Comprehensive Proteomic Characterization Reveals Subclass-Specific Molecular Aberrations within Triple-negative Breast Cancer

Summary

Triple-negative breast cancer (TNBC) is the most aggressive subtype of breast cancer lacking targeted therapies. This is attributed to its high heterogeneity that complicates elucidation of its molecular aberrations. Here, we report identification of specific proteome expression profiles pertaining to two TNBC subclasses, basal A and basal B, through in-depth proteomics analysis of breast cancer cells. We observed that kinases and proteases displayed unique expression patterns within the subclasses. Systematic analyses of protein-protein interaction and co-regulation networks of these kinases and proteases unraveled dysregulated pathways and plausible targets for each TNBC subclass. Among these, we identified kinases AXL, PEAK1, and TGFBR2 and proteases FAP, UCHL1, and MMP2/14 as specific targets for basal B subclass, which represents the more aggressive TNBC cell lines. Our study highlights intricate mechanisms and distinct targets within TNBC and emphasizes that these have to be exploited in a subclass-specific manner rather than a one-for-all TNBC therapy.

Graphical Abstract

Highlights

• •Proteome profiling reveals functionally distinct subclasses within TNBC
• •Kinases and proteases underlie unique functional signatures among the subclasses
• •Kinase-protease-centric networks highlight subclass-specific molecular rewiring
• •Protein association dysregulations reveal TNBC subclass-specific protein targets

Biological Sciences; Cancer Systems Biology; Cancer; Proteomics

Introduction

Triple-negative breast cancer (TNBC) that represents 10%–15% of all breast cancers is the most aggressive breast cancer subtype characterized by high recurrence and poor prognosis (Ovcaricek et al., 2010). Unlike receptor positive subtypes that include estrogen- and/or progesterone-receptor positive (ER⁺/PR⁺) and the human epidermal growth factor receptor 2 positive (HER2⁺) subtypes, TNBC is a highly heterogeneous breast cancer subtype, making it challenging to understand its underlying disease mechanisms and design targeted treatments (Anders and Carey, 2009, Badve et al., 2011, Bauer et al., 2007). Several research groups have attempted subclassifying TNBC toward the mutual goal of finding targeted therapies. TNBC has been categorized under the basal-like category according to the PAM50 classification (Parker et al., 2009, Perou et al., 2000, Sørlie et al., 2001) and has been further divided into two distinct subclasses as basal A and B based on gene clustering (Neve et al., 2006). Another study based on gene expression analysis of TNBC tumors has identified six subcategories with distinct gene ontologies as two basal-like (BL1 and BL2), two mesenchymal (M and MSL), one immunomodulatory (IM), and one luminal androgen receptor subtype (LAR) (Lehmann et al., 2011). Apart from these, subclassification based on specific marker expression, such as claudin-low subtype characterized by low expression of claudin-3, -4, and -7 (Herschkowitz et al., 2007, Prat et al., 2010) or based on elevated expression of positive markers such as cytokeratins, mainly 5, 6, 13, and 17, have also been reported (Bianchini et al., 2016, Foulkes et al., 2010, Nielsen et al., 2004). The heterogeneous nature of TNBC is also accentuated at the protein level by large proteomic profiling of breast cancer tissues and cell lines (Lapek et al., 2017, Lawrence et al., 2015, Mertins et al., 2016). Although all these studies pointed to existence of subclasses within TNBC, their specific mechanistic differences or pathway dysregulations were not explored in detail.

With no current effective therapy, there has been a huge interest in deconvoluting the oncogenic events that lead to the development of TNBC, and several aberrantly activated pathways have been identified. Among them, TGF-β signaling is linked to aggressive behavior such as TNBC cell migration, early metastasis, as well as invasion and extravasation (Massagué, 2008, Padua et al., 2008, Papageorgis et al., 2010, Xu et al., 2018). Other signaling pathways mediated by neurotrophic tyrosine kinase receptor (TrkB), Erk/NF-κB, PI3K/AKT, and HIF-1α have also been associated with TNBC metastasis (Costa et al., 2018, Delaloge and DeForceville, 2017, Kuo et al., 2017, Neophytou et al., 2018, Ponente et al., 2017, Tsai et al., 2017, Wong et al., 2011). Despite several ongoing trials targeting such aberrant pathways, the clinical outcome has been disappointing with limited benefit in TNBC patients. This underscores the complexity of TNBC at the molecular level and emphasizes the need for elucidating TNBC subclass-specific mechanisms and pathways for effective personalized therapies.

Given the distinct genomic alterations and clinicopathological characteristics within different subtypes of TNBC (Bareche et al., 2018), we hypothesized that an in-depth analysis of subclass-specific proteome changes associated with pathway aberrations would reveal molecular vulnerabilities within TNBC, thus identifying novel therapeutic targets. To this end, we carried out comprehensive proteomic profiling of an array of breast cancer cell lines, employing multiplexed quantitative high-resolution mass spectrometry (MS)-based proteomics followed by extensive peptide fractionation to record comprehensive quantitative protein and peptide information across all cell lines. We interrogated the data extensively for network and pathway aberrations, particularly associated with perturbed functional protein classes. Through this strategized and systematic analysis, we identified TNBC inherent proteome-based subclasses, which were also reflected by distinct profiles of kinases and proteases, and their associated functional networks. Finally, using correlation networks, we inferred that dysregulations of kinases and proteases are centric in defining molecular traits unique to each breast cancer subtype and validated kinase AXL as an effective target only in the aggressive TNBC subclass. Altogether, this study provides in-depth proteomic profiles of TNBC cell line panel and reveals subclass-specific mechanisms and potential dysregulated targets, thus offering immense potential for clinical exploration.

