Factors affecting the physical stability (aggregation) of peptide therapeutics

Abstract

The number of biological therapeutic agents in the clinic and development pipeline has increased dramatically over the last decade and the number will undoubtedly continue to increase in the coming years. Despite this fact, there are considerable challenges in the development, production and formulation of such biologics particularly with respect to their physical stabilities. There are many cases where self-association to form either amorphous aggregates or highly structured fibrillar species limits their use. Here, we review the numerous factors that influence the physical stability of peptides including both intrinsic and external factors, wherever possible illustrating these with examples that are of therapeutic interest. The effects of sequence, concentration, pH, net charge, excipients, chemical degradation and modification, surfaces and interfaces, and impurities are all discussed. In addition, the effects of physical parameters such as pressure, temperature, agitation and lyophilization are described. We provide an overview of the structures of aggregates formed, as well as our current knowledge of the mechanisms for their formation.

1. Introduction

The aggregation of peptides is one of the most common and troubling processes encountered in almost all phases of biological drug development (table 1) [32]. Aggregation can take several different forms and the term is used to describe a number of different processes during which peptide molecules associate into larger species consisting of multiple polypeptide chains (figure 1). Aggregates can be amorphous [33] or highly structured, e.g. amyloid fibrils [34], and can form in solution or on surfaces due to adsorption [35–37]. They can arise as a result of the non-covalent association of polypeptide chains, or from covalent linkage of chains. In some cases, aggregation is reversible while in others it is effectively irreversible. In either case, it reduces the physical stability of the peptide in question, not only leading to a loss in activity but also other critical problems such as toxicity and immunogenicity [38–40].

Table 1.

Some examples of peptide and protein pharmaceuticals that form different types of aggregate.

peptide or protein	length (residues)	conditions	type of aggregate	references
GLP-1(7–37)	31	pH 7.5–8.2	fibril	[1]
GLP-1(7–36)NH2	30	pH 7.4	fibril	[2]
glucagon	29	pH 2.0	fibril	[3–5]
glucagon	29	pH 8.5–9.7	fibril	[6,7]
salmon calcitonin	32	neutral pH	fibril	[8–11]
amylin	37	under different conditions	fibril	[12–16]
insulin	51 (31 chain A, 20 chain B)	pH 2.0	fibril	[17–19]
		pH 1.1–3.2	fibril	[20–23]
		pH 1.5–7.5	fibril	[24]
insulin	51 (31 chain A, 20 chain B)	pH 1.75–2.0 conc < 5 mg ml−1 conc > 5 mg ml−1	spherulite fibril	[25]
insulin analogues	51 (31 chain A, 20 chain B)	high temperature and mechanical agitation	precipitation	[26]
human parathyroid hormone	84	—	fibril	[27]
human interferon a2b		chemical degradation-oxidation	covalent aggregate	[28,29]
Proleukin		+SDS	formation of micellar-like structures containing approx. 27 molecules	[30]
Met-G-CSF		pH 6.9, 37°C	precipitation	[31]

Figure 1.

Amorphous and amyloid: atomic force microscopy images of aggregates of GLP-1 formed under different conditions. (a) Amorphous aggregates of 300 µM GLP-1 in water formed after incubation at room temperature for one week. (b) Fibrils of 25 µM GLP-1 in 25 mM sodium citrate at pH 3.5 formed after 100 h of incubation at 37°C. (c) Fibrils of 150 µM GLP-1 in sodium phosphate buffer at pH 7.5 formed after incubation for 120 h at 37°C. (d) Oligomers formed after dilution of GLP-1 fibrils in ddH₂O. Fibrils of 150 μM GLP-1 were first formed in sodium phosphate buffer at pH 7.5 formed after 200 h incubation at 37°C then diluted 100 fold in ddH₂O prior to imaging. (e) Fibrils of 150 µM GLP-1 in 25 mM Tris at pH 8.5 formed after incubation for 144 h at 37°C. (f) Fibrils of 150 µM GLP-1 in 25 mM Tris, pH 8.5 formed after incubation for 144 h at 37°C, and then diluted 10-fold in ddH₂O prior to imaging. The scale bars represent 1 µm. (Online version in colour.)

In recent years, numerous and extensive research programmes have been undertaken and significant progress been made increasing our understanding of the mechanisms by which peptide and protein aggregation occurs and many of the factors that influence the reaction. However, much of the research has focused on peptides and proteins where aggregation in vivo is associated with disease states, particularly neurodegenerative diseases [41,42]. A considerable amount of the research on disease-related peptide and protein aggregation can be applied to the aggregation of peptide-based therapeutic agents currently being developed by the pharmaceutical industry. In general, however, the aggregation pathways of peptides currently coming through drug pipelines have not been characterized in depth.

In this review, we discuss the factors that are known to affect the aggregation of peptides rather than proteins. We focus here mainly on peptides that do not adopt a stable tertiary structure in the absence of their receptor/target proteins. The main differences between peptides, usually relatively unstructured as monomers, versus proteins, which usually form highly structured monomers sometimes within a larger oligomeric state/complex, is that for the latter any factor known to affect the thermodynamic or kinetic stability of the native structure of the monomer and/or complex also affects the aggregation propensity of the protein. There is extensive literature covering this area and for interested readers the following reviews are recommended: influence of thermodynamic and kinetic stability of native states [43–45], the effect of oligomer formation [46–49], the binding of ligands [50,51]. Within this review, wherever possible, we use therapeutically relevant peptides as examples, but we also refer to peptides such as Aβ or intrinsically disordered proteins, such as α-synuclein and tau, whose aggregation is associated with diseases which have been extensively studied. In the occasional case, where there are no suitable peptide-based examples, we use a highly structured protein to illustrate the point.

First, we describe the types of aggregation processes that have been observed and the structure and properties of the different aggregates, as well as some of the fundamental insights that have been gained on the mechanisms of aggregation. Then, we discuss in more detail the factors known to affect aggregation, and how these factors might influence the aggregation of peptide therapeutics. Finally, at the end of the review, we briefly discuss strategies for preventing or minimizing aggregation and also highlight those areas where research has been lacking and many questions remain.

