Comparability and Potency Assays for Biologics and Cell & Gene Therapies

Two of the hardest CMC problems for biologics and for cell and gene therapy (CGT) products live next to each other: proving that product made after a change is still the same product (comparability), and proving that a living or complex molecule actually does what it is supposed to do (potency). They are linked. A comparability exercise that lacks a meaningful potency assay is a comparability exercise that cannot conclude anything about the mechanism that matters to the patient. A potency assay that drifts or fails to discriminate makes every future change harder to justify. This article walks through both, with the regulatory basis, the procedure, the acceptance logic, worked examples, and the findings inspectors keep writing up.

The audience is CMC, analytical development, QA, and regulatory practitioners who own or review these packages. If you sit in any of those seats and you can explain the matrix approach, the difference between comparability and biosimilarity, and why a single potency number is rarely enough, you are ahead of most candidates in an interview and most reviewers in an inspection.

Part 1: Comparability

What comparability is and why it is required

Comparability is the structured demonstration that a biological product remains highly similar in quality after a manufacturing change, such that the change has no adverse impact on safety or efficacy. It is not a claim that pre-change and post-change material are identical. Biologics are heterogeneous by nature, so identity at the molecule level is the wrong bar to aim at. What you are actually asked to establish is that the pre-change and post-change populations sit close enough together on the attributes that carry clinical consequence, and that whatever small differences remain have been looked at deliberately and shown to be without consequence for how the product behaves in a patient. Sameness is not the test. Predictable equivalence of behavior on the attributes that drive safety and efficacy is the test, and the word predictable is carrying weight there, because the prediction rests on what you already know about your own product.

The governing guidance is ICH Q5E, “Comparability of Biotechnological/Biological Products Subject to Changes in Their Manufacturing Process” (2004). ICH Q5E sets the framework: you assess the change, you characterize pre- and post-change product, and you decide whether quality data alone are sufficient or whether nonclinical or clinical bridging data are needed.

Read the guideline itself for its exact wording, because both the general principles section and the glossary carry the operative definition and you want the current text in front of you when you write a protocol. Described in my own terms, the logic runs like this. Identity between old and new material is not what you are being asked to prove, and for a heterogeneous biological product it would be an impossible target in any case. The question is whether the two sit close enough on the attributes that matter, and whether you already understand the product and the process well enough to say with confidence that any residual difference will not change how the drug behaves in a patient. Accumulated knowledge is doing real work in that formulation. The strength of your conclusion depends on how much you have already learned about which attributes move, how far they move under normal process variation, and what happens clinically when they do. A young program with thin characterization history cannot lean on that knowledge the way a mature one can, which is why the same numerical difference can be acceptable for one product and a stop signal for another.

Comparability does not stand alone. It pulls in:

• ICH Q5C (stability of biotech products), because you usually compare stability profiles.
• ICH Q6B (specifications for biotech products), because the specification is one anchor of the comparison.
• ICH Q8(R2), Q9(R1), Q10, Q11, Q12, Q14, because change management, risk, lifecycle, and analytical procedure lifecycle all touch comparability.
• For post-approval changes in the US, 21 CFR 601.12 (changes to an approved BLA) defines which changes need prior approval (PAS), CBE-30, CBE-0, or annual report, and the comparability data drive that classification.
• In the EU, the variations framework (Regulation (EC) No 1234/2008 and Commission guidelines) plays the equivalent role.

The quality rationale: a process change can shift glycosylation, aggregation, charge variants, impurity profile, or, for CGT, vector identity, transduction efficiency, or cell phenotype. Some of those shifts are clinically silent. Some are not. Comparability forces you to find the shifts before the patient does.

For CGT products specifically, FDA’s CBER finalized a guidance, “Chemistry, Manufacturing, and Controls Flexibilities for Developing Human Cellular and Gene Therapy Products for a Biologics License Application” (issued 6 May 2026, for immediate implementation), that endorses a risk-based approach to comparability studies during development, alongside phase-appropriate CGMP and investigational-phase release acceptance criteria. The thrust is that the size and timing of a comparability exercise should scale with the risk of the change and the stage of development, rather than every change triggering the full commercial package. It does not lower the bar at licensure; it formalizes proportionality earlier in the lifecycle. Confirm the current text and scope against the published guidance before you rely on it.

