Parameters of clustered suboptimal miRNA biogenesis

Significance

A substantial number of animal microRNAs (miRNAs) are transcribed as operons bearing two or more primary miRNA (pri-miRNA) hairpins. It was long believed that the biogenesis of individual miRNAs within a genomic cluster was independent of each other. However, it was more recently shown that nuclear cleavage of a suboptimal pri-miRNA hairpin could be enhanced by a neighboring optimal miRNA hairpin. In this study, we systematically assess the structural parameters for clustered miRNA processing and show that there can be both biogenesis enhancement as well as competition between clustered miRNAs. We exploit the cluster assistance phenomenon to functionally evaluate suboptimal miRNA substrates. Finally, we show how a disease-associated mutation can affect the biogenesis of multiple members of a cluster.

Abstract

The nuclear cleavage of a suboptimal primary miRNA hairpin by the Drosha/DGCR8 complex (“Microprocessor”) can be enhanced by an optimal miRNA neighbor, a phenomenon termed cluster assistance. Several features and biological impacts of this new layer of miRNA regulation are not fully known. Here, we elucidate the parameters of cluster assistance of a suboptimal miRNA and also reveal competitive interactions amongst optimal miRNAs within a cluster. We exploit cluster assistance as a functional assay for suboptimal processing and use this to invalidate putative suboptimal substrates, as well as identify a "solo" suboptimal miRNA. Finally, we report complexity in how specific mutations might affect the biogenesis of clustered miRNAs in disease contexts. This includes how an operon context can buffer the effect of a deleterious processing variant, but reciprocally how a point mutation can have a nonautonomous effect to impair the biogenesis of a clustered, suboptimal, neighbor. These data expand our knowledge regarding regulated miRNA biogenesis in humans and represent a functional assay for empirical definition of suboptimal Microprocessor substrates.

microRNAs (miRNAs) are ~22 nucleotide (nt) RNAs derived from hairpin precursors and mediate extensive gene regulatory networks. The bulk of well-expressed miRNAs in animals are generated via a canonical, stepwise, pathway. Following transcription of a primary miRNA (pri-miRNA), the nuclear Microprocessor complex (composed of the RNase III enzyme Drosha and its partner DGCR8) cleaves near the hairpin base to liberate the pre-miRNA hairpin. This is exported to the cytoplasm, where it is cleaved near the terminal loop by the RNase III enzyme Dicer to produce the miRNA duplex. The duplex is loaded into an Argonaute effector protein and one of the duplex species (the miRNA* or passenger strand) is removed to yield the single-stranded complex that is able to identify complementary targets for regulation (1, 2). In addition to the canonical pathway, numerous noncanonical miRNA biogenesis strategies also exist. These include miRNA substrates that are Drosha-independent [e.g., mirtron hairpins generated by splicing (3, 4)] or are Dicer-independent [e.g., the short, Argonaute2-dependent mir-451 hairpin (5–7)].

In vertebrates, 30~40% miRNA hairpins reside within local genomic clusters, where two or more miRNA hairpins reside in a single primary transcript (8). Experimental tests show that clustered miRNAs can typically still be processed when removed from their normal operon context into a "solo" context; i.e., (9–12). This might suggest that there is no particular preference as to their order of biogenesis. However, studies on the mir-17~92a cluster revealed regulated, stepwise, processing of the inner miRNA members of this cluster (13, 14). Additional careful analyses revealed that efficient biogenesis of certain miRNAs from mammals, flies, and viruses can depend on their normal clustered context (15–18). However, the molecular strategies for such cluster assistance are not fully understood. On the other hand, the existence of miRNA clustering might impact miRNA evolution, since the biogenesis of young miRNAs that lack optimal features might be enhanced if they happen to be born near existing miRNAs (19, 20).

As gatekeeper for a multitude of RNA hairpins that might access the miRNA pathway, the Microprocessor complex has high affinity and processing efficiency of only a subset of pri-miRNA hairpins bearing defined features. Over the past years, many features of optimal pri-miRNAs were revealed, including single-stranded flanking regions, a double-stranded stem of ~35 basepairs (bps), a single-stranded terminal loop of >10 nts, and specific motif sequences within the pri-miRNA hairpin and flanking regions. These data derive from analyses of individual miRNA constructs as well as high-throughput processing assays (21–34). While most conserved miRNAs bear these structural features along with one or more auxiliary motifs, a number of suboptimal miRNAs exist that lack optimal feature.

Recently, a new layer of miRNA biogenesis regulation was uncovered through studies of suboptimal miRNA hairpins (35–37). In particular, studies of the conserved vertebrate mir-144/451 cluster revealed that mir-451, which has a short hairpin stem, requires Ago2 but not Dicer to cleave pre-mir-451 and generate mature miR-451. Although the pri-mir-451 hairpin is a highly suboptimal Microprocessor substrate, due to its short stem and small terminal loop, it can be efficiently processed by Drosha/DGCR8 to release pre-mir-451 (5, 7, 38). This contradictory phenomenon is intimately connected to its clustered neighbor, mir-144 (35, 36). Unlike the suboptimal pri-mir-451 hairpin, pri-mir-144 has typical hairpin features that enable efficient recruitment of Microprocessor. After cleavage of pri-mir-144, the released Microprocessor transfers to nearby pri-mir-451 hairpin to promote its processing (35, 36). Interestingly, another study focusing on the mir-15a/16 cluster used an elegant genome-wide CRISPR-Cas9 screen to identify how SAFB1/2 and ERH selectively promote cluster assistance of suboptimal miRNA hairpins (37). In parallel, a biochemical approach identified ERH as a DGCR8-associated factor that enables suboptimal mir-451 maturation from the mir-144/451 operon (36). Although the mechanism of these factors is still unclear, it is intriguing to imagine scenarios for their action. For example, they might operate to facilitate Microprocessor transfer, by stabilizing Microprocessor on suboptimal hairpins, or by forming Microprocessor dimers in the miRNA cluster (39).