1.1. Aggregation: the formation of highly structured amyloid fibrils

Many peptides and proteins, under the right conditions, have been shown to self-assemble into highly structured amyloid fibrils (figure 2) [34]. These fibrils contain a cross-β-sheet structure comprising tightly interacting intermolecular β-sheets [52]. Each sheet is formed from many molecules of the peptide each molecule adopting a very similar and highly regular β-strand conformation (figure 2). The sheets are stabilized by backbone hydrogen bonds that result in a spacing between the β-strands of 4.8 Å and are parallel to the fibril axis [34,52]. A pair of β-sheets is known as a protofilament. Two or more protofilaments interdigitate and intertwine with a spacing of 10–11 Å [53–55] to form a mature fibril. These fibrils are stabilized by the hydrophobic effect, van der Waals interactions, π–π stacking and hydrogen bonds. A number of different cross-β-structures can form and are classified according to whether their β-strands are parallel or anti-parallel, and whether the packing of the β-sheets occurs face-to-face or face-to-back, and whether the β-sheets are organized parallel or anti-parallel to each other (figure 2). For an excellent review on the different structures that have been observed and further details of the packing of the side chains within the different structures see Riek & Eisenberg [34].

Figure 2.

Structure of amyloid-like fibrils. (a) Structure of the mature fibril formed by Aβ (pdb code: 3ZPK). (b) Two views of the atomic resolution β-structure within the amyloid fibrils formed by the GYMLGS peptide from human prion protein (pdb code: 3NHC). The top view shows the juxtaposition of side chains looking down the main axis of the fibril. The bottom view shows the anti-parallel nature of the β-sheet formed and packing of side chains from adjacent β-strands. (c) Two views of the atomic resolution showing the cross β-structure within amyloid fibrils formed by the AIIGLM peptide from Aβ (pdb code: 2Y3J). The top view shows the stacking of side chains of the same type from the same residue looking down the main axis of the fibril while the bottom view shows the parallel arrangements of the β-strands in the structure. See [34] for an excellent review on atomic resolution structures of amyloid fibrils. Figures were made using PyMol. (Online version in colour.)

It is interesting to note that, although many amyloid fibrils have been shown to be associated with disease states and are therefore termed pathological fibrils, nature has exploited this type of structure and there are now a considerable number of functional fibrils [56].

1.2. Aggregation to form other types of structures

1.2.1. Higher-order fibrillar structures

Higher-order structures comprised of amyloid-like fibrils have been observed for a number of proteins. Insulin and β-lactoglobulin have both been shown to form spherulites, spherical structures consisting of radiating fibrils from a central core [57]. In addition, a number of peptides that self-assemble into fibrillar structures have also been shown to form hydrogels [58]. These types of higher-order fibrillar structures are beyond the scope of this review and will not be discussed further.

1.2.2. Amorphous aggregates

Amorphous aggregates, also known as disordered aggregates, are unstructured and there are no ordered intermolecular interactions [33]. They are often granular in appearance when imaged by atomic force microscopy, figure 1, and can vary in size and solubility. When insoluble they form precipitates which present a major challenge in biotechnology.

1.3. Mechanisms of aggregation

Aggregation is a complex phenomenon and can occur from many different states (figure 3). In many cases, aggregation occurs from an unfolded or largely unstructured state, i.e. in the case of many peptides and intrinsically disordered proteins. For amyloid fibril formation, the monomeric peptides first self-associate to form oligomeric structures which go on to form a critical nucleus (figure 4). Elongation of the nucleus is rapid relative to its formation giving rise to the sigmoidal aggregation kinetics often observed (figure 5a). For structured proteins, aggregation can also occur from partially structured folding intermediates or, as is the case for a number of monoclonal antibodies, self-association of the native state can take place which gives rise to inactive oligomers [59,60].

Figure 3.

General aggregation processes from relatively unstructured peptides. Peptides can aggregate to form amorphous (left) or highly structured amyloid-like fibrils (right). This process can be induced by a solid–water (surface), or water–air interface or occur in solution. Many intrinsic and extrinsic factors affect the conformational equilibria (β-strand shown as a green arrow, α-helix shown as a red cylinder), including chemical degradation/modifications (shown as a purple circle). Fragmentation can also lead to peptides with a high propensity to aggregate. (Online version in colour.)

Figure 4.

Mechanisms of amyloid fibril formation from monomeric peptides. (a) Shows a classic nucleation–polymerization mechanism where a nucleus (shown by three light blue circles in a rectangular box) which may be thermodynamically unfavourable and/or slow to form from monomer (red circle) is followed by a relatively rapid elongation step where additional monomers add to the end of the nucleus or fibril. In some cases, this is followed by secondary nucleation steps—nucleation on the surface of an existing fibril to form an additional nucleus (top) which is then followed by further elongation steps. Alternatively, a fibril can fragment to create two shorter fibrils thus increasing the number of nuclei and therefore elongation/growth sites (bottom). Fragmentation is followed by further elongation steps. In this mechanism, a number of on-pathway oligomers may be formed prior to the nucleus. In this case, classic kinetics where the lag time and t_1/2 decrease with increasing monomer concentration are usually observed (figure 5b–d). (b) Shows an alternative mechanism in which the on-pathway nucleation–polymerization (shown by the light blue species) is in competition with the formation of off-pathway oligomeric species (green circles). If there is a unimolecular step such as a conformational change on the direct pathway to fibrils, then unusual aggregation kinetics can be observed (figure 5e–g). (Online version in colour.)

Figure 5.

