Preventing Protein Self-Association Through Strategic Covalent Modification

Abstract

Protein self-interaction leading to aggregation is a major challenge facing protein pharmaceuticals. It leads to a range of problems including increases in immunogenicity and loss of activity. In this work we describe an approach for blocking or antagonizing the quaternary interactions that drive self-association. We applied the approach to glucagon, a therapeutic peptide known for its propensity to form fibrils due to self-interaction. We synthesized a regio-pure common feedstock that allowed easy modification with potential blocking peptides that represented a range of chemical types (anionic, cationic, polar, non-polar). From these synthesized materials we identified two modified glucagons that showed significant stabilization against fibril formation compared to unmodified glucagon. This was confirmed by three complementary biophysical techniques. Both successful modifications introduced excess net charge to glucagon, consistent with overall electrostatic repulsion being at the root of the observed fibrillation resistance. This approach can potentially be applied to other therapeutic proteins which suffer from the problems associated with self-association.

Entry for the Table of Contents

We have attached a wide range of blocking groups to glucagon with the intention of interfering with the quaternary interactions that drive self association. In so doing, we have identified multiple groups that are able to significantly block fibril formation.

Introduction

Protein quaternary interactions are a challenge for a majority of protein pharmaceuticals, as they can cause loss of bioactivity through the formation of both regular structures (e.g. fibrils) and irregular structures (e.g. aggregates).^[1] Both of these classes of structures can limit accessibility of these proteins to their targets. We are interested in developing general ways to prevent these types of interactions, in other words to antagonize quaternary interactions.

In previous work aimed at bringing the peptide hormone glucagon under the control of light, we found that the photocleavable modifications we introduced into glucagon were also able to prevent fibrillation.^[2] Specifically, we created a glucagon trimer in which three glucagon monomers were linked to a central hub via photocleavable linkers. This species exhibited no fibril formation even after 70 hours in solution, whereas control glucagon was observed forming fibrils at ~11 hours. We speculated that this reduction in fibrillation was brought about by the nature of the linkage between the individual glucagon molecules. This lead to the current work which focuses on the incorporation of other non-light-activated protein modifications that can achieve the same effect of blocking quaternary interactions. Here we have sought to understand what features in the blocking group are most effective at preventing fibril formation (Figure 1). In the long term, once we have determined the principles that determine the most effective blocking of fibril formation, we seek to make these modifications biochemically labile, such that they release native active protein after administration in-vivo.

Figure 1.

Overall approach. Protein (glucagon) self-interaction leads to aggregate (fibril) formation and inactivation. Incorporation of blocking groups (red circles) blocks protein self-interaction and fibril formation.

In this work we have explored this approach using glucagon as a test system. Glucagon is a peptide hormone that increases blood glucose levels and can be used both to rescue diabetics experiencing a hypoglycemic episode or as a component of a two hormone artificial pancreas system.^[3] In both applications, once in solution, glucagon suffers from rapid fibril formation.^[4] This fibrillation leads to insolubilization and inactivation of the glucagon.^[5] To address this rapid fibril formation we created a family of glucagons incorporating a wide range of peptides that were linked via a carbamate group through glucagon’s lysine amine. These peptides were varied to explore the effect of different physico-chemical properties (e.g. charge, polarity) on fibrillation. Using a panel of fibrillation assays we demonstrated that two of our synthesized species showed substantial and significant resistance to fibril formation over other modified glucagons and glucagon itself. We interpret these improvements as being rooted in the introduction of excess charge which then increases inter-protein electrostatic repulsion, making fibrillation consequently more difficult. We believe that this and related approaches can potentially be used to create a general solution to the significant problem of protein self-association.

Materials and Methods

All synthetic and characterization details, as well as assay conditions and methods may be found in the Supporting Information.

