Translocation of Intracellular CD24 Towards Cell Membrane Constitutes a Triggering Event for Drug Resistance in Breast Tumor Cell: Correlation With Dynamics of p38MAPK Activation

adaptation capacity of to shift be responsible for and to the of we have provided concerning the role of the in the of cancer can overcome adaptive resistance to doxorubicin treatment. the under drug and new cell cell


Introduction
Plasticity endows cancer cells with the capacity to shift dynamically between different states which could be responsible for chemoresistance. The adaptation capacity of tumor cells has been commonly associated with stemness when the acquisition of Cancer Stem Cell (CSC) properties by non-CSCs has been reported under different conditions. Indeed, CSCs would be responsible for tumor initiation, maintenance, metastasis formation, phenotypic plasticity and drug resistance [1][2][3][4][5][6][7]. CD24, a cell surface adhesion glycoprotein, was identi ed as a CSC marker and is present in various types of cancer including breast, pancreatic and lung [8] and has often been associated with more aggressive diseases in ovarian, breast, lung and prostate cancers [9,10]. This explains the considerable interest for this marker in tumor biology and also in treatment outcome. In this line, relevant studies have revealed that the phenotype switching associated with the detection of surface CD24 could be responsible for chemoresistance [11,12]. For instance, CD24 expression level has appeared to be a signi cant molecular phenotype of cisplatin-resistant residual cells in laryngeal carcinoma lines which corresponds to a differential expression of critical apoptotic and drug resistance [13]. Goldman et al. have observed an enrichment of CD24 + cells following treatment with docetaxel in different breast tumor cell lineages, which corresponded to the generation of new CSCs from non-CSCs [12]. Importantly, the mechanism leading to an increased surface CD24 expression in tumor cells under drug stress remains unknown.
Recent ndings support the hypothesis that epigenetic mechanisms are key players in the phenotypically transition of tumor cells [14,15]. Epigenetic can result in drastic changes not only in cell plasticity but also in drug resistance [16,17]. Besides, MAPK signaling pathways, which are deregulated in tumor [18,19], have been constantly associated with chemoresistance in different type of cancer [20][21][22]. For instance, it has been reported that p38 and JNK MAPK pathways play a role in the control of the balance of autophagy and apoptosis in response to genotoxic stress [20]. Others have explored the role of ERK and p38 MAPK in breast cancer chemotherapy [21]. But, little is known concerning the involvement of MAPK in tumor plasticity. Indeed, the molecular mechanisms that control cellular plasticity upon drug treatment remain to be fully established.
Here, using different breast cancer cell lines, we have evaluated the impact of drug stress on the immediate phenotype change. We have shown that the rapid translocation of the CSC marker CD24 from cytosol to cell membrane was the triggering event for the acquisition of chemoresistance. In drugresistant MDA-MB-231 cells, we have identi ed a tandem constituted by CD24 and MAPK p38, where the continuous p38 activation controlled the tri-methylation of H3K9 and the overexpression of the survival marker Bcl-2. This phenotype enabled the cells to enter in slow-down of cell cycle, after which several weeks later, the dormant cells retrieved their capacity to proliferate, and were characterized by an increased resistance to drug and migratory capacity. Importantly, the use of p38 inhibitor was able to block the acquisition of drug resistance by impeding the upregulation of the anti-apoptotic protein Bcl-2.
The treatment with the TLR7 agonist Imiquimod, which reduced cell proliferation without affecting phenotype switching neither Bcl-2 expression allowed con rming a direct correlation between translocation of CD24 and increased expression of Bcl-2. Finally, we propose that the association of the p38 inhibitor, SB203580, with doxorubicin, a drug leader in clinic, could open up a new strategy in the ght with cancer.

Cell Culture Conditions
We obtained the breast tumor cell lines MDA-MB-231 from ATCC (USA). MACL-1 and MGSO-3 were obtained from Brazilian patients with breast cancer [51]. Cells were grown at 37°C in a humidi ed atmosphere of 5% CO 2 in DMEM supplemented with 10% heat inactivated fetal bovine serum (FBS). For starvation conditions, cells were incubated with serum-free DMEM for 2h before the addition of the respective stimuli.