Results

In-depth Proteomic Profiling

We performed high-resolution MS-based deep proteomics profiling of 18 breast cancer cell lines (12 TNBC, 3 ER⁺/PR⁺, and 3 HER2⁺) by multiplexed tandem mass tag (TMT) approach that allows reliable and robust quantification of proteins across the different cell lines. Two complimentary offline peptide fractionation methods, namely iso-electric focusing (IEF) and offline basic pH reverse-phase liquid chromatography (bRP-LC), were employed prior to MS analysis to obtain a comprehensive proteome by enhancing peptide coverage. Two biological replicates, accounting for 156 LC-MS runs, were included to ensure reproducibility across the panel. The overall workflow is depicted in Figure 1A. Protein quantification between the biological replicates as well as between the two fractionation approaches was highly correlated (Figure S1A), indicating good reproducibility of our MS dataset. In total, we identified a total of 233,910 peptides derived from 11,709 proteins at a false discovery rate (FDR) of 1%, of which 10,400 were quantified across all 18 cell lines (Figure 1B and Tables S1A and S1B), highlighting consistent quantification with few missing values. By capturing unique peptide clusters, the implementation of the two different fractionation approaches in parallel boosted the overall proteome coverage by 25% compared with the proteome profiles obtained using only either of the two approaches (Figure 1C). With an average of 22 peptides identified per protein, of which 16 peptides on an average were also isoform specific, the improved peptide coverage contributed to reliable protein identification and quantification. To our knowledge, this is the most comprehensive quantified proteomes of breast cancer cells, with the highest average number of unique peptides per protein reported to date. Notably, our dataset also reports the highest numbers for quantified proteins belonging to important functional classes such as transcription factors, kinases, proteases, and phosphatases that have not been captured and profiled extensively thus far by MS methods in the breast cancer landscape (Figure 1B). This high proteome coverage was obtained mainly because of our strategized peptide fractionations that enhanced detection capability of peptides by changing their chemical nature (high pH and isoelectric ambient).

Figure 1.

Comprehensive MS-based Proteomics Analysis of Breast Cancer Cells

(A) Schematic of the bottom-up proteome profiling pipeline for screening 18 breast cancer cell lines employing TMT multiplexing with two peptide fractionation strategies.

(B) Number of identified (and overall quantified) proteins, unique peptides, and average number of peptide spectrum matches (PSM) in total, along with their distribution in four functional protein classes. Percentage represents completeness of identification of each protein class.

(C) Total number of peptides identified in samples fractionated by IEF or bRP-LC.

Proteome-Based Clustering

To evaluate the quality of our MS dataset, we mapped expression levels of known molecular markers to cell lines representing each breast cancer subtype. We observed high expression levels of ESR1 and PGR in all ER⁺/PR⁺ cells, high levels of ERBB2 in HER2⁺ cells, and no/low expression of all these receptors among the TNBC cell lines, adding confidence to our proteome dataset (Figure 2A upper histogram). Next, we performed unsupervised clustering of the whole proteome expression data that revealed three main clusters with all receptor positive cell lines grouped into one cluster and the TNBC cell lines bifurcating into two distinct clusters (Figures 2A top tree and S1B). The two TNBC clusters corresponded to well-established TNBC classification based on claudin and cytokeratin expression patterns (Figure 2A middle/lower histogram). Our clustering also aligned with PAM gene expression classification (Neve classification) for breast cancer cell lines that is based on 305 genes (Neve et al., 2006). On comparing our proteome-based categories with the TNBCtype classification (Lehmann et al., 2011), we observed that mesenchymal (M) and mesenchymal stem-like (MSL) were representative of basal B, whereas basal-like 1 and 2 (BL1 and BL2) resembled basal A subclass (Figure 2A middle panel). The BT20 cell line, which could not be distinctly categorized by the TNBCtype classification, appeared to align with the basal A category in our analysis. A similar grouping of BT20 was also reported by Lawrence et al. (2015) in their label-free proteomics study that assigned them to a cluster resembling the BL1 category, thus validating our assignment. However, in contrast to observations by both our study and that of the TNBCtype classification, the MSL MB-436 cell line was categorized within a group most similar to the BL2 group by Lawrence et al. (2015). Furthermore, our analysis, for the first time, also distinctly assigned the uncategorized TNBC cell line HCC1395 to the basal B subclass based on the proteome pattern. We also compared our proteome profile data with published transcriptome data (Klijn et al., 2015) and noted modest correlation, consistent with previously reported observations (Figure 2A lower panel) (Lapek et al., 2017, Sacco et al., 2016, Zhang et al., 2014). Principal component analysis (PCA) of the whole proteome distinctly demarcated the cell lines into the luminal group with clear differentiation of ER⁺/PR⁺ and the HER2⁺ groups and the TNBC group showing two subclasses with discrete patterning based on their proteome landscape (Figure 2B).

Figure 2.

Proteome Profiles of Breast Cancer Cell Lines

(A) Clustering of the cell lines based on the whole proteome expression profiles is shown (top). Relative expression of ER⁺/PR⁺ and HER2⁺ and known claudin and cytokeratin protein markers in breast cancer cell lines. Data are represented as mean ± SEM, n = 4 (upper panel). Comparison to TNBCtype and Neve classification of breast cancer cell lines based on gene expression studies is shown (middle panel). Correlation of protein abundance with mRNA expression is shown (lower panel). The color schema for the respective proteins/categories is shown on the right.

(B) Principal component analysis (PCA) of the whole proteome distinguished cell lines into the luminal subtype (HER2⁺ and ER⁺/PR⁺) and TNBC subtype (basal A and basal B).

(C and D) (C) PCA based on only kinases or (D) proteases showed distinct patterns in cell line classifications.

Next, we sought to ask whether sub-proteomes of major functional protein classes, which are important regulators of cellular signaling, also show similar clustering pattern as the whole proteome. For this, we considered selected functional groups that included kinases, proteases, phosphatases, and transcription factors. PCA of the 321 overall quantified kinases [72% of the human kinome (Manning et al., 2002)] revealed that the expression patterns of only these kinases were sufficient to reflect the same clustering pattern among the different breast cancer cell lines as that obtained using the whole proteome, iterating distinctive kinase expression patterns among the different breast cancer groups (Figure 2C). A similar clustering was also observed for the 247 proteases [51% of human proteases (Rawlings et al., 2018)], with only two cell lines (HCC1806 and MB-231) appearing as outliers based on their protease expression patterns (Figure 2D). However, PCA of 133 phosphatases [71% of human phosphatases (Liberti et al., 2012)] and 1034 transcription factors [61% of human transcription factors (Ravasi et al., 2010)], which represents the largest protein class, showed no obvious distinctive clustering among the different subclasses as by the kinases or proteases (Figure S1C). These results suggested that kinases and proteases have differential signatures among the different subgroups and potentially play prominent roles in orchestrating unique mechanisms within the groups. Hence, in this study, we focused on a kinase- and protease-centric perspective to reveal subclass-specific aberrations.

