A snoRNA–tRNA modification network governs codon-biased cellular states

Significance

The dynamic equilibrium between tRNA supply and codon usage demand is a fundamental mechanism in gene expression, yet the regulators and consequences remain poorly understood. On the other hand, the targets and functions for the vast majority of the large family of snoRNAs (>2,000 in humans) remain unknown. In this study, we used multiple approaches to identify a large snoRNA interactome, including nearly all nuclear-encoded tRNAs. These interactions control tRNA modifications, stability, and levels and affect dichotomous codon-biased gene expression programs in proliferation vs. development in human HEK293 cells and a mouse embryonic stem cell differentiation model. Together, our work revealed a snoRNA-controlled cellular translation economy: specific snoRNAs regulate target tRNA “supply”, which influences the corresponding mRNA codon usage “demand”.

Abstract

The dynamic balance between tRNA supply and codon usage demand is a fundamental principle in the cellular translation economy. However, the regulation and functional consequences of this balance remain unclear. Here, we use PARIS2 interactome capture, structure modeling, conservation analysis, RNA–protein interaction analysis, and modification mapping to reveal the targets of hundreds of snoRNAs, many of which were previously considered orphans. We identify a snoRNA–tRNA interaction network that is required for global tRNA modifications, including 2′-O-methylation and others. Loss of Fibrillarin, the snoRNA-guided 2′-O-methyltransferase, induces global upregulation of tRNA fragments, a large group of regulatory RNAs. In particular, the snoRNAs D97/D133 guide the 2′-O-methylation of multiple tRNAs, especially for the amino acid methionine (Met), a protein-intrinsic antioxidant. Loss of D97/D133 snoRNAs in human HEK293 cells reduced target tRNA levels and induced codon adaptation of the transcriptome and translatome. Both single and double knockouts of D97 and D133 in HEK293 cells suppress Met-enriched proliferation-related gene expression programs, including, translation, splicing, and mitochondrial energy metabolism, and promote Met-depleted programs related to development, differentiation, and morphogenesis. In a mouse embryonic stem cell model of development, knockdown and knockout of D97/D133 promote differentiation to mesoderm and endoderm fates, such as cardiomyocytes, without compromising pluripotency, consistent with the enhanced development-related gene expression programs in human cells. This work solves a decades-old mystery about orphan snoRNAs and reveals a function of snoRNAs in controlling the codon-biased dichotomous cellular states of proliferation and development.

Codon and amino acid usage is nonrandom in messenger RNAs (mRNAs) across three domains of life (1). The transfer RNA (tRNA) pool is dynamically regulated in development and in response to the environment via tissue-specific expression, chemical modification, splicing, and charging (aminoacylation) (2). Biased usage of codons and amino acids in mRNAs and corresponding changes in the tRNA pool create regulatory mechanisms for translation speed, protein function, and gene expression programs in development (3, 4). Disrupted balance underlies various human diseases (5). For example, methionine (Met) is highly enriched in proteins involved in basic cellular processes, such as translation, splicing, and mitochondrial respiration (6). The ability of Met to scavenge reactive oxygen species (ROS) by reversible oxidation and repair is essential in protecting important proteins with long half-lives and/or close to ROS source. Both nuclear and mitochondrial genome–encoded mitochondrial proteins are enriched in Met, reflecting a convergent evolution in the usage of codons and amino acids. However, factors that regulate global tRNA supply and mRNA codon demand remain largely unknown.

Small nucleolar (sno)RNAs are a large family of noncoding (nc)RNAs in eukaryotes and archaea that often use antisense guide sequences to recognize RNA targets (7–12). The human genome encodes ~2,000 small nucleolar RNAs (snoRNAs), many of which are differentially expressed in cell types and development (13, 14). Most snoRNAs are classified into two types, where C/D snoRNAs guide the 2′-O-methyltransferase (MTase) Fibrillarin (FBL) to catalyze 2′-O-methylation (Nm), and H/ACA snoRNAs guide the pseudouridine synthase Dyskerin (DKC1) to catalyze pseudouridylation (Ψ) (Fig. 1 A and B). Evidence for additional snoRNA targets, such as mRNAs, tRNAs, and other ncRNAs, remains limited and sometimes controversial (15–21). The vast majority of snoRNAs have no known targets and are called orphans. Genetic studies have linked snoRNAs to many physiological and pathological conditions (10), such as the neurodevelopmental disorder—Prader–Willi syndrome, metabolic disorders, viral infections, and cancer (22–25); however, our limited knowledge of snoRNA targets made it difficult to study their functions.

Fig. 1.

PARIS2 identifies snoRNA targets. (A and B) Models of C/D and H/ACA box snoRNP complexes. (C and D) Strategies for the PARIS2 and validation. (E) PARIS2 captured three major types of snoRNA targets. (F) RMscores for 98 known rRNA Nm sites. PARIS2 captured 95, missing 18S-Cm463, 28S-Um4499, and 28S-Gm4500. (G) Same as panel F. PARIS2 identified guide snoRNAs for all 6 known Nm sites, for which guide snoRNAs were previously unknown in snoRNABase (26–28). (H) Identification of rRNA Nm sites from RMS (29). Nm sites are defined as P < 0.05 in RMscore differences between siCtrl and siFBL (n = 121), regardless of whether the changes were positive or negative. (I) Alignments of chordate and plant D101 D’ guide sequences, based on Rfam clan CL00074. (J and K) PARIS2 chimeric reads (J, 90 reads) and structure model (K) supporting the 28S-D101 interaction. (L) RMS data supporting the predicted Gm3628 and its loss upon FBL KD. P values: unpaired two-sided t tests.

Here, we use PARIS2 to identify a large number of snoRNA targets, including an extensive and conserved snoRNA–tRNA interaction network that controls tRNA modifications, stability, activity, and codon-biased gene expression. The biased codon usage, especially for Met, tips the balance between the dichotomous cellular states of proliferation and differentiation and skewed germ layer potential of mouse embryonic stem (mES) cells in favor of mesoderm and endoderm fates. Together, we found a snoRNA-guided tRNA modification mechanism governing codon-biased cellular states.

Results

Global Identification of snoRNA Targets Using PARIS2 and dRMS (Denatured RMS).

