MicroRNA biogenesis pathways in cancer

Abstract

MicroRNAs (miRNAs) are critical regulators of gene expression. Amplification and overexpression of individual ‘oncomiRs’ or genetic loss of tumour suppressor miRNAs are associated with human cancer and are sufficient to drive tumorigenesis in mouse models. Furthermore, global miRNA depletion caused by genetic and epigenetic alterations in components of the miRNA biogenesis machinery is oncogenic. This, together with the recent identification of novel miRNA regulatory factors and pathways, highlights the importance of miRNA dysregulation in cancer.

MicroRNAs (miRNAs) repress gene expression by binding to complementary sequences in the 3′ untranslated region (3′ UTR) of mRNAs to target them for degradation and thereby prevent their translation¹. Considering that more than 1,000 individual miRNA genes have been identified, that an individual miRNA can target hundreds or thousands of different mRNAs, and that an individual mRNA can be coordinately suppressed by multiple different miRNAs, the miRNA biogenesis pathway therefore has an important role in gene regulatory networks. Over the past decade, it has emerged that miRNAs have crucial roles in cancer. Propelled by the original publication that described the deletion of the miR-15 and miR-16 loci in the majority of samples from patients with B cell chronic lymphocytic leukaemia (B-CLL), a plethora of subsequent publications described altered miRNA expression in diverse types of cancer^2,3. Functionally, it has been shown through both loss-of-function and gain-of-function experiments in human cancer cells, mouse xenografts, transgenic mouse models and knockout mouse models that miRNAs have key roles in cancer initiation, progression and metastasis^4,5. The first example was provided by enforced expression of the miR-17~92 cluster, the so-called oncomiR-1, that acted with MYC to accelerate tumour development in a mouse model of B cell lymphoma⁶. Certain other miRNAs can function as tumour suppressors: for example, the let-7 family of miRNAs targets important oncogenes such as MYC, RAS family members (HRAS, KRAS and NRAS) and high-mobility group AT-hook 2 (HMGA2) to suppress tumour growth^7–9. Therefore, cancer-associated changes in miRNA expression patterns are emerging as promising diagnostic markers that often correlate with disease progression and patient survival. This pathway might also represent a new therapeutic target for multiple types of cancer². Mechanistically, miRNAs can control cell proliferation, differentiation, survival, metabolism, genome stability, inflammation, invasion and angiogenesis to affect tumour development.

Although individual miRNAs can have either oncogenic or tumour-suppressive function, several studies have shown that miRNA expression is globally suppressed in tumour cells compared with normal tissue, suggesting that miRNA biogenesis might be impaired in cancer^10,11. Indeed, the expression levels of miRNAprocessing machinery components such as the ribonuclease III (RNase III) DROSHA and DICER1 are decreased in some cancers, such as lung cancer, ovarian cancer and neuroblastoma^12–14. Additionally, low DROSHA or DICER1 expression levels are associated with advanced tumour stage and poor clinical outcome in patients with neuroblastoma and patients with ovarian cancer^13,14. Support that this global suppression can have a causative role in cancer was initially provided by the demonstration that genetic deficiency of components of the miRNA biogenesis pathway can accelerate tumour growth in a mouse model of lung cancer¹⁵. Although this work provided proof-of-concept that the miRNA biogenesis pathway can have an important role in cancer progression, it is the recently reported mutations in and dysregulation of miRNA biogenesis pathway components that highlight the pathophysiological relevance of the miRNA biogenesis machinery in human tumours^16–24. Moreover, the recent discovery of certain molecular and cellular mechanisms that control miRNA biogenesis provided compelling evidence that disruption of this pathway is crucially important for a wide variety of paediatric and adult cancers.

In this Review, we discuss what is known about dysregulation of the miRNA biogenesis pathway in cancer, summarize the growing evidence that germline mutations and somatic mutations in core components of the miRNA biogenesis machinery promote oncogenesis, and provide specific examples of how certain RNA-binding proteins and cell signalling pathways contribute to cancer through their control of miRNA expression. With these examples, we aim to highlight emerging themes and the relevance of the miRNA biogenesis pathway in cancer.

