Separation of Aspidiotes species using morphometric analysis ( Coleóptera : Curculionidae )

The efficacy of morphometric characters for separating the species of the genus Aspidiotes Schoenherr, 1847, was evalu­ ated. Thirty characters were analyzed. Multivariate and univariate analyses of variance, and discriminant function analysis, all dem­ onstrated that each species is morphometrically distinguishable. The lengths of rostrum, scape, onychium, pronotum, and width and length of elytra have the maximum discriminatory power. Males and females are also morphometrically distinguishable, mainly due to differences in the widths of rostrum between pterigia and at base of pronotum, and width and length of elytra. The classification functions provided by discriminant gave the correct identification of every single specimen by sex and species. Mahalanobis’ dis­ tances between species were calculated and subjected to UPGMA clustering, to construct a dendrogram reflecting the morphometric relationships between species. This dendrogram did not correspond to the phylogenetic relationships depicted by a cladogram based on discrete characters (Sânchez-Ruiz & Alonso-Zarazaga, 1994). Some hypotheses are reviewed, which might explain this discrep­


INTRODUCTION
The systematics of the genus Aspidiotes Schoenherr, 1847 was recently revised by Sánchez-Ruiz & Alonso-Zarazaga (1994), who found the genus was made up of six species (A. anatolicus, A. cottyi, A. gonzalezi, A. larbii, A. thalassimis and A westringii).Their cladistic analysis, based on qualitative characters of the external morphology and genitalia, suggested the existence of two subgenera within the genus, Aspidiotes and Phaenognathus.The genus Aspidiotes, as presently delimited, shows an East-West Mediterranean disjunction, well known for other taxa of plants and animals (Oosterbroek & Amtzen, 1992).
Previously, Alonso-Zarazaga & Sánchez-Ruiz (1990) used multivariate morphometric analysis to divide the Iberian species A. westringii into two allopatric species: A. westringii, which occurs in southeastern Spain, and a northeastern population, later described as A. gonzalezi Sanchez-Ruiz & Alonso-Zarazaga, 1994.Differences in the external morphology and genitalia of these species were corroborated by metric characters, thus indicating the usefulness of these characters in the taxonomy of As pidiotes.In this paper we propose a new approach to the taxonomy of the genus based on an analysis of mor phometric characters.This analysis reveals new taxo nomic characters.
Multivariate morphometries has proved useful not only in resolving taxonomic problems but also in coevolution studies (Flouck, 1992), and even for phylogenetic infer ence.Sorensen & Foottit (1992) argue that multivariate morphometric methods, in particular discriminant analy sis, can be used for estimating phylogeny because, "as with cladistic methods, they evaluate partitioned variance and reflect polarity or apomorphic character states" (Sites & Willing, 1994).
Unlike in other groups of insects, there is usually no striking (visible) sexual dimorphism in shortnosed Curcu lionidae.Differences between sexes in Aspidiotes involve the entire form of the insect and are thus difficult to de fine in terms of qualitative characters.Multivariate mor phometric analysis considers simultaneously many differ ent intercorrelated characters, thus it might help to visual ize these latent, non obvious differences.
Aims of this work is: (i) evaluate morphometric charac ters for separating Aspidiotes species; (ii) determine the discriminatory characters for (a) every pair of species, and (b) males and females; (iii) to compare the classifica tion of species based on this morphometric analysis with that of Sanchez-Ruiz & Alonso-Zarazaga (1994), based on a cladistic analysis of qualitative characters.

