MODOMICS: a database of RNA modifications and related information. 2023 update

1RNA modifications are chemical alterations to RNA nucleotides that have a profound impact on RNA structure and function. To date, >170 RNA modifications have been identified in all classes of RNA molecules. In living organisms, RNA modifications are introduced post- or co-transcriptionally by a variety of enzymes, and modification reactions can form complex pathways, leading to hypermodified residues. RNA molecules can also acquire modifications from external sources, such as the environment or other organisms. Moreover, specific RNA modifications are reversible, wherein the enzymatic removal of modified residues permits a rapid post-transcriptional adaptation to changing cellular or environmental contexts. RNA modifications have diverse effects on RNA molecules themselves, impacting their stability, structure, function, processing, and regulatory roles within the cell and they can be recognized by specialized reader proteins to fulfill their purpose. RNA modifications play important roles in all aspects of RNA metabolism, including RNA splicing, polyadenylation, transport, localization, translatability and stability. They can also be important for RNA interactions with other molecules, particularly proteins and ribonucleoproteins.

2A growing area of interest lies in the deliberate introduction of artificial chemical modifications to RNA. These synthetic alterations are designed to bestow RNA molecules with functionalities that are not naturally occurring, serving specific experimental or therapeutic purposes. One prominent application of such modifications is in structural studies, where they provide enhanced stability to RNA molecules, ensuring that the RNA maintains its desired conformation during the analysis.

3Over the past two years, the field of RNA modification has witnessed advancements in the novel technological approaches for detection and quantification of modified residues, in particular based on the nanopore technology. Advancements in cryo electron microscopy (cryo-EM) have significantly impacted RNA structure determination. Recent cryo-EM studies have achieved resolutions in the 2–3 Å range or better, enabling the identification of modified nucleotides for which the Coulomb density maps facilitate the direct visualization and atomic-level assignment of modified residues.

4As in the previous release, MODOMICS hosts a catalog of modified residues, enzymes and guide RNAs responsible for individual reactions, RNA modification pathways, sequences of modified RNAs, a catalog of ‘building blocks’ for chemical synthesis of modified RNA, links of RNA modifications to different diseases, and other associated data such as relevant publications. The updated MODOMICS introduces new data types, including a coding system that expands the catalog of modified residues to include synthetic molecules, and an annotation system that indicates the nature and reliability of the data interpretation (Figure 1).

5The catalog of modifications in MODOMICS includes ribonucleosides and ribonucleotides present in RNAs existing in nature, which are introduced into RNA post- or co-transcriptionally by various enzymes. In the previous release, we added sequences of experimentally determined RNA structures that contained modifications present in nature. However, some of these RNAs also contained deoxyribonucleotide residues, and various chemically synthesized residues that had no counterpart among the naturally present modifications, e.g. locked nucleic acids (LNAs), phosphorothioate-modified nucleic acids, fluorine substitutions, and many others, which weren’t included in MODOMICS. For the current release, the catalog of modified ribonucleotide residues was expanded to include 79 modified residues found in experimentally determined RNA structures, including ‘unnatural’ ones (not found in nature). Currently, MODOMICS enables the inclusion of modified residues that are not a part of natural modification pathways. This expansion improved the correspondence between MODOMICS and the entries in the RCSB Protein Data Bank, covering both natural and synthetic residues.

6The extension of MODOMICS to handle potentially a very large number of synthetic residues required an update of the nomenclature system, which was originally developed to deal with only a limited number (e.g. a few hundred) of naturally existing nucleoside and nucleotide residues. Apart from full chemical names, MODOMICS uses several codes to denote modified residues, including:

7short names, traditionally used in literature, e.g. Am, m1A, io6A, cm5s2U, pms2i6A etc.

8a single character code, this system is particularly useful for the presentation of sequence alignments, where each column must contain only a single character.

9MODOMICS code, introduced in the 2017 update that uses one or more digits and one letter (A, G, C, U) as a unique identifier to represent each modified residue. It was designed to be readable both for humans and computer software. For example, modified residues Am, m1A, io6A, cm5s2U, pms2i6A were indicated as 0A, 1A, 60A, 2540U and 2161551A, respectively.

10In this update of MODOMICS, modifications had to be introduced into all three codes. For short names of unnatural residues that are represented in the RCSB database, MODOMICS uses the RCSB LIG codes, e.g. 5BU for 5-bromo-uridine-5′-monophosphate, LCC for [(1R,3R,4R,7S)-7-hydroxy-3-(5-methylcytosin-1-yl)-2,5-dioxabicyclo[2.2.1]hept-1-yl]methyl dihydrogen phosphate. For natural residues, MODOMICS continues to use traditional names, regardless of whether they have RCSB LIG codes or not. For unnatural residues not available in the RCSB PDB that may be included in MODOMICS in the future, relevant short names will be provided. The single character code was extended to Unicode UTF-8 characters, including those from the Chinese alphabet. We also replaced some of the previously used characters, which commonly caused problems for computational parsing or displaying plain text files (such as ‘, ’, “, ”, or =). MODOMICS presents both the previous (now obsolete) and the new updated, computer-friendly single-character code.

