The diversity of Shine-Dalgarno sequences sheds light on the evolution of translation initiation

ABSTRACT

Shine-Dalgarno (SD) sequences, the core element of prokaryotic ribosome-binding sites, facilitate mRNA translation by base-pair interaction with the anti-SD (aSD) sequence of 16S rRNA. In contrast to this paradigm, an inspection of thousands of prokaryotic species unravels tremendous SD sequence diversity both within and between genomes, whereas aSD sequences remain largely static. The pattern has led many to suggest unidentified mechanisms for translation initiation. Here we review known translation-initiation pathways in prokaryotes. Moreover, we seek to understand the cause and consequence of SD diversity through surveying recent advances in biochemistry, genomics, and high-throughput genetics. These findings collectively show: (1) SD:aSD base pairing is beneficial but nonessential to translation initiation. (2) The 5ʹ untranslated region of mRNA evolves dynamically and correlates with organismal phylogeny and ecological niches. (3) Ribosomes have evolved distinct usage of translation-initiation pathways in different species. We propose a model portraying the SD diversity shaped by optimization of gene expression, adaptation to environments and growth demands, and the species-specific prerequisite of ribosomes to initiate translation. The model highlights the coevolution of ribosomes and mRNA features, leading to functional customization of the translation apparatus in each organism.

Introduction

Protein biosynthesis is catalysed by the ribosome, an ancient and gigantic protein-RNA complex accounting for up to 20% of the proteome and 95% of total RNA in Escherichia coli[1], [2]. Ribosomes carry out mRNA translation in four phases: initiation, elongation, termination, and recycling [3]. Translation initiation is a highly regulated process and often represents the rate-limiting step in protein production [4,5]. In bacteria and archaea, it generally proceeds through binding of the small (30S) ribosomal subunit to the ribosome-binding site (RBS) in the 5ʹ untranslated region (5ʹUTR) of mRNA (Fig. 1). There formylmethionyl-tRNA (fMet-tRNA) is positioned in the peptidyl site (P site) of the 30S subunit and forms codon-anticodon base pairing with the start codon of an open reading frame (ORF). Three initiation factors (IF1, IF2, and IF3) assist the reaction through assembly with the 30S subunit and conformational changes at various substeps. Finally, the large (50S) subunit joins and all initiation factors dissociate, leading to formation of the 70S initiation complex and transition into the elongation phase [6–10].

Figure 1.

Architecture of prokaryotic ribosome binding sites. The 3ʹ tail sequence (1530 to 1542 nt) of E. coli 16S rRNA beyond helix 45 is shown. The illustration manifests the base-pairing interaction between the SD region of mRNA and aSD (bold text) of the 16S rRNA. IF, initiation factors; ORF, open reading frame; 5ʹUTR, 5ʹ untranslated region; RBS, ribosome binding site; D_toStart, the distance between the start codon and the 3ʹ end of 16S rRNA.[21]

Discovered by Shine and Dalgarno in 1974, the prevalence of AG-rich sequences at the centre of E. coli RBSs promotes translation initiation [11,12]. This region of mRNA, later named as the Shine-Dalgarno sequence (SD), facilitates ribosome recruitment through base-pairing interaction with the CU-rich anti-Shine-Dalgarno sequence (aSD) at the 3ʹ tail of 16S rRNA (Fig. 1). Structural studies demonstrated the formation of an RNA duplex by SD:aSD base pairing at the mRNA channel of the 30S subunit [13–16]. Over the years, the molecular basis of SD:aSD-dependent translation initiation has been studied in great detail because mRNA translation in E. coli and the majority of bacteria apparently rely pre-dominantly on this pathway [17–19]. Nevertheless, other translation-initiation mechanisms also exist [3,6]. Particularly, recent surveys of prokaryotic genomes unravel a tremendous diversity of 5ʹUTR and SD sequence composition. Besides bearing an SD region capable of pairing stably with aSD (termed SD(+) mRNA), some prokaryotic transcriptomes are dominated by mRNA whose SD region lacks SD:aSD pairing capacity (termed SD(-) mRNA), as well as ‘leaderless’ (LS) mRNA lacking a 5ʹUTR. Since initiating translation of the three mRNA types involves different mechanisms, the finding suggests distinct usage of the initiation mechanisms and functional divergence of ribosomes in each species. Here we summarize known translation-initiation mechanisms in prokaryotes (Fig. 2), review 5ʹUTR diversity and translation initiation in non-model organisms, explore the utility of high-throughput technology in elucidating ribosome function, and outline a coevolution model for ribosomes and transcriptomes based on current evidence.

