Reviewer says the model system is not appropriate

“The model system is not appropriate” arrives late in review and is answered with a sentence more often than authors expect. The reviewer is saying that the system you used cannot support the conclusion you drew from it — a statement about the fit between model and claim, not a statement that the model itself is bad.

That distinction is the whole of the response. If the claim narrows to what the model supports, the objection dissolves.

This page sets out the five mismatches that produce almost every version of the comment, which immune compartment each host strain is missing, why species mismatch breaks crosstalk claims specifically, how to tell from the wording which of the three responses the reviewer wants, what to do when the model they named is unavailable, how to find the mismatch in your own abstract before a reviewer finds it, and the sentences reviewers write when the system does not fit the claim.

The recurring mismatches

Five mismatches account for almost every instance of this objection, and each one attaches to a particular kind of sentence rather than to a particular kind of experiment.

Immune claims in an immunodeficient host. Conclusions about immune evasion or immunotherapy mechanism drawn in nude, SCID or NSG mice. Which compartment is missing matters: nude mice retain B cells, natural killer cells and innate immunity while lacking mature T cells, so an innate claim may survive where an adaptive one cannot. NSG mice lack substantially more. The objection is fatal to a claim about T-cell-mediated killing in a nude host and irrelevant to a claim about tumour-intrinsic proliferation in the same animal, which is why the compartment has to be named before the response is written.

Microenvironment claims from a subcutaneous implant. Flank tumours sit in subcutaneous tissue, not in the organ of origin. Conclusions about stromal composition, tissue-resident immune populations or metastatic tropism generally need an orthotopic model. The subcutaneous site supplies dermal fibroblasts and skin vasculature rather than hepatic stellate cells, pancreatic stellate cells, alveolar macrophages or Kupffer cells, so a manuscript that describes “the tumour microenvironment” from a flank implant is describing a microenvironment the disease does not occur in. Metastatic tropism is the sharpest case: a flank tumour’s route to a distant organ is not the route a primary tumour takes from its own organ.

Species mismatch in xenografts. Human tumour cells in a mouse host communicate with mouse stroma, and many ligand-receptor pairs do not function across species. Claims about crosstalk that assume cross-species signalling are the specific failure.

One cell line standing for a disease. A result in a single line supports a claim about that line. Generalisation to a tumour type needs more lines, or a narrower sentence. The line’s own provenance is part of the objection: a reviewer who doubts the system will also ask whether the line was authenticated by short tandem repeat profiling, at what passage the experiments were performed, and whether mycoplasma testing was recorded — see data availability and resource identification for the reporting form those answers take.

Overexpression standing for physiology. A protein driven far above its endogenous level can do things the physiological protein does not. Mislocalisation, saturation of a degradation pathway, and interactions that never occur at endogenous stoichiometry are all consequences of level rather than of function, and the check a reviewer expects is a measurement of the expressed protein against the endogenous one in the same blot, not an assurance that the construct is the wild-type sequence.

Which compartment the host is missing

The correct objection depends on the strain, and so does the correct answer. Immunodeficiency is not one state: the standard hosts differ in which lymphoid and innate compartments are absent, and each absence forecloses a different set of sentences.

Two features of that table decide most responses. First, the Il2rg null allele carried by NSG mice removes signalling through the common gamma chain shared by interleukin-2, -4, -7, -9, -15 and -21, which is why natural killer cells are absent rather than merely reduced — an NSG experiment cannot be rescued by arguing that innate immunity remained. Second, the NOD background used by NOD-scid and NSG mice is haemolytic-complement deficient, so a complement-dependent mechanism is unsupportable in those hosts even though it is a claim about innate rather than adaptive immunity, and authors regularly miss that because they sorted the objection into the wrong compartment.

Why species mismatch breaks crosstalk claims

Species mismatch is a mechanism, not a general caution, and stating the mechanism is what turns the reviewer’s objection into a specific limitation you can write down. In a xenograft, human tumour cells sit in mouse stroma, and a signal only reaches its target if the mouse ligand engages the human receptor or the reverse.