Results and Discussion

Our aim in this work was to systematically incorporate a wide range of modifications to the surface of glucagon, and then explore how effectively the modifications blocked quaternary interactions leading to fibril formation. To most efficiently do this, we made a central feedstock that we could then modify with any cysteine containing peptide or thiol terminated molecule (Figure 2). For this purpose we used the 2-(2-pyridyldithio) ethanol (PDE) linker. This has been used previously by others to modify proteins.^[6] While this linker has the potential to be reductively self-immolating, this was not our purpose in using it. Instead, we sought a modification from which we could rapidly make a wide range of modified glucagons from a single starting material (i.e. “Central Feedstock”).

Figure 2.

Synthetic approach. Glucagon modified with PDE at primary amines. Central feedstock reacted with cysteine containing peptides of different sequence.

We reacted the PDE hydroxyl group with carbonyl diimidazole (CDI) in DMSO for 60 minutes. We confirmed by HPLC and MS that the activated species formed readily. This activated PDE mixture was then reacted with glucagon over a period of 14 days. We followed the reaction over this period by HPLC, observing ~90% of the glucagon being consumed, and three new species appearing. At the same time, we treated a control sample of glucagon to identical conditions. For every subsequent step in the synthesis of the modified glucagons, we treated the control glucagon in an identical manner, to insure a fair comparison when fibrillation was ultimately examined.

We expect the activated PDE to react most rapidly with the two primary amines found within glucagon, the N terminus and the single lysine side chain. Thus, the three new species are consistent with the two possible mono-modification and the single possible di-modification. The resulting link between glucagon and PDE is a carbamate.

Both the main product (isomer I ~24%) and the secondary product (isomer II, 7%) were purified to >95% purity using semi-preparative HPLC and were confirmed by ESI-MS to be the PDE modified glucagon (figure 3). We assumed that isomer I was modified on the lysine epsilon amine based on our prior experience with this reaction. To confirm this assignment, we further examined the identity of isomer I using protease treatment and HPLC-MS. Pure isomer I was treated exhaustively with chymotrypsin which is expected to cleave after aromatic residues (F,Y,W), which would result in six peptides total. We were able to identify all of the resulting peptide masses in the resulting HPLC MS run, except for the SKY tripeptide. Instead we found the mass of SKY with the PDE incorporated (SKY-PDE expected mass: 610.2, observed mass: 610.0). Thus, we confirmed that isomer I was lysine modified PDE (Figure 4). In equivalent treatment of unreacted glucagon, we identified all six peptides, including the unmodified SKY peptide.

Figure 3.

Synthesis of glucagon PDE lysine regio-isomer: a) HPLC of crude reaction mixture. * = unreacted glucagon, & = mono-modified glucagon, minor isomer II, # = di-modified glucagon b) HPLC of purified lysine regio-isomer-I c) Deconvoluted ESI-MS confirming mass of regio-isomer-I.

Figure 4.

Identification of lysine modification site of major regio-isomer-I. a) HPLC of crude regio-isomer-I after chymotrypsin treatment b) purified new species c) ESI-MS showing mass consistent with SKY peptide modified by PDE. (see SI section for characterization of other peaks in this and control glucagon).

For the subsequent studies, we solely used isomer I (the “central feedstock”), to make pure single-isomer products. This was based on the hypothesis that this lysine, being centrally located in the helix, would have a greater impact on glucagon self-association that would the terminal amine modification. We then synthesized a series of modified glucagons by reacting the central feedstock with cysteine terminated tetra-peptides. The N terminal cysteine thiol readily reacts with the PDE group, to form a disulfide bond as the pyridyl thiol leaves. We wanted to systematically explore the effect of different physico-chemical modifications, so our peptides included anionic (CE3), cationic (CK3), polar uncharged (CQ3), neutral (CG3), mildly non-polar (CA3) and strongly non-polar (CF3). The peptides used were either synthesized via solid phase or purchased. All were acetylated on the N terminus and were C terminal carboxamides, so that the amino acid side chains were the sole determinants of the peptide charge properties. Part of the rationale of the charged peptides was to increase the net charge on the glucagon. Native glucagon has an isoelectric point near neutral and consequently has limited solubility at pH 7.^[2] Therefore it is typically formulated at acidic pH, both for use as a blood glucose rescue therapy, and for biophysical studies. We anticipated that adding positively or negatively charged peptides would shift the net charge from near zero at pH 7 to strongly anionic or cationic. Furthermore we believed that this excess charge could help individual modified glucagons repel each other.

