CHAPTER 1.1: Disulfide Bonds in Protein Folding and Stability

Early protein folding studies were interpreted as suggesting that the native state of a protein corresponds to one well-defined conformation, whereas the unfolded state corresponds to a random coil.¹  If no other states than either the native or unfolded state are kinetically or thermodynamically stable, we speak of a two-state folding mechanism for a protein.²  This scenario set the stage for early analyses of the role that disulfide bonds play in protein stability.

An ideal random coil is devoid of any long-range interactions except excluded volume effects. It behaves as a freely joined chain with segments of defined length.³  In such a system, the impact of a covalent crosslink between two defined residues of the polypeptide chain, such as a disulfide bond, would be greatest on the unfolded state, significantly decreasing the conformational freedom of the random coil. Restricting the conformational space of the unfolded state reduces its entropy. Hence, in the presence of a disulfide bond, the entropy change for the reaction to the ordered native state is less negative, with net stabilization of the folded protein as a result. This model will be called the chain-entropy model in the following sections. A quantitative description was developed by Flory,⁴  Schellman⁵  and Poland and Scheraga.⁶  The decrease in entropy of the unfolded state is derived from the probability that two otherwise free elements of the chain are now found in a defined volume element (v). The mathematical description of the problem, based on polymer theory, can be found in the equation

where R is the gas constant and l the average length of a statistical segment of the chain composed of N segments; in proteins, l is assumed to be 3.8 Å, corresponding to one amino acid. A major point of discussion has been the suitable choice of v. A value of 57.9 ų based on the closest possible approach of two thiols is mostly in use.⁷  Hence eqn (1.1.1) can be simplified to

where n is the number of amino acids bridged by the disulfide bond. Based on a study of ribonuclease (RNase) T₁ with no, one and two intact disulfide bonds, eqn (1.1.2) was developed by Pace et al.⁷  They not only found a good correlation between n and ΔΔG upon removal of disulfide bonds in RNase T₁, but also observed agreement between the predictions from these equations and the experimental data for lysozyme, RNase A and the antibody C_L domain.

The above equations have two main consequences. Conceptually, the stabilization of a protein is thought to be an entirely entropy-driven process with an impact exclusively on the unfolded state. In theory, the stabilization achieved by a disulfide bond should therefore always increase with increase in the number of amino acids between the two cysteines. Despite its appealing simplicity, in practice this theory falls short in important aspects of real proteins. It treats the unfolded polypeptide chain as a system devoid of any intra- or intermolecular interactions. Furthermore, the water surrounding a protein is an important factor in shaping the free energy landscape of the polypeptide chain, and also in the native state, but is neglected in the equations.^8–10  The impact of these considerations regarding disulfide bonds was addressed by Doig and Williams in 1991.¹¹  They argued that disulfide bonds may significantly decrease the solvent-accessible surface in the unfolded state of a protein. As a consequence, hydrophobic residues and also hydrogen-bond donors and acceptors may become buried. Burial of hydrophobic residues would lead to less ordering of water and thus a higher entropy of the solvent surrounding disulfide-containing proteins. Consequently, the hydrophobic effect, a major driving force in protein folding, should be less pronounced. On the other hand, hydrogen bonding with the solvent will be less extensive for a more compact unfolded state, thus reducing this competition. As a result, folding to the native state will be enthalpically more favorable. According to the authors, this enthalpic contribution must be considered as the major stabilizing factor of disulfide bonds. This model will therefore be called the solvent-enthalpy model in the following sections. As in the chain-entropy model, effects on the native state are also neglected in this model. Both models are summarized in Figure 1.1.1.

A variety of experimental evidence argues either for or against these two theories, rendering the problem much more complex, but the data obtained from the various experiments also provide a chance for further insights. The advent of site-specific mutagenesis offered the possibility of introducing artificial disulfide bonds at defined positions within a protein and allowed investigators to explore the effects of disulfide bonds on protein folding more rigorously.^12–15  Many of the results obtained in these studies were not compatible with predictions based on the prevailing theories concerning the effects of disulfide bonds on protein stability. Most discrepancies arose from efforts to stabilize proteins by the introduction of a new disulfide bond that was not present in the original protein. In contrast to expectations, destabilization of the protein was the result of a significant number of these attempts.^16–18  High-resolution structures of the engineered proteins helped to explain some of these unexpected effects. It turned out that strain was imposed on the native state by the disulfide crosslinks in several cases.¹⁹  Two particularly insightful examples are studies on barnase^16,20  and staphylococcal nuclease.¹⁷  Clarke et al. introduced three artificial disulfide bonds into barnase: one destabilized the native state, one stabilized it to an extent predicted by the chain-entropy model and one had a much weaker effect on stabilization than predicted.¹⁶  Furthermore, the bond connecting fewer residues was more stabilizing than the one encompassing more residues. An NMR H–D exchange analysis of the barnase mutants revealed altered dynamics of the native state, imposed by the presence of the covalent crosslinks.²⁰  For staphylococcal nuclease, no stabilization was found for any of the disulfide-introduced constructs, but in this case the cis–trans equilibrium of a peptidyl–prolyl bond was shifted in the native state and the catalytic activity of all the mutants was reduced.¹⁷  Accordingly, strain on the native state, as reflected by the alteration of the cis–trans equilibrium, was evoked to explain the unexpected observations.

