Getting a peptide to crystallise: screens, seeds and salt tests

Crystallisation asks for a solution that is supersaturated in the macromolecule while the conditions remain mild enough not to significantly perturb its natural state; supersaturation is produced by adding mild precipitating agents such as neutral salts or polymers and by manipulating parameters including temperature, ionic strength and pH ^[1]. Both halves of that requirement bind. Push the peptide out of solution hard and it precipitates amorphously; push gently and nothing leaves solution at all. A screen is a search for the narrow band between those outcomes, and the peptide-specific difficulty is that reaching it can demand a higher peptide concentration than a protein would, for reasons set out below.

The geometries, and what each one controls

Vapour diffusion, dialysis, batch and liquid–liquid diffusion are among the most widely used approaches ^[1]. The distinction that matters at the bench is whether the drop travels through the phase diagram or begins at its destination.

Vapour diffusion travels: the drop and the reservoir exchange water through the vapour space, so peptide and precipitant concentrate together along a trajectory. How fast that happens is a variable you control rather than a property of the plate. Measured on model salt drops — 1.00 M NaCl equilibrating against 2.00 M NaCl reservoirs at room temperature — equilibration was slower the greater the distance separating drop from reservoir over the range 7.6 to 119.4 mm, the dependence was non-linear, and it was most sensitive to that distance at short separations; the data fitted a model in which the rate-limiting step is transit of water across the vapour space ^[2]. Increasing the separation is therefore a cheap, direct way to slow the approach to supersaturation — the first thing to reach for when a condition delivers a shower of microcrystals instead of a few.

Batch does not travel. The drop is at its final composition from the moment it is mixed, so it interrogates one point without a path. Under oil, the oil is not inert packaging: it can play a part in the control of nucleation, affect the rate of equilibration and consequently determine the size of the crystals that form, and it protects trials over their duration and helps maintain the stability of the resulting crystals — the same reagent being used for vapour diffusion as well as microbatch ^[3]. Batch is the geometry to move to once screening has located a region, because it removes the trajectory and leaves only an endpoint to reproduce.

Dialysis moves the small ions and leaves the peptide behind, which makes it the only common geometry that changes ionic strength, pH or counterion at constant peptide concentration. Its limitation here is not crystallographic. Dialysis membranes are specified by molecular-weight cut-off, and a chain of one to three kilodaltons will cross membranes that comfortably retain a protein, so retention has to be verified before any dialysis result is trusted.

What a sparse-matrix screen samples, and what a hit is

Sparse-matrix conditions are not laid out on a grid. They are derived empirically from known or published crystallisation conditions of various proteins, selected so as to sample as large a range of buffer, pH, additive and precipitant variables as possible while using small amounts of material ^[4]. Coverage is bought by confounding: neighbouring wells differ in several variables at once, so a hit tells you where to look and not which component did the work.

The remedy is to spend part of the plate on a screen designed to decouple. One published strategy pairs a cherry-picked sparse-matrix set intended to maximise coverage of crystallisation parameter space with no redundancy against a systematic pH-, anion- and cation-testing screen whose stated aim is to decouple the components of each condition and to provide information about the protein even in the absence of crystals ^[5]. That second aim is the one to hold on to for a peptide: a systematic block that yields no crystals still reports which anions salt the sample out, at what pH it stops being clear, and whether the behaviour is monotonic. A sparse-matrix plate cannot report any of that.

It also gives the honest definition of a first-round hit. A plate of uniformly clear drops says the sample never reached supersaturation anywhere sampled; a plate of uniformly heavy precipitate says it overshot everywhere. Both are readings of where the boundary is not, and neither is a property of the peptide. The useful output is the set of conditions bracketing the transition, together with anything within it scored as needles, plates, spherulites or birefringent grains — conventional scoring categories, not sourced findings, but the leads that optimisation and seeding then work on.

Why the peptide is the hard part

Data mining of large-scale structural-genomics crystallisation results, combined with experimental folding studies, led to the conclusion that crystallisation propensity depends primarily on the prevalence of well-ordered surface epitopes capable of mediating interprotein interactions and is not strongly influenced by overall thermodynamic stability ^[6]. That analysis was run on protein constructs; the numbers and sequence correlates it produced are not peptide results and should not be quoted as such. The framing, though, is one a peptide fails on two counts simultaneously. A chain of a few dozen residues presents a small surface, so there are few candidate epitopes to begin with, and without a folded core there is little holding any of them in fixed geometry. A lattice is one conformer repeated; a peptide that samples many in solution has to be persuaded to commit, and the crystal contact is frequently the only thing doing the persuading.

