The interaction of biological macromolecules is a fundamental attribute of cellular life. Proteins, in particular, often form stable complexes with one another. Although the importance of protein complexes is widely recognized, we still have only a very limited understanding of the mechanisms underlying their assembly within cells. In this article, we review the available evidence for one such mechanism, namely the coupling of protein complex assembly to translation at the polysome. We discuss research showing that co-translational assembly can occur in both prokaryotic and eukaryotic organisms and can have important implications for the correct functioning of the complexes that result. Co-translational assembly can occur for both homomeric and heteromeric protein complexes and for both proteins that are translated directly into the cytoplasm and those that are translated into or across membranes. Finally, we discuss the properties of proteins that are most likely to be associated with co-translational assembly.
Proteins within the cell must carry out their important functions in an environment that is highly crowded and are in constant physical contact with various other proteins, metabolites and macromolecules [1,2]. Apart from many transient interactions, which may or may not have important functional duties, e.g. cellular signalling [3,4], many if not most intracellular proteins function as subunits of more long-lived protein complexes [5,6]. Despite the fact that complex formation is crucial for understanding the function (and malfunction) of many proteins, the fundamental mechanisms behind the assembly of individual proteins into complexes with defined quaternary structure have often been neglected and little is known about the in vivo assembly process. One interesting aspect of protein complex assembly is the degree to which it is coupled to the cellular translation machinery. Are the subunits fully translated before finding their interaction partners and forming their defined quaternary structure in a post-translational assembly pathway? Or do some protein interactions form as the individual subunits are still being translated, through co-translational assembly?
The general process of the maturation of a functioning protein involves folding and translocation of the polypeptide as well as complex assembly. In recent years, there has been a growing body of evidence showing that the protein folding process can often occur while the polypeptide is being translated, i.e. co-translationally involving the nascent polypeptide chain [7–12]. Although this could arise due to the basic energetics of protein folding, there are also potential functional benefits to co-translational folding. For example, it may provide a means of tuning the potential energy landscape by lowering the energy of folding intermediates. A co-translational folding process also contributes to the earlier formation of secondary and tertiary structures making unfavourable inter-domain aggregation events less likely.
The assembly of protein complexes can, in many ways, be considered analogous to protein folding, in that it typically follows a specific pathway via energetically favourable assembly intermediates [13,14]. It is therefore natural to envisage that assembly could also sometimes occur co-translationally and even be functionally advantageous. Many protein complex subunits are highly dynamic or unstable in isolation  and so rapid assembly during translation minimizes the opportunities for misfolding or aggregation. If the subunits assemble through a series of lower energy intermediates of nascent polypeptides, it would lower the overall energy of the assembly, analogous to the tuning of folding. Co-translational protein interactions can also be viewed as a means of ensuring a precisely ordered assembly process and for avoiding unfavourable inter-subunit aggregation. Contrary to co-translational folding however, the kinetics of co-translational assembly are not only a function of the rate of assembly, but also highly dependent on the concentration of available assembly partners in close proximity to the polysome.
Although co-translational assembly has received much less attention than co-translational folding, numerous cases have been reported over the years, beginning with observations of specific proteins associating co-translationally with the cytoskeleton [16–19]. Furthermore, recent evidence suggests that the phenomenon might be widespread . Here we review several examples of co-translational assembly, discussing reasons why it occurs and the functional benefits that it can provide. Although there are many transient protein–protein interactions that occur co-translationally, e.g. involving chaperones  or targeting signal sequences for translocation , here we will focus on the assembly of stable protein complexes.
Co-translational assembly of homomers
Protein complexes can be broadly split into two categories based upon their quaternary structure: homomers, formed from the self-assembly of a single subunit type and heteromers, formed from multiple distinct subunits. There are two ways in which co-translational homomer assembly can occur. In one, a newly translated subunit is released and then interacts with another still-translating copy of itself, most probably on the same polysome from which it was translated (Figure 1A). Alternatively, co-translational assembly could involve interaction between two nascent chains on the same polysome (Figure 1B).
