Regulation of translation by site-specific ribosomal RNA methylation

Ribosomes are complex ribozymes that interpret genetic information by translating messenger RNA (mRNA) into proteins. Natural variation in ribosome composition has been documented in several organisms and can arise from several different sources. A key question is whether specific control over ribosome heterogeneity represents a mechanism by which translation can be regulated. We used RiboMeth-seq to demonstrate that differential 2′-O-methylation of ribosomal RNA (rRNA) represents a considerable source of ribosome heterogeneity in human cells, and that modification levels at distinct sites can change dynamically in response to upstream signaling pathways, such as MYC oncogene expression. Ablation of one prominent methylation resulted in altered translation of select mRNAs and corresponding changes in cellular phenotypes. Thus, differential rRNA 2′-O-methylation can give rise to ribosomes with specialized function. This suggests a broader mechanism where the specific regulation of rRNA modification patterns fine tunes translation. Dynamic changes in 2′-O-methylation of rRNA in human cells lead to ribosome heterogeneity and result in altered translation of select mRNAs, correlating with changes in cellular phenotypes.

Using unsupervised hierarchal clustering, we examined the RMS scores of 18S and 28S rRNA across the datasets (Fig. 1c). Around two-thirds of all sites are close to being fully methylated and are generally invariant, while the remainder are fractionally methylated and exhibit a higher degree of variation between cell types and conditions. The most variable methylated sites tend to be poorly conserved with those occurring at equivalent positions in the yeast ribosome ( Fig. 1c and Supplementary Table 1). Furthermore, the 2′-O-me profiles of the fibroblasts after MYC induction cluster alongside the cancer cell lines for both the 18S and 28S rRNAs (Fig.1c). Methylation levels at 18S:C174 and 28S:U2031 are among the principal determinants of variation in the 18S and 28S rRNA clusters, respectively.   From in silico predictions, SNORD45C was proposed to guide methylation at 18S:C174 (Cm174) and also at 18S:A159, using two different antisense elements 20 (Extended Data Fig. 2o). SNORD45A, however, may also guide methylation at 18S:A159, as well as at 18S:U172. The SNORD45 family members are processed from introns of their host gene RABGGTB, which interestingly has been implicated as a MYC transcriptional target 21 (Fig. 1d). Notably, expression of both SNORD45C and SNORD45A is elevated along with the host gene, following MYC induction in fibroblasts (Fig. 1e). So, in this case there appears to be a direct link between induction of SNORD45C by MYC and increased Cm174 levels. We therefore selected Cm174 as the primary candidate for a proof-of-principle study. These results support the idea that discrete cellular pathways can influence rRNA methylation signatures.

Differential 2′-O-methylation of rRNA influences translation.
We selected the cervical carcinoma-derived cell line HeLaS3 to perform characterization of a role for Cm174 in ribosome function, since we required a system where the manifold pleiotropic effects of MYC expression on ribosome biogenesis, translation or transcription 18 would not confound analysis of any changes occurring following alteration of Cm174-modified ribosome levels in the cell.
Typical of many cancer cells, MYC expression is elevated in HeLaS3. The proportion of ribosomes modified at 18S:C174 is similar to that of BJ MYC at 120 hours post MYC induction (Extended Data Fig. 3a-d). Akin to the situation in fibroblasts, MYC expression also influences SNORD45C, SNORD45A and RABGGTB in HeLaS3 cells, since short interfering RNA (siRNA)-mediated knockdown of MYC reduced their expression (Extended Data Fig. 3e).
To generate a system in which we could investigate the function of the MYC-responsive rRNA 2′-O-me at 18S:C174, we abolished it by using CRISPR-Cas9 gene editing to delete SNORD45C in HeLaS3 (HeLaS3 45CKO ). Expression of SNORD45C was lost in the knockout (KO) cells, with modest reductions in RABGGTB and SNORD45A (Extended Data Fig. 3e). RMS analysis showed that Cm174 was completely absent in HeLaS3 45CKO cells, while methylation at 18S:A159, 18S:U172 and other sites remained largely unaffected (Fig. 2a,b and Extended Data Fig. 3f-h). This confirms SNORD45C as the guide for Cm174 and implies that Am159 is predominantly guided by SNORD45A.
Loss of Cm174 did not affect global translation, as determined by both [ 35 S]methionine incorporation and by examining ribosome distribution in polysome gradients ( Fig. 2c-e). Furthermore, the levels of rRNA processing intermediates were also unaltered, indicating no defects in rRNA biogenesis (Extended Data Fig. 3i).
To ascertain whether Cm174 status might influence the translation of specific mRNAs, we performed ribosome profiling of HeLaS3 and HeLaS3 45CKO cells (Extended Data Fig. 3j-q). Transcripts were assigned to one of five categories depending on their differential regulation in HeLaS3 45CKO cells: those that displayed significant change in mRNA expression (transcription); ribosome occupancy (translation); both transcription and translation together; opposite change in transcription and translation; or were unchanged. While the majority of transcripts were unaffected (8,388), abrogation of Cm174 resulted in a large number of transcripts changing at the level of translation alone (2,071), while few changes in transcription (106) were detected ( Fig. 2f and Supplementary Table 2). A total of 1,168 transcripts displayed decreased translation (TL-DN) and 903 showed increased translation (TL-UP) in HeLaS3 45CKO (Fig. 2f). This predominantly translational regulation implicates modulation of ribosome function by Cm174 as a probable mechanism. mRNAs in the TL-DN and TL-UP sets are associated with different gene ontology (GO) categories for biological processes (Fig. 2g,h). For example, cell cycle-and mitosis-related terms are significantly enriched in the TL-DN set. On the other hand, metabolism-related, oxidation-reduction and intracellular transport terms are over-represented in the TL-UP set. In contrast to the translationally regulated sets, no enriched GO categories were found associated with transcripts displaying increased mRNA expression (TC-UP), and only four significant categories were found for those with decreased mRNA expression (TC-DN) (Extended Data Fig. 3r).
Metagene read coverage analysis indicates that ribosome occupancy is globally reduced in TL-DN and increased in the TL-UP sets, respectively, in accordance with altered ribosome density in the translationally regulated transcripts (Fig. 2i). No major changes in the distribution of ribosomes over the coding sequence (CDS) regions are apparent, although we noticed a slight reduction in ribosome density over the CDS in the not significant (NS) set for HeLaS3 45CKO , along with a tendency for increased ribosome density around the translation start site, which is more pronounced for TL-UP transcripts (Fig. 2i). These observations are compatible with differences in either translation initiation or elongation rates.
Taken together, these findings indicate that the proportion of ribosomes methylated at 18S:C174 affects how efficiently different subsets of mRNAs are translated.
Features of translationally altered mRNAs. We began searching for features common among the mRNA transcripts comprising each respective regulated set that might explain their altered translation. Several studies have reported that other potential sources of ribosome specialization may influence the preference for 5′ cap-or Internal ribosome entry site (IRES)-dependent modes of translational initiation 22 IRES-mediated translation, we performed gene-set enrichment analysis (GSEA) on the translationally regulated mRNAs using a list of high-confidence IRES-regulated genes, and found no signifi-cant enrichments (Extended Data Fig. 4a). Additionally, in reporter assay experiments, no differences in the activities of two cellular IRES elements were observed between HeLaS3 45CKO Fig. 4b). Together, these results suggest that Cm174 has no major influence on IRES-dependent translation. Next, we evaluated the codon compositions of mRNAs in the TL-DN, TL-UP and NS sets. Principal component analysis (PCA) indicated a strong variance in the use of elongation-related codons present in the CDSs of TL-DN and TL-UP transcripts (Fig. 3a). Comparison of the individual codon frequencies in these two sets revealed a striking two-pronged distribution pattern (Fig. 3b). In all, we identified 18 codons in TL-DN and 17 in TL-UP that were significantly enriched with log 2 (fold frequency changes) >1 (Fig. 3b,c and Supplementary Table 3). Codons enriched in TL-DN tend to be AU-rich, whereas those in TL-UP are more GC-rich, and this is especially apparent at the third nucleotide position (AT3 or GC3) (Fig. 3d).

