The mitotic checkpoint is a targetable vulnerability of carboplatin-resistant triple negative breast cancers

Triple-negative breast cancer (TNBC) is the most aggressive breast cancer subtype, lacking effective therapy. Many TNBCs show remarkable response to carboplatin-based chemotherapy, but often develop resistance over time. With increasing use of carboplatin in the clinic, there is a pressing need to identify vulnerabilities of carboplatin-resistant tumors. In this study, we generated carboplatin-resistant TNBC MDA-MB-468 cell line and patient derived TNBC xenograft models. Mass spectrometry-based proteome profiling demonstrated that carboplatin resistance in TNBC is linked to drastic metabolism rewiring and upregulation of anti-oxidative response that supports cell replication by maintaining low levels of DNA damage in the presence of carboplatin. Carboplatin-resistant cells also exhibited dysregulation of the mitotic checkpoint. A kinome shRNA screen revealed that carboplatin-resistant cells are vulnerable to the depletion of the mitotic checkpoint regulators, whereas the checkpoint kinases CHEK1 and WEE1 are indispensable for the survival of carboplatin-resistant cells in the presence of carboplatin. We confirmed that pharmacological inhibition of CHEK1 by prexasertib in the presence of carboplatin is well tolerated by mice and suppresses the growth of carboplatin-resistant TNBC xenografts. Thus, abrogation of the mitotic checkpoint by CHEK1 inhibition re-sensitizes carboplatin-resistant TNBCs to carboplatin and represents a potential strategy for the treatment of carboplatin-resistant TNBCs.


Results
Carboplatin resistance in TNBCs leads to global proteome rewiring to resist carboplatin-induced oxidative and genotoxic stress. To understand the mechanisms underlying carboplatin resistance, we generated a carboplatin-resistant model of TNBC. Consistent with its basal-like 1 phenotype, the TNBC cell line MDA-MB-468 showed high sensitivity to carboplatin treatment (Fig. 1A). By continuously exposing MDA-MB-468 cells for 10 months to gradually increased concentrations of carboplatin, we generated MDA-MB-468 resistant cells (468-R) that are able to proliferate in the presence of 2 µM carboplatin (Fig. 1A,B). The resistant cells showed a tendency to grow at higher rate compared to parental MDA-MB-468 cells even in the absence of carboplatin. Thus, the carboplatin resistance in our model could not be explained by decreased growth rate. The dose of carboplatin needed to inhibit cell growth by 50% (IC50) was more than five times higher for 468-R cells (10.75 µM) than parental MDA-MB-468 (2.00 µM) (Fig. 1C). Five days of incubation with carboplatin significantly impaired the growth of sensitive MDA-MB-468 cells, as detected by a reduced fraction of cells in G1 and S phases of the cell cycle and accumulation in the G2/M phase. On the other hand, carboplatin did not affect the cell cycle distribution of 468-R cells (Fig. 1D), indicating that 468-R cells are capable of proliferating in the presence of carboplatin. Furthermore, five days of carboplatin treatment led to apoptosis of parental MDA-MB-468 cells, but not of resistant 468-R cells as detected by Annexin V staining (Fig. 1E).
To elucidate the underlying mechanics of carboplatin resistance, we performed proteome profiling of parental MDA-MB-468 and carboplatin-resistant 468-R cells in the presence or absence of 2 µM carboplatin. To interrogate the differences between parental and resistant TNBC cells, we performed unsupervised hierarchical clustering of differentially expressed proteins in MDA-MB-468 and 468-R cells. Carboplatin resistance was associated with a global proteome re-wiring of carboplatin-resistant cells. We identified 6 clusters of proteins with similar expression patterns, highlighting differences between parental and resistant cells (Cluster 3 and 4), or linked to response to carboplatin treatment in parental (Cluster 1 and 6) and resistant cells (Cluster 2 and 5) ( Fig. 2A). www.nature.com/scientificreports/ We next performed an enriched pathway analysis of these protein clusters using Ingenuity Pathway Analysis (IPA) (Fig. 2B). The pathway analysis revealed that carboplatin resistance was accompanied with a rewiring of the translation-related pathways, such as EIF2 Signaling and Regulation of eIF4 and p70S6K. Carboplatin-resistant cells also exhibited several metabolic alterations, affecting the Sirtuin Signaling Pathway, Mitochondrial Dysfunction, and Oxidative Phosphorylation ( Fig. 2A,B). Strikingly, the same pathways were significantly altered in response to carboplatin treatment in the resistant cells, highlighting the contribution of these pathways to carboplatin resistance. These pathways not only allow adaptation to carboplatin-induced oxidative stress [21][22][23] , but also up-regulate nucleotide biosynthesis 21-23 , a process crucial to coping with DNA damage, while maintaining proliferation [24][25][26] . Concordantly with the proteome analysis, we detected dramatically lower levels of ROS, mitochondrial-specific superoxide as well as a lower ratio of oxidized Nicotinamide Adenine Dinucleotide (NAD) to reduced NADH in carboplatin-treated 468-R cells compared to sensitive carboplatin-treated MDA-MB-468 cells (Fig. 2C-E).
