Cell culture, siRNA transfection, and actinomycin D treatment

293T cells, human embryonic kidney cells transformed by adenoviral and SV40 oncogenes, were maintained in DMEM with L-glutamine, 10% fetal bovine serum, and 1% penicillin/streptomycin. 293T was obtained from ATCC (ATCC #CRL-3216) and were grown in humidified incubators with 5% CO₂ at 37°C. For duplex siRNA transfections, cells were first reverse transfected, then forward transfected 24 hours later using RNAiMax following the RNAiMax protocol. siCon sequence is 5′-CGUACGCGGAAUACUUCGA-3′/5′-UCGAAGUAUUCCGCGUACG-3′, siXRN1 5′-AAUUAUUCCUCAAUGAUAGUG-3′/5′-CUAUCAUUGAGGAAUAAUUAC-3′, siXRN2 5′-UUUACAAAAGCUUAAUUCCAA-3′/5′-GGAAUUAAGCUUUUGUAAAGC-3′. Wildtype and catalytic dead XRN2 293 cells were a kind gift from the David Bentley laboratory [28].

Northern blot

Total RNA was collected using Trizol following the manufacturer’s protocol. 30ug RNA was run on a 15% TBE-Urea gel and stained with SYBR gold to determine equal loading. RNA was then transferred onto Amersham HyBond-N+ membrane and probed using ExpressHyb hybridization solution following manufacturer instructions. Specific probes for northern blot were: 5′-BIO-AAAATAAGAGCACCCGCTTC-3′ for precursor tRNA SerTGA and tRF-1001 detection, and 5′-BIO-AAACGAGGTAACTCCGGAGCACA-3′ for precursor tRNA CysGCA and tRF-1015 detection.

RT-qPCR

For small RNA detection by RT-qPCR, total RNA was collected using Trizol following the manufacturer’s protocol. 30ug RNA was run on a 10% TBE-Urea gel, stained with SYBR gold, and small RNAs between 15-50nt were excised from the gel. Small RNA was eluted overnight in elution buffer containing 1M Tris-HCl (pH 7.5), 2.5M sodium acetate, 0.5M EDTA, and 10% SDS. RNA was collected using phenol-chloroform extraction. RT-qPCR was performed using the miScript primer assay kit (Qiagen) following manufacturer instructions using tRF-1001 (5’-GAAGCGGGTGCTCTTATTTT-3’) and miR-21-5p (5’-TAGCTTATCAGACTGATGTTGA-3’) probes for qPCR.

Small RNA-seq library preparation and analysis

Small RNA-seq library preparation was similar to previously described [29] with addition of spike-in. Briefly, 1 ul (500 ul total volume) of Qiaseq miRNA library QC spike-in (Qiagen #331535) were mixed with 1 ug total RNAs before library preparation. The resulting libraries were size selected for 15–50 nt inserts and pooled for sequencing on Illumina NextSeq500 according to validated standard operating procedures established by the University of Virginia Genome And Technology Core, RRID: SCR_018883.

HDMYZ cells are an acute myeloid leukemia cell line. Small RNA sequencing data from HDMYZ cells was downloaded from GSE46968. There are three biological replicates for each actinomycin D time point. For actinomycin D small RNA sequencing, if the small RNA decreases to a size below the sequencing library cutoff, these small RNAs will not be cloned and counted. This does not affect the conclusions because the shortening is part of the degradation processes. Arabidopsis small RNA sequencing data was downloaded from GSE133461. There are two biological replicates for each total small RNA sequencing library and three biological replicates for each Ago RNA immunoprecipitation sequencing library. Small RNA sequencing data was first trimmed to remove adapters using cutadapt [30]. Only reads longer than 14 nucleotides were kept for mapping. Small RNAs were mapped using unitas 1.7.5 [31] allowing up to one mismatch. Isoforms of small RNAs were collapsed. Unnormalized mapped reads for miRNA and tRFs were imported into R and differential expression analysis was performed using DESeq2, with batch correction where necessary [32]. We required that at least 10 reads per small RNA across all samples to pass an expression threshold for differential expression analysis. Small RNAs were considered highly expressed if present at 50 normalized counts or higher.

