Process characterization is the systematic experimental study that defines how critical process parameters (CPPs) affect critical quality attributes (CQAs) and process performance. It is the scientific foundation of Stage 1 (Process Design) under the FDA 2011 Process Validation Guidance and the mechanism by which CPP-CQA relationships are established for regulatory filings. A well-executed process characterization study typically evaluates 15-30 parameters across 40-80 DOE runs in a qualified scale-down model, culminating in defined proven acceptable ranges (PARs), normal operating ranges (NORs), and a design space per ICH Q8(R2).
This guide walks through the complete process characterization workflow, from risk-based parameter classification through DOE execution to design space establishment, with a worked mAb example and real parameter ranges.
What Is a Process Characterization Study?
A process characterization study is a structured series of experiments that identifies which process parameters significantly affect product quality and process performance, quantifies those relationships, and defines the acceptable operating ranges. It is the most resource-intensive activity in biologics process development, typically consuming 6-12 months and 40-80 scale-down bioreactor runs.
The study answers three questions that regulators require in every BLA or MAA submission:
- Which parameters matter? Of the 20-40 parameters that could theoretically affect product quality, which ones actually do (CPPs) and which are non-critical (non-CPPs)?
- How much can each CPP vary? The proven acceptable range (PAR) defines the experimentally demonstrated window within which all CQAs meet specification.
- How do CPPs interact? The design space captures the multidimensional combination of CPPs that provides quality assurance, including parameter interactions that narrow the acceptable range beyond what univariate studies would predict.
A seven-step workflow diagram showing risk assessment using FMEA and ICH Q9, CPP classification into high, medium, and low risk categories, scale-down model qualification matching P/V and kLa parameters, screening DOE using fractional factorial designs with 15 to 30 runs, optimization DOE using CCD or BBD with 20 to 40 runs, and PAR NOR and design space definition per ICH Q8 R2. Below shows outputs feeding into BLA regulatory submission modules. Four regulatory framework boxes annotate the applicable ICH Q9, Q11, Q8 R2, and FDA Process Validation Stage 1 guidance documents.
Regulatory Framework: ICH Q8, Q9, Q11, and FDA Stage 1
Process characterization sits at the intersection of four ICH guidelines. ICH Q8(R2) defines the design space as "the multidimensional combination and interaction of input variables and process parameters that have been demonstrated to provide assurance of quality." ICH Q9 provides the risk management tools (FMEA, PHA) used to prioritize which parameters to study. ICH Q11 describes the enhanced development approach where process understanding drives parameter classification. The FDA 2011 Process Validation Guidance frames process characterization as Stage 1 (Process Design), preceding Stage 2 (PPQ) and Stage 3 (Continued Process Verification).
The EMA's 2017 Q&A document on NOR, PAR, and design space provides the clearest regulatory distinction between these three concepts. It explicitly states that the NOR should always be narrower than the PAR, and that the design space may be narrower still when parameter interactions are significant.
| Concept | Definition | Width | How established | Regulatory change control |
|---|---|---|---|---|
| Edge of failure | Parameter value where at least one CQA fails specification | Widest | DOE extreme conditions | Not filed; internal knowledge |
| PAR | Range where all CQAs meet spec (experimentally proven) | Wide | DOE + statistical prediction intervals | Filed in BLA; changes need prior approval |
| Design space | Multidimensional PAR accounting for interactions | Medium | Response surface modeling of DOE data | Filed in BLA; movement within is not a change |
| NOR | PAR minus routine process variability (2-3 sigma) | Narrow | Historical manufacturing data + control capability | Filed; defines setpoint ± tolerance |
| Target setpoint | Nominal operating value within NOR | Point | Process development optimization | Filed in batch record |
Risk Assessment and CPP Classification
Risk assessment reduces the experimental burden by 40-60% by focusing DOE resources on parameters most likely to affect product quality. An FMEA-based approach scores each process parameter on three scales: severity (impact if the parameter deviates, 1-10), occurrence (likelihood of deviation during manufacturing, 1-10), and detection (ability to detect the deviation before it affects product, 1-10). The risk priority number (RPN = S x O x D) ranks parameters into high (>100), medium (40-100), and low (<40) risk categories.