11The MODOMICS code was subjected to a major overhaul. To handle multiple unnatural residues, we extended it to use uniformly 10 digits and a letter. To keep the code both human- and computer-friendly, the first digit serves as a barcode to indicate the type of sugar moiety in nucleoside and nucleotide residues i.e. D-deoxyribose, d-ribose, or another (‘unnatural’) moiety, and the presence of a regular phosphate group or a modified phosphate group in nucleotide residues (see Table 1 for examples and explanation). The letter in the code is A, U, C or G if the base moiety is: (i) an unmodified adenine, uracil, cytosine, or guanine (regardless of the biosynthetic pathway, e.g. a result of C to U editing is simply U); (ii) a chemically modified (naturally or unnaturally) derivative of adenine, uracil, cytosine or guanine that keeps all the non-hydrogen atoms of A, U, C or G; (iii) if it is a result of a naturally occurring chemical modification of adenine, uracil, cytosine or guanine, even if non-hydrogen atoms of the respective natural unmodified base are not maintained after the modification (e.g. G for queuosine, A for inosine). In all remaining cases, in particular for residues that do not possess all the non-hydrogen atoms of A, U, C or G (e.g. for 2-amino-9-[2-deoxyribofuranosyl]-9H-purine-5′-monophosphate or d-ribofuranosyl-benzene-5′-monophosphate), the letter is X. The entire string of 10 digits and a letter is a unique identifier for a chemical moiety entry in MODOMICS.

12All the new chemical moieties imported to MODOMICS were checked by experts in our team. This process included the standardization of structures and correcting/adding stereochemical information. As for the moieties previously available, various physicochemical properties were calculated. Each chemical structure is available for download in multiple formats.

13The RNA sequences section in MODOMICS was updated. From RCSB we imported 900 new sequences with naturally/unnaturally modified residues and updated two existing entries previously imported to introduce such residues absent in earlier MODOMICS versions. We added 15 tRNA sequences and 15 rRNA sequences. Users can download individual sequences in the FASTA format, with modifications represented as the MODOMICS code, single character code, or short names.

14The MODOMICS database used to have a section with sequence alignments of tRNAs and rRNAs (LSU and SSU). It was removed during one of the previous updates when a new relational data model for handling sequences was introduced. We now restored this section. ‘Unmodified’ versions of sequences from MODOMICS were aligned to covariance models from Rfam using Infernal. They were then replaced with the modified versions and the alignments were curated manually, taking the 3D structure information into account. New functions were added for data retrieval. Users can download the alignments in the multiple-sequence gapped FASTA format (which is relevant only when modifications are represented in a single Unicode character code), or in a tab-separated values (tsv) format, which enables the presentation of modified residues in various formats, without breaking the correspondences in columns.

15The MODOMICS section on proteins involved in RNA modification pathways was updated to add 57 proteins involved in RNA editing. Annotations of various protein entries were updated based on the literature review. We linked MODOMICS protein entries to over 1000 experimentally verified structures of proteins involved in RNA modification from the RCSB PDB database (including multiple RCSB entries for a number of individual proteins). Whenever available, we linked the MODOMICS entries to 3D structural predictions from the AlphaFold database. The interface was updated to enable visualization of structures and downloading the atomic coordinates or exporting their sequences. Protein sequences can be downloaded in FASTA format.

16We enhanced the data structure to facilitate more detailed annotations. Specifically, we introduced two annotation features: evidence indicates the level of experimental or computational support for a given data feature, along with its source (e.g. specific publication, database import, method used, computational prediction or human inference from metadata). Estimated reliability represents the likelihood of the data's accuracy based on the available evidence.

17For evidence, we defined six classes: (i) multiple direct experimental evidence (validated using at least two different direct methods), (ii) direct experimental evidence (characterized by one direct method); (iii) inferred based on indirect experimental evidence; (iv) predicted computationally; (v) unknown (evidence not yet annotated); (vi) irrelevant (evidence cannot be estimated for a given data type). For estimated reliability, we defined six classes as well: (i) highly reliable; (ii) solid (large errors are unlikely, but cannot be completely excluded, minor errors should not undermine the meaning of the data); (iii) speculative (errors expected); (iv) questionable (e.g. conflicting reports in the literature and in databases); (v) unknown (assessment not yet performed); (vi) irrelevant (cannot be estimated for a data type). This annotation system applies to all MODOMICS entries, encompassing data obtained experimentally, computationally, or both. Examples include:

18Presence of a modified residue in a specific RNA position. For instance, if identified by mass spectrometry and visualized in a high-resolution cryo-EM Coulomb density map, it would be classified as evidence class 1 and reliability class 1. Inference from cDNA sequencing mutation patterns would be evidence class 3 and reliability class 3.

19Determining a protein's role in a modification reaction. An in vitro activity without in vivo confirmation would be evidence class 2 and reliability class 2. Computational prediction based on similarity to a characterized homolog would be evidence class 4 and reliability class 3.

20Associating an RNA-modifying enzyme with a disease. Correlation of expression patterns with disease symptoms, lacking a mechanistic explanation, would be evidence class 3 and reliability class 3.

21Computational alignment of RNA family sequences. Alignments for extant sequences, being computational, would be evidence class 4. Manually curated alignments would typically be reliability class 2, while automated alignments of divergent sequences would be class 3.

22In this update, we primarily focused our annotation efforts on modifications in rRNA and tRNA sequences. Our goal is to systematically annotate all RNA types and data categories in the upcoming future. At present, any unannotated data is labeled under evidence and reliability class 5, which signifies pending annotation and reliability assessment. As always, MODOMICS users are encouraged to provide feedback to enhance the annotations.

23In upcoming MODOMICS updates, we plan to include the annotation of prevalence for individual positions of modified residues. This will detail the frequency of specific features at given positions in RNA molecules, shedding light on modifications present in only a fraction of an RNA population. We also plan to expand the collection of synthetic modifications, including ones beyond those entries in RCSB. We plan to cover new therapeutic strategies that are being developed to target RNA modifications, in particular small molecule inhibitors of RNA modification enzymes and RNA-modifying proteins. We will continue cross-referencing with other databases to ensure comprehensive coverage of RNA modification data. Enhancements to the user interface are also planned to streamline data access, visualization and feedback.