Cooperative control of tumor suppressor genes by a network of oncogenic microRNAs

Abstract

Individual microRNAs (miRNAs) have been implicated as oncogenes in experimental cancer models and their expression may affect clinical outcomes. To gain a more comprehensive view of miRNA action in leukemia, we analyzed miRNA expression patters in T-cell leukemia ALL (T-ALL) and cross-referenced the results with an unbiased genetic screen and computational analyses.¹ We found that multiple microRNAs contribute to leukmogenesis and act as multi-targeted regulators of several tumor suppressor genes. The oncomirs form a network of overlapping and partially redundant interactions that stabilize the malignant phenotype though coordinate repression of cellular failsafe programs. The emerging network pattern of oncomir action is distinct from the notion of single oncogenic “driver” mutation. We will discuss experimental, diagnostic and therapeutic implications of this concept of miRNA action in cancer.

MicroRNAs (miRNAs) are small RNA molecules that do not encode proteins and instead appear to regulate the expression of many genes. Accordingly, miRNAs exert myriad roles in normal development and diseases including cancer.² In cancer, certain microRNAs have been implicated as oncogenes or tumor suppressors.^3–6 Some specific examples include miR-155, which lies within the bic gene,^7–9 miR-125b, which is a target of a chromosomal translocation in B-ALL,^10,11 the 17∼92 cluster of microRNAs that is upregulated in many blood cancers,^12–14 or miR-15/16 which are often lost in chronic lymphocytic leukemia.^15–17 Moreover, miRNA expression studies have revealed broad mis-regulation of many micro-RNAs in diverse cancers, which may indicate a broader role for additional miRNAs in tumor biology. While individual miRNAs have been studied in some detail, a comprehensive analysis of the interactions between miRNAs in oncogensis has been lacking.

We study oncogene and tumor suppressor networks in leukemia and lymphoma and expect that insights into genetic interactions will lead to rational therapies and molecular diagnostics. For example, the clinical heterogeneity among patients with seemingly identical diseases may result, in part, from the underlying genetic diversity and enhanced diagnostics should appreciate these differences to deliver better care. Genomic tools have now become widely available that permit a rapid descriptive assessment of tumor genomes and provide insights into the complexity of genetic changes that occur during tumor evolution. A new challenge lies in discerning the relevant changes from the accompanying noise and others and we have begun to use unbiased genetic screens and also computational approaches to separate the wheat from the chaff.¹⁸ Moreover, to understand in detail the impact of genetic lesions on treatment of leukemia and lymphoma we use genetically versatile mouse models based on the adoptive transfer of hematopoietic progenitor cells.¹⁹ These cancer models enable the rapid analysis of genetic interactions and genotype-response relations in a highly controlled experimental setting. We consider genomic data, unbiased screens and accurate in vivo models as complementary approaches that enable identification and functional annotation of genetic lesions in cancer.

We recently reported a comprehensive analysis of oncogenic miRNAs in T-cell leukemia (T-ALL).¹ Briefly, we identified several candidate oncomirs by comparing miRNA expression data form leukemia samples with an unbiased miRNA library screen designed to pinpoint potentially transforming miRNAs. Notably, only a few miRNAs (the “top ten”) were abundantly expressed in leukemic cells, while most of the >400 miRNAs we analyzed were barely detectable. We assume that oncogenic miRNAs should be abundant. Strikingly, our unbiased screen revealed that the “top ten” highly expressed miRNAs were significantly enriched for transforming activities. We directly tested these candidate oncomirs in a murine model of Notch1-driven T-cell leukemia and confirmed five oncomirs (miR-19b, miR-20, miR-26, miR-92, miR-223) that were able to promote leukemia development in vivo. Hence, we identify a small group of miRNAs that are highly expressed in T-cell lymphoblastic leukemia (T-ALL) and act as oncomirs in a model of that leukemia.

