Distinct IDH1/2-associated Methylation Profile and Enrichment of TP53 and TERT Mutations Distinguish Dedifferentiated Chondrosarcoma from Conventional Chondrosarcoma

Dedifferentiated chondrosarcoma (DDCS) is a rare high-grade chondrosarcoma characterized by a well-differentiated chondrosarcoma (WDCS) component that abruptly transitions to a high-grade, noncartilaginous sarcomatous component. To date, the molecular pathogenesis of DDCS and its distinction from conventional chondrosarcoma remain poorly understood. By targeted sequencing, we examined the mutational and copy-number profiles of 18 DDCS, including macrodissected WDCS components, compared with 55 clinically sequenced conventional chondrosarcomas. In conjunction with publicly available external data, we analyzed the methylation and expression profiles of 34 DDCS and 94 conventional chondrosarcomas. Isocitrate dehydrogenase 1/isocitrate dehydrogenase 2 (IDH1/IDH2) mutations were present in 36% conventional chondrosarcomas and 71% DDCS. Compared with conventional chondrosarcomas, DDCS had higher frequencies of TP53 and TERT promoter mutations and CDKN2A/B copy-number losses. Paired analysis of macrodissected WDCS and the high-grade components revealed TERT promoter mutations as early events. Despite phenotypic similarities, the percentage of genome with copy-number alterations in DDCS was significantly lower than that in other high-grade sarcomas. Differential methylation analysis revealed reduction of IDH1/IDH2-associated global hypermethylation characteristically seen in conventional chondrosarcoma and a distinct methylation profile in DDCS. The WDCS and high-grade components in DDCS showed similar methylation profiles. These CpG sites were associated with upregulated expression of genes involved in G2–M checkpoints and E2F targets. Genomic profiling revealed enrichment of TP53, TERT promoter, and CDKN2A/B alterations in DDCS. Integrated methylation and gene expression analysis revealed distinct IDH1/IDH2-associated methylation and transcriptional profiles as early events in DDCS, which may underlie the pathogenesis of dedifferentiation in chondrosarcomas. Significance: DDCS is a rare, high-grade chondrosarcoma with a dismal prognosis. About 50%–80% of DDCS harbor IDH1/IDH2 mutations. We uncover a significant alteration of IDH-associated methylation profile in DDCS, which we propose is key to the progression to dedifferentiation. In this context, the potential effect of the use of IDH inhibitors is unclear but important to address, as clinical trials of selective IDH1 inhibitors showed worse outcome in DDCS.

Early molecular genetic studies have demonstrated that in the same tumor, both WDCS and the high-grade, noncartilaginous sarcomatous components of DDCS share common genetic alterations in TP and copy-number alterations (CNA) at focal chromosomal loci, with the high-grade component displaying additional genomic aberrations such as aneuploidy and more extended CNAs, implying the two components are derived from a single precursor (11). The same group of investigators demonstrated that 50% and 85% of DDCS harbor alterations in CDKNA and TP, respectively, which are nonspecific oncogenic alterations seen in many sarcoma types (12).
Subsequent studies focusing on dissecting the molecular distinction between the WDCS and high-grade sarcomatous components of DDCS cases have shown variable results. One study showed that the two components shared TP and PTEN alterations in an isolated case of DDCS (13). A recent study of paired WDCS and high-grade components of 11 DDCS cases showed that both components harbor IDH/, COLA, and TERT mutations, whereas TP and large-scale CNAs were more common in the high-grade component (14). This was concordant with an early study showing TP mutation and LOH in TP and RB (15). About 50%-80% of DDCS also harbors IDH/IDH mutations, with no clear prognostic impact (12,16,17). A recent phase I clinical study of ivosidenib (AG-120), a selective mutant IDH1 inhibitor, in patients with advanced IDH-mutant CS showed worse outcomes in DDCS compared with conventional CS, suggesting a different biology between these two histotypes despite sharing IDH mutations (18).
To date, the molecular pathogenesis of DDCS and its distinction from conventional CS remain poorly understood. Using MSK-IMPACT, we examined the mutational and copy-number profiles of 17 DDCS cases, including macrodissected WDCS components when available (eight cases), in comparison with 55 conventional CS cases. As methylation is a marker of cell lineage, and conventional CS and WDCS have similar histologic features, we wanted to assess whether they share similar or distinct cell lineage or undergo alterations predisposing to a dedifferentiated phenotype through methylation analysis. To this end, we analyzed the methylation patterns in conventional, WDCS, and DDCS. Using an in-house Illumina EPIC array platform for DDCS, in conjunction with external publicly available methylation and gene expression data, we analyzed the methylation profiles of overall 33 DDCS and 94 conventional CS.