miRNAs and their biogenesis

miRNAs are a group of short non-coding RNAs that mediate post-transcriptional gene silencing. The first miRNA was reported in Caenorhabditis elegans in 1993 (REF. ²⁵); however, the general regulatory function of miRNAs was not well appreciated until 2001 (REFS 26–28). Since then, thousands of miRNAs have been identified in various species²⁹. Binding of the ~22-nucleo tide miRNA to target mRNA mediates mRNA degradation and blocks translation³⁰. The majority of miRNA genes are transcribed by RNA polymerase II (Pol II) in the nucleus, and the primary miRNAs (pri-miRNAs) are capped, spliced and polyadenylated³¹. Approximately 30% of miRNAs are processed from introns of proteincoding genes, whereas most other miRNAs are expressed from dedicated miRNA gene loci. An individual primiRNA can either produce a single miRNA or contain clusters of two or more miRNAs that are processed from a common primary transcript. Nonetheless, these long pri-miRNAs are cleaved by Microprocessor, which comprises the double-stranded RNase III enzyme DROSHA and its essential cofactor, the double-stranded RNA (dsRNA)-binding protein DiGeorge syndrome critical region 8 (DGCR8)^32,33. DROSHA contains two RNase III domains, each of which cleaves one strand of the dsRNA towards the base of stem–loop secondary structures contained within pri-miRNAs to liberate ~60–70-nucleotide hairpin-shaped precursor miRNAs (pre-miRNAs)^32–35. Microprocessor recognizes the single-stranded RNA (ssRNA)–stem junction as well as the distance from the terminal loop region. It specifically cleaves the dsRNA ~11 bp from the junction with the flanking ssRNA to produce hairpin-shaped pre-miRNAs with an overhang at the 3′ end of either 2 nucleotides (group I miRNAs) or 1 nucleotide (group II miRNAs)^36–39. Although the core components, DROSHA and DGCR8, are required for the biogenesis of almost all miRNAs in the cell, and Microprocessor activity can be reconstituted in vitro with recombinant DROSHA and DGCR8 proteins^32,35, numerous accessory factors are known to have a role in pri-miRNA processing in cells (discussed in more detail below). The pre-miRNAs are then exported from the nucleus to the cytoplasm by exportin 5 (XPO5)^40–42 and further processed by DICER1, an RNase III enzyme that measures from the 5′ and 3′ ends of the pre-miRNA⁴³. DICER1 binding to the end of the pre-miRNA positions its two catalytic RNase III domains so that asymmetrical cleavage of the dsRNA stem, close to the terminal loop sequence, produces the mature ~22-nucleotide miRNA duplex with 2-nucleotide 3′ overhangs⁴⁴. DICER1 associates with transactivation-responsive RNA-binding protein (TRBP; also known as TARBP2), which binds to dsRNA⁴⁵. Although it is not required for pre-miRNA processing by DICER1, TRBP enhances the fidelity of DICER1-mediated cleavage of a subset of pre-miRNAs in a structure-dependent manner and alters miRNA guide-strand selection by triggering the formation of isomiRNAs, which are 1 nucleotide longer than the regular miRNAs^46,47. TRBP also physically bridges DICER1 with the Argonaute proteins (AGO1, AGO2, AGO3 or AGO4) to participate in the assembly of the miRNAinduced silencing complex (miRISC)⁴⁵. One strand of the mature miRNA (the guide strand) is bound by an Argonaute protein and retained in the miRISC to guide the complex, together with members of the GW182 family of proteins, to complementary target mRNAs for post-transcriptional gene silencing. This occurs in processing bodies (P-bodies), which are the cytoplasmic foci that are induced by mRNA silencing and decay but are not necessarily required for miRNA-mediated gene silencing^48–50 (FIG. 1).

Figure 1. Overview of miRNA biogenesis pathway.

MicroRNA (miRNA) genes are transcribed as primary miRNAs (pri-miRNAs) by RNA polymerase II (Pol II) in the nucleus. The long pri-miRNAs are cleaved by Microprocessor, which includes DROSHA and DiGeorge syndrome critical region 8 (DGCR8), to produce the 60–70-nucleotide precursor miRNAs (pre-miRNAs). The pre-miRNAs are then exported from the nucleus to the cytoplasm by exportin 5 (XPO5) and further processed by DICER1, a ribonuclease III (RIII) enzyme that produces the mature miRNAs. One strand of the mature miRNA (the guide strand) is loaded into the miRNA-induced silencing complex (miRISC), which contains DICER1 and Argonaute (AGO) proteins, directs the miRISC to target mRNAs by sequence complementary binding and mediates gene suppression by targeted mRNA degradation and translational repression in processing bodies (P-bodies). TRBP, transactivation-responsive RNA-binding protein.

Pri-miRNA transcription in cancer

miRNA biogenesis initiates with the transcription of the pri-miRNA, and this step is dysregulated in multiple human cancers. A considerable number of human miRNA genes are located at fragile sites or in genomic regions that are deleted, amplified or translocated in cancer⁵¹. These genomic variations alter pri-miRNA transcription and miRNA expression, which leads to the aberrant expression of downstream target mRNAs that can promote cancer initiation and progression^51,52. For example, the locus including miR-15 and miR-16 on chromosome 13q14 is frequently deleted in B-CLL, resulting in the loss or reduced expression of these two miRNAs in ~70% of B-CLLs³. miR-15 and miR-16 normally control apoptosis by targeting BCL-2 mRNAs⁵³. In another example, a point mutation in the miR-128b (also known as miR-128-2) gene blocks the processing of pri-miR-128b and reduces the levels of mature miR-128b, thus leading to glucocorticoid resistance in acute lymphoblastic leukaemia (ALL) cells with the mixed-lineage leukaemia (MLL)–AF4 (also known as KMT2A–AFF1) translocation⁵⁴.

In addition to genomic alterations, dysregulated miRNA expression can arise from alterations in tumour suppressor or oncogenic factors that function as transcriptional activators or repressors to control pri-miRNA transcription. For example, expression of the miR-34 family of miRNAs is driven by p53 and reflects the status of p53 in human cancers^55–59. The miR-34a, miR-34b and miR-34c miRNAs repress growth-promoting genes and coordinate with other members of the p53 tumour-suppressive network to inhibit uncontrolled cell proliferation and to promote apoptosis^55–59. In addition, the proto-oncoprotein MYC activates expression of oncogenic miRNAs, including the miR-17~92 cluster, in cancer^60,61. These MYC-target miRNAs promote cancer progression by controlling the expression of E2F1, thrombospondin 1 (THBS1), connective tissue growth factor (CTGF) and other target mRNAs to regulate cell cycle progression and angiogenesis^60,61. MYC can also contribute to the widespread repression of tumour-suppressive miRNAs in B cell lymphoma⁶². Expression of the miR-200 family (miR-200a, miR-200b and miR-200c) is frequently suppressed in human tumours. These miRNAs are known to directly target the mRNAs encoding the zinc-finger E-box-binding homeobox (ZEB) transcription factors, ZEB1 and ZEB2, which suppress the expression of epithelial genes to promote the epithelial–mesenchymal transition (EMT)⁶³. Interestingly, ZEB1 and ZEB2 directly bind to a regulatory element at the miR-200 promoter to repress transcription of miR-200 as part of a negative regulatory feedback loop that promotes EMT⁶⁴. Many other cancer-associated transcription factors also aberrantly regulate miRNA transcription in cancer. Therefore, transcriptional dysregulation — through either genetic loss of miRNA genes or aberrant transcription factor activity — is an important mechanism for altered miRNA expression in cancer.