Although these new findings on miRNA cluster processing bring new insights to miRNA field, several layers of this regulation remain unclear, or only explored in limited cases. In this study, we systemically investigate features that underlie miRNA cluster assistance, including diverse hairpin parameters and impacts of clusters with multiple miRNAs. We also use cluster assistance as functional assay to reveal the existence of solo suboptimal miRNAs. Finally, we provide evidence that a cancer mutation that renders one miRNA suboptimal can have a nonautonomous effect to impair processing of its suboptimal cluster neighbor. These data improve our knowledge of regulated miRNA biogenesis and its consequences during disease.

Results

Sensitivity of a Suboptimal miRNA to Structural Variation of Its Neighboring miRNA.

The structures of optimal pri-miRNAs are heterogeneous. For example, their terminal loop size can vary from several nts to >20 nts (35). Although nuclear processing of a suboptimal miRNA hairpin can be enhanced by an optimal miRNA neighbor, the minimal functional requirements for the partner miRNA hairpins are not fully known. We investigated this using a series of structural mutations on neighboring miRNA hairpins and study their effects on processing and function of suboptimal miRNAs. We focused on structural features instead of motif sequences on pri-miRNAs, as hairpin structures are usually more critical for microprocessing; most miRNA genes lack some of motifs, indicating that they likely play auxiliary roles in pri-miRNA processing.

Based on our previous studies of the mir-144/451 cluster as a model for cluster assistance and regulated biogenesis (35, 40), we made a series of loop size mutants of pri-mir-144, from 2 to 14 nts (Fig. 1A). We replaced the wild-type mir-144 loop with artificial sequences that maintained single-stranded structures, to avoid potential regulation from wild-type loop sequences (Fig. 1B). To investigate the potential effects on the RNA stability of these pri-mir-144/451 variants caused by loop mutations, we first generated and validated a new Drosha-knockout HEK293T cell line bearing biallelic frameshift mutations within N-terminal exon 5 (SI Appendix, Fig. S1A). We then transfected the panel of pri-mir-144/451 variants into the Drosha-KO cells, inhibited RNA polymerase II activity by actinomycin D treatment, and then assayed pri-miRNA levels at different time points following transcription arrest (SI Appendix, Fig. S1B). Although pri-mir-144/451 levels decreased over time, we did not observe substantial RNA stability differences amongst the pri-mir-144/451 variants (SI Appendix, Fig. S1C). These data indicate that loop mutations did not substantially affect the stability of pri-mir-144/451 variants.

Fig. 1.

Sensitivity of suboptimal miRNA processing to its neighboring miRNA hairpin. (A) Schematic of the mir-144/451 cluster. (B) Loop size variants (from 2 nts to 14 nts) of pri-mir-144. The nts mutated from 144-WT are marked in red. Note the asterisk near the apical loop of mir-144-WT, as this U/A pair was determined not to be base paired in prior SHAPE-MaP experiments (35). (C) Processing of wild-type and different mir-144/451 variants by Northern blotting in HEK293T cells. Cotransfected mir-375 and endogenous let-7a and U6 snRNAs were probed as controls. RNA size markers (nt) are shown on the left. Microprocessing of both pri-mir-144 and pri-mir-451 is gradually increased with the loop size expanding of mir-144. Note that dicing of pre-mir-144 loop variants are totally blocked due to short dsRNA stem, caused by loss of the structure reshaping in the terminal loop region. (D) Activity of wild-type and variant mir-144/451 constructs on luciferase sensors shows that increased miR-451 biogenesis correlated with increased activity, independent of miR-144 maturation. Unpaired two-tailed Student’s t test was applied (*P < 0.05, ***P < 0.001).

Next, we cotransfected the mir-144/451 variants with plasmids encoding control mir-375 and a non-miRNA control (renilla luciferase) into HEK293T cells. We note that cotransfected mir-375 showed variable maturation between samples, which negatively correlated with maturation of mir-144/451 cluster members (Fig. 1C). This effect has been observed in other studies utilizing miRNA construct transfection (28, 36) and may involve biogenesis competition between cotransfected constructs due to limiting Argonaute availability (41). However, we can be certain about equivalent transfection efficiency because we detected comparable renilla luciferase activity across different samples in two independent replicates (SI Appendix, Fig. S2). Moreover, we included additional loading controls by probing for endogenous let-7a miRNA and U6 snRNA in our Northern blots, neither of which was substantially affected by miRNA transfection.

With these validations in hand, we analyzed the biogenesis of miRNA cluster variants. Northern blotting showed an increased accumulation of pre-mir-144 with loop size expansion, but no mature miR-144 was generated due to the short Dicer-incompetent stem of these pre-mir-144 variants (Fig. 1C). Meanwhile, the accumulation of both pre-mir-451 hairpin and mature miR-451 was well coordinated with that of pre-mir-144 (Fig. 1C). This indicated that microprocessing of pri-mir-451 was responsive to the structures of its neighbor and could be uncoupled from maturation of its clustered neighbor. The mir-144 variants with ≥10 nt loops dramatically enhanced pri-mir-144 and thus pri-mir-451 processing, which is consistent with previous study that a relatively large terminal loop promotes pri-miRNA cleavage (22). Sensor assays for miR-451 activity showed similar trends as with its processing patterns, including robust capacity for target repression when mir-144 neighbor variants bore terminal loops of ≥10 nts (Fig. 1D).

Another important structural feature of optimal pri-miRNAs is that the total length of their dsRNA stem should be ~35 bps. To test the cluster assistance sensitivity to the stem length of neighboring miRNA hairpins, we made a series of shortened stems on pri-mir-144, which impair (SI Appendix, Fig. S3A). Northern blotting and sensor assays showed that these stem variants are deleterious to the processing of both miR-144 and miR-451 members within the cluster (SI Appendix, Fig. S3 B and C). This confirmed that proper dsRNA stem length is essential for efficient pri-miRNA processing. Previous studies showed that alteration of pri-mir-144 stem length also affects processing of suboptimal pri-mir-451 (36, 40). For example, a series of lengthened stem variants of pri-mir-144 showed modest effects on their processing, but a variant with 5 bp stem insertion significantly attenuated processing of both pri-mir-144 and pri-mir-451 (40). These results clearly indicate that microprocessing of suboptimal miRNAs is quite sensitive to the structures of their neighbors.

