Background

Osteosarcoma is a malignant bone forming neoplasm with a broad spectrum of morphologies [1]. Conventional high-grade osteosarcoma is the most common subtype and typically arises in the long bones of the extremities with a predilection to the metaphysis, although any bone of the skeleton may be affected. The incidence of osteosarcomas has a bimodal age distribution with a first peak in the second decade of life and a second smaller peak in older adults [1, 2].

The etiology of osteosarcomas is still largely unknown. However, several genetic syndromes (e.g. Li-Fraumeni and hereditary retinoblastoma) greatly increase the risk of developing osteosarcomas, especially in younger patients [3, 4]. Osteosarcomas in older patients often develop from a precursor bone lesion that is prone for malignant transformation (i.e. Paget disease) or may occur after radiation treatment for other diseases, typically after a latency period of several years [1, 5, 6].

The genomic landscape of osteosarcomas is complex [7]. A remarkable attribute of primary and secondary conventional osteosarcomas is their chromosomal instability leading to high aneuploidy [8]. Chromosomal regions frequently affected by somatic structural variations and copy number alterations often contain cell-cycle regulating genes such as TP53, RB1, or c-MYC [9, 10]. Sequencing studies revealed recurrent mutations affecting these genes, supporting their major role in osteosarcoma development [8]. However, the mutational make-up of osteosarcomas is versatile, as some osteosarcomas also present with molecular signatures reminiscent of BRCA1/2 deficient tumors [10]. Chromothripsis, a single cellular catastrophic event resulting in extensive DNA rearrangements and DNA copy number changes, has been observed in roughly one-third of osteosarcomas [11].

Alternative lengthening of telomeres (ALT), a telomere maintaining mechanism compensating progressive telomere attrition by homologous recombination of telomere sequences, is associated with chromosomal instability in osteosarcomas [12]. Almost all osteosarcomas utilize ALT, highlighting its particular importance in these tumors [8, 13]. Somatic mutations in the H3.3-ATRX-DAXX chromatin-remodeling pathway have been linked to ALT in different tumor entities [14, 15]. ATRX (α-thalassaemia/mental retardation syndrome X-linked) mutations, which result in a loss of its nuclear expression, were found in osteosarcomas at a frequency of 20–30% [8, 16]. Loss of DAXX (death-domain associated protein) expression has not yet been observed in any sarcoma [17]. However, a high frequency of mutations in the two genes encoding histone 3 variant 3 (H3.3) was revealed in certain bone tumors. Giant cell tumors of bone (GCTBs) almost always carry mutations in H3F3A at codon 34, whereas chondroblastomas are characterized by a high frequency of H3F3B mutations at codon 36 [18,19,20]. Little is known about H3.3 mutations in osteosarcomas. Previous studies described single high-grade osteosarcomas carrying H3.3 mutations, whereas other series did not. So the question arises whether osteosarcomas carrying H3.3 mutations really exist or might rather be malignant variants of GCTBs [8, 18,19,20,21].

The present study addresses these issues by determining the frequency of H3.3 mutations across the entire age spectrum in de novo high-grade osteosarcomas of the conventional type. Genome-wide DNA methylation and DNA copy number profiling was applied to gain further molecular insight into osteosarcomas carrying H3.3 mutations.

Methods

Tissues specimens

Tissue samples of 106 de novo high-grade osteosarcomas of the conventional type (86 primary tumors, 3 recurrences, 17 metastases) and 14 GCTBs, one of them with signs of malignancy in the local relapse two years after surgical treatment, serving as H3.3 G34 mutant bone tumor control group were collected from the Institutes of Pathology of the University Hospital Heidelberg, Cologne, Jena (Germany), and Nijemegen (the Netherlands). Patients’ characteristics are outlined in Additional file 1: Table S1. Tissue samples were banked in the archives of the Department of Pathology and administered by the tissue bank of the National Center for Tumor Diseases (NCT), Heidelberg. Initial diagnoses were established by interdisciplinary oncology boards and were based on radiological findings, standard histopathological criteria in conjunction with immunohistochemical and molecular analyses according to the current WHO classification [1]. H3.3 mutant osteosarcomas were correlated with clinical information and were histologically and radiologically re-evaluated where available. The local ethics committee approved these analyses.