Figure 2.

Translation initiation mechanisms in prokaryotes. The SD regions of mRNA able or unable to form stable base pairing with the aSD in 16S rRNA are indicated as SD(+) and SD(−), respectively. Open reading frames in mRNA are shown in grey. The 70S scanning mode is shown as the representative for translational coupling

Translation-initiation mechanisms in prokaryotes

SD:aSD-dependent mode for SD(+) mRNA

SD:aSD base-pairing promotes translation initiation through stabilizing the mRNA-30S pre-initiation complex and positioning the 30S subunit right upstream of the translation start site (Fig. 1). The physical location of SD regions has been defined in two ways. Bioinformatically, comparison of RBS sequence features in E. coli (wherein an RBS was defined as the 20–30 nt region upstream of the start codon) revealed a significant enrichment of adenine and guanine at the 7–12 nt region upstream of the start codon [17,20]. Formation of SD:aSD pairing at this place aligns the start codon to the P site of the 30S subunit and keeps a roughly constant distance between the start codon and the 3ʹ terminus of 16S rRNA (termed D_toStart) [21]. Experimentally, deep sequencing of RBS variants consisting of completely random sequences found translation efficiency correlating with the abundance of adenine and guanine at the 5–15 nt upstream region [22,23]. Both results were consistent with the initial discovery of the SD region [12]. Interestingly, in silico prediction showed the majority of SD:aSD base-pairing merely 4–5 bp in E. coli [21,24]. Given limited base-pairing energy, it is thermodynamically challenging for free ribosomes to directly locate the SD region in mRNA, especially when RBSs are buried in the secondary structure. Experimental evidence showed that the presence of a single-stranded region upstream of RBSs enhanced ribosome binding and translation efficiency [25–28]. It has been proposed that the 30S subunit may land on the upstream region of RBSs in a sequence-independent manner, where it awaits for transient relaxation of structured mRNA followed by sliding into the SD region [29,30]. Consistent with this ‘standby’ model, Del Campo et al. found the 13–22 nt region upstream of the translation start site less structured than other parts of endogenous mRNA in E. coli, suggesting it as an evolved standby site to accommodate ribosomes [31]. Several studies showed that ribosomal protein (r-protein) S1, a single-stranded RNA binding protein, facilitated attachment of the 30S subunit to the standby site and helped unwind the secondary structure trapping the SD region [32–35].

SD:aSD-independent mode for SD(-) mRNA

SD(-) mRNA in this review refers to 5ʹUTR-bearing mRNA whose SD region is incapable of pairing stably with aSD. Without SD:aSD interaction, binding of the 30S subunit to SD(-) mRNA appears to proceed in a sequence-non-specific manner. Bioinformatic analysis of the sequence surrounding the start codon predicted significantly weaker secondary structure in SD(-) mRNA relative to SD(+) mRNA [36,37]. Moreover, this region in SD(-) mRNA displays a symmetrical nucleotide frequency bias, potentially disfavouring base pairing and local RNA folding. In contrast to the start codon, the flanking sequence of internal AUG codons does not show these characteristics [37,38], suggesting the lack of secondary structure as a universal feature to guide translation initiation of SD(-) mRNA at the correct start site. On the translation machinery side, r-protein bSl and IF3 were shown to support translation of SD(-) mRNA in E. coli [39]. R-protein bS1 facilitates binding of the 30S subunit to SD(-) mRNA through its high affinity with the U-rich or A/U-rich sequence commonly present in 5ʹUTR of SD(-) mRNA [40]. IF3 here is potentially involved in the selection of correct start codons at the P site [41]. Compared with the SD:aSD-dependent mode, translation initiation of SD(-) mRNA has received less attention, and our current knowledge is insufficient to explain how ribosomes efficiently distinguish the correct start codon from internal AUG aside from the cue of secondary structure [38].

