Reviewer questions whether the experiment was replicated

A reviewer who writes “it is unclear how many times this experiment was performed” is, in most cases, naming a reporting gap rather than a missing experiment. You almost certainly repeated it. The manuscript almost certainly does not say so where a reader can check. The Nature Portfolio Reporting Summary asks for that statement in writing, and most figure legends still dodge it.

It is worth fixing carefully rather than quickly, because the phrase most people reach for is the one that caused the comment.

The phrase that causes this

“Representative of three independent experiments.”

It appears in thousands of figure legends and it is ambiguous in a way reviewers have learned to distrust. It does not say whether the three experiments agreed. It does not say whether the panel shown was chosen because it was typical or because it was best. And it does not say what was quantified across the three, or whether anything was.

A reviewer reading it cannot tell the difference between an experiment reproduced three times with consistent results and one performed three times with the cleanest instance shown.

The ambiguity is structural, not stylistic. The phrase compresses four separate facts — the count, the independence of the units, the selection rule for the displayed panel, and the agreement between repeats — into one adjective, “representative”, which asserts the conclusion of all four without reporting any of them. Vaux, Fidler and Cumming named the underlying confusion in “Replicates and repeats—what is the difference and is it significant?” (EMBO Reports 2012;13:291–296), where their worked example turns on the flat statement that “we have made just one comparison, so n = 1, no matter how many replicate plates we count.” A legend that says “representative” without saying which kind of repeat it is representative of leaves the reviewer to assume the worse reading, and reviewers who have been burned assume it.

What the reviewer is actually asking

A reviewer who questions replication is asking three separable questions, and answering only one of them does not close the comment.

How many, and of what unit. The count is worthless without the unit it counts. Three is three cultures, three cohorts, three days — or three wells, three lanes, three fields of view. ARRIVE 2.0 makes this a reporting item rather than a matter of taste: the Study design entry in its Essential 10 asks authors to state what the experimental unit is — a single animal, a litter, a cage — and not merely how many measurements were made. The same discipline applies to a dish as to a mouse; see pseudoreplication for the general form.

Whether the repeats agreed. Repetition without concordance is not evidence of anything. A reviewer wants to see the spread across repeats, which means seeing each repeat’s value, not a mean with a standard error computed across a mixture of repeats and subsamples.

Whether the panel shown was selected on the outcome. Choosing the clearest of three blots is normal practice and needs a stated rule. Choosing the one that matches the hypothesis is selection on the outcome, and it is the reason the comment carries an edge.

Answer all three in the same place, and the comment does not survive to the second round. Answer only the first — “the experiment was performed three times” — and you will receive it again, differently worded.

Legend phrases and what a reviewer infers

What to report instead

How many independent experiments, and what makes them independent — separate cultures, separate animal cohorts, separate days with fresh reagents. Independence matters more than the count. Write the independence criterion in the methods once, in a sentence a reader can audit: “An independent experiment is one initiated from a separate vial of cells thawed on a separate day, using a separately prepared lysis buffer and a different antibody dilution made that morning.” That sentence takes one line and pre-empts the entire objection.

What is shown. State plainly whether the panel is one representative experiment or pooled data. If representative, say how it was chosen. “The experiment whose control band intensity was closest to the median of the three” is a rule. “A representative blot” is not.

The quantification across all of them. This is the part that actually answers the objection. A blot from one experiment plus densitometry across all three, with each experiment’s value visible, converts an unverifiable claim into evidence. Many journals now require the individual data points rather than a bar with error bars, and reviewers increasingly expect it regardless. Weissgerber and colleagues gave the reason in “Beyond bar and line graphs: time for a new data presentation paradigm” (PLOS Biology 2015;13:e1002128), a review of 703 articles in top physiology journals in which 85.6% included at least one bar graph: “Many different data distributions can lead to the same bar or line graph.” Three points plotted individually cannot.

What varied and what did not. If the magnitude differed between experiments but the direction held, say so. Reviewers are far more comfortable with honest variability than with three experiments implied to be identical. Give the numbers: “the effect was 2.1-, 1.8- and 2.4-fold in the three experiments” is a stronger sentence than “the effect was consistent”, and it is the same information.

A worked example: one legend, rewritten

A western blot legend can close this objection in three added clauses, and the difference is worth seeing side by side.

The version that draws the comment: “Figure 3B. Knockdown of PLK1 reduces cyclin B1. Representative of three independent experiments. *p < 0.05.”

The version that does not: “Figure 3B. Knockdown of PLK1 reduces cyclin B1. One of three independent experiments, each begun from a separately thawed vial of HeLa cells on a separate day with freshly prepared lysates; the blot shown is the experiment whose loading control was closest to the median. Panel 3C shows cyclin B1 band intensity normalised to GAPDH for all three experiments, individual values plotted; fold reduction 0.42, 0.51 and 0.38. Two-sided paired t-test on the three experiment-level means, t(2) = 9.4, p = 0.011.”

