Effects of localized interactions and surface properties on stability of protein-based therapeutics

Abstract

Objectives

Protein-based therapeutics garner significant attention because of exquisite specificity and limited side effects and are now being used to accomplish targeted delivery of small-molecule drugs. This review identifies and highlights individual chemical attributes and categorizes how site-specific changes affect protein stability based on published high-resolution molecular analyses.

Key findings

Because it is challenging to determine the mechanisms by which the stability of large, complex molecules is altered and data are sparse, smaller, therapeutic proteins (insulin, erythropoietin, interferons) are examined alongside antibody data. Integrating this large pool of information with the limited available studies on antibodies reveals common mechanisms by which specific alterations affect protein structure and stability.

Summary

Physical and chemical stability of therapeutic proteins and antibody drug conjugates (ADCs) is of critical importance because insufficient stability prevents molecules from making it to market. Individual moieties on/near the surface of proteins have substantial influence on structure and stability. Seemingly small, superficial modification may have far-reaching consequences on structure, conformational dynamics, and solubility of the protein, and hence physical stability of the molecule. Chemical modifications, whether spontaneous (e.g. oxidation, deamidation) or intentional, as with ADCs, may adversely impact stability by disrupting local surface properties or higher order protein structure.

Introduction

Antibody drug conjugates (ADCs) are the primary means of achieving site-specific delivery of cytotoxic agents, but the impact of conjugation on antibody structure and stability may limit the molecule from making it to market. ADCs currently in clinical development typically utilize conjugation through surface-accessible lysine residues,^[1,2] which alters the surface charge of the molecule, or through selective reduction or amino acid mutation to conjugate to cysteine residues.^[3–5] Altering the surface charge through lysine conjugation directly affects the stability of the molecule.^[1,6] Amino acid mutation to achieve site-specific conjugation also causes changes in surface properties, but site-specific placement can be optimized to avoid deleterious effects on physical stability caused by local structural changes that perturb the stability of the entire molecule.^[3,5] Because high-resolution data are difficult to obtain due to the large size of antibodies and are in limited supply in the literature, mutational studies with smaller, therapeutically relevant proteins such as insulin and erythropoietin have been considered herein as useful model systems from which to extract general principles of stability. In the studies selected, the effects have been investigated at a detailed chemical level, making it possible to assess the effect on structure and of surface properties, as occurs with conjugation, on the physical stability of the molecule. Mutations are akin to modifications required in ADCs because they alter the surface properties of the molecule, which in turn affects physical stability. Although conjugation attaches a larger and often more hydrophobic appendage than any natural amino acid, and as such the extent or mechanism by which stability is affected may not be the same in all cases, these smaller proteins can serve as model systems for understanding the impact of modification on protein structure and stability within ADCs.

Protein structure is modulated by different types of interactions among residues, with some interactions exhibiting greater contribution to protein structure than others. Secondary structural elements, such as α-helices and β-sheets, are formed by a network of hydrogen-bonding interactions among atoms located within the polypeptide backbone. The hydrogen bonds are essential in maintaining secondary structure, and a lack of hydrogen bonds is associated with or may lead to disordered structures.^[7] Formation of the three-dimensional fold of the protein requires additional interactions among amino acid side chains. Hydrophobic association is critical to achieving the condensed state of a folded protein, but ionic interactions between charged residues on the surface of the molecule are important contributing factors in protecting the core to stabilize the protein fold. Covalent bond formation between Cys side chains in the form of disulfide bonds cross-bridge the structure, which also can contribute significantly to the overall protein fold and its stability. The core structure, packing interactions and surface composition impact the physical and structural stability of proteins. The surface properties determine solvent accessibility to key structural elements, which contribute to the overall three-dimensional conformation and aggregation propensity of the molecule.^[7]

Secondary structural features and disulfide bonds are the main structural elements that are monitored when assessing changes in protein structure. These features and changes in them may be observed using high-throughput approaches, which enable rapid comparison among analogues. When coupled with thermal titration, this approach has been particularly useful because as the largely hydrophobic tertiary contacts are broken, individual secondary structure elements are disrupted.^[7] Amino acid mutation can be used to modulate these elements in a site-specific manner to investigate chemically specific positional contributions to stability and in protein engineering to generate a more stable structure. An individual mutation may cause diverse outcomes ranging from local structural perturbations that lead to structural stabilization to far-reaching alterations that affect the global conformation of the molecule. Structural stabilization may be increased by the addition of stabilizing interactions, such as salt bridges, hydrogen bonds or hydrophobic associations, but amino acid mutation may also be destabilizing to protein structure due to the elimination of a disulfide bond or glycosylation site, as observed in antibodies.

Addition of non-protein entities to the protein also may be pursued to develop a protein therapeutic that exhibits the desirable pharmacological properties or therapeutic use, such as a drug molecule in ADC development. A larger molecular weight species such as a polyethylene glycol (PEG) or lipid molecule can be attached to the protein to increase the molecular weight of the entire molecule, thus increasing the circulating half-life and improving the pharmacokinetics of low molecular weight proteins that are otherwise too quickly cleared from the bloodstream to accomplish optimally their desired function.^[8–12] Glycans have been shown to assist in stabilizing protein structure^[13,14] by interacting with the protein directly through hydrogen bonds with the polypeptide backbone or through interactions with surface-accessible amino acids,^[15–17] although the extent of stabilization is highly protein dependent. Alternatively, glycans and PEG molecules can modulate the physical stability of the protein by shielding hydrophobic regions of the surface that may otherwise be prone to self-association.^[10,18–20]

Even though amino acid mutation and addition of larger molecules to a protein may seem only to affect nearby residues, alterations in the local structure of the protein can modify the structure and physical stability of the entire molecule. As has been observed in many studies, these changes can have detrimental effects on protein function, which is undesirable for the use of a protein as a therapeutic entity (Table 1). As such, assessing the impact of local alterations on protein structure is critical during the development of protein therapeutics. This review provides an overview of how the structural stability of current smaller protein therapeutics such as insulin, erythropoietin and the interferons is affected by amino acid mutation and chemical coupling as a means of better understanding how chemical conjugation in ADC generation may affect both antibody structure and physical stability.

