Gene expression profiles relate to SS18/SSX fusion type in synovial sarcoma

Authors


Abstract

We applied 27k spotted cDNA microarray slides to assess gene expression profiles in 26 samples from 24 patients with synovial sarcomas (SS). The data were analyzed in relation to histopathologic type, cytogenetic aberrations, gene fusion type and development of distant metastases. Supervised analysis based on gene fusion type in 12 SS with SS18/SSX1 and 9 with SS18/SSX2 revealed significant differences in gene expression profiles. Among the discriminators were several genes that have previously been found to be upregulated in SS, including AXL, ZIC2, SPAG7, AGRN, FOXC1, NCAM1 and multiple metallothioneins. Histopathology and degree of cytogenetic complexity did not significantly influence expression, whereas a genetic signature that related to development of metastases could be discerned, albeit with a high false-positive rate. In conclusion, our findings demonstrate differentially expressed genes for the 2 major gene fusion variants in SS, SS18/SSX1 and SS18/SSX2, and thereby suggest that these result in different downstream effects. © 2005 Wiley-Liss, Inc.

Synovial sarcomas (SS) account for 5–10% of adult soft tissue sarcomas, are more common than other soft tissue sarcoma types in young adults, and are often located distally and within the deep parts of an extremity. The tumors are highly malignant; almost half of the patients develop metastases. Based on histopathologic appearance, 2 major subtypes of SS can be distinguished: monophasic tumors with a spindle cell component only, and biphasic tumors with a mixture of spindle cell components in varying proportions and epithelial cells organized in glandular formations. In addition, a subset of the tumors is poorly differentiated and has an unfavorable prognosis. More than 90% of the SS, irrespective of subtype, are characterized by the cytogenetic aberration t(X;18) (p11;q11) or variants thereof.1, 2 This chromosomal translocation fuses the 396 N-terminal amino acids of the SS18 gene on chromosome 18 to the 78 C-terminal amino acids of 1 of 3 highly homologous SSX genes: SSX1, SSX2 and rarely SSX4.3, 4, 5 Both the SS18 and the SSX proteins regulate gene transcription, although their modes of action differ; SS18 appears to act as a transcriptional coactivator and is also involved in cell adhesion, whereas the SSX proteins act as corepressors of transcription and the chimeric fusion protein contains both transcriptional activating and repressing domains.6, 7 A correlation between gene fusion type and histopathology has been described; SS18/SSX2 tumors tend to show monophasic histology, whereas biphasic tumors tend to carry the SS18/SSX1 fusion.8, 9, 10

Since morphology and gene fusion type may identify clinically relevant tumor subsets and since the chimeric fusion proteins influence transcriptional regulation, SS is an interesting model system for studying gene expression profiles. Previous studies in which expression arrays have been applied to soft tissue sarcomas have demonstrated that SS form specific subclusters and have also suggested that the expression profiles allow distinction of subclasses that correlate with histopathologic features, i.e., monophasic and biphasic differentiation.11, 12 We applied cDNA microarray to assess the gene expression profiles of 26 SS with the aim to evaluate whether gene expression profiles relate to gene fusion type, cytogenetics, histopathology and development of metastasis.

Material and methods

Tumors and reference RNA

Tumor tissue from 26 SS samples, including 19 primary tumors, 3 local recurrences and 4 metastases, was obtained from 24 patients at the Lund University Hospital and the Karolinska Hospital, Stockholm (Table I). The samples were frozen and stored at −80°C until use. The histopathology had been reevaluated by reference pathologists and the cytogenetic and molecular genetic characteristics of the tumors have previously been reported.13 Tumors were classified as carrying simple alterations, including the X;18-translocation, or a more complex karyotype. Clinical follow-up included chest X-rays and clinical examination every 3 months during the first 3 years and thereafter biannually for at least 5 years. Follow-up was complete and was minimum 9 years for the survivors. Three of these patients received preoperative treatment; chemotherapy in 2 cases and radiotherapy in 1 case. Metastases developed in 16 patients, and from 12 of these the primary tumors were available for analysis (Table I). Among the 12 metastasizing primary tumors, 9 had the SS18/SSX1 fusion, 2 had the SS18/SSX2 and 1 had an unclassified fusion variant, whereas among the 7 nonmetastasizing tumors, 4 had SS18/SSX2, 1 SS18/SSX1, 1 SS18/SSX4 and 1 had an unclassified fusion variant (Table I). The study was approved by the Lund University research ethics committee.

