Targeting autophagic cancer stem-cells to reverse chemoresistance in human triple negative breast cancer

There is growing evidence for the role of cancer stem-cells in drug resistance, but with few in situ studies on human tumor samples to decipher the mechanisms by which they resist anticancer agents. Triple negative breast cancer (TNBC) is the most severe sub-type of breast cancer, occurring in younger women and associated with poor prognosis even when treated at a localized stage. We investigated here the relationship between complete pathological response after chemotherapy and breast cancer stem-cell characteristics in pre-treatment biopsies of 78 women with triple negative breast carcinoma (TNBC). We found that chemoresistance was associated with large numbers of breast cancer stem-cells, and that these cancer stem-cells were neither proliferative nor apoptotic, but in an autophagic state related to hypoxia. Using relevant pharmacological models of patient-derived TNBC xenografts, we further investigated the role of autophagy in chemoresistance of breast cancer stem-cells. We demonstrated that hypoxia increased drug resistance of autophagic TNBC stem-cells, and showed that molecular or chemical inhibition of autophagic pathway was able to reverse chemoresistance. Our results support breast cancer stem-cell evaluation in pre-treatment biopsies of TNBC patients, and the need for further research on autophagy inhibition to reverse resistance to chemotherapy.


Proliferation and cancer stem-cells in human tumor samples
Tumor cell proliferation rate was assessed on pre-treatment biopsies performed before treatment using monoclonal mouse anti-human Ki67 antibody (cloneMIB-1, DakoCytomation/France) as primary antibody. For each biopsy, five different fields at ×400 magnification were analyzed. The proliferation rate was the percentage of positive nuclei in 100 cancer cells analyzed.
To assess cancer stem-cell proliferation, we performed double immunofluorescent staining using the same anti-human Ki67 antibody labelled with Apex-Alexa-Fluor488, and anti-human CD133 antibody labelled with Apex-Alexa-Fluor555. Since CD133-expressing cells were few, the proliferation rate of cancer stem-cells was determined by the ratio of CD133/Ki67-coexpressing cell numbers to CD133-expressing cell numbers. Results were expressed as mean ± SEM.

Apoptosis and cancer stem-cells in human tumor samples
In situ apoptosis was assessed on pre-treatment biopsies using TUNEL assay [1].
Apoptotic cell counts were performed as above for Ki67-expressing cells. The percentage of apoptotic cells in 100 tumor cells was determined, and results expressed as mean ± SEM.
To assess cancer stem-cell apoptosis, TUNEL assay was combined with anti-human CD133 immunostaining on the same tissue section. After TUNEL assay, tissue sections were incubated with mouse monoclonal antihuman CD133 antibody labelled with Apex-Alexa-Fluor555. For each biopsy, five different fields were analyzed. Since CD133-expressing cells were few, the percentage of apoptotic cancer stem-cells was determined by the ratio of CD133-expressing/TUNEL-positive cell numbers to CD133-expressing cell numbers. Results were expressed as mean ± SEM.

Autophagy and cancer stem-cells in human tumor samples
The expression of three cellular markers of autophagy, BECLIN1, BNIP3L, and LC3B was assessed on pre-treatment biopsies. An indirect immunoperoxydase method was performed using rabbit polyclonal anti-BECLIN1 antibody (clone ab16998,Abcam/Cambridge/ USA, dilution1:100), or anti BNIP3L antibody (clone ab8399,Abcam, dilution1:50), or rabbit monoclonal anti-LC3B antibody (clone D11, Cell Signaling/France, dilution: 1:3200), [2], as primary antibodies. For each biopsy and each marker, five different fields were analyzed, the percentage of positive cells in 100 cancer cells was determined, and results were expressed as mean ± SEM.
To assess the number of autophagic cancer stemcells, we performed double immunofluorescent staining using anti-BECLIN1, or anti-BNIP3L, or anti-LC3B antibody labelled with Apex-Alexa-Fluor488, and antihuman CD133-antibody labelled with Apex-Alexa-Fluor-555. For each biopsy, on five different fields, the autophagic stem-cell percentage was determined by the ratio of the numbers of CD133/BECLIN1, or CD133/ BNIP3L, or CD133/LC3B coexpressing cells to the numbers of CD133-expressing cells in each field. Results were expressed as mean ± SEM.

