An endogenous cluster of target-directed microRNA degradation sites induces decay of distinct microRNA families

SUMMARY

While much is known about microRNA (miRNA) biogenesis and targeting, relatively less is understood about miRNA decay. Target-directed miRNA degradation (TDMD) is a mechanism in metazoans where certain RNAs can “trigger” miRNA decay. All known TDMD triggers base pair with the miRNA seed, and extensively base pair on the miRNA 3′ end, a pattern that is believed to be a requirement for miRNA turnover. Using Ago1-CLASH, we find that the Drosophila transcript Kah contains at least two triggers, a “trigger cluster,” against miR-9b and the miR-279 family. One trigger contains minimal/non-canonical 3′ end base-pairing but is still sufficient to induce TDMD of the miR-279 family. We find that these clustered triggers lack cooperativity, that minimal 3′ pairing is required for miR-279 family turnover, and probed the in-cell structure of the Kah trigger cluster. Overall, we expand the list of endogenous TDMD triggers, which revealed unexpectedly complex regulation in miRNA turnover.

In brief

Hiers et al. reports the identification of a TDMD trigger with non-canonical 3′ base-pairing within the Kah transcript. This transcript contains at least two triggers, a “trigger cluster,” that each appears to independently induce decay of miRNAs from two distinct miRNA families.

Graphical Abstract

INTRODUCTION

MicroRNAs (miRNAs) are a class of small ∼22 nucleotide (nt) non-coding RNAs that induce post-transcriptional gene silencing.¹ To do this, miRNAs are bound by one of the Argonaute (AGO) family proteins and serve as guides for AGO association with target RNAs.^1–3 Typically, a target-bearing sequence complementarity to miRNA seed (nt 2–8) is sufficient to predict downregulation.^4,5 With such short targeting requirements and hundreds of conserved miRNAs, they are considered “master” regulators of post-transcriptional gene expression. Indeed, loss of individual miRNA genes has been shown to induce a variety of phenotypes including developmental abnormalities and embryonic lethality.^1,6,7

While extensive research has been given to the study of miRNA biogenesis and functional consequences, relatively less is understood about miRNA decay.⁸ For many years, researchers had observed that certain miRNA targets bearing extensive 3′ complementarity (in addition to seed matching) could “trigger” rapid miRNA turnover, a process collectively referred to as target-directed miRNA degradation (TDMD).^9–15 In 2020, it was revealed that the endogenous TDMD mechanism in mammals is catalyzed via ZSWIM8, a Culin-RING E3 ubiquitin ligase.^16,17 Mechanistically, it appears that when miRNAs bind to TDMD targets, hereafter simply referred to as TDMD “triggers,” extensive base-pairing induces an AGO conformational change, enabling ZSWIM8-mediated ubiquitination and proteasomal degradation, exposing the miRNA to ribonucleases.^16–19 Since this revelation, several studies have identified the ZSWIM8 ortholog in Drosophila (Dora), and Caenorhabditis elegans (Ebax-1) also carry out TDMD in their respective systems.^17,20–23 In fact, several recent reports suggest TDMD plays a fundamental role in both mammalian and Drosophila development.^21,24,25 TDMD of miR-3 is essential for Drosophila embryogenesis,²¹ and degradation of miR-322/503 was found to be critical for mammalian growth.²⁴ In total, loss of ZSWIM8 (or Dora/Ebax-1) increases the abundance of nearly 100 miRNAs; however, the triggers that induce TDMD for most of these miRNAs are still unknown.^{17,20–22,24–26} Therefore, it is essential to identify each of these triggers to better understand the TDMD molecular mechanism and phenotypic consequences of TDMD in metazoans.

Triggers have been identified through computational and biochemical screening approaches.^21,27–30 In either case, these studies typically employ a stringent extensive 3′ complementarity requirement for putative triggers, given that all known TDMD examples contain at least seven consecutive base-pairings between triggers and the miRNA 3′ end.^26,31 Interestingly, it was recently revealed that there is likely a “seed-sufficient” trigger in C. elegans that requires no 3′ end complementarity to degrade the miR-35 family.²⁰ Relatedly, extensive target complementarity with miRNAs can out-compete the miRNA 3′ association with the AGO PAZ domain, thereby exposing the miRNA 3′ end to non-templated nucleotide addition by terminal nucleotidyltransferases (TENTs).^{10–13,18,19,32,33} Given that all known triggers extensively base pair with their targets, they have also been reported to induce tailing/trimming of associated miRNAs.¹⁸ However, it was recently observed that several examples of miRNAs are stabilized following loss of ZSWIM8 (ZSWIM8-sensitive) without any change in tailing.²⁴ Together these observations suggest that there are potential triggers in metazoans that base pair less extensively with the cognate miRNA but are still sufficient to induce TDMD.

We recently have had success in the application of AGO-crosslinking, ligation, and sequencing of hybrids (AGO-CLASH),³⁴ to screen for TDMD triggers.^29,30 AGO-CLASH, a modified CLIP method, uses UV crosslinking, AGO immunoprecipitation, and ligation of miRNAs to their targets.^34–37 Sequencing of the subsequent miRNA-target hybrid enables identification of high-confidence miRNA-target interactions. Our application of this method in Drosophila S2 cells allowed for the identification of five triggers: Ago1:miR-999, h:miR-7, Kah:miR-9b, Wgn:miR-190, and Zfh1:miR-12.³⁰ Given that many other miRNAs are stabilized following loss of Dora in S2 cells, there are additional triggers remaining that require identification.^17,21,23

In the initial studies, we employed stringent extensive base-pairing criteria for the screening of potential triggers,^29,30 though AGO-CLASH data could presumably be used to identify even non-canonical triggers by simply relaxing the screening criteria. Here, we find that the Kah transcript, which we originally identified as a miR-9b trigger, contains a secondary trigger sufficient to induce decay of the entire miR-279 family, consisting of miR-279, miR-286, and miR-996. Interestingly, this quality makes Kah an endogenous example of multiple triggers within the same transcript, a “trigger cluster.” Surprisingly, this miR-279 family trigger contains non-canonical TDMD base-pairing, in that relatively little 3′ complementarity is present, though this minimal 3′ complementarity of three base pairs still appears required for miR-279 turnover. Our study suggests that the types of base pairs that induce TDMD are far from full comprehension. Overall, our results shed light on the existence of both non-canonical and clustered triggers and highlight the ability of AGO-CLASH to aid in identifying additional unexpected triggers.

RESULTS

Ago1-CLASH suggests a cluster of TDMD triggers in Kah

In our initial screen for triggers in Drosophila S2 cells, we validated each of our highest confidence candidates via CRISPR-Cas9-mediated deletion.³⁰ In all but one case, loss of the trigger increased the abundance of the predicted miRNA specifically. With knockout (KO) of the Kah miR-9b trigger (Kah-9b), there is a significant increase in the abundance of both the miR-9b and miR-996 guide strands but not their co-transcribed passenger strands (Figures 1A and 1B). This quality is crucial to differentiate miRNA increase in abundance due to loss of TDMD (increase guide abundance) from an increase in biogenesis (increase both guide and passenger abundance).12,17,20–22,24,25,29,30 Puzzlingly, miR-9b and miR-996 belong to distinct miRNA families and therefore have different seed sequences and target repertoires: miR-9b is a member of the miR-9 family (miR-9a/b/c), while miR-996 is a member of the miR-279 family (miR-279/286/996). Upon further scrutiny, we observed that the Kah-9b trigger KOs broadly upregulated the guide strands of the whole miR-279 family (Figures 1A and 1B). Given the possibility of ligation biases that could disproportionally reflect miRNA abundance in our previously outsourced small RNA sequencing (RNA-seq),^38–41 we validated the upregulation of both miR-279 and 996 via near-infrared northern blot (from here on, northern blot).⁴² As in the Dora-KO, disrupting the Kah-9b trigger elevated the levels of miR-9b, miR-279, and miR-996, while the control miRNA bantam did not increase abundance (Figure 1C). It should be noted that miR-286 is likely undetectable by northern blot in S2 cells due to its relatively low abundance and therefore was excluded in this analysis. Since the miR-279 family guides are upregulated upon loss of Dora¹⁷ and are similarly stabilized following loss of Kah-9b (Figures 1C and S1A), we therefore considered whether we may have unintentionally perturbed a potential miR-279 family trigger.

