Process Analytical Technology Explained | Cell Culture Technology

Back to Technology, Institution & Company Guides

TECHNOLOGY EXPLAINER

Process Analytical Technology (PAT)
— reading what is happening inside a culture without drawing a sample

Cell culture is a process where testing at the end is often too late. PAT is a framework for measuring quality-related conditions on the spot, during manufacture, and controlling the process with them. What gets measured is the broth on the far side of an optical window, so the materials of the window and sensor directly determine how good the measurement is.

Built from primary sources: the US FDA PAT guidance (2004), ICH Q8(R2), European Commission GMP guidelines and peer-reviewed papers / Last updated September 2026

Conceptual image of the faintly glowing tip of a plain cylindrical optical probe inserted into a clear liquid — **AI-generated concept.** An impression of reading the inside of a liquid with light through a window placed in it. **It does not represent any real product, probe shape or colour of light.**

What this article covers

What PAT is (the short version)
Where the 2004 FDA guidance stands
Four places to measure — in-line, on-line, at-line and off-line
What is measured during culture, and with what
Raman spectroscopy — light that water barely gets in the way of
Chemometrics — turning spectra into concentrations
Our calculation: how large are the prediction errors, and how much broth do samples take?
A materials engineer's view (1): the window decides the instrument's performance
Single-use sensors — the materials problem of making them disposable
A materials engineer's view (2): a sensor patch is both a product-contact part and an instrument
The link to real time release testing
Open problems and what is still undecided
Glossary / References / Claim-to-source audit

How claims are labelled in this article

Sourced = stated in guidelines, peer-reviewed papers or other published material (link given)
Our calculation = a figure this article derived, with the assumptions spelled out
Not yet confirmed = not yet established, or described as a task for the future
Explanations of measurement principles and readings from a materials angle are marked separately as Commentary. This article explains measurement in manufacturing processes; it says nothing about the efficacy of any individual product or about treatment.

1. What PAT is (the short version)

Definition: ICH Q8(R2) defines PAT as “a system for designing, analyzing, and controlling manufacturing through timely measurements (i.e., during processing) of critical quality and performance attributes of raw and in-process materials and processes with the goal of ensuring final product quality”Sourced
Starting point: the guidance FDA issued in September 2004, “PAT — A Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance”Sourced
The idea: the guidance says “quality cannot be tested into products; it should be built-in or should be by design”Sourced. It is a shift from inspecting at the end to checking as you make (commentary)

The single most important line in this article

On installing analysers, FDA's PAT guidance says: “Design and construction of the process equipment, the analyzer, and their interfaces are critical to ensure that collected data are relevant and representative of process and product attributes.”Sourced

Whether PAT succeeds depends not only on what is inside the analyser, but on the boundary that connects the process to the analyser. That boundary is, in most cases, an optical window, a sensor membrane, a port or a seal — in other words, a material (commentary).

2. Where the 2004 FDA guidance stands

FDA's PAT guidance sorts the tools used for PAT into four groupsSourced.

Multivariate tools for design, data acquisition and analysis
Process analyzers
Process control tools
Continuous improvement and knowledge management tools

The guidance notes that most pharmaceutical processes are based on “time-defined end points (e.g., blend for 10 minutes)”, which in some cases do not consider the effects of physical differences in raw materials. It adds that products can fail to meet specifications even when raw materials conform to pharmacopeial specifications, because those specifications “generally address only chemical identity and purity”Sourced.

For biologics, the guidance says “talk to us”

The guidance was issued by FDA's Center for Drug Evaluation and Research (CDER) and others, and a footnote says: “For products regulated by the Center for Biologics Evaluation and Research (CBER), manufacturers should contact CBER to discuss applicability of Process Analytical Technology.”Sourced Cell and gene therapy products fall under CBER, so the thinking is shared, but how it applies is worked out case by case (commentary).

3. Four places to measure — in-line, on-line, at-line and off-line

FDA's PAT guidance defines three kinds of measurement in the processSourced. Adding conventional off-line testing in a laboratory gives a four-way split that is easier to follow (commentary).

Fig. 1 Conceptual diagram (vector drawing). The definitions of in-line, on-line and at-line follow FDA's PAT guidance [Ref. 1]. The off-line box and the band at the bottom are this article's commentary, and all equipment shapes are schematic.