Case Selection and Study Cohort
Cases were identified from the Memorial Sloan Kettering Cancer Center (MSKCC) surgical pathology archives from 2013 to 2021. Written informed consent was obtained from patients for use of genomics data for research. This study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board. Criteria included cases with an explicit diagnosis of DDCS and conventional grade 1-3 CS. Clinical charts and pathology reports were reviewed to document patient age, sex, tumor site, tumor grade, and outcome data. For DDCS cases, the hematoxylin and eosin sections were manually reviewed to select and circle areas of the dedifferentiated and WDCS components, which were then macrodissected for downstream DNA sequencing and methylation profiling. Because most blocks containing well-differentiated cartilaginous components were subjected to acid-based decalcification, only in seven cases of DDCS cases were the WDCS components available for further downstream processing, with an additional case of WDCS without the corresponding high-grade dedifferentiated component.

Survival Analysis
Survival analysis by comparison of HRs using log-rank P testing and visualization of Kaplan-Meier curves were performed using R packages "survminer" version 0.4.9 and "survival" version 3.2.13. Clinical charts were manually reviewed to document date of initial presentation, disease progression, and survival status. Median time (in years) to disease progression was defined as the time interval between initial presentation (presence of tumor seen radiographically or on physical examination) and the first instance of tumor recurrence or distant metastases after initial surgical resection and/or chemoradiation therapy with radiographically negative evidence of residual tumor.

Targeted DNA Sequencing, Copy-number and Mutational Profiling and Data Analysis
Detailed descriptions of MSK-IMPACT workflow and data analysis, a hybridization capture-based targeted matched tumor-normal DNA NGS assay targeting 341 to 505 genes for solid tumor were described previously (19,20). After excluding metastatic tumors, a total of 18 DDCS cases were sequenced (17 dedifferentiated components, of which seven also had macrodissected nondecalcified WDCS component sequenced concurrently, and one DDCS case with only WDCS component without the corresponding dedifferentiated component) and analyzed by the MSK-IMPACT pipeline. In all cases, the tumor is macrodissected and we were able to obtain a relatively pure population, estimated at 50%-70% based on pathologist assessment and variant allele frequency (VAF). Between WDCS and DDCS, the DDCS component has a higher tumor cellularity and content. In addition, clinical MSK-IMPACT data were available for 55 conventional CS.
We use the reference standard sample that is a mixed positive control pool consisting of known positive samples with different classes of alterations previously validated by MSK-IMPACT as our positive quality control sample. For every run, this reference sample was run to verify that we are able to detect hotspot gene mutations, including insertions/deletions and single-nucleotide variants at an expected VAF of 5%-20%, copy-number variants, and structural variants. For negative control, DNA from 10 normal diploid blood samples were pooled in equimolar ratios to create a mixed negative control sample, verified in previous runs to be free of tumor contamination and germline copy-number variants in target genes. We also use a PCR no template control (Qubit measurement < 1.0 ng/μL; ref. 19).
For somatic mutation calling, MSK-IMPACT uses genomic DNA from tumor samples with matched patient-derived normal samples from peripheral blood.
When there is no matched normal sample or the coverage of matched normal is below 50X, tumor samples will be compared against a standard, in-batch pooled normal control derived from 10 normal formalin-fixed paraffin-embedded (FFPE) samples for variant calling (19).  (21). A set of normal FFPE samples were used for reference diploid genome comparison. Normalized coverage values from tumor samples were divided by corresponding values in normal samples, and log-transformed to yield log-ratios. The criteria for gene amplification and deletions are as follows: if the fold change is greater than 2, it is reported as amplification. If the fold change is −2 or below, it is reported as a deletion. The IMPACT assay targets a total of 6,729 exons across 505 genes. In addition, the panel contains probes that tile the positions of 1,042 common SNPs, which mimic a low-density SNP tiling array with locations evenly distributed across the genome-coverage values. These positions are used to supplement the copy-number analysis in genomic regions where few targeted genes are located (19). Overlapping segments were derived using the "CNTools" package version 1.52.0. CNA was considered present if the absolute segmentation mean is greater or equal to 0.5. Fraction of CNA across the genome was calculated by dividing the sum of all segments with CNA by the sum of all segments across the genome.
Mutations and gene-level CNAs were visualized and summarized using the R package "ComplexHeatmap" version 2.8.0 (22).