Comparability is not biosimilarity

This distinction comes up in every serious interview. Comparability (ICH Q5E) is a single manufacturer comparing its own product before and after its own change, with full access to the historical process, samples, and clinical data. Biosimilarity (a different regulatory pathway, ICH-aligned but governed by region-specific frameworks such as the US 351(k) pathway and EMA biosimilar guidelines) is a different sponsor trying to match an originator product they did not make, with no access to the originator process. The analytical toolbox overlaps heavily, but the evidentiary burden and the statistical approaches differ. If an interviewer asks “is this comparability or biosimilarity,” the answer hinges on who owns the reference and the process history.

The EU comparability framework

ICH Q5E is not a US-only document. The European Medicines Agency adopted it directly as a CHMP scientific guideline (EMA reference CPMP/ICH/5721/03, effective from June 2005), so the quality-comparability logic described above is the same text a European reviewer is working from. What differs is the companion guidance for the non-clinical and clinical questions a change can raise once the quality comparison is done. The EU has a dedicated document for that: “Guideline on comparability of biotechnology-derived medicinal products after a change in the manufacturing process: non-clinical and clinical issues” (EMEA/CHMP/BMWP/101695/2006, effective 1 November 2007), which replaced an earlier guideline on the same topic. It sets out when the quality data from the ICH Q5E exercise are enough on their own and when additional non-clinical or clinical bridging should be considered, following the same principle as the US approach: the size of the bridging package tracks the risk and the residual uncertainty left after the quality comparison, not a fixed rule tied to the type of change. Confirm the current version and any subsequent revision of both documents before you cite them in a filing, since guidelines in this area are periodically updated.

When comparability is triggered

Any change to the manufacturing process of a licensed or clinical-stage biologic can trigger a comparability assessment. Common triggers:

• Cell bank change (new MCB or WCB, host cell line change).
• Scale change (for example 200 L to 2000 L bioreactor).
• Site change or addition of a second manufacturing site.
• Raw material or reagent change (media component, resin, filter, single-use assembly).
• Process parameter or sequence change (purification step added or removed, formulation change).
• Container closure change.
• For CGT: change in vector production, plasmid source, transfection method, cell selection, cryopreservation, or fill format.

The change itself does not tell you the size of the comparability exercise. The risk of the change does. A buffer supplier change with identical specification is low risk. A change in the affinity resin or in the viral vector packaging system is high risk. ICH Q5E expects a risk-based scope.

Multi-site and scale-up comparability

A common trigger from the list above is a second manufacturing site or a change in scale, and it deserves its own treatment because it gets confused with technology transfer and conflated with process validation when the two are related but distinct exercises. Technology transfer and process performance qualification at the receiving site or the new scale answer “does the process work here.” Comparability answers a narrower, harder question: “is the product made here, or at this scale, still the same product on the attributes that matter.” A site can pass every transfer acceptance criterion and still produce material that sits outside the historical quality range on an attribute nobody thought to watch, because equipment, environment, and operators are never perfectly identical between two locations.

What changes across a site or scale move that a single-site process change does not:

• Equipment and facility differences. Different vessel geometry, mixing characteristics, filter housings, or single-use assembly vendors, even when the specification on paper is the same. See assessment-equipment-facility-equivalence-transfer for a structured way to document equipment and facility equivalence ahead of the comparability lots.
• Scale-dependent parameters. Mixing time, shear, oxygen transfer, and heat transfer do not scale linearly. For cell culture and fermentation processes, this can shift glycosylation, aggregation, or metabolite profiles even when every set point on the batch record reads identical.
• Raw material and utility differences. A second site rarely sources every raw material, gas, or water system from the same qualified supplier chain. Confirm material comparability separately; it is a common blind spot when the comparability plan assumes “same process” also means “same inputs.”
• Environmental and personnel differences. Different classified-area design, different operators, different local interpretation of the same master batch record. None of these show up in a paper gap assessment.

