The 12T-10 line of the LPB-Tag mouse model of prostate cancer was generated using a transgene that consists of the rat probasin promoter driving the SV40 large T antigen with deletion of the small T antigen 3. These mice develop low-grade and high-grade prostatic intraepithelial neoplasia at 2-5 month of age, with progression to invasive and metastatic, high-grade, androgen-independent carcinoma demonstrating neuroendocrine differentiation at 6-14 months of age. A primary prostate tumor from the ventral prostate of a 12T-10 transgenic mouse was used to establish an allograft model by implantation subcutaneously in male athymic nude mice 4. After 18 weeks, the allograft was passaged to another male nude mouse and the process was repeated to establish the NE-10 line 4. Initial passages in all mice developed grossly visible metastases to liver and micrometastases to lung by twelve weeks after implantation. All metastases from the allografts were histologically similar to the metastasis seen in the 12T-10 mice. Later allograft passages showed histologically identical features to the early passages; however, fewer allografts developed metastases. By passage 15, metastatic potential was completely lost, at least up to the point where it was no longer feasible to maintain the mice due to the size of the subcutaneous tumors. Tumors consisting of non-metastasizing subcutaneous allografts (SQnon-met), metastasizing subcutaneous allografts (SQmet), and liver metastases (LiverMet) were collected at 12 weeks post-implantation and snap-frozen for array CGH. All procedures involving mice were approved by the Vanderbilt University Medical Center and Fred Hutchinson Cancer Research Center Institutional Animal Care and Use Committees.

Array CGH was initially carried out using mouse bacterial artificial chromosome (BAC) arrays produced by the Genomics Shared Resource, Fred Hutchinson Cancer Research Center as described 6. The mouse BAC clone set was obtained from A. Bradley, Wellcome Trust Sanger Institute 7 and provides an average resolution of 1 Mb. DNA was isolated using the Gentra Puregene genomic DNA purification kit (Qiagen, Valencia, CA). For the BAC arrays, three SQnon-met tumors (passage 19), three SQmet (passages 4, 9, and 12), and four LiverMet (two each from passages 9 and 12) were analyzed. One matched SQmet and LiverMet from the same nude mouse at passage 9 was included in the analysis. Pooled DNA obtained from normal kidney from four different CD-1 males was used as a reference. DNA labeling, hybridization, scanning, and data analysis was performed as described previously 6.

Array CGH was repeated on a subset of two tumors each from the SQnon-met and LiverMet groups using the Agilent mouse 105K oligonucleotide CGH arrays. These arrays were designed using UCSC mm7 (NCBI build 35, August, 2005), and have an average probe spacing of 15 Kb. One ug of RsaI/AluI digested DNA was labeled with either Cy3 or Cy5 using the BioPrime Array CGH genomic labeling system (Invitrogen Corp., Carlsbad, CA). Approximately 4 ug each of Cy5 labeled test (tumor) DNA and Cy3 labeled reference (normal female C57Bl/6 liver) DNA was combined with 25 ug of mouse Cot-1 (Invitrogen Corp.), Agilent 10× blocking agent and 2× hybridization buffer (Agilent Technologies, Santa Clara, CA) to a final volume of 260 ul. Hybridization and washes were carried out according to the Agilent oligo array-based CGH protocol v. 4.0. Scanning was performed on an Agilent scanner and data extraction was carried out using Agilent feature extraction software v.9.1, employing linear and Lowess normalization. Results were analyzed and chromosome plots generated using CGHanalytics software v. 6.0. The Z-score algorithm with the threshold set at 2.0 and a 2 Mb window was used for determining gains and losses. The LPB-Tag tumors were generated on an outbred CD-1 background, and we did not have non-neoplastic DNA available from the mouse whose tumor was used to establish the allograft. As C57Bl/6 reference DNA was used for the oligonucleotide array CGH experiments, gains and losses smaller than 2 Mb were ignored in order to avoid mistaking benign copy number variants between mouse strains from acquired copy number alterations in the tumors.

To compare array CGH profiles between the different groups of tumors, the Differential Gene locus MAPping software (DIGMAP version 2.0) was used to analyze array CGH data sets 8. Using the default parameters specified in the program, DIGMAP displayed array CGH data as a heat map based on log₂ ratio between test and reference DNA for each tumor. The chromosome regions with significantly different log₂ ratios were marked as differential flagged regions (DFRs) by direct visualization and computational screening using a T-test based sliding window (TTSW) analysis. For this analysis, a window size of 40 genes was used and DFRs represent regions greater than 3 standard deviations from the whole genome average T score.

