Expression of endoplasmic reticulum stress proteins is a candidate marker of brain metastasis in both ErbB-2+ and ErbB-2- primary breast tumors.

The increasing incidence of breast cancer brain metastasis in patients with otherwise well-controlled systemic cancer is a key challenge in cancer research. It is necessary to understand the properties of brain-tropic tumor cells to identify patients at risk for brain metastasis. Here we attempt to identify functional phenotypes that might enhance brain metastasis. To obtain an accurate classification of brain metastasis proteins, we mapped organ-specific brain metastasis gene expression signatures onto an experimental protein-protein interaction network based on brain metastatic cells. Thirty-seven proteins were differentially expressed between brain metastases and non-brain metastases. Analysis of metastatic tissues, the use of bioinformatic approaches, and the characterization of protein expression in tumors with or without metastasis identified candidate markers. A multivariate analysis based on stepwise logistic regression revealed GRP94, FN14, and inhibin as the best combination to discriminate between brain and non-brain metastases (ROC AUC = 0.85, 95% CI = 0.73 to 0.96 for the combination of the three proteins). These markers substantially improve the discrimination of brain metastasis compared with ErbB-2 alone (AUC = 0.76, 95% CI = 0.60 to 0.93). Furthermore, GRP94 was a better negative marker (LR = 0.16) than ErbB-2 (LR = 0.42). We conclude that, in breast carcinomas, certain proteins associated with the endoplasmic reticulum stress phenotype are candidate markers of brain metastasis.

Biana: a software framework for compiling biological interactions and analyzing networks.

BACKGROUND: The analysis and usage of biological data is hindered by the spread of information across multiple repositories and the difficulties posed by different nomenclature systems and storage formats. In particular, there is an important need for data unification in the study and use of protein-protein interactions. Without good integration strategies, it is difficult to analyze the whole set of available data and its properties. RESULTS: We introduce BIANA (Biologic Interactions and Network Analysis), a tool for biological information integration and network management. BIANA is a Python framework designed to achieve two major goals: i) the integration of multiple sources of biological information, including biological entities and their relationships, and ii) the management of biological information as a network where entities are nodes and relationships are edges. Moreover, BIANA uses properties of proteins and genes to infer latent biomolecular relationships by transferring edges to entities sharing similar properties. BIANA is also provided as a plugin for Cytoscape, which allows users to visualize and interactively manage the data. A web interface to BIANA providing basic functionalities is also available. The software can be downloaded under GNU GPL license from http://sbi.imim.es/web/BIANA.php. CONCLUSIONS: BIANA's approach to data unification solves many of the nomenclature issues common to systems dealing with biological data. BIANA can easily be extended to handle new specific data repositories and new specific data types. The unification protocol allows BIANA to be a flexible tool suitable for different user requirements: non-expert users can use a suggested unification protocol while expert users can define their own specific unification rules.

ModLink+: improving fold recognition by using protein-protein interactions.

MOTIVATION: Several strategies have been developed to predict the fold of a target protein sequence, most of which are based on aligning the target sequence to other sequences of known structure. Previously, we demonstrated that the consideration of protein-protein interactions significantly increases the accuracy of fold assignment compared with PSI-BLAST sequence comparisons. A drawback of our method was the low number of proteins to which a fold could be assigned. Here, we present an improved version of the method that addresses this limitation. We also compare our method to other state-of-the-art fold assignment methodologies. RESULTS: Our approach (ModLink+) has been tested on 3716 proteins with domain folds classified in the Structural Classification Of Proteins (SCOP) as well as known interacting partners in the Database of Interacting Proteins (DIP). For this test set, the ratio of success [positive predictive value (PPV)] on fold assignment increases from 75% for PSI-BLAST, 83% for HHSearch and 81% for PRC to >90% for ModLink+at the e-value cutoff of 10(-3). Under this e-value, ModLink+can assign a fold to 30-45% of the proteins in the test set, while our previous method could cover <25%. When applied to 6384 proteins with unknown fold in the yeast proteome, ModLink+combined with PSI-BLAST assigns a fold for domains in 3738 proteins, while PSI-BLAST alone covers only 2122 proteins, HHSearch 2969 and PRC 2826 proteins, using a threshold e-value that would represent a PPV >82% for each method in the test set. AVAILABILITY: The ModLink+server is freely accessible in the World Wide Web at http://sbi.imim.es/modlink/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Functional clustering of metastasis proteins describes plastic adaptation resources of breast-cancer cells to new microenvironments.