A typical mAb upstream process starts with 20-35 candidate parameters. After FMEA scoring, 8-12 emerge as high/medium risk and enter the screening DOE. The remainder are classified as non-CPPs, documented in the risk assessment report, and confirmed via univariate verification runs at extreme conditions.
| Parameter | Severity (1-10) | Occurrence (1-10) | Detection (1-10) | RPN | Classification |
|---|---|---|---|---|---|
| pH setpoint | 8 | 3 | 2 | 48 | Medium |
| Temperature setpoint | 9 | 3 | 2 | 54 | Medium |
| Temperature shift timing | 7 | 4 | 3 | 84 | Medium |
| DO setpoint | 7 | 3 | 2 | 42 | Medium |
| Feed start day | 6 | 4 | 3 | 72 | Medium |
| Feed rate multiplier | 8 | 5 | 4 | 160 | High |
| Seed density | 7 | 4 | 3 | 84 | Medium |
| Agitation speed | 5 | 3 | 2 | 30 | Low |
| Glucose concentration | 6 | 5 | 5 | 150 | High |
| Media lot | 7 | 6 | 7 | 294 | High |
| CO2 overlay % | 4 | 3 | 3 | 36 | Low |
| Antifoam addition | 3 | 3 | 2 | 18 | Low |
Scale-Down Model Qualification
Process characterization studies are executed in a qualified scale-down model, not at manufacturing scale, because running 40-80 experiments in a 2,000 L bioreactor would be prohibitively expensive and time-consuming. The scale-down model (typically 2-15 L) must be statistically demonstrated to reproduce manufacturing-scale performance and product quality.
Key engineering parameters to match within defined tolerances:
- P/V within 20% of manufacturing scale (preserves mixing intensity and kLa)
- kLa within 20% (ensures equivalent oxygen delivery)
- pCO2 profile matched via 5-15% CO2 overlay (mammalian cells; manufacturing-scale vessels accumulate 100-200 mmHg)
- Geometric similarity where feasible (H/T ratio, D/T ratio, impeller type)
Qualification requires a minimum of 5 independent set-point runs compared to manufacturing data using TOST equivalence testing. Practical difference thresholds are typically 10% for VCD and viability, 15% for titer and metabolites, and 25% for growth rate. A predictiveness classification (Case A through D) grades the model's suitability for characterization work.
DOE Design and Execution
Process characterization uses a two-phase DOE strategy. Phase 1 (screening) identifies which parameters significantly affect CQAs from the high- and medium-risk list. Phase 2 (optimization) maps the response surface for confirmed CPPs and defines their acceptable ranges.
Phase 1: Screening DOE
Screening designs evaluate 8-15 parameters in the fewest possible runs. The two most common options:
- Fractional factorial (Resolution IV or V): 16-32 runs for 8-12 factors. Resolution IV resolves main effects but confounds two-factor interactions. Resolution V resolves both main effects and two-factor interactions.
- Definitive screening design (DSD): 2k+1 runs for k factors (e.g., 21 runs for 10 factors). Estimates main effects, quadratic effects, and some two-factor interactions in fewer runs than a fractional factorial. Increasingly favored for biologics characterization.
Each parameter is tested at two levels (low and high) spanning a range wider than the expected NOR but within a safe operating envelope. Ranges are typically set at the edge of historical experience plus a safety margin. Center points (3-5 replicates) provide an estimate of pure error and a check for curvature.
Responses measured include both performance (VCD, viability, titer, growth rate) and quality attributes (glycosylation profile, charge variants, aggregation, HCP, potency). Statistical significance is assessed at p < 0.05, though parameters with p-values between 0.05 and 0.10 may be retained for Phase 2 if the biological mechanism supports their inclusion.
Phase 2: Optimization DOE
The confirmed CPPs (typically 3-6 parameters) enter a response surface design:
- Central composite design (CCD): factorial + axial + center points. 2^k + 2k + 3-5 center points. For 4 factors: 16 + 8 + 5 = 29 runs. Estimates full quadratic models including interactions.
- Box-Behnken design (BBD): 3 levels per factor, avoids extreme corners. For 4 factors: 27 runs. Preferred when extreme combinations are physically unrealistic or could damage cells.
The optimization DOE generates response surface models (typically second-order polynomial: Y = b0 + sum(bi*xi) + sum(bii*xi^2) + sum(bij*xi*xj)) that predict each CQA as a function of the CPPs. Model fit is assessed by R-squared (aim >0.80), adjusted R-squared, predicted R-squared (should be within 0.20 of adjusted), lack-of-fit test, and residual analysis.
DOE Experiment Generator
Generate fractional factorial, CCD, and Box-Behnken designs for process characterization. Calculate required run counts and randomization schedules.
How to Define PAR and NOR from DOE Data
The proven acceptable range is the widest parameter range, demonstrated experimentally, within which every CQA simultaneously meets its specification. For non-interacting parameters, PAR is read directly from the DOE: the parameter range tested where all responses passed. For interacting parameters, PAR must be narrowed to account for the interaction, because a parameter at the edge of its univariate PAR may push a CQA out of specification when another interacting parameter is also at its edge.