ENCODE XRN2 eCLIP bigwig files (hg38) were downloaded from the ENCODE website (experiments: ENCSR657TZB, ENCSR655NZA, files: ENCFF323ABS, ENCFF305SDT, ENCFF782AAU, ENCFF723HML) [33,34]. Tracks were either screenshots from the UCSC genome browser, or deeptools was used to create metagene plots for tRNAs and miRNAs [35].

Half-life analysis

To determine half-life in HDMYZ cells at a sequence level, we first performed decay profile normalization on normalized counts as described in Sorenson et al 2018 [36]. Briefly, all normalized counts were normalized to the zero time point (T0). Then, stable small RNAs were identified as those RNAs that had an apparent increase over time. This happens because total cellular RNA decreases with transcription inhibition over time making stable small RNAs appear to increase if normalizing by total RNA. A decay factor was calculated for each time point. The decay factor is the average T0 normalized count for each time point for the stable small RNAs. All small RNAs counts were then divided by their respective decay factor for each time point. Decay profile normalized counts per small RNA were fit using the following equation:

Where A is abundance, A₀ is the initial normalized count at time zero (steady state abundance), t is time, and k is the decay rate. K was then used to find the half-life using the following equation:

For plotting purposes, small RNAs with half-lives predicted to be beyond 12 hours (the duration of the experiment) had their half-lives set to 12 hours. Phastcons scores were obtained using the bioconductor package phastCons100way.UCSC.hg19.

Ago2 RNA-immunoprecipitation

Stable Flag-HA-Ago2 293T cells were generated by 5-day puromycin selection after transduction with lentivirus-containing supernatant. Lentivirus expression plasmids pLJM1-Flag-HA-Ago2-WT (Addgene #91978) was a gift from Joshua Mendell [37], pMD2.G (Addgene #12259) and psPAX2 (Addgene #12260) were gifts from Didier Trono. Flag-HA-Ago2 was immunoprecipitated using either Flag-M2 magnetic beads from Sigma or anti-MYC mouse antibody 9E10 as a negative control in RIP lysis buffer containing 50mM Tris pH 7.4, 150mM NaCl, 0.5% Triton X-100, and RNase and proteinase inhibitors. Complexes were immunoprecipitated overnight at 4 degrees Celsius. Complexes were washed three times with RIP lysis buffer and RNA eluted with Trizol. Small RNAs were either processed for small RNA sequencing or northern blotting. Northern blotting was performed as previously described [6].

Statistical analysis

Statistical analyses are described in figure legends.

tRF-1s are highly abundant and unstable by half-life analysis

To test whether endogenous tRF-1s are inherently unstable, we determined half-life of each small RNA sequence in HDMYZ cells following actinomycin D (actD) treatment [38]. The small RNA sequencing data captured 1260 unique sequences (S1 Table). Most small RNAs were stable for the duration of the experiment (12 hours) (Fig 1B and S1 Table). Overall length distribution of small RNAs was unchanged by actD treatment (S1A Fig). The large number of small RNAs with half-lives of at least 12 hours is primarily driven by miRNAs (Fig 1C), while there are significantly fewer small RNAs that have half-lives of less than 12 hours. Surprisingly, there were very few tRF-1s detected after 12 hours of actD, and the seven tRF-1s that were detected have very short half-lives (median half-life of 0.28 hours) compared to stable miRNAs (median half-life greater than 12 hours), tRF-3s, and tRF-5s (Fig 1C). There was no clear pattern for tRF-1s that remained at 12 hours. The tRF-1s ranged from 14 to 25 nucleotides in length, with an average length of 21 nucleotides. The tRF-1s arose from various parental precursor tRNAs: tRNA-Leu-TAG-3-1, tRNA-His-GTG-1-7, two 3′-poly-U tail isoforms from tRNA-Val-TAC-1-2, tRNA-Thr-AGT-1-2, tRNA-Asn-GTT-2-2, and tRNA-Ser-GCT-4-1. Therefore, it is currently unclear what feature contributes to these tRF-1s relative increased stability in this cell line. The miRNA half-life is consistent with what is in the literature, which is more than 24 hours for the guide strand [39]. There was only one tRF-leader that was captured in this small RNA sequencing library. The size ranges of tRF-leaders are less than standard small RNA sequencing size selection cutoffs (~15nts) and so are not accurately captured by sRNA seq. Therefore, in this paper we have mainly focused on other highly abundant tRFs.

