Lower and upper stem–single-stranded RNA junctionstogether determine the Drosha cleavage siteHongming Maa, Yonggan Wub, Jang-Gi Choia, and Haoquan Wua,1

aCenter of Excellence for Infectious Diseases, Department of Biomedical Sciences, Paul L. Foster School of Medicine, Texas Tech University Health SciencesCenter, El Paso, TX 79905; and bDepartment of Biological Sciences, Texas Tech University, Lubbock, TX 79409

Edited by Narry Kim, Seoul National University, Seoul, South Korea, and accepted by the Editorial Board November 7, 2013 (received for review June 18, 2013)

Microprocessor [Drosha–DGCR8 (DiGeorge syndrome critical regiongene 8) complex] processing of primary microRNA (pri-miRNA) is thecritical first step in miRNA biogenesis, but how the Drosha cleavagesite is determined has been unclear. Previous models proposed thatthe Drosha–DGCR8 complex measures either ∼22 nt from the upperstem–single-stranded RNA (ssRNA, terminal loop) junction or ∼11 ntfrom the lower stem–ssRNA junction to determine the cleavage site.Here, using miRNA-offset RNAs to determine the Drosha cleavagesite, we show that the Microprocessor measures the distances fromboth the lower and upper stem–ssRNA junctions to determine thecleavage site in human cells, and optimal distances from both struc-tures are critical to the precision of Drosha processing. If the distan-ces are not optimal, Drosha tends to cleave at multiple sites, whichcan, in turn, generate multiple 5′ isomiRs. Thus, our results also re-veal a mechanism of 5′ isomiR generation.

moR | DcRNA | alternative Drosha processing

In mammals, the canonical pathway of miRNA biogenesis isinitiated by the Drosha–DGCR8 (DiGeorge syndrome critical

region gene 8) complex (the Microprocessor), which processeslong primary miRNAs (pri-miRNAs) into ∼60-nt pre-miRNAsfor further processing by Dicer into a duplex ∼22 nt long. One orboth strands of the duplex are loaded into the RNA-inducedsilencing complex (RISC) to repress target gene expression(1, 2). Although the major players of miRNA biogenesis are mostlyknown, no model precisely predicts how the miRNA biogenesismachinery recognizes and processes a novel pri-miRNA or anmiRNA-mimicking shRNA and which strand will finally beloaded into RISC. The missing details in miRNA biogenesis aremajor obstacles to rational design of miRNA-based shRNA orartificial miRNA (amiRNA) that generates predictable results,as this kind of amiRNA must be processed by the endogenousmiRNA machinery to be functional. The lack of a precise modelof miRNA biogenesis also makes it difficult to accurately predictnovel miRNA genes in the genome. In fact, a recent study foundthat more than 150 miRNAs annotated in miRBase (of 564miRNAs checked) were not miRNAs at all (3). Thus, there isstill a need to improve our understanding of how the miRNAbiogenesis machinery recognizes and processes pri-miRNA intomature miRNA.The secondary structure of a canonical pri-miRNA usually

consists of four parts: a terminal loop, an upper stem thatencompasses the mature miRNA duplex of approximately twohelical turns, the lower stem, which is an approximate one-heli-cal-turn extension of the miRNA duplex, and the basal segments,which are single-stranded flanking sequences (4, 5). In 2005,Zeng et al. proposed that a large terminal loop is required toguide Drosha cleavage to occur at ∼22 nt from the junction ofthe terminal loop single-stranded RNA (ssRNA) and upper stem(5). In addition, they showed that single-stranded flanking sequen-ces around the miRNA hairpin structure are also required forefficient Drosha processing (6). However, in 2006, Han et al.proposed an alternative model in which the major component ofthe Microprocessor, DGCR8, recognizes the lower stem–ssRNA(basal segment) junction, guiding Drosha to cut ∼11 nt from the

junction, although they also acknowledged that the terminal loopregion might also be recognized by the Microprocessor (4). Arecent report from Zeng’s laboratory reinforced their previousconclusion that the terminal loop is critical for Microprocessorprocessing (7). Auyeung et al. recently analyzed a large numberof variants derived from four different miRNAs and identifiedthree primary sequences that are critical for Microprocessorrecognition of pri-miRNA. In addition, they showed that, al-though the terminal loop region contributes to recognition andprocessing for some pri-miRNAs, it appears to be less importantthan the lower stem region (8). However, how the lower stemand terminal loop region coordinate selection of the Droshaprocessing site is not clear.Another interesting question in miRNA biogenesis concerns

the end-heterogeneity of mature miRNAs. The end of maturemiRNAs is not absolutely fixed, but subject to variation (9–12),and we and other groups have previously discovered the patternof these variations (13, 14). Although 3′ end variation is a wide-spread phenomenon, the 5′ end of mature miRNA is highlyuniform for most miRNAs. However, in some miRNAs, such asmiR-142, the 5′ end is also highly variable (13). These 5′ isomiRsare of particular interest because positions 2–7 of the 5′ end ofmiRNAs have been called the “seed sequence,” which deter-mines the target pool of the miRNA (15). Although 5′ isomiRshave only a one- or two-nucleotide difference in their 5′ end,they have different seed sequences and different pools of targetgenes, as shown in a recent report (16). Thus, 5′ isomiRs appearto be a very efficient way for an miRNA to expand its pool oftarget genes. To understand why 5′ isomiRs occur in somemiRNAs but not in others, we previously cloned pre-miRNAsgenerated in human cells and shown that in most miRNAs that

