Cross Species Genomic Analysis Identifies a Mouse Model as Undifferentiated Pleomorphic Sarcoma/Malignant Fibrous Histiocytoma

Undifferentiated pleomorphic sarcoma/Malignant Fibrous Histiocytoma (MFH) is one of the most common subtypes of human soft tissue sarcoma. Using cross species genomic analysis, we define a geneset from the LSL-KrasG12D; Trp53Flox/Flox mouse model of soft tissue sarcoma that is highly enriched in human MFH. With this mouse geneset as a filter, we identify expression of the RAS target FOXM1 in human MFH. Expression of Foxm1 is elevated in mouse sarcomas that metastasize to the lung and tissue microarray analysis of human MFH correlates overexpression of FOXM1 with metastasis. These results suggest that genomic alterations present in human MFH are conserved in the LSL-KrasG12D; p53Flox/Flox mouse model of soft tissue sarcoma and demonstrate the utility of this pre-clinical model.


Introduction
Malignant Fibrous Histiocytoma was first described in the 1960s and quickly became the most commonly diagnosed adult soft tissue sarcoma [1]. Because these tumors do not appear to arise from histiocytes, the term Malignant Fibrous Histiocytoma has recently fallen out of favor and many pathologists now classify these tumors as undifferentiated pleomorphic sarcomas. Despite this change in nomenclature, undifferentiated pleomorphic sarcoma (referred to here as MFH) remains one of the most common adult soft tissue sarcomas encountered in the clinic. However, the cell(s) of origin of MFH is unknown. Indeed, some have suggested that MFH is a collection of undifferentiated mesenchymal tumors sharing a common morphology rather than a single clinical entity [2,3,4]. This debate could be clarified by identifying the cell(s) of origin of a mouse model for MFH. Whether MFH describes a cancer that is a single pathogenic entity or an undifferentiated state shared by several sarcoma subtypes, the survival of patients with MFH has not improved for decades. Therefore, identifying a mouse model of MFH may also lead to better treatments for patients with this diagnosis. Human MFH is characterized by a propensity to metastasize to the lungs and by a range of histologic appearances including spindle and pleomorphic cells. Although these features are recapitulated in a mouse model of soft tissue sarcoma initiated by conditional mutations in Kras and Trp53 [5], it is not clear whether this model is most similar to human MFH or another soft tissue sarcoma. Therefore, we sought to more accurately classify the sarcoma subtype for this mouse model using gene expression profiling.

Mouse Genotyping, Tumor Generation, and Determination of Metastatic Potential
Both mouse genotyping and generation of tumors was carried out as described previously [5] in accordance with Duke University and MIT Institutional Animal Care and Use Committee approved protocols. Sarcomas were induced in the lower left limb and allowed to grow until ,200 mm 3 in volume. Tumors were then surgically excised via amputation of the limb and animals followed for a minimum of 4 months to determine the metastatic potential of the primary tumor.

Microarray Processing and Analysis
Full details can be found in supplementary methods S1. Briefly, gene expression was determined using Affymetrix 430A 2.0 arrays (Affymetrix) as described in detail online (http://www.genome. duke.edu/cores/microarray/). CEL files were processed using the RMA algorithm [6,7] to normalize the data. Genesets were identified using a signal-to-noise metric.
Human and Mouse datasets [8,9,10] were downloaded from GEO (GSE6461, GSE6481, GSE2553, and GDS1209), normalized with RMA (when appropriate). Genesets and array data were used in GSEA as described previously [11]. Classes were defined as one soft tissue sarcoma subtype versus controls (other sarcoma or normal muscle) present in their respective datasets.

Database Accession Numbers
Microarray data was generated in conformity to MIAME guidelines and has been deposited in the GEO database under accession number GSE16779.

Oncogenic Pathway Predictors
Human soft tissue sarcoma datasets [9,10] were combined using ComBat [12] and normal tissue samples removed from the combined dataset. An oncogenic pathway classifier for Ras pathway activity was developed as described previously [13]. This classifier was used to compare undifferentiated pleiomorphic sarcoma/MFH samples (n = 29) against all other soft tissue sarcomas. Significance was determined using a non-parametric Mann-Whitney test.