DNA extraction and Sanger sequencing of H3F3A/B mutational hotspots

Source material for DNA extraction was formalin-fixed and paraffin-embedded (FFPE) tumor tissue. Representative tumor tissue with highest available tumor content was chosen for genomic DNA isolation, as microscopically documented on H&E stained paraffin sections. In GCTBs, which are known for their prominent non-tumoral giant cell component, areas with predominant tumor cells (at least 70%) were selected for DNA extraction. The Maxwell® 16FFPE Plus LEV DNA Kit was applied on the automated Maxwell device (Promega, Madison, WI, USA) according to the manufacturer’s instructions. Polymerase chain reaction was performed with 20 ng of DNA under conditions as previously described [19]. Two microliters of the amplification product were submitted to bidirectional sequencing using the BigDye Terminator v3.1 Sequencing Kit (Applied Biosystems, Foster City, CA, USA). Sequences were determined using an ABI 3500 Genetic Analyzer (Applied Biosystems) and the Sequence Pilot version 3.1 (JSI-Medisys, Kippenheim, Germany) software.

Genome-wide DNA methylation data generation and pre-processing

The Illumina Infinium HumanMethylation450 (450 k) array was used to assess the DNA methylation status of 482,421 CpG sites (Illumina, San Diego, USA), according to the manufacturer’s instructions at the Genomics and Proteomics Core Facility of the German Cancer Research Center (DKFZ) Heidelberg. DNA methylation data were normalized by performing background correction and dye bias correction (shifting of negative control probe mean intensity to zero and scaling of normalization control probe mean intensity to 10,000, respectively). Probes targeting sex chromosomes, probes containing multiple single nucleotide polymorphism and those that could not be uniquely mapped were removed. In total, 438,370 probes were kept for analysis.

Unsupervised clustering, copy number profiling and identification of differentially methylated regions

For unsupervised hierarchical clustering, we selected 20,000 probes that showed the highest standard deviation (SD) of beta values across the entire dataset. Samples were hierarchically clustered using the 1-Pearson correlation coefficient as distance measure and average linkage. DNA methylation probes were reordered using Euclidian distance and complete linkage.

Copy number profiles were generated from the 450 k raw data using the ‘conumee’ R package and assessed manually. The dmpFinder method of the R package minfi was used for the identification of differentially methylated regions [22]. Both packages are available from the Bioconductor repository (http://www.bioconductor.org).

Statistical analysis

A two-sided t test was used to test for differences between the mean values for continuous variables. P values of less than 0.05 were considered significant.

Results

H3F3A mutations occur in high-grade osteosarcomas of elderly patients

Targeted sequencing of the two genes encoding histone 3 variant 3 (H3F3A and H3F3B) in 106 de novo high-grade osteosarcomas revealed six H3F3A mutant cases (Fig. 1a). One case carried a c.80A>T heterozygous mutation leading to an amino acid exchange of lysine by methionine at codon 27 (K27M). Another osteosarcoma presented with a c.100G>C heterozygous mutation leading to an exchange of glycin (G) by arginine (R) at codon 34. Four cases carried a c.100G>T mutation, one hemizygous and three heterozygous, leading to an exchange of glycin (G) by tryptophan (W). No mutation was detected in H3F3B. Patients with H3F3A mutant osteosarcomas were diagnosed at higher age (median age 65 years; range 34–75 years vs median age 18 years; range 2–86 years) and were significantly older (p < 0.05) compared to the control group of H3F3A mutant GCTBs (median age 29 years, range 18–75 years). Hence, the incidence increased from overall 5% (6/106) to 15% (6/40) in osteosarcoma patients over 30 years of age (Fig. 1b, c). The six H3F3A mutant osteosarcomas affected the long bones and were distributed similar to H3.3 wild-type osteosarcomas and GCTBs. They occurred in the proximal and distal femur, proximal tibia, proximal fibula, or the distal radius (Fig. 1c; Table 1).

