Expression of the BAD pathway is a marker of triple-negative status and poor outcome

Triple-negative breast cancer (TNBC) has few therapeutic targets, making nonspecific chemotherapy the main treatment. Therapies enhancing cancer cell sensitivity to cytotoxic agents could significantly improve patient outcomes. A BCL2-associated agonist of cell death (BAD) pathway gene expression signature (BPGES) was derived using principal component analysis (PCA) and evaluated for associations with the TNBC phenotype and clinical outcomes. Immunohistochemistry was used to determine the relative expression levels of phospho-BAD isoforms in tumour samples. Cell survival assays evaluated the effects of BAD pathway inhibition on chemo-sensitivity. BPGES score was associated with TNBC status and overall survival (OS) in breast cancer samples of the Moffitt Total Cancer Care dataset and The Cancer Genome Atlas (TCGA). TNBC tumours were enriched for the expression of phospho-BAD isoforms. Further, the BPGES was associated with TNBC status in breast cancer cell lines of the Cancer Cell Line Encyclopedia (CCLE). Targeted inhibition of kinases known to phosphorylate BAD protein resulted in increased sensitivity to platinum agents in TNBC cell lines compared to non-TNBC cell lines. The BAD pathway is associated with triple-negative status and OS. TNBC tumours were enriched for the expression of phosphorylated BAD protein compared to non-TNBC tumours. These findings suggest that the BAD pathway it is an important determinant of TNBC clinical outcomes.


Materials and Methods
patients. This study was performed as part of a University of South Florida IRB-approved protocol and in accordance with the relevant guidelines and regulations, including Code of Federal Regulations Title 45 Part 46 Protection of Human Subjects. Following IRB approval, patient samples and molecular and clinical data stored in the Moffitt Cancer Center (MCC) Total Cancer Care (TCC) clinico-genomic tissue and data repository were accessed (MCC 14690/Liberty IRB #Pro00014441). All patients whose samples and data are included in the TCC protocol have provided prospective written informed consent for their use in research. Breast cancer samples from TCC were limited to those with complete clinical information and Affymetrix gene expression data. To ensure balance within the samples, only breast cancer patients whose carcinomas did not express HER2 receptors were included. Using these criteria, samples from 53 non-TNBC and 53 TNBC patients were available in the TCC database and analysed in this study. Chart abstractions were used to collect the following clinical elements: age, stage, grade, body mass index (BMI), gravida, tumour size, surgery, lymph-node status, and OS. The non-TNBC and TNBC groups were well balanced with no significant differences in age (non-TNBC, mean = 51.84 ± 1.65; TNBC, mean = 52.34 ± 1.55; t test P = 0.83), stage (chi-square test, P = 0.33), tumour size (non-TNBC, 2.95 ± 0.24; TNBC, 2.65 ± 0.25; t test P = 0.38), or BMI (non-TNBC, 28.32 ± 0.88; TNBC, 28.76 ± 1.1; t test P = 0.76); however, the TNBC group included more patients with grade 3 disease (non-TNBC, 38%; TNBC, 83%; chi-square test, P = 2.55 × 10 −5 ).

RnA extraction and microarray expression analyses.
In accordance with TCC protocol, all tissues were snap frozen within 15 minutes of collection, macro-dissected to ensure >80% tumour content, and quantified for the percentage of malignancy, cellularity, stroma, normalcy, and necrosis. Approximately 30 mg of tissue for each sample were pulverized in BioPulverizer H tubes (Bio101) using a Mini-Beadbeater (Biospec Products). Total RNA was collected using the Qiagen RNeasy Mini kit in accordance with manufacturer's instructions. An Agilent Bioanalyzer was used to assess RNA quality via the 28S:18S ribosomal RNAs. Ten micrograms of total RNA were used to develop the targets for Affymetrix microarray analysis, and probes were prepared according to the manufacturer's instructions. Briefly, biotin-labelled cRNA was produced by in vitro transcription, fragmented, and hybridized to customized Human Affymetrix HuRSTA gene chips (HuRSTA-2a520709). Expression values were calculated using the robust multi-array average algorithm implemented in Bioconductor (http://www.bioconductor.org) extensions to the R statistical programming environment. The gene expression data discussed in this publication have been deposited in National Center for Biotechnology Information's Gene Expression Omnibus (GEO) and are accessible through GEO series accession number GSE62931 33 .
