Conditional mouse osteosarcoma, dependent on p53 loss and potentiated by loss of Rb, mimics the human disease

Osterix-Cre-mediated deletion of p53 and pRb results in OS

Based on the near ubiquitous loss of function of p53 and pRb pathways in human OS, we hypothesized that osteoblast-specific loss of these genes may be sufficient to initiate OS development (murineOS, mOS). We made use of conditional (floxed) alleles of both p53 (p53^fl/fl) and pRb (pRb^fl/fl) to allow for the tissue-restricted inactivation of these genes following Cre expression (Jonkers et al. 2001; MacPherson et al. 2003; Sage et al. 2003). To achieve osteoblast-restricted deletion, we utilized the Osterix-Cre transgenic mouse (Osx-Cre) to direct Cre expression to committed osteoblast progenitors (Rodda and McMahon 2006). Osterix is a bone-specific transcription factor required for osteoblast development, and the Osterix-Cre transgene expresses Cre recombinase in a manner that follows that of endogenous Osterix (Nakashima et al. 2002). Animals were intercrossed to develop Osx-Cre⁺p53^fl/flpRb^fl/fl animals with all intermediate genotypes also generated in addition to Osx-Cre⁺ animals bearing only a single conditional allele. All genotypes were born at the expected ratios and mutant animals are viable, fertile, and developmentally normal with the exception of a slight growth delay in Osx-Cre⁺ animals compared with Osx-Cre⁻ animals, although this did not affect fertility and largely resolved with age.

We have not observed the development of mOS in Osx-Cre⁺pRb^fl/fl and Osx-Cre⁺pRb^fl/+ mice with an observation time of ∼18 mo of age (Fig. 1B; Table 1). Unlike osteoblast-restricted deletion of pRb, both Osx-Cre⁺p53^fl/fl and Osx-Cre⁺p53^fl/+ animals develop mOS. Osx-Cre⁺p53^fl/+ mice develop disease with low penetrance and long latency. However, deletion of both alleles of p53 resulted in mOS development with complete penetrance and an average survival to 292 d of age (Fig. 1B; Table 1; Supplemental Table 1). Heterozygosity for both p53 and pRb shortened the latency and increased the penetrance of mOS compared with either heterozygous mutant alone (Fig. 1B; Table 1; Supplemental Table 1).

Osx-Cre⁺p53^fl/flpRb^fl/+ and Osx-Cre⁺p53^fl/flpRb^fl/fl animals develop mOS at ∼4 mo of age, a time point within the first quarter of the expected lifespan of a laboratory mouse. Completely penetrant mOS development was observed in both genotypes (Fig. 1B,C; Table 1; Supplemental Table 1). Interestingly, Osx-Cre⁺p53^fl/+pRb^fl/fl animals are much more susceptible to OS development than double heterozygous animals with 77% penetrance, whereas complete loss of p53 resulted in 100% development of mOS (Fig. 1B; Table 1). We also observed the development of adipogenic tumors, restricted to the Osx-Cre⁺p53^fl/flpRb^fl/fl animals (five of 24 animals to date, ∼20%), which occurred on the outer chest or abdomen (data not shown). All mice presenting with these tumors also had OS. Primary mOS tumors and mOS cell lines derived from them were serially transplantable when injected subcutaneously into Rag2^−/− immunodeficient animals (data not shown). Thus, Osx-Cre-dependent deletion of p53 and pRb results in the development of completely penetrant malignant OS, and this disease is strictly dependent on disruption of p53.

