TUT7 controls the fate of precursor microRNAs by using three different uridylation mechanisms

Terminal uridylyl transferases (TUTs) function as integral regulators of microRNA (miRNA) biogenesis. Using biochemistry, single-molecule, and deep sequencing techniques, we here investigate the mechanism by which human TUT7 (also known as ZCCHC6) recognizes and uridylates precursor miRNAs (pre-miRNAs) in the absence of Lin28. We find that the overhang of a pre-miRNA is the key structural element that is recognized by TUT7 and its paralogues, TUT4 (ZCCHC11) and TUT2 (GLD2/PAPD4). For group II pre-miRNAs, which have a 1-nt 3′ overhang, TUT7 restores the canonical end structure (2-nt 3′ overhang) through mono-uridylation, thereby promoting miRNA biogenesis. For pre-miRNAs where the 3′ end is further recessed into the stem (as in 3′ trimmed pre-miRNAs), TUT7 generates an oligo-U tail that leads to degradation. In contrast to Lin28-stimulated oligo-uridylation, which is processive, a distributive mode is employed by TUT7 for both mono- and oligo-uridylation in the absence of Lin28. The overhang length dictates the frequency (but not duration) of the TUT7-RNA interaction, thus explaining how TUT7 differentiates pre-miRNA species with different overhangs. Our study reveals dual roles and mechanisms of uridylation in repair and removal of defective pre-miRNAs.

In addition to the canonical miRNA biogenesis pathway, noncanonical cleavage of pre-miRNA has been reported (Yang & Lai, 2011;. Pre-miRNAs are often heterogeneous at their 3 0 ends, indicating that they are cleaved or trimmed after Drosha processing (Newman et al, 2011;Burroughs et al, 2012;Heo et al, 2012;Li et al, 2013). Ago2 contributes to the production of truncated species by cleaving pre-miRNAs in the middle of the 3 0 strand. This results in truncated hairpins called Ago-cleaved pre-miRNAs (ac-pre-miRNAs) (Diederichs & Haber, 2007). There is little evidence that ac-pre-miRNAs generate mature miRNAs with an exception of ac-pre-miR-451 that is shorter than others and trimmed further into mature miRNA Cifuentes et al, 2010;Yang et al, 2010;Yoda et al, 2013). Thus, it remains unclear whether acpre-miRNAs have a certain biological role in general or whether they are mostly degradation intermediates. Additional nucleases have been reported to cleave pre-miRNAs (Suzuki et al, 2011;Upton et al, 2012;Asada et al, 2014). However, the molecular mechanism of how the truncated pre-miRNAs are removed is largely unknown.
In this study, we delineate the molecular mechanism of uridylation of pre-miRNAs with various structures. By mapping out the interactions between TUT7 and pre-miRNA, we show that the overhang of a pre-miRNA is the key structural element that TUT7 recognizes. Sensing the overhang structure, TUT7 preferentially uridylates 3 0 truncated pre-miRNAs as well as group II pre-miRNA. Uridylation leads to two opposing consequences. Mono-uridylation of intact group II pre-let-7s (with a 1-nt 3 0 overhang) restores functional pre-miRNAs (with a 2-nt 3 0 overhang). On the contrary, recognition of pre-miRNAs with 5 0 overhang (ac-pre-miRNA or trimmed decay intermediates) leads to oligo-uridylation and RNA degradation. Our single-molecule study further reveals that TUT7 employs a distributive mode for both uridylation pathways and that TUT7 discriminates its substrates by interacting with them at different frequencies.

TUT7 domains required for mono-uridylation
To map the interactions between TUTs and pre-miRNA, we first sought to identify the minimal domains used for pre-miRNA recognition and mono-uridylation. TUTs share a common catalytic motif consisting of a nucleotidyl transferase (Ntr) and the poly(A) polymerase-associated (PAP) domain (Kwak & Wickens, 2007;Martin & Keller, 2007). The Ntr domain contains three catalytic aspartates, whereas the PAP domain provides nucleotide specificity through its contact with the base in the active site (Martin & Keller, 2007;Lunde et al, 2012;Munoz-Tello et al, 2012;Yates et al, 2012). While TUT2 has only one catalytic motif, TUT7 and TUT4 (TUT7/4) have a duplication of the catalytic motif at their N-terminus although it is inactive due to the lack of one of the catalytic aspartates ( Supplementary Fig S1A). Additionally, TUT7/4 possesses a CCHH zinc finger motif at their N-terminus and three CCHC zinc finger motifs around the catalytic motif.
We focused on TUT7 as it is the major enzyme for pre-let-7 mono-uridylation (Heo et al, 2012). We generated three deletion mutants of TUT7 by deleting domains from the N-terminus of TUT7 (DZF1, DNtr1, and DPAP1) (Fig 1A). In addition, we produced Ntr2-PAP2 (NP) mutant that consists of only the active catalytic motif. The truncated proteins were immunopurified and incubated with unmodified pre-let-7a-1 (with a 1-nt 3 0 overhang) or its monouridylated counterpart (+U, with a 2-nt 3 0 overhang) ( Fig 1B). The DZF1, DNtr1, and DPAP1 mutants mono-uridylated unmodified prelet-7a-1 selectively and as efficiently as the full-length (FL) TUT7. However, NP mutant did not show any detectable activity in spite of its higher expression level than that of full-length TUT7 (Fig 1B and Supplementary Fig S1B). It seems that NP mutant cannot uridylate RNA substrates, possibly because surrounding regions of catalytic motif are required to bind RNA and/or to maintain proper protein structure. These results indicate that the N-terminal half of TUT7 is dispensable while the C-terminal domains including the catalytic motif are required for pre-miRNA mono-uridylation. As the C-terminal half of TUT7 (DPAP1) is fully active, we generated recombinant TUT7 (rTUT7) encompassing 951-1,495 a.a ( Fig 1C)  . In vitro uridylation assays demonstrated that rTUT7 can mono-uridylate pre-let-7a-1 and that it has the same substrate preference as the immunopurified full-length TUT7 does ( Fig 1D). This suggests that the residues 951-1,495 are sufficient for monouridylation activity by TUT7. Note that prolonged incubation leads to 'oligo lanes 4 and 8) due to multiple rounds of distributive 'mono'-uridylation (see Discussion for further explanation). Given that the recombinant protein was produced in E. coli and purified to homogeneity, our result indicates that TUT7 does not require any additional cofactors for pre-miRNA monouridylation (Fig 1B and D).

