Peptide Aggregation and Fibrillation: Why Solutions Go Cloudy

Formulation science

Peptide aggregation is the most common way a peptide solution fails without anything obviously reacting: the molecules do not break, they simply find each other. In its most ordered form aggregation becomes fibrillation — the peptide abandons its native fold, stacks into a cross-beta lattice, and becomes amyloid. Insulin has been caught doing this since the 1940s, and almost everything the field knows about preventing it was learned from that one molecule.

Chemical degradation and physical degradation are different problems

Peptides fail in two structurally separate ways. Chemical degradation changes covalent bonds — deamidation, oxidation, hydrolysis. Physical degradation changes only shape: the chain unfolds, exposes surfaces it normally buries, and sticks to copies of itself. Nothing is cut, nothing is added, and the primary sequence you would read off a mass spectrum is unchanged.

Exenatide is the cleanest published demonstration that these are separate processes with different optima. In a 2021 stability study the peptide was relatively stable at pH 4.5 at 37 degrees C; between pH 5.5 and 6.5 degradation was driven by oxidation; between pH 7.5 and 8.5 it was driven by deamidation; and significant aggregation appeared at pH 7.5 and 8.5. The molecule carries two likely deamidation sites (N28, Q12) and two likely oxidation sites (M14, W24), while its physical failure traces to disruption of the C-terminal Trp-cage at residues 29 to 39. Same vial, three different clocks running at once.

How peptide aggregation actually starts

Fibril formation is nucleation-dependent polymerisation, and it plots as an S-curve: a flat lag, a steep elongation phase, then a plateau. The intuitive reading of that curve — nothing happens, then nucleation happens, then growth happens — is wrong, and the arithmetic is not close.

Take the worked example used by Arosio, Knowles and Linse: 100 microlitres of 4 micromolar amyloid-beta 42, entirely monomeric at time zero, with a measured lag of 33 minutes. By the end of that lag roughly 600 million primary nuclei have formed. The average time to the first nucleus is about 3 microseconds — one ten-millionth of one percent of the lag. The flat stretch is not an absence of nucleation. It is exponential amplification running below the detection limit of the dye.

The amplifier is secondary nucleation: existing fibrils present a catalytic surface on which new nuclei form far faster than they do in free solution, so once fibril concentration passes a threshold, secondary events dominate. Two consequences follow. First, as the authors put it, the lag phase should not be identified with a nucleation period at all — nucleation is more frequent during the growth phase. Second, changing the elongation rate constant shifts the lag far more than changing the primary nucleation constant does. This is also why seeding matters: adding pre-formed fibrils supplies growth-competent ends and catalytic surface directly, skipping the amplification the lag was measuring.

Before any of that is visible, the peptide has to loosen. In stirred monomeric insulin, species that bind the hydrophobic probe ANS appear at 2.5 hours, before any fibrillation is detectable, and hydrogen-deuterium exchange shows ten previously buried protons becoming solvent-accessible. The secondary structure then converts: the infrared amide band moves from 1654 per cm (alpha-helix) to 1628 per cm, reaching about 52 percent beta-sheet by 28 hours. Because that end state depends on backbone geometry rather than sequence, structurally unrelated proteins converge on fibrils of the same 10 to 18 nanometre width with the same cross-beta core.

Insulin: the case with the most data

Insulin exists as an equilibrium of hexamers, dimers and monomers, and which one dominates is set entirely by conditions — predominantly monomeric in 20 percent acetic acid, dimeric in 20 mM HCl, hexameric at pH 7.5 in the presence of zinc. Fibrils grow from the monomer, so the working model is hexamer to modified monomer to fibril, with hexamer dissociation as the rate-limiting step. That single fact explains most of an insulin formulation: zinc coordinates and stabilises the hexamer, and the phenolic preservatives that keep the vial sterile also lock in the hexamer conformation.

Concentration and mechanical energy do the rest. Human insulin stirred at 600 rpm at 37 degrees C fibrillates with a lag of 11.4 hours at 2 mg/mL and 5.0 hours at 4 mg/mL — same temperature, same stirring, half the wait for twice the concentration. Anyone who has watched a reconstitution go wrong after vigorous shaking is watching the extrinsic half of that equation.

What a cloudy vial actually tells you

Less than you would like. When commercial insulin formulations were deliberately stressed, physical agitation produced precipitate that was in most cases amyloid, and it formed preferentially near neutral pH. Dropping the pH toward the isoelectric point also produced a precipitate — but that one was not amyloid. Drop the pH further and amyloid returned, so both the positively and negatively charged states were amyloid-prone while the neutral-charge state simply fell out of solution intact.

Two chemically different failures, one identical visual cue. That is precisely why regulated labelling does not ask anyone to distinguish them. FDA-approved insulin labelling says to inspect the product before use and not to use it if particulate matter or coloration is seen; the patient-facing text is blunter still — do not use it if it is thick, cloudy, coloured, or contains lumps or particles.

The pharmacopoeial standard behind that language is more nuanced than it looks. USP General Chapter 790 defines essentially free of visible particulates as an inspection outcome, not a claim of zero particles, and it distinguishes extraneous particulate matter — fibres, glass, metal, elastomer, precipitates — from inherent particles or agglomerates, which some products legitimately contain and which are handled in the individual monograph. USP 1790 adds the honest caveat that visual inspection is a probabilistic process whose detection rate varies with particle size, shape, colour, density and container design.

