A peptide in solution is not a static object. Chains sample conformations, collide, and occasionally stick. Most encounters come apart again; a small fraction do not, and if the associated state adds further chains faster than it falls apart, material begins to leave solution. It can leave in two forms: disordered amorphous clumps, or ordered cross-beta fibrils. This guide covers the chemistry of both, how they are measured, and why a vial on a shelf obeys the same rules as a plate in a kinetics reader.

Aggregation is a family of processes, not one process

The word aggregation covers several mechanistically distinct things. Amorphous aggregation is fast, largely non-specific, and produces particles with no long-range internal order. Fibrillation is slower, sequence-specific, and produces filaments built from a repeating cross-beta architecture in which the backbone hydrogen bonds run parallel to the fibril axis. The two compete for the same monomer pool, and in the systems where this competition has been mapped, changing salt and temperature alone is enough to switch which one dominates.

That competition has been mapped directly. Working with acid-denatured beta-2-microglobulin, one group combined thioflavin fluorescence, circular dichroism, light scattering and atomic force microscopy into a unified temperature-and-salt phase diagram, and found amorphous aggregates forming rapidly at ambient temperatures under high salt while the conversion of those amorphous aggregates into amyloid occurred only as temperature increased [1]. They also report that the diagram is deformed by how fast the salt or heat is titrated, with amyloid regions disappearing at fast titration rates [1]. Aggregation state, in other words, is not a pure function of the final buffer composition. It carries history.

The sigmoid and what its parts mean

Plot aggregate mass against time for most fibrillating peptides and you get a sigmoid: a flat lag region, a steep growth region, a plateau. Reading those three regions as nucleation, elongation and equilibrium describes the shape, not the mechanism, and treating it as a mechanism is a common error.

The modern approach fits a chemical master equation for the whole reaction network to a family of curves collected at different starting monomer concentrations, rather than extracting a lag time from a single trace. A widely used protocol describes determining constraints on the mechanism from the concentration dependence of the reaction, choosing among fundamental models for assembly into linear aggregates, and globally fitting to yield reaction orders and rate constants; it opens with extensive data quality control before any fitting [2]. A lag time on its own carries very little mechanistic information. Its scaling with concentration carries considerably more.

A lag phase also does not require a rare nucleation event. In a study of human insulin at low pH across a range of temperatures and concentrations, the authors combined time-lapse atomic force microscopy, quasi-elastic light scattering, small- and large-angle static light scattering, thioflavin T fluorescence and optical microscopy, and concluded that fibrillar aggregates grow by a thermally activated heterogeneous coagulation mechanism involving both tip-to-tip elongation and lateral thickening, with the simultaneous association of fibrils into bundles and larger clusters responsible for what they call an effective lag time [3]. In other words, an apparent lag can be produced by a growth process that was running the whole time but was invisible to the readout until the assemblies got large enough.

Secondary nucleation makes the curve autocatalytic

If new aggregates could only form from free monomer, the concentration dependence would be simple. In many systems it is not, because existing fibrils generate new ones by two routes. Fragmentation breaks a filament and thereby changes the total number of filament ends, which act as growth sites; an analytical treatment of the master equation for breakable filament assembly derives the time evolution of the filament length distribution and applies it to experimental length distributions of insulin amyloid fibrils, with the authors arguing that this yields microscopic insight beyond what the conventional analysis of filament mass alone provides [4].

The second route is surface-catalyzed secondary nucleation, in which monomers nucleate on the sides of existing fibrils. This has been observed directly. Using stochastic optical reconstruction microscopy with separate fluorophores in seed fibrils and in monomers, one group watched monomers of the amyloid beta 42 peptide grow into relatively large aggregates on fibril surfaces along the length of the fibrils before detaching [5]. When the same monomers were presented with mutant fibrils that do not catalyze their nucleation, the monomers were still seen to interact with those non-cognate surfaces, but no growth along them was detected; the authors take this to support a templating role for secondary nucleation, and note it implies the failure is more likely a lack of structural conversion than a lack of association [5]. That is a carefully bounded inference and it is worth keeping it bounded.

