Profiling the tyrosine phosphoproteome of different mouse mammary tumour models reveals distinct, model-specific signalling networks and conserved oncogenic pathways

Neu-pMIL was used to generate the HER2 breast cancer model. This expresses a truncated form of Neu (approximately 647 to 1,260 amino acids, the rodent orthologue of Her2) with elevated activity due to truncation of the extracellular domain [27]. Truncated Neu was subcloned from Neu-pLJ by digestion using Sal1 followed by subcloning into pENTR2B (Invitrogen, Mulgrave, Victoria, Australia). The gene was subsequently cloned into pMIL, a derivative of pMig [28] that expresses a luciferase marker, to generate Neu-pMIL, in which expression of Neu is driven by the murine stem cell virus (MSCV) promoter. The Gateway LR recombination reaction was used as per the manufacturer's instructions (Invitrogen).

Resected mouse mammary tumours were lysed in 8 M urea, 20 mM 2-[4-(2-hydroxyethyl)piperazin-1-yl]ethanesulfonic acid (HEPES), 2.5 mM sodium pyrophosphate, 1 mM β-glycerol phosphate, 1 mM sodium orthovanadate, 1 mM ethylenediaminetetraacetic acid and 1 mM Tris(2-carboxyethyl)phosphine, pH 8.0. Lysates were cleared by sonication and centrifugation prior to protease digestion, phosphotyrosine (pY) immunoprecipitation (IP) and nano-liquid chromatography tandem mass spectrometry (nano-LC-MS/MS). Lysates were quantitated, and 20 mg of each sample were diluted to a final concentration of 1 M urea with 20 mM HEPES and digested overnight with trypsin at room temperature. Heavy proline (+6 Da)– and alanine (+4 Da)–labelled synthetic standard pY peptides of MK14, elongation factor Tu and EGFR were spiked into each sample at 5 pmol, 500 fmol and 10 pmol, respectively, to enable normalisation of label-free quantitative values. Peptides were then desalted and concentrated using C₁₈ Sep-Pak columns (Waters, Milford, MA, USA), and lyophilised. pY peptides were subject to IP using the PhosphoScan procedure (Cell Signaling Technology, Beverly, MA, USA) with pY100 antibodies as previously described [16]. pY-enriched peptides were dried using a vacuum centrifuge and stored at −80°C prior to nano-LC-MS/MS analysis. All samples were analysed by nano-LC-MS/MS using two technical replicates.

pY peptides were resuspended in 15 μl of MS buffer containing 1% formic acid, 2% acetonitrile (ACN) and 0.05% heptafluorobutyric acid and subjected to nano-LC-MS/MS. Peptides were separated by nano-LC using an UltiMate 3000 high-performance liquid chromatography and autosampler system (Dionex, Sunnyvale, CA, USA). Samples were concentrated and desalted onto a C₁₈ micro-precolumn (500 μm × 2 mm; Michrom Bioresources, Auburn, CA, USA) with H₂O:ACN (98:2, 0.05% trifluoroacetic acid) at 15 μl/min for 4 minutes. The micro-precolumn was then switched online with a nano-C₁₈ column (75 μm × about 10 cm, 5 μm, 200 Å Magic; Michrom) and the reverse phase nano-eluent was subjected to positive nanoflow electrospray analysis in information-dependent acquisition (IDA) mode (250 nl/min for 30 minutes). Mass spectra were acquired on an Orbitrap Velos mass spectrometer (Thermo Electron, Waltham, MA, USA). In IDA mode, survey scans in the mass-to-charge ratio (m/z) range 350 to 1,750 were acquired with lock mass enabled. Up to the 15 most abundant ions (>5,000 counts) with charge states greater than +2 were sequentially isolated and further subjected to MS/MS fragmentation within the linear ion trap using collisionally induced dissociation. MS/MS spectra were accumulated with an activation time of 30 milliseconds at a target value of 30,000 ions and a resolution of 60,000. m/z ratios selected for MS/MS were dynamically excluded for 30 seconds.