Aggregation kinetics: formation of amyloid-like fibrils. (a) Typical sigmoidal kinetics frequently observed in the aggregation of peptides and proteins into amyloid-like fibrils. Aggregation is usually probed by the increase in fluorescence of the dye thioflavin T (ThT) on binding to fibrils containing regular β-sheet structure. The curve shows the key kinetic parameters that can be obtained from a fit of the aggregation data to a simple sigmoidal function. Three kinetic parameters are obtained from the fit of the data to a simple sigmoidal function: t_1/2, the time at which the ThT fluorescence is 50% of its final value (red dashed line), the apparent growth rate which corresponds to the gradient of the steepest part of the rapid growth phase (blue dotted line), and t_lag, the lag time at which point sufficient nuclei are present in solution for elongation and growth of fibrils to become rapid. (b) Typical kinetics for a reaction following a simple nucleation–polymerization mechanism with only on-pathway oligomeric intermediate states. This figure illustrates how the t_1/2 and t_lag decrease with increasing peptide/protein concentration (dark purple, low concentrations, to light blue, high concentrations, of peptide; the increasing peptide concentration is also shown by the grey arrow). The data are simulated using KinTek using Model 1 in [1]. (c,d) Kinetic parameters for the aggregation of GLP-1 into amyloid fibrils at pH 8.2 taken from [1] under conditions where the peptide follows a simple nucleation–polymerization mechanism with on-pathway oligomers only. Dependence of (c) t_1/2 and (d) t_lag on peptide concentration. (e) Simulated data to show how the aggregation kinetics change when off-pathway oligomeric intermediates are populated as well as on-pathway intermediate species and where there is a unimolecular step on the pathway to fibril formation (dark purple, low concentrations, to light blue, high concentrations, of peptide; the increasing peptide concentration is also shown by the grey arrow). KinTek is used, see Model 3 in [1]. (f,g) Kinetic parameters for the aggregation of GLP-1 into amyloid fibrils at pH 7.5 taken from [1] under conditions where the populates on and off-pathway oligomers. Dependence of (f) t_1/2 and (g) t_lag on peptide concentration. (Online version in colour.)

2. Factors affecting the physical stability of peptides

2.1. Peptide concentration

One of the most important factors influencing the physical stability of peptides is peptide concentration. This is long-established through numerous kinetic studies on the formation of amyloid fibrils [61–65], amorphous aggregates [66,67] and adsorption on surfaces [68]. With respect to the formation of amyloid-like fibrils, extensive kinetic studies have been performed over wide ranges of peptide concentrations and, in the majority of cases, the kinetics have been shown to be consistent with a nucleation–polymerization mechanism (figure 4) [69]. In this case, the monomeric starting material is converted into oligomers, and ultimately a species that can act as a nucleus for rapid fibril growth. Addition of usually monomers to a nucleus or growing fibrils occurs rapidly to form protofibrils and mature fibrils. The essential features of such systems are that the kinetics show a lag phase, a rapid elongation or growth phase and a plateau. During the lag phase, a number of oligomeric species, including nucleating species, and short amyloid fibrils form until a point is reached at which the growth of fibrils is rapid (figure 4). The formation of fibrils then slows as a plateau is reached which represents either the point where monomers are completely depleted or an equilibrium point at which mature fibrils are in equilibrium with starting monomers. The sigmoidal kinetic traces can be fitted to obtain key kinetic parameters such as lag time t_1/2 and the growth rate (figure 5a). How these kinetic parameters depend upon peptide concentration has been established for many systems, including Aβ [70], α-synuclein [71], tau [72], glucagon-like peptide-1 (GLP-1) [1] and other peptides. The majority of peptides and conditions show kinetics typical of nucleation–polymerization mechanisms, and in these cases, generally both t_1/2 and t_lag decrease with increasing peptide concentration as shown in figure 5b–d. The growth rate has been shown to decrease or increase, or be independent of peptide concentration, depending upon the peptide/conditions and rate-limiting step. Cases where there is an increase in growth rate with peptide concentration point towards association of monomer and fibril being rate limiting to the growth. In contrast, if the growth rate is independent of peptide concentration this can indicate some other step such as a conformational change of the monomeric unit is rate limiting. The growth rate is also dependent upon secondary nucleation steps, if they exist, such as fragmentation of the fibrils to form small fibrils with a net increase in reactive fibril ends on which growth can occur, or secondary nucleation sites on the surface of a fibril (figure 4a) [62].

This highlights the issues of physical stability of peptides, which may not be a significant issue at low peptide concentrations, but which can become a serious problem at high concentrations such as those frequently used in therapeutics.

In stark contrast to the systems and kinetics described above, there are an increasing number of peptides and conditions under which very different kinetics have been observed. In these cases, increasing concentrations of peptide leads to a decrease in t_1/2 and t_lag [1,73,74] (figure 5e–g). This unusual kinetic behaviour has been attributed to the formation of a different form of oligomeric species where the oligomers formed are off-pathway, and either cannot directly be converted into on-pathway oligomers or nucleating species, or where the interconversion is extremely slow. One such system is GLP-1, which is a particularly interesting system as in this case there is a switch in the kinetic behaviour with pH [1]. At pH > 8.0, GLP-1 follows the more common nucleation–polymerization kinetics with t_1/2 and t_lag decreasing with increasing peptide concentration (figure 5b–d); however, at pH 7.5, the kinetics reverse such that both lag time and t_1/2 decrease with increasing peptide concentration indicating off-pathway oligomers have formed [1]. At an intermediate pH, 8.0, there is little dependence on peptide concentration as there is a delicate balance between on- and off-pathway species [1]. Although off-pathway oligomers have been observed now for a number of peptides and proteins [1,73,74] these systems have not been studied in the detail that peptides and proteins showing simpler nucleation–polymerization kinetics have been and therefore not much is known about the nature of off-pathway oligomers, for example, their size, structure and cellular toxicity.

2.2. Amino acid sequence

How the amino acid sequence of a peptide affects it intrinsic propensity to aggregate, to form either amorphous or amyloid-like fibrils is now very well understood, such that there are numerous prediction programmes available [33,75]. Factors such as hydrophobicity, charge state, β-sheet forming propensity along with other properties are all known to contribute [33,75–77]. A series of five or more residues which all have a high intrinsic propensity form what is known as an aggregation-prone region, APR [75,78–81]. In some cases, these methods have been used to predict the effect of an amino acid substitution on aggregation propensity, for example, the effect of pathogenic mutations on the aggregation of Aβ in vitro and in vivo [75].

A number of approaches have been developed to predict aggregation propensity some of which are primarily based on our knowledge of the intrinsic propensity of amino acids to aggregate and form fibrils. As our understanding of the factors that affect aggregation in terms of amino acid sequence has improved, methods have also been developed that not only take into account APRs in peptide sequence but also other factors such as the presence of gatekeeper residues that help prevent aggregation [82–84], and the flexibility and solvent accessibility of the peptide chain [85]. Other algorithms also take into account the propensity of amino acids to form other forms of stable structure such as α-helices, as aggregation is always in competition with the formation of other conformational states [86].

