Recent developments, opportunities, and challenges in the study of mRNA pseudouridylation

CHARTING THE PSEUDOURIDINE LANDSCAPE IN mRNA—PROGRESS AND ONGOING CHALLENGES

The study of mRNA pseudouridylation was launched by the development of technically demanding sequencing methods that exploit the reactivity of pseudouridine with carbodiimides, specifically N-cyclohexyl-N′-(2-morpholinoethyl)carbodiimide methyl-p-toluenesulfonate (CMC). Stable Ψ-CMC adducts block the processivity of reverse transcriptase enzymes to create truncated cDNAs whose 3′ boundaries can be mapped to identify Ψ locations with single-nucleotide resolution (Carlile et al. 2014; Lovejoy et al. 2014; Schwartz et al. 2014). An advantage of this first-generation of pseudouridine sequencing techniques was that they produced relatively unbiased maps. This was in contrast to m6A mapping with antibodies, which were found to be biased to detect the modification in specific RNA sequence contexts resulting in false positives (Schwartz et al. 2013), but also in false negatives (Garcia-Campos et al. 2019; Liu et al. 2023). However, m6A mapping using antibodies to pull down modified RNA fragments was relatively straightforward and the method was rapidly adopted by many laboratories. Technical challenges of CMC-based approaches to map Ψ include large RNA input requirements (due to harsh treatments that degrade RNA), as well as nonstandard methods of library construction. These difficulties have likely slowed progress in the field by limiting the number of laboratories that undertake pseudouridine profiling.

Beyond the inherent difficulty of the CMC-based Ψ mapping technique, the main limitations were twofold. First, these maps were incomplete because only highly expressed mRNAs could be monitored for pseudouridine with reasonable sequencing depth. This problem was severe for profiling large transcriptomes such as nascent pre-mRNA from human cells. The study that discovered pseudouridine installation in pre-mRNA was limited to inspecting ∼1% of uridines in very highly expressed transcripts (Martinez et al. 2022). Even in the smallest mRNA transcriptome profiled for pseudouridine to date (Escherichia coli), sequencing depth limited the detection of pseudouridine sites in mRNA. The authors estimated that they surveyed ∼25% of mRNA uridine positions in the growth condition profiled (Schaening-Burgos et al. 2023).

Second, these CMC-based methods did not reveal the occupancy of the pseudouridine modification, that is, what fraction of mRNA molecules contained pseudouridine at the mapped site. Two newer approaches are showing promise to address the problem of determining pseudouridine occupancy in high-throughput: a bisulfite-based chemical biology approach using Illumina sequencing, and nanopore direct RNA sequencing. Khoddami et al. (2019) discovered that sequencing bisulfite-treated RNA resulted in deletion signatures at Ψ positions. Subsequent improvements to this approach modified the buffer conditions (pH) to prevent deamination of cytosine (Fleming et al. 2019), which improved the accuracy of read mapping (Dai et al. 2023; Zhang et al. 2023). Although the relationship between the percentage of mRNA molecules containing a Ψ and the percentage of reads with a deletion was not one-to-one, the inclusion of pools of synthetic RNAs of known percent Ψ in all possible flanking sequence contexts allowed the generation of standard curves to estimate occupancy. Another quantitative approach to pseudouridine sequencing that exploits a distinct chemistry to introduce U-to-C mutations at Ψ was reported while this review was in preparation (Xu et al. 2024). Significantly, each of these quantitative profiles identified numerous high-occupancy mRNA sites with >30% of transcripts modified with pseudouridine, as well as many more sites with low levels of modification. Intriguingly, mouse tissues showed many more mRNA pseudouridines compared to cultured cells (Dai et al. 2023).

Nanopore direct RNA sequencing of pseudouridine has the potential to provide direct, single-molecule evidence for the presence of pseudouridine and enable quantification of occupancy (the fraction of sequenced mRNA molecules that are modified at a given site) (Begik et al. 2021; Huang et al. 2021; Piechotta et al. 2022). Data analysis approaches vary substantially. Imperfect compromises between specificity (avoidance of false positives) and sensitivity (avoidance of false negatives) are evident in maps of pseudouridine by nanopore that fail to detect some high-occupancy pseudouridine sites in rRNA where complete pseudouridine maps have been independently obtained by mass spectrometry. Some problems may be due to training data from in vitro transcribed RNA with 100% pseudouridine, which is very different from the sparse distribution of pseudouridine in cellular mRNA. A promising approach is the use of unmodified cellular transcriptomes prepared by in vitro transcription from amplified cDNA templates as negative controls to compare to cellular (modified) mRNA to identify candidate-modified uridines by comparison (Tavakoli et al. 2023). A current challenge of nanopore mRNA analysis without accompanying genetic manipulation of PUS is unambiguously assigning a detected modified uridine as Ψ and not another modification of U, such as dihydrouridine, which is also present in eukaryotic mRNA (Draycott et al. 2022; Finet et al. 2022).

