Widespread 3'UTR capped RNAs derive from G-rich regions in proximity to AGO2 binding sites

doi:10.21203/rs.3.rs-4809688/v1

Download PDF

Research Article

Widespread 3'UTR capped RNAs derive from G-rich regions in proximity to AGO2 binding sites

https://doi.org/10.21203/rs.3.rs-4809688/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

The 3’ untranslated region (3’UTR) plays a crucial role in determining mRNA stability, localisation, translation and degradation. Cap analysis of gene expression (CAGE), a method for the detection of capped 5’ ends of mRNAs, additionally reveals a large number of apparently 5’ capped RNAs derived from locations within the body of the transcript, including 3’UTRs. Here we provide direct evidence that these 3’UTR-derived RNAs are indeed capped and widespread in mammalian cells. By using a combination of AGO2 enhanced individual nucleotide resolution UV crosslinking and immunoprecipitation (eiCLIP) and CAGE following siRNA treatment, we find that these 3’UTR-derived RNAs likely originate from AGO2-binding sites, and most often occur at locations with G-rich motifs bound by the RNA-binding protein UPF1. High-resolution imaging and long-read sequencing analysis validate several 3’UTR-derived RNAs, showcase their variable abundance and show that they may not co-localise with the parental mRNAs. Taken together, we provide new insights into the origin and prevalence of 3’UTR-derived RNAs, show the utility of CAGE-seq for their genome-wide detection, and provide a rich dataset for exploring new biology of a poorly understood new class of RNAs.

3’UTR

CAGE

capping

AGO2

UPF1

3’UTR-derived RNA

G-rich

subcellular localisation

In all eukaryotes, mRNA molecules contain an evolutionarily conserved m7G cap (N7-methylated guanosine), which is incorporated at the 5’ end of nascent transcripts. Co-transcriptional capping is the first modification made to nascent RNA in the nucleus, which protects it from exonuclease cleavage while promoting cap-related biological functions such as pre-mRNA splicing, polyadenylation and nuclear export [1]. In addition to the co-transcriptional capping, there is also evidence for a post-transcriptional capping mechanism, in which an m7G cap is added to newly exposed 5’ ends of RNA fragments created upon endonucleolytic cleavage or decapping [2–5]. However, little is known about the extent and biological role of this post-transcriptional capping, and of its relation to other post-transcriptional RNA processing mechanisms.

Cap analysis of gene expression and deep-sequencing (CAGE-seq) was originally designed to precisely determine transcription start site (TSS) positions by capturing and sequencing 5’ ends of capped mRNA transcripts, and it can also be used to measure gene expression [6]. However, several studies have detected the unexpected, reproducible, and so-far unexplained presence of CAGE signals (~ 10–15% of total reads) and/or an enrichment of RNA-seq reads mapping to 3’UTRs, far away from the usual TSS [7–15]. Previous studies have shown an absence of active promoter marks (i.e. no enrichment histone modifications or RNA polymerase II (RNAPII) occupancy) around these 3’UTR signals [7, 9, 16], arguing against the possibility that they are unannotated transcription start sites. Moreover, the expression of some of these capped 3’ UTRs is tissue-specific and regulated in mouse embryonic development, whilst their subcellular localization can be separated from the associated protein-coding transcript, suggesting that their generation is a regulated process [16]. In addition, specific isolated 3'UTRs have been implicated in a growing number of physiological processes, such as cell signalling or oxidative stress [7, 17, 18]. Moreover, several capped 3’UTRs have been reported to play important roles in regulating protein expression in trans, similar to long non-coding RNAs [7, 9, 10]. Truncated mRNAs and RNA decay intermediates can be subject to post-transcriptional, cytosolic capping [2, 3, 15] and it has been suggested that a similar mechanisms may underlie the generation of some 3’UTR CAGE signals, referred to as 3’UTR-associated RNAs (uaRNAs) [16]. To avoid potential misunderstandings we refer to these as 3’UTR-derived RNAs, as these newly generated RNAs are not known to be physically associated with 3’ UTRs.

Here, we thoroughly examine the presence of these 3’UTR-derived RNAs across the transcriptome, and the molecular basis of their generation and characteristics. We perform a genome-wide identification of 3’ UTR-derived RNAs based on their capped 5’ ends, and proceed to investigate the mechanisms involved in their formation. To this end, we combine CAGE, RNA-seq and cross-linking immunoprecipitation (CLIP)-based techniques from ENCODE and FANTOM consortia. We show that 3’UTR-derived RNAs present biochemical properties similar to their 5’ capped counterparts and that they can be as abundant as the protein-coding version of the host transcript, or even more so. We reveal that the 5' ends of 3'UTR-derived RNAs are enriched in G-rich motifs and tend to form strong secondary structures, while the immediately upstream region of these 5' ends is bound by UPF1 and/or AGO2. Moreover, some of these abundant 3’ UTR-derived RNAs exhibit a markedly different subcellular localisation profile than their protein-coding counterparts. Finally we show, for the first time, that capped RNAs can also emerge following mRNA cleavage by small interfering RNAs (siRNAs).

CAGE-seq identifies non-promoter associated capped 3'UTR-derived RNAs

We and others [7–12] have previously reported the presence of CAGE-seq signals outside of annotated promoter regions in thousands of protein-coding genes. However, their origin or biological relevance has not been thoroughly interrogated. Here we first confirmed the prevalence of these signals in human cell lines using CAGE data provided by the ENCODE consortium. As expected, we detected a similar proportion of CAGE signals per genomic region in two different human cell lines (HeLa and K562) and showed that the CAGE signal is highly reproducible across replicates. This included the library size, number of uniquely mapped CAGE reads and distribution of the 5’ CAGE reads mapping to different genomic regions (Fig. 1a, S1a, b, c, d). A similar genomic distribution has also been detected by other groups before, using the same CAGE-seq protocol [19].

The relative intensities of CAGE signal detected at different genomic regions depend on the priming method for reverse transcription (oligo-dT, random hexamers, or mixtures thereof in different ratios) [12]. Oligo-dT priming quantitatively favours shorter transcripts, while the reverse is true for random priming. We subsequently verified CAGE signals within 3’UTRs are most detected when a combination of Oligo-dT and random primers is used, with the optimal inclusion ratio of 1 to 4 ratio of Oligo-dT to random primers [19, 20] (Fig. S1e). Notably, the same ratio was used in ENCODE CAGE samples analysed in this study.

The CAGE signal is the strongest at 5'UTRs of known protein-coding genes [19] (Fig. 1a, ~ 65% of total reads). While low-level non-promoter CAGE signal (sometimes referred to as “exon painting” [4] [21]), can be detected along the entire length of transcripts, the signal at 3’UTRs is consistently present and occurs in localised clusters, similar to CAGE signals at promoters (see Fig. S1l for examples). We focused on the 3’UTR region, which contains a substantial proportion (~ 11%) of the total CAGE reads (Fig. 1a), the implications of which are unknown. To identify robust CAGE signals with sufficient sensitivity, we used a 20nt window requiring at least two 5' reads overlapping from two different replicates for each cell line separately. This revealed 32,065 3’UTR CAGE clusters across all samples (Table S1). As expected, correlation between technical replicates was high (Pearson correlation (PC) > 0.9) for all CAGE signals, independently of genomic location (Fig. S1g-j). Moreover, expression of the 3’UTR CAGE clusters was highly reproducible between HeLa and K562 samples (PC ~ 0.98, Fig. S1f), suggesting biological relevance. Correlation across cell types was much higher for 3’UTR clusters than that for the 5’ UTR CAGE (PC ~ 0.79, Fig. S1g) and CDS CAGE (PC of 0.61, Fig. S1h) signals and comparable to intronic CAGE signal (PC ~ 0.93, Fig. S1i). This may suggest that 3’UTR CAGE signals originate from more stable and/or less tissue specific subset of transcripts. Together these analyses show that the transcripts whose 5’ end map to 3’UTR ends of protein-coding genes are highly reproducible across cell types, and that CAGE is a robust method for their quantitative detection.

3'UTR-derived RNAs are confirmed by RNA-seq, qPCR and long-read CAGE

Next, we aimed to confirm the existence of these 3’UTR capped RNAs using independent methods. First, we asked if these fragments could be identified in transcriptomic (RNA-seq) data. For this we compared the CAGE signal with the RNA-seq signal of two different cell lines. To categorise CAGE peaks we first used the paraclu [22] peak caller to identify clusters of 5’ ends of capped RNAs, and within each cluster we selected the highest signal as the dominant CAGE peak position. For comparison, we processed paired-end RNA-seq data from the same K562 and HeLa cell lines, then plotted read-starts and read-ends relative to the dominant 3’UTR CAGE peak per transcript (Fig. 1b - in blue, and S1J). Both RNA-seq samples showed highly reproducible enrichments of read ends coinciding with dominant 3’UTR CAGE peaks. This reveals that the 3’UTR CAGE peaks are confirmed by the read-ends from reverse-stranded RNA-seq data, which suggests that the signal could be originating from post-transcriptional cleavage sites. Notably, there is also a small enrichment of RNA-seq read-starts downstream from the 3’UTR CAGE peaks, which could represent the same RNA fragments detectable by the CAGE samples (Fig. 1b in yellow). More importantly, these findings demonstrate that 3’UTR capped fragments identified by CAGE can also be detected by other, methodologically independent, high-throughput sequencing methods such as RNA-seq.

