Modeling Heterogeneity of Triple‐Negative Breast Cancer Uncovers a Novel Combinatorial Treatment Overcoming Primary Drug Resistance

Abstract Triple‐negative breast cancer (TNBC) is a highly aggressive breast cancer subtype characterized by a remarkable molecular heterogeneity. Currently, there are no effective druggable targets and advanced preclinical models of the human disease. Here, a unique mouse model (MMTV‐R26Met mice) of mammary tumors driven by a subtle increase in the expression of the wild‐type MET receptor is generated. MMTV‐R26Met mice develop spontaneous, exclusive TNBC tumors, recapitulating primary resistance to treatment of patients. Proteomic profiling of MMTV‐R26Met tumors and machine learning approach show that the model faithfully recapitulates intertumoral heterogeneity of human TNBC. Further signaling network analysis highlights potential druggable targets, of which cotargeting of WEE1 and BCL‐XL synergistically kills TNBC cells and efficiently induces tumor regression. Mechanistically, BCL‐XL inhibition exacerbates the dependency of TNBC cells on WEE1 function, leading to Histone H3 and phosphoS33RPA32 upregulation, RRM2 downregulation, cell cycle perturbation, mitotic catastrophe, and apoptosis. This study introduces a unique, powerful mouse model for studying TNBC formation and evolution, its heterogeneity, and for identifying efficient therapeutic targets.


Introduction
Genetically engineered mouse models (GEMMs) of breast cancer have been proven as a powerful tool for gaining mechanistic insights into tumor initiation, progression, and metastasis as well as for developing innovative cancer therapy. [1] GEMM models for breast cancer commonly use mammary-gland specific promoters, including MMTV (virus long terminal repeat), WAP (whey acidic protein), and C3 to ensure expression of transgenes in the mammary epithelium. More than 25 different murine GEMMs for breast cancer expressing different genes/oncogenes such as, PyMT (polyoma middle T antigen), SV40 T antigen, ErbB2/Neu, cyclinD1, Ras, Myc, TGF-, and Wnt1 have been established. [2] The most widely used models are MMTV-Neu and MMTV-PyMT, which result in the development of multifocal adenocarcinoma and metastatic lesions in the lungs and/or lymph nodes. MMTV-Neu mice have been used for modeling epidermal growth factor receptor 2 (HER2)positive breast cancer, whereas MMTV-CyclinD1 for estrogen receptor (ER)-positive breast cancer. MMTV-PyMT mice lose the expression of ER and progesterone receptor (PR) as they progress and concomitantly gain androgen receptor (AR) expression, therefore could be used for modelling luminal AR (LAR) positive Triple-Negative Breast Cancer (TNBC). [3] However, only a small fraction of TNBC patients (≈15%) are positive for AR, while the majority have been classified into different molecular subtypes, including basal-like (BL1 and BL2) and mesenchymal (M). [4,5] TNBC, which accounts for ≈10-15% of all breast cancer patients, is defined by the lack of ER and PR expression, as well as by the absence of HER2 amplification/overexpression. Compared to the other breast cancer subtypes, TNBC is characterized by the earliest age of onset, a high propensity for metastasis, and the worst prognosis in terms of relapse and survival rate. [6][7][8] Over 80% of TNBC patients exhibit alterations in the TP53 locus, [9] whereas a smaller fraction has mutations in genes controlling the PI3K pathway and the homologous recombination machinery (BRCA1/2). A molecular feature of TNBC is the dependency of cancer cells on signals that are rarely mutated, a phenomenon defined as "nononcogene addiction." [10] Collectively, these traits are among the leading cause of limited efficacy of current TNBC therapies. Radiation therapy and chemotherapy, applied before and after surgery, are the mainstay of treatment, although frequently associated with drug resistance and recurrent disease. [6][7][8] Extensive efforts have been made to search for molecular targeted therapies effective for TNBC treatment. Although some targeted therapies approved for treatment of other cancer types have been proposed in TNBC, they rarely turned out to be clinically relevant. [11] These limited responses are associated with the high heterogeneity of the disease and the lack of suitable immunocompetent preclinical models that recapitulate the molecular diversity of TNBC. Among potential targets for TNBC subsets are Poly(ADP-Ribose) Polymerase 1 (PARP1), AR, vascular endothelial growth factor receptor, epidermal growth factor receptor (EGFR), MET, PI3K/mTOR, MEK, Cyclin-dependent kinases (CDKs), heat shock protein 90 (HSP90), histone deacetylase (HDAC), hypoxia-inducible factor 1-(HIF1-), and integrins. [11,12] Inhibition of WEE1 kinase has been proposed as a promising treatment option for TNBC and several other types of solid cancer. [13,14] WEE1 plays a central role in the G2/M checkpoint and controls DNA synthesis as part of the S phase checkpoint. Therefore, inhibition of WEE1 is associated with accumulation of DNA damage and aberrant mitosis. Coinhibition of WEE1 with either radiotherapy or anticancer drugs such as cisplatin, gemcitabin, paclitaxel, or inhibitors of CDC25, ATR, or PARP causes death of breast cancer cells. [15][16][17][18][19][20][21][22] The rational of these combined treatments is to associate DNAdamaging therapies together with perturbation of DNA damage checkpoint gatekeepers through WEE1 targeting. Nevertheless, the consequences of WEE1 targeting may be broader than cell cycle regulation, in view of recent studies showing that WEE1 inactivation increases CDK-dependent firing of dormant replication origins thereby leading to replication