Metabolic rewiring and low ROS levels in carboplatin-resistant cells could explain similar levels of DSBs in untreated and carboplatin-treated 468-R cells, as detected by flow cytometry analysis and immunocytochemistry of phosphorylated histone H2AX (γ-H2AX) (Fig. 2F, Supplementary Figure S1A,B). We also did not find any alterations in the DNA damage or checkpoint pathways upon carboplatin treatment of 468-R cells ( Fig. 2A (Fig. 2B). Altogether, these data indicate that acquisition of carboplatin resistance in TNBC is linked to drastic proteome rewiring, leading to alterations in metabolism and upregulation of anti-oxidative stress responses to overcome DNA damage and support cell proliferation.
The mitotic checkpoint is essential for survival of carboplatin-resistant cells. To    www.nature.com/scientificreports/ on average, for each of 752 kinases or kinase-related genes (9421 shRNAs in total) in carboplatin-sensitive parental MDA-MB-468 cells and their carboplatin-resistant 468-R counterpart. After puromycin selection (refers as T0), the cells were exposed to either vehicle, or carboplatin at a concentration that inhibited their growth by 70%; 0.3 µM for MDA-MB-468 cells and 2.0 µM for 468-R cells for more than 10 doublings (T10 vehicle or T10 carboplatin) (Fig. 3A). To identify genes, for which targeting shRNAs showed overall depletion, we calculated a score for each gene using normalized shRNA expression. Using the ANOVA method, we identified essential genes by comparing T0 and T10 scores, whereas the comparison of gene scores in T10 vehicle and T10 carboplatin identified carboplatin-survival genes. Consistent with the global proteome rewiring of carboplatin-resistant cells, we observed only a partial overlap between the genes identified for parental MDA-MB-468 and resistant 468-R cells (Fig. 3A,B).     TPK1  DYRK1A  PHKA2  CDKL5  PIK3CG  GRK1  RIPK2  PACSIN1  ATR  TRRAP  BRAF  MARVELD3  NME3  MPP6  PKD2L2  NPR2  CHKA  CRKL  CLK1  TRAD  CDKL3  CALM2  CDK17  PI4KA  RAB32  CLK4  YSK4  DLG1  IP6K2  ERBB3  TK2    www.nature.com/scientificreports/ by Anaphase Promoting Complex/Cyclosome (APC/C) dysfunction in several cisplatin-resistant models of epithelial ovarian cancer 27 .
On the other hand, String pathway analysis of carboplatin-survival genes for 468-R cells showed enrichment of the G2/M checkpoint components, G2 Checkpoint Kinase WEE1, Cell Division Cycle 7 (CDC7), and CHEK1 (Fig. 3E). The G2/M checkpoint pathway regulates the cell cycle in response to replication stress and DNA damage, including the DNA-crosslinks caused by platinum agents 28 . The kinase CHEK1, a major component of the pathway, facilitates the continuation of cell division by stabilizing the replication fork and activating DNA damage repair mechanisms 26 . The WEE1 kinase, downstream of CHEK1, is a CDK antagonist that slows the cell cycle to adapt to genotoxic stress [29][30][31] . Even though platinum agents are well known to induce DNA damage and replication stress, carboplatin-treated 468-R cells exhibited low levels of DNA damage. Moreover, we found similar cell cycle distributions and comparable levels of 5-Ethynyl-2´-deoxyuridine (EdU)-incorporation in 468-R in the absence or presence of carboplatin (Fig. 1D, Supplemental Figure S1C). This suggests that depletion of CHEK1 or WEE1 re-sensitizes 468-R cells to carboplatin by mechanisms other than abrogation of the G2/M checkpoint control.