cells (Extended Data
These patterns apply only to translationally regulated transcripts, as no such differences in codon usage are apparent between the differentially expressed mRNA sets (Extended Data Fig. 4c-e).
Despite large energetic differences between GC-and AU-rich codon-anticodon duplexes, binding and accommodation rates of aminoacyl-transfer RNAs (tRNAs) and the complex stabilities at the ribosomal A-site appear to be remarkably similar in vitro 24 . This is due to evolution of intricate mechanisms in both ribosomes and tRNAs that minimize differential thermodynamic contributions of codon-anticodon interactions to maintain translation fidelity. The complex ribosome structural rearrangements associated with conformational changes and activation of eukaryotic translation elongation factor 1 alpha 1 (EEF1A1, or EF-Tu in prokaryotes) during translation elongation constitutes one such example.
A revised energy-based interpretation of the genetic code was previously described 25 . Interestingly, codons with increased frequencies in TL-DN transcripts show a robust tendency to have thermodynamically weaker interactions with their cognate tRNAs, while those with increased frequencies in TL-UP tend to be stronger (Supplementary Table 3). Although the third codon position carries less weight in this energy-based theoretical model, recent work has linked GC3 and AT3 codons with optimal and suboptimal translation, as well as mRNA stability, respectively 26  however, we do not observe significant changes in mRNA levels associated with altered stability between the differentially translationally regulated transcripts carrying GC3/AT3 codons. In addition to the general GC/AT bias, all codons we find enriched in the TL-UP set are GC3 codons, while all in TL-DN are AT3 ( Fig. 3b and Supplementary Table 3). Moreover, there is considerable overlap between the TL-DNand TL-UP-enriched codons with corresponding anticodons previously shown to be utilized in translation of proliferation-or differentiation-associated genes, which appear to segregate according to GC/AT and also GC3/AT3 content 27,28 (Extended Data Fig. 4f).
This suggests that certain factors contributing to ribosome heterogeneity, in this case methylation at 18S:C174, can mediate differential translation of such codons by modulating the capacity of the ribosome to compensate for differences in codon-anticodon interactions, for instance.
Regulation of translation by rRNA 18S:C174 methylation. The observations above are particularly interesting when considering the location of Cm174 within the ribosome tertiary structure (Fig. 4a). 18S:C174 is situated at the base of an internal loop of helix 8 (h8), a highly flexible region of the small ribosomal subunit that forms part of the intersubunit bridge B8 (B8), together with helix 14 (h14) 29 . While this area is poorly resolved in human ribosome structures that capture dynamics of the translation elongation cycle 30 , h8 is evolutionarily well conserved (Extended Data Fig. 4g), and in bacteria is known to participate in substantial conformational changes during translation elongation 31,32 . Furthermore, in Escherichia coli, h8 is a hotspot for ribosomal ambiguity mutations (ram) that alter the helix structure and disrupt B8 (refs. 33,34 ). ram mutations in h8 result in translation errors characterized by a decreased stringency of decoding and an increased rate of GTP hydrolysis by EF-Tu, which plays a central role in translation elongation [33][34][35] . It is therefore plausible that methylation status at C174 could affect ribosome function, albeit in a less drastic way than ram mutations, potentially by regulating the conformational flexibility of h8. We speculate that this in turn may modulate h8:h14 interaction, the dissociation of which is known to induce far-reaching ribosomal structural rearrangements that culminate in domain closure, EEF1A1/ EF-Tu activation and subsequent decoding 33 .
It has also been proposed that lower translational efficiency, due to the presence of nonoptimal codons, can result in a back-queuing of ribosomes towards the translation start site 36 . We imagined that similar situations could arise due to regulation of ribosome function, resulting from differential 2′-O-methylation in this case, where altered translation of mRNAs with distinct codon compositions is a factor (Fig. 4b). With this in mind, we examined the average ribosome occupancy, which is considered a measure of ribosome dwell time or pausing 37 , in both HeLaS3 45CKO and parental cells. Transcripts comprising the TL-UP and TL-DN sets were compared (Fig. 4c). Curiously, the average ribosome occupancy rates over the first five codons appear higher in HeLaS3 45CKO for both sets. Over codons 6 to 40, TL-DN transcripts display higher average occupancy in HeLaS3 45CKO , whereas this difference is not observed for the TL-UP set. This apparent queuing of ribosomes on TL-DN transcripts in HeLaS3 45CKO is consistent with slower decoding at AU-rich codons by ribosomes lacking Cm174, and that reduction in elongation rates can result in accumulation of ribosomes towards the translation start site.
Cm174 affects cellular phenotypes. Consistent with decreased translation of mRNAs linked to cell division processes, the proliferation rate and percentage of cells in S phase is reduced in HeLaS3 45CKO (Extended Data Fig. 5a-c). To exclude the possibility that clonal artifacts might contribute to this growth phenotype, we reinstated SNORD45C expression in the knockout background (HeLaS3-GFP 45CKO+45C ). This completely rescued the decrease in cell proliferation, and, crucially, Cm174 levels were also restored ( Fig. 5a-c and Extended Data Fig. 5d-f). Further analysis suggests the growth defect is due to SNORD45C and concomitant Cm174 loss, rather than any reduction in RABGGTB or SNORD45A expression, since these were not restored to wild-type (WT) levels in HeLaS3-GFP 45CKO+45C cells (Extended Data Fig. 5g).
To assess if the increased translation of metabolic-related genes gave rise to any phenotypes, we analyzed respiration and key parameters of mitochondrial function. Cells lacking Cm174 displayed higher basal and ATP-coupled respiration, suggestive of altered metabolism ( Fig. 5d and Extended Data Fig. 5h-j). Reintroduction of SNORD45C fully reversed the altered metabolic phenotype (Fig. 5d).
We then set out to ascertain whether the observed translational regulation resulted in changes at the protein level, which might explain these cellular phenotypes. Quantitative tandem-mass-tag MS was used to analyze the proteomes of HeLaS3 and HeLaS3 45CKO cells (Fig. 6a, Extended Data Fig. 6a and Supplementary Table 4). Peptides derived from 7,893 distinct proteins were detected in both conditions and used for protein abundance (PA) measurements. A total of 936 proteins were identified with significantly reduced abundances (PA-DN) in HeLaS3 45CKO , and 631 with increased (PA-UP) abundances (Fig. 6a). Supporting the ribosome profiling data, GO analysis of the PA-DN and PA-UP sets revealed striking similarities in the over-represented terms with those of the corresponding translationally regulated mRNA sets. Proteins in the PA-DN set were enriched for functions related to cell cycle and cell division, while those in PA-UP were associated predominantly with various metabolic processes and cellular transport (Fig. 6b,c).
We then compared the protein abundances measured in both wild-type and HeLaS3 45CKO cells to ribosome occupancy or mRNA expression for equivalent protein-coding genes. The patterns observed were highly similar for the two cell lines and both correlations were statistically significant, although that between protein and ribosome occupancy was stronger (Fig. 6d,e and Extended Data   Fig. 6b,c). This affirms that, globally, the ribosome footprint measurements are strongly predictive of protein level, and are a more accurate indicator than mRNA levels alone.
Examining those hits identified by both proteomic and ribosome profiling analyses, we found 332 genes in common that were altered significantly in both datasets. While we were initially surprised not   . Points represent mean RPF count (CPM) per transcript (n = 3 libraries from independent cultures, from ribosome profiling) and corresponding mean protein abundance for the same gene. Color scale indicates plotting density. n denotes number of genes/proteins. Spearman's ρ and associated P value shown. e, Correlation between mRNA translation and protein abundance for those genes with significant changes in both ribosome profiling and MS analysis of HeLaS3 45CKO (45CKO) and HeLaS3 (WT). Points represent log 2 FC translation and log 2 FC protein abundance ratio. n denotes number of genes. The number of genes displaying concomitant log 2 FC changes are indicated by PA-DN and TL-DN (down in both protein abundance and translation) and PA-UP and TL-UP (up in both protein abundance and translation). Spearman's ρ and associated P value shown. f, Diagram illustrating codon usage reporter assay. The mRNAs produced are depicted. Shown below are the relative amounts of the proteins produced in two possible situations: normal and less efficient translation through the AT-or GC-rich codon insert. g, High-content imaging assay of codon reporter cell lines (replicate 1), containing either AT-or GC-rich codon inserts. mTagBFP2:mCherry ratio after 24 h expression of the reporter cassette. Box plots: centerline is the median; hinges are first and third quartiles; whiskers are 1.5× interquartile range; outlier data points are omitted. Bonferroni P adj values from Wilcoxon tests, two-tailed, are given. Number of double-positive mTagBFP2/mCherry cells assayed, from left to right: n = 10,321, 6,707, 3,518, 16,028, 8,262, 5,659. to have detected an even larger overlap, even considering the differences between the two methodologies, it is plausible that some translational changes could be buffered by alterations in protein stability. Also, secondary effects may be detected by proteomics. The vast majority of the overlapping genes (281) underwent concomitant changes, resulting in a robust correlation (Fig. 6f). We reasoned that these high-confidence overlap sets of up-(PA-UP and TL-UP) and downregulated (PA-DN and TL-DN) genes, where Cm174-dependent changes in translation were also manifest at the protein level, should be able to explain the phenotypic differences between HeLaS3 and HeLaS3 45CKO cells. We therefore again performed GO analysis and found that the enriched categories were highly similar to both the previous individual analyses of translational regulation and of protein abundance, which is consistent with reduced cell proliferation and altered metabolism in HeLaS3 45CKO (Extended Data Fig. 6d,e). Furthermore, we examined the codon usage in the PA-UP and TL-UP and PA-DN and TL-DN sets, and once more detected clear codon usage bias associated with GC content in particular enriched codons (Extended Data Fig. 6e,f). This is highly similar to the analysis performed on TL-UP and TL-DN sets that we identified by ribosome profiling. Interestingly, for those genes where an opposite correlation was observed in the PA-UP and TL-UP and PA-DN and TL-DN comparison, the codon -usage bias was not apparent (Extended Data Fig. 6f).
To experimentally assess whether translation of AT-and GC-rich codons is specifically altered by the presence of Cm174 in ribosomes, we adapted a reporter assay approach previously developed to study ribosome stalling 38 . Instead of sequences promoting stalling, inserts composed of codons that were found to be significantly enriched in transcripts comprising either the TL-UP or TL-DN sets were placed between two fluorescent reporter genes (mTagBFP2, followed by mCherry) under the control of a doxycycline-inducible promoter (Fig. 6f). Once again, we utilized the HeLaS3-GFP WT , HeLaS3-GFP 45CKO Table 7). In this way it was possible gauge the influence of Cm174 on the ability of the ribosomes to translate through the AT-and GC-rich codon sequence inserts. The ratio of mTagBFP2:mCherry fluorescence intensity was reduced in HeLaS3-GFP 45CKO-GC-UP cells compared to HeLaS3-GFP WT-GC-UP , which is consistent with more efficient translation through the GC-rich codon insert in cells where ribosomes lack Cm174. In HeLaS3-GFP 45CKO+45C-GC-UP , where Cm174 is reinstated, the ratio is restored to a similar level to that in HeLaS3-GFP WT-GC-UP cells. Where we would predict less efficient translation through the AT-rich codon insert in HeLaS3-GFP 45CKO-AT-DN , the mTagBFP2: mCherry ratio was increased, indicating that ribosomes without Cm174 translate this less well than in HeLaS3-GFP WT-AT-DN cells. Again, this effect was rescued in HeLaS3-GFP 45CKO+45C-AT-DN cells. In addition, the largest differences were observed between HeLaS3-GFP 45CKO-GC-UP and HeLaS3-GFP 45CKO-AT-DN cells, reinforcing the idea that transcripts enriched in AT-and GC-rich codons are differentially translated in the absence of Cm174.
In combination, these results show that absence of ribosomes methylated at 18S:C174 results in considerable phenotypic changes. Furthermore, they imply that these changes arise due to the impact of differential translation on the levels of specific proteins and that the presence of Cm174 on the ribosome can influence how efficiently mRNAs, enriched in AT-or GC-rich codons are translated.  context. elevated MYC expression leads to increased ribosome biogenesis and induction of distinct specialized ribosome types, including a higher proportion of Cm174-modified ribosomes (due to increased expression of SNORD45C) and potentially differential production of other types. The transcriptome is also altered. Combined, the changes in the cellular ribosome pool along with differential mRNA expression, lead to differences in the proteome and cell phenotype.

