Molecular characterization of breast cancer cell lines through multiple omic approaches

A panel containing 40 breast cancer cell lines and 4 non-tumorigenic breast cell lines was obtained from the American Type Culture Collection (ATCC, Manassas, Virginia, USA 30-4500 K; [11]). Cells were cultured according to ATCC recommendations for fewer than six months from the time of resuscitation. All cell lines were authenticated by the supplier.

Protein expression in the breast cancer cell lines was quantified by immunoblot analysis as previously described [12]. Briefly, SDS-PAGE was performed loading an equal amount of total protein from cell lysates in each lane as determined by Lowry assay (Sigma Aldrich, Oakville, ON, Canada TP0300). Samples were transferred to nitrocellulose membranes and probed with primary antibodies. Antibodies were obtained from Santa Cruz Biotechnology (Dallas, TX, USA) for ER alpha (sc-8002), PR (sc-538), HER2 (sc-284), phosphatase and tensin homolog (PTEN) (sc-7974) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (sc-25778), and from Cell Signaling Technology (Danvers, MA, USA) for p110α (4249) and p110β (3011). Additional primary antibodies included: EGFR (BD Biosciences, Mississauga, ON, Canada; 610017), p53 (ProteinTech, Rosemont, IL, USA; 10442-1-AP) and p85α (Cedarlane, Burlington, ON, Canada; 05-212). Blots were then probed with infrared 680 nm or 800 nm dye-tagged secondary antibodies (LI-COR Biosciences, Lincoln, NE, USA; 200 ng/ml) and were imaged with the Odyssey Infrared Imaging System (LI-COR Biosciences, Lincoln, NE, USA). The four gels required for each experiment were resolved, transferred, probed, washed, scanned and processed together to minimize technical artifacts. Blots shown are representative of at least three independent experiments, each using a fresh cell lysate.

The ATCC website (http://​www.​ATCC.​org; accessed May 14, 2015) provided basic information regarding the cell lines, including the sources used to generate the cell line, the type of breast cancer, and in some instances, the molecular classification or subtype, and partial data on the expression or absence of some genes. The catalogue of somatic mutations in cancer (COSMIC; http://​cancer.​sanger.​ac.​uk/​cell_​lines) database of cell line mutations was accessed January 30, 2016 (version v75). For Additional file 1: Table S3, the list of mutated genes was filtered using the online tools to provide only the cancer genes considered to be pathogenic. The cancer cell line encyclopedia (CCLE; https://​www.​broadinstitute.​org/​software/​cprg/​?​q=​node/​11; 2012 September version [13]) dataset containing the robust multi-array average (RMA) and quantile-normalized mRNA expression was used. The values for normalized mRNA expression were then divided into four groups for each gene product as follows: ER and PR (>9 = +++, 7-8.9 = ++, 5-6.9 = +, <5 = -); HER2/ErbB2 (>10 = +++, 9-9.9 = ++, 8-8.9 = +, <8 = -); EGFR, TP53, BRCA1, BRCA2, p110α, p110β, p85α and PTEN (>9 = +++, 8-8.9 = ++, 7-7.9 = +, <7 = -).

Cells were cultured under serum-starvation conditions (0.5% fetal bovine serum (FBS) containing medium) for 24 hours to analyze the signaling proteins/pathways activated within these cell lines. Cells were lysed and used to probe a PathScan RTK Signaling Antibody Array (Cell Signaling Technology, distributed by New England Biolabs, Whitby, ON, Canada; 7949) using reagents provided within the kit and according to the supplier’s instructions. These slide-based antibody arrays enable the simultaneous quantification of the extent of phosphorylation of 28 receptor tyrosine kinases and 11 key signaling proteins. The nitrocellulose-coated glass slides have highly specific capture antibodies that selectively bind target proteins within the cell lysate. A biotinylated detection antibody mixture and streptavidin-linked DyLight 680 molecule allow for the detection of bound protein using the Odyssey Infrared Imaging System (LI-COR Biosciences, Lincoln, NE, USA). Quantification was carried out using Odyssey V3.0 software. The raw intensity data for each target (mean of duplicate target measurements on each slide) had the background subtracted (mean of the 2 negative control spots), and was reported as a percentage of positive control spots (mean of 10 positive control spots). Data for the targets are reported for each cell line as the mean ± standard deviation of at least two (but for most three) independent experiments containing duplicate measurements, each using a fresh lysate. Hierarchical clustering was performed on both the phosphoproteins and cell lines using Ward's method (minimize cluster variance) with the Euclidean distance set as the distance metric. The heatmap and dendograms were generated using the Matplotlib and SciPy libraries for Python.