The nano-LC-MS/MS .raw files were processed with MaxQuant software (version 1.1.1.25), which uses the Andromeda algorithm for data processing, database searching and protein identification [33]. Extracted peak lists were searched against the UniProtKB/Swiss-Prot Mus musculus database (Version 2010_10) containing 35,052 entries (including common contaminants) and a proportionally sized decoy database for false discovery rate (FDR) generation. The following search parameters were selected: fixed cysteine carbamidomethylation modification, variable methionine oxidation modification, variable protein N-acetylation, and variable phosphorylation of serine, threonine and tyrosine. A minimum peptide length of six amino acids and up to two missed cleavages were allowed. The initial first search mass tolerance was 20 ppm for precursor ions and 0.5 Da for fragment ions. Matches between runs were enabled with default settings and label-free quantitation enabled. The FDR was limited to 1% for both protein and peptide identification. For identification of the standard peptides, additional variable modifications of heavy proline (+6 Da) and alanine (+4 Da) were enabled.

Raw pY peptide spectral intensities were extracted from the 'Evidence' output files generated in MaxQuant. These intensities were then normalised against pY peptide intensities for the heavy-labelled spiked-in peptide standards. These normalised intensities were then log₁₀-transformed prior to all bioinformatics analyses, and their direct comparison enabled the relative quantitation of pY sites between the p53, PyMT and HER2 GEM models of breast cancer.

Raw MaxQuant output files were subjected to the following filtering criteria prior to their inclusion in further bioinformatics analysis: (1) contaminants and reversed matches were removed, ensuring that all identifications were reported with an FDR of <1%; and (2) pY sites were filtered for a minimum localisation probability confidence of 0.75.

Antibodies against the following proteins were used, with immunoblotting carried out as previously described [44]: IRS1 (Upstate Biotechnology, Lake Placid, NY, USA); IRS1 pY612 (Invitrogen), Stat3, Crkl, Jak-1, P85α, CDK16, Cav1, ErbB3, phospho-ErbB3 (Tyr1328; Tyr1325 in mouse), ErbB2, Met, Met pY1234/1235 and pY (P-Tyr-100) (Cell Signaling Technology); phospho-p85 (Tyr467) (GeneTex, Irvine, CA, USA); platelet-derived growth factor receptor α (PDGFRα) and β-actin (Santa Cruz Biotechnology, Santa Cruz, CA, USA); ERBB receptor feedback inhibitor 1 (Errfi1) (Sigma-Aldrich, St Louis, MO, USA).

Mouse tumour samples and liver tissue samples were harvested, snap-frozen and processed using the QIAamp DNA Mini Kit (QIAGEN, Valencia, CA, USA) according to the manufacturer's instructions. Briefly, up to 25 mg of frozen tissue was ground and digested with proteinase K overnight at 56°C. DNA was purified on a QIAamp spin column, eluted in 200 μl of water and assessed for quality using a NanoDrop spectrophotometer (Thermo Scientific, Waltham, MA, USA). Samples were cleaned up by sodium acetate/ethanol precipitation to ensure 260/280 and 260/230 absorbance ratios >1.8. DNA integrity was checked using gel electrophoresis.

Array comparative genomic hybridisation (aCGH) was performed by The Ramaciotti Centre for Gene Function Analysis (Randwick, NSW, Australia) on differentially labelled tumour DNA (Cy5) and sex-matched reference liver DNA (Cy3) using the Agilent SurePrint G3 mouse CGH 4 × 180 K microarray platform (Agilent Technologies, Santa Clara, CA, USA). The aCGH data were analysed using circular binary segmentation [45] to translate intensity measurements into regions of equal copy numbers. The raw array data were median-normalised; duplicate probes were averaged; and the resulting data were smoothed to remove outliers prior to segmentation. The gain and loss status for each region was assigned by determining plateaus in an ordered plot of all segments by their mean log₂ ratios. By this method, we denoted deletions to be segments with a mean value less than −0.5, amplifications to have mean value between 0.5 and 1.5 and highly amplified regions to have a mean value greater than 1.5. Using this information, we merged segmental values across the genome to create a common set of copy number levels for each individual tumour. Analysis was carried out using the Bioconductor DNAcopy package [46] in R Project R2.15.2 software [47].