Autophagy and hypoxia in cancer stem-cells of human tumor samples
To characterize hypoxic areas on non-pCR pretreatment biopsies, we assessed the in situ expression of a hypoxic-related protein, CAIX [3,4].
To control these immunostaining results, we used a completely different method combining morphological analyses and molecular markers. On frozen pretreatment biopsies, CD133-expressing cells were lasermicrodissected and their BNIP3L, BECN1, LC3B, CAIX and HIF1A-expression levels assessed using quantitative PCR. For each biopsy, a minimum number of 100 CD133expressing cells was required for molecular analyses, therefore biopsies of only 35 non-pCR patients could be studied. Laser-microdissection was performed using a PALM-Microbeam/Zeiss-system, on 7 μm-thick frozen sections after immunofluorescent staining with anti-CD45 (LCA, Dako/France) antibody. Laser-microdissection was only performed on CD45-negative cells. Using PALM-Robot software for quantification on the 35 biopsies, a mean number of 2000 CD45-negative cells (mean surface 4.10 5 μm 2 ) were microdissected for each case and catapulted into tubes for RNA extraction.

Patients with metastatic TNBC and gene expression profiling
Four women with metastatic TNBC participated to this study. For each patient, five tumor samples of a metastasis were obtained during the procedure of imageryguided biopsy at the time of relapse, before any medical treatment. Informed written consent was obtained from the patients. The Clinical Research Board Ethics Committee (in French "Comité de Protection des Personnes") approved this study (CPP Ile-de-France N°13218) Among the five tumor samples, i) two were formaldehyde-fixed and paraffin-embedded for histological analyses, ii) two were immediately snap-frozen in liquid nitrogen and stored in Hôpital-Saint-Louis Tumorbank for molecular analysis, iii) and one was set aside in culture medium for xenografting.
Total RNA was extracted from the frozen tumor sample as above, and transcriptomic analyses were performed using MiltenyiBiotec Microarray service. A linear T7-based amplification step was performed from 0.5 µg of all RNA samples. To produce Cy3-labeled cRNA, the RNA samples were amplified and labeled using Agilent-Quick-labeling kit. The yields of cRNA and the dye-incorporation rate were measured with ND-1000 Spectrophotometer (NanoDrop, LabTech, France). Hybridization was performed according to the Agilent 60-mer oligo-microarray processing protocol: 1.65 µg Cy3-labeled cRNA was hybridized overnight at 65°C to Agilent-Whole-Human-Genome-Oligo-Microarrays 4x44K, and fluorescence signals were detected using Agilent's Microarray-Scanner. Agilent-FE-Software determined feature intensities and quantile normalization was performed with Agi4x44PreProcess R package. Subsequent analyses were carried out with R 3.01 software (Foundation for Statistical Computing, Vienna, Austria) and based on log 2 single-intensity expression data. Classification was provided by correlating gene expression profiles with the centroids for each of the 6 TNBC subtypes described by Lehmann et al [5], and with Parker centroids for PAM50 classification ( [6], https:// genome. unc.edu/pubsup/breastGEO/pam50_centroids.txt.).

Patient-derived breast cancer xenografts and treatments
Four patient-derived xenografts of human TNBC were studied (XBC1 to XBC4). They had been established for a pilot study on personalized treatment for women with metastatic triple negative breast carcinoma [7], before any medical treatment of the metastatic disease.
For each xenograft model, after a successful engraftment of the metastastic sample, a clinical score was recorded daily for the mice and tumor growth was measured in two perpendicular diameters with a calliper. Tumor volumes were calculated as , L being the larger diameter (length), l the smaller (width). After mouse euthanasia, the tumor was resected, cut into small pieces of 1 mm 3 , and grafted again in 20 other nude mice. The day when tumors reached a volume of 300mm3 -i.e. 100% tumor volume -was considered as Day0. Then the mice were treated over 28 days with intra-peritoneal injections of three types of chemotherapy (n=5 mice per treatmentgroup): epirubicin at 1 mg/kg once a week, paclitaxel at 20mg/kg twice a week, and cisplatin at 3mg/kg once a week (Supplementary Table 1 for drugs tested in TNBC xenografted mice). A daily clinical score was recorded and tumor growth measured weekly until tumor weight reached the ethically recommended limit of less than 10% of mouse weight (Directive 2010/63/EU of the European Parliament and the Council of 22 September 2010 on the protection of animals used for scientific purposes; Official Journal of the European Union L 276/33).
Ultrasonography was performed twice a week on treated and untreated mice with an AplioXT ultrasonograph (Toshiba, Japan) to assess tumor response.