DNA Methylation Profiling
Details on methylation profiling were published previously (23). Briefly, genomic DNA was extracted from FFPE tissue sections. Next, 250 ng of genomic DNA was subjected to bisulfite conversion and processed on the Illumina methylation EPIC/850k platform according to manufacturer's instructions. For methylation analysis, we included a total of 18 in-house DDCS cases, of which seven samples were macrodissected WDCS components and 11 were macrodissected dedifferentiated components, using leftover DNA following MSK-IMPACT sequencing. We also recorded the IDH R132 and IDH R172 mutational status of these cases (14 IDH/IDH-mutated, three wildtype, one unknown).
In addition, we downloaded the following external, publicly available methylation data generated by the Illumina 450k or EPIC/850k platforms: (i) 89 cases (73 conventional CS, 16 DDCS) from an integrated molecular characterization of CS study by Nicolle and colleagues: (24) ArrayExpress # E-MTAB-7263, and (ii) 21 conventional CS cases from the Heidelberg Sarcoma Classifier study by Koelsche and colleagues: (25) Gene Expression Omnibus (GEO) # GSE140668.
IDAT processing and data analysis on all 128 samples was performed using R version 4.1.0 and the "minfi" package version 1.38.0.24 (26). Normalization was performed using the preprocess Illumina function and probes with a detection P value > 0.01 were filtered, as were SNP-related probes, and probes on sex chromosomes. After probe filtering and intersecting internal and external datasets, 256,431 CpG probes remained for downstream analysis. Methylation levels were measured using beta values (ratio of the methylated probe intensity to the overall intensity-sum of methylated and unmethylated probe intensities) for all cases. CpG probes were annotated using the "IlluminaHumanMethylationEPICanno.ilm10b4.hg19" package version 0.6.0.
Gene expression microarray data were imported using R version 4.1.0 and the "ArrayExpress" package version 1.56.0. Normalization and background correction were performed using the preprocess Robust Multichip Average (RMA) algorithm from the "oligo" package version 1.60.0. Probes were annotated using the Affymetrix hugene20 annotation data from the "hugene20sttranscriptcluster.db" package version 8.8.0.
A flowchart outlining the samples and respective sources for each analytic workflow is show in Fig. 1.