Discussion
Several important conclusions can be drawn from our study, and these may conceptually enhance our understanding of how ribosome function can be regulated. Ribosome 2′-O-me sites can be generally grouped into two distinct categories, highly and fractionally methylated positions. There is little variation in 2′-O-me levels between different cell lines and conditions tested with regard to the highly methylated sites, suggesting that these are important for ribosome biogenesis and optimal function. In contrast, much greater variability is seen in the fractional set, indicating that methylation levels at these sites may be less constrained, making them better suited for regulatory functions. The existence of fractionally methylated positions demonstrates the presence of distinct pools of differentially 2′-O-methylated ribosomes within specific cell populations, and reinforces the idea that ribosome heterogeneity exists at the level of RNA modifications. In addition, the observation that methylation profiles can differ substantially between cell types suggests a role for 2′-O-me in regulating translation to meet the specific requirements of each cell type. Interestingly, the sites found to exhibit the greatest degree of variation in human cell lines are not conserved in yeast, although they are present in murine ribosomes. This raises the possibility that regulation of ribosome function via differential 2′-O-me has manifested more recently in evolutionary terms. However, there are also 15 2′-O-methylated nucleotides present in yeast rRNA that have no obvious counterparts in human or mouse cells. Overall, the number of rRNA modifications has steadily increased during evolution, together with increased ribosome size and complexity 3 . While previous work in yeast demonstrated that depletion of clusters of ribosomal modifications had various effects on growth, ribosome biogenesis and translation efficiency, it was thought that individual 2′-O-me were of little consequence in terms of their impact on translation 11,39 . Other studies, however, suggest that disruption of single sites could potentially affect ribosome function, for example, knockouts of several individual snoRNAs displayed phenotypes with respect to proliferation and drug sensitivity in yeast 40 .
Excitingly, we show that discrete rRNA 2′-O-me sites can actually respond dynamically to distinct upstream signaling pathways. Furthermore, as demonstrated in the case of Cm174, site-dependent rRNA 2′-O-me levels can influence translation of mRNAs associated with functionally distinct categories. Importantly, this resulted in altered cellular phenotypes.
The study of SNORD45C knockout cells, however, represents an artificial situation. Importantly, however, it facilitates dissection of Cm174 function from the pleiotropic effects of MYC expression (Fig. 7a), which results in increased ribosome biogenesis together with regulation of various translation factors, as well as considerable transcriptional changes 18 . In a biological setting, we expect that SNORD45C induction by MYC would occur in concert with an increase in ribosome number and other changes promoting or regulating global translation (Fig. 7b). In this sense, the production of specialized ribosomes can act concomitantly with changes in ribosome abundance, which alone may influence translation of distinct mRNAs 41 . As such, an additional newly synthesized ribosome pool, harboring Cm174, would act to more efficiently translate proliferation-related mRNAs and reinforce a MYC-activated program. Yet, the other pools of ribosomes lacking Cm174 would also still be present for some time, and efficiently translate other mRNA classes involved in maintenance of different cellular processes.
Our findings supplement the accumulating evidence that distinct variations resulting in heterogenous ribosomes exist 6 . These include sequence variability in rRNA alleles and, particularly, expansion segments 5 , differential association and modification of ribosomal proteins 2,42 , as well as other types of rRNA post-transcriptional modification 4,43 . Despite the abundant evidence of ribosome heterogeneity, so far only a few studies demonstrate that certain transcripts can be differentially translated due to the specific changes in ribosome composition 22,44 .
It is intriguing to contemplate a role for rRNA modifications, or indeed other means of ribosome specialization, in regulating translation of specific mRNAs on the basis of their codon compositions. Indeed, there may be a role for Cm174 in fine-tuning ribosome structural rearrangements related to balancing energetic differences, and those intrinsic to GC3/AT3 codons, in tRNA-mRNA interactions during translation elongation. Further in-depth studies would be required to fully substantiate this, however.
The dynamics of the MYC-responsive 2′-O-me sites are not rapid, with changes occurring over time scales in line with turnover of the cellular ribosome pool 45 . Thus, this type of regulation may be more suited to tuning translation in long-term cellular programs, such as during cell differentiation. Also, the possibility exists that in diseases such as cancer, mechanisms generating differentially 2′-O-methylated ribosomes are positively selected, to produce ribosome populations favoring translation of disease-promoting mRNAs. It will be of great interest to determine the exact mechanisms by which different regulatory 2′O-me sites impact specific aspects of ribosome structure and function to elicit effects on translation. Likewise, elucidating the distinct consequences that the diverse signals converging on the ribosome have, and how they differently influence modification of its function, will prove exciting. Overall, our findings promote the idea that the core ribosome itself is a direct translational regulator, and they point to 2′-O-methylation of rRNA being a potentially powerful means to tune ribosome function and impact translation.