We next aimed to confirm the presence of transcripts initiating at the 3’UTR CAGE peaks by an alternative experimental approach, not dependent on RNA library creation or high-throughput sequencing. We focussed on two genes, CDKN1B/p27kip1 (p27) and JPT2, which contain a dominant CAGE peak located within the 3’UTR region, demonstrated with highly reproducible read coverage for CAGE and RNA-seq in both K562 and HeLa cells (see Fig. S1m for example). Separate sets of primers were designed to quantitatively PCR-amplify ~ 150bp sequences within 300 nucleotides upstream and downstream of the 3’UTR CAGE peaks in CDKN1B and JPT2 (see Methods, Fig. S1k). In agreement with CAGE and RNA-seq data (Fig. S1m), RT-qPCR detected higher levels of these transcripts with the downstream primers (Fig. S1K) than with upstream primers. A similar enrichment in RT-qPCR signal (Fig. S1k) was observed with downstream primers in comparison to primers designed to amplify a ~ 150bp region spanning the CAGE peak in CDKN1B, suggesting an accumulation of 3’UTR fragments in comparison to full-length mRNAs.

Treatment of the samples with TerminatorTM 5’-Phosphate-Dependent Exonuclease (TEX), a 5′→3′ exonuclease that digests RNA with a 5′ monophosphate, but not RNA with 5′-triphosphate, 5′-cap or 5′-hydroxyl group had no or little effect on the amount of JPT2 and CDKN1B transcripts detected with primers amplifying either side of the 3’UTR CAGE peak within these cells. This was in sharp contrast with the known uncapped 3’ fragment of SLC38A2 mRNA, previously described by Malka et al. [8], which was, as expected, sharply reduced upon TEX treatment (Fig. 1c). These results lend further support that all the quantified transcripts, including the 3’UTR fragments, are capped.

We further confirmed that 3’UTR-derived RNAs could be detected by long-read Nanopore-sequencing CAGE (Fig. 1d). We were provided with data in cortical neuron samples by the FANTOM6 consortium for 10 genes that contain HeLa and K562 3’UTR CAGE peaks (Fig. 1d, S1n). In all of the 10 examples, the full length read sequencing CAGE identified reads spanning from the start of our identified CAGE 3’UTR peaks till the end of the annotated transcripts, whereas for most of these genes, reads spanning between the 5’CAGE and the 3’CAGE signal were absent (Fig. S1n). These observations suggest that the capped 3’UTR derived RNAs originate from the full length mRNA whilst fragments upstream of the 3’CAGE may not be stable. Notable exceptions are DDX17 and GHITM but it is unclear whether these 3’CAGE upstream sequences result from alternative polyadenylation or are products from the cytosolic cleavage of the full-lenght.

Capped 3’UTR-derived RNAs are evolutionarily conserved and generated post-transcriptionally

We next wanted to investigate whether the 3’UTR CAGE signals originate from post-transcriptionally capped RNA fragments. First, we explored whether there is evidence of nuclear Cap Binding Complex (CBC) binding to the capped 5’ ends of 3'UTR fragments, as this protein is known to bind to 5’ ends of nascent protein-coding mRNA transcripts in the nucleus. Individual-nucleotide resolution UV crosslinking and immunoprecipitation (iCLIP) is a method that identifies protein-RNA crosslinking interactions with nucleotide resolution in a transcriptome-wide manner. We examined CBC-iCLIP data from HeLa cells, where the authors targeted nuclear cap-binding subunit CBP20 protein [23]. CBP20 is a nuclear component of cap-binding complex (CBC), which binds co-transcriptionally to the 5' cap of pre-mRNAs and interacts directly with the m7-G cap [24, 25]. The CBP20 RNA binding data was analysed using a standard iCLIP processing pipeline, where the nucleotide preceding the cDNA-start position after PCR duplicate removal is reported as the crosslinking position (see Methods). The CBP20 crosslinking positions were then screened across all dominant 5’UTR and 3’UTR CAGE peaks per transcript. As expected, CBP20 crosslinks were enriched around the dominant 5’UTR CAGE peaks where the TSS of full-length transcripts is positioned. However, the enrichment was very weak at the non-promoter 3’UTR CAGE peaks (Fig. S2a). This strongly indicates that the 3’UTR capped fragments identified by CAGE are not part of nuclear CBC, further suggesting that they are likely a product of an independent post-transcriptional processing pathway.

To further explore whether the 3’UTR capped fragments were generated co-transcriptionally, we investigated the location of cap signals in nascent RNAs identified in global nuclear run-on sequencing experiments of 5’ capped RNAs (GRO-cap) [26]. As anticipated, strong GRO-cap signals overlapped with CAGE peaks in 5’UTRs and, to a lesser extent, with introns and upstream CDSs but were notably absent around 3’UTR CAGE peaks (Fig. S2b). This also indicates that capping of 3’UTR fragments occurs post-transcriptionally.

Additionally, we analysed capCLIP data from HeLa cells. capCLIP is a version of CLIP that targets the translation elongation factor eIF4E, a cytoplasmic protein which binds the 7-methyl-GTP moiety of the 5′-cap structure of RNAs to facilitate the efficient translation of almost all mRNAs [27, 28]. The capCLIP data was analysed following the same methodology as CBP20-iCLIP. The enrichment of capCLIP signal at the non-promoter 3’UTR CAGE peaks was much stronger than in the CBC-iCLIP (Fig. S2a, S2c), which demonstrates that the cap of the 3’UTR-derived RNAs is strongly bound by cytoplasmic eIF4E, but not the nuclear cap binding protein CBP20, suggesting that these RNAs are predominantly cytoplasmic. Furthermore, we investigated ribosome footprinting data [29] to interrogate whether the 3'UTR-derived RNAs, which are bound by eIF4E, are translated. However, we did not find evidence of ribosomal binding to these RNAs that suggested active translation (data not-shown).

Next, we investigated the evolutionary conservation of 3’UTR-derived RNAs. Utilising UCSC conservation tracks, we computed conservation scores around 3’UTR CAGE peaks. To exclude the influence of coding regions and transcript termination sites, we specifically selected 21,831 3’UTR CAGE peaks positioned at least 150 bps away from the 3’UTR bordering region (≥ 150 bps downstream from CDS and ≥ 150 bps upstream from transcript termination). Remarkably, our findings reveal that the exact 3’UTR CAGE peaks exhibit lower conservation compared to the surrounding regions. However, the region immediately downstream of the 3’UTR CAGE peaks, corresponding to the "body" of 3’UTR-derived RNAs shows a notable enrichment in conservation scores, suggesting a potential functional contribution (Fig. S2d).

Altogether, these analyses confirm the presence of abundant, evolutionarily conserved, capped 3’UTR-derived non-coding RNAs that may originate from cytosolic cleavage of full-length mRNAs.

5' ends of 3'UTR-derived RNAs are enriched for G-rich motifs and strong secondary structures

Next, we wanted to understand the sequence features that distinguish the CAGE peaks corresponding to co-transcriptional capping of TSS from those originating from post-transcriptional capping of 3'UTR-derived RNAs. We first explored the possibility that 3’UTR fragments might be a by-product of nuclear polyadenylation and associated endonucleolytic cleavage. If this were the case, the identified 3'UTR CAGE peaks should be preceded by enrichment of the canonical polyA signal (A[A/U]UAAA hexamers), which recruit the nuclear polyadenylation machinery. However, we only found such enrichment at the annotated 3'UTR ends, and not upstream of the 3'UTR CAGE peaks (Fig. S2e). We observed a notable enrichment downstream of the 3’UTR CAGE peaks (Fig. S2e - red line), which most likely corresponds to the canonical polyA site as some of the 3'UTR-derived RNAs are relatively short and their 5’ ends are close to the annotated 3'UTR ends.

Next, we explored whether there were additional distinctive sequence characteristics between the two types of CAGE peaks. Consistent with previous studies [9, 13], we detected a strong G-enrichment overlapping the 5’ end of the CAGE reads present in non-promoter regions (Fig. 2a, S2f), distinct from the YR dinucleotide characteristic of signals at 5’ ends of genes. More surprisingly, CAGE peaks within the 3’UTR region showed a strong increase in internal pairing probability (see Methods: Secondary structure) in comparison to CAGE peaks in other regions (Fig. 2b, S2g), suggesting that structural preference may be important for the generation of 3’UTR-derived RNAs. Notably, the surrounding (within 100 bps) region of CAGE peaks in 5’UTRs is more structured (light blue line in Fig. 2b, S2g), representing the higher GC content that is present around all 5’UTRs in vertebrates [30], with a distinctive drop at -25 bps coinciding with the canonical TATA box position.

Motifs with G-rich repeats in the transcriptome can form non-canonical four-stranded structures (G4s) implicated in transcriptional regulation, mRNA processing, the regulation of translation and RNA translocation [31]. Similar to web-logo motif analyses of CAGE peaks from different mRNA regions (Fig. 2a, S2f), the nucleotide enrichment plot of GGG sequences showed the highest enrichment overlapping 3'UTR CAGE peaks (Fig. 2c, S2f). This raises the possibility that the sequence around the 3’UTR CAGE peaks may have an increased propensity to form RNA-G4 structures via the canonical G4 motif (G₃-N_1 − 7-G₃-N_1 − 7-G₃-N_1 − 7-G₃) [32]. To further explore the RNA G-quadruplexes formation profile, we integrated RNA-G-quadruplex sequencing (rG4-seq) data from HeLa cells [33] and ran G4-Hunter predictions [34] around CAGE peaks. Both the rG4-seq data (HeLa) and G4-Hunter predictions (K562) showed the highest G4s enrichment around CAGE peaks in the 3'UTR region (Fig. S2h-l) with the highest percentage of rG4-seq hits within 3’UTRs (Fig. S2m). Nevertheless, it is worth noting that the number of 3’CAGE sites overlapping with rG4-seq sites was relatively small (~ 3800 out of ~ 133900). In sharp contrast, 8 of the 10 gene examples with 3’UTR CAGE peaks and validated with long-read CAGE explored here (selected due to highest 3’UTR CAGE signal) contained rG4-seq clusters coinciding with 3’UTR CAGE peaks (Fig. S1n). It is important to note that the determination of whether these sites are genuinely in the G4-folded state remains uncertain, as the rG4-seq method employs G4 stabilisers to artificially enhance G4 structures.