Previously, we showed that nuclear processing of pri-mir-451 could also be enhanced when the pri-mir-144 neighbor was substituted with other optimal miRNA hairpins (35, 36). Here, we further replaced pri-mir-144 with the pri-mir-7a hairpin or with a chimeric hairpin combining mir-144 terminal loop and mir-7a mature duplex (SI Appendix, Fig. S3 D and E). These miRNA cluster constructs generated mature, active miR-451 (SI Appendix, Fig. S3 F and G), further confirming that biogenesis of miR-451 can be efficiently enhanced by an arbitrary neighboring optimal miRNA, even a synthetic one.

Structural Features of a Suboptimal Microprocessor Substrate.

The pri-mir-451 hairpin is particularly suboptimal owing to both its short stem and small terminal loop, which render it both Dicer-independent and Ago2-dependent (42). Other Dicer-dependent, Ago2-independent suboptimal miRNA hairpins have qualitatively less deleterious features, whose effects cannot always be predicted. To evaluate the structural parameters of suboptimal pri-miRNA processing, we constructed a series of pri-mir-451 mutants by varying the terminal loop sizes or dsRNA stem length. These variants extend our previous assays of mir-451 processing (43), with one notable difference. For these tests, we introduced a loop deletion in the mir-144 neighbor (mir-144LD) that renders it Microprocessor-incompetent (40). Therefore, these new miRNA operon constructs isolate the effects of structural changes of pri-mir-451 on its processing.

First, we evaluated the effect of terminal hairpin loop size. We fixed the dsRNA stem of pre-mir-451 to an optimal Dicer-substrate length (24 bp, “451S24”) and made successive loop size variants ranging from 4 to 9 nts (Fig. 2A). We termed the parent structure as 144LD-451S24, with L# suffix to designate the terminal loop length. With successive loop expansion, both pre-mir-451 and mature miR-451 products increased progressively, indicating enhanced biogenesis (Fig. 2B). In parallel, sensor assays showed progressive increase in target repression with larger loop variants (Fig. 2C). The 9nt loop variant (144LD-451S24L9) generated the most mature miR-451 and had the greatest repression activity. Interestingly, as the loop size of pri-mir-451S24 was increased, the microprocessing of neighboring suboptimal pri-mir-144LD was coordinately and progressively increased (Fig. 2B). This “reverse” enhancement occurred despite the fact that no mature miR-144 was detected, consistent with the deletion of loop sequences needed for Dicer processing of pre-mir-144 (40). Overall, these data emphasize that miRNA cluster assistance is not directional, as it can enhance the nuclear cleavage of either a downstream or an upstream hairpin. Moreover, the neighbor does not have to be an optimal miRNA per se, since the process can be uncoupled from pre-miRNA cleavage by cytoplasmic Dicer.

Fig. 2.

Parameters of suboptimal miRNA structures. (A and A') Variant constructs of mir-451 terminal loop and stem structures. All these mir-451 variants are clustered with a loop-deleted, nonfunctional mir-144 (144LD). (A) Loop variants (from 4 nts to 9 nts) of mir-451 with a defined long upper stem (24 bp, “451S24”). (A') Stem variants (from 18 bp to 24 bp) of mir-451 with a defined large terminal loop (9 nts), in a clustered context with mir-144LD. Note that the final constructs in these series are identical (S24-L9 and L9-S24). (B) Processing of wild-type and different mir-144/451 variants shows that biogenesis of miR-451 increases with loop expansion. As well, elongation of the mir-451 stem increased the accumulation of miR-451, although the processing of pre-mir-451 variants changed from being Ago2-dependent to Dicer-dependent. Note that long-stem mir-451 variants exhibit Dicer processing pattern for mature miR-451, and reciprocally enhance the accumulation of loop-deleted pre-mir-144LD. Nevertheless, pre-mir-144LD is blocked from further maturation, separating effects of nuclear and cytoplasmic biogenesis. Endogenous let-7a and U6 snRNAs were probed as loading controls. (C) Activity assay of mir-451 from these loop variants. Unpaired two-tailed Student’s t test was applied (*P < 0.05, **P < 0.01, ***P < 0.001). (D) Activity assay of mir-451 from these stem variants. Unpaired two-tailed Student’s t test was applied (***P < 0.001).

Second, we evaluated the effect of hairpin stem length. We fixed the terminal loop of pri-mir-451 to an optimal size (451L9) and varied the miRNA duplex region from 18 to 24 bps (Fig. 2A'). Note that as mature miR-451 sequences extend into the loop region of wild-type pri-mir-451, we extended mature miR-451 (the blue sequences) into the terminal loop when the variants bear short stems (Fig. 2A'). This design ensured a fair comparison of target repression capacity of different constructs. Note that the final constructs in each of these series are accordingly identical (stem-24 and loop-9), although we tested them independently for each of these series (S24L9 and L9S24) for consistency. Although the cytoplasmic processing pattern of these mir-451 variants converted from being Ago2-dependent to Dicer-dependent as the stem increased from 18 bps to 24 bps, their nuclear processing (Fig. 2B) and target repression capacity (Fig. 2D) were gradually elevated. Also in line with the loop size variants, microprocessing of neighboring suboptimal pri-mir-144LD hairpins was enhanced by mir-451 variants with longer stems (Fig. 2B).

Taken together, these results establish parameters for suboptimal terminal loop size and stem length and reveal high sensitivity of cluster assistance to structures of neighboring helper miRNAs and of suboptimal miRNAs themselves.

Complex Effects of miRNA Biogenesis in Multimember Clusters.

Most studies on suboptimal miRNA processing focused on miRNA clusters with two members, but many miRNA clusters contain three or more members. We previously showed that introducing a second suboptimal miRNA hairpin into mir-144/451 operon can inhibit mir-451 biogenesis, potentially due to competition effects (35). On the other hand, we used tethering assays to demonstrate that local increase in Microprocessor recruitment, independent of miRNA biogenesis, can improve miR-451 biogenesis (35). Thus, it also seemed plausible that multiple optimal miRNA neighbors might recruit more Microprocessor to further enhance suboptimal miRNA processing.