MATERIAL AND METHODS
Table 1 shows the geographic distribution and number of specimens measured for each of the six species of genus As pidiotes.In all, 177 specimens were measured.
Thirty characters from all parts of the body of male and fe male adults were measured (Table 2), measurements were made with the aid of an ocular micrometer attached to a binocular mi croscope.Ratios or indices were not included in the analysis be cause of the difficulties created by using of ratios in mor phometric analysis (Albretch et al., 1993).Statistical analyses were performed using STATISTICA 5.1 for WINDOWS (StatSoft, 1996).The basic principles of the analyses used may be found in Sneath & Sokal (1973).
(i) We used an analysis o f variance to evaluate the efficacy of morphometric characters in the separation of Aspidiotes species.First, a one-way multivariate analysis of variance (MANOVA) was performed on the six species to determine whether statisti cally significant differences existed between species based on the entire set of characters.One-way analyses of variance (ANOVAs) were then computed for each character to evaluate whether it contributed significantly to species differences.How ever, before ascribing statistical significance, we applied the se quential Bonferroni correction to each character to avoid overes timating the significance o f particular characters in a large suite of attributes (Rice, 1989).
(ii) (a) The most important characters for discriminating be tween species in species pairs were determined by multiple dis criminant function analysis (DFA).DFA demonstrated the degree of separation in multivariate space defined by the main patterns of morphological variation among species (the discrimi nant functions).It also showed which characters contribute more to the discrimination between species.The standardized dis- the larger the standardized coefficient, the larger the weight of the variable in the function.
In addition, DFA allows for the predictive classification of specimens.The attribution of specimens to species was checked by computing the classification functions: an individual was al located to the species for which it had the highest classification score.The percentage of specimens properly classified is a measure of the diagnostic value of the set of characters (Foottit & Sorensen, 1992).
(ii) (b) A similar methodology was followed to determine the existence of sexual morphometric dimorphism in the genus As pidiotes.First, MANOVA was used to determine if significant differences existed between males and females within the genus, regardless of species.A MANOVA test was then performed on each analysed species to test whether it presented sexual dimor phism (Planned Comparisons).An exception was made for A. larbii, where statistical comparison was not possible (7 males/ 1 female).Finally, discriminant analysis was used to determine the most important characters for identifying males and females within the genus.
(iii) Finally, Mahalanobis' generalized distances (D:) were computed between all pairs of species.The square root of Ma halanobis' distance (D) for any two species represents the length of the line between the centroids of the two species in the dis criminant space (Sneath & Sokal, 1973).Mahalanobis' distances were then subjected to the clustering method UPGMA (Un weighted Pair-Group Method Arithmetic average) to construct a dendrogram reflecting the morphometric relationships between Aspidiotes species.This dendrogram was then compared with the cladogram based on qualitative characters given by Sanchez-Ruiz & Alonso-Zarazaga (1994).

RESULTS
The mean value, standard deviation, and range of varia tion for each of the thirty characters in the six species are listed in Table 3.
(i) MANOVA showed highly significant differences (Wilks' lambda150.702= 0.00004; p < 0.001) between the six species based on the entire set of characters.Moreover, subsequent ANOVAs resulted in highly sig nificant differences (p < 0.00001) between the species in each character, even after applying sequential Bonferroni correction.Thus, all characters contributed significantly to the separation of species.
(ii) (a) Discriminant function analysis provided five significant functions (x2 = 1585.089;p < 0.000001).About 94% of the variability in the data is attributable to between-species differences when all thirty variables were considered (R2 = 0.94).The first four functions ex plain 97% of the total variation in the data, which is suffi cient for the analysis (Table 4).Individual specimens are projected onto the first two discriminant functions in Fig. 1, and onto the first and third functions in Fig. 2. Because all species were clearly separated in the discriminant space defined by the first three functions, the fourth function was not used.
The first discriminant function explains 44% of total variation (Table 4), and mainly separates A. thalassinus from A. cottyi.The other four species, clustered in the middle, cannot be discriminated by this function.From the standardized coefficients (Table 4), the characters that have the greatest weight on this function (characters most discriminatory) are the lengths of scape (LSC) and pronotum (LP) and, to a lesser extent, the elytral length (LE).LSC is opposite in sign to LP and LE.In general A. tha lassinus is larger than A. cottyi.The length of the pronotum (LP), however, is very similar in both species: aver age 1. 99 mm in A. thalassinus and 1.91 mm in A. cottyi (Table 3).Likewise, the differences between the elytral length (LE) in both species (5.284-6.295mm) do not cor respond to the general differences in size.The scape, however, is much longer in A. thalassinus (1.174) than in A. cottyi (0.734).That is A. thalassinus is characterized by a proportionally shorter pronotum, shorter elytra, and a very long antennal scape, whereas A. cottyi has a longer pronotum, elongated elytra, and a shorter scape.
The second discriminant function accounts for 27% of total variation.A. larbii and A. anatolicus are clearly dis criminated by this function while the other four species are clustered in the middle (Fig. 1).The contrast between the length of rostrum (LR) and length of elytra (LE) is re sponsible lor this discrimination.A. larbii is larger than A. anatolicus, but in proportion A. anatolicus has a longer  rostrum and shorter elytra than A. larbii.This shape dif ference is reinforced by contrasts in three other variables: frons width between the eyes (WF) and elytral width (WE) against length of pronotum (LP).
As seen in Fig. 1, the discriminant space defined by the first two functions allows us to clearly separate four spe cies of the genus: A. thalassinus, A. cottyi, A. anatolicus, and A. larbii.The third discriminant function explains 17% of total variation.This function morphologically separates the two remaining species (Fig. 2): A. gonzalezi from A. westringii, by an increase in the length of rostrum (LR) and decrease in the length of onychium (LON).
The DFA classification functions, based on linear com binations of the original variables, correctly identified all specimens, thus demonstrating the efficacy of this set of morphometric characters for identifying species of As pidiotes.
(ii) (b) MANOVA revealed highly significant differ ences (Wilks' lambdaso.i'ji= 0.081765; p < 0.001) be tween males and females within the genus Aspidiotes.Subsequent MANOVAs revealed sexual dimorphism in five species: A. thalassinus, A. cottyi, A. anatolicus, A. westringii and A. gonzalezi (p < 0.001).A discriminant analysis, to separate specimens by sex, provided one dis criminant function, allowing for complete discrimination of males and females within the genus (Fig. 3).Based on the standardized coefficients, the weight of this function is mainly dependent on: width of pronotum at base (WPB), width of elytra (WE), length of elytra (LE), and width of rostrum on pterigia (WRP) (Table 5).Females have wider, longer elytra and a wider pronotum than males whereas males have a wider rostrum.
Table 6 shows classification functions that separate specimens by sex and the percentage of correct attribu tions (100%).These classification functions can serve as an additional diagnostic tool for determining the sex of specimens.This may substitute the study of genitalia, which has been often necessary in order to distinguish males from females in Aspidiotes.

