Insights

TSA Multiplex Immunofluorescence QC Checklist

A TSA multiplex immunofluorescence QC checklist: separate bleed-through from carry-over, autofluorescence and tissue damage, and gate every round.

TerryDR

A TSA multiplex immunofluorescence panel produces readable images long before it produces trustworthy ones, and the difference is almost always a quality control step that was skipped. Four different problems make an image look wrong — spectral bleed-through, carry-over from the previous round, tissue autofluorescence, and tissue damage — and each one needs a different control to diagnose and a different remedy. This checklist works through those four separately, then gives you a gate to run before every round, the control set that supports it, and the part most guides skip entirely: which downstream measurements a given failure invalidates, and whether you can still report them.

One note before starting, because searching for “TSA” in this field surfaces the wrong science. Here TSA means tyramide signal amplification, in which a fluorescent tyramide is deposited locally where an enzyme sits. It has nothing to do with trichostatin A, the histone deacetylase inhibitor, which shares the abbreviation and dominates unrelated search results.

Sequential tyramide signal amplification staining workflow on a tissue section with concentric fluorescent rings across multiple channels

Four different problems look like signal overlap

The most expensive mistake in TSA quality control is treating a merged image as the evidence. A merged image is the one view in which four separate failures are indistinguishable, because blending hides exactly the differences you need to see. A dim channel that seems to reproduce the shapes of a bright neighbouring channel looks identical whether the cause is optical spillover, a surviving detection system from the previous round, genuine co-expression, or elevated autofluorescence.

The way out is to stop asking what the picture looks like and start asking which control would prove each cause. Optical bleed-through is confirmed by comparing the single-stain controls against the merged image, and ruled out by comparing against the channel itself. Carry-over is confirmed by the control that omits the primary antibody while keeping the detection enzyme and the tyramide, so that anything appearing in an earlier channel is unambiguously residue. Autofluorescence is what remains in tissue with no antibody at all. Tissue damage is diagnosed from morphology rather than intensity, which is why it is usually the last thing checked and the most expensive one to have missed.

One dim fluorescence channel ghosting the shapes of a brighter neighbouring channel across stained tissue

What you observe Optical bleed-through Round carry-over Autofluorescence Tissue damage
A dim channel reproduces the shapes of a bright neighbouring channel Check the single-stain controls Not the cause Not the cause Not the cause
The previous round’s outline lights up in the current round Not the cause Check the omit-primary carry-over control Not the cause Not the cause
One channel looks hazy across the whole section Check exposure and dynamic range Not the cause Check unstained and no-primary controls Possible if structure is lost
A weak target fades round over round Not the cause Not the cause Not the cause Check antigen damage and tissue retention
The merged image looks correct but the channels disagree Check the single-stain controls Check the carry-over control Check the unstained controls Check registration and tissue retention
The result changes when you move to another instrument Check the spectral library and exposure Not the cause Not the cause Not the cause

Read the table as a decision order rather than a lookup. The sequence that works is staining side first, because carry-over and tissue damage are the causes you can still act on, then the optical side, then autofluorescence last because it is often the most expensive to fix and the least often the actual culprit. Two causes can be present at once, which is exactly why each column names the control that confirms it rather than a symptom you can eyeball.

A quality gate before every round

Sequential and cyclic staining share a property that makes quality control structural rather than occasional: every round inherits the previous one. A panel that runs four rounds is not four independent stains, it is one stain with four opportunities to contaminate the next round. That is why the useful unit is the round, and why the gate belongs before each round rather than at the end of the experiment.

Three gates cover most of what matters. The pre-round gate asks whether the previous round left anything behind, because that is the last moment the answer is cheap. The post-round gate asks whether this round behaved, before it is buried under the next one. The pre-scan gate asks whether the image is acquirable at all, which is a different question from whether the staining worked, and which is usually the only gate that gets checked because it is the one that stops the run.

Gate When it runs What you check Acceptance criterion If it fails Can you continue
Pre-round Before starting round N Previous round’s carry-over control, previous channel shape and intensity, and termination of the detection enzyme Carry-over control shows no signal in the new round Rework the removal step or repeat the round No
Post-round Immediately after round N Target localisation, background level, and tissue integrity Signal sits where the biology predicts, with no edge loss Rework the round before proceeding No
Pre-scan Before acquisition Channel saturation, dynamic range, and background ceiling No saturated pixels in the target channel, and the weakest target is still detectable Adjust exposure or re-stain Only with documented justification

The acceptance criteria matter more than the checks themselves. “Check the background” is a task; “the carry-over control shows no signal in the new round” is a criterion, and a criterion can be failed deliberately rather than argued about. Note also the final column, which is where most protocols go quiet. Continuing after a failed gate converts a recoverable, cheap failure into an unrecoverable one, because every subsequent round compounds whatever went wrong.