RNA motifs that are recognized by TUT7
To investigate which parts of pre-miRNA are recognized by TUT7, we generated mutants of pre-let-7a-1, a pre-miRNA that belongs to group II. Pre-let-7a-1 is divided up into three parts; a 27-nt terminal loop (green), a 21-bp base-paired stem (black), and a 1-nt 3 0 overhang (red) (Fig 2A, left). First, we designed a terminal loop mutant (L4) by reducing the loop size from 27 nt to 4 nt (Fig 2A, center). Immunopurified full-length TUT7 failed to uridylate the L4 mutant efficiently, which suggested that TUT7 recognizes the terminal loop for mono-uridylation (Fig 2A, lanes 1-4). Next, to test whether a stem of a certain length is necessary, we generated a stem mutant (S14) by shortening the stem from 21 bp to 14 bp (Fig 2A, right). The S14 mutant was uridylated as efficiently as the unmodified prelet-7a-1, indicating that the overall length of the stem is not critical for pre-miRNA mono-uridylation by TUT7 (Fig 2A, lanes 1-2  and 5-6).
To find out how the overhang structure influences the monouridylation activity of TUT7, we designed six overhang variants of pre-let-7a-1 by shortening nucleotides from the 3 0 end (Fig 2B, top).
We included ac-pre-let-7a-1 (Ac-pre) with a 10-nt 5 0 overhang, which is known to be uridylated in humans and mice (

Domain organization
In vitro uridylation by TUT7

Domain organization
In vitro uridylation by rTUT7 Figure 1. C-terminal half of TUT7 is sufficient to mono-uridylate pre-let-7a-1 specifically.
Deletion mutants except for NP showed mono-uridylation activity and the same substrate preference as that of full-length TUT7; they mono-uridylate unmodified pre-let-7a-1 with a 1-nt 3 0 overhang more efficiently than pre-let-7a-1 +U with a 2-nt 3 0 overhang. NP mutant lost its uridylation activity. C Domain organization of recombinant TUT7 951-1,495 (rTUT7). D In vitro uridylation of unmodified pre-let-7a-1 and +U by rTUT7. rTUT7 exhibited mono-uridylation activity and the same substrate preference as full-length TUT7.
Source data are available online for this figure.
In vitro uridylation of pre-let-7a-1 by TUT7 In vitro uridylation of pre-let-7a-1 overhang variants by TUT7  A Top: Structure of unmodified pre-let-7a-1, terminal loop mutant (L4), and stem mutant (S14). Green, terminal loop; black, stem; red, overhang. Asterisks mark radiolabeled terminal phosphates. Red arrows indicate the length of the terminal loop or the stem of pre-let-7a-1 L4 or S14, respectively. Bottom: In vitro uridylation of unmodified pre-let-7a-1 and mutants by immunopurified full-length TUT7 (13 min reaction). Ablating the terminal loop (L4) dramatically reduced the monouridylation efficiency, while shortening the stem (S14) did not affect the mono-uridylation efficiency. B Top: Structure of unmodified pre-let-7a-1 and six overhang variants. Red, overhang. Asterisks mark radio-labeled terminal phosphates, and red arrows indicate the net length of overhang. Bottom: In vitro uridylation of unmodified pre-let-7a-1 and overhang variants by immunopurified full-length TUT7 (15 min reaction). TUT7 uridylated blunt end or 5 0 overhang variants efficiently and showed enhanced activity to the long 5 0 overhang variants (DCUUUC and Ac-pre). The average length of U-tail is shown below each band. See Materials and Methods for quantification method.
Source data are available online for this figure.
The EMBO Journal Vol 34 | No 13 | 2015 ª 2015 The Authors The EMBO Journal Precursor microRNA uridylation by TUT7 Boseon Kim et al 2013). The substrates with a blunt end or a 5 0 overhang were uridylated with comparable efficiency to (if not more efficiently than) unmodified pre-let-7a-1 by immunopurified full-length TUT7 ( Fig 2B, bottom). To our surprise, RNAs containing a long 5 0 overhang (DCUUUC and Ac-pre) were strongly oligo-uridylated. Similar results were obtained with rTUT7 951-1,495 (Supplementary Fig  S2A and B). These data indicate that TUT7 acts efficiently on 3 0 truncated pre-miRNAs in the absence of any cofactor. As TUT7/4/2 can act redundantly to mono-uridylate group II pre-let-7s (Heo et al, 2012), we also performed in vitro uridylation using immunopurified full-length TUT4 and TUT2 (TUT4/2) to compare their substrate preferences. By and large, TUT7/4/2 are highly similar to each other in specificity but they also displayed some distinct characteristics ( Fig S2C). Moreover, while TUT7 and TUT4 showed a strong oligo-uridylation activity on pre-miRNAs with a long 5 0 overhang (pre-let-7a-1 DCUUUC and Ac-pre), TUT2 did not show such activity ( Fig 2B and Supplementary Fig S2D). Taken together, the primary cis-acting element recognized commonly by TUT7/4/2 is the overhang structure of pre-miRNA.