Figure 1. Ago1-CLASH suggests a cluster of TDMD triggers in Kah.

(A) Small RNA-seq³⁰ showing impact of Kah-9b trigger KO on miRNA abundance. x axis: mean miRNA counts in Scr and KO; y axis: log₂ fold-change (log₂FC). Significantly upregulated guide strands (red), non-significant (pink), passenger strands (navy/cyan). Significance: FDR-adjusted p < 0.05 via DESeq2 (n = 2).

(B) log₂FC between guide and passenger strands after Kah-9b trigger KO.

(C) Northern blot measuring miRNA abundance in WT (wild-type), Scr (scramble/non-target sgRNA control), Kah-9b trigger KO, and Dora KO S2 cells. Bantam used as loading control.

(D) Ago1-CLASH miRNA-target hybrids for the miR-279 family screened for additional Kah triggers.

(E) Summary of Kah triggers (miR-9b, miR-279, and miR-996) from CLASH. miR-286 lacked hybrids and is separated by a dashed line. Red text marks the miRNA seed region. Base-pairing: CLASH hybrids (Hyb pipeline); miR-286:Kah (RNAcofold).

(F) Schematic model of potential clustered TDMD cooperativity.

We previously utilized the Ago1-CLASH method (miRNAs primarily bind Ago1 in Drosophila) in both Dora-KO cells and a scramble (Scr) sgRNA control.³⁰ Ideally, CLASH ought to preferentially isolate Dora-sensitive miRNAs with their cognate triggers under Dora-KO conditions. To filter these miRNA-target hybrid data, we looked solely at hybrids from miRNAs that were stabilized following loss of Dora and employed three screening criteria: (1) extensive 3′ complementarity (>7 bp) separated from the seed base-pairing by at least a single mismatch (>0 nt central bulge), (2) enrichment of the hybrid following loss of Dora (>4-fold increase), and (3) overall abundance of the hybrid. The miR-279 family members have divergent 3′ end sequences, meaning that a transcript that could extensively base pair with one member would be unlikely to extensively base pair with another (Figure S1B). Given that we would need to remove the extensive base-pairing criteria to screen for non-canonical Kah triggers, we took a closer look at how hybrid enrichment, and abundance could aid our efforts in identifying them.

We initially determined hybrid enrichment based on an increase in hybrid abundance in the Dora-KO as compared to the Scr control.³⁰ However, this criterion may have potential issues as all Dora-sensitive miRNA hybrids, regardless of their contribution to TDMD, ought to increase due to the increase in miRNA abundance upon loss of TDMD. We therefore then simply looked at all the hybrids from Dora-sensitive miRNAs with validated triggers in S2 cells (miR-7, −9b, −12, −190, and −999) and considered the proportion each hybrid occupies in either Scr or Dora-KO to account for the dramatic change in miRNA abundance. To our surprise, validated triggers were not enriched when viewed in this fashion, meaning that loss of Dora merely increases the absolute number of Dora-sensitive miRNA hybrids but did not specifically enrich bona fide targets/triggers (Figure S1C; Table S1A). Thus, we next considered hybrid abundance as a readout for potential triggers. Strikingly, we found that each of our previously validated triggers, h:miR-7, Kah: miR-9b, Zfh1:miR-12, Wgn:miR190, and Ago1:miR-999, occupied the largest proportion of their respective mRNA hybrids even without considering base-pairing (e.g., seed-match, extensive complementarity) (Figure S1D; Table S1B).

With these qualities in mind, we looked for any abundant miR-279 family hybrids with Kah (Figures 1D and S1D). To our surprise, we found Kah hybrids for both miR-279 and 996 at a highly conserved site ∼300 nt upstream of the miR-9b trigger (Figures 1E, S1E, and S1F). This site did not have the extensive 3′ complementarity currently believed to be required for TDMD^26,31; instead, the miR-996/286 3′ ends were only predicted to contain five base pairs and miR-279 only three (Figure 1E). We did not identify any Kah hybrids with miR-286, again presumably due to the low expression level of miR-286 in these cells. All known transcripts that induce TDMD only contain a single site sufficient to induce decay of the associated miRNA; this secondary site against the miR-279 family would make Kah an endogenous example of multiple triggers clustered within its transcript, a so-called “trigger cluster.” Since this site would contain non-canonical TDMD base-pairing, we considered a model where this miR-279 family trigger may be “suboptimal,” and only effective with the coupled “canonical” downstream miR-9b trigger, in a mechanism mirroring the clustered biogenesis of the suboptimal miR-451a coupled with canonical miR-144 primary transcript^43,44 (Figure 1F). If so, then the Kah: miR-279 family interaction may have been disrupted by our Cas9-mediated deletion of the miR-9b trigger. Thus, we set out to interrogate the efficacy and molecular mechanism of the Kah trigger cluster.

The Kah transcript regulates the abundance of distinct miRNA families

The most direct means for interrogating the efficacy of a potential trigger is to delete the endogenous region via CRISPR-Cas9. To do this, we generated new sgRNAs targeting either side of the putative Kah miR-279 family trigger to induce nucleolytic cleavage of its genomic locus (Figures 2A, S2A, and S2B). Genomic DNA from these cells were then subject to high-throughput sequencing to validate the KO efficiency. We estimate that 50%–60% of cells contain complete deletions of the targeted locus (Figures S2C and S2D; Table S1D). Interestingly, these new KO populations (Kah-279-KO1/2) again showed upregulated miR-279 family and miR-9b in tandem when observed via northern blot, comparable to levels observed in a Dora-KO and our original Kah-9b-KO2 population (Figure 2B). To ensure that the observed miRNA stabilization was indeed due to loss of miRNA degradation, we employed the “accurate quantification by sequencing” (AQ-seq) method developed by Narry Kim’s group to generate minimally biased small RNA-seq libraries containing spike-in standards for count normalization³⁸ (Figure 2C). In total, we found five miRNA guide strands (miR-279, −286, −996, −9b, and −92b) that were significantly upregulated (p adj. <.001) compared to their co-transcribed passenger strands upon loss of the miR-279 family trigger (Figures 2C and 2D). We did observe significant upregulation of the miR-980 passenger strand, though its guide was also upregulated to a similar degree, a quality likely indicative of increased biogenesis (Figure 2C).

Figure 2. The Kah transcript regulates the abundance of distinct miRNA families.

(A) CRISPR-Cas9 deletion strategy for Kah-279 trigger. Predicted cut sites (red triangles), trigger sequence (red text), and PAM sites (cyan text).

(B) Northern blot post-Kah-279 trigger KO. Relative miRNA levels are shown as mean ± standard deviation (SD) (n = 3 biological replicates). miR-7 used as Dora-sensitive control.

(C) AQ-seq following Kah-279 trigger KO. miRNA abundance (x axis) represents the mean miRNA counts per million (CPM) in Scr libraries. Highlighted are miRNAs upregulated following Kah-279 KO (FDR-adjusted p value <0.001, n = 2 biological replicates).

(D) log₂FC of guide vs. passenger strands post-KO.

(E) RNA-seq of Kah transcript (TPM) after KO. **p < 0.01, p values were calculated using DEseq2. Error bars indicate ± SD (n = 3 biological replicates).

(F) Candidate miR-92b trigger in Kah (CLASH), compared to Marge:miR-92b interaction.