Multifocal nature of disease and metastatic potential

The most frequent site of mOS was the jaw and head, followed by the hind leg/hip and ribs and vertebra (Fig. 2a; Supplemental Table 2). This contrasts with human OS, in which jaw tumors account for only 9% of all tumors (Unni et al. 2005). When the tumor site data are analyzed exclusive of those tumors arising on the head, the site distribution more closely approximates that observed in human OS (Tan et al. 2006). The tumors presented with histology most consistent with that of human medullary OS. Tumor histology ranged from almost purely fibroblastic tumors with minimal mineralization/osteoid to tumors with marked mineralization of the osteoid (Fig. 2b–i). The distribution of the mOS was typically metaphyseal, with growth into the central medullary cavity and with extraosseous extension into the soft tissues. As in human OS, the tumor bone was woven bone formed by the tumor cells. Areas of cartilage formation were also observed within the mOS, a finding also seen in human tumors. The tumor cells were predominately spindle shaped with moderate to marked atypia (Fig. 2d,e). Mitotic figures were variable in number, and abnormal mitoses are seen among the more atypical cells. Occasional multinucleated giant cells were seen among the tumor cells. In nearly all respects, the mOS bears a close histopathological resemblance to human OS (Unni et al. 2005).

A major cause of mortality in patients with OS is metastasis, which is often present at initial diagnosis. Human OS shows a preference for metastasis to the lung and brain. We observed metastasis in 9.4% of all mOS-bearing mice. The metastatic frequency was >39% among animals with primary tumors at sites other than the head. The most frequent site of metastasis was the lung (seven of nine) followed by the liver (six of nine), and one animal presented with disseminated metastasis throughout the intestinal tissues (Fig. 2j–q). Several animals had metastatic disease in both the lung and the liver, which were apparent by microPET (Fig. 2k; Supplemental Fig. 1). Metastatic disease was observed across genotypes, however it most often presented with a longer latency than the genotype median and rarely in animals that developed a tumor on the jaw or snout, which significantly decreased lifespan (Table 1). We did not observe metastasis in Osx-Cre⁺p53^fl/flpRb^fl/fl animals, perhaps due to extremely rapid tumor development in these animals, which may not provide sufficient time for metastatic disease to become evident. The expression of Cre from the Osx-Cre transgene is able to be temporally repressed by the administration of doxycycline (dox). When Osx-Cre⁺p53^fl/flpRb^fl/fl animals were maintained on dox from conception until 4 wk of age, preventing gene inactivation until the removal of dox, the time to development of OS was delayed by ∼100 d and all animals presented with metastatic OS but no animals had tumors on the head (Supplemental Fig. 2). Cell lines derived from the metastatic lesions were transferred subcutaneously to Rag2^−/− immunocompromised recipients, and mOS readily ensued (data not shown).

Multifocal lesions were observed in a significant number of animals (22.3%). Their appearance did not correlate with genotype. Tumors most frequently presented as multiple lesions on the jaw and snout (Supplemental Fig. 3; Supplemental Movie 1). To enable more accurate assessment of tumor burden, we utilized micro-PET/CT imaging to determine noninvasively the numbers of tumors. For microPET imaging, fluorine-18 sodium fluoride (Na¹⁸F) was used as the imaging agent (Brenner et al. 2004). Na¹⁸F binds to the hypoxanthine matrix of bone with highest deposition at sites of turnover (Even-Sapir et al. 2004; Lim et al. 2007). mOS showed highly avid uptake of Na¹⁸F, which directly corresponded to lesions as visualized by microCT (Fig. 3A–C). Histology was performed verifying the detected lesions as mOS. Interestingly, Na¹⁸F uptake was most pronounced in regions of the tumor that contained less areas of mineralized osteoid as determined by microCT (Fig. 3C). The capacity of microPET to detect small lesions, such as that on the calvaria in Supplemental Figure 3, which would not otherwise be readily detectable on conventional autopsy, suggested that the estimate of tumor burden may underrepresent the true burden due to the microscopic, multifocal nature of some lesions.

Given the sensitivity of Na¹⁸F microPET, we imaged several 2-mo-old animals to determine when lesions developed. The Osx-Cre⁺p53^fl/flpRb^fl/fl animal we assessed had a detectable jaw lesion at 2 mo of age, which was not externally visible, while two Osx-Cre⁺p53^fl/flpRb^fl/+ animals did not have detectable lesions at this time point (Fig. 3D; Supplemental Movies 2, 3). By 3 mo of age, one of the Osx-Cre⁺p53^fl/flpRb^fl/+ animals had also developed a PET positive lesion in the tibia (Supplemental Movie 4). Collectively, these studies demonstrate that the tumor burden may be underrepresented using conventional autopsy coupled with histology and that the mOS can develop by at least 2 mo of age, consistent with the pediatric or adolescent nature of human OS.