online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41594-021-00669-4.

Generation of SNORD45C knockout cells.
HeLaS3 45CKO clones were generated using a CRISPR-Cas9 gene-editing approach. To remove the SNORD45C, sequences for sgRNAs were cloned into pX335 and pX458 plasmids 47 . Each plasmid was co-transfected into HeLaS3 cells using Lipofectamine 3000 (ThermoFisher). Single cells positive for GFP and crimson fluorescence were sorted into 96-well plates by FACS. After 2 weeks, clonal colonies were screened for successful deletion using PCR, verified by DNA sequencing and expanded.

Generation of SNORD45C overexpression and control cells.
SNORD45C was cloned into intron1 from host gene RPL23 replacing the endogenous snoRNA. The RPL23 intron containing SNORD45C was then inserted into a two-exon EGFP sequence derived from pGINT (provided by C. Bellodi), which was used to replace the EGFP from the AAVS1-targeting vector pAAV-PuroCAG-EGFP, obtained from the Vallier laboratory 48 . To generate the overexpression HeLaS3-GFP 45CKO+45C cells, the artificial EGFP-SNORD45C host gene was inserted into the AAVS1 locus of HeLaS3 45CKO using zinc-finger nucleases (according to the protocol from Bertero et al.) 48 . To generate the HeLaS3-GFP 45CKO and HeLaS3-GFP WT control cells, the two-exon EGFP without SNORD45C was similarly inserted in HeLaS3 45CKO and HeLaS3, respectively. The rescue and control cells were FACS-sorted for EGFP expression and sequenced to ensure correct insertion. SNORD45C expression was verified by reverse transcription quantitative PCR (RT-qPCR).

RNA isolation.
Large-scale and small-scale total RNA preparations were performed using QIAZOL (Qiagen) or the Direct-zol RNA MiniPrep kit (Zymo Research), respectively, according to the manufacturers' instructions.

RiboMeth-seq.
RiboMeth-seq library construction and sequencing were performed as previously described 3,14 . Triplicate libraries were produced for each cell line or condition analyzed, and grown to ∼70-80% confluence before collection. A portion of 4-5 µg of total RNA or purified 18S and 28S subunit rRNA (combined at 1:2.6 weight ratios to give an approximate 1:1 molar ratio) were used for input. Note that we have observed no major differences in RMS scores for canonical 2′-O-me sites between libraries constructed from total RNA (as in the latest version of the RMS protocol) or purified subunit rRNA. RNA was partially degraded in alkali at denaturing temperatures. The 20-40-nucleotide fragments were purified by PAGE and linkers added using a system relying on a modified Arabidopsis tRNA ligase joining 2′,3′-cyclic phosphate and 5′-phosphate ends. The libraries were sequenced on the Ion Proton platform using Ion PI Chip Kit v.3 (Life Technologies).
RiboMeth-seq data treatment. Data were analyzed as previously reported 3 using custom Python scripts (https://github.com/lundlab/RiboMeth-seq-public). Briefly, sequencing reads were mapped to a corrected human rRNA reference sequence. To facilitate comparison with other studies, we have used the human rRNA sequence numbering according to snoRNABase 49 throughout this study. An alignment table of these rRNA sequences is provided in ref. 3 . The RMS score represents the fraction of molecules methylated at each nucleotide position, calculated by comparing the number of read-end counts at the queried position to six flanking positions on either side. Quantifications were performed on 39 and 68 methylation sites in 18S and 28S rRNAs, respectively, for which both RMS plus MS evidence exist and are reliably detected in at least one of the cell lines examined in this study. For comparison plots of RMS signatures between two cell lines or conditions, for sites exhibiting significant change, in pairwise comparisons (P < 0.05, two-tailed unpaired Welch's t-test and ≥0.2 difference in RMS score) are labeled with nucleotide position, and P values are given in bar plots for specific sites. Heat map representations were produced using the heatmap.2 function in R by unsupervised clustering. RMS data have been deposited to GEO under accession code GSE153476. Note, we have previously published two of the ten RMS datasets included in this study, the data for HelaS3 (WT) and HCT116 (WT) cells 3 are accessible from Gene Expression Omnibus (GEO) GSE76393.

