Finding priority bacterial ribosomes for future structural and antimicrobial research based upon global RNA and protein sequence analysis

Significance Statement

The growing number of antibiotic resistance pathogenic bacteria are of critical concern to the health profession. Many of the current classes of antibiotics target the bacterial ribosome, the protein making factory for these species. However, much of our knowledge of the bacterial ribosome is based upon non-pathogenic bacteria that are highly divergent from the major pathogens of concern. We have analysed the genetic variation of the RNA and protein components of all available bacterial ribosomes. This has led us to identify the highest priority groups of bacteria that would provide the most benefit most from further analysis of their ribosome structures, from both a medical and evolutionary perspective.

Introduction

A global rise of antibiotic resistant bacteria has become an increasingly urgent problem in recent years, with the bacterial ribosome, particularly the ribosomal RNA (rRNA) component, being a common antibiotic target (Tacconelli et al., 2018; Lin et al., 2018; Wilson, 2014). Due to the limitations and requirements of X-ray crystallography, antibiotic binding studies initially used extremophiles Thermus thermophilus (Polikanov et al., 2015), Deinococcus radiodurans (a 50S structure) (Harms et al., 2008; Kaminishi et al., 2015) and the archaeon Haloarcula marismortui (Ban et al., 2000), with the pathogenic Escherichia coli (Fischer et al., 2015; Noeske et al., 2015; Watson et al., 2020) ribosome structure introduced in later years (Lin et al., 2018; Wilson, 2014; Fox, 2010). However, improvements in electron cryomicroscopy have allowed more diverse ribosome structures to be analysed more recently, such as from non-pathogenic bacteria Bacillus subtilis (Sohmen et al., 2015; Crowe-McAuliffe et al., 2021a), Flavobacterium johnsoniae (Jha et al., 2021) and Mycobacterium smegmatis (Hentschel et al., 2017) (recently renamed “Mycolicibacterium smegmatis”), or from pathogens including Acinetobacter baumannii (Morgan et al., 2020; Zhang et al., 2021), Enterococcus faecalis (Murphy et al., 2020; Crowe-McAuliffe et al., 2021b), Listeria monocytogenes (Crowe-McAuliffe et al., 2021b; Koller et al., 2022), Mycobacterium tuberculosis (a 50S structure) (Yang et al., 2017), Pseudomonas aeruginosa (a 50S structure) (Halfon et al., 2019) and Staphylococcus aureus (Crowe-McAuliffe et al., 2021b; Eyal et al., 2015; Halfon et al., 2019). Each pathogen ribosome study highlights species-specific differences in comparison to both non-pathogenic and pathogenic structures, implying that non-pathogenic ribosome structures are not always optimal for inferring antibiotic binding across all bacteria (Morgan et al., 2020; Yang et al., 2017; Halfon et al., 2019; Eyal et al., 2015). As these structural differences could hinder antibiotic research it is important that solved ribosome structures, which form the basis of our understanding of the bacterial ribosome and ribosomal antibiotic binding, are suitably representative of pathogenic species (Lin et al., 2018; Wilson, 2014).

We further note that the specific target sites of antibiotics correspond to highly conserved, functional ribosome components, and that a survey focusing on these sites will not yield significant diversity. However, we also note that ribosome function has been found to be influenced by less well conserved components of structure (Lin et al., 2018; Eyal et al., 2015), and even by elements of structure quite distant from conserved antibiotic binding sites (Belousoff et al., 2017). In addition, there is evidence in S. aureus and P. aeruginosa suggesting that changes to the overall ribosome can lead to antibiotic resistance, in addition to direct mutations in the target sites (Halfon et al., 2019; Belousoff et al., 2017).

Therefore, the aim of this study is to broadly survey ribosome sequence space to determine how representative current ribosome structures are, ranging from the extremophiles (e.g., D. radiodurans, T. thermophilus) for human pathogenic species (e.g., S. aureus, L. monocytogenes), when compared to all bacterial species. Using this analysis, we propose new representative bacteria, including some surprising and diverse species, with the potential of yielding valuable new ribosome structural information that could be prioritised for future ribosomal structural studies.

