A Transcriptome-proteome Integrated Network Identifies Endoplasmic Reticulum thiol oxidoreductase (ERp57) as a Hub that Mediates Bone Metastasis

Abstract

Bone metastasis is the most common distant relapse in breast cancer. The identification of key proteins involved in the osteotropic phenotype would represent a major step toward the development of new prognostic markers and therapeutic improvements. The aim of this study was to characterize functional phenotypes that favor bone metastasis in human breast cancer. We used the human breast cancer cell line MDA-MB-231 and its osteotropic BO2 subclone to identify crucial proteins in bone metastatic growth. We identified 31 proteins, 15 underexpressed and 16 overexpressed, in BO2 cells compared with parental cells. We employed a network-modeling approach in which these 31 candidate proteins were prioritized with respect to their potential in metastasis formation, based on the topology of the protein-protein interaction network and differential expression. The protein-protein interaction network provided a framework to study the functional relationships between biological molecules by attributing functions to genes whose functions had not been characterized. The combination of expression profiles and protein interactions revealed an endoplasmic reticulum-thiol oxidoreductase, ERp57, functioning as a hub that retained four down-regulated nodes involved in antigen presentation associated with the human major histocompatibility complex class I molecules, including HLA-A, HLA-B, HLA-E, and HLA-F. Further analysis of the interaction network revealed an inverse correlation between ERp57 and vimentin, which influences cytoskeleton reorganization. Moreover, knockdown of ERp57 in BO2 cells confirmed its bone organ-specific prometastatic role. Altogether, ERp57 appears as a multifunctional chaperone that can regulate diverse biological processes to maintain the homeostasis of breast cancer cells and promote the development of bone metastasis.

Large-scale genomic analysis has provided a wealth of information on biologically relevant systems, and the ability to analyze this information is crucial to uncovering important biological relationships. In breast cancer, microarray gene expression analysis is a promising technique for providing consistent patterns of variation in bone metastasis gene expression; the most common metastasis (80%) in those women who progress to an advanced stage of disease (1–4). However, a large number of genes with many diverse functions are identified as prognostic markers, without revealing much about the underlying biological mechanism.

Genes that enhance or suppress bone metastasis are associated with multiple cellular processes that normally occur during metastasis progression, including survival and proliferation in the bone marrow microenvironment, and modification of bone structure and function (1). Many genes in this group encode secretory or cell surface proteins involved in cell homing to bone, angiogenesis, invasion, and osteoclast recruitment (1, 2, 5). Moreover, emerging evidence from murine models suggests that tumor-specific endocrine factors systematically stimulate the quiescent bone marrow compartment (BM), resulting in a BM-derived tumor microenvironment that promotes metastasis initiation (6). Although genes associated with bone metastasis can readily be identified by screening techniques, their validation and characterization require sophisticated animal models that closely mirror the pathophysiology of bone metastasis in humans (7, 8).

Animal models have successfully been used to select variants of cell or tumor lines that have an increased incidence of metastasis to bone (9). Cells with a bone metastatic gene profile are present in the parental population and become selected in vivo as highly metastatic entities. The identification of key proteins involved in the osteotropic phenotype would represent a major step toward the development of new prognostic markers and therapeutic improvements.

Large protein-protein interaction networks are now available, thanks to the recent explosion of high-throughput experimental technologies for characterizing protein interactions between thousands of proteins (10). These networks provide a way to relate genome-wide expression profiles to function (11–13). Protein-protein interaction networks are modeled as undirected graphs in which the nodes represent proteins and the links represent the physical interactions between proteins (14). By revealing the context of a given protein in the interaction network, the systems-level view can yield useful insights into molecular and cell function (15). These cellular network models are obtained through a combination of mRNA expression profiles and curated protein-protein interaction data, which have recently become abundant (16). Identifying subnetworks induced in a certain phenotype using such models can facilitate biological validation (17). Considering that a systems-level study of the mechanisms underlying breast cancer bone metastasis and organ-specificity may improve our understanding of the biology of secondary tumors, here we attempt to characterize organ-specific protein taxonomies of bone metastatic breast cancer cells. We improved the transcriptomic information using a complementary strategy based on integration of expression profiles with protein interactions (14). An initial two-dimensional differential in-gel electrophoresis assessed the distinct expression of proteins in MDA-MB-231 (231) and its bone metastatic variant, BO2 cells (7, 15). To describe the protein-protein interaction network (PPIN) in breast cancer cells that metastasize in bone, we used bioinformatics tools such as the Biological Interactions and Network Analysis (BIANA)¹ (18) and GUILD (19) in combination with data from proteomic and transcriptomic analysis. BIANA creates and analyzes biological networks and GUILD prioritizes the biomolecules in the network according to their relevance to a given phenotype. We enriched the network created by BIANA with the BO2 transcriptomic profiling performed previously, which revealed the characteristic osteoblast-like phenotype compatible with an osteomimetic phenotype (4) and prioritized the proteins in the network for bone metastatic breast cancer phenotype using GUILD. Analysis of the prioritized bone metastatic breast cancer subnetwork (active subnetwork) and biological validation showed 4 down-regulated nodes involved in antigen presentation, associated with human leukocyte antigen class I (HLA class I) molecules and linked with ERp57: an endoplasmic reticulum (ER) thiol oxidoreductase. We also found a direct link between ERp57 and cytoskeleton proteins such as vimentin. The aim of our work is to find the cause-effect relevance of ERp57 in bone metastasis.

MATERIALS AND METHODS

Cellular Models

The MDA-MB-231 (231) human breast cancer cell line was obtained from the European Type Culture Collection (ECACC 92020424). The BO2 cell line was established from bone metastases caused by 231. This subclone of 231 was selected after six in vivo passages in nude mice using a heart injection model. BO2 cells are characterized by their unique predilection for bone metastasis (5, 20). Both cell lines were routinely cultured in DMEM/F12 medium (Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum at 37 °C in a humidified 5% CO₂ incubator. MDA-MB 435 (435-P) breast cancer cells and their metastatic variants to brain (435-Br1), lung (435-L3, 435-L5), bone (435-B1), and liver (435-Lv1) obtained from successive in vivo/in vitro passages were also used for comparative purposes in some experiments (21). Cells maintained in standard culture conditions were used to generate orthotopic primary tumors in 7-week-old athymic nude Balb/c female mice by inoculation of 10⁶ 435-P cells in 0.05 ml of medium without serum in the right inguinal mammary gland (i.m.f.p.). Metastasis was tested for in paraffin sections by hematoxylin-eosin stain.

In the functional experiments, 15 × 10⁴ 231 cells were seeded in six-well plates with coverslips. After 6 h, the cells were starved of glucose (0 or 4.5 mg/ml) using DMEM without glucose (Invitrogen) supplemented with 10% SBF and l-Gln 2 mm and left for 48 h.