Leaderless mode for LS mRNA

LS mRNA contains a short sequence (<8 nt) upstream of ORF or lacks the entire 5ʹUTR [8]. It is present abundantly in many archaea and some bacteria, in contrast to E. coli where the mRNA of 18 genes appears to lack a 5ʹUTR [42–44]. Current mechanistic understanding of LS mRNA translation initiation remains quite limited [3,45]. As the 5ʹ coding sequence does not exhibit distinctive features between SD(+), SD(-), and LS mRNA [46], the start codon located at the 5ʹ end of LS mRNA appears to be the most important translation-initiation signal [47,48]. Experimental studies showed that removal of the start codon or even the 5ʹ terminal phosphate of LS mRNA significantly compromised its translation [49,50]. Insertion of an extra sequence upstream of the start codon of LS mRNA also reduced the efficiency of translation initiation [51,52]. Both the 30S subunit and 70S ribosome of E. coli were found capable of binding to the 5ʹterminus of LS mRNA in vitro, where the 70S-LS mRNA complex was more stable [52,53]. In both cases, the base pairing between fMet-tRNA and the start codon played an important role in stabilizing the ribosome-mRNA complex. Furthermore, IF2 was shown to facilitate recognition of the start codon in LS mRNA [45].

Translational coupling in polycistronic mRNA

For polycistronic mRNA, translation initiation of an internal cistron could be enhanced by translation of its upstream cistron, a phenomenon known as translational coupling [30,54]. Translational coupling occurs in two ways. First, ribosomes translating the upstream cistron help destabilize the secondary structure blocking the RBS of the internal cistron, making it accessible to the free 30S subunits. Second, ribosomes terminating translation of the upstream cistron may be recruited to continue translation of the downstream cistron. This translation re-initiation would happen when ribosomes terminate translation, dissociate from mRNA, and diffuse to the nearby RBS of the downstream cistron. Alternatively, translation re-initiation would occur if ribosomes transiently stay associated with mRNA after termination and ‘scan’ to locate the downstream start codon [55]. Using a reconstituted in vitro translation system, Yamamoto et al. showed that the scanning mode was operated by the 70S ribosome instead of the 30S subunit and required fMet-tRNA as a stimulant [56]. The success of translation re-initiation of internal cistrons was found to depend on SD:aSD base pairing at the internal RBS, the presence of IF1 and IF3, and the distance to the stop codon of the upstream cistron (<50 nt) [57,58]. Nevertheless, a recent study showed that depletion of the ribosome recycling factor in E. coli did not increase the ribosome occupancy of the downstream cistron, suggesting the scanning mode uncommon in vivo [59].

RNA folding and SD:aSD pairing strength correlate with the translation efficiency of endogenous mRNA

Biochemical studies probe the mechanisms of translation initiation through characterizing the interaction of prokaryotic ribosomes with specific mRNA under in vitro or in vivo conditions. Bioinformatic analysis, on the other hand, inspects translation initiation from a genome-wide perspective by examining the mRNA sequence features surrounding the start codon. Relative to other parts of mRNA, the 5ʹUTR and the 5ʹ coding sequence overall show reduced folding potential in favour of the accessibility to ribosomes as described above [24,31,37,60,61]. Furthermore, the RNA folding energy surrounding the start codon correlates negatively with the translation efficiency of endogenous mRNA in E. coli, reaffirming the adverse effect of RNA secondary structure on translation initiation [62–64].