Both nanopore and bisulfite-based approaches are limited in their ability to detect pseudouridine in low-abundance transcripts. This is a major limitation now that we know pseudouridine is installed in introns in unspliced pre-mRNA (Martinez et al. 2022). Pre-mRNA targets could be relevant to the biology of several pseudouridine synthases that are implicated in disease. For example, various pseudouridine synthases are elevated in tumors compared to healthy tissue, and patients whose cancers express higher levels of some PUS have worse outcomes. There is no basis for assuming that the disease-relevant target RNAs will be particularly abundant ones. Thus, the field needs methods to produce complete maps of pseudouridine.

Enrichment-based profiling of pseudouridine is underdeveloped but would address this challenge for the field. The first reported method to specifically enrich for pseudouridine-modified RNA was a chemical biology approach, CeU-seq (Li et al. 2015). However, the reagents required are not commercially available, and this approach has not been duplicated in other laboratories.

The m6A field has successfully combined antibody-based enrichment of modified fragments (increasing the ability to detect modifications in lower abundance mRNAs) with steps to give single-nucleotide resolution maps of the modified site within the enriched fragment (Linder et al. 2015; Roberts et al. 2021). Antibodies against pseudouridine are commercially available. However, the characterization of these antibodies remains limited. Specificity for pseudouridylated RNA fragments is critical to establish conclusively. Furthermore, it is unclear whether antibodies are equally sensitive to detect pseudouridine in the varied RNA sequence and structural contexts in which this modification occurs in mRNA.

Substrate trapping methods are another promising approach to enrich for mRNA targets of specific RNA-modifying enzymes. This class of method relies on feeding cells nucleoside analogs that randomly incorporate into mRNA and become covalently linked to specific enzymes based on their chemical activity. The chemically cross-linked target RNAs are identified by immunoprecipitation of the enzyme followed by sequencing of the linked RNA (Khoddami and Cairns 2013; Dai et al. 2021). This strategy could efficiently identify all the targets of a given PUS, which would be useful in cases where PUS is linked to disease (e.g., overexpressed in cancers with worse prognosis).

A robust two-tiered pseudouridine mapping approach would be valuable to obtain complete maps with paired information about modification occupancy. The first step would use an enrichment strategy (either chemical biology based, with commercially available reagents, or antibody based) to capture all mRNA pseudouridine sites. Then a second, targeted approach could be applied to obtain quantitative information on sites of interest. Possible approaches that would advance the field include pre-enrichment of target transcripts (based on an initial comprehensive map of pseudouridine using an enrichment step) before nanopore sequencing. Alternatively, improvements in medium throughput methods for targeted resequencing of bisulfite- or bromoacrylamide-treated RNA would allow quantification of pseudouridine occupancy.

Progress toward complete and quantitative maps of mRNA pseudouridines will be helped by embracing incremental improvements. Incentivizing authors to claim “victory” prematurely after every technical advance, thereby overstating the capabilities of the newest method, discourages the further innovation that is necessary to advance the field. The current trajectory of technical advances in pseudouridine profiling is cause for optimism. Progress is likely to continue for years. The m6A research community is much larger than the pseudouridine field currently, and improved methods for m6A mapping have been reported in the last year (Liu et al. 2023).

LARGE FAMILIES OF PSEUDOURIDINE SYNTHASE ENZYMES WITH KNOWN OR LIKELY ROLES IN mRNA PSEUDOURIDYLATION

Unlike mRNA m6A, which is predominantly installed by a single, conserved RNA methyltransferase complex (METTL3/METTL14) that mainly modifies mRNA, pseudouridine is installed in mRNA by a large family of enzymes with 13 members in humans, nine in budding yeast, 20 in Arabidopsis, and 11 in E. coli. This large repertoire of known and potential mRNA-modifying pseudouridine synthases could support intricate pseudouridine regulons, given the tissue- and cell-specificity of PUS expression (Borchardt et al. 2020). However, this diversity also complicates pseudouridine profiling studies because multiple PUS must be tested to validate putative pseudouridine sites.