Table I. Clinical, Histopathologic and Cytogenetic Data
CaseSex/AgeTumor siteSpecimenHistotypeGene fusionKaryotypeMetastasis
  • 1, 2

    Tumors not used in the analysis of expression in relation to metastasis.

  • 2

    Karyotypically normal cells were observed.

  • 3

    dic(13;22) also present.

1M/11Upper extremityPrimaryMonophasicSS18/SSX2t(X;18)No
2F/72Upper extremityPrimaryMonophasicSS18/SSX2t(X;18)No
3F/84Lower extremityPrimaryMonophasicSS18/SSX4ComplexNo
4F/52Upper extremityMetastasisMonophasicSS18/SSX2ComplexYes1
5M/44Upper extremityPrimaryBiphasicSS18/SSX1t(X;18)No
6M/39Lower extremityPrimaryMonophasicUnclassifiedNot informative2Yes
7F/46Lower extremityPrimaryMonophasicSS18/SSX1ComplexYes
8M/17Lower extremityPrimaryMonophasicSS18/SSX1Not informative2Yes
9F/50Lower extremityPrimaryBiphasic, poorSS18/SSX1ComplexYes
10F/45Lower extremityLocal recMonophasicSS18/SSX1ComplexYes1
11M/67MediastinumLocal recMonophasicSS18/SSX2ComplexNo1
12M/37Lower extremityPrimaryMonophasicSS18/SSX1ComplexYes
13M/40Upper extremityPrimaryMonophasicSS18/SSX1ComplexYes
14M/25Lower extremityPrimaryMonophasicSS18/SSX1ComplexYes
15M/25Upper extremityPrimaryMonophasic, poorSS18/SSX1Not informative2Yes
16M/59Upper extremityPrimaryMonophasicSS18/SSX2ComplexYes
17F/31Upper extremityPrimaryMonophasic, poorSS18/SSX2t(X;18)No
18aF/42ThoraxLocal recBiphasicSS18/SSX2ComplexYes1
18bF/45LungMetastasisBiphasicSS18/SSX2ComplexYes1
19aM/41AbdomenMetastasisBiphasicSS18/SSX1ComplexYes1
19bM/43Lower extremityMetastasisBiphasicSS18/SSX1ComplexYes1
20F/65Lower extremityPrimaryMonophasic, poorSS18/SSX1Not informative2Yes
21M/35Lower extremityPrimaryMonophasicSS18/SSX1ComplexYes
22M/73Lower extremityPrimaryMonophasicSS18/SSX2ComplexNo
23M/68Upper extremityPrimaryMonophasic, poorSS18/SSX2ComplexYes
24F/42Lower extremityPrimaryMonophasicUnclassifiedt(X;18)3No

