Carine Tisné

The advent of deep-learning-driven tools such as AlphaFold has revolutionized the prediction of biomolecular structures, offering unprecedented accuracy and accessibility for proteins, RNA, and their complexes. While these tools have demonstrated remarkable success in benchmarking competitions and enabled experimentalists to generate models with ease, their widespread use has also highlighted persistent challenges. These include difficulties in assessing model confidence, limitations in predicting transmembrane domains, nucleic acids, conformational diversity, and interactions with ions or ligands, as well as the tendency to misfold intrinsically disordered regions (IDRs). In this perspective, we critically evaluate the strengths and limitations of current AI-based structure prediction tools through illustrative examples with a particular emphasis on the impact of explicit ion modeling. We notably report that the explicit addition of a few potassium cations to the prediction of IDRs or G-quadruplexes can substantially shift AF3’s predicted conformation compared to “dry” predictions. On this basis, we suggest modeling sequences both “dry” and in the presence of explicit potassium cations as a simple, practical way to sample alternative conformations and to expose disordered regions that current predictors tend to overfold. We discuss the importance of reporting confidence metrics in publications to avoid overinterpretation. Furthermore, we address the unique challenges of RNA structure prediction, where data scarcity and structural complexity limit the performance of both classical and deep learning methods. Our analysis underscores the need for continued methodological advancements, integration of complementary computational tools, and expansion of high-quality experimental data sets.

Ribosome stalling compromises proteome integrity and cellular fitness, and requires efficient mechanisms to eliminate problematic mRNAs and rescue blocked ribosomes. We previously identified Rae1 as an endoribonuclease that cleaves mRNAs at the tail end of stalled ribosomes to promote mRNA degradation in Bacillus subtilis . Here, we show that Rae1 has a broader role in the quality control of mRNAs that encounter translation problems. Global 5′-end sequencing identified 31 previously unknown Rae1 cleavage sites, many of them associated with ribosome pausing at polyproline motifs in the absence of elongation factor P (EF-P). We also demonstrate that Rae1 cleaves mRNAs with ribosomes stalled on truncated transcripts lacking a stop-codon (non-stop mRNAs), provided that the transcript is sufficiently long to accommodate multiple collided ribosomes. Rae1 cleavage depends on a cytosine-poor context at the cut site, showing that both ribosome collisions and RNA sequence at the cleavage site govern Rae1’s ability to promote the removal of stalled ribosomes from mRNA. Significance: Cells have evolved mRNA quality-control mechanisms to identify and eliminate defective messenger RNAs. Ribosome stalling, which occurs when ribosomes encounter difficult-to-translate sequences, can contribute to the production of truncated proteins and compromise cellular fitness. We have identified Rae1 as an endoribonuclease that detects and cleaves mRNAs associated with collided ribosomes, allowing stalled ribosomes to be rescued and transcripts to be eliminated. We show that this mechanism operates broadly across different types of problematic mRNAs and depends on both ribosome collisions and a cytosine poor sequence at the cleavage site. These findings reveal how bacteria couple translation surveillance to mRNA degradation and provide new insights into how cells recruit RNases to promote ribosome rescue and recycling.

The human mitochondrial tRNA methyltransferase TRMT10C contains a 41-residue long positively charged α-helix in its N-terminal domain, interacting with tRNAs from the anticodon to the T-loops. Here, we investigated the flexibility of this N-terminal long α-helix using comparative molecular dynamics simulations in explicit solvent with three different force fields/water model combinations, alongside two helices of identical length but different sequences: a polyalanine helix and an AEEEKRK repeat helix. We also expanded our simulations to three additional helices belonging to the Trm10 family. Five independent 1-μs simulations were performed for each system. The human TRMT10C helix turned out to display heterogeneous dynamics, with a stable α-helical segment (residues 132–139) embedded within more flexible regions and this trend was also observed in the human Trm10 family members. This relatively rigid segment is more hydrophobic, whereas the remaining portion is enriched in charged residues and shows higher conformational variability. The rigid region corresponds to the portion interacting with the cavity formed by tRNA and TRMT10C N-terminal domain. Additional simulations using alternative force-field and water-model combinations supported the robustness of the overall flexibility ranking for the AEEEKRK repeat helix and the TRMT10C helix, but it gave force-field-dependent results for the structural dynamics of the polyalanine helix. Simulations of the N-terminal helices of other TRMT10C homologues revealed similar flexibility profiles, suggesting that the observed trends are generalizable across the family. Based on our results, we suggest that the human TRMT10C helix is tuned to combine localized rigidity to provide localized structural support that helps present the fragile mitochondrial tRNA to the maturation enzymes with overall flexibility for conformational adaptation, facilitating tRNA positioning and efficient catalysis.