Recognizing that there are far more tRF-1s at steady state than are present at 12 hours of actD, we next compared the one-hour and zero-hour time points in HDMYZ cells treated with actD to get a more global view of the turnover kinetics of tRF-1s. tRF-1s were selectively depleted by one hour with a median log2 fold change of -3.425 (Fig 1D). miRNAs, tRF-5s, and tRF-3s changed very little by one-hour, with a median log2 fold change of 0.0409, 0.273, and 0.0573, respectively. Note that tRF-1s have the highest median expression among tRFs and miRNAs at steady state or zero-hour actD treatment (median expression of 388.61 normalized counts), and is depleted the most by 12 hr. (S1B Fig). We selected three annotated tRFs [40] in each known tRF class to analyze turnover kinetics for the duration of the actD treatment. The turnover kinetics of tRF-1037 (Fig 1E) has rapid, single step turnover kinetics compared to tRF-5003 and tRF-3011a (Fig 1F and 1G). To our knowledge, this is one of the shortest half-lives for an abundant sRNA detected in human cells.

To determine if tRF-1s are unstable across different cell types, we treated 293T cells with actD for up to 12 hours and measured tRF-1 and miRNA levels by RT-qPCR. We analyzed the turnover kinetics of tRF-1001 because it is the most abundant tRF-1 in 293T and has been previously shown to have biological function [25]. Indeed, tRF-1001 is highly unstable in 293T as well (Fig 1H). These data suggest that tRF stability is differentially regulated based on tRF class and that instability of tRF-1 is not cell-type specific.

Co-depletion of XRN1 and XRN2 increases tRF-1 levels globally

tRF-1 biogenesis is well understood [21], however, tRF-1 decay is uncharacterized. In order to identify the ribonuclease(s) (RNases) responsible for the very rapid tRF-1 turnover, we examined RNases in the XRN and DIS3 families, because they have been implicated in turnover of some sRNAs [41–43]. Knockdown of the XRN but not DIS3 families showed the tRF-1001 level was increased by Northern blot (Fig 2A). These data pointed towards the 5′-3′ exoribonuclease 1 and 2 (XRN1 and XRN2) as being putative tRF-1 regulators. To determine if the effect of XRN1 and XRN2 on tRF-1s was global, we performed small RNA sequencing after co-transfection of siXRN1 and siXRN2 (siXRN1/2) in 293T cells. RNA was size selected to capture small RNAs shorter than 50 nucleotides, which we expect to cover the majority of annotated tRF-1s. Consistent with the hypothesis that XRN1/2 degrades tRF-1s, several tRF-1s were significantly increased following siXRN1/2 (Fig 2B and S2 Table). Analysis of small RNA changes by subclass revealed specific increases in intron-tRF and tRF-1 classes at a global level following XRN1/2 depletion (Fig 2C). Consistent with XRN1/2 regulating tRF-1s in several cell types (and species, as discussed later), reanalysis of small RNA sequencing after knockdown of XRN1/2 in HeLa cells also shows an increase in reads from the trailer region, consistent with an increase in tRF-1 levels (Fig 2D) [44].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. XRN1/2 depletion increases tRF-1 expression.

A) tRF-1001 levels by Northern blot after co-knockdown of DIS3, DIS3L, DIS3L2, or co-knockdown of XRN1 and XRN2 in 293T. B) MA plot of small RNA alterations following co-knockdown of XRN1 and XRN2 (siXRN1/2). Red points are statistically significant. Differential expression determined by DESeq2. C) Boxplot of small RNA alterations following siXRN1/2 by class. Red points are statistically significant. D) Metagene plot of tRNA trailer region in HeLa cells after siXRN1/2 (data reanalyzed from GSE29116). Shown is the 3′ end of the tRNA genic region. The trailer region shown begins just downstream of the 3′ end of the tRNA gene and ends 100 bases downstream. tRF-1s are generally 15–22 bases long and are shown in yellow brackets.