Epigenetic modification of histone proteins and DNA controls local chromatin structure and has an important role in the regulation of both coding and non-coding gene expression. Indeed, epigenetic alteration is a common feature of cancer pathogenesis that drives the dysregulation of miRNA expression. The CpG islands at the gene promoters of tumour-suppressive miRNAs are frequently hypermethylated in cancer, thereby leading to the epigenetic silencing of these miRNAs. Treatment of cancer cells with DNA-demethylating agents can reactivate the expression of tumour-suppressive miRNAs, such as miR-148a, miR-34b, miR-34c and miR-9, that inhibit tumour growth and metastasis⁶⁵. In addition to DNA methylation, histone modifications have important roles in chromatin remodelling and cooperate with DNA methylation to suppress miRNA expression in cancer⁶⁶. Overall, epigenetic silencing is an important mechanism underlying miRNA repression in cancer.

Defective Microprocessor in cancer

The nascent pri-miRNA generated by Pol II forms a typical secondary structure consisting of a stem–loop hairpin flanked by ssRNA that is a substrate for cleavage by Microprocessor to generate pre-miRNA intermediates. A negative feedback mechanism involving the Microprocessor-mediated cleavage and destabilization of DGCR8 mRNA operates to help to control the relative DGCR8 expression level and to maintain the homeostatic control of miRNA biogenesis in cells^67–69. The expression and function of the Microprocessor components are often dysregulated in cancer. For example, copy-number gain or overexpression of DROSHA occurs in more than 50% of advanced cervical squamous cell carcinomas⁷⁰. In addition, DROSHA expression levels are upregulated in multiple types of cancer (TABLE 1). The increased expression of DROSHA alters the global miRNA expression profile and promotes cell proliferation, migration and invasion, which contributes to cancer progression^70,71. Conversely, DROSHA expression levels have been shown to be downregulated in many other types of cancer. DROSHA downregulation results in decreased miRNA expression¹³ and is correlated with metastasis, invasion⁷² and poor patient survival^13,14,73,74 (TABLE 1). Knockdown of DROSHA in lung adenocarcinoma cells results in increased proliferation and tumour growth in vitro and in vivo¹⁵, suggesting that DROSHA can function as a tumour suppressor to inhibit cancer progression in some contexts. Why DROSHA is upregulated in certain types of cancer but downregulated in others is not well understood, but one possibility is that different cancers have different genetic or epigenetic mechanisms controlling DROSHA expression, thus resulting in the abnormal expression of oncogenic or tumour-suppressive miRNAs in a given cancer type.

Table 1.

Dysregulation of miRNA biogenesis machinery in cancers

Protein	Dysregulation	Cancer type	Clinical correlation	Refs
DROSHA	Upregulation	Cervical SCC	Altered miRNA profile; associated with neoplastic progression	70,162
		Oesophageal cancer	Regulates cell proliferation; associated with poor patient survival	71
		BCC	Not determined	163
		SCC	Not determined	163
		Triple-negative breast cancer	No clinical correlation	164,165
		Smooth muscle tumours	Associated with tumour progression	166
		Gastric cancer	Associated with pathological characteristics and patient survival	167
		Serous ovarian carcinoma	Associated with advanced tumour stages	168
		Non-small cell lung cancer	Associated with poor prognosis	169
	Downregulation	Bladder cancer	Altered miRNA profile	170
		Ovarian cancer	Associated with poor patient survival	14
		Endometrial cancer	Correlated with histological grade	171
		Nasopharyngeal carcinoma	Correlated with shorter patient survival	73
		Breast cancer	Not determined	172
		Gallbladder adenocarcinoma	Correlated with metastasis, invasion and poor prognosis	72
		Neuroblastoma	Correlated with global downregulation of miRNAs and poor outcome	13
		Cutaneous melanoma	Associated with cancer progression and poor survival	74
DGCR8	Upregulation	Oesophageal cancer	Associated with poor patient survival	71
		Bladder cancer	Altered miRNA profile	170
		SCC and BCC	Not determined	173
		Prostate cancer	Associated with dysregulated miRNA	174
		Colorectal carcinoma	Not associated with any clinical parameters	175
		Ovarian cancer	Required for cell proliferation, migration and invasion	176
DICER1	Upregulation	Smooth muscle tumours	Associated with high-grade disease and tumour progression	166
		Gastric cancer	Correlated with gastric tumour subtype	167
		Serous ovarian carcinoma	Associated with advanced tumour stages	168
		Prostate cancer	Dysregulated miRNA expression; correlated with tumour stage	174,177
		Oral cancer	Required for proliferation	178
		Colorectal cancer	Correlated with tumour stage and associated with poor survival	179–181
		Precursor lesions of lung adenocarcinoma	Associated with histological subtypes and stages	182
		Cutaneous melanoma	Correlated with clinical stage	183
	Downregulation	Triple-negative breast cancer	No clinical correlation	165,184
		Bladder cancer	Altered miRNA profile	170,185
		BCC	Not determined	163
		Ovarian cancer	Associated with advanced tumour stage and poor patient survival	14,186, 187
		Endometrial cancer	No association with histological grade detected	171
		Nasopharyngeal carcinoma	Correlated with shorter patient survival	73
		Neuroblastoma	Associated with global downregulation of miRNAs and poor outcome	13
		Breast cancer	Associated with cancer progression and recurrence	172,188
		Gallbladder adenocarcinoma	Correlated with metastasis, invasion and poor prognosis	72
		Non-small cell lung cancer	Low levels of DICER1 expression correlate with shortened survival	12,169
		Hepatocellular carcinoma	Not associated with clinical characteristics	189
		Chronic lymphocytic leukaemia	Associated with progression and prognosis	190
		Colorectal cancer	Associated with tumour stage and shorter survival	191
PACT	Upregulation	AK, SCC and BCC	Not determined	173
XPO5	Downregulation	Bladder cancer	Associated with altered miRNA profile	170
AGO1	Upregulation	AK, SCC and BCC	Not determined	173
		Serous ovarian carcinoma	Associated with advanced tumour stages	168
AGO2	Upregulation	AK, SCC and BCC	Not determined	173
		Serous ovarian carcinoma	Correlated with advanced tumour stages and associated with shorter survival	168