The advent of deep learning-driven tools such as AlphaFold has revolutionized the prediction of biomolecular structures, offering unprecedented accuracy and accessibility for proteins, RNA, and their complexes. While these tools have demonstrated remarkable success in benchmarking competitions and enabled experimentalists to generate models with ease, their widespread use has also highlighted persistent challenges. These include difficulties in assessing model confidence, limitations in predicting transmembrane domains, nucleic acids, conformational diversity, and interactions with ions or ligands, as well as the tendency to misfold intrinsically disordered regions (IDRs). In this perspective, we critically evaluate the strengths and limitations of current AI-based structure prediction tools through illustrative examples with a particular emphasis on the impact of explicit ion modelling. We notably report how the explicit addition of a few potassium cations to the prediction of IDRs or G-quadruplexes can trigger massive conformational switches compared to "dry" predictions. On this basis, we suggest modelling sequences both "dry" and in the presence of explicit potassium cations as a simple, practical way to sample alternative conformations and to expose disordered regions that current predictors tend to over-fold. We discuss the importance of reporting confidence metrics in publications to avoid overinterpretation. Furthermore, we address the unique challenges of RNA structure prediction, where data scarcity and structural complexity limit the performance of both classical and deep learning methods. Our analysis underscores the need for continued methodological advancements, integration of complementary computational tools, and expansion of high-quality experimental datasets.

The human mitochondrial genome is transcribed into two RNAs, containing mRNAs, rRNAs and tRNAs, all dedicated to produce essential proteins of the respiratory chain. The precise excision of tRNAs by the mitochondrial endoribonucleases (mt-RNase), P and Z, releases all RNA species from the two RNA transcripts. The tRNAs then undergo 3′-CCA addition. In metazoan mitochondria, RNase P is a multi-enzyme assembly that comprises the endoribonuclease PRORP and a tRNA methyltransferase subcomplex. The requirement for this tRNA methyltransferase subcomplex for mt-RNase P cleavage activity, as well as the mechanisms of pre-tRNA 3′-cleavage and 3′-CCA addition, are still poorly understood. Here, we report cryo-EM structures that visualise four steps of mitochondrial tRNA maturation: 5′ and 3′ tRNA-end processing, methylation and 3′-CCA addition, and explain the defined sequential order of the tRNA processing steps. The methyltransferase subcomplex recognises the pre-tRNA in a distinct mode that can support tRNA-end processing and 3′-CCA addition, likely resulting from an evolutionary adaptation of mitochondrial tRNA maturation complexes to the structurally-fragile mitochondrial tRNAs. This subcomplex can also ensure a tRNA-folding quality-control checkpoint before the sequential docking of the maturation enzymes. Altogether, our study provides detailed molecular insight into RNA-transcript processing and tRNA maturation in human mitochondria.

