miRNAs in human cancer^†

miRNA profiling

The main methods currently used for miRNA profiling are sequencing, microarray and real-time RT–PCR-based approaches (reviewed in 15-17). The input material initially used for these studies comprised high-quality preserved fresh-frozen samples, but recently it has been possible to obtain reproducible and comparable profiles using formalin-fixed paraffin-embedded tissues (FFPE), making these archived tumour collections accessible for study 18-20.

Microarrays generally provide fold changes in miRNA expression between samples, with members of miRNA sequence families prone to cross-hybridization 21-25. More recently, calibration cocktails of synthetic miRNAs were used in array experiments to derive absolute abundance of miRNAs 23. RT–PCR methods are lower throughput and require normalization 26; if external miRNA standards are used for quantification, the most abundant miRNA, which may vary in length due to 3′ end heterogeneity, should be used as a calibration standard. Sequencing methods, besides their obvious potential to identify new miRNAs, editing and mutation events, estimate miRNA abundance based on the frequency of sequence reads (eg 5, 7, 8, 27-30). Given the dramatic increase in sequencing power, bar-coding samples can allow multiple specimens to be processed at the same time, reducing the cost and effort of profiling and paving the way for large specimen studies 30, 31. Ligation biases exist in sequencing methods, which may affect comparable representation of certain miRNAs obscuring absolute expression levels, but are irrelevant when monitoring fold changes between samples. A study with a synthetic pool of 776 miRNA sequences showed that overall these biases do not prevent identification of miRNAs and allow estimation of these biases for the small subset of sequences affected (Hafner, unpublished data). Finally, choosing the appropriate statistical analysis to evaluate the data depends on the methodology used to obtain the profiles, ranging from established Significance Analysis of Microarrays (SAM) 32, to newly developed techniques for sequencing data 30, 33, 34. Recent in situ hybridization (ISH) advances allowed sensitive detection of miRNAs in heterogeneous tissues, defining miRNA cellular localization 35-37. The potential of miRNA localization to suggest function for a subpopulation of cells was demonstrated early on, as in the case of lsy-6, expressed in < 10 neurons in Caenorhabditis elegans, controlling left/right asymmetry 38.

miRNA databases and validation

It is critical to know which miRNAs are validated and have the potential to regulate cellular functions, especially given the frequent revisions of the miRNA database, miRBase (www.mirbase.org) 39 and the dramatic increase in the number of novel and re-annotated miRNAs through the use of deep-sequencing technologies. It is extremely challenging to establish the validity of novel miRNAs, particularly when their definition is based on a handful of sequence reads. The latest release of miRBase (version 16) includes 121 novel human miRNA precursors, 13 miRNA mature and precursor name changes, four miRNA precursor sequence revisions and the removal of 13 miRNA precursors. A recent study of 60 million small RNA sequence reads, generated from a variety of adult and embryonic mouse tissues, confirmed 398 annotated miRNA genes and identified 108 novel miRNA genes but was unable to find sequencing evidence for 150 previously annotated mouse miRNAs. Ectopic expression of the confirmed and newly identified miRNA hairpin sequences yielded small RNAs with the classical miRNA features but failed to support other previously annotated sequences (of the 17 tested miRNAs with no read evidence, only one yielded a single sequence read, while of 28 tested miRNAs with insufficient number of reads, only four were verified) 40. Deep sequencing of large human tissue sample collections will allow us to assess the validity of the human miRBase entries in a similar fashion.

General principles of miRNA genomic organization

miRNAs are frequently expressed as polycistronic transcripts. To date, 1048 human miRNA precursor sequences have been deposited in miRBase 39. Approximately one-third (390) of these miRNAs are located in 113 clusters, each measuring ⩽ 51 kb in the human genome (51 kb being the longest distance between miRNAs belonging to the same cluster; Figure 2). These miRNA clusters are co-expressed based on evidence from miRNA profiling data from a variety of tissues and cell lines 22, 29, 30, 40. The genomic organization of representative oncogenic (mir-17-92 and mir-21) and tumour suppressor (let-7/mir-98 and mir-141/200) families is illustrated in Figure 2. Presentation of miRNA profiles in the form of expression clusters provides a readily interpretable summary of expression data and stresses the importance of cistronic expression regulation; dysregulation of one member of the cluster should be accompanied by similar dysregulation of other cluster members. Since miRNA genes are frequently multi-copy, determining the relative contribution of each genomic location to mature miRNA expression is challenging.