Histology and Immunohistochemistry and Image Analysis
All human samples were obtained from tissue repositories at Duke and MD Anderson. These samples were used in accordance with Duke and MD Anderson Cancer Center Institutional Review Board (IRB) approved protocols under a waiver of consent. Five micron thick sections were cut from formalin fixed paraffin embedded samples. Samples were subjected to standard hematoyxlin and eosin staining or immunohistochemistry. Immunohistochemistry was performed with the following antibodies: phospho-ERK (Invitrogen 29-2389) and FOXM1 (Abcam ab47808), using the Vectastain ABC Rabbit IgG kit with Vectastain Elite ABC Reagent (Vector Labs).
Brightfield images of slides taken at 40x were used for analysis using Image Pro AMS v6.1. The counting module was trained using both positive and negative nuclear staining for phospho-ERK. A minimum of 3000 nuclei were counted per sample and a ratio between total nuclei with positive nuclei to total nuclei was determined using a minimal and maximal area of 100 and 1000 pixels respectively.

Tissue Microarrays (TMAs)
TMAs were generated at MD Anderson Cancer Center and contained a clinically annotated set of 214 soft tissue sarcoma samples including: 166 MFH/Unclassified sarcomas, 19 synovial sarcomas, 6 leiomyosarcomas, 8 pleomorphic liposarcomas, 8 myxoid liposarcomas, 6 atypical lipomatous tumors, and 1 dedifferentiated liposarcoma. TMAs were stained as above and scored semiquantitatively on a scale from 0-3+ by a musculoskeletal pathologist (L.D.) blinded to patient outcome. Scores were correlated with both diagnosis and clinical outcome.

Statistical Analysis of TMAs
Scoring of TMAs was correlated with diagnosis, and metastasisfree survival.
Correlation of immunohistochemical staining with diagnosis was tested for normality using a chi-square test. This did not reach statistical significance, therefore comparison between MFH and other soft tissue sarcomas was performed using the non-parametric Mann-Whitney test.
Metastasis free survival analysis was performed on MFH samples comparing 3+ staining to 0-2+ staining. Survival was determined by Kaplan-Meier analysis.

Cross-Species Genomic Analysis of LSL-KRas G12D ; Trp53 Flox/Flox Mouse Model of Soft Tissue Sarcoma
We hypothesized that genes most differentially expressed between mouse sarcoma and normal muscle would provide a useful molecular signature to interrogate human sarcoma datasets (Fig. 1A). To test this approach, we first analyzed published gene expression data from a previously validated mouse model of synovial sarcoma [14]. After identifying a geneset of 100 genes highly overexpressed in synovial sarcoma compared to normal muscle (Table S1) we used Gene Set Enrichment Analysis (GSEA) [11] to probe gene expression data from three studies of human soft tissue sarcomas [8,9,10]. We demonstrated strong statistical enrichment (p,0.001, FDR,0.02) of this geneset in human synovial sarcomas, but not in other subtypes of soft tissue sarcoma (Fig. S1, Table S2). This result is in agreement with GSEA for the mouse model of synovial sarcoma that was previously reported [14].

Ras Pathway Activity Is Enriched in Human MFH
The initiating events of human MFH are not well understood. Previous studies have shown mutations in p53 occur in 36% of human MFH [15] while the rate of canonical RAS mutations in human MFH varies from 0-50% [16,17]. We hypothesized that the RAS pathway may be activated in human MFH even in the absence of canonical RAS mutations. To explore a link between MFH and Ras, we utilized previously described oncogenic pathway predictors that correlate with RAS activity [13]. The Ras oncogenic signature is enriched in human MFH samples compared to a panel of other soft tissue sarcomas (p = 0.002) ( Fig. 2A). Moreover, in human MFH (n = 8) lacking canonical RAS mutations, we observed nuclear staining of phospho-ERK by immunohistochemistry in greater than 30% of tumor cells in 7 of 8 samples (Fig. S2).

FoxM1 Is a Novel Marker of Metastasis in MFH
Because human MFH and the LSL-Kras G12D ; Trp53 Flox/Flox mouse model of soft tissue sarcoma share common genomic features, we wanted to determine if these shared features could be used to identify diagnostic or prognostic factors for human MFH. As MFH is considered a diagnosis of exclusion, we initially attempted to identify a marker that is specific to MFH. We identified a panel of 10 candidate  (Table S3) (Table S4)  biomarkers based on their common upregulation in both human MFH and the mouse sarcoma model (Table S5). Expression of 9 of these candidates was validated in an independent cohort of mouse sarcomas by Q-RT-PCR (Fig. S3).
FOXM1, which is a member of the forkhead transcription factor family that enhances tumorigenesis in other solid tumors [18], was selected for further analysis because it is downstream of Ras [19]. Immunohistochemistry for FOXM1 in a panel of 8 human MFH samples demonstrated nuclear staining for FOXM1 in all 8 tumors (Fig. S2). We next analyzed the expression of FOXM1 in a clinically annotated tissue microarray (TMA) containing 166 MFH samples and 48 other soft tissue sarcomas. Although 84% of the MFH samples stained positive for FOXM1 (p = 0.02, Fig. S4), this marker was also expressed, but to a lesser degree, in other sarcoma subtypes. Therefore, FOXM1 may not be a useful diagnostic marker for human MFH. Because FoxM1 has previously been shown to regulate the expression of matrix metalloproteases MMP-2 and -9, which are key mediators of cell invasion [20], we hypothesized that high FOXM1 expression may correlate with metastasis. We measured Foxm1 gene expression in murine soft tissue sarcomas and found a correlation with the development of lung metastases (Fig. 2B, p = 0.01). Likewise, human MFH with high FOXM1 expression correlated with decreased metastasis-free survival compared to sarcomas with low to no FOXM1 expression (Fig. 2C). In contrast, overexpression of another candidate marker MELK (Table S5) did not correlate with metastasis-free survival (Fig. S5).