Two well-documented cases carry most of the weight. Mice have no CXCL8 gene — there is no murine interleukin-8, and the functional murine analogues are different chemokines — so a human tumour engineered to depend on interleukin-8 signalling from stroma is in a host that cannot produce it. And human CD47 engages mouse SIRPα poorly, which is the reason human cell engraftment depends on the NOD background: the NOD Sirpa polymorphism binds human CD47, and Takenaka and colleagues reported that difference in Nature Immunology (2007) as the determinant of human haematopoietic engraftment. A “don’t eat me” signal that a mouse macrophage cannot read is not a finding about tumour immune evasion in patients.

The stromal composition of a patient-derived xenograft changes for the same reason. Human stroma engrafted with the tumour is progressively replaced by mouse stroma over passage, so a late-passage patient-derived xenograft is human tumour in a mouse tissue compartment rather than a preserved human tumour. That makes a patient-derived xenograft strong evidence about tumour-intrinsic drug response and weak evidence about stromal interaction, and a manuscript that treats the model as uniformly closer to the patient is making a claim about the wrong axis.

The broader argument that mouse immunology does not map cleanly onto human immunology is old and well cited — Mestas and Hughes set out the compartment-by-compartment differences in “Of mice and not men: differences between mouse and human immunology” in the Journal of Immunology (2004). Treat that paper as the reason to name your compartment precisely, not as a reason to abandon mouse work: the same literature contains a live disagreement, since Seok and colleagues reported in PNAS (2013) that genomic responses in mouse models poorly mimic human inflammatory diseases, and Takao and Miyakawa reached the opposite conclusion in PNAS (2015) reanalysing the same datasets with different gene selection. A reviewer who cites either paper is making an argument about your specific claim, and the response is about your claim rather than about mouse models in general.

Reading the objection before answering it

The wording tells you which of the three responses the reviewer wants, and misreading it is what turns a one-sentence revision into a six-month one.

Three tells do the sorting. First, the verb: “cannot support”, “does not permit” and “is not an appropriate system for” mark a claim the reviewer considers unsupportable, which means narrowing or a second system; “would be strengthened by”, “the authors should discuss” and “it is unclear whether” mark a discussion request answered in text. Second, where the comment lands: a comment quoting the abstract or the title is aimed at a load-bearing sentence, while a comment attached to one supplementary panel is aimed at that panel’s interpretation. Third, whether the reviewer named a compartment, a site or a species — a reviewer who wrote “in an immunodeficient host” has told you which paragraph to rewrite, and a reviewer who wrote “the model is not appropriate” has not, in which case ask the editor which claim the comment is aimed at rather than guessing across five mismatches.

Read the whole review before planning. A request for a second cell line in one comment and a remark about generalisation to the tumour type in another are one objection, and answering them separately produces two weak answers. Reviewer comments on methods and design sets out the sorting exercise across the whole review, and reviewer wants more controls covers the version aimed at the number of systems rather than at the system itself.

The three responses

Three responses cover the objection, and they differ enormously in the work they require — one is an afternoon of editing, another is a new set of animal experiments.

Narrow the claim. Usually correct and usually sufficient. “In this syngeneic model, loss of X reduced tumour growth in an immune-competent host” is precise, defensible and still interesting. “X drives immune evasion in cancer” is neither, if it came from one flank model. Narrowing has to reach every instance of the claim: the title, the abstract’s final sentence, the first line of the discussion and the discussion’s closing paragraph are the four places an over-broad sentence survives a revision, and a claim narrowed in the discussion and left intact in the abstract draws the same objection in the second round — check the result with the overclaim check before resubmitting.

Add a second system. A second cell line, an orthotopic version, an immune-competent counterpart. Slow and resource-intensive, and worth it when the generalisation is the paper’s point rather than an incidental sentence. Choose the system that fails for a different reason than the first: a second line from the same molecular subtype tests reproducibility rather than generality, and an orthotopic version of the same line answers the site objection while leaving the single-line objection untouched. Where the concern is immunological, the syngeneic counterpart — an allograft such as MC38 or CT26 in a wild-type host — answers it directly, with the caveat that highly mutated syngeneic lines respond to checkpoint blockade more readily than most human tumours, so the new model supports a mechanistic claim more securely than a claim about expected clinical benefit.