Significance

MicroRNA genes are transcribed as long primary microRNAs(pri-miRNAs) and then processed by the Drosha–DGCR8(DiGeorge syndrome critical region gene 8) complex into ∼60-nt pre-miRNAs, which is the key step of miRNA biogenesis.How exactly the Drosha–DGCR8 complex recognizes pri-miR-NAs and selects the processing site is not well understood.Here we developed a unique approach to examining the detailsof how this complex chooses the processing site in human cells.Our study yields new knowledge of how the Drosha–DGCR8complex determines its processing sites and also provides anexplanation for how 5′ isomiRs are generated in some miRNAs,which is another interesting but not clearly understood phe-nomenon in miRNA biogenesis.

Author contributions: H.W. designed research; H.M., Y.W., J.-G.C., and H.W. performedresearch; H.W. analyzed data; and H.M. and H.W. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. N.K. is a guest editor invited by the EditorialBoard.1To whom correspondence should be addressed. E-mail: haoquan.wu@ttuhsc.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1311639110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1311639110 PNAS | December 17, 2013 | vol. 110 | no. 51 | 20687–20692

GEN

ETICS

Dow

nloa

ded

by g

uest

on

Mar

ch 2

7, 2

021

have highly uniform 5′ ends, their pri-miRNAs are processedprecisely at one site to generate one pre-miRNA, which gives riseto one or two (in the case of both strands of the duplex beingloaded into RISC) dominant mature miRNAs. By contrast, insome other miRNAs, such as miR-142, that have a high fre-quency of 5′ end variation, the pri-miRNAs are cleaved at dif-ferent sites to generate multiple distinct pre-miRNAs, which giverise to multiple 5′ isomiRs (17). Thus, 5′ isomiRs are caused byalternative Drosha processing in some miRNAs, such as miR-142. A recent report showed that 5′ isomiRs can also be gener-ated by Dicer partner protein binding in some other miRNAs,such as miR-132 (16). Why alternative Drosha processing occursonly in some miRNAs but not in others has not been clear.Here we made a series of mutations to various pri-miRNAs to

change the distance from the lower stem–ssRNA junction and/orupper stem–ssRNA junction to the original processing site anddetermined the Drosha processing site in human cells. Our datashow that Drosha cleavage is coordinated by both the lower andupper stem–ssRNA junctions. Reduced coordination by alteringthe stem length by 1 or 2 bp in either direction causes alternativeDrosha processing in which Drosha alternatively cleaves atmultiple sites and generates multiple 5′ isomiRs.

ResultsmiRNA-Offset RNA Can Be Used to Reliably Determine DroshaCleavage Sites. miRNA-offset RNAs (moRs) were first de-scribed in Ciona intestinalis and described as a class of smallRNAs ∼20 nt long and immediately adjacent to a pre-miRNA(18). It has been proposed that moRs might be the byproducts ofDrosha cleavage, and the end of the moR might indicate theDrosha processing site (18, 19). We analyzed the small RNAscloned from 293FT cells to see whether it is feasible to use moRsto determine the Drosha cleavage site. In a few miRNAs, forexample, miR-18a, the moRs were cloned in small numbers andusually precisely indicate the Drosha cleavage site. However, formost miRNAs, such as miR-16, miR-17, and miR-92a, which arethree of the most abundant miRNAs in 293FT cells, not evena single read of moR sequence was cloned, although the maturemiRNAs yielded more than 30,000 reads each (Dataset S1),which is consistent with a previous report that moRs are notabundant in human cells (19). This fact dampened our enthusi-asm for using moRs to determine Drosha cleavage sites in humancells. However, a recent finding from our study on miRNA-basedshRNAs prompted us to believe that moRs could be used toreliably identify Drosha processing sites in human cells. Pre-viously, we designed two amiRNAs in which the mature miRNAduplex sequence was replaced by an siRNA duplex, and tran-scription was driven by the strong pol III promoter U6. Theconstructs were transfected into 293FT cells and the small RNAproducts were cloned to check how exactly these amiRNAs areprocessed in human cells. Interestingly, in addition to maturesiRNA sequences, large numbers of moR sequences were alsocloned, especially 5′ moRs. The 3′ end of the 5′ moR and the 5′end of the 3′ moR were highly uniform (Datasets S2–S4). Theends could be perfectly aligned with representative matureproducts of the amiRNA, and these ends formed a perfect 2-ntoverhang, which is a typical structure generated by an RNase IIIenzyme like Drosha (Fig. S1 A and B), suggesting that the endsperfectly indicate the exact Drosha cleavage site in human cells.Encouraged by these results, we transfected pri-miR-150 and

-122 into 293FT cells, which do not express these miRNAs en-dogenously, and cloned the small RNA products. Consistent withthe amiRNA experiments, large numbers of moR sequencereads were cloned, and the ends appear to indicate the exactDrosha cleavage site (Fig. S1 C and D). In all four cases, the 3′end of the 5′ moR, the 5′ end of the 3′ moR, and the 5′ end ofthe 5′ arm mature species appear to be highly uniform andperfectly indicate the exact Drosha cleavage site (Fig. S1 A–D).