How to run it: treat the second site or new scale exactly as any other change under the risk-based scope from step 2 of the procedure above, but widen the lot selection so the pre-change population is the originating site or scale and the post-change population is the receiving site or the new scale, run head to head where retains allow, and add a statistical process-capability comparison, not just a mean comparison, because a second site can shift variance without shifting the mean. Extend the stability comparison across sites, because a packaging or storage difference at the new site can change the degradation pathway even when release results match. Where the receiving site is a contract manufacturer, pair this work with a formal technology transfer protocol; see quality-in-technology-transfer and protocol-technology-transfer-cdmo for the transfer side of the exercise.

A recurring finding here: a site or scale change is treated purely as a validation exercise (three PPQ batches, no comparability protocol), and the comparability conclusion is inferred from validation success rather than assessed against the pre-change quality range. Passing PPQ shows the new site or scale is in control. It does not, by itself, show the new site’s product is comparable to the old site’s product.

What goes in a comparability package

A comparability protocol and report typically contain:

1. Description of the change and the rationale.
2. Risk assessment of the change against quality attributes, ideally tied to your Quality Target Product Profile (QTPP) and Critical Quality Attributes (CQAs). See quality-by-design-and-doe and quality-risk-management.
3. Comparability strategy: which attributes, which methods, how many lots, what acceptance criteria, and whether the comparison is at drug substance, drug product, or both.
4. Analytical methods list with their validation or qualification status and discriminating power.
5. Pre-change and post-change lot selection and justification (representative, recent, relevant).
6. Acceptance criteria per attribute, set in advance.
7. Stability comparison plan (real-time and accelerated/forced degradation).
8. Results, statistical analysis, and conclusion, including any out-of-expectation differences and their assessment.
9. Decision on whether quality data are sufficient or whether nonclinical/clinical bridging is required, and the regulatory reporting category.

The comparability attribute matrix

The core deliverable is an attribute-by-method matrix. Each row is a quality attribute, each column an analytical dimension. A worked example for a monoclonal antibody:

For a CGT product (for example an AAV gene therapy), the matrix shifts:

The matrix is the heart of the package. Orthogonality matters: more than one independent method should support each high-criticality conclusion, because no single method sees everything.

How to run a comparability exercise, step by step

1. Define the change and its boundaries. Exactly what is changing, and what is held constant.
2. Run the risk assessment. Map the change to each CQA. Rate impact and detectability. This tells you which attributes need the closest look and where extra orthogonal methods are warranted. Tie this to change-control-validated-systems and your formal change record.
3. Write the comparability protocol BEFORE generating data. Pre-specify attributes, methods, lot numbers, sample sizes, and acceptance criteria. Post hoc acceptance criteria are a classic finding.
4. Confirm method suitability. Each method used in the comparison should be validated or at least qualified and demonstrably stability-indicating and discriminating. A method that cannot detect a real difference will pass everything, which is worse than failing.
5. Select lots. Use representative pre-change lots (recent, in-control, ideally spanning the historical range) and the available post-change lots. CGT autologous products complicate this because each lot is one patient; you may have to compare process-representative or engineering runs.
6. Execute side by side where possible. Same analyst, same instrument, same reference standard reduces between-run noise. Head-to-head testing of retained pre-change samples against fresh post-change samples is the strongest design.
7. Run forced degradation and stability comparison. Comparable release data with divergent degradation pathways is a red flag. ICH Q5C and side-by-side accelerated/stress studies expose that.
8. Analyze against pre-set criteria. Where you used ranges, confirm overlap. Where you used statistical equivalence, apply the pre-specified test.
9. Adjudicate differences. Any observed difference is assessed for clinical relevance, not just statistical significance. Document the rationale.
10. Conclude and classify. State comparable / not comparable / comparable-with-conditions, decide if bridging data are needed, and assign the regulatory reporting category (PAS, CBE-30, CBE-0, annual report, or EU variation type).

The decision path, from change to filing, follows one logic regardless of modality:

Risk-based testing tiers: a scope-setting matrix

The decision flow above tells you the sequence: assess risk, pre-specify, execute, decide. It does not tell you, attribute by attribute, how deep to go. A second tool answers that question directly: for a given attribute, combine how severe a shift would be for the patient with how well your current methods would actually catch it, and read off a testing tier. This is a scope-setting matrix, not a per-change score sheet; use it alongside, not instead of, the formal change-to-CQA risk assessment in step 2, which assigns each attribute in your specific change to one of the three tiers below.

