The regulation of translation is critical in almost every aspect of gene expression. Nonetheless, the ribosome is historically viewed as a passive player in this process. However, evidence is accumulating to suggest that variations in the ribosome can have an important influence on which mRNAs are translated. Scope for variation is provided via multiple avenues, including heterogeneity at the level of both ribosomal proteins and ribosomal RNAs and their covalent modifications. Together, these variations provide the potential for hundreds, if not thousands, of flavours of ribosome, each of which could have idiosyncratic preferences for the translation of certain messenger RNAs. Indeed, perturbations to this heterogeneity appear to affect specific subsets of transcripts and manifest as cell-type-specific diseases. This review provides a historical perspective of the ribosomal code hypothesis, before outlining the various sources of heterogeneity, their regulation and functional consequences for the cell.
At almost every layer of gene expression, a plethora of regulatory mechanisms are co-ordinated to allow extensive, yet nuanced, changes to the proteome that orchestrate a range of cellular processes including development [1,2] and cell survival during stress [3,4]. The contribution made by each phase of gene expression in dictating protein levels has proved somewhat controversial [5,6]. Many estimates suggest that the control of messenger RNA (mRNA) translation can make a considerable contribution to proteomic variation [7,8]. Numerous conserved translational regulatory mechanisms, many affecting the initiation step, represent a compelling explanation for these figures (reviewed in refs [9–12]). More specifically, a range of translation factors and RNA-binding proteins are thought to impact directly or indirectly on the recruitment and activity of ribosomes. Nonetheless, what has been remarkable is the apparent passive role of the ribosome in this process. The ribosome has typically been perceived as a constitutively active, compositionally uniform molecular machine . However, more recent evidence, which will be described in this review, challenges this notion supporting potential regulatory roles for the ribosome.
Intriguingly, at the birth of molecular biology, Crick  speculated on a possible inconstant ribosome pool, even contemplating that each protein could be translated by a unique ribosome, with a unique ribosomal RNA (rRNA) sequence. Subsequently, the finding that compositional differences existed in bacterial ribosomes under different nutrient conditions supported such a view . However, rRNA size was found to be constant across ribosomes , posing the question of how a seemingly uniform rRNA could encode for unique, differentially sized proteins. In addition, an unknown RNA species necessary for protein production was identified; it possessed high turnover rates and was strikingly similar to DNA protein-coding sequences [17,18]. This protein-encoding RNA species was ultimately termed the mRNA , and as a result, the ‘one gene, one ribosome, one protein’  view shifted, instead attributing all heterogeneity in protein sequence to this mRNA . Numerous other observations, such as the fact that both the rRNA sequence  and the structure of the ribosome itself  are remarkably well conserved across three kingdoms of life, have since cemented the vision of a ribosome as a constitutive, highly conserved, homogeneous machine and this remains the predominant textbook view.
The prevalence of this relatively static view of the ribosome may be, in part, due to its complexity. For many years, the ribosome was essentially viewed as a black box , necessitating decades of scrutiny in order to uncover its structure and mechanism (reviewed in ref. ). To this date, the complexity of the ribosome continues to hamper its study. For instance, little is known regarding the degree of sequence variation within the multiple repeats of the rDNA locus, due to their size. Even the most advanced long-read sequencing platforms, which are currently capable of generating single reads spanning ∼900 kb and assembling contigs as long as 3 mb , are unable to shed light on the variation across these repeating units. Thus, despite inklings of functional variance within these rDNA regions across a range of organisms and tissues [24–26], these genes have remained elusive and largely unmapped on most genome databases.
Nonetheless, evidence suggesting that the protein component of ribosomes may be specialised has been accruing for many years, spurred on by the finding that several organisms, particularly plants and fungi, possess multiple paralogues of r-protein-coding genes [27–29]. For example, in Arabidopsis thaliana, all r-proteins are encoded by at least two paralogous genes , whereas in Saccharomyces cerevisiae, 59 of the 79 ribosomal proteins have retained their paralogous gene [31–33]. In terms of the current yeast genome, it represents an evolutionary vestige reached by the gradual loss of duplicated genes after a whole genome duplication event some 200 million years ago . As a consequence, the vast majority of yeast genes are single copy, and even where duplicated genes have been maintained, they have often diverged considerably [34,35]. Thus, the retention of ∼75% of ribosomal protein paralogues is suggestive that these paralogues may possess specific functions . Indeed, yeast strains carrying individual deletions of ribosomal protein genes often exhibited distinct phenotypes to the matched strain bearing the paralogous deletion. Such differential phenotypes were observed across a whole range of yeast biological processes including propagation of the M1 satellite dsRNA of the L-A virus , bud-site selection , secretion of carboxypeptidase Y , sensitivity to killer toxin , telomeric control , ASH1 mRNA localisation , the configuration of actin , replicative life span [43,44], Ty1 retrotransposon mobility [45,46], adaption to distinct environments  and mitochondrial mRNA translation .
