Page 5 - Renaud_Michaux_2007

Mandibles and molars of the wood mouse, Apodemus sylvaticus (L.)

between size estimators, retaining a sign indicator to show a
Reduction of multivariate shape variation on synthetic axes
size increase or decrease. To estimate the amount of
The shape of each outline was described by a set of Fourier morphological difference in a fully multivariate way, the
coefficients: 24 for the mandible (EFT: four FCs per six shape difference was estimated as Euclidean distance on the
harmonics, the FCs of the first harmonic being excluded from FCs. Data from Marettimo islet were excluded, firstly since
the analysis) and 18 for the M1 (RFT: two FCs per nine the population was sampled by a single animal and second
harmonics). These sets of coefficients have been analysed using because it is an intricate case of isolation from the insular
multivariate statistics. The differences between localities were population of Sicily.
tested using a multivariate analysis of variance (manova, test Finally, the physical and ecological parameters were categ-
considered: Wilks’ lambda). Canonical axes were estimated orized and their influence on the morphological divergence
using the manova analyses, and these allowed us to display the investigated using a one-way analysis of variance. As sample
morphological variability on a few synthetic shape axes. size was small (N ¼ 12) we used the Kruskall–Wallis test,
The biological interpretation of these axes relies on visual- which is a nonparametric analogue.
izing the shape changes they describe. Reconstructions can be All statistical analyses were performed using Systat 10 and
obtained from a set of FCs corresponding to scores on the NTsys-pc 2.1 (Rohlf, 2000).
synthetic axes, thereafter using an inverse Fourier transform.
Such a set of coefficients was obtained by calculating the
RESULTS
multivariate regression parameters of FCs (dependent varia-
bles) upon the relevant synthetic axis (independent variable).
Patterns of shape differentiation of the mandible and
In this case, the multivariate regression corresponds to the set
molars
of all single regressions of FCs upon the synthetic axis (Monti
et al., 2001). Significant morphological differentiation was found for both
The relationship of the synthetic shape axes with (geo- molars and mandibles (manova on Fourier coefficients:
graphical) latitude was investigated using simple linear P < 0.001). The percentage of among-group variance dis-
regression on the first two canonical axes from the manova. played by the first two axes was limited (35.3% for the
mandibles and 44.1% for the molars), but most of the
divergence expressed on subsequent axes corresponds to some
Multivariate analysis of the morphological variability
local differentiation. Only the results based on the first two
Although clearly useful for visualizing the pattern of morpho- axes are shown and discussed in detail (Fig. 2).
logical shape variability, expressing this variation on a few The pattern of mandible differentiation is shown in
synthetic axes leads to a loss of information. Therefore, the Fig. 2(a). The mandibles of the two mainland clades appear
relationships of shape with clade, size, latitude and insularity to be very similar, although some difference appears on axis
were also investigated using a range of complementary CA2. The Italian mandibles are grouped on the range of the
approaches. variation exhibited by the western European group, possibly
The relationship of shape and univariate parameters such as because of a more restricted latitudinal range.
size and latitude was investigated using multivariate regression, Insular mandible morphologies scatter around the mainland
with the Fourier coefficients as the dependent variables. As populations. Except for Port-Cros, Porquerolles and the
a significant relationship between size and shape may cause a Sicilian islet Marettimo, islands tend to display negative values
latitudinal shape variation through allometry, the residuals of along axes CA1 and CA2 (Fig. 2a). The most divergent insular
a multivariate regression of the FCs on size were also regressed populations on the first canonical plane are Porquerolles
onto latitude. A similar approach was used on mainland (along CA1), Sardinia (along CA2), Corsica and the single
samples, in order to identify a possible differentiation among specimen from Elba (along CA1). Reconstructed outlines allow
the two clades. The residuals of a regression of the FCs onto a visual interpretation of the canonical plot. The mandible
latitude were tested for clade differences. traits characterizing the insular populations are subtle. Positive
Euclidean distances were calculated among group means for differences along CA1 include mandibles with a smooth
shape and size. The relationships between various data sets alveolar region, a reduced coronoid process, and a wide and
were investigated using a comparison of the distance matrices rounded angular process. Negative deviations are displayed by
(Mantel t-test). mandibles with a pronounced alveolar region, prominent
The amount of morphological divergence on each island coronoid process and a slender angular process. The second
was compared with different physical and ecological charac- axis has massive mandibles with a straight lower border
teristics of the islands. In order to take into account the displaying positive divergence, whereas negative divergence
effect of latitude and of the various mainland source was seen in slender mandibles with an undulating outline and
populations (Michaux et al., 2002a; Gou ¨y de Bellocq et al., prominent processes.
2003), the amount of divergence was estimated as the The differentiation between the two mainland clades is
difference between the mean value of the island and the better expressed by the shape of the first upper molar (Fig. 2b).
closest mainland relative. Simple differences were calculated Sample means from mainland Italy and western Europe are

Journal of Biogeography 34, 339–355 343
ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd