Technology Transfer for Advanced Therapies: How a Transfer Actually Gets Done

A technology transfer moves a manufacturing process, and the analytical methods that control it, from one place of established knowledge to another place that has to execute it with the same result. For a tableted small molecule, process and product are reasonably separable, and a transfer is mostly a logistics and training exercise layered on solid science. For cell therapy, gene therapy, and RNA therapeutics, process and product are much harder to pull apart. A viral vector’s infectivity, an autologous T-cell product’s potency, and an oligonucleotide’s impurity profile are all shaped by parameters a small-molecule transfer would treat as adjustable within a wide operating range. Move the process a little and you may have moved the product.

What follows is how to run that kind of transfer without finding out about the gaps during the validation batch, or worse, during an inspection.

Two Transfers, Not One

Two different situations both get called “tech transfer,” and they carry different risk profiles.

New Product Introduction (NPI) transfer moves a process into a site that has never run it before, often from development or clinical-scale manufacturing into a CDMO or a newly commissioned commercial suite. There is no historical performance at the receiving site to compare against, and the receiving team learns the process and the product at the same time.

Site-to-site transfer moves an established, characterized process between two sites that both already know how to manufacture something, just not this something. The receiving site brings manufacturing maturity but still has to absorb product-specific knowledge. The sending site’s tribal knowledge, the operating detail that never made it into the batch record, is usually the actual content of the transfer.

Both situations get harder for advanced therapies for the reason above, which is why a comparability assessment under ICH Q5E, Comparability of Biotechnological/Biological Products Subject to Changes in Their Manufacturing Process, belongs in the plan from the start, not as an afterthought triggered by a deviation.

The Document Spine

A transfer that runs on tribal knowledge and good intentions eventually produces a batch nobody can defend. The document spine is what makes the transfer reconstructable, for your own use and for an inspector’s.

Technology transfer plan. States what is transferring (process, method, or both), sending and receiving sites, roles and responsibilities, the acceptance criteria for calling the transfer complete, and the schedule. Written and approved before any process step moves. A plan written after the engineering run has already happened is a narrative, not a plan.

Gap assessment. A side-by-side comparison of sending-site and receiving-site facilities, equipment, utilities, personnel qualification, and quality systems against what the process requires. This is where a stainless-steel sending process meeting a single-use receiving suite gets caught before it becomes a deviation during the engineering run. Do it early enough that its findings can still change the equipment order, not just the risk assessment. A real gap assessment row states the factor, the sending-site value, the receiving-site value, and a disposition someone can act on: closed-system processing, sending site uses a functionally closed bioreactor with manual sterile connections, receiving site runs a fully closed, automated tubing-welded system; disposition, more closed than the sending process so no new contamination risk, but re-verify weld integrity test frequency against the new consumable set before the first engineering run. “Equipment differs” is not a disposition; what someone is going to do about the difference is.

Risk assessment. Takes the gap assessment’s findings and the process’s known critical process parameters and critical quality attributes, and scores what could go wrong in this specific sending-to-receiving move. ICH Q9(R1), Quality Risk Management, is the framework; the output should point directly at what the engineering run needs to challenge and what the process control strategy needs to hold tighter.

Master production instructions at the receiving site. Not a copy of the sending site’s batch record with the site name changed. Equipment, room, and material identifiers change, and steps manual at a development-scale sending site sometimes become automated at a receiving commercial site. That difference has to be qualified, not just documented.

Process control strategy. The living description of which parameters are monitored, which are controlled, and why, with ranges tied to the criticality assessment. ICH Q10 Section 3.2.3 treats technology transfer as a lifecycle stage this document has to survive, and ICH Q12’s tools for managing post-approval changes both assume it stays current after the transfer, not filed away once process qualification passes.

Technology transfer report. Closes the loop: what was compared, what met acceptance criteria, what didn’t and how it was resolved, and whether the transfer is complete or complete with follow-up commitments. If a release decision depends on this transfer, quality wants the report, not the raw data set.

Skip any one of these and the gap shows up later as a deviation investigation reconstructing, after the fact, a decision that should have been documented before the batch ran.

Roles and Responsibilities

A transfer with no named owner for each document in the spine defaults to whoever is available that week, which is how a risk assessment ends up written by someone who never read the gap assessment it should be built on. The table below is a starting allocation; adjust it to the organization, but keep the principle that every document has exactly one accountable owner and quality retains sign-off across all of them.

Phase-Appropriate Rigor

The FDA’s 2011 Process Validation guidance frames manufacturing in three stages: Process Design, Process Qualification, and Continued Process Verification. A transfer inherits whichever stage the process is in, and the rigor has to match. ICH Q9(R1)‘s formality principle, that the depth and documentation burden of a risk-based decision should scale with the risk and the state of process knowledge rather than apply uniformly, is the underlying justification for treating these as different jobs with the same shape, not one procedure applied at one intensity regardless of stage.

Running a clinical-stage transfer with commercial-stage documentation overhead slows the program without buying real risk reduction. Running a commercial transfer with clinical-stage looseness is how a site-to-site move ends up generating a 483.

Decision Tree: Scoping Rigor by Stage

Receiving-Site Platform Experience: A Second Risk Axis Besides Stage

The phase table above scopes documentation rigor against one axis: where the process sits in its own lifecycle, design, qualification, or commercial. A second axis runs independently of the first and gets missed more often than it should: whether the receiving site has ever actually run this modality before, regardless of what stage the process itself is in.

A commercial-stage transfer into a site with six other sponsors’ CAR-T programs already running carries a different risk profile than a commercial-stage transfer into a site that has run biologics fill-finish for a decade but is handling its first replication-incompetent lentiviral vector. The phase table, read on its own, would score both transfers identically: full ICH Q7 and ICH Q10 discipline, judged against sending-site CPV trends. That is the right call for the documentation burden, but it says nothing about whether the receiving site’s people, procedures, and containment infrastructure have ever actually been tested against this modality’s specific failure modes, which is exactly what a platform-experienced site has already worked through on someone else’s program and a platform-novice site has not worked through at all.

