A career pathway in protein folding: From model peptides to postreductionist protein science

Abstract

This review discusses the inherent challenge of linking “reductionist” approaches to decipher the information encoded in protein sequences with burgeoning efforts to explore protein folding in native environments—“postreductionist” approaches. Because the invitation to write this article came as a result of my selection to receive the 2010 Dorothy Hodgkin Award of the Protein Society, I use examples from my own work to illustrate the evolution from the reductionist to the postreductionist perspective. I am incredibly honored to receive the Hodgkin Award, but I want to emphasize that it is the combined effort, creativity, and talent of many students, postdoctoral fellows, and collaborators over several years that has led to any accomplishments on which this selection is based. Moreover, I do not claim to have unique insight into the topics discussed here; but this writing opportunity allows me to illustrate some threads in the evolution of protein folding research with my own experiences and to point out to those embarking on careers how the twists and turns in anyone's scientific path are influenced and enriched by the scientific context of our research. The path my own career has taken thus far has been shaped by the timing of discoveries in the field of protein science; together with our contemporaries, we become part of a knowledge evolution. In my own case, this has been an epoch of great discovery in protein folding and I feel very fortunate to have participated in it.

Introduction

Proteins are selected to fold and perform their functions under the constraints of their in vivo environments, which are characterized by complexities such as macromolecular crowding, spatial organization, weak yet specific interactions, and more. There is a growing recognition that those of us seeking to decipher the information in protein sequences must admit these physiological realities and adopt a “postreductionist” view of protein science. However, the rules of physical chemistry obtain, even in the face of physiology. The consequence of a raised consciousness about the cellular realities of protein folding is the necessity to move readily from detailed atomic-level analyses to physiological principles.

Reductionist Roots

The work of Chris Anfinsen established that amino acid sequence encodes the native structure of a protein and that folding could be reconstituted in the test tube as a self-assembly process, so long as appropriate conditions were provided.¹ This launched a generation of protein scientists, myself included, on the quest to learn “the rules” by which given sequences favor particular backbone conformations. In my own case, my graduate work in Elkan Blout's lab was focused on the sequence determinants of reverse turns, notably β-turns. The expertise of the Blout lab made it possible to use organic chemistry in solution to make peptides of any desired sequence, including, if desired, non-native amino acids. Moreover, we exploited cyclization as a strategy to make a peptide less flexible so that it behaved more like a piece of a protein.^2,3

We extended this work to compare sequence preferences and spectral signatures of β-turns and γ-turns; cyclic pentapeptides were ideal model systems for this purpose as they were in essence fused β- and γ-turns (Fig. 1). Both types of reverse turns were strongly favored by combinations of proline and glycine, two amino acids with high frequency of occurrence at the surface of proteins between repeating secondary structure units. These studies of constrained model peptides led to “rules” for turn formation^7,8 that were entirely consistent with those emerging from statistical analyses of residue preferences in the growing number of crystal structures of proteins, for example those carried out by Chou and Fasman^9,10 and Sibanda and Thornton,^11–13 thus validating the local sequence-driven nature of folding information in turns, as well as illustrating the advantages of the use of constrained peptides to mimic structural features in proteins.

Figure 1.

Structure of the cyclic pentapeptide, cyclo(Gly-l-Pro-Gly-d-Ala-l-Pro), as determined by X-ray crystallography⁴ and NMR.^5,6 Note that this canonical cyclic pentapeptide is essentially a fusion of a β-turn (green) and a γ-turn (red) with their stabilizing 1-4 and 1-3 hydrogen bonds, respectively. Adapted from Ref. 4.