Proteomic Analysis

Differential Expression of Proteins

Whole-cell extracts of 50 μg were labeled with 400 pmol of Cy3 or Cy5 (EttanTM DIGE, Amersham Biosciences AB, Uppsala, Sweden) and analyzed by two-dimensional gel electrophoresis (2DE). Three gels from three independent experiments were scanned with a Typhoon Variable Mode Imager 9.400 (Amersham Biosciences AB). The digitalized 2DE images were standardized and compared with the DeCyder Differential Analysis Software program, version 6.0 (Amersham Biosciences AB) according to the manufacturer's recommendation using the DeCyder biological variation analysis module. An average of 1780 spots were visualized in the 2DE from three independent experiments. Of these, 821 spots were further analyzed to assess the differential expression of proteins between parental cells (231) and bone metastatic variant BO2 cells, and simultaneously to match all 56 protein spot maps from three experiments. Student's t test was used for statistical analysis of the protein spot data and a p value of <0.05 was taken to indicate that a protein spot was significantly different across the three experiments. The ratios were standardized in line with: Ri = log10 (V2i/V1i), where the spot volumes (Vi) were in gel sample 1 (V1i) and gel sample 2 (V2i). Ratios were expressed within the range of a twofold increase and decrease (2 and -2), where a log2ratio of 1 represents twofold up-regulation and -1 represents a twofold down-regulation. The selected spots were excised from 2DE silver-stained gels and subjected to trypsin in-gel digestion. The supernatants containing tryptic peptides were purified in a Poros R2 column (Applied Biosystems, Framingham, MA, USA) packed in a constricted Gel-Loader Tip (Eppendorf, Hamburg, Germany). Peptides were eluted with 15–25 μl of 70% acetonitrile, 0.1% formic acid.

For matrix assisted laser desorption ionization-time of flight (MALDI-TOF) MS, 2.5 μl of tryptic digest was deposited onto Mass Spec Turbo 192 type 1 peptide chips prespotted with CHCA (Qiagen, Venlo, Netherlands). Each spot was then washed with finishing solution (Qiagen). Peptide mass fingerprinting spectra were recorded on a Voyager short tandem repeat MALDI-TOF (Applied Biosystems) using a 337 nm nitrogen laser operated in the reflector mode, with an accelerating voltage of 20 kV. Samples were analyzed in the m/z 800–3000 range and were calibrated externally with the Sequazyme Peptide Mass Standards kit (Applied Biosystems). Peptides from trypsin autolysis were used for the internal calibration. Peak lists were generated with the MoverZ software (Knexus edition, Genomic Solutions, Ann Arbor, MI, USA). Monoisotopic peaks with S/N ≥ 3 were selected and common contaminants were eliminated using the Peak Erazor software (v 2.0.1; Lighthouse Data, Odense, Denmark). Annotated spectra for each protein were uploaded at the Peptide Atlas - PASSEL database (dataset identifier PASS00228).

Samples for liquid chromatography-tandem MS (LC-MS/MS) were run on a Q-Star Pulsar (Applied Biosystems) instrument fitted with a nanoESI source, after separation by nanoLC in an Ultimate II system (LC Packings, Bannockburn, US). Peptides were separated by reverse phase chromatography (RP Atlantis dC18 NanoEase Column, 75 μm × 150 mm, Waters, Milford, MA, USA), using a linear 5–50% acetonitrile gradient into 0.1% formic acid over 30 min. An electrospray voltage of 2400 V was used. Samples were analyzed in the information-dependent acquisition mode with one full scan followed by three MS/MS scans of the three most intense peaks. Peak lists were generated from the raw data using the Analyst QS 1.1 software (Applied Biosystems), with the MASCOT script 1.6 b21 (MatrixScience, Boston, MA, USA). MS/MS data were centroided and deisotoped, and only precursor charge states of +2, +3, and +4 were considered.

Protein identification from either the MALDI-TOF or the LC-MS/MS results was performed with an in-house MASCOT server v 1.3 (Matrix Science, Boston, MA, USA), using the human proteins from NCBI (January 2013, 243695 sequences). For database searches, Cys carbamidomethylation and one missed cleavage, plus methionine oxidation were allowed as fixed and variable modifications, respectively. Mass tolerance was set at 50 ppm for MALDI-TOF-derived searches; whereas for LC-MS/MS based searches, tolerance was set at 0.15 Da for both MS and MS/MS data. In all cases, a significant MASCOT probability score (p ≤ 0.05) was used as the condition for positive identification.

Proteomics analyses were performed at the Proteomics Unit of the Universitat Pompeu Fabra (UPF) and Center of Genomic Regulation (CRG).

Protein-Protein Interaction Data

We used BIANA (18), a bioinformatics tool for biological data integration, to combine human protein-protein interaction data from various interaction databases. BIANA is a framework that integrates interaction data from multiple databases into a single database and maps identifiers (i.e. cross-references) in different databases to each other. Using BIANA, we integrated data from the following databases: DIP, MIPS, HPRD, BIND, IntAct, MINT, and BioGRID (all downloaded in May, 2011). The entries in these databases were integrated on the basis of how common the following features were: UniProt Accession, amino acid sequence and Entrez Gene Identifier. That is, two proteins annotated with the same identifier (one of the three features mentioned above) in two distinct databases were considered the same protein.

Prioritization of Bone Metastasis Candidates Using Protein-Protein Interactions

First, a PPIN of the whole human-interactome was built around a set of proteins that are potentially crucial for bone metastatic growth, as identified by proteomics analysis (i.e. root proteins). All the interactions of the root proteins were retrieved from BIANA, provided that the interaction was identified with an experimental method other than a pull-down method. Though suitable for defining protein complexes, interactions from pull-down methods might introduce spurious binary interactions between proteins. Then, gene expression levels were mapped onto the network. A protein was considered to be differentially expressed if the gene encoding for it was differentially expressed in the microarray experiment. This mapping allowed us to find the active subnetworks: clusters of the network with a significant proportion of proteins produced by up-regulated or down-regulated genes. Next, we calculated a bone metastasis likelihood score for all the nodes in the human interactome using GUILD (19), which is network-based disease-gene prioritization software (although only the scores of the PPIN subnetwork were used). GUILD assigns a disease-implication score to each node in the network by disseminating information about roots (potential metastatic growth proteins identified by proteomics analysis) to other nodes through the links in the network. We used 28 out of the 31 root proteins for which we found interactions in the human interactome. We applied the NetCombo algorithm implemented in GUILD, which produces a consensus score that incorporates the distance from the nodes to the roots in the network. We first ranked the root proteins with respect to the number of differentially expressed genes in their neighborhood. Next, we ranked the root proteins with respect to the average NetCombo score of their neighbors in the network. We then combined these rankings to produce an ordered list of bone metastatic candidates. The ranks were obtained using the rank function of R software (http://www.r-project.org/).

We applied a Monte Carlo (22, 23) permutation test to verify the significance of matching proteins of differentially expressed genes in the neighborhood of root proteins. We calculated an empirical p value by sampling random sets of 28 proteins from the whole human interactome. We then counted the number of products of differentially expressed genes in the neighborhood of these proteins. We repeated this analysis 1,000,000 times and computed the p value as the ratio of the times we had at least as many proteins of differentially expressed genes in the neighborhood as in the original network.