Tumours were harvested and cellular preparations made using mechanical disruption and collagenase digestion as described previously [30]. Cells were cultured in vitro for two passages in mouse mammary epithelial cell media consisting of Dulbecco's modified Eagle's medium/Ham's F12 nutrient mixture supplemented with foetal calf serum, mouse epidermal growth factor, hydrocortisone, HEPES, insulin, gentamicin and penicillin/streptomycin [30], then trypsinised and seeded into 96-well plates at a density of 6,000 cells/well. The following day, cells were treated with the Met inhibitor PHA-665752 (Selleck Chemicals, Houston, TX, USA) at concentrations of 0.25 μM, 0.5 μM and 1 μM or vehicle treated with 0.1% dimethyl sulphoxide (DMSO; Sigma-Aldrich). Cell proliferation was assessed and recorded using the IncuCyte ZOOM system (Essen BioScience, Ann Arbor, MI, USA). Briefly, cells were cultured for 5 days while images of cells inside each well (three images per well across four replicate wells) were captured every 2 hours. Proliferation was measured at each time point as a percentage of confluence, as determined using the IncuCyte software on the basis of automated analysis of captured images. Data are representative of two replicate experiments. For immunoblotting studies, cells at 70% to 90% confluence were treated with either vehicle control (DMSO) or Met TKI (1 μM final concentration) for 1 hour and then lysed in 1% (vol/vol) Triton X-100 lysis buffer as previously described [48].

Met knockdown was performed using either individual (ON-TARGETplus Met siRNA 1 and 2) or SMARTpool siRNA targeting mouse Met (Dharmacon, Lafayette, CO, USA). The ON-TARGETplus nontargeting pool was used as a negative control. Briefly, cells were seeded into 96-well plates at a density of 4,500 cells/well 16 hours prior to transfection with 20 nM of siRNA using Lipofectamine 2000 reagent (Invitrogen). Cell proliferation was assessed by MTS assay ([3-(4,5-dimethyl-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium, inner salt; Promega, Madison, WI, USA), according to the manufacturer's protocol, 72 hours posttransfection.

Because the mammary gland develops postnatally, murine mammary epithelium can be transferred into the surgically cleared fat pads of juvenile isogenic naïve recipients. The use of GEM as donors, or viral transduction of cells after extraction, allows the rapid generation of chimeric models in which implanted transgenic or gene-knockout mammary epithelium develops in an otherwise wild-type host [21],[26]. Exploiting this strategy, we performed phosphoproteomic profiling using a combined immunoaffinity/LC-MS/MS approach across multiple MSCV-Neu (HER2), Tp53-null (p53) and PyMT tumours, as illustrated in Figure 1A. The lineage of each tumour from its original clone is outlined in Additional file 1: Figure S1A. In total, 856 unique pY sites were identified and quantitated from the tumour sample population with a FDR <1%. Following high-confidence localisation site filtering at ≥0.75, 763 unique pY sites from 506 unique phosphoproteins remained, with approximately 63% being common across all tumour types (Figure 1B and Additional file 2: Table S1).