“Biologics experienced” is not the same claim as “experienced with this modality.” A site that qualifies as broadly biologics-capable, trained personnel, a working quality system, a track record of commercial batches, can still be encountering, for the first time, a shear-sensitive autologous cell population, a biosafety containment requirement for a replicating or replication-incompetent viral vector, or a microfluidic LNP mixing step, none of which its existing biologics experience prepared it to run without new procedures, new training, and a new round of the failure modes this article covers elsewhere getting discovered live rather than anticipated.

Assess platform experience directly in the gap assessment and risk assessment, as its own named factor, not folded into a general facility-fit score. A useful assessment asks, for the specific modality moving: how many batches of this modality, not biologics generally, has the site manufactured; how many personnel are trained and independently qualified on this modality’s specific unit operations, not just generically qualified on the site’s biologics SOPs; does the site’s quality system already have modality-specific procedures proven out through its own deviation and CAPA history, or is a procedure being written for the first time to support this transfer; and does the site carry the containment, segregation, or biosafety infrastructure the modality requires already commissioned and qualified, or is that infrastructure part of what this transfer is standing up.

An invented illustration. A commercial-stage RNAi therapeutic transfers into a CDMO with ten years of sterile fill-finish experience and a strong CPV program on other products, but no prior LNP mixing step on its floor. The phase table correctly calls for full ICH Q7 and ICH Q10 discipline, judged against sending-site CPV trends, the same as any commercial transfer. A risk assessment that stops there treats the LNP mixing step like any other unit operation the site has run before. The platform-experience gap, no prior microfluidic mixing run of any kind at this site, means the engineering run has to do double duty: prove the specific process transferred correctly, and prove the site’s mixing platform generally, including operator proficiency on hardware nobody there has touched in production before. Scoping the engineering run and the person-in-plant presence to the stage alone, without separately scoring the platform-novelty risk, is how a site passes its documentation review and then struggles operationally through its first several real batches.

The receiving site’s own quality and manufacturing leadership self-report platform history as a matter of course during a technical assessment; the gap is treating that self-report as sufficient. The sponsor’s transfer lead and, for a contract site, person-in-plant presence verify it independently: ask for the actual batch count and deviation history on this specific modality, not the site’s marketing description of its capabilities, before scoring the risk assessment’s platform-experience factor.

The Second Transfer: Moving a Commercial Process Between Contract Sites

Everything above assumes, even in the site-to-site case, that the process is moving from the site that generated the original process knowledge. A second transfer, a commercial process moving from its current manufacturing site, often a CDMO that has run it for years, to a different site, another CDMO or an insourcing facility, breaks that assumption and carries a risk profile the phase table and the standard two-transfer framing both under-describe.

The process is commercial and, by the phase table’s logic, should carry the lightest relative burden of the three rows: it is reasonably locked, methods are validated, and Continued Process Verification data exists. That data existing is exactly the problem a second transfer has to solve for, because the CPV trends, the accumulated change history, and the tribal knowledge that actually describe how the process runs today all live at the current manufacturing site, not at the sponsor, and not at the original sending site from years earlier. A transfer plan that treats a second transfer as a repeat of the first one, same document spine, same gap assessment against the original process description, is working from a baseline that has already drifted.

Knowledge erosion since the original transfer. A commercial process accumulates minor, individually approved changes over its life: a filter supplier substitution, a hold-time adjustment, a reagent lot qualification, each one dispositioned and closed under the current site’s own change control. Each change was almost certainly communicated to the sponsor as a routine notification at the time, as the quality agreement’s change-notification clause requires, but a routine notification reviewed and filed is not the same as that change being reflected in the sponsor’s own master technical file, the document a second transfer actually needs as its starting point. A sponsor whose technical oversight during the commercial period consisted of reviewing and filing notifications, rather than actively folding each one into an internally maintained process description, arrives at a second transfer holding a technical file that describes the process as it was transferred in, not as it is actually run today.

The cooperation problem. A first transfer usually runs with both sides motivated toward a shared goal: the receiving site wants the business, the sending site wants the program to succeed. A second transfer triggered by a terminated or ending commercial relationship removes that alignment on one side. The departing site has reduced commercial incentive to invest senior technical staff time in a smooth handoff, and the people who carry the process’s undocumented operating detail, the actual tribal knowledge, are often the first to be reassigned once the relationship’s end is known. The sponsor’s own technical team, not either contract site’s goodwill, has to be the party that actually owns enough process knowledge to drive the transfer, which only works if that ownership was built during the commercial period rather than assumed to be available on demand once the relationship ends.

The baseline shifts from characterization data to CPV trends. A comparability assessment for a second transfer should be scored against the departing site’s actual Continued Process Verification trends, the real distribution of results the process has produced commercially, not against the original process characterization ranges from the first transfer. A result that would have looked unremarkable against wide characterization-stage ranges can sit outside a control limit the CPV program has since tightened from years of real data; see CPV for autologous, patient-specific manufacturing for how that trend data gets separated from patient-driven variability in the first place, the same trend data a second transfer’s comparability work depends on being trustworthy.

Contractual access to data and know-how. The right to use the departing site’s batch records, deviation history, and process know-how in a second transfer is a contract question, not a quality-system question, and it is far easier to negotiate while the commercial relationship is still active than after a notice of termination has been served. A quality agreement amendment or a technology transfer agreement that specifies data and know-how rights while the relationship is still cooperative protects a transfer that will otherwise happen under worse conditions; see the CDMO technical assessment’s treatment of the technical agreement alongside the quality agreement for how that documented responsibility gets built in the first place, the same structure a second transfer needs to already have in place before it needs it.

