The islands of Bioko, Príncipe, São Tomé and Annobón constitute the only insular system offshore the sub-Saharan Atlantic coast of Africa (Fig. 1). The islands are aligned along a volcanic fracture stretching NE–SW from Mont Kupe in inland Cameroon to St. Helena island (Terán, 1962; Eisentraut, 1963). Príncipe, São Tomé and Annobón are typical oceanic islands, whereas Bioko is a ‘land-bridge’ island, which has been periodically connected to the mainland during past climatic cycles. As a result, Bioko’s flora and fauna are relatively species-rich and poor in endemisms. The oceanic islands, in contrast, show low richness, but high levels of endemism at the specific and generic levels (Exell, 1968; Jones, 1994). The small area and relatively long distances among oceanic islands (Fig. 1) have predominantly determined colonization events from the mainland to each island, generating distinct insular communities. Following this general trend, bat communities show a high proportion of endemisms on each of the oceanic islands, and only three species (Hipposideros ruber, Eidolon helvum, and Rousettus egyptiacus) are shared by more than two of them (Juste & Ibáñez, 1994). Samples from insular populations of the fruit bats Eidolon helvum and Rousettus egyptiacus were examined. R. egyptiacus occurs in all the islands except the most-distant Annobón, whereas E. helvum is spread throughout the system. For each species, two populations from the mainland were used as reference populations, one from the Mount Cameroon area, originated from the same volcanic fracture than the islands, and the other from Río Muni, located south-west of Cameroon, and approximately equidistant to all islands (Fig. 1).
Figure 1. Geographical setting of populations of the fruit bats Eidolon helvum and Rousettus egyptiacus examined, and geographical distances between them. CA=Cameroon (Mount Cameroon area); RM=Río Muni; B=Bioko island; P=Príncipe island; ST=São Tomé island; A=Annobón island. R. egyptiacus does not occur in Annobón.
Download figure to PowerPoint
Landmark coordinates were digitized from a total of 224 skulls of Eidolon helvum (Cameroon, 44; Río Muni, 49; Bioko, 37; Príncipe, 38; São Tomé, 29; Annobón, 27) and 241 Rousettus egyptiacus (Cameroon, 66; Río Muni, 54; Bioko, 44; Príncipe, 39; São Tomé, 38) using a 3-D Reflex Microscope (Reflex Measurement Ltd, Butleigh, UK). From each skull, a set of 82 landmark points was taken across all areas and on three dimensions. From these landmarks, 74 pairs of bilateral distances were calculated following the protocol described in Juste et al. (2001).
Significance of AA and DA patterns were examined for each of the 74 characters prior to all multivariate analyses. A subsample of 31 Eidolon helvum and 30 Rousettus egyptiacus drawn at random from the total sample was analysed following the methodology recommended by Palmer (1994). Presence of AA in the samples was tested on the raw data (right-minus-left differences), by species, using the Shapiro–Wilk’s test for normality (SAS Institute, 1995). The subsamples were measured twice, and a mixed-model sides-by-individuals analysis (Palmer & Strobeck, 1986) was performed for each species. Significance of DA (the ‘sides’ component), as well as the magnitude of measurement error with respect to nondirectional asymmetry (the sides-by-individuals interaction component) were tested on each of the 74 variables. DA was statistically significant only for 10 characters of E. helvum and 10 of R. egyptiacus. The sides-by-individuals interaction term of the ANOVA was not significant for nine characters of E. helvum and 11 of R. egyptiacus. Additionally, correlation analysis found no significant relationship between asymmetry and size, and AA was not significant for any character. Results of univariate analyses are discussed in detail in Juste et al. (2001).
The main objective of this paper is to examine patterns of variation in multivariate fluctuating asymmetry across island populations and between species. Therefore, only variables with no significant DA, and significant interaction term in the sides-by-individuals ANOVA (Palmer, 1994) were used in further analyses (Juste et al., 2001). Thus the data set consisted of 55 variables for E. helvum and 57 for R. egyptiacus. Because no significant relationship with size was detected, data were not scaled by size, but only log-transformed to linearize possible allometric relationships. In order to keep adequate sample sizes in multivariate comparisons, missing values were estimated using the expectation-maximization algorithm of Little & Rubin (1987).