Obtaining of the resistant cell clones from parental MDA-MB-231 The procedures to obtain MDA-MDA-231 cells resistant to doxorubicin (named CD24 + /DxR cells) are schematized in gure 4a. Brie y, MDA-MB-231 cells were incubated with doxorubicin at 0.6mM for 48h.
The remaining cells, named CD24 + /DxR, were trypsinized and used in the respective experiments. These cells were characterized as non-proliferative. DxR/30 achievement is schematized in gure 5a. After obtaining CD24 + /DxR cells, these cells were cultured with DMEM 10% of fetal bovine serum for 15 days, then cells were cultured under stress condition (DMEM 1% of fetal bovine serum for 48h) following by incubation in normal culture conditions until they recover their proliferative activity. From then, cells were
Then, cells were centrifuged at 2000rpm for 10 minutes at 4°C and the supernatant was discarded. In the next step, cells were washed in half the volume of TEB and centrifuge at before. The pellet was ressupended in 0.2N HCl at a cell density of 4x10 7 cells per ml over night at 4°C. Then, samples were centrifugated at 2000rpm for 10 minutes at 4° and the aliuots were stored -20°C.

RNA silencing
MDA-MB-231 cells were seeded in 24-well plates at a density of 2x10 4 cells/well. After 24h of incubation, the cells were transiently transfected with short interfering RNA (siRNA) speci c for CD24 or control siRNA at a nal concentration of 100nM by using Lipofectamin 3000 Transfection Reagent according to manufacturer's instructions. Brie y, siRNA and Lipofectamin were diluted separately in serum-free OPTI MEM. Then, the diluted Lipofectamin and siRNA were mixed (1:1 v/v). After 5min, the mix was added drop-wise onto the cells under their normal growth conditions and after 48h silenced cells were used in the different assays. CD24 silencing was con rmed by western blotting.

Western Blotting
For protein expression, cells were harvested, counted and lysed in RIPA buffer supplemented with phosphatase and protease inhibitor cocktail according to the manufacturer's instructions. For the evaluation of the phosphorylated form of MAPKs, cells were seeded in 6-well plates (10 6 cells/well) and were starved for 2h in free-serum medium (SFM). Then, cells were treated for 30~60 min with medium containing 10% FBS as stimuli. At the end of treatment cells were lysed in RIPA buffer supplemented with phosphatase and protease inhibitor cocktail according to the manufacturer's instructions. Protein lysates were separated by polyacrylamide gel electrophoresis on 10% or on 15% gels when concerned histone extract, and electrotransferred to PVDF-0,45mM membranes (Milipore) or nitrocellulose-0,22mM (BioRad) when concerned histone extracts. Membranes were blocked overnight with 5% dry milk and were incubated with primary antibodies in 5% BSA also overnight. After incubation with the peroxidaseconjugated secondary antibody for 1h, protein expression was detected using Luminat HRP reagent (Milipore) and analyzed using LAS-4000 imaging system (Fuji) or C-DiGit Blot Scanner (Li-Cor).
Magnetic sorting The different cell subpopulations (CD24 + and CD24 -) were sorted from parental MDA-MB-231 cells using CD24 MicroBead Kit (MACS Miltenyi Biotec) following manufacturer's instructions. Brie y, the CD24 + cells were indirectly magnetically labeled with CD24-Biotin antibodies and Anti-Biotin MicroBeads. Then the cell suspension was loaded onto a MACS Column, which is placed in the magnetic eld of a MACS Separator. The magnetically labeled CD24 + cells were retained within the column. The unlabeled cells (CD24cells) ran through; this cell fraction was thus depleted of CD24 + cells. After removing the column from the magnetic eld, the magnetically retained CD24 + cells can be eluted as the positively selected cell fraction. The purity of the sorted populations was veri ed by Flow Cytometry.

