Many genes carry information for making proteins. To make a protein, a working copy of the information stored in DNA is first copied into a molecule of messenger RNA. These RNA messages are then interpreted by the ribosome, the molecular machine that makes proteins. Many messages are produced from each gene, and each message can be read multiple times. Thus, it should follow that the number of messages produced dictates the number of proteins made. However, this is not the case and the number of proteins produced cannot be completely predicted from knowing the number of messenger RNAs.

Cells control how much of a given protein they produce through interactions between the messenger RNAs and other regulatory RNAs. The regulatory RNAs bind directly to a message and impede protein production. Because there are millions of RNAs in a cell, these interactions have evolved to be highly specific. Nevertheless, it seems inevitable that messenger RNAs would encounter other RNAs too, which could short-circuit gene regulation and lead to less protein being produced.

Umu et al. have now asked if such short-circuit events are selected against during evolution. Computational tools were used to predict the strength of binding between the RNAs found in the dominant forms of microbial life on Earth: the bacteria and the archaea. This approach revealed that the majority of messenger RNAs bind more weakly to the most common RNA molecules found in cells than would be expected by chance. Weakened binding should prevent the RNA molecules from becoming tangled with each other and ensure that protein levels are not perturbed by unintended interactions between highly expressed messages and other RNAs.

To test this hypothesis further, Umu et al. generated versions of the gene for a green fluorescent protein that differed only in how well their messenger RNAs could avoid interacting with the most abundant RNAs in E. coli cells. Those messengers that were designed to avoid interacting with other RNAs yielded far more protein than those that were not. The findings show that taking this kind of avoidance into account can improve predictions about how much protein will be produced and should therefore make it easier to control protein production in experimental systems.

Finally, the messenger RNAs of some bacteria do not show such clear avoidance. However, these bacteria have a more complex internal cell structure. This finding hints at an alternative means for avoiding short-circuiting events that could be used by more complicated cells, such of those of animals and plants, which also contain much larger numbers of RNAs.

It should in principle be possible to predict protein abundance from genomic data. However, protein and mRNA levels are not strongly correlated (de Sousa Abreu et al., 2009; Vogel and Marcotte, 2012; Kwon et al., 2014; Maier et al., 2011; Lu et al., 2007; Taniguchi et al., 2010; Chen et al., 2016), which is a major barrier to precision bioengineering and quantification of protein levels. mRNA secondary structure (Pelletier and Sonenberg, 1987; Chamary and Hurst, 2005), codon usage (Ikemura, 1981; Sharp and Li, 1987; Andersson and Kurland, 1990), and mRNA (and protein) degradation rates (Maier et al., 2011) are commonly invoked to explain this discrepancy (Boël et al., 2016-21). Yet, at best, these features account for only 40% of variation, and in some instances explain very little of the observed variation (Kudla et al., 2009; Maier et al., 2011; Plotkin and Kudla, 2011; Goodman et al., 2013; Chen et al., 2016). Here we show that crosstalk interactions between ncRNAs and mRNAs also impact protein abundance, and that such interactions have a greater effect than either mRNA secondary structure or codon usage. We measured interactions between a set of evolutionarily conserved core mRNAs and ncRNAs from 1700 prokaryotic genomes using minimum free energy (MFE) models. For 97% of species, we find a reduced capacity for interaction between native RNAs relative to controls. Furthermore, by generating synonymously variant GFP mRNAs that differ in their potential to interact with core ncRNAs, we demonstrate that GFP expression levels can be both predicted and controlled. Our results demonstrate that there is strong selection for the avoidance of stochastic mRNA:ncRNA interactions across prokaryotes. Applying this knowledge to mRNA design will enable precise control of protein abundance through the incorporation or exclusion of inhibitory interactions with native ncRNAs.

To examine if avoidance of stochastic mRNA:ncRNA interactions is a feature of transcriptomes in bacteria and archaea, we estimated the strength of all possible intermolecular RNA interactions using a minimum free energy (MFE) model (Mückstein et al., 2006) using core ncRNAs and mRNAs. In this work the core ncRNAs are six well conserved and highly expressed tRNA, rRNA, RNase P RNA, SRP RNA, tmRNA and 6S RNA families annotated by Rfam (Gardner et al., 2011; Nawrocki et al., 2015), the core mRNAs are 114 well conserved mRNAs found across bacteria, 40 of which are also conserved across archaea (Wu et al., 2013).

