Genomic Profiling of Chondrosarcoma: Chromosomal Patterns in Central and Peripheral Tumors

Purpose: Histologic grade is currently the best predictor of clinical course in chondrosarcoma patients. Grading suffers, however, from extensive interobserver variability and new objective markers are needed. Hence, we have investigated DNA copy numbers in chondrosarcomas with the purpose of identifying markers useful for prognosis and subclassification.

Experimental Design: The overall pattern of genomic imbalances was assessed in a series of 67 chondrosarcomas using array comparative genomic hybridization. Statistical analyses were applied to evaluate the significance of alterations detected in subgroups based on clinical data, morphology, grade, tumor size, and karyotypic features. Also, the global gene expression profiles were obtained in a subset of the tumors.

Results: Genomic imbalances, in most tumors affecting large regions of the genome, were found in 90% of the cases. Several apparently distinctive aberrations affecting conventional central and peripheral tumors, respectively, were identified. Although rare, recurrent amplifications were found at 8q24.21-q24.22 and 11q22.1-q22.3, and homozygous deletions of loci previously implicated in chondrosarcoma development affected the CDKN2A, EXT1, and EXT2 genes. The chromosomal imbalances in two distinct groups of predominantly near-haploid and near-triploid tumors, respectively, support the notion that polyploidization of an initially hyperhaploid/hypodiploid cell population is a common mechanism of chondrosarcoma progression. Increasing patient age as well as tumor grade were associated with adverse outcome, but no copy number imbalance affected metastasis development or tumor-associated death.

Conclusion: Despite similarities in the overall genomic patterns, the present findings suggest that some regions are specifically altered in conventional central and peripheral tumors, respectively.

Chondrosarcoma is the collective term for a biologically and clinically heterogeneous group of malignant cartilage-producing tumors. They may be subdivided according to grade and morphologic features, with the so-called conventional subtype accounting for ∼85% of the cases; the other variants (dedifferentiated, mesenchymal, clear cell, and myxoid chondrosarcoma) are individually rare (1). Conventional chondrosarcoma of bone is further trichotomized according to the location in the bone. Primary conventional central chondrosarcoma originates from the medullar cavity of the bone, whereas secondary peripheral and periosteal ones develop from the surface of the bone, although with distinct clinical and morphologic characteristics. Finally, chondrosarcomas may either develop de novo or through malignant transformation of a preexisting benign lesion, termed primary and secondary chondrosarcomas, respectively (2).

The incidence of chondrosarcoma increases with age, hence primarily affecting adults. It is also known that patients suffering from multiple osteochondromas and enchondromatosis are at an elevated risk of developing chondrosarcoma. Multiple osteochondromas is an autosomal dominant disorder characterized by multiple benign cartilage-capped bony outgrowths (osteochondromas) at the surface of the long bones. The disease is caused by mutations in the EXT1 and EXT2 tumor suppressor genes and these patients have a 0.5% to 5% risk of developing chondrosarcoma (3). In fact, most peripheral chondrosarcomas are thought to arise within the cartilaginous cap of preexisting benign osteochondromas. The cause of enchondromatosis is presently unknown. Patients affected by this disease have multiple benign cartilage tumors (enchondromas) in the medulla of the bone, which in as many as 15% to 30% of the patients transform into secondary central chondrosarcoma (4).

Chondrosarcomas are most often found in the ribs, pelvis, and the long bones of the extremities, but any bone may be affected (2). Because these tumors are largely insensitive to radiotherapy and currently available chemotherapy, the treatment of chondrosarcoma is based on surgery. The single most important prognostic marker is histologic grade but, unfortunately, the reliability of malignancy grading has been shown to be low (5). New prognostic markers are therefore urgently needed. Ideally, such markers should differentiate not only high-grade from low-grade tumors but also between the different morphologic types of chondrosarcomas and between benign chondromas and low-grade malignant chondrosarcomas.