Flow cytometry analysis
To evaluate the proliferative activity of cells, an assay using the Apoptosis, DNA Damage, and Cell Proliferation Kit from BD Biosciences was realized in accord with the manufacturer's instructions. Cells were cultured as indicated in the gure legends and washed/blocked in staining buffer (PBS 4% fetal bovine serum -v/v). Cells were xed and permeabilized using CitoFix/CitoPerm reagent for 20 min on ice, nucleus was permeabilized using CitoFix/CitoPerm Plus for 10 min on ice and nally treated with DNAse (30mg/10 6 cells) for 1h at 37ºC in humidi ed chamber. Then, cells were simultaneously stained with anti-BrdU/PerCP-Cy5.5 and anti-gH2AX/Alexa647 (BD Biosciences) uorescent antibodies for 20 min at room temperature. Between every step cells were washed with 1 x PBS/PermWash (BD Biosciences).
Concerning CD24 and p38 staining, the following step by step was realized. For extracellular staining, MDA-MB-231 cells were washed/blocked in staining buffer (PBS 4% fetal bovine serum -v/v) and were labeled using anti-CD24/Pe-Cy7 (eBioscience) or anti-CD24 (BD Pharmigen) for 20 min on ice, and when necessary, followed by FITC-conjugated equivalent secondary antibody (BD Pharmigen) for more 20 min on ice. For intracellular staining, after extracellular labeling, cells were xed and permeabilized using CitoFix/CitoPerm reagent for 20 min on ice. Then, cells were incubated with anti-CD24/Pe-Cy7 (eBioscience) and/or with anti-pp38 (Cell Signaling) for 20 min at room temperature, and when necessary, followed by FITC-conjugated secondary antibody (BD Pharmigen) for 20 min at room temperature. Between every step cells were washed with 1 x PBS/PermWash (BD Biosciences). After labeling protocols, cells were xed in PFA 4% overnight (4ºC lightless). Cells were re-suspended in isoton buffer and analyzed by ow cytometry. Single-stain controls were used to set gating parameters and any compensations. All ow cytometry results were analyzed by FlowJo software following a rigorous doublet discrimination based on FSC:A versus width as well as FSC:A versus height. Data were collected by the cell analyzing LSRFortessa (BD Biosciences -Immunocytometry Systems) using "BD FACSDivaTM Software" (BD Biosciences) and analyzed with "FlowJo (Tree Star) Software".
Wound Healing Assay MDA-MB-231 and DxR/30 cells were seeded (8x10 5 ) in 6 wells plate and let in growing conditions (DMEM 10%FBS, at 37 o , 5%CO 2 ) for 24 hours or until cells get con uent. Then, using a 100ml tip, a scratch was performed in the cell monolayer and cells were cultured in DMEM supplemented with 2% of FBS in order to avoid cell proliferation. To obtain the same eld during the image acquisition, pen markings were performed at the bottom of the culture plates as reference points close to the scratch. Cell migration was registered by light microscopy (Evos Skedda) at 24, 48 and 72 hours post scratch. Using ImageJ software the wound healing was measured and a closing-time percentage was calculated based on the initial scratch size of each cell type.

Statistical analysis
The data were presented as mean of triplicates + / -SD or as means of triplicates. Statistical signi cance was determined using Student's t-test, or Two-way ANOVA followed by Bonferroni post-test. The criterion for statistical signi cance was p<0.05.

-The translocation of CD24 from cytosol to membrane is an early event in breast tumor cells under drug stress
Phenotype switching, also commonly referred to as cell plasticity, is an important process observed during treatment of cancer which was repeatedly associated with stemness. Here, using breast cancer cell line, we explored the dynamics of the CSC marker CD24 after doxorubicin treatment. At rst, we sought to de ne the localization of CD24 in MDA-MB-231 cells by extra and intracellular staining. As shown by our results obtained by ow cytometry (Fig 1a,b) only about 5% of cells expressed CD24 in cell membrane corroborating with other studies [23] which explains why MDA-MB-231 is considered as CD24 low/-. By contrast, a signi cant intracellular pool of CD24 was encountered in all cells. Fluorescence microscopy con rmed the presence of both extracellular (yellow arrows) and intracellular CD24 (white arrows) (Fig 1c). After the treatment with doxorubicin at 0,6mM -concentration representing the EC50 after 24h treatment calculated in MDA-MB-231 cell line -a cell phenotype switching occurred which corresponded to an enrichment of the CD24 + subpopulation (Fig 1d). Notably, MFI analysis showed an increase of surface CD24 density during drug treatment (Fig 1e). This phenomenon occurred rapidly sincẽ 42% of cells converted into CD24 + after 2h to nally reach ~96% after 48h of treatment, as visualized by ow cytometry (Fig 1d). Importantly, the majority of cells remained positive even after a pause in the treatment (incubation in drug-free medium for 48h after treatment) as visualized by the last pseudocolor plot in the gure 1d. The fact that this event was detected in the rst hours of treatment excludes the possibility of a Darwinian selection of CD24 + cells. These results led us to hypothesize that the intracellular pool of CD24 immediately available might play a role in this process by translocating from the cytosol toward cell surface under doxorubicin treatment. To support these data, MDA-MB-231 cells were sorted into CD24 + and CD24subpopulations (Fig 1h) by using magnetic beads and the CD24 localization was evaluated in CD24cells after doxorubicin treatment. As shown in the gure 1i, the translocation of CD24 occurred even in CD24population obtained after cell sorting since CD24cells were able to rapidly convert in CD24 + cells. Such data reinforce the idea that CD24 + cells enrichment during drug treatment does not correspond to a pre-selection of clones but to a drug-induced phenotype switching. In order to con rm this theory, we took the opportunity of using brefeldin A, an inhibitor of protein transport from endoplasmic reticulum to Golgi apparatus, to disturb the CD24 tra c after drug treatment. After ow cytometry analysis, we observed that when MDA-MB-231 cells were treated with brefeldin A before doxorubicin, the translocation of CD24 was reduced (Fig 1f). These results were consistent with the uorescence microscopy images obtained from doxorubicin-treated cells and stained with anti-CD24 without permeabilization to solely detect surface CD24 (Fig 1g). To discard the hypothesis that the increased membrane CD24 expression could be partially due to an increased protein synthesis, we used actinomycin D, a DNA-transcription inhibitor. The incapability of actinomycin D to reduce the CD24 + phenotype enrichment in presence of doxorubicin indicated that the intracellular pool of CD24 was the main source of CD24 tra c in the presence of doxorubicin (data not shown).
Concerning MACL-1 and MGSO-3 breast cell lines, obtained from Brazilian patients, the percentage of CD24 + cells was about 5% and 46% respectively, corroborating with their classi cation in other study [24]. The presence of intracellular CD24 was also detected in the whole population of both cell lines, like observed in MDA-MB-231 population (Fig 2a,b). As observed in MDA-MB-231 cells under treatment, phenotype switching occurred in both cell lines which corresponded to an enrichment of the CD24 + subpopulation (Fig 2c,d) since ~27% and ~45% of MACL-1 and MGSO3 cells, respectively, converted into CD24 + after 4h of treatment with doxorubicin. After 24h of treatment, the conversion rate in CD24 + cells reached ~90% for MDA-MB-231 and MACL-1 and ~70% for MGSO-3 cells.
Therefore, we propose a new dynamic model of cell transition phenotype under drug stress that could allow each cell of breast cancer to convert into CD24 + cells which involves the translocation of intracellular CD24 accompanied with its increased expression at cell surface.