Evolutionary conservation of 2′-O-methylated rRNA nucleotides.
Lists of validated 2′-O-methylated positions in 18S and 28S rRNAs, as described for human 3 , yeast 14 and mouse 50 , were collated and compared. Conserved sites at equivalent nucleotide positions were determined by examination of the respective location in secondary structures of rRNA from each organism. The equivalent sites were then categorized as being present or absent in each organism. For visualization and comparison, positions were rank-ordered according to the RMS score and unsupervised clustering analysis performed on multiple human cell lines. Heat map representations were produced using the heatmap.2 function in R.
rRNA MS analysis. rRNA fragments for MS analysis were isolated and subsequently digested with RNase A or T1, as previously described 14,51 . Digested RNA fragments were subjected to MALDI-TOF MS analyses on an Autoflex Speed MS instrument (Bruker Daltonics) using reflector and positive-ion modes.
Western blotting and antibodies. Western blotting was performed as previously described 52 Table 5. For individual experiments, reactions were performed in triplicate or duplicate per target gene for each condition. For biological replicates (n = 3), the mean Ct values from individual experiments were normalized to the control condition and standardized as previously described 53 . Statistical significance was tested using two-tailed unpaired Welch's t-test.

35
S Metabolic labeling assay. For each of three biological replicates, HeLaS3 and HeLaS3 45CKO were seeded at equal density in six-well plates in triplicate. After 24 hr media were replaced with methionine-and cysteine-deficient media, supplemented with 33 µCi of EasyTag EXPRESS Protein Labeling Mix (containing both [ 35 S]l-methionine and [ 35 S]l-cysteine, Perkin Elmer) and plates were returned to the incubator for 1 h. The cells were then washed and collected by scraping into 1 ml ice-cold PBS, pelleted, snap-frozen and stored at −80 °C until further use. Lysis, and quantification of 35 S incorporation by densitometry, were performed as previously described 54 . Vinculin protein levels were used to normalize 35 S incorporation measurements in each gel lane. Statistical significance was tested using two-tailed unpaired Welch's t-test.
Northern blotting analysis of rRNA processing. Total RNA (7.5 μg) was separated on a formaldehyde denaturing 1% agarose gel and transferred to a BrightStar-Plus membrane (Ambion) using capillary blotting. followed by UV cross-linking. The probes (10 pmol each) were radiolabelled with [γ-32 P]ATP using T4 PNK (Thermo Scientific) and hybridized to the membrane overnight in hybridization buffer (4× Denhardt's solution, 6× SSC, 0.1% SDS) at T m of the probe of −10 °C. The membrane was subsequently washed four times in 3× SSC supplemented with 0.1% SDS, followed by exposure to a propidium iodide screen and scanned on a Typhoon scanner (GE Healthcare). Gel images were analyzed using Fiji software.
Ribosome profiling. Ribosome profiling was performed essentially as previously described 55 , following the protocol given in TruSeq Ribo Profile Mammalian (Illumina), with minor modifications. Three individual replicates for each of the two cell lines were collected. A single 15-cm dish corresponding to one replicate was collected at a time. For each replicate, cell media was aspirated, and cells washed with ice-cold PBS. No cycloheximide pretreatment was performed. After thorough removal of PBS, the dish was fully immersed in liquid nitrogen and placed on dry ice. For cell lysis, 1 ml of 1× Mammalian Lysis Buffer (Illumina) containing 100 μg ml −1 cycloheximide was added dropwise to the dish, which was then placed on wet ice. Cells were then scraped off to the lower portion of the dish and allowed to thaw in the lysis buffer. Lysate was homogenized by pipetting and triturated ten times through a 25-gauge needle. The lysate was then transferred to a DNA LoBind 1.5-ml microfuge tube (Eppendorf) and incubated on ice for 5 min. The lysate was cleared by centrifugation at 20,000g, 4 °C for 10 min and the supernatant transferred to a fresh microfuge tube. Aliquots were prepared for each replicate, flash-frozen in liquid nitrogen and stored at −80 °C until further use. The steps detailed in TruSeq Ribo Profile Mammalian protocol (Illumina) were followed to generate total RNA and ribosome protected fragment (RPF) RNA-seq libraries corresponding to the three individual replicates from each of the two cell lines. For RPF libraries, following nuclease digestion, monosomes were purified using Illustra MicroSpin S-400 HR Columns. Ribo-Zero Gold Kit (Illumina) was used to deplete ribosomal rRNA. The libraries prepared from total RNA or RPF for both conditions were pooled and sequenced on a NextSeq 500 System (Illumina).
Ribosome profiling data analysis. The sequencing data were demultiplexed using Illumina bcl2fastq. Quality of the sequencing files was controlled with fastqc. Adapter sequences were removed with cutadapt. Reads derived from RPF and total RNA were aligned to human rRNA and tRNA sequences with bowtie2 (v.2.2.9) and the mapped reads discarded. The remaining reads were aligned to GRCh38. p12 (Ensembl v.97) with Spliced Transcripts Alignment to a Reference (STAR) software to both transcriptome and genome coordinates. Reads mapping to Human Genome Organisation (HUGO) approved genes were used for downstream analyses. RPF read lengths were analyzed for trinucleotide periodicity using Ribotaper 56 . RPF reads with lengths between 29 and 34 nucleotides were selected and the optimal P-site offset was defined as position 12 from 5′ read ends. RPF read alignment files were filtered with samtools to retain only 29-34-nucleotide read lengths, and no read length filtering was applied to the total RNA alignment files. The transcript coordinate alignment files obtained from STAR were converted into genomic coordinate alignments with RSEM. A single canonical transcript representing each protein-coding gene was selected from the GRCh38, v.97 Ensembl annotation file (Supplementary Table 6). FeatureCounts was used to generate counts of reads mapping to exons of these transcripts for both total RNA and RPF. RPF reads with ribosome P-site positions mapping within transcript coding region sequences (CDS) were again counted using FeatureCounts and, along with the mRNA exons mapped reads, used for further measurements of differential translation and mRNA expression. Ribosomal investigation and visualization to evaluate translation (RIVET) 57 was used for translation and expression analysis of the representative transcripts (similar results were obtained for gene-level analysis). No fold change cut-offs were directly applied so that more subtle changes in translation could also be detected. Regulated transcripts were therefore nominally identified by statistical significance. Translation regulation categories were defined according to RIVET on the basis of mRNA expression and ribosome occupancy, derived from normalized total RNA read counts or RPF read counts mapping to protein-coding mRNA transcripts, respectively (Supplementary Table 2). Plots from the resulting RIVET output files were generated using the ggplot2 package in R. The RNA sequencing data have been deposited to GEO under accession code GSE153476.
GO and GSEA. All GO analyses of ribosome profiling and proteomic data were performed using WebGestalt using the over-representation test against the GO biological process database 58 . GSEA was performed using GSEA software (v.4.0.3). Genes for analysis were preranked according to log 2 FC in translation from RIVET analysis (P adj < 0.05) and compared to a custom gene set (containing 112 distinct human genes) compiled from those previously described to contain experimentally validated IRES elements, or that associated with polysomes following cap-dependent translation inhibition 59 .
Metagene analysis. For metagene analyses, bam files containing exon-mapped reads for each library were converted to normalized reads per kilobase per million (RPKM) or counts per million reads (CPM) single-nucleotide resolution coverage bigwig files, with bamCoverage from the deepTools suite 60 . WiggleTools 61 (Ensembl) and wigToBigWig (Encode, kentUtils) were then used to merge these and create mean coverage files per condition. These were input to deepTools computeMatrix, together with an annotation file containing the exon coordinates for the selected mRNA transcripts. For RPF coverage over all transcripts, a count matrix was then generated for library RPKM RPF coverage over the coding regions (CDS), scaled to size 100 nt, flanked by unscaled regions before and after the translation start (TSS) and end (TES) sites. For further analysis, the scaled coverages of transcripts comprising the different translationally regulated categories were extracted from this matrix and median values at each position plotted. For average ribosome occupancy, CPM normalization was used and offset applied using bamCoverage, so as to use only the nucleotide position representing the ribosome P-site for each read as the signal (see 'Ribosome profiling data analysis' , above). The P-site coverage files were input to computeMatrix and a count matrix generated for −30 to +330 or −330 to +30 nucleotides, relative to the CDS start or end site respectively for each transcript (unscaled). The resulting counts at each position were divided by the total RPF count in CDS for each corresponding transcript to give the average ribosome occupancy per nucleotide position in each transcript. The mean values at each equivalent nucleotide position relative to the translation start site were plotted after extreme outlier removal (>3x interquartile range), no smoothing was applied. For P-site CPM the same matrices were used, although here the counts at each position were summed at each nucleotide position. For plotting, extreme outliers (>3x interquartile range) were removed. Plots were produced using ggplot2 in R.
Cell proliferation assay by crystal violet staining. Cells were seeded in triplicate at equal density in 12-well plates. At each time point, cells were washed and then fixed with 4% paraformaldehyde. Subsequently, crystal violet staining was performed for 30 min at room temperature, followed by three washes in PBS. The staining was dissolved in 10% acetic acid and absorbance (at 600 nm) measured using a GloMaxMulti Detection plate reader (Promega).

