Abstract
Mutations in the RNase IIIb domain of DICER1 are known to disrupt processing of 5p-strand pre-miRNAs and these mutations have previously been associated with cancer. Using data from the Cancer Genome Atlas project, we show that these mutations are recurrent across four cancer types and that a previously uncharacterized recurrent mutation in the adjacent RNase IIIa domain also disrupts 5p-strand miRNA processing. Analysis of the downstream effects of the resulting imbalance 5p/3p shows a statistically significant effect on the expression of mRNAs targeted by major conserved miRNA families. In summary, these mutations in DICER1 lead to an imbalance in miRNA strands, which has an effect on mRNA transcript levels that appear to contribute to the oncogenesis.
Brief Communication
MicroRNAs (miRNAs) are small non-coding RNA molecules that regulate expression of their transcript targets [1] DICER1 is a key enzyme that is responsible for cutting the 5p and 3p strands of the pre-miRNA in the early stages of the miRNA biogenesis. Processing of the 5p and 3p strands, which is carried out by the RNase III domains of DICER1, is necessary for loading the functional miRNA strand into the RISC complex. Previous studies have identified recurrent mutations in the RNase IIIb domain in different cancer types [2, 3, 4, 5, 6, 7, 8, 9]. These mutations (at residiues E1813, D1810, D1709, E1705 and R1703) were shown to be in the active site of the enzyme and were proven to disrupt the processing of the 5p stand of the miRNA [10]. Others have shown that hotspot mutations in the RNase IIIb domain cause depletion of 5p strands relative to their corresponding 3p strands, leading to an asymmetry in the abundance of the two [11, 7].
Although the asymmetry in the miRNA processing due to hotspot mutations has been characterized using model organisms; the effect of this miRNA depletion on the mRNA levels have not been studied extensively in the context of the human tumors. It is, for example, unknown whether it is the 5p-strand depletion or increased 3p-strand accessibility that promotes the cancer. In either of the cases, it is also unknown whether there is any particular miRNA or miRNA family of which depletion or over-expression drives this phenotype. In this study, using human tumor data from the Cancer Genome Atlas (TCGA) project, we wanted to better characterize the effects of DICER1 mutations on miRNA and mRNA profiles of the patients.
We first asked whether we could observe the asymmetry in the miRNA processing using the miRNA-Seq data. For this, we looked whether any of the previously identified hotspot mutations were present in the TCGA data set (14 cancer types, 5535 sequenced samples). We found that 15 out of 123 DICER1 mutants carried a mutation in the RNase IIIb domain of the protein at a previously identified hotspot (Figure 1a). After filtering out cases that were hyper-mutated and samples that did not have miRNA-Seq data available, we were left with 8 DICER1 hotspot mutants. We then compared the miRNA levels in these hotspot mutants to the miRNA levels in 3171 DICER1 wildtype tumors across multiple cancers. Confirming the results of the previous studies, we saw 5p strand miRNAs were relatively down-regulated in mutants and the changes in the expression of 5p strands were significantly different than the 3p strands (Wilcoxon rank sum test; p < 10−29; Figure 1b-c).
Having observed a phenotype characterized by relative 5p strand depletion in hotspot RNase IIIb mutants, we asked whether any of the other DICER1 mutants had a similar phenotype. To investigate this, we first estimated the abundance of 5p strands relative to 3p strands for each patient: , where is the median expression of the x-strand miRNAs in patient i. As expected, the majority of the hotspot mutants had exceptionally low 5p-strand abundance compared to DICER1 wildtypes (Figure 1d).
In addition to the known hotspots mutants, we identified three more DICER1 mutant cases that had relatively low 5p abundance . One of these three DICER1 mutants had a hotspot mutation in its RNase IIIb domain, but was excluded from the initial analysis because it was a in hyper-mutated sample (Table S2). Surprisingly, the other two cases with low 5p abundance had an S1344L mutation in the RNase IIIa domain that is responsible for processing the 3p strand of the miRNA.
As the observation of recurrent mutations in cancer samples is consistent with a selective functional impact of the mutation, the question arises as to the effect of the S1344L mutations on the catalytic function of the RNase domains. Inspection of the 3D structure (or model) of the individual domain reveals that residue S1344L (in domain IIIa) and its homologous residue T1733 (in domain IIIb) are far from the active site residues (19.60±2.62Å distance) in their respective domains (Figure 1e). However, evolutionary couplings [12] between S1344L/T1733 and the active site residues, as deduced from co-evolution patterns in the multiple sequence alignment of RNase III-like domains, are fairly strong. The contradiction is resolved by inspection of the model of the RNase IIIa - IIIb heterodimer (as inferred from the crystal structure of the RNase IIIb homodimer) [10]. In the heterodimer, S1344L in domain IIIa is close (11.72±1.98Å distance) to active site of domain IIIb (residues E1813, D1810, D1709, E1705 and R1703) and T1733 in domain IIIb is close to the active site residues of domain IIIa. These residue arrangements and functional couplings are beautifully consistent with the observation that mutations in S1344L in domain IIIa affect 5p processing, as observed in our analysis of the effect of these mutations on the balance of 3p/5p miRNA expression profiles in cancer samples. This is consistent with the earlier observations that mutations in the active site residues of domain IIIa affect 3p processing, while mutations in the active site residues of domain IIIb affect 5p processing. The subtly of the difference between the earlier and current observation lies in the residue interactions across the heterodimer interface [13] and in fact the earlier observation of 3p/5p asymmetry are confirmed here by completely independent observation in human cancer samples.