Table 1.

Compiled list of relevant papers

Type of modification	Significant findings	References
Hydrophobicity	PheB24 vital in maintaining insulin core fold	[28,46]
	Presence of Gly151 in Epo allows for correct arrangement of core	[47,48]
	Non-polar residues within the 63–70 segment of Epo responsible for making hydrophobic contacts that hold helices in correct conformation	[47,48]
	Aromatic interactions critical in maintaining core fold in smaller proteins	[29,47,49]
	Hydrophobic surface area of ADCs increases upon conjugation due to conformation adjustments upon conjugation	[40]
	Increasing DAR ratio of ADCs leads to a decrease in colloidal stability before stressing the molecule due to increased surface hydrophobicity	[40]
	Substitution of an aromatic residue for a charged residue more significantly alters the core of Epo than simply substitution with a polar residue	[48]
	4SS-insulin structure is stabilized (increased thermal stability) due to decrease in hydrophobic surface area	[56]
Disulfide bonds	Conjugation through disulfide cysteine residues leads to increased solvent accessibility and structurally flexibility within CH2 domain and CH2/CH3 interface	[39]
	Lower CH2 thermal stability and higher fragmentation observed when conjugation occurs through disulfide cysteine residues	[3,5,40]
	Unfolding at CH2 domain occurs even when conjugation occurs in Fab region, illustrating the far-reaching effects conjugation can have on antibody structure	[5]
	Altering disulfide pattern of IgG4 to be more similar to IgG1 increases the thermal stability of IgG4	[62,63]
Surface charge	Mutation of charged residues within Epo may cause alterations in 3D fold due to charge–charge repulsion of mutated residues or salt bridge elimination	[47,48]
	Electrostatic interactions responsible for correctly orienting insulin chains	[67]
	Substitution of Arg103 to a negatively charged residue increases the thermal stability of Epo, whereas substitution to polar, non-negatively charged residues has no effect	[48,68,69]
	Increase in positive charge on surface of Epo improves thermal stability and decreases the aggregation rate	[18,47,71]
	Presence of charge-dense regions with mAbs may lead to preferentially alignment of regions of opposing charge and increased aggregation propensity	[40,74]
	Conjugation to lysine residues within T-DM1 structurally destabilizes CH2 domain	[1]
Hydrogen bonding	Folding stability of insulin increased by addition of non-native hydrogen bonds	[36,49,56]
	Additional hydrogen bond may not impart stability increases if non-polar side chains are exposed to solvent	[49]
	Mutation of Met252 to ox-Met, Tyr or Gln within mAb leads to disruption of hydrogen bond within beta-sheet, thus decreasing physical stability	[8,77,79]
Glycosylation	Structure of Epo not modulated by glycosylation	[18,97]
	Native glycosylation at Asn80 of IFN-β-1 necessary for forming 3D fold	[19]
	Glycan entities on protein surface increase structural stability by limiting mobility	[18,19,81,83]
	Addition of four non-native glycans within IFN-α-2 substantially increases the thermal stability	[82]
	Glycan within IgG1-Fc directly interacts with the protein surface to stability conformation	[16,17,104]
	Glycosylation may increase structural stability by shielding hydrophobic patches	[18,103]
PEGylation	PEGylation of B29 within insulin decreases structural flexibility more so than modification at N-terminus	[84,85]
	Random lysine PEGylation of IFN-α-2b decreases temperature-induced aggregation twofold by sterically limiting molecule interaction	[127]
	PEGylation increases physical stability of antibody fragments	[128]