One major conclusion can be drawn from these two studies and a variety of others:^{12–14,19,21–25}  the native state of a protein is often affected by the presence of a disulfide bond, in particular if it was artificially introduced and not naturally evolved. Thus, a major assumption of the chain-entropy model and the solvent-enthalpy model frequently does not hold. Covalent crosslinks often impose strain on the native state. This is then reflected in structural changes in some cases,^17,19  or in more subtle ones, such as variations in the dynamics of the native state. As a consequence, both the enthalpy and the entropy of the native state are likely altered when two residues in the polypeptide are crosslinked covalently. The alterations in dynamics and structure are not always global, but sometimes local and context specific. For example, β-sheets and loops are thought to be more suited to dissipate induced strain than α-helical elements, and more dynamic parts of the structure are influenced to a greater extent. Changes in global dynamics, e.g. the vibrational normal modes of a protein, will also influence the entropy of the native state, as was shown in a molecular dynamics study by Tidor and Karplus.²⁶  Consequently, contrary to simple models, the introduction of disulfide bonds into a protein may well lead to net destabilization due to a loss of native-state entropy. In addition, changes in the solvent-exposed hydrophobic surface, for example in disulfide mutants of interleukin 4, have a significant effect on the stability of the native state.²⁵  In summary, both experimental and theoretical insights clearly argue against the simple chain-entropy model, at least pertaining to one side of the equation, the native state. The same holds true for the solvent-enthalpy model, which is also at odds with thermodynamic parameters derived for some disulfide-bonded proteins, revealing that disulfide bonds do not necessarily stabilize the native state enthalpically.

But what about the unfolded state? Is it correct to assume that a random coil is either completely devoid of any interactions or is dominated only by hydrogen bonding to the solvent? Clearly this is not so in all cases. Residual structure has been detected in a variety of proteins.^27–31  In particular, residual hydrophobic interactions or fluctuating α-helical elements seem to be more of a general feature of proteins than an exceptional one. This is often the case under mildly denaturing conditions,^30–34  and therefore can be expected to be even more pronounced under physiological conditions in the absence of denaturants. Importantly, the structural features of the unfolded state, such as residual structure in the unfolded state of barnase, have in some cases been shown to be influenced by the presence or absence of disulfide bonds.³⁵  Hence disulfide bonds very likely not only influence the conformational freedom of the unfolded chain but can also introduce structure, which, for example, may be protective against irreversible aggregation.^36,37  Even apparently minor structural changes, such as the clustering of some hydrophobic residues, will influence the enthalpy and entropy of the unfolded state. These aspects – together with strain induced on the native state and an impact of disulfide bonds on the dynamics of the native structure, as outlined above – must all be considered to reflect the real complexity of proteins.

Is a comprehensive theory possible that quantitatively describes the effect of disulfide bonds on a protein’s stability? Most likely it will remain an approximation, but many of the factors that need to be taken into account have been identified. The decreased entropy of an unfolded and crosslinked polypeptide chain needs to be considered as developed in the chain-entropy model. Effects on the hydrophobic effect and more prominently hydrogen bonding in the unfolded state as reflected in the solvent-enthalpy model should be included. Interactions induced in the unfolded state due to the disulfide bond, which may not be localized directly around the bond but can be present as long-range residual structure,³⁵  must also be included in a comprehensive model. This will clearly have an effect on the enthalpy and entropy balance for the reaction to the native state. The same holds true for decreased dynamics of the native state, locally or globally, and also enthalpically unfavorable strain or enthalpically favorable induced proximity of interacting residues. In summary, the effect of a disulfide bond on the stability of a protein may have been readily assessed experimentally, yet its molecular explanation can be almost as diverse as the protein under investigation. The key factors giving rise to the net effect are most likely known in many cases, but the individual contributions of these factors to the overall effect are blurred in their sum, as is their influence on each other. Exact stability data, combined with structural data on the native and the unfolded states of the protein under investigation, are required for a detailed understanding of the effect of a disulfide bond. They should be complemented by theoretical approaches, such as molecular dynamics simulations of the native and unfolded states, to obtain a more complete picture. Although this will be unique in detail for different proteins, in summary these analyses should reveal general principles about the effects of disulfide bonds connected to their specific location within a protein.

One of the most fundamental questions in biophysical chemistry is how a linear polypeptide chain specifically adopts its intricate three-dimensional structure within a reasonable amount of time. In vitro, a variety of mutational approaches,^38–46  high-resolution structural techniques^47–50  and ultrafast perturbation and analysis methods^51–55  have provided deep insight into this phenomenon of biological self-organization. It is best understood for two-state folders, proteins that populate only the completely folded or the completely unfolded state. Major concepts, as discussed below, have been developed by studies on these two-state folders.