To identify snoRNA targets, we applied PARIS2 to total RNA, chromatin-associated RNA, and antisense-oligo enriched snoRNAs in human cell lines and induced pluripotent stem cell-derived lineages (Methods and Fig. 1C) (30, 31). In PARIS2, psoralen cross-linking of RNA duplexes, proximity ligation and sequencing reveal transcriptome-wide RNA interactions. Specifically, cells were first cross-linked with psoralen, and then, total RNA or chromatin-associated RNA was extracted and fragmented for PARIS2 library preparation. Alternatively, 46 snoRNA families, including 36 orphans, were enriched from the cross-linked total RNA by biotinylated antisense oligos for PARIS2 experiments. Together, these three approaches allowed us to identify the targets of a broader group of snoRNAs with higher sensitivity (Methods).

To validate these interactions, we employed multiple alternative approaches, including RiboMeth-seq (RMS) to map 2′-O-methylation (29, 32) and crosslinking and immunoprecipitation (CLIP) to map protein–RNA interactions and protein-bound RNA–RNA interactions (Fig. 1D) (17, 33). The commonly used RMS hydrolyzes RNA at high pH, where 2′-O-methylation protection of the phosphodiester bond is detected by sequencing (34–36). However, stable RNA structures and dense modifications strongly skew the fragmentation and reverse transcription, impeding its application to many ncRNAs. We developed a dRMS method, where stronger denaturation in the presence of 95% dimethyl sulfoxide (DMSO) during fragmentation increased the efficiency, uniformity of RNA fragmentation, and detection efficiency (Methods and SI Appendix, Fig. S1).

PARIS2 experiments revealed thousands of target sites for hundreds of snoRNAs, including rRNAs, snRNAs, nearly all nuclear-encoded tRNAs, and many other ncRNAs (Fig. 1E and Dataset S1, n = 7,531 interactions after CRSSANT clustering). C/D snoRNA targets were captured more efficiently than H/ACA snoRNA targets; therefore, we focused on C/D snoRNAs for initial validation. PARIS2 captured significant fractions of PLEXY-predicted low-energy snoRNA–rRNA interactions and known targets in published databases at various minimal free energy (MFE) cutoffs (SI Appendix, Fig. S2 A–C). Known interactions are ranked among the top PARIS-derived contacts (SI Appendix, Fig. S2D) (31). EZH2 is not only a transcription repressor and lysine MTase but also a chaperone that facilitates the assembly of C/D box snoRNAs by interacting with FBL (32). Out of 98 known rRNA Nm sites in snoRNABase, the vast majority have reduced Nm levels upon disruption of snoRNPs by both FBL and EZH2 knockdown (KD) (29, 32), and PARIS2 captured 95 of them (Fig. 1F). For known Nm sites in rRNAs, guide snoRNAs were either unknown or only predicted in the snoAtlas database (n = 6), which were all identified by PARIS2 (Fig. 1G and Dataset S2, n = 98 for known Nm sites). De novo identification of potential sites where Nm levels are reduced after FBL KD also revealed previously unknown sites, a subset of which are supported by PARIS2 (blue, n = 13, Fig. 1H and Dataset S2, n = 121 for de novo determined Nm sites). For example, the D’ guide of the orphan D101 is highly conserved in animals and plants (Fig. 1I and SI Appendix, Fig. S2E). PARIS2 and structure modeling revealed a target at 28S Gm3628 (Fig. 1 J and K). FBL is essential for cell survival; therefore, only partial KD is possible, leading to modest reduction of methylation level at G3628 (Fig. 1L), consistent with most other Nm sites on rRNAs as described before (Fig. 1 F and G) (29). Similarly, analysis of published RMS data confirmed the reduction of G3628 methylation level after EZH2 KD (SI Appendix, Fig. S2F) (32).

Combining PARIS2 and dRMS, we validated known and further identified multiple new Nm sites on several small RNAs, including spliceosomal snRNAs U1, U2, and U6, 7SL in the signal recognition particle, and snoRNAs (SI Appendix, Figs. S3 and S4 and Dataset S3, n = 40,701 nucleotide positions in these ncRNAs). Together, these studies confirmed the accuracy of PARIS2 and dRMS and revealed by far the largest numbers of snoRNA targets across multiple ncRNA types.

A Global and Conserved snoRNA–tRNA Interaction Network.

The PARIS2 dataset expanded known eukaryotic snoRNA–tRNA interactions from two (16) to more than 900, including nearly all nuclear-encoded tRNAs (Fig. 2A and SI Appendix, Fig. S5A). For C/D snoRNA-target chimeras, fragments mapped to snoRNAs piled around the D/D’ guides, as expected (Fig. 2 B and C). To test whether PARIS2-captured interactions are energetically favorable, we shifted tRNA-mapping fragments. The PARIS2 chimeras, but not randomly shuffled ones, produced a deep MFE valley at the target sites (Fig. 2 D and E). Furthermore, snoRNA–tRNA chimeras often extend beyond mature tRNA transcripts, suggesting that the interactions occur on pre-tRNAs prior to its folding into stable 3D structures and processing (Fig. 2F). The precise order of snoRNA-guided modifications and processing events, such as removal of leader and trailer sequences, remains to be determined. We noticed that a subset of snoRNAs bind both tRNAs and rRNAs, suggesting coregulation of these two components in the translation machinery (Fig. 2G and SI Appendix, Fig. S5B). For example, the rRNA-targeting D101 also binds multiple tRNAs, primarily Pro and Glu tRNAs, using the same conserved D’ guide (Fig. 2 H–J). Despite the lower cross-linking efficiency, we found several interactions between H/ACA snoRNAs and tRNAs that may guide Ψ, including the highly conserved TΨC motif (SI Appendix, Fig. S5 C–F). MFE-based prediction of individual snoRNA–tRNA interactions and comparison with chimeric reads from PARIS2 further confirmed the validity of a large number of them, revealing specific modification hotspots in various tRNAs (SI Appendix, Figs. S6 and S7). In particular, several snoRNAs encoded in the introns of Rpl13a, i.e., U32A, U33, U34, and U35A, were identified as mediators of cellular stress (18, 37). Our PARIS2 analysis revealed that several tRNAs, especially for Gly and Val, are major targets of these snoRNAs, in addition to rRNAs (Fig. 2G and SI Appendix, Fig. S6).

Fig. 2.