While the upregulation of the miR-279 family was expected due to trigger KO, the upregulation of miR-9b and miR-92b was puzzling for a couple of reasons: 1) Under a cooperativity model, the Kah-279 trigger KO should only upregulate the miR-279 family, suggesting our model is likely wrong. 2) miR-9b and miR-92b are from distinct miRNA families, the miR-9 family and the miR-310 family, respectively. Several of the miR-310 family members (miR-310, 311, and 313) were reported to be sensitive to loss of Dora in vivo and directed for degradation by the lncRNA Marge.²¹ However, this trigger was not observed to induce degradation of the remaining miR-310 family (miR-312, −92a, and −92b) despite them also being Dora-sensitive, suggesting that at least one other trigger exists inducing decay of the miR-310 family in addition to Marge.²¹ An alternative hypothesis to trigger cluster cooperativity would be that CRISPR-Cas9-mediated deletions aberrantly induced Kah knockdown, decreasing its abundance and therefore reducing TDMD of any triggers localized within the transcript. To test this, we performed RNA-seq to monitor the change in Kah abundance in our Kah-279 KOs compared to the Scr control. Curiously, our Kah trigger KOs significantly reduced the abundance of the Kah transcript, on average having a >50% reduction, and we observed a similar trend with our original Kah-9b trigger KOs (Figures 2E and S2E). With this in mind, we considered whether the observed upregulation of miR-92b may indeed be due to a loss of TDMD catalyzed via another trigger within the Kah transcript. When reconsidering the miR-92b hybrids within our Ago1-CLASH dataset, we found a highly conserved site within Kah for miR-92b with only five 3′ end base pairs, a pattern mirroring what we see with miR-286/996 on the miR-279 family trigger (Figures 1E, 2F, and S2F). Puzzlingly, miR-92b is predicted to base pair with Kah less than it is with Marge, yet Marge does not induce miR-92b degradation, suggesting that more extensive base-pairing may not necessarily be more efficacious for TDMD, depending on the miRNA sequence (Figure 2F). Though, due to the low expression level of miR-92b in S2 cells, it is unlikely to be an ideal model for further study of this potential trigger.

The Kah trigger cluster differentially influences miRNA tailing, trimming, and function

A common feature observed of many miRNAs sensitive to TDMD is alteration to tailing (non-templated nucleotide addition) and trimming (shortening of the miRNA) on the miRNA 3′ end.^{10–12,16–18,24} Tailing and trimming of these miRNAs are usually more pronounced upon loss of ZSWIM8 (or its ortho-log).^16,17,24 The major hypothesis associated with these observations is that triggers ought to extensively base pair with the interacting miRNA, potentially outcompeting the AGO PAZ domain binding of the miRNA 3′ end, leaving the end solvent exposed for non-templated nucleotide addition by TENTs or trimming by non-specific ribonucleases.^{10–13,18,19,32,33} Given the less extensive 3′ base-pairing for the miR-279 family with Kah (Figure 1E), we sought to take a detailed look at how loss of this trigger may affect the accumulation of miRNA isoforms (isomiRs) for miR-279, −996, and −9b. To do this, we reanalyzed our AQ-seq data and categorized miRNAs based on the number of nucleotides and sequence. In this analysis, miRNAs mapping to a certain miRNA gene ought to have 100% sequence identity with the genome from nucleotides 2–18, allowing for mixed sequences at position 1 and 19–26 (Figure S3A). These criteria will capture the bulk of miRNA sequences, as it accounts for the normal size range for miRNAs, non-templated additions and trimming, as well as potential alterations in processing of the miRNA precursors by Drosha/Dicer.

We observed a clear accumulation of shorter (trimmed) isomiRs of the miR-279 family, with little to no increase in tailing upon loss of its trigger (Figures 3A and 3B). This observation mirrors what we see for both miR-279 and miR-996 by northern, with these shorter isoforms becoming the bulk of miR-279 signal (Figure 2B). In contrast, miR-9b displayed a mix of accumulating trimmed and tailed isomiRs (Figure 3C). We next sought to define what proportion of our miRNA counts were from sequences that match the genome (templated). When only considering templated counts, we observed a consistent but minimal reduction, with these sequences being reduced by ∼1%–3% in Kah-279 KOs (Figures S3B, S3C, and S3D). We conclude that while loss of Kah-mediated TDMD alters the average length of all three miRNAs, this alteration is unlikely to be mediated by tailing. We noted that shorter miR-279 isomiRs were upregulated to a much larger degree than longer miR-279 isomiRs upon trigger KO, in a trend that was highly correlated (Figure S3E). Taken together, these data may suggest that shorter miR-279 isomiRs are preferential substrates for Kah-mediated TDMD compared to longer isomiRs. Even so, we did not observe a similar trend for miR-996 or miR-9b, though isomiRs shorter than 22 nt were modestly directed for turnover to a higher degree (Figures S3F and S3G).

Figure 3. The Kah trigger cluster differentially influences miRNA tailing, trimming, and function.

Relative proportions (top) or fractions (bottom) of isomiRs (18–26 nt) in Scr (gray) vs. Kah-279 trigger KO (red) for (A) miR-279, (B) miR-996, or (C) miR-9b; error bars indicate ± SD. The increased repression of the predicted targets of the (D) miR-279 family and (E) miR-9 family following loss of the Kah-279 trigger. Plotted is the cumulative change (log₂FC) in TargetScan-predicted mRNA target abundance following Kah-279 trigger KO compared to Scr control. Targets were classified into all conserved targets, top conserved targets, or all other targets (non-targets), with the number of transcripts considered for each cohort listed within the plot. Dots at the bottom of the graphs represent the median expression level of each target cohort. p values were calculated using the Mann-Whitney U test, n = 3 biological replicates.

(F) The change in miRNA family (miR-279 or miR-9) abundance following Kah-279 trigger KO (n = 2 biological replicates).

In any case, loss of Kah-mediated TDMD for the miR-279 family and miR-9b dramatically increases the overall abundance of these miRNAs and therefore ought to increase their ability to induce repression of target mRNAs. To address this, we categorized our RNA-seq data based on TargetScan-predicted mRNA targets of the miR-279 family and the miR-9 family.⁴⁵ The predicted targets of both groups were significantly repressed compared to the non-target controls, in line with the idea that the function of TDMD lies in its ability to limit miRNA-mediated silencing (Figures 3D and 3E). We did note that the overall repression of the miR-279 family targets was greater than the miR-9 family targets, a quality that becomes more understandable when considering that the absolute increase in miRNA molecules was greater for the miR-279 family (miR-279/286/996) compared to the miR-9 family (miR-9a/b/c) (Figure 3F). These results highlight how Kah controls the endogenous levels of distinct miRNA family targets via clustered TDMD.

The Kah trigger cluster specifies miRNA decay with little crosstalk

Since the aberrant knockdown of Kah following trigger KOs ultimately muddles our ability to unambiguously classify predicted triggers, we next sought to further address if each trigger does indeed specify miRNA decay. To do this, we first attempted to transiently inhibit miRNA-Ago1 complex association with endogenous Kah using morpholino oligonucleotides (or simply “morpholinos”) against either trigger (Figure 4A). Morpholinos are short non-ionic nucleic acid analogs that can associate with RNAs based on sequence identity⁴⁶ and have been adopted in several studies to inhibit AGO binding to targets and triggers alike.^11,30,47 This assay can help to address two outstanding questions: (1) Does the miR-279 family trigger specify decay? (2) Is the miR-279 family trigger suboptimal and dependent on the downstream miR-9b trigger? When we incubated S2 cells with 10 μM of anti-Kah-279 trigger morpholinos, we observed a significant stabilization of miR-279/996 in wild-type (WT) cells but not Dora-KO (Figures 4B [compare lane 6 to lane 4 and lane 12 to lane 10] and S3H). Interestingly, when we attempted the same experiment with anti-Kah-9b trigger morpholinos, we only observed stabilization of miR-9b, with no concomitant increase in miR-279/996 (Figures 4B [compare lane 5 to lane 4] and S3H). These changes are also unlikely to be the result of increased miRNA biogenesis or Kah destabilization, as we did not observe altered Kah or pri-miRNA abundance (Figure S4A). In line with this, we validated the upregulation observed in WT cells via miR-qPCR (Figure S4B). The magnitude of change we observed for trigger inhibition was modest but specific, which can largely be attributed to the transient and passive association of morpholinos with transcripts. These results suggest that our predicted miR-279 family trigger specifies miR-279 family decay and is sufficient to recruit Dora to Ago1 on its own.