Cyclic imaging makes this more demanding rather than less, because each round’s signal has to be removed, inactivated, or isolated before the next, and because the rounds only become one dataset if they can be registered back into a common coordinate frame. TerryDR’s TR-ASTRA Flow is one current implementation of that closed loop, and any platform you evaluate should be able to state what its per-round acceptance criteria are, not only how many channels it can run.

Four sequential quality control gates shown as glowing panels checking tissue signal before each staining round

Antibody removal and fluorescence removal are different checks

This distinction is where sequential staining quality control most often goes wrong, and it is worth stating plainly because the two are routinely conflated.

Removing the antibody and its detection complex is one operation. Removing, inactivating, or otherwise isolating the fluorescence that was already deposited is a different one. A protocol can succeed completely at the first and still fail at the second, and the visible result is the confusing case: the previous round’s signal is still glowing while the antibody is demonstrably gone. Conversely, a scheme that keeps the deposited fluorophore and removes only the detection complex needs a completely different verification, because now the question is whether the detection system survived rather than whether the dye did.

The practical consequence is that you cannot design one control and call it evidence for both. Antibody removal is verified with a control that omits the primary antibody but keeps the detection system. Fluorescence removal is verified by comparing the current round against the image of the previous round for the same tissue, looking for residue in the earlier channel. These are different observations, on different channels, requiring different controls, and a QC checklist that lists them as one item has silently halved the number of things it actually verifies.

The control set and what each control actually proves

Most protocols list control types. Far fewer state what each control is capable of proving, which channel you should read, and what you do when it fails. That last part is what turns a control list into a decision, because a control that fails at minute forty of a long run is only useful if you know at minute forty what it means.

Control What it proves Which channel to read Acceptance criterion If it fails
Unstained The native background and autofluorescence of the tissue itself Every channel Establishes the baseline for all other comparisons Background may be too high for this panel; reconsider the tissue or the chemistry
Single-stain The spectral profile of each fluorophore on this instrument The stained channel only Matches the expected spectral library entry Recalculate the spillover matrix
No-primary Non-specific binding contributed by the detection system Every channel No signal above the tissue baseline Background is reagent-driven; rework the detection chemistry
Carry-over control (round N omits the primary, keeps enzyme and tyramide) Whether the previous round’s detection system survives into round N Every channel except N-1 No signal in the earlier channels Rework removal and termination before continuing
Known-positive That the reagents and the detection system are working The positive channel Expected positivity at the expected location The failure is upstream of the panel
Process control What each round of staining does on its own All channels Records the cumulative effect round by round Use it to locate where signal is being lost

The carry-over control is the one specific to this chemistry, and it is the one most often absent. Its logic is simple to state and awkward to run: in round N, omit the primary antibody for the marker you want to test for residue, but keep everything else, including the enzyme and the tyramide. Anything that then appears is residue, and it belongs to the previous round. Without that control, a positive result in an earlier channel is genuinely ambiguous between carry-over and biology, and no amount of downstream analysis can resolve it.

A note on scope: some of the checks above live at the microscope rather than the bench. Saturation, exposure consistency, and spillover mathematics are handled better in dedicated imaging guides than in a staining protocol, and it would be dishonest to pretend otherwise. The split is worth stating plainly, because it decides what you look at first. Check the staining side before the imaging side, since a stain problem masquerades as an imaging problem far more often than the reverse.

What a quality failure invalidates downstream

Here is the layer that most troubleshooting guides stop before reaching, and the one that determines whether a failed run is an inconvenience or a lost dataset. The question is not “can I still use the images” but “which of my planned measurements are no longer supported”.

Quality failure Downstream metric affected Can it still be reported Salvageable by Must be redone
Carry-over in one channel Co-expression rate and phenotype percentage Only with a documented threshold on the contaminated channel Restricting the phenotype to unambiguous cells The affected round
Background above the usable ceiling Cell density and every positive call No Re-quenching or re-staining The affected channel
Saturated target channel Signal intensity and any intensity-based metric No Re-acquisition with different settings Acquisition only
Tissue loss at the edge Distance metrics and edge-corrected density Only inside the retained region Restricting the analysis area Nothing, if the region is already lost
A weak target fading across rounds Any metric involving that marker No Substituting a validated alternative marker The whole panel run
Missing no-primary control Every claim about non-specific background No Nothing retrospectively The affected sample
Registration error between rounds Every spatial metric Only for coarse, region-level description Restricting to regions with confident correspondence The affected round pair

The third column is the one that changes behaviour. A saturated channel can be recovered by re-acquiring, because the tissue is intact and only the settings were wrong. A faded marker across rounds usually means re-running the panel, because the information was destroyed during staining rather than during imaging. Telling these two apart is worth a great deal more than any single troubleshooting tip, because it decides whether you are correcting a setting or scheduling another run.