If stochastic interactions are selected against, because of the capacity for abundant ncRNAs (Lindgreen et al., 2014; Deutscher, 2006; Giannoukos et al., 2012) to impact translation (Waters and Storz, 2009; Storz et al., 2011), such negative selection would be most comparable between species and readily detected for broadly conserved ncRNAs and mRNAs. Under-representation of interactions has been considered for the specific case of Shine-Dalgarno-like (SD-like) sequences and the ribosome (Li et al., 2012; Woolstenhulme et al., 2015; Borg and Ehrenberg, 2015; Diwan and Agashe, 2016) and between microRNAs and 3' UTRs (Bartel and Chen, 2004; Farh, 2005; Stark et al., 2005; van Dongen et al., 2008). We computed the free energy distribution of interactions between highly conserved mRNA:ncRNA pairs and compared this to a number of negative control interactions, which serve to show the expected distribution of binding energy values (Figure 1A). The initiation of translation has been shown to be the rate limiting step for translation (Tuller and Zur, 2015; Plotkin and Kudla, 2011; Nakahigashi et al., 2014), therefore, we focus our analysis on the first 21 nucleotides of the mRNA coding sequence (CDS). This has the further advantage of reducing computational complexity. We also test a variety of negative control mRNA regions, which are unlikely to play a functional role in RNA:ncRNA interactions. The mRNA controls include (1) di-nucleotide preserving shuffled sequences (Workman, 1999) (orange, Figure 1A), (2) homologous mRNAs from another phylum (with a compatible guanine-cytosine (G+C) content) (purple), (3) downstream regions 100 base pairs (bps) within the CDS (pink), (4) the reverse complement of the 5´ of CDSs (green), and lastly (5) unannotated (intergenic) genomic regions (yellow). Our interaction predictions in a single model strain show that native interactions consistently have higher (i.e. less stable) free energies than expected when compared to the five different mRNA negative controls: that is, there is a reduced capacity for native mRNAs and native ncRNAs to interact. We also compared different energy models and confirm that the MFE shift is a result of intermolecular binding (Figure 1—figure supplement 1A–C). We subsequently deployed the most conservative negative control (i.e. di-nucleotide preserving shuffle) and free energy model (Figure 1—figure supplement 1C) to detect if this shift for less stable binding of mRNA:ncRNA is true of all bacteria and archaea.

Figure 1 with 2 supplements see all

Download asset Open asset

In terms of stoichiometry, the model we use assumes that ncRNA expression levels are vastly in excess of mRNA expression levels (i.e. [ncRNA] >> [mRNA]) (Giannoukos et al., 2012; Deutscher, 2006). This is generally a biologically reasonable assumption when focussing on core genes based upon past analysis and our own work with RNA-seq data from a range of bacteria and archaea (Figure 4) (Lindgreen et al., 2014). Consequently, any potential mRNA interaction regions are saturated with ncRNA, therefore a summative model of interaction energies is a reasonable approximation to the estimated impact of excess hybridization. If modelling ncRNAs that are not so abundant, then a model weighted by expression level may be advantageous, but it is difficult to assess these across all conditions and developmental stages that are evolutionarily relevant. In order to ensure that our analysis is comparable across all bacteria and archaea we have focussed on just the most highly conserved ncRNA and protein-coding genes. Although, many of the ncRNAs are highly structured and are bound by RNA-binding proteins this is not the case during either synthesis and degradation of these products, furthermore, a fraction of the RNA components of these genes will be exposed. Therefore, we expect these will form useful datasets for initial testing of our hypothesis.

In order to assess whether mRNA:ncRNA avoidance is an evolutionarily conserved phenomenon, we calculated intermolecular binding energies for conserved ncRNAs and mRNAs from 1,582 bacterial and 118 archaeal genomes and compared these to a negative control dataset derived using a di-nucleotide frequency preserving shuffling procedure (Workman, 1999). This measures a property that we call the ‘extrinsic avoidance‘ of mRNA:ncRNA interactions, yet this approach may fail to identify genuine avoidance in cases when the G+C content differences between interacting RNAs is extreme. Measuring only extrinsic avoidance (using shuffled mRNAs as negative controls), we found that stochastic mRNA:ncRNA interactions are significantly underrepresented in most (73%) of the prokaryotic phyla (p<0.05, one-tailed Mann-Whitney U test) (Figure 1B,C and Figure 1—figure supplement 2). This indicates that there is selection against stochastic interactions in both bacteria and archaea.

We next sought to establish the degree to which intrinsic G+C features of RNAs lead to avoidance of stochastic interactions (Figure 1D). A similar idea has been proposed which suggests that purine loading in thermophilic bacteria may limit mRNA:mRNA interactions (Lao and Forsdyke, 2000). A test of G+C composition revealed a significant difference (p<0.05, two-tailed Mann-Whitney U test) between mRNAs and ncRNAs for 95% of bacteria and archaea (Figure 1D,E). Therefore, either extrinsic or intrinsic avoidance signals indicate that selection against stochastic interactions and it is near-universal for the prokaryotes (97% of all strains) (Figure 1E and Supplementary file 1A and B).