stress and increased dNTP demand. [23,24] Moreover, WEE1 was reported to modulate Histone H2B phosphorylation to inhibit transcription of several histone genes in yeast and humans and to function as a histonesensing checkpoint in budding yeast. [25,26] We have previously reported the engineering of a unique mouse genetic model in which subtly increased wild-type MET receptor tyrosine kinase (RTK) levels can be triggered in a temporal and spatial manner (R26 stopMet mice). This model has illustrated the vulnerability of specific cells to slightly enhanced RTK levels, notwithstanding the resilience of most cell types. [27][28][29] For example, at adulthood enhanced MET levels in the liver trigger spontaneous formation of hepatocellular carcinoma, recapitulating several features of human patients. [29][30][31][32] Here, we report the generation of a unique mouse model (MMTV-R26 Met mice) in which a subtle increase in the expression levels of the wild-type MET RTK leads to spontaneous TNBC formation. The tumorigenic switch correlated with a critical threshold of MET expression, whereas aggressiveness was associated with high MET levels and discrete signaling reprogramming. Proteomic profiling, signaling network analysis, and machine learning indicated that the MMTV-R26 Met mice not only model different tumorigenic stages of TNBC, but also largely recapitulates heterogeneity of the human disease as well as primary resistance to treatment. We used this unique model to identify potential therapeutic targets for TNBC through signaling reprogramming analysis and provide strong evidence that combination treatment with BCL-XL and WEE1 inhibitors could be a promising therapeutic approach with high clinical impact.

Enhanced Wild-Type RTK MET Expression Levels in the Mouse Mammary Gland Induce Spontaneous TNBC Development
Previous studies showed that expression of oncogenic MET led to the development of diverse mammary tumors with basal characteristics. [33] We assessed the sensitivity of the mammary gland to slightly increased wild-type MET levels by crossing the MMTV-Cre transgenic with R26 stopMet mice (referred to as MMTV-R26 Met ). The specificity of the LacZ-stop cassette deletion obtained by the MMTV-Cre mice [34] was evaluated using the R26 stopMet-Luc mice, [29] in which Met tg is followed by an internal ribosome entry site-Luciferase reporter (Figure 1a). In vivo imaging of female MMTV-R26 Met-Luc mice revealed a strong luciferase signal in mammary glands only after the first lactation (Figure 1b), consistent with the expression of the Cre recombinase following MMTV promoter activation by prolactin. [35,36] This led to removal of the stop cassette, and thus Met tg expression in the mammary gland of MMTV-R26 Met mice ( Figure 1a). Consistently, the Luciferase-positive domains further increased after the second lactation and were significantly reduced in postlactating females, in agreement with involution of the mammary gland occurring when the lactation phase is over (Figure 1b). This imaging analysis exemplifies the remodeling of the mammary gland overtime. In view of a dynamic regulation of the HGF/MET system in mammary gland morphogenesis previously reported, [37,38] we assessed Met and Hgf mRNA levels in MMTV-R26 Met and control mice from the virgin to the postlactation state. RT-qPCR analysis revealed comparable dynamics of Met and Hgf transcript expression in both MMTV-R26 Met and control mice: high levels at virgin state, a progressive downregulation during pregnancy, reaching almost undetectable levels during lactation, and a restoration of Met and Hgf levels at the postlactation stage (Figure 1c). Whereas Met tg expression was undetectable in virgin animals, it became evident starting from the pregnancy stage, coherent with MMTV promoter activation by prolactin, [35,36] and remained expressed during subsequent phases. Western blot analysis confirmed MET tg expression in the mammary gland of MMTV-R26 Met mice ( Figure S1a, Supporting Information). The presence of HGF in the stroma surrounding the gland likely ensures full signaling competence of the MET tg , as we previously reported in embryonic hepatocytes. [27,29,32] We therefore hypothesized that the MMTV-R26 Met mice could be an appropriate genetic setting to assess the vulnerability of the mammary gland to subtle perturbation of wild-type MET levels overtime. In view of remarkable changes occurring during mammary gland morphogenesis, illustrated by our bioluminescence imaging and transcriptional analyses, and the susceptibility of parity-induced mammary epithelial subtypes to signaling perturbations, [39] MMTV-R26 Met mice were kept under repeated cycles of pregnancy. Overtime, a proportion of MMTV-R26 Met mice spontaneously developed mammary gland tumors ( Figure 1d). Remarkably, the kinetic of tumor formation was similar to that of MMTV-ErbB2 mice generated in the same genetic background we used as reference (MMTV-ErbB2 mix ; Figure 1d; Figure S1b, Supporting Information). The percentage of mice with tumors correlated with the severity in RTK alteration: 16% of MMTV-R26 Met mice (with enhanced wild-type MET) developed tumors (32/196) compared to 58% of MMTV-ErbB2 mix mice (with oncogenic HERBB2 overexpression; 11/19; Figure 1d; Figure S1b, Supporting Information). A proportion of MMTV-R26 Met mice with mammary gland tumors also developed lung metastasis (19%; 6/32; Figure S1c, Table S1, Supporting Information). Histological analyses of the MMTV-R26 Met tumors revealed highly aggressive and infiltrating breast carcinomas, which have been histologically identified as being exclusively TNBC (24 tumors analyzed; Figure 1e; Table S1, Supporting Information).