https://doi.org/10.1371/journal.pgen.1010755.g002

XRN2 degrades tRF-1s

Since XRN1 and XRN2 may regulate tRF-1s at the level of production or degradation, we directly measured pre-tRNA and tRF levels by Northern blot in XRN1/2 depleted conditions before and after treating cells with actinomycin D (actD) for one hour. tRF-1s were clearly increased after co-depletion of XRN1 and XRN2, without an increase of its precursor, pre-tRNA SerTGA, indicating that XRN1 and/or XRN2 are involved in the degradation of tRF-1s (Fig 3A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. XRN2 degrades tRF-1s.

A) Northern blot analysis of tRF-1001 and parental pre-tRNA SerTGA. Cells were co-depleted of XRN1/2, then treated with actinomycin D for 1 hour. B) Northern blot analysis of tRF-1001, tRF-1015, and parental pre-tRNAs following depletion of XRN1 and XRN2 separately. Western blot analysis showing knockdown efficiency below. C) Northern blot analysis of degradation kinetics for tRF-1001, parental pre-tRNA, and let-7a-5p after XRN2 knockdown. Quantitation of northern blot below (n = 2). Error bars represent standard error. D) XRN2 eCLIP genome browser tracks for tRNA-Ser-TGA-1-1 (tRF-1001 parental pre-tRNA) and let-7a1 in HepG2 and K562. The tRNA trailer begins just downstream of the 3′ end of the tRNA-Ser-TGA-1-1 gene. E) Northern blot of tRF-1001 and parental pre-tRNA following doxycycline induced expression of wildtype or catalytic dead (D235A) XRN2 in 293 cells.

https://doi.org/10.1371/journal.pgen.1010755.g003

We initially knocked down XRN1 and XRN2 together due to concerns that one enzyme could compensate for the other [45], but proceeded to knock down each exoribonuclease separately to determine which exoribonuclease is responsible for tRF-1 degradation. Depletion of XRN2 alone was sufficient to increase tRF-1001 and tRF-1015 expression (Fig 3B). Furthermore, depletion of XRN2 stabilizes tRF-1001, tRF-1007, and tRF-1015 also on Northern blots (S3A Fig). We conducted a more refined actD time course to assess alterations in tRF-1 half-life when XRN2 is depleted. Depletion of XRN2 resulted in a half-life about twice as long as control knockdown (Fig 3C, including the quantitation from two separate experiments).

To determine if XRN2 can directly bind tRF-1s, we analyzed enhanced crosslinking and immunoprecipitation (eCLIP) from ENCODE [46]. Since XRN2 is a 5′-3′ exoribonuclease, we expect to find eCLIP reads near the 5′ end of tRNA trailer sequences and/or 3′ end of tRNAs. Indeed, we found XRN2 eCLIP reads at the tRNA SerTGA-1-1 trailer sequence, the sequence for tRF-1001, in both K562 and HepG2 cells (Fig 3D). The number of XRN2 eCLIP reads was higher for tRF-1001 compared to let-7a-1, the latter showing very few XRN2 eCLIP reads in both HepG2 and K562 cells (Fig 3D, lower panel). We also analyzed XRN2 eCLIP binding to tRNA and miRNA genes at a global level. XRN2 binding was highest on the 3’ halves of tRNAs, near the beginning of tRNA trailers for both K562 cells and HepG2 cells (S3B and S3D Fig). There was no clear pattern of binding of XRN2 to miRNAs (S3C and S3E Fig). Further evidence for XRN2 binding to tRNA trailers was recently published in Cortazar et al 2022, while our manuscript was under review. Cortazar et al 2022 show a strong peak for catalytic dead XRN2 at tRNA trailers [47]. Together, these data support a direct role for XRN2 binding to precursor tRNAs and subsequent degradation of tRF-1s.