AGO, Argonaute; AK, actinic keratoses; BCC, basal cell carcinoma; DGCR8, DiGeorge syndrome critical region 8; miRNA, microRNA; PACT, interferon-inducible double-stranded RNA-dependent protein kinase activator A; SCC, squamous cell carcinoma; XPO5, exportin 5.

Mutational analysis revealed that DROSHA is frequently mutated in Wilms tumour samples^21–24 (FIG. 2; see Supplementary information S1 (table)). More than 70% of the DROSHA mutations occur at E1147, a metalbinding residue in the RNase IIIb domain. The recurrent somatic missense mutation E1147K interferes with metal binding and therefore affects the function of DROSHA in the processing of pri-miRNAs through a dominantnegative mechanism^21–24. As a result, mature miRNAs are globally downregulated in DROSHA-mutated Wilms tumours^21–24. Several missense mutations and a splicesite mutation of the DROSHA gene have been found in ovarian cancer; however, these mutations do not affect DROSHA expression levels. Therefore, it remains to be characterized whether the functions of DROSHA are affected by these mutations¹⁴. In addition, DROSHA was found to be alternatively spliced in melanoma and teratocarcinoma cells⁷⁵. The splice variants encode carboxy-terminal-truncated DROSHA proteins that partially lack the RNase IIIb domain and the dsRNAbinding domain (dsRBD). These truncated proteins fail to interact with DGCR8 and are deficient in pri-miRNA processing in vitro. However, the splice variants have little effect on mature miRNA expression, which might be due to the relatively low expression level of the splice variants in the cells⁷⁵.

Figure 2. Mutation of the miRNA biogenesis pathway in cancer.

Mutations of the microRNA (miRNA) biogenesis pathway genes identified in cancer are summarized and represented by their relative locations in the protein and the type of mutation. The detailed mutational information (mutation locations, mutation types and tumour types) is provided in Supplementary information S1 (table). ATF, armadillo-type fold; DGCR8, DiGeorge syndrome critical region 8; Dimer, dimerization domain; dsRBD, double-stranded RNA-binding domain; IBN_N, importin-β amino-terminal domain; NLS, nuclear localization signal; PAZ, Piwi–Argonaute–Zwille domain; RNase, ribonuclease; pre-miRNA, precursor miRNA; TRBP, transactivation-responsive RNA-binding protein; WW, WW domain (also known as WWP-repeating motif); XPO1, exportin 1/importin β-like domain; XPO5, exportin 5.

DGCR8 expression is also dysregulated in cancer (TABLE 1). In addition, mutations of DGCR8 were reported in Wilms tumours: a recurrent mutation (E518K) in dsRBD1 results in the reduced expression of crucial miRNAs in the tumours^22–24 (FIG. 2; see Supplementary information S1 (table)). Similar to knockdown of DROSHA, knockdown of DGCR8 also promotes cellular transformation and tumour growth¹⁵, further confirming the important role of Microprocessor in cancer.

Pre-miRNA export in cancer

Pre-miRNAs are exported into the cytoplasm to be processed into mature miRNAs. The export of pre-miRNAs is mediated by XPO5 and its cofactor, RanGTP⁴¹. Three recurrent heterozygous XPO5-inactivating mutations were identified in sporadic colon, gastric and endometrial tumours with microsatellite instability⁷⁶ (FIG. 2; see Supplementary information S1 (table)). These XPO5 mutations impair pre-miRNA export and result in an accumulation of pre-miRNAs in the nucleus, leading to defects in miRNA biogenesis. In addition, genetic and epigenetic association studies revealed that XPO5 genetic variation and expression level are associated with the risk of breast cancer⁷⁷. Therefore, XPO5 dysregulation contributes to miRNA processing defects and tumorigenesis.

Pre-miRNA processing in cancer

DICER1 mutations

After being exported to the cytoplasm, pre-miRNAs are then processed by DICER1 to form ~22-nucleotide mature miRNAs⁷⁸. DICER1 is a large multi-domain nuclease that contains two helicase domains, a dimerization domain, a Piwi–Argonaute–Zwille (PAZ) domain, two RNase III domains (RNase IIIa and RNase IIIb) and a dsRBD (FIG. 2; see Supplementary information S1 (table)). In addition to its function in pre-miRNA cleavage, DICER1 is required for the assembly of the minimal miRISC that executes miRNA function in repressing target gene expression⁴⁸. Depletion of DICER1 in cancer cells or mouse models promotes cell growth and tumorigenesis, indicating the important function of DICER1 in oncogenesis^15,79. Furthermore, Dicer is considered a haploinsufficient tumour suppressor gene, as loss of a single Dicer1 allele reduces survival in a mouse model of lung cancer⁷⁹.