Three facts changed hands: the unit of independence, the selection rule for the displayed panel, and the values of every repeat. The degrees of freedom now state the sample size out loud — t(2) can only come from three paired units — which is itself a defence, because a reviewer who sees t(46) under a claim of three experiments knows immediately that the test counted lanes. The same arithmetic works as a self-check: for an unpaired comparison of three versus three independent experiments, df = 4, and any larger number in your own manuscript means something other than the experiment was counted.

What journals already require

Journals already require the statement most manuscripts omit, which means the reviewer’s comment is usually a compliance failure rather than a scientific dispute. The Nature Portfolio Reporting Summary, submitted with every life-sciences manuscript, contains a Statistics checklist whose second item asks authors to confirm the presence of “A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly”, and whose first asks for “The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement”. The Life sciences study design section then carries five items — Sample size, Data exclusions, Replication, Randomization, Blinding — under the instruction that “All studies must disclose on these points even when the disclosure is negative.”

Two consequences follow. First, the reviewer asking how many times you did it is asking a question the journal already put to you in writing, and the editor can see your answer. Second, a Reporting Summary that says the finding was reproduced, alongside a legend that says only “representative”, is an internal inconsistency inside your own submission — one of the fastest ways to lose a reviewer’s trust in a methods section.

Assay-specific standards say the same thing in field language. MIQE (Bustin et al., Clinical Chemistry 2009;55:611–622) keeps biological replicates and technical replicates as separate reporting items for qPCR, and asks separately about run-to-run variation — three questions that the phrase “performed in triplicate” answers for none. ARRIVE 2.0 does the equivalent for animal work; see the ARRIVE guidelines.

If the answer is that you did it once

Say so. A single well-controlled experiment reported as such is a legitimate, if limited, result. A single experiment described as “representative of three” is misconduct.

For a preliminary or descriptive finding, an honest n=1 with the limitation stated is publishable in many venues. For a central claim it usually is not, and the choice is between repeating it and demoting it.

Demoting is a real option and it is under-used. Moving a claim out of the title and abstract into a results paragraph, changing “demonstrates” to “is consistent with”, and adding one sentence to the discussion converts an indefensible central claim into a defensible supporting one — see the overclaim check and how to write a limitations section. The paper survives; the sentence that could not be supported does not.

There is one thing never to do here: report the single experiment’s technical replicates as the n and let the statistics stand. That converts an honest limitation into a statistical error a reviewer can name, and it is the exact failure the independence objection names.

The nested trap

Be careful that your three “independent experiments” are independent. Three flasks split from one culture on one day are technical replicates. Three passages of the same line in the same week share more than they differ. Neither supports the generalisation that three genuinely independent repeats supports.

This is the same structure as the sample size objection, and if you pool measurements across the repeats and test on the combined total you have created a second problem — see reviewer says my n is not independent.

The trap has a second mouth. Three experiments that are genuinely independent in material can still be dependent in everything that matters analytically: one antibody lot, one plasmid prep, one batch of serum, one instrument calibration, one operator. Independence for the purpose of this objection is independence with respect to the sources of variation your conclusion needs to survive. If a reagent lot changed between repeats and the effect changed with it, that is information a reviewer needs and a discussion paragraph you should write, not a nuisance to average away. Where repeat-to-repeat variation is itself the phenomenon under discussion, see my replicates disagree.

How to find this before a reviewer does

Audit the manuscript for the five specific defects that produce this comment, which takes about twenty minutes and needs no new data.

Search the text for “representative”. Every hit is a place where a selection rule is missing. Either add the rule or delete the word.

Check that every figure legend states an n with its unit. A legend with error bars and no n is the single most common trigger, and the fix is typing six characters.

Reconcile the legend n against the methods. Legends claiming three experiments while the methods describe one cohort of animals, or one thaw, are contradicting each other in the same PDF.

Check the degrees of freedom against the claimed number of experiments. This is the arithmetic a suspicious reviewer performs first, and it is decisive. A t-statistic reported as t(58) beneath a legend claiming n = 3 has already answered the reviewer’s question, in the wrong direction.

Confirm that a quantification exists for every claim carried by an image. A representative micrograph supporting a sentence about magnitude, with no measurement anywhere in the figure, is the version of this objection that arrives from microscopy reviewers and flow cytometry reviewers alike.

When the obvious remedy is unavailable

Repeating the experiment is sometimes impossible, and the honest routes out are narrower than authors assume but they exist. Name the reason in the response letter, then take one of them.

The material is gone. A discontinued antibody lot, an exhausted patient sample set, a contaminated or drifted cell line, an expired animal protocol. Substituting a different reagent and calling the result a repeat is not a repeat; say what changed, and present the new experiment as independent corroboration rather than replication.

