Response to comment on Mutualism weaken the latitudinal diversity gradient among oceanic islands

Camille Delavaux; Evan Gora; James Bever; Thomas Crowther

REVIEW 4 major objections 4 minor 5 references

Response to comment on Mutualism weaken the latitudinal diversity gradient among oceanic islands

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The original conclusion that mutualisms weaken the latitudinal diversity gradient on oceanic islands is not undermined by the reanalysis, which the authors argue relies on a biologically unjustified longitude predictor and overfit…

desk verdict The rebuttal raises a fair point about the critique's code and shows a concerning map of predictions, but the central argument that longitude is biologically meaningless conflates mechanism with prediction and goes untested; as a defense of the original result, it does not hold up. read the letter →

arxiv 2505.21006 v1 pith:SNTS5VIV submitted 2025-05-27 q-bio.PE

Camille Delavaux , Thomas Crowther , James Bever , Evan Gora This is my paper

classification q-bio.PE

keywords latitudinaldiversitygradientoceanicislandsmutualismfilterplantspeciesrichnessmycorrhizalfunginitrogen-fixingbacteriadeficitglobalbiodiversitymodels

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This response defends an earlier finding against a reanalysis that claimed the result was a statistical artifact. The authors argue that the reanalysis adds longitude as a predictor of plant species richness, which has no physical or biological basis at a global scale and is a biased measure of distance. They also say the reanalysis's way of assigning expected mainland richness to islands is unjustified, and that its machine-learning model overfits sparse training data and produces implausible predictions. In the authors' view, the critique's central concerns fail, so the original conclusion stands: plants that depend on mutualists are disproportionately filtered from oceanic islands, weakening the latitudinal diversity gradient worldwide.

What carries the argument

The latitudinal diversity gradient (LDG) is the tendency for species richness to decline from the tropics toward the poles. The central object in this response is the mutualism filter: plant species that associate with pollinators, arbuscular mycorrhizal fungi, or nitrogen-fixing bacteria are hypothesized to be less likely to establish on oceanic islands, so islands show a weaker LDG and a disproportionate loss of mutualist-associated species. The response's argumentative machinery is the baseline model of continental richness against absolute latitude, defended against the reanalysis by three moves: longitude has no global biological mechanism; non-linear model terms can express the mutualism filter's interaction effects; and the random forest's predicted mainland richness is demonstrably unrealistic when mapped. The authors also lean on several independent analyses that find mutualism-related filtering without using latitude.

What would settle it

Fit the comment's random forest to global mainland plant richness data and check whether its predicted richness for well-sampled low-latitude regions such as Ecuador or Amazonia matches observed values; if those predictions are accurate rather than severe under-predictions, the overfitting claim fails and the reanalysis's longitude inclusion would be vindicated.

Watch

Extended reading notes

Core claim

The paper's claim is that the critique does not undermine the original result that mutualisms weaken the latitudinal diversity gradient among oceanic islands. To make this claim, the authors defend their original approach of modeling continental plant richness against absolute latitude and comparing that baseline to island richness. They reject the addition of longitude as a predictor because longitude does not correspond to any consistent global physical gradient and because one degree of longitude shrinks from the equator to the poles, potentially creating spurious covariance with latitude and distance. They further argue that non-linear relationships in the reanalysis's models could actually be capturing the mutualism filter's interaction with area, distance, and latitude rather than disproving it, and that the random forest predictions are overfit, as shown by implausibly low predicted richness in known tropical mainland and island regions. The conclusion is that the reanalysis fails to justify overturning the original finding.

Load-bearing premise

The response's rejection of the critique depends on the assumption that longitude has no meaningful global relationship to plant species richness; if longitude instead captures real spatial differences in mainland richness that the latitude-only baseline misses, the critique's inclusion of it is justified and the original deficit estimates could be biased.