The synthesis of final materials was performed by reacting the central feed stock isomer I with the specific peptide in DMSO overnight and monitoring via HPLC and MS. We observed rapid reaction nearing completion during this time. All proceeded smoothly with the exception of CF3, which formed an insoluble gel-like material during the overnight reaction. This is likely due to aggregate/fibril formation taking place. This is notable, given that the reaction took place in DMSO, and not aqueous solution. All of the resulting species (except glucagon-CF3) were purified using semi-prep HPLC. As previously described, control glucagon was generated by treating glucagon in an identical manner to our synthesized materials. This was done by setting up mock reactions, purification, dialysis and lyophilization (see supporting information for details). The intention here was to have a glucagon control that had undergone identical processing as had our target products, to insure that differences between modified and un-modified glucagons were due purely to their modifications and not to the environments they were exposed to. All of our control glucagon was prepared in this fashion, as it makes for a fairer comparison with our modified materials than would unprocessed glucagon. The fibrillation timing of this processed glucagon is similar to completely unprocessed glucagon (~7.5 hours versus ~4.5 hours respectively).

All purified fractions were dialyzed for 1.5 hours against 500ml of pH 7.0 100mM phosphate buffer. This was to insure that species were in the protonation state present at pH 7. It was confirmed that the pH of the protein sample was at 7, at which point they were then dialyzed against 1 L volume of deionized water for four hours, followed by a second overnight dialysis against 1 L of deionized water to eliminate excess buffer and salts. The materials were then lyophilized and stored as dried powders at −80°C until required for study. Portions were analyzed via analytical HPLC and ESI-MS. Masses were determined for multiple charge states, and the final mass determined using mass deconvolution. All final products have purities ≥95% and masses within 0.2% of the expected. Representative HPLC and MS are shown for CE3 in figure 5 and the HPLC and MS of the remaining species are detailed in supporting information. These results are also summarized in Table 1.

Figure 5.

Sample product (CE3) purification and characterization: a) HPLC of crude b) HPLC of purified Glucagon-CE3 c) Deconvoluted ESI-MS confirming molecular weight. Comparable data for all tested species may be found in Supporting Information.

Table 1.

Summary of species synthesized. All primary data may be found in the Supporting Information.

Species	% Purity by HPLC	Expected mass	Observed mass
Glucagon-CE3	99.0%	4136.2	4136.0
Glucagon-CK3	99.5%	4133.3	4132.0
Glucagon-CQ3	97.3%	4133.2	4133.0
Glucagon-CG3	96.5%	3920.1	3919.0
Glucagon-CA3	95.8%	3962.2	3960.0

We used a panel of fibrillation assays to determine the kinetics of fibrillation. These were the thioflavin-T (ThT) assay, tryptophan fluorescence and turbidity. Each of these assays assesses fibrillation in different and complementary ways, that combine to give an accurate composite assessment of fibrillation resistance. For example, thioflavin-T is a fibril specific fluorescent reporter, that increases in signal with increasing amounts of fibril present.^[7] Because of this, there is a potential that the reporter binding itself can influence fibril formation. Tryptophan fluorescence has no such issue, as it is based on the inherent fluorescent signal of tryptophan contained within the protein. The fluorescent signal is reduced in the fibril form.^{[7a, 7b, 7d, 8]} Finally turbidity is a gross measure of the amount of fibril formation, as fibrils are large and scatter light (resulting in lower “absorbance”) compared to monomeric glucagon.^{[5b, 7b, 9]} With each of these techniques, we observe the signal as a function of time and use the time of the first inflection point as the measure of fibril formation.