PARIS2 and dRMS reveal a global network of snoRNA–tRNA interactions. (A) The global snoRNA–tRNA network supported by >=5 reads. Read numbers are listed after the RNA name. (B) For chimeras connecting C/D snoRNAs, the coverage was averaged over the snoRNAs. The example chimeras have one arm (left side) mapped to the D’ and D guide motifs. For the metagene analysis, the position after the D’ guide is set to 0. (C) For all snoRNA-target pairs, coverage of the arms mapped to snoRNAs is summed up in red. Blue lines are the average length distribution of snoRNAs (x axis: nucleotides). Positions for the average 10 nt D’ and D guides are labeled. (D) For all C/D snoRNA–tRNA duplexes detected by PARIS2 and CLIP, each duplex was shifted to the left or right of the target Nm site. PARIS2 DGs were shifted for MFE calculation (Top, n = 490). MFE medians were calculated for each position (Bottom). (E) Same as panel D, except that randomly positioned snoRNA–tRNA interactions on the experimentally determined snoRNA–tRNA pairs (n = 297 successful shuffles) were used to generate the matrix (Top) and medians (Bottom). (F) Coverage of all PARIS2 reads mapped to tRNAs is normalized to max = 1. (G) Numbers of reads supporting snoRNA–rRNA (blue dots) and snoRNA–tRNA (red dots) interactions compared side by side. (H) SNORD101–tRNA interactions. (I) Alignments of PARIS2 reads supporting D101 interactions with Pro and Glu tRNAs. (J) Structure model of the D101 interactions with Pro-HGG and Glu-CUC. In HGG, H stands for A, C, or U.

CLIP Confirms snoRNP Interactions with tRNAs.

To validate the global snoRNA–tRNA network, we analyzed PAR-CLIP and eCLIP data for human snoRNP proteins (17, 38, 39). Earlier studies failed to identify snoRNP-tRNA interactions due to lack of normalization. Using proper normalization and false-positive controls (Methods and SI Appendix, Fig. S8 A–E), we found that human FBL, NOP56, and NOP58 bind between 60% and 93% of cyto-tRNAs, while human DKC1 binds 25 to 99% cyto-tRNAs, in addition to the known rRNA and snRNA targets (Fig. 3 A and B and SI Appendix, Fig. S8 F–H). CLIP occasionally produces hybrid reads from interacting RNAs (40, 41). Reanalysis of CLIP data using CRSSANT (42) revealed a few snoRNA–tRNA chimeras, most of which are consistent with PARIS2 data (Dataset S4, n = 954 interactions supported by either PARIS2 or CLIP). Similarly, yeast CLIP of FBL/NOP1, NOP56, and NOP58 enriched between 59 and 85% of cyto-tRNAs (Fig. 3 C–E). Together, PARIS2, structure modeling, and CLIP analysis revealed nuclear tRNAs as a major group of targets of snoRNAs.

Fig. 3.

CLIP confirms interactions between snoRNP proteins and tRNAs. (A) RNA enrichment ratios in PAR-CLIP over 20 to 200 nt sRNA-seq input, and eCLIP over size-matched input, normalized to the median of 21 mt-tRNAs. Numbers in parentheses indicate genes in each type. Data were plotted in as violins and box plots (when n > 10) or individual dots. miRNAs served as negative controls for snoRNP CLIP where vertical dash lines indicate the ratio above which 5% miRNAs are considered enriched (FP = 0.05). For AGO2, the FP = 0.05 was defined using SNORD or SNORA RNAs. The blue-colored numbers are the % of RNAs above the FP = 0.05 cutoff. (B) Pairwise Pearson correlations among human snoRNP CLIP experiments. (C) Enrichment ratio of yeast RNA relative to input control. Dashed line: the twofold cutoff for calculating % of enriched RNAs. (D) Venn diagram of enriched tRNAs from the 3 yeast CLIP experiments, for tRNAs with RPM > 1 and enrichment ratio >=2. (E) Pairwise Pearson correlations among all yeast snoRNP CLIP experiments.

FBL is a Master Regulator of tRNA Modification.

To determine the functions of FBL and DKC1, we performed mass spectrometry on purified tRNA and 18S/28S rRNAs. Nm levels were reduced in both 18S/28S rRNAs and tRNAs upon FBL KD, while DKC1 KD did not change tRNA Nm levels except Cm (Fig. 4 A and B). Both FBL and DKC1 are essential for cell survival (43–45); therefore, absolute measurement of FBL and DKC1-dependent Nm and Ψ sites is impossible in the partial KD cell lines. Interestingly, we observed larger reductions in Nm levels in tRNAs than in rRNAs after FBL KD, suggesting stronger dynamic regulation of tRNA Nm levels. Ψ level was reduced in rRNAs but not tRNAs after DKC1 KD, suggesting either interactions that do not guide tRNA modifications, or only few tRNA Ψ sites are catalyzed by DKC1 (Fig. 4C). Surprisingly, Ψ was reduced after FBL KD in tRNAs, even though FBL does not have Ψ synthase activity, indicating cross-regulation among RNA modifications. The reduction of Cm upon DKC1 KD (Fig. 4B) and reduction of Ψ upon FBL KD (Fig. 4C) are unexpected and could be due to several indirect mechanisms. C/D and H/ACA snoRNAs may bind and guide modifications on each other (Fig. 3A) (17). Therefore, loss of C/D snoRNP activity may compromise H/ACA snoRNP functions and vice versa. Alternatively, some of the modifications on tRNAs may be necessary for other modifications (e.g., Nm on tRNAs may be required for Ψ modification). Furthermore, Nm and Ψ can also be installed by stand-alone protein enzymes independent of snoRNAs (46); therefore, some of the modification reductions may be due to secondary defects of other tRNA modification enzymes.

Fig. 4.

FBL is a master regulator of tRNA modification and stability. (A–G) Quantitative LC/MS analysis of Nm and Ψ levels from total tRNA and 18S/28S rRNAs after DKC1 and FBL KD. P values: two-sided unpaired t tests. n.s.: not significant. (H–K) SYBR Gold stained total RNA after FBL or DKC1 KD and 4 h of 0.25 mM sodium arsenite (As) treatment. (Top) shorter exposures of tRNA bands. (L–O) Quantification of RNA fragments (panels L and M). tRNA halves were quantified for the bars (N and O). Error bars are ±SD of n = 2 independent experiments. (P and Q) tRNA in vitro cleavage assay. tRNA bands, ~70 to 90 nucleotides, were isolated from total RNAs after electrophoresis, and digested by recombinant human ANG. (R) Melting temperature of total tRNAs measured on a thermocycler. Replicates n = 5. First derivative was calculated. (S and T) Models for the role of FBL in tRNA modification and stability. The precise order of events in processing, folding, and other modifications is unclear.

To identify Nm sites in tRNAs, we applied dRMS to control and siFBL HEK293 cells (SI Appendix, Figs. S9 and S10 and Dataset S3). Reduced Nm levels upon FBL KD were observed in 149 sites, among which, PARIS2 captured guide snoRNAs for 15 sites (SI Appendix, Figs. S9 and S10). KD of EZH2, a known oncogene required for snoRNP assembly, also reduced tRNA modifications, suggesting a connection of snoRNA–tRNA interactions to cancer (29, 32). Further quantification showed that multiple other tRNA modifications, such as m¹A, m³C, m⁵C, and dihydrouridine (D), were also reduced in tRNAs upon FBL KD, but not DKC1 KD, indicating that snoRNA-guided Nm sites are needed for some of the other modifications (Fig. 4 D–G).