What each tier means in practice:

Reading the grid: an attribute with a validated, orthogonal, stability-indicating method and a low clinical consequence if it shifts (Tier 1) does not need a special campaign; the routine release data carry the argument. An attribute that only a single non-discriminating method covers, where a shift would matter clinically, lands in Tier 3 regardless of how minor the manufacturing change looks on paper, because the risk sits in not being able to see the problem, not in how big the change appears. This is the same logic that pushes host cell protein, potency, and vector-related attributes into the highest tier on most CGT comparability packages: high clinical consequence, and detectability that is only ever as good as the last method qualification.

Setting acceptance criteria

This is where comparability lives or dies. Three common approaches:

• Specification-based: post-change results meet the registered specification. Necessary but weak on its own, because specifications are often wide.
• Range-based / quality-range: define an acceptable range from the historical pre-change dataset (for example mean ± 3 SD, or min-max of historical lots). Post-change lots must fall inside. Document how the range was derived.
• Statistical equivalence: for attributes with enough lots, pre-specify an equivalence margin and test (for example a two-one-sided-test approach). More rigorous, needs more lots, and the margin must be justified on clinical relevance, not convenience.

A worked example for the aggregate attribute (%HMW by SEC):

If a post-change lot had read 2.1%, it would still pass the specification (NMT 3.0%) but fail the quality range. That is exactly the situation comparability is meant to catch: in-spec but not comparable. You would then investigate, and you might need to characterize the new aggregate species and assess clinical relevance before concluding.

Roles and responsibilities for comparability

QA owns the gate. Analytical and process SMEs own the science. Regulatory owns the filing consequence.

Part 2: Potency assays

What potency is and why it is required

Potency is the measure of the biological activity of the product using a relevant, quantitative biological assay, based on the attribute or attributes linked to the relevant biological properties. For US biologics, potency is a statutory requirement. 21 CFR 600.3(s) defines potency as “the specific ability or capacity of the product, as indicated by appropriate laboratory tests or by adequately controlled clinical data obtained through the administration of the product in the manner intended, to effect a given result.” 21 CFR 610.10 requires that tests for potency consist of either in vitro or in vivo tests, or both, specifically designed for each product to indicate its potency.

“Tests for potency shall consist of either in vitro or in vivo tests, or both, which have been specifically designed for each product so as to indicate its potency in a manner adequate to satisfy the interpretation of potency given by the definition in 600.3(s) of this chapter.” (21 CFR 610.10)

The key FDA guidance for CGT is “Potency Tests for Cellular and Gene Therapy Products” (FDA, January 2011) and the more recent draft “Potency Assurance for Cellular and Gene Therapy Products” (draft, Dec 2023) which moves toward a potency assurance strategy (a system of complementary measures) rather than reliance on a single assay. FDA supplemented that draft on 20 Aug 2026 with a companion draft, “Potency Assessment of Active Immunotherapy Products” (docket FDA-2026-D-8561, comments through 18 Nov 2026), which extends the same potency assurance strategy thinking beyond CGT to active immunotherapy products (ACTIMPs): peptide- and protein-based, vectored, personalized, and cell-based products that are not themselves cell or gene therapies. Neither document is finalized. ICH ties in through ICH Q6B, which addresses biological activity in the specification, and the analytical procedure lifecycle now sits under ICH Q2(R2) (2023) for validation and ICH Q14 (2023) for development.

Why it is required, in quality terms: potency is the assay that links your product to its mechanism of action. Identity, purity, and quantity tell you what is in the vial. Potency tells you whether it works. For a “living drug” (a cell therapy, a gene therapy), the active ingredient is the biology, and a chemical quantity measurement (mg, vector genomes) does not, on its own, prove function.

The potency problem for living drugs

Conventional small molecules have a defined structure and a content assay that essentially equals potency. Biologics are harder. CGT products are the hardest case:

• Mechanism may be complex or partly unknown. A cell therapy may act through several pathways (cytotoxicity, cytokine secretion, persistence, differentiation). Which one do you measure?
• The product is heterogeneous and live. Cell composition varies. Viability matters. The product changes over its short shelf life.
• Short shelf life and small lot size. An autologous product may need to be released and infused before a slow cell-based potency assay can read out. You cannot wait two weeks for a bioassay if the product expires in 48 hours.
• No relevant animal model in some cases, and reference standards are difficult to establish for a product that is unique per patient.
• Assays are inherently variable. Cell-based assays routinely carry high variability, so the assay must be designed and qualified to keep that variability controlled.

This is why FDA’s current thinking favors a potency assurance strategy: a matrix of complementary measurements (for example a quantitative cell-based functional assay plus a marker-based assay plus identity and viability) where the whole set, together with process controls, assures potency, rather than betting everything on one slow, variable bioassay.

Types of potency assay

A binding assay alone is rarely accepted as the sole potency assay when the mechanism involves a downstream effect. The expectation is that the potency measure reflects the relevant biological activity, not just an upstream surrogate.

The matrix approach to potency

For complex products, a single assay seldom captures the full mechanism. The matrix (or multi-assay) approach combines complementary assays so that together they cover identity, quantity, and the relevant function. A worked example for a gene-modified cell therapy:

The functional assay is the anchor. Marker assays are fast and support release timing, but the strategy must show that the markers correlate with function. Over time you build the data that justify which assays release the product and which inform the broader potency assurance picture. FDA expects that at licensure you have a quantitative assay measuring a relevant biological activity, even if early clinical lots were released on a broader matrix.

What goes in a potency assay (design elements)

A defensible potency method specifies:

• The biological activity measured and its link to MOA.
• Reference standard (and a documented qualification and bridging program, two-tier where appropriate: primary reference plus working standard).
• Assay format: relative potency against a reference standard, reported as a percentage of reference, using a parallel-line or 4-parameter logistic (4PL) dose-response model.
• System suitability criteria: curve fit quality (R squared), parallelism between sample and reference, slope, EC50 range, asymptote ranges, plate controls.
• Replication strategy: number of plates, wells, independent assays averaged to a reportable result.
• Acceptance criteria for the reportable value (the specification range, for example 50% to 150% relative potency).
• Validation characteristics per ICH Q2(R2): specificity, accuracy, precision (repeatability and intermediate precision), linearity, range. For bioassays, relative accuracy and the dilutional linearity of relative potency are central.

See bioassay-and-elisa-validation and method-validation-essentials for the validation mechanics.

Relative potency and parallelism: a worked example

Bioassays almost always report relative potency: the activity of the sample as a ratio to a reference standard run on the same plate. The validity of the ratio depends on parallelism, meaning the sample and reference dose-response curves have the same shape. If they are not parallel, the ratio is not meaningful and the result is invalid.

A worked snippet using a 4PL fit:

Reading: the sample is 123% as potent as the reference. Because the curves are parallel and system suitability passes, the 123% is a valid reportable value, and it sits inside the 50% to 150% release specification. If the parallelism equivalence test had failed, the run is invalid and gets repeated or investigated; you do not report a potency number from non-parallel curves. Modern practice uses an equivalence test for parallelism (pre-set equivalence bounds on slope/asymptote ratios) rather than the older difference test, because a difference test perversely rewards an imprecise assay by failing to detect non-parallelism. Knowing that distinction is a strong interview signal.

Deriving the potency specification range: a worked example

The 50% to 150% relative-potency range used above is a common convention in early clinical development, not a number pulled from nowhere. Setting it, and later tightening it toward licensure, follows a defined logic: pool the potency results already generated on manufactured lots, layer in whatever the clinic has actually seen, account for how much noise the assay itself contributes, and set a range wide enough to accommodate real process and assay variability but no wider than the data justify.

A worked example. Suppose 15 manufacturing lots (a mix of process-development, engineering, and early clinical lots) have been tested on the qualified relative-potency assay:

The reasoning runs in three steps. First, look at manufacturing variability alone: mean plus or minus 3 standard deviations gives roughly 56% to 140%, the range routine, in-control lots would be expected to occupy. Second, add the assay’s own contribution: a %GCV of 12% means a meaningful part of the spread in the lot data is assay noise, not a true product-to-product difference, so a specification set exactly at the observed 3-SD process range would fail lots for assay imprecision alone, not for a real potency problem. Third, check the clinical anchor: lots spanning 68% to 126% relative potency have already been dosed without an observed exposure-response signal in that band, which supports treating that span as clinically adequate rather than a boundary to fear.