Figure 4. The Kah trigger cluster specifies miRNA decay with little crosstalk.

(A) A schematic of anti-trigger morpholino experimental design for endogenous Kah.

(B) Northern blot after treatment with NT, anti-9b, or anti-279 morpholinos (5/10 μM). Relative miRNA levels are shown as mean ± SD (n = 3 biological replicates).

(C) Schematic of GFP reporter constructs: GFP, Kah-WT, Kah-MutA, and Kah-MutB.

(D) Northern blot following introduction of reporters shown in (C). Relative miRNA levels are shown as mean ± SD (n = 3 biological replicates). AQ-seq following expression of (E) Kah-WT, (F) Kah-MutA, or (G) Kah-MutB compared to the GFP control. Significance was determined using an FDR-adjusted p value <0.01, n = 2 biological replicates.

To further characterize the Kah trigger cluster, we next sought to induce miRNA degradation by transiently expressing a portion (∼700 nt) of the Kah 3′ UTR within a GFP reporter driven by a constitutively active Drosophila actin promoter (Figure 4C). In this experiment, we will express four constructs: GFP alone (GFP), GFP with the WT Kah 3′ UTR (Kah-WT), this same construct with mutated miR-279 family trigger (Kah-MutA), or an alternative with mutated miR-9b trigger (Kah-MutB) (Figures 4C and S4C). In this way, we can observe if each predicted trigger is sufficient to induce miRNA decay while monitoring transfection efficiency via GFP. Ideally, we would like to express these triggers in Kah-279 KO cells to achieve a clear and robust reduction in miR-279, −996, and −9b by northern blot. However, attempts to express constructs containing the Kah 3′ UTR backbone in Kah-279 KOs were sharply repressed compared to Dora-KO (Figure S4D), presumably because of the constitutive expression of CRISPR-Cas9 complex targeting Kah-279. We also considered generating Kah 3′ UTR constructs with mutated PAM sequences adjacent to the sgRNA target sites. However, these constructs were also aberrantly repressed in the Kah-279 KOs (Figure S4E). We instead settled on utilizing the Scr and Dora-KO lines for further experiments, as they offered similar robust expression of Kah reporters, with our Kah-WT reporter increasing Kah 3′ UTR abundance by ∼20-fold (Figures S4F–S4H). When observed by northern blot, only Kah-WT was able to induce miRNA decay of miR-279, −996, and −9b in tandem (Figure 4D, compare lanes 1 and 2). Mutations to either trigger (Kah-MutA or MutB, Figure 4C) relieve repression of either miR-279/996 or miR-9b, respectively, with no observable crosstalk between the triggers (Figure 4D, compare lane 1 to lanes 3 and 4). For all trigger reporters, no consistent trend of reduction was observed when expressed in the Dora-KO, again highlighting the Dora-dependence of these triggers (Figure 4D, compare lane 5 to lanes 6, 7, and 8).

To directly quantify miRNA change in abundance induced by our reporters, we generated new AQ-seq libraries from trigger-expressing S2 cells and compared them to the GFP-only control (Figures 4E–4G). We observed clear and specific repression of the miR-279 family, miR-9b/c, and miR-92b guide strands when expressing the Kah-WT construct in Scr S2 cells (Figure 4E). However, the only significantly changed guides (p adj.< .01) compared to their passenger strands were miR-279, −996, and −9b (Figures 4E and S5A). Consistent with this, it should be noted that we and the Bartel lab observed only a small increase in miR-9c following KO of the Kah-9b trigger, suggesting it is only minimally directed for turnover via Kah.²¹ The overall abundance of miR-286 and miR-92b may limit reliable detection of their change responding to Kah reporters, given that these guides are only represented by dozens or hundreds of counts per million (CPM), respectively. Despite this, the miR-286 guide was considerably more repressed compared to its passenger strand (Figure S5A). Consistent with our northern blots, we observed no TDMD of the miR-279 family or miR-9b/c when their respective triggers were mutated in the Kah reporters (Figures 4F, 4G, S5B, and S5C). In total, these results demonstrate that both the endogenous and ectopic expression of the Kah trigger cluster is sufficient to induce degradation of the miR-279 family and miR-9b.

The Kah-279 trigger requires 3′ base-pairing to induce miR-279 family turnover

With the successful utilization of our reporter system to induce miR-279 family TDMD, we next sought to address if the three base pairs at the 3′ end of miR-279 with Kah is required for turnover (Figure 5A, Kah-WT) or if the Kah-279 trigger is an example of a hypothesized “seed-sufficient” trigger.²⁰ We first generated a new reporter that introduces a seven-nucleotide mutation into the Kah-279 trigger, which removes the predicted 3′ base pairs for both miR-279 and 996 while leaving the seed-match intact (Figure 5A, Kah-Seed). Expression of Kah-Seed did not induce decay of miR-279/996 when observed by northern blot (Figures 5B and 5C, compare lane 1 to lane 3). Interestingly, converting the Kah-279 trigger to a seed-match sequence with the miR-279 family dramatically repressed the reporter transcript compared to GFP alone or Kah-WT (Figure 5D). While tempting to attribute the loss of TDMD efficacy of the Kah-Seed to its reduced transcript abundance, Kah-Seed still induced robust decay of miR-9b when observed by miR-qPCR (Figure 5E). We rationalize that our WT reporters increase Kah 3′ UTR abundance by ∼20-fold (Figure S4G), and a ∼50% reduction in mutant triggers is still sufficient to induce robust TDMD.

Figure 5. The Kah-279 trigger requires 3′ base-pairing to induce miR-279 family turnover.

(A) A schematic of the predicted base-pairing of miR-996 and miR-279 with Kah-279 GFP reporters: Kah-WT, Kah-Seed, Kah-3ʹ Mut, Kah-2nt Mut, Kah-GAG1, and Kah-GAG2.

(B) Northern blot following introduction of GFP reporters to Scramble S2 cells described in (A); relative miRNA levels are shown as mean ± SD (n = 3 biological replicates).

(C) Quantitation of northern blot signal in (B). RT-qPCR measurement of relative GFP reporter transcript (D) and miR-9b-5p (E) following transfection of the reporters. Error bars indicate ± SEM (n = 3 biological replicates), p values were generated with unpaired t tests, p value *<0.05, n.s., not significant, n = 3 biological replicates.

While these data are suggestive of the 3′ base pairs being required for miR-279 family turnover, the mutation in Kah-Seed is quite extensive and may alter the trigger in ways not easily predicted. We therefore more precisely altered the 3′ base pairs of miR-279/996 with the Kah-279 trigger. Kah-3′ Mut introduces a 3-nucleotide mutation that ought to ablate the 3′ end association of both miR-279 and −996 with Kah, and Kah-2nt Mut merely reduces the number of 3′ base pairs with miR-996 without affecting miR-279 base-pairing (Figure 5A). Expression of the Kah-3′ Mut reporter did not induce decay of the miR-279/996 (Figures 5B and 5C, compare lane 1 to lane 4) and showed a relatively low level of trigger transcript like the Kah-Seed reporter (Figure 5D). Conversely, the Kah-2nt Mut reporter still induced decay of the miR-279/996, but the magnitude of miR-996 degradation was reduced compared to Kah-WT, whereas miR-279 degradation was comparable (Figures 5B [compare lanes 1 and 2 to lane 5] and 5C). The Kah-2nt Mut reporter showed a marginal reduction in transcript abundance compared to controls (Figure 5D), potentially indicating a middle ground of miR-279/996-mediated target repression and TDMD. Each of these reporters was able to reduce the abundance of miR-9b comparably to Kah-WT despite the altered transcript abundance (Figure 5E). Finally, we noticed that the “GAGAG” sequence within the Kah-279 trigger potentially allows for miR-279 to base pair with either “GAG” repeat. We therefore generated two final constructs with point mutations to force miR-279 base-pairing with either “GAG” (Figure 5A, Kah-GAG1 and Kah-GAG2). Both constructs induced miR-279 family TDMD comparably to that of the WT trigger, suggesting unexpected conformational flexibility within this trigger (Figure 5B, compares lane 1 to lanes 6/7). As before, these reporters were able to reduce the abundance of miR-9b comparably to Kah-WT (Figure 5E). Together, these data suggest the Kah-279 trigger requires the 3′ base pairs to induce miR-279 family decay, and the loss of these base pairs may induce miRNA-mediated silencing.