RNA isolation, labeling and hybridization

Total RNA was extracted from 80–120 mg frozen tissue, using TRIzol (Invitrogen, Carlsbad, CA) followed by the RNeasy Midi kit (Qiagen, Valencia, CA). As reference RNA, we used the Universal Human Reference RNA (Stratagene, La Jolla, CA). cDNA synthesis and CyDye coupling were carried out using the CyScribe cDNA Post labeling Kit (Amersham Biosciences, UK) according to the manufacturer's recommendations. In short, 35 μg of tumor RNA and 25 μg of reference RNA were annealed with Anchored oligo(dT) primer and an extension reaction was performed using CyScript™ Reverse Transcriptase. During the synthesis, amino allyl-dUTP was incorporated into the cDNA. After degradation of remaining RNA templates by NaOH treatment, the amino allyl-modified first strand cDNA was purified using ethanol precipitation. Purified amino allyl-modified cDNA (15 μL) was coupled with an excess of reactive CyDye™ NHS-esters (Amersham Biosciences) to give a labeled cDNA probe. Tumor cDNA was coupled to Cy3 and reference cDNA to Cy5. To maximize the signal-to-noise ratio, the probes were purified using CyScribe GFX purification columns (Amersham Biosciences), whereafter the labeled tumor and reference cDNA were pooled together with blocking reagents, Poly d(A) (Amersham Biosciences), Yeast tRNA (Sigma) and Human Cot-1 DNA® (Invitrogen™life technologies). Hybridization was performed manually using the Pronto!™ Universal Hybridization Kit (Corning Life Sciences, NY) according to the manufacturer's instructions. The labeled pellet was resuspended in the Pronto!™ Hybridization Solution, applied to the slides, processed in the provided washing solutions and sealed in a Corning® Hybridization Chamber at 42°C for 18 hr. The cDNA arrays used were produced at the Swegene DNA Microarray Resource Centre, Department of Oncology, Lund University and were spotted with PCR products from 27,498 sequence-verified cDNA IMAGE clones representing 25,649 distinct reporters (Build 182 of the UniGene human sequence collection, http://www.ncbi.nlm.nih.gov/UniGene/).

Image analysis and data extraction

The slides were scanned using an Agilent DNA Microarray Scanner (Agilent technologies, Palo Alto, CA) at 5 μm resolution and then analyzed in GenePix™ Pro 4.1 (Axon Instruments Inc., Union City, CA). Individual spots were flagged as good, bad, found, not found or absent, using GenePix Pro. The quantified data from the GenePix Pro, including fluorescence signal intensity of each feature relative to the intensity of the background, feature size and feature position was then saved as a GenePix results file. Subsequent data analysis was carried out using a local installation of the web-based BioArray Software Environment (BASE; http://base. onk.lu.se/int/).14 A preliminary filtering was based on the flagged spots in GenePix, and expression values for remaining spots were calculated as the log2 of the ratio of tumor intensity to reference intensity. To normalize expression values, intensity dependent LOWESS fits were used, within groups of 8 print-tip blocks to correct for spatial bias.15 To account for varying spot qualities, represented by quantities such as intensities and background variation, we adopted a weighted approach, in the same spirit as an analysis discussed elsewhere.16 The approach introduces a smooth treatment of quality measures, gradually reducing the importance of the spot as its quality goes down, in contrast to the commonly adopted additional spot filter, based on thresholds in spot quality measures, which introduces an unmotivated difference between spots with quality values close to the threshold but on opposite sides (one is rejected while the other henceforth is trusted blindly for what it is). As an estimate of the uncertainty of the expression value of a spot, we used u = SNRmath image + SNRmath image, where SNRi is the signal to background noise ratio for channel i. Thus, spots with low signal and/or large background variation are assessed a large uncertainty.

Replicate measurements xi of the same reporter on an assay were merged and represented by a weighted mean m = Σiwixiiwi, where the weight wi is exp(−3ui1/2/|xim|). This set of equations was solved numerically by simple iteration. The exponential suppression of spots with large uncertainty ui implies that poor-quality spots do not contribute to the mean. For small uncertainties ui, the weight is 1, which implies that the weighted mean approaches the normal arithmetic mean for high-quality spots. The argument −3ui1/2/|xim| is motivated in the value modification step given later.

The error of the merged value was defined as U = 1/Σi(1/ui) + Σiwi2(xim)2/(Σiwi)2. This construction has a set of desired properties: for high-quality spots, the first term is negligible and U approaches the squared standard error of the mean, which is an appropriate uncertainty measure of the merged value; if the replicates agree, the second term is negligible, and U reflects the uncertainty of the replicates, being smaller than the smallest ui, since confirming measurements reduce the uncertainty; finally, if a spot has much larger uncertainty than others in the merge, it will have a negligible influence on U.