The timeline does not fit. A revision window of 60 to 90 days does not accommodate a five-month mouse cohort. Write to the editor before the deadline with the specific experiment, the specific duration, and a proposed alternative. Editors grant extensions routinely and refuse retroactive excuses.

Corroborate orthogonally instead. A second assay measuring the same quantity by a different physical principle answers the underlying worry — that the observation is an artefact of one system — more convincingly than a third repeat of the same protocol. That is the substance of orthogonal validation, and it is frequently faster than the repeat.

Make the single observation inspectable. Deposit the raw images, the uncropped blots, the gating files and the per-cell measurements, and cite the accession in the paper. A reviewer can be persuaded by a single experiment they can examine in a way they cannot be persuaded by one they must take on trust; see data availability and accessions.

Reduce the claim to what one experiment supports. The last resort is also the most reliable, and it is the same demotion described above.

What reviewers write when this goes wrong

Reviewers rarely write “please replicate”. They write a specific version, and the wording tells you which of the three questions they are asking.

“It is unclear how many times this experiment was performed.” “The figure legend states ‘representative’ but no quantification of the replicates is provided.” “Were these independent biological replicates or technical replicates of a single preparation?” “The n in Figure 2 does not agree with the number of animals described in the Methods.” “Statistics appear to have been performed on individual cells rather than on animals; please clarify the experimental unit.” “The authors should show the data from all replicates, with individual points, rather than a bar graph.” “How was the representative image selected?” “Please state, for each figure panel, the number of independent experiments and the number of technical replicates within each.”

The last of those is now common enough to treat as the default expectation. A reviewer asking for it panel by panel has usually decided the manuscript’s reporting is unreliable in general, which is a harder position to recover from than a single unclear legend — and the reason to fix the legends before submission rather than after.

Answering it in the response letter

Answer this comment with the revised legend text quoted in full, not with a description of the change. The reviewer’s question is about what the manuscript says, so the response must show what the manuscript now says.

A working shape: state the number and the unit, state where the quantification now appears, give the per-experiment values in the letter itself, and point to the line number. For example — “The experiment was performed three times, each from a separately thawed vial on a separate day. We have added the quantification of all three to new Figure 3C with individual values plotted (0.42, 0.51 and 0.38 fold), revised the legend to state the selection rule for the displayed blot, and reported the paired t-test on experiment-level means (lines 214–218).”

Do not argue that “representative of three independent experiments” is standard practice. It is standard practice, the reviewer knows that, and it is the practice being questioned. Where you genuinely did the experiment once, say so in the first sentence of the reply rather than the last; reviewers accept limitations disclosed early and distrust limitations discovered late. For the general form of the letter, see how to write a response to reviewers.

Before submission

The pseudoreplication agent reports every n whose unit cannot be determined from the manuscript, which is exactly the state that produces this comment. The image quantification agent catches the related version for microscopy: a representative image standing in for a quantification that was never performed.

PerfectPaper reads every figure legend against the methods section and reports each panel where the number of independent experiments, the unit those experiments count, and the selection rule for a displayed representative panel cannot all be recovered from the manuscript. For animal and cell-based work that check runs alongside the experimental-unit reporting required by in vivo rigour.

More in this family: reviewer comments on methods and design · reviewer says the sample size is too small.

Review my manuscript

Frequently asked questions

What does “representative of three independent experiments” actually mean?

It is meant to convey that the experiment was repeated three times with consistent results and one instance is shown. Because it does not say how the shown instance was chosen or whether the three agreed, reviewers increasingly treat it as insufficient on its own.

How many times should an experiment be repeated?

Three independent repeats is the common convention for a central claim in cell biology, but the number matters less than independence and than showing the data from all of them. Three technical repeats are worth less than two genuinely independent ones.

What counts as an independent experiment?

One performed on separately derived biological material, typically on a different day with fresh reagents. Aliquots of one culture measured three times, or three lanes of one preparation, are technical replicates.

Can I publish a single experiment?

Sometimes, if reported honestly as a single observation and if it is not carrying a central claim. What is never acceptable is describing one experiment as representative of several.

Should I show all replicates or a representative one?

Show the quantification across all of them, with individual points visible, alongside a representative panel if the assay is visual. That combination answers the objection before it is made.

How do I answer a reviewer who asks how many times the experiment was performed?

Quote the revised legend in the response letter, giving the number of independent experiments, what made them independent, the value obtained in each, and where the quantification now appears. Describing the change without showing the new wording leaves the reviewer to check for themselves.

What if I cannot repeat the experiment during revision?

Say why in the first sentence of the reply, then choose one of four routes: request an extension from the editor in writing before the deadline, corroborate the finding with an orthogonal assay, deposit the raw data so the single observation can be inspected, or reduce the claim to what one experiment supports.