Subgroup-specific Kinase and Protease Expression

Among the 2,440 differentially expressed (DE) proteins (ANOVA p value ≤0.05 and two-sample t-test p value ≤0.05 for both pairwise comparisons), we observed 87 kinases and 62 proteases (Table S1A) that were exclusively differentially expressed in one group in comparison to the other two groups. Mapping these DE kinases onto the human kinome map (Manning et al., 2002) revealed that a majority of these belonged to tyrosine kinases, CAMK, and AGC kinase families (Figures 3A and S2B). We also compared these DE kinases with targets of approved drugs (www.drugbank.ca; Dec 2019, Figure S2A) and found that a few of them including DDR2, ERBB2, EPHA2, PRKCA, CDK6, and AXL overlapped, indicating that these DE kinases could be potential therapeutic targets, especially for the basal B subclass.

Figure 3.

Kinase and Protease Profiles of TNBC Subclasses

(A) Differentially expressed kinases (p ≤0.05) specific to either basal A or basal B compared with the other two subclasses mapped onto the human kinome tree. The size and shade of the node represents the median log₂ fold-change of the kinases within each subclass with red and blue representing up- and downregulated kinases, respectively. Those kinases indicated with asterisk are differentially expressed by at least two-fold against both the luminal group and the other TNBC subclass. Illustration was created using CORAL online tool (Metz et al., 2018) based on the human kinome tree courtesy Cell Signaling Technology Inc (www.cellsignaling.com).

(B) Differentially expressed proteases (p ≤0.05) specific to either basal A or basal B compared with the other two subclasses mapped according to their protease families. The median log₂ fold-change of the protease within each subclass is represented by the node size and shade. Asterisk denotes those proteases differentially expressed by at least two-fold against both the luminal group and the other TNBC subclass. Red and blue represent up- and downregulated proteases, respectively. The protease families include S – serine, C cysteine, M – metallo, A aspartate, T – threonine proteases, and P – peptidases.

We observed 46, the highest number of DE kinases, specifically within basal B cell lines (Figure 3A right). Some of these kinases have been previously described as promising therapeutic candidates in TNBC. For example, PRKCA and CDK6 have been correlated with poor survival outcomes in TNBC and displayed synergy in impeding tumor growth (Hsu et al., 2014). Targeting the hyperactive CDK4/6 axis in breast cancer has attracted traction as a promising therapeutic (Matutino et al., 2018, Yamamoto et al., 2019), and recently, CDK6 inhibition was proven effective for the MSL subclass of TNBC (Pernas et al., 2018). CAMK1 and DDR2 have been linked to basal-like breast cancer wherein they drive metastasis by triggering EMT (Bergamaschi et al., 2008, Corsa et al., 2016, Zhang et al., 2013). EPHA2 has emerged as a new target impairing tumorigenesis in TNBC, and receptor tyrosine kinase AXL has also been reported to play prominent roles in TNBC tumorigenesis (Leconet et al., 2017, Song et al., 2017). Apart from these targets, our study also highlighted other kinases that are yet to be explored for clinical therapeutic interventions. One among them is PEAK1 that modulates TGF signaling responses toward oncogenic progression in breast cancer (Agajanian et al., 2015a, Agajanian et al., 2015b). Another kinase, NUAK1, has never been described in breast cancer and opens possibilities for new therapeutic explorations in aggressive TNBC subtypes. Although the roles of FYN in mediating EMT and promoting oncogenesis has been described in many cancers including breast cancer (Chen et al., 2013, Xie et al., 2016), its specificity to TNBC or even particular subclass has not been considered. In contrast to the vast array of kinases modulated in the basal B group, the basal A displayed fewer kinases that changed significantly in abundance compared with both the luminal and basal B groups. Tyrosine kinase LYN of the SRC family of kinases that has been linked to aggressive phenotypes including EMT, migration, and invasion in TNBC and another tyrosine-like kinase MLKL were overexpressed specifically within this group (Figure 3A left) (Choi et al., 2010, Tornillo et al., 2018).

From our data, we found that a majority of basal A-specific DE proteases were upregulated (Figure 3B left). Significant upregulation of three kallikreins (KLK5/6/8), three caspases (CASP2/10/14), and three proteasomal components (PSMB8/9/10) suggests involvement of complex protease machinery in driving basal A-specific tumorigenesis. Interestingly, the proteases PSMB8/9/10, all are components of the immunoproteasome representing the modified proteasome that elicit T-cell-mediated immune response underpinning immunomodulatory roles specifically in the basal A subclass (Rouette et al., 2016). Although some of these proteases have been associated with breast cancer tumorigenesis previously (Haritos et al., 2018, Yang et al., 2015), they have not been described in a subclass-specific manner except for CASP14 that was reported to correlate with tumors expressing high claudin and EGFR, characteristic of basal A subclass (Handa et al., 2017). As compared with both luminal and basal A groups, basal B subclasses showed the highest number of proteases that were altered in abundance. Moreover, in contrast to the luminal and basal A subclasses, the majority of DE proteases belonged to the families of metalloproteases and serine proteases (Figures 3B and S2C). Among the metalloproteases, four proteases belonged to the ADAM and ADAMTS family of proteins, with all showing basal B-specific upregulation except for ADAM15. Serine protease FAP, which was found to be overexpressed by 3.6-fold in basal B specifically, has been shown to increase proliferation, migration, and invasion of cancer cells (Huang et al., 2004, Jia et al., 2014). A known metalloprotease MMP2 responsible for extracellular matrix (ECM) degradation in invasive breast cancer was also particularly upregulated within this subclass (Jezierska and Motyl, 2009, Talvensaari-Mattila et al., 2003). Interestingly, another serine protease HTRA1, which has been widely reported as a tumor suppressor protein with decreased expression being correlated to EMT, proliferation, and angiogenesis (Klose et al., 2018, Lehner et al., 2013, Wang et al., 2012, Xia et al., 2013), was observed to be highly overexpressed only in the basal B subclass. This suggests that there are context-specific roles of HTRA1 that regulate differential mechanisms depending on the breast cancer subclasses. Among all the altered proteases, UCHL1 was the only cysteine protease to be abundantly expressed in the basal B group. Altogether, our study highlights differential expression patterns of kinase and protease families across different breast cancer subtypes and iterates differential mediation of oncogenic responses across the different subtypes.

Kinase- and Protease-Centric Pathway Aberrations within TNBC Subclasses

Next, we performed expression-based kinase- and protease-centric network analysis in order to understand their involvement in subclass-specific molecular rewiring. For this, all human protein-protein interactions (PPI) associated with the DE kinases and proteases were curated from the Pathway Commons database to identify neighborhood interactors (Cerami et al., 2011). In addition to physical interactions, functional interactions encompassing those that alter phosphorylation, influence expression, and regulate complex formation were also included to dissect functional specifications within each subtype. The network neighborhood was then filtered to retain only those proteins that overlapped with the DE proteins within each subtype to assemble a subtype-specific kinase-protease-centric expression network (Figures 4A and S3A).

Figure 4.

Kinase-Protease-Centric Interaction Network

(A) Proximity network of differentially expressed kinases and proteases in each TNBC subclass. The kinases and proteases are depicted as green and blue nodes, respectively, and the size of the nodes represents the median log₂ fold-change against the other two subclasses, with larger nodes representing increased fold-change and vice versa. The edges represent physical and functional interactions between proteins.

(B) Distribution of protein ratio within each breast cancer subtype for the indicated cancer hallmark gene sets (*, p value ≤0.05; **, p value ≤0.01; n.s., not significant—using KS test).

On comparing the subclass-specific networks with each other, it was evident that the basal B network displayed a denser topology than the basal A kinase-protease network; partly due to the higher number of DE kinases and proteases identified within basal B subgroup. Despite being a sparse network with only 30 kinases and proteases, the basal A kinase-protease network showed preferential ontology enrichment for regulation of apoptotic processes, cell proliferation, and keratinization. We also identified several DE proteins involved in WNT signaling in the neighborhood network, hinting that the modulated kinases and proteases may have profound influence over WNT signaling regulation in basal A subclass. In addition, the network highlighted kinases including STK24 (MST3), CSNK2A2, and LYN as hub proteins based on their network centrality. Some of these including MST3 and LYN have been identified as pro-proliferative and tumorigenic in TNBC (Cho et al., 2016, Tornillo et al., 2018), and hence implicated as targets, although their specificity to possible subtypes have not been explored. The basal B neighborhood network, on the other hand, revealed a close-knit functional association thus emphasizing on the complex interplay between kinases and proteases in this subtype. In agreement with the aggressive nature of the basal B group, the kinase-protease network showed particular inclination toward processes associated with focal adhesion, extracellular matrix organization, ephrin, and TGF signaling (Canel et al., 2013, Drabsch and ten Dijke, 2011, Lu et al., 2012, Wang, 2011), all of which mediate a more migratory and invasive phenotype. In addition, proteins associated with angiogenic and platelet activation were also enriched in the proximity neighborhood of basal B-specific kinases and proteases. The interaction network spanning a total of 75 kinases and proteases included the before-mentioned kinases AXL, PRKCA, PEAK1, CDK6, and EPHA2 and proteases FAP, MMP2, and UCHL1 in close connectivity with each other, emphasizing an intricate rewiring of kinase-protease networks within the basal B subclass. Some of these kinases and proteases have been shown to correlate with poor prognostic outcomes in breast cancer and are currently being explored as therapeutic targets (Hsu et al., 2014, Leconet et al., 2017, Liu et al., 2019, Song et al., 2017). We also inferred crucial hubs from the neighborhood network, a majority of which play key roles in invasion and metastatic behavior. Specifically, we identified components of TGF signaling including SMAD3, EMT-activator ZEB1, ECM regulators MMP2 and FN1, and transcriptional regulators JUN and SP1 as key proteins in basal B subclass. A cluster assessment based on MCODE topological analysis highlighted several intricately connected proteins, with the top enriched clusters showing functional preference for ECM remodeling, focal adhesion, inositol phosphate metabolism, HIF-1, TGF, and ephrin signaling within the basal B kinase-protease proximity network (Figure S3B). All these evidences suggest that the basal B kinases and proteases work in concert with each other to rewire signaling pathways particularly involved in cell migration and invasion, thus explaining their aggressive nature. This is strikingly different from the signaling events and processes influenced by basal A-specific kinases and proteases.

With observations of distinct functional biases for the two subclasses based on ontology from the kinase-protease proximity network, we next looked into cancer hallmark gene sets uniquely enriched within each breast cancer subgroup using gene set enrichment analysis (GSEA) to assess oncogenic signatures (Table S2A). As expected, the luminal cell lines showed enrichment of estrogen response process. In addition, the luminal group also showed preferential enrichment of metabolism-associated processes such as oxidative phosphorylation and fatty acid metabolism. Interferon-induced signaling was particularly enhanced in basal A group, and the proliferation-associated p53 signaling pathway was also positively enriched in basal A compared with basal B group. Along with EMT, which emerged as the top hallmark in basal B group, other cancer hallmarks such as hypoxia, inflammatory response, and coagulation also displayed selective enrichment in the basal B group. To ascertain if the kinase and protease signatures within each subtype underlie such differences, we specifically compared the distribution of only proteins within the kinase-protease neighborhood belonging to these cancer gene sets (Figures 4B and S4A). Excitingly, we found that the proximity proteins of kinases and proteases are representative of such functional processes, and accordingly EMT, TGF signaling, TNFa signaling via NF-κB, coagulation, hypoxia, and inflammatory response, all showed elevated response in basal B subclass compared with luminal and basal A groups. Similar functional biases were also observed for basal A subclass for interferon signaling and luminal group for estrogen and metabolic response suggestive of intricate functional roles for kinases and proteases in driving subclass-specific distinctive processes. We also looked into the KEGG pathway database to assess pathway preferences for proteins within the kinase-protease neighborhood and found that positive regulation of TGF, PI3K/AKT, HIF-1, RAP1, and sphingolipid signaling in basal B cell lines could be effectively represented by the abundances of kinase-protease proximity proteins (Figure S4B and Table S2B).