Putting the three together, a specification of 50% to 150% comfortably contains the observed process range plus the assay’s own imprecision, stays close to what has already been given to patients, and leaves room for the assay and process to be better characterized as more lots accumulate. It is deliberately not the tightest range the data could support at this stage; over the lifecycle, as the lot count grows and the assay moves from qualified to fully validated, the range is expected to narrow toward the true manufacturing capability, justified the same way with the larger dataset, not simply squeezed to whatever the newest lots happen to read.

The finding to avoid: a specification copied from a comparable program or a compendial default, with no lot data, no assay-precision accounting, and no link to what was actually dosed in the clinic behind it. A specification with no derivation logic is one a reviewer can reasonably ask you to defend, and you will not have an answer beyond “that is what we use.”

How to develop and qualify a potency assay, step by step

1. Map the mechanism of action. Identify the biological activity that matters clinically. This drives assay selection.
2. Select assay format(s). Choose the functional readout closest to MOA that is operationally feasible (timing, throughput, variability). Decide early if you need a matrix.
3. Develop the assay under ICH Q14 thinking: define the analytical target profile, the readout, the dose-response model, and the controls.
4. Establish the reference standard and a bridging plan (the qualification and bridging steps are set out in detail below). Without a stable reference, relative potency is impossible.
5. Qualify, then validate per ICH Q2(R2): accuracy, precision, specificity, linearity, range. For early clinical use, a qualified assay may suffice; a validated assay is expected by the BLA. See bla-readiness-data-package.
6. Set system suitability and sample acceptance criteria based on assay performance and historical data.
7. Set the specification for relative potency, justified by clinical and manufacturing experience, and tightened over the lifecycle.
8. Establish a trending and assay monitoring program so you detect reference standard drift and assay degradation early. Link to continued-process-verification-cpv and out-of-trend-investigations.
9. Build the comparability hook: the potency assay you validate is the one you will rely on for future process changes, so its discriminating power is a comparability asset.

Reference standard qualification and bridging, step by step

The reference standard is the yardstick every relative potency result is measured against, so its own qualification deserves the same rigor as the assay it supports. Treat it in two tiers: a primary reference standard that changes rarely, and a working standard used for routine testing that is periodically re-linked to the primary. Losing track of either one, or replacing one without a documented bridge, quietly invalidates every potency result generated afterward, because nothing in the raw data itself shows that the yardstick moved.

Qualifying a primary reference standard

1. Select representative material. Use a well-characterized lot, ideally a clinically relevant lot with a known safety and efficacy history, or, where one exists, bridge to an international or compendial standard.
2. Characterize beyond routine release. Run the full analytical characterization panel, not just the potency assay, so the primary standard’s identity, purity, and structural attributes are documented in depth, not just its assigned potency.
3. Assign the potency value. By convention the primary standard is usually defined as 100% potency against itself; where a prior standard or an international standard exists, assign its potency through a bridging comparison (see below) rather than by fiat.
4. Establish its own stability program. The primary standard needs a storage condition, a retest or expiry interval, and stability data, the same as a product lot. See stability-programs-ich.
5. Document and approve the qualification report. Record the source, characterization data, assigned value and its basis, storage conditions, and the retest interval, and route it through the same approval chain as a validation protocol.

Establishing a working standard against the primary

1. Select and prepare working standard material, typically a larger, more replaceable pool than the primary, made under conditions representative of routine manufacture.
2. Run a linkage study: test the candidate working standard against the primary reference standard across multiple independent runs (different days, and where feasible different analysts), not a single comparison.
3. Assign the working standard’s potency as the mean relative potency observed against the primary across the linkage runs, with the variability of that linkage documented.
4. Set acceptance criteria for the linkage before running it: an acceptable range for the mean relative potency and a limit on the run-to-run variability (for example expressed as %GCV), pre-specified, not chosen after seeing the data.
5. Put the working standard on its own retest and stability schedule, and use it, not the primary, for routine lot release, to preserve the primary standard’s limited supply.