Recurrent genetic aberrations have been shown previously in other tumors types to constitute diagnostic markers and have been used to predict progression of the disease. Cytogenetic studies of chondrosarcomas have revealed a wide variation of chromosomal aberrations, ranging from karyotypes with a single numerical or structural rearrangement to highly complex, hyperhexaploid karyotypes⁹

(6). In most cases, however, conventional chondrosarcomas present a nonrandom cytogenetic profile dominated by numerical changes, findings that agree well with results of comparative genomic hybridization (CGH) studies (7, 8).

In the present study, we used high-resolution array CGH to assess copy number changes in a series of 67 chondrosarcomas that have been characterized with regard to clinical data, morphology, grade, tumor size, and karyotypic features. Furthermore, in a subset of the tumors, the copy number alterations could be compared with the global gene expression pattern.

Patient data and tumor samples. Tumor samples were collected from 67 patients with skeletal chondrosarcoma treated at the Department of Orthopedics at both the Lund University Hospital and the Karolinska Hospital. These tumors represented all skeletal chondrosarcomas collected at the Department of Clinical Genetics, Lund University Hospital, during a 20-year period from 1987 to 2006, and from which frozen material was available. The patients included 44 men and 23 women, with a median age of 60 years. Fifty-nine tumors were diagnosed as conventional, 6 as dedifferentiated, and 2 as clear-cell chondrosarcoma. In 28 patients, the tumors were located in the extremities (21 upper and 7 lower). The remaining 39 tumors were located in the ribs (18), pelvis (12), scapula (5), sacrum (2), and sternum or vertebral columna (1 each). Based on radiographic appearance, 41 tumors were classified as central, 15 as peripheral, 1 as periosteal, and 10 as of unclassifiable origin. The size of the tumors varied from 3 to 35 cm; 48 were ≤10 cm, 17 were >10 cm, and 2 were of unknown size. Using a three-grade scale, 10 tumors were classified as low-grade (grade 1) and 56 as high-grade (43 grade 2 and 13 grade 3) chondrosarcomas according to the Evans grading scheme (9). In 1 case, the grade was unknown. Follow-up ranged from 2 to 300 months. Three patients were lost to follow-up, and of the remaining 64 patients, 21 developed metastases and 22 died from their disease. Detailed patient information can be found in Supplementary Table S1.

Chromosome banding analysis. Fresh tumor samples were processed for G-banding analysis as described previously (10), and karyotypes were described according to the guidelines in ISCN 1995 (11). Twenty-nine of the karyotypes have been reported before (6, 12–14), including metaphase CGH in 8 of them (7).

BAC array CGH. DNA copy number analysis was done using tiling microarrays containing >32,000 partly overlapping BAC clones, generating complete coverage of the human genome. The arrays were produced at the Swegene DNA Microarray Resource Center, Department of Oncology, Lund University,¹⁰

as described previously (15), using BAC clones mapped to the hg17 genome build. Extraction, labeling, and hybridization of genomic DNA from freshly frozen tumor biopsies as well as pretreatment and washing of slides were done as described (16). As a control for normal copy number, a DNA pool derived from multiple healthy male donors was used (Promega).

Oligo array CGH. In 5 cases (cases 2, 4, 34, 36, and 67), the DNA copy number status was assessed by oligo microarrays containing ∼236,000 oligonucleotide probes (Agilent). Digestion, labeling, and hybridization of genomic DNA were done as described previously (17) using the human genome CGH 244A oligo microarrays (Agilent). The samples were hybridized against gender-matched reference DNA.