MicroRNAs are thought to act primarily by destabilizing mRNAs that harbor specific binding sites/seed matches.²⁰ In silico methods exploit the sequence specificity and conservation to predict target genes,^21–24 and mRNA expression and comparative genomics tools can test these predictions and readily reveal global changes on mRNA and protein levels.^25–27 While these approaches do not address the functional role of specific miRNA target genes, we recently reported the use of genetic screens as an unbiased means to determine which targets are directly relevant to a miRNA-driven phenotype.^28,29

How do these oncomirs act in leukemogenesis? A computational analysis showed that the oncomirs we identified in T-ALL were characterized by a common set of predicted target genes. Moreover, these targets included several genes that have been implicated as tumor suppressor genes in T-ALL. Specifically, we used a machine-learning approach based on lasso regression to identify target genes that could discriminate the highly expressed miRNAs from random sets of less abundant miRNAs. This unbiased computational analysis readily pointed to tumor suppressor genes including the E3 ligase Fbxw7, Pten and the apoptotic regulator Bim. An alternative strategy used ranked miRNA targets by the sum of their binding site scores for known T-ALL tumor suppressors. Consistently, the oncomirs were enriched at the top of the ranking, above miRNAs that were less abundantly expressed. The additional targets that ranked the oncomir group above other miRNAs included the tumor suppressor Nf1, Ikaros (Ikzf1), a determinant of T-cell differentiation and Phf6, a transcriptional regulator that is sometimes mutated in T-ALL.³⁰ We validated these targets experimentally and confirmed their tumor suppressive effect by shRNA-mediated knockdown in the same murine model. Some notable effects include the regulation of Bim and Pten by several members of the miR-17–92 cluster. MiR-223 has been considered a myeloidspecific miRNA; however, we observe differential upregulation of this miRNA in T-ALL. MiR-223 is a strong regulator of Fbxw7 and causes an increase of MCL1, an anti-apoptotic protein whose degradation is triggered by Fbxw7.^31–33

Together, our experimental and computational data indicate a network of regulatory interactions between oncomirs and tumor suppressor genes. These interactions were overlapping and partially redundant, such that multiple miRNAs regulate the same tumor suppressor. For example, Pten expression is repressed by miR-19, miR-20, miR-26 and miR-92 and Fbxw7 is regulated by miR-92 and miR-223.³¹ Conversely, most of the oncomirs affect several tumor suppressors, for example miR-20, affects the expression levels of Pten, Phf6 and also Bim. Based on these findings, we propose a network model wherein oncogenic miRNAs act as multi-target regulators that cooperatively suppress the cellular failsafe programs that protect against malignant transformation (Fig. 1).

Figure 1.

Oncomir-NET: individual and cooperative miRNA effects on key tumor suppressor genes in T-ALL. Our study identifies a cooperative network of overlapping and partially redundant miRNA—tumor suppressor gene interactions in T-cell leukemia.

What then is the relevant contribution of each miRNA in this network model of oncomir action? We used an antagomir approach to measure the effect of removing one or multiple miRNAs from leukemia cells. Single antagomirs produced significant effects on target gene expression and leukemia cell viability and proliferation. However, combinations of two or three antagomirs produced strikingly more powerful effects indicating that oncomirs produce individual and cooperative effects on tumor phenotypes.

Are miRNAs an alternative to the mutational inactivation of tumor suppressor genes? A mutually exclusive genetic relation, such that genetic loss of the gene corresponded to lower miRNA levels and vice versa, has been observed for miR-26 and Pten in glioma.³⁴ We analyzed miRNA expression data in T-ALL for similar occurrences by comparing miRNA expression with cytogenetic data and mutational status of key genes including Notch, Pten or Fbxw7. The pattern of miRNA expression in T-ALL was overall similar among cytogenetic subgroups and largely independent of mutational background. In particular, we found no correlation of the experimentally validated oncomirs with the mutational status of one specific target gene. We think this finding argues against a simple model of linear interactions between one miRNA and its target gene. Instead, we favor the concept that the pleiotropic activity of miRNAs is key to their biological activity.