Assigning biological function to mRNA pseudouridylation is also complicated by the lack of “dedicated” mRNA-modifying enzymes in this family. Most characterized human pseudouridine synthases that have been demonstrated to modify mRNA also target multiple classes of RNA including tRNA, snRNA, snoRNA, or rRNA. Thus, long-term genetic depletion of these PUS is either known or reasonably expected, to perturb the function of noncoding RNAs. Given the very long half-lives of these noncoding RNA targets, acute inactivation of PUS has the potential to distinguish direct effects of mRNA pseudouridylation on mRNA metabolism from myriad indirect consequences. In this regard, the development of fast-acting chemical inhibitors of specific PUS would be transformative.

To tease out the potential significance of specific mRNA pseudouridines, it will be very useful to deploy orthogonal means of installing site-specific Ψ in cells where a PUS has been depleted/inactivated. Reports of programmed mRNA pseudouridylation suggest a reasonable approach (Adachi et al. 2023; Song et al. 2023). In this way, the cellular effects of a specific mRNA Ψ could be determined.

VARIED AND INCOMPLETELY UNDERSTOOD MECHANISMS FOR SITE-SPECIFIC mRNA PSEUDOURIDYLATION

Modification of mRNA by multiple pseudouridine synthases potentially allows diverse RNA targets to be modified. This contrasts to m6A, which is enriched in one specific sequence context, DRACH, due to the intrinsic specificity of the METTL3/METTL14 methyltransferase. The target preferences for site-specific mRNA pseudouridylation by the majority of mRNA-modifying PUS are incompletely described at best. The most exhaustively characterized are Pus1 from budding yeast and human TRUB1. The RNA target preferences of these PUS were determined using massively parallel assays to dissect the RNA sequence and secondary structure features that are necessary for site-specific pseudouridylation by these enzymes. Yeast Pus1 was found to recognize uridines in mRNA based on a structural motif—uridines on the 5′ side of a bulged stem–loop structure were recognized—with minimal sequence requirements: a purine at −1 immediately 5′ to the target uridine, and not a G at −2 (HRU) (Carlile et al. 2019). For human TRUB1, a specific sequence, GUUCNA, displayed in the context of a loop, was recognized (Safra et al. 2017).

Even for these best-characterized PUS, questions remain. For both enzymes, mRNA targets were validated that deviated from the characteristic RNA secondary structures, indicating additional modes of target recognition. In addition, only a minor fraction of potential targets of yeast Pus1 or human TRUB1 (uridines located in the preferred RNA sequence and structural context) has been observed to be pseudouridylated in vivo. This suggests pervasive negative regulation of mRNA pseudouridylation. Undermodification of potential targets is a similarly prominent feature of the m6A landscape. The vast majority of DRACH motifs are not methylated in cells and m6A modifications cluster near stop codons. Recent work has uncovered the basis for this profile, showing that methylation of most DRACH motifs is sterically occluded by the exon junction complex (He et al. 2023; Uzonyi et al. 2023).

The mRNA targets and basis for target selection remain almost completely uncharacterized for other human PUS enzymes. Of particular interest are differences among three sets of PUS paralogs that are widely expressed in human cells: PUS1/PUSL1, PUS7/PUS7L, and RPUSD1/2/(3)/4. (RPUSD3 lacks critical catalytic amino acids [Borchardt et al. 2020].) PUS1, PUS7, and RPUSD4 have been reported to have extensive mRNA targets, while little is known about the RNA targets of PUSL1, PUS7L, and RPUSD1 and 2. The expansion of PUS enzyme families suggests the potential for substrate diversification.

COTRANSCRIPTIONAL RECRUITMENT OF PUS TO NASCENT PRE-mRNA

Martinez et al. (2022) demonstrated that nascent, chromatin-associated pre-mRNA contains pseudouridines, both in introns (that are removed) and exons that retain their modification in mature mRNA. The presence of pseudouridines in unspliced pre-mRNA, which is short-lived, may indicate the existence of mechanisms to promote efficient PUS recruitment to target sites before splicing is completed. The basis for PUS recruitment to pre-RNA has not been established. Most human PUS enzymes localize partially or completely to the nucleus in at least some cell lines, so they are in the right compartment to have access to pre-mRNA. PUS7, which was found to modify pre-mRNA, was identified to copurify with protein components of active chromatin (Ji et al. 2015). The basis for this association with chromatin is not known.