Apart from RNA folding, analysis of endogenous mRNA in E. coli surprisingly found an insignificant correlation between the translation efficiency and base-pairing complementarity of the SD region to aSD [21,63]. In silico prediction estimated the range of SD:aSD base pairing among E. coli endogenous mRNA as 2 to 9 bp, 80% of which was ≥4 bp [21,24]. Interestingly, few mutagenesis studies of mRNA reported longer SD:aSD base pairing to inhibit rather than enhancing translation [33,65–67]. These findings have led to the conclusion that intermediate SD:aSD pairing strength is optimal for mRNA translation in E. coli because longer SD:aSD base pairing may stall ribosomes at the SD region, slowing down the transition to translation elongation [21,33,65–67]. The conclusion, however, is debatable as a later kinetic study pointed out that extended SD:aSD base pairing in mRNA characterized above would also shorten the distance between the 16S rRNA 3ʹ terminus and the start codon (D_toStart; Fig. 1) [5]. As such, reduced mRNA translation in earlier studies may result from the suboptimal SD:aSD pairing location rather than excessive pairing strength. The case highlights the difficulty in interpreting the sequence-function relationship of 5ʹUTR since a mutation can affect SD:aSD pairing strength, SD:aSD pairing locations, and/or local RNA folding, all of which determine the efficiency of translation initiation. Taking such complexity into consideration, Hockenberry et al. re-investigated the contribution of SD:aSD pairing strength to translation initiation of endogenous mRNA in E. coli, Bacillus subtilis, and Caulobacter crescentus [64]. By applying a regression model to separate the influence of RNA folding, they found that in E. coli and B. subtilis intermediate-to-strong SD:aSD pairing strength resulted in a ~ 55% increase in translation efficiency relative to those with weak SD:aSD base pairing. The same analysis showed intermediate-to-strong SD:aSD pairing strength-enhancing the translation efficiency to a less extent (~30%) in C. crescentus, whose transcriptome consists of significantly fewer SD(+) mRNA than E. coli and B. subtilis [68]. Notably, in the three organisms mRNA with strong SD:aSD pairing strength was not translated more efficiently than those with intermediate pairing strength [64].

Diversity of the SD region, transcriptomes, and ribosome composition suggests distinct usage of translation initiation mechanisms in prokaryotes

Experimental evidence and the prevalence of SD(+) mRNA indicated the SD:aSD-dependent mode as the major translation initiation pathway in E. coli and B. subtilis [21,24,69]. By contrast, comparative genomics uncovered the 5ʹUTR diversity among thousands of prokaryotic species, where the proportion of SD(+) mRNA could range from 11% to over 90% [18,19,42,44,70]. In terms of gene function, mRNA encoding proteins for energy metabolism and ribosome biogenesis generally shows higher SD:aSD pairing strength [18,70]. With respect to phylogeny, though SD(+) mRNA is overall dominant in bacteria, SD(-) mRNA is present abundantly in Phylum Bacteroidetes, Chlorobi, Cyanobacteria, and LS mRNA is prevalent in Phylum Actinobacteria, Deinococcus-Thermus, Spirochaetes [44]. Archaeal transcriptomes, on the other hand, consist of mainly LS mRNA and SD(+) mRNA. Intriguingly, transcriptome composition can vary greatly even within a taxonomic group. For instance, 72% of the transcriptome of euryarchaeon Haloferax vocanii consists of LS mRNA while other members in the same phylum have around 15% LS mRNA [71]. A transcriptome comparison between Mycobacterium tuberculosis and its non-pathogenic cousin M. smegmatis found 27% of genes encoded by LS mRNA acquiring 5ʹUTR in the other species [72]. Such divergence underscores the dynamic evolution of 5ʹUTR and implicates distinct usage of translation-initiation mechanisms (Fig. 2). Natural variation of the rRNA sequences and the r-protein content also suggests functional differentiation of ribosomes in different prokaryotes [73–75].

Does the transcriptome composition of an organism reflect the functional specialization of its ribosome to translate SD(+), SD(-) and LS mRNA? Compared with E. coli, functional characterization of ribosomes and translation initiation in other prokaryotes is much limited. Here we review the evidence available for representative species showing distinct transcriptome composition. Synechocystis sp. PCC6803 is a model cyanobacterium for metabolic engineering and studying the gene regulation of photosynthetic bacteria [76]. Its 3ʹ tail of 16S rRNA differs from that of E. coli by one nucleotide (Fig. 1; the 3ʹ end residue being uracil). Though phylogenetically and ecologically distant, the genomes of S. sp. PCC6803 and E. coli show a similar GC content (48% and 51%, respectively) and a similar proportion of mRNA capable of forming ≥4 bp SD:aSD base pairing (70% and 80%, respectively) [24,77]. Inferring mRNA translation based on global protein and mRNA abundance reported a significant correlation between SD:aSD pairing strength and translation efficiency in S. sp. PCC6803 [77]. Analysis of the SD:aSD sites suggested a 3ʹ shift of its aSD core (5ʹ-UCCUUU-3ʹ; 5ʹ-CCUCCU-3ʹ in E. coli) and a wider D_toStart distribution relative to E. coli. In general, S. sp. PCC6803 mRNA with <4 bp SD:aSD base pairing has less structured 5ʹUTR than those with ≥4 bp, suggesting the usage of SD:aSD-independent and -dependent translation initiation, respectively. Ribosome profiling of S. sp. PCC6803 directly confirmed its utilization of SD:aSD-dependent translation initiation [78]. Similar to E. coli, S. sp. PCC6803 ribosomes showed an increased footprint on the AG-rich SD regions of mRNA encoding highly expressed proteins [63]. The similarity of translation initiation in E. coli and S. sp. PCC6803 is also highlighted by two experimental studies, which showed the qualitatively comparable translation efficiency of 14 out of 17 synthetic RBSs in two organisms [79,80].