Since high-resolution structures of amyloid fibrils have become available from X-ray crystallographic and NMR studies [34], structure-based methods for predicting APRs have also been developed [33,75]. In these cases, APRs which are compatible with the formation of the cross β-structure seen in the core of amyloid fibrils are identified by threading peptide and protein sequences through templates of different amyloid structures scoring each resultant structure in terms of packing density [87], pairing energies for side chains that pack against each other in the structure [88–90] and other metrics [91,92]. In another case, a combination of the two approaches has been used and a position-specific scoring matrix developed taking into account both intrinsic propensities and also structure [93].

For peptides which do not form stable tertiary or quaternary structures, these approaches work well and predictions are generally reasonably accurate. In addition, while many algorithms which are not structure-based predict both the propensity for a peptide to form either an amorphous or highly structured aggregate, the structure-based methods are largely accurate only for predicting the formation of amyloid-like fibrillar structures.

Given these prediction methods, summarized in table 2, it is relatively straightforward to identify APRs in largely unstructured peptides and to rationally design amino acid substitutions to reduce aggregation propensity. Of course, such changes may also affect activity. At the moment, there are no programs available to predict the effects of incorporating non-natural amino acids nor the effects of chemical modification of a peptide on its aggregation propensity.

Table 2.

Methods for predicting aggregation-prone regions and aggregation propensity in peptides.

name	description of method	references
Zygreggator	APRs and gatekeepers, protein flexibility and solvent accessibility	[85]
TANGO	statistical thermodynamics approach. Identifies APRs but takes into account competition β-aggregated state and other conformations	[86]
SALSA	uses β-propensity for each amino acid and a sliding average to identify APRs	[81]
AGGRESCAN	uses aggregation propensity for each amino acid	[94,95]
FoldAmyloid	packing density	[87]
3D profile method	structure-based method uses amyloid fibril structure as a template	[91]
Pre-AMYL	structure-based method uses amyloid fibril structure as a template	[92]
PASTA	pairing energies calculated for pairs of residues within an amyloid structure. Sequences are scored by energies to identify APRs	[88,89,96]
BETASCAN	β-strand pairing energies using amyloid core structure	[90]

2.3. pH and net charge

Electrostatic interactions are known to play major roles in the self-association of peptides to form all form of aggregates [97–102]. Essentially, for systems carrying an overall net charge, unfavourable electrostatic repulsion needs to be overcome in order for self-association/aggregation to take place. Generally, the higher the net charge the slower the aggregation and the lower the net charge the higher the propensity to aggregate [97–102]. As such, solution conditions such as pH, ionic strength, nature of cations and anions in solution and the presence of polyelectrolytes can all have a dramatic effect on both the rate and extent of aggregation. One of the best characterized systems in terms of the effects of net charge, ionic strength, and counter ions on aggregation is the 37-residue positively charged peptide IAPP which has no acidic groups [103]. A comprehensive and rigorous study has shown that there are multiple effects on aggregation depending upon the pH [103]. At pH 8.0 there is a strong dependence on ionic strength due to Debye screening and increasing ionic strength accelerates aggregation by up to a factor of ten. In this case, specific ion effects were also observed, the nature of the cation being relatively unimportant but the rate depending crucially on the identity of the anion (factors of four were observed with different anions). The effects were observed both on the lag phase and the apparent growth phase. At pH 5.5, where IAPP has a higher overall net charge, the aggregation kinetics were considerably slower than at pH 8.0, but there were similar effects of ionic strength [103]. For lysozyme and acylphosphatase, studies have established that polyelectrolytes can also have significant effects. In these cases, a number of different polyelectrolytes were found to bind to and stabilize mature fibrils [104].

Although it is easy to establish some general rules of thumb with regards to the effects of pH, net charge, ionic strength, ions and polyelectrolytes, these rules should be viewed with caution. There are an increasing number of examples where these rules are not followed due to the complexity of aggregation reactions. For example, around the isoelectric point of peptides and proteins, where net charges are essentially zero, then monodisperse and quasi-amorphous aggregates have been observed instead of amyloid-like fibrils [57]. Thus, aggregation kinetics can be very different from those of amyloid-like fibril formation which is often seen at pH values away from the isoelectric point. In addition, other changes in mechanism have been found at different values of pH and net charge. For example, as already discussed in §2.1, the aggregation kinetics of GLP-1 into amyloid-like fibrils is highly sensitive to pH [1]. In this case, the presence of off-pathway species slows the rate of aggregation considerably particularly at high peptide concentrations. Thus, although the net charge on the protein at pH 7.5 is lower than at pH > 8.0, the rate of aggregation is considerably slower, contrary to what would be expected [1].

2.4. Chemical degradation

Chemical degradation pathways in biological pharmaceuticals have been reviewed many times [105,106]. A number of chemical degradation pathways have been characterized including: deamidation and isomerization [107], oxidation [106,108], hydrolysis, disulfide bond breakage and formation, succinimidation [109], deglycosylation, Maillard reaction, racemization and β-elimination and many of these have been shown to increase the aggregation of biologics [109]. This is in effect because chemical degradation/modification often changes the physical properties such as hydrophobicity, secondary (relevant for relatively unstructured peptides) and/or tertiary structure, and the thermodynamic and/or kinetic barriers to unfold (relevant only for highly structured proteins and not discussed further here) [105,109].

2.4.1. The effect of deamidation on the physical stability of peptides

Deamidation is a chemical reaction in which the amide group in Asn or Gln side chains is hydrolysed at pH < 3 or undergoes intramolecular cyclization at pH > 6 to form the corresponding carboxylate: Asp or Glu [105], where the isomerization of Asp to form iso-Asp involves backbone-to-side chain rearrangement. The deamidation of Asn at pH < 4 is via direct hydrolysis of the side chain [110] leading to a single product, Asp. However, in basic conditions (pH > 6), deamidation generates two products: Asp and iso-Asp [105,109] and it can also lead to peptide and protein fragmentation [111].

