Transcriptomic and proteomic analyses in bone tumor cells: Deciphering parathyroid hormone-related protein regulation of the cell cycle and apoptosis


  • Isabella WY Mak,

    1. Department of Oncology, McMaster University, Hamilton, Ontario, Canada
    2. Department of Surgery, Juravinski Cancer Centre, Hamilton Health Sciences, Hamilton, Ontario, Canada
    Search for more papers by this author
  • Robert E Turcotte,

    1. Department of Orthopaedic Surgery, McGill University Health Centre, Montreal General Hospital, Montreal, Quebec, Canada
    Search for more papers by this author
  • Michelle Ghert

    Corresponding author
    1. Department of Oncology, McMaster University, Hamilton, Ontario, Canada
    2. Department of Surgery, Juravinski Cancer Centre, Hamilton Health Sciences, Hamilton, Ontario, Canada
    • FRCSC, 699 Concession Street, Hamilton ON L8V 5C2, Canada.
    Search for more papers by this author


Giant cell tumor of bone (GCT) is an aggressive skeletal tumor characterized by local bone destruction, high recurrence rates, and metastatic potential. Previous works in our laboratory, including functional assays, have shown that neutralization of parathyroid hormone-related protein (PTHrP) in the cell environment inhibits cell proliferation and induces cell death in GCT stromal cells, indicating a role for PTHrP in cell propagation and survival. The objective of this study was to investigate the global gene and protein expression patterns of GCT cells in order to identify the underlying pathways and mechanisms of neoplastic proliferation provided by PTHrP in the bone microenvironment. Primary stromal cell cultures from 10 patients with GCT were used in this study. Cells were exposed to optimized concentrations of either PTHrP peptide or anti-PTHrP neutralizing antiserum and were analyzed with both cDNA microarray and proteomic microarray assays in triplicate. Hierarchical clustering and principal component analyses confirmed that counteraction of PTHrP in GCT stromal cells results in a clear-cut gene expression pattern distinct from all other treatment groups and the control cell line human fetal osteoblast (hFOB). Multiple bioinformatics tools were used to analyze changes in gene/protein expression and identify important gene ontologies and pathways common to this anti-PTHrP–induced regulatory gene network. PTHrP neutralization interferes with multiple cell survival and apoptosis signaling pathways by triggering both death receptors and cell cycle–mediated apoptosis, particularly via the caspase pathway, TRAIL pathway, JAK-STAT signaling pathway, and cyclin E/CDK2-associated G1/S cell cycle progression. These findings indicate that PTHrP neutralization exhibits anticancer potential by regulating cell-cycle progression and apoptosis in bone tumor cells, with the corollary being that PTHrP is a pro-neoplastic factor that can be targeted in the treatment of bone tumors. © 2012 American Society for Bone and Mineral Research.


Giant cell tumor of bone (GCT) is an aggressive and highly osteolytic bone tumor that is characterized by local osteolysis, regional pain, and the predisposition to pathological fracture.1 Current preferred treatment of GCT consists of limb sparing surgery by the means of extended curettage with the addition of local toxic adjuvant therapies.2 Although anatomy and function are preserved with such an approach, local recurrence rates remain high,3 thus emphasizing the importance of developing an understanding of the biology of the bone tumor and subsequent creation of more effective therapeutic options.

The cellular elements of GCT include both osteoclast-like giant cells formed by merging of monocytes from the hematopoietic lineage, and proliferating osteoblast-like stromal cells from the mesenchymal lineage.4 Although the giant cells are primarily responsible for the osteolysis typical in GCT, evidence suggests that the osteoblast-like stromal cell is true neoplastic component of this neoplasm.5 Previous work in our laboratory has shown that the osteoblastic transcription factors Runx2 and AP-1, as well as cytokines interleukin (IL)-1β and parathyroid hormone-related protein (PTHrP), play an important role in regulating the neoplastic properties of GCT.6–10

PTHrP is present in many organs and tissues, exerting its effects through an autocrine/paracrine action.11 PTHrP shares the same N-terminal end as parathyroid hormone (PTH); therefore, it can simulate most of the actions of PTH, including increases in bone resorption.12 PTHrP was first identified as the primary tumor-derived agent responsible for humoral hypercalcemia of malignancy.13 When produced by tumors, PTHrP, by virtue of its ability to bind to and activate the G protein–coupled PTH/PTHrP receptor (GPCR), is the humoral factor responsible for pronounced bone resorption and hypercalcemia.14, 15 The majority of neoplastic tissues that metastasizes to bone produce PTHrP, and PTHrP expression correlates with skeletal localization of bone tumors.16

Previous work in our laboratory has shown that GCT stromal cells express high levels of PTHrP and GPCR at both the gene and protein levels.10, 17 We also determined via multiple functional assays (mitochondrial dehydrogenase assays, crystal violet assays, apolipoprotein A1 (APO-1) ELISAs, caspase activity assays, flow cytometry, and immunofluorescent immunohistochemistry) that neutralization of PTHrP in the cell environment inhibited cell proliferation and induced apoptosis in GCT stromal cells.10 Thus it appears that PTHrP acts as an autocrine factor to propagate cell proliferation in GCT and is therefore an intriguing therapeutic target in the setting of malignancy, particularly in the bone microenvironment.

Nonetheless, the rudimentary mechanism of how PTHrP neutralization inhibits cell proliferation and induces cell death in bone tumor cells remains unidentified. Hence, characterization of the global pattern of genes and proteins regulated by PTHrP is important for better understanding the mechanism of action of this autocrine-mediated cytokine. The development of microarray technology represents a powerful tool for characterizing such large-scale changes at the transcriptomic and proteomic levels. The objective of this study was to investigate the global gene and protein expression patterns of GCT in order to identify the underlying pathways and mechanisms of neoplastic proliferation provided by PTHrP in the bone microenvironment.

Patients and Methods

Ethics statement

The use of all patient-derived material was approved by the Hamilton Health Sciences and McMaster University Faculty of Health Sciences Research Ethics Board (REB Project #: 05-302). Written patient informed consent was obtained individually. The Hamilton Health Sciences/McMaster University Research Ethics Board operates in compliance with the International Conference on Harmonisation (ICH) Good Clinical Practice Guidelines and the Tri-Council Policy Statement: Ethical Conduct for Research Involving Human and Division 5 Health Canada Food and Drug Regulations.