Structural probing into the Kah trigger cluster

In our efforts to identify the ideal cell lines for Kah reporters, we considered generating reporters with mutated sgRNA-binding sites (Kah-sgMut) in the regions flanking the Kah-279 trigger to allow for its expression in Kah KO cell lines (Figure S5D). These mutations increased reporter expression in KO cells (Figure S5E) but rendered the Kah-279 trigger ineffective, though the Kah-9b trigger remained active (Figure S5F).

Puzzled by these data, we next considered how regions surrounding the Kah triggers may potentiate their efficacy. To address this, we probed into the endogenous secondary structure of the Kah 3′ UTR using the selective 2′ hydroxyl acylation analyzed by primer extension (SHAPE) method.^48–50 SHAPE uses a base-modifying agent (e.g., 2-methylnicotinic acid imidazolide “NAI”) to modify single-stranded RNA regions in live cells^48–50 (Figure S6A). These modifications introduce mutations during reverse transcription of the complementary DNA (cDNA). Since the full length of the Kah 3′ UTR exceeds 1,000 nucleotides, we took advantage of the SHAPE-competent reverse transcriptase MarathonRT popularized by Anna Pyle’s group who demonstrated its ability to reverse transcribe lower abundance RNAs >2.5 kb.⁵¹ We generated Kah-specific SHAPE libraries of ∼1,150 nucleotides excluding adapter sequences (Figures S6A and S6B). From these data, we performed SHAPE analysis of Kah using SHAPEmapper2 and observed a clear trend of increased SHAPE-induced mutations in the NAI-treated cells vs. DMSO control (Figure S6C). This mutational profile was used in combination with minimum free-energy calculations to generate a consensus structure of the Kah 3′ UTR (Figures S6D and S7).

We found Kah to be fairly SHAPE reactive, indicating that the 3′ UTR is likely primarily linear with certain areas containing structured elements (Figures S6D and S7), consistent with a previous report demonstrating highly diverse SHAPE reactivities in mRNAs compared to more uniformly structured non-coding RNAs.⁵² Using this structure, we considered the local secondary structure present at both the miR-279 family and miR-9b trigger. Intriguingly, either trigger was flanked by highly structured sequences, but the seed-binding region was exposed in single-stranded regions (Figures S6D). To investigate the contribution these surrounding sequences may have on Kah-279 efficacy, we first generated a new reporter lacking the native sequences flanking each trigger (Kah-Short) (Figures S8A). Interestingly, Kah-Short lost the ability to induce both miR-279 family and miR-9b degradation and reduced the reporter transcript abundance (Figures S8B–S8D, lane 3). These data mirror what we observed in mammalian TDMD triggers: loss of flanking sequences rendered the construct ineffective.²⁹

We decided to further interrogate if trigger accessibility and/or trigger adjacent sequences may prime the Kah-279 trigger for TDMD. We generated three new reporters, one placing the Kah-279 trigger within an accessible loop atop a stable hairpin (Kah-Loop, S8A, S8E) and two constructs deleting/disrupting secondary structures within conserved regions of Kah (Figures S8A and S8F). The Kah-Loop trigger impaired miR-279 family turnover, suggesting that accessibility of the trigger may be insufficient to predict efficacy (Figure S8B, lane 4). Similarly, deletion to the upstream secondary structures (Kah-Del1, a region that overlaps with the sequences that were mutated in our ineffective Kah-sgMut reporter) (Figures S5D–S5F and S8F) also reduced efficacy of miR-279 family TDMD (Figure S8B, lane 5). Altering sequences adjacent to Kah-279 in Kah-Loop and Del1 reduced efficacy, suggesting that regulatory elements may extend beyond predicted base-pairing (Figures S8E and S8F). In line with this, the second deletion reporter (Kah-Del2) removed predicted hairpin structures ∼60 nt away from the Kah-279 trigger, and this reporter remained similarly effective to WT (Figures S8B [lane 6], S8C, S8D, and S8F). Interestingly, the Kah-Loop and Del1 mutations appear to repress reporter expression, whereas Del2 behaves similarly to WT (Figure S8C). These data indicate that trigger-proximal sequences may aid in converting canonical miRNA target sites into TDMD triggers, suggesting that TDMD triggers may rely on adjacent features for full efficacy.

DISCUSSION

What are the qualities that make an effective TDMD trigger? TDMD research has long emphasized triggers with seed-matching and extensive 3′ complementarity, separated by a central bulge to prevent slicing.^26,31 More recently, non-canonical triggers with minimal or no 3′ pairing²⁰ have emerged (Figure 6A), suggesting additional, though unknown, features that distinguish them from canonical targets.

Figure 6. Proposed model of canonical/non-canonical TDMD triggers and clustered TDMD.

(A) Schematic of proposed trigger classifications: “canonical” triggers with extensive 3′ complementarity; the mammalian TDMD complex is shown since all known mammalian triggers currently belong to this group. “Non-canonical” triggers may require “minimal” 3′ complementarity or the hypothesized “seed-sufficient” triggers that may require no 3′ base-pairing at all. A representative image of the Drosophila and C. elegans TDMD complex is shown since Kah-279, and the hypothesized miR-35 family trigger fits this classification.

(B) Summary and model of clustered TDMD.

(C) Overview of pri-miRNA transcripts regulated by Kah: pri-miR-279∼996, pri-miR-9c∼9b, and pri-miR-309∼6–3. Dora-sensitive miRNAs marked with black triangles; colors denote miRNA families (green, miR-279; orange, miR-9; cyan, miR-3).

The miR-279 3′ end interaction with guide nucleotide 15–17 (g15–17) that we observe with Kah is reminiscent of and partially overlaps with the canonical 3′ supplement (g13–16).¹ We found that this “supplement-like” base-pairing with the Kah-279 trigger only requires three base pairs to induce decay of miR-279 (Figure 5). There are several reports that point out the 3′ end binding sequences, and supplement base pairs may aid in miRNA-target recognition in part by competing with local target secondary structure.^35,53–55 To what degree 3′ end complementarity is required for an AGO TDMD-competent conformational change, and not simply stabilizing association and/or outcompeting other miRNA targets, has still not been thoroughly investigated. Many studies reporting in vitro miRNA-targeting assays have highlighted a reduction in AGO association, with targets containing only a seed match with no 3′ complementarity.^53,56,57 Additionally, the list of validated triggers has grown considerably since the first crystal structures of an AGO-trigger complex were reported.¹⁹ While challenging, it may be crucial to reassess these findings by generating new structures spanning the validated triggers, canonical and non-canonical alike, to identify common structural rearrangements, if any, upon trigger association.

Here, we found that the minimal miR-279 3′ base-pairing with Kah was required for TDMD, but not all isoforms were directed for turnover evenly. The shortest miR-279 isoforms were degraded to a larger extent compared to longer isoforms, suggesting that miRNA length may be a contributing factor for TDMD of certain miRNAs. Interestingly, given that the miR-279 base pairs with Kah end at g17, and isoforms approaching g18 were turned over more quickly, there is the potential that the 3′ end base-pairing proximity to the terminal nucleotide may contribute to TDMD. The CYRANO:miR-7 interaction, with ∼14 bp of 3′ complementarity, promotes TDMD of the unusually long miR-7 (∼24 nt), likely exposing its 3′ end for tailing by Gld2 (TENT2 and TUT2).¹² The Kah trigger cluster does not appear to readily induce tailing of either the miR-279 family or miR-9b, despite extensive 3′ pairing for miR-9b. Kah trigger removal led to accumulation of shorter isoforms for both the miR-279 family and miR-9b. These results highlight that tailing and trimming may be exceptionally dynamic based on the miRNA, trigger RNA, 3′ end base pairs, and the model system used for study.