The filing sequence runs in both directions at once. A second transfer usually requires its own regulatory action, adding the new site to the marketing application and, eventually, removing the departing site, and the two actions do not have to happen on the same filing or the same timeline. Plan a defined overlap or bridging period where both sites could in principle supply product, rather than assuming the new site’s approval and the old site’s exit can be scheduled as a single clean handoff; a supply interruption is more expensive than running two approved sites briefly at once.

An invented illustration. Acme Therapeutics moves ATX-207, its commercial lentiviral-vector CAR-T product, from CDMO A, the site that has manufactured it commercially for four years, to CDMO B, for capacity reasons rather than any quality concern. The sponsor’s technical file, built from the original transfer five years earlier, describes a 0.45 micron pre-filtration step; CDMO A’s own change control history shows that step was requalified to a 0.22 micron filter two years into commercial manufacturing, approved under CDMO A’s internal change control and communicated to the sponsor at the time as a routine notification. The notification was filed, not folded into the sponsor’s working technical file. The gap surfaces when CDMO B’s engineering team, working from the sponsor’s file, orders the 0.45 micron filter originally specified and only catches the mismatch during a routine cross-check against CDMO A’s actual current batch record, requested as part of the transfer’s reconciliation step. Caught during that reconciliation, the fix is a corrected equipment order and a one-line update to the gap assessment. Caught instead during the receiving site’s first engineering run, the cost would have included a failed run on a scarce vector lot and a reopened question over which filter specification the comparability package should even be tested against.

A checklist for the handoff items a second, contract-site-to-contract-site transfer needs beyond the standard document spine is at CDMO-to-CDMO transfer handoff checklist. Whether that second transfer runs on a short timeline or a comfortable one often traces back to a contracting decision made years earlier, at the original CDMO’s signature, not at the moment the move becomes necessary; CDMO exit readiness, dual-sourcing, and IP and data ownership assessment scores whether that groundwork was actually laid.

Scale-Up and Scale-Down: Translating Critical Process Parameters Across Equipment

A transfer that changes scale, a 1 or 2 liter benchtop bioreactor feeding a 50 or 200 liter single-use production vessel, a lab-scale chromatography column feeding a process-scale column, almost always gets described in the transfer plan as “the same process, bigger,” and that description is the trap. Two vessels can share the same nominal recipe, the same agitation speed written as revolutions per minute, the same flow rate written in liters per minute, and still subject the product to a physically different environment, because the quantities that actually drive mixing, mass transfer, and shear do not scale linearly with vessel volume. A gap assessment built around equipment lists, and a risk assessment that names “scale” as a single generic risk factor, both miss the point unless something in the transfer plan identifies which physical parameter the process actually depends on and re-derives its value on the receiving site’s own equipment.

Geometric similarity, the same height-to-diameter ratio and impeller type, scaled proportionally, is necessary but not sufficient for dynamic similarity, the actual physical behavior the cells, vector, or product experience. Holding one engineering parameter constant across scales usually means at least one other parameter changes, often by a large factor, and a transfer plan has to decide, with a stated rationale, which parameter the process cannot afford to let drift and which parameters are allowed to move.

The parameters that actually matter, and what each one captures. Power per unit volume (P/V), the mechanical energy the impeller delivers per liter of broth, is the most common default basis for scaling a stirred-tank bioreactor, because it correlates reasonably well with overall mixing intensity and, for many mammalian and insect cell systems, with shear exposure at a bulk level. Impeller tip speed, the linear velocity of the impeller blade tip, tracks localized shear at the impeller itself more directly than P/V does, and is usually the parameter of record for a shear-sensitive cell population, an activated T cell culture, a stem cell expansion, or a cell line newly transduced with a viral vector, where damage at the blade tip can matter even when the bulk P/V looks acceptable. Volumetric mass transfer coefficient (kLa) describes how quickly oxygen moves from a sparged gas bubble into the liquid phase, and it does not scale the same way P/V or tip speed does: a scale-up that holds P/V constant, the common default, can still under-supply or over-supply oxygen at the larger scale, because kLa depends on bubble size, sparger design, and vessel geometry in ways a P/V match does not automatically preserve. Mixing time, how long it takes an addition, a feed bolus, a base addition for pH control, to fully homogenize across the vessel, lengthens at larger scale under most practical scale-up approaches even when P/V is held constant, because a larger vessel has a longer bulk circulation path; a process sensitive to a local pH or nutrient spike at the addition point can see an effect at production scale that bench scale, with its much shorter mixing time, never revealed. Reynolds number signals whether the fluid regime is turbulent or laminar at the relevant geometry, most often a consideration in single-use mixing bags, tangential flow filtration manifolds, and tubing runs rather than the stirred-tank vessel itself, where a geometry change at the receiving site can shift the flow regime even when the nominal flow rate is unchanged. Linear velocity and bed height, not column volume, are the governing parameters for a chromatography scale-up; holding linear velocity and bed height constant while the column diameter grows is the standard approach, because resolution and dynamic binding capacity track those two parameters, not the absolute flow rate or total volume processed. Membrane loading, volume processed per square meter of membrane, and transmembrane pressure or crossflow rate per unit area, are the governing parameters for tangential flow filtration scale-up, for the same reason: a larger membrane module processing a proportionally larger volume at the same loading and per-area crossflow conditions behaves like the qualified small-scale system, while matching absolute flow rate alone does not.

No single rule picks the right parameter for every process step; the right parameter is whichever one sits closest to the actual mechanism that sets the critical quality attribute at risk, and a transfer plan that names a parameter without stating why it is the governing one has usually copied the choice from a template rather than derived it from the process.