GCT sample collection

The diagnosis of GCT of bone was established by biopsy prior to surgical excision. Specimens were obtained at the time of surgery from patients undergoing tumor resection and a bone pathologist verified the diagnosis of GCT postoperatively. Tissue samples from 4 cases of GCT of bone were used in this study and all experiments were performed in triplicate for all patient specimens or as otherwise stated. Microarray and proteomics studies were done in vitro with 4 patient samples and we validated these expression signatures with real-time PCR in these 4 patient samples and in 6 additional patient specimens. The results are the mean of triplicate experiments for the average of all patient samples.

Primary cell lines and cultures

Primary cell cultures of GCT stromal tumor cells were isolated, characterized, and established from 4 fresh GCT patient samples as we have described.6 Following several successive passages, the mesenchymal stromal cells became the predominant cell type, whereas the multinucleated giant cells were eliminated from the culture. Primary cultures of the proliferating homogenous stromal tumor cell population obtained after the fifth or sixth passage (without any hematopoietic markers) and up to the tenth passage were used for experiments. Human fetal osteoblast (hFOB) 1.19 cells (American Type Culture Collection [ATCC], Manassas, VA, USA; ATCC#CRL-11372) were used as a control cell line.

Anti-PTHrP neutralization and PTHrP peptide treatment

Anti-PTHrP (1-34) neutralizing antiserum (T-4512; Bachem Americas, Inc., Torrance, CA, USA) was used to neutralize secreted PTHrP. Cells were seeded into 96-well plates at 1 × 103 cells/well in medium containing 10% fetal bovine serum (FBS). After cell attachment, cells were transferred to serum-free medium treated either with 10 µg/mL anti-PTHrP neutralizing antibody (PAB) or 10 µg/mL immunoglobulin G (IgG) control (Control). This concentration was determined in our previous study on the effect of toxicity and dosage response of the anti-PTHrP neutralizing antibody in stromal cells.10 One micromolar (1 µM) of PTHrP peptide (1-34; Bachem Americas), as optimized in our laboratory, was used to compare with the genomic expression from cell death initiated by the neutralization of PTHrP. Total RNA of the cell culture was collected after 24 hours for the microarray assay and real-time PCR validation, and cell lysates were studied in a proteomic antibody microarray assay.

RNA purification and reverse transcription

We isolated total RNA from GCT stromal cells using the RNeasy mini kit (Qiagen, Toronto, ON, Canada) as optimized in our laboratory. To ensure complete removal of contaminating genomic DNA prior to first-strand synthesis, RNase-free DNase I treatment was applied on the RNeasy column during total RNA isolation. Amounts, purity, and integrity of RNA were evaluated by a UV spectrophotometer (DU530; Beckman, Mississauga, ON, Canada) and an Agilent 2100 Bioanalyzer (Agilent, Palo Alto, CA, USA). For real-time PCR, single-strand complementary DNA (cDNA) was synthesized from 1.0 µg of total RNA using the SuperScripts III First-Strand Synthesis System for reverse transcription (RT)-PCR (Invitrogen) and oligo(dT) 12-18 primer, following the manufacturer's instructions.

cDNA microarray experiments

We extracted RNA at the designated time and quantified it for microarray analysis. Target RNA was reverse transcribed into cDNA and in vitro transcription was performed to generate biotin-labeled cRNA for subsequent hybridization. The synthesis of the cDNA and cRNA were assessed using the Agilent 2100 Bioanalyzer. Hybridized target cRNA was stained with streptavidin phycoerythrin on the array. Triplicates of gene transcript profiles in both control and treated cultures were studied by high-density oligonucleotide microarray GeneChip Human Gene 1.0 ST Array (Affymetrix, Santa Clara, CA, USA). Arrays were scanned at a resolution of 2.5 µm using an Affymetrix GeneChip Scanner 3000 (Affymetrix) at an excitation wavelength of 488 nm. Light emissions at 570 nm are proportional to the bound target at each oligonucleotide's position on the GeneChip array. Affymetrix CEL files for each array were generated using the Affymetrix GeneChip Operating Software. Sample labeling, hybridization of test arrays, and hybridization of full-size arrays were performed using protocols described in the Affymetrix manual. The data were subjected to normalization, background correction, and data summarization using Affymetrix Expression Console software.

Microarray data analysis

The microarray image data were processed in the Centre for Functional Genomics facility at McMaster University according to standard procedures. CEL data was analyzed with Partek Genomic Suite v6.5 (Partek Inc., St. Louis, MO, USA), which can normalize and process multiple datasets simultaneously. Data normalization was performed by the robust multiarray analysis (RMA) algorithm. Statistical differences were examined using one-way analysis of variance (ANOVA). This method addresses random oscillation and helps to diminish both false-positive and false-negative rates. In order to account for multiple testing, the significance level was corrected according to a false discovery rate (FDR); ie, the number of false positives divided by the number of selected genes. Significant probe sets were defined with a p value cutoff significant with an FDR of <0.002, and a fold-change ≥2.0 for upregulated genes and ≤ −2.0 for downregulated genes, respectively. Gene expression values were retrieved from four GCT samples under three conditions each: Control, PAB (neutralizing antibody), and PTHrP peptide treatment (PP) array databases. Genes were annotated according to molecular functions of the Gene Ontology (GO) database using the NetAffx database.

GO enrichment and interpretation

We used gene enrichment analysis to interpret the biological impact of PAB treatment in GCT stromal cell proliferation. GO enrichment analysis was undertaken in the Partek Genomic Suite using a chi-square test comparing the proportion of the transcript list in an ontology to the proportion of the background list in that same ontology. Functional groups with >2 genes and an enrichment score >2 were considered significant. For each cluster and for each GO group defined above, GO enrichment based on Fisher's exact test was calculated separately, so that the resulting p value represents the probability that the observed numbers of selected genes belonging to the cluster and annotated with the GO group have resulted from random sampling. GO groups with p value less than 0.05 were considered as significantly enriched for the cluster.

Functional association and pathways analysis

We uploaded the lists of differentially expressed genes into two different software programs: Exploratory Gene Association Networks (EGAN; The Regents of the University of California)18 and WebGestalt (WEB-based GEne SeT AnaLysis Toolkit; Vanderbilt University, Nashville, TN, USA)19 in order to determine differentially regulated pathways using the full human genome as reference background. Data were analyzed in the “Functional Annotation Clustering” tool using the “High” classification stringency setting for Molecular Function (MF), Biological Process (BP), and Cellular Component (CC) GO terms. Functional annotation clusters with enrichment scores >2 were considered significant, in accordance with EGAN and WebGestalt recommendations. Pathway analysis was also performed using both EGAN and WebGestalt under the nonparametric Gene Set Enrichment Analysis (GSEA) enrichment statistics. GSEA considers and ranks all genes in an experiment (not only those above an arbitrary cutoff). For these analyses, 618 gene sets were used, which comprise the entire PAB vs Control gene sets collection. We investigated associated terms in Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, WikiPathways database, and Pathway Commons analysis, using both the EGAN and WebGestalt programs. Gene sets with 5% FDR were considered significant in accordance with GSEA recommendations.

PCR and real-time PCR

In order to validate the microarray measurements, we performed a quantitative real-time PCR as described.6 PCR experiments were performed in triplicate and included negative no-template controls. Primer pairs (Table 1) that spanned at least one intron-exon boundary and produced amplicons in the range of 100 to 200 bp were designed using the Real-time PCR Primer Design software (VWR GenScript Corp., Piscatway, NJ, USA) and synthesized (Sigma-Aldrich, Oakville, ON, Canada). In addition, a primer pair for the housekeeping/reference gene GAPDH was also included. We verified the amplicon specificity and sensitivity of all primer pairs with PCR before applying them to real-time PCR.

Table 1. Human Primer Sequences Specially Designed for Real-Time RT-PCR Amplification
GeneDirectionPrimer sequenceAccession #Size (bp)Melting temp (°C)
  1. F = forward; R = reverse.

MYD88F5′ AAA GAG GTT GGC TAG AAG GCC ACG G 3′NM_00117256713461

Relative quantification using real-time PCR

We designated GAPDH as the reference gene for relative quantification, through which we normalized the expression of endogenous mRNA from GCT stromal cells. Cycle threshold (Ct) numbers were derived from the exponential phase of PCR amplification. Relative changes in mRNA expression were calculated using the comparative ΔΔCT (crossing point) method.

Proteomic reverse-phase antibody microarray

We used an Antibody Microarray 500 (Clontech Laboratories, Mountain View, CA, USA) to determine the protein expression profiles in PAB- and Control-treated GCT stromal cells. Briefly, cell pellets were thawed and homogenized in nondenaturing extraction buffer without protease inhibitors. Protein concentration of the cell lysate was measured with the BCA Kit (Pierce Biotechnology, Rockford, IL, USA). To determine the difference in labeling efficiency, both PAB and control samples were labeled with either Cy5 or Cy3 (Amersham Biosciences, Piscataway, NJ, USA) to yield four sample mixtures. Unbound dye was removed by gel exclusion chromatography using the PD-10 desalting column (Amersham Biosciences). Again, protein concentrations of the four labeled samples were measured using the BCA protein assay. The average dye-to-protein ratio was estimated following the manufacturer's protocol. These four samples were combined into two final hybridizing mixtures for the two proteomics arrays by swapping the fluorescent Cy5 and Cy3 dyes in the labeling of PAB- and Control IgGs–treated samples. Antibody microarray slides were washed, dried, and scanned under the optimized photomultiplier tube (PMT) settings using a Typhoon Trio+ scanner (GE Healthcare, Piscataway, NJ, USA) to obtain Cy5 and Cy3 images, as described by the manufacturer.

Protein array analysis

We did grid alignment and initial quantification of array images using ImageQuant TL 7.0 software (GE Healthcare). Signal intensities for each coordinate on the array were determined by background-subtracted median fluorescence intensities from Cy5 and Cy3 channels. Log2-transformed normalized fluorescence intensity ratios were median-centered and averaged for the dye swapping experiments. The significance analysis of microarrays (SAM) method was first used to identify antibodies with statistically significant changes in intensity ratios relative to normal, by assimilating a set of t tests. Each antibody is assigned a test score on the basis of its change in intensity relative to the standard deviation of repeated measurements. The array data were also analyzed using a ratio approach for the detection of antibody biomarkers. The derived Cy5/Cy3 ratios from the two arrays were imported into the Antibody Microarray Analysis Workbook (Clontech Laboratories). An average international normalized ratio (INR) is calculated, which represents the abundance of an antigen in PAB-treated sample relative to that of the IgG control sample. As suggested by the manufacturer, INR values that are ≥1.3 or ≤0.77 indicate valid changes that signify differences in protein abundance. Those biomarkers that were identified as significant were then entered into the functional annotation and enrichment analysis. The SAM method with two-class t statistics was performed on the same dataset to identify antibody biomarkers that segregate control sample from the PAB-treated sample.

Functional annotation and enrichment analysis of protein array

We performed functional annotation and enrichment analysis of the biomarkers using the web-based WebGestalt software as described in the section on microarray data analysis (Functional Association and Pathways Analysis section, above). GenBank accession numbers of the high-ranking biomarkers were submitted to the online software to search for significant association of the biomarker clusters with relevant Gene Ontology (GO) terms, KEGG pathway, and WikiPathway.

Statistical analysis

GraphPad Prism software (GraphPad Software, Inc., La Jolla, CA, USA) was used for statistical analysis. All data are presented as mean ± standard error of the mean (SEM), and are representative of measurements that were performed on 10 different GCT patient samples (n = 10). To assess variations in real-time PCR gene expression, ANOVA and post hoc multiple comparison Tukey tests (p < 0.05) were applied. Measurements were normalized to the negative control. Values of p < 0.05 were considered to be statistically significant. Each experiment was performed at least three times for each tumor specimen.


cDNA microarray analysis

At the genome-wide level, we investigated the expression profile changes occurring in GCT stromal cells under various PTHrP treatments. Cultures treated for 24 hours with 10 µg/mL anti-PTHrP neutralizing antibody, 1 µM of PTHrP peptide, or 10 µg/mL IgG control vehicle were included in the microarray studies. We refer to these experiments with treatment of Control IgG, PTHrP peptide, and anti-PTHrP antibody as “Control,” “PP,” and “PAB,” respectively. The microarray data were initially extracted by the Partek Genomic Suite software to examine the differentially expressed genes in cells treated with Control, PAB, and PP. The Venn diagram in Fig. 1A shows a summary of overlaps of differentially expressed genes between the Control (Tumor), PP (Tumor-pp), and PAB (Tumor-ab) examined by cDNA microarrays over ±2 fold changes; p < 0.05). Interestingly, this Venn diagram showed that 33.9% of differentially regulated genes in PAB were specific for this condition (821 genes) when comparing it with the other microarray hybridization experiments done in this study (Fig. 1A). Fewer unique genes were identified in the other two treatment conditions; namely, 10.7% in PP (194 genes) and 12.8% in Control (231 genes) (Fig. 1A). The small percentage differences indicate that the microarray platform is able to detect changes in expression profiles specific to the PP and PAB treatments in GCT stromal cells with high precision and specificity and no bias effect due to the probe representation design.