Genome-wide expression profiling. Samples displaying a tumor content of >70%, as determined from H&E-stained frozen sections, were analyzed using the Human-6 v2 Expression BeadChips (Illumina) containing >48,000 transcript probes. In samples showing a lower amount of tumor versus normal tissue, macrodissection of tumor material was done. Total RNA was extracted from fresh frozen biopsies as described previously (18). RNA quality and concentration were measured using an Agilent 2100 Bioanalyzer and Nanodrop ND-1000, respectively. Samples from which insufficient amounts of high-quality RNA could be extracted were excluded from the analysis. For each hybridization, 200 ng total RNA was processed according to the Illumina TotalPrep RNA Amplification Kit (Ambion) to produce double-stranded cDNA and subsequently to generate biotin-labeled cRNA. Following the manufacturer's specifications (Illumina), 1.5 μg biotin-labeled cRNA was hybridized onto the Human-6 v2 Expression BeadChips overnight and subsequently detected with streptavidin-Cy3 using an Illumina BeadArray Reader. Image analysis was done using the Illumina BeadStudio Application Software.

Microarray data analysis. Primary array CGH data were collected using the GenePix Pro 4.0 (Axon Instruments) and Feature Extraction 9.1 (Agilent) software programs for the BAC and oligo microarrays, respectively. The quantified data matrix was deposited into the Web-based database BioArray Software Environment (19), and image and data analyses were done as described (20). In brief, following background correction, the log₂ ratios were calculated for each spot, and unreliable features and spots not showing signal-to-noise ratios ≥5 for both channels were eliminated. Normalization of data was done using Lowess normalization and single outlier probes were removed. Thereafter, the log₂ ratios for each sample were segmented (21) followed by elimination of segments <500 kb in size. To facilitate cross-platform comparison, the segmented data were transformed into a virtual probe set by associating each platform probe to its closest virtual probe, spaced at 50 kb throughout the entire genome. Copy number alterations were determined by comparing the segmented log₂ ratios with gain/loss thresholds obtained by an adaptive scaling method (22). Segments above gain threshold were set to 1, below loss threshold to -1 and in-between to 0. Gene expression data were normalized using Rank Invariant Normalization with the Illumina BeadStudio Gene Expression Module (Illumina). Microarray data are available at GEO¹¹

using the accession no. GSE12532.

Statistical analysis. Comparisons were made with regard to the distribution of copy number imbalances in conventional central versus peripheral tumors, grade 1 versus grades 2 and 3 tumors, small (≤10 cm) versus large (>10 cm) tumors, tumors in the extremities versus other locations, young (<40 years) versus old (≥40 years) patients, and patients with poor (metastases or death of disease) versus good outcome and in cytogenetic subgroups. Group comparisons were done using the Mann-Whitney U test, adjusting the P values for multiple testing by Benjamini-Hochberg false discovery rate correction. Clustering of array CGH data was done with MeV 4.1.01 software¹²

using unsupervised hierarchical clustering analysis based on Euclidean distance and average linkage clustering. Significance Testing for Aberrant Copy Number (STAC; ref. 23) was used to asses the statistical significance of DNA copy number alterations across the entire data set of abnormal profiles as well as separately in conventional central and peripheral tumors, respectively. Regions displaying a frequency or footprint statistic P < 0.05 were considered significantly altered. Known copy number variable regions were excluded. All statistical computations were done using ternary array CGH data.

Patients with poor (metastases or death of disease) versus good outcome were evaluated with regard to their age distribution, the size of the tumors, if the tumors were of high or low grade, situated in the extremities or other locations, and centrally or peripherally located in the bone. These calculations were done using the Mann-Whitney U test or the Fisher's exact test.

Patients with adverse outcome (metastases or death of disease) were significantly older than patients with good outcome (P < 0.05). The tumors of higher grade metastasized more often, resulting in poor survival, than low-grade tumors (P < 0.01), as did conventional central tumors compared with peripheral tumors (P < 0.05). The size and the location of the tumors did not affect the outcome.

Analysis of short-term cultured cells showed clonal chromosome aberrations in 41 cases and a normal karyotype in 25 cases (Supplementary Table S1). In 1 case, no karyotype was obtained. Using the cell population with the lowest chromosome number in cases with multiple clones, 6 cases were classified as hyperhaploid, 15 as hypodiploid, 6 as near-diploid or pseudodiploid, 5 as hyperdiploid, 5 as hypotriploid, 3 as hypertriploid, and 1 as hypertetraploid. Two of the pseudodiploid cases displayed balanced rearrangements.