To test local competition vs enhancement by multiple miRNA neighbors, we inserted the optimal hairpin mir-545 into the mir-144/451 cluster (Fig. 3A). Not only did this fail to confer further enhancement, the mir-545/144/451 cluster actually generated less mature and active miR-451 than the original mir-144/451 cluster (Fig. 3 B and C). One possible explanation is that the local Microprocessor recruited by one miRNA helper saturates the cluster assistance effect, which cannot be supplemented by another optimal hairpin. We tested this further by lowering Microprocessor levels by siRNA-mediated depletion of endogenous Drosha. We validated functional knockdown of Drosha by strong reduction of its mRNA, concomitant with strong regulation of DGCR8 mRNA (SI Appendix, Fig. S4), as expected from loss of Microprocessor cross-regulation (44). These conditions did not strongly affect endogenous miRNAs that presumably perdured from before knockdown (e.g., let-7, Fig. 3D), but clearly reduced maturation from newly transfected miRNA constructs (e.g. miR-144 and miR-451, Fig. 3D). Nevertheless, upon evaluating miR-451 biogenesis from constructs with 0, 1 or 2 helper miRNAs, we failed to observe enhancement of miR-451 biogenesis from mir-545/144/451 compared to mir-144/451 (Fig. 3D). These data indicate that cluster enhancement does not scale with numbers of optimal miRNA hairpin substrates.

Fig. 3.

Suboptimal and canonical miRNA processing within multimember clusters. (A) Schematic of artificial mir-545/144/451 cluster. (B and C) Biogenesis and activity of suboptimal miR-451 was not further enhanced by adding mir-545 to the mir-144/451 cluster; (B) activity sensor assay and (C) Northern blotting assay. Unpaired two-tailed Student’s t test was applied (*P < 0.05). (D) Comparison of mir-451 processing from mir-451, mir-144/451 and mir-545/144/451 constructs under limiting Microprocessor conditions. Intermediate and mature species from mir-144 and mir-451 are reduced upon Drosha depletion, but the level of miR-451 cluster enhancement is similar between mir-144/451 and mir-545/144/451 constructs. (E) Schematic of mir-23a/27a/24 cluster in which all three miRNAs are canonical. (F and G) Processing competition amongst canonical members of miRNA clusters. (F) Activity sensor assay and (G) Northern blotting show that maturation of miR-24 is highest when produced from the solo context, and maturation of miR-27a is higher when produced from a 2-miRNA cluster compared to 3-miRNA cluster. Unpaired two-tailed Student’s t tests were applied (**P < 0.01, ***P < 0.001). (H) Schematics of additional mir-24 cluster variants. (I) Maturation of mir-24 can be inhibited in cluster with other strong Microprocessor substrates, independently of mature miRNA biogenesis. This is partially rescued by ectopic Microprocessor.

We further investigated the biogenesis of mir-412, which is normally located in a cluster with 10 members (mir-323b~656); mir-412 bears a small terminal loop (SI Appendix, Fig. S5A). To simplify this analysis, we cloned constructs of mir-412 with either or both of its flanking optimal neighbors mir-409 and mir-369 (SI Appendix, Fig. S5B). Compared with mir-412 solo construct, the presence of either mir-409 or mir-369 improved biogenesis of suboptimal mir-412 (SI Appendix, Fig. S5C). However, no further enhancement was observed in the mir-409/412/369 cluster. This result was consistent with the above tests of mir-545/144/451 in that an extra optimal hairpin could not further enhance suboptimal miRNA processing.

Finally, we took the opportunity to test a miRNA cluster bearing only optimal hairpins. The cluster assistance model proposes that a miRNA operon may enhance the local concentration of Microprocessor. Although most canonical pri-miRNAs are not believed to have difficulty to recruit sufficient Microprocessor, it might still be the case that canonical miRNAs might be subject to some degree of cluster assistance. We designed miRNA constructs containing 1, 2, or 3 members of the mir-23a/27a/24 cluster (Fig. 3E), all of which appear to be canonical hairpins. However, sensor tests for miR-24 activity showed that it actually had the highest activity when expressed by itself, compared to the 2- or 3-member operon context (Fig. 3F). We analyzed this further using Northern blotting. These tests clearly showed that the most miR-24 was produced from the solo mir-24 construct, with less from the mir-27a/24 or the mir-23a/27a/24 constructs. Furthermore, more mature miR-27a was generated from mir-27a/24 compared to mir-23a/27a/24 (Fig. 3G). To minimize overall structural differences between the solo and cluster constructs, we made a cluster construct specifically lacking only the terminal loop regions of mir-23a and mir-27a (Fig. 3H). This mir-24 construct still showed higher expression than the wild-type cluster construct (Fig. 3 F and I). We further replaced pri-mir-27a sequences in the mir-23a/27a/24 cluster with pri-mir-144 (Fig. 3H). This variant still showed biogenesis competition for mir-24 and even competition for miR-144 biogenesis, compared with its solo construct (Fig. 3I). Overexpression of Microprocessor can partially increase mir-24 processing from both solo and cluster constructs but does not fully eliminate the biogenesis differences between mir-24 solo and cluster constructs (Fig. 3I). This suggests that these canonical miRNA members might also compete for downstream miRNA processing and/or effective factors. Overall, these data indicate that the presence of multiple miRNA hairpins within a single transcript does not necessarily provide collective enhancement of biogenesis but may instead cause competition.

Application of Cluster Assistance to Interpret miRNA Hairpin Structure.

Even with experimental definition of suboptimal miRNA hairpin parameters, it is challenging to predict miRNAs that are genuinely suboptimal structures. For example, it is recognized that bioinformatically derived miRNA secondary structures often differ from experimentally derived structures (45), principally due to overprediction of base-pairing (35). We realized that the miRNA cluster assistance effect provides a strategy to help interpret endogenous miRNA hairpin structures.