We first attempted to dissolve lyophilized glucagon control or modified glucagons in a typical assay buffer (10 mM phosphate buffer saline with 150 mM sodium chloride). Because native glucagon has very low solubility at pH 7, an acidic pH of 2.5 has typically been used in studies of glucagon fibrillation.^{[7d, 10]} We used a stock concentration of 286.9 μM (this is equivalent to a concentration of 1.00 mg/ml glucagon, a value previously used.^{[7c, 11]} We converted to molarity to insure equal true concentration of compared materials). Glucagon control, CQ3, CG3, and CK3 were soluble at pH 2.5 and at the concentration of 286.9 μM. CE3 and CA3 were not soluble at pH 2.5, so the pH was adjusted to 7.2 for CE3 and 7 for CA3 at which they were soluble.

For fibrillation studies, the materials were added to buffer and gently dissolved by repeated aspiration using a pipettor. They were then immediately filtered through a 0.22 μm filter to eliminate any nuclei existing at the start of the experiment. “Time 0” was counted as the point at which the filtered material was transferred to plates for the assay. We used the plate reader vortexing mode for 5 seconds to agitate samples prior to readings every 15 minutes. This method of agitation has been used previously to accelerate fibril formation.^{[10, 12]} For the thioflavin-T experiments, 50 μM thioflavin-T was included and signals were generated with excitation and emission wavelengths of 440 nm and 482 nm. The observed time based thioflavin-T signals typically show sigmoidal shapes, consistent with slow initial nucleation of monomers, followed by a more rapid addition of subsequent monomers to these nuclei. The resulting curves were fit using a modified “Boltzmann” equation as previously used, giving the time of transition (essentially the inflection point for the curve).^{[7b, 13]} Figure 6 shows one replicate of the transition curves for glucagon and the five modified glucagons. The fit curves are indicated as well as the fit transition time. Each of these analyses was run in triplicate, and fit identically. These additional replicates are shown in the supporting information. To compare materials and determine if modified materials fibrillation times were significantly improved relative to unmodified glucagon, we performed a one way ANOVA analysis. This was made possible by the triplicate determinations of transition times, which allowed us to determine the variance of these values for each material. The ANOVA analysis of these then were used to identify pairs of materials whose increase in transition time had a P value < 0.001. These results are shown in figure 7.

Figure 6.

Fibrillation of glucagon and derivatives, followed by thioflavin-T fluorescence. Representative replicate shown, others in Supporting Information.

Figure 7.

Summarized fibrillation results using thioflavin-T. n=3, standard errors indicated. * indicates p<0.001 by ANOVA

For this run, unmodified control glucagon showed a transition time of 7.1 hours similar to what we have previously observed for unmodified glucagon. This value was very similar to the glycine modified species (CG3) at 7.9 hours as well as the CQ3 species (9.4 hours). On closer inspection, we observed that the curves for both the CQ3 material and the CA3 material suggest that fibrillation has actually started prior to “Time 0”, the point at which the fluorescence measurements started. Specifically, there is little or no initial plateau prior to the transition, again suggesting that fibrillation had already started. This was confirmed visually as we observed opalescence in the CQ3 and CA3 samples prior to being analyzed by plate reader.

The key observation is that both species that strongly shift the net charge on the glucagon, CE3 and CK3, have significantly delayed fibrillation times (38.6 and 28.9 hours respectively for this replicate). Our interpretation of this is that net excess charge makes the close approach required of nucleation more difficult, hence the long delay prior to transition. Consistent with this, the effect of the anionic CE3 species is greater than the cationic CK3 species. Because the carbamate forming reaction removes one positive charge (the amino group of the lysine) the actual net change in charge is only ~ +2 for CK3, whereas the net change in charge is ~ −4 for CE3. Thus the greater resistance to fibril formation.

