Proteomic analysis of Medulloblastoma reveals functional biology with translational potential

Abstract

Genomic characterization has begun to redefine diagnostic classifications of cancers. However, it remains a challenge to infer disease phenotypes from genomic alterations alone. To help realize the promise of genomics, we have performed a quantitative proteomics investigation using Stable Isotope Labeling by Amino Acids in Cell Culture (SILAC) and 41 tissue samples spanning the 4 genomically based subgroups of medulloblastoma and control cerebellum. We have identified and quantitated thousands of proteins across these groups and find that we are able to recapitulate the genomic subgroups based upon subgroup restricted and differentially abundant proteins while also identifying subgroup specific protein isoforms. Integrating our proteomic measurements with genomic data, we calculate a poor correlation between mRNA and protein abundance. Using EPIC 850 k methylation array data on the same tissues, we also investigate the influence of copy number alterations and DNA methylation on the proteome in an attempt to characterize the impact of these genetic features on the proteome. Reciprocally, we are able to use the proteome to identify which genomic alterations result in altered protein abundance and thus are most likely to impact biology. Finally, we are able to assemble protein-based pathways yielding potential avenues for clinical intervention. From these, we validate the EIF4F cap-dependent translation pathway as a novel druggable pathway in medulloblastoma. Thus, quantitative proteomics complements genomic platforms to yield a more complete understanding of functional tumor biology and identify novel therapeutic targets for medulloblastoma.

Keywords: Medulloblastoma; Pediatric brain tumor; Proteomics; SILAC.

Fig. 1

SILAC proteomics workflow and output. a The Labeled Atlas of Medulloblastoma Proteins (LAMP) was generated by combining equal amounts of isotopically labeled (Lysine-¹³C_6, and Ariginine¹³C₆) proteins from 8 primary and established cell lines, representing the four primary medulloblastoma subgroups. Protein lysates from tissues were spiked 1:1 with the SILAC reference atlas (LAMP). The resulting protein lysates were fractionated on a 1-D gel in triplicate, trypsin digested, and further fractionated by HPLC in line with the MS/MS analysis. The spectra were then searched against the Uniprot or custom protein databases to identify peptide sequences and their originating proteins. Protein quantitation is derived from the ratio of the light tissue peptides relative to the heavy SILAC reference peptides. The proteomic data was functionally integrated with tumor-matched transcriptomic, epigenomic, and genomic data. b Venn diagram of quantified proteins along the 4 medulloblastoma subgroups and cerebellum tissues. A total of 2901 proteins were quantified

Fig. 2

Correlation between mRNA and protein abundance. a mRNA and protein were positively correlated for most (87%) pairs of mRNA-proteins with a mean Spearman’s correlation of 0.31, but only 45% showed a significant correlation (p < 0.05). b Density plot of Spearman’s correlation by medulloblastoma subgroup; a significantly lower Spearman correlation mean was found in Group 4, 0.16 compared to 0.30, 0.24 and 0.21 in Group3, SHH and WNT respectively. c Different biological processes displayed differences in mRNA and protein correlation. Genes encoding ribosomal and metabolic functions showed higher mRNA-protein correlation than those involved in the spliceosome or proteasome. d Relationship between mRNA-protein correlation and the stability of the molecules. Human mRNA and proteins are divided into stable or unstable and the distribution of mRNA-protein Spearman’s correlation is represented by box-plots. Unstable proteins have significantly lower correlations than stable ones; however, no differences were found for mRNA. p-values were calculated using the two-sided Wilcoxon rank sum test

Fig. 3

Effect of copy number alteration on MRNA and protein abundance. Frequency of significant correlations between CNA and mRNA (upper panel) or protein (middle panel) across all chromosomes. The heatmap (lower panel) indicates the copy number for each sample

Fig. 4

Proteomic subgroup classification recapitulates genomic subgroups. Non-negative matrix factorization consensus clustering of protein expression data from 36 primary medulloblastoma and five normal cerebellar tissues reveals six (Cophenetic Coefficient, k = 6) subgroups. This result largely recapitulates the subgroups found by genomic data

Fig. 5

CALD1 and HMGA1 isoforms in medulloblastoma tumors. Schematic representation of CALD1 (a) and HMGA1 (b) isoforms. The boxplots show the quantification of each protein isoform group across all medulloblastoma subgroups. p-values for differences between subgroups were calculated based on the Kruskal-Wallis rank-sum test. A protein group is defined as the group of isoforms that are indistinguishable due to the position of identified peptides

Fig. 6

Medulloblastoma subgroup specific upstream regulators. Top upstream regulators predicted by Ingenuity pathway analysis from downstream proteins differentially expressed by subgroup. Upstream regulators are predicted to be active if colored red and inhibited if colored green

Fig. 7

EIF4F inhibitors reduce medulloblastoma cell viability. D556 at normal (10%FBS) and nutrient deprivation (2% FBS) conditions (a) and D556 and primary human cerebellar astrocytes cells (b) were treated with the EIF4F inhibitor 4EGI-1 for 72 h at the indicated concentrations. The 4EGI-1 toxic concentration in medulloblastoma D556 cells is considerably lower than in normal cerebellar astrocytes. Error bars indicate the standard deviation. c Treatment of D556 and primary human cerebellar astrocytes cells with 4EGI-1 in combination with cisplatin