3’UTR CAGE sites are flanked by enriched UPF1 binding

The evidence outlined so far is consistent with our hypothesis that capped 3’UTR-derived RNAs are formed post-transcriptionally. Next, we aimed to determine whether specific RNA-binding proteins (RBPs) were involved in the mechanism of their generation. To that end, we analysed publicly available enhanced CLIP (eCLIP) data for 80 different RBPs in the K562 cell line, produced by the ENCODE consortium [35]. For each RBP, we calculated normalised cross-linking enrichment compared to other RBPs around maximum CAGE peaks per annotated gene region (5’UTR, CDS, intron, 3’UTR). This identified a specific set of RBPs around CAGE peaks, with UPF1 (Up-frameshift protein 1) as the top candidate in 3’UTRs (Fig. 2c), DDX3X (DEAD-Box Helicase 3 X-Linked) in 5’UTRs (Fig. S2n), KHSRP (KH-type splicing regulatory protein) in introns (Fig. S2o), and less protein specific enrichments in CDS with YBX3 (Y-Box-Binding Protein 3) as the top candidate (Fig. S2o,p). UPF1 is involved in a variety of RNA degradation pathways [36], including Nonsense-Mediated Decay (NMD) [37] and the normal mRNA decay where stalled UPF1 at CUG and GC-rich motifs activates decay [38]. KHSRP plays a well-characterised role in pre-mRNA splicing, but has also been involved in several other aspects of RNA biology, such as mRNA decay and editing and maturation of miRNA precursors [39]. On the other hand, the YBX3 has been implicated in regulation of mRNA translation as well as stability, likely in a transcript-dependent manner [40]. As a positive control for our enrichment score approach, we noted DDX3X enrichment around 5’UTR CAGE peaks. This is consistent with known roles for DDX3X in transcription and pre-mRNA splicing through interactions with transcription factors and Spliceosomal B Complexes [41].

Interestingly, the crosslinking of UPF1 is enriched within 20 nt upstream of the 3’UTR CAGE peaks, followed by a steep depletion within ~ 10 bps downstream (Fig. 2d). Additionally, a substantial correlation (R = 0,654) was observed between the 3’UTR CAGE signal and UPF1 binding, but not associated with gene expression or 3’UTR length (Fig. S2q). More specifically, the degree of UPF1 binding coincides with the intensity of the 3’UTR CAGE peaks and proximity to the peaks (Fig. S2r, s). However, transfecting K562 cells with UPF1-targeting small interfering RNAs (siRNAs) for 48 hours did not lead to changes in the enrichment of RT-qPCR signal obtained with primers targeting downstream of the 3’UTR CAGE peak in CDKN1B or JPT2 when compared to upstream-targeting primers (Fig. S2t). Thus, it remains unclear if the precise binding position of UPF1 relative to the re-capping position may be important for the generation of the 3’UTR capped fragments, or if accumulation of UPF1 is an indirect result of the presence of other factors that contribute to the cleavage.

mRNA cleavage by small interfering RNAs generates newly capped RNA fragments

mRNAs can be cleaved post-transcriptionally through RNA interference (RNAi). Indeed, a common way to artificially accomplish gene silencing is to utilise siRNAs to induce endonucleolytic degradation of the target transcripts [42, 43]. SiRNAs are usually 21–23 nt long and their sequence is antisense to their mRNA target sequence. Silencing by siRNAs is induced through the endonuclease activity of Argonaute 2 (AGO2), a subunit of the RNA-induced gene-silencing complex (RISC) in the cytoplasm [44].

We hypothesised that siRNA silencing through AGO2 cleavage could lead to cytoplasmic capping of the cleaved RNA fragments instead of degradation. To test this hypothesis, we first investigated if CAGE-seq could detect cleaved RNA fragments guided by siRNA. We analysed CAGE data from siRNA-treated samples from the FANTOM5 dataset [45], which included samples from the TC-YIK human cell line transfected with siRNAs targeting mRNAs of 28 different transcription factors (20 siRNAs designed by ThermoFisher and 8 by the study authors) and 5 non-targeting control samples, in triplicates. We detected CAGE signal at the exact position targeted by the siRNA in at least two replicates in 20 out of the 28 samples (Fig. 3a). The strongest enrichment in CAGE signal relative to the siRNA target site was detected in the Islet-1 knockdown (ISL1-KD) samples, with no signal detected in control samples (Fig. 3b,c,S3a). More interestingly, the dominant CAGE 5' end signal was present in the middle of the siRNA target sequence (Fig. 3e, S3a), where the AGO2 cleavage is known to take place [46, 47]. As expected, the TSS CAGE signal in the 5'UTR of the corresponding protein-coding gene dropped by ~ 75% compared to the control samples in all 3 replicates (Fig. 3b,c), confirming that the silencing of the ISL1 transcript was efficient. Together these results indicate that siRNA-mediated recruitment of AGO2 can lead to the generation of post-transcriptionally capped RNA fragments following mRNA cleavage.

3'UTR CAGE peaks coincide with AGO2 and UPF1 binding sites alongside G-rich motifs

Since the endonuclease activity of AGO2 facilitates mRNA cleavage guided by siRNAs, we investigated whether AGO2 binding also occurred at the endogenous 3'UTR CAGE peaks. There was no publicly available AGO2 binding data for either HeLa or K562 cells so we produced ‘enhanced individual nucleotide resolution’-CLIP (eiCLIP) [48] data for AGO2 (AGO2-eiCLIP) in HeLa cells. Our analysis revealed that 32.8% of the crosslinking positions mapped to the 3’UTR region (Fig. S3b), with a higher binding enrichment in known microRNA (miRNA)-regulated transcripts, and with a clear miRNA-seed matching-sequence enrichment downstream of the crosslinking site (Fig. S3c,d). Similarly to UPF1, AGO2 crosslinks were enriched immediately upstream from the 3'UTR CAGE peaks but, unlike UPF1, they were not depleted in the downstream region (Fig. 3d, S3e, 3d).

In animals, endogenous RNAi is mainly mediated by microRNAs (miRNAs). MiRNAs are ~ 21–23 nucleotide (nt) long RNAs that, in contrast to siRNAs, recruit the miRNA induced silencing complex (miRISC) containing AGO1-4 to mRNAs with partial sequence complementarity. As a result, miRNA action induces translational repression and/or exonucleolytic cleavage of the target mRNAs [42, 43]. Thus, miRNA-mediated degradation of target mRNAs in animals usually involves deadenylation, decapping, and degradation by the major cytoplasmic 5’to-3’ exonucleases, rather than direct endonucleolytic cleavage by AGO2 [5, 49]. Nevertheless, it has been demonstrated that extensive miRNA-mRNA pairing can also trigger AGO2 catalytic activity [50–52]. We thus hypothesised that AGO2 miRNA-guided cleavage of mRNA targets might lead to the generation of recapping fragments in a similar manner to that observed for siRNAs. To test this we first identified genomic sequences with extensive complementarity (fewer than 2 mismatches) to human miRNAs. We identified 29 such targets that mapped within 3’UTRs but there was no CAGE signal present around any of them (data not shown). In line with this, AGO2 crosslinking enrichment around 3’UTR CAGE signals was considerably weaker for AGO2 binding sites overlapping with predicted miRNA binding sites (see Methods, Fig. S3f). Intriguingly, CRISPR/Cas9-mediated elimination of AGO2 in K562 cells did not change the enrichment in RT-qPCR signal detected for JPT2 and CDKN1B with primers downstream the 3’CAGE versus upstream primers (Fig S3g-i). All together, our observations suggest that AGO2 binds immediately upstream of the site of cleavage that generates 3’UTR-derived RNAs independently of miRNA directed recruitment and that endogenous 3’UTR-derived RNAs are not produced as a result of AGO2 cleavage activity.

Accordingly, we instead explored the binding specificity of AGO2-eiCLIP data, and performed a motif analysis using HOMER motif finder. When analysing the 15 bp flanking region around AGO2-crosslinking peaks (see Methods), one of the most prominent motifs was highly enriched in Gs (Fig. S3j − 2nd and 3rd). Notably, this also agrees with one of the first AGO2-CLIP studies performed on mouse embryonic stem cells, where the authors showed that, without the miRNA present, AGO2 binds preferentially to G-rich motifs [53].

As we had previously demonstrated that G-rich motifs, which have the capability to form RNA-G-quadruplexes, are enriched around 3’UTR-derived RNAs, we next investigated whether AGO2 and UPF1 could be attracted to these specific G-rich motif structures independently of their location to CAGE peaks. We first aligned AGO2-eiCLIP and UPF1-eCLIP cross-linking positions relative to the 3' end of rG4-seq sites in different regions of primary transcripts. Both AGO2 and UPF1 crosslink-binding sites are much more highly enriched at rG4-seq sites in the 3’UTRs relative to 5’UTRs, introns and coding sequence although we noted that the binding of UPF1 occurred at the 3’end of the G4-seq sites and AGO2 bound immediately upstream (Fig. S3k-n). To further explore the relationship between AGO2 and UPF1 binding concerning 3’UTR CAGE peaks and their association with G4-seq signals, we categorised the 3’UTR CAGE peaks into four classes, depending on the presence or absence of these elements. We observed that the majority of 3’UTR CAGE peaks contained both AGO2 and UPF1 binding but not G4-seq signal although, on the other hand, the majority of 3’UTR CAGE sites overlapping with G4-seq also contained AGO2/UPF1 sites (Fig. S3o). Then we further explored the binding position of UPF1 and AGO2 relative to the 3'UTR CAGE peaks, in the presence or absence of the G4-seq site. Interestingly, both proteins exhibited a distinct shift in position influenced by the G4 motif enrichment; whilst AGO2 showed a pronounced shift to the upstream region of the 3’UTR CAGE peak in the presence of G4-seq sites, UPF1 displayed a downstream shift (Fig. S3p-q). An important next direction for future studies will be to experimentally investigate the mechanistic implications of the overlap between sites with the ability to form RNA-G4 structures and AGO2 and UPF1 binding for the generation of the capped 3’UTR-derived RNAs.