Type	FDA definition	What it means in cell culture (commentary)
In-line	Measurement where the sample is not removed from the process stream and can be invasive or noninvasive	A probe or window on the bioreactor measures directly. No extra open handling
On-line	Measurement where the sample is diverted from the manufacturing process, and may be returned to the process stream	Measured in a recirculation loop or similar. The loop's materials are product-contact parts too
At-line	Measurement where the sample is removed, isolated from, and analyzed in close proximity to the process stream	Measured on an analyser beside the bioreactor. Each sample takes broth away
Off-line	— (not defined in FDA's guidance)	Measured in the QC laboratory. Results take time

The “FDA definition” column is Sourced (FDA PAT guidance [Ref. 1]). The right-hand column and the off-line row are this article's commentary.

4. What is measured during culture, and with what

Below are parameters commonly monitored in cell culture, and the methods for which measurements have been reported in the literature this article consulted.

Parameter	Why watch it (commentary)	Measurements this article confirmed
Viable cell density	How much the cells have grown; informs harvest and passaging decisions	Predicted by Raman spectroscopy (in-line)Sourced. Permittivity (capacitance) relies on the principle that only living cells with intact membranes significantly influence the electrical characteristicsSourced
Glucose, lactate	Nutrient consumption and metabolite build-up; informs feeding times	Predicted simultaneously by Raman spectroscopySourced
Glutamine, glutamate, ammonium	Amino acid consumption, and build-up of metabolites that can stress cells	Predicted by Raman spectroscopySourced
Dissolved oxygen	Whether cells have enough oxygen to respire	Optical methods use fluorescence quenching of an indicator molecule by oxygenSourced
pH	A basic condition of the culture environment; falls as lactate builds up	Fibre-optic methods use pH-sensitive fluorescent indicator dyes. Single-use glass electrodes also existSourced

Measurements are Sourced (Raman: Abu-Absi et al. 2011 [Ref. 5], Baradez et al. 2018 [Ref. 6], Yan et al. 2024 [Ref. 7]; permittivity, optical sensors and electrodes: Busse et al. 2017 [Ref. 9], Justice et al. 2011 [Ref. 10]). “Why watch it” is this article's commentary.

Fig. 2 Conceptual diagram (vector drawing). Each dot is a combination for which a measurement was confirmed in the literature consulted [Refs 5, 6, 7, 9, 10]. A blank cell does not mean measurement is impossible, and this is not an exhaustive comparison.

5. Raman spectroscopy — light that water barely gets in the way of

Raman spectroscopy uses the fact that when a substance is lit with monochromatic light, a tiny fraction of the light is scattered with its wavelength shifted by the energy of a molecular vibration (Raman scattering). Because the shift differs from molecule to molecule, several components can potentially be read from one spectrum at the same time (commentary).

Culture broth is mostly water. The review by Busse et al. says that, compared with infrared spectroscopy, Raman has less interference from water molecules in aqueous solutions and a high signal-to-noise ratio, while for near-infrared (NIR) “the high concentration of water molecules also poses a challenge”Sourced.

Report	System	What it reports
Abu-Absi et al. (2011)	Mammalian cell culture bioreactors	An in-line Raman probe simultaneously predicted glutamine, glutamate, glucose, lactate, ammonium, viable cell density and total cell density. Reported as the first demonstration of the technical feasibility of in-line use in mammalian cell culture bioreactors
Baradez et al. (2018)	A model autologous T-cell immunotherapy process (stirred tank)	Chemometric models for glucose, glutamine, lactate and ammonia built from Raman spectra. Tracked the donor-specific rise in nutrient consumption and metabolite production. Univariate models of peak intensity were correlated with cell concentration and viability
Yan et al. (2024)	Commercial-scale (1,500 L) CHO cell culture	Developed a method for in-line monitoring of glucose, lactate and viable cell density. Examined the effects of different measurement channels and of the number of batches used to build the models
Rapala et al. (2026)	Mammalian cell culture (ambr250 to a 3 L vessel)	Named selectivity between similar metabolites, multicollinearity among correlated components and insufficient range in the training data as barriers to Raman chemometric models, and reported a procedure to address them with pure-compound characterisation and spiking

All Sourced (the abstracts of each paper [Refs 5, 6, 7, 8]).