Deamidation of Gln to Glu can also take place but is rarely observed, unless the pH < 3. It is well known that Asn is more prone to deamidation than Gln [112–115]. Moreover, it has been reported that the deamidation of Asn is strongly dependent on primary structure with the adjacent amino acid on the C-terminal side having the greatest influence on the rate. In peptides, an Asn-Gly sequence is particularly prone to deamidation and is considered a ‘hot spot’ of reactivity. Asp degradation is similar to Asn degradation resulting in the formation of iso-Asp and racemic products. Asp residues can also undergo backbone hydrolysis.

There is limited literature on the link between deamidation and physical stability; however, deamidation may alter the local structure of the peptide. Deamidation to Asp results in a change of charge (−1) but Asp is isosteric with Asn. In contrast, iso-Asp, which contains a methylene group in its peptide backbone is likely to be more disruptive to local structure. In some cases, deamidation has been shown to accelerate amyloid formation and alters amyloid fibril structure, e.g. amylin_20–29 [116] and amylin [117]. Moreover, deamidation of Asn76 in human γS-crystallin promotes dimer formation [118] and deamidation of lens protein, βA3-crystallin destabilizes and triggers aggregation of that protein [119].

2.4.2. The effect of oxidation on the physical stability of peptides

It is well known that some amino acid side chains can be oxidatively modified during any stage of peptide production, purification [120], formulation [121], storage [105,106,122]. This includes aromatic amino acids His, Trp and Tyr and sulfur-containing side chains Met and Cys [105,109,111]. Moreover, complicated oxidation reactions can modify the primary sequence of a protein and may cause aggregation [123].

There is rather limited literature on the link between oxidation and the physical stability of peptides, such that the effects are still poorly understood. However, oxidation of the side chain of Met has been discussed in a few papers. Interestingly, studies on the peptide Aβ_1–40 have shown that oxidation of the methionine residue reduces lag-times of fibril formation but extends lag-times for Aβ_1–42 [124]. It has been shown, for both peptides, that Met35 oxidation affects fibril morphology and length [124]. Studies on methionine oxidation also in α-synuclein at physiological pH show that oxidation inhibits fibril formation significantly [125].

2.4.3. The effect of hydrolysis on the physical stability of peptides and proteins

Most peptide bonds are stable except those in -X-Asp-Y- sequence which is at least 100 times more labile than other peptide bonds in dilute acid. The Asp residue undergoes backbone hydrolysis of peptide bonds both N- and C-terminal to the residue as discussed earlier. In addition, the side chain of Trp is known to undergo hydrolysis and the primary degradation product is called kynurenine [105,109]. N-terminal cyclization reactions are closely related to hydrolytic reactions, both degradation pathways involving nucleophilic attack [105]. They have been described as occurring by two general pathways: diketopiperazine (DKP) formation and pyroglutamic acid (pGlu) formation [105]. An example of a peptide forming DKP is the undecapeptide substance P [126]. Pyroglutamic acid formation follows a mechanism similar to that for DKP formation described earlier [105] and forms a cyclized, N-terminal structure which is has been observed in some peptide hormones [127]. Unfortunately, there is very little literature describing how these hydrolysis reactions in peptides affect the physical stability of the system apart from in a very few cases. For example, it has been shown that pGlu-Aβ peptides contribute to the formation of amorphous and fibrillar deposits in Alzheimer's disease, familial British dementia, and familial Danish dementia [128].

2.4.4. The effect of β-elimination and racemization on the physical stability of peptides and proteins

β-Elimination reactions typically require high pH and elevated temperatures [105,111], and thus there are few reports of β-elimination in proteins of pharmaceutical relevance, except for β-elimination at Cys being observed in insulin in the solid state [129]. The racemization reaction is very similar to β-elimination and has been observed in Asp residues in crystallin as been described in a number of studies [130]; however, the impact of racemization on subsequent physical stability is not well known. It has been suggested that isomerization and/or racemization of Asp at position 1 or 7 is associated with the pathological role of Aβ_1–42 peptide [131]. On the other hand, racemization and/or isomerization of Asp23 in the same system is thought not to be related to pathogenesis, but to be a consequence of chemical reactions during long-term deposition of fibrils [131].

2.5. Chemical modifications

Chemical modifications of different types are increasingly used to optimize the chemical/physical and biological properties of therapeutic peptides. This includes the incorporation of non-natural amino acids [132], engineering of disulfide bonds [133–137] and lipidation or acylation of the peptide [138,139].

Lipidation/acylation of peptides is becoming increasingly important in the field of therapeutic peptides [140,141]. Inspired by endogenous lipidation processes, it has been shown to improve many properties of a peptide, particularly its ability to bind to human serum albumin increasing the half-life of the peptide in vivo dramatically and also enhancing its ability to permeate body tissues [140,141]. Current therapeutics which use this approach include liraglutide [142] (a lipidated form of GLP-1), and insulin detemir, a lipidated conjugate of insulin [143]. Recently, GlaxoSmithKline have developed this idea further and created a GLP1-albumin fusion to achieve similar effects [144,145].

Although considerable research has been undertaken on how this class of chemical modification affects the activity of the peptide in vitro and in vivo, rather little is known about how these modifications affect the physical stability of the system. One recent biophysical study on liraglutide by Wang and co-workers showed that this lipidated peptide can form oligomers in solution [146]. In this case, a pH-sensitive equilibrium between octamers and dodecamers was observed in the vicinity of pH 7 [146]. In another case, acylation of the anti-microbial peptide novicidin resulted in an increased tendency to form α-helical structure when a C12 acyl group was attached. The concentration dependence of this conformational change was attributed to the increased tendency of the peptide to form micelles as shown by NMR [147]. Although not peptides with any therapeutic potential, the Prive group has published extensively on the properties of lipopeptide detergents (LPDs) which have been developed for structural studies of membrane proteins [148]. LPDs, which consist of an α–helical peptide with alkyl chains at either end of the helix, have been shown to self-assemble into cylindrical micelles that have a densely packed hydrophobic core [148].