Fig. 1
figure 1

H3F3A mutations and patients characteristics of the study group. a Chromatograms showing the H3F3A sequence spanning codon 27 and 34 illustrate the wild-type status and mutations that exchange either the amino acid lysine with methionine (K27M), glycin (G34) with tryptophan (G34W) or with arginine (G34R) b Charts show the overall incidence of H3G3A G34W mutations in our series of osteosarcomas over all age groups and in the group of patients older than 30 years. c Osteosarcomas are separated in wild-type and mutant regarding their H3F3A status. Giant cell tumors of bone act as control. The age distribution is demonstrated for each case. The black bar indicates the median age. Gender distribution is illustrated in the form of a diagram. The site of occurrence is only shown for cases that were further analysed for methylome and copy number profiling. Each ring/dot represents a case. Red color codes for cases that were further analyzed for methylome and copy number profiling

Full size image
Table 1 Clinical characteristics of the six H3.3 mutant osteosarcomas
Full size table

Radiological features and histological patterns are indistinguishable between H3F3A mutant and H3.3 wild-type osteosarcomas

Radiological data was available for re-evaluation in five of six H3F3A mutant osteosarcomas (Fig. 2; Additional file 2: Figure S1). At radiographic examination, the tumors to a variable extent demonstrated a mixed pattern of sclerotic opacities with mineralization and lytic areas violating the cortex. The infiltrative growth pattern was associated with expansive growth into the surrounding soft tissue. Some cases presented with prominent soft-tissue masses (e.g. Fig. 2b). On MR imaging, where available, these masses partly wrapped around the circumference of the affected bone (Additional file 2: Figure S1).

Fig. 2
figure 2

Radiologic appearance and morphological phenotypes of H3F3A mutant osteosarcomas. a Anteroposterior radiograph of the left wrist shows a lesion in the epi- and metaphysis of the distal radius (left). Histology revealed an osteoblastic osteosarcoma with pronounced nuclear atypia, brisk mitotic activity and a compact, extensive sclerosing osteoid matrix. b Lateral radiograph of the right knee shows a lesion in the distal femoral metaphysis with infiltration of the epiphysis and prominent extension into the soft tissue. The tumor was composed of pleomorphic, relatively plump and focally spindle-shaped tumor cells depositing osteoid in plaques (mixed osteoblastic/fibroblastic osteosarcoma). c Lateral radiograph of the right knee shows a lesion in the distal femoral meta- and epiphysis with subchondral extension and infiltration of the joint. The tumor (mixed osteoblastic/fibroblastic osteosarcoma) was predominantly composed of plump tumor cells producing coarse plaque-like osteoid. d Lateral radiograph of the right knee shows a lesion in the proximal fibula (caput fibulae). Histology revealed an osteoblastic osteosarcoma with highly pleomorphic tumor cells focally producing coarse and lace-like osteoid deposits. e Anteroposterior radiograph of the hip joint shows a lesion in the intertrochanter and subtrochanter region of the proximal femur. Histology revealed a mixed osteoblastic/fibroblastic osteosarcoma with coarse osteoid production. f X-ray imaging of the proximal tibia was not available in case 79,428. Histologically this bone lesion presented as osteosarcoma (mixed osteoblastic/fibroblastic subtype) with neoplastic bone production that was focally arranged in a coarse lace-like pattern. Foci with necrosis and dystrophic calcification were noted. White arrows indicate the corresponding lesion (a–e). Scale bars equal 50 µm (a–f)

Full size image

All six H3F3A mutant cases showed histological features indistinguishable from other conventional high-grade osteosarcomas (Fig. 2; Additional file 3: Figure S2). Mitotic activity was variable with presence of atypical mitotic figures. Neoplastic bone formation was observed in all H3F3A mutant osteosarcoma, although to a variable extent. Of note, giant cells were sparse to absent in all six H3F3A mutant osteosarcomas.