suggesting that acquisition of drug tolerance controlled by Bcl-2 expression may also require cells to exit the cell cycle (Fig 3c). In previous studies, we have focused on the role of p38 MAPK and ERK1/2 in the proliferation of MDA-MB-231 cells [25,26], so we analyzed their activation pro le in the resistant CD24 + /DxR cell. Interestingly, this phenotype switching was accompanied by a strong and continuous activation of p38 MAPK at the detriment of another MAPK, ERK1/2 (Fig 3d). Then, we veri ed the relationship between CD24 and Bcl-2 by silencing CD24 using interference-RNA. In CD24-silenced cells (SiCD24), a decreased Bcl-2 and p38 expression was observed (Fig 3f). This may explain the reduced capacity of SiCD24 cells to resist to drug treatment in all the doxorubicin concentrations tested when compared with control-silenced (SiC) and parental MDA-MB-231 cells (Fig 3e). By contrast, CD24subpopulation, obtained from magnetic sorting, presented a similar drug sensibility when compared with CD24 + subpopulation and parental MDA-MB-231 cells, probably due to the phenotype switching after drug treatment thanks to CD24 translocation as shown in the gure 1j (Fig 3e).
Next, we sought to treat MDA-MB-231 cells with a drug capable to reduce cell proliferation without inducing immediate death, like doxorubicin does. We tested the e cacy of the TLR7 agonist Imiquimod, previously used in skin cancer treatment and in the treatment of cutaneous metastatic breast cancer [27][28][29], on MDA-MB-231 cell proliferation. At the concentration of 1mM, a signi cant decreased in cell proliferation was observed while a total blocking of cell replication was noted at 10mM during the period of experiment (Fig 3g). No signi cant cell death was observed in the rst 48 hours in the presence of both concentrations of Imiquimod. In such context, we tested the capacity of cells to respond to a second treatment 96h after the rst dose of Imiquimod. As shown in the gure 3h, where the results are expressed in % of cell viability, we observed a similar pattern when an unique dose or two subsequent treatments with the TLR7 agonist were used indicating that no phenotype change occurred after the rst treatment and consequently avoid chemoresistance. When we used CD24 as a phenotypic marker, we con rmed that no switching phenotype of MDA-MB-231 population leading to an enrichment of CD24 + cells occurred (Fig 3i). In accord with this, no upregulation of Bcl-2 expression was detected in cells treated with Imiquimod (Fig 3j).
Taken together these data indicated that translocation of CD24 is a triggering event leading to phenotype change and upregulation of Bcl-2 expression.
3 -CD24 and p38 work in tandem in the chemoresistance acquisition phenotype in breast cancer cells.
According to the gure 3d, p38 phosphorylation was stronger, constitutive and independent on serum in CD24 + /DxR cells contrasting with the serum-dependent activation of p38 in MDA-MB-231 cells under normal culture conditions. This suggests that p38 activation in different con gurations can cause different outputs and may participate in the phenotype switching.
The next question was whether there was a privileged relationship between CD24 and p38. At rst, we explored this point in MDA-MB-231 cells under growing culture conditions. After magnetic sorting, we evaluated by western blot the status of MAPK activation after cell stimulation with serum according to the kinetic presented in the gure 4a. As clearly shown, CD24 + cells phosphorylated p38 in a more pronounced way than CD24subpopulation. In contrast, a higher phosphorylation of ERK1/2 was observed in the CD24and parental MDA-MB-231 cells. The results obtained by ow cytometry con rmed the correlation between surface CD24 expression and preferential p38 phosphorylation, as observed by the fact that ~70% of the CD24 + cells phosphorylated p38, while in only ~15% of the CD24cells the activation of this MAPK was observed (Fig 4b).
Another evidence demonstrating that CD24 and p38 work together is presenting in the gure 4c. According to the western blotting, siRNA-mediated knockdown of CD24 decreased the phosphorylation of p38 when cells were submitted to doxorubicin treatment indicating that the absence of CD24 jeopardized the cell capacity to induce activation of p38 MAPK.
The sustained p38 activation in CD24 + /DxR cells makes it a prime target. In this context, we used SB203580, a p38 activity inhibitor [30] to evaluate its impact on drug resistance. According to the results obtained by cell counting, the combination of SB203580 and doxorubicin was more e cient in reducing cell number than doxorubicin alone (Fig 4d). Concerning the results observed by MTT assay, the inhibition of p38 was bene t from two aspects: rstly, SB203580 sensitized MDA-MB-231 cells to the therefore doxorubicin treatment (blue line vs black line) and at second, SB203580 disrupted the resistantphenotype acquired by the cells that received two consecutives doxorubicin treatments (green line vs red line). The effect of the drug association may be considered as synergistic since the total effect of the combination of SB203580 and doxorubicin was greater than the sum of the individual effects of each drug. Importantly, SB203580 alone was unable to impact on cell viability in the same experimental conditions (Fig 4e).