Which Parameter to Hold Constant, by Process Step

Worked Example: A Dissolved Oxygen Sag Caught During the Engineering Run

An invented illustration. A CAR-T manufacturing process is transferred from a 2 liter benchtop bioreactor, where process characterization was run, to a 50 liter single-use production bioreactor at the receiving site. The scale-up basis stated in the risk assessment holds power per unit volume constant between the two vessels, the common default, and the receiving site’s engineering team confirms the agitation setpoint that reproduces the same P/V on the larger, differently proportioned vessel.

During the engineering run, dissolved oxygen in the 50 liter vessel sags below the process’s established lower control limit during the peak growth phase, a point the 2 liter characterization runs never approached. Investigation finds the root cause in kLa, not agitation: the larger vessel’s sparger design and bubble size distribution deliver proportionally less oxygen transfer at the matched P/V than the bench-scale vessel did, a mismatch the P/V-based scale-up basis never checked because it assumed matching P/V implied matching oxygen transfer by extension.

The fix was a sparging-rate adjustment, confirmed against a repeat dissolved oxygen trace showing the parameter held comfortably inside range through peak growth, not a change to agitation speed, which would have reopened the P/V basis the mixing-sensitive part of the process actually depended on. Caught in the engineering run, the cost was a short delay and a sparging-rate change written into the receiving site’s master production instructions. Caught instead during a PPQ batch, the cost would have been a batch disposition question over a cell therapy’s actual patient-specific material, the same asymmetry the earlier titer-gap example describes for an assay rather than an engineering parameter.

Qualifying the Scale-Down Model Itself

A transfer rarely challenges the commercial process directly for every question it needs answered; raw material variability studies, filter and resin lifetime studies, and failure-mode or edge-of-failure characterization almost always run on a small-scale model instead, because running them at commercial scale would consume product, time, and equipment capacity the program does not have. That small-scale model is only useful if it is itself qualified as representative: a side-by-side comparison of critical quality attribute data, not just matched critical process parameters, between the small-scale model and the actual commercial-scale process, documented before the model is trusted to stand in for commercial scale in any study that feeds a filing. ICH Q11’s expectation that a scale-down model supporting process characterization be shown to represent the commercial process, not assumed to, is the standard a reviewer checks this against, and FDA’s 2011 process validation guidance treats PPQ run on a properly qualified scale-down model, with a documented justification, as an accepted alternative to running PPQ at full commercial scale, the same qualification burden PPQ readiness and prerequisites assumes is already settled before a batch count or sampling plan gets decided.

Scale-down model qualification is itself a transfer deliverable whenever the receiving site plans to run characterization, matrix, or bracketing studies on anything other than full commercial-scale equipment, and it has to be refreshed whenever the commercial-scale equipment or operating range changes, not qualified once and treated as permanent. Mixing equipment selection and scale-down qualification for a formulation step specifically, rather than an upstream bioreactor step, is covered in formulation development patterns for advanced therapies; the physics are the same, the unit operation is different.

A worksheet for identifying the governing engineering parameter at each process step, recording the sending-scale and receiving-scale calculated values, and documenting the scale-down model qualification basis is at bioreactor and unit operation scale translation worksheet.

Engineering Runs and After-Action Discipline

An engineering run, sometimes called a demonstration run, is a batch made at the receiving site under close scrutiny, before anything is called a validation or process performance qualification batch. Its purpose is to fail safely. Define entry criteria before it starts: facility qualified, equipment IQ/OQ complete, personnel trained on the receiving site’s version of the procedure, materials released. Define exit criteria too: what data gets reviewed, which deviations can close out versus escalate, and who signs off that the run supports moving to the next stage.

The discipline that gets skipped under schedule pressure is the after-action review. Hold it within days of the run, while operators still remember what felt different from the batch record. A good after-action review captures three things separately: what happened that the documentation predicted correctly, what happened that surprised the team, and what the team would do differently next time. Each surprise gets a disposition: it changes the master production instructions, changes the risk assessment, or gets an explicit, documented decision that no change is needed.

A working after-action review agenda runs in a fixed order, every time: a step-by-step walk of the batch record against what was actually planned; every deviation opened during the run, with an immediate call on whether root cause is already known or needs a formal investigation; operator and analyst observations that never made it into the record, the moments that felt different from the procedure but didn’t trigger a deviation; a look at any in-process or release result trending toward a limit, even one that technically passed; and a closing round where each attendee names one thing to change before the next run. Assign an owner and a due date to every action before the meeting ends, not in a follow-up email.

The trap is treating the engineering run as a rehearsal that doesn’t count, then discovering during the actual validation batch that the receiving site’s operators are still learning the process live. If the engineering run’s job was to surface the unknowns, use it. A validation batch is not the place to find your first unknown. For campaigns at a contract site, person-in-plant presence through the engineering run and early validation batches catches these surprises while they are still cheap to fix; see person-in-plant oversight.

Worked Example: A Titer Gap Caught Before It Became a Deviation

Acme Therapeutics is transferring ATX-207, a lentiviral-vector CAR-T process, from its clinical manufacturing site to a CDMO for the pivotal program. The risk assessment flagged viral vector titer as a change driver because the receiving site’s qPCR-based titer assay runs on a different instrument platform than the sending site’s, even though both sites call it “the same method” in the transfer plan.

The engineering run produced a titer result roughly 30 percent below the sending site’s historical range, on a split sample of the same vector lot tested at both labs. Because reviewing any result trending away from the sending-site range is a fixed after-action review item, not something raised only if someone happens to notice, the gap surfaced immediately. Root cause: the receiving lab’s qPCR standard curve used a different reference material lot than the one qualified at the sending site, a detail that lived in an analyst’s working notes, not the transferred method document.