Integrated Methylation and Expression Differential and Gene Set Enrichment Analysis
Differential methylation and differential gene expression analysis, respectively, between DDCS and conventional CS adjusted by IDH mutation status were performed using a two-factor (DDCS vs. conventional CS and IDH-mutant vs. IDH-wildtype) contrast matrix design with the "limma" package version 3.52.1 (27). Differentially methylated or expressed genes were computed using the empirical Bayes statistics and selected with the following parameters: P-value cutoff 0.05, adjusted by the Benjamini-Hochberg correction method. We defined hypomethylated and hypermethylated genes based on the t-statistic from the output of the "decideTests" function: a t-statistic of −1, 0, or 1 is classified as significantly negative, not significant, or significantly positive, respectively.
The same applies for differential gene expression analysis. Thereafter, genes corresponding to differentially methylated probes were matched to differentially expressed genes by intersecting hypermethylated genes to downregulated genes (decreased expression), and hypomethylated to upregulated genes (increased expression).
Gene set enrichment analysis (GSEA) was performed using gene sets downloaded from the Molecular Signatures Database (28,29). Gene ontology analysis and gene sets testing for methylation data were conducted using the "missMethyl" package version 1.30.0. Gene sets testing and visualization of mRNA expression data were performed using the "clusterProfiler" package version 4.4.4 using default parameters (30).
Unsupervised hierarchical clustering and heatmap generation were performed using the "ComplexHeatmap" R package version 2.12.0 with Euclidean distance for clustering of rows and columns (21). All data analysis was performed using R version 4.1.0.

Data Availability
The data generated in this study are publicly available in GEO GSE214180.

Clinicopathologic Summary
Among in-house cases from the MSKCC archives, 55 conventional CS and 18 DDCS cases (total 73 patients) were included in the cohort for MSK-IMPACT profiling. Patients with DDCS had a median age of 64 years old (range,  and were significantly older than those with conventional CS, who had a median age of 52 years old (range, 17-79; Student t test, P = 0.006). DDCS tended to arise from proximal appendicular skeleton (50%), for example, femur, humerus, followed by the axial skeleton (39%). In contrast, conventional CS most commonly arose from the axial skeleton (60%) followed by proximal appendicular skeleton and the head and neck (mostly skull base and larynx) at equal proportions (both at 18%). Detailed breakdown of patient characteristics is shown in Table 1.   Of the 18 DDCS cases, seven had available WDCS and high-grade components for MSK-IMPACT profiling, one had only WDCS but not the high-grade component. The WDCS (n = 8) components showed mutational and CNAs more similar to DDCS than conventional CS (Fig. 2). Among DDCS cases, those who harbor TERT promoter mutations showed borderline worse overall survival (OS) but not progression-free survival (PFS) compared with those without TERT promoter mutations (log-rank P = 0.089; Supplementary Fig. S1). Other genetic alterations were not prognostic of worse PFS or OS in DDCS, most likely due to the already dismal prognosis of this tumor type.

Mutational and Copy-number Profiling of DDCS versus Conventional CS
Paired analysis of macrodissected WDCS and the high-grade sarcoma components from 7 patients with DDCS revealed TERT promoter mutations as common, early events. On the other hand, in 3 patients, TP mutations were detected only in the dedifferentiated but not the WDCS components. Furthermore, there were acquisition of nonrecurrent additional copy-number gains and losses in the high-grade dedifferentiated component not seen in the WDCS component (Fig. 3).
Next, we analyzed and compared the percentages of genome-wide copynumber alterations among conventional CS and DDCS cases in comparison with other high-grade sarcomas. The percentage of genome involved by CNAs in DDCS was significantly lower than those in other high-grade sarcomas (osteosarcoma, LMS, UPS; Wilcoxon P < 0.0001; Fig. 4).    Fig. 5A and B), suggesting both a reduced and altered IDH-dependent methylation landscape in DDCS. In contrast, among IDH-wildtype cases, there were no significantly differentially methylated CpG sites comparing DDCS versus conventional CS cases ( Fig. 5A and B). In other words, IDHmutant DDCS exhibit methylation patterns that are distinct from IDH-mutant conventional CS.

Methylation Profiling and Differential Methylation in DDCS versus Conventional CS
On the other hand, the macrodissected WDCS and high-grade sarcoma components in DDCS cases showed similar methylation profiles, regardless of IDH/IDH mutational status, suggesting that the observed IDH/IDHassociated methylation pattern is an early event in DDCS (Fig. 6A).

