Umbra

Robert E. Schmidt & Robert A. Daniels, 2006, Hybridization in Umbridae in the Hudson River, New York, with Designation of Neotypes for Umbra limi and Umbra pygmaea., Zootaxa 1113, pp. 1-20 : 5-9

[[ Genus Umbra ]]

Results

We examined 47 Umbra pygmaea specimens, 77 Umbra limi , and compared them to 35 Umbra sp. collected from Manitou Marsh. Umbra limi and U. pygmaea are very similar species, distinguishable (Table 1) by mean meristic (t-test) and morphometric (MannWhitney U test) measurements, all of which were statistically significant except for the predorsal length.

Using ANOVA, we found significant differences among the groups in all mensural and meristic variables (Table 2). An examination of the means indicated that the observed differences were due to the larger size of U. limi specimens. Because the mensural variables were highly correlated with SL (Table 2), the ANOVA may have established significance based on this size relationship. The differences among the meristic characteristics are significant and not related to size. There also were differences (p<0.05) among groups in four of the six ratios using the Kruskal-Wallis ANOVA (Table 2). By using the ratios, we moderate, to some extent, the effect of SL. These means show that the differences are not due to the larger U. limi , and indicate that there are shape differences in the taxa unrelated to size.

A second way to control for the effect of size is to use ANCOVA, with SL as the covariate. The overall test was significant (Wilks’ λ = 0.31, p <0.001); SL is related to the dependent variables. Differences among the taxa were significant (p <0.01) for each mensural variable except anal-fin base. This also suggests that there are differences in shape among the taxa irrespective of size. Umbra limi has a relatively larger head than U. pygmaea and the hybrid is intermediate. This relationship holds for all four head measurements (head length, snout length, and interorbital width, and, to a lesser extent, orbit diameter) and predorsal length.

The first two principal components explain 90% of the variance in the data (Table 3). Separation of the taxa is incomplete, but visible (Fig. 2A). Individuals of U. limi tended to have higher, often positive scores on PCA2 whereas the scores of U. pygmaea were generally negative. The hybrids are scattered throughout. All variables had positive loadings on PCA1, with snout length having the highest loading. PCA1 explained 82% of the variation in the data. This component can be interpreted as a generalized size component (Jolliffe 1986), however when the component scores are plotted against SL some separation among the groups emerges (Fig. 2B). This suggests that some shape information is present in PCA1, i.e., the taxa differ in shape at similar life stages. The second principal component explains 8% of the variation in the data and anal-fin base is the characteristic with the highest loading, which is negative. This component also has characteristics of a size component (Pimentel 1979). When the arcsine-transformed ratios, in an attempt to reduce the influence of SL, are used in the analysis, slightly better separation among the groups is achieved (Fig. 2C). The combination of these two principal components accounts for 59% of the total variance (Table 3). Head length/SL and snout length/SL have high positive loading on the first component; anal-base length/SL has a high positive loading and predorsal length/SL loads with a high negative score on the second component. The sheared PCA2 when plotted against PCA1 offers little more discrimination (Fig. 2D). In short, the PCA suggests that the two species, based on their morphometrics, are very similar. Despite their similarity, the mensural characteristics can separate the two species into two slightly overlapping groups. Characteristics of the hybrid are intermediate and the differences are related to changes in shape that are not the result of changes in size.

Because PCA suggested that we had three groups, although very closely related, represented in the data, we entered nine variables into a forward stepwise discriminant function analysis to assess whether the groups were different. Both discriminant functions 1 and 2 were significant (χ2 = 255 and 80, respectively, p <0.005). The first discriminant function is weighted most heavily by head length and, to a lesser extent, interorbital width and predorsal length (Table 3). This function also accounts for 76% of the discriminatory power of the analysis. Head length and interorbital width make the greatest contribution to the second function as well, which accounts for the remaining 24% of the variance. The first discriminant function largely separated U. limi , with negative scores, from U. pygmaea and the hybrids, which have positive scores. The second function discriminates between the hybrid, with negative scores, and the two species, which generally have positive scores (Fig. 3).

The discriminant function correctly classified 89% of the individuals used in the study (Table 4). Hybrid individuals were the ones most likely to be misclassified and only 79% were correctly assigned. Separation of the two species was much better, with only 4 individuals placed into the wrong species. The results were similar using the jackknife technique: 85% of the individuals were assigned to the correct group (Table 4). This suggests that the bias in estimating these statistics resulting from using the values of our study sample in the DFA is low.

We interpret these data as demonstrating hybridization between these closely related species in Manitou Marsh. The tidal Hudson River is the only place where the range of these species overlaps and therefore the only place where hybridization could occur.