Despite recent successes in identifying endogenous triggers, there are still many left to be found. AGO-CLASH and similar methods appear to be a powerful tool for capturing trigger RNAs interacting with their cognate miRNAs. Here, we established that highly abundant miRNA-target hybrid reads in our Ago1-CLASH datasets were predictive of bona fide triggers, aiding in our screen for non-canonical triggers within Kah. These results align with other work suggesting TDMD trigger abundance contributes to efficacy.²¹ Interestingly, canonical triggers were the most abundant hybrids in Ago1-CLASH, while Kah:miR-279/996 hybrids were less abundant (Figure S1D). However, the Kah:miR-279/996 interactions were the eighth and fourth most abundant mRNA hybrid, respectively (Figure S1D). These results may indicate that while CLASH is useful for the identification of triggers, both canonical and non-canonical, there may be a bias toward the extensively base-pairing canonical triggers in CLASH data. Validation of the non-canonical Kah-279 trigger revealed that the Kah transcript contains a cluster of triggers, each inducing miRNA degradation independently of the other (Figures 4 and 6B). Since Kah-279 is sufficient to induce decay of all the entire miR-279 family, it was more potent in its ability to influence miR-279 targets compared to miR-9 targets upon loss of Kah-mediated TDMD (Figures 3D–3F). These results highlight how clustered TDMD provides an additional layer of complexity to the tissue-specific gene regulation, in that tissues expressing Kah may also express the miR-279 family and miR-9b, only one of these miRNAs, or none of these. Such combinations would imply that Kah could simply act as an mRNA, a trigger, or a trigger cluster depending on the context. The idea of clustered TDMD also highlights how it was recently shown that the primary transcripts of miRNAs directed for TDMD in mammals preferentially organize into clusters, with TDMD acting as a tool for cells to augment the abundance of select miRNAs derived from polycistronic transcripts.^24,25 Interestingly, Kah induces decay of the miR-279 family in addition to miR-9b, suggesting Kah post-transcriptionally augments the abundance of several miRNAs derived from three distinct polycistronic transcripts (Figure 6C).

A prior study using synthetic clustered triggers reported reduced efficacy,¹³ suggesting that trigger number and spacing may require optimization. Consistently, we found that several triggers require flanking regions for them to be effective²⁹ (Figures S5D–S5F and S8). More broadly, expanding the list of endogenous triggers may aid in the development of potent synthetic triggers able to induce decay of individual miRNAs, entire miRNA families, or several miRNA families via clustered TDMD. Such tools would be key for probing the influence of specific miRNAs during development and disease alike.

Limitations of this study

The research described here uses only a single Drosophila cell line (S2) for all in-cell experiments, potentially obscuring any cell-type-specific effects and limiting broader applicability. While we provide a detailed molecular analysis of the Kah trigger cluster in S2 cells, how integral the TDMD of these miRNAs are to the Drosophila life cycle was not examined. Interestingly, Kah (Kahuli) is predominantly expressed in the Drosophila mesoderm during development,⁵⁸ but the miR-279 and miR-9 families are often neuronal.^59–62 There are reports of the miR-279 family being integral to the development of Drosophila sensory organs, with miR-279 and 996 likely functioning redundantly.^60,61 In line with this, we find that not only do miR-279 and 996 target redundantly but they also appear to be concurrently degraded. Further in vivo analyses would be key in providing a spatiotemporal view of Kah expression, miR-279 family, and miR-9b TDMD and any phenotypic consequences therein.

An unexpected observation made here was reduced Kah levels upon trigger KO. While it appears that perhaps mutations in and around the Kah-279 trigger limit its ability to induce miRNA decay and potentially convert the trigger to a canonical miRNA target site, it is unclear how this switch may occur. In a similar vein, the contribution of RNA structural features to TDMD efficacy is not well understood. Here, we report only a handful of the many potential permutations that may yield insight into the specific regulatory elements surrounding triggers. More targeted studies using a combination of mutations and RNA structural probing methods would aid in the discovery of regulatory elements and aid in the discovery and design of potent TDMD triggers.

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Mingyi Xie (mingyi.xie@ufl.edu).

Materials availability

All unique/stable reagents generated in this study are available from the lead contact without restriction.

Data and code availability

• All original RNA-seq (e.g., sRNA-seq, mRNA-seq, SHAPE) data have been deposited to the NCBI Sequence Read Archive (SRA) database under SRA accession number PRJNA1189499.

• The scripts used to analyze various data are available at https://github.com/UF-Xie-Lab/TDMD-in-Drosophila and https://zenodo.org/records/7737958.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact, Mingyi Xie, upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell lines and cell culture

Drosophila S2 cells (sex: male) were maintained in Schneider’s Insect Medium (Sigma, S9895) supplemented with 10% heat-inactivated FBS (Cytiva, SH3054103HI) and 1% penicillin-streptomycin (Gibco, 15140163) at 28^◦C. Cells were passaged 1:5 every 4–5 days when the culture reaches confluency. Cells tested negative for mycoplasma upon arrival to the lab.

METHOD DETAILS

Ago1-CLASH prediction of non-canonical Kah triggers

All miRNA-RNA hybrids were extracted from previously reported Ago1-CLASH data.³⁰ These hybrids were summarized based on miRNA and target sequences, target RNA type (e.g., mRNA, rRNA, lncRNA), and mean abundance of this hybrid in Scramble and Dora-KO cell replicates (Tables S1A and S1B). The scripts used to perform these analyses are available on GitHub (https://zenodo.org/records/7737958; https://github.com/UF-Xie-Lab/TDMD-in-Drosophila). From this summary, hybrids were manually reorganized based on miRNA identity (e.g., miR-999 hybrids, miR-279 hybrids), and sorted based on mean hybrid abundance in either Scramble or Dora-KO cells. Hybrid enrichment in the Dora-KO was assessed by taking each miRNA hybrid cohort (e.g., miR-279 hybrids) and determining the overall proportion each individual hybrid occupied within the cohort to account for miRNA change in abundance following Dora-KO. Fold changes between the Scramble and Dora-KO were compared to assess hybrid enrichment (Table S1A). The abundance of individual hybrids in Dora-KO, as this data should contain the most abundant Dora-sensitive miRNA-RNA hybrids, was similarly reorganized into miRNA cohorts. The overall proportion each individual hybrid occupied within the cohort was calculated, with non-mRNA hybrids being excluded to remove spurious miRNA hybrids (e.g., miRNA-miRNA, miRNA-rRNA, or miRNA-tRNA hybrids) (Table S1B).

Plasmid construction

Drosophila gene-specific knockout plasmids were generated using the pAc-sgRNA-Cas9 (Addgene, 49330) vector. Three sgRNA expression plasmids were used in pairs flanking the Kah-279 trigger sequence. sgRNA sequences inserted into the expression vector are as described in the supplementary table (Table S1C). For GFP and GFP-Kah expression vectors, the pAC-sgRNA-Cas9 (Addgene, 49330) plasmid was used as a backbone wherein the Cas9 CDS was swapped for GFP with or without insertion of the Kah 3′ UTR. The Kah 3′ UTR was amplified by PCR from genomic DNA extracted from S2 cells using standard organic nucleic acid extraction. Mutants at specific Kah locations were introduced via overlap extension PCR. Primers for the construction of these vectors are listed in the supplementary material (Table S1C).

Transfection and stable cell line generation

Transfection of S2 cells was performed according to the manufacturer’s protocol using either lipofectamine 3000 (Invitrogen, L3000015), or the SF Cell Line 4D X Kit (Lonza, V4XC-2024) on the 4D-Nucleofector X Unit (Lonza, AAF-1003X) platform using the default S2 cell settings. All transfections were performed using 1 μg plasmid/1×10⁶ S2 cells and allowed to grow for 72 h following transfection before being collected. GFP signal was captured from reporter-transfected S2 cells using either an EVOS FL (Invitrogen, AMF4300) or EVOS M5000 (Invitrogen, AMF5000). Stable lines were generated by antibiotic selection using 5 μg/mL puromycin (Gibco, A1113803) for at least 3 weeks. Genomic DNA from these stable cells were extracted and subjected to high throughput amplicon sequencing of the Kah locus. Libraries were sequenced on the Illumina MiSeq by the University of Florida Interdisciplinary Center for Biotechnology Research (ICBR). Allele frequency of these stable cells were analyzed via CRISPResso2⁶⁸ using a 50% minimum homology for alignment, 30 bp quantification window, and a minimum single base quality score >20. A summary of the allele frequency observed in Kah-279 KOs is available in Table S1D.