Consequently, this ‘ribosomal code’ hypothesis, whereby functionally specific, compositionally heterogeneous ribosomes exist, that are responsible for the translation of unique subsets of mRNA (Figure 1) [41,47,49–54], has resurfaced. While alternative hypotheses to specialised ribosomes should be noted , evidence suggestive of a ribosomal code has been gradually accumulating over many years. The remainder of this review follows the evidence for functionally diverse ribosomes and for the regulation of translation as well as detailing potential alternative hypotheses.
Sources and functional consequences of ribosomal heterogeneity.
Ribosomal protein variation
The role of the protein component of the ribosome can be both structural and regulatory. For example, ribosomal proteins promote the correct folding of rRNA during ribosome biogenesis and in the complete ribosome . Moreover, these proteins interact with numerous factors to facilitate ribosome recruitment to appropriately processed mRNA [57,58], stalled ribosome rescue  and ribosome recycling . As the non-catalytic component of the ribosome , ribosomal proteins represent an obvious point at which ribosomal content could be modulated. Specifically, ribosomal proteins could vary in the incorporation of paralogous ribosomal proteins, with unique amino acid compositions or even the presence/absence of specific proteins altogether (Figure 1).
Several organisms possess paralogous genes for many ribosomal proteins, as outlined above. The presence of multiple ribosomal proteins, each with discrete amino acid composition, could represent a vast source of heterogeneity. As stated above, in S. cerevisiae, 59 of 79 ribosomal protein genes have retained a paralogous gene, and of these 59, 38 encode some form of amino acid sequence divergence, which could impart specific function. The presence of 38 paralogous gene sets which encode proteins with specific biochemical properties or functions would confer 238 (2.75 × 1011) unique combinations of ribosomal proteins, providing the yeast cell with an extraordinary additional potential complexity to form diverse and specialised ribosomes.
In mammals, ribosomal protein gene paralogues are typically far less frequent and only a small subset exhibit dissimilarities in the encoded amino acid sequence. Here, structural insights have provided crucial evidence in teasing out how these paralogues may exert their specific functions. For instance, human RPS4 (uS4) possesses three paralogous genes, each encoding distinct residues, namely RPS4X, RPS4Y1 and RPS4Y2, so-called because of their genomic loci . Interestingly, while RPS4X and RPS4Y1 proteins are expressed in the majority of tissues, RPS4Y2 is solely expressed in the testis and prostate gland. Here, models suggest that incorporation of RPS4Y2 into the mature ribosome facilitates the generation of a novel RPS4 conformation, reliant upon the unique hydrogen bonding arrangement afforded by the 12 amino acid alterations specific to RPS4Y2 . This novel conformation may be vital in recruiting specific mRNAs, or protein cofactors that allow RPS4Y2-containing ribosomes to augment the translation of specific mRNAs. However, most ribosomal proteins in mammals are encoded by a single gene, so it appears, at least superficially, that the scope for ribosomal variation is much less here than for organisms such as S. cerevisiae and A. thaliana. Although an additional consideration is that compositionally heterogenous ribosomes could also arise via the complete exclusion, or inclusion of specific ribosomal proteins [50,64].
Early connotations of this mechanism were found when mutations in the RPS19 gene (eS19) were shown to underlie Diamond–Blackfan anemia (DBA) , a disease characterised by a congenital erythroid aplasia . Since this early work, numerous other ribosomal proteins have been implicated in the disease, including RPS26 (eS26) . RPS26 is located in the mRNA exit channel on the ribosome , where it promotes the translation of Kozak sequence containing mRNAs . Perhaps then, alterations in the translation of these mRNAs upon mutation of RPS26 are underpinning DBA. Interestingly, other so-called ribosomopathies have also been identified, where mutations in single ribosomal protein genes appear to result in a range of diseases, each with diagnostic aetiologies .