High solubility compounds the problem arithmetically. Supersaturation is a ratio of concentration to solubility, so a freely soluble peptide must be driven to a much higher absolute concentration than a protein to reach the same supersaturation. That means more material per drop, more precipitant, or both — and a screen that stops short of the concentrations required reads as uniformly clear however many conditions it contains.

Racemic and quasi-racemic crystallisation

Chemical synthesis gives access to the mirror-image form of a polypeptide, and a racemic mixture of the two enantiomers can crystallise in ways natural proteins cannot; experimental data support a theoretical prediction that this should make racemic mixtures highly amenable to crystallisation, and crystals obtained from them also offer advantages in structure determination ^[7]. For a peptide already made by solid-phase synthesis the marginal cost is one further synthesis using D-amino acids, which is a far smaller obstacle than the equivalent would be for a recombinant protein.

It is worth knowing what the resulting lattice looks like before committing to it. L- and D-plectasin, a 40-residue molecule, were each prepared by total chemical synthesis; the racemate crystallised in an achiral centrosymmetric space group with one L and one D molecule forming the unit cell, synchrotron data were collected to atomic resolution, and the structure was solved by direct methods. Dimer-like interactions between the enantiomers were observed and may account for an extremely low solvent content of 13–15% and for the more highly ordered character of the racemic crystals, while the resulting conformation was generally similar to the previously determined NMR structure, suggesting minimal impact of the packing ^[8].

The approach reaches down to genuinely short chains. A racemic mixture of the 12-residue designed peptide Alpha-1 (acetyl-ELLKKLLEELKG) gave a 0.75 Å structure with two L-helices and two D-helices in the unit cell, and equivalent solutions of the centrosymmetric bilayers were found by two independent direct phasing programs. Even at that resolution the model (R = 13.1%, R_free = 14.5%) had to incorporate much of the discrete disorder of peptide and solvent; side-chain disorder propagated into the backbone of one helix but not the other, and no side chain was rigid, although the crystal environment restricted some to a single active torsion or none ^[9]. A peptide lattice restrains conformation rather than fixing it. That structure also located ethanolamine and 2-methyl-2,4-pentanediol from the crystallisation liquor bound near the helix termini — a reminder that screen components are candidate lattice constituents, which is the theme of the next section.

Where a true racemate is unavailable or unwanted, the quasi-racemate — the D-form of a close analogue rather than of the exact molecule — serves the same end. Bracelet cyclotides, cyclic cystine-knotted peptides, had resisted crystallographic characterisation because of very low in vitro folding yields; a single substitution of Ile-11 to Leu or Gly raised those yields substantially, and mirror-image enantiomers synthesised on that basis were used in quasi-racemic crystallography to give the first crystal structures of the subfamily ^[10]. Note where the bottleneck sat: upstream of the drop, in obtaining enough correctly folded material to screen with.

Counterion, residual TFA and purity

Freshly synthesised peptides can arrive contaminated with trifluoroacetic acid ^[11], and TFA is not a spectator in the crystallising system. A large cocrystallisation screen on TFA-contaminated L-Leu-L-Leu dipeptide found a competing range of multiple crystal forms accessible, running from the target cocrystal through to the TFA salt, and isolated transient ternary and quaternary phases in which the peptide appeared together with a coformer plus TFA, or with solvent alcohol or water plus TFA ^[11]. TFA also alters solution phase behaviour. Although not usually regarded as a Hofmeister reagent, it has been shown to modulate biocoacervation, and in a coacervating bioinspired peptide labelled with ¹⁹F at a single phenylalanine it altered liquid–liquid phase separation behaviour; solid-state NMR resolved two dynamically distinct binding modes between TFA and that residue, and quantum-chemical modelling of the chemical-shift differences indicated a structured complex in which one TFA molecule intercalates between two stacked Phe rings ^[12].

The operational consequence is that two lots of the same sequence carrying different residual TFA are not the same input to a screen, and a plate set up before counterion exchange is not comparable with one set up after. Fix the salt form first, record it, and treat a change of counterion as a change of experiment.

