Reviewer wants more controls

“Additional controls are needed” is not one request. It is a family, and its members range from a paragraph of text to six months of bench work. Before planning anything, work out which control is missing — the rest of the review usually says, and the answer is often one you already ran.

This page sets out the five classes of missing control and what each establishes, how to identify which class a comment belongs to, where to look before assuming an experiment is needed, what counts as a specificity control under the five pillars of antibody validation, why a positive control is non-negotiable for a negative result, what to do when the control a reviewer named is unavailable, and the sentences reviewers write when the control set does not convince them.

The classes of missing control

Five classes cover almost every version of this comment, and the work they require ranges from an afternoon at a desk to half a year at the bench. Sort the request into one of them before deciding anything.

A control you ran and did not report. The most common case. Vehicle-only conditions, untreated baselines, secondary-antibody-only panels, no-template amplification controls — all routinely performed and routinely omitted from the manuscript because they were unremarkable. The fix is a supplementary figure and a methods sentence.

ARRIVE 2.0, the 21-item animal-research reporting guideline published by the NC3Rs in PLOS Biology in 2020, treats that silence as non-compliance rather than as a neutral absence. Sub-item 1a asks for “The groups being compared, including control groups. If no control group has been used, the rationale should be stated.” A reviewer reading a manuscript with no vehicle arm on the page cannot distinguish a vehicle arm that was flat from a vehicle arm that never existed, and the guideline instructs them not to try. The ARRIVE guidelines sets out the items one at a time.

Check your raw data before assuming an experiment is needed. Often the control exists already and only the plot is missing.

A specificity control. The reviewer accepts your reagent works but not that it works only on your target. Second independent siRNA, second inhibitor with a different scaffold, a rescue, a knockout validation of the antibody. This is the class most often genuinely missing, and it is usually the one worth doing.

A condition control. Something that changed alongside the variable of interest: a solvent, a temperature, a handling difference, a time in culture. Often answerable from existing data, sometimes with a small new experiment. The most frequent instances in submitted work are an unmatched DMSO concentration between treated and untreated wells, a passage-number difference between arms, treatment and control plated on different days, and treated and control samples processed in separate runs — the last of which is a batch effect rather than a biological result, and is detectable in the existing data by testing whether run or plate predicts the outcome.

A positive control. Evidence that the assay can detect the effect it reported as absent. Essential whenever you claim no difference — a negative result without a positive control is uninterpretable, and reviewers are right to insist.

A biological control. A second cell line, a second model, a second cohort. The most expensive class, and the one where narrowing the claim is most often the better answer. A finding in one immortalised line is a finding in that line; the claim that survives without a second model is the mechanistic one, not the general one. Reviewer says the model system is not appropriate covers the version of this objection aimed at the system itself rather than at the number of systems.

Working out which class the reviewer means

The class is usually named elsewhere in the same review, in a comment the author reads as separate. Read the whole review before planning, and match the control request to the claim it is aimed at.

Three tells do most of the sorting. First, the verb: “please state”, “please clarify” and “it is unclear whether” point at a control that exists and was not reported; “does not exclude”, “is required to establish” and “cannot be distinguished from” point at one that does not exist. Second, the figure the comment lands on: a comment naming a specific panel is nearly always a reporting request, while a comment naming a claim in the abstract is a design request. Third, the neighbours: a request for a second siRNA in comment 4 and a remark about off-target effects in comment 7 are one objection, and answering them separately produces two weak answers instead of one strong one.

When the comment names no class at all — “the study would benefit from additional controls” — the reviewer is generally repeating a concern stated more precisely in the significant-weaknesses paragraph or in the confidential comments summarised by the editor. Ask the editor which specific control the reviewer had in mind rather than guessing across five classes; editors answer that question, and a wrong guess spends weeks on the wrong bench work.

Checking whether the control already exists

Search the primary records, not the figures you assembled. The figures are a selection; the instrument export is not.

Six places hold controls that never reached the manuscript. Raw plate exports frequently retain no-template control wells that were deleted from the summary table. Flow cytometry .fcs files retain unstained, single-stain and fluorescence-minus-one tubes that were not exported as plots — and fluorescence-minus-one, not an isotype tube, is what a reviewer of a polychromatic panel is asking for; flow cytometry reporting covers the panel-level version. Uncropped blot images usually contain the loading control and the secondary-only lane. Microscopy sessions retain the secondary-only and unstained slides acquired for setting exposure. Sequencing submissions retain spike-ins and negative extraction controls. Animal study records retain the vehicle cohort even when only two arms were graphed.

MIQE, the qPCR reporting guideline published by Bustin and colleagues in Clinical Chemistry (2009;55:611–622), treats these as per-run records rather than per-project ones: no-template controls belong on every plate or batch of samples, and the results of those controls sit among the items its checklist marks as essential to report. The paired no–reverse transcription control is the check for residual genomic DNA in the RNA preparation, and is run against the same samples. If your assay followed MIQE, the control the reviewer wants is already in the run files.