To visualize these results, we performed the capture of light microscopy images of cells treated with drug pair. These experiments were performed under sub-con uence or con uence conditions to exclude the in uence of uctuating environment. A direct impact of doxorubicin on MDA-MB-231 cells was observed after 24h treatment marked by a decreased cell number and changes in morphology. The association of SB203580 and doxorubicin exacerbated the cell phenotype changes under con uent and sub-con uent conditions. The results con rmed that the e ciency of the drug pair constituted by doxorubicin and SB203580 was superior in killing cells in both plating conditions (Fig 4g). Consistent with the above results, western blots showed that SB203580 prevented the augment of Bcl-2 expression induced by doxorubicin (Fig 4f). Further, in MDA-MB-231, which has high levels of a mutant p53, it has been described that p53 mutants can contribute to the suppression of apoptosis [31]. In line with this, SB203580 was also able to reduce the expression of p53 in doxorubicin treated cells (Fig 4f).
Epigenetic events drive cell reprogramming and tumor cell plasticity [14,15]. In this context, we have, evaluated whether the state of histones could be altered in CD24 + /DxR cells. We focused on the lysine 9 at the histone H3 (H3K9), which can turn genes on or silence them by getting acetylated or methylated [32]. Our western blot analyzes have shown that doxorubicin treatment increased the tri-methylation of H3K9 (H3K9me3) which combined with its deacetylation. Importantly, SB203580 was able to prevent the increase of H3K9me3 induced by doxorubicin treatment indicating that p38 can regulate the methylation state of H3K9 observed under drug pressure (Fig 4h).
Taken together, these results suggest that targeting p38 during the chemotherapy-induced phenotypic cell state transition can overcome adaptive resistance to doxorubicin treatment.
4 -CD24/DxR cells become proliferative after a long-lasting period in dormancy.
The capacity of slow cycling cells to reentry into cell cycle has been a topical debate for quite a time. As reported above, CD24 + /DxR cells have adopted a slow-down cell cycle after doxorubicin treatment evidenced by a reduction of cyclin D1 (Fig 3c) and were characterized by an enlarged morphology that are hallmarks of dormant cells (Fig 5b).
So, we sought to monitor CD24 + /DxR cells to evaluate the reversibility of their dormant state according to the scenario presented in the gure 5a. CD24 + /DxR cells were cultured in drug-free medium for a long period and then were submitted to a serum deprivation for two days followed by the culture in medium contained 10% of serum. In such conditions, we observed the emergence of revertant cells (named DxR/30) which reacquired the ability to proliferate as con rmed by their capacity to incorporate BrdU (Fig  5c).
The percentage of CD24 + cells in the revertant-population recovered to levels observed in naïve MBA-MB-231 cell population (Fig 5d). Importantly, even in the absence of drug, the phosphorylation of p38 in DxR/30 remained strong and constitutive indicating that DxR/30 cells have conserved some features of their precedent states while having eliminated others (Fig 5e).
When we evaluated the drug resistance of DxR/30 cells, more than one month after the rst treatment the cells remained tolerant to doxorubicin. Indeed, as shown in the light microscopy images, the morphology of these cells appeared little affected after treatment when compared to naïve MDA-MB-231 treated cells. More surprisingly, DxR/30 cells retained their capacity to proliferate even in the presence of drug (5 days) without entering in slow cycling stage like CD24 + /DxR, con rming that DxR/30 cells have acquired a new identity. After 12 days of treatment, they have re-colonized the plastic dishes (Fig 5f). One of the hypothesis is that DxR/30 cells might have a competitive advantage over naïve cells under drug stress due to their constitutive phosphorylation of p38. Finally, we evaluated the migratory capacity of these cells by using the in vitro wound healing assay based on the creation of an arti cial gap, so called "scratch", on a con uent cell monolayer. Images were captured every 24 hours during cell migration, the scratch was measured and a comparison of time required to close the scratch between naïve cells and DxR/30 was performed. The incubation time was determined at 48h when the faster moving cells DxR/30 were just about to close the scratch. The con rmation of the migratory capacity of both cells was made by using the software Image J (Fig 5g).