Differentiation of pre-miRNAs at the binding step
To further investigate the molecular mechanism by which TUT7 recognizes structural elements of its substrates, we employed singlemolecule fluorescence spectroscopy. For long-term single-molecule observations, rTUT7 fused to a 6×-His tag was immobilized on a PEGylated quartz surface using anti-His tag antibodies ( Fig 3A). Fluorescently labeled pre-let-7a-1 molecules were introduced to the microfluidic chamber, and the interactions between TUT7 and RNAs were monitored in real time. As shown in representative time traces (Fig 3B and C), the interaction of rTUT7 with RNA molecules was marked with a sudden increase and subsequent rapid decrease in the fluorescence intensity. This brief interaction suggests that uridylation by TUT7 is distributive. This is similar to Lin28-independent mono-uridylation by TUT4, which we previously reported to be distributive (Yeom et al, 2011). Control experiments showed that neither dye-labeling of pre-let-7a-1 nor immobilization of rTUT7 affected its uridylation efficiency (unpublished observations). Using this experimental system, we questioned at which kinetic step TUT7 discriminates between different RNA substrates. We first determined the dissociation rate (k off ) by analyzing the dwell time of interaction (Ds, the inverse of k off ) between TUT7 and unmodified pre-let-7a-1. The dwell time distribution from a total of 8,943 binding events followed a single-exponential decay with a time scale of 0.36 AE 0.03 s (Supplementary Fig S3A). This indicates that dissociation of an RNA substrate from TUT7 is a single-step process. To gain more insights into the molecular mechanism of TUT7, we repeated this measurement for mono-uridylated pre-let-7a-1 (+U) and the terminal loop mutant (L4). Intriguingly, the dwell time of the +U substrate (0.35 AE 0.05 s, 17,003 events; Fig 3D) was similar to that of unmodified substrate within an error, whereas that of the terminal loop mutant showed a slight increase (0.54 AE 0.03 s, 8,012 events; Fig 3D). These results suggest that the binding strength between TUT7 and RNA ( Supplementary Fig S3B) is not a dominant factor in distinguishing between different RNA substrates, although the terminal loop might play a role in the release of the substrate.
Next, we asked whether the binding rate (k on ) might govern the substrate preference of TUT7. For this measurement, unmodified pre-let-7a-1 and a variant (e.g. +U) were labeled with spectrally separated fluorescent dyes (Cy3 and Cy5, respectively) and introduced together into a microfluidic chamber. Unmodified pre-let-7a-1 served as a reference. By monitoring the interactions between immobilized rTUT7 and two RNA substrates simultaneously, we were able to compare k on of a variant to that of unmodified pre-let-7a-1 (Fig 3A and C). This frequency measurement revealed that the mono-uridylated substrate (+U) binds less frequently than unmodified substrate does (k +U on /k Unmodified on = 0.44 AE 0.07, Fig 3E), which indicates that the addition of a single uridine suppresses the TUT7-RNA interaction. Stronger suppression was observed with the terminal loop mutant (L4) (k L4 on /k Unmodified on = 0.24 AE 0.01). This is consistent with the decrease in uridylation efficiency observed in Fig 2A. Taken together, TUT7 may discriminate between the substrates during the binding step rather than after binding to RNA (Supplementary Fig S3B).
Our biochemical study ( Fig 2B) indicated that TUT7 oligouridylates truncated pre-let-7a-1 effectively under the condition where unmodified pre-let-7a-1 is mono-uridylated. We questioned whether uridylation is changed from a distributive to a processive mode in the presence of the 5 0 overhang, or whether the 5 0 overhang increases the frequency of the distributive interaction. We repeated A Schematic overview of the single-molecule assay. Recombinant TUT7 951-1,495 (rTUT7) fused to a 6×-His was immobilized on a PEGylated surface using anti-His tag antibodies. Afterward, fluorescently labeled RNA substrates were introduced into the chamber. B Representative time trajectory for the dwell time analysis. Ds, dwell time of interaction; k on , binding rate. C Representative time trajectory for binding frequency measurements. k on , binding rate. D Average dwell time of pre-let-7a-1 +U and L4 mutants (n = 3). Pre-let-7a-1 +U showed similar dwell time to unmodified pre-let-7a-1. Pre-let-7a-1 L4 yielded slightly increased dwell time (P-value 9.0 × 10 À3 , two-tailed t-test). Error bars represent standard error. E Binding frequency of pre-let-7 +U and L4 mutants relative to unmodified pre-let-7a-1 (n = 3). Pre-let-7a-1 +U and L4 mutants showed much lower binding frequency compared to unmodified pre-let-7a-1 (P-values 3.7 × 10 À3 and 7.8 × 10 À8 , respectively, two-tailed t-test). Error bars represent standard error. F Average dwell time of pre-let-7a-1 DCUUUC and Ac-pre mutants (n = 3). Pre-let-7a-1 DCUUUC and Ac-pre had similar dwell times compared to unmodified pre-let-7a-1. Error bars represent standard error. G Binding frequency of pre-let-7 DCUUUC and Ac-pre mutants relative to unmodified pre-let-7a-1 (n = 3). Pre-let-7a-1 DCUUUC and Ac-pre mutants displayed much higher binding frequency compared to unmodified pre-let-7a-1 (P-values 2.4 × 10 À3 and 4.0 × 10 À5 , respectively, two-tailed t-test). Error bars represent standard error.
Data information: All data sets (D-G) are normally distributed (Shapiro-Wilk test, P > 0.1). All the datasets of the binding frequency measurements showed equality in variance (F-test). For the dwell time measurements, some data sets did not show equality in variance (+U and -CUUUC) in F-test, and we have adjusted our two-tailed t-tests accordingly. The EMBO Journal our single-molecule kinetic measurements for two pre-let-7a-1 mutants with different 5 0 overhang (DCUUUC with 4-nt 5 0 overhang and Ac-pre with 10-nt 5 0 overhang). Intriguingly, the dwell times (Ds = 1/k off ) of pre-let-7a-1 DCUUUC and Ac-pre were within the error comparable to that of unmodified pre-let-7a-1 (DCUUUC, 0.37 AE 0.02 s, 3,495 events; Ac-pre, 0.38 AE 0.03 s, 10,308 events; Fig 3F). This suggests that the uridylation mode remains distributive ( Fig 3F). Thus, in the absence of the processivity factor Lin28, TUT7 exclusively uses a distributive mode for both mono-uridylation and oligo-uridylation. We next assessed k on and compared the binding frequency of these RNAs with that of unmodified pre-let-7a-1. Indeed, TUT7 interacted The EMBO Journal Precursor microRNA uridylation by TUT7 Boseon Kim et al with DCUUUC and Ac-pre substrates with a higher frequency than with the unmodified substrate (k DCUUUC on /k Unmodified on = 3.33 AE 1.01, k Ac-pre on /k Unmodified on = 5.91 AE 1.00; Fig 3G). Moreover, binding frequency increased as the length of 5 0 overhang got longer (from 4 nt to 10 nt). These results collectively hint that TUT7 distinguishes between its substrates at the binding step.
We further validated the distributive mode of TUT by an in vitro uridylation assay with dilution. The reaction mixture with recombinant TUT4 267-1,312 (rTUT4, Supplementary Fig S4A and B) and RNA (unmodified pre-let-7a-1 or ac-pre-let-7a-1) was either not diluted or diluted four times with reaction buffer after 20 s, which should lower the uridylation efficiency in case of distributive uridylation. As a control for processive reaction, uridylation assay with rTUT4 and rLin28b was included. Of note, we used rTUT4 (267-1,312) instead of rTUT7 (951-1,495) because rTUT4 interacts with Lin28, while rTUT7 lacks the first zinc finger motif known to mediate the interaction with Lin28 (Thornton et al, 2012). We failed to produce soluble full-length rTUT7 protein. In the presence of Lin28b, rTUT4 oligo-uridylated unmodified pre-let-7a-1, which was not affected by dilution ( Supplementary Fig S4C,  lanes 7-8). This is consistent with our previous single-molecule data that Lin28-mediated oligo-uridylation is a processive reaction (Yeom et al, 2011). In contrast, oligo-uridylation of ac-pre-let-7a-1 ( Supplementary Fig S4C, lanes 1-3) and mono-uridylation of unmodified pre-let-7a-1 (Supplementary Fig S4C, lanes 4-6) was strongly reduced after dilution (Supplementary Fig S4C, lanes  1-6). This result suggests that TUTases act distributively in the absence of Lin28, supporting our conclusion from single-molecule measurements (Fig 3).

