Historical connectivity and environmental filtering jointly determine the freshwater fish assemblages on Taiwan and Hainan Islands of China

Abstract

The biotas of Taiwan and Hainan Islands are of continental origin, but the manner with which historical and ecological factors shaped these insular species is still unclear. Here, we used freshwater fish as a model to fill this gap by quantifying the phylogenetic structure of the insular faunas and disentangling the relative contribution of potential drivers. Firstly, we used clustering and ordination analyses to identify regional species pools. To test whether the insular freshwater fish faunas were phylogenetically clustered or overdispersed, we calculated the net relatedness index (NRI) and the nearest taxon index (NTI). Finally, we implemented logistic regressions to disentangle the relative importance of species attributes (i.e., maximum body length, climatic niche dissimilarity [ND], and diversification) and historical connectivity (HC) in explaining the insular faunas. Our results showed that the most possible species pools of Taiwan are Zhejiang and Fujian provinces, and those of Hainan are Guangdong and Guangxi provinces. These insular faunas showed random phylogenetic structures in terms of NRI values. According to the NTI values, however, the Taiwanese fauna displayed more phylogenetic clustering, while the Hainanese one was more overdispersed. Both the standard and phylogenetic logistic regressions identified HC and climatic ND as the 2 top explanatory variables for species assemblages on these islands. Our reconstruction of the paleo-connected drainage basins provides insight into how historical processes and ecological factors interact to shape the freshwater fish fauna of the East Asian islands.

Keywords: biogeography; community assemblage; paleo-drainage; sea level change.

Figure 1.

The paleo- and current landmasses, drainage basins, and rivers of Taiwan, Hainan Island, and Mainland Asia. The map of current landmasses was downloaded from the Resource and Environment Science and Data Center of China (http://www.resdc.cn/). The map of paleo-landmass was reconstructed from ETOPO1 1 arc‐minute global relief models, with paleo-landmass indicating the exposed land bridge during the Last Glacial Maximum when the sea level was −120 m below the current sea level.

Figure 2.

Classification maps of freshwater fish communities of Taiwan, Hainan Island, and Mainland Asia. (A) Dendrograms are yielded using the UPGMA hierarchical clustering based on the phylogenetic Simpson’s dissimilarity index. The phenon line indicates the optimal cut-off location determined by the “optimal_phyloregion” function in the R package “phyloregion”. The colors of the UPGMA clusters are identical to those on (B) maps and in (C) NMDS ordinations.

Figure 3.

Phylogenetic community assemblage structure of freshwater fish on Taiwan and Hainan Islands. The phylogenetic structures were estimated using (A) the NRI and (B) the NTI within 3 (large, medium, and small) potential source pools.

Figure 4.

Regression coefficients of predictors and model diagnostics for logistic regression models explaining the occurrence of freshwater fish species on Taiwan and Hainan Islands. (A and B) Results of standard (non-phylogenetic) logistic regression for species assemblages on Taiwan (A) and Hainan Island (B). Bar lengths show that the relative importance value of each variable is calculated by the sum of the Akaike weight of models including them. Plus and minus symbols indicate the sign of model-averaged logistic regression coefficient estimates (β). Asterisks indicate the level of significance of the variables in logistic regression analyses (*P < 0.05, **P < 0.01, and ***P < 0.001). R² indicates the variance explained by the best model. (C and D) Results of phylogenetic logistic regression for species assemblages on Taiwan (C) and Hainan Island (D). Bar lengths show the mean relative importance values with a vertical bar representing standard deviations based on the calculations of 100 phylogenetic trees. Filled dots represent the median values of regression coefficient (β) estimates for phylogenetic models with an inference interval as the 2.5–97.5th percentiles for 100 phylogenetic trees. The phylogenetic correlation (α) and the variance (R²) explained by a model are shown as the mean values and standard deviation of the best models among 100 repetitions.