Editorial extensions

If this is right

If the response is correct, the original conclusion holds: oceanic islands worldwide have a reduced latitudinal diversity gradient, and mutualist-associated plants contribute disproportionately to island species deficits.
A reanalysis that includes longitude as a global richness predictor would need a physical mechanism; without one, any improvement in model fit is likely spurious.
Non-linear fits of area, distance, and latitude are not automatically evidence against the mutualism filter, because interactions between the filter and those variables can appear as non-linearities.
The critique's species-deficit estimates should be treated skeptically unless its random forest predictions for mainland richness are validated against observed diversity in well-sampled tropical regions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

One implication not developed in the response is a general test: adding an arbitrary coordinate with no physical gradient to any global biodiversity model should degrade or destabilize predictions; studies could test this by adding longitude to unrelated richness models and checking prediction error.
The authors' reasoning implies a scale distinction: longitude may be a useful regional predictor where coastlines, mountains, or ocean proximity create coherent longitudinal gradients, so the critique might be rescued by restricting longitude to such regions rather than globally.
A direct extension would be to rerun the deficit calculation using an alternative source-pool definition that averages mainland richness across all potential source environments, which the authors suggest would approximate their latitude-only baseline; comparing those estimates would clarify whether the reanalysis's longitude projection causes its divergent result.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Dear colleague,

Quick take: this response is a rebuttal, not a new scientific result, and it fails to land the decisive blow it claims. The one genuinely useful thing is that they flag the critique's code and figure ambiguities — that is a fair and important point. They also offer a map of the random forest's mainland predictions, which visually does look odd in some data-sparse regions (e.g., very low predicted richness in parts of the tropics). That is a legitimate red flag worth following up.

But the core of the rebuttal doesn't hold up. Section 1A argues longitude is not a meaningful predictor because it "does not correlate with any physical difference at the global scale." That conflates mechanism with prediction. In the critique's model, longitude is used as a spatial coordinate to estimate mainland source-pool richness at island locations; it doesn't need a direct biological effect to be useful, because it encodes continentality, land-sea distribution, and proximity to source pools. The right way to test spuriousness is to compare in-sample vs. out-of-sample error, or use spatial cross-validation, and to ask whether the longitude-augmented model changes the estimated mutualism filter. The response does none of that; Figure 1 is anecdotal, not a quantitative test.

Their second argument — that non-linear fits could absorb mutualism interactions — is plausible but untested. They could have fitted the critique's model with mutualism interactions or shown that the non-linear terms actually capture the mutualism gradient. They didn't.

I also take issue with the "independent analyses" paragraph. Some of those studies are not independent (Delavaux et al. 2019, 2021, 2022 are self-citations), and more importantly, they support an ecological mutualism-filter pattern but do not address the specific statistical concern: whether the latitude-only baseline in the original deficit calculation is biased by omitted spatial structure.

The call for better documentation from Pitcher and Hartig is reasonable — the code and figure descriptions do seem unclear — but requesting clarity is not a substitute for a quantitative rebuttal.

Bottom line: as a response, this is weak. It asserts rather than demonstrates. Still, it is part of an active, high-stakes exchange about a Nature paper, and the methodological question (longitude in global richness models) is genuinely important. I would send it to peer review — paired with the original critique — but I would expect a competent referee to reject it or demand major revision. It does not deserve to stand as a convincing defense.

Bring it to reading group if the group wants to discuss the causality/prediction confusion; otherwise skip. My recommendation: accept for peer review as part of the exchange, but be clear that the response, as written, does not undermine the critique's statistical case.

Referee Report

4 major / 4 minor

Summary. This manuscript is a formal response to a critique by Pitcher and Hartig (arXiv:2411.15105) of the authors' previous study (Delavaux et al., Nature 2024). The authors defend their original conclusion that mutualisms weaken the latitudinal diversity gradient on oceanic islands by arguing that: (1) longitude is not a biologically meaningful predictor of global species richness, so its inclusion creates spurious fits; (2) non-linear terms for latitude, distance, and area could absorb the mutualism-filter interaction, so the critique's reanalysis does not rule out their mechanism; (3) the critique's assignment of expected mainland richness to islands is unjustified and likely overfitted; and (4) the critique's code and figures are insufficiently documented. Based on these arguments plus several cited prior studies, the authors conclude that the critique does not undermine their original conclusion.

