Different traits dominate evolution at early and late stages of adaptive radiation

Adaptive radiation (AR), a process of rapid speciation and ecomorphological diversification, played an important role in generating past and contemporary global biodiversity. An unsolved question is what maintains high rates of speciation during AR, a phenomenon we call “speciation paradox”. One possible explanation for resolving this paradox is a sequential trait evolution, i.e., a series of ecological diversifications, which enables evolving lineages to fully and more effectively exploit the ecological space. We tested this hypothesis using the highly diverse subterranean amphipod genus Niphargus. Niphargus shows distinct signatures of adaptive radiation both at the genus level and at the level of four larger clades. Our analysis revealed decoupled evolution of habitat-related traits and trophic-biology-related traits. Moreover, on a genus level, we found the evidence that AR commences with a tight association between speciation rates and the dynamics of habitat-related traits. At a later stage, speciation dynamics become associated with diversification of trophic-biology-related traits. This suggests that the dependence of macroevolutionary rates in this group switches among niche axes before saturation, resulting in prolonged high speciation rates during AR.

Methods
We assembled an extensive dataset on species phylogeny, ecology, and functional morphological traits across the adaptive radiation of Niphargus. We used the molecular taxonomic units (MOTUs) and phylogeny produced in Delić et al. (2025), and morphological and ecological data of analysed species, produced in Premate et al. (2024).

The analyses consist of three parts: decoupled diversification, states-dependent speciation and extinction, and speciation rate modelling. All analyses were first conducted across the entire genus. Next, we looked for independent replicates of diversification within the entire AR. We selected four large monophyletic clades with diversification patterns consistent with AR predictions (Borko et al., 2021).

1) DECOUPLED DIVERSIFICATION

To test whether diversification bursts of the functional trait classes differ in time, we analyzed the dynamics of trait space filling through time for each functional trait class separately (i.e., traits associated with habitat (namely habitatPC1 and habitatPC2) and traits linked to feeding abilities (namely trophicP and trophicD)). For each trait class separately, we first reconstructed ancestral states using the best model of trait evolution, using R packages phytools (Revell, 2024) and mvMORPH (Clavel et al., 2015). Next, we used the best model of trait evolution to reconstruct ancestral states of traits (Clavel et al., 2015). We analysed the course of evolution for each functional trait class by reconstruction of morphospace expansion through time, following the procedure of Ronco et al. (2021).

2) STATES-DEPENDENT SPECIATION AND EXTINCTION

We fitted maximum likelihood SSE models (State-dependent Speciation and Extinction; Maddisson et al., 2007, Herrera-Alsina et al., 2019) where per-lineage speciation rates depended on the particular state of a given trait (i.e., traits associated with habitat (namely habitatPC1 and habitatPC2) and traits linked to feeding abilities (namely trophicP and trophicD)). We modelled trait-class dynamics as the change from one state to another over time.

3) SPECIATION RATE MODELLING
To model speciation rates through time, we used birth-death diffusion (BDD) models, using Julia package Tapestree (Quintero et al., 2024).