Cytogenetic complexity of mOS

A feature of human OS is cytogenetic complexity with the absence of recurrent clonal chromosomal translocations (Bridge et al. 1997; Helman and Meltzer 2003). To determine whether the mOS also displayed cytogenetic complexity, we performed spectral karyotyping (SKY) on early passage cell lines derived from primary tumors. mOS cells were isolated by enzymatic digestion of primary tumor tissue, placed into culture, and then following metaphase arrest, cells were assessed for cytogenetic integrity using SKY. We assessed ploidy and the presence of translocations in cell lines derived from six independent tumors. mOS cells consistently exhibited marked aneuploidy, a characteristic of human OS (Supplemental Fig. 4; Supplemental Table 3). Two of the tumors assessed showed aberrations of chromosomes 8 and 14, with multiple small chromosomes and fragments. Translocations were present, although these were nonclonal and, in general, nonreciprocal (Supplemental Fig. 4).

OS cell of origin

The mOS we generated is based on the deletion of p53 and pRb in cells that express Osterix. Osterix expression identifies a stage of osteoblast commitment after Runx2 expression that defines the first commitment step to the osteoblast lineage from a multi/bipotential precursor (Aubin 2001; Nakashima et al. 2002). Thus, OS arises from a cell at, or after, the Osterix stage of osteoblast differentiation. To determine the most likely stage of OS development in this model, we undertook expression analysis of a panel of genes representative of different stages of osteoblast differentiation on the cell lines derived from the primary tumors. The mOS cell lines were utilized as they are relatively homogenous, while they retain the capacity for efficient in vivo tumor formation and display an expression profile consistent with human OS (Supplemental Fig. 4; data not shown).

Markers of osteoblast progenitors (Runx2, Osterix), preosteoblasts (alkaline phosphatase, Collagen1a1, Ebf2, BMP4), and mature osteoblasts (osteoprotegerin, bone sialoprotein, osteocalcin) were assessed by quantitative real-time PCR, and the expression in the OS cell lines was compared with that of in vitro differentiated primary osteoblasts (Aubin 2001; Franz-Odendaal et al. 2006). mOS cell lines express markers of osteoblast progenitors and preosteoblasts but fail to express markers of mature osteoblasts. These results indicate that the mOS cell lines are arrested at the preosteoblast stage of differentiation and that the candidate target cell for transformation in this model is a lineage-committed immature osteoblast (Fig. 4A).

We also analyzed the expression of other genes associated with human OS. We found significantly increased expression of the proto-oncogene c-Fos in the mOS cell lines compared with primary osteoblasts. Constitutive transgenic overexpression of Fos results in the development of osteogenic tumors in mice and is frequently observed in human OS and in human OS cell lines (Ruther et al. 1989; Wu et al. 1990; Wang et al. 1995; Thomas et al. 2004). The expression levels of both the fibroblast growth factor receptor 2 (FGFR2) and Bub3 were reduced. Their expression is reduced frequently in human OS, likely due to allelic loss at 10q26 (Mendoza et al. 2005). Reduced expression of the p53 target gene Mdm2 was also detected (Fig. 4B). Lowered Mdm2 expression in osteoblasts in vivo was associated with reduced differentiation and Runx2 expression (Lengner et al. 2006). Changes in the expression from the INK4 locus were noted with both p16^Ink4a and p19^Arf expression increased nearly 10-fold compared with control osteoblasts (P < 0.04); however, in the absence of pRb and p53, both p16^Ink4a and p19^Arf lack their respective molecular targets.