Bone Metastatic Cell Transcriptome Data (BMCT)

The protein-network-based approach to identifying markers of bone metastasis was performed on the results of the previously analyzed microarray hybridization, using the standard Affymetrix protocol (4). The probes were hybridized in duplicate to the Human Genome U133 (HG-U133) set, consisting of two GeneChip® arrays (U133A and B).

The final data set consisted of a total of 50 scan files. Each one was obtained using the Affymetrix GeneChip® software, which assigns an intensity that represents a measure of the corresponding transcript abundance to each qualifier in the file. The scan files were concatenated into a single file with duplicates forming adjacent columns and qualifiers forming the rows. Finally, replicates were combined by computing the median of the replicate intensities. The output files were further processed into a format that adds an estimate of the standard deviation of the noise for each intensity (4). Only the genes with the most significant variation (twofold regulation, p value <0.05) were retained in the analysis using profile filtering.

We did not make any adjustments by multiple testing because the protein network itself was used to filter genes. We assumed that those genes that match a node in the PPIN were possibly real positive genes involved in bone metastatic processes.

Protein Expression

Western Blotting

The 231 and BO2 cells were lysed in a 1% SDS (v/v) extraction buffer containing an antiprotease mixture (Roche, Vilvoorde, Belgium). Protein concentrations were determined using the Bradford assay (MicroBCA, Pierce, Belgium). After resolution by SDS-PAGE, electrophoresed proteins were transferred to polyvinylidene fluoride membranes that were blocked and probed with each antibody (Table I) and the corresponding peroxidase-conjugated secondary antibody.

Table I. Antibodies used in each kind of analysis. * undiluted hibridoma supernatant.

Antibody	Company	IF/IHCa	FCa	WBa
ERp57	ACris, Germany	1/400, 1/1000	1/300	1/2000
HLA I, clone MEM-147	ExBio, Praha, Czech Republic	1/200	1/200	1/100
Alpha-tubulin, clone B-5–1-2,	Sigma, Saint Louis, Missouri			1/10000
Beta-tubulin, clone H-235	Santa Cruz Biotechnology, Santa Cruz, CA			1/500
Beta-actin	Abcam			1/50000
HSP60, clone LK1	Abcam, Cambridge, UK	1/3000		1/200
Cathepsin D, clone C-20	Santa Cruz Biotechnology			1/500
Galectin-1, clone N-16	Santa Cruz Biotechnology	1/2000		1/100
Hsp27, clone C-20	Santa Cruz Biotechnology			1/500
Peroxiredoxin 2, clone LF-PA0007	Lab Frontier, Seoul, Korea			1/1000
Vimentin, clone RV 202	Abcam			1/500
Stathmin 1, C-terminus	Sigma, Saint Louis, Missouri	1/500		1/500
CD3 PE, clone UCHT1	BD Biosciences		20 μl/test
CD56 FITC, clone MEM-188	BioLegend, San Diego, CA		20 μl/test
PCNA, clone PC10	Santa Cruz Biotechnology	1/50
Caspase 3	Cell Signaling	1/200
W6/32, anti HLA-ABC	(72)	*
l-368, anti β2m	(73, 74)	*
1082C5, anti HLA-A locus	(75)	*
42-IB5, anti HLA-B locus	(76)	*
HC-10, anti HLA-BC free heavy chain	(77)	*
SRF-8, anti HLA-Bw6	(78)	*
Peroxidase conjugated anti-rabbit secondary Ab	Amersham			1/2000
Peroxidase conjugated Antimouse secondary Ab	Pierce, Perbio Science Ltd., Cheshire, U.K			1/2000
Alexa Fluor® 555 anti-rabbit	Invitrogen	1/1000
Alexa Fluor® 488 anti-mouse	Invitrogen	1/500	1/500
Alexa Fluor® 647 anti-rabbit	Invitrogen		1/2000

^a IF, immunofluorescence; IHC, immunohistochemistry; FC, flow cytometry; WB, western blot.

Immunoreactive bands were viewed on a VersaDoc™ (Bio-Rad) Imaging System using the Super Signal west-Pico (Pierce). MWs were established with See Blue Plus2 prestained Standford (Invitrogen, San Diego, CA, USA).

Immunofluorescence Analysis

The 231 and BO2 cells were analyzed for the expression of ERp57 and HLA class I molecules under basal conditions or after ERp57 knockdown. Briefly, 15 × 10⁴ or 6 × 10⁴ cells were seeded in 6- or 24-well plates with coverslips. After 24 h of transfection, immunofluorescence was performed by adapting the technique in each case, as follows. For flow cytometry, 10⁶ cells were seeded in Petri dishes and analyzed at 24 h. For membrane antigen detection, cells were first incubated with the appropriate primary antibody in PBS 1 × and SBF 5%, and then fixed with paraformaldehyde at 4% in PBS 1 × for 15 min at 4 °C. For intracellular antigen detection, cells were first fixed as described and then blocked and permeabilized with fetal bovine serum 20% in PBS 1 × containing 0.2% Triton X-100% for 1 h at room temperature, beforeincubation with the primary antibody and the appropriate secondary antibody (Table I).

Flow cytometric analysis was carried out with an FACSCalibur (Becton Dickinson, San Jose, CA) flow cytometer using standard methods. Data were analyzed with CellQuest Pro and FlowJo software.

For IF analysis, coverslips were mounted on slides using Vectashield (Vector laboratories, Burlingame, CA) with DAPI, which was used for nucleus visualization. Preparations were analyzed with an Olympus BX60 microscope (Olympus Optical Co., Ltd., Tokyo, Japan) and images were taken and analyzed using a digital camera and Spot 4.2 software (Diagnostic Instruments, Inc., Sterling Heights, MI, USA). Confocal analysis was performed when necessary using a Nikon C1-Si confocal laser-scanning imaging system (Nikon Inc, Kawasaki, Japan), built onto a Nikon inverted research microscope Ti-E. Quantification was performed using Image J software.

Immunohistochemical Analyses

Cytospin analyses of cell lines were carried out by loading 500 μl of a 5 × 10⁴ cells/ml suspension and centrifuging the sample at 1000 rpm for 3 min. Slides were dried and fixated in acetone for 5 min before storage. Immunocytochemical techniques were performed with the Novolink Polymer Detection System (Leica Microsystems Inc., Bannockburn, IL, USA). Stained cells were analyzed by two independent examiners and HLA class I expression was determined for each antibody according to the following parameters: the reactivity with different mAbs was graded (according to the HLA & Cancer Component of the 12th IHW) (25) as negative when < 25% cells were stained, as heterogeneous when between 25 and 75% of cells were stained, or positive when > 75% of cells were stained. We scored the results as highly positive (+++), medium positive (++), and weakly positive (+) when the immunostaining of tumor cells was significantly weaker but still stronger than in the negative control, in which we normally used non-immune serum instead of a specific antibody.