Other studies have shown that DICER1 hotspot mutations are biallelic in cancer, where a disabling mutation acts as the second hit to the enzyme [5, 6, 8] Based on this observation, the relative 5p depletion phenotype of RNase III mutants in our analysis suggested that these patients also had a second event disabling the other DICER1 allele. To address this question, we re-analyzed the sequencing data available for DICER1 mutant cases, this time using a different pipeline that can better identify insertions or deletions. In a majority of the DICER1 RNase III hotspot mutant samples, we were able to identify a secondary disabling genomic event affecting the other DICER1 allele (Table S3). Furthermore, we found that these biallelic mutated cases had lower 5p abundance than the other DICER1 mutants in our earlier analysis.
Having identified possibly functional mutations in DICER1 and their effect on the miRNA profiles, we tested whether these mutations lead to functional changes in the mRNA profiles. Others have previously characterized DICER1 hotspot mutations using mouse-derived cell lines as in vitro models [7, 8, 11] These studies have shown that the mRNA profiles of cell lines with different DICER1 RNase IIIb hotspot mutations had different mRNA signatures compared to the DICER1-wildtype cell lines. They further found an association between the down-regulated miRNAs and their differentially-expressed target transcripts, which suggests a differential regulation of the mRNA levels due to asymmetric miRNA processing in DICER1 hotspot mutants.
Although there is in vitro evidence that the asymmetry in the miRNA processing lead to significant changes in the mRNA profiles; there are no previous reports that describe the differential mRNA expression in accordance with the miRNA expression data from human tumors. To this end, we identified 12 cases across four cancer types that both had RNA-Seq data available and carried a hotspot RNase III mutation either in the IIIa or IIIb domains of the DICER1 protein. We then wanted to check whether we could identify a common mRNA expression signature for these DICER1 RNase III hotspot mutants in comparison to 1212 DICER1 wildtype cases in those four cancer studies. For this, we decided to restrict our analysis to the Uterine Corpus Endometrial Carcinoma (UCEC) study where the RNA-Seq data set contained 8 DICER1 RNase III mutants and 222 DICER1 wild-types. We found 10 genes to be significantly up-regulated and none to be down-regulated in the hotspot mutated cases when compared to wildtypes (p < 0.05 after Bonferroni correction; Table S4). Notably, we found higher expression of HMGA2, a well-known oncogene and target of let-7 miRNA family, in mutants [14, 15, 1].
Following up on this, we asked whether the up-regulated genes in mutants were targets of particular miRNA families. To answer this question, we conducted a gene set enrichment analysis (GSEA) using well-known biological pathways and well-conserved miRNA family target genes as our query gene sets [16]. Our analysis showed strong enrichment of both let-7/98/4458/4500 and miR-17/17-5p/20ab/20b-5p/93/106ab/427/518a-3p/519d target genes in RNase III mutants (Table 1; FDR < %10). For both families, 5p strand of the miRNA is the predominant strand and as expected, in RNase III mutant cases, 5p-strand miRNAs that belong to these families were relatively down-regulated. Results from the GSEA also suggested that there was relatively weaker enrichment for other miRNA families and NOTCH-related pathways (Table 1; FDR < %15). A majority of the enriched gene sets (5 out of 7) represented miRNA family targets, which suggests the gene expression signature associated with these RNase III hotspot mutants is more likely to be mediated by depleted miRNA families rather than a common biological pathway. In accordance with the 5p strand depletion phenotype, a majority of these miRNA families (3 out of 5) were 5p-strand dominated. For the other two families, miR-29abcd and miR-101/101ab, although 3p is the pre-dominant miRNA strand, we saw that members of these families were down-regulated as a family in DICER1 mutants compared to wildtype, which might be due to an indirect regulatory effect of 5p miRNA depletion.
In summary, we showed that biallelic DICER1 RNase III hotspot mutations, although infrequent across cancers, lead to relative depletion of 5p stand of miRNAs. In addition to known hotspot mutations, we were able to identify a previously unknown recurrent DICER1 mutation, S1344, that also leads to the 5p depletion phenotype. In accordance with the miRNA depletion phenotype, we saw up-regulation of genes that are well-known targets of the 5p-dominant miRNA families in mutant samples. It still remains unclear whether up-regulation of a particular gene, such as HMGA2, or activation of a particular pathway, such as NOTCH, is contributing to the oncogenesis as a result of the 5p miRNA depletion in these cells.
Competing Financial Interests
The authors declare no competing financial interests.
Acknowledgments
We would like to thank Kjong Lehmann, Andre Kahles, Gunnar Rätsch, Özgün Babur, Pınar Aksoy, Ed Reznik, Nils Weinhold, Ruomu Jiang, Berkin Elvan for helpful discussions on the manuscript. This work was supported by US National Cancer Institute funding of the TCGA Genome Data Analysis Center (U24 CA143840).