Another well-described function of CHEK1 is its role in the maintenance of spindle checkpoint function 30 , delaying anaphase until proper alignment of chromosomes 31 . CHEK1 is also required for stable metaphase arrest 32 , whereas CHEK1 deficient cells show decreased localization of the checkpoint kinase BUBR1 to kinetochores 30 . WEE1-mediated inhibition of CDK1 could block mitotic entry, whereas CDC7 kinase was reported to stimulate Aurora B kinase activity during mitosis 33,34 . Given that we observed mitotic checkpoint aberration in carboplatin-resistant 468-R cells (Figs. 2B, 3C), this suggests that abrogation of mitotic control through blockade of WEE1 or CHEK1 might induce death of carboplatin-treated 468-R cells. Moreover, we found increased levels of CHEK1 activity in carboplatin-resistant 468-R cells compared to sensitive parental cells, further confirming the dependency of carboplatin-resistant cells to CHEK1 activity (Fig. 3F). In summary, our results indicate that targeting of the mitotic checkpoint could re-sensitize carboplatin-resistant TNBCs to carboplatin.

Pharmacological inhibition of CHEK1 or WEE1 re-sensitizes carboplatin-resistant cells.
To validate the results of the shRNA screen, we assessed the growth of carboplatin-resistant 468-R cells after suppression of CHEK1 or WEE1 by two independent short hairpins. In agreement with the screen results, we found that depletion of either CHEK1 or WEE1 inhibited the growth of carboplatin-treated 468-R cells, confirming the dependence of carboplatin-resistant cells on these checkpoint kinases (Fig. 4A,B).
Next, we assessed the efficiency of pharmacological inhibition of CHEK1 or WEE1 in re-sensitization of carboplatin-resistant cells to carboplatin treatment. We chose several CHEK1 and WEE1 inhibitors, for which clinical data are available. AZD7762, a CHEK1/CHEK2 inhibitor, showed a complete response in a patient with metastatic small-cell renal cancer when combined with irinotecan 35 . The more selective CHEK1 inhibitor rabusertib was well tolerated in combination with gemcitabine 36 . Prexasertib, an even more selective CHEK1 inhibitor, has been tested as monotherapy 37 . It showed only transient adverse events, and was able to induce modest responses as monotherapy in patients with advanced cancer 37,38 . For WEE1 inhibition, we selected adavosertib as it was shown to effectively inhibit WEE1 activity in advanced cancer patients 39 , and the combination of adavosertib with carboplatin has already shown encouraging results in treating carboplatin-resistant ovarian cancer 40 .
We tested the ability of the selected inhibitors to inhibit growth of carboplatin-resistant 468-R cells. As a single agent, all tested CHEK1 and WEE1 inhibitors impaired the growth of 468-R cells only at the highest concentrations or not at all (Supplemental Figure 2A-C). On the other hand, when combined with even low doses of carboplatin all CHEK1/WEE1 inhibitors re-sensitized carboplatin-resistant 468-R cells (Fig. 4C-F). These results indicate that inhibitors of CHEK1 or WEE1 in combination with carboplatin could represent an effective strategy for the treatment of carboplatin-resistant TNBCs.

Low dose prexasertib is non-toxic in combination with carboplatin. Even though all tested
CHEK1 or WEE1 inhibitors suppressed the growth of carboplatin-resistant cells in combination with carboplatin, for in vivo validations we decided to focus on prexasertib, which is currently in phase II of clinical trials. First, we assessed the toxicity of the combinational treatment of carboplatin and prexasertib compared to carboplatin alone. We treated NMRI nude mice with either a low (8 mg/kg) or a high (20 mg/kg) weekly dose of prexasertib in combination with 50 mg/kg carboplatin. We found that the higher dose of prexasertib led to weight loss and yellow skin rash in some mice, whereas we did not observe this in mice treated with the low dose regimen (Fig. 5A).
To assess whether the combinational treatment induces liver injury, we measured the activity of aspartate-(AST) and alanine (ALT) aminotransferases in serum of the treated mice. The activity of both liver enzymes was not increased in the serum of mice treated with the low dose of prexasertib in combination with carboplatin (Fig. 5B,C). Furthermore, standard histological analysis did not reveal prominent changes in the liver of mice following the administration of the combinational treatments as compared with those of mice in the carboplatin-treated group (Fig. 5D). Livers from all mice did show scattered apoptotic or mitotic hepatocytes, probably resulting from carboplatin treatment. The presence of megakaryocytes in the spleens of these mice was also unaffected, which could indicate sustained thrombopoiesis. Moreover, we did not find treatment-related abnormalities in kidney, pancreas, heart and lung (Supplemental Figure S3). Thus, a combination of a low dose of prexasertib with carboplatin is well tolerated by mice. www.nature.com/scientificreports/

CHEK1 inhibition sensitizes carboplatin-resistant TNBC to carboplatin in vivo.