The MMTV-R26 Met Tumor Model Recapitulates Heterogeneity and Primary Drug Resistance of TNBC Human Patients
To further characterize the MMTV-R26 Met mammary tumors, we applied a semiquantitative proteomic profiling through re-verse phase protein array (RPPA), a high-throughput antibodybased technique to analyze protein activities in signaling networks. Analysis of expression and/or phosphorylation levels (247 signals, Table S2, Supporting Information) displayed that the MMTV-R26 Met tumors (n = 24) clearly segregate from control mammary glands (n = 3; Figure 2a). Interestingly, the MMTV-R26 Met tumors form four distinct clusters, highlighting heterogeneity in signaling levels, including the MET phosphorylation status (Figure 2a,b; Figure S1d, Table S3, Supporting Information). Heterogeneity was also observed at Met transcript levels, as revealed by RT-qPCR ( Figure S1e, Supporting Information), reflecting the heterogeneity of MET levels among TNBC patients. [40][41][42] Thus, a slight increase in Met levels in the mouse mammary glands is sufficient to trigger the tumorigenic program of TNBC.
Next, we explored the possibility to classify the MMTV-R26 Met tumors to TNBC subtypes by analyzing the RPPA data applying the Random Forest machine learning algorithm previously used with transcriptomic data. [43] As subtype classification is usually done on transcriptomic data, we first used the RPPA data of 152 TNBC patients from the TCGA dataset to build a model for subtyping prediction. We trained the model using tenfold cross validation to optimize the method parameters ( Figure S1f, Supporting Information). The model was sensitive to the M class (balanced accuracy 0.89), and had lower sensitivity in distinguishing between BL1 and BL2 classes (balanced accuracy of 0.69 and 0.55, respectively). This was done to take into account that most of the patients are BL1, with a consequent 30% correct BL1 prediction called "no information rate." Our Random Forest model had accuracy of 57% (or 71% without distinguishing BL1 and BL2), with a significance of p-value = 0.002 compared to the "noinformation rate." We then applied the model on the RPPA data of MMTV-R26 Met tumors to predict their classification. Remarkably, we found that all TNBC subtypes are represented by the MMTV-R26 Met tumors with an enrichment of the mesenchymal subtype (Figure 2c,d). Collectively, these results showed that a moderate increase of MET levels in the mammary gland is sufficient to perturb tissue homeostasis, is able to initiate the TNBC program including the formation of lung metastasis, and that the resulting tumors recapitulate the heterogeneity characteristic of TNBC patients.
To further exploit the MMTV-R26 Met cancer model, we established and molecularly/biologically characterized six mammary gland tumor (MGT) cell lines from individual MMTV-R26 Met tumors (Figure 3a). Four cell lines, MGT4, MGT9, MGT11, and MGT13 exhibited tumorigenic properties in vivo, illustrated by the formation of tumors when injected heterotopically into the flank of nude mice, whereas the two other lines, MGT2 and MGT7 did not (Figure 3b). The four tumorigenic cell lines exhibited oncogenic features, whereas the MGT2 and MGT7 cell lines did not. In particular, we observed increased MET mRNA and protein levels ( Figure 3c; Figure S2a,b, Supporting Information), with a heterogeneity similar to that observed among MMTV-R26 Met tumors and reported in TNBC patients. [41,42] MGT4, MGT11, and MGT13 (not MGT9) cell lines were capable of forming tumor spheroids when grown in self-renewal conditions (Figure 3d,e). MGT9 cells did not form spheroids, rather adhered to the low-attachment culture dish, correlating with downregulation of proteins involved in stemness such as Notch/Jagged and PI3K/AKT/mTOR pathways ( Figure S2c, Table S4, Supporting Information). Additionally, the tumorigenic cell lines were characterized by a high proliferation index, with a low proportion of cells in the G0 cell cycle phase ( Figure S3a-d, Supporting Information). Cells of the tumorigenic lines also exhibited increased motility, particularly for MGT13 cells that display a rather mesenchymal-like morphology compared with the other cell lines (Figure 3f; Figure S2a, Supporting Information). Furthermore, these cell lines also recapitulated the heterogeneity of p53 alterations observed in TNBC patients: p53 overexpression (likely oncogenic) in MGT4 and MGT9, decreased expression of p53 in MGT13, and comparable p53 levels in MGT11 ( Figure S3e, Supporting Information).