Of the 67 chondrosarcomas analyzed with array CGH, 59 displayed abnormal DNA copy numbers (Fig. 1). The typical case showed copy number imbalances affecting almost one third of the genome (median, 28%; range, 4-85%), with a median total size of the imbalances of 826 Mb (range, 113-2,536 Mb) and a median of 26 events (range, 3-120). Frequently affected regions detected in >25% of the cases are listed in Table 1. Significantly altered regions identified by STAC are listed in Table 2. Common breakpoints (located within 500 kb and detected in ≥3 cases) were found in 2q33.2 (204.12-204.60 Mb), 3p21.31 (47.57-47.88 Mb), 4q12-q13.1 (59.32-59.74 Mb), 5q31.1 (137.11-137.14 Mb), 9q13.1 (103.34-103.67 Mb), 9q33.3 (123.72-124.16 Mb), 12q14.1 (57.24-57.68 Mb), and 21q22.11 (31.05-31.29 Mb). In 8 tumors (cases 17, 24, 29, 33, 39, 42, 49, and 66), no acquired aberrations were detected; the few imbalances visible were considered noise or normal copy number variation.

Amplification of genomic regions. In total, 15 amplicons (defined as segments with log₂ ratios >5 times the threshold or >1) were found in 12 cases affecting chromosomes 2, 4, 5, 7, 8, 11, 12, 15, 19, 22, and X. Two amplicons were recurrent, with sequences 128.32 to 132.45 Mb in 8q24.21-q24.22 (cases 2 and 65) and 100.83 to 102.85 Mb in 11q22.1-q22.3 (cases 60 and 67), as the shared amplified region in 2 cases each.

Homozygous deletion of genomic regions. Homozygous deletions (defined as segments with log₂ ratios <7 times the negative threshold) were found in 1 case each in 1q34 (233.77-234.82 Mb; case 54), 8q24.11-q24.12 (118.80-119.33 Mb; case 54 peripheral chondrosarcoma), 11p11.2 (44.04-44.69; case 44 peripheral chondrosarcoma), and 13q33.1 (100.54-101.40 Mb; case 59). The region 21.33 to 21.97 Mb in 9p21.3 was homozygously lost in 2 cases (cases 8 and 22; both were central chondrosarcomas), and the neighboring region 21.97 to 22.71 Mb in 9p21.3 was homozygously deleted in 3 cases (cases 35, 36, and 59; 2 were central chondrosarcomas and 1 was unclassifiable).

DNA copy number imbalances associated with cytogenetic findings. Unsupervised hierarchical clustering of all cases, based on array CGH data from all autosomes, distinguished two groups of tumors with primarily large regions of losses and gains, respectively (Fig. 2). The total size of the imbalances was larger in both these groups compared with the rest of the tumors (P < 0.001; Fig. 3). The group of 6 cases with losses included 5 tumors with hyperhaploid karyotypes and 1 hypodiploid. This group presented a specific pattern of genomic alterations with significant loss of chromosome arms 1p and 18q and chromosomes 3, 4, 6, 8, 10 to 14, and 17 (P < 0.05). Thirteen cases constituted the group of tumors with a significant gain of chromosomal regions (P < 0.05). Gains included chromosomes 2, 7, 8, 15, 16, and 19 to 22, chromosome arms 5p, 6p, and 9q, and parts of chromosomes 1 (0-170 Mb) and 12 (0-54 Mb). The cytogenetic findings in this group were heterogeneous; 1 case was near-diploid, 6 were near-triploid, 1 was near-tetraploid, and 5 displayed normal karyotype. Of the 27 cases, which displayed a normal or balanced karyotype, 5 showed normal DNA copy numbers.