On the other hand, the observation that recombinant purified human PUS can site-specifically modify their (pre-)mRNA targets in vitro (without additional factors) suggests that the dominant mode of regulating mRNA pseudouridylation is negative—factors prevent pseudouridylation of potential target sites. The nascent pre-mRNA transcriptome is large, and there are 13 PUS that together are capable of recognizing a wide diversity of RNA sequences and structures. And yet the landscape of pre-mRNA pseudouridines is sparse. These observations together suggest that many potential pseudouridine sites are not modified in human cells, similar to m6A, where a tiny fraction of DRACH motifs are methylated. If this is true, how is mRNA pseudouridylation restricted to a subset of potential modification sites? Answering this question awaits a better understanding of the mRNA sequence and structural features required for site-specific modification by PUS.

VAST POTENTIAL FOR PSEUDOURIDINE TO AFFECT INTERACTIONS OF (PRE-)mRNA WITH RNA-BINDING PROTEINS, BUT PHYSIOLOGICAL EXAMPLES ARE LACKING

Pseudouridine affects chemical properties of the nucleobase that are known, or likely, to affect interactions with RNA-binding proteins (RBPs). Several examples demonstrate an altered affinity for RNA upon incorporation of a single pseudouridine including the RPBs PUM (Vaidyanathan et al. 2017), MBNL (Delorimier et al. 2017), and U2AF (Chen et al. 2010). These RBPs differ in their RNA-binding domains and modes of RNA recognition. The fact that Ψ affects each of them indicates the potential for this mRNA modification to impact many, diverse RBPs.

Consistent with this possibility, pre-mRNA pseudouridines were found within the binding sites of 103 human RBPs based on a comparison of pseudouridine locations and RBP binding sites determined by eCLIP in the same human cell line (Martinez et al. 2022). These mapped overlaps include U2AF, which was previously shown to be sensitive to the presence of pseudouridine in some cases. However, the effect of Ψ on U2AF binding was previously reported to change depending on the location of the Ψ within a polypyrimidine tract (Chen et al. 2010), and it was not determined whether any of the endogenous U2AF binding sites with Ψ in fact altered the affinity of U2AF for these sites. Thus, the functional consequences of endogenous pseudouridylation of splicing factor binding sites remain to be determined.

The effects of intronic pre-mRNA pseudouridylation on the efficiency of splicing were investigated by testing modified and unmodified pre-mRNA splicing in nuclear extracts (Martinez et al. 2022). This was tested for just two intronic pseudouridines out of thousands mapped in pre-mRNA, but it established an important precedent—a single pseudouridine modification installed at a site that is modified endogenously in cells was sufficient to significantly change the extent of splicing. The mechanisms were not determined for these two introns.

Many pseudouridines have been mapped to human 3′ UTRs, including overlapping binding sites for proteins that are known to affect mRNA export, localization, translation, and/or decay. There are many leads to investigate! The challenge may be identifying the (few?) functionally significant mRNA Ψ sites that affect RBP binding. There could be widespread “junk” pseudouridylation, analogous to so-called “junk” protein phosphorylation—sites that are modified because they can be modified, and there is no selective pressure to avoid their modification because there is no functional consequence. Nevertheless, it is likely that many Ψ in RBP binding sites will affect the protein–RNA interaction given that three out of three RBPs tested to date showed effects of single pseudouridines on K_d ranging from threefold to 100-fold. The field needs more careful biochemistry of endogenously modified mRNA sites tested with RBPs that are known to bind in cells.

The ability of a Ψ to affect a single transcript by affecting affinity for an RBP will likely not matter if only a minor population of mRNA is modified at that site. Thus, the improvement and deployment of quantitative methods to determine the occupancy of mRNA Ψ sites will be critical to identify sites that are likely to affect a substantial fraction of mRNA via effects on RBP binding. Quantitative approaches have recently identified dozens to hundreds of mammalian mRNAs with Ψ occupancy >30% (Dai et al. 2023; Zhang et al. 2023). Intriguingly, high-occupancy Ψ sites were more frequent in primary mouse tissues compared to cultured human cell lines.