While folding, proteins explore a multi-dimensional energy landscape (hypersurface), where loss of chain entropy is compensated by a gain in enthalpy-driven backbone or side-chain interactions, most prominently hydrogen bonds, ionic interactions and hydrophobic interactions.^56,57  Folding is not a random exploration of all possible conformers, because interactions formed early will restrict the conformational space for downstream exploration of the energy landscape. For naturally occurring proteins, this gives rise to a funnel-like energy landscape.^58,59  This funneled energy landscape, which allows protein folding to occur on biologically relevant time scales, is a product of evolution and not an intrinsic characteristic of a heteropolymer.⁶⁰  In many cases, readily foldable polypeptide sequences have been selected by evolution for minimal frustration, thus lessening competition between individual interactions and instead providing cooperativity in their formation, which reduces the ruggedness of the energy hypersurface.^61,62  Often, formation of the native topology, which can depend on a very small number of residues coming into contact, is the rate-limiting step and defines the overall nature of the transition state that separates the unfolded and native states.⁶³  Accordingly, the average sequence separation of amino acid residues that ultimately need to interact in the native state of a protein is a critical factor affecting the folding rate for two-state proteins.^64,65  In agreement with the theory of protein-folding funnels, two-state folders can take multiple pathways to the native state – yet only the native state and the unfolded state will be kinetically and/or thermodynamically stable.^59,66 

Structure in the unfolded state is regarded as an important element in protein folding. As outlined before, interactions in the unfolded state will not only influence the net stability balance of a protein, but are also likely to have an impact on the pathways taken through the energy landscape on the way to the native state. Preformed interactions in the unfolded state may have opposing effects on a folding reaction. For instance, if these interactions are present in the native state and do not adversely influence the ability of the remaining polypeptide chain to explore the necessary conformational space, they may lead to faster and more efficient folding to the native state. Conversely, if they are non-native or too stable, they may thwart the folding process and acquisition of the native state. This is where disulfide bonds prominently come into play and examples for both cases have been reported. In the case of RNase T₁, the preservation of a disulfide bond during denaturation decelerated the refolding kinetics, which was attributed to an influence on peptidyl–prolyl isomerization, over-stabilization of partially folded states and decreased chain flexibility.⁶⁷  In this context, it is important to note that RNase T₁ possesses two disulfide bonds, one connecting a small N-terminal β-turn and the other connecting the C-terminus to this N-terminal β-turn (Figure 1.1.2). Accordingly, the protein will be an almost completely looped structure in its unfolded state if the long-range disulfide bond is preserved. This will clearly affect the overall dynamics of the polypeptide chain and may have an impact on the formation of certain topologies.

Conversely, in the case of the constant domain (C_L) of the antibody light chain (Figure 1.1.2), formation of its single disulfide bond accelerated folding up to ∼100-fold.^68,69  In this case, the disulfide bond is found in the hydrophobic core of the protein and is part of the immunoglobulin folding nucleus.^70,71  For other proteins that have been investigated, the presence of their natural intrinsic disulfide bonds can either enhance or reduce their respective folding reactions.^18,72–77  Hence disulfide bonds are in general far from being inert in kinetic terms. Very often, different disulfide bonds within the same protein have distinct effects on the folding rates. A comprehensive study in this respect has been carried out for the all-β-sheet protein CD2, where 13, artificially introduced, disulfide bonds showed markedly different effects on the folding behavior,¹⁸  and also the unfolding kinetics of the altered proteins.^18,78  The experimental findings are in agreement with simple lattice-based simulations, where disulfide bonds inside the folding nucleus were found to accelerate the folding reaction, whereas those outside decelerated folding.⁷⁹ 

Despite their heterogeneity, the effects of disulfide bonds on protein folding/unfolding kinetics can be more easily rationalized than their effects on native-state stability. The key lies in the transition states for folding and unfolding. As outlined above, establishment of the native state topology is often rate limiting for protein folding. Hence acceleration of folding is expected when residues that need to come into contact early in a protein-folding reaction are crosslinked, if not outweighed by entropy/enthalpy compensation in the transition state. Analogously, deceleration of unfolding is expected when two residues, whose interactions need to be broken in the transition state for unfolding, are covalently linked. Both effects can also provide structural information about the otherwise hardly accessible transition states. Examples are the immunoglobulin domain-containing proteins C_L and CD2, where the transition states for folding are stabilized by disulfide bonds,^18,68  and barnase, where the transition state for unfolding is destabilized by a disulfide bond.⁷⁸  Often, disulfide-bonded proteins are found to fold less cooperatively.⁷⁶  This may be caused by the population of disulfide-stabilized, partially folded intermediates. If stabilization of partially folded structures becomes too strong, by either native or non-native interactions, this can even result in a net deceleration of protein folding, as was reported for CD2, where attempts to increase the folding rate by multiple disulfide bonds had the opposite effect, namely deceleration of the folding reaction by over-stabilization of a partially folded state.⁷⁵  This highlights the aforementioned role of cooperativity for efficient protein refolding, which can be beneficially or detrimentally influenced by maintaining disulfide bonds during the denaturation process. In addition, if erroneous disulfide bonds form during the refolding of a reduced substrate, they may trap and stabilize non-native folding intermediates, thus inducing covalent frustration, so to speak, which is more difficult to overcome. When multiple disulfide bonds can form in a protein, these may lead to parallel folding pathways, further complicating protein folding. These scenarios are of particular relevance for protein folding in vivo and will be discussed below.

