In bioproduction the medium is alive, sterility keeps sampling to a minimum, and every batch counts. So the measurement gets in without opening anything, and returns its result while the decision is still open.
A fourteen-day culture steered on one sample a day is steered on interpolation. A continuous measurement replaces that interpolation with the real trajectory, glucose as it runs down and lactate as it builds, while there is still time to feed differently.
Before inoculation, the composition of a medium is known and declarable: calibration on pure components applies, with no design of experiments. In the living culture, glucose and lactate call for calibration on samples.
A probe that served for one campaign? A Raman added to a bioreactor, whose model did not follow the change of medium or of scale. Before buying anything new, the diagnosis: often the hardware is sound, and it is everything else that is missing. What to do with an instrument that has fallen out of use.
Where to start, depending on your role
- You run production: start from the culture steered on one sample a day. Scoping.
- You develop the process: pure components or calibration on samples, what a proof of concept has to settle. Feasibility.
- You own quality: what software declared 21 CFR Part 11 covers, and what stays with you. 21 CFR Part 11.
- You decide the investment: the cost of a control, calculated on your batches. Cost of a control.
- You run a multi-product CDMO or CMO: what carries over from one product to the next, and what gets recalibrated. Calibration burden.
- You want examples: eight real situations, presented without names. Experience.
Why bioproduction is not a dry form
This is the substantive difference from dry forms. Elsewhere, what is measured is a stable material that can be sampled again. Here the material changes while it is being measured, and the window for action is counted in hours. In bioproduction, measurement is first of all a batch-security tool.
The two calibration routes part company at inoculation. On a declared medium, a buffer or a formulation solution, decomposition onto pure components applies. In a living culture, metabolites appear, biomass scatters the light and the product accumulates: following glucose or lactate there calls for calibration on samples, with the reference campaign that goes with it. Knowing which side of inoculation your question sits on settles the scope in one meeting.
The dossier stays yours: how far we go with your CMC teams
What we carry. The method, its validation and the pieces that support it, through to the analytical part of a comparability dossier where a process change is involved: protocol, sampling plan, statistical analysis, report. Our experts can carry the regulatory writing in support of your CMC teams, who keep the filing and the relationship with the authority.
A process change calls for a comparability exercise. In biologics the question is not only whether the method is valid, but whether the product after the change resembles the product before it closely enough. ICH Q5E puts it plainly, at § 1.4: comparability “does not necessarily mean that the quality attributes of the pre-change and post-change product are identical, but that they are highly similar”. And the text asks that in-process controls be confirmed, modified or created after a change.
That is where an in-line measurement pays. The same § 1.4 provides that “if a manufacturer can provide assurance of comparability through analytical studies alone, nonclinical or clinical studies with the post-change product are not warranted”. An analytical package that holds is therefore what avoids a bridging study. An in-line measurement contributes twice: it documents the process batch by batch on both sides of the change, and it follows attributes continuously where the laboratory gives only points.
What the authority looks at. The attributes are those of a biological product: titre, aggregate profile, charge variants, glycosylation, activity. The dossier has to show that the change did not move them outside what was known, and that the trial plan was able to see it. ICH Q5E asks that the product be evaluated at the process step most likely to reveal a difference — which is exactly the argument for a measurement placed in line rather than at the end. What validating a multivariate method asks for.
The right moment is the change itself. A comparability exercise is prepared before the change, not after: a measurement installed once the new process is running says nothing about the old one. And once the method is named for release, going back to the laboratory test on a batch that surprises calls for a full documented investigation.
Increasing production capacity: what in-line measurement changes
Three levers serve to produce more: scaling up, intensifying the process, or entrusting manufacturing to a CDMO or a CMO. Each raises its own question. How do you keep what development has built when moving to industrial scale? How do you intensify, with intensified fed-batch, perfusion or continuous operation, and keep control of the quality attributes? How do you follow a process run on another site?
Intensify. An intensified process runs closer to its limits. In-line measurement, RAMAN or NIR, follows substrates, metabolites and biomass continuously, and lets feeding be controlled on the real state of the bioreactor. It is the foundation of a Quality by Design approach, and of a move to continuous manufacturing, whose frame is set by ICH Q13.
Scale up. The proof of concept runs at pilot scale or on the process itself, under the conditions of final deployment: the resulting model is built on production material and equipment, and is used there.
Outsource. Objective process data, shared between the sponsor and the CDMO or CMO, makes technology transfer and batch-by-batch monitoring easier. For a CDMO, it is also a capability to offer its own clients.