Expression values for reporters associated to the same gene symbol were merged similarly. It is possible to continue the weighted approach including weights to the statistical analysis,16 but we adopted a value-modification approach, where poor-quality values xi, now representing the value merged on gene symbol, were moved toward the wieghted mean m across assays for that gene.17 The modified expression value was given by xi′ = m + wi(xi + m). The chosen form of wi implies that high-quality merged measurements were moved ∼3 standard errors towards the mean across assays, while low-quality measurements thus got a value xi′ essentially equal to m, reducing its importance when searching for differential expression. After reducing the importance of low-quality measurements in this way, the following analysis should be unweighted.17 After value modification, presence- and variation filters were applied to the data, rejecting genes with known measurement in less than 90% of the samples or with a standard deviation across assays of modified expression values smaller than 0.2.

Statistical analysis

Discriminating gene lists that differentiate between different histopathologic types (monophasic versus biphasic), gene fusion type (SS18/SSX1 versus SS18/SSX2), karyotype (X;18-translocation as the sole change versus a complex karyotype, albeit including the X;18-translocation), and development of metastases were generated based on the ∼3,500 genes that passed the preprocessing filters. The relative expression ratios of each gene in the subclasses were used to calculate the mean (m) and the standard deviation (σ) for that particular gene across all of the samples. By using the mean and standard deviation calculated, each gene was assigned a discriminative weight, a Golub-score18

equation image

where m1 and m0 are the (unweighted) mean values for subgroups 1 and 0, respectively, whereas σ1 and σ0 are the standard deviations for the same subgroups. A high Golub-score implicates minor variation in gene expression within the group, but large variation between the subgroups. A random permutation test with 1,000 permutations was performed so as to assess the discriminating power of the score to differentiate the groups. For each score, the average number of genes in a permutation list above that score was divided by the number of genes in the true list to get the false-discovery rate.

Results

Hierarchical cluster analysis

Unsupervised hierarchical cluster analysis based on the 26 SS did not reveal any distinct clusters related to histotype (monophasic versus biphasic), gene fusion type (SS18/SSX1 versus SS18/SSX2), karyotype (X;18-translocation only versus as part of a complex karyotype), or development of metastases. When Golub-scores were estimated and supervised hierarchical cluster analysis19, 20 was performed on genes with top scores, differences in expression profiles were found in relation to fusion type (Table II and Fig. 1). Development of metastases gave a weak discrimination; the top-30 genes discriminated between primary tumors that did and did not metastasize (Fig. 2). No differences were found relative to histotype or karyotype. Regarding the latter, only 5 tumors had t(X;18) as the only change, and this small number of samples reduces the possibility to establish statistical significance.

Table II. Genes Discriminating The Different Gene Fusion Types
RankGolub scoreSign1Image IDGene nameGene symbol
  • 1

    +, upregulated in SS18/SSX2 tumors; −, upregulated in SS18/SSX1 tumors.