As an example, we tested the solo locus mir-128-1, which is predicted by RNAStructure (46) to contain two G:U pairs within its distal stem region (Fig. 4A). If true, this would render it a highly suboptimal miRNA. However, when mir-128-1 is coexpressed with the ectopic partner mir-144, we did not observe enhancement of miR-128 biogenesis or activity (Fig. 4 B and C). This suggested that mir-128-1 may not actually be suboptimal. To evaluate this more definitively, we altered the two G:U pairs into strong G:C pairs to force a small loop (128LM, Fig. 4A). This indeed strongly compromised biogenesis and function of miR-128-1. However, this small loop mutant was now susceptible to enhancement by neighboring mir-144. Conversely, we changed the two G:U pairs to be definitively unpaired by introducing G:G mismatches (128LM2, Fig. 4A). This variant exhibit similar biogenesis and function as endogenous mir-128-1, and it was now insensitive to the introduction of an ectopic miRNA neighbor (Fig. 4 B and C). Therefore, miRNA cluster assistance is a simple functional test to assess relative miRNA biogenesis capacity in vivo.

Fig. 4.

Biogenesis of solo suboptimal miRNA loci. (A) Schematic of predicted pri-mir-128-1 structure (from RNAStructure) with two G:U pairs within its terminal stem (highlighted yellow). A question mark (?) indicates that it is unclear whether these genuinely form a continuous stem, which would render it a very small loop. We tested this using variants that either force a small loop (by converting putative G:U pairs into G:C pairs, 128LM) or by forcing a large terminal loop (making G:G unpaired nts, 128LM2). We also tested these mir-128-1 variants in the context of an artificial cluster with canonical mir-144. (B) Northern blotting and (C) activity sensor assay show that biogenesis of wild-type mir-128-1 is not affected by neighboring mir-144, suggesting that it is not functionally suboptimal. The small loop mutant 128LM has reduced biogenesis, which is rescued by a canonical miRNA neighbor, indicating the parameter space exists to detect potential suboptimality. Consistent with this, the large-loop mutant 128LM2 behaves similarly to wild-type. Unpaired two-tailed Student’s t test was applied (***P < 0.001). (D) Schematic of solo suboptimal mir-491 bearing a predicted small apical loop and an artificial cluster construct with mir-545. (E) Northern blotting and (F) activity sensor assays show that biogenesis of mir-491 is enhanced by an optimal neighbor, suggesting that it is normally subject to Microprocessor regulation. Enlarging the terminal loop of mir-491 (491LM) enhances its maturation in the solo context and reduces its receptivity to neighbor enhancement. Unpaired two-tailed Student’s t test was applied (**P < 0.01).

Experimental Validation of a Suboptimal Solo miRNA.

Although characterized suboptimal miRNAs reside in miRNA clusters for efficient processing, some apparently suboptimal miRNAs exist as solo loci. However, as we could not validate mir-128-1 as suboptimal, this raised the question whether suboptimal, solo miRNAs truly exist. We next evaluated the well-conserved solo locus mir-491, whose terminal loop is only 5 nts in length (Fig. 4D). A solo mir-491 construct yielded detectable mature miRNA (Fig. 4E), but it was marginal in target repression (Fig. 4F). To test whether mir-491 definitively bears a suboptimal terminal loop, we artificially enlarged it in the 491LM construct (Fig. 4D). Indeed, this greatly enhanced the production of pre-mir-491 and mature, active miR-491 (Fig. 4 E and F). Moreover, addition of an ectopic helper miRNA (mir-545) also greatly enhanced miR-491 biogenesis and activity (Fig. 4 E and F).

Importantly, the small suboptimal loop of mir-491 is retained across mammalian orthologs (SI Appendix, Fig. S6). Thus, impaired biogenesis of miR-491 is not an incidental feature, but seems to be selected for its biological function. Since mir-491 does not have an endogenous helper miRNA, it may be regulated by other strategies that can enhance its nuclear biogenesis in the appropriate setting. Indeed, its entire terminal loop is highly conserved across mammals (SI Appendix, Fig. S6) and thus seems well-positioned to be recognized by some trans-acting factor(s), as has been shown for other regulated miRNAs (40, 47–50). Thus, cluster assistance is a functional assay that can identify regulated miRNAs.

Potential Impact of miRNA Cluster Assistance in Disease-Related miRNA Mutations.

Although miRNA cluster assistance has been documented as a regulatory phenomenon, it has thus far not been connected to disease situations. Since the normal impact of miRNA suboptimality on biogenesis is hidden within a clustered context, we hypothesized that certain mutations deleterious to miRNA biogenesis might be buffered within an operon. In this situation, the biogenesis of such a miRNA mutant might benefit from cluster assistance, if positioned appropriately. If true, this might contribute to the high frequency of clustered miRNAs in metazoans.

To test the plausibility of this scenario, we selected the tumor suppressor mir-34b/34c cluster, for which we generated several expression constructs (Fig. 5A). Both miRNA hairpins have optimal structures and their mature sequences (miR-34b-5p and miR-34c-5p) differ only at nts 10 and 11 (Fig. 5B). This makes it difficult to distinguish their processing and function using Northern blotting and seed pairing–dependent sensor assays. However, base pairing of nts 10 to 11 between miRNA sequences and targets is essential for Ago2-mediated target cleavage. Thus, we considered that perfect miRNA sensors might be differentially sensitive to regulation by miR-34b-5p and miR-34c-5p. Indeed, we found that mir-34b-5p and mir-34c-5p could efficiently repress their cognate-perfect sensors but had only mild activity on their noncognate seed-matched sensors (Fig. 5C).

Fig. 5.

Cluster assistance buffers a miRNA against a deleterious biogenesis mutation. (A) Schematic of mir-34b/34c cluster and truncated constructs. (B) The seed regions of mir-34b and mir-34c are identical, but positions 10 and 11 (red sequences) are different, allowing them to exert differential repression of perfect target sensors. (C) The sensor assays validate that mir-34b and mir-34c repress their cognate-perfect sensors much more strongly than the noncognate sensors. Unpaired two-tailed Student’s t test was applied (***P < 0.001). (D) Structure of pri-mir-34c and a small loop variant bearing a point mutation. (E) Activity sensor assay shows that the small loop mutation impairs mir-34c function as a solo miRNA, but neighboring mir-34b allows it to maintain wild-type activity. Unpaired two-tailed Student’s t test was applied (*P < 0.05, **P < 0.01). (F) Model for how cluster assistance may buffer the effect of deleterious mutations that impair miRNA biogenesis.

