Laserfiche WebLink
McELROY AND DOUGLAS-MORPHOLOGICAL VARIATION IN GILA 641 <br />tion error rate was 0.042, with nine of 215 in- <br />dividuals misclassified. No pattern was evident <br />among the misclassifications. <br />Cluster analysis provided little evidence of a <br />geographic component to hierarchical relation- <br />ships of populations (Fig. 4a). Population sam- <br />ples from Black Rocks and Westwater Canyon <br />were most similar in discriminant space; these <br />localities are separated by less than 15 river miles <br />(Fig. 1). However, the Black Rocks/Westwater <br />Canyon group formed part of a larger cluster <br />containing populations occurring in different <br />river basins and separated by over 600 river <br />miles (Yampa River/Rifle; Fig. 1). In fact, there <br />was no correlation between geographic prox- <br />imity and morphological similarity of popula- <br />tions (Mantel Z = -0.066, t = -0.215, one- <br />tailed P = 0.475). The Cataract and Desolation <br />canyon populations were distinct from other <br />populations (and from each other) under all <br />clustering methods. <br />Variation in G. cypha.-As in G. robusta, popu- <br />lations of G. cypha could be differentiated using <br />CVA, despite the presence of significant mor- <br />phological variability. PCA results for G. cypha <br />were comparable to those from G. robusta. PCI <br />explained 80.1% of the total variation, and <br />loadings on this axis reflected primarily size <br />variation among individuals. PCII-IV combined <br />accounted for only 48.6% of the variation not <br />attributable to size (i.e., not explained by PCI), <br />and at least 15 latent shape factors contained <br />significant information. There was no evidence <br />of population structure in these lower dimen- <br />sions. <br />CVA of individuals of G. cypha revealed sig- <br />nificant shape divergence among populations: <br />41 of 56 univariate comparisons showed signif- <br />icant among-group differences (P < 0.0009), <br />and multivariate tests were highly significant <br />(Wilk's Lambda = 0.0014, FYB0,439.1 = 4.264, P <br />< 0.0001). The largest univariate F values were <br />in characters associated with relative position <br />of fins and head and body depth (OA-vIA, Pu- <br />dPu, OA-IA). Each of the five canonical roots <br />captured a significant component of among- <br />group variation. The Cataract and Desolation <br />canyon populations were separated from all <br />others along CVI, whereas the Grand Canyon <br />population was distinct on CVII (Fig. 3b). As in <br />G. robusta, specimens could be accurately as- <br />signed to groups using DFA. Only three of 148 <br />individuals were misclassified (error rate = <br />0.020). <br />Hierarchical relationships of populations of <br />G. cypha did not reflect geographic proximity <br />of localities (Fig. 4b). As in G. robusta, samples <br />11 <br />Black Rocks <br />Westwater Canyon <br />Yampa River <br />Rifle <br />Debeque Canyon <br />Desolation Canyon <br />Cataract Canyon <br />Black Rocks <br />Westwater Canyon <br />Grand Canyon <br />Yampa River <br />Desolation Canyon <br />Cataract Canyon <br />Fig. 4. Hierarchical relationships among popula- <br />tions of (a) Gila robusta and (b) G. cypha. Dendrograms <br />represent the strict consensus topology derived from <br />UPGMA, complete linkage, and single linkage clus- <br />tering of generalized-squared Mahalanobis distances <br />among group means. Branch lengths are not indica- <br />tive of the true phenetic distances among nodes. In <br />both species, there is no relationship between canon- <br />ical distance and the geographic proximity of sampled <br />populations. <br />of G. cypha from Black Rocks and Westwater <br />Canyon were most similar, whereas both Des- <br />olation and Cataract canyon populations were <br />distinct from all other localities. Morphological <br />similarity was uncorrelated with geography <br />(Mantel Z = 0.003, t = 0.010, one-tailed P = <br />0.493). <br />Distinctiveness of species.-Despite high variabil- <br />ity, individuals of G. robusta and G. cypha could <br />be discriminated readily. As in intraspecific <br />comparisons, the first three shape factors ex- <br />tracted from PCA (PCII-IV) accounted for only <br />50.2% of the variation not attributable to size, <br />and at least 16 latent roots explained a signifi- <br />cant component of the total shape variation. <br />Although some structure was evident in PCA, <br />within-species clouds overlapped considerably <br />(Fig. 5a). <br />By contrast, CVA clearly separated groups <br />(Fig. 5b). Here, 46 of 56 univariate character <br />comparisons were significant (P < 0.0009), as <br />were multivariate tests (Wilk's Lambda = 0.207, <br />Fae,soe = 20.961, P < 0.001). Characters asso-