Insights into the impact of disulfide bonds on protein folding are necessarily biased by the available experimental studies. As described before, these mostly rely on using unfolded proteins with disulfide bonds already formed – or studying folding under reducing conditions to prevent the formation of disulfide bonds (Figure 1.1.3). Of particular relevance for protein folding in the cell, however, is the more complex scenario: folding of proteins while all possible disulfide bonds can form, break and isomerize (Figure 1.1.3). Recent single-molecule,⁸⁰  theoretical⁸¹  and in vivo studies⁸²  have provided insight into these processes and argue that conformational folding, which brings correct cysteines into proximity, drives disulfide bond formation. However, these studies also rely on a limited set of proteins and many in vivo studies argue that these are only part of the picture.

The same biophysical principles outlined in the preceding sections also form the basis of oxidative protein folding in vivo, which in eukaryotic cells occurs primarily in the endoplasmic reticulum (ER). However, a number of major differences exist between how protein folding is studied in vitro and what we have learned about protein maturation in the ER. For instance, in vitro folding reactions usually employ small, single-domain, two-state folders that are not modified with glycans. Conversely, in the mammalian ER, most proteins are large, comprised of multiple domains, often oligomeric, and glycosylated. In cases where their folding trajectory has been elucidated, multiple intermediates have been identified. Perhaps one of the most prominent difference is that folding reactions in vitro generally start by transferring a denatured protein into a solution that will allow it to refold to a native conformation (Figure 1.1.3). As such, the full-length polypeptide chain will be present from the beginning of the protein folding reaction. All cysteines are available and free to react with each other. To deal with this complexity, many in vitro studies start with either a denatured protein with disulfide bonds intact or a denatured and completely reduced protein. In the latter case, reducing agents are included in the folding reaction. In contrast, most proteins that will ultimately form disulfide bonds in vivo will emerge from the Sec61 translocon co-translationally into the disulfide-promoting environment of the ER lumen and cysteines will be exposed vectorially during biosynthesis. Co-translational folding therefore may simplify disulfide bond formation when sequential cysteines are the native ones to bond. However, sequential emergence of cysteines can also complicate folding when cysteines that form native disulfide bonds are far apart in the linear sequence or are separated by interjacent cysteine residues (Figure 1.1.3). Both scenarios occur and will be discussed in the following sections.

A second major difference between in vitro and in vivo oxidative folding reactions is that the speed and conditions at which disulfide bonds form and the native structure is achieved can differ vastly. Early experiments by Anfinsen’s group revealed that urea-denatured ribonuclease (RNase) spontaneously oxidizes to form four intramolecular disulfide bonds over a protracted period, but this did not represent the native form of the protein, as the oxidized RNase had no enzymatic activity.⁸³  In contrast, if a molar excess of β-mercaptoethanol was added to the reaction, nearly 80% of activity could be spontaneously achieved within 24 h, or considerably faster when much lower concentrations of RNase were used.⁸⁴  In both cases, however, non-physiological conditions of pH and temperature were required. This is in dramatic contrast to RNase folding in cells, where it has been estimated to occur within about 3 min of entry into the ER.⁸⁵  Although many proteins synthesized in the ER achieve a folded state sufficient to pass ER quality control within 30–60 min, it is important to note that not all proteins fold this rapidly in vivo.^86–88  The binding of molecular chaperones and complex folding pathways can lead to very long maturation times. For instance, a portion of the HIV-1 envelope glycoprotein gp160 is still folding after 24 h of synthesis, resulting in a significant steady-state ER localization. Folding is still very efficient, leading to high yields of the mature protein.⁸⁹  Anfinsen and co-workers found that the addition of purified microsomes from a variety of tissue sources dramatically increased the rate at which activity was restored to the denatured RNase in their in vitro refolding assays and allowed this to be achieved under more physiological conditions.⁹⁰  Isolation of the molecule(s) responsible for the increased folding activity led to the identification of the enzyme protein disulfide isomerase (PDI), which is the founding member of the large PDI family of enzymes found in the mammalian ER and is discussed in detail in Sections 3 and 4 of this book. Subsequent studies using an in vitro translation system coupled with dog pancreas microsomes demonstrated that microsomes depleted of PDI were dramatically deficient in supporting the formation of intramolecular disulfide bonds in the wheat storage protein γ-gliadin.⁹¹ 

The formation of disulfide bonds in proteins is largely restricted to proteins of the secretory pathway and specifically those portions of the protein that reside within the lumen of secretory pathway organelles. In addition, disulfide bonds are critical to the biogenesis of a number of proteins present in the intermembrane space of mitochondria,⁹²  and their formation is dependent on Mia40, an oxidoreductase, and Erv1, a sulfhydryl oxidase (see Chapter 3.2 for details).^93–95  Although the reducing environment of the cytosol is likely to be largely responsible for the scarcity of disulfide bonds in proteins expressed there, it is noteworthy that one study examined the cysteine content of intracellular and extracellular mammalian proteins and found that cysteines are much rarer in intracellular proteins.⁹⁶  Cysteines therefore appear to be selected against even in the face of a reducing cytosolic environment owing to their high potential reactivity. The intricate connection between protein folding and disulfide bond formation that exists in vivo is highlighted by the fact that reducing agents such as dithiothreitol (DTT), which inhibit disulfide bond formation, are among the most potent inducers of the ER unfolded protein response (UPR), which is activated by the accumulation of unfolded proteins in the ER.⁹⁷  This argues that for at least some proteins, their folding/stability in the cell is driven by or dependent on disulfide bonds.