What are the implications of this network pattern of oncomir action for therapy? Our analyses indicate a network of abundantly expressed oncomirs that produce a robust repression of tumor suppressive mechanisms and stabilize the malignant phenotype through additive and redundant activities. There is much interest in targeting oncomirs for cancer therapy and in some cases the problems of delivery have been overcome.³⁵ The targeted repression of an oncogenic miRNA is based on the notion that it acts as a “driver mutation.” This is in analogy to the activating mutations in classical oncogenes, and the therapeutic concept is best described as “shoot the driver.” Data from experimental models indeed suggest that the enforced expression of individual miRNAs can drive tumor phenotypes, and that these tumors continue to depend on that activity.^12,28,36 However, our data indicate that in human cancer cells oncomirs behave in a manner that is quite distinct from classical oncogenes with regard to their redundancy. As described above, multiple highly expressed miRNAs produce overlapping repressive effects on a set of target genes in leukemic cells. Accordingly, we cannot define the driver, and instead are faced with multiple potential drivers. Hence, the therapeutic blockade of one oncomir may be partially compensated by other highly expressed miRNAs that have overlapping target profiles. One way to overcome this redundancy would be to target multiple cooperating miRNAs. We speculate that in most instances no one particular oncomir may be absolutely required to maintain a cancer, although in some cases an individual driver may be identified. A detailed map of the oncomir—tumor suppressor gene interactions can identify non-redundant miRNA effects and help develop effective combinations.

What about miRNAs as molecular diagnostic and prognostic markers? MiRNAs are relatively stable and can be accurately measured from very small tissue samples, making them potentially ideal diagnostic and prognostic markers. By comparison, protein-based diagnostics by immunohistochemistry requires large tissue blocks, the interpretation is highly subjective and variable, and the scope is limited to a few well-characterized antibody assays. In our study, we did not directly examine the diagnostic and prognostic value of miRNAs in T-ALL. However, we provide a short list of the most abundantly expressed miRNAs that have demonstrated roles in that leukemia. Heterogeneous expression of these oncomirs between patients may contribute to clinical outcomes and should be explored in suitable sample collections. We are also beginning to decipher the control of signaling pathways by specific miRNAs. For example, miR-19 affects multiple regulators of PI3K/Akt signaling.²⁸ In T-ALL, activation of this pathway has been linked to resistance to chemotherapy and therapeutic gamma-secretase inhibitors.^37,38 Conversely, activation of this pathway may indicate sensitivity to targeted inhibitors of Akt or mTOR. MiRNAs need to be considered as potential markers in clinical studies so that we can begin to address these questions.

Are our findings on miRNA action in T-cell leukemia relevant to other cancers? T-cell leukemia is a comparatively rare disease, and we speculate that similar principles may apply to other cancers. Clearly, miRNAs can have diverse effects in different cellular and genetic contexts, and individual miRNAs may behave as an oncogene in tissue but not another. A striking example of the context dependence of miRNA function is miR-26, which can enhance gliomagenesis by affecting Pten levels.³⁴ On the other hand, exogenous delivery of miR-26 can suppress the development of liver cancers and may downregulate cell cycle genes.³⁵ We define a genetic network of oncomir action in T-ALL that provides some insight into the action of miRNAs in oncogenesis. Most likely, similar miRNA networks are active in other cancers, although they may involve other miRNAs and target genes. Perhaps our study can provide a blueprint for more comprehensive miRNA studies in other cancers.

Outlook

We provide a compelling example of the genetic interactions of oncogenic miRNAs with the cellular failsafe mechanisms that guard against malignant transformation. Important questions remain. First, what is the role of tumor suppressive miRNA? Several miRNAs loci are targets of recurrent genomic deletions and appear to behave as tumor suppressor genes.^15–17 Redundancy of target repression may be key to understand tumor suppressive miRNAs. For example, to affect oncogenesis loss of one miRNA must have unique effects on gene expression that are not compensated for by other miRNAs. Systematic knockout studies of individual miRNAs in different model organisms indicate that most individual miRNAs are quite dispensable in development and their loss produces minor or no phenotypes.^39,40 These studies may imply that only a few individual miRNAs can act as tumor suppressors. Second, we have not touched on the mechanisms controlling miRNA expression or activity in cancer. Recent studies have revealed coordinate transcriptional regulation of several miRNAs by c-Myc,⁴¹ and we can speculate about regulated decay or RNA editing as additional means of controlling miRNA activity. Finally, how will we translate these findings to benefit cancer patients? With some exceptions the efficient delivery of miRNA-based drugs may remain problematic. On the other hand the ease of measuring miRNAs in clinical samples indicates a very significant potential for diagnostic applications that is yet to be explored.