Appearance
25.11 — Making It at Scale
In 1998 the HIV drug ritonavir was on the market and working. Then, in one manufacturing lot, the drug crystallised into a different arrangement — the same molecule, packed differently in the solid. The new form was less soluble. Capsules made from it did not dissolve properly and did not reach the blood at the required level.
Worse, the new crystal form then appeared in other batches and in other facilities, and the original form became difficult to produce at all. The company pulled the capsules from the market and had to reformulate the product while patients depending on it waited.
Nothing about the molecule had changed. Nothing about the recipe had changed. The way the solid packed itself had changed, and that was enough.
This chapter is about the gap between "we have a molecule that works in a trial" and "we can make three hundred million doses of it, identical, for the next twenty years". It is the part of the industry that software engineers most often underestimate, and it is where a large share of quality and supply-chain project work actually lives.
The two worlds again
Chapter 25.3 split the industry into small molecules made by chemistry and biologics made by living cells. Manufacturing is where that split becomes physical: different buildings, different equipment, different staff, different failure modes.
Both share one governing idea, and it is worth stating before the detail. You cannot test quality into a product. Testing a sample tells you about that sample. Quality has to come from a process that is understood well enough to produce the same thing every time, and that is what all the machinery below is for.
Making a small-molecule drug substance
The chemists in Chapter 25.6 made grams. The factory needs tonnes, and the route that worked at gram scale usually cannot be used.
Why a laboratory route fails at scale is worth understanding, because it explains why process chemistry is a separate profession.
Heat does not behave the same. A reaction that releases heat is easy to control in a flask sitting in a water bath, because the surface area is enormous relative to the volume. Scale it to a two-thousand-litre reactor and the same reaction can run away, because the heat is generated in the volume and removed only through the walls.
Mixing does not behave the same. In a flask everything is uniform in seconds. In a large vessel, one region can be at a different concentration for minutes, producing by-products that never appeared in the laboratory.
Cost and safety change what is allowed. A reagent that is merely expensive at gram scale is prohibitive at tonne scale, and a solvent that is acceptable in a fume hood may be unacceptable in an industrial plant.
So process chemists redesign the route: fewer steps, cheaper starting materials, safer reagents, and crucially fewer purification steps, since each one loses material. A route with eight steps at 90 percent yield each delivers about 43 percent overall — which is why removing a single step can change a product's economics.
Then comes the step the ritonavir story was about: crystallisation. The drug substance is normally isolated as a solid, and how it is crystallised determines the crystal form, the particle size and the purity. The same molecule can crystallise in different arrangements, called polymorphs, and these differ in how fast they dissolve — which changes how much drug reaches the blood. Polymorph screening and control is therefore a formal, regulated part of development, and the crystallisation step is often the most tightly specified operation in the whole route.
Turning drug substance into a product people can take
The active ingredient is a powder. A patient takes a tablet, and everything else in that tablet is there for a reason.
Excipients are the inactive ingredients: fillers to make a tiny dose into a tablet you can pick up, binders to hold it together, disintegrants that swell on contact with water and break the tablet apart, lubricants that stop powder sticking to the machine, and coatings that mask taste, protect from moisture, or delay release until the tablet has passed the stomach.
The manufacturing sequence for a typical tablet is short to describe and full of ways to fail: weigh and blend the powders; granulate them, which means turning fine powder into small free-flowing granules so it flows evenly into the press; dry the granules; blend in the lubricant; compress at high force into tablets; coat them; and package.
The two most common failure modes are worth knowing because they turn up in deviation reports. Content uniformity — if the blend separates, some tablets carry more drug than others, which for a potent drug is dangerous. Dissolution — if the tablet does not release the drug at the expected rate, the product fails specification even though the correct amount of drug is present.
Injectable products are a different discipline again, because they must be sterile. Sterility is not a test you pass at the end; it is designed into the process. The product is either sterilised in its final container by heat, which is the most reliable method and impossible for anything heat-sensitive, or made from sterile-filtered components under strictly controlled conditions — aseptic processing, where every intervention by a human is a risk and the modern answer is to remove the human with isolators and robotics.
And some products are freeze-dried — lyophilised. The solution is frozen and the ice removed under vacuum, leaving a dry cake that is reconstituted with water before use. It is used when a molecule is not stable in liquid, and it is slow, expensive and capacity-limited, which is why lyophilisation slots are a real constraint in supply planning.
Making a biologic: the upstream half
Here the factory is a population of living cells, and the process starts from the frozen cell bank created in Chapter 25.6.
One vial is thawed. The cells are grown in progressively larger vessels — a seed train — from millilitres to litres to hundreds of litres, until they fill a production bioreactor which may hold anywhere from a few hundred to twenty thousand litres.
Inside the vessel, conditions are held within narrow bands: temperature near body temperature, dissolved oxygen, pH, stirring rate, and a nutrient feed added on a schedule. The cells grow, and as they grow they secrete the antibody or protein into the liquid around them.
Two operating styles dominate. Fed-batch runs for one to three weeks, feeding nutrients as they are consumed, then harvests everything at once. Perfusion continuously removes product-containing liquid and adds fresh medium, keeping the cells producing for much longer.