Our results clearly establish a signature of selection that acts to minimise stochastic mRNA:ncRNA interactions. However, with thousands of potential interacting RNA species in even simple prokaryotic systems (Vivancos et al., 2010; Sharma et al., 2010), the complete avoidance of stochastic interactions is combinatorially unlikely. Therefore, there ought to be a tradeoff between avoidance and optimal expression. To assess this, we examined the relationship between potential stochastic interactions and the variation between mRNA and cognate protein levels for four previously published endogenous mass spectrometry datasets from Escherichia coli (E. coli) and Pseudomonas aeruginosa (P. aeruginosa) (Laurent et al., 2010; Kwon et al., 2014; Lu et al., 2007). We computed Spearman’s correlation coefficients between protein abundances and extrinsic avoidance, 5´ end internal mRNA secondary structure and codon usage. Of the three measures, avoidance is significantly correlated in all four datasets (Spearman’s rho values are between 0.11–0.17 and corresponding p-values are between 0.01 and 1.3 × 10^–12). In contrast, 5´ end mRNA structure significantly correlates in two datasets, and codon usage significantly correlates in all four datasets. This indicates that, despite strong selection against stochastic interactions, such interactions do significantly impact the proteome (Figure 2A and Supplementary file 3). We have also conducted an ‘outlier analysis’ on one of the E. coli datasets (Laurent et al., 2010). We have selected the top and bottom-most expressed genes relative to mRNA expression levels and computed Z-scores for each of codon-usage, internal secondary structure and avoidance measures. We found that avoidance measures show the most extreme shifts downwards for the bottom-most expressed genes and is shifted the highest for the top-most genes (Figure 2—figure supplement 4).

Figure 2 with 5 supplements see all

Download asset Open asset

We also test how mRNA:ncRNA crosstalk impacts the translation of transformed mRNAs that have not coevolved with the ncRNA repertoire (low avoidance mRNAs are rare in native datasets). We examined two available E. coli-based GFP experimental datasets (Goodman et al., 2013; Kudla et al., 2009), where synonymous mRNAs are generated for a GFP reporter gene. This enables the assessment of the impact of synonymous changes on protein abundance using fluorescence. Avoidance and mRNA secondary structure are both significantly correlated with fluorescence, whereas codon usage is not (Spearman’s rho values are 0.11 and 0.65, the corresponding p-values are 3.17 × 10⁻⁴¹ and 1.69^–20) (Figure 2A). Note that one of the GFP datasets (Goodman et al., 2013) uses native E. coli mRNA 5´ ends for their constructs, whereas the other GFP dataset (Kudla et al., 2009) is randomly generated. We observe that the influence of avoidance on gene expression for randomly sampled synonymous mRNAs is strong (Figure 2—figure supplement 3), while endogenous gene expression is limited. Presumably, due to negative selection pruning low avoidance mRNAs from the gene pool (Figure 2A).

For each of the seven datasets described above we have tested linear models of measures of mRNA levels, codon usage, internal secondary structure and avoidance (Figure 2—figure supplement 3 and Supplementary file 5). Avoidance alone explains around 35% of variance in GFP datasets where extreme mRNA compositions can be explored, whereas in native mass-spec derived datasets 2–3% of the variance is explained by avoidance alone. Codon usage describes 2% to −0.5% of variance in GFP data, and 19% to 0.3% of variance in mass-spec derived datasets. Internal secondary structure 33% to 10% in GFP datasets, and 0.2% to 0% of the variance in mass-spec derived datasets. Using all four measures in combination across the seven datasets between 70% and 42% of variation in protein levels can be explained, removing avoidance from the model reduces these estimates by between 56% and 0.7%. Thus, avoidance is at least as good an explanation of variation in protein abundance as either codon usage and internal mRNA secondary structure.

Our results indicate that crosstalk between mRNAs and ncRNAs can impact protein expression levels. We therefore predict that taking crosstalk into account will enable the design of constructs where protein expression levels can be precisely controlled. To test this, we generated GFP constructs based on the following constraints: codon bias, 5´ end mRNA secondary structure stability and crosstalk avoidance (see Materials and methods). Our constructs are designed to capture the extremes of one variable, while controlling other variables (e.g. high or low avoidance and near-average codon bias and mRNA secondary structure). The G+C content, a known confounding factor, was also strictly controlled for each construct. We selected a commercial service to perform our GFP transformations to avoid possible bias and increase the robustness of our approach (Ioannidis and Khoury, 2011). We predicted that a construct where all three parameters are optimised will result in a higher expression. Consistent with predictions, our optimised construct had maximal expression (Figure 2—figure supplement 1). Of the three parameters, avoidance showed the largest range, suggesting that tuning this parameter permits expression levels to be finely controlled (R_s = 0.56, p=6.9 × 10⁻⁶) (Figure 2B–D and Figure 2—figure supplements 1–4).