Heterozygous germline DICER1 mutations were first identified to be responsible for pleuropulmonary blastoma (PPB), a rare paediatric lung tumour that arises during fetal lung development and is often part of an inherited cancer syndrome (Online Mendelian Inheritance in Man (OMIM) #601200)¹⁶. Germline frameshift or nonsense mutations mainly affect DICER1 upstream of the region encoding RNase III domains (FIG. 2), resulting in truncated DICER1 proteins lacking the C-terminal catalytic domains. DICER1 loss of heterozygosity (LOH) is almost never observed in human tumours, and homozygous Dicer1 loss is generally selected against in mouse cancer models⁷⁹. Although more than 50% of heterozygous germline DICER1 mutation carriers are clinically unaffected, the tumours that develop in PPB patients are typically associated with another important group of DICER1 mutations: recurrent somatic mutations in the RNase IIIb domain^18,80. The mutation hot spots of the RNase IIIb domain occur in the metal-binding residues (E1705, D1709, G1809, D1810 and E1813)¹⁸ (FIG. 2); this domain is responsible for the cleavage of the 3′ end of the miRNAs derived from the 5′ side of the pre-miRNA hairpin called 5p miRNAs. These mutations do not change DICER1 protein expression but instead cause defects in the function of the RNase IIIb domain. As a result, the maturation of 5p miRNAs is specifically blocked, while the processing of 3p miRNAs (miRNAs derived from the 3′ side of the pre-miRNA hairpin) remains unaffected, leading to the global loss of 5p miRNAs in cancer^17,18. Particularly, DICER1 RNase IIIb mutations strongly reduce the expression of the members of the let-7 tumour-suppressive miRNA family (that are all 5′ derived), which probably helps to explain the selective pressures that give rise to this specific mutation spectrum in cancers. Interestingly, modelling of PPB in mice supports the idea that Dicer1 deletion in the distal airway epithelium causes non-cellautonomous tumour initiation, whereby Dicer1 loss in the epithelium causes the underlying mesenchymal cells to be malignantly transformed⁸¹. DICER1 mutations are frequently found in different types of inherited tumours: PPB^16,80–84, non-epithelial ovarian cancer^18,84,85, Wilms tumour^22,86,87, pituitary blastoma⁸⁸, cystic nephroma⁸⁹, rhabdomyosarcoma⁹⁰ and others⁹¹ (see Supplementary information S1 (table)). As a result, patients harbouring these DICER1 mutations have reduced DICER1 expression and/or impaired DICER1 function, which cause the abnormal expression of miRNAs and contribute to the pathogenesis of cancer. As such, DICER1 mutation is considered a tumour predisposition syndrome known as DICER1 syndrome²⁰. This topic has recently been reviewed in detail¹⁹.

In addition to genetic mutations of DICER1, DICER1 expression is often dysregulated in cancer. Similar to that of DROSHA, DICER1 expression can be increased or decreased in cancer, depending on the cancer type (TABLE 1). Many oncoproteins and dysregulated tumour suppressors regulate cancer progression by targeting DICER1 expression. For example, the p53 family member TAp63 directly binds to the promoters of DICER1 and miR-130b and drives their expression to suppress tumorigenesis and metastasis⁹². Overall, both genetic mutation and dysregulation of DICER1 can result in aberrant miRNA expression and tumorigenesis.

TRBP mutations

Impaired function of TRBP also contributes to miRNA dysregulation in cancer. Sequencing of the genes encoding the miRNA processing machinery revealed two frameshift mutations of TRBP in sporadic and hereditary carcinomas with microsatellite instability^93,94 (FIG. 2; see Supplementary information S1 (table)). These mutations cause reduced TRBP and DICER1 expression as well as defective processing of pre-miRNAs. Re-introduction of wild-type TRBP in the mutated cell lines rescued TRBP and DICER1 expression, restored miRNA processing and suppressed cancer cell growth in vitro and in vivo⁹³. Interestingly, the expression of TRBP is repressed in the cancer stem cell (CSC) population of Ewing sarcoma family tumour (ESFT), which results in the miRNA profile of ESFT CSCs that is required for CSCassociated self-renewal and tumour growth⁹⁵. Therefore, TRBP-mediated miRNA processing has an important tumour-suppressive role in normal cells.

Other miRNA regulators in cancer

Aberrant expression of or mutations in the genes encoding key components of the miRNA biogenesis pathway contributes to the global repression of miRNAs in cancer. However, a widespread suppression of miRNA expression has been observed in cancers with normal expression of the miRNA biogenesis machinery. This suggests that other pathways regulating miRNA processing are dysregulated in cancer. We highlight below recent discoveries of selected cancer-relevant pathways involved in the regulation of miRNA biogenesis.

Regulators of Microprocessor

The original characterization of a large DROSHA-containing complex identified multiple classes of RNA-binding proteins, including the DEAD (Asp-Glu-Ala-Asp) box helicases DDX5 (also known as p68) and DDX17 (also known as p72), Ewing sarcoma family proteins and heterogeneous nuclear ribonucleoproteins (hnRNPs)³². These Microprocessorassociated proteins can directly affect Microprocessor activity, and alterations in this regulation can result in aberrant miRNA biogenesis in cancer⁹⁶. Other factors might also regulate Microprocessor activity in cancer: for example, the tumour suppressor BRCA1 interacts with multiple Microprocessor regulators to facilitate miRNA biogenesis⁹⁷. Moreover, RNA-binding proteins such as KH type-splicing regulatory protein (KSRP; also known as FUBP2)⁹⁸, serine/arginine-rich splicing factor 1 (SRSF1)⁹⁹, hnRNP A1 (REFS ^100,101) and FUS (also known as TLS)¹⁰² bind to certain regions of primiRNAs (stem or terminal loop) and facilitate DROSHA recruitment and function (FIG. 3).