Deriving a BAD pathway principal components analysis score. The BAD pathway gene expression signature (BPGES) was developed from the GeneGo Metacore-defined BAD Apoptosis and Survival Pathway using the genes that showed importance in the PCA model. These included BAX, BCL2, EGFR, PDK1, PIK3CA, PIK3CB, PPP1CA, PPP2CA, PPP3CA, PPM1A, YWHAB, YWHAE, YWHAG, YWHAH, YWHAQ, and YWHAZ. All genes in the BPGES have previously been shown to directly or indirectly influence the phosphorylation status and/or apoptotic activity of BAD protein; these are BAX 34,35 , BCL-2, EGFR 36,37 , PDK1 (PDPK1) 38 , PI3 kinase (PIK3CA, PIK3CB) 36 , PP1 (PPP1CA) 39 , PP2A (PPP2CA) 40 43 . The PCA methodology was used to derive a BPGES pathway score that would represent overall gene expression levels for these BAD-pathway genes. Genes and probe sets used in the PCA model for the different datasets are listed in the Supplemental Table S1. Only 1 probe set was used per gene, which was selected on the basis of the highest expression value in the TCC dataset samples.
PCA is a well-established technique for unsupervised data analyses and dimensional reduction, as described by Joliffe and Ma 44,45 . We and others have previously shown that the first component of a PCA model, defined as PC1, can successfully compare the expression of gene signatures and describe pathway activation in tumour samples. It can also be used for survival analyses 29,38 . In brief, the first step when using PCA to compare signature expression in clinico-genomic datasets is to create a subdataset by selecting only the probesets in the given gene expression signature. To calculate the BPGES, probesets representing 16 genes within the BAD pathway were reduced to a set of uncorrelated principal components. After removing the column mean (mean centring) and scaling each column-to-unit variance, the PC1 score can be calculated. That is, the pathway score is ∑w i x i , a weighted-average expression among the BAD-pathway genes in which x i represents gene i expression level, w i is the corresponding weight (loading coefficient) with ∑w 2 i = 1, and the w i values maximize the variance of ∑w i x i . PC1 describes the direction in the N-dimensional space (in which N is the number of genes) that maximizes the explained variance. By not using all the variation in the data, PC1 is stable to random noise, can handle missing values, and provides a simple score for each sample. Thus, the PC1 score is a numeric value that summarizes the expression pattern of the entire signature and represents an overall similarity measurement between samples on the basis of their expression profile for the selected genes. The corresponding loadings are related to the variables (e.g., Affymetrix probe sets) and reflect relative importance in the PCA model.

Statistical considerations.
Principal components analysis (PCA) was used to derive the BPGES.