We also performed an assessment by flow cytometry to determine the cell surface phenotype of both primary mOS and mOS cell lines. Limited information is available regarding the cell surface phenotype of murine osteoblasts. However, previous studies demonstrate that osteoblasts have a surface phenotype negative for hematopoietic (CD45 and Ter119) and vascular (CD31) markers and positive for Sca-1 and CD51 (Integrin α_v) (Semerad et al. 2005). Both primary mOS and mOS cell lines contain a prominent population of Sca-1⁺CD51⁺ double positive cells that are negative for hematopoietic and vascular markers (Fig. 4C). Other markers associated with mesenchymal cells and osteoblasts, such as SSEA-4, CD34, and VCAM, showed variable, if any, expression, on both primary mOS and cell lines (Gang et al. 2007). Taken together, these data suggest that OS development in this model arises from a cell stage at or near the preosteoblast and that many of the gene expression changes observed in human OS are mirrored in mOS (Fig. 4C).

Recapitulation of the transcriptional profile of human OS

The mOS we describe recapitulates many of the characteristic biologic features of human OS. To determine whether mOS reflects the molecular features of human OS, we undertook expression profiling using Affymetrix A430 2.0 gene chips. To enable a direct comparison between murine and human samples, we performed metagene projection analysis (Isakoff et al. 2005; Tamayo et al. 2007). Metagene projection analysis permits a robust cross-species and cross-platform comparison enabling an assessment of the degree to which mOS displays a transcriptional profile comparable to human OS, as distinguished from other human sarcomas. In the absence of metagene analysis, unsupervised hierarchical clustering of the mouse and human tumor data sets is uninformative, as it results in the separation of these data based on both species and microarray platform, consistent with that previously described (Fig. 5A).

To define the metagene for human OS, we utilized a previously published set of human sarcoma data (Henderson et al. 2005). This data set contained 11 OS samples, as well as a variety of other defined human sarcoma types of mesenchymal origin. All samples, both human and mouse, were normalized together and then the metagenes were defined from the human data set alone. A metagene was defined for each of the human sarcoma types (Fig. 5B; Supplemental Fig. 5). Once the metagene that defined human OS was identified, the transcriptional information from our primary mOS (n = 10 primary tumors, two independent data sets) and also five previously published primary murine synovial sarcoma samples (Haldar et al. 2007) were added, and the metagene information was used to cluster the mouse samples within the defined human tumor types (Fig. 5C; Supplemental Fig. 6). Mouse synovial sarcomas (green squares) associated with their human counterparts (light green circle), providing independent validation of this approach as this model was previously defined using an alternative comparison technique to be a faithful model of human synovial sarcoma (Haldar et al. 2007). Strikingly, the mOS expression signature (Fig. 5C, red squares) closely resembled that of the human OS samples (Fig. 5C, salmon circles). The mOS samples cluster within the human OS samples, which also associate with the human chondrosarcoma samples, reflecting the common origin of osteoblasts and chondrocytes. Also, as for human OS, we observed areas of cartilage formation within the mOS. These analyses demonstrate that mOS arising in our engineered mice exhibit a transcriptional profile very comparable with human OS.

Conservation of cytogenetic rearrangements between mOS and human OS

Numerous alterations and rearrangements in human OS, targeting many regions throughout the genome, have been observed by conventional cytogenetics (Bridge et al. 1993; Batanian et al. 2002). To query the mOS model independent of metagene analysis, we sought to determine whether the mouse tumors display altered expression of genes contained within regions defined cytogenetically in human OS.

To achieve this, we devised a novel in silico approach, based on the gene set enrichment analysis (GSEA) methodology, which we term cytogenetic region enrichment analysis (CREA) (Mootha et al. 2003; Subramanian et al. 2005). Cytogenetic regions of interest were chosen based on those frequently reported in the literature as rearranged in human OS (Bridge et al. 1993; Batanian et al. 2002). The human genes contained within these chromosomal bands were compiled from the consensus human genome sequence (UCSC Genome Browser, release March 2006, and Ensembl human genome sequence, release 46.36h), and GSEA gene lists were constructed for each region containing all known genes (outlined in Fig. 6A). GSEA was then performed using the primary mOS expression profiles, the expression profile of mOS cell lines, and also the murine synovial sarcoma data set that we previously utilized for metagene analysis. The mOS samples (both primary tumor and cell line) were compared with the gene expression signatures of in vitro differentiated primary osteoblasts, while the synovial sarcoma samples were compared with wild-type muscle. Enrichment of a given region was defined as a P-value of <0.05 and a false discovery rate (FDR) Q-value of <0.15.