Alterations in genomic miRNA copy numbers and location

Changes in miRNA expression between normal and tumour specimens are often attributed to the location of miRNAs in regions of chromosomal instability (amplification, translocation or deletion) or nearby chromosomal breakpoints, initially locating 52.5% of miRNA genes in cancer-associated regions or fragile sites 12. The miRNA cluster 15a/16-1 is located in a frequently deleted genomic locus containing a putative tumour suppressor-containing region in chronic B cell lymphocytic leukaemia (B-CLL) 41. Other examples include deletion of let-7g/mir-135-1 in a variety of human malignancies 12, amplification of the mir-17-92 cluster in lymphoma 42, translocation of mir-17-92 in T cell acute lymphoblastic leukaemia (T-ALL) 43 and amplification of mir-26a in glioblastoma 44.

Alterations in miRNA transcriptional regulation

Some autonomously expressed miRNA genes have promoter regions that allow miRNAs to be highly expressed in a cell type-specific manner, and can even drive high levels of oncogenes in cases of chromosomal translocation. The mir-142 gene, a marker of haematopoietic cells, is located on chromosome 17 and was found at the breakpoint junction of a t(8;17) translocation, which causes an aggressive B cell leukaemia due to strong up-regulation of a translocated MYC gene 45. The translocated MYC gene, which was also truncated at the first exon, was located only four nucleotides from the 3′ end of the mir-142 precursor, placing the translocated MYC under the control of the upstream mir-142 promoter. In an animal model for hepatocellular carcinoma (HCC), a similar event placed MYC downstream of the mir-122a promoter active only in hepatocytes 46.

Many transcription factors regulate miRNA expression in a tissue-specific and disease state-specific fashion and some miRNAs are regulated by well-established tumour suppressor or oncogene pathways, such as TP53, MYC and RAS (reviewed in 47). The miRNA and its transcriptional regulators can participate in complex feedback regulation loops. Examples include the TP53-regulated miR-34a 48, 49, the RAS-regulated miR-21 29, 50, 51 and the MYC-regulated miR-17-92 cluster 52, 53.

miRNA dysregulation has also been linked to changes in epigenetic regulation, such as the methylation status of miRNA genes, which results in alterations in their expression levels 54, 55. Examples of methylated miRNAs include mir-127 in bladder cancer cells 56 and mir-9-1 in breast cancer 57.

miRNA biogenesis pathway in tumorigenesis

miRNA biogenesis has been reviewed extensively 47, 58-64 (Figure 1). miRNA pathway components could either be mis-expressed in tumours or mutated (reviewed in 65, 66). Post-transcriptional regulation of miRNAs themselves through RNA editing or terminal modifications was shown to alter miRNA targeting, processing and stability, but connection of these modifications to tumorigenesis has not yet been definitive (reviewed in 47, 66, 67).

miRNA function/mechanism

As described above, miRNAs function through the Ago proteins, containing both RNA-binding domains and RNase H domains (reviewed in 91). The four human Ago genes are co-expressed and bind to miRNAs irrespective of their sequence. Ago2, in contrast to the other members, retains an active RNase H domain, able to directly cleave target RNAs with extensive complementarity to the bound miRNAs. The assembly of the miRNP complex involves multiple Ago conformational transitions captured in a series of crystal structures (reviewed in 92). The mRNA target is recognized by pairing of the miRNA seed region (positions 2–8) to complementary sequences located mainly in the target 3′ UTR but also in the coding regions. Target mRNA recognition and regulation involves members of the GW182/TNRC6 family. Moreover, Ago and TNRC6 proteins co-localize in mRNA-processing bodies (P bodies) that are important for mRNA storage and RNA turnover. Proteomic approaches identified additional Ago-interacting proteins, some of which likely represent mRNA-interacting partners that co-purified with miRNA-targeted mRNPs; their function in RNA silencing processes and potentially tumorigenesis remains to be established.