By contrast, Bacteroides thetaiotaomicron, an anaerobic bacterium dominant in human gut microbiota, shows a translation-initiation profile distinct from E. coli. The genome of B. thetaiotaomicron has a slightly lower GC content (43%), and the majority of its mRNA is SD(-) [81]. Unlike E. coli, mRNA translation efficiency in Phylum Bacteroidetes correlated poorly with the SD:aSD pairing strength and appeared to be more sensitive to secondary structure [36,82]. Though the 3ʹ tail of B. thetaiotaomicron 16S rRNA (5ʹ-GAACACCUCCUUUCU-3ʹ) retains the conserved CU-rich aSD core (Fig. 1), its endogenous SD regions and RBSs are enriched for adenine and uracil but short of guanine [82]. Aiming for development of genetic toolkits, two studies examined the translation efficiency of hundreds of synthetic RBSs in B. thetaiotaomicron [83,84]. Results of both studies showed AU-rich and AG-rich RBSs to enhance and inhibit translation initiation, respectively, contrasting the SD sequence-function rule in E. coli and B. subtilis [21,24,37,85]. Similar results were reported by studies of other bacteroides [82,86].

Organisms expressing a significant amount of LS mRNA are to be addressed last but not least. Saccharolobus solfataricus, a thermophilic crenarchaeon inhabiting volcanic hot spring, produces mainly LS mRNA (69%) [87]. Although SD regions in the 5ʹUTR of the remaining mRNA showed weak base complementarity with the aSD core (3ʹ end of 16 rRNA as 5ʹ-GAUCACCUCAUA-3`) of S. solfataricus [88], bioinformatics suggested a high SD:aSD pairing capacity for the SD regions of internal cistrons in polycistronic mRNA [87,89]. Through deleting the pre-existing SD region, an in vitro study demonstrated the essentiality of SD:aSD interaction for translating the downstream cistron of a bicistronic mRNA [90]. The same study found that translation of this cistron could be restored by complete removal of the upstream mRNA segment, i.e. turning it into LS mRNA. Two later studies further showed that attachment of a SD(+) sequence to the 5ʹ end of LS mRNA directed binding of the 30S subunit to this location and enhanced translation of the leading cistron [91,92]. Together, results revealed the preference of S. solfataricus ribosomes to use the leaderless and SD:aSD-dependent modes for translation initiation. Similar to S. solfataricus, RNA-seq studies showed the transcriptomes of M. tuberculosis and M. smegmatis made of roughly 25% LS, 35% SD(+), and 40% SD(-) mRNA [93,94]. Genome-wide analysis found similar translation efficiency of LS and SD(+) mRNA but lower translation efficiency of SD(-) mRNA [72,95,96].

Current experimental evidence, despite being limited, has revealed functional differentiation of ribosomes among prokaryotes. The transcriptome composition of an organism to some extent predicts the capacity of its ribosomes to translate LS, SD(+), and SD(-) mRNA, but the relative abundance of the three mRNA types does not necessarily reflect the order of translation efficiency.