DNA copy number imbalances in conventional central versus peripheral tumors. The number and total size of the aberrations did not differ between conventional central and peripheral chondrosarcomas (Fig. 3). Among the 15 peripheral chondrosarcomas, 13 showed abnormal array CGH profiles. The EXT1 and EXT2 genes were homozygously deleted in 1 case each. In the haploid case 55, both EXT1 and EXT2 were hemizygously lost. In 3 additional cases (cases 43, 46, and 47), EXT2 was hemizygously deleted. In the remaining 7 abnormal profiles, no deletion of either of these genes was detected. Of the 41 central chondrosarcomas (of which 36 showed abnormal array CGH profiles), 4 and 7 cases showed hemizygous deletion of the EXT1 and EXT2 loci, respectively.

DNA copy number imbalances associated with clinical features. No differences in copy number imbalance patterns were seen with regard to tumor grade, size, or location or to patient age or outcome. Furthermore, the number and total size of the altered regions were similar in the tumors that metastasized compared with nonmetastasizing tumors (Fig. 3).

Genome-wide expression profiling was available for 17 of the 67 cases. The small number of cases investigated precluded clustering analysis based on gene expression profiles. Instead, the expression levels were used to provide information on candidate genes in selected regions showing copy number alterations. For the recurrent amplicon in 8q24.21-q24.22, gene expression data were available for one of the affected cases. The amplified region contained five annotated genes (MYC, MLZE, FAM49B, DDEF1, and ADCY8), none of which showed increased expression compared with the other cases. Expression data were not available for any of the 2 cases with an amplicon in 11q22.1-q22.3. However, this region contains a cluster of MMP genes (MMP1, MMP3, MMP7, MMP8, MMP10, MMP13, MMP20, and MMP27), the expression of which varied widely; for example, the expression level of MMP3 varied >4,000 times between the tumors with the highest and lowest expressions. The gene expression pattern was not analyzed in the 2 cases presenting homozygous deletions of the regions in 8q24.11-q24.12 and 11p11.2 containing the EXT1 and EXT2 genes, respectively. The only 2 peripheral tumors investigated did not show altered expression of these genes compared with the rest of the tumors, 13 of which were classified as central and 2 were unclassifiable. Homozygous deletions in 9p21.3, encompassing the CDKN2A and CDKN2B genes, were not detected in any of the cases with gene expression data available.

Previous cytogenetic and molecular genetic studies of skeletal chondrosarcomas have, in general, found complex alterations lacking any obvious correlation with specific subtypes.⁹ Typically, gains and losses of whole or large parts of chromosomes have been detected (6–8) as well as an increasing level of genomic alterations with increasing malignancy grade (8, 14). Here, we found copy number imbalances in as many as 90% of the tumors, and in agreement with previous studies, the alterations were generally large with, on average, one third of the genome being affected. Low-level imbalances (single copy loss and gain of one or a few copies of genomic regions) predominated; amplification and homozygous deletion were uncommon. Of the 8 tumors without detectable, acquired genomic aberrations, 5 displayed normal or balanced karyotypes, whereas 3 showed highly complex karyotypes; the latter clearly show that lack of copy number alterations at array CGH analysis is sometimes due to poor sampling.

All 6 patients with a dedifferentiated chondrosarcoma died from their disease. These tumors had a hypodiploid to hypertetraploid chromosome number and presented a heterogeneous pattern of copy number imbalances. No obvious aberration distinguishing these chondrosarcomas from the conventional central ones was seen and the number and size of the aberrations were not different between the two groups. Nor could any distinguishing feature between the 2 clear-cell chondrosarcomas be discerned.