Expression arrays were performed on two SQnon-met tumors (passages 11b, 18c) and four SQmet tumors (passages 2, 5, 14a, 14c). Allograft passages were labeled a, b, c, etc. when the same graft passage was implanted into more than one mouse. As a control, mouse reference RNA from two postnatal day 1 mice was pooled. RNA was isolated using the RNeasy kit (Qiagen Valencia CA), including treatment with DNase. RNA was quantified using a NanoDrop (ThermoScientific). All samples analyzed had a A260/230 ratio greater than 1.8. RNA quality was analyzed on an Agilent 2100 BioAnalyzer. All samples analyzed had a 28S peak greater than 18S peak, or an RIN number greater than 7. The 16 k mouse arrays used for the study were printed at the Vancouver Microarray Facility using Operon's 16 K 70 mer oligomers printed on aminosaline slides. Ten ug of total RNA was labeled using Genisphere's Dendrimer 350 expression array detection kit for microarrays according to the manufacturer's protocol. Samples were co-hybridized with ten ug of the above described mouse reference RNA. Arrays were pre-hybridized in 5 × SSC, 0.1% SDS and 0.2% BSA for 45 minutes. Pre-hybridization buffer was washed off with 3 × 30 second water washes and a 2 minute wash in isopropanol. Slides were spun dry at 2000 rpm for 4 minutes. The samples were applied to arrays containing 60 × 22 mm lifter slips (Erie Scientific). Slides were subsequently treated as described in the Genisphere Array 350 expression array detection kit for microarrays. Arrays were scanned on a Axon 4200AL scanner (Molecular Devices). Image intensities were extracted using ImaGene V8.0 software (BioDiscovery).

To identify chromosome regions that demonstrate differential gene expression between metastatic and non-metastatic tumors, expression array datasets were analyzed between SQnon-met and SQmet mice using DIGMAP as described above for array CGH data. A TTSW genome scan was also performed as described above.

A total of 146 known genes in the chromosome 2 differential region of deletion (DFR) between metastatic and non-metastatic tumors (nucleotides 122,316,740-139,585,560) were retrieved from the UCSC mm7 database http://genome.ucsc.edu. Two bioinformatics filters were designed to determine which of these 146 genes might function as candidate metastasis suppressor genes. A functional filter utilized the Gene Ontology (GO) database http://www.geneontology.org, PubMed http://www.ncbi.nlm.nih.gov/PubMed/, and Ingenuity Pathway Analysis (IPA, http://www.ingenuity.com) to evaluate the 146 genes for potential metastasis suppressor function. A parallel filter employing our recently developed human cancer expression signature database (EXALT) was used to independently validate the genes as candidate metastasis suppressors in silico 9. A weighted score was assigned to each gene for both filters (see Additional file 1).

Total RNA was isolated using the RNeasy Mini kit (Qiagen, Valencia, CA) according to the manufacturer's protocol, including the recommended DNase treatment step. Three ug of RNA was reverse transcribed using the ReactionReady First Strand cDNA Synthesis Kit (SuperArray Bioscience Corp., Frederick, MD). Primer sets for mouse Hprt1, Slc27a2, Mal, Snrpb, and Rassf2 were purchased from SuperArray (proprietary primers, sequence not disclosed).

PCR reactions were carried out in triplicate in a 25 μL volume using SYBR Green Master Mix (SuperArray). The standard two-step amplification with an annealing temperature of 60°C was performed in an ABI 7900 PCR machine. Hprt1 was chosen as an endogenous control gene because Hprt1 expression values were stable among test samples in the microarray expression data set.

To allow for a comparison between samples and the two groups, quantities of all target genes in the test samples and a common reference RNA (Mouse XpressRef Universal Total RNA, SuperArray) were normalized to the corresponding Hprt1 levels. Relative expression levels (fold changes) were calculated using the relative standard curve method as outlined in the manufacturer's technical user manual (SuperArray). A standard curve was generated using the fluorescent data from 10-fold serial dilutions of the common reference RNA sample. Four tumors each from the metastasizing and non-metastasizing allografts were analyzed. Statistically significant differences in expression between the two groups were determined using a Student's T-Test.