Uridylation of 3 0 trimmed pre-miRNAs in cells
It is interesting that TUT7 is capable of oligo-uridylating pre-miRNAs with a blunt end or 5 0 overhangs in vitro (Fig 2B). To investigate whether the 3 0 truncated pre-miRNAs are indeed uridylated by TUT7 in cells, we carried out pre-miRNA deep sequencing in HeLa cells with or without TUT7/4/2 knockdown (Fig 4A and  Supplementary Fig S5). We depleted TUT7, TUT4, and TUT2 simultaneously due to their redundant activities (Heo et al, 2012).
Pre-miRNA library was constructed by size fractionation, 3 0 adapter ligation, reverse transcription followed by PCR using primers specific to pre-miRNAs (Fig 4A). We selected 55 pre-miRNAs whose miRNAs are abundant in HeLa cells and/or those reported to produce ac-pre-miRNAs (Diederichs & Haber, 2007;Burroughs et al, 2012;Li et al, 2013) (Supplementary Table S1).
Knockdown of TUT7/4/2 resulted in a decrease in uridylation in the vast majority of pre-miRNAs, suggesting that TUT7/4/2 can uridylate most pre-miRNAs to some degrees (Fig 4B and Supplementary Table S2). Adenylation was not affected significantly, confirming that TUT7/4/2 work mainly as uridylyl transferases on pre-miRNAs (Fig 4B and Supplementary Table S2). Note that any terminal residue matching genomic sequence was considered as templated, so the modification rates are underestimated. To observe the effect of TUT7/4/2 knockdown, we first investigated monouridylation pattern of group I and group II pre-let-7s (Fig 4C). Consistent with the previous study (Heo et al, 2012), a significant portion (36.2%) of group II pre-let-7s is mono-uridylated in control cells, while mono-uridylation decreased to 7.0% upon TUT7/4/2 knockdown. On the other hand, a proportion of group I pre-let-7s was rarely mono-uridylated (1.08%) in the control HeLa cells (Fig 4C). Notably, trimmed pre-miRNAs accumulated more than 2-fold in both group I and group II pre-let-7s when TUT7/4/2 are depleted, suggesting that TUT7/4/2 may act to facilitate removal of trimmed pre-miRNAs.
To observe trimming and uridylation pattern of pre-let-7s in detail, we drew dot plots which show the fractions of pre-miRNAs that were trimmed of specific length (x-axis) and gained a U-tail of certain size (y-axis) (Fig 4D). Although the majority of pre-let-7a-1 (group II) was mono-uridylated in control cells, uridylation reduced dramatically upon TUT7/4/2 knockdown (Fig 4D, right). We also observed that shorter pre-let-7a-1 species increased upon TUT7/4/2 knockdown. Pre-let-7e (group I) was not strongly affected by TUT7/4/2, yet we detected a modest decrease in uridylation and an accumulation of trimmed pre-let-7e (Fig 4D,  left).
Next, we analyzed 54 pre-miRNAs that yielded sufficient reads for analysis (> 400 total reads), including pre-let-7s. For most pre-miRNAs, a substantial fraction of reads corresponded to the 3 0 A Scheme of pre-miRNA deep sequencing. B Box plot of uridylation ratio and adenylation ratio of 78 pre-miRNAs, which are sequenced in both control sample and TUT7/4/2 knockdown cells. Dark red line indicates the median. TUT7/4/2 knockdown suppressed uridylation specifically. C Mono-uridylation of group I (let-7a-2, let-7c, and let-7e) and group II (let-7a-1, let-7a-3, let-7b, let-7d, let-7f-1, let-7f-2, let-7g, let-7i, and miR-98) pre-let-7s in control and TUT7/4/2 knockdown (TUTKD) HeLa cells. A considerable portion of intact group II pre-let-7s was mono-uridylated ("Mono-U") in control. Upon TUT7/4/2 knockdown, mono-uridylation of intact group II pre-let-7s substantially decreased. The reads whose length of trimming is > 0 were defined as "Trimmed". The reads whose length of trimming is = 0 with none non-templated addition were defined as "Intact". The reads whose length of trimming is = 0 with one non-templated uridine were defined as "Mono-U". The rest reads were defined as "Other". The percentage was calculated by normalizing with total read. D Dot plots of pre-let-7e (group I) and pre-let-7a-1 (group II). The status of 3 0 trimming and 3 0 U-tailing for each pre-miRNA read is represented by a circle on a twodimensional matrix. The x-axis represents the length of 3 0 trimming, and the y-axis represents the length of U-tail. Area of a circle is proportional to the relative abundance of the pre-miRNA reads. Reference sequence of the hairpin is shown below the dot plot. Position 0 indicates the 3 0 end of genomic sequence of most abundant read in control HeLa cells. Pre-let-7e was rarely mono-uridylated in control and was not affected by TUT7/4/2 knockdown. In contrast, about half of pre-let-7a-1 was monouridylated in control, and the mono-uridylated reads almost disappeared upon TUT7/4/2 knockdown. The style of the figure was adopted from Zhao et al (2012). E Box plot of trimmed ratio and uridylation ratio of trimmed species of 54 pre-miRNAs, which have more than 400 reads. Dark red line indicates the median.