A Natural Progression: Peptide Sequences that “Do Biology”

Many peptide hormones were discovered over the 1970s and 1980s as modulators of wide-ranging physiological processes from water balance in the body to reproductive cycles to appetite. These molecules emerged as exciting targets for drug development, as it seemed engineered peptides or peptidomimetics could be found that displayed improved bioactivity and controlled pharmacological properties. Of particular interest was the strategy of enhancing affinity for the target receptor and simultaneously improving biostability by the use of conformational constraints. A collaborative project we participated in illustrates this approach: The goal was to develop conformationally constrained analogs of gonadotropin-releasing hormone (GnRH), also known as luteinizing hormone-releasing hormone. Without knowledge of the bioactive conformation of this linear decapeptide, we reasoned that tests of activity among a number of cyclic analogs could be used to elucidate the pharmacophore, and in turn the conformation of the native hormone when productively bound to its receptor. Excitingly, a synergistic collaboration combining the synthetic prowess of Jean Rivier, powerful computational exploration of favored conformations spearheaded by Arnie Hagler, and detailed NMR studies in our lab yielded several cyclic and bicyclic GnRH analogs (Fig. 2).^15–18

Figure 2.

The structures of (A) the 10-residue linear peptide hormone, GnRH, and (B) a representative potent bicyclic analog designed by a combination of bioactivity assays, molecular dynamics simulation, and NMR¹⁴ [panel (B) adapted from Ref. 14]. All amino acids are in the L configuration unless noted, and 2Nal, 3-(2′-naphthyl)alanine; pClPhe, 4-chlorophenylalanine; 3Pal, 3-(3′-pyridyl)-alanine; Dbu, 2,4-diaminobutyric acid; Dpr, 2,3-diaminopropionic acid.

Related molecules can act either as GnRH antagonists or agonists and are in current use as treatments for a variety of diseases of the reproductive system including importantly prostate cancer.¹⁹ It is gratifying that the “rules” we and others developed for β and γ turn formation came into play in the GnRH analog project.

Another example of biologically crucial information present in relatively short sequences caught our attention in the mid-1980s and, in retrospect, drew us to the complexity of protein folding and maturation in the cell: Critical information for targeting of newly synthesized proteins to their correct destinations is encoded in relatively short segments of polypeptide sequence. Proteins that are destined to be secreted to extracellular sites or to be integrated into the plasma membrane carry a 15–25 amino acid long extension called a signal sequence, normally at their extreme N-termini, and normally cleaved from the final mature product.^20,21 Most provocative was the high sequence variability of signal sequences but with retention of a pattern of residue properties.²⁰ Moreover, in large measure signal sequences could be swapped from one protein to another with retention of function, arguing that their roles were self-contained and did not rely on interaction with their passenger protein. The Nobel Prize-winning work of Blobel²² showed that a single set of components of the secretory pathway handled many different secreted proteins. Clearly, the properties of signal sequences were sufficient to specifically engage the machinery of secretion and facilitate correct targeting. This led us to examine what these properties were, using synthetic signal peptides corresponding to native and defective mutant sequences, which could be compared so as to correlate proper functional capability with essential features.²³ We found that functional signal sequences were required to have some well-defined properties: an ability of the hydrophobic core to fold as an α-helix, a detergent-like tendency to interact favorably with a membrane, but an inability to stably incorporate into a bilayer as a transmembrane helix.^20,24–31 In subsequent work, we and others have been able to reconcile how these essential properties allow the secretory machinery to specifically and productively bind and manage signal peptides.^31,32

In Vivo Folding Helpers: The Molecular Chaperone Era

Because of our research project on signal sequences, one could say that my laboratory was “enlightened”: We knew that bacteria had two pathways for protein secretion: one post-translational and one co-translational. To traverse a membrane in either system a polypeptide had to be unfolded, as it did to enter the mitochondrion or chloroplast. Additionally, we noted that many of the proteins used to demonstrate the principle of self-assembly via Anfinsen-like refolding experiments were actually secreted from cells and folded by a significantly more complex process either in the lumen of the endoplasmic reticulum or in the bacterial periplasm. Hence, the idea that there existed species in the cell that modulated and facilitated folding was not surprising to us. Consequently, we caught the chaperone wave and became fascinated with the early life of a protein in the cell (Fig. 3).^33,34

Figure 3.

Illustration of from a brief review published as the chaperone era was launched. Cartoon reprinted from Trends in Biochemical Sciences, Vol 16, Samuel J. Landry and Lila M. Gierasch, Recognition of nascent polypeptides for targeting and folding, 159–163, Copyright (1991), with permission from Elsevier.