To determine whether yeast tRNAs are modified by snoRNPs, we reanalyzed published RMS data in three yeast mutants (SI Appendix, Fig. S11 A and B). Bcd1 encodes an essential factor in snoRNP assembly. The bcd1-D72A mutation causes cells to have low steady-state levels of box C/D snoRNAs, resulting in significant loss of Nm levels (47). The Dbp3 RNA helicase participates in snoRNA processing and recycling (48). The Dbp7 RNA helicase is required for snoRNA-dependent ribosome assembly (49). Loss of Bcd1 resulted in greater reduction of Nm levels in rRNAs than KD of Fbl/Nop1 and the Dbp3 or Dbp7 KO yeast (SI Appendix, Fig. S11C). Analysis of RMS data in bcd1-D72A and Dbp3 KO yeast revealed several sites in tRNAs with reduced Nm (Dataset S5, n = 26,747 nucleotide positions with calculated Nm levels, SI Appendix, Fig. S11 D–G).

FBL is Required for Global tRNA Stability.

To determine whether snoRNPs affect tRNA stability, we knocked down FBL and DKC1 in HEK293 and A549 cells using siRNAs (SI Appendix, Figs. S1I and S12A). FBL, but not DKC1, KD increased fragments in the 15-50 nt range, either in the absence or presence of oxidative stress (arsenite, Fig. 4 H–K). The fragments include tRNA halves (~34 and 40 nts) and shorter ones below 20 nts from D and T loop cleavage. Stable shRNA KD of FBL and DKC1 in HEK293, A549, and HepG2 cell lines and exposure to various stresses, such as arsenite oxidative stress, alkaline pH 9.0, and heat shock (SI Appendix, Fig. S12 B–G), confirmed the siRNA KD results, demonstrating the general role of FBL in tRNA stability.

To determine whether the increased fragmentation was due to intrinsic defects on tRNAs, we purified tRNAs from wild-type and siFBL cells and incubated them with the purified endonuclease angiogenin (ANG, Fig. 4 P and Q). tRNAs from FBL KD cells are more susceptible to cleavage, generating a wide range of sizes, most of which are tRNA halves (50). The melting temperatures of purified total tRNAs were not changed after FBL KD, indicating that tRNAs were mostly folded properly (Fig. 4R). RNA-seq of fragments in the 15–50 nt range revealed increased global levels of cytosolic tRFs in siFBL, compared to siCtrl and siDKC1 cells (SI Appendix, Fig. S12 H–K). Together, these studies showed that FBL and the C/D snoRNPs act as a master regulator of global tRNA modification and stability (Fig. 4 S and T), consistent with their early binding to pre-tRNAs (Fig. 2F).

D97/D133 snoRNAs Target an Extensive Set of tRNAs.

The conserved snoRNAs D97/D133, eMet-CAU tRNA, and their partners, form the strongest snoRNA–tRNA subnetwork (Fig. 5A and SI Appendix, Fig. S13 A and B). Clustering resolved two duplex groups (DGs) connecting the D’/D guides to two distinct regions on eMet-CAU (Fig. 5B). These DGs form strong duplexes, predicting Nm sites at Gm22 and Cm34 (Fig. 5 C and D). PARIS2 also revealed Leu-CAA-5-1 as a target for D97/D133, likely due to its close homology to eMet-CAU (Fig. 5 E–G and SI Appendix, Fig. S13C). Exhaustive search revealed D97 homologs in archaea, some of which were previously predicted to guide archaeal Met tRNA modification at C34 (12, 51). Alignments of tRNAs and guides in human, plant A. thaliana, and 4 archaeal species revealed a conserved duplex of at least 11 base pairs (SI Appendix, Fig. S13D). Interestingly, the target eMet-CAU tRNAs in plants and archaea have introns in the anticodon loop and participate in the extended duplex, further supporting that pre-tRNAs are snoRNA targets, and splicing likely occurs after the snoRNA-guided modifications. R-scape and CaCofold (52, 53) revealed two significantly covaried base pairs, in addition to 4 invariable base pairs, confirming deep functional homology among archaeal and eukaryotic guide RNAs for eMet Cm34 (SI Appendix, Fig. S13 E and F). This analysis also revealed a conserved function for the poorly studied tRNA introns in guiding tRNA modifications (15, 54).

Fig. 5.

D97/D133 snoRNAs target an extensive set of tRNAs. (A) D97/D133 and eMet-CAU snoRNA–tRNA subnetworks. (B) PARIS2 reads supporting D97/D133 interactions with eMet-CAU tRNA. Blue vertical lines represent the start of the mature tRNA. D’ and D guide sequences are highlighted in gray boxes. Each DG is one group of gapped reads that support one RNA duplex. (C and D) Models of D97 interactions with eMet-CAU (MFE in kcal/mol). DG1 supports two alternative conformations with the D’ guide. (E–G) PARIS2 gapped reads and models of D97 interactions with Leu-CAA-5-1 tRNA (MFE in kcal/mol). (H) PARIS2 gapped reads for additional D97 snoRNA targets, secondary structure models, and MFE (kcal/mol). (I) dRMS analysis of D97/D133 target tRNAs in siCtrl, siFBL, and D97/D133 double KO HEK293 cells. RMscore: RMS score. P values: unpaired two-sided t tests. n.s.: not significant. n = 4 samples for each group. (J and K) Heatmap for PLEXY predicted (J) and PARIS2/CLIP identified (K) Nm sites guided by D97. Each row is a tRNA gene (n = 430 rows), grouped by anticodon. Shades of blue represent D guide targets, whereas shades of red represent D’ guide targets. All tRNAs are aligned to a standard model of 73 nts plus insertions at the D-loop (not to be confused with the snoRNA D box/guide) and V-loop. Three hotspots in the tRNAs are labeled: 10, 22, and 34, and targeted by D’ and D guides, respectively.