When you recover a control this way, report it as what it is. Give the date or the run identifier, state that it was performed as part of the original experiments, and put it in the supplement with the same figure legend discipline as new data. A recovered control still needs its clone, catalogue number, lot and RRID to be checkable.

What counts as a specificity control

Specificity controls establish that the readout tracks the intended target and nothing else, and the standard for what counts is published rather than a matter of taste. The International Working Group for Antibody Validation set out five conceptual pillars in Uhlén and colleagues, “A proposal for validation of antibodies”, Nature Methods 2016;13(10):823–827: genetic strategies, orthogonal strategies, independent antibody strategies, expression of tagged proteins, and immunocapture followed by mass spectrometry. The paper’s own framing is that these are applied in an application-specific manner — a pillar that validates an antibody for western blot does not validate it for immunohistochemistry.

The same logic transfers to the other reagent classes, and the operative test in each is whether the second piece of evidence fails for a different reason than the first.

RNA interference. Two independent siRNA or shRNA sequences targeting non-overlapping regions, plus a rescue with an RNAi-resistant construct carrying silent mutations in the targeted region. Two sequences with the same seed region are not two controls. The rescue is the pillar that answers off-target effects directly, because an off-target phenotype does not revert when the intended target is restored.

CRISPR. Two independent guides targeting different exons, plus a rescue with a guide-resistant cDNA. Knockout and knockdown of the same gene frequently disagree, and that disagreement is a finding rather than a failed control — see my knockdown and knockout give different results before treating one as the error.

Small molecules. A second compound of a different chemical scaffold, an inactive structural analogue of the same compound, or a drug-resistant target allele. Two inhibitors from the same series share the series’ off-target profile and validate nothing about the target.

Antibodies. Knockout or knockdown validation is the genetic pillar and is the strongest single piece of evidence; a second antibody against a different epitope is the independent-antibody pillar and is weaker. Two antibodies raised against the same immunogen are one antibody for this purpose.

Measurements rather than perturbations. When the doubt is about the readout rather than the reagent, the request is for an orthogonal method, not a control — reviewer wants an orthogonal method sets out which pairings actually break the shared failure mode.

Deciding what to do

For each requested control, ask what it would change. A control that could overturn the central claim should be run, or the claim narrowed. A control that would tidy a secondary figure can be addressed by acknowledging the limitation.

The operational form of that test is to name the sentence the control bears on. Write each requested control on one line and beside it the exact manuscript sentence that becomes unsupportable if the control comes out badly. A control that maps to a sentence in the title or abstract is load-bearing and should be run or the sentence changed. A control that maps to a sentence in a results paragraph describing a secondary figure is a limitation. A control that maps to no sentence at all is usually a reviewer thinking aloud, and it is answered in one line.

Say which is which in the response letter, explicitly. Reviewers accept “we did not perform X because the conclusion does not depend on it, and we have limited the claim accordingly” far more readily than silence about X.

Kaelin made the standard argument for spending the effort where it is load-bearing in “Publish houses of brick, not mansions of straw” (Nature 2017;545:387): a smaller set of claims each supported by adequate controls is worth more than a larger set supported by one experiment apiece. That is also the argument for narrowing. If the control set supports three claims and the manuscript makes five, the revision that removes two claims is stronger than the revision that adds two rushed experiments — and it draws fewer comments in the second round. Check the result against the overclaim check before resubmitting, because a claim narrowed in the discussion and left intact in the abstract draws the same objection again.

If you claimed a negative result

Treat the positive control as non-negotiable. If you reported that a treatment had no effect, the reviewer needs evidence the assay would have detected an effect had one existed. Without it, “no effect” and “assay did not work” are indistinguishable, and no amount of statistical argument separates them.

Altman and Bland made the general form of this point in “Absence of evidence is not evidence of absence” (BMJ 1995;311:485): a non-significant result is not a demonstration that the effect is zero. The positive control is the bench version of that argument, and it is the only version that addresses assay failure, because assay failure is not a statistical property and no statistical procedure detects it.

Three things belong in a manuscript reporting a negative result, and reviewers ask for all three. A positive control run in the same experiment, not a historical one — a known-positive sample, a spike-in at a defined concentration, or a treatment with an established effect in the same system. A confidence interval on the effect estimate rather than a p-value alone, so the reader can see which effect sizes the data exclude. A statement of the smallest effect the experiment could have detected, decided before the experiment where possible rather than reconstructed from the result.

What does not work is computing power from the observed effect after the fact. Post-hoc power is a deterministic function of the p-value and adds no information about whether the assay worked — reviewer asks for post-hoc power covers the response when a reviewer requests it anyway.

If you have no positive control and cannot generate one, the honest revision reports the result as inconclusive rather than negative.

When the control is unavailable

State the constraint plainly and supply a substitute. A named limit with an alternative beside it is accepted far more often than a defence that leaves the manuscript unchanged.

The reagent, line or cohort is gone. The line failed authentication, the antibody lot is discontinued, the cohort is closed, the person who ran the work has left. Say so factually, then supply what remains: banked lysates, archived slides, a deposited dataset containing the comparison, or the same control performed in a related system with the difference named.

