The microarray dataset GSE14827 was downloaded from the GEO database (National Center of Biotechnology Information, Bethesda, MD, USA; www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE14827). The dataset included biopsy samples from 9 patients with OS who developed pulmonary metastases ≤4 years following neoadjuvant chemotherapy and curative resection, and 18 patients who did not relapse in this time frame (7). All tumor samples in the dataset were obtained through diagnostic incisional biopsies from primary sites of OS prior to neoadjuvant chemotherapy at the National Cancer Center Hospital (Tokyo, Japan) between March 1996 and September 2007 (7).

Raw microarray data and probe annotation files from the dataset were downloaded for analysis. The data downloaded included the expression levels of 54,613 probe sets across 27 samples from patients with OS (age, 8–38 years; 13 males and 14 females) that were analyzed using the GeneChip^® Human Genome U133 Plus 2.0 array (Affymetrix, Inc., Santa Clara, CA, USA). In the present study, the fold change values for each probe were determined. The fold change value for each probe is the average expression value of OS samples divided by that of normal samples. Subsequently, each probe was converted into an Entrez Gene ID (www.ncbi.nlm.nih.gov/gene). If a gene mapped onto >1 probe, the mean expression value of all corresponding probes was considered the expression value of the gene. Genes with a fold change value of >1.5 or <0.667 were identified as DEGs.

The Subpathway Miner R package (version 1.0; cran.r-project.org/src/contrib/Archive/SubpathwayMiner), which is a flexible subpathway identification software, was used to obtain subpathways associated with OS metastasis (12). DEGs between patients with OS who did or did not develop metastases were imported into Subpathway Miner, which identified significantly enriched subpathways using hypergeometric tests (12). This software converts pathway structure data from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (www.genome.jp/kegg) into undirected R graph objects. In the pathways, proteins are considered nodes and any two nodes belonging to the same reaction are connected by an edge. Finally, Subpathway Miner divides the entire pathway into subpathways using the ‘k-clique’ method, which is defined as a sub-graph in which the distance between any two nodes is <k (12). In the present study, k was set at 3, thus the distance between proteins in a single subpathway was <3. P<0.05 was considered to indicate a statistically significant subpathway.

The global associations between drugs and metabolic subpathways were obtained from a study performed by Li et al (13), in order to identify drugs that serve a role in OS metastasis-associated subpathways. In the above study (12), microarray data from cancer cells treated with or without small molecule drugs were obtained from the CMap database (Broad Institute, Cambridge, MA, USA) (10) and DEGs were identified using a fold change threshold of >2 or <0.05. Subsequently, drug-affected subpathways were identified for each drug if the corresponding drug-affected genes could be significantly (P<0.01) enriched in the subpathways using Subpathway Miner (13). A total of 3,925 associations were identified between 488 drugs and 403 subpathways in this study (12). These associations were downloaded and the intersecting subpathways between OS metastasis and the drugs were obtained for use in the present study. Among the 403 subpathways, 9 subpathways corresponding to 98 drugs were associated with OS metastasis. Detailed information and Anatomical Therapeutic Chemical classification for these drugs were obtained from the Drug Bank database (www.drugbank.ca).

U2OS cells (obtained from the American Type Culture Collection, Manassas, VA, USA) were starved through culturing in serum-free medium [RPMI-1640 (MGC-803); Gibco; Fisher Scientific, Inc., Waltham, MA, USA] for 12 h under standard conditions, as described by the American Type Culture Collection. The cells were subsequently seeded into 6-well plates at a density of 2.5×10⁵ cells/well. Following treatment with 100 µmol lansoprazole for 24 h at 37°C, cells were fixed and stained and their invasiveness was investigated using a Transwell chamber (Corning Incorporated, Corning, NY, USA). BD Matrigel™ Basement Membrane Matrix (50 mg/l; BD Biosciences, San Jose, CA, USA) was diluted with serum-free medium at a ratio of 1:8, and each Transwell chamber was coated with 60 µl of this solution. Prior to use, the polycarbonate membrane (pore size, 8 µm) was hydrated with 50 µl of serum-free medium containing 10 g/l bovine serum albumin (Beyotime Institute of Biotechnology, Haimen, China) at 37°C for 30 min. Treated cells (8×10⁵cells/well) were subsequently seeded into the upper chamber in 200 µl of serum-free medium, and 600 µl of medium containing 10% fetal bovine serum (Stemcell Technologies, Inc., Vancouver, BC, Canada) was added to the lower chamber as an attractant. Following incubation for 24 h at 37°C in 5% CO₂, the cells that had invaded the lower surface of the filter were fixed with 75% ethanol and stained with 0.1% crystal violet. Cells were counted by eye in three random areas in each chamber (magnification, ×200) using a light microscope.

The migration ability of U2OS cells was determined using a wound healing assay. U2OS cells were seeded into 6-well plates (2×10⁵). When the cells reached 70% confluence they were treated with lansoprazole (200 µM). The control group was treated with DMSO.A sterile 10 µl pipette tip was used to draw straight lines on the confluent monolayer. Photographs of the wounds were taken at 0 and 24 h post-treatment. The width of the wound at these time points was measured to measure cell migration.

All data were from ≥3 independent experiments and are presented as the mean ± standard deviation. The differences between groups were analyzed using the Student's t-test or one way analysis of variance with SPSS 19.0 software (IBM SPSS, Armonk, NY, USA) and Graph-Pad Prism 5.0 (GraphPad Software, Inc., La Jolla, CA, USA). P<0.05 was considered to indicate a statistically significance difference.

The microarray dataset GSE14827 from the GEO database was downloaded and used to identify DEGs between metastatic and non-metastatic OS samples. A total of 546 genes were identified as DEGs (data not shown).

To investigate the underlying mechanism of OS metastasis, dysregulated metabolic subpathways were identified through the use of Subpathway Miner (12). Following integration of the DEGs with metabolic subpathways (k=3), 9 enriched metabolic subpathways corresponding to 6 entire metabolic pathways were identified (Table I). The 9 metabolic subpathways were considered significantly associated with the development of OS metastasis (all P<0.05; Table I).

To determine if any existing drugs could be repositioned to treat metastatic OS, candidate small molecule drugs that were capable of targeting OS metastasis-associated metabolic subpathways were identified. The global associations between drugs and metabolic subpathways were downloaded from a study performed by Li et al (12), in which 403 metabolic subpathways affected by 488 known drugs were identified through the use of Subpathway Miner (k=3).

A total of 9 overlapping subpathways associated with OS metastasis and affected by small molecule drugs were identified. In addition, 98 small molecule drugs were identified in these subpathways (Table II). Detailed information and Anatomical Therapeutic Chemical classification for these drugs were obtained from the Drug Bank database (www.drugbank.ca).

A bipartite network of the drugs identified and the overlapping metabolic subpathways was built (Fig. 1). In this network, certain drugs could affect several metabolic subpathways. For example, lansoprazole and omeprazole perturbed 8 and 7 subpathways, respectively, while others affected fewer subpathways.

In order to investigate the effect of the drugs identified on the invasiveness of OS cells, a Transwell assay was performed on U2OS cells following treatment with lansoprazole, due to it affecting the most subpathways. As illustrated in Fig. 2, lansoprazole significantly inhibited the invasion of U2OS cells compared with the control group (P=0.02) and did so in a dose-dependent manner. Furthermore, cell migration was assessed using a wound healing assay. The results of this assay demonstrated a marked reduction in cell migration following lansoprazole treatment (Fig. 3).