The 3′ ends of the 3′ arm mature species can be varied dra-matically, as seen in miR-122 (Fig. S1D), which is not surprising,as it is well known that the 3′ ends of mature miRNAs are underactive modification (13, 20–23). Thus, to avoid confusion, the 3′end of the 3′ arm of mature species will not be used to identifyDrosha cleavage sites in the following experiments. It is note-worthy that the 3′ end of the 5′ moRs appears to be the mostuniform end (Fig. S1 A–D), suggesting that the 3′ end of the 5′moRs provides a clearer picture of where Drosha cleaves.To ensure the accuracy of moR prediction of Drosha pro-

cessing, we also cloned the ends of the pre-miRNAs for miR-150and miR-122. Unlike mature miRNAs, which go through mul-tiple steps after Drosha processing, pre-miRNAs are the directproducts of Drosha cleavage and therefore should be better atindicating the Drosha cleavage site than mature miRNAs.However, the 3′ end of pre-miRNAs is also subject to extensivemodification (3, 17, 24); thus, we chose to clone the 5′ end ofpre-miRNAs to determine the Drosha cleavage site. Despite thehigh noise, the results are consistent with the results for moRsand mature miRNAs (pre-miR-150 and pre-miR-122 in TableS1). These data further support our hypothesis that moRs can beused to identify the Drosha cleavage site in human cells.Previously, it has been reported that two hairpin structures of

DGCR8 mRNA, hairpin A and hairpin B, can be processed bythe Drosha–DGCR8 complex in vitro, but no mature miRNAsequences were detected from these two hairpins in cells (25,26). A pre-miRNA-like product could be detected for hairpin Aby Northern blot in human cells, suggesting that the hairpincould be processed by the Drosha–DGCR8 complex in humancells (26). Thus, we chose hairpin A to determine whether moRscan be used for identifying the Drosha cleavage site in thiscontext. The construct harboring hairpin A was transfected into293FT cells, and the small RNAs were cloned. As expected, nomature miRNA products were generated from the hairpin. Mostof the reads that were cloned were 5′ moRs, whereas no 3′ moRreads were cloned (Fig. S1E). According to the previous report,the Drosha–DGCR8 complex can process hairpin A at site A1 orsite A2 in vitro; however, only one band that correlates with siteA2 has been detected in human cells, suggesting that site A2might be the Drosha cleavage site in human cells (26). Indeed,most reads cloned from the transfected hairpin A indicated thatDrosha cleaved at site A2 (Fig. S1E). The data therefore suggestthat 5′ moR sequences can be used to predict Drosha cleavagesite of mRNAs, even when they do not generate miRNAs.Thus, we will mainly use 5′ moRs in combination with 3′ moRsand the 5′ arm mature miRNAs to identify Drosha cleavagesites in this study.

Extending the Lower Stem Causes Alternative Drosha Processing. Tounderstand how the lower stem–ssRNA junction regulates Droshaprocessing, we constructed a series of mutations of the pri-miR-150with altered distances from the lower stem–ssRNA junction tothe Drosha cleavage site. We started by mutating one nucleotidein the lower stem to increase the distance between the lowerstem–ssRNA junction and the original Drosha cleavage site by1 bp (miR-150 LS+1 in Fig. 1A). According to Han et al.’s model(4), this change should shift the Drosha cleavage site toward thelower stem–ssRNA junction by 1 bp. Interestingly, this singlemutation instead caused alternative Drosha processing in whichDrosha cuts at the original site (site 1) and also at a site 2 ntupstream (site 2), judging from the 5′ moR sequences (Fig. 1Aand Table S2). Consistent with the 5′ moR results, two dominantmature miR-150 5′ isomiRs corresponding to two Drosha cleavagesites were cloned (Fig. S2A and Table S2), supporting the idea thatDrosha indeed cuts at these two sites.Extending the lower stem by 2 bp (miR-150 LS+2) also caused

alternative Drosha processing that cuts at the same sites as miR-150 LS+1, but it appeared that extending the lower stem by 2 bp

20688 | www.pnas.org/cgi/doi/10.1073/pnas.1311639110 Ma et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