For a final confirmation of the avoidance hypothesis, we tested the Thermus thermophilus (T. thermophilus) HB8 SSU ribosomal RNA, which is a component of one of the most complete prokaryotic ribosomal structures available in the PDB (Rozov et al., 2015). We identified the regions of the SSU rRNA that had the least capacity to interact with T. thermophilus core mRNAs and found that these regions were generally not bound to either ribosomal proteins or other ncRNAs, such as the LSU rRNA (p=2.49 × 10⁻¹⁷, Fisher’s exact test) (Figure 3; see ‘Materials and methods’). The influence of internal SD-like regions on translation pausing have been described elsewhere (Li et al., 2012), in addition we note that the anti-SD region on SSU rRNA is one of the RNA avoidance regions (Figure 3A).

Figure 3 with 3 supplements see all

Download asset Open asset

This study focusses on the 5´ ends of the CDS as this region is important for the initiation of translation (Plotkin and Kudla, 2011; Tuller and Zur, 2015) and is a consistent feature of all the genomic, transcriptomic, proteomic and GFP expression datasets that we have evaluated in this work. In smaller-scale tests we have observed similar conserved avoidance signals within the entire CDSs (Figure 3—figure supplement 1) and within the 5´UTRs (Figure 3—figure supplement 2). Furthermore, we predict that similar signals can be observed for mRNA:mRNA and ncRNA:ncRNA avoidance. Although the impacts of these features are challenging to validate, interactions between clustered regularly interspaced short palindromic repeats (CRISPR) spacer sequences (Bhaya et al., 2011) and core ncRNAs are good candidates to test ncRNA:ncRNA avoidance.

In conclusion, our results indicate that the specificity of prokaryotic ncRNAs for target mRNAs is the result of selection both for a functional interaction and against stochastic interactions. Our experimental results support the view that stochastic interactions are selected against, due to deleterious outcomes on expression. We suspect avoidance of crosstalk interactions has several evolutionary consequences. First, as transcriptional outputs become more diverse in evolution, we expect that the probability of stochastic interactions for both new ncRNAs and mRNAs becomes higher. This will impact the emergence of new, high abundance RNAs, since selection for high abundance may be mitigated by deleterious crosstalk events. Second, we predict that stochastic interactions limit the number of simultaneously transcribed RNAs, since the combinatorics of RNA:RNA interactions imply that eventually stochastic interactions cannot be avoided. This may in turn drive selection for forms of spatial or temporal segregation of transcripts. Finally, taking codon usage, mRNA secondary structure and potential mRNA:ncRNA interactions into account allows better prediction of proteome outputs from genomic data, and informs the precise control of protein levels via manipulation of synonymous mRNA sequences (Figure 2—figure supplement 5).

Here we summarize the data sources, materials and methods corresponding to our manuscript. We performed all statistical analyses in R, and all other computational methods in Python 2.7 or Bash shell scripts. We explicitly cite all the bioinformatics tools and their versions. All tables (Supplementary files 1–5) are available as supporting online material. All of our own sequences, scripts and R workspace images are available on Github including the supplementary files (http://github.com/UCanCompBio/Avoidance). The other datasets are cited in the manuscript (Supplementary file 3).

Green fluorescence protein (GFP) mRNA design

Request a detailed protocol

We sampled 537,000 synonymous mRNA variants of a GFP mRNA (the 239 AA, 720 nucleotide long, with accession AHK23750, can be encoded by 7.62 × 10¹¹¹ possible unique mRNA variants). In brief, these mRNA variants were scored based upon (1) CAI, (2) mRNA secondary structure in their 5´ end region, and (3) mRNA:ncRNA interaction avoidance in their 5 end region. The genome of E. coli BL21 encodes 52 unique core ncRNAs (Gardner et al., 2011; Nawrocki et al., 2015), to estimate the level of ncRNA avoidance for each GFP mRNA, we sum the binding MFEs. For example, for each GFP mRNA we compute 52 independent binding MFE values for each ncRNA. In short, a higher summed MFE score for a GFP mRNA implies a higher avoidance, while a lower summed MFE score implies a lower avoidance. This approach assumes that the ncRNAs are expressed at much higher levels than GFP mRNAs (i.e. [ncRNA] >> [mRNA]) (Figure 4). Consequently, any potential interaction site on GFP mRNAs are likely to be saturated with ncRNA. Finally, we selected 13 GFP mRNA constructs, while controlling the range of G+C values. These GFP mRNAs were designed to have four different aspects; extreme 5 end secondary structure (2 minimum and 2 maximum folding MFE constructs), extreme codon bias (2 maximum and 2 minimum CAI constructs), extreme interaction avoidance (2 minimum and 2 maximum binding MFE constructs) and an ‘optimal’ construct. The optimal construct was selected for a high CAI, low 5´ end structure and high avoidance. All extreme GFP mRNA constructs have near identical G+C content (between 0.468–0.480) and identical G+C contents at the 5´ end (0.48). Each of the sampled GFP mRNAs is separated from other mRNAs by at least 112 nucleotide substitutions and 122 nucleotide substitutions on average (Figure 2—figure supplement 1).