For a multi-product CDMO, the question becomes: what carries over from one product to the next? A medium of declared composition is calibrated on pure-component spectra, with no sample campaign for each new client. Culture metabolites call for calibration on samples. The proof of concept establishes what transfers from one molecule to another, and what has to be redone.
The three questions that come up most often
Where are glucose and lactate
Substrates and metabolites, glucose, lactate, glutamine, ammonium, are the variables that govern feeding and the end of culture. They are concentrations in a living medium, so calibration on samples, not decomposition onto pure components. NIR-HPTLS, MIR and Raman share the work depending on the range, the matrix and the variable.
When to stop, and when to cut a fraction
End of culture, the switch from a growth phase to a production phase, the end of elution in chromatography: these are endpoints. Often the question is at what moment the signal changes regime rather than which value it reaches, and that form is far lighter to instrument.
What size are the particles, in flow and undiluted
For an injectable emulsion or lipid nanoparticles, the sample keeps its state when it is measured undiluted. Methods for sizing in flow need no multivariate model at all: the quantity comes out of the physics. They ask for three conditions, all buildable: laminar flow, a bubble-free line, a mounting free of vibration.
Where the drug substance ends and the drug product begins
In biologics the drug substance is the purified bulk: culture, harvest, chromatography, tangential flow filtration. The drug product starts at formulation and runs to fill-finish — often on another site. The attributes change on either side: titre, aggregation and glycosylation upstream; concentration, osmolality and container integrity downstream.
For measurement the consequence is clear: upstream it follows a living process whose trajectory can still be corrected, downstream it serves a unit check at high throughput. The equivalent reasoning for a chemically synthesised active substance, ICH Q11 and ICH Q7, is set out on the drug substance page.
Where the measurement sits, operation by operation
| Unit operation | What is measured | What it changes |
|---|---|---|
| Media preparation | Identity of the components and titre after reconstitution | A weighing slip or a swapped powder shows up before inoculation, while the medium can still be remade. |
| Cell culture, fermentation | Glucose, lactate, glutamine, ammonium, biomass | The trajectory replaces interpolation between two samples. Batch-to-batch differences show early. |
| Substrate feeding | Residual concentration of the limiting substrate | Feeding follows the real state of the bioreactor instead of a preset profile. Requires continuous measurement. |
| Harvest | The tipping point: substrate exhaustion, product accumulated. Viability is inferred rather than measured directly | The harvest date becomes a measured criterion rather than a protocol duration. |
| Purification, chromatography | Product and buffer concentration at the column outlet, start and end of collection. Inside the column, electrical tomography gives access to the elution front, to channelling and to packing quality | Fractions are cut on composition rather than on volume, so more product is recovered at the edges of the peak. And a bed irregularity is located by height rather than inferred from an outlet profile. |
| Final formulation | Excipient and product concentration, plus particle size for dispersed forms, and the viscosity of concentrated formulations | The final mix is checked before filling, on the volume rather than on one sample. |
| Filling | Identity of the filled unit through the vial, by NIR and RAMAN, without opening it | A doubtful unit is rejected in line, and the container stays closed. |
| Freeze-drying | Residual moisture of the cake, read by NIR through the bottom of the vial | The batch is judged vial by vial, not only on the vials that are opened. |
The real constraint is the sterile barrier. Any measurement has to happen without breaking the closed system. Three routes exist, and they do not carry the same lead times. A steam-sterilisable probe, fitted to a bioreactor port, is cleaned and sterilised with the vessel. That is the classic stainless steel route. An optical window on an aseptic port avoids direct contact. And a sensor built into a single-use system travels with the consumable.
The third is the most convenient, and the one decided earliest: the sensor is irradiated, qualified and documented with the bag, so the choice is made when you choose your consumable, often months before the measurement project. Raising it at scoping is what keeps both options open.
For a sterile product, this choice becomes part of the contamination control strategy required by EU GMP (Annex 1, point 2.3). A measurement that does not break the closed system removes a sampling step, and with it an intervention that the strategy would otherwise have to justify. Annex 1 also asks manufacturers to consider rapid or alternative methods and continuous monitoring systems to protect the product from contamination (point 2.1).
The regulatory frame is the one used in pharma, where it is most developed.