Statistical reliability of our results
Because the specimen to variable ratio is relatively low (=6), our results may be statistically unreliable.Typically, a minimum ratio of 10-20 is considered necessary for re liability.The number of specimens was limited and any a priori selection among the 30 characters, obtained after a preliminary study, was not feasible, as all of them dif fered significantly between species (p < 0.00001).
As a check of the accuracy of our results, we performed a multivariate analysis using a reduced set of 15 charac ters.This included those characters that proved the best discriminators among species in the previous analysis; we also excluded those characters that were redundant (as judged by the tolerance values), and summed others into a single character (e.g., LT1 + LT2 + LT3).In addition, re sults from this analysis (having greater reliability but poorer resolution), if consistent with the first 30characters study, would provide strong support for our conclusions.The results of this analysis conformed with those of the previous analysis.Although there was a loss of resolution (i.e., morphometric differences between spe cies were lower), both the characters and the species dis criminated were the same (Fig. 5).We believe that this two-step procedure gets over the problem of limited num bers of specimens.

Character efficacy
On the basis of this study it is evident that morphomet ric characters can be used to separate species within the genus Aspidiotes.The set o f morphometric characters herein analysed has proved effective in species discrimi nation, in particular, lengths of rostrum, scape, onychium and pronotum, and width and length of elytra.
Despite differences in their average values, the ranges in morphometric characters overlapped to some extent be tween species (Table 3).No character alone can be used for full discrimination.That is, species can only be sepa rated on the basis of all the characters.Thus, taxonomic discrimination requires a multivariate approach: species overlap when characters are used individually but become distinct entities when many characters are considered jointly (Foottit & Sorensen, 1992).This is a consequence of the intercorrelation of characters, itself derived from the epistatic or pleiotropic interactions among genes cod ing them (Leamy & Sustarsic, 1978;Atchley et al., 1982).
The genus Aspidiotes shows clear sexual morphometric dimorphism.Differences between sexes occur in the width of rostrum on pterigia, width of pronotum at base, and width and length of elytra.The last three characters differ in size between males and females, as in other gen era of Curculionidae.Females are generally larger than males.This size difference, however, is more difficult to appreciate in Aspidiotes without the help of multivariate analysis.
In addition, the classification functions provided by dis criminant analysis can be used as a diagnostic tool when discrete characters are unreliable.They have the advan tage of identifying "problematic" specimens objectively, and their efficacy can be evaluated by the percentage of correct identifications.