How could mutations in a single ribosomal protein gene affect predominantly on a specific cell type and not simply impair organism-wide translation? One theory is that specific mRNAs involved in the differentiation of the impacted cell lineages are preferentially translated by ribosomes containing the necessary ribosomal proteins . Seminal work by Maria Barna supports this theory, providing evidence for ribosomal protein heterogeneity and offering mechanistic insights into how these specialised ribosomes elicit their effects. Specifically, through the use of a large-scale screen, Kondrashov et al.  identify cell- and tissue-specific defects in mice upon mutation of the large ribosomal subunit gene, RPL38 (eL38). The RPL38 protein is specific to eukaryotic ribosomes (absent in prokaryotes) and had previously been shown to display a tissue-specific pattern of expression . Mice with defective RPL38 show stark skeletal abnormalities, a phenotype that was later linked to the role of RPL38 protein in enhancing the translation of homeobox (Hox) mRNAs (Figure 1) , a group of genes that regulate a plethora of developmental programmes . In this context, RPL38 is recruited to Hox mRNAs via internal ribosome entry site (IRES) elements within their 5′-UTR . Together, the evidence from studies on RPS26 (eS26) and RPL38 (eL38) demonstrates direct mechanisms, through which compositionally heterogeneous ribosomes may mediate translational regulation, either through specific nucleotide sequences or structured mRNA segments [69,73].
Subsequent studies have used selected reaction monitoring mass spectrometry to quantitate ribosomal protein stoichiometry in an unbiased manner . In doing so, it is now known that ribosomal heterogeneity is prevalent in cytoplasmic ribosomes [50,64]. Furthermore, assessing the mRNAs bound to ribosomes containing sub-stoichiometric ribosomal proteins revealed that the heterogeneous ribosomes produced are functionally specialised . For example, RPL10A (uL16) is present at sub-stoichiometric levels in endothelial stem cells, enhancing the translation of numerous proteins involved in vitamin B12 synthesis . These observations show that the loss/presence of particular ribosomal protein subunits can provide specificity, and expand the repertoire of translational control mechanisms available to the cell.
Despite the abundance of proteins in the ribosome, generating some 1.7 MDa of the ribosome molecular mass, the catalytic unit is, in fact, rRNA [61,74]. In eukaryotes, the 15–18S, 5.8S and 25–28S rRNA that come together with 5S rRNA to form the ribosome are encoded on the genome as tandem repeating rDNA genes . In humans, these repeating units are present with an average of ∼300 repeats [76,77] spanning up to 6 mb . Archetypally, these rDNA sequences are thought to display particularly low sequence variation .
In the most recent genome sequencing studies, however, marked variability between individuals and across tissues in rDNA sequences are beginning to be recognised. By analysing >2000 whole genomes with a novel read alignment algorithm, Parks et al.  found substantial variation in rDNA sequences. Importantly, such variation appears to be functionally significant, as the precise sites of this heterogeneity are linked to established functional regions on the ribosome, such as eIF3 contacting residues, ribosome-bridging regions and the RPL38-binding pocket . In light of the known role of RPL38 in ribosomal specificity shown above, this finding offers an attractive mechanism by which cells may selectively modulate RPL38 incorporation, by modulating the expression of rRNA isoforms. In this case, differential rRNA incorporation may facilitate the formation of ribosomes with heterogeneous ribosomal protein composition.
Changes in rRNA sequence may also affect mRNA translation in a more direct manner, through base-pairing directly with specific mRNAs. Such a mechanism is reminiscent of the Shine-Dalgarno (SD) sequence in prokaryotes, whereby 3–9 nucleotide (nt), purine-rich mRNA sequences upstream of the start-codon hybridise with a complementary 9 nt sequence in the prokaryotic 16S rRNA [80,81]. The resultant tethering of the small ribosomal subunit to the mRNA dramatically enhances translation . Interestingly, while this rRNA–mRNA base-pairing mechanism in prokaryotes is common , similar examples of are comparatively rare in eukaryotes and often are utilised by viral transcripts [84–87]. Nonetheless, instances of eukaryotic mRNA recruiting ribosomes via rRNA are evident. For instance, the 18S rRNA is known to directly modulate translation of specific transcripts by base-pairing with a complementary 9-nucleotide (nt) element in the 5′-UTR sequences of Gtx domain-containing mRNAs (Figure 1) [88,89]. Furthermore, recent structural data have revealed that 18S rRNA contacts a specific nucleotide sequence slightly downstream of the start codon in the Histone H4 mRNA, which facilitates start-codon recognition and ribosome recruitment . Also, recent in silico work studying zebrafish suggests that the expression of sequence-specific maternal 5S rRNA may promote preferential ribosome binding to maternal mRNAs .