Chemical purity acts on the growing face rather than on the phase diagram. Atomic force microscopy combined with X-ray diffraction on hen egg-white lysozyme grown with covalently bound lysozyme dimer as a deliberate impurity showed the (101) face becoming clearly rough at 5% impurity and its growth steps disappearing as the impurity reached 10%, while maximum resolution limit, I/σ(I), R_merge and overall B factor all tended to degrade as impurity concentration rose ^[13]. Those are protein measurements with a protein-derived impurity and the numbers do not transfer. The mechanism does: an impurity structurally close to the target adsorbs at growth steps and is incorporated, and for a synthetic peptide the closest structural relatives — deletion sequences, single-residue truncations, oxidised or epimerised forms — are precisely what the synthesis generates.

Seeding

Microseed matrix screening was established with the aim of influencing the nucleation event in standard crystallisation screens ^[14]. A seed stock produced by a simple seed-bead method is dispensed together with the sample and the reservoir solution — in the published implementation, simultaneously through a three-bore tip into drops of 0.6 µl total volume — across an entire screen rather than back into the condition the seeds came from. The number and type of hits increased for the proteins tested, and the method was expected to be generally applicable where little or no nucleation is normally observed ^[14]. For a peptide that gives one poor crystal in ninety-six conditions this is often a productive next move: it converts a single nucleation event into a source that the rest of the chemical space can be tested against.

Telling a peptide crystal from a salt crystal

Screens are built on concentrated salts, so salt crystals are a routine outcome. Intrinsic ultraviolet fluorescence has been investigated as a rapid non-invasive discriminator: in both an epi-fluorescence geometry and a straight-through geometry, protein crystals fluoresced brightly and gave high contrast against the background fluorescence of the surrounding solution, allowing them to be identified and distinguished from non-protein crystals ^[15].

Two cautions apply to peptides. First, the signal is aromatic. In two-photon excited fluorescence imaging of protein crystals, the blue-green emission was attributed mainly to oxidation of tryptophan residues, and non-specific fluorescent dyes were shown to bind crystals and enhance their detection where the intrinsic signal was inadequate ^[16]. A peptide with no tryptophan, perhaps no tyrosine, will not light up, and darkness is then not evidence that the object is salt. Second, second-harmonic imaging is not the fallback it looks like. The method reports second-harmonic generation by noncentrosymmetric crystals; an inverse correlation was documented between the level of symmetry in the crystals examined and their SHG intensity, and SHG-silent protein crystals were instead identified by two-photon fluorescence ^[16]. A racemate crystallising in a centrosymmetric group, the outcome described above, therefore falls outside what the technique selects for. That is an inference from the selection rule, not a measurement anyone has reported on such crystals.

The decisive test is diffraction, and the discriminator is geometric rather than experimental: a large unit cell produces closely spaced reflections, whereas a small-molecule salt lattice gives a few widely separated, very strong spots persisting to high angle. That follows from the reciprocal relation between cell dimensions and reflection spacing and is not a result from any one paper.

Cryoprotection

Data are routinely collected near 100 K because radiation damage is mitigated there. That requires cryoprotecting the crystal and cooling it rapidly enough to take a significant solvent fraction below the glass-transition phase of water, and the reagents and procedures for doing so — including those aimed at preventing the stresses that damage crystals when flash-cooling — are the substance of the technique ^[17]. None of that is peptide-specific in principle; two things are in practice. The solvent fraction of a peptide lattice is not predictable in advance, and the racemate described earlier sits well below what such procedures presuppose, so the cryoprotectant concentration that works for a typical protein crystal is a guess here rather than a starting point. And the cryoprotectant is itself a change in composition applied to a lattice held together by a small number of contacts, so it should be screened on expendable crystals with the same discipline the condition itself was screened with, rather than applied once to the only crystal you have.

When there is nothing to see

A peptide screen scored by eye can return a false negative. Crystals of NACore, an 11-residue segment of α-synuclein, have dimensions smaller than the wavelength of visible light and are thus invisible by optical microscopy, and are thousands of times too small for structure determination by synchrotron X-ray diffraction; the structure was determined instead by micro-electron diffraction, at 1.4 Å resolution ^[18]. Short self-assembling sequences are exactly the class that behaves this way, so a drop that reads as fine amorphous precipitate is not proof that nothing ordered formed, and a diffraction-based check is the only way to exclude it.

What a screening campaign licenses you to say is narrow: that under these conditions, this preparation of this sequence with this counterion produced ordered material. It does not establish that the sequence crystallises in general, that another lot will behave the same way, or that the packing obtained reflects the conformation the peptide adopts free in solution — the last of which has to be argued separately, as the comparison with an independent structure above illustrates. Getting the crystal and interpreting the diffraction are different problems, and the second begins only once the first is solved.