Caught in the engineering run, the fix was a bridging study comparing both reference material lots and, once the discrepancy was confirmed to sit in the reference material rather than the vector itself, a documented equivalency rather than a full assay revalidation. Cost: one bridging study, about two weeks. The same gap found after a pivotal batch had already shipped for release testing would have meant a batch on hold, an OOS investigation, and a comparability question over every prior pivotal batch’s titer result. Luck did not catch this. A standing agenda item did.

Capturing Tribal Knowledge Before It Walks Out the Door

“Tribal knowledge” gets named earlier above as the actual content of a site-to-site transfer, but naming the problem is not the same as having a method for it, and most transfer plans stop at naming it. A concrete capture method turns a vague intention into work someone can actually do and someone else can actually check.

Structured knowledge-capture interviews, run by the receiving site’s process owner against the sending site’s operators and analysts before the engineering run, not during it, surface the working detail a batch record was never written to hold: which valve someone always opens a quarter turn past the written setting because the gauge reads low, which reagent lot’s viscosity runs thick enough that the pipetting technique has to change, which alarm the team has learned to treat as a nuisance versus a real signal. A written interview guide, covering setup, in-process handling, common deviations and how they were actually resolved, and anything the operator would tell a new hire that isn’t in the SOP, turns this from a casual conversation into a document with a name and a date that the risk assessment and the master production instructions can both cite.

Video documentation of manual, hands-on steps, aseptic connections, visual in-process checks, anything where the written procedure compresses a skilled motion into a single sentence, captures what a paragraph of text cannot. It is not a replacement for training, but it is a durable reference the receiving site’s trainers can return to months after the sending site’s expert has moved on to the next program, which matters specifically because advanced-therapy programs run lean and a single departure can otherwise take a critical piece of undocumented know-how with it.

A short secondment or shadowing period, receiving-site operators physically present at the sending site for one or more real batches before the engineering run, or sending-site operators present at the receiving site through the engineering run and first qualification batches, catches the moments a written interview or a video misses entirely: the ambient judgment calls a skilled operator makes without narrating them, because the action feels too routine to mention out loud. Person-in-plant presence through this period, covered in person-in-plant oversight, serves the same function from the sponsor’s side when the receiving site is a CDMO.

Whatever methods a transfer uses, the discipline that makes them worth the effort is a tribal-knowledge capture log: a running list of every non-written detail surfaced through interviews, video review, or shadowing, each entered with a disposition, folded into the master production instructions, folded into the risk assessment, or explicitly accepted as an undocumented judgment call that training alone will carry forward. A capture effort that produces a pile of interview notes nobody dispositions is only marginally better than not running the interviews at all; the log, not the raw notes, is what an inspector or a future transfer team can actually use.

Where Advanced Therapies Break the Standard Approach

The approach above is generic process validation and transfer discipline. Several things about cell, gene, and RNA therapeutics stress it in specific, predictable ways.

Potency assays rarely transfer like a compendial method. A cell-killing bioassay or a functional potency assay carries biological variability a chromatographic purity method does not. Analytical method transfer for these assays needs a wider acceptance window built on real inter-site, inter-analyst precision data, not borrowed from a chemistry method’s tighter expectations. USP General Chapter <1224>, Transfer of Analytical Procedures, still governs the transfer mechanics, comparative testing, co-validation, or a transfer waiver where scientifically justified, even though the underlying assay behaves differently than most of that chapter’s examples assume.

Chain of identity has no retest. For an autologous cell therapy, the material moving through the process belongs to one patient, with no bank of extra starting material to rerun if identity or traceability breaks. The risk assessment has to weight identity and chain-of-custody failure modes far more heavily than a multi-lot or small-molecule transfer would, because a mixed-up identity cannot be corrected by retesting a retain; see chain of identity versus chain of custody, and the redundant-identifier controls that make the distinction real for how that control gets built at every physical handoff, a discipline a transfer’s receiving site inherits along with the process itself. The same one-patient-per-batch reality reaches past the transfer itself into commercial monitoring: see CPV for autologous, patient-specific manufacturing for how a Stage 3 program has to separate patient-driven variability from process-driven variability once the receiving site is running the process on its own.

Viral vector titer and infectivity assays are transfer-fragile. Small changes in cell substrate lot, assay timing, or detection method can move a titer result meaningfully, which makes bridging studies between sending-site and receiving-site results a required part of the transfer, not an optional cross-check, exactly the mechanism the worked example above shows.

Shelf life compresses the whole schedule. A short-dated intermediate or drug product limits the time between manufacture and release testing at the receiving site’s qualified lab. A transfer plan that does not explicitly map logistics and testing turnaround against shelf life will discover the conflict during the first real batch, not during planning.

Comparability carries more formal weight. ICH Q5E was written for biologics because process and product are entangled. Every advanced-therapy transfer should ask, on paper, what comparability package the move requires, analytical alone or analytical plus nonclinical or clinical bridging, and get alignment on that scope before generating data that doesn’t answer the question.

A lipid nanoparticle drug product carries a transfer risk the modality table below treats separately from a naked or conjugated oligonucleotide, because the manufacturing step that defines an LNP, mixing an ionizable lipid, cholesterol, a helper lipid, and a PEGylated lipid with the nucleic acid cargo across an aqueous-organic interface, is itself the critical process step, not a downstream formulation add-on. Two sites running “the same” microfluidic mixing process on different mixer geometries, a T-junction device at one site and a staggered herringbone or multi-inlet vortex mixer at another, are not automatically running an equivalent process even if the chemistry, the lipid ratios, and the nominal flow rates all match on paper. Particle size, polydispersity index, and encapsulation efficiency are all sensitive to the actual mixing physics, not just the recipe, so a transfer plan that copies flow rate ratio and total flow rate numbers from the sending site’s mixer onto the receiving site’s different mixer geometry without re-deriving them is making an assumption the underlying fluid dynamics do not support. A defensible LNP transfer treats mixing hardware equivalency, or the lack of it, as its own named risk in the risk assessment, re-characterizes flow rate ratio and total flow rate on the receiving site’s actual hardware rather than inheriting the sending site’s numbers, and confirms particle size, polydispersity index, and encapsulation efficiency against acceptance criteria before calling the receiving site’s process equivalent, not after the first engineering run surprises everyone. Mixing equipment selection for shear-sensitive formulations generally, including the bench-to-commercial scale-down qualification this equivalency work depends on, is covered in formulation development patterns for advanced therapies; a technology transfer inherits that qualification burden a second time whenever the receiving site’s hardware is not a scaled version of the same mixer geometry.