ConspectusRNA modification is a co- or post-transcriptional process by which specific nucleotides are chemically altered by enzymes after their initial incorporation into the RNA chain, expanding the chemical and functional diversity of RNAs. Our understanding of RNA modifications has changed dramatically in recent years. In the past decade, RNA methyltransferases (MTases) have been highlighted in numerous clinical studies and disease models, modifications have been found to be dynamically regulated by demodification enzymes, and significant technological advances have been made in the fields of RNA sequencing, mass spectrometry, and structural biology. Among RNAs, transfer RNAs (tRNAs) exhibit the greatest diversity and density of post-transcriptional modifications, which allow for potential cross-talks and regulation during their incorporation. N¹-methyladenosine (m¹A) modification is found in tRNAs at positions 9, 14, 16, 22, 57, and 58, depending on the tRNA and organism.Our laboratory has used and developed a large panel of tools to decipher the different mechanisms used by m¹A tRNA MTases to recognize and methylate tRNA. We have solved the structures of TrmI from Thermus thermophilus (m¹A58), TrmK from Bacillus subtilis (m¹A22), and human TRMT10C (m¹A9). These MTases do not share the same structure or organization to recognize tRNAs, but they all modify an adenosine, forming a non-Watson-Crick (WC) interaction. For TrmK, nuclear magnetic resonance (NMR) chemical shift mapping of the binding interface between TrmK and tRNA^Ser was invaluable to build a TrmK/tRNA model, where both domains of TrmK participate in the binding of a full-length L-shaped tRNA and where the non-WC purine 13-A22 base pair positions the A22 N1-atom close to the methyl of the S-adenosyl-l-methionine (SAM) TrmK cofactor. For TRMT10C, cryoEM structures showed the MTase poised to N¹-methylate A9 or G9 in tRNA and revealed different steps of tRNA maturation, where TRMT10C acts as a tRNA binding platform for sequential docking of each maturation enzyme. This work confers a role for TRMT10C in tRNA quality control and provides a framework to understand the link between mitochondrial tRNA maturation dysfunction and diseases.Methods to directly detect the incorporation of modifications during tRNA biosynthesis are rare and do not provide easy access to the temporality of their introduction. To this end, we have introduced time-resolved NMR to monitor tRNA maturation in the cellular environment. Combined with genetic and biochemical approaches involving the synthesis of specifically modified tRNAs, our methodology revealed that some modifications are incorporated in a defined sequential order, controlled by cross-talks between modification events. In particular, a strong modification circuit, namely Ψ55 → m⁵U54 → m¹A58, controls the modification process in the T-arm of yeast elongator tRNAs. Conversely, we showed that m¹A58 is efficiently introduced on unmodified initiator tRNA_i^Met without the need of any prior modification. Two distinct pathways are therefore followed for m¹A58 incorporation in elongator and initiator tRNAs.We are undoubtedly entering an exciting period for the elucidation of the functions of RNA modifications and the intricate mechanisms by which modification enzymes identify and alter their RNA substrates. These are promising directions for the field of epitranscriptomics.

The human mitochondrial genome is transcribed into two RNAs, containing mRNAs, rRNAs and tRNAs, all dedicated to produce essential proteins of the respiratory chain. The precise excision of tRNAs by the mitochondrial endoribonucleases (mt-RNase), mt-RNase P and Z, releases all RNA species from the two RNA transcripts. The tRNAs then undergo 3’-CCA addition. In metazoan mitochondria, RNase P is a multi-enzyme assembly that comprises the endoribonuclease PRORP and a tRNA methyltransferase subcomplex. The requirement for a tRNA methyltransferase complex in mt-RNase P cleavage activity is not fully understood. Moreover, the mechanisms of pre-tRNA 3’-cleavage and CCA addition have also yet to be uncovered. Here we report cryo-EM structures that visualise these events, revealing the molecular basis for the first steps of RNA-transcript processing in human mitochondria. Altogether, the structures visualise four steps of mitochondrial tRNA maturation: 5’ and 3’ tRNA-end processing, methylation and 3’-CCA addition, and explain the defined sequential order of tRNA processing. The methyltransferase subcomplex recognises the pre-tRNA in a distinct mode that can support tRNA processing and 3’-CCA addition, likely resulting from an evolutionary adaptation of mitochondrial tRNA maturation complexes to the structurally-variable mitochondrial tRNAs. This subcomplex can also ensure a tRNA-folding quality-control checkpoint before the sequential docking of the maturation enzymes.

As essential components of the protein synthesis machinery, tRNAs undergo a tightly controlled biogenesis process, which include the incorporation of numerous posttranscriptional modifications. Defects in these tRNA maturation steps may lead to the degradation of hypomodified tRNAs by the rapid tRNA decay (RTD) and nuclear surveillance pathways. We previously identified m¹A58 as a late modification introduced after modifications Ψ55 and T54 in yeast elongator tRNA^Phe. However, previous reports suggested that m¹A58 is introduced early during the tRNA modification process, in particular on primary transcripts of initiator tRNA_i^Met, which prevents its degradation by RNA decay pathways. Here, aiming to reconcile this apparent inconsistency on the temporality of m¹A58 incorporation, we examined its introduction into yeast elongator and initiator tRNAs. We used specifically modified tRNAs to report on the molecular aspects controlling the Ψ55 → T54 → m¹A58 modification circuit in elongator tRNAs. We also show that m¹A58 is efficiently introduced on unmodified tRNA_i^Met, and does not depend on prior modifications. Finally, we show that m¹A58 has major effects on the structural properties of initiator tRNA_i^Met, so that the tRNA elbow structure is only properly assembled when this modification is present. This observation provides a structural explanation for the degradation of hypomodified tRNA_i^Met lacking m¹A58 by the nuclear surveillance and RTD pathways.