Our next method of analysis was turbidity, which was monitored as “absorbance” at 405nm. Monomeric glucagon has no absorption at this wavelength, so the appearance of absorption is due to light scattering off of increasingly large fibrils. As nucleation progresses fibrils form and grow, scattering increasing amounts of light. This leads to similar sigmoidal plots similar to those observed with the thioflavin T method. Figure 8 shows the time dependence of this signal for glucagon and the tested derivatives in a single replicate. The remaining two replicates for all 6 species can be found in the supporting information. The glucagon control shows a typical sigmoidal transition and a fibrillation time of 6.8 hours, comparable to that found using ThT. The CG3 species is similar with a 4.8 hour transition time. Both CQ3 and CA3 glucagon show poor fits to the sigmoidal function, as also seen with thioflavin-T, and consistent with fibrillation beginning before the start of measurements. The fit transition times of 3.5 and 9.1 hours can not be considered reliable.

Figure 8.

Fibrillation of glucagon and derivatives, followed by turbidity. Representative replicate shown, others in Supporting Information.

Paralleling the ThT results, CE3 and CK3 glucagons give robust increases in fibrillation resistance with transition times of 62.8 hours and 28.6 hours respectively. Transition times for all three replicates were determined, averaged and tested for significant differences using an ANOVA analysis as shown in figure 9. As with the results from the thioflavin-T assay, CE3 and CK3 showed significant increases in resistance to fibrillation by turbidity analysis.

Figure 9.

Summarized fibrillation results using turbidity. n=3, standard errors indicated. * indicates p<0.001 by ANOVA

Our final method of analysis was tryptophan fluorescence using excitation and emission frequencies at or near 295 nm and 355 nm respectively (wavelengths were adjusted by a small increment for samples going off scale. See Supporting information for more information). Samples were prepared and fibrillation observed in a similar fashion as with the other methods used. Figure 10 shows the results from one replicate of these runs. In a similar fashion to the thioflavin T and turbidity results, we observe sigmoidal curves, only “inverted”, as tryptophan fluorescence decreases as the monomer is incorporated into fibrils. These results closely parallel the thioflavin-T results. The glucagon control has a similar transition time of 6.5 hours. The CG3 sample has a similar transition of 7.4 hours, and the CA3 derivative is relatively fast at 3.9 hours. As with the thioflavin-T and turbidity results, both the CQ3 and CA3 derivatives have poorly fit equations. The transition time determined from this fit for CQ3 is 64.1 hours, inconsistent with the other two methods used. This may be an artifact of the fitting process for this material, linked to the observation that this material starts fibrillating before the start of the assay. The data for CQ3 and CA3 again also show signs of pre-fibrillation (prior to the start of the run) and this was confirmed visually by observing opalescence in the samples.

Figure 10.

Fibrillation of glucagon and derivatives, followed by trypotphan fluroresence. Representative replicate shown, others in Supporting Information.

We again observe that the CE3 and CK3 derivatives are strongly resistant to fibril formation, with fibrillation times of 48.7, 28.4 hours respectively. The average of the transition times of three replicates are shown in figure 11. Transition times for all three replicates were determined, averaged and tested for significant differences using an ANOVA analysis as shown in figure 11.

Figure 11.

Summarized fibrillation results using tryptophan fluorescence. n=3, standard errors indicated. * indicates p<0.001 by ANOVA

Thioflavin-T, tryptophan fluorescence, and turbidity give a consistent pattern of fibrillation resistance for the six tested species. With the exception of CQ3 in tryptophan fluorescence, all techniques gave a consistent picture of fibrillation resistance in the following order CE3 > CK3 > CG3 ≈ glucagon > CQ3 ≈ CA3. CE3 and CK3 consistently show significant fibrillation resistance, and our working hypothesis is that the greater the excess charge on the modified protein, the greater the difficulty of nucleating a fibril, hence the delay in transition time. As mentioned previously, glucagon’s pI is near neutral, so we would expect the addition of both cationic groups or anionic groups to increase the overall net charge, and hence lead to fibrillation resistance. Also consistent with our hypothesis is the greater effect of the CE3 addition over the CK3 addition: The carbamate linkage already removes one + charge (the lysine amine) from glucagon, so the net effect of the CE3 addition is approximately −4, whereas the net effect of the CK3 addition is approximately +2. Hence the CE3 addition has a greater stabilizing effect.

