The second law of thermodynamics dictates that all things tend towards disorder. Yet molecules and other microscopic particles in liquids frequently arrange themselves into perfectly ordered arrays — crystals — without violating this law. Moreover, a given molecule can often arrange itself into more than one type of array, producing different crystal forms known as polymorphs. These polymorphs can have remarkably different properties despite being composed of the same building blocks. In a paper in Nature, Van Driessche et al.1 report experimental observations of protein molecules as they begin to assemble into clusters that then evolve into distinct polymorphs. Their findings bring fresh insight to the important processes of crystal formation and growth, and polymorph selection.
The everyday consequences of crystal polymorphism are perhaps highlighted best by pharmaceutical drugs, most of which are administered as crystalline solids2. Polymorphs of drug molecules often exhibit considerable variation in their ease of manufacture, their shelf life and — crucially — their physical and chemical properties, which greatly influence their physiological efficacies2. The selection and manufacture of appropriate polymorphs is a major and costly component of the drug-development process, yet processes for polymorph selection are largely conducted on the basis of trial and error, rather than through molecular design.
The development of rational approaches for the design and control of crystal growth, as well as for polymorph selection, requires an understanding of nucleation — the initial stages of crystallization, in which the building blocks begin to form clusters known as nuclei. Unfortunately, there are two main hurdles to capturing and characterizing crystals at birth. First, the nuclei are typically too small to be visualized in 3D space using most experimental methods, especially when they consist of atoms or small molecules. Second, such nuclei are, by definition, unstable and therefore form only transiently.
To address the first issue, Van Driessche and colleagues used the protein glucose isomerase (GI) as a building block, the box-like shape and nanometre dimensions of which make it relatively easy to identify using a technique called cryo-transmission electron microscopy (cryo-TEM). And to overcome the second issue, they used a protocol in which protein samples were rapidly frozen to about −183 °C, a temperature at which essentially all motion by GI molecules stops. This enabled them to take snapshots of the crystallization process at time intervals ranging from seconds to minutes. The authors thus captured the nucleation and growth of GI crystals with sufficiently high temporal and spatial resolution for them work out how the emergent crystal morphologies depend on specific interactions formed between GI molecules.
To initiate crystallization, the authors mixed GI with ammonium sulfate or polyethylene glycol (PEG), which are commonly used as agents for modulating protein solubility and the strength of interactions between proteins. They observed that, at high concentrations of ammonium sulfate, GI molecules rapidly line up into rods that are a few proteins in length. These rods then align side by side, while also growing longer, to yield macroscopic crystals with a rectangular, prism-like shape that resembles the rod-shaped nuclei (Fig. 1a). By contrast, low concentrations of PEG cause the GI nuclei to grow more slowly and evenly in all dimensions, to yield rhombic crystals with a diamond-like shape (Fig. 1b). And at high concentrations of PEG, the GI molecules become locked into structures that are best described as disordered gels rather than crystals (Fig. 1c).
Van Driessche and colleagues’ cryo-TEM images of the GI nuclei are detailed enough to be compared with known 3D atomic structures of GI crystals. Such comparisons enabled the authors to propose plausible models for the arrangement of GI molecules in the nuclei, as well as for the specific interactions between the molecules. The authors then designed mutants of GI in which a key amino-acid residue at each interaction site in the various nuclei was replaced with another residue. The mutant proteins were unable to form their corresponding polymorph and either produced the alternative crystal polymorph or aggregated into gels, depending on the conditions. These observations validated the proposed structural models and demonstrated that polymorphs could be selected predictably.
Classical nucleation theory (CNT) posits that crystallization must start with the formation of a nucleus that has the same molecular order and arrangement as do the macroscopic crystals, and that the building blocks are added one by one to the nucleus3. In the past two to three decades, CNT has been largely superseded by two-step or multistep nucleation models in which an amorphous, high-density liquid phase forms, and then transitions into either a single crystalline domain (as occurs in some inorganic nucleation processes4) or numerous small, locally ordered clusters, which align to form growing crystals by a process known as oriented attachment. In light of these competing views, characterization of the nucleation events for various systems has been the subject of much experimental and theoretical work3.
Van Driessche and co-workers’ observations, particularly of the prismatic GI crystals, indicate that elements of both CNT and multistep nucleation might be at play. The authors uncover no evidence of an amorphous liquid phase, and the smallest GI rods, which they captured at very early time points (just 20 seconds after the addition of ammonium sulfate), have the same crystalline registry as the mature crystals — findings that are in accord with CNT. However, the rods undergo oriented attachment during crystal maturation, as in the multistep model. The picture is less clear for the rhombic GI crystals, the precursors of which become observable only after several minutes — at which point they are already bigger than a nucleus. It could be that an amorphous, high-density liquid phase does form in this case but escapes detection because of its instability or the low contrast of the cryo-TEM images.
We can, however, be certain that rapidly advancing techniques such as cryo-TEM, liquid-cell transmission electron microscopy5,6 and in situ atomic force microscopy7, complemented by theory and computational modelling8, will continue to provide intriguing results with which to refine our understanding of crystal nucleation and growth. More practically, Van Driessche and colleagues’ study shows that the crystallization or self-assembly pathways of proteins can be rationally engineered at the molecular level to obtain a desired polymorph. This feat is particularly notable, given that the number of protein-based agents being used as pharmaceutical drugs is increasing9, and that synthetic protein assemblies and crystals are being designed and constructed to have unusual and potentially useful properties10.
Nature 556, 41-42 (2018)