Dysregulated Kinase and Protease Associations within TNBC Subclasses

We expanded our analysis to investigate differences in the protein co-regulation networks between the subclasses. For this, protein correlations were obtained between all protein pairs of the 10,400 quantified proteins, and a high-confidence correlation network was obtained after filtering for significant associations. The positive correlation network included a total of 92,643 co-regulated protein pairs, and the negative correlation network comprised of 61,061 associating pairs, both corresponding to a minimum Spearman's rank correlation coefficient of ∼0.76 (Table S3). To infer cell-line-specific network aberration, we specifically identified protein pairs displaying significant deviations from correlation in each cell line by performing outlier analysis on all protein pairs in the significant protein-association networks (Lapek et al., 2017) (Table S4). The median dysregulated associations based on positively and negatively correlating protein pairs across all cell lines were 1,444 and 860, respectively. On mapping these dysregulated protein associations, we observed differential patterning of proteins with deviations that aligned with the breast cancer subtypes, particularly for those inferred from positive correlations. This suggested that protein dysregulations are more likely to be conserved across different subtypes, and investigating such dysregulations can offer insights on subtype-specific pathways and mechanisms. Thus to identify dysregulated proteins specifically enriched within each breast cancer subtype we carried out enrichment analysis using hypergeometric testing (Figures 5A and S5A and Tables S5A and S5B). This analysis highlighted that the basal B subclass included the most number of dysregulated proteins (672), followed by luminal subtype with 455 and basal A subclass with 171 dysregulated proteins. Through ontology-based enrichment analysis of these dysregulated proteins, we observed subtype-specific dysregulated pathways (Figure 5A and Table S6). For instance, proteins mainly involved in metabolism and mitochondrial translation were dysregulated in the luminal subgroup. In basal A, proteins involved in epidermis development/keratinization, innate immune response, interferon signaling, and those regulating proliferation and apoptosis showed dysregulation, whereas for basal B, those associated with ECM organization, migration, angiogenesis, PI3K/AKT, and TGF signaling pathway were dysregulated. Thus the dysregulations inferred captured functional distinctions between the two subclasses. In several instances, the associated proteins also displayed consistent pairwise dysregulation within subclasses, with basal B displaying the most number of deviated protein pairs (Figure S5B and Tables S5C and S5D). For example, protein associations involving EPHA2, TGFBI, DKK3, AXL, SERPINE1, and several collagens were dysregulated in a majority of the basal B cell lines. In contrast, the basal A cell lines revealed far less number of conserved protein pairs, with associations involving SLPI, KLK5, and keratins showing perturbations in many cell lines in a basal A-specific manner.

Figure 5.

Dysregulated Proteins in Breast Cancer Subclasses

(A) Proteins with dysregulated associations in each breast cancer subclass mapped onto the background protein-protein association network derived from co-regulation analysis of protein abundances from all breast cancer cell lines. The network includes only positive correlations (for negative correlations see Figure S5A). The corresponding functional annotation of the dysregulated proteins for each subclass is indicated below.

(B) Dysregulated kinases and proteases specifically enriched within each TNBC subclass are depicted as heatmap based on -log₁₀ transformed p values derived from hypergeometric testing. Kinases or proteases derived from dysregulations of positive associations and negative associations are depicted as red and blue, respectively. K denotes kinases and P denotes proteases.

(C and D) (C) Perturbed association network of kinases and (D) proteases specifically dysregulated in basal B subclass. Red edges denote perturbations inferred from positive associations and blue edges represent perturbations inferred from negative associations.

We posit that kinases or proteases displaying association perturbations with several other proteins in a subtype-specific manner reflect their activity modulation in that particular subtype. Interrogating the dysregulated kinases and proteases among different breast cancer types, we found that the basal B group had the highest representation of perturbed kinases and proteases (Figures 5B and S6 and Tables S5A and S5B). Among them, AXL, DDR2, PEAK1, EPHA2, and FYN were also specifically overexpressed in basal B stressing on possible key roles played by these kinases in the aggressive basal B subclass. In addition, our analysis also identified several other kinases that were not significantly differently expressed but whose associations have been dysregulated in a basal B-specific manner (Figures 5B and 5C). For example, receptor kinases FGFR1 and TGFBR2 were predicted to be dysregulated only in basal B subclass. FGFR1 is regarded as a prognostic factor in TNBC, and TGF-β pathway components have been considered promising targets for impairing tumorigenesis in TNBC (Bhola et al., 2013, Cheng et al., 2015, Jamdade et al., 2015). Associated kinases that regulate these pathways including the neurotrophic tyrosine kinase ROR1 that potentiates FGFR signaling in basal breast cancer, and activin signaling receptor-like kinase AVCR1 that activates SMAD pathway were also predicted to be dysregulated in a basal B-specific manner (Bashir et al., 2015, Pandey et al., 2019, Ramachandran et al., 2018) A member of the IRAK family of kinase, IRAK1, which mediates toll-like receptor (TLR) signaling, has been particularly proposed as a therapeutic target driving TNBC metastasis (Wee et al., 2015). Our dysregulation analysis predicted another member IRAK2, which is crucial in late-phase TLR signaling, to be modulated in basal B. It is interesting to note that many of the dysregulated kinases including TGFBR2, PEAK1, ACVR1, EPHA2, EPHA4, and AXL shared perturbed associations with multiple proteins within the basal B subclass, suggesting prominent roles for these kinases in driving oncogenic aberrations (Figure 5C). Apart from these kinases inferred from positively correlated dysregulated associations, we also identified others that displayed deviations in their negative correlation especially in the basal B group. Among them, the non-receptor tyrosine kinase TNK1 (KOS1) has been reported as a tumor suppressor in breast cancer that downregulates RAS signaling (Hoare et al., 2008). Another transcriptional kinase, CDK13, however, is associated with aggressive features in breast cancer and is regarded as a potential target for TNBC (Quereda et al., 2019). In comparison to the extensive modulation of kinase association network in basal B group, only two kinases were identified to be dysregulated in basal A class. The dysregulated MST1R kinase, also known as RON receptor tyrosine kinase, is frequently overexpressed in TNBC, and anti-RON targeted therapy has been advocated based on promising outcomes from preclinical TNBC models (Suthe et al., 2018).