The above evidence suggests that compositionally heterogeneous ribosomes exist within the cell. However, further scope for ribosome variation is provided by modifications to the rRNA or ribosomal protein content. Roles for such modification and therefore for specialised ribosomes have been suggested in highly dynamic processes, such as adaptation to growth conditions [15,48], stress responses  and developmental processes . Indeed, post-translational modification of RPS6 (eS6) via mTOR in response to nutrient signalling is well documented (Figure 1)  and appears to affect upon key stress–response mechanisms, such as stress granule formation .
Post-translational modifications to ribosomal proteins would provide the necessary plasticity for the cell to reversibly modulate its ribosome content in response to environmental cues. For instance, interferon gamma signalling results in the phosphorylation and subsequent loss of all detectable RPL13A (uL13) from 60S ribosomes . Other examples of post-translational modifications to ribosomal proteins are less extreme, providing functional specificity to the ribosome, without resulting in the removal of any ribosomal proteins. For example, dichotomous methylation of RPS24 (eS24), or RPS31 (eS31) in Dictyostelium discoideum, modulates the ability of the ribosome to bind r-protein mRNAs, thus modulating cell growth. Similarly, Simsek et al.  identified the presence of UFMylation on human ribosomes, a metazoan-specific post-transcriptional modification, similar to ubiquitin, which could modulate ribosome interaction networks, providing specificity through unique protein partners.
Typically, modifications to rRNA are much more common place: 2% of rRNA bases are modified in some way . Canonical examples of rRNA modification include pseudouridylation and 2′-O-methylation [97–99], although numerous different types of rRNA nucleotide modifications have been identified . These modifications effectively broaden the physico-chemical potential of rRNA, enabling specific interactions that cannot be fulfilled by regular nucleotides [100,101]. For example, N6 methylation of adenine (m6A)  blocks Watson–Crick base-pairing via steric hindrance [103,104], but concomitantly increases base stacking and can promote binding to hydrophobic regions found on particular proteins . Alternatively, many modifications can enhance hydrogen binding, effectively increasing RNA rigidity and stabilising secondary structure [100,106]. RNA modifications are thus capable of subtly altering structure to facilitate recruitment of particular proteins or mRNAs through specific interactions. Such effects could impact upon communication across the ribosome transferring information from external ribosomal proteins to the decoding centre .
Typically, rRNA modifications are not evenly distributed across the rRNA, but rather they are concentrated at regions of interest, such as the peptidyl transferase centre and intersubunit interface [101,108]. Importantly, this pattern of clustering is markedly conserved across a range of organisms , suggestive that these modifications may co-ordinate ribosomal activity [110,111]. However, many rRNA modifications are vital for appropriate ribosome biogenesis  and as such, their capacity to be variably modulated would seem minor at best.
Despite these concerns, recent structural studies using electron cryomicroscopy, RiboMethSeq and mass spectrometry have identified numerous novel rRNA modifications that are also highly dynamic, sub-stoichiometric and potentially cell-type-specific [101,112–114]. Cell-specific rRNA modifications could perhaps enable specialised translational programmes, as genetic studies demonstrate that the loss of particular rRNA modifications only affects specific subsets of mRNAs, without any discernable effects on ribosome biogenesis. For instance, alterations in dynamic rRNA pseudouridylation and methylation result in the translational down-regulation of IRES-containing mRNAs, with no effect on cap-dependent translation [113–116]. Similar mechanisms may also be employed in response to environmental changes, as mutations in Rcm1, the enzyme, that catalyses 25S rRNA C2278 methylation inhibit the recruitment of specific transcripts into polysomes following oxidative stress (Figure 1) .
Many of the studies detailed above rely on the use of genetic approaches to observe the impact of deleting single genes. While the deletion of specific genes is often associated with idiosyncratic phenotypes, little work has been performed to demonstrate that these observations are specifically due to alterations in translation, or ribosome structure, particularly in vivo. As such, it is possible that alternative hypotheses could explain these observations.