11.137483402Transcription factor 7 (T-cell specific, HMG-box)TCF7
21.07051635709Deltex 3 homolog (Drosophila)DTX3
30.9754415161Homo sapiens mRNA; cDNA DKFZp686J21109DKFZp686J21109
40.94051604674Zic family member 2 (odd-paired homolog, Drosophila)ZIC2
50.9349449126G protein-coupled receptor 153GPR153
60.9173+788524Syntaxin binding protein 6 (amisyn)STXBP6
70.9129138991Collagen, type VI, alpha 3COL6A3
80.87431455249Coatomer protein complex, subunit epsilonCOPE
90.8710129883Notch homolog 2 (Drosophila) N-terminal like|Insulin induced gene 1NOTCH2NL|INSIG1
100.86563546928Insulin induced gene 1INSIG1
110.8206+134748Glycine cleavage system protein H (aminomethyl carrier)GCSH
120.8027756533Basigin (OK blood group)BSG
130.7916+240711Neural precursor cell expressed, developmentally down-regulated 4NEDD4
140.78932568378Family with sequence similarity 50, member AFAM50A
150.772649318AXL receptor tyrosine kinaseAXL
160.7695+155050Hypothetical protein MDS025MDS025
170.7546+359701Adrenergic, beta-1-, receptorADRB1
180.7444412975Golgi autoantigen, golgin subfamily a, 2GOLGA2
190.7422+307471Transmembrane 4 superfamily member 2TM4SF2
200.7395138693Hypothetical protein LOC349114LOC349114
210.7323191664Thrombospondin 2THBS2
220.7298809981Glutathione peroxidase 4 (phospholipid hydroperoxidase)GPX4
230.7283+252382EST 
240.7231898218Insulin-like growth factor binding protein 3IGFBP3
250.7202245990Metallothionein 1F (functional)MT1F
260.713880707Sperm associated antigen 7SPAG7
270.71301292136SEC24 related gene family, member D (S. cerevisiae)SEC24D
280.7013898262Ubiquitin-activating enzyme E1UBE1
290.69891456120G protein-coupled receptor kinase 5GRK5
300.6915+504915Cofilin 2 (muscle)CFL2
310.6899417867X-box binding protein 1XBP1
320.6837950710Propionyl Coenzyme A carboxylase, alpha polypeptidePCCA
330.6818202535Metallothionein 1G|H19MT1G|H19
340.6766+897884KIAA1573 proteinKIAA1573
350.6756+234736GATA binding protein 6GATA6
360.6754897233Polypyrimidine tract binding protein 1PTBP1
370.67542548367Aldolase C, fructose-bisphosphateALDOC
380.6750417393Similar to contains PTR5.b2 PTR5 repetitive element, mRNA sequence 
390.67432056049Histone 1, H2blHIST1H2BL
400.6700346009Phosphofructokinase, liver|Triosephosphate isomerase 1PFKL|TPI1
410.6699430235Histone 2, H2beHIST2H2BE
420.6657590150Metallothionein 2AMT2A
430.663045587DKFZP564I1171 proteinDKFZP564I1
440.6626757487CDNA FLJ13598 fisN/A|RRBP1
450.6596+276962Hypothetical protein A-211C6.1LOC57149
460.6579856454Solute carrier family 3, member 2SLC3A2
470.65431460224Serine/threonine kinase 32CSTK32C
480.6533280156FLJ00012 proteinFLJ00012
490.6533810801AgrinAGRN
500.6505700857Cell division cycle 2-like 1|Cell division cycle 2-like 2CDC2L1|CDC2L2
510.64942014382Nuclear receptor subfamily 2, group F, member 1NR2F1
520.6450878571G protein-coupled receptor 56GPR56
530.64361472735Metallothionein 1E (functional)MT1E
540.6414248095Hypothetical protein LOC92497LOC92497
550.6388+877827Ribosomal protein S27aRPS27A
560.6361814998Similar to contains Alu repetitive element, mRNA sequence 
570.6341759184Solute carrier family 35, member A4SLC35A4
580.6338+896914Mitochondrial ribosomal protein S33MRPS33
590.63322544116GLI-Kruppel family member HKR3HKR3
600.63292556946Villin 2 (ezrin)VIL2
610.630415740581-acylglycerol-3-phosphate O-acyltransferase 2AGPAT2
620.6300+358885Forkhead box C1FOXC1
630.62901474337Phosphorylase, glycogen; brainPYGB
640.6282813675Stimulated by retinoic acid 13 homolog (mouse)STRA13
650.6273295939Chromosome 6 open reading frame 85C6orf85
660.6249624627Ribonucleotide reductase M2 polypeptideRRM2
670.62302049639Cerebral cavernous malformations 1CCM1
680.6222725405Early estrogen-induced gene 1 proteinMGC50853
690.62221475659Protease, serine, 8 (prostasin)PRSS8
700.6206504308Chromosome 10 open reading frame 3C10orf3
710.61932312170Tubulin-specific chaperone dTBCD
720.6184214162Metallothionein 1HMT1H
730.6153+277165Transmembrane protein with EGF-like and two follistatin-like domains 1TMEFF1
740.6148449112Hypothetical protein LOC157567LOC157567
750.6147+506165CDC42 effector protein (Rho GTPase binding) 3CDC42EP3
760.6146125187Excision repair cross-complementing rodent repair deficiency, 2ERCC2
770.6132269791Dopachrome tautomeraseDCT
780.6127725680Transcription factor AP-2 gammaTFAP2C
790.6123+52990Solute carrier family 1, member 2SLC1A2
800.6103701751Cut-like 1, CCAAT displacement protein (Drosophila)CUTL1
810.6094826204Flightless I homolog (Drosophila)FLII
820.6086810504Proteolipid protein 2 (colonic epithelium-enriched)PLP2
830.60832248780Likely ortholog of mouse la related proteinLARP
840.6072399456Syntaxin 4A (placental)STX4A
850.6058281476AspartylglucosaminidaseAGA
860.6051290841Histone 1, H2bkHIST1H2BK
870.6049213176EST 
880.6043815535Treacher Collins-Franceschetti syndrome 1TCOF1
890.6042+448068Hypothetical protein FLJ10204FLJ10204
900.60312565353MLL septin-like fusionMSF
910.6016202154Homo sapiens transcribed sequences 
920.6011+365826Growth arrest-specific 1GAS1
930.59812460159Tyrosine kinase, non-receptor, 1TNK1
940.5966108377Tubulin, gamma 1TUBG1
950.596650887Ral guanine nucleotide dissociation stimulatorRALGDS
960.5966+1883065Neural cell adhesion molecule 1NCAM1
970.5954284115Solute carrier family 6 (neurotransmitter transporter), member 15SLC6A15
980.59531675553Histone 1, H2bjHIST1H2BJ
990.5945843036Microtubule-associated protein 7MAP7
1000.5937+810899CDC28 protein kinase regulatory subunit 1BCKS1B
Figure 1.