Modulation of structural elements through mutation

Mutations that alter key structural elements

The likelihood of amino acid side chains to associate with water is a primary factor governing protein folding and physical stability. Non-polar amino acids are preferentially excluded from the protein surface,^[21–24] resulting in a hydrophobic association among these residues to form the core of the molecule and drive the formation of the tertiary structure. Therefore, mutation of residues that contribute to the hydrophobic core alters the structure and stability of the molecule with the degree of alteration dependent on the amount of energy contributed to maintaining the core of the molecule. Mutations within insulin have been shown to alter the hydrophobic surface area of the molecule. The effect of mutations on the monomeric structure of insulin is commonly evaluated using two mutants of insulin, 4E and DKP insulin, as insulin primarily exists in either dimeric or hexameric form without these mutations.^[25–35] In 4E insulin, ValA3Gly and ThrA8His/Arg mutation results only in local structural alterations that have opposite effects on the hydrophobic content of the region (Figure 1).^[36] The A3 mutations within DKP insulin also suggest that non-polar interactions are essential for maintaining structural stability. Although ValA3Thr mutation did not affect protein conformation, it did alter the denaturation stability of the molecule due to the introduction of a polar moiety in a region of high hydrophobicity,^[37] which alters the packing efficiency and solvent accessibility in the region. Even if a residue seems to contribute minimally to protein structure or the hydrophobic core of the molecule, it may exert a large effect on protein structure by protecting buried, stabilizing interactions. Elimination of the protection provided by specific residues leads to greater solvent accessibility and disruption of the stabilizing interaction, which adversely affects protein stability.^[38] As conjugation of the toxin to the antibody can lead to the elimination of the hinge-region disulfide bonds, structural perturbations postconjugation would be expected. Pan et al.^[39] illustrated that the C_H2 domain and C_H2/C_H3 interface exhibit increased solvent exposure and structurally flexibility in comparison with the antibody itself. This increase in solvent accessibility to the core of the molecule could provide explanation as to why cysteine conjugation has been shown to result in conjugates with lower thermal stability within the C_H2 domain and higher fragmentation propensity.^[3,5,40] Similar decreases in physical stability upon disruption of the protein core are observed in smaller proteins. Within insulin, PheB24 is important for maintaining the hydrophobic core of the molecule as it packs against ValB12 and LeuB15.^{[25,27,33,34,41–43]} Substitution of this residue with non-aromatic l-amino acids or aromatic residues alters the solution structure of the molecule, which leads to decreases in bioactivity.^[31,44,45] In the PheB24His mutant, the His side chain is more tightly packed into the B24 hydrophobic pocket than Phe due to the ability to form additional hydrogen bonds.^[45] The PheB24 to Gly mutation also drastically decreases the folding stability of the molecule. NMR illustrates that the core structure and α-helical content of the PheB24Gly mutant remain, but the mutant is more flexible, which leads to instability and unfolding of the B20-B30 region.^[28,46] Within erythropoietin (Epo), mutation of Gly151, which contributes substantially to core formation by bringing the side chain of Lys152 into contact with Val63, Trp51 and Phe148,^[47] results in global rearrangements within the hydrophobic core of the molecule.^[48] In addition to maintaining the core of the molecule, hydrophobic residues are also important for forming interactions that stabilize different domains of Epo. Mutation of Leu69, Leu70 and Ala73, which make important hydrophobic contacts with Ile39 and Val41,^[47] alters protein conformation. Leu69Ala resulted in only slight conformational changes, whereas Leu70Ala, but not Leu70Ile, substitution resulted in drastic conformational changes of Epo,^[48] illustrating the necessity of a larger non-polar side chain at this position. Substantial conformational changes are also observed upon substitution of Val63, Leu67 and Val74 with Ser.^[48] These residues are buried within the helices and make hydrophobic contacts holding the helices together, thus confirming their importance.^[47]

Figure 1.

A3 and A8 mutations affect surface hydrophobicity by altering helicity of local segment. Mutation of ValA3 or ThrA8 only causes local structural alterations in the A2–A8 helix of the A-chain. The mutated residues (ValA3 and ThrA8) are shown in orange in all three structures, and the side-chain of the cysteine residues are shown in yellow. ValA3Gly mutation decreases the α-helical character in the A2–A8 region, which causes decreased hydrophobic surface area in this region due to elimination of the ValA3 side chain and shifting of the GluA4 side chain to encompass a region of the protein that was predominantly hydrophobic in nature in the native structure. ThrA8His/Arg mutation leads to tightening of the A2–A8 helix due to the formation of a non-native hydrogen bond between the His/ArgA8 and GluA4 in the mutant and increased exposure of the hydrophobic residues IleA2 and ValA3 without adversely affecting the global structure. Modified and reproduced with permission from Olsen et al.^[36] Elsevier, Philadelphia, PA.

The presence of aromatic side chains, as opposed to solely hydrophobic side chains, may be necessary to form the most stable structure of a protein. Within insulin, removal of PheB1 eliminates the packing interaction with LeuA13 and causes destabilization of the B1–B4 region.^[29] Mutation of TyrB26 to Thr also decreases folding stability due to the elimination of stabilizing aromatic interactions.^[49] The importance of aromatic interactions in maintaining global structure is also observed within Epo and the interferons (INFs). The side chains of Phe142/Tyr145 and Phe148/Tyr156 interact with hydrophobic residues located on opposite helices, to form the hydrophobic core of the Epo molecule (Figure 2).^[47] Substitution of these residues causes disruption of the core, leading to protein unfolding.^[48,50,51]

Figure 2.

Four aromatic residues within Epo make large contributions to protein core. Substitution of the four primary residues responsible for maintaining the hydrophobic core leads to substantial conformational rearrangements. Phe142Ile mutation causes slight perturbations in global conformation, whereas mutation of Tyr145 to Ile or Phe causes substantial conformational rearrangements. Therefore, Phe142 must primarily make hydrophobic interactions, so Ile will not alter the structure, but placement of Tyr at position 145 is necessary for both aromatic interactions, as determined by the Ile substitution, as well as a type of dipole interaction or stabilizing hydrogen bond because substitution with Phe at this site also resulted in conformational rearrangement. Phe148Val or Tyr156Ile/Phe mutation confirms that a larger hydrophobic entity is required at this site because Val/Ile substitution results in substantial alterations in global conformation, but Phe substitution does not. Phe148Tyr substitution results in slight conformational changes, which could be caused by the addition of the hydroxide moiety into the hydrophobic pocket, whereas elimination of the moiety in the Tyr156Phe substitution would not be expected to alter core structure. PDB file 1BUY was used to generate this figure.