EdU incorporation assay and flow cytometry analysis.
For each of three biological replicates, HeLaS3 and HeLaS3 45CKO were seeded in 6-cm dishes in triplicate at densities resulting in equal confluence (approximately 70%) at collection. At 48 h later, cells were pulsed with 30 μM EdU for 40 min and then collected by trypsinization and washed 3× in PBS with 1% BSA. EdU detection was performed using the Click-iT EdU Alexa Fluor 647 Flow Cytometry Assay Kit (ThermoFisher) according to the manufacturer's instructions, and incorporation was assessed by flow cytometry analysis of 20,000 cells per condition. Statistical significance of cell cycle phase (G1, S, G2/M) data was determined using two-way analysis of variance (ANOVA) with Sidak correction for multiple comparisons. For propidium iodide staining, BJ MYC cells were collected, washed, fixed in 70% ethanol, washed again and counted before incubation in FxCycle PI/RNase Staining Solution (ThermoFisher). Analysis of all FACS data was performed using the FlowJo software package (Tree Star).
Metabolic analysis of oxygen consumption rate. Cellular oxygen consumption rate (OCR) was measured using a Seahorse 96XF analyzer, utilizing the mitochondrial stress test according to the manufacturer's protocol. Cells were plated 24 h before the experiment. HeLaS3, HeLaS3-GFP WT and HeLaS3 45CKO+45C were seeded at 10,000 cells per well, while HeLaS3 45CKO and HeLaS3-GFP 45CKO were seeded at 14,000 cells per well (seeding densities were optimized to ensure equal confluence after 24 h, and that OCR measurements were within the linear detection rate). At 1 h before the experiment, the cells were washed and media exchanged into Seahorse assay medium (DMEM with 3 mg l −1 phenol red and 5.5 mM glucose, pH 7.4) and kept at 37 °C. Four measurements were performed for each condition: baseline, oligomycin A (final concentration 1 µM), FCCP (final concentration 0.2 µM) and antimycin A/rotenone (final concentration 1.0 and 0.5 µM, respectively). Concentrations of oligomycin A, FCCP and antimycin A/rotenone were optimized with dose-response experiments to ensure optimal conditions for the assay. OCR measurements were analyzed similarly to previously described 62 . Each condition was performed with 16 technical replicates in each experiment, the data are presented as the mean of four independent experiments. Statistical significance was tested using repeated measures one-way ANOVA with Turkey's correction for multiple comparisons for experiments with three sample groups, or two-tailed unpaired Welch's t-test for two sample groups.
In each experiment, firefly and Renilla luciferase activities were measured for quadruplicate transfections with the Dual-Glo Luciferase Assay System (Promega) using the GloMaxMulti Detection System (Promega). Ratios of firefly to Renilla luciferase activity were then calculated to assess 5′ cap-versus IRES-mediated translation initiation in each condition. Statistical significance was tested using two-tailed unpaired Welch's t-test.

Codon usage analysis.
Coding sequences for transcripts mapped in ribosome profiling analysis were retrieved from Ensembl (GRCh38.p12, v.97). After exclusion of the start codon the codon usage frequency and GC content were calculated using the uco and GC functions from the seqinr R package. The codon usage frequencies or GC content were then grouped according to the regulation categories defined in the analysis of the ribosome profiling data and compared between conditions. PCA was executed with the R function prcomp, with the transcripts as observations and codons as variables. To evaluate which codons contributed the most to the different codon usage profiles, the mean and median of each codon was calculated for the above-mentioned categories. The log 2 fold changes were calculated from the means of the different groups. P values for individual codons and GC content were determined by Wilcoxon test (unpaired, two-sided) and adjusted for multiple testing using the Bonferroni correction. Codons with log 2 (fold change in frequency) > 1, P adj < 0.001, were selected for further analysis. Codons enriched in each translation category were then classified as 'proliferation' , 'differentiation' or 'other' according to a previous study focused on tRNAs 28 . Codons were designated as having strong, weak or intermediate interactions with their cognate tRNAs, as defined in previous analysis 25 (Supplementary Table 3).