Apart from kinases, we also observed several proteases displaying subtype specificity (Figures 5B, 5D, and S6 and Tables S5A and S5B). Strikingly, in contrast to kinase perturbations, protease association dysregulations were more pronounced in basal A subgroup. Proteases such as caspases 10 and 14, kallikreins 5 and 6, and calpain 13, all displayed perturbations with multiple proteins among the basal A cell lines, implicating dysregulation of a wide array of protease families in the basal A subclass. Some of these including KLK5 and CASP14 are prognostic markers for aggressive TNBC (Handa et al., 2017, Yang et al., 2015). In the basal B subclass, we predicted multiple metalloproteases (MMP2 and MMP14), ADAM family proteases (ADAM12 and ADAMTS1), and deubiquitinases (UCHL1 and STAMBPL1) to be specifically dysregulated. Although MMP2 protease was found to be both dysregulated and also overexpressed in basal B group, we predicted MMP14 to be particularly modulated in basal B subclass based on our dysregulation analysis. Of note, ADAM12 has been reported to promote cancer stemness in claudin-low breast cancer (Duhachek-Muggy et al., 2017), and this aligns with our prediction that ADAM12 is dysregulated specifically in the basal B subclass that expresses low claudin. Apart from these, a class of serine proteases encompassing PRSS23 and PRSS8 were also found to have dysregulated associations in the basal B group. PRSS8, also known as prostasin, has been reported to inhibit breast cancer invasiveness (Chen and Chai, 2002), and our analysis predicted negative association perturbations.

AXL Regulates Subclass-specific Migration and Invasion

Our comprehensive protein-protein network analysis revealed several TNBC subclass-specific potential therapeutic targets including kinases and proteases. We validated some of these to highlight protein expression differences within the three breast cancer groups by Western blotting in multiple representative cell lines (Figure S7).

Multiple evidences suggest dysregulation of AXL, and its associations specifically in basal B subclass, and validation experiments confirmed AXL overexpression exclusively in basal B subclass as against the other groups (Figures 3A, 5C & 6A, S3B, and S7A). Hence, we hypothesized that basal B cells should be more sensitive to AXL inhibitors than basal A cells. To test this, we treated seven TNBC cell lines representing basal A (BT20, MB468, and HCC70) and basal B (BT-549, Hs578t, MB-157, and MB-231) subclass with AXL-specific inhibitor R428 that inhibits the autophosphorylation of AXL required for its activity (Korshunov, 2012, Sadahiro et al., 2018). Using an antibody against phosphorylated AXL at tyrosine 779 (pAXL-Tyr⁷⁷⁹), we confirmed effective inhibition of AXL activity upon treatment with R428 inhibitor in basal B cell lines (Figure 6B). Next, we investigated whether inhibition of AXL suppresses the aggressive behavior of TNBC cell lines by assessing the migration and invasion response of the selected TNBC cell lines to AXL inhibitor R428 using modified wound-healing assays. The reduction in AXL activity with inhibitor treatment was accompanied with a significant decrease of invasion for all basal B cell lines, with Hs578t showing the highest sensitivity (∼80% reduction in invasion) (Figure 6C). As expected, the basal A cell lines showed no significant changes in their invasive potential in response to AXL inhibitor (Figure 6C). The invasive feature of basal B cell lines was specifically responsive to AXL inhibition as compared with the migratory potential that displayed cell-line-dependent responses (Figure S7B). We also performed the functional assays using transient AXL knockdown in Hs578 T cell line, and the results showed significant decrease in invasion, whereas migration showed no notable effects (Figures 6D and 6E). Altogether, this suggested that AXL kinase and its associated mechanisms mediate aggressive behavior specifically in basal B subclass and hence is an effective target for basal B but not basal A type tumors.

Figure 6.

Subclass-specific Regulation of AXL Kinase

(A) Western blot analysis of AXL expression in selected basal A and B cell lines (only a section of Figure S7A is shown).

(B) Western blot analysis of AXL and pAXL-Tyr⁷⁷⁹ in selected basal B cell lines with and without AXL inhibitor (R428) treatment.

(C) Invasive potential represented as relative gap closure rates between R428 and DMSO-treated cells. Data shown as mean ± SEM, n = 3. *, p value ≤0.05; **, p value ≤0.01; n.s., not significant (using t-test).

(D) Migratory and invasive potential represented as relative gap closure rates for Hs578t cell line treated with siRNA against AXL (siAXL) or scramble (siSCR) as control. Data shown as mean ± SEM, n = 3. *, p value ≤0.05; **, p value ≤0.01; n.s., not significant (using t-test).

(E) Western blot analysis of AXL expression in Hs578T cell line treated with siRNAs.

Discussion

Deconvolution of TNBC complexity is necessary to resolve molecular mechanisms and devise effective targeted therapies for this highly heterogeneous and aggressive breast cancer subtype. In this study, we aimed to explore patterns of proteome profiles and pertinent pathway dysregulations within TNBC by employing advanced proteomics MS in combination with methodical computational approaches. The coupling of proteomic MS with two peptide fractionation methods boosted the overall quality of our proteome data with higher peptide and proteome coverage along with robust quantification accuracy as revealed by good concordance between the biological replicates. This is exemplified by the identification of regulatory proteins belonging to various functional classes such as kinases, transcription factors, and proteases, which are generally not extensively profiled by conventional MS-based proteomics profiling and hence are often underrepresented. With consistent high coverage of proteome across the panel, we obtained the most comprehensive quantitative proteome profile of breast cancer cell lines reported to date. Unsupervised clustering of the unprecedented high-quality proteome revealed two major subclasses within TNBC, basal A and B, which aligned with previously reported transcriptome-based classification. Moreover, for the first time, we observed distinct expression profile patterning of kinases and proteases that were able to segregate the two TNBC subclasses, stressing on key roles played by these protein classes within TNBC.