Immunohistochemistry analysis of the experimental tissues was carried out on six-micrometer-thick paraffin sections of tissues preserved in formaldehyde. Previous to paraffin inclusion, the bones were treated with HNO₃ 5% for 24 h. Specific antibodies (Table I) that had been incubated overnight at 4 °C followed by addition of the appropriate secondary biotinylated antibodies were used before hematoxylin (Sigma) counterstaining. Slides were visualized in a Nikon 80i epifluorescence microscope/Nikon Ds-Ri1 digital camera controlled by the Nis-Elements (Br) software.

Constructions and Transfections

ERp57 Protein Knockdown

Stealth RNAi oligonucleotide duplexes targeting the open reading frame sequences of ERp57 as well as a nontargeting Stealth RNAi negative control (medium GC content) were obtained from Invitrogen (Paisley, UK). The following sequence was used: 5′-GAAGCUAAAUCCAAAGAAA-3′. The RNA duplexes were introduced into the BO2 cells using Lipofectamine 2000 (Invitrogen) as a transfection agent. A total of 15 × 10⁴ cells were seeded in six-well plates and transfected after 24 h (at 50–60% confluence). Protein knockdown was assessed 24, 48, and 72 h after transfection by semiquantitative PCR and Western blot analysis (supplemental Table S1).

Stable ERp57 Protein Knockdown

ERp57 short hairpin RNA (shRNA) was generated using the siSTRIKE U6 Hairpin Cloning System (Promega, Madison, WI, USA). The coding sequence of ERp57 was analyzed for sites of siRNA targeting using Promega's siRNA Target Designer and the following potential sequence was selected: shERp57 (5′-GAGCTTACTGCATGTTTAT-3′). This corresponds to nucleotides 2378–2396 of the ERp57 transcript (GI:67083697). Short hairpin primers were designed around these sequences and annealed before their ligation into the PstI site of the psiSTRIKE Neomycin vector, following the manufacturer's instructions. In a similar fashion, control constructs were produced using scrambled versions of the shERp57 target sequence indicated above (5′-GGACCCTTAAATGGTTTTT-3′). Both the ERp57 targeted and the control nucleotide sequences were tested against the Genbank database to avoid undesired interactions with other mRNA transcripts. Plasmids were produced via the Qiagen MaxiPrep protocol (Qiagen, Inc., Valencia, CA, USA), and purity and concentration were confirmed by analysis on a NanoDrop Spectrophotometer. BO2 cells were transfected first with a retroviral vector peGFP-CMV/Luc expressing GFP and luciferase and afterward with the shERp57 or sh control respectively, using Lipofectamine 2000 (Invitrogen) as a transfection agent. PCR and Western blot were used to validate the knockdown of ERp57 using the amplification of Neo as a positive control of transfection.

RNA Extraction, Reverse Transcription and PCR

Total RNA was isolated with the TRIzol reagent (Invitrogen) according to the manufacturer's instructions. Purified RNA was reverse transcribed using 300 ng of RNA in a final volume of 20 μl. PCR was performed to evaluate the extent of ERp57 knockdown using cyclophilin as a housekeeping gene. The primers used are indicated in supplemental Table S1. HaeIII was used as a marker.

RT products from the carcinoma cell lines were analyzed for the expression of 10 target genes (HLA class I heavy chain, HLA-A, B, and C specific loci, β2m, TAP1, TAP2, tapasin, LMP2, and LMP7) by quantitative real-time PCR (26, 27). cDNA synthesis was performed with the RNA Reverse Transcription System (Promega Corporation, Madison, WI, USA), following the manufacturer's instructions. These PCR reactions were carried out in a Light Cycler using a DNA Master Probes Kit (Roche Diagnostics, Manheim, Germany). We used commercial kits (Roche Diagnostics and Search LC, GmbH Heidelberg, Germany) to test the housekeeping gene G6PDH and HPRT amplification.

In Vivo Mouse Models, Bioluminescence Imaging and Radiography

Metastasis was induced by injecting 10⁶ cells in 100 μl HBSS into the tail vein of NOD/SCID female mice, under the approval of the animal care committee. The mice were controlled periodically during the experiment until symptoms of metastasis appeared, and metastasis development was monitored by noninvasive bioluminescence imaging. Osteolytic lesions were identified on radiographs as demarcated radiolucent lesions in the bone. The quantification and analysis of the photons recorded in the images was performed using Living Image 4.0 image analysis software (Caliper, Hopkinton, MA). The number of photons was expressed as photon counts per second (ph/s). In vitro regression models were used to correlate cell number and bioluminescence (supplemental Fig. S4) and the final data were expressed as a correlation of the logarithm of cell number versus time after injection. Comparisons were always established one-to-one versus the control transfected cells.

Microsatellite Analysis

To determine the possible loss of heterozygosity in chromosomes 6 and 15, DNA obtained from carcinoma cells was studied with seven short tandem repeats (D6S291, D6S273, C.1.2.C, C.1.2.5, D6S265, D6S105, and D6S276), a mapping HLA region and D6S311 located in 6q. For β2-microglobulin studies, five STR markers that flanked the gene (DS15209, DS15126, DS15146, DS151028, and DS15153) were used. The methods used for the PCR reactions, electrophoresis, and data analysis have been described previously (28, 29). Given that there is no DNA available from normal patient material, additional loss of heterozygosity in BO2 was assigned when a signal reduction of more than 25% in three alleles was seen compared with the original 231.

Coculture Experiments

Buffy coats were processed to extract lymphocytes as previously described (30). NK cells were sorted from purified peripheral blood mononuclear cells by CD3- (BD Biosciences, San Jose, CA) CD56+ (BioLegend, San Diego, CA) selection.

For the cytotoxicity assays, after cell sorting in IMDM 10% human serum and 100 U/ml IL-2, NK cells were kept overnight. The next day, the NK and tumor cells were seeded in 96 U-shaped well plates at a ratio of E:T 9:1 and left overnight. IP incorporation of target cells was measured by flow cytometry. Staurosporine was used as a positive control of cell death at 1 μm.

Statistical Analysis

The statistical analysis of in vitro experiments was performed using SPSS for Windows. Two-way analysis of variance was used to compare the mean expression levels. In all the analyses, differences were considered significant when p < 0.05. Microsoft Excel was used to plot the graphs.

RESULTS

Identification of Bone Metastatic Candidate Proteins

Of the 56 proteins differentially expressed in 2D-DIGE experiments (supplemental Table S2), we identified 35 by MALDI-TOF or LC-MS/MS peptide fingerprint analysis (enlarged with the boxes in supplemental Fig. S1A; supplemental Tables S3 and S4), which revealed that the peptides belonged to 31 different proteins; 15 underexpressed and 16 overexpressed. Because the protein pI range was 4–7, we found mainly acidic proteins with molecular weights of between 11.985 kDa and 65.865 kDa. Most of them were associated with cell structure (tropomyosin 4, vimentin, p16-ARC, calpain, cytokeratin 2, and p23), chaperones and stress (heat shock protein (HSP) 27, HSP 60, TCP1-chaperonin cofactor A, and B23 nucleophosmin-280 AA), metabolism and redox regulation (thioredoxin, PRDX2, ERp57, glyoxalase I, enolase 2, human pyruvate deshydrogenase, pyrophosphatase 1, and dimethylarginine) and intracellular transport and energy (ATP synthase beta subunit and ATP synthase D chain, mitochondrial). Other functions such as transcription (prohibitin), proliferation-death (PCNA and programmed cell death 5), protease (cathepsin D) or transduction and signaling (stathmin 1, galectin 1, human elongation factor and heme-binding protein) were less represented (supplemental Table S5).