Systematic implications
The comparison o f the dendrogram in Fig. 4, portraying the morphometric relationships between species, with the cladogram in Fig. 6, representing their phylogenetic rela tionships (Sanchez-Ruiz & Alonso-Zarazaga, 1994), show notable differences.The species closely clustered in the dendrogram, such as A. anatolicus-A.cottyi or A. thalassinus-A.gonzalezi, are very far apart in the clado gram, and are even included in different subgenera, Phaenognathus and Aspidiotes s. str., respectively (San chez-Ruiz & Alonso-Zarazaga, 1994).Conversely, the species that are closely related in the cladogram and be long to the same subgenus (A.anatolicus-A.thalassiniis) showed the greatest morphometric differences in the phenogram, even greater than in species from different sub genera.The only exception occurs in the pair A. Iarbii-A.westringii, which appear closely related in both figures.
This tends to contradict the claim of Sorensen & Foottit (1992) about the potential use of discriminant function analysis in phylogenetic inference.These authors claim that the DFA-based dendrogram reflects not only mor phometric similarity but also the phylogenetic relation ships between species.Using Lande's (1979) phenotypic model for multivariate evolution, discriminant functions would represent "historical gradients of selective pressure that the species have been exposed to during their com mon evolutionary history" (Sorensen & Foottit, 1992).This approach has been criticized, however, because of the lack of biological meaning of discriminant functions (Crespi & Bookstein, 1989;Crespi, 1992).To date, the DFA model has been used several times (Schluter, 1984;Sorensen, 1987;Wood & Pesek, 1992;Simon, 1992).In these studies, the DFA-based dendrograms largely cor roborated previous cladograms for the same groups.In our case, there was no congruence between the relation ships depicted by the cladogram and the phenogram.Be cause phylogenetic arguments have a more sound theo retical base than phenetics, it is more likely that the clado gram reflects the phylogenetic relationships between spe cies of Aspidiotes whereas the phenogram only depicts their morphometric relationships.
The comparison of the distributions of the species in both figures revealed other important differences.The species that are closely related in the cladogram have similar geographic distributions.This suggests that the species could have originated from a succesion of vicari ance events on a widespread ancestor, with every vicari ance being followed by a spéciation event (Sanchez-Ruiz & Alonso-Zarazaga, 1994).In the phenogram, in constrast, those species occurring in the same or adjacent geographical areas are separated.This is the case for the species pairs A. anatolicus (Turkey)-^, thalassiniis (Greece + Turkey), and A. cottyi (Morocco)-zl.larbii (Morocco).In each pair, both species belong to the same subgenus, Phaenognathus or Aspidiotes, respectively (Fig. 6).The only apparent exception to this pattern is the species-pair A. Iarbii-A.westringii, which appears closely clustered both in the cladogram and in the phenogram.They have allopatric geographic distributions: A. west ringii occurs in southeastern Spain whereas A. larbii is apparently restricted to the westernmost foothills of the High Allas (Mogador) (Sanchez-Ruiz & Alonso-Zara zaga, 1994).
Therefore, two conclusions can be drawn from a com parison of the DFA-phenogram and the cladogram of As pidiotes: (1) The contradictory form of the species relationships depicted by both methods.The most similar species based on discrete characters are in terms of mor phometric characters the most different.(2) The largest morphometric differences were between closely related species of Aspidiotes (i.e., belonging to the same subge nus), which share a similar geographic distribution (sympatric species).
Morphometric changes are evolutionarily less expen sive than discrete changes, which imply the modification or disappearance of a structure.The latter necessarily re quire disassociations or rearrangements of genetic link ages so discrete changes are more expensive from an en ergetic viewpoint, and evolutionarily constrained (Sorensen & Foottit, 1992).Morphometric changes only require arrangements of a few pleiotropically correlated genes.Thus, a large morphometric divergence could sim ply arise from a small number of genetic divergences (Atchley et al., 1982).
For species belonging to the same subgenus, that share very similar genetic covariance matrices and have the same geographic distribution (i.e., subjected to similar se lection pressures), "morphometric variance may be the last and easiest way to diverge during evolution because of lower evolutionary energy constraints" (Sorensen, 1991).Therefore, morphometric differences have proved to be more important for closely related and sympatric species of Aspidiotes whereas divergence in discrete char acters is more important in the case of distantly related species (belonging to different subgenera).Thus, the phenogram in Fig. 4 mainly reflects the recent morphometric divergence between closely related, sympatric species of Aspidiotes whereas the cladogram in Fig. 6 would actu ally reflect the evolution of the genus.
Fig. 1.Plot of all Aspidiotes specimens onto the first and second discriminant functions based on a set of 30 morphometric charac ters.Circles include 95% of specimens in each species.

Fig. 5 .
Fig. 5. Plot of all Aspidiotes specimens onto the first and second discriminant functions based on a reduced set of 15 morphomet ric characters (see text).
that a considerable distance (D2 = 130) separates the first two clusters, indicating that A. anatolicus and A. cottyi are very different morphometrically from the other spe cies, and also very different from each other (D2 = 108).

Table 1 .
Distribution and sample size of each species of Aspidiotes.
T a b l e 2. List of characters used in this study.

Table 3 .
Mean value, standard deviation and range of variation of the characters measured in each species.
Taule 5. Standardized coefficients of the Fisher's linear dis criminant function separating specimens of Aspidiotes by sex.In bold, characters with the greatest weight in the function.