Protein folding and misfolding in the cell: towards an atomistic picture

Abstract

Proteins, the molecules that perform the majority of tasks required to sustain life at the molecular level, must generally fold into a specific structure in order to perform their molecular functions. Despite decades of research, we do not fully understand how proteins fold up into their correct structures, starting off as a linear chain of amino acids, while avoiding incorrect misfolded structures linked to diseases such as Alzheimer's, Parkinson's, and various forms of cancer. It was previously believed that most proteins can autonomously fold into their native structures, driven by the physical and chemical interactions between a protein's constituent amino acids. But growing evidence suggests that, for a large number of proteins, these interactions instead cause the chain to misfold into nonnative structures, thus necessitating the assistance of additional cellular mechanisms to ensure the correct native state is attained. For instance, in the cell many proteins can start to fold as they are being synthesized on the ribosome. Previous studies have demonstrated that this process, known as co-translational folding, can significantly improve the native folding efficiency for many proteins that cannot efficiently fold autonomously. Furthermore, recent bioinformatics studies have shown that, in many organisms, co-translational folding tends to begin at nascent chain lengths associated with evolutionarily conserved, slowly translating codons, suggesting that it is widely beneficial to slow down synthesis and give proteins time to fold co-translationally. But the precise molecular mechanisms through which co-translational folding helps proteins efficiently reach their native state and avoid detrimental misfolded states, remains poorly understood, largely owing to immense technical difficulties in studying this highly dynamic process. Such an understanding is necessary if we are to rationally manipulate protein quality control mechanisms in the cell to alleviate misfolding diseases. The goal of this dissertation is to develop and apply novel interdisciplinary pipeline, combining theory, atomistic simulation, and in vitro experiments to elucidate, at the molecular level, why certain proteins which face difficulty folding autonomously can reach their native states much more efficiently via co-translational folding. In Chapter 1, we present a novel algorithm, known as DBFOLD, that uses atomistic Monte-Carlo simulation, machine-learning based analysis, and statistical physics theory to predict detailed folding pathways and rates for large proteins while accounting for the possibility of non-native misfolding--a crucial feature omitted from many existing atomistic simulation algorithms for the sake of computational feasibility. In Chapter 2, we apply the DBFOLD algorithm to predict the co-translational folding mechanisms of certain E. coli proteins with conserved clusters of slow codons and to explain why these proteins benefit from folding co-translationally. We find that, for these proteins, there is a narrow window of intermediate translation lengths at which native-like folding is both thermodynamically favorable and kinetically fast. But beyond these lengths, folding kinetics slow down by orders of magnitude due to deep nonnative traps stabilized by newly-synthesized C-terminal residues. Thus, co-translational folding is predicted to help these proteins circumvent deep kinetic traps and rapidly reach their native state--strategically-evolved slow codons at these lengths can give the nascent chain additional time to take advantage of these optimal folding windows. A key advantage of our atomistic simulations is that they generate highly, specific, experimentally testable predictions. In Chapter 3, we test these predictions as they apply to E. coli MarR, one of the proteins predicted to circumvent deep folding traps via co-translational folding. Using in vitro refolding and mutagenesis experiments, we confirm the existence of these trapped states and preliminarily show that the simulations can accurately predict their structure and underlying molecular interactions. Our experiments thus lend support to our atomistic model for the MarR folding landscape, and indirectly support the predicted mechanism by which co-translational folding may allow folding traps to be circumvented. The work also sheds light on evolutionary tradeoffs that MarR faces between various biophysical properties under selection. Finally in Chapter 4, we apply our combined computational/experimental methodology to investigate the folding mechanism of the receptor binding domain (RBD) of the SARS-CoV-2--the virus behind the Covid-19 pandemic--with the ultimate goal of linking biophysical folding properties to viral fitness and pathology. We find that the RBD can only refold reversibly if its disulfides are kept intact during denaturation, whereas their disruption leads to spontaneous misfolding into a molten-globule like nonnative state which is highly aggregation-prone. But our simulations predict that the RBD can solve this problem by folding co-translationally during secretion in to the endoplasmic reticulum--this process is predicted to increase the odds that the correct disulfides form and ultimately lock the protein into its native state, thus minimizing nonnative misfolding. Together, these results present and validate a novel interdisciplinary pipeline that significantly advances our detailed molecular understanding of co-translational protein folding in the cell--a process long known to be beneficial albeit through poorly understood mechanisms. We expect that future work will continue probing the detailed molecular models generated here, along with their crucial evolutionary and biomedical implications.