Finally, we tested the role of the XRN2 catalytic activity in regulating tRF-1 levels. To do this, we determined tRF-1 levels in 293 cells with constitutive shXRN2 expression and either doxycycline inducible wildtype or catalytic dead (D235A) XRN2 [28]. Cells with constitutive knockdown of XRN2 expressed higher levels of tRF-1001 compared to 293T cells with intact XRN2 expression (Fig 3E). tRF-1001 was reduced below baseline upon overexpression of wildtype XRN2, while tRF-1001 increased when a catalytic dead form of XRN2 was expressed. Expression of shXRN2 and D235A XRN2 resulted in an even higher accumulation of tRF-1001 due to the additive effects of this condition. D235A XRN2 is a dominant negative protein [28], therefore overexpressing D235A XRN2 will further decrease the activity of any endogenous XRN2 that persists after shXRN2. Thus, XRN2 catalytic activity is required for XRN2 mediated degradation of tRF-1s.

XRN2 specifically blocks tRF-1s from entering RISC

Some classes of tRFs, particularly tRF-3s, have been identified in argonaute protein complexes and have been shown to reduce gene-expression via a miRNA-like mechanism [6,26]. tRF-1s, however, are generally absent or found at low levels in argonaute proteins [24,26]. We hypothesized that XRN2 mediated degradation prevents tRF-1s and potentially other small, single stranded RNAs such as intron-tRFs, from entering Ago2. To test this hypothesis, we immunoprecipitated Flag-HA tagged Ago2 (Flag-Ago2) after knocking down XRN2 in 293T cells and performed small RNA sequencing. First, we checked the amount of overlap in significant small RNA changes between siXRN2 input and siXRN1/2 from Fig 2. Of the 72 small RNAs that significantly increased in siXRN2 input, 31 overlapped with the small RNAs that significantly increased in siXRN1/2 small RNA sequencing (Fig 4A). Out of the 31 small RNAs that significantly increased in both datasets, one was a miRNA, one mapped to a piRNA locus, one was a misc-tRF, three were intron-tRFs, and twenty-five were tRF-1s. Only two small RNAs decreased in siXRN2 input and siXRN1/2. These data suggest that XRN2 alone was primarily responsible for tRF-1 instability.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. XRN2 depletion leads to an accumulation of tRF-1s, in-tRFs, and tRF-leaders in Ago2.

A) Overlap of significant alterations in siXRN1/2 total RNA and Flag-Ago2 input siXRN2 from small RNA sequencing. B) Barplot of total tRF-1s in Flag-Ago2 immunoprecipitation. C) MA plot of small RNA alterations in Flag-Ago2 siCon compared to Flag-Ago2 siXRN2. Red points are significantly increased small RNAs, blue points are significantly decreased small RNAs. Differential expression determined by DESeq2. D) Significant alterations in Flag-Ago2 siXRN2 compared to Flag-Ago2 siCon by small RNA class. Red points are significantly increased small RNAs. E) Heatmap of top 30 differentially expressed small RNAs in Flag-Ago2 siXRN2 relative to Flag-Ago2 siCon. Top 30 differentially expressed small RNAs are based on the DESeq2 adjusted p-value. A row-wise z-score transformation was performed. F) Barplot of tRF-1015 from Flag-Ago2 immunoprecipitation. G) Northern blot validation of tRF-1015 Ago2 association.

https://doi.org/10.1371/journal.pgen.1010755.g004

After confirming that depletion of XRN2 alone increases tRF-1s and not miRNAs in our input data (S2A and S2B Fig), we determined whether XRN2 depletion significantly alters small RNAs that bind Ago2 (S3 Table). Flag-Ago2 RNA immunoprecipitates pulled down miRNAs as expected (S2C Fig), indicating that our Ago2 immunoprecipitation was successful. tRF-1s were globally increased in the Flag-Ago2 bound fraction (Fig 4B). We note that tRF-1001 (tRF-1-tRNA-Ser-TGA-1-1) and tRF-1015 (tRF-1-tRNA-Cys-GCA-2-3) are among the tRFs that had the highest mean expression and show a large fold change in the Ago2 IP following siXRN2 (Fig 4C).