Morpholino oligonucleotide treatment

S2 cells were passaged normally prior to introduction of custom vivo-Morpholino oligos (GeneTools, Table S1C) which was performed as described previously.³⁰ Briefly, 6×10⁶ S2 cells/well were seeded into 6-well plates and incubated with either non-target, Anti-279-trigger, or Anti-9b-trigger vivo-morpholinos at either 5 or 10 μM. The cells were then cultured for additional 48 h, at which point cells were collected for total RNA extraction.

RNA isolation and RT-qPCR

To collect total RNA, cells were pelleted by centrifugation at 300xg for 2–3 min, washed once with 1X phosphate buffer saline (PBS) (Gibco, 10010023) and the cell pellets were resuspended in and extracted using the TRIzol Reagent (Invitrogen, 15596018) following the manufacturer’s protocol. For RT-qPCR experiments, cDNA was generated from total RNA samples using the QuantiTect Reverse Transcription Kit (QIAGEN, 205313). qPCR experiments were performed using SsoAdvanced Universal SYBR Green Supermix (BioRad, 1725275) and data was normalized to Actin. For miR-qPCR experiments, miRNA cDNA was generated using made-to-order TaqMan miRNA Assay kits (ThermoFisher, 4440886) and the TaqMan MicroRNA Reverse Transcription kit (ThermoFisher, 4366596). qPCR was performed using Taqman Universal Mastermix II with UNG (ThermoFisher, 4440038). Quantified miRNAs were normalized to D. melanogaster U27 or bantam. All primers used in this study are listed in Table S1C.

Northern blot

Near-infrared northern blots were performed as previously described.^29,30,42 Briefly, 10–20 μg of total RNA per sample was separated on a 20% denaturing polyacrylamide 7M urea gel. RNA was transferred (semi-dry) to the Nytran N (Cytiva, 10416196) nylon membrane. Blots were then chemically crosslinked to the membrane via 1-Ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride (EDC) (Thermo Scientific, 22981). Crosslinking reagents were then rinsed off the membrane with water, and preincubated in ExpressHyb hybridization buffer (Takara, 636833). Blocked membranes were then incubated with IR-dye labeled antisense oligonucleotide probes or azide-labeled oligonucleotides which can be conjugated to IR dyes (Table S1C) against the desired RNAs for at least 6 h. IR signal was captured with an Amersham Typhoon (Cytiva, 29238583) and images were analyzed with ImageQuant TL (v7.0).

RNA-seq library preparation

Total RNA samples were depleted of genomic DNA contamination using Turbo DNase (Invitrogen, AM1907) according to the manufacturer’s protocol. RNA quality was assessed via Agilent TapeStation (Agilent, G2992AA) on RNA ScreenTapes (Agilent, 5067–5576). RNA samples with an RNA integrity number (RIN) above 9 were used for library preparation. RNA sequencing was outsourced to MedGenome for library prep using the KAPA mRNA HyperPrep kit (Roche, 08098123702) and subsequently sequenced on the Illumina NovaSeqX Plus platform.

AQ-seq library preparation

AQ-seq small RNA libraries were generated as described previously.³⁸ Briefly, 10–20 μg of total RNA per sample was mixed with 1 μL 3.33 nM synthetic spike-in small RNAs and was size selected for small RNAs on a 15% polyacrylamide urea gel. Gel purified small RNAs are ligated to a pre-adenylated 3′ adapters with RNA Ligase 2 KQ (NEB, M0373L). Ligated small RNAs were again size selected via urea-PAGE. 3′ ligated small RNAs were then ligated to the 5′ adapter with RNA Ligase 1 (NEB, M0437M). Following the final ligation, small RNAs were directly reverse transcribed using SuperScript III Reverse Transcriptase (Invitrogen, 18080085). cDNA libraries were amplified using NEBNext High-Fidelity 2X PCR Master Mix (NEB, M0541L) and library sizes/concentrations were estimated via Agilent Tapestation (Agilent, G2992AA) on DNA ScreenTapes (Agilent, 5067–5582). Libraries were sequenced either by Admera Health or the University of Florida ICBR. Adapters, primers, and spike-in sequences are listed in Table S1C.

In vivo RNA SHAPE modification

In vivo S2 RNA modification was performed as previously described with minimal modifications.⁵¹ Briefly, WT S2 cells were passaged normally for 2–4 days prior to collection to ensure cells were growing in log phase. Per SHAPE replicate, 3×10⁷ cells were collected and resuspended in the SHAPE modification mixture (0.4 U/μL SUPERaseIn [Invitrogen, AM2694], 200 mM NAI [Millipore Sigma, 03–310]) brought up to 3 mL total volume with 1X PBS. Control samples were collected in tandem with the NAI volume being swapped for 100% LC-MS grade DMSO (Thermo Scientific, 85190). Cells were incubated in the mixture at room temperature (∼24^◦C) for 10 min turning constantly (10 RPM), pelleted at 250xg for 2 min, the supernatants were removed, and pellets were resuspended in 2 mL TRIzol Reagent (Invitrogen, 15596018) per replicate. Biological replicates were performed on separate days using fresh materials. Total RNA from these cells was collected as previously described.

Poly-A RNA selection

To enrich Kah, poly-A selection was performed on SHAPE-modified (NAI) and control (DMSO) RNA samples using Dynabeads Oligo (dT)₂₅ (Invitrogen, 61005) using the manufacturer’s protocol with minor modifications. 225 μg total RNA (per condition/replicate) was resuspended up to 300 μL with 10 mM Tris-HCl, pH 7.5. 300 mg of beads were used per condition/replicate. Beads were washed with 1 mL of Binding Buffer (20 mM Tris-HCl, pH 7.5; 1.0M LiCl; 2 mM EDTA), separated onto a magnet stand, and the supernatant was removed. Beads were then resuspended with Binding Buffer equal to the original volume of beads aliquoted. An equal volume of Binding Buffer was added to diluted total RNA (300 μL), and briefly mixed. Washed beads were added to total RNA and mixed by gentle agitation briefly. Bead/total RNA mixture was incubated in a pre-heated thermomixer set to 80^◦C for 3 min, then allowed to cool down slowly to 37^◦C (∼10–15 min). Tubes were then placed onto magnet stand, and poly-A depleted supernatant was discarded. Beads were washed with 600 μL of Washing Buffer B (10 mM Tris-HCl, pH 7.5; 0.15M LiCl; 1 mM EDTA) by pipetting. Tubes were placed on the magnet stand and washed again, with the supernatant discarded. Beads were then resuspended in 60 μL of 10 mM Tris-HCl, pH 7.5. The samples were then heat-denatured at 77^◦C for 2.5 min to elute poly-A RNAs off the beads. Eluate was separated from the beads on a magnet stand, eluted poly-A RNAs were ethanol precipitated and monitored via high-sensitivity RNA ScreenTape (Agilent, 5067–5579) before continuing with SHAPE library preparation.

Long-read Kah SHAPE library construction

To generate the cDNA for long-read Kah SHAPE libraries, the UltraMarathon Reverse Transcriptase (uMRT) kit (RNAConnect, N/A) was used according to the manufacturer’s recommendations. In brief, a fresh uMRT master mix (50 mM Tris-HCl, pH 7.5; 200 mM KCl, 1 mM MnCl₂, 20% Glycerol, 0.25 μM Kah-RT primer, 0.25 mM dNTPs, 1U/μL uMRT Enzyme, 1U/μL SUPERaseIn, 2.5 mM DTT) was mixed with the Poly-A RNA samples and allowed to reverse transcribe for 3 h at 42^◦C. cDNA was cleaned up from short primers via 1X AMPureXP (Beckman Coulter, A63881) according to the manufacturer’s protocol. cDNA was eluted and amplified using NEBNext High-Fidelity 2X PCR Master Mix (NEB, M0541L) and indexed primers until sufficient material was obtained for PacBio library construction. Library size and concentration was monitored via D5000 ScreenTape (Agilent, 5067–5588). PCR amplicons were sent to the University of Florida ICBR for PacBio SMRTbell library construction and sequenced on the PacBio Sequel IIe system.