Cluster analysis based on the top-100 discriminators for gene fusion type, SS18/SSX1 versus SS18/SSX2. Clustering was done using the TMeV application from the TM4 microarray software suit.21 The Pearson correlation distance metric was used.

Figure 2.

Cluster analysis based on the top-30 discriminators for development of metastasis. Clustering was done using the TMeV application from the TM4 microarray software suit.21 The Pearson correlation distance metric was used.

Identification of differentially expressed genes

Comparison between the 12 tumors with SS18/SSX1 and the 9 tumors with SS18/SSX2 (excluding duplicate tumor samples obtained from 2 patients, 2 tumors for which no fusion gene had been identified and 1 tumor with the SS18/SSX4 fusion) identified different gene expression patterns; when genes were ranked according to Golub-score, the false-discovery rate remained fairly constant around 40%, ∼30–100 genes (Fig. 3). Among the genes upregulated in SS with the SS18/SSX1 fusion were the transcription factor 7 (TCF7), the Zic family member 2 (ZIC2), the insulin growth factor binding protein 3 (IGFBP3), the sperm-associated antigen 7 (SPAG7), agrin (AGRN), ezrin (VIL2) and the genes AXL, RALGDS and CDC2L1, which have been implicated in oncogenesis. Furthermore, metallothioneins, histones, and G protein-coupled receptors were also upregulated in tumors with the SS18/SSX1 fusion type (Table II). In tumors with the SS18/SSX2 fusion, the genes forkhead box C1 (FOXC1), growth arrest specific-1 (GAS1) and the neural cell adhesion molecule 1 (NCAM1) appeared as upregulated (Table II). Whereas cytogenetic complexity and histopathology did not reveal distinct gene expression patterns, analysis of the 12 primary SS that metastasized compared to the 7 primary tumors that remained free of metastasis, using the Golub-score and a random permutation test, revealed 30 discriminating genes, of which ∼9 were expected to appear by chance. Among these genes were syntaxin binding protein 6 (STXBP6), survivin (BIRC5) and topoisomerase IIα (TOP2A) (data not shown).

Figure 3.

False discovery rates for metastasis (filled line) and gene fusion type (dotted line). Based on these results the study focused on the top-30 genes for metastasis and the top-100 genes for gene fusion type.

