CAR-T Transduction Efficiency by Flow Cytometry: Which Control Sets the Gate

The run sheet says 41% CAR-positive. The process development scientist wants to know whether that number is the dose, and the QA reviewer wants to know which control set the gate. Those are the same question asked twice, and the answer depends almost entirely on what you ran alongside the test article. This walks the gating chain for a transduction-efficiency measurement, names the control that belongs at each step, and works the arithmetic that turns a percentage into the two numbers a batch record actually carries.

What a CAR-T transduction efficiency number is used for

Transduction efficiency leaves the process record and ends up in two places that carry regulatory weight. FDA’s January 2024 guidance Considerations for the Development of Chimeric Antigen Receptor (CAR) T Cell Products recommends that dose be based on the number of viable transduced CAR T cells rather than total cell number. The reason is lot-to-lot variation: at a fixed total-cell dose, a swing in transduction efficiency delivers a variable active dose. The same guidance ties the percentage into the vector copy number calculation, recommending VCN be reported per CAR-positive cell rather than per total cell — using total cells puts CAR-negative cells in the denominator and lowers the reported integration rate.

So the percentage feeds both the dose and a safety attribute, in opposite directions. A gate that moves a couple of points raises one and lowers the other, and the worked example at Step 4 puts numbers on how far.

Step 1 — Build the parent population before you look at the CAR channel

The CAR-positive frequency is a fraction of something, and half the disagreements about transduction efficiency are actually disagreements about the denominator. The conventional chain is time → scatter → singlets → viability → T cells → CAR.

• Time gate on the acquisition record. A clog, a bubble, or a pressure change mid-tube shifts fluorescence for a stretch of events, and those events then land on whichever side of the CAR boundary the shift pushed them. It costs nothing to add to the template and it is the only check here that looks at acquisition stability rather than at the cells.
• Debris exclusion on FSC/SSC.
• Singlets, on FSC-A versus FSC-H. A transduced cell stuck to an untransduced one is one event that is CAR-positive and two cells that are not both CAR-positive. The mechanics are covered in FSC-A versus FSC-H doublet discrimination.
• Viability, with a fixable dye if the sample will be fixed. Dead cells bind antibody non-specifically and land in the positive gate, which inflates the number in the one direction nobody checks. See choosing between DAPI, PI and fixable viability dyes.
• T-cell identity — CD45 then CD3, with CD4/CD8 if the product specification calls for subset composition.

Report the parent gate with the number. “41% CAR-positive” and “41% of viable CD3+ singlets” are different claims, and only the second one is checkable.

Step 2 — Pick a detection reagent, then justify it

How you detect the CAR determines how much background you are fighting. The 2024 guidance recommends direct detection of the CAR to determine the percentage of CAR-positive cells. Where detection is instead by a surrogate (a co-expressed marker) or a broad-specificity reagent such as protein L, it asks you to evaluate the correlation with CAR expression and to justify the surrogate’s sensitivity and specificity.

There is measured data behind that preference. Schanda et al. compared four CD19.CAR detection reagents and reported median background staining on CAR-negative material of 0.02% for the CD19 CAR detection reagent (range 0.007–0.06%), against 0.80% for F(ab’)₂ anti-human IgG (0.47–1.58%), 0.73% for protein L (0.44–1.23%) and 0.65% for FITC-labelled CD19 protein (0.25–1.35%) (Schanda et al., Cells, 2021).

Step 3 — The control that sets the gate is the untransduced cells from the same donor

This is the step the title of this post is about, and it is the one most often substituted. Three candidates get used, and they are not interchangeable.

The NTD control and the FMO answer different questions, and a panel above about six colours usually needs both: the NTD tells you where unmodified cells sit, the FMO tells you how far the other fluorochromes push events into the CAR detector. That second effect scales with panel size, which is the argument developed in when an FMO control is actually required.

On gate placement itself, Schanda et al. set the positive boundary from the histogram where CAR-positive and CAR-negative formed two distinguishable peaks. That is a resolved-peak placement. If the two peaks do not resolve, you have learned something about the reagent or the construct, and the response is to fix the staining; picking a boundary in the valley that produces a plausible number just moves the problem into the batch record.