Other D97–tRNA interactions are also supported by strong duplexes despite the lower numbers of chimeric reads (Fig. 5H). The Nm levels at 6 sites in 5 tRNA targets are reduced either in FBL KD or D97/D133 double KO, or both, confirming the extended interaction network (Fig. 5I and SI Appendix, Fig. S13G). The differential effects of FBL KD and D97/D133 snoRNA KO on Nm levels suggest additional snoRNA guides for these sites (Fig. 5I and Dataset S6, n = 44,540 for all predicted snoRNA–tRNA interactions). The substoichiometric and variable Nm levels suggest heterogeneity in the tRNA population and potential alternative functions in the differentially modified tRNA molecules. We further used computational prediction to validate the subnetwork captured by PARIS2 (Dataset S6). Predicted D97 targets on all 432 human cyto-tRNA genes were aligned to a standard tRNA (Fig. 5J), revealing many potential sites, and three hotspots (10, 22 and 34), a subset of which were captured by PARIS2 and/or validated by dRMS, such as Gm22 and Cm34 in Leu and eMet tRNAs, Arg-UCU Gm10, and Ile-AAU Am38 (Fig. 5K). Prediction of D133 targets revealed primarily interactions with the D guide due to divergence of sequence of the D’ guide (SI Appendix, Fig. S13H). RNA-seq of RNA fragments in the 15 to50nt range confirmed the elevated levels of tRNA halves after D97/D133 KO (SI Appendix, Fig. S14 A–C). Together, the integration of PARIS2 and optimized dRMS revealed an extensive and deeply conserved D97/D133-tRNA subnetwork.

D97/D133 snoRNAs Balance the tRNA Pool for Efficient Translation.

The extensive set of D97/D133 targets and the deeply conserved interactions with Met/Leu tRNAs across archaea and eukaryotes suggest that this network is essential; however, little is known about their functions. CRISPR KO of either or both D97/D133 in HEK293 cells significantly reduced cell growth (Fig. 6A). snoRNA overexpression (OE) rescued the defects, confirming that the snoRNAs are essential (Fig. 6B). Labeling of nascent peptides by the Met analog HPG revealed dramatically reduced global translation in all KO strains (Fig. 6 C and D). To determine the cause of the defects, we performed RNA-seq and ribosome profiling (ribo-seq) (55) (SI Appendix, Fig. S15 A–D). Single KO strains did not change tRNA levels, even though each snoRNA paralog is necessary for tRNA modifications; however, the double KO reduced expression of several tRNAs, including eMet, Leu, Ile, etc. (Fig. 6E), many of which are targets of D97/D133 (Fig. 5). On the other hand, Pro tRNAs, which are not D97/D133 targets, were significantly up-regulated, suggesting secondary effects of the KOs on the tRNA pool.

Fig. 6.

D97/D133 snoRNAs balance the tRNA pool for efficient translation. (A) Proliferation of WT and single and double KO cells (two clones each). P values from two-sided unpaired t tests. P values for D97 KO vs. WT: 0.30 and 0.38 (day 4), 0.0024 and 0.0084 (day 5). P values forvalues for D133 KO vs. WT: 0.0005 and 0.0004 (day 4), 3.2E-05 and 3.6E-06 (day 5). P values for D97/D133 double KO vs. WT: 0.0001 and 5.8E-05 (day 4), 2.4E-06 and 2.5E-06 (day 5). (B) KO cell lines were rescued by snoRNA OE. P values < 0.01 are indicated in the figure (asterisks *, for all clones), based on two-sided unpaired t tests. (C and D) Nascent proteins were labeled using HPG, detected using Alexa Fluor 647 azide (C), quantified and normalized against total protein on the gel (D). (E) All 432 nuclear tRNA genes were grouped by anticodons, and then, log2-transformed ratios between KO and WT were normalized so that median = 1 for each KO/WT comparison. Error bars: SDs. (F) Rescue of cell growth defects by overexpressing eMet-CAU or 7 tRNAs in WT and KO cell lines. (G) Bicistronic luciferase reporter plasmids. 6xMet-ATG oligo was inserted after the start codon of Fluc. (H) WT and KO HEK293 cells were transfected with plasmids in panel G. The y axis of normalized Fluc/Rluc ratio corresponding control transfected control plasmid in the same cell line. All data are representative of at least three independent experiments. P values were shown for each KO cell line relative to WT.

To confirm that reduced tRNA activity and levels are responsible for the translation defects, we infected HEK293 cells with lentiviruses expressing no insert (control) or tRNA eMet-CAT, or a mixture of 7 lentiviruses expressing eMet-CAT, Arg-CCT, Gly-TCC, Lys-CTT, Ile-TAT, Sec-TCA, and Trp-CCA (7tRNAs), which are reduced in the D97/D133 double KO. After puromycin selection, these tRNAs are expressed at high levels (SI Appendix, Fig. S15 E and F). OE of both eMet-CAT and the 7-tRNA mixture partially rescued growth defects in KO lines (Fig. 6F). To test whether the Met-AUG codon bias is responsible for the reduced translation in KO cell lines, we further constructed reporters (Fig. 6G). The insertion of 6xMet-ATG decreased protein synthesis in KO cell lines relative to WT (Fig. 6H), confirming that the D97/133 KO caused eMet-CAU tRNA defects. Together, these studies demonstrate an essential role of snoRNA-guided modifications in maintaining a balanced tRNA pool for efficient translation.

D97/D133 Govern Met-Biased Gene Expression Programs.

Met is one of the few reversibly oxidizable amino acids in proteins. Dedicated enzymes in all three domains of life, such as Met sulfoxide reductases, repair oxidized Met to protect proteins with long half-lives or close to ROS sources (56). In particular, spliceosomal proteins, nucleic acid binding proteins, and mitochondrial proteins encoded by both the nuclear and mitochondrial genomes have some of the highest ratios of Met residues in the proteome in higher animals (6). On the other hand, the short-lived proteins involved in development, differentiation, and morphogenesis are relatively depleted of Met. Therefore, the D97/D133 targeting of tRNAs, particularly Met, is likely a mechanism to regulate Met codon and amino acid usage and the corresponding gene expression programs.

To test this hypothesis, we measured codon usage in both the transcriptome and translatome (ribosome-associated mRNAs). The usage of many codons, including Met-AUG, changed significantly in the D97/D133 double KO, and less so in the single KO strains (Fig. 7A). There is a clear enrichment of GC-ending codons and depletion of AU-ending ones, which correlate with stem cell self-renewal, differentiation, and multicellular functions (3, 4). Usage of the Met amino acid is the fourth most reduced in the double KO in both the transcriptome and translatome (Fig. 7B). Strong positive correlations were observed between the transcriptome and translatome in the codon and amino acid usage changes in double KO vs. WT (Fig. 7 C and D). The D97 and D133 single KO did not show the same trend of codon and amino acid frequency changes on the transcriptome, but changed codon and amino acid frequency on the translatome level, suggesting dose-dependent tRNA defects in the single vs. double KO cells (SI Appendix, Fig. S15 G–J).