MALDI-TOF analysis of different spots resulted in the same proteins being identified. Four of them belonged to HSP60 and two spots belonged to Stathmin. Western blot analysis of 2DE with whole-cell lysates of metastatic variants confirmed the different expression of both protein isotypes between BO2 and 231 cells (supplemental Fig. S1B). Western blot of total denatured proteins suggested that a precursor and processed forms of Stathmin 1 might be differentially expressed between metastatic and parental cells (supplemental Fig. S1C). Moreover, protein expression analysis confirmed the overexpression of HSP60 (5.98-fold), and the underexpression of PRDX2 (0.46-fold), Stathmin 1 (1.25-fold), Cathepsin D (0.50-fold), and Galectin 1 (0.005) in BO2 cells compared with 231 cells. In addition, immunohistochemical analysis showed expression of these proteins in experimental bone metastasis from BO2-injected mice (supplemental Fig. S1D).

We examined previous reports for further support on the relationship of these proteins with breast cancer and/or metastasis (supplemental Table S5 and supplementary references for supplemental Table S5).

The Protein-Protein Interaction Network Reveals that Down-regulated HLA Class I Molecules Interact with ERp57 in Bone Metastatic Cells

The integration package of BIANA was used to build a PPIN around the 31 proteins identified above (i.e. root proteins; proteins known to be involved in bone metastatic breast cancer). The PPIN contained 28 root proteins (of the initial 31, as interactions not be assigned to the remaining three could) and their protein partners (interactors); that is, all together 1328 proteins (Fig. 1A and supplemental Table S6).

Fig. 1.

Flow chart of the work with the in silico bone metastasis analysis using transcriptomic and proteomic data to find functions that are differentially expressed between 231 and BO2 cells. A, A protein-protein interaction network (PPIN) was built from a target set of proteins identified by mass spectrometry (n = 31). The PPIN connects 28 root proteins (yellow), which interact with 1328 nodes (gray) or interacting proteins. B, The previously performed genomic and proteomic analysis was performed on 93 matching proteins included in the PPIN. Gene expression levels are indicated in green (underexpression) or red (overexpression). The cluster containing protein ERp57 interacted with the products of eight genes, four of which were members of the HLA class I family and the SLC2A1 protein. C, Root proteins interacting with ERp57 (yellow) and interacting proteins whose genes were up-regulated (red) or downregulated (green).

We included the information on differential expression reported in the transcriptomic analysis of the comparison between 231 and BO2 (4) cells to increase our knowledge of the network of interactors obtained in silico. The work flow chart in Fig. 1B shows the strategy followed in the analysis.

The 833 differentially expressed genes between 231 and BO2 cells matched 93 proteins from the PPIN: 34 overexpressed and 59 underexpressed (supplemental Table S7). The number of matched proteins was significantly higher than could be expected by random (based on a Monte Carlo permutation test, p = 7.3e-3, see Methods) which motivated the use of interaction network topology in combination with gene expression data. We then used GUILD, a network-based disease-gene prioritization program, to pinpoint the genes that are most likely to drive metastasis due to the location of their proteins in the PPIN. For each root protein, we counted the number of differentially expressed genes in the neighborhood and calculated their average NetCombo score (supplemental Table S8).

The gene encoding for the ERp57 protein was ranked among the top roots and interacts with an elevated number of differentially expressed genes (i.e. at least 10 differentially expressed genes) with the highest scores obtained with GUILD. The ERp57 protein, an ER thiol oxidoreductase also named glucose regulated protein 58, was a central root protein in the PPIN interacting with products of underexpressed HLA genes (supplemental Table S7), SLC2A1, APP, and CBR1; and overexpressed CCT2, CBFB, PHGDH, ILF3, and PPIB (Fig 1C).

In addition, the association of HLA class I and SLC2A1 genes with breast cancer bone metastasis was validated in the literature (31). We searched for references to bone metastasis signatures in published genomic data from experimental and clinical breast cancer and metastasis analysis (supplemental Table S9). HLA-A was found in three gene signatures, HLA-B in two, and HLA-F in another, which analyzed samples from breast cancer patients that were well-correlated with clinical outcome and osteotropic human breast cancer cells (24, 31, 32). Other genes validated in previous studies were: HSPA1A (24), TXNIP (33, 34), FN1 (1, 34), and TUBA1 (1). The S100A4, KARS, SLC2A1, and TIMP3 genes were also associated with breast cancer progression (24), as was CD44, which was found in two signatures (31, 35).

We performed Western blot and immunofluorescence microscopy analysis of bone metastatic cells using specific antibodies to experimentally validate the relationship between HLA class I underexpression and ERp57 protein overexpression (Fig. 2A and 2B). Both independent procedures showed general HLA class I molecule down-regulation in BO2 cells compared with 231 cells. This was inversely correlated with ERp57 expression, which was higher in BO2 than in 231 cells. Moreover, the immunofluorescence analysis of cells using a primary antibody against HLA class I showed differences between the two cells, either in membrane or intracellular expression (Fig. 2B). This was confirmed by quantitative cytometric analysis, together with an inverse correlation with ERp57 expression (supplemental Fig. S2A, right-hand panel, box). Furthermore, we analyzed the expression of ERp57 and HLA class I proteins in lung and brain metastatic variants from a different breast cancer model and no relationship was found. In addition, we calculated the ratio of intracellular HLA class I proteins with regard to membrane expression and found a higher ratio in bone metastasis (both BO2 and 435-B1) cells than in their parental cells (231 and 435, respectively (supplemental Fig. S2B)). This was associated with high levels of ERp57, 41% more in 435-B1 and 66% in BO2 cells versus their parental cells. Similarly, liver metastatic cells (435-Lv1) showed a higher ratio of HLA class I retention and increased ERp57 levels (15% more than 435-P). In contrast, lower levels of ERp57 in 435-L3 and 435-Br1 metastasis were associated with lower retention of HLA class I in intracellular compartments (supplemental Fig. S2B).

Fig. 2.

ERp57 and HLA class I molecule expression in BO2 cells compared with parental 231 cells. Expression of HLA class I molecules (clone MEM-147) and ERp57 under the conditions described in Table I, by: A, WB and B, immunofluorescence. Representative microscopic images (40×) showing differences in ERp57 expression and HLA class I localization, as either mHLA (membrane) or iHLA (intracellular) class I molecules, between the bone metastatic BO2 cells and 231 cells. C, Immunocytochemistry (40×) of cytospined and acetone-fixed cells carried out with specific antibodies (see Table I) showing the cellular localization of the proteins indicated. The table at the bottom shows the semi-quantitative analysis of positive cells and staining intensity.