G G C C U C C U A G C U U U C C C C A G G
(1) Lin28-dependent oligo-uridylation of pre-let-7 (Processive mode) Ago2 or unknown nuclease A 1-nt 3 overhang 5 overhang or blunt Overhang Essential cis-acting elements for TUT7-mediated mono-uridylation
(1) In embryonic cells and certain cancer cells, Lin28 recruits TUT4 (and TUT7, to a lesser extent) to pre-let-7 for oligo-uridylation. This reaction is processive. Oligo-uridylated pre-let-7 cannot be processed to mature miRNA and is instead degraded by Dis3l2.
(2) Group II pre-miRNA with a 1-nt 3 0 overhang is mono-uridylated by TUT7/4/2, which creates an optimal end structure for Dicer-mediated processing. TUT7 plays a major role in HeLa cells but TUT4/2 also contribute to mono-uridylation. The reaction is distributive, and the interaction between TUT and pre-let-7 is infrequent.
(3) Trimmed pre-miRNAs are uridylated by TUT7 and TUT4, which promote their degradation. TUT2 seems to be less active than TUT7/4. As the interaction between enzyme and substrate is frequent, multiple cycles of distributive uridylation result in oligo-uridylation. B Cis-acting elements of pre-let-7a-1 for mono-uridylation by TUT7. The overhang and the terminal loop are recognized by the C-terminal half of TUT7. C Model of interaction between TUT7 and its RNA substrates. For 3 0 trimmed pre-miRNAs, binding rate to TUT7 increased, resulting in enhanced uridylation activity.