7, 2

021

induced Drosha to cut at a higher frequency at site 2 (63.7%)than at site 1 (35.5%), judging from the 5′ moRs (Fig. 1B andTable S2). The mature miR-150 isomiRs corresponding to site 2(72.6%) were cloned at much higher frequency than those cor-responding to site 1 (24.4%), suggesting that Drosha indeed cutsmore at site 2 in miR-150 LS+2 (Fig. S2B and Table S2). Overall,it appears that extending the lower stem by 1 or 2 bp does notmove the Drosha cleavage site 1 or 2 nucleotides toward thelower stem–ssRNA junction but instead causes alternative Droshaprocessing in which Drosha cleaves at two sites alternatively andgenerates two 5′ isomiRs. To further confirm that the alternativeDrosha processing was caused by the distance change from the

lower stem–ssRNA junction to the original Drosha processingsite, we extended the lower stem by inserting 1 bp (miR-150 LS+A)or 2 bp (miR-150 LS+AU), as shown in Fig. 1 C and D. Althoughthe pattern was slightly different, extending the lower stem byinserting 1 or 2 bp also caused alternative Drosha processing,just like miR-150 LS+1 and miR-150 LS+2 (Fig. 1 C and D andFig. S2 C and D). These results further support the idea thatextending the lower stem causes alternative Drosha processing.The number of 3′ moR reads in this set of data did not cor-

relate well with the number of 5′ moRs and the mature miRNAdata, although the cleavage sites were mostly well correlated(Fig. 1 A–D and Fig. S2 A–D). Therefore, to confirm the Droshacleavage site, we also cloned the 5′ end of the pre-miRNA forthese mutated miRNAs. In all four mutations, two dominantpre-miRNA isoforms were cloned, and the number roughlycorrelated with the number of 5′ moR reads and the maturemiR-150 data (Fig. 1E). We also performed a Northern blot todetect pre-miRNAs generated from the mutations. As shown inFig. 1F, by comparing with WT miR-150, which has one domi-nant band, all mutations showed an extra band that is ∼4 ntlonger. Moreover, the density of bands roughly correlated withthe number of 5′ moRs and the mature miRNA data. The pre-miRNA data supported the idea that Drosha indeed cleaves atthe sites predicted from 5′ moRs and mature miRNAs to gen-erate two major pre-miRNAs. Thus, it appears that the 3′moR isnot a reliable indicator for determining the Drosha processingsite. In addition, the number of cloned 3′ moRs is generallymuch fewer than the number of 5′ moRs, which, consistent withprevious reports (18, 19), makes using 3′ moR to determine thecleavage site more vulnerable to experimental noise. We believethat this might be the reason for the inconsistency between 3′moR and 5′ moR results here and also in several cases in thefollowing experiments.To test whether extending the lower stem causes alternative

Drosha processing is a general phenomenon or occurs only inmiR-150, we repeated the experiments with miR-122. Weinserted 1 or 2 bp in the lower stem of the pri-miR-122, trans-fected the constructs into cells, and cloned the small RNAproducts. The moR results were consistent with the results formiR-150, where inserting 1 bp (miR-122 LS+U) caused Droshato alternatively cut at the original site (site 1, 40.6%) and at a site2 nt upstream (site 2, 56.1%), whereas inserting 2 bp (miR-122LS+UG) caused Drosha to cut more at site 2 (67.8%) than atsite 1 (31.1%) (Fig. 1G and H and Table S2). However, althoughthe moR results show that the Drosha cleavage site is localizedmore at site 2, very few reads of mature products were generatedfrom this cleavage site—more than 200-fold fewer reads than forthe native pre-miR-122 in both mutants (Fig. S2 E and F andTable S2). The reason could be that the pre-miRNA generatedfrom site 2 could not go through the processing steps, includingtransporting, Dicer processing, and loading, as efficiently as theoriginal pre-miR-122 generated from site 1. Thus, mature miRNAis not a reliable resource for studying Drosha processing events insome cases, and we believe that this might be the reason for theinconsistency between mature miRNA and 5′ moR results inseveral cases in the following experiments.

Altering the Upper Stem Length Also Causes Alternative DroshaProcessing. Next, we tested how the upper stem–ssRNA (termi-nal loop) junction regulates Drosha–DGCR8 complex action.The distance between the upper stem–ssRNA junction and theoriginal Drosha cleavage site was reduced by deleting 1 or 2 bp ofthe upper stem at a position close to the upper stem–ssRNAjunction in miR-150 and miR-122 (Fig. 2). Surprisingly, reducingthe distance from the upper stem–ssRNA junction to the originalDrosha cleavage site by deleting 1 or 2 bp also caused alternativeDrosha processing (Fig. 2, Fig. S3, and Table S2). It appears thatthe upper stem–ssRNA junction contributes to the control of

Fig. 1. Increasing the distance between the lower stem–ssRNA junction andthe Drosha cleavage site causes alternative Drosha processing. (A) A singlemutation from C to G that extends the lower stem by 1 bp in pri-miR-150.Mature miRNA sequences are highlighted red. The mutations are indicatedin green. Representative reads of cloned small RNAs are found in Table S2.Arrows represent the Drosha cleavage sites, as judged from the moRs. Thelength of the arrows represents roughly the frequency of moRs. The dottedline indicates the position of the original dominant Drosha cleavage site. (B)miR-150 LS+2, with mutations that extend the lower stem by 2 bp. (C) miR-150 LS+A, with insertion of 1 bp in the middle of the lower stem. (D) miR-150LS+AU, with insertion of 2 bp in the middle of the lower stem. (E) Clonedpre-miRNA 5′ ends generated from mutated pri-miR-150s. Only the domi-nant reads are presented. The raw data are found in Table S1. (F) Northernblot of pre-miRNAs generated from mutated pri-miR-150s. (G) miR-122 LS+U, with a 1-bp insertion. (H) miR-122 LS+UG, with a 2-bp insertion.

