Material and methods
A total of 420 adult Chioglossa lusitanica was collected from 20 localities (Fig. 1). Sample sizes ranged from 7 to 51. For each individual the dorsal colouration pattern was classified on two accounts i) prominence of colour spots on the dorsum and ii) proportion of colour patches relative to colour stripes, with six ranked (0–5) colour pattern types as follows: type 0, coloured patches absent i.e., melanic; type 1, few coloured patches arranged into incomplete stripes; type 2, two uninterrupted coloured stripes, separated by a dark area of approximately the same width (typical pattern); type 3, stripes wider; type 4, central area with coloured patches diffusely distributed; and type 5, central area with coloured patches densily packed in a mosaic pattern. For a general impression of colour pattern variation in C. lusitanica see Vences (1990). Seven morphometric measurements were taken on 275 individuals representing all populations except nos. 14 and 17: snout-vent length (SVL), head (snout-gular) length (HL), head width (HW), forelimb length (FLL), hindlimb length (HLL), third ﬁnger length (TFL) and fourth toe length (FTL) (Fig. 2). Measurements were taken to the nearest 0.1 mm with digital callipers by a single observer (JA) on animals anaesthetized with MS-222. Subsequently, salamanders were released on their localities of origin.
Morphometric data were log-transformed to reduce deviations from normality and distortion effects caused by allometric relationships. Population means were compared by multivariate analysis of variance (MANOVA) with the variable ‘sex’ nested under the variable ‘locality’. All measurements showed signiﬁcant differences between populations (Wilks’ lambda = 0.098, P< 0.001; univariate results all with P< 0.001) and HL and FLL showed signiﬁcant differences between sexes (respectively, F = 2.865, P< 0.001 and F = 4.437, P< 0.001). Subsequent morphometric analyses were performed for males and females separately by principal component analysis (PCA) with the software package
Statistica/w 4.5 (StatSoft, 1993). Following conﬁrmation that size explained most of the observed variability (ﬁrst axis explained c. 54% of the variance, with variable negative correlations of 0.65– 0.83), the residuals of the regression of log<character> against logSVL were used, hereafter denoted by the symbol # linked to the variable name. Trend surface maps were generated by kriging under default settings in Surfer 6.0 (Golden Software, 1996) geostatistical software.
Morphological differentiation, expressed by pairwise euclidean distances between population means (both of size adjusted variables and scores of size adjusted PCA), was tested against the independent variables geographic distance (measured on 1:100,000 maps) and group membership by partial Mantel tests (RT 2.0; Manly, 1996), with 10,000 randomizations. Similarly, morphometric variability (population distances calculated from the variance on the PCA axes) and colour pattern variability (population distances calculated from the coefﬁcient of variation) were tested against genetic heterozygosity and hybridity (Alexandrino et al., 2000). Hybridity distinguished populations located within and outside the putative contact zone (nos. 3, 4, 6 and 7 versus the others). To evaluate the a posteriori classification of individuals in each population to either group 1 or group 2, Discriminant Analysis (DA) was performed on size adjusted morphological measurements with Statistica/w 4.5 (StatSoft, 1993). The hypothesis of association between morphological variation and latitude was tested through linear regression of PCA1 scores with distance along a south to northaxis.