Foreign surfaces do the same job. Copolymer particles, cerium oxide particles, quantum dots and carbon nanotubes were all reported to shorten the nucleation lag phase of beta-2-microglobulin, by an amount depending on the quantity and nature of the particle surface; the authors observed exchange of protein between solution and the surface, with multiple layers forming on the particle, and proposed that the locally increased protein concentration promotes oligomer formation by surface-assisted nucleation [6]. A dust particle, a filter fiber or a scratch on a vial is a surface too.

Thioflavin T reports rigidity, not fibril mass

Thioflavin T is the default readout, and understanding what it measures explains most of the ways it fails. ThT is a molecular rotor. Studies of its quantum yield and fluorescence lifetime as a function of temperature and glycerol content in water-glycerol mixtures established that the low fluorescence intensity in low-viscosity solvents is caused by non-radiative deactivation of the excited state through torsional motion of the benzothiazole and aminobenzene rings relative to one another, taking the molecule into a non-fluorescent twisted internal charge transfer state; the rate of that process is set by solvent viscosity, and quantum-chemical calculations put the ground-state torsional minimum near 37 degrees and the excited-state minimum near 90 degrees [7]. The authors take these data to support the idea that the characteristic intensity increase on incorporation into fibrils is determined by the rigidity of the dye environment, which prevents the ring rotation in the excited state [7].

So the signal is a report on how tightly the dye is held. It is not a stoichiometric measure of fibril mass, it is not calibrated across different fibril polymorphs, and it says nothing about species that do not bind the dye.

Where the ThT assay misleads

Two well-documented failure modes deserve naming. The first is optical interference from co-solutes. Compounds including Basic Blue 41, Basic Blue 12, Azure C and tannic acid were shown to decrease the fluorescence of ThT bound to insulin fibrils at submicromolar or micromolar concentrations, yet kinetic experiments showed they neither decreased the fibrillization rate nor increased the lag period, and soluble insulin was absent at the end of the experiment; the authors concluded these compounds are not fibrillization inhibitors at the concentrations studied and that interference with the ThT test is a general phenomenon [8]. Anything colored, anything that absorbs near the ThT excitation or emission bands, and anything that binds the dye can produce an apparent inhibition that is purely optical.

The second is adsorption. Amyloid fibrils adsorb to the plastic surfaces of multiwell plates and Eppendorf tubes, and one group reported that this adsorption is markedly increased in the presence of complex biological mixtures subjected to a denaturing air-water interface; a time-dependent loss of ThT fluorescence that had previously been interpreted as disaggregation was traced to increased adsorption as the added proteins denatured at the interface under agitation, and the authors concluded that their original interpretation had been invalid [9]. A falling ThT signal can mean less fibril, or it can mean the fibril has left the optical path.

Light scattering sees a different population

Scattering readouts are orthogonal to ThT: they respond to particle size and mass without requiring a dye-binding site, so they detect amorphous material that ThT is blind to, and they are dominated by the largest species present. The insulin study above used quasi-elastic and static light scattering alongside ThT precisely because the techniques report on different length scales, from monomeric protein through elongated fibrils to aggregate networks tens of microns across [3]. Disagreement between a ThT trace and a scattering trace on one sample is informative rather than embarrassing: it usually means ordered and disordered material are present in different proportions, or that something has sedimented.

Interfaces and mechanical stress

Interfaces dominate aggregation behavior far more than most protocols acknowledge, and the arithmetic explains why. Consider a flat-bottomed well 6.4 mm across holding 100 microliters. The air-liquid area is pi times 3.2 mm squared, about 32.2 square millimeters, over a volume of 100 cubic millimeters: roughly 0.32 square millimeters of interface per cubic millimeter of solution. Now take 1 mL in a vial of 11 mm internal diameter. Pi times 5.5 mm squared is about 95.0 square millimeters, over 1000 cubic millimeters: about 0.095 square millimeters of interface per cubic millimeter of solution. The denominator in both cases is the solution volume, and the well presents roughly 3.4 times more air-liquid interface per unit volume. Wall contact scales the same way. A plate assay and a vial are not the same experiment.

Mechanical stress compounds this. In a study of the peptide hormone glucagon, spectroscopic and microscopic data showed that fibrillar polymorphism could be induced by mechanical stress alone, producing two classes of fibril, low-stress and high-stress, differing in kinetic profile, secondary structure and morphology, each able to self-propagate in a template-dependent fashion; the authors proposed a model in which interfaces play a fundamental role in the switch between the two classes, and raised the cautionary note that mechanical perturbation is a potential source of variability in the study of fibrillation mechanisms and fibril structures [10]. Shaking does not simply speed things up. It can change what is made.