When separately examining classes of short RNAs, tRF-1s, tRF-leaders, and intron-tRFs significantly increase in Ago2 association following XRN2 depletion, while tRF-5s were generally unaffected (Fig 4D). A few tRF-3s significantly decreased in Ago2 association upon XRN2 depletion, but the fold changes were smaller than the fold change increase seen for tRF-1s, intron-tRFs, and tRF-leaders. A small number of miRNAs are altered in Ago2 following siXRN2, but there was no consistent trend of increase or decrease of the association (Fig 4D). Next we determined the enrichment of the top 30 significantly altered sRNAs in Ago2 RIP versus input. Several tRF-1s and intron-tRFs were enriched in Ago2 relative to input after XRN2 depletion (Fig 4E). For example, tRF-1015 went from 0.5 fold enrichment in control conditions to 1.6 fold enrichment upon loss of XRN2. We also used a Northern blot to validate the increased association of tRF-1015 in Ago2 after siXRN2 (Fig 4G). Together, these data support the hypothesis that XRN2 is necessary to prevent the aberrant accumulation of tRF-1s and potentially other single stranded small RNAs in RISC.

XRNs decrease tRF-1s and prevent tRF-1 entry into ago in Arabidopsis thaliana

XRN2 is a highly conserved and essential enzyme. The plant Arabidopsis thaliana (A. thaliana) has two nuclear XRNs, XRN2 and XRN3 (XRN2/3). Interestingly, in yeast and plants, XRNs can be inhibited by the toxic metabolite 3′-phosphoadenosine 5′-phosphate (PAP), which accumulates as a by-product of the sulfur assimilation pathway [48–50]. PAP is converted to AMP and inorganic phosphate by FIERY1 (fry1) in A. thaliana. We hypothesized that fry1 mutant (fry1mut) plants will accumulate tRF-1s in Ago1/2 due to inhibition of the XRN2/3 by PAP (Fig 5A). Using small RNA sequencing data from You et al. 2019, we focused on the relationship between tRF-1s, -5s, -3s, -leaders, and misc-tRFs and XRN inhibition in A. thaliana. Very few intron-tRFs were detected in the wildtype small RNA libraries. Manual inspection of the few introns that were greater than or equal to 14nts identified a total of 3 reads for the two wildtype small RNA sequencing libraries, perhaps due to 5′-hydroxyls and 3′ cyclic phosphates preventing cloning of more intron-tRFs during library preparation. Regardless, we found that tRF-1s are the most globally increased tRF class in two different fry1mut insertion lines (Fig 5B and 5C). tRF-1s were also increased in Ago1 and Ago2 IPs in the fry1mut lines. Only tRF-1s were globally increased in Ago1 RIP when comparing fry1mut 1–6 to wildtype (Fig 5D), with 36 tRF-1s showing significant accumulation in Ago1 RIP (Fig 5E). tRF-1, tRF-leaders and tRF-3s were all increased in the Ago2 RIP when comparing fry1mut to wt (Fig 5F), with 22 tRF-1s showing significant accumulation in Ago2 RIP (Fig 5G). These results are consistent with a role of XRNs in degrading tRF-1s and excluding them from Ago in plants, suggesting that this regulatory mechanism is important for controlling the specificity of RISC in at least two different species.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Arabidopsis thaliana FRY1 prevents inhibition of XRN2/3, which leads to reduced tRF-1 expression and prevents tRF-1 Ago1/2 entry.

A) Schematic of hypothesis: Normally, fry1wt plants degrade PAP, ensuring the proper function of XRN2/3. tRF-1 levels are kept low in this condition in both total and Ago1/2. In fry1 mutant plants, PAP accumulates and inhibits XRN2/3. tRF-1 levels increase in total RNA fractions and Ago1/2 immunoprecipitates. Chemical structure drawn using PubChem Sketcher [88] B,C) Small RNA alterations by class in fry1mut vs fry1wt plants. Red points are statistically significant. D,F) Median log2 fold change of fry1mut vs fry1wt Ago1 and Ago2 RNA immunoprecipitation followed by RNA sequencing. E,G) Scatter plot of tRF-1 expression in fry1wt and fry1mut plants after Ago1 and Ago2 RIP. Red points are statistically significant.

https://doi.org/10.1371/journal.pgen.1010755.g005