To examine the molecular mechanisms underlying breast cancer metastasis in liver and search for potential markers of metastatic progression in soft-tissue, we analyzed metastatic variants developed from the highly metastatic MDA-MB 435 cell line through in vivo stepwise selection in the athymic mice. Comparative proteomic analysis using two-dimensional electrophoresis (2DE-DIGE) revealed that 74 protein spots were reproducibly more than doubled in liver metastatic cells compared to parental counterpart. From 22 proteins identified by MALDI-TOF, belonging to intermediate filaments, intracellular transport and ATP synthesis, we generated a protein-protein interaction network containing 496 nodes, 12 of which interacted. GRP 75 was connected with four other proteins: prohibitin, HSP 27, elongin B and macropain delta chain. After functional classification, we found that pathways including hepatocyte growth factor receptor (p = 0.014), platelet-derived growth factor (p = 0.018), vascular endothelial growth factor (p = 0.021) and epidermal growth factor (p = 0.050) were predominant in liver metastatic cells, but not in lung metastatic cells. In conclusion, we suggest that GRP 75 is involved in cell proliferation, tumorigenesis and stress response in metastatic cells by recruiting signals in which the transmembrane receptor protein tyrosine kinase signaling pathway (p-value FDR = 1.71 x 10(-2)) and protein amino acid phosphorylation (p-value FDR = 3.28 x 10(-2)) might be the most significant biological process differentially increased in liver metastasis.

Predicting cancer involvement of genes from heterogeneous data.

BACKGROUND: Systematic approaches for identifying proteins involved in different types of cancer are needed. Experimental techniques such as microarrays are being used to characterize cancer, but validating their results can be a laborious task. Computational approaches are used to prioritize between genes putatively involved in cancer, usually based on further analyzing experimental data. RESULTS: We implemented a systematic method using the PIANA software that predicts cancer involvement of genes by integrating heterogeneous datasets. Specifically, we produced lists of genes likely to be involved in cancer by relying on: (i) protein-protein interactions; (ii) differential expression data; and (iii) structural and functional properties of cancer genes. The integrative approach that combines multiple sources of data obtained positive predictive values ranging from 23% (on a list of 811 genes) to 73% (on a list of 22 genes), outperforming the use of any of the data sources alone. We analyze a list of 20 cancer gene predictions, finding that most of them have been recently linked to cancer in literature. CONCLUSION: Our approach to identifying and prioritizing candidate cancer genes can be used to produce lists of genes likely to be involved in cancer. Our results suggest that differential expression studies yielding high numbers of candidate cancer genes can be filtered using protein interaction networks.

Biological pathways contributing to organ-specific phenotype of brain metastatic cells.

Secondary to the increased survival following chemotherapy, brain metastases have recently become a significant clinical problem for breast cancer patients. The aim of this study was to characterize those functional phenotypes that might enhance brain metastasis in breast cancer cells. We first analyzed by two-dimensional electrophoresis (2DE-DIGE) differences in protein expression between parental MDA-MB 435 cells and the brain metastatic variant 435-Br1, obtaining 19 identified proteins by peptide mass fingerprinting, 11 under-expressed (<2-fold) and 8 overexpressed (>2-fold) in 435-Br1. We created and analyzed protein interaction networks with a bioinformatic program (PIANA) from protein data, and it allowed us to associate 34/67-laminin receptor functionally with HSP 27, through a chaperone glucose-regulated protein GRP 94. Moreover, HSP 27 had the largest amount of direct and indirect protein interactions, forming a cluster of chaperones and cochaperones, associated through kinases to a set of intermediated filament proteins. In addition, functional groups of proteins identified were peptidase, DNA binding transcription factors, ATP synthase complex, anion transporters, and carbohydrate metabolism. Further functional analyses in cells, expression analyses in experimental tissues, and in human brain metastasis were addressed to validate the biological pathways contributing to organ-specific phenotype of brain metastasis.