Significance. If the response were correct, it would salvage a high-profile Nature result from a quantitative challenge, and it would clarify that the critique's reanalysis is not a valid artifact test. The response does provide a small piece of reproducible evidence by linking to code for its own Figure 1, and it correctly identifies some ambiguities in the critique's presentation. However, the response is largely qualitative: its central assertions about longitude, non-linear absorption, and overfitting are not backed by quantitative tests, and its appeal to 'independent' support partly relies on the authors' own prior papers. The significance is therefore moderate: the manuscript could be useful if its claims were substantiated, but in its current form it does not provide a rigorous refutation.

major comments (4)

[Section 1A] The rejection of longitude as a predictive variable is not a valid statistical argument. The statement that 'longitude is perpendicular to latitude and does not correlate with any physical difference at the global scale' conflates mechanistic causation with predictive conditioning. In the critique's random forest, longitude serves as a spatial coordinate that can encode continentality, land-sea distribution, and proximity to mainland source pools; it does not need to have a direct biological effect to improve predictions. The response provides no quantitative comparison of out-of-sample predictive performance, no cross-validation, and no demonstration that the longitude-augmented model is biased while the original latitude-only model is not. Because this assertion is the cornerstone of the rejection of the critique, it is load-bearing and currently unsupported.
[Section 1B] The claim that non-linear terms for latitude, distance, and area 'could capture the effect of the mutualism filter strength' is plausible but untested. The authors state that they 'explicitly tested for interactions between mutualism filter strength and all other variables' and found them significant, but they do not report any analysis showing that a non-linear model without the mutualism interaction reproduces the mutualism effect. Without such a test, this assertion does not demonstrate consistency between the critique's results and the original mechanism; it only raises a possibility.
[Section 2] The assessment of the random forest predictions in Figure 1 is qualitative and does not establish overfitting. The response identifies selected underpredictions (e.g., Ecuador) and states that fine-grain differences are 'likely' due to sparse data, but it does not compare in-sample versus out-of-sample error, quantify prediction bias, or show that the original latitude-only baseline is not itself biased by omitted spatial structure. The claim that a higher R-squared is 'misleading' because a visualization shows errors is not a substitute for a formal diagnostic, especially when the original model's performance is not similarly scrutinized.
[Overall/conclusion] The response repeatedly cites 'independent analyses' (Simonsen et al. 2017, Delavaux et al. 2019, Taylor et al. 2019, Delavaux et al. 2021, König et al. 2021, Delavaux et al. 2022) as external support for the mutualism filter. Several of these are authored by Delavaux and co-workers and rely on similar data and assumptions, so labeling them 'independent' overstates their evidential weight. Moreover, even fully independent evidence that mutualist-filter patterns exist elsewhere would not directly address the critique's specific claim that the original island-deficit calculation is confounded by longitude. This weakens the appeal to external support as a defense of the original analysis.

minor comments (4)

[Section 1A] There is a typo in the text: 'do es not' should be 'does not'.
[Section 2] The caption of Figure 1 does not specify the color scale, the source of the 'known' diversity values used for comparison, or the exact grid and data subset used; this makes the qualitative visual claim difficult to evaluate.
[Section 1B] The sentence 'Interactions between two linear variables can alternatively be fit as a non-linear relationship between one of those variables and the response' would benefit from a citation or a small simulation demonstration, since it is a key premise of the authors' argument.
[Section 2] The response complains that the critique's code is not fully repeatable, but does not provide a detailed log of the errors encountered or a minimal working example; adding such details would strengthen this process concern.