What changes when it is used in cell therapy processes

Baradez et al. note that in cell therapy processes, measurements of nutrient consumption, metabolites and cell concentration are in many cases made off-line and only at set time points, and showed that in-line Raman can feed the state of the process back immediatelySourced. In autologous products, where the starting cells differ from donor to donor, deciding by the state of this particular culture rather than by the clock matters all the more (commentary).

6. Chemometrics — turning spectra into concentrations

A Raman spectrum is not a concentration as it stands. Spectra have to be paired with concentrations measured at the same time points by a reference method, and a statistical model (a calibration model) built from the pairs. This multivariate data analysis is called chemometrics (commentary). FDA's PAT guidance, too, lists multivariate tools first among PAT toolsSourced.

Fig. 3 Conceptual diagram (vector drawing). This article's structuring of the procedure of Baradez et al. [Ref. 6] and of the pitfalls described by Rapala et al. [Ref. 8] and Yan et al. [Ref. 7]. The spectrum drawn is schematic, not a real Raman spectrum.

Yan et al. examined how the number of batches used to build the model affects performance in a commercial-scale processSourced. Rapala et al. reported that a model built on small-scale equipment (ambr250) could be transferred to a 3 L bioreactorSourced. In other words, adopting Raman is not a matter of buying an instrument; it includes building, maintaining and transferring models (commentary).

7. Our calculation: how large are the prediction errors, and how much broth do samples take?

Our calculation (1): prediction error as a share of the measured range

Yan et al. (2024) report test-set prediction errors (RMSEP) for a 1,500 L commercial processSourced. We divide them by the width of the measured rangeOur calculation.

Glucose: 0.22 g/L ÷ (3.53 − 1.66) g/L = 0.22 ÷ 1.87 = about 11.8%
Lactate: 0.08 g/L ÷ (1.19 − 0.15) g/L = 0.08 ÷ 1.04 = about 7.7%
Viable cell density: 0.31 ÷ (5.68 − 0.96) = 0.31 ÷ 4.72 = about 6.6% (units of 10⁶ cells/mL)

Assumptions and limits: dividing by the range width is a view this article chose, not an evaluation metric used in the paper. The paper states that the method met the analytical purpose of the studySourced; this calculation does not judge the errors as good or bad. The values are for one process and one product.

Fig. 4 Drawn from our calculation (vector drawing). RMSEP and measured ranges are published values from Yan et al. (2024) [Ref. 7]. The percentages (about 11.8%, 7.7% and 6.6%) were calculated by this article and are not published values. They do not apply to other processes or products.

Our calculation (2): sampling takes a bigger bite out of a smaller culture

At-line and off-line measurement means drawing broth. We compare orders of magnitude across scalesOur calculation.

Assumption: 1 mL drawn once a day for 10 days = 10 mL in total
100 mL of broth: 10 ÷ 100 = 10%
1,500 L of broth: 10 mL ÷ 1,500,000 mL = about 0.0007%

Assumptions and limits: broth volume, sample volume and frequency are all illustrative assumptions, not values for any particular product or process. The 1,500 L figure is borrowed from the Yan et al. process above for comparison. The calculation exists to show that in a small-scale culture, the samples taken for measurement can themselves be a non-negligible part of the process. Each sample also adds a step of opening and closing the vessel, which can be a burden from an aseptic-handling point of view (commentary).

8. A materials engineer's view (1): the window decides the instrument's performance

Why this matters for materials engineers: the optical window is a part shared by the broth and the analyser

FDA's PAT guidance says the design and construction of the interfaces between analyser and process are critical to how representative the data are, and that installing analysers on existing process equipment should be done after risk analysis to ensure it does not adversely affect the process or product qualitySourced.

For fitting a Raman probe to a single-use bioreactor, Busse et al. say “a special adaptor, composed of a glass window without inner hardware, could be used”Sourced. The same review points out that at present there is no standard interface between single-use bioreactors and sensorsSourced.

Seen from the materials side, a window has to meet several conditions at once (commentary).

Optically: pass the excitation and scattered light, without the window material's own signal interfering with the spectrum
As a product-contact part: as GMP requires, not react with, leach into or absorb from the broth (21 CFR 211.65; see our explainer on GMP)
Over long cultures: protein or cell deposits on the surface can change the optical path and shift predicted values
Through sterilisation and sealing: be sterilised with the bioreactor and stay sealed afterwards

So the window surface is at once a component of analytical chemistry and a product-contact part under GMP. It is an area where optical materials, surface treatment and sealing technology translate directly into measurement reliability (commentary).