Conclusion

In this work we have demonstrated that strategic covalent modifications significantly blocks the formation of fibrils in glucagon. We don’t know if our modified glucagons are bioactive as is, as the free lysine may be needed for activity. Given this, our long term aim is for our modified proteins to be rapidly converted into native proteins when used in-vivo, through the action of an endogenous metabolic process on the linker between the protein and the modification (the prodrug approach as others are pursuing).^[14]. In our system the intention behind the carbamate linkage was to act as a cleavable linker, as there are multiple examples of carbamate prodrugs being cleaved by esterases to reveal the constituent amine. We have preliminarily examined this potential cleavage of the glucagon-CE3 species using human plasma, but to date have inconclusive evidence of native glucagon release. In the long term, we hope to explore this and other, potentially more cleavable linkers. The conclusion from our work is that such modification can dramatically prevent fibrillation. The pursuit of cleavability is the next phase in this work.

The overall goal of the described work is to develop an approach for blocking quaternary interactions. We refer to this process as “Quaternary Antagonism”. These interactions are at the root of multiple processes that reduce or eliminate the efficacy of protein drugs. Examples include fibrillation in glucagon, hexamer formation in insulin^[15], and aggregation in antibodies.^[16] We are using peptides initially, because they represent a relatively easy route to diverse physico-chemical properties. There can potentially be many ways in which a surface modification could antagonize quaternary interactions: through steric interference at the monomer-monomer interface, through increased gross electrostatic repulsion between monomers, or though more subtle interactions rooted in the specific sequence of peptides employed. We chose to initiate these studies by using peptides that divide molecule space grossly (anionic, cationic, polar, neutral, non-polar) and in so doing revealed the importance of net charge in reducing the rate at which fibrillation takes place. Our observation of the influence of excess charge on the fibril forming tendency of glucagon is consistent with the observations of prior authors who showed that net charge slowed fibril formation in other peptides (amyloid, calcitonin e.g.).^[17] Our aim is to accomplish this, not by changing the sequence, or by manipulating pH, but by the temporary addition of charge shifting (or other) peptides.”

The CK3-modified glucagon, which resulted in a net charge increase of approximately +2, exhibited significant resistance to fibrillation. This resistance was even more pronounced in the CE3-modified glucagon, which had a net charge change of approximately −4. The “neutral” modification of CG3 showed fibrillation that was similar to unmodified glucagon, suggesting that sterics alone with these tetra peptides is insufficient to antagonize quaternary interactions and slow fibril formation. The fact that the CQ3 and CA3 modifications showed signs of early fibril formation is not surprising. Glutamines are notorious for being present in beta sheet forming peptides (e.g. the Huntingtons peptide), and can drive this formation with bi-dentate intermolecular hydrogen bond interactions.^[18] The mildly non-polar CA3 modification could similarly drive fibril formation through intermolecular hydrophobic interactions. Indeed with the even more non-polar CF3 modified glucagon, signs of fibril formation were present, even during synthesis of the product.

The results presented suggest that the overall approach of quaternary antagonism is sound. Depending on the assay used, we have increased fibril resistance in the model protein glucagon between 3 and 6 fold. Given this success, the system is highly un-optimized, and many potential avenues remain for significant improvements, including ultimately biochemical lability of the attached groups. The ultimate benefit of these gains will be the development of a general approach that can increase the utility and performance of protein drugs, arguably the most important class of therapeutics for the foreseeable future.^[19]

Supplementary Material

Supporting information for this article is given via a link at the end of the document

This file contains details of synthetic and analytical procedures, as well as all mass spectrometry and HPLC traces. In addition, all primary data replicates of assays are included. The authors have cited additional references within the Supporting Information.