Co-translational assembly of protein complexes
An early example of co-translational assembly of a homomer came from investigations into the well-characterized bacterial homotetramer β-galactosidase [23,24]. The enzymatic activity of β-galactosidase is only evident after the conformational changes that are required for tetramer formation have taken place. In these experiments, it was shown that the enzymatically active form of the complex could be observed at the same time as nascent polypeptide chains. It became clear that not only the folding, but also the assembly into the functioning enzyme occurred in a co-translational manner. The authors suggested that this was due to the proximity of the nascent polypeptides, as monomers from adjacent ribosomes dimerized before forming the final tetrameric structure.
The reovirus attachment protein σ1 forms a homotrimer and can be divided into two segments: an N-terminal tail that is anchored in the virion and a globular C-terminal domain that is responsible for virion attachment (Figure 2). Curiously it was found that the trimerization of the two regions takes place using two different mechanisms . Assembly of the N-terminal region, which is translated first, was found to occur co-translationally, at neighbouring ribosomes that had passed the midpoint of the mRNA strand. In contrast, trimerization of the globular C-terminal region, which is translated last, was found to be highly chaperone- and ATP-dependent and occurs post-translationally.
Structure of the reovirus attachment protein σ1
The tumour suppressor p53 forms a homotetramer with dihedral symmetry. Although both alleles of p53 are often mutated or non-functional in cancer cells, mutations in a single allele often display a dominant-negative effect. Depending on the location of the mutation, numerous factors contribute to this effect, but a key aspect relates to the mechanism of p53 tetramer assembly and the extent to which it is coupled to translation. It was demonstrated that this process occurs by an initial co-translational dimerization of p53, with tetramers forming separately and post-translationally . The suggested driving force for this assembly mechanism was the stabilization of the dimer through hydrophobic interactions between the N-termini. A direct effect of this co-translational assembly is that the possible stoichiometries of the fully assembled complex are constrained: p53 dimers will always be homomers of either the wild-type or the mutant version of the protein. Consequently, one-fourth of the resulting homotetramers will be fully wild-type, as opposed to only one-sixteenth in a situation where co-translational dimerization does not occur. This suggests that the co-translational dimerization step has a strong influence on the magnitude of the dominant-negative effect observed.
NF-κB1 (nuclear factor of kappa light polypeptide gene enhancer in B-cells 1) is a member of the NF-κB family of transcription factors and is involved in regulation of several cellular processes, particularly the inflammatory response . The complex exists predominantly as a heterodimer of p50 and p105 subunits, with the p50 subunit being a truncated form of p105. Full-length p105 comprises an N-terminal Rel homology domain (RHD) and a larger C-terminal ankyrin-repeat domain that functions as an I-κB-like inhibitor of mature NF-κB1. Between these two domains lie a nuclear-localization signal and a glycine-rich region that acts as the site of endoproteolytic cleavage by the 26S proteasome . Active NF-κB1 requires cleavage and degradation of the C-terminal domain of p105 to form mature p50 . A key question arising from this observation is how the proteasome degrades p105 while sparing p50. Building on early observations that free p50 rapidly associates with other Rel family proteins in vitro, Lin et al.  demonstrated in vivo that p50–p105 heterodimers assemble on the same polysome via co-translational homodimerization of p50 RHDs  (Figure 3). This is coupled with co-translational processing by the proteasome; crucially, it is the process of dimerization that appears to act as a physical barrier to degradation of p50. In support of this, it was shown that deletion of the second sub-domain of RHD (essential for dimerization) led to a significant reduction in the amount of p50 observed upon expression of the mutant NF-κB1 gene. This suggests that in the absence of dimerization, p105 is completely degraded. If so, this provides a clear example of how co-translational assembly can be functionally important; in this case, co-translational assembly is essential for the production of mature NF-κB1, with the active p50 subunit being placed under immediate control of the inhibitory p105 subunit by the process. Subsequent post-translational activation then depends on phosphorylation and ubiquitin-mediated cleavage of the remaining p105 ankyrin-repeat domain.