Two-tier bridging protocol outline (replacing a standard)

Every reference standard eventually runs out, expires, or needs replacing for cause. Bridge the replacement before switching:

1. Trigger the bridging study ahead of depletion or expiry, never after the old standard has already run out, so there is no gap where testing has no valid reference.
2. Test the old and new standard side by side across multiple independent runs, using the same design (days, analysts, plates) as the original working-to-primary linkage.
3. Apply the same pre-specified acceptance criteria used for the original linkage: an acceptable mean relative potency range and a run-to-run variability limit.
4. Assign the new standard’s potency from the bridging data (either against the prior working standard, or against the primary if the primary itself is being replaced), and document the basis.
5. Adjudicate and document the decision. If the bridging study meets criteria, qualify the new standard, set its effective date, and archive the old standard’s data and remaining inventory per the retention policy. If it does not meet criteria, investigate before making any switch, because failing to bridge cleanly can force a re-basis of historical trending and even the specification itself.
6. Assess the trending impact. Confirm whether historical potency trend charts and control limits need a documented shift or annotation at the changeover, so a step change from the reference standard alone is never mistaken for a process or product change.

Worked numeric example: bridging an old and new reference standard

Reading: six independent runs is enough to characterize both the average shift and the run-to-run noise of the bridge, not just a single comparison that could be an outlier. The mean sits close to 100% and the variability is well inside the pre-set limit, so the bridge is clean, there is no step change to explain in the historical trend, and the new standard can be adopted on its effective date without re-basing the specification. Had the mean landed outside the acceptance range, for example at 70%, the switch would trigger an investigation into whether the old standard had drifted, the new standard was mischaracterized, or both, before any release testing relied on the new material.

Acceptance criteria: what good looks like

• The assay reads on a relevant biological activity, not just binding or quantity, unless you can prove the surrogate equals function.
• System suitability (parallelism, curve fit, control wells) gates every run; failing runs are invalid, not just out of spec.
• Validated per ICH Q2(R2) for commercial; intermediate precision quantified and controlled.
• A reference standard exists with a qualification and replacement (bridging) protocol.
• The specification is set on clinical relevance and is defensible, typically a relative potency range.
• The assay is stability-indicating: it detects loss of activity on degraded or stressed material.
• For CGT: a potency assurance strategy is documented, showing how the matrix of measures plus process controls assures potency given assay and product constraints.

Roles and responsibilities for potency assays

As in Part 1, QA owns the gate on the assay and the reference standard record. Analytical Development and the reference standard custodian own the science. Regulatory owns the filing consequence of the specification and the assurance strategy.

Common mistakes and recurring inspection findings

These patterns appear in regulatory observations, deficiency letters, and complete response letters for biologics and CGT, described generically:

• Potency assay not measuring a relevant biological activity. Relying on a binding assay or a content/quantity measure as the sole potency test when the mechanism involves downstream function. For CGT this is the single most common potency deficiency at BLA stage.
• No quantitative potency assay at licensure. Carrying a qualitative or matrix-only approach too long without a validated quantitative functional assay measuring relative potency.
• Reference standard problems. No two-tier standard, no bridging protocol, reference standard drift uncontrolled, or a new reference qualified without demonstrating equivalence to the prior one. Reference standard mismanagement quietly corrupts every relative potency result.
• Parallelism handled wrong. Reporting potency from non-parallel curves, or using a difference test for parallelism that an imprecise assay passes by default.
• Comparability acceptance criteria set after seeing the data. Post hoc ranges, or “the result met specification” used to declare comparability when the specification is far wider than the historical experience.
• Methods not discriminating. Using assays in a comparability exercise that cannot detect a meaningful difference, so everything passes and the conclusion is hollow.
• No forced degradation or side-by-side stability comparison, so a difference in degradation pathway is missed even though release data look identical.
• Statistical inadequacy. Too few lots for the equivalence claim made, or an equivalence margin chosen for convenience rather than clinical relevance.
• Change control disconnected from comparability. A process change implemented without a comparability assessment, or the comparability conclusion not feeding the correct regulatory reporting category. See change-control-validated-systems and deviation-management.
• Data integrity gaps in the supporting data. Missing audit trails, unjustified reintegration of chromatograms, or selective use of passing runs. See chromatography-data-system-integrity, audit-trail-design-and-review, and data-integrity-in-gene-therapy.
• Aggregate or impurity shifts in-spec but out of historical range dismissed without characterization or clinical-relevance assessment.