Differential expression of BPGES between groups was determined using Student's t test analyses of derived PCA scores. Log-rank (Mantel-Cox) significance tests of Kaplan-Meier curves were used to compare the associations of BPGES PCA scores and phospho-BAD protein expression scores with OS. OS was defined as the period from the date of diagnosis to the date of death from any cause. Median OS was used to define the high-versus-low BPGES. Pearson's correlation test was used to evaluate the correlations between phospho-BAD isoform expression scores and other phospho-BAD isoforms as well as the BPGES. The Mann-Whitney test and unpaired t tests were used to compare the expression of individual genes within datasets and the BPGES between breast cancer subtypes.  29,46,47 . Antibodies recognizing these phosphorylated serine residues are cross reactive between human and mouse BAD protein. For simplicity, only the mouse designations are used in this manuscript. Before samples were stained, they were deparaffinised and rehydrated using xylene, followed by serial dilutions of ethanol. Slides were stained using a Ventana Discovery XT automated system (Ventana Medical Systems, Tucson, AZ, USA) with proprietary reagents, in accordance with the manufacturer's protocol. Briefly, slides were deparaffinised on the automated system with EZ Prep solution (Ventana). A heat-induced antigen retrieval method was used in Cell Conditioning 1 (Ventana). Primary antibody optimization and tissue staining was performed under the direction of the study pathologist using manufacturer-suggested methods and control tissues. Antibodies were diluted in PSS antibody diluent (Ventana), incubated for 60 minutes at room temperature, rinsed with diluent, and incubated with the appropriate horse radish peroxidase-(HRP-) linked secondary antibodies for 16 minutes. Antibody binding was detected using the Ventana ChromoMap kit, and slides were counterstained with haematoxylin. Slides were then dehydrated and coverslipped in accordance with normal laboratory protocol. To compare phospho-BAD isoform levels, expression scores were determined by the study pathologist using the product of intensity and cellularity, in which an intensity of 1 was considered weak, 2 moderate, and 3 strong, and a cellularity of 1 was ≤33%, 2 was 34% to 65%, and 3 was ≥66% 48,49 . In addition, Definiens software was used to generate a histo-expression score (H-score). Stained TMA slides were scanned using the Aperio (Vista, CA, USA) AT2 with a 200x/0.8NA objective lens at a rate of 3 minutes per slide via Basler trilinear-array detection. Each core was then segmented for individual analyses using Spectrum's TMA block software. Image analyses were performed to segment positive staining of various intensities using an Aperio Positive Pixel Count v9.0 algorithm with the following thresholds: hue value = 0.1, hue width = 0.5, colour saturation threshold = 0.04, intensity weak positive (high) = 220, intensity weak positive (low) = intensity positive (high) = 175, intensity positive (low) = intensity strong positive (high) = 100, and intensity strong positive (low) = 0. The algorithm was applied to the entire digital core image to determine the percentage of positive biomarker staining by applicable area. Tissue cores on the TMA were manually inspected by the study pathologist and removed from the analysis when less than 50% of the tissue was remaining, tissue folds were evident, or obvious artefacts were present. (2019) 9:17496 | https://doi.org/10.1038/s41598-019-53695-0 www.nature.com/scientificreports www.nature.com/scientificreports/ cell culture and reagents. All cell lines were obtained from the American Type Culture Collection (Manassas, VA, USA) and cultured in RPMI (MDA-MB-157, MDA-MB-436, BT549, and Hs578t) supplemented with 10% foetal bovine serum, 1% sodium pyruvate, and 1% nonessential amino acids or DMEM (MCF-7, T47D, MDA-MB-231, and MDA-MB-468) supplemented with 10% foetal bovine serum. All media were supplemented with Mycozap Plus-CL (Lonza, Rockland, ME, USA) for the prevention of mycoplasma and bacterial contamination. All tissue culture reagents were obtained from Fisher Scientific (Pittsburgh, PA, USA). Mycoplasma testing was performed every 6 months in accordance with the manufacturer's protocol (Lonza). The AKT inhibitor perifosine was acquired from SelleckChem (S1037; Houston, TX, USA). The PKA inhibitor H89-dihydrochloride Hydrate (H89) was purchased from Sigma-Aldrich (SKU 19-141; St Louis, MO, USA). Both inhibitors were solubilized in dimethyl sulfoxide at a stock concentration of 20 mM and stored at −20 °C until use.