Whereas exquisite methods have been developed to study the refolding of proteins in vitro in great detail, owing to the complex and crowded environment of cells that express a large number of different proteins in various states of folding, the methods for examining folding in vivo are more limited and often rather indirect. These include the failure of a protein to be expressed in the correct cellular location,⁹⁸  the unusually rapid degradation of a protein⁹⁹  or prolonged association of the protein with molecular chaperones.^100,101  More direct methods for studying protein folding in vivo include detergent solubility experiments,¹⁰²  interaction with conformation-specific antibodies,^103,104  and sensitivity to proteolysis.^105,106  However, even if a protein fails to fold and as a result, for instance, becomes insoluble in non-ionic detergents, the portion of a protein that has failed to fold or the effects of unfolded segments on other regions of a protein are not readily revealed. By far the most direct method of examining oxidative protein folding in a cell is the determination of the redox status of cysteines in various regions of a protein.

As a protein unfolds, its hydrodynamic radius often increases, which can result in a slower mobility on sodium dodecyl sulfate (SDS)-polyacrylamide gels.¹⁰⁷  Correspondingly, in some cases, the formation of intramolecular disulfide bonds can confer increased mobility to a protein on SDS-polyacrylamide gels, if the cysteine pairs are a sufficient distance from each other, owing to the stabilization of more compact conformers of the protein.¹⁰⁸  This realization provided the basis for two methods to study the folding of a protein during translation and translocation into the ER lumen. The first employs cDNA constructs of proteins in which ribosome-stalled intermediates are generated by introducing restriction sites at various locations along the reading frame.¹⁰⁹  The cDNA is linearized with the restriction enzyme and used to generate mRNA transcripts that are translated in the presence of ER microsomes to the end of the mRNA generated by the cut. The absence of an encoded stop codon stalls and stabilizes the bound ribosome at this site, allowing the N-terminal region of the protein to enter the ER lumen, while ∼20 amino acids span the microsomal membrane and ∼40–60 amino acids remain in the ribosome channel. The stalled, labeled construct can be examined directly on non-reducing polyacrylamide gels, or the addition of either puromycin or RNase can be used to force the release of ribosomes, allowing an additional ∼60–80 amino acids to enter the ER before analysis by SDS-PAGE.¹¹⁰  A second method relies on pulse-labeling conditions that are shorter than the time required to synthesize the full-length protein, which is followed by chase periods of varying length. The isolated polypeptide is then analyzed by two-dimensional SDS-PAGE (polyacrylamide gel electrophoresis) in which the first dimension is run under non-reducing conditions and the second under reducing conditions. The formation of a disulfide bond on the nascent chain is visualized by a shift in migration below the diagonal due to the induced increase in mobility.¹¹¹  The demonstration that DTT could act as a cell-permeable reductant¹¹²  paved the way for studies by Braakman et al. to use this small molecule to synthesize metabolically labeled, full-length proteins in a completely reduced state. Subsequent removal of the DTT resulted in reoxidation of the ER and allowed oxidative folding of the protein of interest in the natural environment of the ER.^113,114  This method provided an in vivo correlate to some of the in vitro methods that employed full-length, reduced proteins. Together, these methods have allowed fairly detailed delineations of both rather simple and amazingly complex pathways of oxidative folding. These include the folding of proteins in which disulfide bonds occur through oxidation of sequential cysteines, ones in which long-range, non-sequential cysteines are used, and even to pathways that rely on the initial production of non-native disulfide bonds that are subsequently reduced during the folding process to allow the formation of native bonds in order to achieve the mature, functional structure (Figure 1.1.3). Examples of each type of oxidative folding are discussed below.

Some of the earliest attempts to examine the formation of disulfide bonds in nascent chains as they entered the ER lumen came from the work of Bergman and Kuehl,^115–117  who used heavy and light chains of antibody molecules for their studies. This client is particularly interesting to discuss, as it has continued to be used for in vitro and in vivo folding studies as described in more detail below. The early studies by this group utilized a mouse plasmacytoma making very large quantities of an immunoglobulin (Ig) light chain (LC) that is composed of two ∼100 amino acid Ig domains organized in tandem, each of which folds independently into a twisted β-barrel structure that is secured with a single disulfide bond as already described for the C_L domain (Figure 1.1.2). Nascent chains that remained associated with a tRNA were biochemically isolated and tryptic fragments were examined. They found that Ig LC in which the entire first domain (V_L) had entered the ER, but in which only a small portion of the second domain (C_L) had been synthesized, had already formed the V_L disulfide bond and they calculated that this occurred within ∼1 s after the second cysteine of the bond entered the ER lumen.^115,116  Additional studies by this group revealed similar co-translational oxidation of the N-terminal V_H domain of an Ig γ heavy chain and even an interchain disulfide-linked assembly of Ig subunits before full translation of the proteins was completed.¹¹⁷ 