Strikingly, primary mOS showed significant enrichment at 10 of 17 regions tested by CREA. Moreover, mOS cell lines showed enrichment at 7 of 17 regions when compared with primary in vitro-differentiated osteoblasts (Fig. 6B; Supplemental Figs. 7 and 8). In contrast, we did not observe enrichment for any of the 17 regions for the murine synovial sarcoma samples compared with control muscle samples (Fig. 6B; Supplemental Fig. 9). We further analyzed the gene expression changes at the 17 regions in two additional data sets derived from either breast or lung cancer data sets (Sweet-Cordero et al. 2005; Li et al. 2007), observing 3 and 4 significantly altered regions, respectively (Supplemental Table 4). Thus, changes in these regions are not a common occurrence in tumors per se but appear specific for OS. These results underscore how closely mOS resembles human OS. Furthermore, our findings suggest that combining cytogenetic changes with gene expression signatures via CREA may be a useful strategy for validation of other mouse tumor models as faithful representations of the corresponding human disease.

p53-dependent disease

Inherited mutation of RB in humans predisposes patients to OS, the second most frequent tumor in this population with an incidence increased 500-fold compared with the general population (Gurney et al. 1995). Moreover, sporadic OS also frequently harbors alterations in the Rb gene (Wadayama et al. 1994). These observations have led to the hypothesis that the retinoblastoma gene plays a central role in osteoblast biology, and in OS genesis in particular (Thomas et al. 2003). Loss of Rb was reported to impair terminal osteoblast differentiation due to a loss of coactivator activity from CBFA1, which was not observed in the absence of either of the related pocket proteins p107 or p130 (Thomas et al. 2001). However, mice bearing germline heterozygous Rb mutation have not been reported to develop OS but do develop other malignancies (Jacks et al. 1992). We did not observe OS development in Osx-CrepRb^fl/fl animals up to 18 mo of age, nor did we observe any gross skeletal abnormalities in these animals. These in vivo results contrast with those derived from in vitro analysis of cell lines in which pRb expression was shown to be required for osteoblast differentiation. We propose that Rb acts as a potentiator in the context of OS, but its mutation alone is insufficient to act as an initiating event.

The results of Rb deficiency contrast with the clearly established link between p53 mutation and OS development both in human and murine malignancies (Jacks et al. 1994; Lang et al. 2004; Olive et al. 2004; Lengner et al. 2006). Consistent with the importance of p53 loss or mutation to the genesis of OS, we observed loss of heterozygosity of p53 but did not observe this for pRb (Supplemental Fig. 10). The effects of missense mutation of p53, as distinct from the loss of function allele utilized here, would be interesting to determine if this alters any characteristics of the OS generated in these animals. Other cancer-associated mutations in the “Rb pathway,” such as p16^INK4a and p15^INK4b, have been reported in human OS at low frequency, and animals mutant for these genes can develop OS at low penetrance (Miller et al. 1996, 1997; Sharpless et al. 2001, 2004; Krimpenfort et al. 2007). In the model we report that Rb mutation is unable to initiate OS in the absence of mutation of p53, consistent with observations from radiation-induced OS in retinoblastoma patients (Gonin-Laurent et al. 2006, 2007). These results argue strongly that loss of p53 is the rate-limiting step in initiation of OS and that pRb loss cooperates in the steps leading to frank malignancy.

Cell of origin of OS

Lengner and colleagues previously reported OS following conditional p53 deletion with Collagen1a1-Cre (Col3.6-Cre transgenic) (Lengner et al. 2006). The Col3.6-Cre transgenic resulted in 60% penetrance of OS with a survival time similar to that observed in the Osx-Cre model. Within the osteoblast lineage, Osterix expression precedes the expression of Collagen1a1 expression, which is thought to occur at the preosteoblast stage of differentiation (Aubin 2001; Nakashima et al. 2002). Despite the difference in stages of Cre expression, both models develop OS with a very similar latency for p53^fl/fl animals. Analysis of OS arising in c-Fos transgenic animals also suggested that the cell of origin was contained within the committed osteoblast population (Grigoriadis et al. 1993). Additionally, we observed tumors in the Osx-Cre animals from intramembranous bone, such as the calvaria and periosteal surface of the femur, demonstrating that OS can arise from osteo-restricted precursors independently of chondrogenic and adipogenic cells. Collectively, these independent lines of evidence support a model in which OS arises as a result of transformation events that occur at or near the preosteoblast stage of osteoblast development (Fig. 5D).

Identification of the cell of origin of the majority of human tumors has remained elusive. The tumor initiating cell, or cancer stem cell as it also termed, has been postulated to arise from within the normal tissue stem cell pool given the similarities that exist between these cells and cells capable of sustaining the tumor (Reya et al. 2001; Kim et al. 2005b; Polyak and Hahn 2006). The mesenchymal stem cell (MSC) is the stem cell from which osteoblasts arise. Our data are inconsistent with the MSC as the cell of origin of OS and are most consistent with disease arising from an osteoblast lineage-committed progenitor, unlike that recently postulated for Ewing sarcoma (Tirode et al. 2007). Interestingly, despite the deletion occurring in an Osterix-dependent manner, we have observed areas of chondrogenic differentiation within some of the primary OS and have observed adipogenic tumors on several double mutant animals. These results suggest either that the expression of the Osx-Cre transgene is not absolutely restricted to the osteoblast population or that, once immortalized, the OS cells are able to undergo multilineage differentiation in response to appropriate environmental cues. We also cannot exclude that more differentiated cells than preosteoblasts are also capable of initiating OS, and this will need to be experimentally determined. Therefore, sarcomas, like hematopoietic tumors and also recently proposed for breast cancer, may arise from the lineage-committed fraction of cells and do not require transformation of the multipotential tissue-restricted stem cell (Huntly et al. 2004; Ince et al. 2007).

Molecular profiling demonstrates similarity to human OS

In addition to the similar transcriptional profiles of human and mOS as observed using metagene projection analysis, we documented transcriptional changes in murine tumors that were reflective of the common cytogenetic alterations observed in human OS using a novel in silico approach, CREA. CREA is based on the gene set enrichment analysis methodology and utilizes cytogenetic information to define genomic regions of interest for which gene expression information can be assessed. Conserved cytogenetic regions that display expression alterations are of particular interest. These may represent chromosomal regions of importance in the genesis, maintenance, or progression of OS. Regions, such as 10q26, that show loss of heterozygosity in human OS and contain both FGFR2 and Bub3 (which have been implicated as important in OS biology) show significantly reduced expression in both primary mOS and mOS cell lines by CREA (Mendoza et al. 2005). Reduced expression of FGFR2 and Bub3 was also confirmed by quantitative real-time PCR. Analysis of the genes within the other significantly altered cytogenetic regions, and those that show alterations in both the primary mOS and cell lines, may yield important insights into the molecular genetics of OS and highlight conserved genomic regions of importance in OS. One intriguing observation from analysis of the gene expression data is that independent of the initial genotype of the animals the OS that arises appears to be the same disease based on the correlation coefficient of gene expression of the tumors as measured by microarray (Supplemental Fig. 11).

The genetically engineered mouse OS we describe recapitulates multiple defining features of human OS, a devastating malignancy for which new therapies have not been developed in more than two decades. Metastasis is a major cause of mortality in OS and with temporal control of Cre expression it is possible to generate a model in which all animals develop metastatic disease. Murine models of human cancer serve as valuable platforms for probing cancer genetics and preclinical therapeutics. The model we report provides new opportunities for exploration of the molecular pathogenesis of OS and should constitute a convenient platform in which to test or screen for novel therapeutic agents to treat the disease.