DNA methylation profiling separates H3F3A G34 mutant osteosarcomas from H3.3 wild-type counterparts

We next investigated the six H3F3A mutant and 28 H3.3 wild-type osteosarcomas and 14 H3F3A G34W/L mutant GCTBs for genome-wide DNA methylation patterns applying the 450 k DNA methylation array (Fig. 3). Unsupervised hierarchical clustering identified two distinct DNA methylation clusters (Fig. 3, upper panel). The first cluster was composed of H3.3 wild-type osteosarcomas and the single osteosarcoma carrying a H3F3A K27M mutation. This particular case was indistinguishable from that of H3.3 wild-type osteosarcomas. The second methylation cluster was composed of the five H3F3A G34 mutant osteosarcomas and the H3F3A G34 mutant GCTB control cases. Within this cluster, the 14 GCTBs showed the most homogeneous DNA methylation patterns and accordingly grouped together (Fig. 3, heatmap with dendrogram).

Fig. 3
figure 3

Methylome and copy number profiling. The heatmap depicts unsupervised hierarchical clustering of methylation levels of the top 20,000 most variant probes (highest standard deviation). Each row represents a probe and each column a sample. The level of DNA methylation (beta value) is represented with a colour scale as depicted. Fourteen giant cell tumors of bone carrying H3F3A G34 mutations served as control group. For each sample (n  =  48), histological diagnosis, mutational status of H3F3A and selected chromosomal copy number alterations are indicated

Full size image

We next analyzed copy number data derived from the DNA methylation arrays (Fig. 3, lower panel). H3F3A mutant osteosarcomas frequently showed a gain of chromosome arm 8q involving the c-MYC locus (3/6) and a distinct deletion of 3q13.31 involving LSAMP (2/6). However, c-MYC amplification, 6p12–21 amplification or chromothripsis, both hallmark copy number alterations in osteosarcomas, were absent in all six H3F3A mutant osteosarcomas. GCTBs lacked any copy number alterations except for the single GCTB with a secondary malignant transformation (Additional file 4: Figure S3).

H3F3A G34 mutant osteosarcomas present with a promoter methylation phenotype

Given the distinct DNA methylation patterns of H3F3A G34 mutated osteosarcomas we next examined differences in DNA methylation in specific genomic regions (Fig. 4). H3F3A G34 mutant and H3.3 wild-type osteosarcomas showed widespread hypomethylation across the whole genome, especially in non-promoter regions, which was particularly prominent at centromeres and telomeres (Fig. 4a). However, H3F3A G34 mutant osteosarcomas had higher methylation levels in promoter regions compared with their H3.3 wild-type counterparts. Specifically, we found that the promoter region of the genes HIST1H2BB (p < 0.00005) and KLLN/PTEN (p < 0.0005) showed significantly higher methylation levels in H3F3A G34 mutant osteosarcomas (Fig. 4b).

Fig. 4
figure 4

Distinct DNA methylation patterns in osteosarcoma subgroups and giant cell tumors of bone. a Upper part the empirical cumulative distribution function for DNA methylation levels (beta-values) is plotted individually for each subgroup. Lower part DNA methylation levels (mean beta values) regarding different genomic regions of individual osteosarcoma methylation subgroups. b Further DNA methylation analyses identified the HIST1H2BB and KLNN/PTEN promoter as the most differentially methylated region when comparing the H3F3A G34 mutant osteosarcoma subgroup (OS G34) with the H3.3 wild-type osteosarcoma group (OS H3.3 wt). Significant difference to the H3F3A G34 mutant osteosarcoma subgroup (OS G34) is indicated

Full size image

Discussion

H3.3 mutations in bone tumors occur with an exceedingly high frequency in GCTBs and chondroblastomas [18, 19]. We herein present for the first time a clinical and molecular characterization of six H3F3A mutant de novo osteosarcomas and suggest that H3.3 mutations might define a distinct molecular subtype in the heterogeneous group of osteosarcomas.