Western blot analysis. Cells were harvested in media using a Cell Lifter (Thermo Fisher Scientific, Pittsburgh, PA, USA) and washed with cold phosphate-buffered saline containing 1x phosphatase inhibitor cocktail (Sigma-Aldrich, St. Louis, MO, USA). Lysates were prepared with sodium dodecyl sulphate (SDS) lysis buffer (2% SDS, 10% glycerol, 0.06 M Tris, pH 6.8) and evaluated for protein concentration using the bicinchoninic acid method (Pierce, Rockford, IL, USA). Proteins (50-100 μg) were separated on the same day as collection on 12% to 15% SDS-polyacrylamide gel electrophoresis gels and transferred to polyvinylidene fluoride membranes. Membranes were blocked with 5% nonfat milk in Tris-buffered saline containing 0.05% Tween 20 (TBST) and incubated with primary antibody in SuperBlock blocking buffer (Thermo Fisher Scientific) with 0.05% Tween 20 overnight at 4 °C. Membranes were washed 3 times for 5 minutes with TBST and incubated with the appropriate secondary antibody in 5% nonfat milk in TBST for 60 minutes at room temperature. cell viability assays. The CellTiter96 MTS assay kit (Promega, Madison, WI, USA) was used to assess cancer cell survival in the presence of the AKT inhibitor perifosine and the PKA inhibitor H89 dihydrochloride hydrate [50][51][52][53] . For the assays, 3 to 5 × 10 4 cells in 100 µL were plated to each well of a 96-well plate and allowed to adhere overnight at 37 °C and 5% CO 2 . The following day, cells were incubated with increasing concentrations of the chemotherapeutic agent for 72 hours. Three replicate wells were used for each drug concentration, and an additional 3 control wells received a diluent control without drug. One well (blank) was devoid of cell and was used to assess the background optical density readings of media and reagents. After drug incubation, the optical density of each well was read at 490 nm using a SpectraMax 190 microplate reader (Molecular Devices Inc., Sunnyvale, CA, USA). Percent cell survival was expressed as control-treated/control-blank × 100. All experiments were performed at least 3 times to ensure reproducibility and accuracy of the results. To compare sensitivities, IC 50 values for cisplatin and cisplatin and perifosine or H89 hydrate were computed using sigmoidal dose-response algorithm implemented in GraphPad (GraphPad Software, Inc, San Diego, CA, USA). BAD protein depletion. Pooled RNA duplexes for BAD (SMARTpool:On Target Plus BAD siRNA ID#572, Dharmacon/GE Life Sciences, Lafayette, CO, USA target sequence: UCGGAAGUUUUGGGUUUUC) were used to transfect breast cancer cells (4 × 10 6 ) using the Nucleofector transfection kit containing Transfection Buffer V (#VCA1003, Lonza), in accordance with manufacturer's protocols. The nontargeting Silencer-negative control #2 small interfering RNA (siRNA; Ambion/ThermoFisher Scientific) was used as a control. Subsequent depletion of BAD protein was determined by Western blot analyses.

Results
BAD pathway expression associates with triple-negative status and overall survival. To evaluate the overall expression and activation of the BAD pathway in breast cancer, PCA was used to generate a BPGES score. The genes comprising the BPGES included BAX, BCL-2, EGFR, PDK1, PIK3CA, PIK3CB, PPP1CA, PPP2CA, PPP3CA, PPM1A, YWHAB, YWHAE, YWHAG, YWHAH, YWHAQ, and YWHAZ. PCA modelling of the BPGES in a gene expression dataset, comprised of 106 breast cancer patient samples (53 non-TNBC and 53 TNBC), indicated that a high BPGES score was associated with breast cancers displaying the triple-negative phenotype, with a mean BPGES score of −1.7 ± 0.24 for non-TNBC and 1.7 ± 0.19 for TNBC (unpaired t test, P = 4.8 × 10 −19 ; Fig. 1a). As the TNBC group contained a significantly higher percentage of grade 3 tumours, we evaluated whether the BPGES was associated with disease grade. The mean BPGES score was 1.46 ± 0.58 for lowgrade disease (grades 1 and 2) versus 1.92 ± 0.24 for high-grade disease (grades 3 and 4; t test, P = 0.4973); therefore, it was not associated with grade of disease. The association between the BPGES and triple-negative status was confirmed by using hormone receptor positive www.nature.com/scientificreports www.nature.com/scientificreports/ BPGES in the TCC dataset (n = 106). Similar performance of the BPGES genes was observed in the TCGA dataset ( Supplementary Fig. S1).