Co-translational assembly of p50–p105 heterodimer
Co-translational assembly of heteromers
The assembly of heteromers is inherently more complex than for homomers, due to the fact that it involves interactions between distinct proteins that are usually encoded by different genes. Those interacting proteins must somehow find each other within the cell. Co-translational interaction, in which a fully translated protein finds its way to the nascent chain of another protein (Figure 1C), provides a way to minimize the stochasticity of assembly by increasing the chance of subunit encounter.
One example of a heteromeric interaction with both co- and post-translational assembly mechanisms is the covalent disulfide bond formation between heavy and light chains in the immunoglobulin molecule. Despite earlier evidence of post-translational formation of the disulfide linkers, it was shown that over-production of light chains in the endoplasmic reticulum (ER) in certain cell types lead to light-heavy chain heterodimerization on the nascent heavy chain, purely due to light chain abundance in the proximity of the translating heavy chain transcript . Thus protein expression levels and abundance are likely to be important regulators of whether or not heteromeric assembly occurs co-translationally.
The yeast histone methyltransferase, COMPASS (complex proteins associated with Set1p), is comprised of eight different subunits. In work originally designed to investigate the role of mRNA in COMPASS function, a four-member sub-complex of COMPASS (formed by the proteins Swd1p, Spp1p, Shg1p and Set1p) was found to interact with SET1 mRNA . Crucially, formation of the mRNA-associated sub-complex was found to be dependent on active translation, indicating that the subunits are binding to the nascent Set1p protein as it is translated. Furthermore, whereas structural data are not available for the full complex, it appears that the binding sites of Shg1p, Swd1p and Spp1p are localized to the N-terminal or central region of Set1p; this is consistent with co-translational assembly as these regions are translated earlier [33,34].
Previously, the first systematic analysis of co-translational assembly has been performed. Duncan and Mata  used ribonucleoprotein immunoprecipitation-microarray (RIP-Chip) experiments to identify mRNA sequences associated with 31 proteins from Schizosaccharomyces pombe . Here they found that 12 of these (38%) co-purified with the mRNAs of known interaction partners. Importantly, as for the COMPASS example, co-purification was found to be dependent on active translation, indicating that interactions are probably occurring between proteins and nascent peptides, rather than protein and mRNA. These interactions were also found to be highly specific; Cdc2p, for example, was found to co-purify with just two mRNAs, despite having a large number of known and hypothesized protein interaction partners. Interestingly, the fact that these mRNA–protein interactions are so specific has since been used by the same group to predict novel protein–protein interactions .
Co-translational assembly of secreted and membrane complexes
The above examples involve co-translational assembly within the cytoplasm, but many proteins are directly translocated into or across membranes during translation. In eukaryotes, membrane and secreted proteins are translated at the rough ER. In investigations into the assembly of the extracellular human tenascin protein, responsible for cell adhesion, it was shown that the hexameric complex is formed without any assembly intermediates being observed . As soon as the tenascin protein is experimentally detectable, it appears to be co-translationally assembled into its active hexamer structure. In this case, the authors suggested that the arrangement of the membrane-bound polysome at the ER, where the ribosomes have been seen to form various circular loops and spirals, directly resulted in the homomer acquiring its circular hexamer shape.
A further example of ER membrane influence on co-translational assembly is seen in voltage-gated potassium cation channels. These channels are tetrameric with interfaces located at the N-terminal region of the subunits [referred to as the tetramerization (T1)-domains]. In experiments using Xenopus laevis oocytes, it was shown that T1–T1 association occurred between ribosome-attached subunits and the ER membrane was postulated to regulate the local concentration of the interacting domains . In the related human ether-à-go-go related gene (hERG1), responsible for the potassium channel hERG, the two subunits hERG1a and hERG1b are isoforms of hERG1, arising from two mRNA splice-variants . The isoforms are identical apart from the important N-termini and it was observed that the two N-termini localize and bind to each other co-translationally. This mechanism is crucial to avoid unfavourable aggregation events involving the hERG1b subunits and is mediated by the ER, which ensures co-localization of the transcripts.