Functional pathways shared by liver and lung metastases: a mitochondrial chaperone machine is up-regulated in soft-tissue breast cancer metastasis.

Genes that mediate breast cancer metastasis to lung are different from those which mediate bone metastasis. However, which markers accounts for the diversity of breast cancer metastasis remains unknown. The aim of this study was identify proteins associated with the soft-tissue metastatic ability of breast cancer tumors in metastases, coupling microarray data from clinical metastases and immunohistochemistry, for further screening for early detection at the first diagnosis in patients. We use a bioinformatic program to create and analyze protein interaction networks from protein experimental data, and to translate RNA expression analysis of breast cancer human metastases to protein, in a search for the phenotype associated with soft-tissue metastases. The pre-validated proteins constituted the protein signature for each metastasis: 37 (8.9%) from liver, 92 (8.5%) from lung and 167 (13%) from bone. Pleiotrophin, BAG 2, HSP 60 and vinculin were pre-validated in liver and lung metastases performing the soft-tissue phenotype. After IHC validation, we conclude that HSP 60, one of the best-known mitochondrial chaperone machines, is a key protein in soft-tissue metastases phenotype interacting with BAG 2, which competes for binding to GRP 75, the other mitochondrial chaperone. The relationship between HSP 60/GRP 75 and BAG 2 might result in the activation of several transcription pathways, different in liver from in lung metastases, as a nodal point coupling positive and negative actuators in the multiple survival-signal pathways and so achieving metastatic growth.

Characterization of protein hubs by inferring interacting motifs from protein interactions.

The characterization of protein interactions is essential for understanding biological systems. While genome-scale methods are available for identifying interacting proteins, they do not pinpoint the interacting motifs (e.g., a domain, sequence segments, a binding site, or a set of residues). Here, we develop and apply a method for delineating the interacting motifs of hub proteins (i.e., highly connected proteins). The method relies on the observation that proteins with common interaction partners tend to interact with these partners through a common interacting motif. The sole input for the method are binary protein interactions; neither sequence nor structure information is needed. The approach is evaluated by comparing the inferred interacting motifs with domain families defined for 368 proteins in the Structural Classification of Proteins (SCOP). The positive predictive value of the method for detecting proteins with common SCOP families is 75% at sensitivity of 10%. Most of the inferred interacting motifs were significantly associated with sequence patterns, which could be responsible for the common interactions. We find that yeast hubs with multiple interacting motifs are more likely to be essential than hubs with one or two interacting motifs, thus rationalizing the previously observed correlation between essentiality and the number of interacting partners of a protein. We also find that yeast hubs with multiple interacting motifs evolve slower than the average protein, contrary to the hubs with one or two interacting motifs. The proposed method will help us discover unknown interacting motifs and provide biological insights about protein hubs and their roles in interaction networks.

PIANA: protein interactions and network analysis.

UNLABELLED: We present a software framework and tool called Protein Interactions And Network Analysis (PIANA) that facilitates working with protein interaction networks by (1) integrating data from multiple sources, (2) providing a library that handles graph-related tasks and (3) automating the analysis of protein-protein interaction networks. PIANA can also be used as a stand-alone application to create protein interaction networks and perform tasks such as predicting protein interactions and helping to identify spots in a 2D electrophoresis gel. AVAILABILITY: PIANA is under the GNU GPL. Source code, database and detailed documentation may be freely downloaded from http://sbi.imim.es/piana.