Global SD sequence-function mapping resolves debates over translation initiation in E. coli

The development of ribosome profiling, cryo-EM, and massively parallel reporter assays (MPRA) in the last decade has yielded unprecedented insights into ribosome dynamics at each stage of mRNA translation [97–99]. MPRA provides means to quantify the translational output of hundreds of thousands of synthetic mRNA variants in a single experiment. This high-throughput approach has an advantage over ribosome profiling in that it characterizes synthetic mRNA with focal features (e.g. RBS sequence) more diverse than endogenous mRNA, allowing researchers to uncover the biophysical propensity of ribosomes in an unbiased manner. MPRA has been applied to explore the sequence-function relationship of SD regions [22,23,85,100], dissect the influence of RNA secondary structure on translation initiation [22,101–104], and compare the species-specific translation efficiency of a reference RBS collection in multiple bacteria [105]. Recently, our group applied MPRA to determining the translation efficiency of all possible SD sequence variants in E. coli cells in order to quantify the functional contribution of SD:aSD base pairing, estimate the influence of RBS contexts on SD sequence-function mapping, and investigate how this mapping may impact SD evolution [85]. As depicted in the following, the experimental datasets also provide vital clues for two debates over translation initiation of E. coli in the literature: (1) SD:aSD base pairing is nonessential [37,38,66,106]; (2) extended SD:aSD base-pairing causes inhibition [21,33,65–67].

Our MPRA experiments characterized three SD variant libraries using a cellular expression of the green fluorescent protein (GFP) as the proxy for translation efficiency. Each of the SD libraries contained 262,144 (4⁹) sequence variants in the 9-nt SD region surrounded by a unique RBS context (Fig. 3A; the location of defined SD regions differed by one nucleotide, but this experimental design did not affect the results and conclusions [85]). The three RBS contexts, RBS_fepB (endogenous), RBS_arti (artificial), RBS_dmsC (endogenous), exhibit a twofold difference in the GC content (14%, 29%, 57%, respectively), corresponding to low, medium, and high potential of forming RNA secondary structure. The translation efficiency of SD variants in the three libraries showed very different distribution (Fig. 3B), underscoring the impact of RBS contexts and secondary structure in translation initiation. To examine the significance of SD:aSD base pairing, we assigned SD variants in each library to 10 groups (1–10, corresponding to low to high translation efficiency) and inspected the distribution of SD:aSD pairing strength and D_tostart in each group (Fig. 4). From Group 1 to 10, we observed increasing SD:aSD pairing strength and centralization of the D_tostart distribution. Demonstrated by all possible SD variants under three distinct RBS contexts, the significant correspondence between SD:aSD pairing strength and translation efficiency argued against inhibition of translation initiation by excessive SD:aSD base pairing.

Figure 3.

Sequence and function of the SD region in E. coli. Shown is extended analysis of a published dataset [85]. (A) The RBS contexts of three SD variant libraries. The start codon of the GFP reporter is in bold, and the 9-nt SD region is underlined. (B) Functional distribution of 4⁹ SD variants in the three RBS contexts. CA, canonical SD; 9 G, nine-guanine; 9 U, nine-uracil. (C) Functional distribution of the 1,024 5-nt motifs in the SD region under the three RBS contexts. The impact of a motif on translation initiation is computed as the mean GFP expression of all SD variants bearing this motif. The 1,024 motifs are classified and displayed in terms of the guanine content. Numbers in parentheses indicate the total amount of motifs each category. In (B) and (C), SD variants and 5-nt motifs are ranked by GFP expression and grouped into 10 equal-sized bins in the logarithmic scale. (D) Functional ranking of 5-nt motifs in the SD region under the three RBS contexts. Out of the 1,024 motifs, only the top ten and the canonical motifs are shown. Guanine residues are stressed in bold. Identical motifs present in different RBS contexts are connected by lines

Figure 4.

Distribution of SD:aSD pairing strength and D_tostart with respect to translation efficiency. Shown is extended analysis of a published dataset [85]. All SD variants in the three RBS contexts are assigned to ten groups (1–10, corresponding to low to high translation efficiency in Fig. 3B). The SD:aSD pairing strength (kcal/mol) is estimated by the ViennaRNA Package [142], and D_tostart is calculated based on a described method [21]. The distribution of these two characters is shown as density plots