In mammalian cells miRNAs destabilize targeted transcripts, without significantly perturbing translation of mRNAs, through various mechanisms, such as de-capping and de-adenylation, in addition to P-body localization 93, 94. mRNA profiling and proteomic studies, following the introduction of miRNAs, inhibition of miRNAs using antagomirs or miRNA knockout in mouse models, have shown that miRNAs destabilize mRNAs through binding sites located mainly in their 3′ UTRs, without a measurable effect on translation 95-103. Ribosome profiling studies demonstrated that the ribosome density of miRNA targets was unaltered, while changes in miRNA levels were inversely correlated to mRNA and protein abundance, emphasizing the role of miRNAs in regulation of mRNA stability but not translation 104.

Organization of miRNAs into sequence families

Certain miRNAs share sequence similarity in regions that are critical for mRNA target recognition, specifically the seed region, and are best viewed as a family when considering mRNA target regulation and the functional consequences of altered miRNA expression. miRNAs can be grouped in sequence families, based not only on their seed sequence similarity but also on overall sequence similarity, given that the miRNA 3′ end also contributes to miRNA targeting, although to a lesser extent (reviewed in 59) (Figure 2). Changes in the overall abundance of miRNA sequence families relate directly to target regulation. In a MYC-driven B cell lymphoma mouse model, a conditional knockout of the oncogenic mir-17-92 cluster induces apoptosis, which can be reduced by reintroduction of only one of the four sequence families produced from the cluster 105.

miRNA–mRNA stoichiometry

The majority of miRNA profiling studies do not provide an estimate of miRNA abundance, which is critical in our understanding of the role of miRNA–mRNA-mediated regulation in tumorigenesis. Only the most abundantly expressed miRNAs occupy a substantial fraction of their available mRNA target sites and affect target mRNA stability 100. Abundant miRNAs that behave as ‘switches’, turning on or off during the tumorigenesis process, as shown in developmental processes, have the most significant regulatory potential, given that miRNAs usually lead to modest 1.5–4-fold regulation of their target expression 95, 97, 102. However, miRNAs that have lower expression can, in combination, fine-tune target expression, since a given mRNA may be subject to regulation by multiple miRNAs of unrelated families 106-108. To conclude, the interplay between miRNAs expressed in particular tissues, the levels of their respective expressed targets, as well as other post-transcriptional gene regulatory mechanisms (such as regulation by RBPs or other competing interactions—see below) is likely responsible for balancing miRNA conferred regulation.

Changes in the miRNA targets

The binding sites of miRNAs in mRNAs can be altered through a variety of mechanisms, such as point mutations, translocations, shortening of the 3′ UTR, competition with other RBPs or decoy molecules for mRNA binding. Point mutations in miRNA targets can both create or destroy a miRNA binding site 109-111. Chromosomal translocations can remove miRNA binding sites from their regulated oncogenes, such as in the case of let-7 targeting of the 3′ UTR of the Hmga2 gene 112. Shortening of the 3′ UTR through alternative polyadenylation can relax miRNA-mediated regulation of known oncogenes, such as IGF2BP1/IMP1, and lead to oncogenic transformation 113, as does use of decoy pseudogenes, as in the case of PTEN, by saturating miRNA binding sites 114. Finally, cooperativity or competition of miRNAs for mRNA target site binding with other RBPs, such as ELAVL1 (HuR), DND1 and PUM1, can also de-repress target expression 115-118.

miRNA cancer database

The development of miRNA microarrays, RT–PCR platforms and deep sequencing methodologies have resulted in an exponential acquisition of miRNA profiles. Some of the published miRNA profiles are available in the NCBI Gene Expression Omnibus, similarly to mRNA profiles (other resources include: www.microrna.org; http://www.mirz.unibas.ch). However, there is no database or viewer that allows for cross-platform comparison of existing data.