Because antisymmetry was not significant, univariate asymmetry in this sample represents ‘true’ fluctuating asymmetry sensuPalmer (1994), an apparently random pattern of variation between sides for a character within a population, normally distributed, and with a mean not significantly different from zero. When considering multivariate fluctuating asymmetry we extend the concept of FA to more than one variable and their correlations, with no a priori association of pattern to any process.
Absolute deviation of asymmetry values (log r – log l) from their mean was calculated for all individuals, by character, population and species. These transformed asymmetry values were used in all further analyses. Although the mean absolute deviation for any one character is not equivalent to a variance (mean squared deviation of asymmetry values), it is a measure of spread around the mean that allows for parametric univariate and multivariate comparisons of variances via Levene’s test (Schultz, 1985). A preliminary two-way multivariate analysis of variance (MANOVA) was carried out to test for significant differences in FA between subadults and adults (specimens with fully grown dentition and with bone sutures fused) and between sexes, pooling all populations within each species. No significant differences were detected, either for main factors or interaction term for any species (E. helvum: sex, F65,152=0.82, P=0.78; age, F65,152=1.29, P=0.05; interaction, F65, 152=1.02, P=0.43. R. egyptiacus: sex, F63, 172=0.82, P=0.80; age, F63, 172=1.32, P=0.08; interaction, F63,172=0.9, P=0.64). Therefore, sexes and ages were pooled in further tests.
To estimate the degree of association (integration) of FA in the skull, Spearman’s rank correlation and its significance was estimated for all possible pairs of characters, by population. Correspondence of character asymmetries across populations was assessed by calculating the Spearman’s rank correlation of mean asymmetry values of all characters for all possible pairs of populations within a species. Significance of correlations indicates the presence of a population asymmetry parameter (PAP). Additionally, we tested for pairwise associations in multivariate FA between populations using Mantel’s tests, significance of these tests also is indicative of a PAP. MANOVA was used to test for differences in levels of multivariate FA between populations. Analyses were run by species, with populations as treatments. To obtain a visualization of the trends of asymmetry being compared, the grand mean of the asymmetry values for each character was calculated, by population and species.
For species comparisons, populations were pooled into one sample by species (excluding Annobón, in which R. egyptiacus does not occur). Concordance of character asymmetries was tested on the variables common to both species (48) by calculating the mean asymmetry value of each character for each species and then the rank correlation between them. Using the same set of characters, matrices of pairwise correlations of characters were compiled for each species and the significance of their association tested using Mantel’s tests. Differences in levels of asymmetry between species were tested via MANOVA (multivariate Levene’s test); a two-way analysis species-by-population was performed to control for differences between populations within each species.
We assessed the relationships of asymmetry variation with other morphological (average trait size), genetic (allozymic variation) and geographical patterns of variation across the islands. General patterns of multivariate FA were evaluated by species by calculating pair-wise Mahalanobis distances among all possible pairs of populations. From these, a distance matrix was constructed and used with a neighbour-joining algorithm (Swofford et al., 1996) to construct unrooted trees, which summarize relationships of multivariate FAs among populations for each species. In these trees, branch lengths are proportional to the distance among populations. Similar matrices of genetic distances between pairs of populations, by species, were built from allozymic data using Rogers’ (1972) distances (Juste et al., 1996, 2000). To assess variation in mean trait size, a matrix of morphological distances was generated by taking the average of the raw measurements on right and left sides of the skull (r + l/2) for each observation and calculating the Mahalanobis distances between populations for each species. Finally, a matrix of geographical distances among localities was constructed. Genetic, morphometric and geographical distance matrices were compared against multivariate FA matrices using Mantel’s tests. A sequential Bonferroni adjustment (Rice, 1989) of α (expected Type I error for a number of simultaneous hypotheses being tested) was used to decide on the statistical significance of multiple simultaneous tests at a global α level of 0.05 by set of tests and species. Unless otherwise specified, all statistical analyses were performed in Matlab for Windows ver. 4.2c (The MathWorks Inc., 1994).