Chondroblastoma (CB) and chondromyxoid fibroma (CMF) are benign tumors of bone morphologically recapitulating cartilage differentiation. CMF can resemble high-grade central chondrosarcoma (HGCCS) because of its cellular atypia. The mechanism that drives this morphologic spectrum of cartilage differentiation is unclear.
CMFs and CBs were hybridized on a complementary DNA microarray that was enriched for cartilage-specific genes. Data were analyzed by Linear Model for Microarray Analysis and were compared with previous data on osteochondromas and HGCCS. Verification was performed in an extended series.
None of the 68 genes that were differentially expressed in CB versus CMF, including several extracellular matrix (ECM) and ECM-degradation genes, were related specifically to cartilage. Perlecan, versican, collagen 4A2 (Col4A2), and cell-cell adhesion genes, such as CD166, were significantly higher in CMF. Sixty genes were expressed differentially in CMF versus HGCCS. Higher expression levels of CD166, cyclin D1 (CCND1), and p16INK4A were observed in CMF.
Intrinsic to its name, cartilaginous tumors are so-called because of their resemblance to normal cartilage in terms of cellular morphology and quality of the produced extracellular matrix (ECM).1 Chondroblastoma (CB) and chondromyxoid fibroma (CMF) are 2 rare, benign cartilaginous tumors that are characterized by distinct clinical and histologic features.2, 3 CB usually affects skeletally immature individuals, with a typical location in the epiphysis of long bones. Its name reflects the presence of neoplastic cells that resemble chondroblasts (Fig. 1A). Because a clear-cut cartilaginous matrix is lacking, the pattern of signaling molecules and the quality of ECM have led to the hypothesis that the neoplastic cells are committed toward bone/hyperthrophic chondrocyte differentiation.4, 5
CMF has a wide age range of occurrence and most frequently affects the metadiaphysis of long bones.3 It is made up of lobules of extracellular myxochondroid matrix in which spindled and stellated cells are embedded (Fig. 1B).3 We demonstrated previously that, within the histology of CMF, the different steps of in vitro chondrogenesis are reflected by the presence of round cells that are more similar to mature chondrocytes and the presence of less differentiated spindle-cell precursors that show partial myofibroblastic transdifferentiation.6, 7 The polygonal-atypical cells in CMF sometimes closely resemble the cells of high-grade central chondrosarcoma (HGCCS). This can make the differential diagnosis difficult in clinical practice, especially in biopsy specimens.
Thus, despite their cartilaginous phenotype, cartilage tumors of bone display a wide spectrum of differentiation. CB and CMF represent the extremes of the spectrum: from CMF, which features spindle cells/partially myofibroblastic precursors and chondrocyte-like, neoplastic cells with an accordingly myxochondroid extracellular matrix, to CB, which has features more suggestive of hypertrophic chondrocyte/immature bone differentiation. Osteochondroma (OC), a benign, cartilage-capped, bony outgrowth of bone, bears a close resemblance to the growth plate, whereas, in conventional chondrosarcoma, the cells resemble proliferating chondrocytes.8, 9
To identify the molecules that drive this wide spectrum of differentiation among the different cartilaginous tumors of bone, we performed a genome-wide expression array of CB and CMF. Expression profiles were compared with those that had been obtained previously for OC and HGCCS.10, 11
MATERIALS AND METHODS
Snap-frozen tumor material was available from 8 patients with CB and from 7 patients with CMF. These samples were used for complementary DNA (cDNA) microarray analysis. For verification, 4 and 5 extra snap-frozen samples of CMF and CB, respectively, were obtained through a European Network of Excellence for studying the genetics and pathology of bone tumors (EuroBoNeT). Formalin-fixed, formic acid (pH 2.1)-decalcified, and paraffin-embedded archival tumor tissue was available for 20 cases of CMF (12 of which were biopsies), 12 cases of CB (4 of which were biopsies), and 36 HGCCS (grade 2 and 3 according to the Evans grading system; all from resected specimens12). The other cases were retrieved from the surgical pathology and consultation files of the Leiden University Medical Center. Clinical data are given in Table 1. All specimens were handled according to the ethical guidelines, as described in the Code for Proper Secondary Use of Human Tissue in the Netherlands of the Dutch Federation of Medical Scientific Societies.