Interestingly, the tumorigenic MGT cell lines (MGT4, MGT9, MGT11, and MGT13) were resistant to conventional chemotherapeutic agents, such as Docetaxel, Cisplatin, 5-Fluorouracil (5′FU), and only partially sensitive to Doxorubicin, although only at high doses ( Figure 3g). Furthermore, all these MGT cell lines were resistant to three drug combinations previously reported to be effective for TNBC treatment: combined inhibition of EGFR+MET, PI3K+MEK, and EGFR+PYK2 [44][45][46] (Figure 3h). Together, these results show that the MMTV-R26 Met -derived cell lines are a relevant model to study as well drug resistance, an important feature of TNBC.

Signaling Network Analysis of MMTV-R26 Met Tumor Derived Cells
To further characterize MMTV-R26 Met MGT cells, we examined their signaling status by RPPA and subsequent bioinformatics analysis (247 epitopes, listed in Table S2, Supporting Information). The signaling profiles highlighted two major features, as illustrated by principal component analysis (PCA). First, the tumorigenic MGT4, MGT9, MGT11, and MGT13 cells clearly segregate from the two types of nontumorigenic cells: MGT7 and   (Figure 4a; Figure S4a, Table S4, Supporting Information). MGT2 cells, which express very low level of MET, can be considered as pretumorigenic. It is therefore tempting to speculate that critical levels of MET might establish a threshold for a tumorigenic switch, while higher MET levels are associated with aggressiveness. Second, the four tumorigenic MMTV-R26 Met MGT cell lines fall into two distinct TNBC subtypes that we named "subtype A" for MGT4, MGT9, MGT11, and "subtype B" for MGT13 (Figure 4a; Figure S3a, Table S4, Supporting Information). These two subtypes display distinct phenotypic features and MET levels ( Figure S2a,b, Supporting Information). Strikingly, by PCA analysis we could segregate into "subtype A" and "subtype B" both MMTV-R26 Met MGT cells and tumors (Figure 4b). Additionally, we identified ARID1A, Claudin-7, and E-Cadherin as hallmark of the "subtype A" (Figure 4c; Table S5, Supporting Information).
To obtain insights on molecular and cellular functions characterizing "subtype A" from "subtype B," we performed a series of enrichment analyses by applying Enrichr, a web-based tool to highlight enrichments based on gene sets. Both subtypes showed an enrichment in pathways related to DNA repair, cell cycle regulation, and metabolism ( Figure 4d; Figure S4b, Supporting Information). These enrichments are consistent with enhanced proliferation capacity of MGT4, MGT9, MGT11, and MGT13 cells versus nontumorigenic cells. Moreover, "subtype B" is enriched in pathways related to stemness properties (Figure 4d; Figure S4b, Supporting Information), consistent with the enhanced capability of MGT13 to form tumor spheroids in vitro (Figure 3d,e). Further analysis of RPPA data using the Limma package highlighted differences between the RPPA profiles of subtypes A and B versus the nontumorigenic MGT cells. In particular, we detected upregulation of (a) Bim, which might sensitize the cells to antiapoptotic drugs, (b) CDK1 and RAD50, which are implicated in cell cycle regulation and DNA damage response ( Figure 4e; Table  S6, Supporting Information). Biochemical studies supported the RPPA results and revealed consistent upregulation of oncogenic signals in MGT4, MGT9, MGT11, and MGT13 compared with control cells. This included phosphorylation of MET, EGFR, of their downstream adaptor GAB1, of MEK/ERKs, AKT, RB, and elevated antiapoptotic signals such as MCL1, BCL-XL, and XIAP ( Figure 4f).

Combined Targeting of WEE1 and BCL-XL Is Deleterious for TNBC Cells
Inspired by the signaling profiles of MGT cells and tumors, we designed a drug screen aiming at identifying combinatorial treatments effective for the two subtypes of TNBC cells modeled by the MMTV-R26 Met mice. Among all treatments tested in the MGT4 cell line (single or combined drugs), we uncovered that the simultaneous inhibition of BCL-XL and WEE1 drastically reduced tumor cell viability ( Figure S5a MGT7; Figure 5a,b). Importantly, this highlights lack of toxic effect of the newly identified drug combination. Combined inhibition of BCL-XL and WEE1 was synergistic (for 3 out of 4 MMTV-R26 Met MGT cell lines), as shown by the Bliss score and by the Chou-Talalay combination index score calculation (Figure 5a; Figure S5c, Supporting Information). Furthermore, BCL-XL+WEE1 targeting was detrimental for all six human TNBC cells tested (Figure 5c). Intriguingly, when this drug combination was tested on human non-TNBC cells, we found that inhibition of BCL-XL did not exacerbate the effects elicited by WEE1 targeting (Figure 5d), indicating that WEE1 inhibition is particularly detrimental in TNBC cells.