Step 4 — Subtract the background, then convert to the numbers the batch record wants

Background-corrected transduction efficiency:

Now watch the denominator, because this is where the number usually goes wrong. The 33.3% is a percentage of viable CD3+ singlets, so it must be multiplied by the viable CD3+ count — not by the total viable count, which includes the 5% that are not T cells:

Multiplying 33.3% by the total viable count instead would have given 4.00 × 10⁸ — a 5% overstatement of the dose quantity, arrived at by pairing a percentage with the wrong denominator. That is the same mismatch the parent-gate rule at Step 1 exists to prevent, and it survives into the batch record because both numbers look reasonable.

VCN per CAR-positive cell, from bulk VCN and the same percentage:

Notice what the third line does. The bulk figure of 0.75 copies per cell looks comfortably low; expressed per CAR-positive cell, where the integrations actually are, it is 2.25. Both numbers are arithmetically correct and they support different conclusions about insertional-mutagenesis risk, which is exactly why the guidance specifies the denominator.

The same arithmetic shows why the gate boundary matters. Suppose the gate is placed loosely enough that the test article reads 37.0% instead of 34.2%. Subtracting the same 0.9% background gives a corrected 36.1%, so the dose count rises to 4.12 × 10⁸ and VCN per CAR-positive cell falls to 2.08. One gate decision, 2.8 percentage points wide, moved a dose figure up by 8% and a safety attribute down by 8%, with no new measurement behind either.

One caveat on that sensitivity run, since it is the kind of thing a reviewer will ask about: a looser gate also admits more events in the NTD control, so holding the background at 0.9% while the test article moves is a simplification. Re-read both from the same gate, every time.

Before you trust a high number: check when it was measured

Surface CAR signal soon after transduction is not all integrated vector. Working on a 24-hour manufacturing process using non-activated T cells, Ghassemi et al. saw CAR-marker expression persist even when integrase or reverse-transcriptase inhibitors were present — pseudotransduction, where vector-associated material carried in with the prep produces detectable surface signal with no genomic integration behind it (Ghassemi et al., Nature Biomedical Engineering, 2022).

Take the scope with the finding: that experiment was in non-activated cells on a rapid process, and a conventional bead-activated, expanded product is a different system. What transfers is the mechanism and the diagnostic. An early flow percentage can read high against a PCR measurement of the same material, and cells carrying marker without integration have no antigen-specific activity, so the flow number is the optimistic one.

Two practical consequences. Measure CAR expression once the signal has had time to resolve, well after the earliest point a number is available; and read a flow-versus-VCN discrepancy as a timing question first, before you go looking for an assay fault.

How many events the number is built on

A transduction-efficiency measurement on a well-transduced lot is a common-population measurement and the counting statistics are forgiving. They stop being forgiving on a low-transduction lot or a residual-CAR-cell measurement in a patient sample, where the population of interest may be a fraction of a percent and the confidence interval on the frequency is set by how many positive events you actually acquired. That arithmetic — and the event counts it implies — is worked through in how many events a rare population needs before its frequency means anything.

What this does not cover

Everything above is assay design for a transduction-efficiency readout. A release assay in a regulated setting carries requirements this post does not address. FDA’s guidance calls for a comprehensive validation study conducted per ICH Q2 for lot-release flow cytometry, including validation of each labelled marker on the specific cytometer used for release. It also calls for robustness work defining maximum sample holding times before staining and between staining and acquisition. Antibody titration is a development-stage recommendation there too — a reagent used at the vendor’s default dilution has not been optimised for your matrix, and the resulting separation affects exactly the gate boundary this post is about. Titration method and the stain index arithmetic are covered in antibody titration and the stain index.

Before the panel is built, the constraint worth checking first is whether the CAR detection channel has a clean detector available given the identity markers already in the panel. The fluorophore spectrum viewer shows what each dye emits into each detector for a given instrument configuration, which is where a CAR channel carrying an unnecessary spillover burden gets caught — at panel design, while it still costs nothing to move.