Given our interest in how information is encoded in sequence, we found a central aspect of chaperone function—the question of how unfoldedness can be recognized by chaperones, which seemed to ignore native states—particularly intriguing. We used short flexible peptides as mimics of floppy unfolded protein substrates, and we deployed a powerful NMR approach, transferred nuclear Overhauser effects,³⁵ to explore the features of their chaperone-bound states, in hopes that general principles would emerge. Even though structures were not yet known for the two major classes of Escherichia coli chaperone, GroEL with its partner GroES, and DnaK, the Hsp70 family member, the trNOE results provided compelling pictures of how differently they bound unfolded protein substrates (Fig. 4).³⁶

Figure 4.

Transferred nuclear Overhauser (trNOE) data supporting the different modes of binding of the same peptide to the E. coli Hsp70, DnaK, and GroEL. Adapted from Ref. 36. For details, see the original article, but note the lack trNOEs for the free peptide (A), the absence of NH-NH trNOEs when the peptide is bound to DnaK in an extended conformation (B), and the clear NH-NH trNOEs when the peptide is bound to GroEL and adopts a helical conformation (C).

Peptides bound to DnaK were extended with both side chain and backbone interactions with the chaperone-binding site, whereas GroEL-bound peptides could be folded into any backbone conformation so long as a cluster of hydrophobic side chains was present for binding to the chaperone.^37,38 These observations can now be understood in terms of the structures of these two major chaperones.

After the initial recognition that molecular chaperones assisted folding in the cell as well as helping to protect against aggregation and precluding premature folding in processes like membrane translocation, there ensued a very active period of research on their molecular mechanisms. In our lab, taking advantage of its narrow NMR resonances, we identified the mobile loop on GroES responsible for the interaction of this co-chaperone with GroEL (Fig. 5).³⁹ In collaboration with the Deisenhofer lab, we later solved the crystal structure of GroES,⁴⁰ and by crystallography the mobile loop was essentially invisible, showing the complementarity of the two structural tools. The subsequent tour-de-force solution of the structure of a GroEL/ES complex by the Horwich and Sigler labs confirmed the role of the mobile loop⁴¹; we now know that formation of a complex between GroEL and GroES serves to cycle GroEL from its substrate-bound state to a low-affinity state in which the encapsulated substrate explores conformational space in a protected space.

Figure 5.

The sequence of the mobile loop of GroES (A) and two-dimensional NMR spectra (TOCSY in bold, NOESY in thin lines) supporting the surprisingly high degree of flexibility of this loop in the context of the 70 kDa GroES heptamer, as indicated by the narrow linewidths of its resonances (B). Adapted from Ref. 39.

We continue to work actively on the mechanism of Hsp70 chaperones.^42–48 This ubiquitous family not only acts on unfolded proteins to facilitate folding or to pass them on to GroEL or other partners, but also protects nascent chains, shepherds proteins that are to be translocated across membranes, facilitates formation and disassembly of complexes, and acts with partner chaperones to rescue proteins from aggregates. This active research project in our lab is emblematic of the challenge of connecting detailed physical chemistry with physiological complexity. Hsp70 chaperones are paradigmatic allosteric machines, and deep understanding of their molecular mechanism is both challenging and highly important. They work in complex networks with other chaperones and co-chaperones and understanding their roles in protein homeostasis in the cell demands that we incorporate the full roster of their interacting partners. We accept the importance of including this complexity in our research, but we are cognizant of the technical challenges and the inevitably less deep and detailed descriptions we must accept when we treat the more complex system.