Recent studies have reported the sensitivity of TNBC to WEE1 targeting in the presence of either PARP or ATR inhibitors, or Cisplatin. [17,19,20,22,47] However, we found that MMTV-R26 Met MGT cells were either resistant or only partially sensitive to these drug combinations (Figure 5e), recapitulating other mechanisms of primary resistance beside those reported in Figure 3g,h. Thus, BCL-XL targeting is a preferable strategy to exacerbate WEE1 essentiality in TNBC.
Finally, we assessed in vivo the potency of BCL-XL+WEE1 cotargeting on tumor growth. We engineered a cell line for in vivo imaging by stably transfecting the MGT11 cells, characterized by strong tumorigenic properties, with a Luciferase reporter vector (defined as MGT11 Luc ). We confirmed that the MGT11 Luc cells have comparable biological properties as the parental cells, and maintain sensitivity to combined BCL-XL+WEE1 targeting (Figure S5d,f, Supporting Information). Orthotopic studies showed that combinatorial BCL-XL+WEE1 inhibition reduced in vivo tumor growth of MGT11 Luc cells injected into the mammary fat pad of mice (Figure 5f-i; Figure S5g, Supporting Information). No obvious effects on mouse viability or murine weight indicated the lack of toxicity. Thus, the MMTV-R26 Met model recapitulating heterogeneity and resistance of TNBC led us to uncover a new potent drug combination based on BCL-XL and WEE1 inhibition, effective on human TNBC cell lines characterized by distinct features.

BCL-XL Inhibition Exacerbates WEE1 Requirement in TNBC Cells
While BCL-XL primarily has an anti-apoptotic function by inhibiting cytochrome C release, WEE1 acts on multiple regulatory circuits. Besides its well-established involvement in regulating the G2/M transition through phosphorylation of CDK1, recent studies have highlighted additional mechanistic roles of WEE1 in DNA replication stress and regulation of histone synthesis and levels. Therefore, we thoroughly examined the signaling changes occurring following BCL-XL and WEE1 inhibition in MGT cell lines by RPPA analysis (426 epitopes were analyzed, Table S2, Supporting Information); some changes were validated by Western blot studies. As examples, the profile of proteins differentially expressed and/or phosphorylated in MGT4 is displayed in Figure 6a (Table S7, Supporting Information). Interestingly, while BCL-XL inhibition alone had modest effects, WEE1 inhibition had marked effects on the cells, many of which were accentuated by treatment with the BCL-XL inhibitor. Among identified changes, some were consistently observed in all MGT cell lines, whereas others were specific to individual MGT lines. These changes covered a broad range of signaling/cellular functions, such as those associated with cell survival/death, cell cycle  (Table S5, Supporting Information). Unpaired Student's t-test was used. * P < 0.05; *** P < 0.001. d) Proteomic profiles of cells belonging to the different clusters (shown in (a)) were compared to identify enrichments by applying the Enrichr software. Histograms report the enriched cell signaling pathways, using the Reactome database, ordered according to the combined score. The 20 top ranked enrichments are highlighted. Note that the majority of the changes are related to signals (indicated by arrowheads) involved in DNA repair (red), metabolism (blue), cell cycle regulation (yellow), and stemness (green) (indicated by colored arrowheads). e) Graph showing the fold change (Log 2 ) of protein phosphorylation or expression between "subtype A" (x-axis) or "subtype B" (y-axis) cell lines versus the nontumorigenic cells (MGT2, MGT7). Proteins among the highest significant differentially expressed in "subtype A" (in blue), "subtype B" (in red), or both (in purple) are indicated (Table S6, Supporting Information). p-values were determined by the Limma package in R. f) Western blot analysis of total protein extracts from MMTV-R26 Met MGT cells. Actin and Ponceau S stainings were used as loading controls in all studies. At least two independent experiments were performed. www.advancedsciencenews.com www.advancedscience.com regulation, DNA damage/repair, histone levels, and oncogenic properties (Figure 6b; Table S7, Supporting Information).