Fig. 7.

D97/D133 govern Met-biased gene expression programs. (A) For nuclear-encoded mRNAs in RNA-seq (Upper) or ribo-seq (Lower) from WT and D97/D133 double KO HEK293 cells, ratios of expression levels were ranked. Usage of 66 codons—64 plus iMet and Sec—was calculated for the top 10% (up-regulated) and bottom 10% (down-regulated) mRNAs and weighted by expression levels. Blue and red: A/U- and G/C- ending codons. Asterisk (*): stop codons. M and m: eMet and iMet. U_uga: Sec. (B) Same as panel A, except that amino acid usage was calculated. n = 23 for 20 amino acids plus iMet, stop, and Sec. (C) Changes in codon and anticodon usage between D97/D133 double KO and WT. RNA-seq codon: usage was calculated for 66 codons in mRNAs from KO and WT. Ribo-seq codon: same as RNA-seq codon, except that ribo-seq was used. RNA-seq tRNA anticodon freq: tRNA levels were measured in the RNA-seq. (Left) RNA-seq vs. ribo-seq codon usage. The Inset x axis ratio = 1.276 indicates ratio of average NN[GC] vs. average NN[AU] in RNA-seq. The inset y axis ratio = 1.098 indicates the ratio of average NN[GC] vs. average NN[AU] in ribo-seq. P values after the ratios: two-sided unpaired t tests between the two codon groups. (Middle) RNA-seq mRNA codon freq. vs. RNA-seq tRNA codon frequency. Codons recognized by the same tRNA anticodons were merged. (Right) same as the Middle panel, except that ribo-seq mRNA codon frequency is the y axis. (D) Same as panel C, except that the amino acid usage was plotted. (E) Altered mRNAs based on log ratios of D97D133 KO vs. WT were median-normalized and tested for gene set enrichment using gene set enrichment analysis (GSEA) and MSigDB c5.all collection. Enriched gene sets were clustered based on term similarity and tested for enrichment of terms. GOBP term clusters consistent across RNA-seq, ribo-seq, and gene-specific codon usage (GSCU) were highlighted in blue boxes and linked by gray lines. (F) Same as panel E, except that the analysis was performed on ribo-seq mRNA levels. (G) Same as panel E, except that the analysis was performed on Met-AUG codon enrichment in mRNAs. (H) Same as panel E, except that the analysis was performed on translation efficiency (TE, mRNA levels in ribo-seq over RNA-seq). (I) Example GOBP terms with significantly depleted Met-AUG usage (n = 15,423 genes with APPRIS principal isoforms). (J) Same as panel E, except that the analysis was performed on TE. (K) Example GOBP terms with significantly enriched Met-AUG usage. (L) After inhibition of cytosolic translation by cycloheximide, the alkyne-containing Met homolog HPG was used to label nascent peptides (57). Then mitochondria were isolated from cells and labeled proteins were reacted with Azide–Alexa 647 and visualized on a gel (Left). Mitochondrial peptides were quantified (Right).

Gene ontology (GO) analysis (58, 59) of KO cell lines revealed significant downregulation of genes with basic cellular functions, such as ribosome, spliceosome, mRNA metabolism, and oxidative phosphorylation in both the double and single KO cell lines (Fig. 7 E and F and SI Appendix, Fig. S15 A and C), consistent with the reduced growth and translation (Fig. 6). Up-regulated GO terms on both the transcriptome and translatome levels include differentiation, development, and morphogenesis, among others (Fig. 7 E and F). Interestingly, Met is depleted in the up-regulated GO terms on the transcriptome and translatome levels and vice versa (Fig. 7G and SI Appendix, Figs. S16 and S17). GO analysis of translation efficiency did not reveal similar terms, suggesting that the gene expression alteration for these GO terms is primarily on the transcriptome levels (Fig. 7 H–K), likely due to changes in transcription and/or stability.

Both nuclear and mitochondrially encoded mitochondrial proteins are enriched in Met, suggesting convergent evolution to cope with oxidative stress (6). However, only the nuclear-encoded tRNAs are targeted by snoRNAs (Fig. 3). The reduction of mitochondrial translation as measured by ribo-seq (SI Appendix, Fig. S17C) is likely a secondary effect of cellular adaptation to coordinate translation in the two subcellular compartments (60). Nascent translation of mitochondrial genome encoded peptides, measured after cycloheximide inhibition of cytoplasmic translation, was all reduced in both the single and double KO cells, confirming the RNA-seq and ribo-seq measurements (Fig. 7L). Together, these studies revealed a critical role of Met codon usage in controlling the dichotomous gene expression programs.

Transcriptome and Translatome Codon Usage Depends On snoRNA Dose.

As shown above, the D97/D133 single and double KO cell lines reprogram the transcriptome and translatome to different extents (Fig. 7 A and B vs. SI Appendix, Fig. S15 G and H). To further quantify the gene expression reprogramming, we measured SDs of expression fold changes in KO vs. WT. Single KO primarily affected translation (Fig. 8 A, Top and Middle, higher variation on the translatome level), while double KO affected both the transcriptome and translatome (Fig. 8 B, Bottom, higher variation on the transcriptome level). Double KO induced bigger changes in relative tRNA levels (Fig. 8B, panels 1-2). On the transcriptome level, double KO induced larger differences in codon and amino acid usage, than single KOs (panels 3-6). On the translatome level, double KO induced similar changes in codon and amino acid usage as single KOs (Fig. 8B, panels 7-10). Comparing variations between RNA-seq and Ribo-seq, the single KO exerted more effects on the translational level, whereas the double KO already showed large differences on the transcriptome level, which persisted in the translatome level (Fig. 8B, panels 3 vs.7, 4 vs. 8, 5 vs. 9, and 6 vs. 10, and Fig. 8C).

Fig. 8.

Transcriptome and translatome codon usage depends on snoRNA dose. (A) Ratios of mRNA levels in KO vs. WT for RNA-seq and ribo-seq. mRNAs plotted: n = 1,682. Linear regression results on the right. (B) Nuclear tRNA anticodon groups (n = 48), mRNA codon (n = 66), and mRNA amino acid (n = 23) usage changes upon D97/D133 single and double KO. Panels 1 and 2: abundance of tRNA anticodons in KO vs. WT (WT set to 1). Panels 3-6: mRNA codon and amino acid usage on the transcriptome level. Panels 7-10: mRNA codon and corresponding amino acid usage on the translatome level. (C) SD of anticodon, codon, and amino acid usage plots in panel B. (D) Model for the effects of snoRNA-guided modifications on the tRNA pool, mRNA codon usage, translation, and cellular states. Single and double KO affect the transcriptome and translatome to different degrees.