How long do you have to decide
Every operation leaves a window in which to decide before moving to the next one. Set it beside the time the method meant to answer actually takes: the ratio of the two says whether the question can be settled in the laboratory, or not. The table runs from the hardest case to the easiest. The durations are indicative orders of magnitude: replace them with your own.
| Operation | Decision window | Attribute followed | Reference method, and its duration | Ratio |
|---|---|---|---|---|
| Chromatography | 0.5 h | Aggregation | HPLC, 1 h | 0.5 |
| Centrifugation | 1 h | Recovery | HPLC, 1 h | 1 |
| Microbial fermentation | 2 h | Misincorporation | HPLC, 1 h | 2 |
| Refolding | 2 h | Misfolds | HPLC, 0.5 h | 4 |
| Mammalian cell culture | 10 h | Glycosylation profile | Oligosaccharide profile, 1 h | 10 |
| Lyophilisation | 1 h | Residual moisture | NIR, 0.1 h | 10 |
The ratio measures how easy the implementation is. At the top of the table the laboratory route does not return its answer inside the window: on a chromatography step the decision is taken in half an hour and the assay asks for one. That is where in-line measurement is at once the most needed and the most demanding — succeeding there means reworking the process, the equipment and the method together, and that is the authors’ own wording. At the bottom the window is wide and the laboratory keeps up: installation is straightforward there. Lyophilisation shows it, and the reason has to be read the right way round: its ratio is ten because NIR is already the method.
These figures are optimistic, and it helps to know why
The duration above is the instrument’s analysis time. It counts neither the sampling, nor the trip to the laboratory, nor the queue, nor the preparation, nor the review of the result. On a real site, an “HPLC, one hour” is half a day.
The quantity that actually decides is the one we put a figure on at scoping: the time between the state of the medium and the command that follows from it. Every ratio above is therefore an upper bound, and more operations fall below the threshold than this table shows.
What NIR and RAMAN measure in a culture, and what they do not
On the cell culture line the attribute is a glycosylation profile, and it comes out of a released-glycan assay. Neither NIR nor RAMAN returns it.
What they do return in line are the variables that govern it: glucose, lactate, glutamine, ammonium, biomass, titre. The distinction changes the claim, not the interest. Steering the conditions that determine glycosylation is an ordinary project; measuring the profile in line is not.
Durations and ratios after A. S. Rathore, R. Bhambure and V. Ghare, Process analytical technology (PAT) for biopharmaceutical products, Analytical and Bioanalytical Chemistry, Springer, 2010, DOI 10.1007/s00216-010-3781-x, table 3 and figure 5. Values given by the authors for illustration. The choice of rows, their order and the commentary are ours.
The two questions particular to this sector
What calibration on pure components sees, and what completes it
This is the exact counterpart of its advantage. The model decomposes the spectrum onto the components you declared, so the declaration is what sets the field of view. An additive left off the list, an impurity in a raw material or a degradation product appearing late in the culture is read across the other components, and where the library of pure spectra misses an interaction a stable bias can settle in. Two habits cover both, and they are cheap: watch the spectral residual produced at each acquisition, and keep one comparison point in the laboratory.
Matching the loop time to the bioreactor
A result returned in ten minutes is excellent for understanding a process, and substrate feeding is decided by the minute. So the figure to settle before buying is the time between the state of the medium and the command that follows from it: sampling, transit, acquisition, calculation, writing the setpoint. Add them up and you know which of the two jobs the installation will do, and you can choose it deliberately. A deployment point, and it is settled on a whiteboard in ten minutes.
What this is based on
Three sources, and we keep them apart.
Our instrument manufacturers. Composition monitoring in liquid media is documented and quantified, in ultrafiltration and diafiltration among others. Ranges by analyte, coefficients of determination and repeatability are published. Sizing without dilution is in production in the pharmaceutical industry.
The literature. Monitoring of substrates and metabolites in cell culture is widely covered. It is what sets what is reachable, and at what calibration burden.
Our own expertise. Twenty-five years of non-destructive measurement and chemometrics on complex media. It bears on the arbitration. Which variable belongs to which calibration route, which sterile access route suits your equipment, and whether your decision loop is fast enough.
What it changes, in practice
The gain rarely shows up in the cost of the analysis. It shows up on three lines: batches saved, process development time, and the time to transfer from one scale to another. A process whose real trajectory you know transfers better than one where you know fourteen points per batch.
On a robust culture, where every batch overlays the last and feeding runs itself, the gain shows elsewhere: measurement brings the documentary evidence a Quality by Design file rests on. That is a good reason of its own, and it is a different one from return on investment.
Tell us what is in your medium. The answer starts there.
Forty-five minutes is enough to settle decomposition onto pure components against calibration on samples, the sterile access route your equipment allows, and the real speed of your decision loop.