We modeled a potentially deleterious mutation by introducing a single C-to-G point mutation in the terminal loop of pri-mir-34c. This minor change is predicted to render mir-34c as a suboptimal small loop hairpin (Fig. 5D). To test this expectation, we assayed the activity of mir-34c when expressed from clustered or solo contexts, including a longer construct that was deleted for pre-mir-34b (34c/Δpre-34b) or a shorter construct bearing only mir-34c (Fig. 5A). When we used normal hairpin sequences, mir-34c activity was insensitive to its genomic context (Fig. 5E). By contrast, the small loop mir-34c mutant had poor capacity for target repression when expressed from either construct lacking mir-34b but was rescued when paired with its miRNA neighbor (Fig. 5E). These data provide support to the notion that miRNA clustering can buffer the deleterious impact of certain hairpin mutations that may arise in disease (Fig. 5F).

Reciprocally, we wondered whether the loss of miRNA cluster assistance might cause an individual miRNA mutation to impact other clustered members. The mir-15a/16 locus was previously shown to be subject to cluster assistance, in which optimal mir-16 enhances the processing of its suboptimal neighbor mir-15a (37). Previous studies showed that chronic lymphocytic leukemia (CLL) patients exhibit recurrent deletion and/or downregulation of miR-15 and/or miR-16 (51, 52), suggesting these as tumor suppressors. One such mutation affects a highly conserved sequence downstream of the pri-mir-16 hairpin (Fig. 6 A and B). While its significance was not known at the time, it was later realized that this destroys the downstream CNNC motif, which recruits SRSF3 (SRp20) to enhance pri-miRNA processing (26).

Fig. 6.

Nonautonomous effect of a cancer mutation on clustered miRNA biogenesis. (A) Schematic of the tumor suppressor mir-15a/16 cluster and a CLL patient mutation in the CNNC motif downstream of mir-16 hairpin. (B) The CNNC motif of pri-mir-16 is highly conserved across vertebrates, similar to mir-16 duplex sequences. (C) Northern blotting shows that the C to T mutation in the mir-16 CNNC motif not only impairs its biogenesis but also reduces maturation of its suboptimal neighbor mir-15a. (D) Model for how a single-point mutation in cancer can impair both its associated miRNA hairpin (autonomous effect) as well as its neighboring miRNA (nonautonomous effect).

Our tests confirm that this mutation in the CNNC motif indeed impairs the nuclear biogenesis of mir-16, from both a solo context and the normal mir-15a/16 cluster context. This was particularly evident in the reduced level of pre-mir-16 (Fig. 6C, dotted box); the endogenous expression of mature miR-16-5p in HEK293T cells partly masks the effect of miRNA expression from the transfected constructs. However, since mir-15a requires optimal nuclear biogenesis of its neighbor, we also tested its processing. Interestingly, processing of mir-15a was also clearly impaired by mutation of the mir-16 CNNC motif (Fig. 6C). Therefore, a single nt change outside of the pre-miRNA can reduce the biogenesis of both members of a tumor suppressor miRNA cluster, due a direct effect on cofactor recruitment to mir-16 and an indirect effect on mir-15a, its dependent suboptimal miRNA neighbor (Fig. 6D). We recognize that direct physiological or clinical impacts of these miRNA cluster mutations need further investigation under specific disease contexts. Still, these precedents lead us to speculate that additional impacts of cluster assistance on miRNA expression may be uncovered from directed analysis of human SNPs and somatic mutations recovered from largescale genome sequencing.

Discussion

Regulated Biogenesis of Clustered miRNAs.

When miRNAs were first recognized as an extensive gene family (53–55), it was observed that their hairpins generally exhibit similar lengths of dsRNA stems and sizes of ssRNA loops. Although each miRNA hairpin has different sequences, their overall similar structures implied that these were relevant for their recognition and processing by cellular machineries. Indeed, a wealth of information from molecular assays of individual miRNAs and variants, high throughput processing assays, and structural biology, have revealed comprehensive information on how the nuclear Microprocessor (Drosha/DGCR8) complex and cytoplasmic Dicer interrogate and cleave hairpin substrates (1). For the most part, such studies focused on how the miRNA machinery addresses each individual miRNA hairpin. General structural attributes, along with a selection of specific auxiliary motifs, enable efficient recognition and processing of pri-miRNAs by Microprocessor (26, 28, 32–34).

However, biogenesis capacities determined from in vitro substrate assays and/or individual miRNA hairpin contexts do not always reflect in vivo biogenesis capacities. One general reason involves the impact of chromatin context of cotranscriptional miRNA processing (56–60), which might differ for different miRNAs and is not modeled in widely applied massively parallel substrate processing assays. Another reason involves whether there is one or more than one miRNA hairpin on a given primary transcript (35–37). In particular, the nuclear biogenesis of certain miRNA hairpins with suboptimal features can be rescued if they reside near another strong substrate of the Microprocessor complex. In practice, this corresponds to a canonical miRNA hairpin, but experimental manipulations demonstrate that the helper does not have to be a functional miRNA substrate per se (35). This process, termed “cluster assistance”, suggests that increasing local concentration of Microprocessor can improve its action on suboptimal nuclear hairpin substrates, in a manner that involves the Microprocessor cofactors ERH or SAFB2 (35–37, 39). However, the precise role of these cofactors in clustered miRNA enhancement remains unclear. Yet another reason for the difficulty in assessing miRNA biogenesis, despite abundant available data, regards the challenges in prediction of their secondary structures. While short inverted repeats can be fairly reliably predicted as hairpins, their specific details may differ from presumed in vivo structures, in ways that can strongly impact biogenesis. For example, while mir-144 is the helper canonical miRNA for suboptimal mir-451, the biogenesis of mir-144 is itself highly regulated and inferred to adopt dynamic structure in cells that controls its dicing (40).