In their simplest form, antibodies are heteromeric proteins covalently assembled from two Ig heavy chains (HC) and two Ig light chains (LC). Each chain is comprised of multiple Ig domains (four and two, respectively) that are ∼100 amino acids in length and form a twisted β-barrel structure, which is stabilized by a single intradomain disulfide bond between two sequential cysteines (Figure 1.1.2). It is noteworthy that antibodies were one of the very first proteins that PDI was shown to bind in cells,¹¹⁸  and the addition of recombinant PDI to purified Ig HC and LC enhanced the formation of heteromeric HC₂LC₂ antibodies.^119,120  In vitro studies performed with single Ig domains^69,121  and full-length LC¹²²  provided evidence that the isolated Ig domains are able to fold independently of each other, that the presence of a slower folding domain does not impede folding of other domains and that the maintenance of the intradomain disulfide bond during unfolding can accelerate refolding of a domain. These studies also revealed that the oxidized form of the isolated C_L domain could be readily distinguished from a reduced form on non-reducing gels. DTT washout experiments, as described above, revealed that LC synthesized in cells in the presence of this reducing agent rapidly formed the intrachain disulfide bond in the C_L domain upon reoxidation of the ER, with the V_L forming its bond more slowly.¹²³  This is contrary to in vitro studies with a Bence Jones protein, in which the V_L domain was found to fold more rapidly than the C_L domain,¹²²  and points to the fact that the sequences of V_L domains can vary considerably. In fact, some V_L domains are unable to fold to a native structure in isolation and instead become substrates for ER-associated degradation in cells.¹²⁴  The in vitro refolding of most domains of the Ig HC have been studied individually, which revealed that the intradomain disulfide bond significantly increased the stability of the folded state of the C_H3 domain,¹²⁵  or for the less stable C_H2 domain, needed to be left intact in order to allow refolding.¹²⁶  However, the C_H1 domain, which forms the cellular basis for Ig quality control, is unique in that it remains unoxidized in cells when expressed without LC and is bound to BiP (immunoglobulin-binding protein).¹²⁷  The post-lysis release of BiP with ATP allows the C_H1 domain to oxidize, but it is not properly folded and aggregates quickly.¹²⁸  In vitro studies performed on an oxidized C_H1 domain confirmed that it binds to BiP and thus is still unfolded, and NMR analyses established that the C_H1 domain is unstructured in isolation.¹²⁹  The introduction of LC into cells expressing BiP : HC complexes led to dissociation of BiP and the concomitant native oxidation of the C_H1 domain.^127,128  More refined NMR analyses provided a molecular mechanism for this observation, where it was observed that interaction of the C_L domain of the LC with several residues in the C_H1 domain initiated a folding nucleus in this domain.¹²⁹  The Ig domain represents a basic structural module that is widely used in a large number of functionally diverse proteins, because it contains a core with an anti-parallel β-sheet structure in which the multiple loops emanating from this very stable fold are modified to encode other functions. Hence the rules for oxidative folding of Ig domains elucidated for antibody molecules are likely to apply to many other proteins. A study employing a combination of in vitro refolding assays and in vivo expression data examined the oxidative folding of the extracellular portions of the T-cell receptor (TCR) α and β chains, each of which possesses two Ig-like domains in tandem. Indeed, for the A6 TCR, a single domain in each of the chains folded well in isolation and formed its intradomain disulfide bond, whereas the complementary domain in the partner chain remained unstructured and reduced prior to assembly. In both cases, the well-folded domain induced folding of its unstructured partner domain and formation of the intradomain disulfide bond upon heteromeric assembly.¹³⁰ 