Figure 1.

Overall manifestation of differentially expressed genes in cells treated with Control, PAB and PP. (A) Venn diagram shows overlap of genes differentially expressed in Control (Tumor), PAB (Tumor-ab), and PP (Tumor-pp) treatment in GCT stromal cells. Numbers of genes whose expression was significantly altered by PTHrP treatments by at least twofold (p < 0.05) are shown. (B) Overall gene expression level in terms of up- and downregulation in GCT stromal cells treated with PAB and PP compared to Control (p < 0.05). More genes were downregulated than upregulated.

In a total of 28,869 genes, the transcripts of 384 and 126 genes were downregulated in GCT stromal cells treated with PAB and PP, respectively, compared to Control, while the transcripts of 234 and 87 genes were upregulated by PAB and PP treatments, respectively (Fig. 1B). The variation across the samples in relative gene expression to the PAB and PP treatment can be visualized using two-way hierarchical cluster analysis (Fig. 2A). The non-normalized ratios of sample-specific to reference-specific fluorescence from the three independent experiment sets were grouped and clustered, together with hFOB, a human fetal osteoblastic cell line, for comparison. Unsupervised hierarchical clustering algorithm permits the clustering of individual tumor profiles based on their similarities to their coexpression of the PTHrP-affected genes. Each column represents a sample and each row a single gene on Fig. 2A. Red indicates upregulation; blue, downregulation; and gray, no changes in relative gene expression. The cluster includes only genes that gave reproducible measurements in all three data sets, defined by a 99% confidence threshold for correlation. We identified clusters of coordinately regulated genes that were common to the three experiments, demonstrating that the expression profile is similar in all samples: PAB, PP, Control, and hFOB, but still retains detectable changes. The top dendrogram shows that PP and Control from each individual GCT sample are clustered together and exhibit greater homology to the control osteoblast cell line (hFOB) than to the PAB-treated cells from the same patient.

Figure 2.

Two-way hierarchical cluster and principal component analyses. (A) Hierarchical clustering of data from the microarray analysis of gene expression in hFOB cells and GCT stromal cells treated with Control (GCT), PAB (GCT-Pab), and PP (GCT-Pp). The genes were identified after unsupervised and supervised analyses of the expression values, statistical analysis, and ANOVA with an FDR < 0.002. Samples with similar patterns of expression of the genes studied are clustered together, as indicated by the dendrogram. Each column represents data from each sample, with shades of red and blue indicating up- or downregulation of a given gene according to the color scheme shown below. Gray indicates no changes. Genes represented by rows were clustered according to their similarities in expression patterns for each treatment and cell type. The dendrogram showing similarity of gene expression among the treatments/cells is shown on top of the overview image, and relatedness of the arrays is denoted by the distance to the node linking the arrays. The gene tree shown at the left of the image corresponds to the degree of similarity (Pearson correlation) of the pattern of expression for genes across the experiments. (B) Three-dimensional PCA plot of all expression data showing the separation of samples is represented: hFOB (control = red), GCT stromal cells treated with Control (tumor = orange), PAB (tumor-ab = light blue), and PP (tumor-pp = brown). Each color dot represents serial triplicate specimens from each sample group. Ellipsoids are highlighted accordingly to each group. All samples in each treatment settle closely on the same ellipsoid.

Likewise, an alternative projection method such as the principal component analysis (PCA) can also be used to link samples to genes (Fig. 2B). The matrix created by the genes and experimental conditions with expression values as entries defines a multidimensional space where each gene represents a coordinate axis and the treatments are points located by their relative expression of these genes. Treatments with similar expression profiles occupy nearby positions in this space, and as more experiments are added to the data matrix, the number of points increases but the number of dimensions (genes) remains fixed. PAB, PP, and Control are on different ellipsoids on the PCA graph, but hFOB is spatially removed from all the GCT samples (Fig. 2B).

GO enrichment analysis

Due to the vast scope of this study, only analysis of the PTHrP neutralization treatment with the antibodies is reported in this section. In order to understand the biological annotations of differentially expressed genes, differentially expressed gene information of PAB and Control was imported into three software packages for GO analysis: Partek Genomic Suite, EGAN, and WebGestalt. A total of 618 differentially expressed genes (p < 0.05) associated with these GO clusters were ranked by fold change within the p value range (see Supplementary Table S1). Only 429 genes were identified by the published data (PubMed) and 371 of them matched to a GO term. The two main categories, biological processes (BP) and molecular function (MF), are shown in Fig. 3A, B. Within the BP GO categories, 179 genes (48.2%) were linked to metabolic process, 170 genes (45.8%) to biological regulation, and 138 genes (37.2%) were related to response to stimulus (Fig. 3A). Interestingly, 44 genes, 42 genes, and 10 genes were associated with cell proliferation, cell death, and cell growth, respectively. Within the MF GO categories, 179 genes (48.2%) were affiliated with protein binding and 80 genes (21.6%) were involved in ion binding (Fig. 3B). The overall BP GO graph is depicted in Fig. 3C.

Figure 3.

Gene ontology (GO) enrichment and annotation of differentially expressed genes. (A) GO Biological Process bar chart for the enriched GO categories. Gene ontology assessment and division of genes identified in GCT stromal cells in response to anti-PTHrP antibody treatment are grouped into similar Biological Process categories, and ranked with the total number of genes in each group as shown below each bar. Only the enriched GO categories with p < 0.01 and at least two genes are reported. (B) GO Molecular Function bar chart is shown in the same fashion. (C) GO tree for the enriched GO categories. The tree indicates the GO structure in only Biological Process. Terminal GO categories with p < 0.01 and at least two genes are identified as enriched and color-filled.

