Morphological and allozyme variation of

1994). The island populations of E. helvum are compared with a population from

R´ıo Muni, a small continental region southwest of Cameroon that is included in the Central African rainforest, and is approximately equidistant to all the islands.

A total of 12 wing, 13 cranial, and 14 dental variables (Appendix 1) were measured by the senior author ( JJ) using a digital caliper with a precision of 0.1 mm. Measurements follow Bergmans (1979), with the exception of the mandibular angle (MA) defined as the angle between the mandibular ramus and the coronoid process, and obtained using a protractor with a precision of 2°. For each data set (wing, cranial and dental), data were log-transformed to linearize possible allometric relationships. Missing values were estimated using the expectation-maximization method of Little & Rubin (1987), which optimizes the missing values as a set by stabilizing the covariance matrix. Only full-grown individuals—according to wing bones and cranial suture criteria (Anthony, 1988)—were considered in the following statistical analyses. After inspecting the data for normality, overall geographic variation and presence of sexual dimorphism in the populations were evaluated using 2-way MANOVAs by data set. Further analyses dealt with the sexes separately. Principal component analysis (PCA) on the covariance matrices was used to assess the variation among populations, since it does not presume any grouping of data. The relative discriminatory power of each variable for each data set was assessed by a stepwise discriminant analysis (DFA) designed to introduce variables so as to maximize at each step Lawley–Hotelling’s trace (=Rao’s V). This measure is proportional to the mean Mahalanobis distance. For each data set, the four variables with the highest discriminant power were pooled by sex into the final matrix of the

12 variables selected. PCAs on these matrices were used to confirm and visualize the differences among populations. Mahalanobis’ D² measured the multivariate distances among them, and the values were tested for significance (H₀: D²=0) based on an F statistic. Neighbour-joining trees on the D² distance matrices were checked for patterns of geographic relationships. A sequential Bonferroni procedure (Rice,

1989) was used to adjust the statistical significance of multiple simultaneous tests to a table-wide or level of 0.05. All statistical analyses were performed in Matlab for Windows ver. 4.2c (MathWorks, Inc., 1994). Matlab statistical functions and script files are available from the authors upon request.

Homogenates of the samples were run in horizontal starch gel electrophoresis, according to standard procedures (Pasteur et al., 1987), to assess genetic variation at 34 presumptive loci encoding 25 enzymatic systems (Appendix 3). The loci Est-

Data were analysed with the BIOSYS-1 package (Swofford & Selander, 1989). Genetic variability was estimated as the mean number of alleles per locus, percentage of polymorphic (95% criterion), loci (P) observed (H_o) and expected heterozygosities per locus. Since no comparisons yielded observed values of heterozygosity deviating significantly from expected Hardy–Weinberg equilibrium, only values for observed heterozygosity are presented. Genetic relatedness among populations was evaluated using Nei’s (1978) unbiased distances (D_N) and modified (Wright, 1978) Rogers’ (1972) distance (D_R).

Differences in genetic structure among populations were estimated from F_ST values (Wright, 1965). Gene flow between populations (Nm) was estimated directly from the F_ST values using Wright’s method under the island model without selection or mutation (Wright, 1965), and from Slatkin’s (1985) private alleles method after correcting for sample sizes. A mid-point rooted phenogram was constructed from the matrix of the modified Rogers’ genetic distances using the distance-Wagner procedure, which allows for the detection of different rates of genetic evolution. Levels of association between the matrices of genetic (D_R) distances, of morphological (D²) distances, and of the minimum geographic distances (km) among populations, were assessed by Mantel’s tests.

RESULTS
Morphological analyses

71.1% (dental variables for females) and 84.2% (wing variables for males) of the variation of the data. According to the cumulative values of Rao’s V, the skull data set showed the largest differences among groups. The values of Rao’s V were larger in females for all data sets (Table 2), which indicates that between-group differences were consistently larger among females than among males.

The first three components of the PCAs on the 12 selected variables explained

74.5% and 75.6% of the total variation for males and females respectively. For both sexes, the plots of the scores on the first two axes placed the Annobo´ n population clearly apart from the rest (Fig. 2). The remaining populations overlapped to various degrees. Nevertheless, differences in centroid position indicate a certain degree of differentiation among them (Fig. 2). For both sexes, the largest Mahalanobis’ distances

Effect	Wilks’ 2	F	DF	P
Wing Population	0.211	4.052	48/391.1	<0.001∗
Sex	0.832	1.693	12/101	0.0792
Population∗Sex	0.619	1.078	48/391.1	0.3424
Skull Population	0.122	4.950	56/387.2	<0.001∗
Sex	0.638	3.998	14/99	<0.001∗
Population∗Sex	0.587	1.014	56/387.2	0.4514
Dentition Population	0.248	2.997	56/387.2	<0.001∗
Sex	0.779	2.002	14/99	0.0249
Population∗Sex	0.525	1.247	56/387.2	0.1205

were between Annobo´ n and all other populations; all these values were highly significant (Table 3). The shortest distance was between the mainland population and that of Bioko. The Neighbour-joining method produced identical topologies for males and females. The mainland and Bioko clustered together, and were fol- lowed—in a geographically sound pattern—by the other islands, with the population of Annobo´ n branching away from the rest at the base of the tree (Fig. 3).

