Is there a robust effect of mainland mutualism rates on species richness of oceanic islands?
Preprint
Abstract
In island biogeography, it is widely accepted that species richness on island depends on the area and isolation of the island as well as the species pool on the mainland. Delavaux et al. (2024) suggest that species richness on oceanic islands also depends on the proportion of mutualists on the mainland, based on the idea that mutualists require specific interaction partners for their survival and thus have lower chances of establishment after successful immigration. As the proportion of mutualists increases towards the tropics, this effect could explain a weaker latitudinal diversity gradient (LDG) for oceanic islands. However, after re-analyzing their data, we have doubts if these conclusions are supported by the available data.
What the paper shows and why it matters
(AI-generated)
Island biogeography explains species richness on oceanic islands by area, isolation and the mainland species pool. Delavaux et al. (2024) added the mainland proportion of mutualists, species that depend on specific interaction partners and may therefore establish less often on islands. Because their share rises towards the tropics, this could weaken the island latitudinal diversity gradient. Their evidence, however, rests on a mutualism measure known only at some mainland sites. For the islands it was filled in from a model using latitude alone, and it entered the regression as a straight-line effect next to latitude. The paper questions whether such a covariate can be separated from latitude and from nonlinear effects of the other predictors.
The paper argues that the reported effect of mainland mutualism on oceanic-island species richness may be a statistical artefact of how the measure was built and modelled, so the data give no convincing evidence of an ecological effect. The latitude-only model explained little of the variation (R2 about 0.14), and because the filled-in values are predicted from latitude alone, they are by construction a curve of latitude and nonlinearly confounded with it. The linear regression ignores nonlinearities, so misfit in the latitude effect can appear as a mutualism effect. Re-estimating the island values with a random forest (a flexible tree-based method) from latitude and longitude raised R2 to about 0.39, and the effect became small and non-significant. Smooth curves instead of straight lines for the other predictors also removed the significance for either version of the measure; in the main comparison only the original combination was significant. If mutualists established less often, their island share should fall below the mainland share; instead it is about the same as the predicted mainland proportion, which does not fit a strong establishment disadvantage.
The paper concludes that the data give no convincing statistical evidence for a robust effect of mainland mutualism on island richness; it does not claim the effect is absent. The weaker island gradient is thus not shown to result from mutualism. More generally, a poorly predicted covariate confounded with another predictor, combined with unmodelled nonlinearity, can create an apparent ecological effect. The original authors replied (Delavaux et al. 2025), rejecting the conclusion and arguing that the random forest overfits unevenly distributed training data.
References
- Delavaux et al. (2024). Mutualisms weaken the latitudinal diversity gradient among oceanic islands. Nature 627, 335–339. doi:10.1038/s41586-024-07110-y
- Delavaux et al. (2025). Response to comment on Mutualism weaken the latitudinal diversity gradient among oceanic islands. arXiv. doi:10.48550/arxiv.2505.21006