Table 1. Clinical Data and Inclusion of Cases in the Experiments Performed
Only samples that consisted of ≥70% tumor cells, as estimated on hematoxylin and eosin-stained frozen sections, were used. RNA was isolated as described previously.13 The amount of RNA was determined with a spectrophotometer (Ultrospec 3100 pro), and the integrity of the RNA was assessed by gel-electrophoresis.
cDNA Microarray Experiments
Gene expression profiling was performed as described previously.11 Briefly, an in-house, spotted cDNA array containing 8696 cDNA clones that were enriched for cartilage-specific genes was used.10, 11 All spots were printed in duplicate. This platform had been used successfully before and enabled a comparison with data previously published by our group.10, 11 It was chosen originally to overcome problems related to the low yield of RNA in cartilage tumors.10, 11, 13 Slides were hybridized with labeled cDNAs (indocarbocyanine and indodicarbocyanine) that were synthesized from 1 μg of total RNA from the samples as well as from a reference panel that consisted of 15 different cell lines.10, 11 The between-spot correlation was used as a correction factor in the final model to find differentially expressed genes. For 2 cases each of CB (CB3 and CB6) and CMF (CMF3 and CMF16), slides were hybridized twice, either as duplicates or using dye-swap. Using hierarchical clustering, these technical replicates clustered directly adjacent to each other, implying reproducibility of the technique used. Therefore, we randomly selected 1 of the technical replicates of each sample to avoid spurious effects of averaging replicates.
Statistical Analysis of Microarray Data
For the statistical analysis, Linear Model for Microarray Analysis (limma) was used.14 The procedures applied were according to the limma user guide. In the current study, the limma approach was preferred because of its ability to handle a limited number of arrays and the advantage of using within-array duplicates correlation, which is more precise than just averaging the duplicate spots.14 A conservative approach with a moderated, empirical Bayes t statistics adjusted with a false-discovery rate procedure for multiple testing was used. Because of the prevalence of missing data through selection in preprocessing, the B-statistic or log odds was used to select differentially expressed genes.14 For this calculation, a cut-off value of >1.5 was used, which grossly corresponds to a probability of 82% that the gene is expressed differentially. It turned out in the actual analysis that this cut-off value corresponds to a Benjamini and Hochberg adjusted P value of approximately .005.15 CB and CMF results were compared further with available data on OC10 and HGCCS.10, 11 Paired intersections of the sets of differentially expressed genes were used for OC, CB, and CMF. Specifically, these were chosen to represent the spectrum of cartilaginous differentiation in benign lesions and, thus, in the absence of more complex genetic abnormalities. Using intersections of gene lists from 2 comparisons that involved the same group (ie, CB vs CMF and CB vs OC), we were able to identify a more specific profile of genes that specifically were up-regulated or down-regulated in CB and CMF. This approach is visualized by using a Venn diagram.
Quantitative reverse transcriptase-polymerase chain reaction (Q-PCR) analysis was performed as described previously.6 ECM genes, such as collagen 4A2 (COL4A2), versican, perlecan, and procollagen-lysine 2-oxoglutarate 5-dioxygenase 2 (PLOD2), which is involved in posttranslational modification of ECM, were investigated further because of their more likely causative role in determining the spectrum of differentiation. Primers are described in Table 2. For each gene, a calibration curve of the same reference panel of cell lines that were used for microarray analysis was included to allow calculation of relative starting quantities of each gene. These were used in the normalization and statistical analyses. Genes for normalization were identified previously.10 Normalization was based on geometric averaging of these normalization genes.16
Table 2. Sequences of the Primers Used for Messenger RNA Expression Studies
To investigate whether gene up-regulation also led to protein up-regulation and to determine which cell type expressed the protein, we performed immunohistochemistry. Four-micrometer sections of formalin-fixed, paraffin-embedded material were used according to standard laboratory procedures.4 Antigen retrieval was performed by boiling tissue sections in citrate buffer, pH 6.0, in a microwave oven for 10 seconds. Sections from CB and CMF series were then incubated overnight with anticathepsin K (polyclonal; Santa Cruz, Heidelberg; dilution 1 : 200); and sections from CB, CMF, and HGCCS series were incubated with a primary antibody against CD166 (monoclonal, clone MOG/07; Novocastra, Newcastle; dilution 1 : 80). The slides were evaluated by 2 pathologists independently (S.R. and P.C.W.H.), as described previously in detail.17, 18 It has been demonstrated that this scoring system is highly reproducible on decalcified bone tumor specimens.4, 18 The cellular localization (nuclear, cytoplasmic, and membranous) of immunopositivity was also noted. Scores that were generated with the same evaluation system for p16INK4A and cyclin D1 (CCND1) in CMF and HGCCS were available from previous studies.7, 19, 20 A cut-off value >2 was applied for the sum of intensity and the number of positive cells for p16INK4A and CCCND1, and a cut-off value >3 was applied for activated leukocyte adhesion molecule (ALCAM/CD166). Cut-off selection was set as the minimal value for the sum of intensity and percentage that best discriminated CMF versus HGCCS. The Fisher exact test was applied to evaluate significant differences.
cDNA Microarray Expression Analysis
CB versus CMF
Limma analysis generated a list of 68 genes with expression levels that differed significantly between CB and CMF (Fig. 2A), including 31 genes that were up-regulated significantly in CB and 37 genes that were up-regulated significantly in CMF. The expression of genes that were considered specific for cartilage differentiation (such as SOX9, collagen II, aggrecan, cartilage-linking protein, fibromodulin, and cartilage oligomeric matrix protein) did not differ significantly between CB and CMF. However, several ECM and ECM-degradation genes were expressed differentially: Heparan sulphate proteoglycan 2 (HSPG2/perlecan), chondroitin sulphate proteoglycan 2 (CSPG2/versican), collagen 4A2 (COL4A2), laminin βa 2 (LAMB2), ADAM metallopeptidase with thrombospondin type 1 motif 6 (ADAMTS6), matrix metallopeptidase 2 (MMP2), and matrix metallopeptidase 13 (MMP13) had greater expression in CMF; whereas PLOD2 expression was higher in CB (Fig. 2A). Genes known to be highly expressed in osteoclastic giant cells appeared to be expressed at significantly higher levels in CB, including superoxide dismutase 1 (SOD1), acid phosphatase 5-tartrate resistant (ACP5), and cathepsin K (CTSK), reflecting the substantial number of osteoclastic giant cells that can be found in CB. Genes involved in cell-cell adhesion, such as cadherin 11 type 2 (CDH11), FYN oncogene related to SRC, FGR, YES (FYN), FAT tumor-suppressor homolog 1 (FAT), and ALCAM/CD166, were expressed at significantly higher levels in CMF.
CMF versus HGCCS
Sixty genes had significantly differential expression levels in CMF compared with HGCCS, including 41 genes that were up-regulated in HGCCS and 19 genes that were up-regulated in CMF (Fig. 2B). Among these were the cell cycle regulators CCND1 and cyclin-dependent kinase 10 (CDK10), which were up-regulated in CMF, and E2F transcription factor 5 (E2F5), which was up-regulated in HGCCS. ALCAM/CD166, which had higher expression in CMF compared with CB, also was increased compared with HGCCS.
Twenty genes specifically were characterized CB (Fig. 3). Among these, versican and perlecan specifically were down-regulated. CMF had a specific gene expression profile that consisted of 10 genes (Fig. 3).
Q-PCR confirmed the results of microarray analysis on an extended series. mRNA expression levels of perlecan, versican, and COL4A2 were significantly higher in CMF compared with CB (P < .001). PLOD2 had higher mRNA expression in CB versus CMF; however, the difference did not reach statistical significance (P = .1) (Fig. 4).
Cathepsin K protein appeared to be expressed strongly in giant cells, which are far more numerous in CB than in CMF. Some weak staining was observed in tumor cells surrounded by matrix both in CMF and in CB (Fig. 1C,D). Higher ALCAM/CD166 expression was observed in CMF (P < .05) compared with both CB and HGCCS (Fig. 1E-G), confirming the 21-fold and 28-fold difference, respectively, that we observed in the microarray experiments. In particular, in CMF, staining was more prominent in the spindle-cell component at the periphery of the lobules. Staining decreased in the matrix-rich areas in the center of the lobules (Fig. 4C). A meta-analysis of CCND1 immunohistochemistry confirmed the significantly higher expression levels in CMF versus HGCCS: 67% of CMF samples versus 20% of HGCCS samples were positive for CCND1.7, 19 In addition, the expression of p16INK4A, a molecule that is involved in cell cycle progression (although with an inhibitory action),7, 20 was significantly higher in CMF versus HGCCS (P < .05). Specifically, 67% of CMF samples versus 27.5% of HGCCS samples were positive for p16INK4A. The pattern of staining was homogenous throughout the evaluated slides for HGCCS. No significant differences were observed from testing biopsy slides or excision specimens of CMF or CB.
The current results demonstrated specific gene expression profiles for CB and CMF, supporting the notion that these are 2 distinct entities. The genes involved in cartilage formation were expressed similarly. This is in line with previous reports, which demonstrated the homogeneous expression of cartilage-specific genes in both benign and malignant cartilaginous tumors, including the expression of SOX9, collagen II, aggrecan, cartilage-linking protein, fibromodulin, and cartilage oligomeric matrix protein.10, 11, 21, 22 The results confirm that these tumors display a “real” cartilaginous nature, which is intrinsic to their classification. If the expression of cartilage-specific genes is shared between all types of cartilaginous tumors, then the differences in their spectrum are likely to be driven by “noncartilaginous” genes (ie, genes that are not identified strictly as pertinent to cartilage only), as confirmed by our results. For instance, the adhesion molecules and ECM/ECM-degrading molecules that resulted from our comparison may interfere with or modulate cartilage formation.