Capped 3’UTR fragments of CDKN1B and JPT2 transcripts do not co-localise with the parental mRNAs

Finally, we examined the potential implications of 3’UTR-derived RNAs. Specifically, we sought to understand how 3’UTR-derived RNAs might localise either together or independently from the parental mRNAs. To test this, we designed smFISH (single molecule fluorescence in situ hybridization) probes to simultaneously image the RNA upstream and downstream of the proposed post-transcriptional cleavage and capping site in CDKN1B and JPT2 using hybridisation chain reaction RNA-fluorescence in situ hybridization (HCR-FISH 3.0) [54]. To account for technical biases in detection, we also designed probes against the coding sequence (hereafter upstream) and 3’UTR (hereafter downstream) of a control mRNA, PGAM1, which does not contain CAGE peaks in the 3’UTR and contained a similar 3’UTR length to our targets.

We performed HCR-FISH in HeLa cells to determine whether putative 3’UTR-derived RNAs can be found independently of the RNA upstream of the cleavage site (Fig. 4a, b). In the control transcript, PGAM1, we observed that 17.3% of upstream signals did not have a colocalising downstream signal and 21.3% of downstream signals did not have a colocalising upstream signal (Fig. 4c, S4a). However, the mRNAs that contain a 3’UTR CAGE signature were significantly more likely to show independent signals from the RNA downstream of the proposed cleavage site (CDKN1B: 53.3%, p adj. < 0.05; JPT2: 52.3%, p adj. < 0.05; Fig. 4c). In the case of JPT2, we also observed significantly more independent signals from the upstream probes (29.3%, p adj. < 0.05; Fig. 4c). These observations are consistent with the existence of cleaved 3’UTR fragments in the cell, and they reveal that these products may localise differently from their host transcripts.

Previous studies had identified 3’UTR-derived RNAs via enrichment of RNA-seq reads-starts or CAGE signals mapping at 3’UTRs [7, 16, 18]. Nevertheless, most 3’UTR-derived RNAs may have remained undetected until now due to technical limitations inherent to these approaches, including reliance on fragmented-based sequencing methods and potentially biassed library preparations. Here, we provide multiple lines of evidence that complement CAGE and RNA-seq data, including RNA structural features, RBP interactions around 3’UTR CAGE signals and long-read nanopore CAGE to validate the widespread presence of capped 3’UTR-derived RNAs in human cells. We show that capped 3’UTR-derived RNAs are generated post-transcriptionally at positions characterised by the presence of G-rich motifs and specific RBP binding sites. We also demonstrate that, consistent with a functional role, capped 3’UTR-derived RNA sequences are evolutionary conserved and can localise to different subcellular regions than the parental mRNAs.

The role of AGO2, UPF1 and G-rich motifs in the generation of 3'UTR-derived capped RNAs

One of the key findings of our work is that siRNA action can result in the generation of capped RNAs downstream of the cleavage site (Fig. 3a,c, S3a). However, with the available data we could not quantify the efficiency of such capping, or identify all the factors that might be involved in the process. It is well established that AGO2 cleaves the double stranded RNA formed by the reverse complementary binding of siRNAs to their target mRNAs in the cytosol, pointing to a model in which AGO2 cleavage products can be recapped. The strong enrichment of AGO2 to a region with a modest increase in sequence conservation (Fig. S2d) immediately upstream of the endogenous 3’UTR CAGE peaks (Fig. 3d) suggested that AGO2 also plays a role in the generation of endogenous capped 3’UTR-derived RNAs.

Endogenous RNA interference in mammalian cells is mainly mediated by miRNAs. MiRNAs drive AGO2 to their targets through partial complementarity only and thus, in contrast to siRNAs, do not trigger AGO2 catalytic activity [55]. On the contrary, miRNA action in mammals relies mostly on translational repression and/or exonucleolytic degradation of their targets through the recruitment of other protein partners [55]. This raises two interesting possibilities: (1) that the mechanism by which AGO2 is involved in the generation of endogenous 3’UTR-derived capped RNAs is different to its role in the generation of capped fragments following siRNA action. In this scenario, AGO2 role will likely be independent of its catalytic activity and possibly require the recruitment of other nucleases, or (2) that AGO2 endonucleolytic activity is also important for the generation of endogenous 3’UTR-derived capped RNAs and therefore likely independent of its role in miRNA-mediated silencing of gene expression. The latter model is supported by the finding of a stronger enrichment in AGO2 binding in 3’UTR CAGE peaks in the absence of miRNA binding sites (Fig. S3f). We also failed to observe CAGE peaks in the vicinity of 29 mRNA targets whose genomics sequences contained miRNA target sites of extensive (< 2 mismatches) complementarity to known miRNAs. Interestingly, we found that AGO2 binds to potential RNA-G4s in 3’UTR CAGE sites (Fig. S3h-i). Early AGO2-CLIP experiments had already identified an enrichment of a G-rich motif in sequences cross-linked to AGO2 likely in a miRNA-independent manner [53]. AGO2 binding sites neighbouring G-rich sequences may be less likely to be guided by miRNAs, as RNA-G4s can prevent miRNA binding from their target sites [56]. Thus, our analysis suggests that the role of AGO2 in the generation of capped 3’UTR-derived RNAs is independent of its role in miRNA-mediated gene silencing. This is in agreement with previous findings by Andreassi et al. [57]. Moreover, elimination of AGO2 did not affect the enrichment of JPT2 and CDKN1B RT-qPCR signals corresponding to 3’UTR-derived RNAs (S3i) further arguing against a direct role of AGO2 catalytic activity in the generation of these species. In the future, long-read CAGE analysis in models of AGO2 loss of function, possibly in conjunction with the elimination of other AGO proteins (1, 3–4) present in the cells, may help clarify the specific role of AGO2 in the global generation of capped 3’UTR-derived RNAs.

Our work identified UPF1 as the RBP with the strongest binding enrichment around 3’UTR CAGE peaks and uncovered an overlapping of AGO2, UPF1 and G-rich sequences in 3’UTR CAGE sites. Interestingly, a significant overlap between UPF1 and AGO2 binding sites as well as preferential UPF1 binding to structured G-rich regions had been previously reported [58]. Whilst it is well-established that UPF1 plays an essential role in mRNA degradation [36] and binds to GC-rich motifs in 3’UTRs [38], the main trigger of UPF1-mediated mRNA decay remains unknown. It has been suggested that G-enrichment in 3’UTRs plays a vital role in triggering UPF1-mediated mRNA decay [38]. It was therefore tempting to hypothesise a causal role for UPF1 in the generation of 3’UTR-derived RNAs in connection with its role in mRNA decay [18]. Nevertheless, arguing against this possibility, siRNA-mediated downregulation of UPF1 did not change the relative amount of CDKN1B and JPT2 mRNA detected with primers targeting downstream versus upstream of the 3’CAGE peaks (Fig. S2t). Additional experimental data will be needed to reach a more definitive conclusion, but it is is also likely that the presence of RNA-G4 sequences (or strong secondary structures) around 3’UTR CAGE causes the stalling of UPF1 as the helicase translocates in a 5’-3’ direction [59].

Our analysis also shows that the G-rich sequences around 3’UTR CAGE peaks have a strong pairing probability and thus have the potential to form G4 structures (Fig. 2b, S2g-l). Nevertheless, although RNA-G4s are known to form stable structures in vitro, recent studies have suggested that they may be less stable in vivo due to active unwinding by RNA helicases [60, 61]. However, Kharel et al. (2022) [62] demonstrated that 3’UTR-G4s are dynamically regulated under cellular stress conditions and may play a role in mRNA stability for several transcripts, including the 3’UTR-G4 in the amyloid precursor protein (APP) mRNA. The G4 motif in the 3’UTR of APP mRNA was found to suppress overproduction of APP protein, but the underlying mechanism remained unclear [63]. Analysis of rG4-seq and CAGE data from HeLa cells showed that the 3’ end of this G4 motif in APP 3’UTR precisely coincided with a 3’UTR CAGE signal (Fig. S1n - APP, Fig. 2l). Moreover, long-read sequencing CAGE from cortical neuron samples, confirmed that abundant 3’UTR-derived capped RNAs are present in these samples that span from the identified 3’UTR CAGE peak to the end of the annotated APP gene (Fig. S1n - APP). This raises the possibility that the RNA-G4 regulates APP protein through an unknown mechanism that involves the generation of 3’UTR-derived capped RNAs.

Future studies will focus on investigating whether these G-rich sequences form stable G4 structures in other genes and whether they directly contribute to the formation of 3’UTR-derived RNAs. Another important open question that warrants further investigation is whether these G-rich motifs are essential for the recruitment of AGO2 and/or other proteins responsible for the generation of endogenous 3’UTR-derived RNAs.

Independent localisation of capped 3’UTR fragments from the parental mRNAs.

CAGE-seq, RNA-seq, RT-qPCR and long-read CAGE experiments suggested that 3'UTR-derived RNAs in CDKN1B and JPT2 are capped and highly expressed in cells. In line with these experiments, HCR-FISH showed that RNAs derived from CDKN1B and JPT2 3’UTRs can be detected at separate cytosolic locations than their parental mRNAs. In agreement with the CAGE data, higher ratio of downstream vs. upstream probes is present in CDKN1B (Fig. 4a-c, S1n - CDKN1B).

