Integrated molecular characterization of chondrosarcoma reveals critical determinants of disease progression

Abstract

Chondrosarcomas are primary cancers of cartilaginous tissue with highly contrasting prognoses. These tumors are defined by recurrent mutations in the IDH genes and other genetic alterations including inactivation of CDKN2A and COL2A1; however, these have no clinical value. Here we use multi-omics molecular profiles from a series of cartilage tumors and find an mRNA classification that identifies two subtypes of chondrosarcomas defined by a balance in tumor differentiation and cell cycle activation. The microRNA classification reveals the importance of the loss of expression of the 14q32 locus in defining the level of malignancy. Finally, DNA methylation is associated with IDH mutations. We can use the multi-omics classifications to predict outcome. We propose an mRNA-only classifier to reproduce the integrated multi-omics classification, and its application to relapsed tumor samples shows the progressive nature of the classification. Thus, it may be possible to use mRNA-based signatures to detect patients with high-risk chondrosarcomas.

Fig. 1

mRNA expression classification. a Cophenetic correlation coefficient for increasing number of clusters. Red point shows the chosen number of clusters (two) selected as the number previous to the largest decrease in the cophenetic correlation coefficient. b Heatmap of the co-classification matrix resulting from the consensus clustering approach. The matrix is symmetric and shows the frequency of co-classification of all 102 by 102 samples in the 1000 resampling iterations of the consensus clustering procedure at the selected cut of 2 clusters. c Characterization of the identified mRNA-based transcriptomic classification of chondrosarcoma using: subtype-specific marker gene expression, grading and histology features, metagenes extracted from the transcriptome profiles, general characteristic of genetic stability and gene-specific genetic alterations. When relevant, the association between the features shown as a heatmap and the two-class transcriptomic classification is shown (Student’s t-test for continuous variables or chi-squared test for discrete variables Significance of an FDR correction of the p-values are shown on the right of each line using the following encoding: ***FDR < 0.1%, **FDR < 1%, *FDR < 5%, and no symbol for FDR > 5%. d Gene Set Enrichment Analysis (GSEA) comparing E2 versus E1 samples. e Overall survival comparison of E1 and E2 tumors. Source data are provided as a Source Data file. CIN: Chromosomal instability index. Dediff Dedifferentiated chondrosarcoma, NES Normalized enrichment score

Fig. 2

microRNA classification. a Cophenetic correlation coefficient for increasing number of clusters. Red point shows the chosen number of clusters (four) selected as the number previous to the largest decrease in the cophenetic correlation coefficient. b Heatmap of the co-classification matrix resulting from the consensus clustering approach. The matrix is symmetric and shows the frequency of co-classification of all 102 by 102 samples in the 1000 resampling iterations of the consensus clustering procedure at the selected cut of four clusters. c Characterization of the identified microRNA-based transcriptomic classification of chondrosarcoma using: subtype-specific microRNA expression annotated based on their genomic position, median expression of all microRNA in the 14q32 locus, genetic state of the 14q32 locus, and grading and histology. When relevant, the association between the features shown as a heatmap and the four-class micro-RNA classification is shown (Student’s t-test for continuous variables or chi-squared test for discrete variables). Significance of an FDR correction of the p-values are shown on the right of each line using the following encoding: ***FDR < 0.1%, **FDR < 1%, *FDR < 5%, and no symbol for FDR > 5%. d Sample-centered expression of microRNA plotted by their genomic position in the 14q32 locus. Each expression value is colored by the microRNA subtype and smoothened subtype-specific mean expression value with 95% confidence interval is shown. e Gene Set Enrichment Analysis (GSEA) comparing 14q32^high versus 14q32^low samples. f Overall survival comparison of the four microRNA subtypes. g Overall survival comparison of the 14q32^high samples, corresponding to the Mir1 subtype, versus the 14q32^low subtype comprising the Mir2, Mir3 and Mir4 subtypes. Source data are provided as a Source Data file. log-FC: log fold-change. LOH loss of heterozygosity, Dediff Dedifferentiated chondrosarcoma, NES Normalized enrichment score, ECM Extracellular matrix

Fig. 3

DNA methylation classification. a Cophenetic correlation coefficient for increasing number of clusters. Red point shows the chosen number of clusters (3) selected as the number previous to the largest decrease in the cophenetic correlation coefficient. b Heatmap of the co-classification matrix resulting from the consensus clustering approach. The matrix is symmetric and shows the frequency of co-classification of all 102 by 102 samples in the 1000 resampling iterations of the consensus clustering procedure at the selected cut of three clusters. c Characterization of the identified methylation-based classification of chondrosarcoma using: grading and histology features, subtype-specific CpG methylation level, median level of all CpG found in CpG islands, components extracted from the methylation profiles, gene-specific genetic alterations and age of the patient at diagnosis. When relevant, the association between the features shown as a heatmap and the three-class methylation classification is shown (Student’s t-test for continuous variables or Chi-squared test for discrete variables). Significance of an FDR correction of the p-values are shown on the right of each line using the following encoding: ***FDR < 0.1%, **FDR < 1%, *FDR < 5%, and no symbol for FDR > 5%. d Gene Set Enrichment Analysis (GSEA) comparing IDH^mut versus IDH^wt samples. (e) Overall survival comparison of M1, M2, and M3 tumors. Source data are provided as a Source Data file. Dediff Dedifferentiated chondrosarcoma, NES Normalized enrichment score

Fig. 4

Multi-omics classification. a Schematic of a multi-omics classification of chondrosarcoma based on the three single-omics classifications. b Multi-omics classification of chondrosarcoma along each of the single-omics classification as well as its characterization using grading and histology features, and gene-specific genetic alterations. All of the 102 patient samples are represented as a column in the same order in each of the lines of the heatmaps. When relevant, the association between a feature shown as a heatmap and the 6-class multi-omics classification is shown (Student’s t-test for continuous variables or chi-squared test for discrete variables). Significance of an FDR correction of the p-values are shown on the right of each line using the following encoding: ***FDR < 0.1%, **FDR < 1%, *FDR < 5%, and no symbol for FDR > 5%. c Relative quantification of T lymphocytes cell population infiltration using MCP-counter and of d of the PDL1 immune checkpoints. Boxplots show the median, the first and third quartile and whiskers extend to 1.5 times the interquartile range. e Overall survival comparison of the 6-class multi-omics classification. f Forest plot of the multi-variate analysis of survival including grade and the multi-omics classification after the simplification of the three alteration low subtypes into one (C1, C2, and C4). g. Follow-up study of patients with mRNA profiled relapse sample. x-axis shows time after initial diagnosis and each dot corresponds to a sample, including the initial sample. Dots are colored depending on the mRNA-based multi-omics subtype prediction. The right panel summarizes the main molecular events identified in at least one the relapse sample as compared to the initial sample, if any. Dediff Dedifferentiated chondrosarcoma