Discussion
Our work demonstrates that there are three distinct pathways of pre-miRNA uridylation (Fig 5A). (1) In embryonic cells and certain cancer cells, TUT4 (and TUT7, to a lesser extent) associates with Lin28 and oligo-uridylates pre-let-7 specifically. Lin28-mediated oligo-uridylation blocks pre-let-7 processing and promotes degradation by DIS3L2.
(3) In this study, we uncover another pathway in which oligo-U tails are added by TUT7/4 to truncated pre-miRNAs with a 5 0 overhang. TUT2 seems to be less active than TUT7/4 ( Fig 2B and  Supplementary Fig S2D). The oligo-U tails on the trimmed pre-miRNAs may promote rapid degradation of non-functional pre-miRNA species. This study reveals the molecular mechanism of uridylation of group II pre-miRNA (Fig 5). Our results indicate that TUT7 distinguishes pre-miRNA substrates at the binding step by recognizing the cis-acting elements. Pre-let-7a-1 with a 2-nt 3 0 overhang is not efficiently uridylated due to infrequent binding of RNA to TUT7. Ablation of the terminal loop of pre-let-7a-1 also reduced the binding frequency, indicating that TUT7 recognizes both the overhang and the terminal loop. Given that TUT4/2 do not recognize the terminal loop ( Supplementary Fig S2C), TUT7 may be used preferentially to uridylate pre-let-7s while TUT4/2 may have a broader specificity.
We have drawn the energy landscape that explains kinetics of group II pre-miRNA mono-uridylation ( Supplementary Fig S6A and  B). RNA binding is represented as a transition from a 'free TUT7' state to 'RNA-bound TUT7' (termed RNA+TUT7). The structural motif of RNA is probed by TUT7 at the transient state, (RNA+TUT7)*. The energy barrier (DG, then DDG = À RT ln(k variant on /k Unmodified on )) between the 'free TUT7' state and the transient state becomes higher by 2.0 AE 0.5 or by 3.5 AE 0.1 (kJ/ mol) if pre-let-7a-1 is mono-uridylated (pre-let-7a-1 +U) or if the terminal loop is removed from the RNA substrate (pre-let-7a-1 L4), respectively (Supplementary Fig S3B). In contrast, the energy barrier becomes lower when a pre-let-7a-1 variant contains a long 5 0 overhang (DDG(DCUUUC) = À2.7 AE 0.9 kJ/mol; and DDG(Ac-pre) = À4.3 AE 0.6 kJ/mol) (Supplementary Fig S3B). However, the energy barrier between (RNA+TUT7) and (RNA+TUT7)* does not appear to change as significantly as the energy barrier of binding does. In summary, in the energy landscape of the TUT7-RNA interaction, it is the transient state at which TUT7 probes two cis-acting elements (the terminal loop and the overhang) and discriminates its RNA substrates.
Non-canonical overhang structures of pre-miRNAs (e.g. 1-nt 3 0 overhang or 5 0 overhang) increase uridylation efficiency. TUT7 oligo-uridylates 3 0 trimmed pre-miRNAs at a much higher rate than unmodified pre-let-7a-1 due to enhanced binding (Figs 2B, 3G and 5C). Intriguingly, unlike the processive reaction observed with Lin28-dependent oligo-uridylation of pre-let-7 (Yeom et al, 2011), oligo-uridylation of 3 0 trimmed pre-miRNAs results from successive uridylation in a distributive mode (Fig 5A and C). This explains why apparent 'oligo'-uridylation is observed even for unmodified pre-let-7a-1 when a large amount of TUT enzyme is used or when reaction time is extended (Fig 1D and Supplementary Fig S2). The distributive activity of human TUT7 is consistent with a recent structural study, which showed that Cid1, homolog of TUT7 in S. pombe, uridylates single-stranded RNA in a distributive manner (Yates et al, 2015).
Deep sequencing results suggest that TUT7 (and its paralogues) uridylates 3 0 trimmed pre-miRNAs in vivo as well as in vitro and that the uridylation is likely to induce degradation of the 3 0 trimmed pre-miRNAs. This result is consistent with a recent finding by Mourelatos and colleagues that defective Ago-bound pre-miRNAs are uridylated by TUT7/4 and degraded by exosome (DIS3 and RRP6) in mouse embryonic fibroblasts (Liu et al, 2014). They also reported that TUT7/4 associate with exosome, and this interaction may facilitate degradation of pre-miRNA. Our study confirms and expands the role of uridylation in removal of defective pre-miRNAs and further provides with mechanistic insights into the differential uridylation by TUTs. Moreover, our work explains their intrinsic preference for trimmed pre-miRNAs at the molecular level. Sensing the overhang structure, TUTs can employ multiple modes of action and thereby have versatile consequences of either repairing or removing pre-miRNAs depending on the molecular and cellular contexts.