Ma et al. PNAS | December 17, 2013 | vol. 110 | no. 51 | 20689

GEN

ETICS

Dow

nloa

ded

by g

uest

on

Mar

ch 2

7, 2

021

Drosha processing in a way that is similar, but not exactly thesame, as the lower stem–ssRNA junction. Both data sets suggestthat the Drosha–DGCR8 complex measures the distance fromboth the lower stem–ssRNA and upper stem–ssRNA junction todetermine the Drosha cleavage site. Changing the distance byeven 1 or 2 bp on either side of the cleavage site causes alter-native Drosha processing.

Lower and Upper Stem–ssRNA Junctions Coordinate MicroprocessorProcessing. WT pri-miR-150 and pri-mir-122 are mostly pro-cessed precisely by the Drosha–DGCR8 complex. Small distancechanges from the lower stem–ssRNA junction or upper stem–

ssRNA junction to the original Drosha cleavage site cause im-precise Drosha cleavage. We hypothesized that precise Droshacleavage requires coordinated signals from both directions. Theimprecision caused by an altered distance in one direction mightbe corrected by changing the distance from the other direction incompensation. To test this hypothesis, we changed the distancefrom both directions simultaneously by increasing the distancefrom the lower stem–ssRNA junction by 1 or 2 bp and reducingthe distance from the upper stem–ssRNA junction by 1 or 2 bp atthe same time to see whether it would shift Drosha cleavageprecisely to another site from the original site.As shown in Fig. 3A, altering the distance in both directions by

1 bp (miR-150 LS+A/D-C) caused significantly more cleavage atsite 2 compared with changing the distance in only one direction(miR-150 LS+A or D-C; Figs. 1C, 2A, and 3A, Fig. S4A, andTable S2), suggesting that the cleavage shift toward site 2 couldbe enhanced by adding the two signals from both directions.When we inserted 2 bp in both directions at the same time (miR-150 LS+AU/D-CA), the Drosha cleavage site was preciselyshifted to site 2 (Fig. 3B, Fig. S4B, and Table S2). For miR-122,altering by only 1 bp in both directions (miR-122 LS+U/D-U) isenough to shift the Drosha cleavage site 2 bp away from theoriginal site 1 to site 2 exactly (Fig. 3C, Fig. S4C, and Table S2).Changing the distance in both directions by 2 bp (miR-122 LS+UG/D-UU) shifted the cleavage site mostly to site 2 but alsoa small fraction to site 3 (Fig. 3D, Fig. S4D, and Table S2). Thus,the imprecision caused by an altered distance in one directioncould be rescued by adjusting the distance from the other

direction in compensation, and we conclude that optimal dis-tances in both directions are required for precision of the Dro-sha–DGCR8 complex processing.

Naturally Occurring Alternative Drosha Processing Can Be Reversedby Changing the Distances of the Lower and Upper Stem–ssRNAJunctions from the Cleavage Site. miR-142 is one of the mostabundant miRNAs in T cells (13). We have previously shownthat the high frequency of 5′ isomiRs in miR-142 is caused byalternative Drosha processing (17). The pattern of naturallyoccurring alternative Drosha processing in miR-142 is similar tothe pattern of alternative Drosha processing caused by reducedcoordination between the lower and upper stem–ssRNA junc-tions by experimentally altering their distances from the Droshacleavage site in miR-150 and miR-122. We hypothesized thatnaturally occurring alternative Drosha processing might also becaused by reduced coordination between the two signals.To test this hypothesis, we made a series of mutations to the

pri-miR-142 and checked the position of the Drosha processingsite. We first checked the processing of WT pri-miR-142. Judgingfrom the moR reads, Drosha cuts at three different sites (Fig. 4Aand Table S2). Mature miR-142 reads also showed that Droshaindeed cuts at these three sites (Fig. S5A and Table S2). The datasupport our previous conclusion that the 5′ isomiR in miR-142 iscaused by alternative Drosha processing. The pattern of miR-142that we observed here in 293FT cells is consistent with ourpreviously reported observation in naïve T cells, suggesting thatthe system we used here faithfully recapitulates the Droshaprocessing of miR-142 in naïve T cells (13, 17).The majority of Drosha cleavages of the native pri-miR-142

were at site 2, which is 1 nt upstream of site 1 and 2 nt down-stream of site 3 (Fig. 4A). To induce Drosha to precisely cut atsite 1, we deleted 1 bp in the lower stem and/or inserted 1 bp inthe upper stem (Fig. 4B). The 1-bp deletion in the lower stem(miR-142 LS-A) increased Drosha cleavage at site 1 from 35% to75%, whereas the 1-bp insertion in the upper stem (miR-142 D+U) increased the cleavage to 77% (Fig. 4B and Table S2), sup-porting our previous observation that both the lower stem–

ssRNA and the upper stem–ssRNA (terminal loop) junctionregulate Drosha–DGCR8 complex action. The combination ofboth (miR-142 LS-A/D+U) induced Drosha to cut precisely at

Fig. 2. Altering the distance from the upper stem–ssRNA (terminal loop)junction to the Drosha cleavage site can also cause alternative Drosha pro-cessing. (A) miR-150 D-C, with a 1-bp deletion in the upper stem close to theterminal loop. Deleted base pairs are indicated in gray. (B) miR-150 D-CA. (C)miR-122 D-U. (D) miR-122 D-UU.