And the equipment itself has changed the economics of the industry. Single-use systems — sterile plastic bags and tubing sets replacing fixed stainless steel — removed the cleaning and cleaning-validation burden between batches and let a facility switch products in days rather than weeks. This is why contract biologics manufacturing became a viable business at moderate scale, and it is a good example of a process innovation that changed an industry structure without changing any science.
The downstream half: getting the product out clean
What comes out of the bioreactor is a soup: your protein, plus cells, cell debris, host cell proteins, DNA, and spent medium. Downstream processing removes everything that is not the product.
The sequence is standard for antibodies. Cells and debris are removed by centrifugation and filtration. Then affinity chromatography — the liquid passes through a column packed with a material that binds antibodies specifically, everything else washes through, and the antibody is released by changing the conditions. This one step does most of the purification and is also historically the most expensive raw material in the process.
Then, because the cells used to make biologics can in principle harbour viruses, there are dedicated viral clearance steps: holding at low pH to inactivate enveloped viruses, and filtration through membranes with pores small enough to retain virus particles. These steps are validated by deliberately spiking known viruses into a scaled-down model and measuring how much is removed, and the total clearance must exceed the worst-case contamination by a very large margin.
Further chromatography steps polish out remaining impurities, the buffer is exchanged for the final formulation, the product is concentrated, and it is filtered sterile.
Now the sentence from Chapter 25.3 becomes concrete: the process is the product. Because a protein's exact form depends on how the cells were grown and how it was purified, a change to the process can change the molecule's sugar attachments, its charge variants, or its tendency to clump — any of which can change how it behaves in a patient. So a manufacturing change requires a comparability exercise: analytical evidence, and sometimes clinical evidence, that the product after the change is equivalent to the product before it. That requirement is why biologics manufacturers are so reluctant to change anything, and why "we cannot just move that step" is a real answer rather than obstruction.
Fill-finish: the last and most dangerous step
Fill-finish is putting the sterile product into its final container — vial, prefilled syringe, cartridge or autoinjector — and sealing it.
It is the step with the highest contamination risk in the entire chain, because the product is exposed at the moment of filling. Modern lines run inside sealed isolators with filtered air moving in one direction, with operators outside the barrier working through glove ports or not present at all.
The European Union's revised Annex 1, the international reference for sterile manufacturing, made a formal contamination control strategy an explicit expectation: a documented, plant-wide analysis of every contamination route and what controls it, rather than a collection of individual procedures. A great many recalls and shortages of injectable medicines originate here — a particle in a vial, a container closure that did not seal, a sterility test failure — and fill-finish capacity is one of the industry's genuine bottlenecks.
Technology transfer: moving a process without breaking it
Technology transfer is moving a manufacturing process from one place to another — development to commercial, one plant to another, or a company to a contract manufacturer. It is one of the most common project types in the industry and it fails routinely.
What is transferred is far more than a recipe: the process description, every parameter and its allowed range, the analytical methods and their validation, the specifications, the equipment requirements, the training, and the reasoning behind every choice.
And the reason it fails is almost always equipment differences that look trivial. A different impeller shape mixes differently. A dryer with different geometry produces a different particle size. A filter of the same nominal rating from a different supplier retains slightly different material. The receiving site must show, with data, that its version of the process produces the same product — and if it does not, the fix may take a year.
Proving the process works: validation in three stages
Regulators think of process validation as a lifecycle, in three stages, and this vocabulary is used verbatim in client documents.
Stage 1, process design. Development work establishes which parameters matter and what ranges they must be held within. The modern framing, from the ICH quality guidelines, is quality by design: identify the product attributes that matter to the patient, work out which process parameters affect them, and control those. The set of parameter ranges shown to give acceptable product is called the design space, and moving within it is not a change requiring approval — which is a considerable operational freedom, earned by upfront work.
Stage 2, process qualification. The commercial-scale process is run — traditionally three consecutive successful batches — with intensive sampling and testing, to demonstrate that it reliably produces conforming product.
Stage 3, continued process verification. Production data is monitored for the life of the product, using statistical trending to detect drift before it becomes a failure.
And process analytical technology is the toolkit that makes this smarter: sensors that measure the process while it runs — particle size, moisture, blend uniformity, product concentration — rather than testing a sample afterwards. This is the direction the industry has been moving for twenty years, and continuous manufacturing, where material flows through the process rather than moving in discrete batches, is its logical end point.
What all this means for the systems
The manufacturing side of a pharmaceutical company runs on software you may well be asked to build, integrate or validate.
The manufacturing execution system, or MES, delivers the electronic batch record: it presents each step to the operator in order, refuses to let them skip a step or exceed a parameter, and records who did what and when. The laboratory information management system, LIMS, handles samples, tests and results. Historians record the continuous stream of sensor data from equipment. Enterprise planning systems handle materials, lots and orders. And the automation layer controls the equipment itself.
Every one of these is a GxP system in the strictest sense, because its records are the evidence that a batch was made correctly, and a patient may be harmed if that evidence is wrong. Chapter 25.12 is what those records must contain, and Chapters 25.20 and 25.21 are the rules for the systems that hold them.
Next: Chapter 25.12, the factory floor — GMP, the batch record, the quality control laboratory, and what has to happen before anyone is allowed to sell what was made.