In this study, we conduct a number of detailed molecular assays that reveal parameters of suboptimal miRNA biogenesis. We emphasize that despite many largescale studies of miRNA biogenesis to date, which clearly define features of optimal miRNA substrates (28, 33, 61), it remains difficult to divine which substrates are functionally suboptimal, for reasons described above. Our detailed studies of a defined suboptimal miRNA substrate (mir-451), along with manipulations of other miRNA clusters, provide an empirical basis for suboptimal miRNA biogenesis and yields unexpected features of this process. Beyond characterizing how stem length and terminal hairpin loop impact the susceptibility of a miRNA hairpin to cluster assistance, we clearly uncouple miRNA biogenesis per se from cluster enhancement. That is, our data show how Microprocessor substrates that are incompetent to generate mature miRNAs can efficiently participate in cluster assistance and, in some settings, be responsive to enhancement of Drosha/DGCR8 cleavage. With other recent findings on the diversity of stable hairpins that can exist in mammalian cells, independent of small RNA biogenesis (62), these data spur greater appreciation for how miRNA processing pathways can be impacted by substrates that do not make functional small RNAs.

Complexity of Biogenesis of Clustered miRNAs and Suboptimal miRNAs.

Importantly, we also reveal complexities in the efficacy of miRNA biogenesis, which were not prospectively appreciated from current methods. First, examination of two- and three-member miRNA clusters revealed competitive effects on biogenesis. Since large miRNA operons can concentrate Microprocessor (63), it is reasonable to imagine that clustered arrangements might generally enhance miRNA biogenesis. However, we observe suboptimal miRNA enhancement does not scale with the number of canonical neighbors, and even observe competitive effects on the biogenesis of canonical miRNAs when transcribed from clustered contexts. One scenario that might explain this nonintuitive result is that the amount of Microprocessor recruited to each individual transcript might be limiting. Currently, it is not possible to visually resolve the number of Microprocessor complexes at individual miRNA loci in cells using microscopy, so further tests of this concept will continue to rely on correlative approaches.

Perhaps more nonintuitively, we use the phenomenon of cluster enhancement as a proxy to define suboptimal miRNAs by empirical means. We not only invalidate a seemingly suboptimal miRNA (mir-128-1), we provide compelling evidence that solo mir-491 truly has impaired Microprocessor substrate features. This begs the question of whether it is subject to an unknown strategy for setting-specific enhancement of its nuclear biogenesis. This certainly seems plausible given the unusually constrained terminal loop sequences of mir-491. It is recognized that a number of miRNA loci bear atypically conserved loop regions (40, 47, 64), which augur the existence of uncharacterized regulatory inputs. It is relevant to note that many biological processes do not operate constitutively or at optimal rate. Instead repression and suboptimality are critical aspects of biological regulation, ranging from miRNA biogenesis (1, 48, 50, 65), signaling (66), transcription factor binding sites (67, 68), and protein–protein interactions (69), to name a few.

On the other hand, identification of other solo suboptimal miRNAs is intriguing but it is hard to operate without reliable experiment-based global pri-miRNA structure datasets, since the predicted structures are usually inaccurate to interpret pri-miRNA structures, especially for the lower stem region. Anyway, we indeed found several solo miRNAs with suboptimal small loop (mir-320a, mir-320b-2, mir-320c-2, mir-320d-1 and mir-320d-2, SI Appendix, Fig. S7), but the processing of these miRNAs are reported Drosha-independent (70). So, it’s plausible that more suboptimal solo miRNAs might use different ways to enhance their biogenesis, for example, through special RBPs binding, using Drosha-independent pathway, or using the newly reported noncanonical microprocessing (71).

Roles of miRNA Cluster Assistance in Disease.

The phenomenon of cluster assistance seems poised to favor the evolutionary emergence of de novo miRNA hairpins within clusters, since newly born hairpins are likely to be deficient in some aspects of miRNA biogenesis, but this may be compensated by residence near an existing canonical miRNA hairpin (19, 72). Over time, miRNA hairpins that incorporate into beneficial regulatory networks might acquire features that enhance their accumulation of small RNAs and capacity to repress targets. Conversely, we may consider mutations of conserved miRNAs to run this process in reverse. We show several ways in which this process might have impact on disease situations. First, we show that point mutations that impair miRNA biogenesis can be masked, if they affect a locus within a genomic cluster. This buffering effect may help to stabilize target regulatory networks of clustered miRNAs. On the other hand, it may be beneficial for abnormal cells to inhibit certain miRNAs. For example, the related members of the mir-15a/16-1 cluster are tumor suppressors that repress several oncogenes, which accounts for their deletion or downregulation in various cancers (73). Here, we extend previous observations that a cancer point mutation outside of the miR-16 sequence (51, 52) affects it biogenesis by impairing the auxiliary CNNC motif (26). By reducing the efficacy of mir-16-1 as a Microprocessor substrate, we show that this has an additional consequence to impair biogenesis of suboptimal mir-15a.

Our data emphasize that the larger genomic context is needed to appreciate regulated miRNA biogenesis in vivo. Numerous SNPs and mutations in miRNA loci have been documented in disease and cancer (74–76). Coupled with burgeoning amounts of whole-genome sequencing data of healthy and diseased individuals (77, 78), it seems certain that additional unsuspected impacts of sequence variants on miRNA biogenesis can be revealed using the strategies outlined in this study.

Materials and Methods

Plasmid Constructs.

Plasmids for expression of all the miRNAs or miRNA clusters used in this study were constructed by inserting amplified DNA fragments containing the miRNA precursors (200 bp ~1 kb) from genomic DNA of HEK293T cells between Bgl II and Xho I sites downstream of a CMV promoter. All the miRNA mutants were constructed using overlapping PCR method based on the wild-type miRNA constructs. The luciferase plasmids containing bulge (2× or 4×) or perfect (1×) miRNA sensors were constructed by inserting annealed DNA oligonucleotides containing miRNA sensor sequences between Nhe I and Xba I (for bulge sensors) or Xho I and Xba I (for perfect sensors) sites in the 3′ UTR (untranslated region) of the firefly luciferase gene. All the details and oligonucleotide sequences used to clone these constructs are listed in SI Appendix, Table S1.