Class I molecules consist of a polymorphic 45 kDa transmembrane heavy chain, a 12 kDa subunit termed β2-microglobulin (β2m), and an 8–9 residue peptide ligand (36). We typed by immunocytochemistry the expression of HLA-ABC and β2m, the HLA-A and B loci and the free BC heavy chain in 231 and BO2 cells (Fig. 2C). We found general underexpression of all HLA molecules in BO2 cells compared with 231, at the protein level (Fig. 2C). There was a lower percentage and less staining intensity (25%, ++) of intracytoplasmic free B, C heavy chain in BO2 cells than in 231 cells (75%, +++). In addition, BO2 cells showed general underexpression of HLA class I molecules both intracellularly and in the membrane: HLA-ABC, β2m, HLA-A and B loci, and the free BC heavy chain.

ERp57 Regulates HLA Class I Protein Membrane Expression in Bone Metastatic Cells

Tumor cells can use multiple mechanisms to partially or totally down-regulate the expression of HLA class I antigens (37). To analyze the functional relationship between ERp57 and HLA class I expression in bone metastatic cells, we down-regulated ERp57 in BO2 using a small interfering RNA (Fig. 3A). After 24 h, ERp57BO2 knockdown cells showed lower expression of HLA class I proteins in the intracellular compartment (iHLA class I) and higher expression of HLA class I proteins in the membrane (mHLA class I) than the control cells (Fig. 3B). This was statistically significant by Image J quantification (intensity/cell p < 0.05 in all cases).

Fig. 3.

ERp57 knockdown in BO2 cells leads to an increase in membrane and a decrease in intracellular HLA class I molecules. A, We verified gene expression by PCR (upper panel) and protein expression by WB (bottom panel) in cells with scrambled oligo (SiBO2) and ERp57 (siERp57BO2) at different times. The efficiency of the transfection reduced the protein expression at 24 h by more than 50%. B, Immunofluorescence validated the efficient down-regulation of ERp57 in siERp57BO2 at 24 h, with a consequent increase in HLA class I molecules in the cell membrane and a decrease in intracellular compartments. C, Regulation of ERp57 expression by glucose levels was tested by immunofluorescence. Glucose starvation for 24 h increased ERp57 leading to a decrease in membrane (mHLA) and retention of intracellular (iHLA) class I molecules in parental 231 cells. D, Confocal microscopy to verify in 231 and BO2 cells the colocalization of HLA class I proteins (Alexa ® 488) with either an ER-tracker (ERp57) or Golgi tracker (golgin), both labeled with Alexa ® 555. Quantification of the colocalization of HLA class I molecules with golgin and ERp57 was performed and represented as a percentage versus the total amount of each one (right-hand panel). Significant differences between 231 and BO2 cells in the HLA class I molecules localized in the Golgi apparatus is observed, with an increase in BO2 cells (p < 0.0001). No differences between 231 and BO2 cells were observed for HLA class I molecules colocalized with ERp57 in ER (p = 0.090).

Because ERp57 is a thiol oxidoreductase of the ER, whose expression is induced by low glucose concentrations, similar results were obtained when we used the parental ERp57 low-expressing 231 cells thereby confirming that glucose starvation induced overexpression of ERp57 with secondary underexpression of mHLA class I (Fig. 3C). These results suggest that overexpression of ERp57 enables cells to adapt to the low nutrient availability in the bone microenvironment.

Because ERp57 is overexpressed under low glucose conditions, the low glucose burden of bones (38) might determine the selection of the most appropriate phenotype to live in the bone microenvironment. We hypothesized that bone metastatic cells may present specific metabolic behavior that justifies the role of ERp57 in the bone metastatic phenotype. Therefore, we measured both glucose intake and lactate production rates by the cell as an assessment of glycolytic flux (Supplemental Fig. S2C). The 435-B cells showed a ratio between lactate production and glucose consumption of ≈1.5, whereas lung metastatic cells had a ratio of ≈2, which is an indication of higher glycolytic capacity (mean ± S.E., **p < 0.01, ***p < 0.001; Student's t test). These results strongly suggested that the survival advantage of bone metastatic cells might lie in the specific adaptation of breast cancer cells to the low glucose conditions of the bone microenvironment.

Moreover, we used confocal microscopy to verify the intracellular association of ERp57 andHLA class I molecule localization. The higher expression of ERp57 in the ER of BO2 cells was inversely related with HLA class I expression in membrane, in contrast to the low ERp57 expression in 231 cells (Fig. 3D) with HLA class I membrane protein expression (Fig. 3D, left-hand panels). Because the colocalization of ERp57 and HLA class I molecules in the ER was the same in 231 and BO2 cells (p = 0.090), we used the tracker golgin to verify the colocalization of ERp57 and HLA class I molecules in the Golgi apparatus (Fig. 3D, right-hand panels). Indeed, HLA class I molecules were retained mainly in the Golgi apparatus (p < 0.0001) in BO2 cells. This suggests that ERp57 plays an indirect role in the transport of HLA class I proteins from ER to the Golgi apparatus and so to the cellular membranes.

The role of ERp57 in the folding and assembly of HLA class I molecules has been reported (39, 40). Therefore, we studied which HLA class I molecules and/or subunits of the antigen-processing machinery (APM) could be affected by ERp57 expression and whether the processing machinery might be involved.

We performed RT-PCR analysis of 231 and BO2 cells, including siERp57BO2 and control BO2 cells, to establish the functional relationship between ERp57 and HLA class I gene expression in bone metastatic cells. We analyzed the expression of HLA-A, B and C loci in the HLA-A, B and C heavy chain/B2m complex (Fig. 4A). These experiments showed that the mRNA levels for all gene products were constitutively lower in BO2 cells than in the parental cell line. In the case of β2m, this could be because of a loss of heterozygosity in this gene. Because similar amounts of HLA class I molecules were observed in siERp57BO2 cells (Fig. 4A), which suggests that ERp57 regulates HLA class I molecule expression at the posttranscriptional level.

Fig. 4.

Transcriptional expression of HLA class I molecules and the antigen-processing machinery (APM) in BO2 cells and their relationship with ERp57. Normalized mRNA levels of transcripts in tumor cell lines are measured by RT-QPCR and absolutely quantified using GPDH as a housekeeping gene. A, The expression levels of the indicated HLA class I genes show underexpression in BO2 cells compared with 231, as levels were maintained in siBO2 or siERp57BO2 cells. B, The APM shows general underexpression in BO2 cells compared with 231. Only tapasin is inversely correlated with ERp57 expression when we compare siBO2 or siERp57BO2 cells, with a significant increase in siERp57BO2 cells.

It has been stated that the covalent tapasin/ERp57 association affects the stability of the APM and the recruitment of HLA class I. In the PPIN, we found tapasin interacting with ERp57 (as shown below in Fig. 5A). Consequently, we hypothesize that an imbalance between tapasin and ERp57 can induce kinetically unstable HLA class I molecules that cannot transit the cell surface. This decreases their expression in the cellular membrane. We used RT-PCR to explore the expression of APM proteins in 231 and BO2 cells (Fig. 4B). General underexpression of TAP1, TAP2, LMP2, LMP7, and tapasin was found in BO2 cells, compared with 231. Then, we knocked down ERp57 in BO2 cells to analyze the cause-effect on APM protein expression. We found specific up-regulation of tapasin expression in siERp57BO2 compared with the control cells.