The genes we identified are closely interconnected in the human interactome, generating a “PTHrP interactome.” The connected components between PAB and Control that we highlighted contains 434 nodes and 478 edges with an averaged connectivity degree of >1 (Fig. 4A). The analysis revealed connectivity between many of the genes associated with PTHrP treatment in GCT stromal cells, but mapping these relationships showed that this connectivity is mediated through a few key factors in several important signaling pathways. The key pathways included both cellular metabolic and catabolic processes, followed by cell-cycle and apoptosis pathways. Each gene associated with the cell-cycle and apoptosis pathways (Fig. 4B, C) is plotted, respectively, according to the mean log2 ratio from the cDNA microarray data.

Figure 4.

Clusters of interacting genes. (A) Network analysis of PTHrP-regulated genes in GCT stromal cells. The PTHrP neutralization treatment network consists of 434 genes (nodes) significantly altered by a cutoff of twofold and 478 edges incident on them. The proteins are encoded by differentially expressed genes that interact directly with them in the “PTHrP interactome.” Topological connections of GO-enriched terms are displayed and nodes directly connected by a pathway in the GO graph are grouped together into GO main annotation groups. (B,C) Cell cycle and apoptosis are the two GO pathways shown here. The expression value for each gene is plotted. All other interacting connections and edges are hidden for enhanced readable presentation.

Pathways association

To reveal biological mechanisms and to identify pathways potentially involved in PAB-treated GCT stromal cells, we extracted from original expression data for all genes that are found to be differentially expressed at unadjusted p < 0.05 and showing expression change > ± 2.0. All the genes containing an Entrez Gene ID (434 genes) were mapped to the KEGG pathway, WikiPathways, and Pathway Commons analysis databases to define the signaling pathways that possibly participate in the observed cell death caused by PAB treatment. Results are provided in Table 2. Some important pathways that consistently appeared in all three pathway databases are in bold. As anticipated, two major intracellular pathways were involved in the PAB-stimulated cell death process in our model: cell cycle and apoptosis.

Table 2. Pathway Analyses of Differentially Expressed Genes Using Multiple Databases
Pathway nameaGenes (n)Entrez IDs
  • a

    Some important pathways that consistently appeared in all three pathway databases are in bold.

Pathway commons
 G1/S transition148318 9837 5696 51659 5427 5721 4174 7298 5720 5698 4176 1017 5557 26271
 Synthesis of DNA128318 9837 5696 51659 5427 5721 4174 5720 4176 5698 1017 5557
 DNA replication128318 9837 5696 51659 5427 5721 4174 5720 4176 5698 1017 5557
 DNA replication preinitiation128318 9837 5696 51659 5427 5721 4174 5720 4176 5698 1017 5557
 Unwinding of DNA58318 9837 51659 4174 4176
 M/G1 transition128318 9837 5696 51659 5427 5721 4174 5720 4176 5698 1017 5557
 S phase128318 9837 5696 51659 5427 5721 4174 5720 4176 5698 1017 5557
 E2F transcriptional targets at G1/S128318 5696 5427 5721 4174 7298 5720 4176 5698 1017 5557 26271
 E2F-mediated regulation of DNA replication128318 5696 5427 5721 4174 7298 5720 4176 5698 1017 5557 26271
 Cell cycle, mitotic199837 5721 64105 5720 9088 5698 4176 79682 5557 26271 8318 5696 23421 5427 51659 64946 4174 7298 1017
 DNA strand elongation69837 4176 5557 8318 51659 4174
 G1 phase115721 5720 4176 5698 5557 8318 5696 5427 4174 7298 1017
 Cyclin Dassociated events in G1115721 5720 4176 5698 5557 8318 5696 5427 4174 7298 1017
 Assembly of the prereplicative complex105721 5720 4176 5698 5557 8318 5696 5427 4174 1017
 IL27-mediated signaling events56775 3606 6773 3569 6772
 IL23-mediated signaling events77128 6347 6775 3606 3569 6772 2919
 Activation of the prereplicative complex64176 5557 8318 5427 4174 1017
 Pyrimidine biosynthesis (interconversion)47296 7358 2936 7298
 Metabolism of water-soluble vitamins and cofactors680025 1528 10135 3620 2936 8942
 G2/M checkpoints95721 5720 4176 5698 26271 8318 5696 4174 1017
 Activation of ATR in response to replication stress85721 5720 4176 5698 8318 5696 4174 1017
 Metabolism of vitamins and cofactors680025 1528 10135 3620 2936 8942
 Orc1 removal from chromatin75721 5720 5698 4176 5696 4174 1017
 Switching of origins to a postreplicative state75721 5720 5698 4176 5696 4174 1017
 Removal of licensing factors from origins75721 5720 5698 4176 5696 4174 1017
 Cyclin A/B1 associated events during G2/M transition75721 5720 9088 5698 26271 5696 1017
 Regulation of DNA replication75721 5720 5698 4176 5696 4174 1017
 Cell cycle checkpoints95721 5720 4176 5698 26271 8318 5696 4174 1017
 Type II interferon signaling (IFNG)143433 3627 5610 8370 6890 6772 10379 554313 5698 4938 2537 6773 3383 9636
 Selenium117296 5743 2936 6347 8942 3569 6289 6414 2876 3383 6291
 Benzo(a)pyrene metabolism51646 1645 8644 1545 2052
 Oxidative stress67296 2936 1728 2729 3162 2876
 Toll-like receptor signaling pathway93627 6696 7098 6373 3569 3576 6772 4615 6352
 Keap1-Nrf242730 1728 2729 3162
 Toll-like receptor signaling pathway - mir103627 6696 7098 6373 7128 3569 3576 6772 6352 4615
 DNA replication68318 5427 4174 4176 1017 5557
 G1 to S cell cycle control79134 8318 5427 4174 4176 1017 5557
 Glutathione metabolism42730 2936 2729 2876
 Proteasome degradation65696 5721 3134 7318 5720 5698
 Cell cycle79134 8318 4174 6502 9088 4176 1017
 Metapathway biotransformation102936 1646 1573 57016 2947 1645 8644 2876 2052 1545
 Tryptophan metabolism53620 8942 1573 1545 7453
 Osteoblast33690 4982 5156
 Osteoclast36696 3690 4982
 Prostaglandin synthesis and regulation45321 5743 306 5729
 DNA damage response59134 5888 9874 5371 1017
 Complement and coagulation cascades KEGG4718 623 2152 719
 Folic acid network37296 2936 2876
 Apoptosis53070 330 8743 834 843
 Senescence and autophagy43569 3576 2919 5055
 Estrogen metabolism21728 1545
 Selenium metabolism/selenoproteins37296 6414 2876
 Endochondral ossification46696 6772 808 4322
KEGG pathways
 NOD-like receptor signaling pathway107128 330 6347 3569 6355 3576 2919 3606 6352 834
 Cytokine-cytokine receptor interaction193976 6347 10673 8743 53833 3606 6352 5156 58191 10344 3627 6373 3569 6355 3576 4254 2919 4982 8793
 Antigen processing and presentation86891 5720 1520 5721 3134 972 3310 6890
 Toll-like receptor signaling pathway93627 6696 6373 7098 3569 3576 6772 6352 4615
 RIG-I-like receptor signaling pathway723586 3627 79132 64135 843 9636 3576
 Chemokine signaling pathway1258191 3627 10344 6373 6347 6355 3576 6772 57580 2919 6352 6773
 Cytosolic DNA-sensing pathway623586 3627 3606 6352 834 3569
 Glutathione metabolism62947 2730 2936 51056 2876 2729
 Systemic lupus erythematosus9718 8370 8342 554313 6737 8350 8358 715 8340
 Metabolism of xenobiotics by cytochrome P45061645 2947 1646 8644 2052 1545
 Complement and coagulation cascades6718 715 629 623 719 2152
 Jak-STAT signaling pathway93976 3569 6775 6772 10379 53833 3598 10253 6773
 Arachidonic acid metabolism55321 5743 8644 2876 1573
 DNA replication45427 4176 4174 5557
 Small-cell lung cancer69134 5743 330 1017 3673 6502
 Tryptophan metabolism43620 8942 7453 1545
 Epithelial cell signaling in Helicobacter pylori infection52919 245972 6352 523 3576
 Cell cycle78318 9134 9088 4176 1017 4174 6502
 Proteasome45720 5696 5698 5721
 Pyrimidine metabolism67296 7298 129607 5427 87178 5557
 Pathways in cancer139134 5743 330 5888 3569 3673 3576 5371 6502 6772 4254 1017 5156
 Linoleic acid metabolism35321 1573 57016
 Alanine, aspartate, and glutamate metabolism32744 9945 8528
 Hematopoietic cell lineage54254 3569 7037 3673 3690
 Apoptosis58743 4615 330 8793 843