The procedure for defining PAR from optimization DOE data:
- Fit response surface models for each CQA as a function of the CPPs.
- Define the CQA specification limits (acceptance criteria from the quality target product profile).
- For each CPP, calculate the 95% prediction interval on each CQA response across the studied range.
- The PAR is the CPP range where the 95% prediction interval for every CQA falls within specification simultaneously.
- When two CPPs interact significantly (p < 0.05 for the interaction term), the PAR for each is conditional on the other's value. Report this as a contour plot or an interaction table.
The normal operating range is derived differently. NOR reflects what the manufacturing process can routinely achieve, not what it can tolerate. It is calculated from historical manufacturing data (or PPQ data) as:
NOR = Target setpoint ± (k x sigma_process)
Where k is typically 2-3 (covering 95-99.7% of routine deviations) and sigma_process is the standard deviation of the parameter as controlled during manufacturing. NOR must always fall inside the PAR. If it does not, either the control strategy needs tightening (reduce sigma_process) or the PAR needs re-evaluation.
Establishing the Design Space
The design space is the multidimensional region within the PAR where all CQA specifications are met simultaneously. It is always a subset of the individual PARs because parameter interactions create corners of the multidimensional space where combinations of extreme values push at least one CQA out of specification, even though each parameter individually is within its PAR.
Two approaches are commonly used to define and present the design space:
Overlay contour approach
For each pair of interacting CPPs, generate contour plots of each CQA response surface with specification limits overlaid. The design space is the region where all contour constraints are satisfied simultaneously. This approach is intuitive but limited to 2D projections; for designs with more than 2 CPPs, multiple pairwise plots are needed (with other parameters held at their target setpoints).
Probabilistic (Bayesian) approach
Rather than asking "do all CQAs meet spec?" (which is a point prediction), the probabilistic approach asks "what is the probability that all CQAs meet spec simultaneously?" A reliability threshold of 95% or 99% is applied, and the design space is defined as the region where P(all CQAs in spec) exceeds that threshold. This approach is more conservative, accounts for model uncertainty, and is increasingly preferred by regulators. Monte Carlo simulation (10,000+ iterations) propagating model uncertainty generates the probability surface.
Filing the design space in a BLA or MAA requires:
- The list of CPPs and their individual PARs
- Response surface model equations (or a clear graphical representation)
- Contour plots showing the design space boundaries for interacting parameters
- A statement that movement within the design space is not considered a change
- The control strategy that ensures manufacturing stays within the design space
From Design Space to Control Strategy
The control strategy translates design space boundaries into actionable manufacturing instructions. It defines the target setpoint, NOR, alarms, and corrective actions for each CPP, ensuring that routine manufacturing stays well within the proven acceptable range.
| CPP | Target setpoint | NOR | PAR | Action if outside NOR |
|---|---|---|---|---|
| Temperature (growth) | 37.0 °C | 36.5-37.5 °C | 35.0-38.0 °C | Investigate; adjust jacket setpoint |
| Temperature (production) | 33.0 °C | 32.5-33.5 °C | 31.0-35.0 °C | Investigate; verify shift executed correctly |
| pH | 7.00 | 6.90-7.10 | 6.80-7.20 | Check probe; verify CO2/base delivery |
| DO | 40% | 30-50% | 20-60% | Check cascade; verify aeration |
| Feed rate (% of target) | 100% | 90-110% | 75-130% | Check pump calibration; verify feed bottle |
| Seed density (106/mL) | 0.5 | 0.3-0.7 | 0.2-1.0 | Adjust N-1 passage timing |
Worked Example: CHO mAb Process Characterization
Worked Example: Defining PAR for Temperature
Setup: A CHO mAb fed-batch process (14-day culture, target 5.0 g/L titer) is undergoing process characterization for BLA filing. A qualified 5 L scale-down model has been validated against the 2,000 L manufacturing process. Temperature was identified as a medium-risk parameter (RPN = 54) and confirmed as a CPP in the screening DOE (p = 0.003 for titer, p = 0.01 for galactosylation).
Optimization DOE: A CCD was run with temperature as one of 4 CPPs. The studied range was 32.0-35.0 °C (production phase, after day-3 shift from 37 °C). Center point: 33.5 °C, 5 replicates.
Results from response surface model:
- Titer: Y = -142.7 + 8.73T - 0.129T2 (R2 = 0.91). Maximum titer at 33.8 °C.
- Galactosylation (G1F+G2F %): Y = -85.2 + 5.31T - 0.078T2 (R2 = 0.87). Galactosylation increases with temperature.