Proteomic MS analysis.
HeLaS3 and HeLaS3 45CKO cells were grown to ∼70% confluency. A single 10-cm dish was collected for each of four biological replicates in each case. Cells were washed, collected by scraping into PBS, pelleted by centrifugation at 100g for 4 min and the supernatant removed. Samples were flash-frozen and stored at −80 °C until further use. Lysate preparation and digestion was similar to that previously described 64 . Briefly, cell pellets were lysed using 300 µl guanidinium hydrochloride lysis buffer (6 M GdnHCl, 100 mM Tris-HCl pH 8.5, 5 mM TCEP, 10 mM CAA). Samples were sonicated on high for 5× 30 s in a Bioruptor sonication water bath (Diagenode) at 4 °C. After determining sample protein concentration with Bradford reagent (Sigma), 10 µg of each was taken forward for digestion. Samples were diluted 1:3 with 10% acetonitrile, 50 mM HEPES pH 8.5 and LysC (MS grade, Wako) was added in a 1:50 (enzyme to protein) ratio, and samples were incubated at 37 °C for 4 h. Samples were further diluted to 1:10 with 10% acetonitrile, 50 mM HEPES pH 8.5 and trypsin (MS grade, Promega) was added in a 1:100 (enzyme:protein) ratio and samples were incubated overnight at 37 °C. Enzyme activity was quenched by adding 2% trifluoroacetic acid (TFA) to a final concentration of 1%. Before TMT labeling, the peptides were desalted on in-house packed C18 StageTips 65 . For each sample, two disks of C18 material (3M Empore) were packed in a 200-µl tip, and the C18 material activated with 40 µl of 100% methanol (HPLC grade, Sigma), then 40 µl of 80% acetonitrile, 0.1% formic acid. The tips were subsequently equilibrated twice with 40 µl of 1%TFA and 3% acetonitrile, after which 10 μg of sample was loaded using centrifugation at 4,000 r.p.m. After washing the tips twice with 100 µl of 0.1% formic acid, the peptides were eluted into clean 500-µl eppendorf tubes using 40% acetonitrile, 0.1% formic acid. The eluted peptides were concentrated in an Eppendorf Speedvac (ThermoFisher) and reconstituted in 50 mM HEPES (pH 8.5) for TMT labeling 66 . Labeling was done according to the manufacturer's instructions, and, subsequently, labeled peptides were mixed 1:1:1:1:1:1:1:1:1:1:1 (11-plex), acidified with 1% TFA and acetonitrile concentration brought down to <5% using 2% TFA. Before MS analysis, the peptides were fractionated using an offline ThermoFisher Ultimate 3000 liquid chromatography system using high pH fractionation (5 mM ammonium bicarbonate, pH 10) at 5 µl min −1 flow rate. Peptides (30 µg portions) were separated over a 120-min gradient (5% to 35% acetonitrile) while collecting fractions every 120 s. The resulting 60 fractions were pooled into 44 final fractions and vacuum concentrated to dryness. Fractions were resuspended in 1% TFA, 2% acetonitrile for MS analysis. Analysis was performed on an Orbitrap Fusion mass spectrometer. For each fraction, peptides were loaded onto a 2-cm C18 trap column (ThermoFisher, catalog no. 164705) connected in-line to a 50-cm C18 reverse-phase analytical column (Thermo EasySpray ES803) using 100% buffer A (0.1% formic acid in water) at 750 bar, using the Thermo EasyLC 1200 HPLC system and the column oven operating at 45 °C. Peptides were eluted over a 60-min gradient ranging from 10 to 60% of 80% acetonitrile, 0.1% formic acid at 250 nl min −1 , and the Orbitrap Fusion instrument (ThermoFisher Scientific) was run in an SPS-MS3 top speed mode. Full MS spectra were collected at a resolution of 120,000, with an AGC target of 2 × 10 5 or maximum injection time of 50 ms and a scan range of 380-1,500 m/z. The MS 2 spectra were obtained in the ion trap operating at rapid speed, with an AGC target value of 1 × 10 4 or maximum injection time of 50 ms, a normalized CID collision energy of 35 and an intensity threshold of 5 × 10 3 . Dynamic exclusion was set to 60 s, and ions with a charge state <2, >6 or unknown were excluded. From the resulting MS 2 scan, ten precursors were selected for SPS-MS3 analysis, fragmented with a normalized HCD collision energy of 65 and ions collected for a maximum of 86 ms or AGC target of 1 × 10 5 . Resulting MS3 spectra were collected at 50,000 resolution and scan range of 100-500 for reporter ion quantification. MS performance was verified for consistency by running complex cell lysate quality control standards, and chromatography was monitored to check for reproducibility. All MS spectra were searched in Proteome Discoverer 2.4 (ThermoFisher) using the SEQUEST algorithm against the human proteome Uniprot database (containing its reversed complement and known contaminants). Spectral matches were filtered to false discovery rate (FDR) < 0.01 using the target-decoy strategy combined with linear discriminant analysis. Proteins were quantified only from peptides with an average reporter S/N threshold of 10, and co-isolation specificity of 0.75. Statistical analysis of protein abundance changes was performed using the DEqMS pipeline for TMT-labeled MS data (https://github.com/yafeng/DEqMS) 67 , with q value <0.05. The log 2 FC changes in protein abundance were determined to be significant at sca.adj.pval (Benjamini-Hochberg method adjusted DEqMS P values) <0.05 (Supplementary Table 4). The MS data have been deposited to the ProteomeXchange Consortium, via the PRIDE repository with the dataset identifier PXD019449.
Generation of codon usage reporter plasmids. The reporter cassette from plasmid pmGFP-P2A-K0-P2A-RFP 38 (Addgene, plasmid no. 105686) was amplified by PCR. The backbone of plasmid pLVX-TetOne-puro (Takara Bio) was amplified by PCR, so as to exclude the puromycin resistance gene, which was then replaced with the G418 resistance gene. The TetOne promoter cassette was amplified by PCR and all segments were assembled together with NEBuilder HiFi DNA Assembly kit (NEB). The resulting reporter plasmid pLVX-TetOne-G418-EGFP-mCherry was amplified by PCR with primer to exclude EGFP, which was replaced with mTagBFP2 to give pLVX-TetOne-G418-mTagBFP2-mCherry. Two separate 40-codon long inserts were designed comprising either eight AU-rich codons enriched in the TL-DN transcript set (AU-DN) or eight GC-rich codons (GC-UP) in the TL-UP set (Supplementary Tables 3 and 5). The order of specific codons was shuffled in each case to minimize secondary structure formation. AU-DN and GC-UP codon cassettes were ordered as gBlocks from Integrated DNA Technologies and inserted into the mTagBFP2-mCherry reporter plasmid, in frame, between the fluorescent reporter genes using the NEBuilder HiFi DNA Assembly kit, to give pLVX-TetOne-G418-AU-TL-DN and pLVX-TetOne-G418-GC-TL-UP plasmids. All assembly products were verified by sequencing. High-content imaging assays of codon reporter cell lines. For each of the six codon usage reporter cell lines, ten wells were seeded in 96-well plates. Media in nine of these wells contained doxycycline (100 ng ml −1 ), with no doxycycline in the remaining well. After 24 h, fluorescent imaging was performed on live cells with an INCell2200 high-content imager (GE Healthcare), using a ×20 objective capturing 60 fields in each well. mTagBFP fluorescence was captured using the DAPI channel, GFP using the FITC channel and mCherry using the TexasRed channel. Image analysis was performed with Incell Analyzer 1000 workstation software (GE Healthcare). Cells were segmented on the basis of the mCherry signal using the Tophat segmentation method, and the mean intensity per cell of both the BFP and mCherry was measured. mCherry segmentation was used rather than mTagBFP, due to higher signal to background levels. For each cell line, thresholds for cells positive for mTagBFP and mCherry expression were defined using intensity measurements from the wells not treated with doxycycline. Further data analysis and plotting was performed using R. Cells that were positive for both mTagBFP and mCherry were selected, and the ratio of mTagBFP:mCherry calculated from the intensity measurements for individual cells per condition. Cells with outlying ratios (>1.5× interquartile range), probably resulting from imaging artifacts, were excluded from downstream analysis. Box plots of mTagBFP:mCherry ratios per cell for each condition were generated using the ggplot2 package in R. Statistical significance was tested using two-tailed unpaired Wilcoxon test with 95% confidence interval, and the resulting P values adjusted for multiple comparisons with the Bonferroni correction method. Statistical analysis was performed using the rstatix package in R.