Notice also which failures are not recoverable at all. A missing control cannot be reconstructed after the fact, no matter how much the data look plausible, and tissue that has been lost from the slide cannot be recovered by any image analysis. That is the argument for treating the gate as a stopping rule rather than a suggestion, and for deciding in advance which failures force a restart.

Recording the decision is what makes the mapping useful to other people. For each run, note which gate failed, which rows of the table apply, and what was done about it. Over a project this produces something more valuable than a pass or fail: a record of how this particular panel behaves on this particular tissue, which is usually the only thing that distinguishes a repeatable method from a lucky one. It also makes an outside reviewer’s judgement much easier, because the reasoning is visible rather than implied by what survived into the final figure.

There is a second-order point about independence. A team that builds its own panels will, over time, know which of its failure modes are common and which are rare for its own tissue and chemistry, and that knowledge is worth more than any generic troubleshooting chart. It is also the reason that a project run by an outside team is not automatically worse: an outside team may see a failure pattern you have normalised, and the record you send with the data is what lets them see it.

When to check the microscope instead of the stain

Some symptoms are conventionally attributed to staining when they are actually optical. If the same tissue looks different on a different instrument, the question is the spectral library and the exposure settings rather than the slide. If one channel alone looks grainy or washed out while the rest are clean, check saturation and the dynamic range before you re-stain anything. If the background is high even on a section with no primaries at all, the chemistry is a less likely culprit than the tissue or the detection system.

The order that respects this is straightforward: run the staining-side checks first, because they are cheaper to correct and because a stain failure will produce optical symptoms that look exactly like instrument problems. Then run the imaging-side checks. TerryDR’s QuPath plugins cover part of the downstream handling, including registration between rounds and noise reduction, but neither can recover a signal that was never deposited, and neither substitutes for a wet-lab control.

This is also the point where it is worth being explicit about the limit. A digital pathology slide scanner that sets exposure per channel automatically, and a standardised automated IHC stainer, remove two common sources of variation between operators, but they do not make quality control optional. They make the gates cheaper to pass and make failures easier to attribute.

The practical version of the order is a ten-minute sequence. Read the unstained and no-primary controls and compare them channel by channel against the stained slide; then read the single-stain controls against the merged image to see whether any channel is carrying a neighbour’s signal; then look at the tissue morphology itself under transmitted light, because edge loss and tearing are visible before they are quantifiable. Only after those three does it make sense to suspect the instrument. Running that sequence in that order, you will usually have named the cause before you have touched a control, and you will have avoided the expensive error of re-running a stain that was never the problem.

Where a project is handed to a team that runs the whole chain, that record is what carries the reasoning across the handover. TerryDR describes the same split between mIF image analysis service delivery and the QuPath plugins used for self-run analysis, and the underlying principle is unchanged: the gate result belongs with the images, not in someone’s notebook.

Once a panel passes the gates, the slides it produces still have to be turned into cells, phenotypes and spatial measurements that survive review. That is the subject of Multiplex Immunofluorescence Image Analysis Workflow.

Frequently Asked Questions

What causes bleed-through in multiplex immunofluorescence?

Usually one of three separate mechanisms, and telling them apart matters. Optical spillover happens when the fluorophores overlap in the detection channels and is diagnosed with single-stain controls. Carry-over happens when the previous round’s detection system survives into the next round. Genuine co-expression looks identical in a merged image but disappears when you examine the single channels. Checking the single-stain controls first separates the first cause from the other two cheaply.

How many controls do I need for a TSA multiplex panel?

At minimum, unstained tissue, a single-stain reference for each channel, a no-primary control, and a carry-over control for each round after the first. A known-positive sample and a process control make the set considerably more useful. The count matters less than the discipline of reading a specific channel against a specific acceptance criterion for each one, because a control nobody interprets is only a slide.

Does every round of TSA staining need its own quality check?

Yes, and this is the point that most checklists miss. Each round inherits the previous one, so a panel that runs four rounds has four opportunities for residue and four for cumulative tissue damage. Running the gate once at the end tells you that something is wrong but not which round introduced it, and by then the cost of fixing it is the whole run rather than one round.

Can I skip the carry-over control if my antibody removal step works?

You can skip it if something else in your process provides equivalent evidence that the detection system does not survive into the next round. In practice most protocols do not, because antibody removal and fluorescence removal are different operations with different failure modes. The control is inexpensive relative to the cost of a panel whose second round silently contaminates the first.

What should I check first when the merged image looks wrong?

Start with the staining side, in this order: whether the previous round left residue, whether the tissue is still intact, and whether the background is within range. Then move to the optical side and check the single-stain controls against the merged image. Autofluorescence comes last because it is the least common cause and the most expensive to address. A ten-minute check in that order usually identifies the cause without re-staining anything.

Sources

✉ Contact Us