Interestingly, some cells exhibited a pronounced perinuclear accumulation of 3’UTR-probes in CDKN1B (Fig. 4a), whereas for the majority of the cells, the signal was dispersed throughout the cytosol. An intriguing possibility is a potential cell cycle-dependence, as observed in other aspects of CDKN1B gene expression regulation, such as mRNA translation [64, 65]. It has been previously shown that defective CDKN1 splicing can be rectified to restore CDKN1B/p27^kip protein production and induce cell cycle arrest [66, 67]. Likewise, the capped 3’UTR fragments of CDKN1B mRNA could regulate p27^kip protein in a cell cycle-specific manner, albeit this possibility remains to be investigated. Of note, a separate study proposed a cell-cycle function for 3’UTR fragments of NURR1 (nuclear receptor related 1 protein) mRNA which were highly expressed in proliferating neuronal cells [17]. Exploring the dynamic nature of these capped 3’UTR fragments and their potential influence on cellular functions in a manner dependent on the cell cycle remains an important area for further investigation.

The different localisation of the 3’UTR-derived fragments to their full-length counterparts raises more questions on the fate and function of these RNAs. It is worth mentioning that analysis of ribosome footprinting data [29] (not-shown) failed to reveal any substantial ribosomal binding to capped 3’UTR-derived RNAs. This argues against their translation and supports a role similar to other cytosolic lncRNAs [68], as previously observed by others for specific 3’UTR-derived RNAs [17, 18, 69]. Nevertheless, the binding of the translation initiation factor eIF4E to these RNAs seems paradoxical. Moreover, these findings are in sharp contrast with those by Sudmant et al. who found evidence for ribosomal binding and for the existence of peptides encoded by comparable isolated 3’UTR fragments in human brain samples[17, 18, 69]. It is important to acknowledge that the coverage of ribo-footprinting in 3’UTRs is limited and it therefore can’t fully rule out that some of the capped, eIF4E-bound, 3’UTR-derived RNAs identified in our study are translated. This will require further investigation, possibly in a gene- and tissue-dependent manner.

Methodological implications

Different abundance and localisation of 3’UTR derived RNAs relative to their parental transcripts and the 5’ cleavage fragment containing protein coding sequence suggests that 3’UTR fragmented-based sequencing methods might be measuring the wrong RNA species in a significant proportion of cases. Even in the case of RNA-seq, quantitating the signal across the entire length of the uncleaved mRNA might measure a combination of protein-coding and non-coding RNA species. To increase the accuracy of quantitation of protein coding transcript levels, as well as those of 3’UTR-derived RNAs themselves, it may be necessary to develop new computational quantitation methods informed by the results of this paper, which will try to estimate the levels of protein-coding and 3’UTR fragments separately. Moreover, many new drugs which are based on siRNA targeting are already in use or under active clinical trials for treating a variety of conditions including neurological diseases [70]. Side cleavage products of the targeted mRNAs from these therapeutic drugs could be subjected to a cytoplasmic capping mechanism and result in unwanted toxic side effects.

Limitations

At the moment we do not have the ability to identify the full-length size of 3'UTR-derived RNAs in a high-throughput manner since the main technique that we used (CAGE-seq) is based on 5’ end sequencing. Also, CAGE-seq method has a limitation on fragment size similar to other HT-sequencing methods with a minimum fragment size of 200 bps. This can be overcome with the long-read CAGE (Fig. S1n), but the whole data is not yet publicly available. Using only experimental datasets has limitations in coverage, organisms, cell lines and can increase the number of false positives as a result of background noise. It is for example important to note that we have limited our study to RBPs with eCLIP available, which may have prevented the identification of other important proteins involved in the cleavage and/or re-capping of 3’UTR mRNA fragments.

This study is limited to human samples and, even though the sequence conservation suggests that capped 3'UTR-derived fragments may be conserved, this will need further experimental validation. New computational methods will need to be developed for these studies which are currently limited by the availability of such large datasets.

3'UTR-derived RNAs are emerging as novel regulatory molecules, with potential implications in broad cellular processes such as cell cycle or neuronal homeostasis [17, 18, 69]. However, the molecular mechanisms involved in the generation of these RNA species had been largely unknown. Our study sheds new light into these mechanisms by revealing that capped 3'UTR-derived RNAs originate from sequences rich in G motifs that contain both UPF1 and AGO2 binding sites. These findings suggest a significant role for these elements in the regulatory mechanism. Overall, our findings provide the framework for further investigations where their functions will surely emerge.

Acknowledgments

This work was funded in part by The Wellcome Trust grants (106954/Z/15/Z) awarded to B.L., by the Medical Research Council (MRC) (MR/P023223/1 and MR/X009912/1) to A.M-S, MRC (grant number MC_UP_1102/18) to S.V, and by (215593/Z/19/Z) to J.U., Medical Research Council (MRC) Core Funding (MC-A652-5QA10), by the Imperial College Research Fellowship awarded to N.H., and by the Francis Crick Institute which receives its core funding from Cancer Research UK (CC0102), the UK Medical Research Council (CC0102), and the Wellcome Trust (CC0102), and by an institutional budget from RIKEN, MEXT (Ministry of Education, Culture, Sports, Science and Technology) and from institutional budget from the Human Technopole. We thank the Crick Advanced Light Microscopy facility, especially Donald Bell, for their support. We thank Sarvesh Nikumbh for help with the CAGEr tool and other members of Lenhard's group for helpful discussions and comments on the manuscript. We thank Ira Iosub for processing the ribosome footprinting data. For the purpose of Open Access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission.

Author contributions

N.H. and B.L. conceived the study. N.H. designed the experiments, analysed the data, and led the project. R.C. and A.M-S. performed and designed qPCR experiments. AGO2-eiCLIP was designed and performed by C.R.S and A.M-S. R.F. and H.D. designed and performed smFISH probe imaging supervised by J.U. Long-read CAGE examples were produced, processed and provided by C.P., K.Y., T.K., C.W.Y., M.K., H.T., P.C., and visualised by A.M.C. N.H. and A.M-S. wrote the manuscript, with contributions from A.M.C., R.F., H.D., A.M.J, S.V., C.R.S., J.U. and B.L.

Declaration of interests

The authors declare no competing interests.