Fig. 3. The lower and upper stem–ssRNA junctions coordinate precise cleav-age of the Drosha–DGCR8 complex. (A) miR-150 LS+A/D-C, with a simulta-neous 1-bp insertion in the lower stem and a 1-bp deletion in the upper stem.(B) miR-150 LS+AU/D-CA. (C) miR-122 LS+U/D-U. (D) miR-122 LS+UG/D-UU.

20690 | www.pnas.org/cgi/doi/10.1073/pnas.1311639110 Ma et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

7, 2

021

site 1 (97%; Fig. 4B and Table S2). The mature miRNA readscorrelated well with the 5′ moR results—reads from site 1 wereincreased from 10% in WT miR-142 to 24%, 77%, and 91% inthe three mutated miR-142s (Fig. S5B and Table S2). Thus,naturally occurring alternative Drosha processing can be re-versed to cut precisely by altering the distance to the lower stemand upper stem junctions.We also tried to induce Drosha to cut precisely at site 3 by

inserting 2 bp in the lower stem and/or deleting 2 bp in the upperstem (Fig. 4C). Inserting 2 bp in the lower stem (miR-142 LS+AC) increased Drosha cleavage at site 3 from 1% to 75%,whereas deleting 2 bp in the upper stem (miR-142 D-UA) in-creased it to 41%, and the combination of both (miR-142 LS+AC/D-UA) induced Drosha to cut precisely at site 3 (97%; Fig.4C and Table S2). Mature miRNA reads correlated well with 5′moR results—reads from site 3 were increased from 35% in WTmiR-142 to 88%, 76%, and 91% in the three mutated miR-142s(Fig. S5C and Table S2). The results appear to be similar to whatwe observed in constructs with a 1-bp deletion in the lower stemand/or a 1-bp insertion in the upper stem, but in the oppositedirection. Thus, naturally occurring alternative Drosha process-ing can be reversed to cut precisely by altering the distance to thelower stem and upper stem junctions.

DiscussionIn this study, we report two main findings. First, the Drosha–DGCR8 complex can sense signals from both the lower andupper stem determine the cleavage site, and coordinated signalsfrom both are required for precise processing. Second, reduced

coordination between the lower and upper stem–ssRNA junc-tions causes alternative Drosha processing, in which Drosha al-ternatively cleaves at different sites and generates multiple 5′isomiRs, thus explaining how 5′ isomiRs are generated insome miRNAs.Currently, there is no method available to reliably determine

the Drosha processing site in human cells, which is one of themajor obstacles to elucidating the mechanism of Drosha–DGCR8recognition and processing of pri-miRNA. Our approach usingmoRs to determine the Drosha cleavage site provides a methodto study Drosha–DGCR8 complex processing in human cells.The Drosha–DGCR8 complex plays a critical role not only inmiRNA biogenesis but also in regulating mRNAs that havea hairpin structure (25–27). Unlike miRNAs, in which themature molecule provides clues to where Drosha cleaves, theprocessing of mRNA substrates that do not generate maturemiRNA is particularly difficult to study. Our approach providesa method to study Drosha–DGCR8 complex processing of thesemRNA substrates, although the name moR is not accurate asthere is no miRNA generated [Drosha cleavage–associatedRNA (DcRNA) might be a better name]. However, becausethe mechanism of how moRs or DcRNAs are generated andpreserved is not understood, exceptions may exist, in which theend of these RNAs might not faithfully represent the Droshacleavage site.How the Drosha–DGCR8 complex senses signals from both

directions and why reduced coordination between the signalscauses alternative Drosha processing are not clear. However,a recent study by Faller et al. showed that the highly cooperativebinding of DGCR8 proteins to a pri-miRNA formed a higher-order structure that is big enough to cover the pri-miRNA, in-cluding the double-stranded region and surrounding ssRNAfragments, such as the basal segments and terminal loop (28).Thus, we propose that the DGCR8 complex could cover threehelical turns of the pri-miRNA double-stranded region and alsoa small part of its surrounding ssRNA and therefore could sensesignals from both the lower and upper stem to determine theDrosha cleavage site (Fig. S6). In miRNAs with optimal signalsfrom both sides, such as in miR-150 and miR-122, the DGCR8complex assembles to form a structure that allows Drosha tocleave precisely at one site, which ensures that only one mature

Fig. 4. Naturally occurring alternative Drosha processing can be reversed byaltering the lengths of the lower stem and upper stem. Illustration of Droshacleavage sites for (A) WT miR-142 and (B and C) the indicated miR-142mutations. The dotted line indicates the position of the original dominantDrosha cleavage site (site 2).

Fig. 5. Validating the model in miR-16. Drosha cleavage sites for the in-dicated miR-16 mutations. Representative reads of cloned small RNAs arefound in Fig. S7A.