UNRESOLVED BIOLOGICAL ROLE OF EFFECTS OF mRNA PSEUDOURIDYLATION ON TRANSLATIONAL FIDELITY

The majority of mapped Ψ in mRNA are located within coding sequences. What happens when the translating ribosome encounters them? The literature supports various effects of Ψ on decoding in some but not other circumstances (Hoernes et al. 2016, 2019; Eyler et al. 2019; Kim et al. 2022). The physiological significance and prevalence of these effects in vivo remain unresolved. Here, I will briefly highlight key findings as well as pressing questions. For a review of the mechanistic effects of mRNA modifications on translation elongation, see Franco and Koutmou (2022).

Pseudouridine can affect the fidelity of translation in diverse systems. Incorporation of pseudouridine into a UUU phenylalanine codon was reported to cause frequent amino acid misincorporation when translated in vitro using a reconstituted bacterial translation system and analysis of dipeptide formation (Eyler et al. 2019). In contrast, another group using a bacterial reconstituted translation system did not detect any mistranslation of UUU Phe codons when translation products were analyzed by mass spectrometry (Hoernes et al. 2016). Subsequent work extended these bacterial in vitro experiments to UAC Tyr codons and reported that Ψ stabilized mispairing of near- and noncognate tRNAs and promoted mistranslation (Kim et al. 2022).

It was suggested that reduced nutrient conditions might be required for substantial levels of Ψ-induced amino acid misincorporation (Eyler et al. 2019). In light of this possibility for starvation to induce site-specific amino acid misincorporation depending on the presence of Ψ in mRNA, it is interesting that nutrient deprivation changed the locations and extent of mRNA pseudouridylation in yeast and human cells (Carlile et al. 2014).

Mistranslation of pseudouridine-modified codons is a curious result given that pseudouridines are found in codons of conserved amino acids in essential genes. However, the relation between findings from reconstituted translation systems and in vivo conditions is not clear. Fully Ψ-substituted mRNA translated in human cells produced some erroneous peptides that could be detected by mass spectrometry in one study, albeit at a greatly reduced level compared to the frequency of misincorporation observed in the reconstituted bacterial system (Eyler et al. 2019). Separate studies using two different Ψ-modified reporters did not detect mistranslation in human cells, whereas one study detected Ψ-dependent mistranslation in wheat germ extracts (Hoernes et al. 2019; Kim et al. 2022).

The physiological relevance and generalizability of results with artificial pseudouridylated reporter mRNAs are unclear. First, the replacement of 100% of Us with Ψ in mRNA creates translation templates that are very different from endogenously pseudouridylated mRNAs, which mostly contain single (or well-separated) pseudouridines. In this regard, it is notable that full substitution of UUU with ΨΨΨ reduced total peptide yield without producing detectable miscoded peptides (Eyler et al. 2019). Furthermore, observed frequencies of mistranslation in bacteria and yeast cells vary widely between codons and between instances of a given codon (Mordret et al. 2019). Thus, the effects of endogenous mRNA Ψ are likely to be context-dependent, if they occur at all. Given evidence for high-occupancy pseudouridine modifications in the coding sequences of hundreds of mRNAs (Dai et al. 2023; Zhang et al. 2023), determining the effects of naturally occurring mRNA pseudouridines on decoding fidelity is an important unanswered question for the field.

FIELD FRONTIER: ESTABLISH BIOLOGICAL FUNCTIONS OF INDIVIDUAL PSEUDOURIDINES IN mRNA

The preceding sections highlight the extensive molecular potential for mRNA pseudouridines to affect mRNA function. Various quantitative assays have demonstrated pseudouridine-dependent effects on pre-mRNA splicing, translation, and binding to regulatory RBPs. However, these examples of molecular consequences of single pseudouridines come from in vitro studies or nonphysiological pseudouridylated mRNAs. Currently, scant evidence is available regarding the biological function of individual, endogenous pseudouridines in mRNAs. Although genetic manipulation of mRNA-modifying pseudouridine synthases produces widespread effects on gene expression, these effects are difficult to interpret due to confounding effects of PUS activity on noncoding RNAs as well as mRNAs. Programmed mRNA pseudouridylation is a promising approach to tease out direct functional effects. Overall, there is a pressing need for deeply researched examples of individual Ψ in specific mRNAs that have biological consequences with molecular mechanisms that are understood. The broad potential for mRNA pseudouridylation to affect gene expression is exciting to contemplate given new evidence of hundreds of high-occupancy mRNA pseudouridines revealed by recent improvements in quantitative pseudouridine profiling.