Ramanathan, A., Robb, G.B., and Chan, S.-H. (2016). mRNA capping: biological functions and applications. Nucleic Acids Res. 44, 7511–7526.
Otsuka, Y., Kedersha, N.L., and Schoenberg, D.R. (2009). Identification of a cytoplasmic complex that adds a cap onto 5’-monophosphate RNA. Mol. Cell. Biol. 29, 2155–2167.
Mukherjee, C., Bakthavachalu, B., and Schoenberg, D.R. (2014). The cytoplasmic capping complex assembles on adapter protein nck1 bound to the proline-rich C-terminus of Mammalian capping enzyme. PLoS Biol. 12, e1001933.
Hestand, M.S., Klingenhoff, A., Scherf, M., Ariyurek, Y., Ramos, Y., van Workum, W., Suzuki, M., Werner, T., van Ommen, G.-J.B., den Dunnen, J.T., et al. (2010). Tissue-specific transcript annotation and expression profiling with complementary next-generation sequencing technologies. Nucleic Acids Res. 38, e165.
Naeli, P., Winter, T., Hackett, A.P., Alboushi, L., and Jafarnejad, S.M. (2023). The intricate balance between microRNA-induced mRNA decay and translational repression. FEBS J. 290, 2508–2524.
Murata, M., Nishiyori-Sueki, H., Kojima-Ishiyama, M., Carninci, P., Hayashizaki, Y., and Itoh, M. (2014). Detecting expressed genes using CAGE. Methods Mol. Biol. 1164, 67–85.
Kocabas, A., Duarte, T., Kumar, S., and Hynes, M.A. (2015). Widespread differential expression of coding region and 3’ UTR sequences in neurons and other tissues. Neuron 88, 1149–1156.
Malka, Y., Steiman-Shimony, A., Rosenthal, E., Argaman, L., Cohen-Daniel, L., Arbib, E., Margalit, H., Kaplan, T., and Berger, M. (2017). Post-transcriptional 3´-UTR cleavage of mRNA transcripts generates thousands of stable uncapped autonomous RNA fragments. Nat. Commun. 8, 2029.
Mercer, T.R., Dinger, M.E., Bracken, C.P., Kolle, G., Szubert, J.M., Korbie, D.J., Askarian-Amiri, M.E., Gardiner, B.B., Goodall, G.J., Grimmond, S.M., et al. (2010). Regulated post-transcriptional RNA cleavage diversifies the eukaryotic transcriptome. Genome Res. 20, 1639–1650.
Affymetrix ENCODE Transcriptome Project, and Cold Spring Harbor Laboratory ENCODE Transcriptome Project (2009). Post-transcriptional processing generates a diversity of 5’-modified long and short RNAs. Nature 457, 1028–1032.
Adiconis, X., Haber, A.L., Simmons, S.K., Levy Moonshine, A., Ji, Z., Busby, M.A., Shi, X., Jacques, J., Lancaster, M.A., Pan, J.Q., et al. (2018). Comprehensive comparative analysis of 5’-end RNA-sequencing methods. Nat. Methods 15, 505–511.
Carninci, P., Kasukawa, T., Katayama, S., Gough, J., Frith, M.C., Maeda, N., Oyama, R., Ravasi, T., Lenhard, B., Wells, C., et al. (2005). The transcriptional landscape of the mammalian genome. Science 309, 1559–1563.
Carninci, P., Sandelin, A., Lenhard, B., Katayama, S., Shimokawa, K., Ponjavic, J., Semple, C.A.M., Taylor, M.S., Engström, P.G., Frith, M.C., et al. (2006). Genome-wide analysis of mammalian promoter architecture and evolution. Nat. Genet. 38, 626–635.
Kiss, D.L., Oman, K., Bundschuh, R., and Schoenberg, D.R. (2015). Uncapped 5’ ends of mRNAs targeted by cytoplasmic capping map to the vicinity of downstream CAGE tags. FEBS Lett. 589, 279–284.
Berger, M.R., Alvarado, R., and Kiss, D.L. (2019). mRNA 5’ ends targeted by cytoplasmic recapping cluster at CAGE tags and select transcripts are alternatively spliced. FEBS Lett. 593, 670–679.
Mercer, T.R., Wilhelm, D., Dinger, M.E., Soldà, G., Korbie, D.J., Glazov, E.A., Truong, V., Schwenke, M., Simons, C., Matthaei, K.I., et al. (2011). Expression of distinct RNAs from 3’ untranslated regions. Nucleic Acids Res. 39, 2393–2403.
Ji, S., Yang, Z., Gozali, L., Kenney, T., Kocabas, A., Jinsook Park, C., and Hynes, M. (2021). Distinct expression of select and transcriptome-wide isolated 3’UTRs suggests critical roles in development and transition states. PLoS ONE 16, e0250669.
Sudmant, P.H., Lee, H., Dominguez, D., Heiman, M., and Burge, C.B. (2018). Widespread Accumulation of Ribosome-Associated Isolated 3’ UTRs in Neuronal Cell Populations of the Aging Brain. Cell Rep. 25, 2447–2456.e4.
Takahashi, H., Lassmann, T., Murata, M., and Carninci, P. (2012). 5’ end-centered expression profiling using cap-analysis gene expression and next-generation sequencing. Nat. Protoc. 7, 542–561.
Djebali, S., Davis, C.A., Merkel, A., Dobin, A., Lassmann, T., Mortazavi, A., Tanzer, A., Lagarde, J., Lin, W., Schlesinger, F., et al. (2012). Landscape of transcription in human cells. Nature 489, 101–108.
Kanamori-Katayama, M., Itoh, M., Kawaji, H., Lassmann, T., Katayama, S., Kojima, M., Bertin, N., Kaiho, A., Ninomiya, N., Daub, C.O., et al. (2011). Unamplified cap analysis of gene expression on a single-molecule sequencer. Genome Res. 21, 1150–1159.
Frith, M.C., Valen, E., Krogh, A., Hayashizaki, Y., Carninci, P., and Sandelin, A. (2008). A code for transcription initiation in mammalian genomes. Genome Res. 18, 1–12.
Giacometti, S., Benbahouche, N.E.H., Domanski, M., Robert, M.-C., Meola, N., Lubas, M., Bukenborg, J., Andersen, J.S., Schulze, W.M., Verheggen, C., et al. (2017). Mutually Exclusive CBC-Containing Complexes Contribute to RNA Fate. Cell Rep. 18, 2635–2650.
Izaurralde, E., Lewis, J., McGuigan, C., Jankowska, M., Darzynkiewicz, E., and Mattaj, I.W. (1994). A nuclear cap binding protein complex involved in pre-mRNA splicing. Cell 78, 657–668.
Schoenberg, D.R., and Maquat, L.E. (2009). Re-capping the message. Trends Biochem. Sci. 34, 435–442.
Core, L.J., Martins, A.L., Danko, C.G., Waters, C.T., Siepel, A., and Lis, J.T. (2014). Analysis of nascent RNA identifies a unified architecture of initiation regions at mammalian promoters and enhancers. Nat. Genet. 46, 1311–1320.
Jensen, K.B., Dredge, B.K., Toubia, J., Jin, X., Iadevaia, V., Goodall, G.J., and Proud, C.G. (2021). capCLIP: a new tool to probe translational control in human cells through capture and identification of the eIF4E-mRNA interactome. Nucleic Acids Res. 49, e105.
Rhoads, R.E. (2009). eIF4E: new family members, new binding partners, new roles. J. Biol. Chem. 284, 16711–16715.
Ferguson, L., Upton, H.E., Pimentel, S.C., Mok, A., Lareau, L.F., Collins, K., and Ingolia, N.T. (2023). Streamlined and sensitive mono- and di-ribosome profiling in yeast and human cells. Nat. Methods 20, 1704–1715.
Zhang, L., Kasif, S., Cantor, C.R., and Broude, N.E. (2004). GC/AT-content spikes as genomic punctuation marks. Proc Natl Acad Sci USA 101, 16855–16860.
Kharel, P., Becker, G., Tsvetkov, V., and Ivanov, P. (2020). Properties and biological impact of RNA G-quadruplexes: from order to turmoil and back. Nucleic Acids Res. 48, 12534–12555.
Lee, D.S.M., Ghanem, L.R., and Barash, Y. (2020). Integrative analysis reveals RNA G-quadruplexes in UTRs are selectively constrained and enriched for functional associations. Nat. Commun. 11, 527.
Kwok, C.K., Marsico, G., Sahakyan, A.B., Chambers, V.S., and Balasubramanian, S. (2016). rG4-seq reveals widespread formation of G-quadruplex structures in the human transcriptome. Nat. Methods 13, 841–844.
Bedrat, A., Lacroix, L., and Mergny, J.-L. (2016). Re-evaluation of G-quadruplex propensity with G4Hunter. Nucleic Acids Res. 44, 1746–1759.
Van Nostrand, E.L., Freese, P., Pratt, G.A., Wang, X., Wei, X., Xiao, R., Blue, S.M., Chen, J.-Y., Cody, N.A.L., Dominguez, D., et al. (2020). A large-scale binding and functional map of human RNA-binding proteins. Nature 583, 711–719.
Staszewski, J., Lazarewicz, N., Konczak, J., Migdal, I., and Maciaszczyk-Dziubinska, E. (2023). UPF1-From mRNA Degradation to Human Disorders. Cells 12.
Kurosaki, T., Li, W., Hoque, M., Popp, M.W.-L., Ermolenko, D.N., Tian, B., and Maquat, L.E. (2014). A post-translational regulatory switch on UPF1 controls targeted mRNA degradation. Genes Dev. 28, 1900–1916.
Imamachi, N., Salam, K.A., Suzuki, Y., and Akimitsu, N. (2017). A GC-rich sequence feature in the 3’ UTR directs UPF1-dependent mRNA decay in mammalian cells. Genome Res. 27, 407–418.
Gherzi, R., Chen, C.-Y., Ramos, A., and Briata, P. (2014). KSRP controls pleiotropic cellular functions. Semin. Cell Dev. Biol. 34, 2–8.