Although similar TNBC subclasses based on proteomics differences have been reported previously (Lapek et al., 2017, Lawrence et al., 2015), our observation of exclusive subclass-specific kinase and protease signatures offered direct phenotypic perspectives on the different TNBC subclasses. In fact, some of the observed basal B-specific kinases and proteases have been previously shown to be linked with migratory and invasive potential of tumors, supporting a more aggressive feature of basal B cell lines. We observed several kinases including PEAK1, ACVR1, and NUAK1 specifically within the basal B subclass, and it is worthwhile to assess their roles in mediating aggressive oncogenic features for exploring newer targets for this subclass. We also found AXL with increased abundance specifically within the basal B subclass. AXL has been previously shown to be associated with TNBC tumorigenesis, particularly in those cancers cells displaying mesenchymal features (Bottai et al., 2016). AXL activates PI3K/AKT and JAK/STAT pathways, resulting in increased cell survival and reduction of immune response, and also mediates EMT and increases invasion (Gay et al., 2017, Goyette et al., 2018, Wilson et al., 2014). With its prominent oncogenic roles, AXL has been proposed as a therapeutic target for TNBC and also for other cancers, and one of its selective inhibitors R428 (BGB324; Bemcentinib) is under clinical trials for acute myeloid leukemia (AML), non-small cell lung cancer (NSCLC) (NCT02488408/NCT02424617) as well as for advanced TNBC and adenocarcinoma of the lung in combination with pembrolizumab (Yule et al., 2018). Despite several studies showing that AXL mediates TNBC tumorigenesis (Holland et al., 2010, Wilson et al., 2014), the possibility of subclass-specific functions or therapeutic implications has not been explored thus far. In fact, in our study, we demonstrated that inhibition of AXL suppresses the migratory and invasive behavior specifically in the basal B subclass and hence, can be an effective target for tumors belonging to this subclass. In comparison to kinases, proteases are far less studied, their repertoire of targets is not well understood, and hence, only a few available specific inhibitors have been exploited in the clinic. With their profound roles in various functional processes, particularly in mediating migration and invasion during tumor progression, there has been a leap in the number of studies focusing on elucidating the mechanistic roles and defining their target substrates. It is therefore not surprising that except for CASP14, none of our observed DE proteases have been reported in a subclass-specific manner, even though some of them including FAP, MMP2, UHCL1, and kallikreins have been reported to be involved in breast cancer tumorigenesis in general (Haritos et al., 2018, Huang et al., 2004, Liu et al., 2019, Talvensaari-Mattila et al., 2003, Yang et al., 2015).

The analysis of subclass-specific kinase-protease-centric signatures and their associated networks represented known functional signatures of each breast cancer type and also revealed several molecular hubs, altered pathways, and signaling crosstalk that may signify intricate molecular rewiring within each type. For the basal B class, the kinase-protease-centric network and signatures displayed functional inclination to a more aggressive phenotype with potential crosstalk between signaling cascades. We found that HIF-1 signaling and hypoxia-driven signatures were more pronounced in basal B as compared with basal A subclass. Hypoxia has been shown to be linked to EMT and cancer progression in many solid tumors (Muz et al., 2015), and its major regulator transcription factor HIF-1α is known to be associated with advanced disease stage and poor clinical outcome in breast cancers (Liu et al., 2015, Yamamoto et al., 2008). Recent studies demonstrated that HIF-1α is particularly overexpressed in TNBC tumors as compared with other subtypes and favors invasive nature (Chen et al., 2014, Ponente et al., 2017). From the basal B kinase-protease network, we inferred that HIF-1α lies within the proximity neighborhood of protein kinase C α (PKCα encoded by PRKCA). PKC isoforms including PKCα have been reported to play crucial roles in regulating the stability, accumulation, and transactivation of HIF-1α during hypoxia (Datta et al., 2004, Lee et al., 2007). PKCα has also been implicated in TNBC aggressiveness, and PKCα inhibitors have been advocated as promising therapeutic agents (Humphries et al., 2014, Lin et al., 2017). With the observed basal B-specific enhanced expression of PKCα, we postulate a robust activation and maintenance of HIF-1 signaling particularly within the basal B TNBC subgroup. Further, HIF-1α stabilization through deubiquitination by UCHL1, a protease we found to be overexpressed within basal B, has also been reported to increase HIF-1 signaling, leading to angiogenesis and distant metastasis (Goto et al., 2015). Taken together, these data suggest HIF-1 signaling to be specifically hyperactivated within the basal B group. The network also revealed transcription factor JUN as one of the key functional hubs within the proximity neighborhood of several altered kinases and proteases including MAPK7, DDR2, EPHA2, and FAP in the basal B subclass. Among breast cancer types, overexpression and activation of JUN transcriptional cascade has been reported especially in TNBC wherein it aggravates proliferation and migratory potential along with triggering a stem cell phenotype (Xie et al., 2017). A potential crosstalk between JUN signaling and EPHA2 receptor is known to promote phenotypic alterations associated with tumorigenesis and acquiring stem-cell-like properties in cancers (Song et al., 2014). With EPHA2 abundantly expressed in basal B, our results support sustained activation of EPHA2 via JUN, which may be conducive for promoting aggressive features within basal B subclass. In addition to pinpointing signaling hubs, we also identified EMT inducer ZEB1 as a key molecular pivot within the basal B kinase-protease network supporting a more invasive and migratory signature for this aggressive subclass (Zhang et al., 2015). Such evidences from our analyses thus add validity to the relevance of unique kinase and protease signatures among different breast cancer subtypes and reinforce their profound influence over key functional processes.