Discussion

Previous studies applying gene expression profiling to soft tissue tumors have revealed distinct clusters formed by the SS and the overexpressed genes have included members of the retinoic acid receptor pathway, members of the WNT and TGF-β pathways, ephrins, genes related to the fibroblast growth factor (FGF) and the insulin growth factor (IGF), CRABP1 and the SSX genes.11, 12, 22, 23, 24 Studies of gene expression in SS have revealed subclusters based on tumor morphology – monophasic versus biphasic.11, 24 Based on the expression pattern of 1,405 genes, Nagayama et al. identified 2 subclasses; biphasic tumors clustered together, whereas monophasic tumors were divided into 2 subsets; 1 of which clustered together with the biphasic tumors and the other formed a separate cluster.24 Among the discriminating genes reported were JUN, TGFB1 inducible early growth response (TIEG) gene and annexin A4.24 Allander et al. used 21 genes to obtain a separation between biphasic and monophasic tumors with keratin-encoding genes being the most frequently upregulated in biphasic tumors, which would agree with an epithelial component being present in these tumors.11 We did not, however, find any distinct gene expression patterns in relation to tumor morphology, but our study only contained 6 biphasic tumors from a tumor type with inherent heterogeneity.

The t(X;18) or variants thereof are found in about 90% of SS and is suggested to represent the underlying cause of tumorigenesis, but secondary alterations including both numerical changes and unbalanced structural alterations are present in about 1/3 of the tumors and occur at increased frequency in recurrent and metastatic SS.25 Only 5 of the tumors in this study carried t(X;18) as the sole change precluding meaningful statistical analysis. Regarding gene fusion type, the SS18/SSX1 has been suggested to correlate with poor prognosis, although the correlation has been questioned, and tumors with this fusion type may also have a higher proliferative rate.8, 26, 27 When comparing the gene expression profiles in tumors with the SS18/SSX1 and the SS18/SSX2 fusion, we identified discriminating gene expression profiles for these 2 groups (Fig. 1), which stand in contrast to the study by Allander et al.11 Among the top-100 genes, 81 were upregulated in tumors with the SS18/SSX1 fusion variant, and included a number of genes that have previously been linked to SS, such as AXL, ZIC2, SPAG7 and AGRN, and in the case of VIL2 to rhabdomyosarcoma and osteosarcoma (Table II).11, 22, 28, 29, 30 Multiple metallothioneins and histones were also upregulated in tumors with the SS18/SSX1 fusion. Metallothioneins have previously been shown to be expressed in SS, and have also been suggested to be of prognostic value in soft tissue sarcomas.31, 32 Since the C-terminal of SSX1 is known to target histones, it is possible that the action of SYT-SSX may be linked to these proteins.33 A high expression of histones and genes within histone metabolic pathways may indicate a high proliferative activity, and interestingly, high proliferation has been linked to SS with the SS18/SSX1 fusion27, 34 Many of the other genes on the list have also been implicated in oncogenesis, such as TCF7, IGFBP3, AXL, RALGDS and CDC2L1. Among the genes upregulated in tumors with the SS18/SSX2 fusion were FOXC1 and NCAM1, which previously have been reported as upregulated in SS22 (Table II). Our findings thus suggest that gene fusion type affects gene expression and identifies transcription factors, growth factor receptors and metallothioneins as discriminators, which may provide important information about the histopathologic and prognostic differences that have been suggested in SS.

Metastases, most commonly in the lungs, develop in almost 50% of the patients with SS. Differentially expressed genes were found when the 12 primary SS that metastasized were compared to 7 primary tumors that did not, but with a high false discovery-rate (Fig. 3). However, the TOP2A gene, which has been demonstrated to correlate with prognosis in SS and which has been associated with metastatic potential also in other tumor types was among the disciminating genes, but this observation needs to be confirmed in a larger sample set.35, 36, 37 Since SS carry a high risk of metastasis, development of a molecular prognosticator would be valuable for decisions on adjuvant chemotherapy.

Acknowledgements

We acknowledge E. Nilsson at the Scandinavian Sarcoma Group Register for contributing with clinical data. P.E. was supported by the Swedish Foundation for Strategic Research through the Lund Center for Stem Cell Biology and Cell Therapy. This study was financially supported by the Swedish Cancer Fund, the Swedish Children's Cancer Foundation, the I-B and A. Lundberg Foundation, the K. and A. Wallenberg Foundation via the SWEGENE program, the Nilsson Cancer Fund, the Kamprad Cancer Fund, and the Lund University Hospital Cancer Funds.

Ancillary