Argue the model is appropriate. Legitimate when the reviewer has misread what the model provides — for example objecting to a xenograft for a claim that is explicitly about tumour-intrinsic biology, where host immunity is not load-bearing. Make the argument explicitly and say which conclusions do and do not depend on host context. Make it once, factually, and pair it with something: the sentence you narrowed elsewhere, the supplementary analysis that separates the host-dependent result from the host-independent one, or the citation showing the model is standard for this question. An argument that leaves the manuscript unchanged is the version editors decline.

When the model the reviewer named is unavailable

Name the constraint and supply a substitute. A stated limit with an alternative beside it is accepted far more often than a defence that changes nothing, and every one of these situations ends in the same fallback: narrow the claim until the missing model is no longer load-bearing.

The orthotopic model is not feasible. Orthotopic implantation in pancreas, lung, brain or bone requires surgical capability, an imaging modality for tumour burden, and an ethical approval that names the procedure. Where any of the three is absent, say so, report tumour burden by the method you have, and limit the microenvironment claim to what a subcutaneous site supports rather than presenting flank data as microenvironment data.

No immune-competent counterpart exists. A human-specific target with no murine orthologue, or a therapeutic antibody that does not cross-react with the mouse protein, leaves no syngeneic version of the experiment. State the reason precisely — the orthologue is absent, or the antibody has no murine cross-reactivity — because that sentence converts an apparent evasion into a specification, and then confine the immune claim to what a humanised or ex vivo human system showed.

The humanised model has a short window. Mice reconstituted with human peripheral blood mononuclear cells develop xenogeneic graft-versus-host disease within weeks, which caps experiment duration and confounds late endpoints. Where the endpoint you need falls outside that window, report the window as a design constraint and choose an endpoint inside it rather than extending the study and attributing the wasting to treatment.

The cohort, line or model is gone. The patient-derived xenograft failed to re-engraft, the line failed authentication, the colony was closed. Say so factually and supply what remains: banked material, archived tissue, a deposited dataset containing the comparison, or the same experiment in a related system with the difference named. How to write a response to reviewers sets out the shape of that answer, and a worked response letter shows it applied.

Finding the mismatch in your own abstract

The mismatch is found by reading claims against the methods, which is a mechanical exercise rather than a judgement, and it can be done in a single sitting.

Write the system down in one line first. Model type, host strain, immune status, implantation site, species of every cell in the experiment. Most authors cannot produce this line from memory, and the sentence they eventually write is already the limitation paragraph the paper needed.

List every claim sentence from the title, abstract and discussion. Put the system line beside each one and mark the sentences that depend on a compartment, a site or a species the system does not have. Sentences in the closing discussion paragraph fail this test most often, because that paragraph is written last and aims outward.

Search the manuscript for the words that carry the claim. Immune evasion, tumour-infiltrating, microenvironment, stromal crosstalk, metastatic tropism, translational, patients, clinically and therapeutic potential are the strings that most often exceed the model. Each hit is either supportable in the system you wrote down or it is the sentence to rewrite.

Read ARRIVE 2.0 item 18 against the discussion. Item 18, generalisability and translation, asks how the findings relate to human biology, and a manuscript that has never answered it in the text has left the reviewer to answer it instead. The ARRIVE guidelines sets out the items one at a time, and in vivo experimental rigour covers the randomisation, blinding and exclusion reporting that the same reviewer usually raises in the neighbouring comment.

Check the dose and schedule alongside the model. A claim of clinical promise depends on the treatment schedule as well as the host; translational dose relevance covers the version of this objection aimed at exposure rather than at the system.

What reviewers write when the model does not fit

Reviewers rarely write “the model system is not appropriate” as the whole comment. They write the specific version, and the wording identifies the mismatch and the remedy.

“The conclusion regarding immune evasion is not supportable in an immunodeficient host.” “Tumours were implanted subcutaneously; the authors’ claims regarding the tumour microenvironment require an orthotopic model.” “The proposed crosstalk assumes signalling between human tumour cells and mouse stroma, which is not established for this ligand-receptor pair.” “All conclusions rest on a single cell line, and the abstract generalises them to the disease.” “The protein is expressed well above endogenous levels; the phenotype may reflect expression rather than function.” “The authors describe this as a translational finding, but the model has no immune component.” “The strain used is not stated, and the immune status cannot be inferred.” “The limitations section acknowledges the model but the abstract does not.”

