Phenotypic Variation in Endangered Texas Salamanders: Application of Model-Based Clustering for Identifying Species and Hybrids

Abstract

The endangered Barton Springs and Austin blind salamanders (Eurycea sosorum and E. waterlooensis, respectively) are micro-endemics to the Barton Springs segment of the Edwards Aquifer and its contributing zone in Central Texas. Although vertically segregated within the aquifer system, both species are known from the same spring outflows and occasionally hybridize. We used geometric morphometrics and model-based clustering applied to a large sample of standardized salamander photographs to evaluate the potential for objective phenotypic assignment to either species, as well as putative hybrids. In addition to characterizing variation in head shape, our analyses inferred sets of clusters corresponding to ontogenetic series in both species but did not infer any distinct hybrid clusters. Eurycea sosorum and E. waterlooensis have distinctive head size to trunk length allometries, which contributed to the effective clustering of species, even at small body sizes. We also observed subtle, but significant, microgeographic variation in E. sosorum, suggesting the possibility of population substructuring, phenotypic plasticity, or undetected hybridization.

Reference‐free discovery of nuclear SNPs permits accurate, sensitive identification of Carya (hickory) species and hybrids

“…× brownii ) are more commonly used as wood for cooking or lumber (Thompson and Grauke, 1991 ). Although hybrid samples have been identified using non‐DNA methods such as biochemical profiling of tree bark (Likhanov et al, 2020 ) and the application of geometric morphometrics techniques (Strom et al, 2020 ), DNA‐based methods are more common, including analysis of length polymorphisms in simple sequence repeat (SSR) markers (Hanson et al, 2020 ), analysis of amplified products from restriction enzyme digestion (i.e., sequence‐characterized amplified region [SCAR] markers) (Anuntalabhochai et al, 2007 ), and even through the application of convolutional neural networks where SNPs are re‐encoded as binary images and analyzed using deep‐learning algorithms (Blischak et al, 2021 ). However, when paired with genome‐scale data sets, we have found that simple SNP profiling (i.e., binary allele matching) is a computationally tractable yet statistically robust method for hybrid identification even at low sequencing depths, and this method requires no bait capture, enzymatic digestion, or specialized processing steps prior to sequencing.…”