To determine whether BAD pathway expression influenced patient survival, the BPGES score was evaluated for associations with patient survival in both the TCC and TCGA datasets. Using the median BPGES score as a cut-off to dichotomize samples for analysis, log-rank testing of Kaplan-Meier curves indicated that a high www.nature.com/scientificreports www.nature.com/scientificreports/ BPGES score was associated decreased OS in the TCC dataset (hazard ratio [HR] = 2.76, P = 0.0072; n = 105, survival data available for 52 of 53 non-TNBC) (Fig. 1d) and the TCGA dataset (n = 757, hazard ratio = 2.045, P = 0.0009) (Fig. 1e). The BPGES was not associated with OS when only TNBC samples were considered in either the TCC dataset (HR = 1.07, P = 0.873, n = 52) (Fig. 1f) or the TCGA dataset (HR = 1.827, P = 0.35, n = 82, survival data available for 82 of 84 TNBC cases) (Fig. 1g). As TNBCs are more aggressive than nonTNBCs, we evaluated the role of the cell cycle in the association between the BPGES and decreased OS among patients with breast cancer. Although we observed a correlation between the BPGES and the Hallmark gene set cell cycle signature (r 2 = 0.544, P < 0.0001) ( Supplementary Fig. 1b), the cell cycle signature was not associated with OS in the TCC dataset (HR = 1.721, P = 0.1284) (Supplementary Fig. 1c). These data indicate that the cell cycle is an aspect of the BPGES but do not explain the association between the BPGES and OS. tnBc cases express higher mean levels of phospho-BAD isoforms than non-tnBc cases. To determine the association between the BPGES and the phosphorylation status of BAD protein, a TMA was generated using triplicate cores of 18 non-TNBC and 36 TNBC samples with both gene expression and available tissue. The TMA was evaluated by IHC staining using antibodies specific to phospho-BAD isoforms. Representative core samples for each of the stains are shown in Fig. 2a. Mean histological expression scores (cellularity × intensity) of all cores/cases based on pathologist assessment of IHC staining (scale of 1-3) suggested that TNBCs expressed higher mean levels of phospho-BAD-Ser 136 (P = 0.0078) than non-TNBC samples. Although phospho-BAD-Ser 112 and phospho-BAD-Ser 155 also appeared to be higher in TNBC, statistical significance was not reached (phospho-BAD-Ser 112 ; P = 0.26, phospho-BAD-Ser 155 ; P = 0.3663, Fig. 2b, top panel). Definiens software-derived Histo-scores of phospho-BAD isoform levels showed similar results. Unpaired t tests of Definiens-derived Histo-scores suggested that TNBCs expressed higher mean levels of phospho-BAD-Ser 136 (P = 0.0012) but not phospho-BAD-Ser 112 (P = 0.593) or phospho-BAD-Ser 155 (P = 0.249) (Fig. 2b, lower  panel). The TMA Map, pathologist expression scores, Definiens-derived Histo-scores, and IHC stains of the phospho-BAD isoforms can be viewed in Supplemental Tables S2 to S4 and Supplemental Fig. S2 to S5.

targeted inhibition of the BAD pathway in tnBc cells potentiates chemotherapy.
To determine whether expression of the BAD pathway was associated with triple-negative status in cell lines, the BPGES was evaluated using expression data downloaded from the CCLE database (www.broadinstitute.org). Similar to our results shown in the tumour samples, gene expression data from 42 breast cancer cell lines (28 TNBC, 14 non-TNBC) indicated a differential expression of the BAD pathway in TNBC cell lines, with a mean BPGES score of −1.81 ± 0.33 in non-TNBC cells versus a mean BPGES score of .90 ± 0.25 in TNBC cells (unpaired t test, P = 4.22 × 10 −7 ; Fig. 3A,B). Figure 3c shows the performance of the individual genes of the BPGES in the CCLE dataset. As shown, the performance of BPGES genes in breast cancer cell lines was similar to their performance in tumour samples. Therefore, cell line models were used to determine whether disruption of the BAD pathway via BAD protein kinase inhibitors could enhance the cytotoxic effects of nonspecific chemotherapy. To do this, a panel of randomly selected ER + /PR + (MCF-7, T47D) and TNBC (Hs578t, MDA-157, MDA-231, BT549, MDA-436, MDA-468) cell lines were treated with inhibitors of BAD protein kinases. The 5-year survival probability HRs suggested that TNBC patients have a 2-fold increased risk of death if their tumours display increased expression of phospho-BAD-Ser 136 (AKT phosphorylation site) or phospho-BAD-Ser 155 (PKA phosphorylation site). As such, we evaluated whether BAD pathway disruption was capable of sensitizing TNBC cells to cisplatin by using the AKT inhibitor perifosine and the PKA inhibitor H89. Figure 3d shows the relative phosphorylated (active)/ native protein expression levels of AKT and PKA as determined by densitometry analyses of replicate Western blots.