The last of those is the one authors most often trigger themselves. A limitations paragraph that concedes the model while the abstract still makes the broad claim reads as an admission that the claim is known to be unsupported, which is worse than either sentence alone. How to write a limitations section covers the form that avoids it.

Saying it in the response letter

Name the constraint rather than working around it. “Our model does not permit conclusions about adaptive immunity; we have removed the claim from the abstract and now limit the discussion to innate mechanisms, which the model does support” reads as competence. A defence that leaves the abstract untouched does not.

A complete answer states the system, names the compartment or the site at issue, quotes the revised sentence, and says where in the manuscript it now appears. “Reviewer 2 comment 1 concerns the immune claim. The experiments used athymic nude mice, which lack mature T cells while retaining B cells, natural killer cells and innate immunity. We agree that a claim about adaptive immune evasion is not supportable in this host. The abstract now reads: ‘Loss of X reduced tumour growth in an athymic host, consistent with an innate or tumour-intrinsic mechanism.’ The discussion, page 14, now states explicitly that the model does not permit conclusions about T-cell-mediated immunity.” Every clause is checkable against the manuscript, which is what makes it short.

Answer every version of the comment, including the ones you decline, and give the reason rather than the fact. Where the same objection appears in two reviews, answer it once and cross-reference, so the editor reads one position rather than two.

Before submission

Model-to-human relevance is built for this: it establishes the system from the methods, identifies which immune compartments are absent or impaired in the host used, and reports every claim in the abstract and discussion that the system cannot support — naming the compartment and the sentence, and proposing the narrower wording. It is the agent most likely to save a revision round, because this is the objection that arrives last.

More in this family: reviewer comments on methods and design · AI peer review for preclinical cancer research · reviewer says the mechanism is not shown · unit of analysis and pseudoreplication

PerfectPaper reads the host strain, implantation site and cell species out of the methods and reports every claim in the title, abstract and discussion that the system cannot support, naming the missing compartment and proposing the sentence that the model does support.

Review my manuscript

Frequently asked questions

What does “the model system is not appropriate” mean in a peer review?

The comment means the system used cannot generate evidence for the specific conclusion drawn, not that the system is poor. Five mismatches produce almost all instances: an immune claim in an immunodeficient host, a microenvironment claim from a subcutaneous implant, a crosstalk claim across species, a disease-level claim from one cell line, and a physiological claim from an overexpression construct.

How do I respond when a reviewer says my model is not appropriate?

Narrow the claim, add a second system, or argue the model was misread. Narrowing is correct and sufficient in most cases: state which compartment or context the model lacks, change the sentence in the title, abstract and discussion, and quote the revised wording in the response letter rather than defending the original.

Is a nude mouse an inappropriate model for an immune claim?

For an adaptive claim, yes. Nude mice lack mature T cells but retain B cells, natural killer cells and innate immunity, so innate claims may hold while adaptive or T-cell-mediated ones do not. Name the specific compartment rather than treating immunodeficiency as all-or-nothing.

Is a subcutaneous tumour model an inappropriate model?

For many questions it is perfectly appropriate. For conclusions depending on organ context — stromal composition, tissue-resident immunity, metastatic tropism — the site does not reproduce the relevant environment and an orthotopic model is usually expected.

Does a single cell line make my model inappropriate?

Only for a claim that outruns it. One line supports a claim about that line, and generalising to a tumour type generally needs several lines or a sentence that does not generalise. The objection is about the sentence, not about the line.

Does narrowing the claim to fit the model look like a weakness?

No. Reviewers read a considered narrowing as evidence the authors understand their system. What reads as weakness is defending a claim the model cannot support.

Why does the model-appropriateness objection arrive so late in review?

Because it depends on reading the discussion against the methods, which reviewers do last. It is also why checking model-to-claim fit before submission catches more than most pre-submission passes.