The percentages of polymorphic loci (P) were very similar among populations and ranged from 23.53% (Pr´ıncipe) to 29.41% (mainland and Bioko). Mean observed heterozygosity (H_o) ranged from 0.045 in the mainland population to 0.066 in the population of Pr´ıncipe (Table 5). There was no significant correlation between P and H_o. The mean Wright’s fixation index (F_ST) among populations was 0.153, and the corresponding Nm was 1.38, suggesting relatively high overall gene flow among populations. In pairwise comparisons, the lowest F_ST values were between Bioko and R´ıo Muni (F_ST=0.029), and between Sa˜ o Tome´ and Pr´ıncipe (F_ST=0.033). The theoretical migrants per generation varied from 8.37 between Bioko and R´ıo Muni to 1.06 between R´ıo Muni and Annobo´ n (Table 6). The populations of Bioko and R´ıo Muni showed the lowest genetic divergence, and Annobo´ n showed the largest values with respect to the rest of the populations (Table 7). The Wagner’s

See Appendix 1 for acronyms of the variables

Males Females Males Females Males Females

Var	Rao’s V	Var	Rao’s V	Var	Rao’s V	Var	Rao’s V	Var	Rao’s V	Var	Rao’s V
FA	1.33	IIMC	1.45	ZB	1.92	GSL	2.57	BM2	0.49	BM2	0.90
IVF2	1.67	IVMC	1.86	C1-C1	2.45	M1-M1	3.49	LP3	0.71	LP3	1.69
IVF1	2.01	IIIMC	2.43	ML	2.96	RL	4.04	LP4	0.85	BM1	2.18
VMC	2.32	IIIF2	2.69	RL	3.40	MH	4.53	LM1	0.99	LM1	3.06
VF1	2.48	FA	2.93	GSL	3.80	ZB	5.11	LP4	1.07	LP4	3.52
IIF3	2.63	IVF1	3.17	PL	4.15	C1-C1	5.65	BP3	1.15	BM1	4.16
VF2	2.75	IIF1	3.46	MH	4.59	BCB	6.17	BP3	1.26	LM2	4.49
IIF2	2.81	VF2	3.73	C1-M1	4.98	C1-M2	6.49	LP3	1.33	BP4	4.76
IIF1	2.87	IVF2	3.92	CBL	5.28	ML	6.96	BP4	1.41	BP3	4.96
IVMC	2.91	VMC	4.00	BCB	5.59	MA	7.62	LM2	1.47	BP3	5.17
IIIMC	2.96	IIIF3	4.07	M1-M1	5.85	CBL	7.85	BM1	1.54	LP4	5.31
IIMC	3.01	VF1	4.09	IOB	6.03	PL	8.08	BP4	1.60	LP3	5.44
—	—	—	—	MA	6.11	C1-M1	8.23	BM1	1.66	BP4	5.56
—	—	—	—	C1-M2	6.19	IOB	8.49	LM1	1.69	LM1	5.63

^B *
_{* + +}
^ST**+ * ^* _P+ RM

*

9.6

9.8 10

PC1

10.2

–0.4 ^B

*

–0.5


–0.6

–0.7

A+
14.5%
_* *

* P₊ ^* _*

B
⁺ + RM

+
_*

^ST *
*
*

–0.8
52.8%
9.5

*

9.7 9.9

PC1

10.1

Figure 2. Representation of the two first PCs of selected variables, by sex, (A, males; B, females) for the populations of Eidolon helvum studied from the Gulf of Guinea. Plots show 95% confidence ellipses and centroids (+) for each population. Percentage of variation explained by each variable is in the lower left corner. (Χ) R´ıo Muni (RM); (∗) Bioko (B); (Ε) Pr´ıncipe (P); (×) Sa˜ o Tome´ (ST); (Μ) Annobo´ n (A).

phenogram based on these distances showed a cluster including the mainland and Bioko populations, and another group comprising Pr´ıncipe and Sa˜ o Tome´. The Annobo´ n population stood by itself at the base of the tree (Fig. 4). Mantel’s tests showed significant correlation between the matrices of Rogers’ genetic distances and morphological distances (males: r=0.84, P=0.015; females: r=0.85, P=0.017); and between matrices of genetic and geographic distances (r=0.75, P=0.010). Correlations were highly significant between geographic and morphological matrices (males: r=0.82, P=0.005; females: r=0.79, P=0.006).