Together, the RNA-seq and ribo-seq in D97/D133 single and double KO HEK293 cells suggest a global reprogramming of the transcriptome and translatome (Fig. 8D). Loss of the snoRNAs resulted in defective target tRNAs, especially eMet-CAU, accompanied by reduced usage of Met and other codons in the transcriptome and translatome levels, leading to an imbalance in two competing gene expression programs: proliferation vs. differentiation/ development/ morphogenesis. The tipped balance between GC- and AU-ending codons, which has been observed in other biological contexts of cellular proliferation vs. development (3, 4), is likely induced by the defects in D97/D133 target tRNAs here in HEK293 cells. The double KO induced larger differences on the transcriptome level, suggesting adaption of the cells to dramatically reduced levels of eMet-CAU and several other tRNAs (Fig. 8D, total mRNAs). Together, this quantification revealed dose-dependent effects of D97/D133 on the transcriptome and translatome programs.

Mouse D97/D133 Regulates Codon-Biased Stem Cell Differentiation.

Given the codon-biased induction of development-related gene expression programs in human HEK293 cells after D97/D133 KO, we tested whether these snoRNAs regulate mES self-renewal and differentiation into embryoid bodies after ASO KD of mouse Snord97 and Snord133 (D97/D133) (Fig. 9 A and B). Pluripotency-related mRNAs increased while markers for the three germ layers were skewed, favoring the mesoderm and endoderm (Fig. 9C). In particular, the cardiomyocyte (CM) Myh6 increased (Fig. 9D). These results are consistent with the upregulation of genes involved in differentiation, development, and morphogenesis in HEK293 cells (Fig. 7). To further analyze the roles of D133 snoRNA in CM differentiation (CMD), we knocked it out using CRISPR (SI Appendix, Fig. S18A). Cell growth slowed down significantly, similar to HEK293 cells (Fig. 9E). mES gross morphology and self-renewal remained the same (SI Appendix, Fig. S18B), yet all KO clones significantly increased the speed and efficiency of CM formation, from ~40 to ~75% (Fig. 9 F–H). The Myh6 mRNA was induced earlier and higher in the D133 KO throughout differentiation (SI Appendix, Fig. S18C). At the same time, mRNA and protein levels of pluripotency markers increased, such as Nanog, Sox2, and Oct4, similar to the ASO KD (Fig. 9 I–K). Together, these phenotypes confirmed the codon-biased gene expression programs in D97/D133 KO HEK293 cells.

Fig. 9.

Mouse D97 and D133 regulate stem cell differentiation. (A) Diagram of mES differentiation into embryoid bodies that contain three germ layers. (B–D) Expression of the snoRNAs, markers for pluripotency and germ layers, and CM mRNA Myh6 from EBs was measured by qRT-PCR. P values: unpaired two-sided t tests. (E) Proliferation of WT and D133 KO cells (clones #4 and #5). (F–H) WT and D133 KO mES cells (3 clones) were differentiated into CMs. Beating CM patches were counted every 3 d. P values are from two-sided t test, between each KO cell line and WT, and color-coded. (I) Expression of pluripotency and CM mRNAs in WT and D133 KO mES cells, measured by qRT-PCR. (J and K) Pluripotency factor protein levels in WT and D133 KO mES cells, measured by western blots. (L and O) Expression dynamics of pluripotency factors (L) (61), cardiac chamber morphogenesis factors (M), and mitochondrial RNAs (N and O) for the two genotypes across ES, EB, and CM. Stage-specific expression of mitochondrial RNAs is summarized in violin and box plots (N and O).

To determine the mechanisms driving the faster and skewed differentiation, we performed RNA-seq in mES, EB (embryoid body), and CM (SI Appendix, Fig. S18D). The mES D133 KO did not change relative tRNA levels between KO and WT (SI Appendix, Fig. S18 E and F), consistent with the HEK293 D97/D133 single KO lines. At each stage, the KO induced distinct differences in the transcriptome (SI Appendix, Fig. S19 A and B). Pluripotency TFs dropped from ES to EB and CM stages while cardiac genes were induced, confirming successful differentiation (SI Appendix, Fig. S19 C–E). Pluripotency factors Pou5f1 (Oct4), Sall4, and Nanog increased after KO in the ES stage (Fig. 9L), while cardiac development factors increased in D133 KO vs. WT (Fig. 9M), consistent with the qRT-PCR and western blots (Fig. 9 I–K). Mitochondrial transcripts significantly reduced in the ES stage (Fig. 9 N and O). The return to normal of mitochondrial gene expression in D133 KO CMs is surprising but consistent with the more efficient differentiation to CM.

Interestingly, several other mitochondrial metabolic processes, such as the one-carbon cycle, are up-regulated, suggesting dysregulation of metabolites with potential roles in altering epigenetic status of D133 KO stem cells (SI Appendix, Fig. S20 A and B). Gene set enrichment analysis revealed consistent upregulation of development-related terms, especially cardiac development (SI Appendix, Fig. S20 C–F). In contrast, mitochondrial electron transport, neurodevelopment-related genes were down-regulated in KO vs. WT, again consistent with the D133 KD studies (Fig. 7 E–G). This analysis revealed enhanced pluripotency and CM gene expression programs in mouse D133 KO, consistent with the skewed and more efficient CMD phenotype. Together, the KD and KO studies in human and mouse cells demonstrated a critical role of the D97/D133 target tRNAs in controlling the dichotomous cellular states of proliferation vs. development (Fig. 8D).

Discussion

This comprehensive study presents important findings and conceptual advances. We identify an extensive snoRNA targetome that includes multiple classes of ncRNAs, solving a long-standing mystery of orphan snoRNAs (Fig. 1). Integrated PARIS2 interactome capture, structure modeling, conservation analysis, normalized analysis of CLIP, and optimized dRMS modification mapping demonstrated a conserved global snoRNA–tRNA interaction network (Figs. 2 and 3). The 2′-O-methylation of pre-tRNAs by FBL controls global tRNA modifications beyond 2′-O-methylation and tRNA stability (Fig. 4). Specifically, we identify a subnetwork of D97/D133–tRNA interactions that are required for a balanced tRNA pool, cellular proliferation, and translation (Figs. 5 and 6). Loss of D97/D133 tipped the balance between the dichotomous programs of proliferation vs. development, as a result of the need for increased usage of the antioxidant Met in proliferation-related proteins (Figs. 7 and 8). Consistently, in mouse ES cells, codon-biased gene expression promoted and skewed stem cell differentiation without compromising pluripotency (Fig. 9). Together, this study revealed a class of regulators for codon-biased gene expression programs and cellular states.