Although the analysis based on differentially regulated proteins, especially kinases and proteases, provided insights on possible functional rewiring of basal A and basal B TNBC subclasses, signaling pathways and networks underlie intricate involvement of several proteins beyond those that are only differentially expressed. There exists a high degree of protein co-regulation and association in various cellular processes, and any imbalance affecting the stoichiometry between proteins or within functional complexes often leads to signaling aberrations making them vulnerable to transformations. We hypothesized that inherent protein-association dysregulations exist within each subtype of breast cancer, and mining for such functional differences at the systemic level will identify subtype-specific molecular aberrations. Indeed, we observed several protein associations that were consistently dysregulated in particular groups of breast cancer subtypes. Although not all dysregulated proteins identified through this analysis were differentially expressed in a subtype-specific manner, the functional differences among the breast cancer subgroups were well represented. The basal B subclass displayed the highest number of dysregulated proteins involved in perturbed associations. Several of these intricately associated proteins are known to play vital roles in invasion and metastatic processes. For example, the specific prediction of MMP14 in the more aggressive basal B group rather than the basal A group is interesting, as MMP14 is implicated in aggressive phenotypes of breast cancer and is involved in mediating EMT, thus contributing to invasion and metastasis (Jiang et al., 2006). We also predicted TGFBR2 and associated TGF signaling to be specifically modulated in the basal B subclass, and indeed TGF-β pathway is well documented in mediating oncogenesis in the mesenchymal-like and claudin-low subgroups of TNBC (Wahdan-Alaswad et al., 2016). Besides triggering EMT mechanisms, TGF-β pathway also enhances cell proliferation and reduces apoptosis (Massagué, 2008, Xu et al., 2018). Although both TGFBR1 and TGFBR2 heterodimerize during transcriptional activation via SMAD signaling (Principe et al., 2014), only TGFBR1 inhibitors are predominant in clinical trials for effectively blocking the pathway. Given the pro-proliferative role of TGFBR2 to activate MAPK and PI3K/AKT pathways independently of TGFBR1 (Bakin et al., 2000, Janda et al., 2002), and the fact that we predict TGFBR2 perturbation within basal B subclass, it is prudent to inhibit TGFBR2 for effective abolishment of downstream oncogenic effects pertaining to this pathway within aggressive tumors belonging to basal B subclass. Indeed, several of the kinases that we observed to vary in abundance within the basal B subclass also have been reported to engage in signaling crosstalk with TGFBR2 and their associated signaling. For instance, PEAK1 has been reported as a molecular switch altering TGFBR2 responses from canonical tumor suppressing to non-canonical tumor promoting, thus driving EMT and invasion (Agajanian et al., 2015a, Agajanian et al., 2015b). For protease FAP, which was both overexpressed and predicted to be dysregulated within basal B, we observed perturbed associations with several EMT-associated proteins such as BGN, CDH2, TAGLN, COL6A3, and VCAN (Chen et al., 2019, Hu et al., 2014, Huang et al., 2018, Mrozik et al., 2018, Sheng et al., 2006). Such dysregulated associations of FAP strongly argue for an important functional role specific to basal B cells; however, little is known about its actual in vivo targets and tumorigenic mechanisms (Hamson et al., 2014). All these assert that protein associations and co-regulations are critical determinants in defining cellular mechanisms and functional alterations. Although some of the dysregulated kinases and proteases including AXL, EPHA2, MMP2/14, and FAP have previously been shown to be possible targets for TNBC, they have not been studied in a subclass-specific manner. While some of these are currently being explored for TNBC therapy, our analyses suggest that targeting these proteins could prove more effective in a particular subclass rather than in TNBC in a broader context. In addition to these, we also unraveled several other novel kinases and proteases that have the potential to be exploited as TNBC subclass-specific druggable targets.

Although the molecular heterogeneity of TNBC is well documented, success with regard to clinical intervention has been disappointing. Despite several studies reporting varied expression patterns of genes and proteins within TNBC, a thorough investigation from a therapeutic perspective to unravel the complexity has been lacking. Our systemic and systematic workflow, with emphasis on protein association dysregulations, opens up new avenues for understanding molecular perturbations at the subtype level and also extracts subclass-specific therapeutic targets for strategized clinical applications. The candidates identified in our study are now at secondary validation stage, where tumor screening for target verification and in vivo biological studies should be performed. At the same time, establishment of robust subclass-specific biomarkers is mandatory for patient stratification for successful targeted treatments. Here, our analysis has been carried out focusing only on two major TNBC subclasses. Nevertheless, we postulate that there could be more than two subclasses within TNBC with different functional signatures. For this, deep proteome profiling of all available TNBC cells as well as tumors, followed by systematic analysis herein reported is required. Altogether, our study uncovers molecular mechanisms within TNBC subclasses and thus holds potential to enhance applications of personalized medicine for TNBC.

Limitations of the Study

This study revealed distinct proteome-based subclasses within TNBC that are functionally discrete and highlighted mechanisms and potential targets unique to the subclasses opening new avenues for further functional explorations. However, certain limitations and challenges remain to be addressed. First, the repertoire of TNBC cell lines profiled by this study is limited and may not represent the complex heterogeneity underlying TNBC. Extensive proteomic screening of a wide array of breast cancer cell lines that represents the entirety of TNBC complexity is likely to identify additional subclasses or better resolve existing subclasses with distinctive functional signatures. These subclasses may refine our current understanding of transcriptome-derived TNBC clusters and offer functional insights on inherent molecular specificities within TNBC. An additional drawback of this study is the use of cell lines for proteomic profiling and validation. Existing cell line models are homogeneous and lack components of the tumor such as infiltrating cells, macrophages, blood vessels, and ECM that represent the tumor microenvironment and hence are not perfect in mimicking actual tumor conditions. Functional interplay between tumors and its environment is crucial in defining oncogenic progression, and such complexities are not considered in the current cell-line-based proteome profiling. Yet, our study shows feasibility of unraveling TNBC complexity using proteomic approaches and extension of such investigations to profile tumors derived from large patient cohorts will benefit from patient stratification based on functional subclasses that hold potential for improved TNBC-targeted therapies. With emphasis laid on subclass-specific roles of kinases and proteases based on our proteomic characterization, more efforts are needed in investigating their oncogenic roles within TNBC subclasses to explore their clinical utility for personalized medicine. This is especially important for targets that are currently under clinical trial where knowledge on subclass specificity will favor successful clinical outcomes.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.

Published: February 21, 2020

Data and Code Availability

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (Perez-Riverol et al., 2019) partner repository with the dataset identifier PXD017025.

Supplemental Information

Table S1. Quantitative Proteome Profiles of Breast Cancer Cell Line Panel, Related to Figures 1, 2, and S1

(A) Normalized median log₂ fold-change values.

(B) Normalized protein abundance (raw Proteome Discoverer 2.2 output).

(C) Experimental design and cell line classification.

Table S2. Functional Processes and Pathways Enriched among Differentially Expressed Proteins within Different Breast Cancer Groups, Related to Figures 4 and S4

(A) Hallmark cancer gene set enrichment of differentially expressed proteins.

(B) KEGG pathway enrichment of kinase-protease network proximity proteins from different breast cancer groups.

Table S3. Protein Association Co-regulations from Breast Cancer Proteome Profiles, Related to Figures 5 and S5

(A) Positively correlating protein pairs inferred from breast cancer cell line proteome.

(B) Negatively correlating protein pairs inferred from breast cancer cell line proteome.

Table S4. Dysregulated Protein Association Perturbations in Breast Cancer Cell Lines, Related to Figures 5 and S5

(A) Dysregulated protein association perturbations based on positive co-regulations.

(B) Dysregulated protein association perturbations based on negative co-regulations.

Table S5. Enrichment of Dysregulated Proteins within Different Breast Cancer Subtypes, Related to Figures 5, S5, and S6

(A) Enrichment of dysregulated proteins inferred from positive co-regulations.

(B) Enrichment of dysregulated proteins inferred from negative co-regulations.

(C) Enrichment of dysregulated protein pairs inferred from positive co-regulations.

(D) Enrichment of dysregulated protein pairs inferred from negative co-regulations.