The ploidy level seen at chromosome banding analysis varied extensively. Notably, none of the 15 peripheral chondrosarcomas was hyperdiploid; 6 were hypodiploid or pseudodiploid, 1 was hyperhaploid, 7 displayed a normal karyotype, and in 1 case no karyotype was obtained. In contrast, 9 of the 41 central tumors displayed a hyperdiploid-hypertriploid karyotype, 13 were hypodiploid or pseudodiploid, 5 were hyperhaploid, and 14 had a normal chromosome complement. These findings are in contrast to previous data (24), suggesting that central chondrosarcomas typically have a near-diploid DNA content. Possibly, this could be explained by the larger proportion of high-grade tumors in the present series. In the array CGH analyses, the results were always, irrespective of the ploidy level detected at G-banding analysis, correlated with the diploid chromosome content of the control DNA. Obviously, this means that the hyperhaploid cases show loss of a vast number of chromosomes, whereas near-triploid tumors will have massive gain of chromosomes. Thus, it was not surprising that hierarchical clustering identified the near-haploid and near-triploid tumors as two distinct groups. Striking, however, were the opposite patterns of chromosomal imbalances detected in the two groups. The deleted chromosomes in the near-haploid group were not affected in the near-triploid group. Likewise, the gained chromosomes in the near-triploid group were normal in the near-haploid group. This means that, irrespective of whether the tumors showed heavy loss or massive gain of chromosomes, the net outcome was the same relative chromosomal imbalances. This finding provides further support for the notion that a substantial subset of chondrosarcomas starts as hyperhaploid/hypodiploid tumors, followed by polyploidization, a hypothesis previously put forward based on cytogenetic as well as loss of heterozygosity data (24–26). However, both near-haploid and near-triploid chromosome numbers were most frequently found in central tumors of high grade. It therefore does not seem as if polyploidization is specific for peripheral chondrosarcomas, as previously suggested based on results from smaller series of chondrosarcoma (24, 27).

Regions affected by chromosomal imbalances in more than one-fourth of the investigated cases included gain of regions on chromosomes 5, 12, and 19 to 22 and loss of parts of chromosomes 1, 4, 6, 9 to 11, 14, and 17 and whole chromosome arm 13q. Although these regions most probably harbor genes of importance for chondrosarcoma development, we extended the analysis using STAC to find specifically targeted regions of gain or loss that would harbor interesting candidate genes. STAC is a statistical method evaluating the significance of copy number abnormalities across an entire sample set or within a subset of samples (23). To be able to find alterations specific for conventional central and peripheral tumors, respectively, these variants were analyzed separately. In addition, to locate alterations of importance for any chondrosarcoma variant, the entire data set consisting of all abnormal cases was investigated. Using this approach, several regions harboring interesting candidate genes were identified (Table 2). Unfortunately, due to lack of gene expression data on a sizeable amount of cases with or without alterations of these regions, we could not evaluate the genomic effects on gene expression levels.

Although amplifications and homozygous deletions were relatively infrequent findings, these mechanisms seem to be important in a subset of chondrosarcomas. An amplicon in 8q24.21-q24.22 was identified in 2 cases and this region was gained in an additional 10 cases, all of which were high-grade tumors. An overlapping region has been shown previously to be gained and amplified in predominantly high-grade chondrosarcomas, and the MYC gene has been suggested as a potential target gene (28). However, in a recent study, nuclear MYC protein expression was found in only 8 of 70 central chondrosarcomas (29). In addition, although included in the amplicon, we did not find overexpression of MYC in the tumors with amplification or gain compared with the rest of the cases (6 of the 17 cases with gene expression data showed gain or amplification of MYC). Thus, although the amplicon at 8q24 seems to affect chondrosarcoma progression, the relevant target gene(s) remains to be identified. A recurrent amplicon was also identified in 11q22.1-q22.3, affecting a region harboring a cluster of MMP genes. These genes encode matrix metalloproteinases, and interestingly, 3 of the genes in the cluster (MMP3, MMP7, and MMP13) have recently been shown to be highly expressed in chondrosarcoma (30).

Of the homozygous losses identified in the present study, the deletions affecting the CDKN2A gene are the ones most obviously affecting chondrosarcoma development. Although both CDKN2A and CDKN2B were homozygously deleted in 3 cases, the deletion in 2 additional cases only targeted CDKN2A. In agreement with this, loss of function of the encoded protein is a frequent finding in chondrosarcoma (29, 31) as well as in other bone tumors (20). Furthermore, the region harboring CDKN2A was hemizygously lost in ∼50% of the remaining tumors with abnormal profiles, with no discernible difference between central and peripheral tumors. In support of its importance for chondrosarcoma treatment, it is noteworthy that the intrinsic resistance of chondrosarcoma cells to radiation therapy has been suggested to in part be due to loss of this protein (32).