These pY sites mapped to 506 unique proteins. These proteins, which were grouped based on Gene Ontology functional categories, are illustrated in Additional file 3: Figure S2A. Additional file 3: Figure S2B shows an expanded section of the kinases identified in the three tumour types. The largest proportion of kinases comprised serine/threonine kinases, followed by non-RTKs and RTKs. The least represented group comprised the dual-specificity kinases. The total pY site intensities were determined for the non-RTKs, which are shown in Additional file 3: Figure S2C. The most intense signal from the pY sites of non-RTKs was derived from SFKs (Yes, Fyn, Lyn and Hck), followed by focal adhesion kinase 1. When this approach was applied to RTKs, the most intense signal was derived from Met, followed by EphA2 (Additional file 3: Figure S2D).

To compare tyrosine phosphorylation patterns between the different tumour types and identify potential similarities, we undertook unsupervised hierarchical clustering. This revealed that two p53 tumour samples formed a distinct subgroup characterised by markedly elevated tyrosine phosphorylation of many pY sites (Additional file 4: Figure S3). The remaining tumour samples were distributed over a second major subgroup and exhibited a complex pattern of relationships. Of note, five PyMT tumours were closely related, and the majority of the p53 and HER2 tumours clustered together (Additional file 4: Figure S3). Also, although some tumours of shared origin, such as p53 tumours 1203 and 1204, clustered together, others, such as HER2 tumours HER 1197 and 3051, were well separated (Additional file 1: Figure S1B), indicating that divergence can occur following serial transplantation.

To generate a list of pY sites characteristic of each tumour type, we first identified pY sites detected in at least 75% of the corresponding samples. Therefore, we selected pY sites that were identified in at least (1) six of eight p53 tumour samples, (2) six of eight PyMT tumour samples or (3) four of five Her2 tumour samples. Following this selection-filtering process, 381 pY sites remained (Figure 1C and Additional file 5: Table S2). These sites mapped to 256 proteins, with 177 of them being found in all three tumour types. Upon unsupervised hierarchical clustering of this refined data set, all of the PyMT tumours were clustered together (Additional file 6: Figure S4). However, p53 tumours 1203 and 1204 still formed a distinct subgroup, and the HER2 tumours were divided into two groups, one related to PyMT tumours and the other to p53 tumours (Additional file 6: Figure S4).

To further refine the list of sites upregulated in a particular tumour type versus the other two, an analysis of variance (ANOVA)–based approach was implemented. Thirty-two pY sites were upregulated in HER2 tumours, eighty-six in p53 tumours and eighty in PyMT tumours (Additional file 5: Table S2). Of note, certain proteins exhibited site-selective variation in phosphorylation between different tumour types, examples being Tln1 and Cbl. Elevated levels of Erbb3 pY1325 and p85 pY467 in PyMT tumours were confirmed by Western blot analysis with appropriate phosphospecific antibodies (Additional file 7: Figure S5). Immunoblotting for selected targets revealed that, for certain pY sites characteristic of a given subtype, elevated phosphorylation was accompanied by increased expression of the given protein. This was exemplified by Irs1 and Pik3r1 (p85) in PyMT tumours (Figure 2A and B). However, for others, the increased phosphorylation must reflect a change in relative phosphorylation, as for Erbb3 in PyMT tumours, Cdk16 in p53 tumours and Erffi1 in HER2 tumours (Figure 2C to E). References for the sites characterised in this figure are provided in PhosphoSite [49].

We also developed and applied a random forest classifier to identify 17 highly variable pY sites that can be used to distinguish between the tumour types (Figure 3). Amongst these, elevated phosphorylation of Pik3r1, Pik3r3, a G protein–coupled receptor (Gprc5c) and an uncharacterised protein (2310030G06Rik) was specific to PyMT tumours, and p53 tumours were characterised by increased phosphorylation of Vim, Ptrf, Vcl, Fgr, Met and a protein related to heat shock protein 70 kDa. Elevated tyrosine phosphorylation of Errfi1 characterised HER2 tumours.