A customized peptide array was developed to consider cancer-associated pathways, proteins and phosphorylation events. Phosphorylation events represented on the array were selected from databases of experimentally defined phosphorylation events and from those predicted by the software program DAPPLE2 [14]. Additional peptide substrates were included to represent proteins, and their associated phosphorylation events, that were shown in the literature to be differentially regulated in a number of cancers including renal cell carcinoma, pancreatic cancer, and prostate cancer [15‐17]. Major signaling pathways involved in proliferation, metabolism, and apoptosis were also included on the array in order to give a general overview of the cell signaling patterns. In total, 1290 15-mer peptides were rationally selected for the array.

Design, construction, and application of the peptide arrays were based upon a previously reported protocol with modifications [18, 19]. Briefly, cells were serum-starved by growing in 0.5% FBS for 24 hours and 10 × 10⁶ cells were collected, pelleted, and lysed by addition of 100 μl of lysis buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100, 2.5 mM sodium pyrophosphate, 1 mM sodium orthovanadate, 1 mM sodium fluoride, 1 μg/ml leupeptin, 1 μg/ml aprotinin, and 1 mM phenylmethylsulfonyl fluoride) (all from Sigma-Aldrich unless indicated). Cells were incubated on ice for 10 minutes and spun in a microcentrifuge for 10 minutes at 4 °C. A 70-μl aliquot of this supernatant was mixed with 10 μl of activation mix (50% glycerol, 500 μM ATP (New England BioLabs, Pickering, ON, Canada), 60 mM MgCl₂, 0.05% (v/v) Brij 35, 0.25 mg/ml bovine serum albumin) and incubated on the array for 2 hours at 37 °C. Arrays were then washed with phosphate-buffered saline (PBS) + 1% Triton X-100. Slides were submerged in phosphospecific fluorescent ProQ Diamond Phosphoprotein Stain (Invitrogen, ThermoFisher, Burlington, ON, Canada) with agitation for 1 hour. Arrays were then washed three times in destaining solution containing 20% acetonitrile (EMD Biosciences, VWR Distributor, Mississauga, ON, Canada) and 50 mM sodium acetate at pH 4.0 for 10 minutes. A final wash was done with distilled deionized water. Arrays were air dried for 20 minutes and then centrifuged at 300 × g for 2 minutes to remove any remaining moisture. Arrays were analyzed using a GenePix Professional 4200A microarray scanner (MDS Analytical Technologies, Toronto, ON, Canada) at 532 to 560 nm with a 580-nm filter to detect dye fluorescence. Images were collected using GenePix software (version 6.0) and the spot intensity signal was collected as the mean of pixel intensity using local feature background intensity calculation with the default scanner saturation level.

All data processing and analysis was done using the Platform for Intelligent, Integrated Kinome Analysis (PIIKA) software [20], which is freely available for non-commercial use at http://​saphire.​usask.​ca/​saphire/​piika. For each peptide within a given array, the chi-square test was performed to determine whether the degree of variability among the technical replicates for that peptide was greater than would be expected by chance. Any peptide that had a P value according to the chi-square test of less than 0.01 was considered to be inconsistently phosphorylated among the technical replicates and was excluded from further analysis.

The preprocessed data were subjected to hierarchical clustering and principal component analysis (PCA) to cluster peptide response profiles across cell lines. For each of the 1290 peptides in a single sample and cell line, the average was taken over the nine replicates that are transformed through variance stabilization and normalization (VSN). For hierarchical clustering, each sample/cell line vector was considered a singleton (i.e., a cluster with a single element) at the initial stage of the clustering. The two most similar clusters were merged, and the distances between the newly merged clusters and the remaining clusters were updated, iteratively. The method, as described by Eisen et al. [21], used the following calculation: average linkage + (1 - Pearson correlation). The method takes the average over the merged (i.e., the most correlated) kinome profiles and updates the distances between the merged clusters and other clusters by recalculating the correlations between them.

InnateDB is a publicly available resource which, based on levels of either differential expression or phosphorylation, predicts biological pathways based on experiment fold change data sets [22]. Pathways were assigned a probability value (P) based on the number of proteins present for a particular pathway and the degree to which they were differentially expressed or modified relative to a control condition.

For hierarchical clustering of kinome profiles, the distance metric used was (1 - Pearson correlation), while McQuitty linkage was used as the linkage method. Colors indicate the average (over nine intra-array replicates) normalized phosphorylation intensity of each target, with red indicating increased phosphorylation and green indicating decreased phosphorylation. The intensity of the color corresponds to the degree of increase or decrease [23].