Despite extensive efforts in the past 3 decades to identify snoRNA targets, the vast majority of snoRNAs remain orphans. Technological advances in our recent work (PARIS2) (31) and presented here (dRMS) are beginning to reveal complex modification networks with nucleotide resolution and provide mechanistic insights. While earlier studies suggested that these interactions typically range between 10 and 21 bps (62), the interactions found here span a bigger range. The frequently detected bipartite duplexes for D/D’ guides likely strengthened the stability of some of the otherwise weaker interactions (e.g., SI Appendix, Fig. S3 and Fig. 5). However, we cannot exclude the possibility that a subset of them represent target scanning intermediates (31) or may function beyond guiding modifications (e.g., folding chaperone, guided processing, and RNA quality control), like the well-studied U3, U8, and U13. Several tRNA Nm sites are substoichiometric, suggesting heterogeneous tRNAs where modification variants may have different functions (Fig. 5I and SI Appendix, Fig. S5).

The D97/D133 single and double KO reprogram the transcriptome and translatome via distinct mechanisms. In single KO strains, the reduced translation of Met-enriched proteins suggests that eMet tRNAs are defective, even though tRNA levels remain constant. The double KO significantly reduced several D97/D133 target tRNAs, especially eMet-CAU, reprogramming both the transcriptome and translatome to adapt to the skewed tRNA pool. Beyond the Met-AUG codon, the transcriptome of double KO cells exhibited a remarkable dichotomy of decreasing A/U ending codons, and increasing G/C ending ones, nominating the D97/D133-tRNA subnetwork as a regulator of the dichotomous programs rooted in wobble position bias of A/U and G/C content (3, 4). Together, our work revealed a snoRNA-controlled cellular translation economy: Specific snoRNAs regulate target tRNA activity and levels—the “supply”, which influences the corresponding codon usage in mRNAs—the “demand” (Fig. 8D).

It is not a coincidence that D97/D133 regulates the dichotomous programs of proliferation vs. development. The D97/D133 snoRNA–tRNA network is conserved in many species across eukaryotes and archaea (SI Appendix, Fig. S13). The antioxidant role of protein-intrinsic Met is conserved across all three domains of life and its usage is highly enriched in proteins that have longer half-lives and/or need higher ROS-resistance, and relatively depleted from genes involved in development, differentiation, and morphogenesis (Figs. 6 and 7). Therefore, the regulation of proliferation vs. development by D97/D133 represents an “evolutionary inevitability”. However, the precise mechanisms of the observed ES differentiation phenotypes remain poorly understood. In addition to the codon usage bias that directly alters levels of pluripotency and differentiation TFs, skewed mitochondrial metabolism is another possibility. The reduced respiration chain components and concurrent increases in other metabolic branches, such as one-carbon metabolism (SI Appendix, Fig. S20), may alter key metabolites that participate in DNA, RNA, and histone modifications to control stem cell epigenetics (63).

Since biased usage is common across all codons and amino acids, and underlies specific gene ontologies, the extensive snoRNA–tRNA network covering nearly all nuclear-encoded tRNAs suggests much broader impacts of snoRNAs in codon-biased gene expression programs in various biological and disease contexts. The large number of snoRNA-guided tRNA modifications may form a combinatorial regulatory code, like the histone modification code.

Methods

A full description of materials and methods is available in SI Appendix.

PARIS2 Library Preparation.

Briefly, cells were cross-linked with psoralen and 365 nm ultraviolet light (UV) and collected to extract total RNA or chromatin-associated RNA (31). snoRNAs and their cross-linked targets were enriched from the total RNAs using biotinylated antisense oligos. All RNA samples were fragmented, and cross-linked fragments were enriched using the DD2D gel method. After proximity ligation, and reversal of cross-links by 254 nm UV, the RNA samples were ligated to barcoded adapters, circularized, and amplified by PCR. The cDNA libraries were sequenced by NovaSeq 6000 (SE 100 bp).

CRISPR/Cas9-Mediated snoRNA Knockout.

All snoRNA guide RNAs were designed by using Broad’s CRISPick algorithm (64, 65). Guide RNAs used in CRISPR/Cas9 were cloned into the lentiCRISPRv2 vector (Addgene, 52961). HEK293T cells were transfected with equal amounts of lentiCRISPRv2 vectors carrying two-guide RNAs flanking the snoRNA to be deleted, followed by clonal selection under puromycin and clone expansion. Genotypic characterization was performed using PCR amplification. The PCR products of positive clones with homozygous deletion were validated by agarose gel and Sanger sequencing. The sequences of guide RNA and PCR primers are listed in SI Appendix, Table S1.

Lentiviral OE of snoRNAs and tRNAs.

The sequence of human snoRNAs and tRNAs was, respectively, constructed into pLV-EF1a-IRES-Puro (Addgene, #85132) using BamHI and EcoRI cloning sites. Lentivirus was packaged by cotransfecting the above constructs with pVSVG and psPAX2 plasmids into HEK293T cells. The medium was changed 24 h posttransfection. The lentivirus supernatants were collected at 48 h and 72 h and clarified with a 0.45 μM filter. Lentivirus was either used directly for experiments or stored at −80 °C.

mES Cell Differentiation.

The CMD procedure is based on a previous study with adjustment for the particular purpose of time course observation of the CM beating activity (66). Briefly, mES cells were plated at a density of 2 × 10⁵ cells/mL in ultra-low attachment plates in CMD [DMEM-High Glucose (CORNING) supplemented with 15% Fetal Bovine Serum (Gibco), 1% Pen Strep (Gibco), 1× GlutaMax (Gibco), and 1 mM Ascorbic Acid (Sigma-Aldrich, A4544)] to induce EB formation. On day 3, the medium was replaced with fresh CMD medium; on day 6, EBs were resuspended in fresh CMD medium and replated on 0.2% gelatin-coated plates. During day 9 to day 24, the number of beating patches of cells was quantified in triplicate for each cell line, and the CMD medium was changed every 3 d.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (XLSX)

Dataset S06 (TXT)

Data, Materials, and Software Availability

The raw and processed total RNA seq, small RNA seq, optimized dRMS, ribo-seq, and PARIS2 data were deposited to Gene Expression Omnibus (GEO) with accession number GSE234689 (67). Code is available at https://github.com/zhipenglu/snoRNA (68) and https://github.com/minjiezhang-usc/snoRNAs_discovery (69).