Circularity Check

1 steps flagged · score 4.0 of 10

Response rejects critique partly by citing the original paper's own Figure 3 as support for the disputed latitudinal deficit pattern; otherwise the rebuttal is non-circular but rests on untested assumptions.

self citation load bearing [Section 2, final paragraph (Assigning expected mainland richness to islands)]
"Moreover, we show that islands support fewer mutualist taxa than equal-latitude mainlands (Figure 3, Delavaux et al. 2024), as do many prior analyses (Delavaux et al. 2019, Taylor et al. 2019, Delavaux et al. 2021, König et al. 2021, Delavaux et al. 2022). We show these trends in our study and therefore find support for the latitudinal structuring of species deficit on islands."

The point under dispute is whether the original island-deficit calculation and the latitudinal structuring of species deficit are statistical artifacts of the latitude-only model. The response answers this by citing the original paper's own Figure 3 and then stating 'We show these trends in our study and therefore find support for the latitudinal structuring of species deficit on islands.' That is using the very result being challenged as evidence for itself. The preceding citations include genuinely external studies for the broader mutualism-filter pattern, but the specific latitudinal deficit structure is supported here by the original study alone, making this a load-bearing self-citation in the rejection of the critique.

full rationale

This is a response to a critique rather than a new derivation, so most circularity categories do not apply. The defense of the latitude-only baseline relies on biological plausibility arguments and a qualitative map of the random forest's underpredictions; those are contestable but not circular. One explicit self-supporting move does appear in Section 2: the authors reject the reanalysis partly by citing their own original paper's Figure 3 and then asserting that because they show these trends in their study, there is support for the latitudinal structuring of species deficit on islands. Since the question is whether that original result is a statistical artifact, invoking it as its own evidence is circular. This step is load-bearing for the final rejection, though the response also cites genuinely external peer-reviewed studies (Simonsen et al. 2017, Taylor et al. 2019, König et al. 2021) for the broader mutualism-filter pattern, so the central claim still has independent content. No fitted-input-called-prediction or uniqueness-imported-from-authors pattern is present; the lack of a quantitative out-of-sample comparison is a weakness of the rebuttal but not itself circularity. Overall score reflects one significant self-citation while the central claim retains independent support.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The response introduces no new fitted parameters or entities. Its central rejection depends on three modeling assumptions: the biological irrelevance of longitude, the unbiasedness of the original latitude-only baseline, and the capacity of non-linear terms to absorb the mutualism signal. None of these is independently verified in this response, and the first two are directly contested by the critique.

assumptions (3)

domain assumption Longitude is not a meaningful predictor of plant species richness at global scales
Section 1A: the authors assert this without empirical support; commonly, geographic coordinates are used as spatial covariates in macroecology.
domain assumption The original latitude-only mainland model provides an unbiased estimate of expected island richness
Section 2: the response defends this baseline by rejecting longitude, but does not test whether longitudinal variation in mainland diversity biases the expected values.
ad hoc to paper Non-linear terms for latitude and distance can fully absorb the mutualism-filter interaction
Section 1B: the claim that a non-linear model can capture mutualism interactions is plausible but not tested with the critique's model, and it is used to dismiss the critique's finding.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Response to comment on Mutualism weaken the latitudinal diversity gradient among oceanic islands." pith.science (2026). https://pith.science/paper/SNTS5VIV

@misc{pith2026250521006,
  author       = {Pith},
  title        = {Pith review of: Response to comment on Mutualism weaken the latitudinal diversity gradient among oceanic islands},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SNTS5VIV}},
  note         = {Machine review of arXiv:2505.21006}
}