Integrated Methylation and Expression Analysis
Next, to interrogate how such changes of methylation in DDCS affect gene expression, using external, publicly available Affymetrix gene expression data, we analyzed differentially methylated genes (P adjusted < 0.05) in DDCS versus conventional CS, adjusted by IDH/IDH mutational status. Among 25,667 genes that were profiled, within the IDH-mutant group, there were 910 genes that were significantly upregulated and 1,602 that were significantly downregulated in expression in DDCS compared with conventional CS. In contrast, within the IDH-wildtype group, there were 243 genes that were significantly upregulated and 374 that were significantly downregulated in expression in DDCS compared with conventional CS.
We then matched these differentially expressed genes to the corresponding differentially methylated CpG sites by correlating the genes upregulated (increased expression) in DDCS to CpG sites hypomethylated in DDCS, and vice versa. Figure 6B depicts the expression of top genes that were hypomethylated and upregulated and those that were hypermethylated and downregulated in DDCS versus conventional CS.
Gene ontology analysis showed that the differentially methylated sites between DDCS and conventional CS within the IDH-mutant group were enriched in genes associated with pathways involved with E2F targets, G 2 -M checkpoint, MYC targets, etc. (Fig. 7A). Concomitantly, GSEA of the genes upregulated in DDCS compared with conventional CS within the IDH-mutant group revealed enrichment of genes associated with pathways in G 2 -M checkpoint, E2F and MYC targets, and inflammatory/cytokine responses ( Fig. 7B; Supplementary

FIGURE 6 A, Methylation profiles of cartilaginous and sarcomatous components of DDCS. Microdissected WDCS components and high-grade
noncartilaginous sarcomatous components of DDCS cases from the MSKCC cohort were subjected to methylation profiling by the Illumina 850k/EPIC array platform. Heatmap represents beta values of the CpG sites clustered by DDCS component subtype. B, Integrated differential methylation and expression analysis in DDCS. Genes corresponding to differentially methylated probes were matched to differentially expressed genes by intersecting hypermethylated genes to downregulated genes (decreased expression), and hypomethylated to upregulated genes (increased expression) in DDCS relative to conventional CS within the IDH-mutated group. Heatmap represents the top genes from this integrated analysis with a minimum log 2 fold change of 1.5.
with conventional CS within the IDH-mutant group revealed enrichment of genes associated with pathways involved in hypoxia, epithelial-mesenchymal transition, and metabolic responses ( Fig. 7C; Supplementary Table S1).