We next assessed the efficiency of the combinational treatment of carboplatin and prexasertib in vivo. Consistent with the results of the in vitro studies, we found that carboplatin did not affect the growth rate of the 468-R xenograft tumors. In contrast, a low dose of prexasertib in combination with carboplatin suppressed 468-R xenograft growth (Fig. 6A). CHEK1 inhibition in the presence of carboplatin also led to decreased number of Ki-67-positive cells (Fig. 6B). As PDX models better reflect patient's responses than cell line models 41 , we tested the carboplatin sensitivity of three PDX models, established from triple negative primary invasive ductal adenocarcinomas. The PDX model, BRC016, derived from a chemotherapy-naïve tumor, showed high sensitivity to carboplatin, achieving a complete response to carboplatin in nine out of ten mice (Fig. 6C). Two other models were generated from two tumors that received non-platinum neoadjuvant chemotherapy consisting of epirubicin, cyclophosphamide, and taxane. The PDX model BRC021 showed only partial response to carboplatin, maintaining a stable disease  www.nature.com/scientificreports/ (Fig. 6D), whereas the HBCx39 xenograft was able to grow slowly in the presence of carboplatin (Fig. 6E). The latter observation is consistent with the poor tumor response when the patient underwent the first line of adjuvant carboplatin. This variability in carboplatin sensitivity of different PDX models reflected the differences in response observed in TNBC patients.
Remarkably, an individual xenograft tumor of the sensitive BRC016 model achieved only a partial response to carboplatin and even resisted a second line of carboplatin. We used this tumor to establish a novel carboplatinresistant PDX model, C4O (Fig. 6F). We found that carboplatin did not inhibit the growth of the C4O xenograft. On the other hand, the combination of carboplatin with a low dose of prexasertib suppressed the growth of this carboplatin-resistant model, resulting in a stable disease (Fig. 6F). Similarly, while carboplatin reduced the growth of the second carboplatin-resistant PDX model, HBCx39 minimally, the combination of carboplatin with low dose prexasertib halted growth almost entirely (Fig. 6G). These results confirmed the efficacy of the combination of prexasertib and carboplatin to treat carboplatin-resistant TNBCs.

Discussion
In this preclinical study, we developed new in vitro and patient-derived models of carboplatin resistance in TNBC. By employing unbiased proteomic and shRNA screens, we uncovered a targetable dependency of carboplatin-resistant TNBC on the mitotic checkpoint. The combination of carboplatin with pharmacological inhibitors of CHEK1 or WEE1 inhibits the growth of carboplatin-resistant TNBCs, including patient-derived resistant tumors.
We found carboplatin-resistant TNBC cells show longer prometaphase and anaphase as well as alterations in the Kinetochore Metaphase Signaling Pathway. The mitotic checkpoint delays mitosis by inhibiting the APC/C complex. In cancer cells, mitotic control is commonly dysregulated to allow excess proliferation, whereas abrogation of this checkpoint could lead to mitotic catastrophe, senescence, or apoptosis 42 . The effect of platinum agents on the mitotic checkpoint is not well understood. A recent report has linked platinum resistance in ovarian cancer to APC/C dysfunction coupled with functional dependency on Polo-like kinase 1 (PLK1) for mitotic exit. Concordantly, we found that the mitotic checkpoint kinases, BUBR1 and AURKA, are essential for the growth    13,16 are well described to induce DNA damage and replication stress 28 , we found only a slight increase in oxidative stress and DNA damage, and no evidence of replication stress in response to carboplatin treatment in our carboplatin-resistant TNBC model, suggesting that depletion of CHEK1 or WEE1 re-sensitizes 468-R cells to carboplatin by mechanisms other than abrogation of G2/M checkpoint control. Given that the G2/M and mitotic checkpoints are tightly intertwined 32   www.nature.com/scientificreports/ that loss of CHEK1 or WEE1 suppresses the growth of carboplatin-treated 468-R cells due to abrogation of the mitotic checkpoint.