When it happens inside a person

Injection-site amyloid is not a theoretical concern; it appears on FDA-approved labelling. The insulin lispro label lists localised cutaneous amyloidosis at the injection site as a post-marketing adverse reaction, notes hyperglycaemia with repeated injection into affected areas and hypoglycaemia on sudden switching to an unaffected site, and instructs rotation within the same region to reduce the risk.

The measured effect is large. In one controlled comparison the area under the serum insulin curve at a previously used site was 320 pmol/L·h against 798 pmol/L·h at a normal site — absorption reduced to at most 40 percent; a CMAJ review puts it at about one-third. In a published case report a patient whose requirement had climbed to 192 units per day at an HbA1c of 9.7 percent came down to 48 units per day after nothing more than changing injection sites. This is reported clinical literature, not guidance for anyone reading it.

It is also routinely mistaken for lipohypertrophy, which occurs in as many as 50 percent of patients; imaging can suggest the difference but definitive diagnosis is histological, and amyloid regresses slowly once injections stop while lipohypertrophy resolves faster. A 2016 review found 86 published case reports, 96.5 percent of which described a palpable subcutaneous mass — and later work confirmed the lesion can be present with no palpable mass at all. Multiple authors state plainly that true prevalence is probably underestimated.

What formulations actually do about it

• Hold the pH away from the aggregation window. Exenatide’s stable point was pH 4.5; commercial semaglutide sits at approximately pH 7.4 with the peptide engineered to tolerate it.
• Keep the peptide assembled. Zinc plus a phenolic preservative holds insulin as a hexamer. In stress testing, m-cresol directly suppressed amyloid formation — the preservative is quietly doing double duty.
• Protect the interfaces. Agitation, air-water interfaces, foreign surfaces and shear are all listed extrinsic accelerants; surfactants exist in parenteral formulations largely to occupy those interfaces.
• Do not rely on sugars alone. Mannitol, sorbitol and sucrose conferred only small protective effects against exenatide aggregation at pH 7.5.
• Remove the water. Lyophilisation stops the whole reaction by removing the solvent it needs — which is why the dry vial is the stable form and the reconstituted one is not.
• Respect the physical limits. Labels say do not freeze, protect from heat and light — see freeze-thaw damage, container choice and light exposure — and the seven-day pump reservoir limit is tied explicitly to loss of the preservative.

The same physics runs the amyloid diseases

Fibril-catalysed secondary nucleation has been observed for amyloid-beta 40 and 42 and for alpha-synuclein. Islet amyloid polypeptide — the pancreatic peptide better known as amylin — deposits in the islets in type 2 diabetes, and those deposits are associated with loss of beta-cell mass, with the cytotoxic species apparently forming in the very earliest stages of aggregation. Calcitonin fibrillates too. The dyes are the same in every case: Congo red with apple-green birefringence, thioflavin S or T under fluorescence.

Frequently asked questions

Does a clear solution mean there is no aggregation?

No. The lag phase is defined by the sensitivity of whatever you are measuring with, and fibrils at nanomolar concentrations can be detected within minutes of a lag phase that has hours left to run. Clarity means aggregation has not crossed a detection threshold, not that it has not started.

Can aggregation be reversed?

Cross-beta fibrils are an extremely stable end state, and the field treats fibrillation as one-way in practice. The isoelectric case is different in principle — that precipitate retains its structure — but nothing about the appearance of a container lets anyone tell which situation they are looking at, which is why the labelled instruction is simply not to use it.

Why does shaking matter so much?

Mechanical shear, air-water interfaces and foreign surfaces all appear on the published list of extrinsic accelerants, and the standard laboratory method for producing insulin fibrils on demand is a magnetic stirrer at roughly 600 rpm. Swirling rather than shaking is not folklore; it is the same variable, turned down.

Does aggregation reduce potency?

Mechanistically it must — fibrils sequester monomer, and monomer is the active species. A reliable universal percentage does not exist in the literature. The one well-measured human number is absorption from tissue containing insulin amyloid, which runs at roughly 40 percent, or about one-third, of a normal site.

References

1. Arosio P, Knowles TPJ, Linse S. On the lag phase in amyloid fibril formation. Phys Chem Chem Phys 2015;17:7606–7618. Open access
2. Ahmad A, Uversky VN, Hong D, Fink AL. Early events in the fibrillation of monomeric insulin. J Biol Chem. Open PDF
3. Halseth T, et al. The effects of pH and excipients on exenatide stability in solution. Pharmaceutics 2021;13:1263. PMC8398870
4. Investigation of factors that cause insulin precipitation and/or amyloid formation in insulin formulations. J Pharm Health Care Sci 2019. Full text
5. Frankær CG, et al. Insulin fibrillation: the influence and coordination of Zn2+. J Struct Biol 2017;199:27–38. Open PDF
6. Kano Y. Insulin-derived amyloidosis. CMAJ 2022;194:E1616. CMAJ
7. Nagase T, et al. Insulin-derived amyloidosis without a palpable mass at the insulin injection site. J Diabetes Investig 2020;11:1002–1005. PMC7378411
8. United States Pharmacopeia. General Chapter 790, Visible Particulates in Injections. USP-NF
9. DailyMed. Insulin lispro injection — FDA-approved labelling. DailyMed

Informational only — not medical advice. VialHelp does not sell peptides and does not recommend any product, treatment or dose. Prescription medicines, including insulin and GLP-1 receptor agonists, should be used only under the direction of a qualified healthcare professional, and nothing here should be used to alter a prescribed regimen. Intended for readers 21+.