As the exposure of hydrophobic entities to a water-based environment is unfavourable and destabilizing,^[33,34] an increase in hydrophobic surface area would be expected to decrease the physical stability of the molecule. Many of the toxins currently used in the formation of conjugates are highly hydrophobic in character. Therefore, it is not surprising that the hydrophobic surface area increases with increasing payload (drug-to-antibody ratio, DAR).^[40] Greater exposure of hydrophobic surface area within ADCs is not due solely to the hydrophobic nature of the payload, but also due to conformational perturbations introduced upon disulfide loss during conjugation, which increase the hydrophobic surface area of the protein as a whole.^[40] On the other hand, substitution of solvent-exposed non-polar side chains for more polar residues decreases the exposed hydrophobic surface area and would be expected to increase the physical stability. Introduction of Asp, His or Ser into the B25 position of insulin increases the folding stability due to subtle rearrangements in the global structure.^[31,49] The side chain of PheB25 is pointed towards solvent, suggesting it does not contribute significantly to maintaining protein structure, and implying that the increases in folding stability are due to the elimination of hydrophobic surface area. The introduction of a polar side chain into the hydrophobic core will have the opposite effect and destabilize protein structure, as is illustrated by the conformation changes observed within Epo upon substitution of core residues Ala98 or Leu105 with Ser.^[48] The Leu105Asp mutation causes more drastic conformational changes than the Leu105Ser mutation,^[48] suggesting that, as expected, the core of Epo is more destabilized by the introduction of a charged residue than simply a polar residue. Charged residues are more water soluble than polar residues and in order for these to be stably contained within the hydrophobic core, salt bridges or hydrogen-bonding requirements must be satisfied.^[52–55] Structural rearrangements in the four-disulfide insulin mutant (4SS-insulin) led to a drastic decrease in the solvent-exposed hydrophobic surface area due to mutation of the hydrophobic residue Ile to Cys and additional shielding of hydrophobic residues,^[56] thus increasing the physical stability of the monomer. The T_m of the four-disulfide mutant is 34.6 °C greater than that of native insulin, and extensive shaking causes complete fibrillation of the native molecule, whereas the mutant exhibits none,^[56] illustrating that improvements in physical stability can be made by decreasing exposed hydrophobic surface area.

In addition to modulation of core structural elements through mutation of hydrophobic residues, protein structure may also be affected by disulfide bond disruption or addition, which is commonly performed in the generation of ADCs. Drug conjugation to native cysteine residues can be performed after breaking intermolecular disulfide bonds made accessible through selective reduction.^[3,5,40] Non-covalent interactions maintain the structure near the hinge region in the absence of the disulfide bonds,^[57] but the greater fragmentation rates observed at higher ionic strength confirm that these interactions are not sufficient to maintain protein structure.^[5] Even though the quaternary structure is well-maintained, elimination of these structural features and introduction of a hydrophobic entity at the site could cause local structural changes that cause the molecule to be more susceptible to unfolding when thermally stressed.^[3] The rapid aggregation of cysteine-conjugated ADCs and lower T_m of the C_H2 domain indicate that structural alterations have occurred within the molecule,^[3,5,40] although they are not easily observable with low-resolution techniques such as CD. It is suggested that the C_H2 domain is the cause of aggregate formation due to its instability in the absence of the disulfides and its presence as a partially unfolded domain within the aggregate.^[3] The number of conjugated entities or DAR has also been shown to influence the physical stability of the molecule, confirming that the decreased physical stability of ADCs is due to both elimination of disulfides and introduction of a hydrophobic entity on the surface of the protein structure.^[40] The decrease in T_m observed within higher DAR conjugates suggests that the number of drug entities, as opposed to disulfide elimination, causes the largest impact on protein structure. Unfolding of the C_H2 domain has also been observed at low DAR,^[5] where conjugation primarily occurs in the Fab region, illustrating the far-reaching effects that surface modification can have on the structural stability of ADCs.

Engineered disulfide bonds have also proved useful in increasing the thermal stability of antibody fragments^[58–61] and whole antibodies. For example, modulation of the disulfide network within the Fab domain of IgG4 leads to increased thermal stability of the entire antibody.^[62] Although the Fab domain of IgG4 has a drastically lower thermal stability than IgG1,^[63] altering the interchain disulfide pattern of IgG4 (DSB between LC Cys214 and HC Cys229 as opposed to HC Cys214) to be more similar to that of IgG1 increases the thermal stability of IgG4 to similar values observed for IgG1. Although CD analysis indicates only minimal structural changes accompany these substitutions, this low-resolution method is not sufficient to probe how local structural perturbations cause increased thermal stability. Additional investigation into the physical stability of the disulfide-mutants could provide details regarding the key structural interactions necessary for stabilization, and if the potential exists to further modulate the disulfide network of other therapeutic proteins to increase their physical stability.

Surface charge alterations

Charged residues help stabilize protein structure through electrostatic interactions pairing residues of opposite charge. This can occur on the protein surface^[53,64,65] or in the form of internally buried salt bridges.^[53,66] Because these interactions contribute to local structural elements within domains or assist in structure formation by connecting domains, mutation of these residues can cause substantial conformational changes. Large changes in protein conformation have been observed due to electrostatic repulsion when an oppositely charged residue is introduced in close proximity to a similarly charged residue. The extreme case of this effect is apparent in many intrinsically disordered proteins (IDPs). Within Epo, alterations in the three-dimensional fold observed upon Gln78Glu, His94Glu or Lys97Glu/Asp substitution^[48] are due in part to charge repulsion between the mutated residues and nearby similar charges (Figure 3).^[48]

Figure 3.

Alterations in the three-dimensional fold of Epo are caused by charge repulsion. Mutation of residues shown in red to Glu causes conformational changes because it leads to charge repulsion between the Glu78 mutant and Asp96, and repulsion between the Glu94 and Glu/Asp97 mutants and Glu18 or Glu21. Ala substitution at position 78 or 97 results in slight structural changes, suggesting that the substantial alterations in protein conformation observed upon introduction of Glu in these regions are not solely due to charge repulsion, but may also be caused by the elimination of stabilizing interactions such as a hydrogen bond or salt bridge. PDB file 1BUY was used to generate this figure.