Validation of microarray analysis

To validate the microarray data, 16 genes that were either upregulated or downregulated in the microarray analysis were analyzed by real-time PCR. These selected genes were CASP1, CASP10, CLU, CCNE2, CDK2, IL6, IL8, MCM7, MYD88, PTGS2, PSMB9, SESN3, STAT1, SKP2, TNFSF10, and TNFRSF10D. These genes were chosen because they are important players in both cell-cycle and apoptosis. The trend in expression levels of the genes detected by real-time PCR was consistent with the results obtained from the microarray expression analysis (Fig. 5A). Scatter plot analysis of the relative changes in expression as determined by real-time PCR and microarray, respectively, revealed a good correlation (Pearson R2 = 0.9466), thereby confirming the validity of the microarray data set (Fig. 5B). To further generalize the microarray data from the 4 patient samples, the expression patterns of PTHrP neutralization were validated with real-time PCR in a further 6 patient specimens. Out of the 16-gene validation list, seven genes (CASP1, CCNE2, CDK2, CLU, SESN3, STAT1, and TNFSF10) were carefully chosen to extrapolate the accuracy of microarray data in the general GCT population. These genes are heavily involved in apoptosis or inhibition of cell proliferation, and their roles in the outcome of cell death in GCT cells treated with anti-PTHrP antibody are discussed further in the Discussion section. The average expression level of the 6 patient samples is reproducible and consistent with the microarray data (Fig. 5C).

Figure 5.

Validation of gene expression levels derived from microarray expression analysis and real-time PCR. (A) Confirmation of significant difference in differentially expressed genes in GCT stromal cells after anti-PTHrP antibody treatment detected by microarray was carried out by real-time RT-PCR. The ΔΔCT method was used to calculate the real-time RT-PCR fold change using GAPDH mRNA for normalization, and all changes in expression are relative to the control without any treatment. Three independent real-time PCR runs were performed on individual samples. For both methods, normalized log2 ratios corresponding to each gene with p < 0.01 were plotted. Although variances in expression values are observed between the microarray and real-time PCR, all selected genes indicated the same highly significant difference in PTHrP counteraction treatment. (B) Scatter-plot of relative changes in gene expression as determined by microarray analysis and by real-time PCR. A high Pearson correlation (close to 1.0) is shown between the microarray and real-time PCR data. (C) Further generalization of the accuracy of microarray results using real-time PCR analysis on an additional 6 GCT patient samples. The average expression level of the 6 patient samples is reproducible and consistent with the microarray data.

Proteomics analysis

Using proteomics analysis as an additional validation tool, the pathways determined by microarray analysis were further confirmed. As shown in Fig. 6A, lysates prepared from GCT stromal cells under PAB or Control treatments were subjected to antibody microarray analysis to examine protein expression. Comparison of the spot patterns of the samples in PAB and Control indicated that, of more than 1000 total spots (equivalent to 500 different proteins), a set of consistently corresponding spots of 62 proteins were identified: 44 proteins that were upregulated and 18 proteins that were downregulated. These proteins have consistently exhibited detectable quantitative changes. None of the proteins identified from this proteomic approach could be mapped to the corresponding genes on the cDNA microarray due to a small number (10%) of the differentially expressed proteins that could be detected compared to the 618 genes detected in cDNA microarray analysis. Fifteen of these proteins that are involved in either cell cycle and apoptosis processes are plotted in Fig. 6B. These proteins are ALDH, Cathepsin D, Cytochrome c/Apaf-2, E2F-1, NEDD-4, Ubc9, UbcH7, AMPK b, G-protein alpha-t, p21 (Cip1/WAF1), Rap 1, APP-BP1, Caspase-6/Mch2, PTP1D/SHP2, and TRADD. A high correlation between the differentially expressed proteins under interchanging Cy3 and Cy5 labels represents the reproducibility of the proteomic microarray (data not shown).