Finally, there is evidence that the plant D1 transmembrane protein assembles co-translationally into the photosystem II complex [39,40]. This is an interesting example as the D1 protein frequently experiences photodamage and experiences a high rate of turnover. Thus the ability of D1 to be translated directly into the chloroplast membrane and co-translationally assemble allows photosystem II to be quickly repaired. It is also notable that translational pausing is known to occur at specific sites during the translation of D1 , potentially allowing time for assembly to occur , analogous to how translational pausing can facilitate protein folding .
Here we have highlighted a number of examples of homomeric and heteromeric protein complexes that assemble co-translationally. However, we still have very little idea about the frequency of the phenomenon. One systematic analysis suggested that it might be quite widespread, yet this work considered only a very small fraction of known proteins in fission yeast . In addition, questions remain about the specific mechanisms by which co-translational interactions occur. For example, it is unclear whether binding events are limited to those occurring between one nascent and one fully-folded chain or whether dimerization ever occur while both chains are being translated. Thus, there is considerable future potential for both large- and small-scale screens looking for evidence of co-translational assembly.
Why do some protein complexes assemble co-translationally whereas others do not? Although possible functional benefits have been discussed here, it is important to remember that co-translational assembly has not necessarily been selected for evolutionarily in all cases. Co-translational assembly could occur simply because a free subunit encounters a nascent chain and their interaction is energetically favourable. In fact, for some proteins, there may be evolutionary pressure to avoid co-translational assembly. Although we only have experimental evidence of co-translational assembly for a fairly small number of complexes, we can make some predictions about which complexes might be most likely to assemble co-translationally:
All things being equal, homomers should be more likely to co-translationally assemble than heteromers, since interacting subunits can be translated from the same polysome and local subunit concentration will be high.
Many prokaryotic complexes are encoded in operons, so that interacting proteins are often translated off the same polycistronic mRNA. This ensures that interacting subunits are in close physical proximity upon translation and facilitates a higher rate of complex assembly . Thus we can predict that operon-encoded heteromers should be more likely to undergo co-translational assembly.
For both homomers and heteromers, the likelihood of co-translational assembly should be greater for highly abundant proteins, as this will increase the chance that an interaction partner encounters and binds a nascent chain still in the process of being translated.
Localization towards N-terminal regions is likely to be a general feature of interfaces that form co-translationally, since this will allow more time for co-translational assembly to occur. Therefore, complexes with N-terminal interfaces should be more likely to have formed co-translationally.
Subunits that are highly flexible or disordered in isolation  could benefit from co-translational assembly, as this would avoid them spending unnecessary time free and susceptible to proteases in the cell.
The first step of a protein complex assembly pathway is the most probable to occur co-translationally. Thus we may be able to use experimental characterization or structure-based prediction of assembly order [13,14] to identify subunits and interfaces that are most likely to form co-translationally.
Finally, there are major questions remaining about how co-translational assembly is regulated and how proteins are localized to polysome, especially for heteromers with subunits translated from different mRNA molecules. This is especially important for eukaryotic complexes, which have a much greater propensity to form heteromers [44,45], compared with bacterial proteins, which are more likely to self-assemble into homomers  or be encoded in operons. Much more work is needed to fully understand how the assembly of heteromeric complexes occurs within eukaryotic cells, both co- and post-translationally and how it is regulated.
We thank Cathy Abbott and Dinesh Soares for helpful comments on the manuscript.
This work was supported by a University of Edinburgh Chancellor's Fellowship to J.M.
Translation UK 2015: Held at the University of Aberdeen, U.K., 7–9 July 2015.