According to our data, slow-cycling cells under stress may reentry into cell cycle leading to cells that possess new properties, including higher drug resistance and higher migratory capacity, rea rming that they have acquired a new identity.

Discussion
Tumor cells have been shown to hijack signaling pathways involved in reprogramming to become phenotypically plastic and evolve towards drug-refractory cell identities. Our results provide new insight about these mechanisms leading to the emergence of resistant subpopulation during chemotherapy treatment. Robust evidence showed that the CSC subpopulation is enriched after chemotherapy, suggesting that this subset is responsible for the majority of treatment failure [33,34]. Here, by monitoring the CSC marker CD24 during doxorubicin treatment, we have shown the uniform conversion of CD24breast cancer cell population into CD24 + cells characterized as drug resistant cells. In the same line, Goldman et al have reported that the treatment of breast or ovarian cancer cells with high concentration of taxanes results in the generation of 'persistent' cells, which are de ned by a transition towards a CD44 Hi CD24 Hi expression status [12]. Others studies have demonstrated an enrichment for CSC-like phenotype after chemotherapy in glioblastoma [35] indicating how the phenomenon is ubiquitous.
Importantly, our data revealed that each given breast tumor cell may convert into CD24 + phenotype thanks to the translocation of CD24 from cytosol to cell membrane made possible by the presence of an intracellular pool of CD24, never described until now. Such ndings suggest that CSC may not be considered as a clonal identity but rather a state as also claimed by Dirske etal [35]. The conversion observed between non-CSC and CSC after chemotherapy demonstrates that cell plasticity emerges as an important contributor to therapy escape. This was veri ed when we used the TLR7 agonist Imiquimod, capable to reduce MDA-MB-231 cell proliferation without inducing translocation of CD24 + , neither resistance to a second treatment, nor changes in the expression of the anti-apoptotic protein Bcl-2 that contrasted with the overexpression of Bcl-2 observed in cells under doxorubicin treatment. In this regard, great efforts have been directed towards nding small molecules to inhibit these anti-apoptotic Bcl-2 family proteins to tackle anti-apoptotic adaptation of tumor cells [36].
Hence, targeting cell plasticity should provide a unique opportunity to improve the e ciency of existing therapies. The identi cation of signaling pathways involved in the cell phenotype changes during treatment may be a key to impede chemoresistance. The tandem constituted by CD24 and the MAPK p38 appears crucial in cell fate of MDA-MB-231 population. A preferential use of p38 in CD24 + cells have been noted under normal conditions which has been ampli ed under drug treatment in accord with the dynamics of CD24. The relationship between CD24 and p38 is supported by the incapability of SiCD24 cells to activate p38 under drug stress. Indeed, we hypothesize that the change in the dynamics of p38 activation that turned sustained early after CD24 translocation is linked to the increased surface CD24 expression observed in doxorubicin treated cells. In this regard, a very recent paper reported that MAPK cascade signaling dynamics (transient to sustained activation) may be controlled by the activation kinetics of a given membrane receptor and not necessarily by the intracellular topology of the kinase networks. In the same study, it was proposed that redirecting signal dynamics may be a more fruitful and effective approach than controlling receptor activation in pathologic situation [37]. This is why we considered the pro le of p38 activation as an important marker of the resistant cell identity (CD24 + /DxR) and a choice target to impede phenotype switching during doxorubicin treatment.