Fig. 5.

ERp57 is linked with the cytoskeleton structure through vimentin. A, Detail of breast cancer bone metastasis PPIN showing the relationship between ERp57 (overexpressed) and vimentin (underexpressed), connected by CCT2 and Prohibitin. Interacting proteins are also shown: in yellow root proteins, and in red and green overexpressed and underexpressed proteins, respectively. The right-hand panel shows the different expression of vimentin in 231 and BO2 cells (60X). It is underexpressed in bone metastatic cells compared with parental cells. B, Immunofluorescence (60×) against vimentin in transient (siERp57BO2 versus siBO2) and stable knockdown of ERp57 in BO2 cells (sh21) and control (Ctr4). Both systems display a similar distribution of vimentin induced by down-regulated ERp57.

In addition, we explored the immunogenicity of 231 and BO2 cells cocultured with NK cells from healthy donors and we measured cell death by propidium iodide staining. As expected, NK cells effectively recognized and killed more BO2 than 231 cells: 1.7- to 2.5-fold more in three independent experiments (p < 0.05). These results show proper NK recognition according to HLA class I levels in the tumor cell membrane (supplemental Fig. S3).

ERp57 is Involved in Conformational Cytoskeleton Changes of Bone Metastatic Cells

We further analyzed the PPIN to look for additional functions regulated by ERp57 that could affect the metastatic ability of breast cancer cells. We found that ERp57 interacted with proteins that function in cytoskeleton and cellular adhesion, such as cytokeratin-19, cytokeratin-2, galectin-1, stathmin-1, and vimentin (Fig. 5A).

The processes required for the metastasis of epithelial cells from primary tumors crucially involve the epithelial mesenchymal transition, a mobile cellular phenotype that is characterized by increased vimentin expression (41). This process is followed by mesenchymal-to-epithelial transition (MET) at the site of arrest. Therefore, we verified the expression of these proteins by specific antibodies. Vimentin expression was found to be lower in BO2 cells than in 231, which validates the in silico data and suggests that BO2 cells have already bypassed the MET process. Moreover, knockdown of ERp57 either by siRNA or shRNA showed a reorganization of vimentin fibers in the cell. This leads to a perinuclear location and the presence of protuberances that are reminiscent of vimentin and microtubule condensation (Fig. 5B). These results support cause-effect involvement of ERp57 expression in vimentin organization and the MET process, a well-known phenotype in established metastases of most cancers, which recapitulate the differentiated phenotype of their primary tumors.

ERp57 Plays a Cause-Effect Role in Bone Metastasis

We therefore decided to study knockdown ERp57 expression in BO2GFP/tTA with the siSTRIKE U6 Hairpin Cloning System (supplemental Fig. S4A) to further examine the consequences of elevated ERp57 in bone metastasis in vivo. We chose stable shERp57BO2 down-regulated clones #21 and #32, and the scrambled version of ERp57 (scbl) for further characterization. Plots of the amount of light versus the number of BO2GFP/tTA cells were linear throughout the range of cells tested, R² = 0.993 (supplemental Fig. S4B). Moreover, ERp57 expression was verified by Western blot and IF analysis (Fig. 6A and Supplemental Fig. S4C).

Fig. 6.

ERp57 down-regulation impairs bone metastasis. A, BO2 cells were stably transfected to underexpress ERp57. IF validation of ERp57 underexpression and HLA class I overexpression in shERp57BO2C21 and shERp57BO2C32 cells versus controls (shBO2/GFPCT4). B, Clones were injected into NOD/SCID mice and tumor formation was followed by bioluminescence analysis. First, in vitro linear regression was performed between bioluminescence (ph/s) and cell number. Then, conversion to cell number was performed after bioluminescence follow-up in injected mice. (* = p < 0.05, ** = p < 0.01, *** = p < 0.001). Bottom panel: representative image showing bioluminescence in animals at day 47. In the case of sh21, the animal shown was the only one affected. Radiographic images of hind legs were taken at day 60 showing osteolytic lesions as radiolucent areas. C, H&E and immunohistochemical analysis of bone metastasis in shERp57BO2C21 and shERp57BO2C32 cells versus shBO2/GFPCT4 using ERp57, PCNA and Caspase 3 antibodies. All images at ×20. (scale = 100 μM)

To examine whether down-regulation of ERp57 in bone metastatic cells affects their metastatic capacity and their propensity to grow in the bones, we induced bone metastasis by endovenous injection in NOD/SCID mice (n = 7 per group) of ERp57 knockdown clones, shBO2 #21 and #32, or control cells (scbl). The diffuse photon accumulation over the entire animal body was monitored periodically by Bioluminescence imaging (BLI). As expected, the first evidence of bioluminescence emission from bones, which is indicative of bone metastasis, was seen at day 14 after endovenous injection of cancer cells in control animals. The follow-up revealed that the tumor burden of shERp57BO2C21 and shERp57BO2C32 animals was statistically significantly less than that observed in control mice injected with shBO2/GFPCT4 cells (p < 0.01 in sh21, p < 0.0001 in sh32 compared with ct4) from day 34 until the end of the experiment at day 60 (Fig 6B, upper panel). A radiographic analysis of hind legs at day 45 and day 60 after tumor cell inoculation showed that mice bearing shERp57BO2C21 and shERp57BO2C32 cells had significantly smaller osteolytic lesions than mice injected with shBO2/GFPCT4 cells (Fig. 6B, bottom). Moreover, the incidence of metastasis in mice bearing shERp57 cells was different from that observed in control animals: shERp57BO2C21, 25% (1/4); shERp57BO2C32 40% (2/5); and shBO2/GFPCT4, 67% (4/6). Lung micrometastasis or macrometastasis was found in all groups: shERp57BO2C21, 100% (5/5); shERp57BO2C32 80% (4/5); and shBO2/GFPCT4, 67% (4/6). These results indicated that ERp57 overexpression improved BO2 cell growth in bones and played an organ specific cause-effect role in bone metastasis development.

We further examined the ex vivo expression of ERp57 in bone metastasis samples from sh21 and sh32 metastasis-bearing mice and control animals by immunohistochemical analysis. As expected, stable knockdown of ERp57 persisted in bone metastasis (Fig. 6C). Furthermore, there were PCNA differences among tumors indicating a decrease in proliferation in shERp57BO2 metastasis with respect to controls (Fig. 6C). Increased cell death as indicated by caspase-3 activation was evident in shERp57BO2 bone metastasis, relative to controls.

DISCUSSION

The integration of expression profiles with protein interactions in a well-known model of experimental breast cancer bone metastasis permitted the analysis of active subnetworks in which the GRP58/ERp57 protein (localized in the ER lumen and functions as a protein chaperone involved in the folding of glycoproteins) was clustered with four down-regulated nodes from HLA class I molecules (HLA-A, HLA-B, HLA-E, and HLA-F). This phenotype might be primed by the limited glucose conditions and hypoxia in the bone microenvironment. Indeed, low glucose under hypoxic conditions induces an unfolded protein response (42). The interactions observed in our model (see Fig. 5A) suggest functional connections between microenviromental metabolism and bone metastasis progression. In addition, the current model for the role of ERp57 is that it functions with calreticulin and calnexin; complexes which directly modulate glycoprotein folding and therefore act as molecular chaperones (43).