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All RNA sequencing data from this study have been deposited in the GEO under series accession numbers GSE153476 (for RNA-seq data) and GSE76393 and BJ MYC cells, where MYC expression is induced for 120 h or not induced. Columns indicate mean RMS score for each condition of n = 3 libraries from individual cultures, points (magenta) denote each value separately. error bars represent ±s.d. P values (Welch's t-test, two-tailed) are shown above brackets. c, As (a), for 28S rRNA. d, As (B) for 2'-O-me at 28S:U2031 and 28S:U2402. e, siRNA-mediated knock-down of MYC in HelaS3 (WT) cells. expression of MYC, and canonical target genes NPM1 and FBL, as well as SNORD45C, SNORD45A and host gene RABGGTB, as determined by RT-qPCR analysis. Columns indicate mean (n = 3 independent experiments) fold-change relative to the control condition, normalized to U6 expression. error bars indicate 95% confidence intervals. P values (Welch's t-test, two-tailed) are shown for pairwise comparisons, indicated by brackets. f, RiboMeth-seq (RMS) scores representing fraction of 2'-O-methylation at each site in 18S (left) and 28S rRNAs (right) from HelaS3 45CKO cells. Nucleotide positions in respective rRNAs are indicated. RMS scores are means of n = 3 sequenced libraries from individual cell cultures. error bars represent ± s.d. g, expression of SNORD45C, SNORD45A and host gene RABGGTB in HelaS3 (WT) and HelaS3 45CKO (45KO) cells, as determined by RT-qPCR analysis. Columns indicate mean (n = 3 independent experiments) fold-change relative to the control condition, normalized to U6 expression. error bars indicate 95% confidence intervals. P values (Welch's t-test, two-tailed) are shown for pairwise comparisons, indicated by brackets. Note, due to high sequence similarity between SNORD45C and -A, a residual signal for SNORD45C is apparent in 45KO. h, As (a), for 28S rRNA in HelaS3 and HelaS3 45CKO cells. No sites exhibit significant change. i, Analysis of rRNA biogenesis pathways in HelaS3 GFP (WT), HelaS3 45CKO-GFP (KO) and HelaS3 45CKO-GFP+45C (Oe) cells, derived from parental HelaS3 and HelaS3 45CKO (please refer to section 'Cm174 affects cellular phenotypes' in the main text where they are formally introduced and described in more detail). Schematic (left) showing rRNA processing steps and location of probes 'a' and 'b'. Levels of pre-rRNA and processing intermediates as assessed by northern blot. Data shown is representative of n = 2 independent experiments. j, Percentage reads mapping to 5′ untranslated regions (5′ UTR), coding sequences (CDS) or 3′ untranslated regions (3′ UTR) of protein-coding genes, for each replicate sequencing library. Both total RNA (left) and ribosome protected fragment (right) derived reads are shown. k, Periodicity of ribosome protected fragment (RPF) reads mapped to mRNA transcripts. Metagene analysis shows normalized mean counts, at single-nucleotide resolution, representing ribosome P-site occupancy relative to start (left) and stop (right) codons from HelaS3 (WT, black) or HelaS3 45CKO (KO, red) libraries (n = 3 libraries from individual cultures). Number of transcripts analyzed after extreme outlier removal is given by 'n'. l, Correlation between reads mapped per mRNA transcript in HelaS3 (WT) and HelaS3 45CKO (KO) RPF libraries (n = 3 libraries from individual cultures). Normalized (CPM) mean counts are plotted. Color scale indicates plotting density. 'n' denotes number of transcripts analyzed. Spearman's rho and associated P value (algorithm AS 89) are shown. m, As (k), for reads mapping to mRNA from total RNA libraries. n, As (k), for correlation between reads mapping to mRNA transcripts from RPF and total RNA libraries in HelaS3 (WT) cells. o, As (m), for HelaS3 45CKO (KO) cells. p, Differential ribosome occupancy on mRNA transcripts in HelaS3 45CKO compared to HelaS3 cells. Log2 fold-change in read counts derived from analysis of RPF libraries (Log2FC RPF) and corresponding -Log10 of Benjamini-Hochberg P adj values are plotted. Transcripts changing significantly between conditions are colored (purple). 'n' denotes total number of transcripts analyzed. The number of transcripts showing reduced (DN) or increased (UP) translation is also shown. q, As (o), for mRNA transcripts displaying altered expression level (Log2FC mRNA), as measured by analysis of read counts mapped to mRNA transcripts derived from total RNA libraries. Those changing significantly between conditions are colored (cyan). r, Gene ontology analysis of mRNA transcripts displaying elevated expression in HelaS3 45CKO cells. GO categories (FDR < 0.05) for biological process are labelled. Number of genes overlapping with each biological process GO category is indicated by the color scale gradient (count). No GO terms were found to be overrepresented for transcripts with reduced expression levels (at FDR < 0.05). Fig. 6 | Ribosome regulation by cm174 alters the levels of proteins involved in distinct cellular processes and affects translation of codons with differing gc content. a, Overview of protein-level quantitation values between replicate mass spectrometry analyses. Coefficient of variation (CV) is below 20% for >96% of protein quantitation values in for both HelaS3 (WT, left) and HelaS3 45CKO (KO, right) samples (n = 4 per condition), indicating high reproducibility between replicates (n = 4 independent protein samples per condition). b, Correlation between mRNA translation and protein abundance in HelaS3 (WT) cells. Points represent mean total RNA (RNA) read count per million mapped reads (CPM) for each transcript (n = 3 libraries from independent cultures per condition, from ribosome profiling) and corresponding mean protein abundance for the same gene (n = 4 independent protein samples per condition, from MS analysis). Color scale indicates plotting density. 'n' denotes number of genes/proteins. Spearman's rho and associated P value (algorithm AS 89) are shown. c, as (b), for HelaS3 45CKO (45CKO) cells. d, Gene ontology analysis of overlap between PA-DN and TL-DN sets. Top 10 GO categories for biological process are labelled. Number of genes overlapping with each GO category is indicated by the color scale gradient (count). e, As (d), for overlap between PA-UP and TL-UP sets. f, Principal component analysis of codon usage frequency in transcripts exhibiting significant Log2FC changes in both proteomic and ribosome profiling analyses of HeLaS3 and HelaS3 45CKO