The relevance of this strategy was veri ed when we compared the protein expression pattern of CD24 + /DxR cells characterized by an overexpression of Bcl-2 and decline of cyclin D1 which are indicators of adaptative cellular reprogramming and also considered as markers of premature senescence [38][39][40] and naïve MDA-MB-231 cells treated with doxorubicin in the presence of the p38 inhibitor SB203580. In such conditions, SB203580 was capable not only to reduce Bcl-2 expression but to decrease the tri-methylation of H3K9 which constitutes a hallmark for stress response leading to cell reprogramming including drug resistance [41][42][43]. The bene t of this association was translated into a synergism of the cytotoxic effect of the combination of SB203580 and doxorubicin. The impact of SB203580 on cell fate may also correlate with the capacity of p38 to control expression of p53. Recently, p53 was also considered as a marker of cell reprogramming and acquisition of stemness [44]. Finally, an interesting point to note is that SB203580 had no effect on cell viability at 24h at the concentration used.
This sustains the idea that p38 is an early active player in drug resistance by inducing cell identity changes, notably increasing cell survival marker expression.
To better assess the long term consequences of the rapid phenotype switching after drug treatment we have evaluated whether the slow-cycling state of CD24 + /DxR cells was reversible after several weeks in drug-free medium since several studies have demonstrated the reversibility of senescence [45][46][47][48]. In fact, the revertant cells (named DxR/30), recovered their capacity to proliferate and importantly, have gained higher drug resistance and stronger migratory properties. This may be put in parallel with the constitutive activation of p38 in these cells. Interestingly, in the absence of the drug pressure and in proliferative conditions, the DxR/30 population retrieved a basal level of CD24 + cells. This follows mathematical models that tend to establish that tumor cell populations always maintain its heterogeneity at xed ratio in dynamic conditions like proliferation [49].
So, here we have established a model to accompany the journey of MDA-MD-231 cells after drug treatment starting from their switching to CD24 + phenotype to their entry in slow-cycling state and reversion into proliferative cells. Notably, the sustained p38 phosphorylation observed in all the posttreatment stages may also indicate the participation of this MAPK in the network that successfully promotes the formation of metastasis at distant organs. Previous studies have shown that the program EMT (Epithelial-Mensenchymal Transition) has been involved in the distant metastases frequently detected following chemotherapy [50] that could also include p38. The monitoring of cell evolution throughout this journey has shown that they have conserved some features of previous state while having eliminated others, in other words, that they have acquired hybrid properties corresponding to a new identity.
The pertinence of using a drug pair, associating a leading drug with an inhibitor of cell reprogramming, was illustrated by the high e ciency of the combination of p38 inhibitor and doxorubicin on the killing of aggressive MDA-MB-231 breast cancer cell line as schematized in the gure 6.

Conclusions
This study rea rms how a better understanding of the biology and molecular drivers of the CSCs plasticity will enable identi cation of new anticancer targets. More speci cally, we have shown here that the cellular localization of CD24 associated with the changes in the dynamics of p38MAPK activation in breast tumor cells under drug stress plays a crucial role in the acquisition of drug resistance and new cell identity. The use of p38 inhibitor was able to disrupt this partnership and consequently to avoid chemotherapy-induced cell state transition. These results suggest that targeting p38 in aggressive breast cancer might be an interesting approach in order to overcome adaptive resistance to doxorubicin treatment. The authors declare no con icts of interest to disclose.

Contributions
HWH was involved in the laboratory experimental work, data analysis and interpretation and writing the manuscript. AMG participated in the study formulation. CR was involved in the supervision of the experimental work, in the study formulation and writing the manuscript. The authors read and approved the nal manuscript.     Translocation of CD24 is associated with chemoresistance and overexpression of Bcl-2 (a) Schematic shows experimental design used to obtain cell populations for MTT ( g 3b) and western blot assays ( g 3c and 3d). dfm = drug-free medium; Dx = doxorubicin; CD24+/DxR = CD24+ doxorrubicin-resistant cells.       the use of a speci c inhibitor, like SB203580, in association with doxorubicin reduces Bcl-2 expression and H3K9 tri-methylation leading to an increased cell death.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. graphicalabstract.pdf graphicalabstract.pdf