Interview-ready questions and strong answers

Q: What is the difference between comparability and biosimilarity? A: Comparability (ICH Q5E) is one manufacturer comparing its own product before and after its own change, with full process history and clinical data access. Biosimilarity is a different sponsor matching an originator they did not make, with no access to the originator’s process, under a separate regulatory pathway. The toolbox overlaps; the evidentiary burden differs.

Q: A post-change lot meets specification but falls outside the historical range. Is the product comparable? A: Not automatically. Meeting specification is necessary but not sufficient. Comparability is assessed against the quality range derived from historical experience, not just the registered limit. An in-spec but out-of-range result triggers investigation, characterization of any new species, and a clinical-relevance assessment before you can conclude.

Q: Why can’t I use a binding assay as my potency assay? A: Potency must reflect the relevant biological activity tied to mechanism of action. A binding assay measures an upstream interaction and may not reflect downstream function. It is acceptable only if you demonstrate the binding equals the functional activity. Otherwise it is a characterization assay, not a potency assay.

Q: What is relative potency and why does parallelism matter? A: Relative potency is the ratio of the sample’s activity to a reference standard run on the same plate, usually from a 4PL or parallel-line fit. The ratio is only valid if the two dose-response curves have the same shape, which is parallelism. Non-parallel curves mean the ratio changes with dose, so the potency number is meaningless. Parallelism is tested by an equivalence approach, not a difference test.

Q: How do you handle potency for an autologous cell therapy that expires before a cell-based assay can read out? A: You use a potency assurance strategy: release on fast, qualified measures (viability, identity, transgene expression, vector copy number) that correlate with function, while the slower functional assay runs in parallel as part of the matrix and characterization. Over development you build the correlation data and move toward a validated quantitative functional assay, supported by process controls. The current FDA draft guidance on potency assurance frames exactly this.

Q: What triggers a comparability exercise, and what determines its size? A: Any manufacturing change to a licensed or clinical biologic. The size is risk-based: the impact and detectability of the change on CQAs drives how many attributes, methods, and lots you need, and whether quality data alone suffice or nonclinical/clinical bridging is required.

Q: Which regulations and guidances govern this area? A: ICH Q5E for comparability; ICH Q6B for biotech specifications including biological activity; ICH Q5C for stability; ICH Q2(R2) and Q14 for analytical validation and development; 21 CFR 600.3(s) and 610.10 for the US potency requirement; FDA’s 2011 potency guidance and the more recent potency assurance draft for CGT; 21 CFR 601.12 for post-approval change reporting categories.

Q: How do comparability and potency relate? A: Comparability needs discriminating methods, and the potency assay is often the most clinically meaningful one. A weak potency assay makes every comparability conclusion weak, because you cannot show function was preserved across the change. Investing in a strong, validated potency assay pays off every time you change the process.

Practical tips

• Write the comparability protocol and lock the acceptance criteria before you run a single comparison sample. Pre-specification is your strongest defense.
• Always test pre-change retains side by side with post-change material on the same plate, same analyst, same reference, where the samples allow it. It removes between-run noise that can mask or fake a difference.
• Treat the reference standard as a product. Two-tier it, qualify it, and bridge every replacement with documented equivalence. Reference drift is silent and corrosive.
• Use an equivalence test for parallelism and for comparability claims, with margins justified on clinical relevance, not on what the assay can pass.
• For CGT, start building the potency matrix early and collect the correlation data between fast markers and the functional readout from the first clinical lots, because you will need it at BLA.
• Keep your quality ranges separate from your specifications in the report. Reviewers want to see both, and the gap between them is where comparability findings live.
• Run forced degradation as part of comparability, not just release stability. Divergent degradation pathways are the differences release data miss.

Related articles

• process-validation-for-biologics
• atmp-gmp-cell-gene-manufacturing
• bioassay-and-elisa-validation
• stability-programs-ich
• ich-q14-q2r2-analytical-lifecycle
• quality-in-technology-transfer
• data-integrity-in-gene-therapy
• bla-readiness-data-package
• analytical-method-transfer
• statistics-in-quality-cpk-control-charts