Among SD variants of high translation efficiency, we discovered the unusual nine-guanine (9 G) variant not present in E. coli endogenous RBSs (Fig. 3B). The 9 G variant yielded GFP expression comparable to the canonical SD (5ʹ-UAAGGAGGU, reverse complement of aSD), likely attributed to its outstanding ability to form extended wobble pairing with aSD in multiple configurations [107]. The beneficial effects of the guanine content and SD:aSD base pairing in translation initiation were also revealed at the motif level. Examination of the average GFP expression of SD variants bearing each of the 1,024 (4⁵) 5-nt motifs found a positive trend between the former and the guanine content of a motif (Fig. 3C). Across the three RBS contexts, 10 motifs that enhanced GFP expression most all had ≥3 guanines and formed ≥4 SD:aSD base pairing (Fig. 3D). Among the top 10 motifs, seven were shared by all RBS contexts, but only two of them were canonical motifs (5ʹ-AGGAAG-3ʹ and 5ʹ-GGAGG-3ʹ, part of canonical SD). The unusual 5 G motif was always among the top 10 and ranked ahead of several canonical motifs, stressing the advantage of G:U wobble pairing in promoting SD:aSD interaction and translation initiation. On the other hand, the trend shown in Fig. 4 also called attention for the antagonism between SD:aSD pairing strength and RNA folding. Relative to RBS_arti and RBS_dmsC, SD variants in the weakly structured RBS_fepB context generally required lower SD:aSD pairing strength to achieve the same translation level. In particular, the majority of SD variants in Groups 3–5 of RBS_fepB showed detectable GFP expression but was predicted unable to pair stably with aSD. Using in vitro and in vivo assays, we confirmed one such variant, made of nine uracils (9 U) in Group 3, capable of binding to ribosomes and initiating mRNA translation at the start codon (Fig. 3B) [85]. The dispensable role of SD:aSD interaction in locating the correct translation start site is supported by a recent study which profiled the footprint of ribosomes with an altered aSD sequence [38]. Collectively, extended analysis of formerly published datasets revealed an overall positive role of SD:aSD base pairing in translation initiation and showed SD:aSD base-pairing nonessential to translation in E. coli whose transcriptome consists of predominantly SD(+) mRNA.

Functional coevolution of ribosomes and transcriptome composition

Comparative genomics revealed high conservation of aSD in 16S rRNA but significant variation in the SD region of mRNA and the transcriptome composition in prokaryotes [18,19,44,70]. Experimental studies, on the other hand, demonstrated the promiscuity of bacterial and archaeal ribosomes to translate SD(+), SD(-), and LS mRNA, though with varied efficiency [85,90,92,96,108]. These findings raise intriguing questions regarding the extent of diversification of translation-initiation mechanisms between species. Here we outline a coevolution model of the translation machinery and transcriptome composition that accounts for the mRNA diversity and the apparent rRNA and r-protein variation in extant organisms, with an emphasis on the bacterial lineage (Fig. 5). Ancient ribosomes were able to translate LS and SD(-) mRNA. Early on aSD evolved at the 3ʹ end of the small subunit rRNA as the binding site of the ancient and highly conserved Era GTPase [109–113]. Era GTPase facilitates the maturation of rRNA and ribosomes as well as coordinating protein expression with metabolism and the cell cycle. The birth of aSD represented a case of exadaptation as it fortuitously provided a new way to control translation initiation. Under mutation and selection, mRNA encoding high-demand proteins evolved SD(+) to enhance its binding to the small ribosomal subunit and assist start codon recognition. As a result, aSD engraved the ancient genome and led to an increase of SD(+) mRNA. Following the divergence from the last universal common ancestor (LUCA), bacteria and archaea evolved distinct mechanisms of transcription. For unclear reasons, perhaps resource conservation or changes in gene regulation, many archaeal promoters were positioned immediately upstream of ORF, resulting in generally more abundant LS mRNA in archaea than in bacteria [44,71]. In accord with the presence of SD(+) mRNA and experimental evidence, archaeal ribosomes retain aSD capable of pairing with SD(+) mRNA [92,114,115]. Eukaryotes, descending from an archaeal ancestor, evolved 5ʹ capped mRNA and larger ribosomes which lost aSD and adopted the scanning mode for translation initiation [116].

Figure 5.