Assessment of tumor response in patients
For each line of chemotherapy, the patient response under treatment was characterized. Metabolic response was assessed according to PERCIST criteria [8]. Briefly, partial metabolic response (PMR) is defined by a reduction in SUVmax of at least 30%, with no new lesions. Complete metabolic response corresponds to disappearance of all lesions in the blood-pool background. Progressive metabolic disease (PMD) is defined by an increase in SUVmax greater than 30%, or appearance of new FDG-avid lesions. Stable metabolic disease (SMD) applies when criteria for other categories (CMR, PMR or PMD) are not met.
For each line of chemotherapy, time-to-progression (TTP) was defined as the time between initiation of treatment and diagnosis of disease progression.
18-Fluoro-deoxyglucose ( 18 FDG) (5MBq/kg; not exceeding 500MBq) was injected intravenously 60 minutes before data were acquired on a Philips Gemini XL Positon Emission Tomography/Computed Tomography (PET/CT) scanner. CT data was acquired first (120kV; 100mAs; no contrast-enhancement). PET 3D data were acquired with 2 minutes per bed position, and images were reconstructed using a 3D row-action maximum likelihood algorithm (RAMLA).
PET/CT images were interpreted by a nuclear medicine physician (LV) blinded to the patient's record. 18 FDG uptake was expressed as standardized uptake value (SUV). A 3-dimensional region of interest (3D-ROI) was drawn around the lesions and SUV max (maximum SUV value within the ROI) was measured. SUV max of the lesions with the highest uptake were recorded and used for the study analysis (five target lesions were assessed). SUV max of the liver was also recorded as a control value. The change in SUV max at each evaluation was expressed as ΔSUV max (%) = 100 × (cycle n SUV max -cycle (n-1) SUV max )/cycle n-1 SUV max . The appearance of new lesions was also recorded.
For cytotoxicity, cells from dissociated spheres were seeded in 96-well tissue culture plates at a density of 5.10 4 cells/well. After 24h of culture, they were exposed to increasing concentrations of drugs (cisplatin, epirubicin or cyclophosphamide) for 24 additional hours. Cytotoxicity was determined by the colorimetric conversion of yellow, water-soluble tetrazolium MTT (3-[4, 5-dimethylthiazol-2-yl]-2,5-diphenyl-tetrazolium-bromide; Sigma, Saint-Quentin, France), to purple, water-insoluble formazan. This reaction, catalyzed by mitochondrial dehydrogenases, is used to estimate the relative number of viable cells [9]. After incubation for 4 hours at 37°C with 0.4 mg/ml MTT, the cells were placed in 0.1 ml of DMSO, and the absorbance was measured at 560 nm using a Fluostar Optima microplate reader (BMG LabTech, France). Experiments were performed in triplicate, untreated cells being used as positive controls, and drug-containing medium without cells as a negative control.

Experimental hypoxia, assessment of stem-cell markers and of mammosphere sensitivity to drugs
For the two xenograft models, XBC1 or XBC4, spheres were separated into two groups maintained in a humidified chamber for 48h, one under experimental hypoxia (1% O2) and the other under normoxia (20% O2). For experimental hypoxia, we chose a pO2 level of 1% as it is now accepted that stem-cell niches are hypoxic with oxygen tension as low as 1% [10].
The relative number of CD133 and CD146 expressing cells in hypoxic and normoxic spheres was assessed by flow cytometry. A total of 3.10 5 cells from hypoxic or normoxic spheres were incubated at 4°C with anti-CD133-APC (clone 293C3, Miltenyi-Biotech, Germany) and anti-CD146-PE (clone 12G5, Beckman-Coulter, France). For each antibody, corresponding isotypes were used as negative controls. Cells were analyzed on BD-Facscalibur (BD Biosciences, France) using Flow Software. Cytotoxicity tests were performed as described above.

Knock-out of autophagy gene expression in spheres
Because of a 16 day-lifetime of spheres derived from patient-derived xenografts (XBC1 and XBC4 models), CRISPR-CAS9 technology was chosen to invalidate BECN1 and BNIP3L gene expression. A dedicated algorithm (see http://crispr.mit.edu website) enabled us to choose the target sequence of 20 bases around the active sites of phosphorylation for each of the two genes. Oligos designed to build the vectors containing the sgRNA1 targeting BECN1 or BNIP3L genes are detailed in Supplementary Table 2. Phosphorylated active site is underlined in yellow. 5 µL of each oligo (forward and reverse) at a concentration of 10 µM was denatured for 5 minutes at 95°C, and they were then hybridized at room temperature. At this stage, hybridized oligo were cloned in a plasmid containing the Cas9 nuclease, according to manufacturer recommendations (System Bioscience Ozyme, France). 2 µg of the plasmid expressing the sgRNA and the Cas9 nuclease with 0.2 µg of a plasmid expressing GFP were added to 500 µL Opti-MEM ® I Reduced Serum Medium (Life Technologies) and left for 30 minutes at room temperature. 15 µL of Lipofectamine ® LTX Reagent (Life technologies) and 7.5 µL de PLUS™ Reagent were added to the medium, which was deposited in a 6-well plate together with 2.10 5 cells from spheres derived from our tumor xenografts, and incubated for 4 hours at 37°C.
After incubation for 24 hours, 1% Kanamycin (G418, Life Technologies) was used to select cells that had integrated the plasmid.