Materials and Methods

Our analysis includes the full-length conserved components of the bacterial ribosome. We have elected not to specifically analyse antibiotic targets (e.g., Spectinomycin targets G1064 and C1992 of 16S rRNA (Brink et al., 1994)) as these are highly conserved functional components of the ribosome, which therefore carry a limited number of phylogenetically informative characters (Petrov et al., 2014; Bernier et al., 2018; Tirumalai et al., 2021). These studies also illustrate the conservation and variation across ribosomal RNA structures. Full-length sequences also allow a global overview of ribosome variation, some of which may change the behaviour of binding sites due to “action at a distance” (Lin et al., 2018; Eyal et al., 2015; Belousoff et al., 2017).

A total of 3,758 bacterial genomes and the H. marismortui genome (Accession: AY596297.1) were obtained from the European Nucleotide Archive (Amid et al., 2020; Baliga et al., 2004). One representative sequence was retained per species for each rRNA and ribosomal protein sequence, which were filtered to only include genomes with annotations covering 80% of the expected sequence length based upon consensus sequences. If paralogues were present, the sequence with the highest corresponding covariance model and hidden Markov model bit score for the species was used (Nawrocki & Eddy, 2013).

Both 16S and 23S rRNA were annotated using barrnap v0.9 (https://github.com/tseemann/barrnap), and INFERNAL 1.1.2 was used to create sequence alignments using Rfam v14.3 covariance models (Nawrocki & Eddy, 2013; Kalvari et al., 2018). Covariance models are the gold-standard for structural RNA alignment (Nawrocki & Eddy, 2013; Freyhult, Bollback & Gardner, 2007), and the Rfam models are derived from the CRW database (Cannone et al., 2002), which is the product of 20 years of manual alignment curation, resulting in some of the highest quality ribosomal RNA structural alignments available to date (Cannone et al., 2002; Gutell, Lee & Cannone, 2002).

Ribosomal protein sequences from 32 universally conserved proteins were identified from six-frame translations of whole genomes, to account for potentially inconsistent or absent genome annotations. The selected sequences were aligned using HMMER v3.1 (hmmer.org), with protein hidden Markov models (HMMs) from Pfam v33.1 (El-Gebali et al., 2019). Profile HMMs are the gold-standard for protein homology search and alignment (Eddy, 2011; Söding & Remmert, 2011), while Pfam is a widely used and highly curated database of protein domain alignments.

The advantages of this approach are that our results are up-to-date and consistent across the current genome databases, contain both RNA and protein components of the bacterial ribosome and utilise the most accurate methods for alignment production. Whilst we acknowledge that there are existing ribosomal datasets (e.g., SILVA, GreenGenes, RDB, etc.), these do not meet all the above requirements. Although there are alternative ribosomal RNA and protein alignments we could have used (e.g., Bernier et al., 2018; Doris et al., 2015; Hug et al., 2016), their use here would likely produce very similar results.

Phylogenetic trees for each alignment were generated using the maximum likelihood method from PHYLIP v3.697, with distance matrices of the pairwise distances between species computed in R v4.2.1 using ape v5.6 (Felsenstein, 2009; Paradis & Schliep, 2019). The distance matrices for each ribosomal gene were summed to create a single unified distance matrix, and this was used for multi-dimensional scaling (MDS) and visualisation (Fig. 1).

Results

Ribosomal sequence clustering suggests that several bacterial phyla are poorly represented by the currently available ribosome structures. To identify phyla that are poorly represented by solved ribosome structures, two rRNA and 32 universally conserved ribosomal protein phylogeny trees were initially created using a representative sequence for each species. To confirm that these universally conserved ribosomal proteins were truly conserved in our dataset, the trees were filtered on whether they contained more than 95% of isolates from each taxonomy class in the dataset. Taxonomy class was used over phyla to take into account variability within larger, overrepresented phyla such as Pseudomonadota (Proteobacteria) (Hug et al., 2016). This resulted in ribosomal proteins uS2, uS4, uL23, uL24 and uL30 being removed because they were completely absent in Epsilonproteobacteria and uL15 being removed as it was only present in 16% of Epsilonproteobacteria genomes. As assessing individual ribosomal sequences is unlikely to reflect structural variation, the remaining two rRNA and 26 ribosomal protein phylogenetic trees were combined into one distance matrix, resulting in a final dataset of 1,396 species. This combined tree was then reduced using MDS, allowing us to visualise the most dissimilarities between species (Hug et al., 2016).

The resulting MDS plot shows four main clusters; an Alphaproteobacteria cluster and a Betaproteobacteria plus Gammaproteobacteria cluster illustrate the non-monophyletic nature of Pseudomonadota (Proteobacteria) (Hug et al., 2016), one is dominated by Actinomycetota (Actinobacteria) and the fourth consists of the remaining bacterial phyla and Proteobacterial classes (Fig. 1). The Pseudomonadota (Proteobacteria) cluster follows observations made in previous studies, such as Beta and Gammaproteobacteria being closely related, and Deltaproteobacteria and Oligoflexia being more related to non-Proteobacteria (Hug et al., 2016; Waite et al., 2020). Due to these non-monophyletic properties, Pseudomonadota (Proteobacteria) classes will be treated as individual phyla for the rest of this study (Hug et al., 2016; Waite et al., 2020). Bacteroidota (Bacteroidetes), Acidithiobacilla and Epsilonproteobacteria formed slightly isolated groups away from the larger clusters, suggesting that these could become more defined clusters if their sample sizes were larger (Fig. 1). Alphaproteobacteria is the only cluster without a solved structure, implying the presence of phyla specific variation that has not been captured by current structures (Fig. 1). Therefore, it is unlikely that all the bacterial phyla are well represented by the available ribosome structures, given that the solved structures in the multiple phyla cluster group together, instead of being evenly distributed throughout the clusters (Fig. 1).

Discussion

Our results reveal significant phylogenetic gaps in the existing cohort of solved ribosome structures. Filling these gaps will help identify clade-specific features of isolated bacterial clades that lack representative ribosome structures. This is important for the analysis of ribosome evolution, pathogen research, and potentially for antibiotic development.

There are two limitations with this approach to ribosome structure prioritisation: there is a bias towards species with available genomes, which are enriched in pathogenic species rather than being the most evolutionarily representative (Wu et al., 2009). In addition, our approach prioritises the most novel ribosomes within each major bacterial clade, with the risk of over-emphasising the most diverse ribosomes. This is somewhat mitigated by selecting the most closely related pathogenic species as representative targets.

While having at least one representative ribosome structure per phyla is expected to capture the bulk of structural variation (Table 1), there is no guarantee that primary sequence differences result in significant changes at the tertiary level. A prominent example of this is H. marismortui which, although it is an archaean ribosome structure, has been used as an alternative to bacterial counterparts (Fox, 2010), despite having no close relatives at the primary level (Fig. 1). Another consideration is that the structural variation observed may be species-specific, rather than shared across a phylum, with species-specific differences having been observed between P. aeruginosa and A. baumannii, yet no phyla-specific differences when compared to E. coli and other available structures (Morgan et al., 2020; Halfon et al., 2019). However, these three structures were observed to be relatively divergent from each other based upon the MDS analysis (Fig. 1), which reinforces the in vivo observation that P. aeruginosa and A. baumannii ribosomes are more similar to each other than to E. coli, suggesting that phylogenetic analyses have the potential to reflect structural variation (Morgan et al., 2020).

Our list of high-priority bacterial ribosomes reveals that a number of bacterial clades are poorly represented by the existing solved structures (Table 1, Table S1). Working down this list gives an opportunity to identify the most variant ribosomes based upon sequence analysis, which in many instances will reflect diverse ribosome structures. These will allow the development of more specific classes of antibiotics, as well as provide further information regarding the specifics of ribosome structure evolution.

Supplemental Information