The experiment takes longer than the revision window. Say what it would establish, say how long it would take, offer a quicker piece of evidence that addresses the same concern, and ask the editor to weigh proportionality before the deadline rather than after it. Editors rarely require an experiment that exceeds what the claim is worth, and they cannot weigh a request you did not make.

The control exists only in a published paper. Cite it as previously published, in the figure legend as well as the text, and state the conditions under which it was performed. A previously published control is usable evidence and unusable as new data, and the distinction is one editors check.

The reviewer’s premise is wrong. Say so once, factually, with the analysis or the citation that shows it, and offer something alongside — the alternative analysis in supplementary material, or the raw data that lets a reader judge. Reasoned disagreement is normal; an ignored comment is what causes trouble. How to write a response to reviewers sets out the shape, and a worked response letter shows it applied.

In every one of these, the fallback is the same: narrow the claim until the missing control is no longer load-bearing. A claim that does not depend on the control does not need it, and saying which claim you removed is a stronger answer than an apology for the control you could not run.

What reviewers write when the control set does not convince them

Reviewers rarely write “the controls are inadequate” as their whole comment. They write the specific version, and the wording identifies the class.

“No vehicle-only condition is shown; please confirm whether one was performed.” “A single siRNA at one concentration does not exclude off-target effects; a second reagent or a rescue is required.” “The antibody has not been validated in this application; knockout or knockdown material would establish specificity.” “The authors report no effect, but no positive control demonstrates that the assay would have detected one.” “Treated and control samples appear to have been processed in separate batches; the comparison cannot be separated from run-to-run variation.” “The DMSO concentration is not stated and may differ between conditions.” “The conclusion rests on a single cell line.” “The secondary-only control referred to in the methods is not shown.”

Two of those are answered with a sentence, three with a supplementary panel, and three with bench work. Sorting them that way before starting is the difference between a six-week revision and a six-month one.

Wording the response

Group the controls rather than answering them one by one where they overlap, and be specific about what each shows. If you are adding a control you already ran, say when it was performed — presenting a pre-existing control as new is a small dishonesty that occasionally gets noticed and never helps.

A grouped answer states the concern, names each control, says what each establishes, and points at the panel. “Reviewer 1 comments 3 and 6 and Reviewer 2 comment 2 all concern the specificity of the knockdown. We have addressed these together: new Figure 3d shows a second, non-overlapping siRNA reproducing the phenotype, and new Figure 3e shows rescue by an RNAi-resistant construct. The vehicle-only condition in comment 5 was performed as part of the original experiments in March 2026 and is now shown as Supplementary Figure 2.” Every clause in that answer is checkable, which is what makes it short.

Quote the revised manuscript text under each comment so the editor can see the change without opening the file. Answer every comment including the ones you decline, and give the reason for declining rather than the fact of it.

Before submission

In vivo experimental rigour checks the design elements this objection targets in animal work, and reports what each gap prevents a reader from concluding rather than only that it is missing. For antibody and reagent validation specifically, data availability and resource identification covers clones, RRIDs and authentication.

More in this family: reviewer comments on methods and design.

PerfectPaper reads each claim in the title, abstract and discussion against the controls that appear in the figures and methods, and reports a missing control as a reporting gap or a design gap — naming the sentence that would need to change in the first case and the conclusion that cannot be drawn in the second.

Review my manuscript

Frequently asked questions

What does “additional controls are needed” mean in a peer review?

The comment means one of five specific things: a control you ran and did not report, a specificity control, a condition control, a positive control, or a second biological system. Which one is usually stated elsewhere in the same review, and the classes differ widely in the work they require — a supplementary figure at one end, months of new experiments at the other.

What controls do reviewers most often ask for?

Specificity controls — a second reagent or a rescue confirming the effect comes from the intended target — and positive controls where a negative result is claimed. Vehicle and baseline controls are asked for almost as often but usually exist already and were simply not reported.

Do I need a positive control for a negative result?

Yes. Without evidence that your assay could have detected an effect, absence of effect is indistinguishable from failure of the assay. This is the least negotiable control request there is.

Can I use a control from a previous paper?

Sometimes, if the experiment was run under the same conditions, and you say clearly that it is previously published and cite it. Presenting it as new data for this study is not acceptable.

The reviewer wants a control that would take six months. What do I do?

Say so, explain what it would establish, and offer either a quicker alternative that addresses the same concern or a narrowing of the claim so the control is no longer load-bearing. Editors weigh proportionality; they rarely require an experiment that exceeds what the claim is worth.

Should I add controls the reviewer did not ask for?

If you have them and they strengthen the paper, yes. A revision that anticipates the next objection often prevents a third round.

How do I know whether the control the reviewer wants already exists?

Search the primary records rather than the assembled figures: plate exports retain no-template wells, .fcs files retain unstained and fluorescence-minus-one tubes, uncropped blot images retain loading and secondary-only lanes, and animal records retain vehicle cohorts that were never graphed. Report a recovered control with the date it was performed.