Mutations that lead to elimination of ionic interactions, even on the solvent-exposed surface, often are destabilizing, and likewise, the introduction of electrostatic interactions is stabilizing for protein structure. Charged residues within the B20–B23 loop of insulin stabilize the tertiary conformation by forming electrostatic interactions with C-terminal A-chain residues to correctly orient the two chains of the molecule. Removal of these interactions through mutation of GluB21 or ArgB22 to Ala results in a 25% decrease in protein expression levels relative to native insulin,^[67] confirming that interactions within this loop are important in maintaining protein structure. The global structure of Epo is also substantially altered upon elimination of a charged residue (Glu23Ser).^[48] As Glu23 is buried between a helix interface^[47] and forms a stabilizing salt bridge, Ser substitution eliminates this interaction. The results observed with insulin and Epo confirm the importance of salt bridges in maintaining protein structure. In contrast to the destabilizing effects observed by salt bridge elimination, addition of a non-native salt bridge through amino acid mutation is stabilizing to protein structure. Deletion of ThrB27 within insulin causes rearrangements that position LysB29/GluA4 side chains and the N-terminus of the A-chain and C-terminus of the B-chain in close proximity to one another. Therefore, the increase in folding stability observed with this mutant may be caused by the addition of two non-native interaction sites.^[49] Modulation of the thermal stability of Epo has been investigated by mutation of Arg103. Ala substitution causes minimal structural changes,^[48,68] whereas introduction of a negatively charged side chain through Glu/Asp mutation results in substantial increases in thermal stability.^[68,69] Multiple mutations were tested at this site to determine their effect on thermal stability with Asn, Asp, and Glu substitution increasing the stability, Ala and His substitution having similar stability as the native molecule, Gln substitution slightly decreasing the stability and Lys substitution substantially reducing the stability.^[69] Even though positively charged residues were nearby, the exact interaction(s) responsible for the increase in thermal stability has not been determined but likely results from electrostatic attraction.^[69] Charged residues can exert a field effect in which one charged moiety interacts with several oppositely charged residues within the bonding radius to further stabilize the protein conformation.^[53,70]

The physical stability of protein therapeutics can also be improved by increasing the charge on the protein surface. This principle has been applied to non-glycosylated Epo, which has a lower stability than its glycosylated counterpart.^[18] Mutation of the N-glycosylation sites to Lys results in a mutant with similar structural properties as native Epo produced in Escherichia coli,^[12,34,35] but improved thermal structural and aggregation stability and solubility in comparison with native, non-glycosylated Epo.^[18,47,71] The denaturation stability of the two variants is similar, and the observed T_m was also determined to be scan-rate dependent.^[71] The data suggest that the increase in net charge does not necessarily rigidify the protein fold to impart stability; rather, it alters the temperature-dependent aggregation kinetics of the unfolded state by decreasing the aggregation rate for the Lys mutant in comparison with the native protein.^[71] Although increasing the net charge within Epo increases physical stability, the effect of modifying the surface charge within antibodies is more complex because of the high-concentration formulations required of antibodies. The effect of surface charge modification on antibody stability is also influenced by the properties of individual antibodies.^[72] At higher concentration, the overall net charge and specific arrangement of charged moieties can greatly impact other important properties such as viscosity. Investigations into multiple antibodies have shown that modification of the net surface charge within MAb-1 in particular to cause unique effects on physical stability not observed with other antibodies.^[40,73,74] Mutation analysis revealed that this antibody contained regions of highly localized surface charge, whereas the other antibodies contained a larger dispersion of surface charge (Figure 4).^[40,74] MAb-1 is more likely to self-associate because the highly charged area is able to align preferentially and interact with a region of opposite charge, which is not possible when the surface charge is more diffuse. Therefore, removing charged residues within charge-dense regions could decrease the self-association propensity of the molecule, allowing for formulation at high concentrations.

Figure 4.

Regions of highly localized surface charge affect the physical stability of antibodies. MAb-1 (a) contains a dense region of negative surface charge, which, when aligned with regions of positive charge on other molecules, can lead to self-association and aggregation. On the other hand, MAb-2 (b), which has a more diffuse charge in this region, is not prone to self-association. Modified and reproduced with permission from Yadav et al.^[129] ACS Publications, Washington, D.C.

Akin to mutation, chemical modification alters the surface properties of a protein. Although the compounds attached induce a significantly larger change in size than mutation, chemical principles still apply. The most noted property is decreased solubility, due to the relatively more hydrophobic character of the linkers and drugs. Self-selection occurs, where higher DAR species precipitate during the reaction, such that an average DAR is typically below 4. The net surface charge commonly becomes more negative upon the formation of conjugates because the most commonly used method relies on coupling with surface-exposed lysine residues, incorporating the basic amine into a neutral amide bond. In addition to generating a heterogeneous mixture of molecules, multiple charge states have been observed with modification of a single lysine, depending on the position of the residue within the protein structure,^[75] further complicating structural analysis. In some cases, this may be due to the formation of different chemical isomers upon linking at the same residue. Conjugation to lysine residues may also impart structural modifications that negatively impact the physiochemical properties of the molecule. Addition of the drug entity to lysine residues in T-DM1 caused destabilization of the structural stability of the C_H2 domain, which is the least thermally stable domain in antibodies and is where a disproportionately high amount of conjugation occurs.^[2] Alterations in the structure of this single domain led to decreased thermal stability of the entire conjugate in comparison with the parent antibody.^[1] In addition, it was shown that the distribution of Lys conjugation sites is not random or always correlated with surface accessibility.^[76] A single Lys in trastuzumab is preferentially modified, likely due specific local interactions that enhance its reactivity. As such attachment at this site may potentially disrupt stabilizing interactions. Although lysine conjugation was shown to alter the thermal stability of T-DM1, it has a more limited effect on thermal stability than thiol conjugation.^[6] This difference may be due to the more severe structural alterations caused by the modification of cysteine residues. In both cases, the number of drug molecules attached is low, particularly with respect to the large size of an antibody and the high number of basic residues.