Cell culture and transfection
HeLa and HEK293T (mycoplasma-free) cells were maintained in DMEM (Welgene) supplemented with 9% fetal bovine serum (Welgene). For RNAi, HeLa cells were transfected with 42 nM of siRNA by using Lipofectamine 2000 (Life Technologies). For simultaneous knockdown of three TUTs, equal amounts (14 nM) of siTUT7, siTUT4, and siTUT2 were combined. Transfection was performed two times over 4 days. For ectopic expression of proteins, HEK293T cells were transfected with plasmids by calcium phosphate method. The sequences of siRNA are listed in Supplementary Table S4.

Quantification of in vitro uridylation data
In vitro uridylation data are quantified as described in . The signal intensity profiles (20 pixels/mm) were calculated from the whole blot phosphor images using Fujifilm MultiGauge v3.0. For each lane, background signal is estimated using the arithmetic mean of the 25 th and the 50 th percentiles of the signal intensities. The signal intensities were subtracted by the estimated background level, then clipped to zero so that all intensities have zero or positive values. For the alignment of size marker bands, the signals from a marker lane were transformed to the first and second derivatives using Savitzky-Golay filter (window = 31 pixels, order = 3). The marker positions were detected by searching points where the sign of first derivative turns from positive to negative, and the second derivative is smaller than À100. The detected positions of marker bands were verified by visual inspection. The function between physical position in the gel and RNA size was defined using cubic spline interpolation. The density of RNA amount in size space was calculated using the first-order discrete differences of equal-width samples (0.1 nt) from cumulative density of the original intensity values. For the average length of extensions, the position having maximum signal intensity in the 0-min sample is used as a reference position. The average length of extension was derived from an equation, where x is the average length of extension, p is a position in the gel (by 0.1 nt-wide intervals), s p is the RNA size in nucleotides count for position p, r is the reference size of unextended RNA, and I p is signal intensity for position p. We excluded signals from degraded products (shorter than the reference size by 3 nt) in the calculation of average extension.

Purification of recombinant proteins
Recombinant TUT7 951-1,495 (rTUT7) and His-rTUT7 were prepared as previously described . For purification of recombinant TUT4 protein, human TUT4 267-1,312 was inserted into a self-modified pMAL expression vector, which fuses a hexa-His tag plus a maltose-binding protein tag at the N-terminus to the target protein. The plasmid was transformed into E. coli BL21(DE3)-RIL strain (Stratagene). The cells were cultured at 37°C until OD600 reached 1.0, and then the protein expression was induced with 1 mM IPTG at 16°C overnight. The hexa-His-MBP tagged protein was purified using a HisTrap FF column (GE Healthcare). The tag was cleaved by TEV protease and further removed by a second step HisTrap FF column (GE Healthcare) purification. The target protein was further purified by a Heparin FF column (GE Healthcare) and a Hiload Superdex G200 16/60 column (GE Healthcare). Recombinant Lin28b was prepared as previously described (Yeom et al, 2011).
Sample preparation and RNA labeling for single-molecule measurements RNA samples were labeled and prepared as previously described (Yeom et al, 2011).
The frequency measurement requires an accurate ratio between the concentrations of a sample of interest and a reference sample. To account for differences in concentration, we adsorbed RNA molecules to a positively charged surface as follows. KOH-etched quartz slides were coated with a layer of positively charged poly-L-lysine. After 5 min of incubation with 20 ll 0.01% poly-L-lysine (P4707, Sigma), the chamber was washed with 100 ll of buffer A. After washing, two fluorescently labeled RNA substrates (Unmodified-Cy3, Variant-Cy5) were introduced into the microfluidic chamber. After 5 min of incubation, the remaining unbound substrate was washed away with 100 ll of imaging buffer A and data were obtained from 20 fields of view. For each construct, this procedure was repeated with three individual dilutions on three different slides. Mean number of counts was used to correct the relative binding frequency for concentration.

Single-molecule data acquisition and analysis
A series of CCD images were acquired with laboratory-made imaging software at a time resolution of 0.03-0.1 s. Fluorescence time traces were extracted with an algorithm written in IDL (ITT Visual Information Solutions) that picked fluorescence spots above a threshold with a defined Gaussian profile. The extracted time traces were analyzed using laboratory-made Matlab algorithms (Math-Works) that selectively picked anticorrelated traces above a defined threshold. These selected traces were further analyzed using a laboratory-made Matlab algorithm to extract dwell times and the number of binding events per trace. The relative binding frequency plots were generated by dividing the total number of binding events of each construct with the correction factor obtained from the poly-L-lysine experiment, after which the variant was normalized against the unmodified construct.
To measure the binding frequency, Cy3 molecules were simultaneously excited over an area of 50 × 50 lm 2 with 16% of the full laser power of the (8 mW) green laser (532 nm) and red laser (640 nm), while the time resolution was set to 0.03 s. Under these imaging conditions, we obtained a high signal-to-noise ratio that facilitated the automated analysis. For dwell time measurements, Cy5 molecules were excited with 8% of the full laser power (4 mW) green laser (640 nm) to minimize photobleaching of the Cy5 dye during our observation time. Meanwhile, the time resolution was set between at 0.1 s to collect a large enough number of photons per time bin.