Figure 4

Download asset Open asset

1. 1. Andersson SG
   2. Kurland CG

   (1990)

   Codon preferences in free-living microorganisms

   Microbiological Reviews 54:198–210.

   • Google Scholar
2. 1. Bartel DP
   2. Chen CZ

   (2004) Micromanagers of gene expression: the potentially widespread influence of metazoan microRNAs

   Nature Reviews Genetics 5:396–400.

   https://doi.org/10.1038/nrg1328

   • Google Scholar
3. 1. Bhaya D
   2. Davison M
   3. Barrangou R

   (2011) CRISPR-Cas systems in bacteria and archaea: versatile small RNAs for adaptive defense and regulation

   Annual Review of Genetics 45:273–297.

   https://doi.org/10.1146/annurev-genet-110410-132430

   • Google Scholar
4. 1. Borg A
   2. Ehrenberg M

   (2015) Determinants of the rate of mRNA translocation in bacterial protein synthesis

   Journal of Molecular Biology 427:1835–1847.

   https://doi.org/10.1016/j.jmb.2014.10.027

   • Google Scholar
5. 1. Boël G
   2. Letso R
   3. Neely H
   4. Price WN
   5. Wong K-H
   6. Su M
   7. Luff JD
   8. Valecha M
   9. Everett JK
   10. Acton TB
   11. Xiao R
   12. Montelione GT
   13. Aalberts DP
   14. Hunt JF

   (2016) Codon influence on protein expression in E. coli correlates with mRNA levels

   Nature 529:358–363.

   https://doi.org/10.1038/nature16509

   • Google Scholar
6. 1. Chamary JV
   2. Hurst LD

   (2005) Evidence for selection on synonymous mutations affecting stability of mRNA secondary structure in mammals

   Genome Biology 6:R75.

   https://doi.org/10.1186/gb-2005-6-9-r75

   • Google Scholar
7. 1. Chen WH
   2. van Noort V
   3. Lluch-Senar M
   4. Hennrich ML
   5. Wodke JA
   6. Yus E
   7. Alibés A
   8. Roma G
   9. Mende DR
   10. Pesavento C
   11. Typas A
   12. Gavin AC
   13. Serrano L
   14. Bork P

   (2016) Integration of multi-omics data of a genome-reduced bacterium: Prevalence of post-transcriptional regulation and its correlation with protein abundances

   Nucleic Acids Research 44:1192–1202.

   https://doi.org/10.1093/nar/gkw004

   • Google Scholar
8. 1. Cock PJ
   2. Antao T
   3. Chang JT
   4. Chapman BA
   5. Cox CJ
   6. Dalke A
   7. Friedberg I
   8. Hamelryck T
   9. Kauff F
   10. Wilczynski B
   11. de Hoon MJ

   (2009) Biopython: freely available Python tools for computational molecular biology and bioinformatics

   Bioinformatics 25:1422–1423.

   https://doi.org/10.1093/bioinformatics/btp163

   • Google Scholar
9. 1. de Sousa Abreu R
   2. Penalva LO
   3. Marcotte EM
   4. Vogel C

   (2009) Global signatures of protein and mRNA expression levels

   Molecular BioSystems 5:1512–1526.

   https://doi.org/10.1039/b908315d

   • Google Scholar
10. 1. Deutscher MP

    (2006) Degradation of RNA in bacteria: comparison of mRNA and stable RNA

    Nucleic Acids Research 34:659–666.

    https://doi.org/10.1093/nar/gkj472

    • Google Scholar
11. 1. Diwan GD
    2. Agashe D

    (2016) The frequency ofinternal shine-dalgarno-like motifs in prokaryotes

    Genome Biology and Evolution 8:1722–1733.

    https://doi.org/10.1093/gbe/evw107

    • Google Scholar
12. 1. Eddy SR

    (2011) Accelerated Profile HMM Searches

    PLoS Computational Biology 7:e1002195.

    https://doi.org/10.1371/journal.pcbi.1002195

    • Google Scholar
13. 1. Farh KK
    2. Grimson A
    3. Jan C
    4. Lewis BP
    5. Johnston WK
    6. Lim LP
    7. Burge CB
    8. Bartel DP