Tissue heterogeneity

Tissues are generally composed of multiple cell types, each with its distinct gene expression programme. Disease not only alters the expression programme of the affected cell type but often also its cell type composition. To best separate these effects in the profiling of heterogeneous tumour samples, it may be useful to profile tumour cell lines and individual cell types that may be present in a tumour sample, or define miRNA cellular localization by performing ISH. Figure 3 compares miRNA abundance profiles of normal breast, an oestrogen receptor-positive invasive ductal breast carcinoma, the oestrogen receptor-positive ductal cell line MCF7, human fat and blood (Farazi and Horlings, unpublished data). Strikingly, we can model the profile of a human cancer by simply combining tumour cell line and human fat profiles at equal ratio. This exercise demonstrates that the MCF7 tumour cell line may be a good disease model for deciphering miRNA regulatory networks, expressing many of the miRNAs present in the predominant tumour-derived cell type and highlighting the need for individual cell type miRNA profiles.

miRNAs as tumour suppressors and oncogenes

miRNA dysregulation could be used as a diagnostic tool, even if the particular miRNAs do not serve any regulatory function. Alternatively, miRNA dysregulation could drive tumorigenesis, through the roles miRNAs can adopt as tumour suppressors or oncogenes. miRNAs that are up- or down-regulated in malignancies are respectively referred to as oncogenic or tumour-suppressor miRNAs, sometimes even if there is no evidence for their causative role in tumorigenesis. Some of the most commonly dysregulated miRNAs are summarized in Table 1 (reviewed in 11).

Functional studies performed in cancer cell lines or mouse models of various malignancies through over-expression or knockdown of miRNAs have supported a role for some of these miRNAs in tumorigenesis. Over-expression of tumour suppressor miRNAs, such as let-7g, reduced tumour burden in a K-RAS murine lung cancer model 156. Over-expression of the oncogenic mir-17-92 cluster led to a lymphoproliferative disorder, and higher level expression of the cluster in MYC-driven B cell lymphomas dramatically increased tumorigenicity 53, 157. Over-expression of another oncogene, mir-21, frequently highly expressed in solid and haematological malignancies, resulted in a pre-B malignant lymphoid-like phenotype, whereas subsequent mir-21 inactivation in the same model led to apoptosis and tumour regression 158. Transgenic mouse models with loss and gain of function of mir-21 combined with a model of lung cancer confirmed the role of mir-21 as an enhancer of tumorigenesis when over-expressed, or a partial protector when genetically deleted 50. Ectopic expression of mir-155 in bone marrow induced polyclonal pre-B cell proliferation progressing to B cell leukaemia or myeloproliferation in mice 159, 160.

Metastasis-related miRNAs have been identified in various malignancies, mainly from cell line and xenograft experiments (reviewed in 161). Examples include breast cancer-related miR-10b, miR-9, miR-31 and miR-335, among others. The interesting regulatory roles of these miRNAs cannot easily be validated in large clinical studies. Two clinical studies with long-term follow-up data instead identified miR-210 associated with tumour aggressiveness 162, 163, pointing to difficulties reconciling cell line, xenograft model and patient materials, due to the tissue heterogeneity discussed earlier, the heterogeneous nature of the malignancy and timing of clinical specimen acquisition. Tumour miRNA profiles cannot dissect contributions from sub-populations of cells that may be important for tumour characteristics, such as metastasis, while cell line miRNA profiles cannot capture the cellular interactions from supporting cell types in the tumour microenvironment. Patient samples are often collected at time of diagnosis, by which time a tumour is already well established and cannot unravel early changes that may be critical in tumour initiation or later changes important in metastasis.