The Full Monty: Exploring Protein Folding in Intact Cells

Many technical and theoretical challenges accompany any efforts to explore protein folding and function under in vivo conditions, as even gentle disruption of the cellular context irreversibly abolishes its essential features, not unlike Humpty Dumpty.⁴⁹ Key factors that are emerging as paramount in cellular folding are the ubiquitous weak intermolecular associations and the networks of chaperones that remodel the folding landscape and impact fluxes of protein to any possible cellular fates: proper folding, degradation, or aggregation.⁵⁰

In the mid-1990s, we launched an effort to explore the folding landscape of proteins in the cell. Our early studies have focused on a heterologous protein expressed in E. coli, and we chose to use cellular retinoic acid-binding protein I (CRABP I) initially because of our extensive experience with the folding of this β-rich protein in vitro.^51–56 A clear advantage of using a non-E. coli protein is that folding can be studied without altering directly the cellular physiology by virtue of the function of the protein under study. CRABP I is also delicately balanced between successful folding and formation of aggregates, both in vitro and in vivo.^57,58 Therefore, we can observe the impact of any perturbations to this delicate balance as a change in the fraction of native protein produced. For example, we can overexpress or underexpress cellular chaperones and read-out whether the percent soluble natively folded CRABP I is changed.

We are designing folding sensors so that we might measure in-cell stability and also directly observe populations of N or U (or even potentially intermediates) in the cell. The first effort along these lines was quite promising and enabled us to carry out urea titrations on live E. coli (Fig. 6).^59,60

Figure 6.

Comparison of urea melts for FlAsH dye-labeled tetra-Cys CRABP I in vitro and in vivo. For details, see Refs. ⁵⁹ and ⁶⁰. Not that both the apparent stability and the urea dependence of the transition region differ markedly between the two environments. Reprinted with permission from Proceedings of the National Academy of Sciences, Vol 101, 523–528, Zoya Ignatova and Lila M. Gierasch, Monitoring protein stability and aggregation in vivo by real-time fluorescent labeling, copyright (2004).

The result was an apparent ΔG_UN° that was lower than that determined in vitro at the same temperature. The apparent m value (urea dependence of ΔG_UN°) was nearly twice that in vitro, and equilibration between N and U was more rapid in vivo than in vitro. Although these observations were tantalizing, we certainly knew that urea would be perturbing other proteins than the one that was labeled, and we are now engineering folding sensors that can be triggered by ligand binding and so avoid the use of a denaturant. Meanwhile, other innovative systems have been reported to explore folding in the cell,⁶¹ and we are beginning to have multiple examples to compare but cannot yet make general conclusions about how folding equilibria are perturbed by cellular conditions.

We have become convinced that a major factor that reshapes the energy landscape for protein folding in the cell is the presence of many chaperone systems, all of which share the propensity to bind unfolded or incompletely folded states of proteins. To explore the combined impact of chaperone networks and to help us select and perform the best experiments, in collaboration with Evan Powers, we have developed a computational model of the E. coli chaperone networks and the flow of protein from biosynthesis, to native or aggregated state, or to degradation. The program, called FoldEco, runs in Mathematica and has proven extremely useful in helping pose experimental questions (E. Powers and L. M. Gierasch, manuscript in preparation).

The Present and the Future: Protein Folding and Physiology Hand-in-Hand

The half century that has passed from Chris Anfinsen's seminal work to the present has witnessed great advances in our understanding of protein folding, from insights into energy landscapes and folding mechanisms, to the recognition that folding in the cell is assisted by molecular chaperones, to the discovery of protein misfolding diseases, and to the realization that aggregation is a near-universal process that competes with folding of proteins both in vitro and in vivo (Fig. 7). These chapters in the evolution of protein folding research are all linked, but the intellectual connections could not have been anticipated. Now, we work in an era when therapeutic strategies are being developed to treat misfolding diseases⁶² and when the idea of chaperone-based therapies is on the horizon.⁶³ What an exciting time to have been privileged to be part of!

Figure 7.

Timeline of protein folding research developments since the seminal work of Chris Anfinsen.

What are the frontiers? The technical challenges? We will need to understand in increasing depth the complexity and linkages of pathways and regulatory circuits affecting folding in the cell. Methods will be required to measure transient interactions, to obtain spatial maps of protein distributions, to follow in time the states visited by a protein, both chaperone-bound and free, and to computationally model this symphony and how it is perturbed by stresses and mutations. The current and future generations of protein scientists must take up the challenge and work with the necessarily detailed and powerful physicochemical rigor, but admit the biological realities and complexities that are inherent to living systems. Daunting, but so cool!