Animals

pRb^fl/fl animals were generously provided by Dr. T. Jacks (Massachusetts Institute of Technology) and were on a C57Bl6/J 129 hybrid background (MacPherson et al. 2003; Sage et al. 2003); p53^fl/fl animals were obtained from the National Cancer Institute Mouse Models of Human Cancer Consortium Mouse Repository, have been described previously, and were on a C57Bl6/J 129 FVB/n hybrid background (Jonkers et al. 2001). Osx-Cre animals have been described previously and were on a C57Bl6/J background (Rodda and McMahon 2006). All animals were genotyped using published protocols. Rag2^−/− immunodeficient mice were used as recipients for transplant of primary tumor samples and mOS cell lines (subcutaneous transplant). All experiments were approved by the Animal Ethics Committee (Children’s Hospital Boston).

Histology

Tissue was fixed in 10% neutral buffered formalin and then paraffin embedded, sectioned, and stained with hematoxylin and eosin. All pathology was performed blinded to the sample genotype.

SKY

SKY was performed as previously described (Franco et al. 2006). Briefly, metaphase spreads of OS cell lines were hybridized using a SkyPaint DNA Kit (Applied Spectral Imaging), following manufacturer’s instructions. Spectral images were captured and analyzed using an interferometer and software from Applied Spectral Imaging.

microPET/CT imaging

PET imaging studies were performed using a Siemens Focus 120 high-resolution, small-animal PET scanner. Each animal was injected via tail vein with a bolus injection, <0.2 mL per animal, of the radiopharmaceutical (¹⁸F sodium fluoride). Between 3.7 and 37 MBq (0.1 to 1.0 mCi) of the radiopharmaceutical was administered. Animals were anesthetized to facilitate imaging by inhalation of isoflurane (2%–4%) through a nose cone for the duration of the imaging study. Image acquisition was performed after an uptake period of 30 min. Each animal was positioned on the imaging table (head first, prone) and imaged for 15 min. MicroCT imaging was performed on a Siemens MicroCAT II scanner at 27-μm resolution. Data were acquired with the system with X-ray tube voltage of 60–80 kVp, tube currents of 200–500 uA, and exposures per view of 300 msec for 200–400 views.

Microarray sample preparation and analysis

Primary tumor samples and cell lines were homogenized in Trizol (Invitrogen) to extract RNA purified using an RNeasy kit (Qiagen) using standard procedures (a more detailed method is provided in the Supplemental Data). Samples were hybridized to Affymetrix A430 2.0 mouse expression arrays, according to the manufacturer’s instructions, at the Dana-Farber Cancer Institute’s Microarray Core Facility. Data from microarray was normalized using dChip software (Li and Wong 2001). The microarray data were deposited in the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) under the accession number GSE9460.

Metagene analysis

Metagene analysis was performed essentially as described by Tamayo et al. (2007) using the available program from http://www.broad.mit.edu/cgi-bin/cancer/publications/pub_paper.cgi?mode=view&paper_id=161. Human sarcoma data were obtained from a previously published data set (Henderson et al. 2005).

CREA

Common recurrent cytogenetic rearrangements reported in the literature were complied and then the consensus human genes contained within these bands were collated from both the Ensembl and UCSC human genome browsers. Lists of consensus genes were compiled, and GSEA was performed using gene expression data obtained from the comparisons of primary mOS versus in vitro differentiated primary osteoblasts, mOS cell lines versus in vitro differentiated primary osteoblasts, and mouse synovial sarcoma versus muscle (Haldar et al. 2007). GSEA was performed as described (http://www.broad.mit.edu/gsea). Each data set underwent 1000 permutations. A complete list of human genes for each chromosomal region can be found in the Supplemental Material (Supplemental Table 5).

Quantitative real-time PCR

Genomic DNA and cDNA were prepared and quantitative real-time PCR performed as described previously (Walkley et al. 2007). Oligonucleotide sequences are listed in the Supplemental Material.