miRNA-regulated pathways

The observed effects of miRNA mis-expression on tumour initiation, maintenance or metastasis can be explained by the mRNA targets and pathways they regulate, which include known tumour suppressors and oncogenes (reviewed in 11). miRNAs regulate a large number of genes, some estimates reporting miRNA regulation of up to 60% of the human genome, making it challenging to attribute a phenotype after mis-expression of a particular miRNA through its action on only a subset of targets 94, 164. If a few of these targets control rate-limiting steps in the studied tumorigenesis processes within the specified tissues and cell types, such as metastasis, then miRNA regulation of a handful of targets could potentially explain the phenotype resulting from miRNA mis-expression 165.

Examples of miRNA-regulated cancer pathways include differentiation, apoptosis, proliferation and stem cell maintenance, a process important for disease relapse and/or metastasis. The skeletal muscle-specific miR-206 blocked human rhabdomyosarcoma growth in mouse xenograft models by inducing myogenic differentiation 150, while the miR-200/141 family was critical in the epithelial to mesenchymal transition (EMT) in various malignancies (reviewed in 166). Sustained expression of the endogenous mir-17-92 cluster was required to suppress apoptosis in MYC-driven B cell lymphomas in a conditional knockout allele of the mir-17-92 cluster 105. TP53-regulated ectopically expressed miR-34 induced cell cycle arrest in both primary and tumour-derived cell lines, down-regulating genes promoting cell cycle progression (reviewed in 49). In a final example of miRNA-regulated cancer pathways, isolation of a subset of highly tumorigenic breast cancer cells that were thought to have stemness properties showed that these cells do not express let-7 family members and that expression of let-7 or its known target RAS leads to loss of self renewal 167.

Alterations of miRNA sequence

miRNA dysregulation could be a result of mutations in miRNA genes in well-conserved regions in their mature sequence affecting mRNA targeting, or the remainder of the miRNA precursor potentially affecting processing and stability of the mature miRNA (reviewed in 66). For example, a mutation in the seed region of mir-96 was shown to lead to hearing loss in a mouse model 168 and was identified in families with non-syndromic progressive sensorineural hearing loss 169, while a point mutation in the viral mir-K5 precursor stem loop was shown to interfere with its processing and reduce mature miR-K5 accumulation 170. If miRNAs are drivers of oncogenic and tumour suppressor pathways, we would expect to find miRNA mutations that can also be causative of the disease. So far the only mutation identified in a miRNA that could lead to malignancy is miR-16, where a germline mutation potentially affects miR-16 biogenesis and abundance in a kindred with familial CLL 171 and New Zealand black mice that naturally develop CLL-like disease 172. Single nucleotide polymorphisms (SNPs) located in both precursor and mature miRNA sequences have been examined in the context of disease risk for various malignancies but have not been validated as causative (reviewed in 66).

miRNAs as therapeutics

Because miRNAs affect the expression of multiple genes and thereby tune multiple points in disease pathways, miRNAs and their regulated genes, represent interesting drug targets. Antisense oligonucleotide targeting experiments in mice 99, 177, 178 and non-human primates 179 have demonstrated the feasibility of manipulating miRNA levels. miR-143 was initially shown to promote adipocyte differentiation and could be a target for therapies in obesity and metabolic diseases 180. Alternatively, ‘miRNA sponges’ have been exploited to reduce miRNA expression in mammalian cells and mouse models by using RNA transcripts expressed from strong promoters containing miRNA-complementary binding sites (reviewed in 181). Systemic administration of antisense oligonucleotide therapeutics to miR-122, a liver-enriched miRNA, in mice and primates was shown to alter lipid metabolism and hepatitis C viral load, resulting in reduced liver damage 99, 177-179, 182, 183. At the same time, systemic delivery of a miRNA mimic for miR-26a in a murine model of HCC reduced tumour size 131. The new and exciting advances in delivery of miRNA inhibitors and mimics hold the promise of quickly translating our knowledge of miRNAs into treating disease.