Consistent with the well-known regulatory activity of WEE1 in cell cycle progression, we observed decreased levels of phosphoY 15 CDK1 (the direct target of WEE1), phosphoS 795 RB, and CHK1 expression upon WEE1 inhibition (Figures 6b and 7a; Table S7, Supporting Information). This was accompanied by an alteration in the distribution of cells in cycle phases as shown by FACS profiles (Figure 7b). Concerning cell survival signals, combined BCL-XL+WEE1 targeting in MGT cells led to a drastic downregulation of MCL1 and XIAP antiapoptotic signals associated with intense cleavage of Caspase3 and PARP (Figures 6a,b  and 7a). Regarding DNA damage and repair, we observed an upregulation of phosphoS 1987 ATM and phosphoS 139 H2AX (histone variant, H2AX), reflecting increased levels of DNA damage in the MGT cells upon treatment (Figures 6a,b and 7a; Figure S6a, Supporting Information). This high proportion of DNA damage is associated to the downregulation of Rad51 (Figure 6a,b), which plays a major role in double-strand break (DSB) repair by homologous recombination and in fork protection, and restart during replication stress, [48] raising the possibility that this downregulation is related to the increase of phosphoS 139 H2AX. Moreover, we observed a drastic downregulation of RRM2, a subunit of the ribonucleotide reductase required to maintain high levels of dNTPs, [23] with an upregulation of Histone-H3 and H3K9me2 levels (Figure 6a-d; Figure S6b, Table S7, Supporting Information). This was further confirmed by cell fractionation studies, showing downregulation of RRM2, and increased pS 33 RPA32 in the chromatin fraction of cells co-treated with WEE1 and BCL-XL inhibitors (Figure 7e). Finally, concerning oncogenic signals, we found a significant downregulation of phosphorylation levels of MET and GAB1 (Figure 7a). Collectively, these results indicated that BCL-XL inhibition exacerbates the dependence of TNBC cells on the overall functions exerted by WEE1: an intact dNTP pool (by stabilizing RRM2 protein levels), appropriate histone levels, and proper cell cycle progression through G2/M.

Combined BCL-XL and WEE1 Inhibition Leads to Mitotic Catastrophe and Apoptosis of MMTV-R26 Met TNBC Cells
We explored at cellular levels the biological events associated with BCL-XL and WEE1 inhibition by immunocytochemistry. In cells experiencing the combined treatment, we found a significant increase in the number of H2AX-positive cells as well as a raise in intensity of H2AX staining per cell, reflecting an accumulation of DNA DSBs (Figure 8a). These findings are in agreement with the above results (Figure 7a; Figure S6a, Supporting Information) and reflect increased DNA damage in a high proportion of cells following BCL-XL+WEE1 targeting. In addition, we found a striking increase in phosphoS 10 Histone H3 (pH3)positive cells when subjected to the combined treatment (Figure 8b), suggesting that a high proportion of cells are in G2/M. [49] This could reflect a premature entry in mitosis due to WEE1 inhibition, but also an accumulation of unrepaired DNA damage in mitosis. [13,50] We investigated the consequences of this premature mitotic entry by performing a double immunostaining with anti--Tubulin and anti-pH3 antibodies in cells treated (or not) with BCL-XL+WEE1 inhibitors. Interestingly, the staining highlighted a marked increase of cells harboring mitotic catastrophe revealed by monopolar, multipolar, or disorganized spindles, and even cytokinesis failure (Figure 8c). The results further showed that BCL-XL inhibition exacerbated the effects of WEE1 targeting by forcing cells to exit mitosis without undergoing complete chromosome segregation, a phenomenon called mitotic slippage. As a consequence, these excessive unscheduled and abnormal mitosis events led to an increased formation of micronuclei in treated cells (Figure 8d).
Finally, we assessed the terminal event associated with combined BCL-XL and WEE1 targeting on TNBC cells. We found that a pan-Caspase inhibitor (Z-VAD-FMK) significantly rescued cell death caused by the combined treatment of TNBC cells (Figure 8e,f). This result was consistent with Western blots analysis showing a strong increase in Caspase3 and PARP cleavage upon combined BCL-XL+WEE1 inhibition (Figure 7a). By contrast, inhibition of ferroptosis, another cell death mechanism to which TNBC cells are highly sensitive, [51] did not prevent cell death (Figure 8e,f). Together, these findings indicate that the combined targeting of BCL-XL and WEE1 exacerbates the dependency of TNBC cells on WEE1 function in a context of low antiapoptotic inputs, leading to mitotic catastrophe and apoptosis.