9. Single-use sensors — the materials problem of making them disposable

As disposable culture bags (single-use bioreactors) have spread, sensors have had to become disposable too. The review by Busse et al. says that for sensors in disposable bioreactors, “long lifetime or resistance to steam and cleaning procedures are less crucial factors, while a requirement of sensors for disposable bioreactors is a cost that is reasonable on a per-use basis”Sourced.

Fig. 5 Conceptual diagram (vector drawing). The layout follows the description in Busse et al. (2017) [Ref. 9], as does the point that per-use cost is the requirement for disposables. Layer thicknesses and shapes are schematic and do not represent any real product. The band at the bottom is this article's summary.

Issue	What Busse et al. (2017) say
Layout	The reader is connected via a reusable fibre optic inserted into a sheath with an optical window at its end. On the other side of the window, inside the bioreactor, a disposable sensor patch is mounted, containing an immobilised indicator dye
Leaching	In-situ sensors cannot leach any extractable compounds into the medium, including after sterilisation
Weak points of pH optodes	Less expensive, but suffer from cross sensitivity to ionic strength, a limited dynamic range, and loss of sensitivity during sterilisation or cleaning
Robustness in manufacturing	Cites concerns that single-use (especially pH) optodes are not robust enough for manufacturing operation because of lot-to-lot variability, long equilibration times and unknown strength of signal drift, among other factors
Radiation sterilisation	Single-use glass pH electrodes retain high accuracy after γ-radiation. Biosensors with biological components are susceptible to steam sterilisation but can tolerate γ-sterilisation
Capacitance sensors	Connectors and ports are integrated into the bioreactor during production and sterilised together with it by γ-radiation. Designs that measure directly through the plastic wall are in development
Calibration	Free-floating wireless sensors could be placed in the bioreactor during production and delivered pre-sterilised and pre-calibrated

All Sourced (Busse et al. 2017 [Ref. 9]). The “Issue” headings are this article's own grouping.

The EU's GMP (Annex 1), for its part, requires that for single-use systems the extractable and leachable profiles and any impact on product quality be evaluated, especially where the system is made from polymer-based materials, and that an assessment be carried out for each component of the applicability of the extractable profile dataSourced. Since sensor patches and ports touch the broth, it is natural to regard them as falling within that assessment (commentary).

10. A materials engineer's view (2): a sensor patch is both a product-contact part and an instrument

Why this matters for materials engineers: “please react” and “please don't leach” in the same membrane

The indicator dye in an optical sensor patch only measures anything if it interacts with the oxygen or hydrogen ions in the broth. Yet that dye and the components of the membrane must not leach into the mediumSourced.

So designing the patch is a problem of selective permeation and immobilisation: let the molecules through, but don't let the dye out (commentary).

Immobilisation chemistry: trap the dye physically in a polymer matrix, or bond it chemically? Bonding may reduce leaching, but can slow the response or change sensitivity
Dyes and polymers that survive γ-rays: if radiation bleaches the dye or crosslinks or degrades the polymer, sensitivity shifts and breakdown products can become extractables. The “loss of sensitivity during sterilization” Busse et al. mention is a problem in exactly this areaSourced
Consistency lot to lot: if patches are calibrated at the factory and shipped, variation in membrane thickness, dye concentration and matrix composition becomes measurement error directly. “Lot-to-lot variability”, “equilibration time” and “drift” are all problems of membrane materials and manufacturingSourced
Per-use cost: as a disposable part, it is thrown away every timeSourced. A high-performance but expensive material is hard to use

Materials that meet all four at once sit at the intersection of optics, polymers, surface chemistry and radiation chemistry. It reads as an area where makers of membrane materials and dyes, rather than sensor makers, could take the lead (commentary).

11. The link to real time release testing

One destination of PAT is real time release testing (RTRT). ICH Q8(R2) defines it as “the ability to evaluate and ensure the quality of in-process and/or final product based on process data”, which typically includes a valid combination of measured material attributes and process controlsSourced. FDA's PAT guidance says it considers real time release “comparable to alternative analytical procedures for final product release”Sourced.