The adhesion molecule ALCAM was up-regulated in CMF. ALCAM is an adhesion molecule that acts through both homophylic and heterophylic interaction by CD6 binding.23 In vitro experiments in articular chondrocytes have demonstrated that the expression of ALCAM and other cell adhesion molecules correlates with efficient cartilage formation,24 and we have demonstrated high expression in CMF, specifically in the spindle-cell component at the periphery of the lobules. This pattern of distribution is strikingly similar to what was observed previously for the adhesion molecule N-cadherin in CMF.7 During cartilage condensation in skeletal development, the mesenchymal stem cells—through both cell-cell interaction and cell-ECM binding—get committed toward cartilage formation.25 This step is characterized by the transient expression of specific adhesion molecules (ie, N-cadherin). In CMF, the sustained cell-cell adhesions through ALCAM may well account for the spindle-cell phenotype resembling ongoing condensation.7
In addition to adhesion molecules, specific ECM molecules, such as versican and perlecan, are also expressed transiently during cartilage condensation.26 The chondroitin-sulphate proteoglycan versican is expressed in several different types of tissue but has a key role in the commitment of precursors toward efficient cartilage formation.26 The heparan sulphate proteoglycan perlecan also has an important role in the condensation and in promoting further cartilage differentiation.27 It is noteworthy that both versican and perlecan expression is specifically down-regulated in CB. The lack of a clear-cut cartilaginous morphology in CB may result from a defective condensation and differentiation caused by a decreased level of “noncartilaginous molecules,” such as perlecan and versican. In CMF, the cartilaginous features may be the result of high expression levels of versican and perlecan. In addition, the different appearance of ECM in CMF also may be driven by a higher level of COL4A2, because collagen Type 4 is not present in normal cartilage.28 Significantly different levels of mRNA expression of COL4A2 in normal cartilage versus malignant cartilaginous tumors have been reported.29–31
SOD1, ACP5, and CTSK are up-regulated in CB, and it is known that they have high expression levels in osteoclastic giant cells. CTSK is a member of the papain cysteine proteinases family, which is involved in a variety of physiologic processes, such as enzyme activation and inactivation, antigen presentation, hormone maturation, tissue remodeling, and bone matrix resorption.32 In particular, it has been demonstrated that CTSK is expressed highly in osteoclasts and in the giant cells from giant cell tumors.32 Accordingly, high CTSK expression levels were observed mainly in the giant cells of CB. The expression in other cell components of both tumors is in line with the reported expression in malignant cartilaginous tumors, for which this molecule seems to have a role in ECM remodeling.33
Although CMF is a benign tumor, and HGCCS is an aggressive and metastasizing malignancy, the distinction between CMF and HGCCS can be difficult at the histologic level.34 In the current study, we demonstrated increased expression of cyclin D1 in CMF compared with HGCCS. It is noteworthy that 2 other molecules that also are involved in the same cell cycle progression pathway are expressed differentially. CDK10 also is up-regulated in CMF, and E2F5 is up-regulated in HGCCS. We investigated p16INK4A because of its inhibitory action on cyclin D/cyclin-dependent kinases complex phosphorylation of retinoblastoma protein. The binding of p16INK4A to the complex induces an allosteric change in CDK4/CDK6, thereby altering the binding site of D-type cyclins and reducing their affinity for adenosine triphosphate, thus inhibiting cell cycle progression.35 High expression of CCND1 in CMF is unusual for a benign tumor; however, the co-occurrence of high p16INK4A expression may balance its effect.7 Moreover, the lower expression of p16INK4A in HGCCS underscores the significance of its loss in chondrosarcoma progression, which was reported previously.20
ALCAM/CD166 also had greater expression in CMF than in HGCCS. This is in agreement with the impression that ALCAM up-regulation is specific for CMF, as discussed above. In HGCCS, ALCAM down-regulation may be the result of loss of cell-cell adhesions in malignant progression. Accordingly, experimental models have demonstrated ALCAM down-regulation in sarcomas with more aggressive behavior.36
In conclusion, we have demonstrated the differential expression of molecules that are not cartilage matrix-specific but have a role in regulating crucial steps of chondrogenesis, such as cell-matrix interaction, condensation, and differentiation. It is likely that these molecules are responsible for the spectrum of intertumoral and intratumoral cartilaginous differentiation. The immunohistochemical assessment of p16INK4A, CCND1, and ALCAM, which are easily applicable in routine clinicopathologic assessment, appears to be a promising tool for the differential diagnosis of CMF versus HGCCS.
We thank H. can Beerendonk, P Wijers-Koster, and I. Briaire-de Bruijn for their technical assistance. We also thank the EuroBoNeT partners for providing additional samples of CMF.