Importance of hydrogen bonding to stability

Amino acid mutation may also result in the introduction of non-native hydrogen bonds that stabilize protein structure. In native insulin, PheB25His/Ser mutation leads to an increase in folding stability due to subtle structural rearrangements that lead to an additional hydrogen bond between the substituted residue and TyrA19.^[31,49] ThrA8His/Arg mutation also introduces an additional hydrogen bond within insulin between A8 and GluA4. A hydrogen bond at this location is not present in native insulin because it would result in the exposure of the non-polar portion of the Thr side chain to water, which is highly unfavourable.^[49] This additional hydrogen bond present in the mutant tightens the helix (Figure 1), not altering the global conformation, but increasing the folding stability of the molecule.^[36,49]

Addition of hydrogen bonds also can lead to increases in structural stability of therapeutic proteins. Mutation of HisB10 to either a Glu or Asp introduces the potential for another hydrogen bond within the molecule (Figure 5). The additional interaction leads to large increases in the chaotropic denaturation stability of the molecule.^[49] Even though HisB10Thr mutation also provides the potential for a similar stabilizing hydrogen bond, Thr substitution decreases the denaturation stability because hydrogen bond formation exposes the methyl side chain of Thr to solvent, which is unfavourable and deleterious to protein stability.^[49] The thermal stability of insulin can also be increased through the addition of non-native hydrogen bonds. Introduction of a fourth disulfide bond within insulin (A10/B4) results in the addition of at least four non-native hydrogen bonds (Figure 6). This mutant has a T_m 34.6 °C higher than the native molecule,^[56] confirming the importance of hydrogen bonds and disulfides in modulating the structural flexibility that can directly impact the stability profile of protein therapeutics.

Figure 5.

HisB10Glu mutation within insulin introduces additional hydrogen bond. Mutation of HisB10 to either Glu or Asp introduces the possibility of an additional hydrogen bond within the molecule between the side chain of Glu/Asp and the backbone amide NH of CysB7. PDB file 1HUI was used to generate this figure.

Figure 6.

Addition of a non-native disulfide within insulin increases thermal stability. With the mutation of A10 and B4 to Cys, an additional disulfide bond is formed between the two strands leading to additional hydrogen bonding interactions. These additional stabilizing interactions drastically increase the T_m of the mutant. Modified and reproduced with permission from Vinther et al.^[56] John Wiley & Sons, Inc., Hoboken, NJ.

Hydrogen-deuterium exchange (HDX) studies on mAbs have shown that chemical modification of Met252 to methionine sulfoxide (ox-Met) in the C_H2 domain and separately mutation of this residue to Tyr result in disruption of hydrogen bonds within the beta-sheet near this position,^[77,78] leading to a decrease in physical stability. Mutation of this residue to Gln, which emulates the chemical properties of ox-Met, produces an equivalent decrease in thermal stability to chemical modification, whereas mutation to Leu stabilizes the mAb and prevents loss of stability in the presence of oxidants.^[79] These detailed studies and the resultant engineering of a more stable molecule reflect clear understanding about the chemical basis of local structure and its implications for stability.

Modulation of stability through chemical modification

Chemical modifications such as glycosylation and PEGylation impact protein stability, but the mechanism of stabilization can be more complex and difficult to elucidate than mutation. The carbohydrate entity or PEG molecule may increase protein stability through direct interactions with the protein or indirectly by shielding the protein from interaction with other protomers, often by occluding regions of hydrophobic surface area.^[15,19,80] Therefore, modulating the glycosylation pattern of proteins, such as Epo and IFN, may afford increased stability, although the means by which this occurs is highly protein dependent.^{[18,19,81–83]} PEGylation provides an additional means of increasing physical stability due to the large size of the PEG molecule, which sterically hinders association among protein molecules.^[84–86]

Glycosylation

Glycosylation, or the covalent attachment of carbohydrate moieties to the protein surface, can occur within a native protein sequence or non-native sites can be engineered into the native sequence. Epo contains three N-linked (Asn24, Asn38, Asn83) and one O-linked (Ser126) glycans that account for nearly 40% of the molecule by weight. Full glycosylation of Epo is reliant upon preservation of the four-helix bundle, as mutants missing helical regions are not fully glycosylated.^[87] Also, the introduction of a bulky carbohydrate chain near the hydrophobic core would be detrimental in achieving the correct fold, which leads to lack of glycosylation near these regions. Glycan addition occurs during protein folding, suggesting it may contribute to folding in vivo with some glycosylation sites contributing more than others.^[81,88–94] Native O- and N-glycosylation in Epo has been shown to have minimal effects on protein structure and secretion.^[95,96] Removal of the O-glycosylation site in Epo through substitution of Ser126 has no impact of the global conformation of the molecule.^[48] Even though elimination of all three N-glycosylation sites causes Epo to be produced mainly in the inclusion bodies in the E. coli system,^[18,87,97] the refolded molecule contains similar secondary and tertiary structural features of the native protein,^[18,97] suggesting that protein structure is not modulated by the glycan moieties, but its stability may be. The addition of non-native carbohydrate groups has also been shown to have minimal influence on the global structure of insulin^[98,99] and IFN-α-2b.^[82] On the other hand, native N-glycosylation of IFN-β-1 at Asn80 assists in the formation of the stable three-dimensional structure.^[19] Two IFN-β variants exist, IFN-β-1a and IFN-β-1b, with the primarily difference between the two variants being the presence and absence of glycosylation. IFN-β-1a is glycosylated at Asn80, whereas IFN-β-1b is produced in an E. coli-based system; thus, it is not glycosylated. IFN-β-1b, which folds in the absence of the glycan, exhibits a T_m 7 °C lower than that of enzymatically deglycosylated IFN-β-1a, which is folded in the presence of the glycan, and then, the glycan is removed.^[19] Removal of the N-glycan in IFN-β-1b also does not affect protein secondary structure once folding is achieved.^[100] These data collectively indicate that the glycan assists in achieving and also stabilizing the desired fold.