Figure 3. Dysregulated miRNA biogenesis in cancer.

Aberrant microRNA (miRNA) biogenesis in cancer occurs at different steps during miRNA maturation. a | Genetic alterations, epigenetic modifications, oncogenes and tumour suppressors negatively or positively regulate primary miRNA (pri-miRNA) transcription in cancer. b | Pri-miRNA processing is regulated in the following ways: hypoxia, genetic mutations and transcriptional regulation control DROSHA and DiGeorge syndrome critical region 8 (DGCR8) expression in cancer; RNA-binding proteins such as DEAD box protein 5 (DDX5), DDX17 and BRCA1 modulate Microprocessor activity in cancer; cell signalling pathways such as Hippo and bone morphogenetic protein (BMP) regulate pri-miRNA processing; and LIN28 proteins selectively block the processing of pri-let-7. c | Genetic mutations in and transcriptional regulation of exportin 5 (XPO5) affect XPO5-mediated precursor miRNA (pre-miRNA) export in cancer. d | Pre-miRNA processing in cancer is regulated in the following ways: hypoxia, genetic mutations and transcriptional regulation modulate DICER1 expression and function to control pre-miRNA cleavage in cancer; LIN28 proteins selectively bind to pre-let-7 and recruit terminal uridylyltransferase 4 (TUT4), TUT7 and DIS3-like exonuclease 2 (DIS3L2) to degrade pre-let-7; and hypoxia-induced and epidermal growth factor receptor (EGFR)-induced phosphorylation of Y393 of Argonaute 2 (AGO2) inhibits pre-miRNA processing. e | miRNA function is regulated in the following ways: competing endogenous RNA (ceRNA) inhibits miRNA function in cancer (high-mobility group AT-hook 2 (HMGA2) blocks let-7 function), as do mutations of miRNA-binding sites in non-small cell lung cancer (mutation of let-7-binding site in the 3′ untranslated region (UTR) of KRAS mRNA). hnRNP, heterogeneous nuclear ribonucleoprotein; KDM6, lysine-specific demethylase 6; KSRP, KH-type splicing regulatory protein; Pol II, RNA polymerase II; RIII, ribonuclease III; SRSF1, serine/arginine-rich splicing factor 1; TGFβ, transforming growth factor-β; TRBP, transactivation-responsive RNA-binding protein; YAP, Yes-associated protein; ZEB, zinc-finger E-box-binding homeobox.

In addition to regulating Microprocessor activity, DDX5 and DDX17 function as bridging factors for important oncoproteins or tumour suppressors to regulate miRNA biogenesis in cancer. For example, the tumour suppressor protein p53 regulates miRNA biogenesis through association with DDX5 and DDX17. In response to DNA damage, the level of p53 expression increases, which enhances the expression levels of tumour-suppressive miRNAs including miR-34a, miR-16-1, miR-143 and miR-145 (REF. 103). In contrast to miR-34a, which is a transcriptional target of p53 (REF. 55), the other miRNAs are post-transcriptionally regulated by p53. Mediated by DDX5 and DDX17, p53 interacts with the DROSHA complex and promotes the processing of tumour-suppressive pri-miRNAs. Accordingly, miRNA processing is hindered in p53-mutant cells¹⁰³. Given that p53 is frequently mutated in human cancer, dysregulation of miRNA biogenesis by p53 mutation might account for the widespread miRNA repression in cancer (FIG. 3).

Cell signalling control

Cell signalling pathways also modulate Microprocessor activity to dynamically control pri-miRNA processing and miRNA expression in cancer⁹⁶ (FIG. 3). For example, SMADs — which transduce transforming growth factor-β (TGFβ) and bone morphogenetic protein (BMP) signalling — associate with DDX5 and promote miRNA processing by binding to a consensus sequence in the stem region of primiRNAs^104,105. Moreover, the core biogenesis machinery components, including DROSHA, DGCR8, DICER1 and TRBP, are subject to post-translational control such as phosphorylation and/or acetylation (reviewed in REFS ^106,107). The effect of these protein modifications, and their possible dysregulation in cancer, remains to be determined.

It was recently found that the Hippo pathway controls Microprocessor activity¹⁰⁸. The Hippo pathway controls organ size by regulating cell proliferation and differentiation in response to cell density¹⁰⁹. Given its key role in regulating organ size and cell proliferation, it is perhaps not surprising that the Hippo signalling pathway is frequently perturbed in a variety of human cancers¹⁰⁹. miRNA biogenesis is activated by cell–cell contact and Hippo signalling^108,110. Mechanistically, it was found that the Hippo downstream effector Yes-associated protein 1 (YAP1) post-transcriptionally regulates miRNA biogenesis by targeting DDX17. In in vitro cell culture systems, at low cell density, the growth-suppressive Hippo pathway is inactive, and nuclear YAP1 binds to and sequesters DDX17 to suppress pri-miRNA processing, whereas at high cell densities, the Hippo pathway is active, which leads to YAP1 phosphorylation and its retention in the cytoplasm. When YAP1 is cytoplasmic, DDX17 is able to bind to a specific sequence motif in pri-miRNA, associate with Microprocessor and enhance miRNA biogenesis. Accordingly, inactivation of the Hippo pathway or constitutive activation of YAP1, which occurs in cancer cells, results in widespread miRNA suppression both in human cancer cell lines and in mouse tumour models¹⁰⁸. It will be interesting to explore whether Hippo signalling is responsible for the widespread repression of miRNA expression in cancer.