Copy number alterations in tumors were identified by array CGH performed using BAC arrays for all tumors, and oligonucleotide arrays for a subset of tumors. Array CGH using both BAC and oligonucleotide platforms detected multiple gains and losses in both non-metastatic tumors (SQnon-met) and tumors with metastatic potential (SQmet and LiverMet). Both array platforms uncovered similar chromosome imbalances, but the oligonucleotide arrays allowed us to more precisely map regions of gain and loss. Many copy number alterations were common to both groups of tumors, including deletions involving regions of chromosomes 1, 2, 3, 4, 8, 13, 16, 17 18, 19, whole chromosome loss of 14, and gains of chromosome 8 (Figure 1 and Table 1); however, differences in copy number changes between the tumors with and without metastatic potential were also uncovered (Figure 1 and Table 1). One notable difference between the two groups of tumors was a differential region of deletion in distal chromosome 2. Both the metastatic and non-metastatic tumors shared a common 27.4 Mb interstitial deletion of distal chromosome 2, from bands F3 to H3 (139,585,560-167,088,181 bp); however, the metastatic tumors (LiverMet) showed a larger 45 Mb deletion (122,316,740-167,088,181 bp) that extended proximally to band E5 (Figure 2). The non-metastatic tumors showed clonal heterogeneity for both the larger and smaller deletions, with low-level mosaicism for the larger deletion (Figure 2 top), but the clone or clones with the larger deletion were enriched in the metastatic tumors (Figure 2 bottom).

To obtain differential gene expression profiles between non-metastatic and metastatic NE-10 tumors, the different groups of tumors (SQnon-met, SQmet, and LiverMet) were profiled using gene expression microarrays and analyzed by DIGMAP. A T-test sliding window whole genome scan was performed and plotted by chromosome (see Additional file 2). Based on this analysis, chromosome 2 had the greatest number of regions showing the largest difference in gene expression between metastatic (SQmet and LiverMet) and non-metastatic (SQnon-met) tumors. A detailed T-test sliding window plot of chromosome 2 revealed the highest DFR expression peak in band H4, from 179 to 181 Mb (see Additional file 2). This region is distal to the common region of deletion found in both the metastatic and non-metastatic tumors. Three other smaller peaks were also located slightly more centromeric, from 155 to 168 Mb, within the common region of deletion. Thus, regions of differential gene expression between metastatic and non-metastatic tumors were located both within and distal to the common chromosome 2 deletion; however, we were unable to detect any clusters of genes that demonstrated down-regulation in the metastatic compared to the non-metastatic tumors. There were also no significant differences in gene expression detected in the differential region of deletion on chromosome 2 between the metastatic and non-metastatic tumors.

Metastasis suppressor genes are expected to show inactivation or decreased activity in metastatic tumors because of genetic or epigenetic changes that result in loss or down-regulation of expression. The finding of a differentially deleted region on chromosome 2, potentially resulting in loss of a metastasis suppressor gene in the metastatic tumors, led us to focus on this region for more detailed analysis. Given that only a subset of genes in the differentially deleted region was represented on the expression arrays, we employed an informatics approach to identify all potential candidate metastasis suppressor genes in the region of interest (chromosome 2 E5-F3; nucleotides 122,316,740-139,585,560). Two parallel bioinformatics filters were designed to carry out this analysis. Out of a total of 146 genes in the region of interest, 11 candidate metastasis suppressor genes were identified based on the function-based filters, while 22 candidates were identified using a filter that employed a cancer expression signature filter using the EXALT database (see Additional file 3). A final list of four candidate genes (Slc27a2, Mall, Snrpb, and Rassf2) was chosen based on having the highest total score.

Quantitative RT-PCR for the four candidate metastasis suppressor genes was performed in tumors with metastatic potential (SQMet) and tumors without metastatic potential (SQnon-Met). Metastatic liver tumors were not analyzed due to the presence of contaminating normal liver tissue in some of the tumors. All four candidate metastasis suppressor genes demonstrated decreased expression in the metastasizing tumors compared to the non-metastasizing tumors, with Slc27a2 and Snrpb showing statistically significant decreased expression in the metastasizing tumors (Figure 3).