Discussion
We used cross species genomic analysis to determine which human sarcoma subtype is represented by the LSL-Kras G12D ; Trp53 Flox/Flox mouse model of soft tissue sarcoma. We identified a geneset in the mouse sarcomas that is highly enriched in human MFH. This geneset was not enriched in other human sarcomas, such as fibrosarcoma or leiomyosarcoma, which can be difficult to distinguish from MFH. Furthermore, we have identified enrichment of Ras pathway activity in human MFH compared to other types of soft-tissue sarcoma. Additionally, the LSL-Kras G12D ; Trp53 Flox/Flox mouse model of soft tissue sarcoma has a propensity to metastasize to the lungs much like human MFH [5]. Based on this genomic analysis, the pattern of lung metastasis, and the similarity of the mouse sarcomas to human MFH at the histological level (Fig. S6), we conclude that this model closely resembles MFH.
We recognize that the diagnosis of undifferentiated pleomorphic sarcoma (MFH) has recently been questioned as a distinct clinical entity [2,3,4]. Our results do not exclude the possibility that MFH is a collection of mesenchymal tumors derived from different cell types that share an undifferentiated state. However, our finding of a sarcoma geneset conserved between mouse sarcomas and human MFH suggests that this subtype of human sarcoma shares an underlying biology beyond a common histologic appearance. Moreover, the use of crossspecies analysis to identify FOXM1 as a marker of metastasis-free survival in human MFH supports the use of this mouse model to understand mechanisms of metastasis, to investigate the cell(s) of origin, and to develop novel therapies for human MFH.

Table S1
Geneset derived from mouse model of synovial sarcoma versus control (normal muscle). Geneset was derived using signal-to-noise metric.  Table S2 GSEA results for synovial sarcoma geneset derived from mouse model of synovial sarcoma (Table S1). The mouse synovial sarcoma geneset was used to examine three datasets of human soft tissue sarcomas.   [9][10]. This list of genes was then cross referenced against the LSL-Kras G12D ; Trp53 Flox/Flox soft tissue sarcoma geneset (Table S3) and ten overlapping genes were identified. Found at: doi:10.1371/journal.pone.0008075.s006 (0.03 MB DOC) Figure S1 The mouse synovial sarcoma geneset (Table S1) shows strong enrichment for human synovial sarcoma (Enrichment Score (ES) = 0.67, Normalized Enrichment Score (NES) = 2.21) in a human soft tissue sarcoma dataset [10] (p = 0.00024; FDR = 0.004). Found at: doi:10.1371/journal.pone.0008075.s007 (0.21 MB PDF) Figure S2 Representative immunohistochemistry for A phospho-ERK and B FOXM1 show strong nuclear staining from the same human tumor sample. All scale bars represent 100 microns. Found at: doi:10.1371/journal.pone.0008075.s008 (9.18 MB PDF) Figure S3 Box and whisker plots of nine candidate genes selected for Q-RT-PCR using an independent set of mouse soft tissue sarcomas (n = 5) and normal muscle samples (n = 3). Relative fold expression determined to lowest expressing normal muscle sample. Significance of differentially expressed genes was determined by two-tailed student's T-test. Found at: doi:10.1371/journal.pone.0008075.s009 (0.31 MB PDF) Figure S4 Immunostaining of tissue microarrays (TMAs) for FOXM1 and MELK identifies expression in MFH. Tissue Microarrays (TMAs) containing 214 soft tissue sarcomas were stained for FOXM1 and scored semi-quantitatively. The degree of staining for A FOXM1 (p = 0.0017, non-parametric Mann-Whitney test) and B MELK (p = 0.02, non-parametric Mann-Whitney test) was correlated with MFH.