The addition of glycan entities on the protein surface increases structural stability by limiting structural mobility.^[81,101] Carbohydrate chain size also contributes to folding stability by increasing the free energy associated with the unfolded state of the protein.^[13,14] This increased folding stability afforded by glycosylation causes measurable differences in thermal stability, as the glycosylated forms of IFN-β,^[19] IL-5,^[83] Epo,^[18] yeast external invertase, bovine serum fetuin and glucoamylase^[81] all exhibit increased thermal stability in comparison with the deglycosylated forms. The N-glycan motifs drastically influence Epo structural stability because their removal caused a decrease in molecular stability, whereas O-glycosylation does not contribute appreciably.^[102] Non-glycosylated Epo is also more susceptible to unfolding in the presence of denaturant, and it is less stable in acidic conditions.^[18,103] Acid-induced denaturation results in elimination of the tertiary structure of the mutant, but not of the glycosylated form, which suggests that the glycan moieties assist in the stabilization of the global structure of the protein at acidic pH.^[18] Glycan molecules also modulate the thermal stability of IFN-α-2, as the variant containing four non-native glycans (Pro4Asn, Arg23Asn, Lys70Asn, Asp77Asn) exhibited a T_m approximately thirty degrees higher than the native O-glycosylated and E. coli generated non-glycosylated variant.^[82] The T_m value of the enzymatically deglycosylated 4N variant is the same as the native protein, confirming that it is the glycan addition and not the amino acid mutagenesis that is responsible for the drastic increase in thermal stability, although a detailed description of the interaction between the protein surface and glycan has not yet been reported. The glycan within IgG1-Fc also directly interacts with the protein surface to stabilize protein conformation. The outer and inner regions of the N-glycan at Asn297 within IgG1-Fc have the largest effect on protein conformation^[104] and are hypothesized to stabilize protein conformation through non-covalent interactions with nearby residues on the protein surface.^[16,17] Removal of the terminal N-acetylglucosamine resulted in the largest reduction in T_m (3.1 °C) with an additional decrease in T_m by 1.4 °C upon complete deglycosylation.^[105] In a separate study, it was also shown that deglycosylated IgG1-Fc exhibits higher physical stability than the non-glycosylated form.^[106] In the solid-state structure, the loop containing the N-glycosylation site and the oligosaccharide are most perturbed by removal of the terminal N-acetylglucosamine or mannose sugar residues,^[104] suggesting an important role of these motifs in maintaining protein conformation.

Even though glycosylation generally increases folding stability, deglycosylation can also increase stability due to alterations in the protein conformation resulting in decreased hydrophobic surface area.^[107] IFN-α-2 contains one O-glycosylation site at Thr106,^[108] and the effect of the O-glycan on the thermal stability of IFN-α-2 is debated. Johnston et al.^[96] reported T_m values of 65.7 and 63.8 °C for the non-glycosylated and O-glycosylated variants of IFN-α-2b, whereas Ceaglio et al.^[82] reported similar T_m values near 66 °C for both forms. The O-glycosylation site is located within a flexible loop, and addition of GalNAc or Gal(β1,3)GalNAc, which are two of the three major glycan moieties found in native IFN-α-2, decreases the conformational dynamics of the loop.^[109] This suggests that loop flexibility is required for structural stability, as the rigidity imparted by the glycan molecule leads to decreased stability. Because the observed difference in thermal stability between the two variants is so small, additional experiments concerning the interactions between the glycan motifs and protein surface would prove useful in determining if the O-glycan is indeed destabilizing in IFN-α-2.

In addition to contributing to protein folding, glycosylation may also increase the physical stability of the molecule. At the high concentrations necessary for protein therapeutics, protein molecules may associate to form aggregates.^[110] Attachment of the glycan to the protein surface increases the solubility of the molecule due to its hydrophilic nature and steric effects,^[13] which leads to decreased aggregation propensity.^[111–113] Introduction of non-native glycosylation sites increases the physical stability of IFN-α-2, as has been shown with other therapeutic proteins.^{[18,98,99,103]}

The carbohydrate moiety may also increase the physical stability by shielding regions of hydrophobic surface area that would otherwise be prone to interact. Severe precipitation of non-glycosylated Epo occurs after incubation at elevated temperatures and the helical content of the remaining soluble portion is non-existent, but the glycosylated protein retains the native structure and remains soluble.^[18,97] ANS fluorescence suggests that non-glycosylated Epo contains more solvent-exposed hydrophobic residues in comparison with the fully glycosylated form.^[18] Simply adding free mimics of the N-glycans to non-glycosylated Epo results in decreased ANS fluorescence, suggesting that the glycans do not need to be chemically attached to have the protective effect.^[103] In Epo, the inner regions of the N-glycans are responsible for interacting with the protein surface.^[103] Removal of only the galactose groups from the glycan resulted in a decrease in denaturation stability and an increase in ANS florescence, suggesting that these groups are primarily responsible for making contacts with the hydrophobic surface area. The glycan also modulates the physical stability of IFN-β-1a.^[19,114,115] Non-glycosylated IFN-β-1a exhibits a T_m 5 °C lower than glycosylated IFN-β-1a.^[19,20] The glycan hydrogen bonds with surface residues, further increasing the structural integrity of the molecule,^[15,19,116] and it also shields a region of uncharged surface area so the observed increases in thermal stability could be due to either of these factors. In native IFN-α-2, glycosylation at this position is not required for maintaining physical stability, as this region contains charged residues that are unlikely to self-associate (Figure 7).^[117,118] In addition, the negatively charged sialic acid residues on the glycan may further decrease the self-associate propensity through electrostatic repulsion between neighbouring molecules, as is observed with Epo.^[119,120] Similarly to the non-glycosylated forms of smaller proteins, the extent of molecule hydrophobicity also modulates ADC stability. As many of the payloads used are highly hydrophobic, the ANS binding rate increases substantially from DAR2 to DAR6 species.^[40] This increase in ANS binding is accompanied by a significant surface tension drop at higher DAR, confirming that the molecule is overall more hydrophobic in nature. The increased hydrophobic surface area within higher DAR species negatively impacts the colloidal stability of the molecules and leads to multimer formation before stressing the molecule.^[40] The use of payloads less hydrophobic in nature could decrease the hydrophobic nature of the molecule and thus increase the colloidal stability future ADCs. Further studies need to be completed using less hydrophobic payloads as most of the reported case studies use some version of the auristatin toxin.