For cell products this idea becomes especially pressing. The EU's GMP for ATMPs (Part IV) says the followingSourced.

Release testing may not be possible, for example when the product needs to be administered immediately after completion of manufacturing, or when the amount of available product is limited to the clinical dose (2.32)
In such cases, in-process controls instead of batch release testing can be considered, if the relevance of their results to the critical quality attributes of the finished product can be demonstrated (2.34)
Real time testing in case of short shelf-life products (2.35)

All Sourced (EU GMP Part IV [Ref. 3]).

How it fits together (commentary)

If there is neither the time nor the quantity to test the finished product properly before release, the basis for quality can only move to data gathered during the process. What then underpins the reliability of those in-process data is the accuracy of the sensors, the validity of the models, and the stability of window and patch materials. For cell products PAT can be not just a tool for efficiency but a technology bound up with how quality assurance itself is constructed. How far in-process data can be used for release decisions for a cell product, however, is something agreed product by product with the regulators (see our explainer on QbD and CMC).

12. Open problems and what is still undecided

(1) Applying PAT to biologics and cell products is decided case by case

FDA's PAT guidance asks manufacturers of CBER-regulated products to discuss the applicability of PAT with CBERSourced. As of this article's research (September 2026), we found no unified guidance setting out the use of PAT as a basis for release decisions for cell productsNot yet confirmed.

(2) No standard interface between sensors and bags

Busse et al. point out that there is no standard interface between single-use bioreactors and sensors, and say manufacturers should address this shortcomingSourced. The review dates from 2017, and whether standardisation has advanced since was not checked against primary sources for this article.

(3) Robustness and transfer of chemometric models

Rapala et al. (2026) describe the adoption of Raman chemometrics in biologics manufacturing as “limited”, citing the difficulty of model developmentSourced. How to maintain models in processes where the starting material changes with every lot, as in cell products, remains an open taskNot yet confirmed.

(4) What Raman can measure

The literature this article consulted confirmed measurements of glucose, lactate, amino acids, ammonium, cell density and the like. We found no primary source stating that attributes tied directly to product efficacy, such as cell function (potency), can be measured in-line.

The article in summary

PAT is a framework for measuring and controlling during the process; FDA issued guidance on it in September 2004Sourced
In-line, on-line and at-line are distinguished by whether and how the sample is taken outSourced
Raman spectroscopy suffers little interference from water, and there are reports of it predicting several broth components in-lineSourced
In reported commercial-scale values, prediction error was about 7 to 12% of the measured rangeOur calculation
Disposable sensor patches must meet four demands at once: no leaching, sterilisation, lot-to-lot consistency and costSourced
Windows and patches are both product-contact parts and instruments — an area where materials makers have a role (commentary)

13. Glossary

PAT: Process Analytical Technology. A system for designing, analysing and controlling manufacturing through timely measurements during processing.

In-line: Measurement where the sample is not removed from the process (FDA definition).

On-line: Measurement where the sample is diverted from the process and may be returned (FDA definition).

At-line: Measurement where the sample is removed, isolated and analysed close to the process (FDA definition).

Raman spectroscopy: A method that identifies components by measuring light scattered with its wavelength shifted by molecular vibrations.

Near-infrared spectroscopy (NIR): A method that measures absorption of near-infrared light. Water absorbs strongly, which is a challenge in aqueous solutions.

Permittivity (capacitance) measurement: A method that measures the amount of viable cells, using the fact that live cells with intact membranes behave like tiny capacitors.

Chemometrics: Statistical methods that derive concentrations and the like from multivariate data such as spectra.

RMSEP: Root mean square error of prediction. A measure of how large a calibration model's prediction errors are.

Multicollinearity: A situation in which explanatory variables are strongly correlated, making it hard to tell which one has the effect.

Optode: A sensor that measures a chemical quantity through an optical change in an indicator dye. Used for pH and dissolved oxygen.

Fluorescence quenching: The weakening of an indicator's fluorescence in the presence of another molecule (here, oxygen).

RTRT: Real time release testing. The ability to evaluate and ensure quality based on process data.

Single-use bioreactor: A bioreactor that uses a disposable plastic vessel (bag).