Cooke, A., Schwarzl, T., Huppertz, I., Kramer, G., Mantas, P., Alleaume, A.-M., Huber, W., Krijgsveld, J., and Hentze, M.W. (2019). The RNA-Binding Protein YBX3 Controls Amino Acid Levels by Regulating SLC mRNA Abundance. Cell Rep. 27, 3097–3106.e5.
Mo, J., Liang, H., Su, C., Li, P., Chen, J., and Zhang, B. (2021). DDX3X: structure, physiologic functions and cancer. Mol. Cancer 20, 38.
Carthew, R.W., and Sontheimer, E.J. (2009). Origins and Mechanisms of miRNAs and siRNAs. Cell 136, 642–655.
Bartel, D.P. (2018). Metazoan MicroRNAs. Cell 173, 20–51.
Lam, J.K.W., Chow, M.Y.T., Zhang, Y., and Leung, S.W.S. (2015). siRNA Versus miRNA as Therapeutics for Gene Silencing. Mol. Ther. Nucleic Acids 4, e252.
Lizio, M., Ishizu, Y., Itoh, M., Lassmann, T., Hasegawa, A., Kubosaki, A., Severin, J., Kawaji, H., Nakamura, Y., FANTOM consortium, et al. (2015). Mapping Mammalian Cell-type-specific Transcriptional Regulatory Networks Using KD-CAGE and ChIP-seq Data in the TC-YIK Cell Line. Front. Genet. 6, 331.
Elbashir, S.M., Lendeckel, W., and Tuschl, T. (2001). RNA interference is mediated by 21- and 22-nucleotide RNAs. Genes Dev. 15, 188–200.
Lin, J., Xu, K., Roth, J.A., and Ji, L. (2016). Detection of siRNA-mediated target mRNA cleavage activities in human cells by a novel stem-loop array RT-PCR analysis. Biochem. Biophys. Rep. 6, 16–23.
Paterson, H.A.B., Yu, S., Artigas, N., Prado, M.A., Haberman, N., Wang, Y.-F., Jobbins, A.M., Pahita, E., Mokochinski, J., Hall, Z., et al. (2022). Liver RBFOX2 regulates cholesterol homeostasis via Scarb1 alternative splicing in mice. Nat. Metab. 4, 1812–1829.
Nishihara, T., Zekri, L., Braun, J.E., and Izaurralde, E. (2013). miRISC recruits decapping factors to miRNA targets to enhance their degradation. Nucleic Acids Res. 41, 8692–8705.
Jung, E., Seong, Y., Jeon, B., Song, H., and Kwon, Y.-S. (2017). Global analysis of AGO2-bound RNAs reveals that miRNAs induce cleavage of target RNAs with limited complementarity. Biochim. Biophys. Acta Gene Regul. Mech. 1860, 1148–1158.
Bracken, C.P., Szubert, J.M., Mercer, T.R., Dinger, M.E., Thomson, D.W., Mattick, J.S., Michael, M.Z., and Goodall, G.J. (2011). Global analysis of the mammalian RNA degradome reveals widespread miRNA-dependent and miRNA-independent endonucleolytic cleavage. Nucleic Acids Res. 39, 5658–5668.
Karginov, F.V., Cheloufi, S., Chong, M.M.W., Stark, A., Smith, A.D., and Hannon, G.J. (2010). Diverse endonucleolytic cleavage sites in the mammalian transcriptome depend upon microRNAs, Drosha, and additional nucleases. Mol. Cell 38, 781–788.
Leung, A.K.L., Young, A.G., Bhutkar, A., Zheng, G.X., Bosson, A.D., Nielsen, C.B., and Sharp, P.A. (2011). Genome-wide identification of Ago2 binding sites from mouse embryonic stem cells with and without mature microRNAs. Nat. Struct. Mol. Biol. 18, 237–244.
Choi, H.M.T., Schwarzkopf, M., Fornace, M.E., Acharya, A., Artavanis, G., Stegmaier, J., Cunha, A., and Pierce, N.A. (2018). Third-generation in situ hybridization chain reaction: multiplexed, quantitative, sensitive, versatile, robust. Development 145.
Gebert, L.F.R., and MacRae, I.J. (2019). Regulation of microRNA function in animals. Nat. Rev. Mol. Cell Biol. 20, 21–37.
Rouleau, S., Glouzon, J.-P.S., Brumwell, A., Bisaillon, M., and Perreault, J.-P. (2017). 3’ UTR G-quadruplexes regulate miRNA binding. RNA 23, 1172–1179.
Andreassi, C., Luisier, R., Crerar, H., Darsinou, M., Blokzijl-Franke, S., Lenn, T., Luscombe, N.M., Cuda, G., Gaspari, M., Saiardi, A., et al. (2021). Cytoplasmic cleavage of IMPA1 3’ UTR is necessary for maintaining axon integrity. Cell Rep. 34, 108778.
Hurt, J.A., Robertson, A.D., and Burge, C.B. (2013). Global analyses of UPF1 binding and function reveal expanded scope of nonsense-mediated mRNA decay. Genome Res. 23, 1636–1650.
- Abstract - Europe PMC Available at: https://europepmc.org/article/pmc/pmc4506499 [Accessed December 11, 2023].
Caterino, M., and Paeschke, K. (2022). Action and function of helicases on RNA G-quadruplexes. Methods 204, 110–125.
Guo, J.U., and Bartel, D.P. (2016). RNA G-quadruplexes are globally unfolded in eukaryotic cells and depleted in bacteria. Science 353.
Kharel, P., Fay, M., Manasova, E.V., Anderson, P.J., Kurkin, A.V., Guo, J.U., and Ivanov, P. (2023). Stress promotes RNA G-quadruplex folding in human cells. Nat. Commun. 14, 205.
Crenshaw, E., Leung, B.P., Kwok, C.K., Sharoni, M., Olson, K., Sebastian, N.P., Ansaloni, S., Schweitzer-Stenner, R., Akins, M.R., Bevilacqua, P.C., et al. (2015). Amyloid precursor protein translation is regulated by a 3’UTR guanine quadruplex. PLoS ONE 10, e0143160.
Hengst, L., and Reed, S.I. (1996). Translational control of p27Kip1 accumulation during the cell cycle. Science 271, 1861–1864.
Cuesta, R., Martínez-Sánchez, A., and Gebauer, F. (2009). miR-181a regulates cap-dependent translation of p27(kip1) mRNA in myeloid cells. Mol. Cell. Biol. 29, 2841–2851.
Kaida, D., and Shida, K. (2022). Spliceostatin A stabilizes CDKN1B mRNA through the 3’ UTR. Biochem. Biophys. Res. Commun. 608, 39–44.
Satoh, T., and Kaida, D. (2016). Upregulation of p27 cyclin-dependent kinase inhibitor and a C-terminus truncated form of p27 contributes to G1 phase arrest. Sci. Rep. 6, 27829.
Noh, J.H., Kim, K.M., McClusky, W.G., Abdelmohsen, K., and Gorospe, M. (2018). Cytoplasmic functions of long noncoding RNAs. Wiley Interdiscip. Rev. RNA 9, e1471.
Ciolli Mattioli, C., Rom, A., Franke, V., Imami, K., Arrey, G., Terne, M., Woehler, A., Akalin, A., Ulitsky, I., and Chekulaeva, M. (2019). Alternative 3’ UTRs direct localization of functionally diverse protein isoforms in neuronal compartments. Nucleic Acids Res. 47, 2560–2573.
Dammes, N., and Peer, D. (2020). Paving the road for RNA therapeutics. Trends Pharmacol. Sci. 41, 755–775.
Chakrabarti, A.M., Haberman, N., Praznik, A., Luscombe, N.M., and Ule, J. (2018). Data Science Issues in Studying Protein–RNA Interactions with CLIP Technologies. Annu. Rev. Biomed. Data Sci. 1, 235–261.
Balwierz, P.J., Carninci, P., Daub, C.O., Kawai, J., Hayashizaki, Y., Van Belle, W., Beisel, C., and van Nimwegen, E. (2009). Methods for analyzing deep sequencing expression data: constructing the human and mouse promoterome with deepCAGE data. Genome Biol. 10, R79.
Pasquier, C., and Robichon, A. (2020). Computational prediction of miRNA/mRNA duplexomes at the whole human genome scale reveals functional subnetworks of interacting genes with embedded miRNA annealing motifs. Comput. Biol. Chem. 88, 107366.
Kuehn, E., Clausen, D.S., Null, R.W., Metzger, B.M., Willis, A.D., and Özpolat, B.D. (2022). Segment number threshold determines juvenile onset of germline cluster expansion in Platynereis dumerilii. J. Exp. Zool. B Mol. Dev. Evol. 338, 225–240.
Stringer, C., Wang, T., Michaelos, M., and Pachitariu, M. (2021). Cellpose: a generalist algorithm for cellular segmentation. Nat. Methods 18, 100–106.
Bahry, E., Breimann, L., Zouinkhi, M., Epstein, L., Kolyvanov, K., Mamrak, N., King, B., Long, X., Harrington, K.I.S., Lionnet, T., et al. (2022). RS-FISH: precise, interactive, fast, and scalable FISH spot detection. Nat. Methods 19, 1563–1567.
Lee, F.C.Y., Chakrabarti, A.M., Hänel, H., Monzón-Casanova, E., Hallegger, M., Militti, C., Capraro, F., Sadée, C., Toolan-Kerr, P., Wilkins, O., et al. (2021). An improved iCLIP protocol. BioRxiv.
Sibley, C.R. (2018). Individual Nucleotide Resolution UV Cross-Linking and Immunoprecipitation (iCLIP) to Determine Protein-RNA Interactions. Methods Mol. Biol. 1649, 427–454.
Kurata, J.S., and Lin, R.-J. (2018). MicroRNA-focused CRISPR-Cas9 library screen reveals fitness-associated miRNAs. RNA 24, 966–981.
Haberman, N., Huppertz, I., Attig, J., König, J., Wang, Z., Hauer, C., Hentze, M.W., Kulozik, A.E., Le Hir, H., Curk, T., et al. (2017). Insights into the design and interpretation of iCLIP experiments. Genome Biol. 18, 7.
DeMario, S., Xu, K., He, K., and Chanfreau, G.F. (2023). Nanoblot: an R-package for visualization of RNA isoforms from long-read RNA-sequencing data. RNA 29, 1099–1107.