Lists of pY sites enriched in each tumour type were then subjected to bioinformatics analysis to identify associated pathways or biological processes (Table 1). Strikingly, marked differences were observed between p53, HER2 and PyMT tumours. The former exhibited an enrichment of processes linked to cytoskeletal reorganisation, with the top five significant terms being leucocyte transendothelial migration, bacterial invasion of epithelial cells, regulation of actin cytoskeleton, focal adhesion and muscle contraction. The only term significantly enriched for HER2 tumours was the ErbB signalling pathway, whereas the PyMT tumours were characterised by processes involving PI3K signalling, including type 2 diabetes mellitus, the FcεRI signalling pathway, acute myeloid leukaemia, Fcγ receptor–mediated phagocytosis and the neurotrophin signalling pathway.

Significantly upregulated pY sites (determined by ANOVA) characteristic of each tumour type were analysed using PINA [37],[38] (Figures 4, 5 and 6). Where available, consensus lists were searched against mouse data sets containing previously reported interaction data; however, in cases where interactions were not reported in mouse data sets, human orthologues were used to model potential interactions. In addition, kinase substrate relationships were extracted from PhosphoSitePlus [39]. A key characteristic of p53 tumours was the presence of a major protein–protein interaction hub composed of several proteins involved in cytoskeletal organisation (Actn1, Actn4, Sept7, Sept 9 and Plec), as well as a further hub centred on Itgb1 (Figure 4). The latter involved numerous proteins localised to focal adhesions, including Bcar1, Tln1, Pxn and Vcl. These findings are in agreement with the corresponding pathway analysis (Table 1). Furthermore, there are notable similarities between this network and that present in human basal and claudin-low breast cancer cell lines [16], including increased tyrosine phosphorylation of Met, Cav1, Bcar1 and Tln1. This is consistent with previous reports that subsets of p53 tumours exhibit gene expression signatures characteristic of these breast cancer subgroups [21],[22]. In contrast, HER2 tumours exhibited a protein–protein interaction network that reflected the relationship between HER2 and its associating substrates (for example, Cbl, Ptpn11, Plcγ1), but it is noteworthy that Egfr and Pdgfrα, as well as the feedback inhibitor Errfi1, are also components of this network (Figure 5). Consistent with the pathway analysis, PyMT tumours exhibit a major interaction hub that centres on the PI3K regulatory subunit Pik3r1 (Figure 6). Interestingly, in addition to PyMT itself, this hub involves two RTKs (Kit and Erbb3) and two docking proteins (Irs1 and Gab1) that are well-characterised binding partners for Pik3r1. Although Src is required for efficient induction of mammary tumours by PyMT [50], increased phosphorylation of Y416 on the Src activation loop (a peptide sequence conserved amongst SFKs) or Src-selective peptides was not detected in PyMT tumours. However, three additional SFKs—Yes, Lyn and Hck—exhibited increased phosphorylation on unique peptide sequences and resided within the network (Figure 6). The former is known to interact with PyMT [19] but is not essential for PyMT-induced mammary tumourigenesis [50].