read the original abstract

In our original paper (Delavaux et al. 2024; https://www.nature.com/articles/s41586-024-07110-y), we find that the latitudinal diversity gradient (LDG) in plant species richness is reduced on oceanic islands worldwide. Moreover, we find that plants that associate with mutualists, including pollinators, AM fungi, and N-fixing bacteria, disproportionately contribute to this reduction. To make this assertion, we fit models of continental diversity to absolute latitude and compared mainland diversity estimated by these models to the observed diversity of individual islands. The critique of our work by Pitcher and Hartig (arXiv:2411.15105) questions our results through a reanalysis of the data. In their reanalysis, they add longitude as a predictor of richness on islands. However, we have several concerns about this reanalysis. In particular, we question (1) the biological/physical basis of predicting species richness with longitude as well as modelling non-linear relationships potentially leading to spurious relationships and (2) the justification for how this comment calculated island deficits from the data. Additionally, (3) described analyses and provided code (https://github.com/MaximilianPi/IslandLDG) are in many cases insufficient to understand or meaningfully engage with this work, precluding productive discussion about cited analytical concerns. Given these major issues with the reanalysis, we reject the conclusions from this critique that our prior results are due to statistical artifacts. We therefore do not have reason to question our original conclusion that the latitudinal diversity gradient is weakened by mutualisms among oceanic islands worldwide.

Figures

Figures reproduced from arXiv: 2505.21006 by the authors.

**Figure 1.** A global projection of predictions based on Pitcher and Hartig’s random forest model including latitude and longitude (A) across a grid of world locations and (B) across islands used in Delavaux et al. 2024. Some regional predictions underpredict known values, raising concerns about the validity of this approach [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

5 extracted references · 5 canonical work pages

[1]

Pyšek, and A

Pergl, P. Pyšek, and A. Stein. 2019. Mycorrhizal fungi influence global plant biogeography. Nature ecology & evolution 3:424

work page 2019
[2]

Delavaux, C. S., P. Weigelt, W. Dawson, F. Essl, M. van Kleunen, C. König, J. Pergl, P. Pyšek, A. Stein, and M. Winter. 2021. Mycorrhizal types influence island biogeography of plants. Communications biology 4:1-8

work page 2021
[3]

Delavaux, C. S., P. Weigelt, S. M. Magnoli, H. Kreft, T. W. Crowther, and J. D. Bever. 2022. Nitrogen-fixing symbiotic bacteria act as a global filter for plant establishment on islands. Communications biology 5:1209. König, C., P. Weigelt, A. Taylor, A. Stein, W. Dawson, F. Essl, J. Pergl, P. Pyšek, M. van Kleunen, and M. Winter. 2021. Source pools and d...

work page 2022
[4]

Simonsen, A. K., R. Dinnage, L. G. Barrett, S. M. Prober, and P. H. Thrall. 2017. Symbiosis limits establishment of legumes outside their native range at a global scale. Nature communications 8:1-9

work page 2017
[5]

Weigelt, C

Taylor, A., P. Weigelt, C. König, G. Zotz, and H. Kreft. 2019. Island disharmony revisited using orchids as a model group. New Phytologist 223:597-606

work page 2019

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Pyšek, and A

Pergl, P. Pyšek, and A. Stein. 2019. Mycorrhizal fungi influence global plant biogeography. Nature ecology & evolution 3:424

work page 2019

[2] [2]

Delavaux, C. S., P. Weigelt, W. Dawson, F. Essl, M. van Kleunen, C. König, J. Pergl, P. Pyšek, A. Stein, and M. Winter. 2021. Mycorrhizal types influence island biogeography of plants. Communications biology 4:1-8

work page 2021

[3] [3]

Delavaux, C. S., P. Weigelt, S. M. Magnoli, H. Kreft, T. W. Crowther, and J. D. Bever. 2022. Nitrogen-fixing symbiotic bacteria act as a global filter for plant establishment on islands. Communications biology 5:1209. König, C., P. Weigelt, A. Taylor, A. Stein, W. Dawson, F. Essl, J. Pergl, P. Pyšek, M. van Kleunen, and M. Winter. 2021. Source pools and d...

work page 2022

[4] [4]

Simonsen, A. K., R. Dinnage, L. G. Barrett, S. M. Prober, and P. H. Thrall. 2017. Symbiosis limits establishment of legumes outside their native range at a global scale. Nature communications 8:1-9

work page 2017

[5] [5]

Weigelt, C

Taylor, A., P. Weigelt, C. König, G. Zotz, and H. Kreft. 2019. Island disharmony revisited using orchids as a model group. New Phytologist 223:597-606

work page 2019