    (2005) The widespread impact of mammalian MicroRNAs on mRNA repression and evolution

    Science 310:1817–1821.

    https://doi.org/10.1126/science.1121158

    • Google Scholar
14. 1. Fisher RA

    (1922) On the interpretation of χ2 from contingency tables, and the calculation of P

    Journal of the Royal Statistical Society 85:87–94.

    https://doi.org/10.2307/2340521

    • Google Scholar
15. 1. Gardner PP
    2. Daub J
    3. Tate J
    4. Moore BL
    5. Osuch IH
    6. Griffiths-Jones S
    7. Finn RD
    8. Nawrocki EP
    9. Kolbe DL
    10. Eddy SR
    11. Bateman A

    (2011) Rfam: Wikipedia, clans and the "decimal" release

    Nucleic Acids Research 39:D141–D145.

    https://doi.org/10.1093/nar/gkq1129

    • Google Scholar
16. 1. Giannoukos G
    2. Ciulla DM
    3. Huang K
    4. Haas BJ
    5. Izard J
    6. Levin JZ
    7. Livny J
    8. Earl AM
    9. Gevers D
    10. Ward DV
    11. Nusbaum C
    12. Birren BW
    13. Gnirke A

    (2012) Efficient and robust RNA-seq process for cultured bacteria and complex community transcriptomes

    Genome Biology 13:R23..

    https://doi.org/10.1186/gb-2012-13-3-r23

    • Google Scholar
17. 1. Goodman DB
    2. Church GM
    3. Kosuri S

    (2013) Causes and effects of N-terminal codon bias in bacterial genes

    Science 342:475–479.

    https://doi.org/10.1126/science.1241934

    • Google Scholar
18. 1. Hoeppner MP
    2. Gardner PP
    3. Poole AM

    (2012) Comparative analysis of RNA families reveals distinct repertoires for each domain of life

    PLoS Computational Biology 8:e1002752.

    https://doi.org/10.1371/journal.pcbi.1002752

    • Google Scholar
19. 1. Ikemura T

    (1981) Correlation between the abundance of Escherichia coli transfer RNAs and the occurrence of the respective codons in its protein genes: a proposal for a synonymous codon choice that is optimal for the E. coli translational system

    Journal of Molecular Biology 151:389–409.

    https://doi.org/10.1016/0022-2836(81)90003-6

    • Google Scholar
20. 1. Ioannidis JP
    2. Khoury MJ

    (2011) Improving validation practices in "omics" research

    Science 334:1230–1232.

    https://doi.org/10.1126/science.1211811

    • Google Scholar
21. 1. Kudla G
    2. Murray AW
    3. Tollervey D
    4. Plotkin JB

    (2009) Coding-sequence determinants of gene expression in Escherichia coli

    Science 324:255–258.

    https://doi.org/10.1126/science.1170160

    • Google Scholar
22. 1. Kwon T
    2. Huse HK
    3. Vogel C
    4. Whiteley M
    5. Marcotte EM

    (2014) Protein-to-mRNA ratios are conserved between Pseudomonas aeruginosa strains

    Journal of Proteome Research 13:2370–2380.

    https://doi.org/10.1021/pr4011684

    • Google Scholar
23. 1. Lao PJ
    2. Forsdyke DR

    (2000) Thermophilic bacteria strictly obey Szybalski's transcription direction rule and politely purine-load RNAs with both adenine and guanine

    Genome Research 10:228–236.

    https://doi.org/10.1101/gr.10.2.228

    • Google Scholar
24. 1. Laurent JM
    2. Vogel C
    3. Kwon T
    4. Craig SA
    5. Boutz DR
    6. Huse HK
    7. Nozue K
    8. Walia H
    9. Whiteley M
    10. Ronald PC
    11. Marcotte EM

    (2010) Protein abundances are more conserved than mRNA abundances across diverse taxa

    Proteomics 10:4209–4212.

    https://doi.org/10.1002/pmic.201000327

    • Google Scholar
25. 1. Li GW
    2. Oh E
    3. Weissman JS

    (2012) The anti-Shine-Dalgarno sequence drives translational pausing and codon choice in bacteria

    Nature 484:538–541.

    https://doi.org/10.1038/nature10965

    • Google Scholar
26. 1. Lindgreen S
    2. Umu SU
    3. Lai AS
    4. Eldai H
    5. Liu W
    6. McGimpsey S
    7. Wheeler NE
    8. Biggs PJ
    9. Thomson NR
    10. Barquist L
    11. Poole AM
    12. Gardner PP

    (2014) Robust identification of noncoding RNA from transcriptomes requires phylogenetically-informed sampling

    PLoS Computational Biology 10:e1003907.