- Titer specification: ≥ 3.0 g/L
- Galactosylation specification: 35-65%
PAR calculation:
- Titer ≥ 3.0 g/L predicted (95% PI) at T = 31.8-35.2 °C
- Galactosylation 35-65% predicted (95% PI) at T = 31.5-36.0 °C
- PAR (intersection) = 31.8-35.2 °C (titer is the binding constraint)
NOR calculation:
- Historical manufacturing sigma for production temperature: 0.3 °C
- NOR = Target ± 2.5 sigma = 33.0 ± 0.75 °C = 32.25-33.75 °C
- NOR (32.25-33.75 °C) falls well inside PAR (31.8-35.2 °C).
pH interaction: Temperature interacted significantly with pH (p = 0.02). At pH 7.2 (high end of pH PAR), the temperature PAR narrows to 32.2-34.8 °C because galactosylation exceeds 65% at high temperature + high pH. This interaction is captured in the design space contour but does not affect the NOR because pH NOR (6.90-7.10) does not reach pH 7.2.
Scale-Up Calculator
Verify that your scale-down model matches manufacturing-scale P/V, tip speed, and kLa. Compare five scale-up criteria side-by-side.
Fed-Batch Calculator
Calculate feeding profiles for your scale-down characterization runs. Exponential, linear, and constant strategies with organism presets.
Related Tools
- DOE Experiment Generator — Generate fractional factorial, CCD, and Box-Behnken designs for process characterization studies.
- Scale-Up Calculator — Match engineering parameters (P/V, kLa, tip speed) between scale-down model and manufacturing scale.
- Clone Scorecard — Rank clones on productivity, quality, and stability for CLD decisions that precede process characterization.
References
- Rathore AS, Winkle H. Quality by design for biopharmaceuticals. Nature Biotechnology. 2009;27(1):26-34. doi:10.1038/nbt0109-26
- Horvath B, Mun M, Laird MW. Characterization of a monoclonal antibody cell culture production process using a quality by design approach. Molecular Biotechnology. 2010;45(3):203-206. doi:10.1007/s12033-010-9267-4
- Abu-Absi SF, Yang L, Thompson P, et al. Defining process design space for monoclonal antibody cell culture. Biotechnology and Bioengineering. 2010;106(6):894-905. doi:10.1002/bit.22764
- ICH Q8(R2). Pharmaceutical development. International Council for Harmonisation. 2009. Available at: ich.org
- FDA. Process Validation: General Principles and Practices. Guidance for Industry. January 2011.
Frequently Asked Questions
What is the difference between PAR and NOR in process characterization?
The proven acceptable range (PAR) is the experimentally demonstrated range for a critical process parameter within which all critical quality attributes remain within specification. The normal operating range (NOR) is a tighter subset of the PAR that accounts for routine process variability and operational control capability. A typical NOR covers the PAR minus 2-3 standard deviations of historical operating data on each side. EMA guidance explicitly states that NOR should always be narrower than PAR.
How many DOE runs are needed for a process characterization study?
A typical biologic process characterization study requires 40-80 total runs across screening and optimization phases. Screening DOEs (fractional factorial or definitive screening designs) use 15-30 runs to evaluate 8-15 parameters. Optimization DOEs (central composite or Box-Behnken designs) use 20-40 runs for 3-6 confirmed CPPs. Including center-point replicates (minimum 3-5 per design) for reproducibility assessment, most programs complete 50-70 runs using a qualified scale-down model over 6-12 months.
What DOE designs are used for process characterization of biologics?
Process characterization typically uses a two-phase DOE approach. Phase 1 (screening) employs fractional factorial designs (Resolution IV or V) or definitive screening designs to evaluate 8-15 parameters in 15-30 runs. Phase 2 (optimization) uses response surface designs such as central composite designs (CCD) or Box-Behnken designs (BBD) to map the response surface for 3-6 confirmed CPPs. Center points are replicated 3-5 times in each design to estimate pure error.
What is a design space under ICH Q8?
ICH Q8(R2) defines the design space as the multidimensional combination and interaction of input variables and process parameters demonstrated to provide assurance of quality. Movement within the design space is not considered a change and does not require regulatory prior approval. The design space is established through process characterization DOE studies, bounded by the proven acceptable ranges of each CPP, and is always a subset of the PAR because it accounts for parameter interactions.
How do you define a proven acceptable range from DOE data?
The PAR is defined by overlaying specification limits for all CQAs on the DOE response surface and identifying the parameter range where every CQA simultaneously meets specification. For interacting parameters, the PAR is narrower because parameter combinations near the extremes may push CQAs out of specification. Statistical prediction intervals (95% confidence) should be used rather than point estimates to account for model uncertainty.