Surfactants are the standard countermeasure, and they are only a partial one. A comparison of polysorbate 20, polysorbate 80, poloxamer 188 and Brij 35 across interfaces of differing hydrophobicity found that all four non-ionic surfactants stabilized the four antibody drugs tested against the air-water interface, but none of them protected against a hydrophilic charged cellulose surface, and poloxamer 188 had a negligible stabilizing effect against cyclo-olefin copolymer and a model hydrophobic material [11]. That study is on antibodies rather than short peptides, so the specific rankings should not be carried over, but the structural point that protection is interface-specific is general.

Supersaturation, and why storage is the same problem

Fibril formation from a soluble peptide is a supersaturation-limited process, closer to crystallization than to a simple bimolecular reaction. Beta-2-microglobulin at neutral pH was shown to form fibrils efficiently when heated with agitation under high-salt conditions, and the authors constructed temperature- and salt-dependent conformational phase diagrams with and without agitation, framing fibril formation as requiring the breakdown of supersaturation; they also observed that the fibrils depolymerized at both lower and higher temperatures, which they compare to cold- and heat-induced denaturation of globular proteins [12]. A solution can therefore sit metastable for a long time and then convert quickly once the barrier to breaking supersaturation is crossed, by agitation, by an introduced surface, or by a seed.

This is why storage is the same problem. A stored solution is a supersaturated solution with time on its hands, exposed to a headspace interface, container walls, thermal cycling on every removal from a refrigerator, and any particulate that came in with the diluent. Freeze-drying removes the solvent all of this requires, which is why research peptides are usually shipped as a solid.

Sample history also propagates into subsequent experiments. Comparing two common preparation routes for the amyloid beta 1-40 peptide by capillary electrophoresis, one group found that size-exclusion-purified material initially contained more small species, including monomer, than freshly dissolved hexafluoroisopropanol-pretreated material, and that the lag time to oligomer formation for the size-exclusion-isolated samples was about 23 hours shorter [13]. Two aliquots of the same peptide, prepared differently, are not kinetically equivalent starting materials.

One more piece of arithmetic makes the storage point concrete. If a 5 mg/mL solution converts 1 percent of its peptide to fibril, that is 0.05 mg/mL, or 50 micrograms per milliliter, of aggregate. Against a purity assay reporting whole percentage points, the denominator being total peptide, that conversion is close to invisible. Against a seeding experiment, 50 micrograms per milliliter of pre-formed fibril is a great deal of catalytic surface. Aggregation can be analytically silent and kinetically decisive at the same time.

What is not settled

Several things here remain genuinely open. The structure of the critical nucleus is not directly observable for most sequences, and rate constants for primary nucleation are quantities extracted from global fits to mass-versus-time data rather than direct counts of species [2]. The relative weighting of fragmentation against surface-catalyzed secondary nucleation differs between systems and between conditions within one system, and a mechanism established at one pH does not transfer to another. Whether an observed rate change is a true chemical effect or an optical artifact of the dye is a live question in any experiment using a fluorescent reporter [8], and the same caution applies to any signal loss that could instead be adsorption [9]. Extrapolation from the well-studied model peptides, chosen for tractability, to an arbitrary synthetic sequence is an assumption rather than a result. For most research peptides no aggregation phase diagram has been published, and the honest position is that the behavior is unknown rather than benign.

Practical implications for handling

The chemistry above translates into a few habits, most of them standard practice rather than findings from any one paper.

  • Treat the container as part of the experiment. Well geometry, plastic type and fill volume change the interfacial area per unit volume, and therefore the result.
  • Record agitation conditions explicitly, because quiescent and shaken samples of one peptide can give different kinetics and, in at least one reported case, different fibril classes [10].
  • Prefer orthogonal readouts. A dye signal and a scattering signal fail in different ways.
  • Keep the preparation route fixed across a comparison series, since it is known to shift lag times on its own [13].
  • Do not read a falling fluorescence signal as dissolution without first excluding adsorption to the vessel [9].

None of this makes a peptide aggregation-proof. It makes the aggregation that does occur legible, which is the realistic goal.