There was no difference between low-grade and high-grade tumors with regard to number or size of genomic aberrations. Adverse outcome was associated with increasing age of the patient as well as higher grade of the tumor, in agreement with previous findings (1, 2). In a previous study comprising 59 cytogenetically investigated chondrosarcomas (partially overlapping with the cases of the present study), we found loss of material from chromosome 13 to be associated with an increased risk of metastasis (6), a finding that could not be confirmed at array CGH analysis. To some extent, this discrepancy can be explained by the inherent tendency of chromosome banding analysis to exaggerate chromosomal losses due to the inability to resolve marker chromosomes in complex karyotypes. In the present series, patients with peripheral chondrosarcomas had a better outcome than patients with central tumors; none of the peripheral tumors metastasized and only one of these patients died from the disease. In contrast, one third of the conventional central tumors metastasized and the same number of patients died from the disease. This can in part be explained by the fact that the conventional central tumors more often were of higher histologic grade. Nonetheless, these findings underline the importance of correctly distinguishing central from peripheral chondrosarcomas, a distinction that is currently based largely on radiologic information. Peripheral chondrosarcomas are believed to almost exclusively arise from the transformation of a benign osteochondroma (24), and this idea is supported by the documented occurrence of osteochondromas in 9 of 15 patients with peripheral tumors of the present study. Osteochondromas can occur either in a hereditary setting, caused by constitutional mutation of the EXT1 or EXT2 genes (33, 34), or as sporadic lesions. Also in sporadic osteochondromas is the EXT1 gene frequently affected; in a recent study of 8 nonhereditary osteochondromas, using a BAC array platform similar to ours, all showed large hemizygous deletions and 5 showed homozygous deletions of this gene (35). Furthermore, in both forms of osteochondroma and in peripheral chondrosarcoma, decreased expression levels of EXT1 and EXT2 have been reported (36). Here, we found homozygous deletions of both EXT1 and EXT2 in peripheral chondrosarcoma, which has not been described previously. Of the 13 peripheral tumors with an abnormal array CGH profile, 1 case each showed homozygous deletion of EXT1 and EXT2, respectively. Four additional cases displayed hemizygous loss of the regions covering one or both of these genes. Thus, in total, 6 of 13 peripheral chondrosarcomas presented deletions affecting either EXT1 or EXT2. We cannot, of course, exclude the possibilities that some of the remaining 7 samples had deletions that were missed due to admixture of normal cells or that some peripheral chondrosarcomas were misclassified. Furthermore, EXT1 or EXT2 may be affected through other mechanisms, such as point mutations combined with copy number neutral loss of heterozygosity or deletions too small to be detected by the array platform used here (35, 36). Still, the fraction of peripheral chondrosarcomas with deletions affecting EXT1 was far from the frequency reported in osteochondromas, and in contrast to previous findings, the expression levels of these genes were not significantly different between central and peripheral chondrosarcomas. Thus, although the findings of homozygous deletions underline the importance of EXT1 and EXT2 in the development of peripheral chondrosarcoma, our molecular genetic results indicate that a subset of these tumors may develop through other mechanisms.

In conclusion, although conventional central and peripheral tumors shared many genomic alterations, some appeared to specifically affect one of the subtypes. It will thus be of interest to study the consequences of these aberrations at the transcript level and, for those associated with peripheral chondrosarcoma, to search for the same genetic alterations in osteochondroma.

No potential conflicts of interest were disclosed.

We thank Prof. Ulf Strömberg (Department of Occupational Medicine, Lund University Hospital) for valuable help with the statistical analyses and Ronald Duim (Department of Pathology, Leiden University Medical Centre) for the expert technical assistance with the expression array.