Data obtained from MTS cell viability assays showed that most TNBC cells treated with a fixed dose of perifosine or H89 were more sensitive to the cytotoxic effects of cisplatin than cells treated with cisplatin alone. In contrast, the sensitivity of non-TNBC cells to cisplatin did not significantly change or was lessened in the presence of perifosine and H89 (Fig. 4a,b). Comparing cisplatin IC 50 values in the presence and absence of perifosine or H89, which were calculated using the sigmoidal dose-response algorithm (as implemented in GraphPad Prism 7), indicated that the potentiation of cisplatin by a BAD protein kinase inhibitor was significant in the majority of TNBC cells but not in non-TNBC cells ( Table 1). The ability of an AKT or H89 inhibitor to enhance cisplatin cytotoxicity depended on the expression of the active phosphorylated isoform of the target kinase in TNBC cells. Treatment of the TNBC cell MDA-468 with perifosine but not H89 potentiated the cytotoxicity of cisplatin. Densitometry analyses of immunoblots suggested that nearly 80% of AKT within this cell line was phosphorylated. In contrast, MDA-468 expressed almost no phosphorylated PKA. However, there were exceptions. In some TNBC cell lines, www.nature.com/scientificreports www.nature.com/scientificreports/ such as Hs578t, which showed high levels of phosphorylated AKT and low levels of phosphorylated PKA, cisplatin cytotoxicity appeared to be enhanced by H89 but not by perifosine. The reason for this was not explored. However, kinases other than AKT, such as PAK-1 and PKC-iota 54 , have been shown to phosphorylate BAD protein at ser-136 and may have more dominant roles in this cell line.
In TNBC cell lines, where potentiation of platinum agents was observed, we sought to determine whether the enhancement of cisplatin cytotoxicity was mediated by BAD protein. The TNBC cell lines BT549, MDA468, and MDA436 were depleted of BAD by using siRNA and evaluated for sensitivity to cisplatin, H89, and/or perifosine. www.nature.com/scientificreports www.nature.com/scientificreports/ As shown, depletion of BAD protein in BT549 (Fig. 5a) and MDA468 (Fig. 5b) did not change cell line sensitivity to cisplatin but resulted in decreased sensitivity to perifosine. Similar results were observed with H89 in BT549 (Fig. 5c) and MDA436 (Fig. 5d) cells. This suggests that BAD protein was required for cisplatin sensitization effects of H89 and perifosine.

Discussion
Although TNBCs account for only about 17% of all breast cancers 1,8,55 , they are heterogeneous, biologically aggressive, and carry a poorer prognosis than the growth factor receptor-driven or hormone receptor-driven breast cancers 9,10,55 . The earlier age of onset and lethality of TNBC underscores the need for the development of more efficacious therapeutic strategies. Few targeted agents have shown effectiveness against this aggressive form of breast cancer, and the majority of TNBC patients are treated with nonspecific chemotherapeutic agents, including platinums, anthracyclines, and taxanes 7 . Therapies that enhance cancer cell sensitivity to cytotoxic agents could significantly improve patient outcomes.