Figure 7.

The glycan moiety within IFN-β-1 increases the structural and physical stability of the molecule. The glycan forms stabilizing hydrogen bonds with residues Gln23 and Asn86 to maintain the structural integrity of the protein, and it also shields a surface-exposed region of uncharged residues. IFN-α does not require a glycan at this location because it contains charged residues in the analogous region, thus eliminating the region prone to self-association. Reproduced with permission from Karpusas et al.^[116] © Springer, New York.

PEGylation

Covalent attachment of PEG to protein therapeutics has been shown to increase serum half-life and solubility, while also decreasing immunogenicity.^[8–12] PEGylation also affects the physical stability of proteins, and PEGylation of smaller protein therapeutics can be used to investigate the structural effects site-specific chemical conjugation at different positions may induce. Conjugation of a PEG molecule or drug entity usually occurs through the amino groups located on the N-terminus of the protein or surface-exposed lysine side chains, but can also be accomplished to thiols, often at non-native, engineered cysteine residues. Conjugation through lysine or histidine residues is usually non-specific and results in a highly heterogeneous product,^{[2,121–124]} whereas conjugation to non-native engineered cysteine residues allows for greater site specificity.^[4,125] The conjugation site is mainly determined by properties of the protein, the number of possible conjugation sites within the protein and the type of conjugation chemistry used.

Site-specific PEGylation using native residues has been accomplished using knowledge of the reactivity of the potential conjugation sites in the protein of interest to develop the reaction conditions necessary for the desired modification site. Insulin contains three potential PEGylation sites (N-terminus of the A-chain, LysB29, N-terminus of the B-chain), and conjugation at either site results in slight structural differences that depend on the conjugation site but not PEG size,^{[84–86,126]}suggesting that the residue modification, and not size of conjugated entity, causes structural perturbations. The disubstituted A-chain N-terminus/LysB29 molecule exhibits a more intense α-helical feature due to the PEG molecule reducing the overall flexibility of native monomeric insulin.^[126] Substitution at LysB29 also causes restriction in the rotation of B-chain aromatic residues,^[84,85] suggesting that conjugation at LysB29 impacts the flexibility of the monomer more so than conjugation at the N-terminus of the A-chain. On the other hand, PEGylation of IFN-α-2 at native lysine or histidine residues has minimal impacts on protein structure. The attachment of a 12-kDa PEG molecule at His34 of IFN-α-2b does not cause substantial changes in the secondary or tertiary structure,^[123,124] and PEGylation at lysine residues is expected to have minimal effects on protein structure, as was the case with IFN-β-1b.^[122]

Although PEGylation has been shown to minimally impact the structure of protein therapeutics, it does modulate the molecule's stability. PEGylation of Epo is hypothesized to occur either at His32 or His94 and drastically increases its thermal stability of unfolding and resistance to aggregation in comparison with the non-glycosylated form.^[97] A twofold decrease in temperature-induced aggregation was also observed with random lysine PEGylation of IFN-α-2b in comparison with the non-PEGylated counterpart.^[127] Because the global structure is not affected by addition of the PEG molecule, additional investigations are needed to determine whether the PEG molecule increases the thermal stability through modifications in local structure due to direct interactions with nearby residues. On the other hand, the thermal unfolding of IFN-β-1b is not influenced by random PEGylation at surface-exposed lysine residues, but the aggregation propensity of the molecule is decreased.^[122] This suggests that in the case of IFN-β-1b, the PEG molecule does not interact with the protein surface; rather it increases the stability of the molecule by sterically interfering with protein self-association. PEGylation has also been shown to increase the physical stability of antibody fragments, allowing for the generation of high-concentration formulations >200 mg/ml without protein aggregation.^[128] Because PEGylation allows for increased physical stability with minimal structural implications, it is a very useful tool for generating protein therapeutics with the desired properties.

Conclusion

The stability of protein therapeutics is modulated by many different factors including structural elements and surface properties, and investigating the effects of changing these factors is of critical importance, particularly for advancing beyond empirical evaluation to implement design of more stable therapeutics. The generation of ADCs requires modulation in both the structure and surface of the antibody so the stability of the conjugate relative to the native molecule must be determined. Because high-resolution data on antibody structure are limited, it is difficult to connect observed changes in molecule's stability to the specific structural elements responsible for altering the stability. As such, smaller protein therapeutics can serve as model systems for assessing the molecular mechanisms behind changes in stability. In this review, the impact of modifying key elements responsible for stabilizing protein structure, such as hydrogen bonds, hydrophobic associations, and ionic interactions, within smaller protein therapeutics was analysed, and the trends observed were considered with respect to larger antibody systems. Mutations within smaller protein therapeutics are akin to chemical conjugation to antibodies because both modulate structural and surface properties, which directly impact stability. Until a significant body of site-specific data is reported regarding the impact of conjugation on the structural stability of ADCs, the trends observed with mutation and modification of smaller protein therapeutics provide foundational principles for understanding structural stability, which can be applied when attempting to design site-specific conjugates and ADCs that exhibit the best possible stability profiles.

Declaration