Stress response

Rapidly growing tumours often experience hypoxia owing to the limited oxygen supply in the tumour microenvironment. Interestingly, miRNA expression and function are dynamically regulated under stress conditions¹¹¹. Oncogenic epidermal growth factor receptor (EGFR) signalling is activated by hypoxia to promote cell growth and oncogenesis¹¹². Identification of the EGFR protein complex in serum-starved EGFtreated HeLa cells revealed that EGFR interacts with AGO2 (REF. 113). In response to hypoxia, EGFR induces the phosphorylation of AGO2 at Y393, which inhibits the interaction between DICER1 and AGO2 and blocks miRNA accumulation. Furthermore, EGFR-mediated AGO2-Y393 phosphorylation is required for cell survival and invasion under hypoxic conditions and is associated with poor survival rates in patients with breast cancer¹¹³. In addition, recent studies uncovered the important role of hypoxia in suppressing DROSHA and DICER1 expression in cancer cells, which results in aberrant miRNA biogenesis and promotes tumour progression^114,115. These studies provide an interesting link between hypoxia and miRNA repression in cancer and uncover a novel oncogenic role of hypoxia in regulating miRNA biogenesis during tumorigenesis^113–115 (FIG. 3).

LIN28-mediated blockade of let-7

The let-7 miRNA family members function as tumour suppressors in multiple cancer types by inhibiting expression of oncogenes and key regulators of mitogenic pathways^116–118. In humans, there are 12 let-7 family members (let-7a-1, let-7a-2, let-7a-3; let-7b; let-7c; let-7d; let-7e; let-7f-1, let-7f-2; let-7g; let-7i; miR-98) located at 8 unlinked chromosomal loci. The let-7 miRNAs are downregulated in numerous cancer types, and low let-7 expression levels correlate with poor prognosis^119–122. The expression of the let-7 miRNA family is coordinately regulated by the paralogous RNA-binding proteins LIN28A and LIN28B during early embryonic development^123–126. Reactivation of this embryonic pathway in adult cells by expression of LIN28A and LIN28B is sufficient to promote cellular transformation and tumorigenesis in vitro and in vivo^127–130. Of note, expression of LIN28B is sufficient to drive neuroblastoma, T cell lymphoma, intestinal adenocarcinoma, Wilms tumour (nephroblastoma) and hepatocellular carcinoma in mouse models^{128,130–133}. LIN28 proteins block cell differentiation, promote cell proliferation and alter cellular metabolism to promote tumorigenesis^134,135. The repression of the let-7 family in these contexts is crucial, as tumour formation is suppressed by enforced expression of let-7g, and genetic deletion of a let-7 locus (let7c2 and let7b) recapitulated the effects of LIN28B overexpression in the intestine^{127–129,133}. Depletion of LIN28A or LIN28B in human cancer cell lines results in decreased cell proliferation, cell invasion and tumorigenicity^129,136, and withdrawal of LIN28B expression can revert liver tumorigenesis in mice¹³⁰. At least 15% of all human cancer samples investigated are characterized by reactivation of either LIN28A or LIN28B, with a corresponding reduction in let-7 levels¹²⁹. Moreover, elevated LIN28A or LIN28B expression correlates with poor prognosis and decreased patient survival^{129,131,137–140}. Considering also that LIN28A and LIN28B expression may characterize distinct tumorigenic subpopulations of cells within the tumour, known as tumour-initiating cells or CSCs¹⁴¹, these studies underscore the importance of the LIN28 proteins in promoting and characterizing various human malignancies and suggest that this pathway represents an important new target for effective cancer therapies.

Mechanistically, LIN28 proteins selectively bind to the terminal loop region of pre-let-7 through RNA–protein interactions through its cold-shock domain and tandem Cys-Cys-His-Cys (CCHC)-type zinc-fingers^142,143. LIN28 proteins recruit two alternative 3′ terminal uridylyltransferases (TUTases), ZCCHC11 (also known as TUT4) and ZCCHC6 (also known as TUT7), to pre-let-7 RNA^144–146. These TUTases are key mediators in the LIN28 blockade of let-7 biogenesis, in which they catalyse the addition of an oligouridine tail to pre-let-7. Uridylated pre-let-7 is resistant to DICER1 processing and is rapidly degraded to prevent let-7 biogenesis in LIN28Aor LIN28B-expressing cells¹²⁵. The enzyme responsible for this decay pathway was recently identified as DIS3L2, a novel 3′–5′ exonuclease that selectively degrades 3′ oligouridylated (>12 uridines) RNA^147–149 (FIG. 3). Intriguingly, DIS3L2 is a tumour suppressor gene that is deleted in Perlman syndrome, which is characterized by fetal overgrowth and cancer predisposition, as well as in ~30% of sporadic Wilms tumours analysed¹⁵⁰. Considering the strong links between DROSHA and DICER1 mutations in Wilms tumours, the demonstrated ability of LIN28A and LIN28B to promote tumorigenesis as well as the tumour-suppressive role of DIS3L2, it is perhaps likely that loss of let-7 expression and/or function is a unifying driver of Wilms tumours and of other types of cancer. This let-7 loss might be accomplished by any of the aforementioned mechanisms as well as by the possible titration of let-7 function via the considerable overexpression of mRNAs containing let-7 binding sites, as was recently suggested for HMGA2 (REF. 151). Another possible mechanism involves mutations in the let-7 binding sites of key downstream targets, thus relieving these mRNAs from let-7 regulation. In support of this, a single-nucleotide polymorphism (SNP) in a let-7 binding site in the 3′ UTR of the KRAS mRNA has been genetically associated with an increased risk of cancer¹⁵² (FIG. 3).