In contrast to the chemical modifications used by the pharmaceuticals industry to optimize the properties of therapeutic peptides, where little is known about the effect of the modifications on physical stability, a great deal has been reported for peptides whose aggregation is associated with disease states. In these cases, the modifications investigated are largely post-translational modifications that occur in vivo and which, in many cases, are thought to affect the disease. For example, the effects of many post-translational modifications on the aggregation of Aβ peptides have been studied. An N-terminal truncation and pyroglutamate modification at residue 3 of Aβ peptide has been shown to have a higher propensity for oligomerization and aggregation than full-length Aβ [149,150]. In addition, phosphorylation of Aβ has been shown to increase oligomer formation [151–153], while citrullination of Aβ decreases the rate of fibril formation but may increase the amount of toxic oligomers [154]. Similarly, nitration has also been shown to increase the amount of oligomers of Aβ in solution [155]. Backbone modification of Aβ at Gly29 also shows decreased fibril formation and increased oligomer formation [156].

The aggregation and fibril formation of α-synuclein is associated with another neurodegenerative disease, Parkinson's disease. In this case, there is also a large body of data that establishes that numerous post-translational modifications in vivo affect the propensity of the protein to aggregate and thus potentially the disease state. For example, N-acetylation has been shown to decrease fibrillation [157,158], post-translational modifications due to oxidative stress such as HNE conjugation, nitration and oxidation have been shown to reduce fibril formation but increase oligomer formation [159,160], and oxidation of Met5 also decreases fibril formation but increases the concentration of oligomers [161].

The intracellular protein tau, associated with Alzheimer's disease, is also known to undergo a large number of post-translational modifications in vivo including phosphorylation, ubiquitination, nitration, truncation, prolyl isomerization, glycosylation and glycation. Many of these are thought to affect neurofibrillar tangle formation, the aggregated form of tau in cells [162]. A few, but not many, of these post-translational modifications on tau aggregation have been studied in detail. For example, cysteine guanylation of tau has been shown to inhibit fibril formation with a corresponding increase in the population of oligomers [162].

The effects of a large number of other modifications on physical stability have also been reported including different glycosylations [163] and fatty acid conjugation [164].

2.6. Surfaces and interfaces

Surface-induced aggregation, often called ‘surface adsorption’, is a physical degradation process whereby there is accumulation and adhesion of peptide molecules to a surface, without penetration of the surface. During this process, peptide molecules change their physical state, i.e. they are no longer in solution [165–168]. There are different mechanisms of surface adsorption of biomolecules and some of the major factors driving adsorption include intra-molecular forces, hydrophobicity [169–171], and ionic or electrostatic interactions [168,172].

Extensive research has been conducted on the surface adsorption of human insulin [171,173–177]. It is well known that insulin aggregates to form fibres in the presence of hydrophobic surfaces [170,171,175,176] and especially in a peristaltic pump system [177]. In this case, the potential formation of aggregation-prone intermediates is very problematic as this may catalyse the aggregation of other molecules in bulk solution. Other studies have established that the aggregation of Aβ_1–40 into fibrils is also influenced by surfaces [178–182]. The surface adsorption was studied for other systems such as GLP-1 [172], GLP-2 [183], acylated glucagon-like peptide-2 [183,184], glucagon [185], poly(ethylene glycol) (PEG)-glucagon [186], and α-synuclein [187].

It has been demonstrated that any interface (between solid, liquid or air) will affect the ability of water molecules to form a dynamic hydrogen-bonding network as in bulk water, and therefore air–water and ice–water interfaces are described in the literature as hydrophobic interfaces, because these interfaces do not allow hydrogen bonding [167]. Adsorption of peptides through their hydrophobic side chains to such hydrophobic interfaces can thus ‘hide’ the hydrophobic surface from the aqueous bulk. Some research shows that, for example, Aβ_1–40 [179,188] and insulin both aggregate at air–water interfaces [189].

Many studies have confirmed that the rate of peptide and protein adsorption is normally governed by diffusion, and is thus a function of concentration [109]. The amount of surface-adsorbed species slowly increases over time, usually due to further peptide aggregation at the site [178]. Of course, there may also be slow structural rearrangements of the adsorbed peptide [178]. The adsorbed layer often changes its properties over time; this is often associated with a change from a reversible to an irreversible process [168]. The final morphologies of the aggregates formed can be sensitive to the surface chemistry [172,174,178].

It is well known that after sufficient time most biomolecules will adsorb to solid surfaces so strongly that they will not (or only very slowly) desorb from the attached surface back into solution [190]. However, it is possible to some degree to control this process by adding surface active agents which can lead to desorption. The degree of adsorption depends on multiple factors, including: concentration, pH, excipients and temperature [167,168].

2.7. Excipients

Excipients of many types have been employed to try and reduce aggregation in numerous systems including salts, surfactants, osmolytes, preservatives, chelators and antioxidants, specific ligands, sugars and carbohydrates. Below is a brief description of some of the excipients that have been used and their effects on aggregation.

2.7.1. Buffering agents

As already discussed, pH strongly influences peptide stability and the potential for biomolecular aggregation. The most commonly used buffers in pharmaceutical development are acetate, citrate, histidine, phosphate, Tris and glycine [105,109,191]. A complete description of the effects of buffer on peptide aggregation is beyond the scope of this review; however, interested readers are directed towards the recent excellent review by Zbacnik and co-workers [192].

2.7.2. Salts

Standard physiological salts have been used as tonicity modifiers in peptide formulations [109,191]. Salts have complex effects on the physical stability of biomolecules affecting both conformational and colloidal stability. Their effects frequently vary according to the surface charge on the peptide or protein, and the overall effect of a salt on physical stability is a balance of different and multiple mechanisms by which salt interacts with water and biomolecules. Salts can influence physical stability by altering the properties of the peptide–solvent system (Hofmeister effects) and by screening electrostatic interactions (Debye–Hückel effects). Many different studies on the dependence of amyloid formation on ionic strength over the past 10 years have shown that ions can influence both the kinetics of aggregation and the structure of fibrils formed. For example, islet amyloid polypeptide (37-residue) [103], human glucagon (29-residue) [193], and Aβ peptide [100,102,194].