Platform knowledge reuse is an emerging path worth tracking for genome-editing gene therapies specifically. FDA issued a draft guidance on June 3, 2026, “Leveraging Prior Knowledge in the Development of Human Gene Therapy Products Incorporating Genome Editing,” describing how a sponsor developing a product that incorporates ex vivo or in vivo genome editing of somatic cells may cite established platform knowledge, including CMC data, nonclinical findings, and clinical information from a related product, to support a new program rather than regenerating the same evidence from scratch. On the CMC side specifically, the draft names five categories of data eligible for this kind of reuse: analytical methods and method qualification or validation data, lot release specifications, stability data, comparability data, and process characterization and process validation data, which line up closely with the document spine a technology transfer already has to produce. The draft still requires a sponsor to provide a scientific rationale demonstrating that the reused data actually applies to the specific product at hand; it is not a blanket waiver. For a technology transfer moving a genome-editing platform process into a new site, this matters directly: a comparability package scoped under ICH Q5E can, once this guidance finalizes, potentially draw on platform-level prior knowledge to narrow what the transfer’s own analytical bridging needs to generate fresh, provided the rationale for applicability is documented as explicitly as the guidance expects. The guidance’s comment period closed September 1, 2026, and it remains a draft, not yet final, as of this writing, so treat it as a direction to plan around rather than a settled rule to cite as binding.

ICH has also started building ATMP-specific comparability guidance as a formal Annex to Q5E rather than a freestanding document. An Expert Working Group concept paper describing the plan, fit-for-purpose comparability approaches for cell therapy, viral vector and nucleic acid-based gene therapy, genetically modified somatic cells, stem-cell-derived products, and gene-editing products, was endorsed by the ICH Management Committee on June 2, 2026, following the ICH Assembly’s endorsement of the concept paper outline in May 2025. The Annex targets exactly the comparability pressure this section has been describing: limited product availability, patient-specific variability in starting material, and analytical methods that cannot always characterize an ATMP the way a conventional biologic gets characterized, pushing toward science- and risk-based comparability judgments that weigh the totality of available evidence together rather than one fixed battery of tests applied identically regardless of modality. This is early in the ICH process; an Expert Working Group has only just been chartered, and Step 4 adoption of a final Annex is not targeted until early 2028, so nothing about how to scope a comparability package under the base ICH Q5E guideline changes today. It belongs on the same watch list as the genome-editing prior-knowledge draft above: a transfer plan written now should expect the comparability section of its governing guidance to get more ATMP-specific over the life of the program, not less.

Modality-Specific Transfer Risk at a Glance

Choosing the Analytical Transfer Mechanism

USP General Chapter <1224> names three mechanisms for moving an analytical method between labs, and a transfer plan that just cites the chapter number without stating which mechanism applies, and why, has not actually made the decision the chapter asks for.

Potency and other biological assays, discussed above as needing a wider acceptance window than a chemistry method, almost always transfer through comparative testing or co-validation, never a waiver, precisely because their inherent biological variability is exactly the thing a waiver skips measuring.

Worked Example: Setting a Comparative Testing Acceptance Criterion

The numbers below are an invented illustration of the method, not a reported result from any real transfer. A potency bioassay reports percent relative potency against a reference standard, with an established method precision, expressed as percent relative standard deviation, of 12 percent RSD at the sending site, built from historical intermediate-precision data. Comparative testing runs six lots, split and tested at both sites, with the transfer’s predefined acceptance criterion set before testing begins: the mean absolute difference between sending-site and receiving-site results, across the six lots, must fall within two times the method’s own established intermediate-precision RSD, a criterion that ties the pass/fail bar to what the method itself is known to be capable of, rather than to an arbitrary round number chosen for the transfer.

Testing produces a mean absolute difference of 9 percent across the six lots, comfortably inside the 24 percent criterion, so the transfer passes on the predefined basis. Contrast that with a transfer that instead adopted a flat, borrowed criterion of “within 10 percent” without checking it against the method’s own historical precision: a method this variable would fail a 10 percent criterion on ordinary assay noise alone, generating a false transfer failure and an unnecessary investigation into a method that was never actually broken, exactly the trap this article warns about above when it says a potency assay’s acceptance window has to come from real inter-site precision data, not a chemistry-method habit.

Raw Materials, Single-Use Systems, and Automation: The Equivalency Traps Outside the Process Itself

A transfer risk assessment built around unit operations, equipment, and scale, the structure this article has built so far, catches most of what actually breaks a transfer. Three quieter categories generate a disproportionate share of the deviations that show up after PPQ has already started, precisely because they look like paperwork rather than process risk until something fails: raw materials, single-use assemblies, and the automation or control system code that actually runs the process.

Raw materials. A critical raw material, a cytokine, a growth factor, a plasmid backbone, a lipid used in an LNP formulation, carries more variability between suppliers, and sometimes between lots from the same supplier, than a transfer plan built around “same specification, same CAS number” assumes. Two suppliers’ material can meet the same nominal specification and still differ on an attribute the specification never tested for, an impurity profile, an endotoxin baseline, a potency-relevant activity level, because the specification was written to a compendial or internal standard that does not capture everything a sensitive cell or gene therapy process actually responds to. The defensible practice is a side-by-side comparison of the certificate of analysis for the sending site’s qualified material against the receiving site’s actual sourced material, attribute by attribute, not specification limit against specification limit alone, and where the two cannot be sourced identically, a bridging study demonstrating the process performs the same with the receiving site’s actual material before it is treated as equivalent.