Conclusions and perspectives

Discoveries over the past 15 years have provided substantial insights into the mechanisms controlling miRNA biogenesis. The identification and characterization of the core miRNA biogenesis machinery provided the framework for recent developments that uncovered cancer-causing mutations in miRNA biogenesis components as well as for the identification of cellular signalling and regulatory pathways that control different subsets of miRNAs. Although clear examples of individual miRNAs with oncogenic function have been described, the net effect of widespread miRNA depletion is to promote tumorigenesis. This was first demonstrated in human cancer cells and mouse models and is strongly supported by the mutations recently identified in core miRNA biogenesis genes.

Analogous to the defective differentiation phenotype of miRNA-deficient embryonic stem cells, it seems that also in the context of cancer the dominant function of miRNAs is to help to maintain differentiated cells in a particular cell state or lineage^153,154. In this model, loss of miRNAs facilitates epigenetic reprogramming, loss of differentiated cell identity and adoption of an undifferentiated cancer phenotype. Indeed, DGCR8 depletion is sufficient to reprogramme human primary keratinocytes to induced pluripotent-like cells¹⁵⁵. Furthermore, miRNA expression is globally elevated in confluent cells, which is consistent with their roles in suppressing cell proliferation and in coordinating the altered metabolic demands of less-proliferative cells and tissues^108,110. Presumably this is how widespread miRNA depletion — through loss of components of the biogenesis machinery or loss of growth-suppressive signalling pathways (for example, the Hippo pathway) — contributes to rapid cancer cell proliferation and tumour growth. In this way, widespread loss of miRNAs functionally cooperates with other cancer hallmarks to regulate cancer progression¹⁵⁶. Is loss of any particular miRNA or miRNA family responsible for these tumorigenic effects? One good candidate is the let-7 family. The let-7 family is required in adult fibroblasts to suppress the expression of a mid-gestation embryonic gene signature that is enriched with oncofetal genes¹⁵⁷. Conversely, antagonizing let-7 with antisense oligonucleotides can enhance reprogramming to induced pluripotent stem cells, suggesting that let-7 has a dominant role in stem cell differentiation¹⁵⁸. Indeed, re-introduction of let-7 into miRNA-deficient mouse embryonic stem cells rescued the stem cell differentiation phenotype¹⁵⁸; similarly, restoration of let-7 expression was shown to effectively inhibit growth of lung and breast cancer cells, as well as in mouse models of hepatocellular carcinoma and Wilms tumours^118,159,160. Thus, let-7 emerges as a key regulator in stem cell biology and tumorigenesis and, as outlined in this Review, there are multiple mechanisms by which cancer cells inactivate this miRNA ‘guardian’ of differentiation, proliferation and metabolic reprogramming.

Future work promises to illuminate the most relevant miRNAs in the context of different cancer types and will probably uncover additional pathways that control the expression of individual miRNAs or of miRNA subsets. Studies in this area will be facilitated by the recent advances in genome engineering using CRISPR–Cas9 (clustered regularly interspaced short palindromic repeat–CRISPR-associated protein 9) technology, in mouse modelling and in the use of organoid culture systems to model cancer¹⁶¹, as well as by the application of high-throughput sequencing technologies that will uncover cancer-causing mutations in patients and that can be applied in the laboratory to examine the effects of possible regulators on global miRNA expression profiles²¹. With this powerful toolkit in hand, the next several years promise exciting discoveries that will help to unlock the secrets of miRNA dysregulation in cancer. Understanding the molecular and cellular pathways controlling miRNA biogenesis and how these mechanisms go awry in cancer will identify promising therapeutic targets that might be readily manipulated by small pharmacological agents to allow restoration of miRNA expression profiles and to bypass the challenges associated with delivering synthetic miRNA mimics or antagomiRs.

Glossary

3′ untranslated region

(3′ UTR). The non-coding region of mRNA between the translation termination codon and the poly(A) tail. The 3′ UTR often contains regulatory elements, such as miRNA binding sites, for post-transcriptional regulation of gene expression

Ribonuclease III

(RNase III). Enzymes that can specifically recognize and cleave double-stranded RNA with their ribonuclease III domains

Germline mutations

Heritable gene mutations that occur in germline tissues

Somatic mutations

Gene mutations that occur in non-germline tissues that are not inherited

Post-transcriptional gene silencing

A gene-silencing effect that controls gene expression after transcription, often mediated by small non-coding RNAs such as small interfering RNAs (siRNAs) and microRNAs (miRNAs)

Epithelial–mesenchymal transition

(EMT). A process that occurs during development or cancer progression in which the epithelial cells lose their cell polarity and cell–cell adhesion to become mesenchymal cells with migratory and invasive characteristics

CpG islands

Genetic regions with high CpG content, often located at the gene promoter, that have important functions in regulating gene expression

Microsatellite

Short (2–5 bp) tandem repeat of DNA that can be used as a genetic marker

Loss of heterozygosity

(LOH). Deletion or mutation of the normal allele of a gene, of which the other allele is already deleted or inactivated, resulting in loss of both alleles of the gene

Cold-shock domain

A protein domain of ~70 amino acids that is often found in DNA- or RNA-binding proteins and that functions to protect cells during cold temperatures

Cys-Cys-His-Cys (CCHC)-type zinc-fingers

Protein domains that are found in RNA-binding proteins or single-stranded DNA-binding proteins

Terminal uridylyltransferases

(TUTases). Enzymes that catalyse the addition of one or more uridine monophosphate (UMP) molecules to the 3′ end of RNA

Oncofetal genes

Genes that are typically highly expressed during fetal development and repressed in adult life, and reactivated in cancers