2.7.3. Surfactants

Surface-active agents are amphiphilic molecules that tend to orient such that the exposure of the hydrophobic part of the molecule to the aqueous solution is minimized. Non-ionic surfactants such as Tween20 and Tween80 are often added to peptide solutions during pharmaceutical development to prevent aggregation or adsorption during purification, filtration, shaking and transportation [191,195–198]. However, the chemical stability of Tween20 and Tween80 in pharmaceutical formulations is very important. Unfortunately both surfactants are known to undergo oxidation and cleavage at their ethylene oxide subunits, as well as hydrolysis of the fatty acid ester bond [199–202]. Thus, their use can be complicated as these chemical processes can potentially promote precipitation in some peptide formulations. The use of Tween80 is also far from straightforward as it has been reported that it can have a dual effect on physical stability. For example, Tween80 in the formulation of the protein IL-2 mutein inhibited shaking-induced aggregation, but also had a dramatic effect on the oxidation and aggregation of the protein during storage of the liquid formulation [203,204]. To counteract this issue, methionine and tryptophan are used in formulation development to prevent Tween from oxidative degradation [205]. Readers interested in the interactions between surfactants and peptides are directed towards two excellent recent reviews by Otzen [206] and Khan and co-workers [196].

2.7.4. Amino acids

A number of free amino acids have been used to stabilize proteins and reduce aggregation. Arginine, histidine, lysine, glycine and aspartic acid have been found to reduce aggregation for a number of biomolecules [191].

2.7.5. Osmolytes

Osmolytes such as sucrose, trehalose, sorbitol and glycine are commonly used in formulation development [191]. Sucrose has been shown to inhibit IL-1ra dimer formation [109]. However, their use is not always beneficial and it has been reported that some osmolytes accelerate the rate of aggregation, for example, of glucagon [207].

2.7.6. Antioxidants and chelators

Peptide oxidation, which was described earlier in this review, is a major cause of chemical instability and also sometimes linked to physical stability. Amino acids such as methionine, cysteine, histidine, tyrosine and tryptophan in peptides are susceptible to oxidation under some conditions encountered during pharmaceutical development. Therefore, a number of antioxidants are used as excipients including ascorbic acid [191]. It has also been reported that sodium thiosulfate, methionine, catalase or platinum, and the chelating agents EDTA and DTPA are also effective in reducing the oxidation of biologics [191,208–210].

2.7.7. Preservatives

Preservatives are used as antimicrobial agents and are very common in liquid formulations [191,211]. Their role is to prevent bacterial growth during storage [191]. The most commonly used preservatives in pharmaceutical development are m-cresol, phenol and benzyl alcohol. However, it has been reported that preservatives can sometimes cause protein aggregation. For example, benzyl alcohol induces aggregation of recombinant human interleukin-1 receptor [212], and m-cresol induces aggregation of cytochrome c [213]. Moreover, destabilizing effects of phenol and m-cresol were reported for insulin lispro and insulin aspart [214]. Little is known about their effects on the physical stability of peptides.

2.7.8. Polymers and proteins

A number of polymers have been shown to stabilize drug products including PEGs. Other polymers that are used as excipients include dextrans, heparin, gelatins type A, poly-l-glutamic acid, poly-l-lysine, poly-Asp and poly-Gl [105,109,191]. For example, it has been reported that different PEGs stabilize the proteins lysozyme and bovine serum albumin [109]. Human serum albumin is also commonly used as an excipient to inhibit protein/peptide adsorption onto surfaces during product development [109]. However, adding protein-based excipients adds more complexity to pharmaceutical formulations [191].

Recently, some excellent reviews on biologics formulation [215,216] and excipients, including protein–excipients interactions [191,216,217] have been published.

2.8. Impurities

The results of in vitro aggregation studies are often controversial and difficult to reproduce by others. One of the reasons for this problem is the variable amounts of intrinsic impurities in peptide and protein preparations that can affect aggregation rates. For example, the aggregation kinetics of synthetic Aβ has been reported to vary not only batch-to-batch but also with storage and solubilization conditions [218]. In line with these findings, kinetics studies revealed that the impurities left in the synthetic peptides of glucagon decreased fibrillation rates dramatically compared to that of pharmaceutical-grade glucagon [193,219,220].

Numerous sorts of impurities may be encountered in synthetic peptides, which either originate from the raw materials, the manufacturing process, formed by degradation during the manufacturing process or during storage. Some examples of potential synthetic peptide process-related impurities are deletions, truncations, diastereoisomers, substitution (Leu/Ile) and disulfide modification. The most common peptide degradation impurities are deamidation of Gln/Asn/C-terminus, methionine oxidation, acetylation of amino functions and disulfide modification (see §2.4). Non-peptide process-related impurities can be present. Some examples are residual solvents and heavy metals, traces of reagents used in the process, such as trifluoroacetic acid (TFA), acetate and hydrochloride salts, inorganic fluoride compounds from protective groups such as t-butyloxycarbonyl- (Boc-). Unfortunately, there are few reports about the impact of degradation impurities and counter ion salts on the kinetics of peptide aggregation and there is a lack of information about the influence of other common impurities in synthetic peptides on either the mechanism or the kinetics of aggregation.

2.8.1. Peptide-related impurities

The addition of 1% oxidized glucagon (Met27) has shown to greatly delay the rate of fibril formation. The authors suggest that glucagon analogues can interfere with fibril formation by either competing with or blocking unmodified glucagon adsorption at the ends of fibrils [3,4]. Another study shows that in the absence of impurities, amylin 20–29 does not spontaneously aggregate and is not amyloidogenic. However, this peptide can spontaneously deamidate, and the presence of less than 5% of deamidation impurities leads to the formation of aggregates that have the hallmarks of amyloid. In addition, small amounts of deamidated material can induce amyloid formation of the purified peptide. These results have fundamental implications for the definition of an amyloidogenic sequence and for the standards of purity of peptides and proteins used for aggregation studies [58,221–227].

2.8.2. Non-peptide-related impurities

The majority of synthetic peptides are still manufactured by solid-phase procedures. Due to the cleavage and standardized purification steps, the resulting product usually contains TFA salts. The presence of TFA in the peptidic material affects its physico-chemical properties. Several investigations have documented the influence of TFA on peptide structure. For example, characterization of synthetic Aβ peptide has shown that there is a strong influence of TFA counter ions on the physico-chemical properties of the peptide [228]. In particular, it has been shown that TFA can change the secondary structure, solubility, and aggregation propensity of synthetic Aβ peptides, and even affect the ultimate size, flexibility, and geometry of fibrils formed [229–231]. Fezoui and co-workers showed that solvation of Aβ synthetic peptide with sodium hydroxide followed by lyophilization produced stocks with superior solubility and fibrillogenesis characteristics [232,233]. Relative to trifluoroacetate or hydrochloric acid Aβ peptide counter ion salts, yields of low molecular weight (monomers and/or dimers) were improved significantly after pre-treatment with sodium hydroxide [232,233].