Droplet digital PCR
Droplet digital PCR was performed to assess the copy number of BECN1 or BNIP3L genes in spheres transfected with empty plasmid and spheres transfected with sg1BECN1 or sg1BNIP3L. 48h after transfection, and after antibiotic selection, spheres were processed for RNA extraction and reverse transcription.
Each droplet digital PCR assay, performed according to the MIQE guidelines (minimum information for publication of quantitative real-time PCR experiment) [11], was conducted in triplicate. Reagent mixes were prepared using standard Taqman primer/probe chemistry with a 2XddPCR Mastermix (BioRad, Laboratories, CA), a 20X primer/probe (900/250 nM), and 5 µL sample cDNA template in a final volume of 20 µL. The reagent mixture was loaded into an eight-channel droplet generator (BioRad, Laboratories, CA). 60 µL of droplet generation oil was loaded for each channel and after generation of water-in-oil droplets the droplets were transferred to a 96well PCR plate and placed in a Biorad thermocycler. An initial denaturation step (95°C, 10 min) was followed by 45 cycles at 95°C for 15 sec and at 60°C for 1 min. The PCR products were streamed through a droplet reader and the results analysed using QuantaSoft software (BioRad Laboratories, CA). All droplets were gated on the basis of detector peak width to exclude doublets or triplets.

Electron microscopy of spheres KO for BECN1 or BNIP3L
Spheres transfected with empty plasmid and spheres transfected with sg1BECN1 or sg1BNIP3L were dissociated and cells in suspension were counted. After centrifugation, a pellet of 10 6 cells was further processed for electron microscopy, and fixed in 2% glutaraldehydebuffered 0.1M cacodylate and embedded in epoxy resin. Ultrathin sections were stained with uranyl acetate and lead citrate.
Ultrastructural analysis, performed on a Hitachi-7650, focused on the cytoplasms of tumor cells to detect characteristic autophagosomes according to the "Guidelines for the use and interpretation of assays for monitoring autophagy" [12].
A quantitative study was performed comparing the numbers of autophagosomes per cytoplasmic area in cells transfected with empty plasmid versus sg1BECN1 or sg1BNIP3L transfected cells (Supplementary Figure 2). We used CellSens Dimension 1.9 software (Olympus) to delineate and calculate the cytoplasmic areas of tumor cells on ultrastructural images at magnification 20.000. The cytoplasmic area was calculated in 25 tumor cells in each series of spheres. Within these cytoplasmic areas, the numbers of autophagosomes were counted, and the results expressed as the mean numbers of autophagosomes per 1000 µm² of cytoplasmic area.

Statistical analyses
The statistical analyses were performed using R 2.15.2 statistical software (R-Foundation for Statistical Computing, Vienna/Austria). The differences between the two groups of patients (pCR and non-pCR), for the pretreatment numbers of CD133-expressing cells, ADLH1expressing cells, ALDH1/CD133-coexpressing cells, and the presence of clustered cells or necrosis, were assessed using Wilcoxon signed-rank tests with the wilcox. test R command. Probabilities were corrected for multiple comparisons using the Bonferroni method with the p.adjust R command. A p value ≤ 0.05 was considered to be significant. Principal component analysis and correlation circle were performed with the ade4 R-package.
Among non-pCR patients, the differences between tumor cells and cancer stem-cells, for the number of Ki67expressing cells, TUNEL-expressing cells, BECLIN1-, BNIP3L-, or LC3B-expressing cells, were assessed using the same test and the same correction. Variables significantly predicting pCR were also selected by multivariate regression with the glm command (binomial model) in the R stats package and by forward selection with the stepAIC algorithm in the R MASS package. Association between pCR and relapse was assessed by Cox regression with the R survival package.
For cell counts in immuno-stained tissue sections, flow cytometry, and MTT results, and for RT-qPCR results, the mean ±SEM (standard error mean) was calculated in each experimental group and shown in bar graphs. Quantitative values were compared using Student's t-test (two-tailed). P values under 0.05 (after Bonferroni correction for multiple comparisons) were considered significant. Comparisons between tumor growth curves were analysed globally with permutation tests with the R statmod package and two-by-two comparisons were carried out for critical points. For correlation studies, the Kendall rank correlation coefficient was calculated between patient ∆SUVmax for one chemotherapy regimen and the coefficient of inhibition for the same regimen in TNBC xenograft.