Ma et al. PNAS | December 17, 2013 | vol. 110 | no. 51 | 20691

GEN

ETICS

Dow

nloa

ded

by g

uest

on

Mar

ch 2

7, 2

021

miRNA is generated (Fig. S6A). In miRNAs, such as miR-142,the signals were evolutionarily selected not to be optimal forcleavage at a single site, so that the DGCR8 complex assemblesinto a structure that guides Drosha to cleave at multiple sites togenerate multiple 5′ isomiRs (Fig. S6B), which could efficientlyexpand the pool of target genes.Alternative Drosha processing was commonly observed in

human cells when we changed the length of the lower/upperstem, which was not observed in previous reports that weremainly based on the in vitro systems (4, 8). Because additionalregulatory factors other than DGCR8 and Drosha could affectthe Drosha processing and each miRNA could be differentiallyassociated with those regulatory factors (24, 29–32), the observeddifference might have been caused by different choices of miR-NAs. To examine this possibility, we generated serial mutationsof miR-16, which has been comprehensively analyzed in previousreports (4, 8). Similar to miR-150 and miR-122, inserting 1 or 2bp in the lower stem (miR-16 LS+A or LS+AG) caused alter-native Drosha processing (Fig. 5 and Fig. S7). However, deleting1 or 2 bp in the upper stem (miR-16 D-G or D-UG) did notchange the Drosha processing site (Fig. 5 and Fig. S7), sug-gesting that the upper stem–ssRNA junction of miR-16 mightnot contribute to Drosha processing, which is consistent withprevious reports (4, 8). However, altering the stem length by 1 bpin both directions (miR-16 LS+A/D-G) induces Drosha to cutmore at an alternative site that is 1 bp upstream of the originalsite (from 7.6% to 27.5%) compared with inserting 1 bp in thelower stem only (miR-16 LS+A), and altering the stem length by2 bp in both directions (miR-16 LS+AG/D-UG) induced Droshato cut precisely at a site 2 bp upstream of the original sitecompared with inserting 2 bp in the lower stem only (miR-16LS+AG, with only 26.5% cleavage at the site; Fig. 5 and Fig. S7),suggesting that the upper stem of miR-16 can contribute to theselection of the Drosha processing site and the lower and upperstem–ssRNA junctions coordinate selection of the Drosha pro-cessing site. These data also suggested that, whereas optimaldistances on both sides are generally required for precise Droshaprocessing, alteration of the distance on one side might be

tolerated in some miRNAs, such as miR-16. Thus, although themodel we proposed might have broad generality, it is possiblethat there might be exceptions.Our results concerning Drosha–DGCR8 complex recognition

and processing of pri-miRNA could provide clues for designingbetter miRNA-based shRNA. It is clear that both distances(from the lower stem and upper stem–ssRNA junction to theDrosha cleavage site) should be optimally maintained to ensureprecise processing of shRNA so that the desired siRNA duplex isgenerated. When the potency of shRNA is in doubt, it is alsopossible to manipulate the distance between the lower and upperstem–ssRNA junctions to purposefully make Drosha cut at dif-ferent sites so that multiple siRNA duplexes are generated,which could double or triple the chance to obtain a high-potencyshRNA, as even a 1-bp change at the end of the siRNA duplexcan dramatically alter siRNA functionality. Additional discussionis available in SI Discussion.

Materials and MethodsSmall RNAs were cloned partly as described previously (13, 33). Here wemademajor modifications to improve adaptor ligation efficiency and therebyminimize cloning bias. Briefly, the small RNAs were purified with the miR-Neasy kit (Qiagen). Small RNA (50 ng) was ligated with 3′ and 5′ linkers (withbarcode) with an improved ligation method. The ligated small RNAs werereverse-transcribed and amplified with the KAPA library amplification kit(KAPA Biosystems) for 10 cycles, and the library was sequenced using theIllumina HiSeq2000 or MiSeq. The deep sequencing data were processed byin-house–developed software. All reads that had been cloned only oncewere discarded to lower the noise level. All reads of cloned small RNAs forindicated miRNAs are included in Dataset S5. Note that for small RNAs to beefficiently cloned with the method described here, the 5′ phosphate and 3′hydroxyl groups must be preserved. Additional methods are available in SIMaterials and Methods.

ACKNOWLEDGMENTS. We thank Drs. Manjunath Swamy and PremlataShankar for discussion and critical reading of the manuscript. We thankJessica Montoya and Junli Zhang for technical support and Dr. DianneMitchell and the staff of the genomics core facility at Texas Tech UniversityHealth Science Center at El Paso for their work on DNA sequencing.

1. Ambros V (2004) The functions of animal microRNAs. Nature 431(7006):350–355.2. Bartel DP (2004) MicroRNAs: Genomics, biogenesis, mechanism, and function. Cell

116(2):281–297.3. Chiang HR, et al. (2010) Mammalian microRNAs: Experimental evaluation of novel

and previously annotated genes. Genes Dev 24(10):992–1009.4. Han J, et al. (2006) Molecular basis for the recognition of primary microRNAs by the

Drosha-DGCR8 complex. Cell 125(5):887–901.5. Zeng Y, Yi R, Cullen BR (2005) Recognition and cleavage of primary microRNA pre-

cursors by the nuclear processing enzyme Drosha. EMBO J 24(1):138–148.6. Zeng Y, Cullen BR (2005) Efficient processing of primary microRNA hairpins by Drosha

requires flanking nonstructured RNA sequences. J Biol Chem 280(30):27595–27603.7. Zhang X, Zeng Y (2010) The terminal loop region controls microRNA processing by

Drosha and Dicer. Nucleic Acids Res 38(21):7689–7697.8. Auyeung VC, Ulitsky I, McGeary SE, Bartel DP (2013) Beyond secondary structure:

Primary-sequence determinants license pri-miRNA hairpins for processing. Cell 152(4):844–858.