Coevolution model of ribosomes and transcriptomes. Pie charts illustrate the relative abundance of SD(+), SD(−), and leaderless (LS) mRNA in transcriptomes. SSU, small ribosomal subunit; LUCA, last universal common ancestor; Kz, Kozak sequence

On the bacterial side, while SD(+) mRNA is dominant overall, bacterial transcriptome composition continues evolving and differentiates in each lineage. Through examining phylogenetically diverse bacteria, Hockenberry et al. identified potential physiological and environmental causes of transcriptome diversification [19]. The transcriptomes of fast-growing species tend to enrich for SD(+) mRNA, which is translated more efficiently to support the rapid growth lifestyle. SD(+) mRNA is also enriched in thermophiles, where the SD:aSD base-pairing stabilizes the 30S pre-initiation complex under high temperatures. By contrast, as the protein production rate is less significant for slow-growing bacteria occupying unique ecological niches, the percentage of SD(+) mRNA declines in their transcriptomes due to genetic drift or negative selection. Aside from these, the process of genome reduction is likely responsible for an increase of LS mRNA and loss of multiple r-protein genes in endosymbionts (including mitochondria) and bacteria with smaller genomes (≤1 Mb) [117–119]. Loss of aSD in these organisms further impairs the SD:aSD pairing function and wipes out SD(+) mRNA. Though the ribosome assumes its central role in protein production, its rRNA and r-protein contents clearly vary across prokaryotes [74,75,109,116,120–122], suggesting functional diversification of ribosomes in different species. Given the interdependence of mRNA and ribosomes, we hypothesize that the capability of ribosomes to initiate translation of SD(+), SD(-), and LS mRNA has been tailored to match up the transcriptome composition and cellular demand through a coevolution process. The ribosome of E. coli, for instance, has adapted to translate SD(+) mRNA efficiently while retaining residual capabilities to handle SD(-) and LS mRNA [37,53,85,108,123]. By contrast, cyanobacterial ribosomes evolved reduced dependence on SD:aSD base pairing to cope with the large amount of SD(-) mRNA in their transcriptome, and vice versa [37,77,78,124].

In contrast to the model presented here, an earlier review proposed the leaderless mode as the ancestral mechanism for translation initiation in LUCA under the assumption that its mRNA was structurally simple, i.e. leaderless and monocistronic [125]. The assumption was based on the prevalence of LS mRNA in archaea (thought to be more similar to LUCA than bacteria at the time) and the shared potency of bacterial, archaeal, and eukaryotic ribosomes to translate LS mRNA [126]. The earlier model postulated the SD:aSD-dependent mode to evolve after the emergence of polycistronic mRNA in order to translate internal cistrons. We consider our model more realistic as it does not assume the ancestral mRNA type, acknowledges the ancient and conserved role of Era GTPase and aSD in ribosome biogenesis, and accounts for ecological and physiological factors leading to the extant prokaryotic transcriptome diversity.

Future perspectives

Divergence of the SD regions, transcriptome composition, rRNA, and r-proteins implicate functional specialization of the translation machinery in each organism. Quoting Theodosius Dobzhansky’s momentous statement, ‘Nothing in biology makes sense except in the light of evolution.’ [127], recent advances by ribosome profiling, MPRA, and cryo-EM start to reveal the species-specific features of prokaryotic ribosomes [15,16,63,78,96,105,115,128–136]. Though experimental evidence for the coevolution of ribosomes and transcriptomes remains limited [137,138], we expect that the application of high-throughput methods in conjunction with detailed genetic and biochemical characterization will shed more light on the functional significance of rRNA and r-protein variation in mRNA translation. Moreover, we anticipate the application of single-molecule imaging techniques, as demonstrated by our and others’ work [85,139–141], will unravel the translational dynamics in living cells under different growth conditions and elucidate how the ribosome of a SD(-) mRNA- or LS mRNA-abundant organism effectively recognizes the start codon and initiates translation of the two mRNA types rare in E. coli. Life scientists are living in an exciting era surrounded by powerful research tools and ever-increasing genome sequences. To unravel the full picture of mRNA translation and genome evolution, future work should explore the diversity of translation initiation and ribosome function beyond the model systems.

Funding Statement

This work was supported by the Ministry of Science and Technology, Taiwan [105-2311-B-002-004-MY2]; Ministry of Science and Technology, Taiwan [106-2628-B-002-002-MY4]; Ministry of Science and Technology, Taiwan [109-2311-B-002-009-MY3]; National Taiwan University (TW) [109L7706]; National Taiwan University (TW) [109L7867]; National Taiwan University (TW) [109L104051].

Disclosure statement

No potential conflict of interest was reported by the authors.

Author contributions

J.D.W. and H.D.C. wrote the manuscript; S.T.K. conducted data analysis; H.D.C. generated figures.