In this study, we show that the BAD pathway is a targetable biomarker to increase the response of TNBC to nonspecific chemotherapeutic agents, such as cisplatin. PCA was used to develop an expression signature based a subset of genes that comprise the GeneGo Metacore-defined BAD pathway. This BPGES) was derived using 16 genes that showed the most importance within the PCA model, including BAX, BCL-2, EGFR, PDPK1 (PDK1), PIK3CA, PIK3CB (PI3 kinase), PPP1CA (PP1), PPP2CA (PP2A), PPP3CA (Calcineurin), (PPM1A) PP2C, YWHAB, YWHAE, YWHAG, YWHAH, YWHAQ, and YWHAZ (14.3.3). The BPGES was capable of delineating TNBC tumours from hormone responsive (non-TNBC) tumours in independent clinic-genomic breast cancer datasets, including TCC (n = 106) and TCGA (n = 408) as well as the breast cancer cell lines of the CCLE (n = 42). Furthermore, the BPGES was associated with decreased survival probability in both the TCC and TCGA cohorts. TNBC is known to be more aggressive than hormone responsive tumours. Therefore, we evaluated whether the www.nature.com/scientificreports www.nature.com/scientificreports/ performance of the BPGES was an artifact of increased TNBC proliferation. As described above, although the BPGES showed a correlation with the Hallmark cell cycle signature (r 2 = 0.54), log-rank tests of Kaplan-Meier curves indicated that cell cycle alone could not explain the association between the BPGES and decreased OS among patients with breast cancer (log-rank P = 0.13). All genes within the BPGES have previously been shown to directly or indirectly influence the phosphorylation status and/or apoptotic function of the BAD pathway, including BAX 34,35 , BCL-2, EGFR 36,37 , PDK1 (PDPK1) 38 , PI3 kinase (PIK3CA, PIK3CB) 36 , PP1 (PPP1CA) 39 , PP2A (PPP2CA) 40 , calcineurin (PPP3CA) 41 , (PP2C) PPM1A 25,27,42 , and 14.3.3 (YWHAB, YWHAE, YWHAG, YWHAH, YWHAQ, YWHAZ) 43 . The performance of the individual genes were similar in the TCC and TCGA cohorts and in the CCLE cell lines despite the fact that different gene expression platforms were used for each dataset. Genes encoding 3 of 4 documented BAD protein phosphatases, PP2C 25,27,42 , PP2A 40 , and calcineurin 41 , showed decreased expression, whereas genes encoding the kinases EGFR 36,37 , PDK1 38 , and PI3 kinase 36 showed increased expression in TNBC compared to non-TNBC cases and cell lines. This suggests an expression profile in TNBC that favoured the phosphorylation and inactivation of BAD protein. In parallel, TNBC cases displayed increased expression of 5 out of 6 genes encoding 14.3.3. compared to non-TNBC cases. 14.3.3 has been reported to inhibit BAD-induced cell death by binding and sequestering BAD when phosphorylated on Ser 136 , a known AKT phosphorylation site 43 . Overall, the performance of the individual genes of the BPGES suggests a prosurvival phenotype in TNBC cases compared to non-TNBC cases. Indeed, immunohistochemistry analyses of a study-generated TMA showed that TNBC cases expressed higher levels of phospho-BAD isoforms, most notably the AKT phosphorylation site BAD-Ser 136 compared to non-TNBC cases. We next sought to determine whether inhibition of BAD pathway signalling would increase the sensitivity of TNBC cells to a nonspecific chemotherapy agent. HRs of 5-year survival curves suggested that TNBC patients were at 2-fold increased risk of death if their tumours expressed increased levels of phospho-BAD-Ser 136 or phospho-BAD-Ser 155 . Therefore, we used the AKT inhibitor perifosine and the PKA inhibitor H89. Treating TNBC cells but not non-TNBC cells with perifosine or H89 increased the cytotoxic effects of cisplatin. Although there were exceptions, the ability of the BAD protein kinase inhibitor to enhance cisplatin cytotoxicity appeared to be dependent on the expression of the target kinase. The inhibition of AKT and PKA did not sensitize all TNBC cell lines to cisplatin. Although the reasons for this were not explored, it must be noted that there are several antiapoptotic BAD protein kinases, including PAK1, PKC-iota, ERK, IKK, C-raf, and CK2, that were not evaluated in this study 54 .
These results are consistent with the work of others. It is well known that AKT phosphorylation of BAD protein is a negative regulator of the proapoptotic function of BAD protein 56 . Consistent with our data, other groups have shown that activated AKT signalling is increased in TNBC 57-60 and that inhibition of AKT signalling can