Deregulation of MYCN, LIN28B and LET7 in a molecular subtype of aggressive high-grade serous ovarian cancers

Abstract

Molecular subtypes of serous ovarian cancer have been recently described. Using data from independent datasets including over 900 primary tumour samples, we show that deregulation of the Let-7 pathway is specifically associated with the C5 molecular subtype of serous ovarian cancer. DNA copy number and gene expression of HMGA2, alleles of Let-7, LIN28, LIN28B, MYC, MYCN, DICER1, and RNASEN were measured using microarray and quantitative reverse transcriptase PCR. Immunohistochemistry was performed on 127 samples using tissue microarrays and anti-HMGA2 antibodies. Fluorescence in situ hybridisation of bacterial artificial chromosomes hybridized to 239 ovarian tumours was used to measure translocation at the LIN28B locus. Short interfering RNA knockdown in ovarian cell lines was used to test the functionality of associations observed. Four molecular subtypes (C1, C2, C4, C5) of high-grade serous ovarian cancers were robustly represented in each dataset and showed similar pattern of patient survival. We found highly specific activation of a pathway involving MYCN, LIN28B, Let-7 and HMGA2 in the C5 molecular subtype defined by MYCN amplification and over-expression, over-expression of MYCN targets including the Let-7 repressor LIN28B, loss of Let-7 expression and HMGA2 amplification and over-expression. DICER1, a known Let-7 target, and RNASEN were over-expressed in C5 tumours. We saw no evidence of translocation at the LIN28B locus in C5 tumours. The reported interaction between LIN28B and Let-7 was recapitulated by siRNA knockdown in ovarian cancer cell lines. Our results associate deregulation of MYCN and downstream targets, including Let-7 and oncofetal genes, with serous ovarian cancer. We define for the first time how elements of an oncogenic pathway, involving multiple genes that contribute to stem cell renewal, is specifically altered in a molecular subtype of serous ovarian cancer. By defining the drivers of a molecular subtype of serous ovarian cancers we provide a novel strategy for targeted therapeutic intervention.

Figure 1. Molecular subtypes of serous ovarian carcinomas with clinical outcome data.

(A) Heatmap of gene expression data taken from AOCS, TCGA, NCI and Norway datasets shows that tumours are classified into 4 molecular subtypes. Genes are clustered by Pearson correlation and samples are ordered by molecular subtype. While the original K-means clustering from Tothill et al. is shown for the AOCS cohort, a supervised learning procedure was used for classification of tumours in other datasets (see Supplementary Methods S1). (B) Kaplan-Meier survival curves of samples are plotted. Overall survival is used as the endpoint in all four datasets. Cox proportional hazard model is used to compute statistical significance of the difference in survival between all four groups. Log-rank test p-value is reported. (C) Samples from all datasets were combined to estimate the survival characteristics. Subtypes were compared to C5 and the log rank test p-value given in the table. (D) Kaplan-Meier curves are plotted to depict the survival function of samples in the four different subtypes after combining the samples.

Figure 2. Oncofetal genes deregulated in the C5 subgroup.

Let-7 target genes, including HMGA2 are specifically deregulated in the C5 molecular subtype of high-grade serous cancers. (A) mRNA expression of HMGA2 is significantly higher in the C5 molecular subtype. (B) Immunohistochemical analysis of HMGA2 expression in ovarian cancer samples showing consistent over-expression in C5 tumours. Example of strong (3+) staining (top) and no staining (bottom) panel. (C) A core set of Let-7 regulated target genes (Oncofetal genes) identified by Boyerinas et al are over-represented in C5 gene signature. (D) Let-7 target genes obtained from TargetScan5.0 are enriched in C5 tumours. The significance of overlap between C5-specific and Let-7 target gene sets was determined by one-sided Fisher's exact test. Bar plot depicting the association is shown. (* indicates p<0.0001).

Figure 3. Amplification of MYCN and over-expression of MYCN and LIN28B in C5 tumours.

(A) Boxplots depict differential expression of LIN28B and MYCN in different molecular subtypes of serous ovarian cancers AOCS dataset. (B) DNA copy number levels and expression levels of MYCN are highly correlated. C5 samples (coded in red) are predominately distributed in the top right corner of the plot. Samples with segmented copy number log ratio greater than 0.3 were considered to have a gain of MYCN and samples with segmented log ratio less than −0.3 were considered to have a loss. Fisher's exact test was used to compute the statistical significance of the association. (C) Association between MYCN targets and C5 gene sets. MYCN targets were extrapolated from the intersection of two gene lists, (1) genes bound by N-myc in mouse embryonic cell lines and (2) genes with 2-fold increase in expression following transfection of N-myc in mouse embryonic cells (see Supplementary Methods S1). Using data from AOCS or TCGA gene expression sets genes were classified as high or low in C5 tumours versus all other tumours (C1–C4) at p<0.001 by two-sided t-test. The significance of overlap between C5 gene sets and MYCN gene sets was determined by a one-sided Fisher's exact test. Bar plot depicting the association is shown.

Figure 4. Amplification and over-expression of MYCN influences a regulatory loop involving LIN28B, Let-7 and HMGA2 in C5 high-grade serous tumours.

Each event is significantly enriched in C5 versus non-C5 high grade serous tumours: MYCN amplification p<0.0001; MYCN over-expression p<0.0001; over-expression MYCN targets p<0.0001; LIN28B over-expression p<0.0001; under-expression Let-7 alleles p<0.01-0.0001; HMGA2 amplification p<0.001, HMGA2 over-expression p<0.0001 (TCGA data). Cumulatively loss of Let-7b, and/or gain of HMGA2, and/or gain of MYCN occur in 77.8% of samples in C5 subtype (p<0.0001, two-sided Fisher's exact test; Table S4).