A common problem associated with deletions in ribosomal protein genes is the acquisition of spontaneous suppressor mutations, especially across deletion collections . Furthermore, a specific deletion could affect expression of upstream or downstream genes , and may affect the expression of snoRNAs , which can be housed in introns present in the ribosomal protein genes [121,122]. All of these side effects of a deletion mutation could potentially confound findings. For instance, much of the phenotypic variation identified in paralogue-specific yeast mutants can be attributed to variation in the expression level of the genes, or their internally housed snoRNA genes . In addition, many paralogous ribosomal proteins are identical in amino acid sequence. In this scenario, it is difficult to envisage how incorporation of either ribosomal protein paralogue confers specificity to the resulting ribosome. Nonetheless, recent data suggest that, even for paralogues where this is the case, ribosome specificity can still be plausible .
Another confounding aspect is that specific ribosomal proteins may have so-called extra-ribosomal functions, outside of the ribosome and protein synthesis . A classic example of an extra-ribosomal function is the autoregulation of ribosomal protein-encoding mRNAs by their cognate ribosomal proteins . In S. cerevisiae, examples of this mechanism are widespread; RPL30 (eL30) , RPL2 (uL2) , RPS28 (eS28) [127,128] and RPS14 (uS11)  all modulate processing, and thus translation, of their mRNA. Importantly, such extra-ribosomal functions are not specific to yeast. For instance, in humans, human RPL5 (uL18) binds to MDM2–p53 complexes under conditions of stress . In this case then, RPL5 effectively functions as a tumour suppressor, inhibiting p53 degradation and cell-cycle progression . In such instances, it is important to distinguish whether the effects of mutations, or deletions to these ribosomal proteins, are a consequence of these extra-ribosomal functions.
Finally, it is possible that the tissue-specific defects observed in a range of ribosomal mutations or deletions [50,67,70] are not a consequence of the loss of specialised ribosomes. Rather, these differences could be due to tissue-specific differences in compensatory processes that either buffer the effect of a mutation or deletion or regulate ribosome quality control . An example where ribosome quality control impacts upon a ribosomal protein mutant has been found in a model system of DBA where RPS19 (eS19) is depleted . Here levels of Pelota, a protein that facilities ribosome recycling and stalled ribosome rescue [133,134], correlate with erythroid cell differentiation state and overexpression of this protein rescues translation defects of the mutant, thus, indicating that cell sensitivity to ribosomal defects may underpin this disease rather than the loss of specialised ribosomes.
Translation of mRNA to generate protein is a pervasive cellular process, generating some 13 000 proteins per second in S. cerevisiae . To facilitate this profuse rate of protein synthesis, the ribosome is present at ∼200 000 copies per cell in yeast [136,137] and 2–3 000 000 copies per cell in human cells [138,139]. Considering the ubiquity of translation and the number of ribosomes required, mRNA translation is an energetically demanding process and is consequently highly regulated [9,140]. This stringent regulation underpins a wide range of processes such as development , dendrite plasticity  and metabolism [143,144]. Furthermore, under conditions of stress, translation initiation is commonly rapidly inhibited for a major portion of transcripts , highlighting the importance of such regulatory mechanisms for cell survival. Interestingly, the majority of this translational regulation is classically thought to be imparted at the translation initiation step through a range of auxiliary protein complexes [9,140]. This review highlights the ribosome as an emerging potential regulator of translation. Indeed, the evidence presented here highlights the possibility that cells utilise specific ribosomal components, or modifications to generate a plethora of unique ribosomes, each with their own characteristic functions. Furthermore, due to the abundance of the translational machinery, even nuanced changes in heterogeneity could result in the formation of relatively large populations of specialised ribosomes. However, quantitative studies examining ribosomal heterogeneity currently require large populations of cells, potentially obscuring these subtle changes that may exist at the single-cell level. Additionally, to date, little is known regarding the exact mechanisms by which the cell mobilises these specialised ribosomal programmes and to what extent this phenomenon dictates biological function.
C.B. is funded by a Wellcome Trust studentship [210002/Z/17/Z]. This project is also funded by Biotechnology and Biological Sciences Research Council (BBSRC) project grants to M.P.A. and S.J.H. [BB/K005979/1 and BB/P018270/1].
The Authors declare that there are no competing interests associated with the manuscript.