14. References

U.S. FDA “Guidance for Industry: PAT — A Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance”, September 2004 https://www.fda.gov/media/71012/download
ICH “Q8(R2) Pharmaceutical Development”, August 2009 https://database.ich.org/sites/default/files/Q8%28R2%29%20Guideline.pdf
European Commission “Guidelines on Good Manufacturing Practice specific to Advanced Therapy Medicinal Products”, 22 November 2017 https://health.ec.europa.eu/system/files/2017-11/2017_11_22_guidelines_gmp_for_atmps_0.pdf
European Commission “EudraLex Volume 4, Annex 1: Manufacture of Sterile Medicinal Products”, 25 August 2022 https://health.ec.europa.eu/system/files/2022-08/20220825_gmp-an1_en_0.pdf
Abu-Absi NR et al. “Real time monitoring of multiple parameters in mammalian cell culture bioreactors using an in-line Raman spectroscopy probe”, Biotechnology and Bioengineering 108(5):1215–1221 (2011), PMID 21449033 https://doi.org/10.1002/bit.23023
Baradez MO et al. “Application of Raman Spectroscopy and Univariate Modelling As a Process Analytical Technology for Cell Therapy Bioprocessing”, Frontiers in Medicine 5:47 (2018), PMCID PMC5844923 https://doi.org/10.3389/fmed.2018.00047
Yan X et al. “Development of an in-line Raman analytical method for commercial-scale CHO cell culture process monitoring”, Biotechnology Journal 19(1):e2300395 (2024), PMID 38180295 https://doi.org/10.1002/biot.202300395
Rapala S et al. “Addressing the elephant in the room: A comprehensive framework to resolve selectivity, multicollinearity, and scalability challenges in Raman chemometric models for mammalian cell culture applications”, Journal of Pharmaceutical and Biomedical Analysis 280:117625 (2026), PMID 42341650 https://doi.org/10.1016/j.jpba.2026.117625
Busse C et al. “Sensors for disposable bioreactors”, Engineering in Life Sciences 17(8):940–952 (2017), PMCID PMC6999375 https://pmc.ncbi.nlm.nih.gov/articles/PMC6999375/
Justice C et al. “Process control in cell culture technology using dielectric spectroscopy”, Biotechnology Advances 29(4):391–401 (2011), PMID 21419837 https://doi.org/10.1016/j.biotechadv.2011.03.002