Discussion
In this study, we have developed a new TNBC mouse model that recapitulates primary drug resistance, we uncovered that the combined inhibition of WEE1 and BCL-XL selectively kills mouse and human TNBC cell lines, and provided mechanistic insights into inhibition of WEE1 and BCL-XL in TNBC cells.  A number of transgenic mice have been engineered to model TNBC, predominantly through drastic genetic manipulations (often combined), such as loss-of-tumor suppressors and overexpression of activated oncogenes. [52] Although they have been instrumental to implicate candidate genes as oncogenes, each model generally recapitulates a fraction of disease features predominantly associated with the genetic manipulation employed. Patient-derived xenografts capture the heterogeneity that characterizes cancers like TNBC. However, transplantation-based models in immunocompromised mice do not report on the reciprocal crosstalk between cancer and immune cells, a limitation that can be overcome, in part, by laborious and expensive "human-ized models." However, most murine TNBC models do not recapitulate the formation of spontaneous cancers occurring in human patients. In this respect, the MMTV-R26 Met model is rather unique for a series of features. i) Tumors developed by MMTV-R26 Met mice are exclusively TNBC rather than covering a range of breast cancer types. This is not the case for other transgenic mice overexpressing MET oncogenic forms in the mammary gland, [33,[53][54][55] generated in the past because of the implication of MET in breast cancer pathophysiology. Indeed, MET is overexpressed in about 40% of breast cancer patients (in luminal A: 36%; in Luminal B: 39%; in HER2+: 48%; in TNBC: 53% [56] ), and its overexpression often correlates with poorly differentiated and aggressive forms of the disease. [41] In TNBC, MET is particularly highly expressed, and implicated in malignancy progression ( Figure S7, Supporting Information), metastasis, and resistance to anticancer therapies. [40,41] Consequently, agents targeting MET are actively being explored for clinical purposes, [57,58] including in TNBC. [59][60][61] ii) In MMTV-R26 Met mice, the TNBC program is driven by a subtle increase of MET levels in the mammary gland, and by the wild-type form of MET rather than oncogenic versions of MET (mutated MET or TPR-MET). [33,53,54] Consequently, MMTV-R26 Met TNBC cells are not addicted to MET, a feature predominantly characterizing cancers induced by driver oncogenes. This might explain why the MMTV-R26 Met model reported here recapitulates the tumor heterogeneity typical of TNBC patients. iii) TNBC heterogeneity is recapitulated by MMTV-R26 Met mice at different levels, and heterogeneity is maintained in MMTV-R26 Met cell lines established from independent tumors. iv) MMTV-R26 Met TNBC recapitulates primary resistance to conventional chemotherapy and to a set of targeted molecular treatments reported in previous studies. [45] Histologically, the 28 MMTV-R26 Met tumors analyzed revealed differences in their grade (although the majority were highgrade), with a range of low to high mitotic index, and the absence or presence of necrotic areas. Heterogeneity is also evidenced by processing our RPPA analysis of 24 MMTV-R26 Met tumors through machine learning, indicating that the 4 TNBC subtypes (BL1, BL2, LAR, and mesenchymal) are indeed represented, with the most aggressive mesenchymal subclass enriched. Diversity was preserved in vitro by the four MMTV-R26 Met MGT cells we established from distinct tumors. Heterogeneity was maintained even concerning MET levels in TNBC, as shown for example by the different levels of MET expression and activation in MMTV-R26 Met tumors and cells (for example, phosphorylation levels of MET and of GAB-1, a MET downstream effector). It is tempting to speculate that although all tumors in MMTV-R26 Met mice originate from a common genetic setting characterized by a slight increase of MET levels, this context does not impose an oncogenic path in which MET would be systematically altered in all tumors to the same extent. Finally, MMTV-R26 Met MGT cells are also heterogeneous in terms of drug sensitivity: for example, while MGT9 and MGT11 are resistant to single WEE1 targeting, MGT4 and MGT13 are partially sensitive.
It is rather surprising that a subtle increase of MET levels, in its wild-type form, spontaneously initiates a destabilization process that fully recapitulates the whole TNBC program. Nevertheless, this sensitivity to MET levels is conditioned by a multiparous context, as females without multiple pregnancies did not develop tumors. Both MET and HGF are dynamically expressed during pregnancy/lactation, as we showed here consistent with previous reports. [37,38] Moreover, the HGF/MET system regulates mammary gland morphogenesis, especially ductal branching and proliferation of ductal end buds. [62] These data illustrate how a tight regulation of the time and signal input levels required for the mammary gland remodeling is critical to prevent transformation. The vulnerability of the mammary gland to a slight increase in MET levels resembles the susceptibility of the liver we reported in previous studies. [29][30][31][32] The vulnerability of the mammary gland and the liver is contrasted by a remarkable resilience of other tissues, in which a tumorigenic event requires additional genetic alterations, as reported for malignant peripheral nerve sheath tumors. [63] Whether such a mild MET perturbation in the mammary gland occurs in specific subgroups of women and/or physiological contexts and can increase susceptibility to tumor development remains an open issue. If this is not the case, such genetic manipulation nevertheless makes it possible to initiate a cascade of molecular events leading to a clinically relevant TNBC context.