Figure 6.

Proteomic antibody microarray analysis. (A) The superimposed dual-color proteomic antibody microarray image is shown. Lysates prepared from GCT stromal cells under PAB or Control treatments were analyzed on an antibody microarray platform. Equal amounts of proteins were labeled with either Cy3 or Cy5, and incubated with the antibody microarray as described in Materials and Methods. Each spot represents an individual antibody spotted in duplicate. (B) The differentially expressed proteins in GCT stromal cells after PTHrP neutralization were plotted against the expression difference in percentage (p < 0.01). Proteins involved in cell-cycle or apoptosis processes are indicated.

Finally, both cDNA microarray and proteomic array data were combined and analyzed for the overall picture in the biological pathways affected by PTHrP neutralization. Functional annotation suggested that anti-PTHrP antibody treatment terminated GCT stromal cell proliferation through targeting genes and proteins in pathways related to cytokine interaction, JAK-STAT, proteasome, ubiquitin, DNA replication, cell cycle, and apoptosis. In particular, probing these combined data in the KEGG pathways database showed significant enrichment of genes and proteins involved in both the cell cycle–mediated and apoptosis-mediated pathways (Fig. 7A, B). TRAIL, TRAIL-R, TRADD, MyD88, CASP6, CASP10, IAP, and CytC are affected in the apoptotic pathway (Fig. 7A), whereas Skp2, p21/Cip1, CycE, CDK2, E2F, Mcm5, Mcm7, and Myt1 are affected in the cell-cycle pathway (Fig. 7B). By integrating all these findings, a concept of the effect of PTHrP in the bone tumor cell has been proposed and is illustrated in Fig. 7C.

Figure 7.

KEGG apoptosis and cell-cycle pathway diagrams and integrated concept of the effect of PTHrP in bone tumor cells. (A) The KEGG representation of the apoptosis signal pathway integrated both the transcriptomic and proteomic expression data in GCT stromal cells treated with PTHrP neutralization. Genes and proteins affected by the PTHrP treatment are highlighted in red. A variety of different cellular signals initiate activation of apoptosis in distinctive ways, depending on the various cell types and their biological states. Solid arrows indicate direct and dashed arrows indirect activation. Crossbars indicate inhibition. Branching arcs go to alternative as well as to concurrent successors. Detailed legend information can be found on the KEGG website. (B) Cell-cycle signaling pathway according to the KEGG database is shown in the same fashion. (C) Schematic diagram depicts signal transduction pathways dysregulated by PTHrP neutralization in GCT. Light arrows indicate activation and dark knobs indicate inhibition from the findings of this study; dark arrows indicate established concepts from the literature.


PTHrP mediates bone disease associated with tumors and metastasis primarily through its effects on bone resorption.20, 21 PTHrP has also been shown to regulate tumor-relevant genes and to play a role in tumorigenesis, modulation of tumor progression, and response to treatment in breast cancer and bone metastases.22, 23 Indeed, our recent findings that PTHrP is expressed in GCT, together with the fact that neutralization of PTHrP in the cell environment inhibits cell proliferation and induces cell death, was demonstrated by various functional and mechanistic assays such as mitochondrial dehydrogenase assays, crystal violet assays, APO-1 ELISAs, caspase activity assays, flow cytometry, and immunofluorescent immunohistochemistry.10 All these functional data suggest that PTHrP serves to propagate proliferation and manifests the neoplastic process in GCT. To investigate the neoplastic role of PTHrP in bone tumor cells using GCT of bone as a model, anti-PTHrP neutralizing antiserum was used to study the pathways and mechanisms of PTHrP involved in GCT cell proliferation, apoptosis, and cell-cycle progression at the transcriptomic and proteomic levels.

A caveat for both transcriptomic and proteomic microarray analyses is the generation of datasets with high dimensionality, which in turn generates a proportion of type I errors.24 Higher-stringency RMA normalization and filtering in this study reduces the frequency of false-positive data, because noise is dependent on the observed signal intensity.25, 26 Another drawback for the proteomic antibody microarray is its limited multiplexing capacity; different emissions of two fluorescent dyes on the array integrate two conditions only. To offset the small sample size, multiple statistical tests, including ANOVA analyses, t tests, and correlation plots, were used to guard the quality of this study. These altogether provide a broader spectrum for integrative hypotheses while maintaining both the statistical confidence and biological relevance of the analyzed expression data.

A high correlation between the microarray analysis and validation using both real-time PCR and proteomic array supports the reliability and accuracy of the data in this project. It is crucial that the microarray results can be generalized in the GCT population and confirmed in 6 other randomly chosen GCT patient samples. It is important to note that the gene expression patterns observed in the data set are similar among all analysis methods. The nonoverlapping identification of genes and proteins between the cDNA and antibody microarray states the differences in expression at the transcriptional and translational levels. These findings highlight the fine balance between the specificity and sensitivity of normalization algorithms and filtering conditions, and the importance of implementing algorithms that coincide with the biological questions of interest.

In this study, we report that tumor cells treated with anti-PTHrP neutralizing antiserum and PTHrP peptide expressed many genes in common to the Control, yet cells treated with PTHrP antibody contained a larger group of distinct sets of genes than the other conditions. We believe these differences allow us to discriminate the outcomes of PTHrP neutralization-induced cell death in GCT, and suggest possible mechanisms of apoptotic control and cell-cycle regulation in the bone tumor cells. The two-way hierarchical cluster analysis revealed a considerable similarity in overall gene expression patterns between PTHrP peptide stimulation and control, with sufficient difference allowing their separation into respective subgroups. However, PTHrP neutralization identified a clear-cut clustering pattern distinct from all other treatment groups, as well as the control cell line hFOB. Likewise, the PCA disclosed a similar trend that multiple individual GCT samples of the same treatment condition assemble at their own ellipsoid planes.