PPINs provide a framework for studying the functional relationships among biological molecules. They can be used to attribute functions to genes whose functions have not been characterized, in this case providing valuable insights into the complexity of metastasis pathogenesis. Aragüés and coworkers (16) postulated that integrating PPINs and genomics data would improve success in predicting genes involved in cancer. They combined several sources of data such as gene expression, protein-protein interactions, and structural, functional and evolutionary features to create a prediction model based on the integration of data (44). This combined prediction model outperformed each of the single-model predictions and was used to produce a reliable list of cancer gene candidates.

There is an elevated likelihood for cancer genes to encode hubs and they usually appear in networks with a high functional and topological centrality (45). The analysis of a structurally characterized network revealed that cancer-related proteins have smaller, more planar, more charged and less hydrophobic binding sites than noncancer proteins. In addition, they are more prone to employ transient interactions and more likely to be involved in multiple pathways (46, 47). In this context, ERp57 emerges as a hub protein with the potential to recruit many proteins that function during critical stages of bone metastasis progression.

ER stands at the crossroads of two fundamental cellular processes in cancer progression: the immunological response through HLA class I antigen presentation and the unfolding protein response. Exposure to ER stress conditions cells for survival, decreases levels of HLA class I proteins at the cell surface and decreases the synthesis of new HLA class I molecules. This prepares cells for subsequent, more severe stress (40, 48). We show evidence of the regulation of ERp57 at the cellular level by the amounts of glucose in the medium. Consequently, the reported low glucose in bone tissue might condition the selection of the most adaptive phenotype by regulating ERp57 expression. Moreover, ERp57 knockdown in BO2 cells impaired bone metastasis without affecting lung metastasis progression. This indicates selective organ-specific pressure on ERp57-expressing cells to live in bones. Thus, ERp57 appears as a multifunctional chaperone that can regulate diverse biological processes to maintain cellular homeostasis, which enhances bone metastasis development.

Because bones are a hypoxic microenvironment, these results are in accordance with the reported inverse correlation between oxygen concentration and ERp57 expression (49). Consequently, cells with high levels of ERp57 are more prone to apoptotic cell death in hyperoxic microenvironments. Indeed, ERp57 is a widely expressed protein disulfide isomerase that regulates protein-protein interactions via a redox mechanism based on the activity of its two thioredoxin-like domains. It acts as a cellular redox sensor to adapt cells to oxidative insults (50). The redox-sensitive mechanism that depends on ERp57 expression activates mTORC1, which mediates the correct assembly of mTOR complexes (51). This is a master regulator of protein synthesis that couples nutrient sensing to cell growth and cancer (52).

ERp57 has an emerging role as a regulator of cytokine-dependent signal transduction and gene expression. STAT3 signaling is dependent on ERp57, which modulates its activity by ER luminal interactions (53). Moreover, ERp57 aberrant expression in carcinomas is a prognostic factor (54). This supports its multifunctional role in cancer progression. Other functions related to the metastatic phenotype are directly activated by ERp57 expression, as is the MET needed for metastatic cell anchorage and growth in bones, which takes place mainly through cytoskeletal reorganization (55). In agreement with our data, recent studies have found a nuclear multimeric complex in which ERp57 is associated with proteins involved in cell division and cytoskeleton, such as β-actin or vimentin. In that complex, ERp57 might exert a redox control on TUBB3 folding and the correct association of microtubules and the kinetochore (56). Further studies have evaluated the importance of this phenotype in paclitaxel resistance in ovarian cancer (57, 58).

ERp57 expression in BO2 cells interferes with HLA class I membrane expression. This induces the strong down-regulation of HLA class I in BO2 cells, which are also found by cell membrane proteomic analysis (31). ERp57 operates in two functions during HLA class I biogenesis: as a thiol oxidoreductase during the early HC oxidative folding stages and as a novel structural component within the peptide loading complex, which includes subunits of the peptide transporter associated with antigen processing and presentation (TAP1 and TAP2), the HLA class I-specialized chaperone tapasin, the lectin chaperones calreticulin and calnexin and HLA class I heterodimers (39, 59, 60). In this scenario, only tapasin is inversely regulated by ERp57, because in siERp57BO2 cells its expression was increased. Moreover, general APM molecule underexpression was found in BO2 cells compared with parental cells.

It is well known that tapasin sequesters 15–80% of the total ERp57, independently of the lectin-like chaperones calnexin and calreticulin (61). This suggests that their covalent interaction affects the stability of peptide loading complex and the recruitment of HLA class I molecules in bone metastatic cells, probably by changing the dynamics of the typically weak and transient interactions observed among calreticulin, ERp57 and their glycoprotein substrate (60). Studies of tapasin-deficient cells have indicated that HLA class I assembly is compromised even with optimal functionality of calreticulin and ERp57 (62). Therefore, the general retention of HLA class I molecules in ER and post-ER compartments is probably because of the low-affinity peptide cargo. Tapasin participates in quality control of HLA class I molecules, restricts the export of peptide-deficient class I molecules beyond the cis-Golgi (63), and functions as a cargo receptor, mediating the packaging of unstable HLA class I molecules in COPI-coated vesicles for retrograde transport from the Golgi complex to the ER (64). Because the mixed disulfide complex that ERp57 forms with tapasin is stable, the covalent complexes of cellular ERp57 and tapasin might make it difficult to export HLA class I molecules (36). Therefore, the BO2 cell phenotype might be compatible with folding and antigen loading defects, which dispose of misfolded proteins for ER-associated retrotranslocation and degradation (65). The retention of HLA class I molecules in the Golgi apparatus is secondary to tapasin down-regulation.

In conclusion, the combination of the GUILD scores and the analysis of active subnetworks provides an explanation of the role of ERp57 in the defects in HLA class I surface expression, with relevant functions involved in bone metastasis development. The total or selective losses of HLA class I antigens is one of the escape mechanisms found most frequently in experimental and spontaneous tumors (66–68), and in bone metastatic cells (31). In addition, HLA class I suppression affects the proper function of a cell independently of the immune system, and has been associated with abnormal p53 activation and ER stress, conferring cellular survival advantage (69). Our results show that bone metastatic cells present lower levels of HLA class I in the cell membrane, due to a general decrease in transcription of HLA class I molecules and an additive accumulation in post-ER compartments, which might be associated with the selective pressure exerted by the microenvironment, suggesting a tumor-immune escape mechanism.

Bone is the preferred organ site of breast cancer metastasis, with an incidence of between 70 and 85% in patients with advanced disease (70, 71). Future studies should focus on analyzing the expression of ERp57 in primary breast carcinomas to assess its putative use as a prognostic factor. In addition, therapies should focus on preventive bone metastasis treatments.