9. Ruby JG, et al. (2006) Large-scale sequencing reveals 21U-RNAs and additional mi-croRNAs and endogenous siRNAs in C. elegans. Cell 127(6):1193–1207.

10. Lim LP, et al. (2003) The microRNAs of Caenorhabditis elegans. Genes Dev 17(8):991–1008.

11. Lagos-Quintana M, Rauhut R, Meyer J, Borkhardt A, Tuschl T (2003) New microRNAsfrom mouse and human. RNA 9(2):175–179.

12. Lau NC, Lim LP, Weinstein EG, Bartel DP (2001) An abundant class of tiny RNAs withprobable regulatory roles in Caenorhabditis elegans. Science 294(5543):858–862.

13. Wu H, et al. (2007) miRNA profiling of naïve, effector and memory CD8 T cells. PLoSONE 2(10):e1020.

14. Ruby JG, et al. (2007) Evolution, biogenesis, expression, and target predictions ofa substantially expanded set of Drosophila microRNAs. Genome Res 17(12):1850–1864.

15. Lewis BP, Burge CB, Bartel DP (2005) Conserved seed pairing, often flanked by adenosines,indicates that thousands of human genes are microRNA targets. Cell 120(1):15–20.

16. Fukunaga R, et al. (2012) Dicer partner proteins tune the length of mature miRNAs inflies and mammals. Cell 151(3):533–546.

17. Wu H, Ye C, Ramirez D, Manjunath N (2009) Alternative processing of primary mi-croRNA transcripts by Drosha generates 5′ end variation of mature microRNA. PLoSONE 4(10):e7566.

18. Shi W, Hendrix D, Levine M, Haley B (2009) A distinct class of small RNAs arises frompre-miRNA-proximal regions in a simple chordate. Nat Struct Mol Biol 16(2):183–189.

19. Berezikov E, et al. (2011) Deep annotation of Drosophila melanogaster microRNAsyields insights into their processing, modification, and emergence. Genome Res 21(2):203–215.

20. Liu N, et al. (2011) The exoribonuclease Nibbler controls 3′ end processing of micro-RNAs in Drosophila. Curr Biol 21(22):1888–1893.

21. Han BW, Hung JH, Weng Z, Zamore PD, Ameres SL (2011) The 3′-to-5′ exoribonucleaseNibbler shapes the 3′ ends of microRNAs bound to Drosophila Argonaute1. Curr Biol21(22):1878–1887.

22. Ameres SL, et al. (2010) Target RNA-directed trimming and tailing of small silencingRNAs. Science 328(5985):1534–1539.

23. Katoh T, et al. (2009) Selective stabilization of mammalian microRNAs by 3′ adenylationmediated by the cytoplasmic poly(A) polymerase GLD-2. Genes Dev 23(4):433–438.

24. Heo I, et al. (2008) Lin28 mediates the terminal uridylation of let-7 precursor Micro-RNA. Mol Cell 32(2):276–284.

25. Triboulet R, Chang HM, Lapierre RJ, Gregory RI (2009) Post-transcriptional control ofDGCR8 expression by the Microprocessor. RNA 15(6):1005–1011.

26. Han J, et al. (2009) Posttranscriptional crossregulation between Drosha and DGCR8.Cell 136(1):75–84.

27. Kadener S, et al. (2009) Genome-wide identification of targets of the drosha-pasha/DGCR8 complex. RNA 15(4):537–545.

28. Faller M, et al. (2010) DGCR8 recognizes primary transcripts of microRNAs throughhighly cooperative binding and formation of higher-order structures. RNA 16(8):1570–1583.

29. Piskounova E, et al. (2011) Lin28A and Lin28B inhibit let-7 microRNA biogenesis bydistinct mechanisms. Cell 147(5):1066–1079.

30. Michlewski G, Cáceres JF (2010) Antagonistic role of hnRNP A1 and KSRP in theregulation of let-7a biogenesis. Nat Struct Mol Biol 17(8):1011–1018.

31. Suzuki HI, et al. (2009) Modulation of microRNA processing by p53. Nature 460(7254):529–533.

32. Viswanathan SR, Daley GQ, Gregory RI (2008) Selective blockade of microRNA pro-cessing by Lin28. Science 320(5872):97–100.

33. Neilson JR, Zheng GX, Burge CB, Sharp PA (2007) Dynamic regulation of miRNA ex-pression in ordered stages of cellular development. Genes Dev 21(5):578–589.

20692 | www.pnas.org/cgi/doi/10.1073/pnas.1311639110 Ma et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

7, 2

021