15. Claim-to-source audit

Claim in the text	Basis	Label
The title and date (September 2004) of the guidance. “Quality cannot be tested into products; it should be built-in or should be by design.” The four groups of PAT tools. That time-defined end points may not consider physical differences in raw materials, and that pharmacopeial specifications generally address only chemical identity and purity. The definitions of at-line, on-line and in-line. That the design and construction of process equipment, analysers and their interfaces are critical, and that installation on existing equipment should follow risk analysis. That real time release is considered comparable to alternative analytical procedures for final product release. The footnote that manufacturers of CBER-regulated products should contact CBER	FDA PAT guidance, Reference 1 https://www.fda.gov/media/71012/download	Sourced
The definitions of PAT and RTRT	ICH Q8(R2) glossary, Reference 2 https://database.ich.org/sites/default/files/Q8%28R2%29%20Guideline.pdf	Sourced
Cases where release testing is not possible (immediate administration, product limited to the clinical dose; 2.32), the option of in-process controls instead (2.34), and real time testing for short shelf-life (2.35)	EU GMP Part IV (ATMPs), Reference 3 https://health.ec.europa.eu/system/files/2017-11/2017_11_22_guidelines_gmp_for_atmps_0.pdf	Sourced
That the extractable and leachable profiles of SUS and their impact be evaluated, especially for polymer-based materials, and the applicability of extractable data assessed for each component (8.136)	EU GMP Annex 1, Reference 4 https://health.ec.europa.eu/system/files/2022-08/20220825_gmp-an1_en_0.pdf	Sourced
That an in-line Raman probe simultaneously predicted glutamine, glutamate, glucose, lactate, ammonium, viable cell density and total cell density, reported as the first demonstration of feasibility in mammalian cell culture bioreactors	Abu-Absi et al. 2011, Reference 5 https://doi.org/10.1002/bit.23023	Sourced
That in a model autologous T-cell immunotherapy process (stirred tank), models for glucose, glutamine, lactate and ammonia were built from reference analyser data and donor-specific changes tracked. That univariate models were correlated with cell concentration and viability. That measurements in cell therapy processes are in many cases off-line at set time points	Baradez et al. 2018, Reference 6 https://doi.org/10.3389/fmed.2018.00047	Sourced
That an in-line method for glucose, lactate and viable cell density was developed for a 1,500 L commercial process. RMSEP (glucose 0.22 g/L, range 1.66 to 3.53; lactate 0.08 g/L, range 0.15 to 1.19; viable cell density 0.31 × 10⁶ cells/mL, range 0.96 to 5.68). That the effects of measurement channel and batch number were examined. That the method met the analytical purpose	Yan et al. 2024, Reference 7 https://doi.org/10.1002/biot.202300395	Sourced
That adoption of Raman chemometrics is limited; that the barriers are selectivity, multicollinearity and insufficient training range; transfer from ambr250 to a 3 L vessel	Rapala et al. 2026, Reference 8 https://doi.org/10.1016/j.jpba.2026.117625	Sourced
Requirements for disposable sensors (per-use cost rather than lifetime or steam resistance). The layout of optical sensors (reusable fibre optic, optical window, patch with immobilised indicator dye). Fluorescence quenching for dissolved oxygen; pH indicator dyes. The prohibition on leaching. Cross sensitivity, dynamic range and loss of sensitivity with sterilisation in pH optodes. Concerns about robustness in manufacturing (lot-to-lot variability, equilibration time, drift). Accuracy of single-use glass electrodes after γ-radiation; γ-tolerance of biosensors. That capacitance sensor ports are sterilised with the bioreactor by γ-radiation. The glass-window adaptor for Raman. Water interference (low for Raman, a challenge for NIR). That only living cells significantly influence electrical characteristics. That there is no standard interface. Pre-sterilised, pre-calibrated free-floating sensors	Busse et al. 2017, Reference 9 https://pmc.ncbi.nlm.nih.gov/articles/PMC6999375/	Sourced
That dielectric spectroscopy can meet some PAT requirements as a process-monitoring tool for cell culture	Justice et al. 2011, Reference 10 https://doi.org/10.1016/j.biotechadv.2011.03.002	Sourced
Dividing RMSEP by range width to get about 11.8%, 7.7% and 6.6%. Putting sampling of 1 mL once a day for 10 days at 10% of a 100 mL culture and about 0.0007% of 1,500 L	Our calculation. Dividing by range width, and the broth volumes, sample volume and frequency, are all premises and assumptions set by this article; they are not the paper's metric or values from any real process	Our calculation
A unified guidance on using PAT as a basis for release decisions for cell products; later progress on standard interfaces; maintaining models in processes where the starting material changes with every lot	No primary source could be confirmed as of this article's research (September 2026), or these are described as future tasks	Not yet confirmed
The explanation of the Raman scattering principle. The off-line category, the four-way split and what each means in cell culture. “Why watch it” for each parameter. The reading that the interface is a material. The four conditions for window materials (optical, product contact, fouling, sterilisation and sealing). The readings on selective permeation and immobilisation in patches, γ-bleaching and degradation, lot-to-lot variation and cost. The framing that membrane and dye makers could take the lead. The reading that sensor patches naturally fall within the Annex 1 assessment. The framing of RTRT and quality assurance for cell products. That no primary source was found stating potency can be measured in-line. The layouts and flows of Figs. 1 to 3 and 5	This article's own structuring and commentary based on the published material. Not views expressed by regulators, authors or companies	Commentary
That Figs. 1 to 3 and 5 are explanatory drawings and Fig. 4 is drawn from our calculation. That the hero image is AI-generated	Our note	Commentary

Last updated 23 September 2026. Sources are limited to primary material (FDA guidance, ICH guidelines, European Commission GMP guidelines and peer-reviewed papers). The content of papers is limited to what could be confirmed from their abstracts and, for open-access papers, their full text. Because the article includes explanations of measurement principles and readings from a materials angle, those are marked as Commentary and kept separate from sourced fact. The percentages and sample-volume comparison in Section 7 are this article's calculations and do not apply to any particular process or product. Figs. 1 to 3 and 5 are vector drawings, Fig. 4 is a vector drawing based on our calculation, and the hero image is AI-generated; none of them shows a real instrument, sensor or spectrum.