Single-use systems. A tubing set, a bag, or a filter assembly ordered under “the same part number” at a receiving site is not automatically the same physical article the sending site qualified, because a distributor, an assembly site, or a resin or film lot change upstream of the part number can shift the extractables and leachables profile even when the printed specification is unchanged. This is the same principle filter validation for biologics and oligonucleotides builds for filter qualification specifically, extended to every single-use, product-contact component a transfer carries: confirm the actual assembly drawing, film or resin lot family, and extractables data behind the receiving site’s sourced component match what the sending site’s qualification assumed, rather than trusting a part number alone to mean the components are interchangeable.

Automation and control systems. Moving a process to a new site usually means moving its control logic too, and that move happens one of two ways: a direct export and import of the sending site’s recipe or program, or a manual re-creation of the same logic on the receiving site’s control platform when the two sites run different automation vendors or software versions. The second path is where transfer-fragile errors live. A control engineer manually re-keying an interlock, a hold time, an alarm setpoint, a batch-step sequence, can transpose a unit, minutes for hours, or a decimal place, and the error will not show up in a functional test that simply confirms the batch ran to completion, because a batch can complete successfully while an interlock silently never triggers or an alarm threshold sits an order of magnitude away from where it should. The defensible verification is an independent, line-by-line comparison of the receiving site’s control logic against the sending site’s, performed by someone who did not write the migrated code, followed by a dry-run or factory acceptance test that deliberately exercises every interlock and alarm condition rather than only the nominal batch path, with the predicted and actual behavior compared and documented before the engineering run relies on the automation working correctly.

Worked Example: An Interlock Transposition Caught Before the Engineering Run

An invented illustration. During a transfer of an autologous cell therapy process to a CDMO running a different automation platform than the sending site, the receiving site’s controls engineer manually re-creates the sending site’s batch recipe rather than importing it directly, because the two platforms use incompatible recipe file formats. One interlock in the sending site’s recipe holds a wash step for a minimum of 90 minutes before the next step can start; in the re-created recipe, the engineer enters the value into a field labeled in hours rather than minutes, setting the interlock to 90 hours instead of 90 minutes.

Because the transfer plan requires an independent logic review before the dry run, a second automation engineer who did not build the recipe compares it line by line against the sending site’s exported configuration and flags the discrepancy before the dry run is even scheduled. Caught at this stage, the fix is a single corrected field entry and a re-review. Had the transfer instead relied only on a functional dry run using representative timing overrides rather than exercising every real interlock value, or skipped the independent review entirely and gone straight to the engineering run, the 90-hour interlock would likely have surfaced as an inexplicable process hang during the first real batch, consuming a scarce, possibly patient-specific, starting material lot while the team traced a control system error the review step was built to catch for free.

A working risk assessment for these three equivalency traps is at single-use system and raw material equivalency assessment, and a checklist for verifying an automation recipe migration before it reaches an engineering run is at automation and control system migration verification checklist.

What “Done” Actually Looks Like

A transfer is not done when the last batch record is signed. It is done when the receiving site can run and control the process on its own, which means all of the following hold, with evidence behind each:

• Process performance at the receiving site meets predefined acceptance criteria against the product’s established critical quality attributes, not just “the batch was released.”
• Analytical methods for release and characterization are validated, verified, or formally transferred, with the transfer protocol executed and closed, not just planned.
• Receiving-site operators and analysts have demonstrated independent performance; sending-site personnel are no longer standing behind them for every run.
• Deviations from the engineering and qualification runs have been trended, not just individually closed, with nothing in the trend pointing at an unresolved systemic issue.
• Any CAPA opened during the transfer is closed with verified effectiveness, not closed on the promise of a fix.
• The process control strategy and master production instructions reflect the receiving site’s actual equipment, reviewed and approved through change control.
• Quality has formally accepted the transfer report, and regulatory has confirmed whether a filing is required and, if so, in motion.
• The quality agreement reflects the receiving site’s actual responsibilities, batch release authority, in-process testing ownership, deviation and change control roles, amended and co-signed, not still describing a role the sending site performed or a role the receiving site has not yet actually assumed.

Anything short of that list is a transfer technically executed but not complete, and it will surface later as a finding, a deviation, or a supply interruption, usually at a worse time than now.

A transfer report can satisfy every item above and still leave the quality agreement, the document FDA’s 2016 guidance, Contract Manufacturing Arrangements for Drugs: Quality Agreements, expects a sponsor and a contract facility to maintain as the controlling statement of who does what, describing a relationship that no longer matches reality. The transfer report closes the technical record; the quality agreement, amended and co-signed by both sites, is what closes the accountability record, and a reviewer checking whether a CDMO transfer is really complete checks both. See the CDMO technical assessment’s quality and technical agreement structure for how that document gets built and kept current once the receiving site is running the process on its own; a transfer plan for a move into or between contract sites should name the quality agreement amendment as its own tracked deliverable, not an assumed side effect of the transfer report’s approval.

Common Failure Modes

What Gets Probed in an Inspection or Audit

An inspector or auditor reviewing a completed transfer works through a short list of pointed questions. Can you produce the gap assessment and show its findings were closed, not just logged? Does the risk assessment’s critical parameter list match what the process control strategy monitors, or did the two documents drift apart? Can you trace a specific engineering-run deviation to the change it drove in the master production instructions? Does the signed transfer report’s conclusion match the batch disposition history? For advanced therapies specifically, expect direct questions about how potency assay comparability was demonstrated and how chain of identity was verified end to end, the two areas these transfers most often cannot answer cleanly.