No competing interests reported.

FigureS1c.pdf
Fig. S1: Related to Fig. 1 (a) Pearson’s correlation of raw CAGE reads counts per TSS or consensus cluster across biological replicates and cell types. (b) Reverse cumulative distribution of CAGE reads after normalisation using CAGEr package [72](CTSS = CAGE Transcription start site). (c) Total number of CAGE reads in each sample. (d) Density of total 5’ CAGE read positions normalised by the length of the correspondent transcript region identified in CAGE-seq libraries of K562 and HeLa samples with two biological replicates each, provided by ENCODE. (e) Percentage of CAGE tags per transcript region using Random primers, Oligod-T primers, and combination of both primers (1:4 oligod(T):Random Primers) in CAGE-seq libraries of THP-1 cells generated by RIKEN. (f) - (i) Pearson’s correlation between CAGE-seq replicates (rep1, rep2) and different cell lines (HeLa, K562) samples in 3’UTRs (F), 5’UTRs (G), CDS (H) and introns (I). (j) Top: Plot of the normalised coverage of the 5’ ends of forward paired-end reads (yellow line) and 3’ ends of reverse paired-end reads (blue line) of RNA-seq relative to 3’UTR CAGE peaks in HeLa cells. Bottom: Schematic representation of paired-end read positioning. Forward and reversed paired-end reads are presented in yellow and blue, respectively. The black box represents the ends of reads that are plotted in the top graph. (k) RT-qPCR data of gene expression using primers designed to amplify sequences located downstream (3’C), upstream (5’C) and overlapping (AC) the 3’UTR CAGE sites of CDKN1B and JPT2. Data represents fold detection using downstream versus upstream/overlapping primers relative to the 3‘UTR CAGE peaks. Each dot represents an independent biological replicate. Primer target sequences relative to the 3’UTR CAGE peak are schematically represented on the top right-hand side and visualised for each gene using IGV genome browser on the bottom. (l) Top gene examples with strongest 3’UTR CAGE peaks present in K562 and HeLa cell lines using IGV-genome browser for visualisation. (m) Visualisation of RNA-seq reads relative to dominant 3’UTR CAGE peaks in CDKN1B and JPT2 gene using IGV-genome browser. (n) Visualisation of 10 gene examples with 3’UTR CAGE peaks (HeLa, K562), rG4-seq clusters (HeLa) and long-read CAGE (Cortical Neuron) reads using IGV-genome browser.
FigureS2c.pdf
Fig. S2: Related to Fig. 2 (a) RNA-map showing normalised density of CBP20-iCLIP (HeLa) crosslink sites relative to dominant 3’UTR and 5’UTR CAGE peaks. (b) Mean score of GRO-cap seq coverage [26]per CAGE peak for 5’UTR, CDS, intron and 3’UTR regions. The heatmap represents GRO-cap seq scores plotted with deeptools. (c) RNA-map showing normalised density of cap-CLIP (HeLa) crosslink sites relative to dominant 3’UTR and 5’UTR CAGE peaks. (d) Mean coverage of conservation score from UCSC phastCons30way track relative to inner 3’UTR CAGE peaks (≥150 bps downstream from CDS and ≥150 bps upstream from transcript termination) and randomised control positions around the same region of 150 nts window for each peak. (e) Normalised motif enrichment of canonical PolyA motifs relative to 3’UTR ends and to the dominant 3’UTR CAGE peaks. (f) Sequence logos around K562 cells’ CAGE peaks across different transcript regions (g) The 75 nt region centred on K562 cells’ CAGE peaks at different transcript regions was used to calculate pairing probability with the RNAfold program, and the average pairing probability of each nucleotide is shown for the 50 nt region around CAGE peaks. (h) GGG-motif enrichment relative to CAGE (HeLa) peaks. (i) GGG-motif enrichment relative to CAGE (K562) peaks. (j) Summarised score from G4-Hunter prediction tool in the region of 50 nts upstream and downstream relative to CAGE (HeLa) peaks. (k) Summarised score from G4-Hunter prediction tool in the region of 50 nts upstream and downstream relative to CAGE (K562) peaks. (l) Enrichment of RNA-G-quadruplex sequencing (rG4-seq) hits from HeLa cells relative to CAGE (HeLa) peaks. (m) Percentage of G-seq sites per transcript region. (n) Enrichment of eCLIP cross-linking clusters surrounding 5’UTR CAGE peaks from 80 different RBP samples (right panel) in K562 cells from ENCODE database using sum of log ratios. The red line represents the threshold of top 10 RBP targets (left panel). (o) Enrichment of eCLIP cross-linking clusters surrounding intronic CAGE peaks from 80 different RBP samples (right panel) in K562 cells from ENCODE database using sum of log ratios. The red line represents the threshold of top 10 RBP targets (left panel). (p) Enrichment of eCLIP cross-linking clusters surrounding CDS CAGE peaks from 80 different RBP samples (right panel) in K562 cells from ENCODE database using sum of log ratios. The red line represents the threshold of top 10 RBP targets (left panel). (q) Pearson’s correlation between the 3’UTR CAGE (K562) tags and RNA-seq (K562) read coverage per gene (top-left). Pearson’s correlation between the 3’UTR crosslink coverage of UPF1-eCLIP (K562) and 3’UTR CAGE (K562) tags (top-right). Pearson’s correlation between the 3’UTR length and 3’UTR crosslink coverage of UPF1-eCLIP (K562) (bottom-left). Pearson’s correlation between the 3’UTR length and 3’UTR CAGE (K562) tags (bottom-right). (r) UPF1-eCLIP (K562) crosslink enrichment relative to the distance from the 3’UTR CAGE (K562) peaks. (s) Heatmap of UPF1-eCLIP (K562) crosslink site enrichment showing the top 500 3’UTR AGO2 targets in 100 nts flanking region relative to 3’UTR CAGE (K562) peaks. The heatmap represents log2 of crosslink counts, normalised by the mean of all counts within 200 nts of the targeting site. (t) K562 cells were transfected with siRNAs targeting UPF1 or non-targeting controls (C). Left-hand side panel: UPF1 expression was measured with RT-qPCR. Data is normalised by the housekeeping gene RPLP0 and presented as fold change of the control. Right-hand side panel: RT-qPCR data of gene expression using primers designed to amplify sequences located downstream (3’C) and upstream (5’C) the 3’UTR CAGE sites of CDKN1B and JPT2. Data represents fold detection using downstream (3’C) versus upstream (5’C) primers. Each dot represents an independent biological replicate. Primer target sequences relative to the 3’UTR CAGE peaks are schematically represented on the right.
FigureS3.pdf
Fig. S3: Related to Fig. 3 (a) Visualisation of 5’ CAGE reads relative to dominant transcription start site (TSS) and relative to small interfering RNA of ISL1 target (in red) for CAGE-ISL1-KD and CAGE-control samples with 3 biological replicates using IGV-genome browser. (b) Percentage of AGO2-eiCLIP (HeLa) binding sites per transcript region. (c) Binding enrichment of AGO2-eiCLIP (HeLa) relative to miRNA-regulated transcripts and non-miRNA-regulated transcript in HeLa (data from [73] ). (d) Heatmap of miRNA-seed sequence enrichment in 30 nt flanking region showing the top 500 AGO2 binding sites relative to AGO-eiCLIP (HeLa) crosslink sites. Meta plot visualises the miRNA-seed sequence composition relative to the 5’ of the AGO2 binding site. (e) Heatmap of AGO2-eiCLIP (HeLa) crosslink site enrichment showing the top 500 3’UTR AGO2 targets in 100 nts flanking region relative to 3’UTR CAGE (HeLa) peaks. The heatmap represents log2 of crosslink counts, normalised by the mean of all counts within 200 nts of the targeting site. (f) RNA-map showing normalised density of AGO2 crosslink sites from AGO2 binding sites that contain (mir+) or are absent (mir-) from predicted miRNA binding sites relative to 3’UTR CAGE peaks. (g) AGO2 mRNA expression was measured with RT-qPCR in 3 clonal subpopulations generated by single cell sorting of K562 cells transfected with a plasmid expressing Cas9 and two gRNAs targeting AGO2 and preselected by genomic DNA sequencing. Both KO1 and KO2 contained edited AGO2 sequences and control (C) wild-type sequences. Data is normalised by the housekeeping gene RPLP0 and presented as a fold change of the control. Each dot represents an independent biological replicate. (h) AGO2 protein was detected by Western Blot in 6 clonal subpopulations generated as in (g). Clon 2 and 6 were selected for further experiments and re-named as KO1 and KO2. This experiment was performed once. (i) RT-qPCR data of gene expression using primers designed to amplify sequences located downstream (3’C) and upstream (5’C) the 3’UTR CAGE sites of CDKN1B and JPT2. Data represents fold detection using downstream (3’C) versus upstream (5’C) primers. Each dot represents an independent biological replicate. Primer target sequences relative to the 3’UTR CAGE peaks are schematically represented below. (j) Sequence logos and statistics of top 12 significantly enriched motifs of AGO2-eiCLIP (HeLa) binding sites using Homer for de novo motif discovery. (k) Enrichment of AGO2-eiCLIP (HeLa) cross-linking sites relative to the 3’end of the rG4-seq site (HeLa). (l) Heatmap for AGO2-eiCLIP (HeLa) crosslink site enrichment to show the top 500 3’UTR AGO2 targets in 100 nts flanking region relative to 3’end of rG4-seq (HeLa) site. The heatmap represents log2 of crosslink counts, normalised by the mean of all counts within 200 nts of the targeting site. (m) Enrichment of UPF1-eCLIP (K562) cross-linking sites relative to the 3’end of the rG4-seq site (HeLa). (n) Heatmap for UPF1-eCLIP (K562) crosslink site enrichment to show the top 500 3’UTR UPF1 targets in 100 nts flanking region relative to 3’end of rG4-seq (HeLa) site. The heatmap represents log2 of crosslink counts, normalised by the mean of all counts within 200 nts of the targeting site. (o) Upset plot of intersection of AGO2-eiCLIP (HeLa) and UPF1-eCLIP(K562) binding sites, G4-seq (HeLa) sites relative to 3’UTR CAGE peaks (K562 & HeLa). (p) RNA-map showing normalised density of AGO2 crosslink sites relative to 3’UTR CAGE peaks intersecting G4-seq site (G4+) or not (G4-) from Fig. S3l. (q) RNA-map showing normalised density of UPF1 crosslink sites relative to 3’UTR CAGE peaks intersecting G4-seq site (G4+) or not (G4-) from Fig. S3l.
FigureS4.pdf
Fig. S4: Related to Fig. 4 (a) Density plots showing the shortest distance per detected signal in pixels to a signal of the opposite colour. The dashed line shows the cutoff used to distinguish colocalising and non-colocalising signals.
TableS1.tab.zip
Tables Table S1: This table contains genomic locations of 32,065 unique 3’UTR CAGE clusters across all CAGE samples from HeLa and K562 cell lines. Each sample contains a normalised value of 5’ read positions for each cluster.
Figure0Graphicalabstract.pdf
SupplmentrayFile.docx

Download PDF

Editor assigned by journal
30 Jul, 2024
Submission checks completed at journal
29 Jul, 2024
First submitted to journal
26 Jul, 2024

You are reading this latest preprint version

Widespread 3'UTR capped RNAs derive from G-rich regions in proximity to AGO2 binding sites

Status:

Version 1

Abstract

Figures

Background

Results

CAGE-seq identifies non-promoter associated capped 3'UTR-derived RNAs

3'UTR-derived RNAs are confirmed by RNA-seq, qPCR and long-read CAGE

Capped 3’UTR-derived RNAs are evolutionarily conserved and generated post-transcriptionally

5' ends of 3'UTR-derived RNAs are enriched for G-rich motifs and strong secondary structures

3’UTR CAGE sites are flanked by enriched UPF1 binding

mRNA cleavage by small interfering RNAs generates newly capped RNA fragments

3'UTR CAGE peaks coincide with AGO2 and UPF1 binding sites alongside G-rich motifs

Capped 3’UTR fragments of CDKN1B and JPT2 transcripts do not co-localise with the parental mRNAs

Discussion

The role of AGO2, UPF1 and G-rich motifs in the generation of 3'UTR-derived capped RNAs

Methodological implications

Limitations

Conclusions

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1