The second group of secretory pathway proteins that undergo oxidative folding includes those that require the formation of long-range disulfide bonds occurring between non-sequential cysteines (Figure 1.1.3). This group of proteins by necessity is likely to require the involvement of molecular chaperones/PDIs to prevent the reduced cysteines from interacting prematurely with other cysteines, leading to the formation of non-native bonds. The oxidative folding of several proteins representing this group have been analyzed in detail and three examples are highlighted in the following. The HIV-1 envelope glycoprotein gp160 possesses 10 highly conserved disulfide bonds, five of which involve non-consecutive cysteines, and the formation of these bonds begins co-translationally with final bonds forming after a full-length translation product has been made.^89,131  Cell-based studies have shown that gp160 is a client of both BiP¹³²  and calnexin/calreticulin,¹³³  and the additional activity of PDI is required to establish proper oxidative maturation of this envelope protein.^134,135  A second member of this group is the cytokine IL-12, which is composed of an α and a β subunit assembled into a covalent heterodimer. The α chain possesses three intramolecular disulfide bonds, all of which occur between non-consecutive cysteines, one of which is particularly long distance.¹³⁶  Their establishment during folding was recently determined through cell-based studies, although the role of molecular chaperones in protecting these vulnerable cysteines was not directly assessed in this study.¹³⁷  However, prior to assembly with the β subunit, the α chain populated multiple oxidation states, one that remained partially reduced and several in which incorrect disulfide bonds formed, arguing that chaperones are likely to play a role in protecting the unassembled IL-12 α chain. In support of this possibility, an independent study found that several ER chaperones, including PDI, associate with the IL-12 α and β subunits.^138,139  Interaction with the β subunit triggered the oxidative folding of the α subunit,¹³⁷  providing a further example of assembly-mediated oxidative folding, similar to that observed for antibodies and T-cell receptors. Lastly, the oxidative folding of the influenza hemagglutinin (HA) ectodomain, which requires the formation of six intrachain disulfide bonds (four involving consecutive cysteines and two that link very distant, non-sequential cysteines), has been determined in exquisite detail. Studies using a combination of techniques, including very short (∼1 min) pulse-labeling conditions,¹⁰³  in vitro translation/translocation assays,¹⁴⁰  DTT washout studies,¹¹³  and the production of stalled translation intermediates¹¹⁰  have provided a detailed understanding of disulfide bond formation during HA biosynthesis. These studies revealed that first relatively short-range disulfide bonds form co-translationally between consecutive cysteines 64/76 and 97/139. Once a full translation product has been made, a long-range disulfide bond forms first between cysteines 52/277, stabilizing the folded head domain. A second bond, connecting even more distant cysteines 14/466, forms after the mid- and stalk regions fold, snapping the native monomeric structure together in a conformation suitable for trimerization and transport to the Golgi. This folding trajectory immediately reveals several N-terminal cysteines that remain reduced and vulnerable until very late in the maturation of HA. Correspondingly, HA maturation relies on the lectin chaperones calnexin and calreticulin to protect these regions,^141,142  and also their dedicated oxidoreductase, ERp57, which is required specifically for the post-translational oxidative events.¹⁴³  Whereas the complex maturation of HA has been almost entirely elucidated through cell- or microsome-based assays, on-column oxidative refolding of the recombinant HA1 top domain to a state that can be recognized by a native conformation-specific antibody has been obtained using GSH/GSSG-containing buffers.¹⁴⁴ 

The third paradigm in oxidative folding is exhibited by proteins in which non-native disulfide bonds are a mandatory intermediate in maturation and must be broken to allow native bonds to form as folding progresses (Figure 1.1.3). As such, this group of proteins should rely on PDI family members with oxidoreductase activity. Because these intermediate bonds cannot be inferred from structural data, it is unclear how many proteins possessing intramolecular disulfides belong to this group. The detailed and complex studies required to detect these intermediates in vivo have been performed for only a handful of proteins. Data on the maturation pathway of the low-density lipoprotein receptor (LDLR) is the most complete. The ligand-binding domain of LDLR is comprised of seven cysteine-rich repeats (LR1–LR7) that ultimately will be stabilized by three intra-repeat disulfide bonds, all of which are formed between non-consecutive cysteines.^145,146  The ligand-binding domain is followed by three EGF-like domains, each with three disulfide bonds that are arranged differently than in the LR repeats, around a β-propeller structure. However, to achieve this native state, the full-length LDLR protein first collapses into a compact structure with many non-native disulfide bonds occurring between cysteines in various LRs and with cysteines in the EGF-like domains.^147,148  The molecular chaperone BiP binds transiently to nascent LDLR¹⁴⁹  and ERdj5, an ER reductase, is required to reduce the non-native bonds so that the correct mature bonds can be formed.¹⁵⁰  The LRs each have a number of conserved amino acid residues that coordinate a calcium ion.¹⁴⁶  Mutation of these acidic residues affects disulfide bonding within the LRs and has been linked to familial hypercholesterolemia.¹⁵¹  Correspondingly, in vitro refolding of isolated LRs can be achieved in the presence of calcium under conditions that allow disulfide exchange.¹⁵²  The Amaranthus α-amylase inhibitor, a member of the cysteine-knot family of proteins that possess three intramolecular disulfide bonds, is a second likely member of this group. Although in vivo studies of the type performed on LDLR have not been conducted on this protein, in vitro refolding studies revealed the formation of non-native disulfides between vicinal cysteines that were ultimately resolved to form the native bonds, all of which occur between non-sequential cysteines.¹⁵³  To promote native bond formation, in vitro refolding was performed in a buffer containing guanidinium with 1 mM cysteine/0.05 mM cystine. Although the aforementioned IL-12 α chain also forms non-native disulfide bonds before heterodimerization, these do not seem to be a prerequisite for correct folding, as mutational studies have shown,¹³⁷  hence care must be taken in interpreting the role of non-native disulfide bonds in protein folding. Lastly, evidence for the presence of non-native disulfide intermediates in protein maturation extend all the way to bacteria. The Escherichia coli lipopolysaccharide transport protein D (LptD) forms a complex with LptE that is critical for transporting LPS from the periplasm to the cell surface of Gram-negative bacteria. In vivo studies found that LptD forms a number of non-native disulfide bonds through the action of the oxidase DsbA and that subsequent formation of native bonds is triggered upon association with LptE.¹⁵⁴  The fact that examples of this class of oxidative folding have been identified from bacteria to humans and in proteins that are members of large families strongly suggests that non-native disulfide bonds may be required for the maturation of many proteins.