Our results also indicate that pharmacological inhibition of WEE1 or CHEK1 re-sensitizes carboplatinresistant TNBC to carboplatin. In current clinical trials, the CHEK1 inhibitor prexasertib showed least toxicities and the most encouraging results as monotherapy 37 . Even so, its effect was modest (stable disease 33.3%; partial response 4.4%) in the treatment of advanced cancers, similar to the single agent efficiency of other CHEK inhibitors, seemingly implying an inherent resistance 37,38 . In our preclinical models, prexasertib alone also showed limited efficacy, whereas the combination of prexasertib with carboplatin achieved a superior response, even with a low dose of prexasertib.
The compounded toxicities of prexasertib and carboplatin could be a cause for concern. Both prexasertib 38 and carboplatin 5 induce toxic haematological effects, in particular, neutropenia and thrombocytopenia, as their major adverse events. Nonetheless, the low dose of prexasertib that we used in our study in combination with an average dose of carboplatin induced significant tumor inhibition in vivo, without signs of toxicity. In vitro, our doses are below the concentrations measured in the serum of patients treated with the clinically recommended phase II dose for prexasertib 37 or a standard infusion of carboplatin 44 , suggesting they could be used at lower clinical doses without losing efficacy. Pending a tolerable dose, the combination treatment could be used for both inherent and acquired resistant TNBC with or without selection for basal biomarkers, such as expression of high molecular weight cytokeratins and EGFR1 overexpression. This combination could increase the fraction of patients achieving the benefits of responders to carboplatin in the neoadjuvant 5,6 , adjuvant 45 , and metastatic 46 settings. In conclusion, we show that low dose prexasertib can sensitize carboplatin-resistant TNBC to carboplatin. We believe that this combination could potentially bring the remarkable benefits of responders to carboplatin 8,47,48 , to the majority of basal-like TNBC, providing more options to these otherwise difficult-totreat tumors. www.nature.com/scientificreports/ diluted in reaction buffer according to manufacturer's protocol. DNA staining was performed in 1% PBS-BSA supplemented with 1 µg/ml FxCycle Violet (Thermofisher).
Total ROS and mitochondrial superoxide were measured by CellROX Deep Red Reagent (Thermofisher, C10422r) and MitoSOX (Thermofisher, M36008) using BD FACSCanto II, BD Biosciences in the APC/A and PE-A channels, respectively, according to the manufacturer's protocols.
Detection of NAD + /NADH ratio. NAD + and NADH levels were determined using NAD/NADH-glo assay (Promega, G9071) according to supplier's instructions. A Victor X4 plate reader (Perkin Elmer) was used to measure bioluminescence.
shRNA screen. The kinome shRNAi library was supplied by Sigma Aldrich (Merck Millipore). 468-R cells were transduced at a multiplicity of infection of 1 to minimize the incorporation of multiple shRNAs per cell. After puromycin selection (2 µg/mL; Thermofisher A11138-03), cells were either collected (T0 control), or grown for over 10 doublings in the absence (T10 control) or presence of 2 µM carboplatin (T10 carboplatin). Cell pellets were shipped to Sigma Aldrich (Merck Millipore) for DNA sequencing and data deconvolution (Sigma Aldrich CSTSEQ).
The shRNA raw read counts were normalized to reads per million (RPM) for each sample, then transformed by log2 (RPM + 1). To identify genes, for which specific shRNAs were overall depleted, a gene-score was computed by R package (GSVA). The GSVA is a non-parametric and unsupervised algorithm that estimates pathway activity using gene expressions, transforming the gene expression matrix to a pathway enrichment score matrix. For shRNA screen analysis, targeting genes were treated as 'pathways' and shRNAs as 'genes' . The GSVA was applied to shRNAs that targeted the same gene to summarize normalized expressions into a single score. The significance of gene-scores between T0, T10 vehicle and T10 carboplatin were determined by ANOVA followed by Tukey's HSD Test with an adjusted P value < 0.1. Significant genes were used as an input for canonical pathway analysis by the String pathway analysis.

Liquid chromatography-tandem-mass spectrometry (LC-MS/MS). Cell-pellets of MDA-MB-468
and 468-R cells, treated for five days with either vehicle or 2 µM carboplatin were analyzed by Liquid chromatography-tandem-mass spectrometry (LC-MS/MS) as described in the Supplemental Methods.
The mass spectrometry proteomics data were deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD021726 (login username: reviewer_pxd021726@ebi.ac.uk and password: 9uD57dKL). www.nature.com/scientificreports/