Looking at the H3F3A G34 mutational status and the age distribution alone raises the possibility that the group of H3.3 mutant osteosarcomas may be composed of GCTBs presenting with Denosumab-therapy induced changes mimicking an osteosarcoma-like histology, may have undergone malignant progression or may represent the very rare de novo malignant variant of GCTBs, the latter two of which are known for their occurrence in elderly [23,24,25]. It is also evident that osteosarcomas in elderly are often of secondary origin, which would further argue in this direction [1].

Indeed, at least for the H3F3A G34 mutant osteosarcoma cases presented here, there is no absolute certainty that they might not represent the most undifferentiated end in a spectrum of GCTB differentiation. However, none of the H3F3A mutant osteosarcomas presented with distinct radiological and histological features resembling GCTBs or its malignant variant, even though the radiologic appearance of malignant GCTBs may be challenging to distinguish from osteosarcomas [26]. We also cannot preclude with absolute certainty that areas reminiscent for a GCTB were missed due to a sampling bias in the H3F3A mutant osteosarcomas, which would argue for the extremely rare de novo malignant variant of GCTBs. However, this can be considered very unlikely. Furthermore, the patients were treated according to standard treatment protocols for osteosarcoma, thereby excluding Denosumab-induced changes. It is also unlikely that a precursor lesion underwent malignant transformation, since medical records were empty in this regard (Table 1). Therefore, taking all information into account, the diagnosis conventional osteosarcoma has been retained in all six H3.3 mutant cases. The H3F3A mutant osteosarcoma cases of this study predominantly occurred at higher patient age. This finding might explain why other osteosarcoma studies focusing on pediatric patients alone did not find H3.3 mutations at all [8]. However, larger future studies will be required to confirm that H3F3A mutant osteosarcomas are rare or even absent in children and adolescent patients. Interestingly, the age distribution of the H3.3 mutant osteosarcomas is in contrast to H3.3 mutant brain tumors, which typically present at younger age than their wild-type adult counterparts [15, 27]. This observation coupled with evidence from studies about H3F3A G34 mutant GCTBs might suggest an entity dependent age distribution of H3.3 mutant tumors [18, 19].

Our integrated approach revealed several molecular alterations in H3F3A mutant osteosarcomas that are typical for high-grade osteosarcomas. A homozygous deletion of 3q13.31 carrying the LSAMP gene was seen in two of the six H3F3A mutant osteosarcomas. One of them additionally carried a TP53 deletion, a prominent tumor suppressor gene known to be involved in the development of osteosarcomas [8]. LSAMP is a tumor suppressor that is frequently inactivated in conventional osteosarcomas and linked to increased proliferation. It is also a predictor for an unfavorable biological behavior [28]. Thus, an aggressive biological behavior typically seen in conventional osteosarcomas might also be assumed for osteosarcomas with H3F3A mutations. Accordingly, three of four H3F3A mutant osteosarcoma patients with follow-up data available showed an unfavorable clinical course. Two patients died of the disease shortly after diagnosis. One patient presented with widespread metastases and progressive disease (Table 1).

A gain of chromosome 8q bearing the c-MYC locus was detected in three of six H3F3A mutant osteosarcomas. Amplification of c-MYC, a member of the MYC transcription factor family tightly involved in the modulation of gene expression and a strong oncogene, has been frequently observed in osteosarcomas [7, 8, 10]. It has also been shown in mice that c-MYC overexpression in mesenchymal bone stem cells may induce osteosarcomas [29]. The gain of 8q might confer a tumor propagating gene dosage effect, although c-MYC is not amplified in the six H3F3A mutant osteosarcomas presented here. In this context, other MYC gene family members may also be pivotal in the development of H3F3A mutant osteosarcomas. Brain tumors with H3F3A G34 mutations have been associated with an increased expression of MYCN, another transcription factor well known for its oncogenic capacity [30]. Expression analysis might give further insights whether c-MYC, MYCN, and/or other oncogenes orchestrate together in H3F3A G34 mutant osteosarcomas. The H3F3A G34 mutation alone will probably not be sufficient to induce osteosarcomas, since GCTBs carrying such mutations usually do not undergo malignant transformation [1].