Cell Culture.

HEK293T cells were grown in DME (Dulbecco’s Modified Eagle)-high glucose media containing 10% FBS (Fetal Bovine Serum), 1% nonessential amino acids, 1% sodium pyruvate, penicillin/streptomycin, and 0.1% 2-mercaptoethanol. Mycoplasma contaminations were regularly tested for the cell lines.

Sensor Assays.

Transient cotransfections of HEK293T cells with miRNA expressing plasmids (150~200 ng/well), firefly luciferase plasmids containing miRNA sensors (15 ng/well), and control renilla luciferase plasmids were performed in 24-well cell culture plates using Lipofectamine2000 (Thermo Fisher) according to the manufacturer’s protocol. Cells were lysed 24 h posttransfection using 70 µL/well lysis buffer (PBS (Phosphate-buffered saline) with 0.2% Triton X-100), and then, 10 µL of the cell lysates were used to measure firefly and renilla luciferase activities according to the Dual-Glo luciferase assay system (Promega).

Northern Blotting.

Cotransfection of miRNA cluster plasmids (2 µg/well for 6-well plate or 1 µg/well for 12-well plate) with control mir-375 plasmid (200 ng/well for 6-well plate or 100 ng/well for 12-well plate) or renilla luciferase plasmid (30 ng/well for 12-well plate) were performed in HEK293T cells using Lipofectamine2000. Two or three days post-transfection, split 1/10 of the cells to check renilla luciferase activities as described above and total RNA was prepared from the remaining cells using Trizol reagent (Invitrogen). Equal amounts of total RNAs (10 to 15 µg) were mixed with 2× RNA loading dye, denatured at 95 °C for 5 min, and then fractionated by electrophoresis on a 20% urea polyacrylamide gel in 0.5× TBE (Tris/Borate/EDTA) buffer, until the bromophenol blue dye run out of the gel. Then, the gel was transferred to GeneScreen Plus nylon membrane (Perkin Elmer) at 300 mA for 1.5 h, UV-crosslinked with 120,000 µJ of energy, baked at 80 °C for 30 min, and then hybridized with γ-³²P-labeled DNA probes against mature miRNA sequences in hybridization buffer (5× SSC (saline-sodium citrate), 7% SDS (sodium dodecyl sulfate), 2× Denhardt’s solution) at 42 °C overnight. Wash the membrane with Non-Stringent Wash Solution (3× SSC, 5% SDS, 10× Denhardt’s solution) and then two rounds of wash with Stringent Wash Solution (1× SSC, 1% SDS). Each wash step is conducted at 42 °C for 30 min. Seal the membrane in plastic wrap and stick in the film cassette and then expose the film for 1~3 d. For reprobing of the same blot with other miRNA probes, wash the blot with 1% SDS at 80 °C for 30 min and then do hybridization. All the probe sequences are listed in SI Appendix, Table S1.

RNAi-Mediated Knockdown of Drosha.

To deplete Drosha in HEK293T cells, miRNA plasmids (1 µg) and 20 pmol of siRNAs targeting endogenous Drosha or control siRNAs were cotransfected into cells using Lipofectamine 2000 (Thermo Fisher) in 6-well plates. Cells were collected 48 h after transfection using Trizol reagent for total RNA extraction. siRNA sequences are listed in SI Appendix, Table S1.

Generation of Drosha Knockout HEK293T Cell Lines Using CRISPR/Cas9.

We cloned a gRNA plasmid by inserting annealed DNA oligonucleotides containing guide RNA sequences against the second coding exon in drosha (exon 5) into the BsmBI site of lentiCRISPRv2-GFP. We transfected 2 µg drosha-sgRNA/Cas9 plasmid into HEK293T cells in 6-well plates using Lipofectamine 2000. After 48 h, the cells were collected for cell sorting to split the GFP (Green Fluorescent Protein)-positive single cells into 96-well plates. After 2 wk of culturing, the cells were used for genotyping and then Western blotting to identify Drosha knockout colonies. Oligo sequences for sgRNA and genotyping are listed in SI Appendix, Table S1.

Western Blotting.

Wild-type and drosha-KO HEK293T cells were harvested and lysed for separation on 4 to 20% Mini-PROTEAN TGX Precast Protein Gels (Bio-Rad) and then transferred to a PVDF (polyvinylidene difluoride) membrane. The blot was probed for 2 h at room temperature with Rabbit monoclonal anti-Drosha (D28B1) antibody (Cat #3364) diluted to 1:4,000 or mouse monoclonal anti-β-tubulin antibodies (Developmental Studies Hybridoma Bank, DSHB) diluted to 1:2,000 and then incubated with a secondary antibody conjugated to horseradish peroxidase diluted to 1:5,000. The signal was detected with Amersham ECL Prime Western Blotting Detection Reagent.

RT-qPCR.

Total RNAs (1 µg) were extracted using Trizol and used for cDNA preparation by DNase I treatment and reverse transcription using SuperScript III Reverse Transcriptase (Invitrogen). qPCR reactions were performed using SYBR Select master mix (Life Technologies). Data were normalized to GAPDH amplification. Three replicates were done for qPCR. Primer sequences for qPCR are listed in SI Appendix, Table S1.

Measurement of pri-miRNA Half-Life.

To measure the half-life of pri-mir-144/451 and mutants in vivo, drosha knockout HEK293T cells were transfected with mir-144/451 and mutant plasmids. At 16 h post-transfection, 2.5 µg/mL of actinomycin D (Sigma) was added to the cells to terminate the transcription of RNA polymerase. Then, the cells were harvested at different time points (0 h, 4 h, and 8 h after adding actinomycin D) for total RNA extraction, reverse transcription, and qPCR analysis as described above.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All study data are included in the article and/or SI Appendix.