Mutations in tryptophans and cysteines have the highest probability of causing diseases in humans among all possible amino acid mutations.¹⁵⁵  This can be attributed to the regulatory roles of cysteines (see, e.g., Chapter 2.3 and ref. 156), but most prominently to their structural roles in proteins of the secretory pathway. The broad spectrum of human diseases associated with cysteine mutations can be seen in Table 1.1.1, which can show only a subset of mutations involving cysteine residues that have been associated with human pathologies.

Although very heterogeneous in the types of proteins affected, and also the resulting disease phenotypes, some general characteristics can be deduced from the analysis of the proteins in Table 1.1.1:

• Mutations involving cysteines generally introduce or delete a single cysteine residue, leading to an uneven number of cysteines in the protein. On the one hand, if the cysteine was part of a disulfide bond, this may destabilize the protein under investigation. On the other hand, this will lead to an unpaired cysteine residue in the mutated protein, which becomes free to interact with other cysteines either in the same protein, another copy of the same protein or other proteins in the ER. If mutations in disulfide bonds lead to partial folding, the problem may be further exacerbated as partially folded states of proteins are generally particularly prone to misfolding and mis-assembly.

• Most of the disulfide mutants listed in Table 1.1.1 are retained in the ER, and in many cases targeted for degradation, consistent with ER quality control recognizing the aberrant proteins. This quality control is exerted by molecular chaperones, which retain their clients by binding to exposed hydrophobic sites in non-native proteins,^157,158  and by PDI family members, which retain proteins in the ER by binding to partnerless cysteines (due to an odd number or misfolding) (see Sections 3 and 4 of this book).

• Mutated proteins can be roughly divided into two classes: loss of function or gain of function. For each, two major scenarios exist. Loss of function is caused by either ER retention and degradation, as outlined above, or by a failing quality control, which releases dysfunctional mutants to their native location (as, for example, patient mutants of classes 3 and higher in cystic fibrosis and familial hypercholesterolemia). Gain of function is caused either by a conformer with aberrant signaling properties (as, for example, growth factor receptors in cancer), or by (disulfide) mutants that are severely misfolded and not easily degraded because of aggregation, leading to UPR induction and possibly cell death, or amyloid formation, leading to (neurodegenerative) disease.

It will be important to study which of these general features are relevant for each particular human disease that involves cysteine mutations. This will provide a basic idea of in which direction therapies may be developed.

A variety of temporal and structural high-resolution in vitro techniques have provided an exquisite understanding of the principles that govern protein folding. By necessity, however, most of the proteins that have been examined in detail are purified, small, full-length and often single-domain proteins, although several more complex folding scenarios have been elucidated, as discussed in this chapter. The folding of a protein in the cell, and particularly in the ER, occurs under vastly different conditions (i.e. time scales, concentrations and protein composition) and begins even before the full-length protein has been translated. Nonetheless, it is important to reiterate that the principles identified in vitro are obeyed by proteins that fold in a cell, although the minute details in their folding trajectories have remained fairly obscure owing to a lack of similar high-resolution in vivo methods. A number of relatively large proteins translated to full length in the presence of DTT have been shown to fold to their native structure when DTT is removed from the cell, despite the fact that these proteins normally fold co-translationally. This argues that the information encoded in the polypeptide chains remains a critical element of protein folding in the cell, similar to and further validating conclusions derived from in vitro studies. Finally, the question remains of whether disulfide bond formation contributes to folding or whether it merely reflects the acquisition of a folded state. We discussed cases where the retention of disulfide bonds during denaturation greatly enhanced in vitro refolding, and also cases where the presence of these bonds hampered the ability to reach a native state. Similarly, there are examples of proteins synthesized in the ER in which disulfide bond formation indicates that a natively folded state has been achieved, but there are also proteins in which disulfides can occur in a domain that does not fold correctly or even proteins with non-native bonds as an intermediate in the folding pathway. Hence examples exist for each scenario in the cell: Folding may drive disulfide formation – but also vice versa. And even disulfide bonding coupled to temporary misfolding may drive the formation of native structure. A combination of complementary in vitro and in vivo assays and the development of higher resolution methods to study the maturation of proteins in a cell will be required to understand oxidative folding pathways fully and how these are compromised in many disease states.

The early sections of this chapter are an update from the book Oxidative Folding of Peptides and Proteins, published by the Royal Society of Chemistry in 2009. M. J. F. is a Rudolf Mössbauer Tenure Track Professor and as such gratefully acknowledges funding through the Marie Curie COFUND program and the Technical University of Munich Institute for Advanced Study, funded by the German Excellence Initiative and the European Union Seventh Framework Program under Grant Agreement 291763. I. B. gratefully acknowledges funding by the Netherlands Science Organization (NWO), Chemical Sciences, and the NWO Graduate Program. L. M. H. is grateful for support from National Institutes of Health Grant R01 GM54068 and by the American Lebanese Syrian Associated Charities of St Jude Children’s Research Hospital.