Crystallization of pharmaceutical APIs

Purification elsewhere in this cluster spends selectivity on resin or membrane; crystallisation gets its selectivity from the lattice itself. A growing crystal accepts molecules that fit its symmetry and rejects the rest, so a single crystallisation can raise purity by orders of magnitude where a chromatography step would struggle. The price is that the driving force — supersaturation — is shared between the process you want and the processes you don’t: nucleation of the wrong polymorph, of fines, of amorphous junk. Running a crystalliser is the discipline of making only one of those things happen at a time.

Supersaturation and the two clocks

Solubility c* is a curve in temperature and composition; supersaturation S = c/c* is the distance above it, and both crystal growth and nucleation feed on it. Growth consumes supersaturation on existing surfaces in an orderly, size-increasing way. Nucleation creates new crystals from nothing and is steeply non-linear: below a threshold the liquid sits metastably supersaturated for as long as you like; above it, crystals appear everywhere at once. The band between is the metastable zone — the entire operating theatre of controlled crystallisation. Inside it, growth happens on the seeds you added; outside it, the process detonates into a burst of fine, impurity-occluding crystals whose size distribution no downstream filter wants. Nucleation is also stochastic — the same cooled solution may nucleate in one vessel and not in another — which is why seeding is not an optimisation but the difference between a process and a lottery. Ostwald’s rule of stages adds the historical trap: the polymorph that nucleates first is often the least stable one, which then transforms slowly through solution — a plant can fill a campaign with the wrong form before anyone sees a problem.

Choosing the mode by the curve

The shape of the solubility curve picks the operation. A steep curve rewards simple cooling: large solubility change per degree, no composition change. A flat curve makes cooling useless and hands the job to evaporation. When thermal sensitivity or a flat, awkward curve rules out both, an antisolvent — a miscible liquid in which the compound is poorly soluble — pulls supersaturation by composition; its cost is local: at the feed point, mixing is imperfect and supersaturation spikes, so antisolvent crystallisation lives or dies on adding slowly against strong agitation, and “oiling out” — a second liquid phase condensing instead of crystals — is the standard failure. A pH-shift (reactive crystallisation) is the same trick through ionisation. Whatever the mode, the control objective is identical: keep the trajectory inside the metastable zone, on purpose-grown surfaces.

Why this is the purity step — and the identity step

The lattice discriminates: molecules with the wrong size, shape or hydrogen-bonding pattern either do not enter the crystal or enter at parts-per-thousand rates, so impurity rejection per pass is enormous. What defeats rejection is not the lattice but the liquor: mother liquor trapped in agglomerates and inclusions carries dissolved impurities inside the crystal cake, which is why washing efficiency at the filter and a clean crystal-size distribution matter as much as the chemistry. And polymorphism turns the crystalliser into the step that fixes the product’s legal identity: same molecule, different lattice — different solubility, dissolution rate, stability, and a separate specification in the dossier. The crystalliser is therefore the place where the API becomes simultaneously pure, processable and officially itself — and the place where scale-up bites hardest, because mixing determines local supersaturation, and a crystallisation validated in a one-litre vessel has no automatic right to survive a ten-cubic-metre one.