The honest test of a transfer program: hand the document spine, gap assessment through transfer report, to someone who was not in the room, and see if they can reconstruct what was compared, what passed, what didn’t, and why the transfer was accepted anyway or wasn’t. If that reconstruction takes a meeting instead of a file review, the documentation didn’t do its job.

Interview Q&A

Walk me through how you would scope a comparability package for a gene therapy technology transfer. Start with the stage: is this transfer feeding early clinical material, a pivotal batch, or a commercial supplement? That answer sets the rigor described in the phase table above. Then apply ICH Q5E: what actually changed, site, scale, or equipment, what quality attributes could plausibly be affected, and whether analytical comparability alone answers the question or nonclinical or clinical bridging is also needed. For a viral vector program, name titer and infectivity as the assays most likely to need a bridging study, and say why: instrument and reagent lot differences move these results in ways a chromatographic purity method rarely does.

What is the difference between a gap assessment and a risk assessment, and why does a transfer need both? The gap assessment is a factual, side-by-side comparison, facility, equipment, personnel, quality systems, against what the process requires, and it produces findings. The risk assessment takes those findings plus the process’s critical parameters and scores what could actually go wrong from this specific move. Skip the gap assessment and the risk assessment scores against assumptions instead of verified differences. Skip the risk assessment and the findings sit in a table with no link to what the engineering run needs to challenge.

Tell me about a transfer where something surprised you. What did you do? A strong answer names a specific technical surprise, an assay drift, a hold-time gap, a titer discrepancy, states how it was caught, ideally a standing after-action review item rather than luck, walks through the root cause investigation, and states the disposition: a bridging study, an MPI change, or a documented no-change decision. The earlier titer-gap example is a template for the shape of that answer: what looked fine on paper, what the data actually showed, and what closed the gap before it became a deviation.

How do you decide how many PPQ batches a transfer needs to support? Tie the answer to process knowledge, not a fixed number: characterization data, the number of raw material lots and donor sources represented, and any flexibility applicable to the modality. The rationale, not the count, is what a reviewer and an interviewer are both actually asking for; see PPQ readiness and prerequisites for how that rationale gets built before the first batch, and validation master planning for where the transfer sits inside the broader validation program it feeds.

How would you decide between comparative testing, co-validation, and a transfer waiver for an analytical method? The strong answer starts from the method’s own state: a waiver needs a genuinely strong documented equivalence basis, typically a compendial method run on the same platform by already-qualified analysts, and is never the default. Co-validation fits a method still being validated when the transfer happens. Comparative testing is the default otherwise, and its acceptance criterion should come from the method’s own established precision, not an arbitrary round number borrowed from a different method’s transfer. A weak answer treats all three as roughly interchangeable paperwork choices.

What would make you trust that a receiving site’s LNP process is really equivalent to the sending site’s, not just similar on paper? The strong answer names mixing hardware geometry as the thing that actually determines particle size, PDI, and encapsulation efficiency, not the recipe alone, and states that flow rate ratio and total flow rate need to be re-derived on the receiving site’s own mixer rather than copied from the sending site, with particle size, PDI, and encapsulation efficiency data confirming the result before the process is called equivalent. A weak answer assumes matching lipid ratios and nominal flow rates are sufficient proof.

How do you verify a receiving site’s automation recipe is really equivalent to the sending site’s, not just similar in the HMI? The strong answer names the actual failure mode: a manually re-created recipe can transpose a unit or a setpoint in a way a batch that completes successfully will never reveal, so verification needs an independent line-by-line logic review by someone who did not write the migrated code, plus a dry-run or factory acceptance test that deliberately exercises every interlock and alarm condition, not only the nominal path. A weak answer treats a successful functional test of the normal batch sequence as sufficient proof.

How do you decide which engineering parameter to hold constant when scaling a bioreactor process up or down? The strong answer names the actual mechanism at risk for the specific process step, bulk mixing and shear for power per unit volume, localized shear for tip speed, oxygen supply for kLa, local concentration gradients for mixing time, rather than defaulting to power per volume because it is the most commonly cited basis. It states that holding one parameter constant usually means others move, and that the transfer has to check the parameters it did not choose as the basis with real data, not assume they moved proportionally. A weak answer treats “match power per volume” as a universal rule applying the same way to every process step and modality.

How is a second transfer, moving a commercial process from one CDMO to another, different from the original transfer into that CDMO? The strong answer names the baseline problem directly: the sponsor’s technical file from the original transfer has almost certainly drifted from the process as the departing site actually runs it today, through years of individually approved changes that were filed as notifications but never folded back into the sponsor’s own working description. It states that comparability has to be scored against the departing site’s CPV trends, not the original characterization ranges, and that contractual access to data and know-how is easiest to secure while the relationship is still cooperative, not after a termination notice. A weak answer treats the second transfer as a repeat of the first with a new site name.

The receiving site for a transfer has years of biologics manufacturing experience but has never run this specific modality before. Does that change your risk assessment? The strong answer separates platform experience from general site maturity: a site can be fully GMP-mature and still be encountering a modality’s specific failure modes, a biosafety containment requirement, a shear-sensitive cell population, a microfluidic mixing step, for the first time, and that gap does not show up in a risk assessment scored only against process stage. It states that platform experience should be scored as its own named factor, verified against the site’s actual batch count and deviation history on this specific modality, with engineering run scope and person-in-plant intensity widened accordingly. A weak answer takes the site’s general biologics track record as sufficient reassurance.

Monday morning, if you are mid-transfer right now: pull the current risk assessment and check whether its critical parameter list still matches what the process control strategy is monitoring. That drift is the single most common gap this approach catches, and it is cheap to fix today and expensive to explain during an inspection.