Underexpression of transcriptional regulators is common in metastatic breast cancer cells overexpressing Bcl-xL.

Bcl-xL gene induces metastasis in the lung, lymph nodes and bone when breast cancer cells are inoculated in Nude Balb/c mice. In an attempt to identify the molecules required for diverse metastatic foci, we compared gene expression levels in tumor cells and metastatic variants with a cDNA GeneFilter containing 4000 known genes. The transcriptional regulators of alpha1-fetoprotein transcription factor, TBP-associated factor 172 (TAF-172) and the human zinc finger protein 5 (ZFP5) were downregulated. The expression of TAF-172 was inversely proportional to Bcl-xL expression (ANOVA P < 0.0001) and metastatic activity (ANOVA P < 0.0001). A protein interaction program allowed us to functionally associate Bcl-xL and TAF through TATA-binding protein (TBP), suggesting that Bcl-xL connects metabolic pathways with transcriptional machinery. The prediction included proteins involved in apoptosis, electron transfer, kinases and transcription factors. These results indicate that the selection of diverse metastatic cells from the broad spectrum of tumor cell leads to the underexpression of certain transcriptional regulators that might act as adaptor molecules to different microenvironments, and indicate that the synergistic activity of several genes is needed for the selection process in several metastatic foci.

Bcl-x(L)-mediated changes in metabolic pathways of breast cancer cells: from survival in the blood stream to organ-specific metastasis.

Bcl-x(L) protein plays a role in breast cancer dormancy, promoting survival of cells in metastatic foci by counteracting the proapoptotic signals in the microenvironment. The aim of this study was to identify phenotypes mediated by Bcl-x(L) in breast cancer cells that enhance in vivo survival of clinical metastases. 435/Bcl-x(L) or 435/Neo human breast cancer cells were injected into the inguinal mammary gland of nude mice, and tumors, metastases in lymph node, lung, and bone, and bloodstream surviving cells were examined. Proteomic analysis identified 17 proteins that were overexpressed (more than twofold) or underexpressed (less than twofold) in metastases. A protein interaction program allowed us to functionally associate peroxiredoxin 3, peroxiredoxin 2, carbonyl reductase 3, and enolase 1, suggesting a role for cellular responses to oxidative stress in metastasis organ selection. The prediction included proteins involved in redox systems, kinase pathways, and the ATP synthase complex. Furthermore, the interaction of redox proteins with enolase 1 suggests a connection between glycolysis and antioxidant pathways, enabling achievement of a high metastatic activity. In conclusion, Bcl-x(L) mediates a phenotype in which redox pathways and glycolysis are coupled to protect breast cancer metastatic cells during transit from the primary tumor to the metastatic state.

Detecting remotely related proteins by their interactions and sequence similarity.

The function of an uncharacterized protein is usually inferred either from its homology to, or its interactions with, characterized proteins. Here, we use both sequence similarity and protein interactions to identify relationships between remotely related protein sequences. We rely on the fact that homologous sequences share similar interactions, and, therefore, the set of interacting partners of the partners of a given protein is enriched by its homologs. The approach was bench-marked by assigning the fold and functional family to test sequences of known structure. Specifically, we relied on 1,434 proteins with known folds, as defined in the Structural Classification of Proteins (SCOP) database, and with known interacting partners, as defined in the Database of Interacting Proteins (DIP). For this subset, the specificity of fold assignment was increased from 54% for position-specific iterative BLAST to 75% for our approach, with a concomitant increase in sensitivity for a few percentage points. Similarly, the specificity of family assignment at the e-value threshold of 10(-8) was increased from 70% to 87%. The proposed method would be a useful tool for large-scale automated discovery of remote relationships between protein sequences, given its unique reliance on sequence similarity and protein-protein interactions.