    https://doi.org/10.1371/journal.pcbi.1003907

    • Google Scholar
27. 1. Lorenz R
    2. Bernhart SH
    3. Höner Zu Siederdissen C
    4. Tafer H
    5. Flamm C
    6. Stadler PF
    7. Hofacker IL

    (2011) ViennaRNA Package 2.0

    Algorithms for Molecular Biology 6:26.

    https://doi.org/10.1186/1748-7188-6-26

    • Google Scholar
28. 1. Lu P
    2. Vogel C
    3. Wang R
    4. Yao X
    5. Marcotte EM

    (2007) Absolute protein expression profiling estimates the relative contributions of transcriptional and translational regulation

    Nature Biotechnology 25:117–124.

    https://doi.org/10.1038/nbt1270

    • Google Scholar
29. 1. Maier T
    2. Schmidt A
    3. Güell M
    4. Kühner S
    5. Gavin AC
    6. Aebersold R
    7. Serrano L

    (2011) Quantification of mRNA and protein and integration with protein turnover in a bacterium

    Molecular Systems Biology 7:511.

    https://doi.org/10.1038/msb.2011.38

    • Google Scholar
30. 1. Mückstein U
    2. Tafer H
    3. Hackermüller J
    4. Bernhart SH
    5. Stadler PF
    6. Hofacker IL

    (2006) Thermodynamics of RNA-RNA binding

    Bioinformatics 22:1177–1182.

    https://doi.org/10.1093/bioinformatics/btl024

    • Google Scholar
31. 1. Nakahigashi K
    2. Takai Y
    3. Shiwa Y
    4. Wada M
    5. Honma M
    6. Yoshikawa H
    7. Tomita M
    8. Kanai A
    9. Mori H

    (2014) Effect of codon adaptation on codon-level and gene-level translation efficiency in vivo

    BMC Genomics 15:1115.

    https://doi.org/10.1186/1471-2164-15-1115

    • Google Scholar
32. 1. Nawrocki EP
    2. Burge SW
    3. Bateman A
    4. Daub J
    5. Eberhardt RY
    6. Eddy SR
    7. Floden EW
    8. Gardner PP
    9. Jones TA
    10. Tate J
    11. Finn RD

    (2015) Rfam 12.0: updates to the RNA families database

    Nucleic Acids Research 43:D130–D137.

    https://doi.org/10.1093/nar/gku1063

    • Google Scholar
33. 1. Nawrocki EP
    2. Eddy SR

    (2013) Infernal 1.1: 100-fold faster RNA homology searches

    Bioinformatics 29:2933–2935.

    https://doi.org/10.1093/bioinformatics/btt509

    • Google Scholar
34. 1. Pain A
    2. Ott A
    3. Amine H
    4. Rochat T
    5. Bouloc P
    6. Gautheret D

    (2015) An assessment of bacterial small RNA target prediction programs

    RNA Biology 12:509–513.

    https://doi.org/10.1080/15476286.2015.1020269

    • Google Scholar
35. 1. Pelletier J
    2. Sonenberg N

    (1987) The involvement of mRNA secondary structure in protein synthesis

    Biochemistry and Cell Biology 65:576–581.

    https://doi.org/10.1139/o87-074

    • Google Scholar
36. 1. Plotkin JB
    2. Kudla G

    (2011) Synonymous but not the same: the causes and consequences of codon bias

    Nature Reviews Genetics 12:32–42.

    https://doi.org/10.1038/nrg2899

    • Google Scholar
37. 1. Rozov A
    2. Demeshkina N
    3. Westhof E
    4. Yusupov M
    5. Yusupova G

    (2015) Structural insights into the translational infidelity mechanism

    Nature Communications 6: 7251.

    https://doi.org/10.1038/ncomms8251

    • Google Scholar
38. 1. Schmittgen TD
    2. Livak KJ

    (2008) Analyzing real-time PCR data by the comparative C(T) method

    Nature Protocols 3:1101–1108.

    https://doi.org/10.1038/nprot.2008.73

    • Google Scholar
39. 1. Sharma CM
    2. Hoffmann S
    3. Darfeuille F
    4. Reignier J
    5. Findeiss S
    6. Sittka A
    7. Chabas S
    8. Reiche K
    9. Hackermüller J
    10. Reinhardt R
    11. Stadler PF
    12. Vogel J

    (2010) The primary transcriptome of the major human pathogen Helicobacter pylori

    Nature 464:250–255.

    https://doi.org/10.1038/nature08756

    • Google Scholar
40. 1. Sharp PM
    2. Li WH

    (1987) The codon Adaptation Index--a measure of directional synonymous codon usage bias, and its potential applications

    Nucleic Acids Research 15:1281–1295.