Pre-miRNA library preparation
To prepare pre-miRNA cDNA library, total RNA was separated on 15% urea-PAGE and RNAs of 40-120 nt were gel-purified. Sizefractionated RNAs were ligated to 3 0 adaptor by using T4 RNA ligase 2, truncated (NEB). The 3 0 adaptor-ligated RNA was separated on 12.5% urea-PAGE and RNAs of 60-140 nt were gelpurified. Size-fractionated RNAs were reverse-transcribed with a RT primer that is complementary to the 3 0 adaptor by using SuperScript III (Life Technologies), followed by two-step PCR amplification. cDNA was firstly amplified with the RT primer and miRNA-specific forward primers for 10 cycles and secondly amplified for 10 cycles (12 let-7 family primers) or 13 cycles (43 other miRNA primers) with Phusion DNA polymerase (NEB). The sequences of miRNA-specific primers are shown in Supplementary  Table S1. The cDNA libraries were separated on 6% native polyacrylamide gel, and DNAs of 150-225 bp were gel-purified. The library was sequenced on Illumina MiSeq (110 × 59 paired end run) with 50% of the PhiX control library (Illumina, FC-110-3001). All adapters and primers are synthesized by IDT. Oligonucleotide sequences except for miRNA-specific forward primers are shown in Supplementary Table S4.
Processing for pre-miRNA sequencing We used miRBase release 21 (Kozomara & Griffiths-Jones, 2014) and the UCSC hg38 genome assembly for the reference sequences of human pre-miRNAs and flanking regions. To reduce misalignments near the ends of miRBase hairpin sequences, we extended the miRNA precursor spans by 10 bp to both ends. The sequences were extended and retrieved from the genome assembly using BEDTools (Quinlan & Hall, 2010) slop and getfasta commands. The cDNA library was sequenced for 110 cycles with the small RNA workflow in Illumina MiSeq. Sequences were processed using Cutadapt (Martin, 2011) to trim the 5 0 -most fifteen nucleotides and clip 3 0 adapter sequences out. The sequence at the 5 0 end was removed because they originate from PCR primers for specific enrichment of pre-miRNAs and often include significant number of mismatches to known pre-miRNA sequences. Short sequences (< 15 nt) after trimming and clipping, and sequences without a 3 0 adapter part were removed from the further analyses. The remaining sequences were aligned to 10-bp-extended miRBase hairpins explained above using BLAT (Kent, 2002) with options "ÀnoTrimA ÀtileSize=8 ÀstepSize=4 ÀminIdentity=70 Àout=pslx". From the output alignments, the best alignments among multi-mapped reads were chosen by following criteria in order: maximum matched bases, minimum mismatch, minimum number of gaps in query of alignment, minimum number of gaps in target of alignment, minimum number of gapped bases in query of alignment and minimum number of gapped bases in target of alignment (preferred first). For the selected alignments, all unaligned bases in 3 0 ends of local alignments were regarded as The EMBO Journal Vol 34 | No 13 | 2015 ª 2015 The Authors non-templated additions. We additionally "rescued" the non-templated additions, which were matched to the reference sequence by ambiguity in A/U-rich sites. First, the regions to be re-examined for the rescue were defined as all subsequent bases containing only A or U immediately starting from the 3 0 end of a sequenced read (/[AU]+$/in the regular expression). Then, all subsequent bases including the first mismatched base in the re-examination region were rescued so as to be regarded as non-templated additions. The source codes, workflows implemented in Snakemake (Koster & Rahmann, 2012), and interactive notebooks in IPython Notebook (Perez & Granger, 2007) used for the analyses in this study are freely available from https://github.com/hyeshik/bskim-2015pre-miRNA.

Determination of length of trimming and length of U-tail
For each pre-miRNA, the most frequent 3 0 end position of templated portions of reads in the control was considered as the "reference end position". Six hairpins (hsa-mir-16-2, hsa-mir-100, hsa-mir-222, hsa-mir-320a, hsa-mir-1248, and hsa-mir-1291) whose reference end positions were offset by more than 3 nt from the 3 0 end of the mature miRNA from 3 0 arm of the hairpin defined in the miRBase were removed from the subsequent analyses to exclude artifacts from the statistics. The length of trimming of each read was calculated by subtracting the position of last templated base in the read from the reference end position (positive for "trimmed", negative for "extended" reads). Length of U-tail was defined as number of U residues in the non-templated additions without any other kind of nucleotidyl additions.

Accession numbers
Sequenced reads have been deposited in the NCBI Gene Expression Omnibus (GEO) database (accession number GSE64482).
Supplementary information for this article is available online: http://emboj.embopress.org