The second major finding of this work is that combinatorial inhibition of WEE1 and BCL-XL kills a panel of heterogeneous MMTV-R26 Met and human TNBC cell lines. For decades, WEE1 has been considered primarily as a key regulator in cell cycle progression. [13,14] In particular, WEE1 regulates the G2/M checkpoint through phosphorylation and inactivation of CDK1, thus preventing entry of cells with unrepaired DNA damage into mitosis. [13] Nevertheless, additional mechanistic functions of WEE1 have recently emerged. Indeed, WEE1 stabilizes RRM2 protein, a regulatory subunit of the ribonucleotide reductase required to maintain high dNTPs levels. [23] In addition, WEE1 was reported in yeast and human to inhibit transcription of several histone genes by phosphorylating Histone H2B at Tyr 37 . [25] In addition, the WEE1 yeast homolog Swe1 WEE1 was recently reported to act as a histone-sensing checkpoint by sensing excess histone levels before cells enter mitosis, thus preventing aberrant chromosomal segregation and polyploidy. [26] Thus, WEE1 targeting  Figure S3e, Supporting Information), to document its upregulation upon drug treatment. At least two independent experiments were performed. b) Graph reporting the distribution of cells treated with the indicated drugs (for 12 h), compared to untreated cells (no), in each phase of the cell cycle as determined by flow cytometry using PI and Ki67 staining. Three independent experiments were performed. Statistical analyses were performed by two-way ANOVA followed by Tukey test, and are reported in Table S11 of the Supporting Information. c) Graphs reporting changes by RPPA in levels of Histone H3, H3K9me2, RRM1, and RRM2 in MGT cells exposed to A1155463, Adavosertib, or in combination. Distinct MGT cell lines are indicated in colors. For multiple comparisons, statistical significance was assessed by one-way ANOVA followed by Tukey test.  might affect several key cellular processes that are particularly relevant in cancer cells as they proliferate at high rates and are more prone to replication stress with higher demands in dNTP and histones. WEE1 inhibition may therefore particularly expose cancer cells to DNA damage. This is reflected by the marked increase of cells harboring mitotic catastrophe upon the combined inhibition of WEE1 and BCL-XL in TNBC cells.
WEE1 is an attractive target for cancer therapies including for TNBC, and strategies are being intensively explored in preclinical studies and clinical trials. It has been recently reported that WEE1 targeting, in combination with either cisplatin or inhibitors of ATR or PARP is effective in human TNBC cells lines. [17,19,20,22,47] By testing them in MMTV-R26 Met MGT cells, mimicking primary resistant treatment contexts, we have shown that these three combinations are indeed effective, albeit to a varying degree and depending on the cell line. Nevertheless, the combinatorial targeting of WEE1 together with BCL-XL elicits superior effects, as shown by the loss of viability of all four very aggressive/highly tumorigenic MMTV-R26 Met MGT cell lines and of the six human TNBC cell lines tested. Interestingly, such vulnerability is specific to TNBC cells as three out of four non-TNBC cell lines were resilient to WEE1 plus BCL-XL inhibition. This resilience, as well as the absence of effects on two nontumorigenic MMTV-R26 Met MGT cell lines (MGT2, MGT7) highlights two relevant points. First, the reduction of the stress support pathway by targeting BCL-XL exacerbates a specific requirement of WEE1 in TNBC. This effect resembles an "essentiality-induced" synthetic lethality, characterized by the essentiality of one gene following the targeting of a second gene. [64] Nevertheless, we cannot exclude that BCL-XL targeting may also contribute to altered cell cycle progression. [65] Second, in addition to the absence of in vivo side effects, BCL-XL plus WEE1 targeting appears to be a rather safe treatment for healthy cells.

Conclusion
We propose that the MMTV-R26 Met genetic setting we have generated is a relevant model for molecular and preclinical studies on TNBC in an immunocompetent context. The usefulness of the MMTV-R26 Met model is further strengthened by its capability to recapitulate TNBC heterogeneity and primary resistance to treatments. As cells derived from MMTV-R26 Met tumors maintain these features, they constitute a valuable cellular model to explore TNBC molecular/biological properties, to screen TNBC therapeutic options, and to investigate the immune-cancer cell crosstalk through syngeneic orthotopic studies. We illustrated how the combination of this unique model with proteomic profiling, signaling network analysis, and machine learning can lead to the identification of a new, potent drug combination for TNBC treatment, based on WEE1 and BCL-XL targeting. Our findings may be particularly relevant from a translational perspective, considering that agents targeting WEE1 or BCL-XL are already in phase I/II clinical trials.
RPPA: Protein lysates of dissected mammary gland tumors (n = 24), control mammary glands (MMTV and MMTV-R26 Met ), and MMTV-R26 Met MGT cells either not treated or treated for 12 h with A1155463 (1 × 10 −6 m), Adavosertib (3 × 10 −6 m), or A1155463 + Adavosertib (1 × 10 −6 m, 3 × 10 −6 m) were prepared according to the MD Anderson Cancer Center platform instructions. Samples were screened with 426 antibodies to identify signaling changes in protein expression and phosphorylation levels. . Statistics refer to cell viability obtained in presence of the apoptosis or ferroptosis inhibitors compared to A1155463+Adavosertib alone. The efficiency of Ferrostatin-1 was assessed in presence of Erastin, a ferroptosis inducer. Three independent experiments were performed. Data are expressed as means ± s.e.m. For multiple comparisons, statistical significance was assessed by one-way ANOVA followed by Tukey test. not significant (ns) P > 0.05; ** P < 0.01; *** P < 0.001.