    https://doi.org/10.1093/nar/15.3.1281

    • Google Scholar
41. 1. Stark A
    2. Brennecke J
    3. Bushati N
    4. Russell RB
    5. Cohen SM

    (2005) Animal MicroRNAs confer robustness to gene expression and have a significant impact on 3'UTR evolution

    Cell 123:1133–1146.

    https://doi.org/10.1016/j.cell.2005.11.023

    • Google Scholar
42. 1. Storz G
    2. Vogel J
    3. Wassarman KM

    (2011) Regulation by small RNAs in bacteria: expanding frontiers

    Molecular Cell 43:880–891.

    https://doi.org/10.1016/j.molcel.2011.08.022

    • Google Scholar
43. 1. Taniguchi Y
    2. Choi PJ
    3. Li GW
    4. Chen H
    5. Babu M
    6. Hearn J
    7. Emili A
    8. Xie XS
    9. Li G

    (2010) Quantifying E. coli proteome and transcriptome with single-molecule sensitivity in single cells

    Science 329:533–538.

    https://doi.org/10.1126/science.1188308

    • Google Scholar
44. 1. Tuller T
    2. Zur H

    (2015) Multiple roles of the coding sequence 5' end in gene expression regulation

    Nucleic Acids Research 43:13–28.

    https://doi.org/10.1093/nar/gku1313

    • Google Scholar
45. 1. van Dongen S
    2. Abreu-Goodger C
    3. Enright AJ

    (2008) Detecting microRNA binding and siRNA off-target effects from expression data

    Nature Methods 5:1023–1025.

    https://doi.org/10.1038/nmeth.1267

    • Google Scholar
46. 1. Vivancos AP
    2. Güell M
    3. Dohm JC
    4. Serrano L
    5. Himmelbauer H

    (2010) Strand-specific deep sequencing of the transcriptome

    Genome Research 20:989–999.

    https://doi.org/10.1101/gr.094318.109

    • Google Scholar
47. 1. Vogel C
    2. Marcotte EM

    (2012) Insights into the regulation of protein abundance from proteomic and transcriptomic analyses

    Nature Reviews Genetics 13:227–232.

    https://doi.org/10.1038/nrg3185

    • Google Scholar
48. 1. Waters LS
    2. Storz G

    (2009) Regulatory RNAs in bacteria

    Cell 136:615–628.

    https://doi.org/10.1016/j.cell.2009.01.043

    • Google Scholar
49. 1. Woolstenhulme CJ
    2. Guydosh NR
    3. Green R
    4. Buskirk AR

    (2015) High-precision analysis of translational pausing by ribosome profiling in bacteria lacking EFP

    Cell Reports 11:13–21.

    https://doi.org/10.1016/j.celrep.2015.03.014

    • Google Scholar
50. 1. Workman C
    2. Krogh A

    (1999) No evidence that mRNAs have lower folding free energies than random sequences with the same dinucleotide distribution

    Nucleic Acids Research 27:4816–4822.

    https://doi.org/10.1093/nar/27.24.4816

    • Google Scholar
51. 1. Wu D
    2. Jospin G
    3. Eisen JA

    (2013) Systematic identification of gene families for use as "markers" for phylogenetic and phylogeny-driven ecological studies of bacteria and archaea and their major subgroups

    PLoS One 8:e77033.

    https://doi.org/10.1371/journal.pone.0077033

    • Google Scholar

Author details

1. Sinan Uğur Umu

   1. School of Biological Sciences, University of Canterbury, Christchurch, New Zealand
   2. Biomolecular Interaction Centre, University of Canterbury, Christchurch, New Zealand

   Contribution

   SUU, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article, Contributed unpublished essential data or reagents

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0001-8081-7819

2. Anthony M Poole

   1. School of Biological Sciences, University of Canterbury, Christchurch, New Zealand
   2. Biomolecular Interaction Centre, University of Canterbury, Christchurch, New Zealand

   Contribution

   AMP, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

3. Renwick CJ Dobson

   1. School of Biological Sciences, University of Canterbury, Christchurch, New Zealand
   2. Biomolecular Interaction Centre, University of Canterbury, Christchurch, New Zealand
   3. Department of Biochemistry and Molecular Biology, University of Melbourne, Parkville, Australia

   Contribution

   RCJD, Conception and design, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

4. Paul P Gardner

   1. School of Biological Sciences, University of Canterbury, Christchurch, New Zealand
   2. Biomolecular Interaction Centre, University of Canterbury, Christchurch, New Zealand
   3. BioProtection Research Centre, University of Canterbury, Christchurch, New Zealand

   Contribution

   PPG, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article, Contributed unpublished essential data or reagents

   For correspondence

   paul.gardner@canterbury.ac.nz

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0002-7808-1213

• 2,679

  views

• 363

  downloads

• 25

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.