Nonetheless, clustering tells us little about the internal mechanisms of the clusters that it finds. Thus, multiple bioinformatics tools were used to analyze changes in gene expression and identify important gene ontologies, processes, and pathways common to this anti-PTHrP–induced regulatory gene network. This analysis is important in that it may add to an understanding of the mechanisms of neoplasia in bone tumor cells. With a pragmatic focus on biological processes and molecular function, this analysis revealed common regulators and associated pathway components within the data set by portraying the network of genes and their functional or pathway-based relationship to one another. Among all pathways identified, we have focused on the apoptosis and cell-cycle linkages that offer relevant mechanistic insight into the orchestration of bone tumor cell survival and death. In particular, genes identified by GO from the microarray analysis as being important in apoptosis and cell proliferation were further validated and generalized to other GCT patient samples, and these genes reflect the observation and functional outcome of cell death upon anti-PTHrP antibody treatment as reported in our previous article.10

Our current study indicates that after a 24-hour treatment with anti-PTHrP antiserum, bone tumor cells are at the late stage of apoptosis. Caspase-1 and -10, upstream proteases in the caspase pathways essential for apoptosis, are downregulated, but their downstream activator caspase-6, responsible for cleaving cellular targets,27 was upregulated. Caspases play a central and posterior role in the direct transduction of apoptotic signals.27 Upstream of caspase-6, TRAIL/TNFSF10 is a death ligand for the extrinsic apoptotic pathway via the TRAIL-R/TNFRSF10D receptor,28 which was tremendously increased after PTHrP neutralization. Interestingly, others have also found that stimulation of the PTHrP receptor in HEK 293 cells triggers TRAIL-R–mediated apoptosis.29 In contrast, the adaptors TRADD and MyD88 of the tumor necrosis factor α (TNFα) and IL-1 extrinsic apoptotic pathways, respectively, were downregulated in this study, possibly in an attempt to rescue the cells from apoptosis.28

Others have shown that PTHrP lies upstream of the anti-apoptosis gene product Bcl-2, with PTHrP increasing Bcl-2 expression both in vitro and in vivo.30 We found that PTHrP neutralization also contributes to apoptosis by increasing expression of proapoptotic proteins on the mitochondria (eg, cytochrome C) and ubiquitin-mediated proteolysis (eg, ubiquitin-conjugating enzyme E2I), while diminishing the JAK-STAT pathway. A lower expression of STAT1, which binds to the Bcl-xL promoter through gp130 to activate its transcription and suppress apoptosis,31 was found in the GCT stromal cells exposed to PTHrP neutralization. Similarly, SHP2, LIF, and IL-6, part of the gp130/JAK-STAT signaling transduction pathway,31 all had significantly lower expression levels in the PAB group, as did the downstream proteasomes (PSMB-8, PSMB-9, PSME-1, and PSME-2) of STAT1. Thus, our findings consistently indentify downregulation Bcl-2 signaling as one of the mechanisms for promoting apoptosis in GCT cells as a result of PTHrP neutralization.

In addition to apoptosis, we have previously found with flow cytometry that the cell cycle is deregulated by anti-PTHrP treatment.10 Progression through the cell cycle is strictly regulated by the periodic and phase-specific abundance and activity of a defined set of two classes of protein-kinase complexes. The cell division kinases (CDKs), also referred to as cyclin-dependent kinases, are the catalytic subunits of these complexes, whereas the cyclins function as the regulatory subunits. The G1 (gap) phase preceding the DNA synthesis (S) phase and the mechanisms that drive the cells across the restriction point are crucial for the cell's fate toward division, differentiation, senescence, and most importantly, apoptosis.32 Cyclin E CCNE2 associates with cell cycle checkpoint kinase CDK2 and the activated complex phosphorylates key substrates such as Rb to facilitate transcription of target genes necessary for DNA replication and progression of the cell cycle from G1- to S-phase.32 Both cyclin E and CDK2 expression significantly decreased after PTHrP neutralization in our experiments. The cyclin E and CDK2 complex is under E2F-dependent regulation,32 and E2F was found to be significantly downregulated at the protein level by PTHrP neutralization.

Cell cycle progression is also controlled by CDK inhibitors, the Cip/Kip and the INK4 families, which negatively regulate G1-phase progression by forming complexes with CDKs and preventing S-phase entry.32 Cip/Kip p21, a critical mediator of antiproliferative signals that arrest the cell cycle and controlled by the tumor suppressor protein p53, was found to be overexpressed following PTHrP neutralization. We have also found that Skp2 of the F-box families, regulated by STAT131 and responsible for ubiquitin-dependent proteolysis of the complex of Cip/Kip p21, cyclin E, and CDK2,33 was downregulated, implying a loss of timed destruction of p21. Thus, our results indicate a central role for PTHrP in G1/S cell-cycle progression in GCT cells. This finding is consistent with previous findings of PTHrP peptide stimulation in enhancing cell proliferation via amplifying cyclin E and CDK2 in human pancreatic islets.34 Others have found that PTHrP-activated GPCR is able to regulate the expression, activity, localization, and stability of the cell-cycle regulatory proteins cyclin E and CDK2 via three intracellular pathways (PKA, PKC, PI3K/AKT/PKB) that mediate regulation of G1- to S-phase progression.35

In this study, we demonstrated by both transcriptomic and proteomic microarray analyses that PTHrP markedly regulates genes involved in cell cycle and apoptosis in bone tumor cells. PTHrP interferes with multiple cell survival and apoptosis signaling pathways by inhibiting both death receptor and cell cycle–mediated apoptosis signaling. We propose that PTHrP neutralization exhibits anticancer potential by the regulation of cell cycle and apoptosis in bone tumor cells. These data suggest that unique cell cycle and apoptosis controls are instituted in the bone tumor cell life span and that understanding of their regulatory mechanisms could lead to directed therapeutic approaches for various bone disorders. Of clinical relevance, these data constitute a starting point for clinical evaluation of anti-PTHrP strategies against primary GCT of bone and other PTHrP expressing bone tumors in vivo.


All authors state that they have no conflicts of interest.


This work was supported by the Canadian Institutes of Health Research (CIHR); Hamilton Health Science New Investigator Fund; Hamilton Health Science Early Career Award; Juravinski Cancer Centre Foundation; and McMaster University Surgical Associates grant. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We are grateful to Dora Ilieva for helpful discussions.

Authors' roles: Study design: IWYM and MG. Study conduct: IWYM. Data collection, analysis and interpretation: IWYM. Drafting manuscript: IWYM and MG. Revising manuscript content and approving the final version: MG, IWYM, and RET. MG and IWYM take responsibility for the integrity of the data analysis.