(Almost) Everything in Cotranslational Folding Makes Sense in the Light of Evolution

Main Text

Thanks to Anfinsen’s classic finding, it is now commonplace that “full-length” protein sequences, unfolded by adding denaturant, can spontaneously refold upon diluting out the denaturant. Perhaps less mainstream, although an observation almost as old as Anfinsen’s original work (see references in (1)), is that “partially synthesized” protein molecules translated on the ribosome can also begin to fold, albeit partially. The process of folding and translation occurring simultaneously is called cotranslational folding (Co-tf). Post-translational folding (Post-tf) denotes folding of the full-length protein after synthesis.

Post-tf embodies the central notion that “static information” encoded in the full-length primary sequence dictates fold and function. Co-tf raises the possibility of a new paradigm that “information production” can influence function. The following question now arises: do Co-tf and Post-tf ultimately serve the same biological purpose?

An impressive array of recent experimental and theoretical efforts provide many insights to this question. First, multiple experiments have now conclusively proven that proteins can begin to fold during translation (1). Experiments have further revealed that Co-tf can increase the yield of folded and functional proteins compared with refolding after denaturation (2). In addition to devising new techniques, experimentalists have cleverly used synonymous mutations to vary the speed of translation by inserting slow or fast-translating codons while maintaining the same amino acid sequence. The overall consensus is that translational speed can be varied to control the folding pathway/kinetics and the formation of intermediate structures, which, in turn, affect the propensity to aggregate and the activity of the protein (3, 4, 5).

A growing list of studies support the important role of Co-tf in biological function. Only a few were noted above because space here is limited. Slow-translating codons in specific locations in protein sequences have been found to be conserved across diverse species, hinting at an evolutionary role for Co-tf. P. L. Clark and colleagues have shown that changing codons can even alter cellular fitness (6). Taken together, there is mounting evidence that proteins have evolved to fold cotranslationally. Yet, these arguments are primarily post hoc, learning from the aftermath of evolution. We lack a “forward model” to understand how protein sequences evolve under Co-tf.

In this issue of Biophysical Journal, Zhao, Jacobs, and Shakhnovich embrace this challenge head on (7). They present a simple yet elegant model incorporating the essential biology to simulate evolution under Co-tf. The model relies on a fitness metric that depends on the amount of available functional protein. The evolutionary advantage of Co-tf over Post-tf is modeled by allowing full-length unfolded proteins (typical in Post-tf) to be stochastically converted to the inactive form instead of folding to the functional form. Irreversible conversion to the inactive form effectively captures the commonly observed deleterious effect of aggregation and/or degradation when proteins fold post-translationally, starting from the unfolded state. Next, new protein sequences are generated via random mutations. These sequences then undergo synthesis, folding (during synthesis), refolding (after synthesis), unfolding (after synthesis), and loss of function (the above noted mechanism), all within a probabilistic framework. Stochastic trajectories are generated for sufficiently long and biologically relevant timescales. Sequence-specific fitness functions are evaluated from these stochastic trajectories to ultimately accept or reject mutations. The model makes a number of interesting predictions.

First, the model predicts that proteins with predominantly local contacts (between amino acids closely spaced in the primary sequence) have evolved to fold cotranslationally. Low contact order can be used as a rough metric to define this set of proteins (class I). The ability to fold cotranslationally is, however, not due to their ability to fold fast, typical of low-contact-order proteins (8). These proteins can begin to fold soon after the start of synthesis because of the hierarchical nature of folding, a feature specific to the native topology, not to folding speed. Simulated evolution of class I proteins shows accepted mutations tend to weaken non-native contacts and strengthen native contacts (primarily local ones). This evolved feature further facilitates the ability of the protein to fold in small steps during the early stages of translation. The second group of proteins (class II) tends to fold late during translation. Proteins in class II typically require formation of nonlocal contacts (between amino acids further apart in the primary sequence), forbidding the early folding observed in class I. Class II proteins are broadly classified as proteins with high contact orders, and they typically fold immediately before being released from the ribosome. These proteins are more likely to fold post-translationally, particularly if their folding speed is slower than their translation speed. Authors Zhao, Jacobs, and Shakhnovich also simulated evolution by ignoring the translation process, mimicking the in vitro folding scenario (7). They found that proteins in class I fold faster when they evolved under this no-translation scenario (similar to in vitro folding) compared with evolution with translation. The no-translation condition requires proteins to fold from the fully unfolded state and offers no protection against degradation or aggregation, unlike Co-tf. Thus, evolution under this scenario forces proteins to fold fast to minimize exposure to the unfolded state.

Besides making new predictions, the “forward model” is supported by multiple well-known observations. In a proteome-wide study, O’Brien and colleagues (9) noted that helical proteins (low contact order) are more likely to fold cotranslationally compared with proteins with α/β or β (high contact order) structures, in agreement with the “forward model.” Zhao et al. predicted that proteins in class I typically fold faster during translation, compared with refolding after translation, from the full-length protein (7). The difference in the two folding rates is consistent with the observation that proteins that readily fold in vivo often take longer to fold in vitro or simply do not fold because of irreversible aggregation. Next, the “forward model” predicts class I proteins tend to accumulate synonymous mutations that encode slow-translating codons. These mutations are most likely to be accepted when introduced in a midsequence location, and they result in enhanced folding efficiency, in agreement with previous studies (10). This finding is further supported by a proteome-wide study by these authors in which protein structures preceding rare codons were curated. Rare codons are also typically slow-translating codons. The curated set statistically had lower contact order than substructures preceding random positions in genes without conserved rare codons. Another outcome of the “forward model” is that evolution, in general, makes folding times faster than the degradation timescale, as previously found (11).

The work of Zhao et al. demonstrates the importance of “forward models” in molecular evolution (7). As with any insightful work, exciting new avenues emerge. In their work, the dynamics of proteins that arise from sequence and structure, often important for proper function, were not allowed to evolve. An intriguing possibility is to explore how nonsynonymous mutations would be gathered (or rejected) in evolutionary responses to the needs for functional dynamics and Co-tf. Would one criteria overrule the other, or would the two evolve in series (i.e., proteins first evolve to a certain function and then respond to the need of Co-tf)? Although their model is highly insightful for small toy lattice proteins, extending the study to larger proteins would be a welcome addition. Larger proteins often exhibit intriguing complexities that may include novel benefits of Co-tf, for example, preferentially delaying formation of some secondary structures while accelerating others to orchestrate their final assembly and overall function (10). It is also timely to think about the roles of other folding machinery, such as chaperones that participate during translation to enhance folding. Moreover, physical interactions with the ribosome could be critical, a feature neglected in their model. Could these considerations be included in a “forward model” to further refine our understanding of evolution?

These are not only questions of intellectual curiosity. They are of immense practical importance too. Understanding Co-tf would significantly contribute to “translational medicine.” Proper folding and function is of prime relevance to ameliorating diseases related to protein homeostasis. Looking ahead, enhanced yield of functional proteins facilitated by Co-tf is critical in several industries, reaching even as far as energy and food production, in which renewable and sustainable alternatives rely on in vivo synthesis instead of traditional chemical syntheses. The time is ripe to further evolve “forward models” to advance the physical biology of evolution.

Editor: Amedeo Caflisch.