We also detected one osteosarcoma with a H3F3A K27M mutation. This mutation was initially found in diffuse pontine midline gliomas and has recently been described also in other brain tumor entities [27, 31]. The H3F3A K27M mutation has been demonstrated leading to an enzymatic inhibition of EZH2 in brain tumors, a catalytic subunit of Polycomb Repressor Complex 2 [32]. EZH2 catalyzes methylation of histone H3 at lysine 27 and mediates gene silencing of target genes via local chromatin reorganization. Dysregulated EZH2 activity is seen in many cancer types [33]. We here report the first H3 K27M mutant tumor outside the nervous system. It therefore remains to be determined whether the underlying molecular mechanisms of H3.3 K27M mutation described in brain tumors may be adapted to other cancer types.

Interestingly, the DNA methylation analysis revealed several differing aspects in H3F3A mutant osteosarcomas compared to their H3.3 wild-type counterparts. Our data indicate that the DNA methylation signature of G34 mutant osteosarcomas is different to their conventional wild-type counterparts, but more closely related to GCTBs. However, this difference was only seen for G34 mutant cases, since the single K27M mutant osteosarcoma of this series was indistinguishable from H3.3 wild-type cases. On the one hand, the close relation between H3F3A G34 mutant osteosarcomas and H3F3A G34W/L mutant GCTBs might suggest a similar cell of origin. On the other hand, their close relation at the methylation level might reflect similar changes in the epigenetic landscape caused by the H3F3A G34 mutation or a responsible higher order mechanism. However, the DNA methylation profile of a dedifferentiated chondrosarcoma, in which we detected a H3F3A G34L mutation, was neither related to GCTBs, nor related to H3F3A G34W/R mutant osteosarcomas (data not shown).

In-depth analysis of the methylome data revealed widespread DNA hypomethylation in both, H3F3A G34 mutant and wild-type osteosarcomas. DNA hypomethylation is generally more pronounced with increased aggressiveness, which might explain this pattern [34]. However, the hypomethylation phenotype was less pronounced within the promoter regions in H3F3A G34 mutant osteosarcomas compared to their wild-type counterparts. The most differentially methylated promoter regions involved the HIST1H2BB and KLNN/PTEN genes. Both showed significantly higher methylation in H3F3A G34 mutant osteosarcomas. The role of HIST1H2BB in cancer is largely unknown. HIST1H2BB is an integral element of the nucleosome complex and hence, impaired expression could disrupt the physiological status and affect nucleosome remodelling [35]. KLNN is located juxtaposed to PTEN on chromosome 10q23 and both genes share the same promoter site [36]. Accordingly, both genes probably might be affected in H3F3A G34 mutant osteosarcomas. An important role of PTEN in the development of osteosarcomas has not been described yet, although PTEN alteration has been observed [8, 10, 37]. The neighboring gene KLLN functions at the pericentric region of chromosomes. KLLN organizes the heterochromatin by maintaining the H3K9 trimethylation and thereby contributes to chromosomal stability. Loss of KLLN resulted in the dysregulation of pericentric heterochromatin with consequent chromosomal instability and numerical chromosomal aberrations in vitro [38]. Interestingly, hypomethylation was significantly pronounced around the pericentric region in H3F3A G34 mutant osteosarcomas compared with GCTBs. Whether the hypomethylation phenotype may be a surrogate for pericentric instability and the role of KLLN in context of H3.3 mutations will be subject of future projects.

Conclusions

Taken together, our data confirm that detection of H3F3A mutations in bone tumors does not exclude malignancy in individual cases, especially in older patients. H3F3A mutant conventional high-grade osteosarcomas may radiologically, histologically and karyotypically be indistinguishable from conventional high-grade osteosarcomas, but present with distinct epigenetic features.