Dawkinsia

Sudasinghe, Hiranya, Raghavan, Rajeev, Dahanukar, Neelesh, Pethiyagoda, Rohan, Rüber, Lukas & Meegaskumbura, Madhava, 2021, Diversification and biogeography of Dawkinsia (Teleostei: Cyprinidae) in the Western Ghats-Sri Lanka biodiversity hotspot, Organisms Diversity & Evolution (New York, N. Y.) 21 (4), pp. 795-820 : 801

Monophyly of Dawkinsia

The ML and BI analyses based on mtDNA did not recover the monophyly of Dawkinsia . Instead, Sahyadria , another genus of Smiliogastrinae confined to west-flowing rivers of southern peninsular India, was recovered as the sister group of the ‘filamentosa group’, while the ‘assimilis group’ was recovered as the sister group of Sahyadria + ‘filamentosa group’ with strong node support. However, there are no synapomorphies, which serve to distinguish the 'assimilis group' and the 'filamentosa group'. Both these groups share with Sahyadria the distinctive barred colour pattern of juveniles, and adults of both Dawkinsia species groups share several plesiomorphies by which they differ from Sahyadria (see Table S3). Therefore, for the purpose of this study, we carry out two different timing analyses: (1) based on the topology obtained from the ML and BI phylogenetic analyses for the mtDNA dataset, where Dawkinsia is paraphyletic with respect to Sahyadria ; and (2) by imposing a prior topological constraint for the monophyly of Dawkinsia .

The divergence timings within Dawkinsia were estimated based on the concatenated mitochondrial cytb + cox1 (1719 bp) dataset in BEAST 2 ( Bouckaert et al., 2014). We used primarily the samples of Katwate et al. (2020b) for which both cytb and cox1 sequences were available, supplemented by the newly generated sequences in the present study. We carried out the divergence timing analysis for two datasets. The first includes only interspecific data (the 15-taxon dataset), which was also used in the ancestral-range estimations. The second dataset includes both intra- and interspecific data (58-taxon dataset) which was used to estimate the divergence timing between the Sri Lankan and Indian lineages of D. filamentosa .

The optimal substitution model for each dataset was determined in PartitionFinder 2, providing the model as “beast”. In accounting for over-parameterization, a single partition per gene was given as the starting prior in PartitionFinder 2. A log likelihood ratio test with and without enforcing the molecular clock in MEGA rejected the null hypothesis for the 58-taxon dataset, though not for the 15-taxon dataset. Therefore, a relaxed clock under lognormal distribution was used for the 58-taxon dataset, while a strict clock was used for the 15-taxon dataset. Given the narrow taxonomic focus of our study and the fact that many of the deeper phylogenetic relationships within Cypriniformes are poorly resolved ( Tan & Armbruster, 2018), we decided against the use of multiple fossils belonging to several cyprinid groups that have been used as calibration points in recent studies and omitted all taxa except for Dawkinsia and Sahyadria from this analysis. We used the average cyprinid cytb substitution rate of 0.0082 substitutions per site per million years to calibrate the cytb clock rate ( Rüber et al., 2004, 2007); the cox1 substitution rate was estimated relative to that of cytb. This substitution rate for cytb has been derived in reference to European cyprinids, based on reliably dated geological events ( Zardoya & Doadrio, 1999). Both the 58- and 15-taxon datasets were recognized as a single partition by PartitionFinder 2: therefore, the substitution models were linked in both analyses, while the clock models were unliked between cytb and cox1. A Yule pure-birth model was used as the tree prior for the 15-taxon dataset. Because it contains a mixture of inter- and intraspecies taxon sampling, the 58-taxon dataset violates the assumptions of both the speciation- and coalescent-based tree priors. Therefore, we used a Bayesian skyline coalescent tree prior, which usually produces stable results across such datasets ( Ritchie et al., 2016). Two independent runs consisting of 100 million generations each were implemented, with the sampling interval of the Markov Chain Monte Carlo (MCMC) chain set to every 5000 generations. Tracer was used to determine the convergence of the two runs and assured that ESS> 200 for the combined run. The first 10% generations were discarded as burn-in. The two runs were then combined using LogCombiner, and a maximum clade credibility (MCC) tree constructed using the posterior sample of trees by TreeAnnotator, visualized using FigTree v1.4.3.