4C). Consistent with our in vitro findings, paclitaxel alone enriched the frequency of CD44^highCD24^low CSCs, which was averted mainly by luteolin alone and in combination (Figure 5G). A similar trend in response to the treatment was observed among CD73^high tumor cells (Figure 5H).