QUANTIFICATION AND STATISTICAL ANALYSIS

Northern blot quantification

Northern blot images were quantified using ImageQuant TL (v7.0). In brief, raw microRNA signal was quantified, and the background signal was subtracted from each value. Relative microRNA quantification was then normalized to the bantam loading control. Normalized values were then compared to the negative control lane. Blots with ±values represent the normalized signal standard deviation, or representative bar graphs with SEM, between 3 biological replicates. Significance was determined via Student’s t-tests, as described in figure legends.

RNA-seq analyses

The RNA-seq differential expression analyses follow a standard pipeline. After adapter trimming using Cutadapt,⁶³ the clean reads were aligned to the reference genome using HISAT2 to generate a mapping file.⁶⁴ Gene-level read counts are then computed using the prepDE.py3 script. Subsequently, the read counts were input into the DESeq2 software for differential expression analyses, enabling the identification of differentially expressed genes (DEGs) between conditions.⁶⁵ For miRNA family target analyses, these data were categorized based on conserved miRNA targets predicted through TargetScanFly (v7.2).⁴⁵ Top conserved targets were separated from all conserved targets based on a criterion of < −.40 conservation context score.

AQ-seq analyses

The analysis of AQ-seq data began with preprocessing raw sequencing reads using Cutadapt to remove adapter sequences.⁶³ The processed reads were collapsed using FASTX Collapser to group identical sequences and reduce PCR-introduced redundancy. Following this, random 4-nucleotide sequences, unique molecular identifiers (UMIs), at both the 5′ and 3′ ends are trimmed to ensure accurate identification of miRNAs. Differentially expressed miRNAs were determined via input of raw miRNA counts into DEseq2.⁶⁵ For miRNA isoform quantification, isoforms were determined by aligning at least 18 nucleotides (nt) of the collapsed reads to a reference miRNA database. The count of such matching reads was used to quantify miRNA expression levels and their isoforms.

SHAPE-MaP analysis

PacBio Kah-specific SHAPE libraries were demultiplexed from one another and converted from .bam to .fastq files using isoseq3. PCR duplicates were collapsed based on a 6-nucleotide UMI included in the Kah-specific SHAPE RT primer as described previously. Deduplicated reads were used as inputs for SHAPE-MaP analysis with SHAPEmapper2 (v2.3)⁶⁷ referencing the Kah sequence amplified. Nucleotide SHAPE reactivities were calculated via SHAPEmapper2, with nucleotides not mapping to the reference being set to 0. SHAPEmapper2-generated consensus structures (.ct) were used as the reference for structure generation in RNAcanvas/RNA2-drawer.⁶⁶ SHAPE reactivities <0.5 were considered non-reactive, 0.5–1 moderately reactive, and >1 highly reactive. All sequencing data for this study have been uploaded to the NCBI Sequence Read Archive (SRA) database under SRA accession number PRJNA1189499.

Supplementary Material

SUPPLEMENTAL INFORMATION

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2025.116162.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
TRIzol reagent	Thermo Fisher	Cat# 15596018
Puromycin	Gibco	Cat# A1113803
NAI	Millipore Sigma	Cat# 03–310
PBS	Gibco	Cat# 10010023
SsoAdvanced Universal SYBR Green Supermix	Biorad	Cat# 1725274
EDC Crosslinking reagent	Thermo Scientific	Cat# 22981
ExpressHyb hybridization buffer	Takara	Cat# 636833
Taqman Universal Mastermix II with UNG	ThermoFisher	Cat# 4440038
Turbo DNase	Invitrogen	Cat# AM1907
RNA Ligase 2 KQ	NEB	Cat# M0373L
RNA Ligase 1	NEB	Cat# M0437M
SuperScript III Reverse Transcriptase	Invitrogen	Cat# 18080085
NEBNext High-Fidelity 2X PCR Master Mix	NEB	Cat# M0541L
SUPERaseIn	Invitrogen	Cat# AM2694
DMSO	Thermo Scientific	Cat# 85190
Critical commercial assays
TaqMan MicroRNA Reverse Transcription kit	ThermoFisher	Cat# 4366596
QuantiTect Reverse Transcription Kit	QIAGEN	Cat# 205313
AMPure XP Beads	Beckman-Coulter	Cat# A63881
Lipofectamine 3000	Invitrogen	Cat# L3000015
SF Cell Line 4D X Kit	Lonza	Cat# V4XC-2024
TaqMan miRNA Assay	ThermoFisher	Cat# 4440886
RNA ScreenTapes	Agilent	Cat# 5067–5576
KAPA mRNA HyperPrep kit	Roche	Cat# 08098123702
D1000 DNA ScreenTapes	Agilent	Cat# 5067–5582
Dynabeads Oligo (dT)25	Invitrogen	Cat# 61005
UltraMarathon Reverse Transcriptase kit	RNAConnect	Cat# R1002M
D5000 ScreenTape	Agilent	Cat# 5067–5588
Deposited data
The raw and processed sequencing data	This paper	SRA: PRJNA1189499
Experimental models: Cell lines
Drosophila Schneider 2 (S2) cells	Laboratory of David Bartel	N/A
Oligonucleotides
RT-qPCR primer sequences	This paper	Table S1C
Northern blot probe sequences	This paper	Table S1C
Primer sequences for SHAPE-MaP library	This paper	Table S1C
Vivo-morpholinos	GeneTools	Table S1C
Recombinant DNA
pAc-sgRNA-Cas9	Addgene	Cat# 49330; RRID: Addgene_49330
pAc-EGFP	This paper	N/A
pAc-EGFP-Kah-WT	This paper	N/A
pAc-EGFP-Kah-noPAMs	This paper	N/A
pAc-EGFP-Kah-MutA	This paper	N/A
pAc-EGFP-Kah-MutB	This paper	N/A
pAc-EGFP-Kah-sgMut	This paper	N/A
pAc-EGFP-Kah-Seed	This paper	N/A
pAc-EGFP-Kah-3′ Mut	This paper	N/A
pAc-EGFP-Kah-2nt Mut	This paper	N/A
pAc-EGFP-Kah-GAG1	This paper	N/A
pAc-EGFP-Kah-GAG2	This paper	N/A
pAc-EGFP-Kah-Short	This paper	N/A
pAc-EGFP-Kah-Loop	This paper	N/A
pAc-EGFP-Kah-Del1	This paper	N/A
pAc-EGFP-Kah-Del2	This paper	N/A
Software and algorithms
Prism 9	Graphpad Software	https://www.graphpad.com/scrientifificsofware/prism/
SnapGene	SnapGene Software	https://www.snapgene.com/
ImageQuant TL (v7.0)	Cytiva ImageQuant Software	https://info.cytivalifesciences.com/image-analysis-software.html
Cutadapt	Martin et al.63	N/A
HISAT2	Kim et al.64	N/A
DESeq2	Love et al.65	N/A
RNA2Drawer	Johnson et al.66	https://rna2drawer.app/
SHAPEMapper2	Busan et al.67	N/A
Custom scripts	Sheng et al.30	https://github.com/UF-Xie-Lab/TDMD-in-Drosophila
Other
Nytran N blotting membrane	Cytiva	Cat# 10416196
Amersham Typhoon	Cytiva	Cat# 29238583
Agilent TapeStation	Agilent	Cat# G2992AA
EVOS M5000	Invitrogen	Cat# AMF5000

Highlights.

• Ago1-CLASH identifies a non-canonical TDMD trigger in Kah

• The Kah transcript contains a cluster of multiple TDMD triggers

• Triggers within the Kah trigger cluster induce decay independently of one another

• Minimal 3′ end base-pairing is required for miR-279 family degradation