2.9. Temperature/pressure/agitation/lyophilization

Numerous other external factors including temperature, pressure, agitation and lyophilization/freeze drying can affect the aggregation of peptides, in many cases, in ways difficult to predict.

2.9.1. Pressure

High hydrostatic pressure (HHP) has been used in a number of studies [234]. It is now well established that HHP pushes the conformational equilibria towards ensembles of structures that occupy smaller volumes, it acting on cavities that are excluded from solvent that can be found in the hydrophobic cores of natively folded proteins and aggregates including fibrils [234]. HHP has been used to dissociate multimers in oligomeric structures and, in some cases, dissociate aggregates [234,235]. The degree to which HHP acts to dissociate aggregated forms of peptides depends upon the structures that are formed and also their inherent thermodynamic stability. For example, if fibrils have cavities within them (cavity defects as they are called) then often low pressures are sufficient to dissociate the aggregated form; for example, early aggregates and protofibrils are generally more readily dissociated by HHP than mature fibrils [236]. For a peptide from transthyretin that can form fibrils, TTR_105–115, 220 MPa dissociates aggregates and protofibrils while mature fibrils remain stable up to 1.3 GPa [236]. For α-synuclein, pathogenic variant fibrils are less resistant to HHP than those of the wild-type protein, hence more toxic oligomers are present [234]. The dissociation of aggregates and fibrils by HHP is reversible and can be enhanced by the addition of cosolvents such as glycerol, methanol and dimethylsulfoxide [236].

2.9.2. Temperature

Aggregation kinetics frequently show non-Arrhenius behaviour even over short temperature ranges [237]. A consequence of this is that it is challenging to extrapolate the results of aggregation kinetics commonly measured at higher temperatures to lower temperatures.

However, cold denaturation of amyloid fibrils has been observed [238,239], and low temperatures have been combined with high pressures to dissociate aggregates and fibrils [236].

2.9.3. Agitation

Agitation, in the form of a magnetic stirring bar or by use of a shaker table, is routinely used in studies of peptide and protein aggregation. It has been employed in order to accelerate otherwise slow aggregation kinetics and also used to improve the reproducibility of experimental results [240]. Despite the frequent use of agitation little is known about the shear forces that are produced and the effects on different steps in the aggregation reaction (figure 4). In addition, as stirring/shaking is often different in different studies, it is challenging to undertake a comparative analysis on systems studied in different research groups [241].

2.9.4. Lyophilization/freeze drying

The stabilization of peptides in a solid form is a very common approach to increase both their chemical and physical stability [242]. Lyophilization and freeze drying are frequently used to prepare peptides in a solid state. However, for a number of systems studies have shown these processes can result in conformational changes leading to increased aggregation after reconstitution [243].

3. Summary

The physical stability of peptide- and protein-based therapeutics remains a huge challenge for the pharmaceutical industry. Table 1 provides just a few examples of the different types of aggregation processes that can occur that lead to low physical stability; however, the number of potential therapeutic agents known to aggregate in some form is very large indeed [30,32,123,196,215]. In contrast to peptides that aggregate and that are associated with disease states, the relative propensity of therapeutic peptides to aggregate may be low. However, the fact that these peptides are frequently used at high concentrations and are required to have long-term physical stability under these conditions makes the problem particularly acute. In addition, therapeutic peptides have to withstand a number of processes during production, formulation and use, that increase their propensity to aggregate and which exacerbate the problem.

Fundamentally, the basic thermodynamic and kinetic parameters that underpin the aggregation of peptides whether they be disease-associated or therapeutic must be the same. In many cases, the intrinsic and extrinsic factors that affect the physical stability of peptides, such as amino acid sequence, pH etc., will be common to both classes. In other cases, factors will differ. For example, the physical stability of disease-associated peptides depends upon many factors in vivo such as the presence of molecular chaperones, proteostasis and membrane surfaces which are not relevant to therapeutic peptides. In contrast, therapeutic peptides are exposed to a range of surfaces and processes, such as high pressures and freeze-drying, that are not relevant to in vivo systems. Despite this, much knowledge is directly transferable between the different classes from studying the fundamentals of aggregation in vitro in either system, particularly from a mechanistic perspective. It might even be that strategies for increasing the physical stability of therapeutic peptides may come from a deeper understanding of how nature deals with the problem.

Given the large number of different factors that can affect the aggregation of peptides, and which are outlined in this review, it is no great surprise that this remains a challenging area in the development and use of peptide-based therapeutic agents. In some cases, great strides have been made in recent years in our understanding of some of the factors, particularly the sequence of the peptide. In other cases, such as impurities, there is almost no knowledge on which and how these species affect aggregation although it is known to be a significant issue. In other cases, there is a large amount of empirical data available, for example on how post-translation modifications affect oligomer and fibril formation in disease-related peptides. Even in this case though, there is a fundamental lack of understanding of how the modifications affect each step on the aggregation pathway. Without such, it is impossible to rationally predict effects. In addition, there are an increasing number of chemical modifications employed by the pharmaceutical industry to enhance the properties of therapeutic peptides, whose effects on physical stability are not known.

The number of factors that affect aggregation reactions to some degree reflects the intrinsic complexity of the reactions and very subtle changes in the physico-chemical properties of the system can result in large changes in aggregation propensity and rates. There is clearly a significant amount of research that is still required in order for scientists to fully understand the complex energy landscape of aggregation and how this is affected by numerous variables. As well as empirical studies, more fundamental research is needed for this major problem to be fully addressed.

Data accessibility

This article has no additional data.

Authors' contributions

All authors co-wrote the manuscript.

Competing interests

We declare we have no competing interests.

Funding

K.L.Z. was funded by MedImmune. F.J.B. is funded by MedImmune and Peterhouse, Cambridge. A.L.G.d.S. is an employee of MedImmune.