In a previous study on the Tp53-null mouse mammary tumour model, researchers demonstrated that the generated tumours exhibit considerable heterogeneity and can be classified by transcript profiling into a variety of molecular subtypes, including one luminal, one claudin-low and two basal-like subtypes [22]. To interrogate this potential heterogeneity at the level of tyrosine phosphorylation patterns, we undertook unsupervised hierarchical clustering of the p53 tumours based on all phosphorylation sites detected in these tumours (Additional file 8: Figure S6). Tumours 1203 and 1204 clustered together and were characterised by markedly elevated (greater than tenfold) tyrosine phosphorylation of approximately 100 proteins compared to the remainder of p53 tumours (Additional file 9: Table S3). Tumour 14845 also exhibited relatively high tyrosine phosphorylation of many proteins, but not to the same extent as tumours 1203 and 1204, and tumours 3049, 1201 and 1202 displayed elevated phosphorylation of more distinct protein subsets. Tumours 1205 and 1206 were characterised by relatively low levels of tyrosine phosphorylation. Hypothesizing that the extremely high levels of tyrosine phosphorylation in tumours 1203 and 1204 might reflect upregulation of a particular tyrosine kinase in these two tumours, we noted that several of the upregulated pY sites were from Met, specifically Y1228, Y1232, Y1233, Y1347 and Y1354 (Figure 7). In addition, Y6 and Y14 on Cav1 also exhibited increased phosphorylation, which was of interest, given the close proximity of the Met and Cav1 genes on mouse chromosome 6 and coamplification of these two genes in particular tumour types [51]-[53]. Immunoblotting for Met and Cav1 revealed markedly increased total protein expression in the p53 samples from tumours 1203 and 1204 (Figure 8A and B). In addition, aCGH was performed on the p53 tumours. This revealed that, though p53 tumours 1201, 1205, 1206 and 14845 exhibited small (two- to threefold) increases in copy number for regions of chromosome 6 harbouring Met and Cav1, tumours 1203 and 1204 displayed 20- and 25-fold amplifications, respectively, of a chromosomal segment spanning these two genes (Figure 8C and D and Additional file 10: Table S4).

Having demonstrated that tumours 1203 and 1204 are characterised by Met gene amplification, we compared the tyrosine phosphorylation profile of these two tumours versus the other p53-null tumours in more detail. Consistent with enhanced Met signalling, tyrosine phosphorylation of Hgs, which regulates Met endocytic trafficking [54], was markedly elevated (Additional file 9: Table S3). These tumours were also characterised by increased tyrosine phosphorylation of Ptrf/cavin, which is localised with Cav1 to caveolae, and of the RTK Axl and the docking protein Dock1 (Additional file 9: Table S3). Interestingly, pathway analysis revealed that tumours 1203/1204 demonstrated significant enrichment for the term glycolysis/gluconeogenesis, reflecting increased tyrosine phosphorylation of several metabolic enzymes, including Pgam1, Pkm2 and Pgk1 (Additional file 11: Table S5).

To gain insight into the functional role of Met in the p53 tumours, we established cell lines from tumours 1201 and 1206 (low-level Met gene amplification) and tumour 1204 (high-level Met amplification). Pretreatment of these cell lines with the Met TKI PHA-665752 decreased Met phosphorylation on Y1234/Y1235 and markedly reduced the elevated cellular tyrosine phosphorylation levels in tumour 1204 cells (Figure 9A). Importantly, though administration of the Met TKI at a concentration (1 μM) that ablated Met tyrosine phosphorylation had no effect on proliferation in tumour 1201 cells (Figure 9B), this treatment markedly attenuated proliferation of tumour 1204 cells (Figure 9C). As an alternative and complementary approach, we knocked down Met expression using selective siRNAs (Additional file 12: Figure S7). Met knockdown had only a modest effect in the 1201 and 1206 tumour cell lines, but markedly attenuated proliferation in the 1204 tumour cell line with high-level Met amplification (P < 0.005 for comparisons of Met knockdowns in the 1201 or 1206 tumour cell line versus the 1204 tumour cell line). These data indicate that the extremely high levels of tyrosine phosphorylation detected in tumour 1204 (Figure 7) are dependent on Met activity, and that cells from these tumours are addicted to Met activation for their proliferation in vitro.

Interrogation of the tyrosine phosphorylation patterns characteristic of the other p53-null subgroups revealed that phosphorylation of the atypical kinase Peak1/SgK269 [55] was elevated in tumour 14845 and to a lesser extent the subgroup containing 3049, 1202 and 1201, whereas phosphorylation of Prex2 characterised tumours 14845 and 3049, where it was accompanied by enhanced phosphorylation of Pik3r2 (Figure 7). Taken in association with our demonstration of Met amplification in tumours 1203 and 1204, these findings highlight the power of tyrosine phosphorylation profiling to identify candidate oncogenic drivers [56].