Discussion
We performed comprehensive, integrated genomic, and methylation profiling of a sizable cohort of DDCS cases, including macrodissected WDCS components, in comparison with conventional CS cases. By targeted DNA sequencing, IDH/IDH mutations were present in 36% conventional CS and 71% DDCS cases. Compared with conventional CS, DDCS had increased tumor mutation burden and higher frequencies of TP and TERT promoter mutations and CDKNA/CDKNB copy-number losses.
In addition, paired analysis of macrodissected WDCS and the high-grade sarcoma components revealed TERT promoter mutations as common, early events in both components, but acquisition of additional copy-number gains and losses in the high-grade sarcoma component not seen in the WDCS component. These observations corroborate previous studies that showed frequent alterations in CDKNA and TP in DDCS (12), and that IDH/ and TERT mutations are shared by both WDCS and high-grade sarcomatous components, while acquisition of TP mutations and additional CNAs are seen mostly in the high-grade component only (14).
Despite phenotypic similarities, the percentage of genome involved by CNAs in DDCS was significantly lower than those in other high-grade sarcomas, that is, osteosarcoma, LMS, and UPS. To our surprise, DDCS does not seem to be  more or as genomically complex compared with other high-grade sarcomas with complex karyotypes, such as osteosarcomas, despite sharing phenotypic features, and poor survival. This suggests that the biology and clinical behavior of DDCS may be driven by epigenetic factors rather than genotype/genomic complexity.
IDH/IDH mutations in cartilage tumors were associated with an aberrant epigenome, leading to global hypermethylation and downregulated expression of genes (8,24,31,32). Expression of mutant IDH2 in mesenchymal progenitor cells led to DNA hypermethylation and an impairment in differentiation (33). This is due to increased production of the oncometabolite D-2-hydroxyglutarate (D-2-HG), an inhibitor of ten-eleven translocation (TET)-mediated DNA demethylation (32,33). In human mesenchymal stem cells, D-2-HG has also been shown to promote chondrogenic over osteogenic differentiation (34). One study suggested that this dysregulation of differentiation by 2-HG-producing mutant IDH was by suppression of histone H3K9 demethylation, which repressed expression of lineage-specific differentiation genes (35). In human mesenchymal stem cells, induction of the IDH R132C mutant led to enhanced expression of SOX and COLA via increased H3K4me3 and inhibited expression of ALPL via increased H3K9me3 (36).
Nonetheless, the prognostic impact of IDH/IDH mutations is unclear (37). While some studies have suggested an association of IDH/IDH mutations with worse overall survival (38), others have not demonstrated such associations (24,29). One study indicated that IDH/IDH mutations are associated with improved PFS but not OS in high-grade CS (39). Another recent study in central CS showed that although TERT mutations occur more frequently in IDH-mutant compared with IDH-mutant tumors, they are associated with worse survival in IDH-mutant but not IDH-mutant tumors (40).
Similar to conventional CS, about 50%-80% of DDCS also harbor IDH/IDH mutations in both WDCS and high-grade sarcomatous components (12,16,17), with no clear prognostic differences between IDH-mutant versus IDH-wildtype DDCS (17). The epigenetic alterations in DDCS, including methylation of CDKN2A isoforms and E-cadherin, shown in an early study could be accounted for by the increase in D-2-HG produced by mutant IDH in DDCS (41). Other studies demonstrated genetic alterations such as SUZ or EED alterations in H3K27me3-deficient DDCS cases (42). Another study showed that L-2-HG, another enantiomer of 2-HG, is increased in hypoxia, leading to the increased methylation of histone repressive marks such as H3K9me3 (43,44). Furthermore, a recent study showed increased expression of the HSPs HSP and HSP in DDCS (45), corroborating the findings of upregulation of HIF2α in CS progression (46).
We analyzed methylation profiling data from conventional CS and DDCS, combining our own cohort with external data. Such alterations of methylation pattern or de novo hypomethylation, we propose, is key to the progression to dedifferentiation, which ultimately leads to alteration of cellular lineage and therefore a dedifferentiated phenotype. Alternatively, rather than DDCS arising from conventional CS, our findings could instead support the theory of distinct cells-of-origin between DDCS and conventional CS, despite histologic similarities between WDCS of DDCS and conventional CS. It is possible that in a different cellular context in DDCS, mutant IDH-dependent inhibition of TET-mediated demethylation occurs at reduced number of and at alternative sites, resulting in epigenetic reprogramming in the progression to dedifferentiation. Further studies are needed to elucidate the mechanism and biological consequences of reduced and altered IDH-associated hypermethylation in DDCS. On the other hand, a recent study in central CS suggested that IDH-mutant tumors displayed significantly higher degree of global hypermethylation compared with IDH-mutant and IDH-wildtype tumors (40). Future studies looking into the differences in the methylation landscapes between IDH versus IDH-mutant DDCS may offer additional insights into the biology of different DDCS subtypes.
Importantly, such alterations in methylation patterns in DDCS also raise the question of the role of epigenetic therapies and IDH1/IDH2 inhibitors in the treatment of DDCS (49). This is especially relevant as preclinical studies have demonstrated activity of DNA methyltransferases and histone deacetylases inhibitors in the treatment of CS cell lines and xenografts (50