Distinguishing between Leptidea sinapis and L. reali (Lepidoptera: Pieridae) using a morphometric...

40 Accepted by M. Toliver: 23 April 2008; published: 9 Jul. 2008

ZOOTAXAISSN 1175-5326 (print edition)

Zootaxa 1819: 40–54 (2008) www.mapress.com/zootaxa/

Distinguishing between Leptidea sinapis and L. reali (Lepidoptera: Pieridae) using a morphometric approach: impact of measurement error onthe discriminative characters

MICHELE FUMIVia Pompili, 34, 06122 Perugia, Italy. E-mail: michele.fumi@tiscali.it

Abstract

A morphometric approach was used to test the possibility of discriminating between L. sinapis and L. reali by taking intoaccount some new genitalic characters in addition to those used in previous surveys. Principal component analysis, per-formed on the size-and-shape data sets and on the size-adjusted data sets, has allowed two completely separate morpho-types to be detected, both in males and in females. Discriminant analysis has confirmed the separation of previouslydetected morphotypes and has correctly classified 100% of the specimens in both sexes with six discriminative charac-ters being identified in males and two in females. However, some of these discriminative characters were not consideredreliable enough because of the high associated measurement error and the scarce discriminative power. Reliable discrim-inative characters were: vinculum (≈ valve) width, length of phallus (≈ aedoeagus ≈ aedeagus), saccus and uncus inmales and ductus bursae length in females. The main topics discussed are: a comparison of the discriminative characterswith previous studies, the sources of measurement error and the devices used to reduce it, as well as the between andwithin-species variability of the characters.

Key words: sibling species, repeatability, size-adjustments, principal component analysis, discriminant analysis

Introduction

Since Lorkovic (1993) reported that males of Leptidea sinapis (Linnaeus, 1758) and those of Leptidea reali(Reissinger, 1990) could be identified by means of genitalic measurements, because they distinguish betweentwo morphotypes, many surveys have dealt with these two species (Vila et al. 2003 and Beneš et al. 2003 pro-vide a full list of the published papers). Many of these surveys, which are based exclusively on a morphologi-cal approach, confirm the existence of two separate morphotypes (interpreted as probable distinct siblingspecies), both in males and females (Laštůvka et al. 1995; Ábrahám 1996; Embacher 1996; Karsholt 1999;Mazel & Leestmans 1999; Nelson et al. 2001; Gianti & Gallo 2002; Freese & Fiedler 2004; Cupedo & Hoen2006). However, some uncertainties in assigning some individuals, mainly males, to L. sinapis or L. reali havearisen in some studies (Neumayr & Segerer 1995; Göhl & Buchsbaum 1996, 1997; Hauser 1997). Theseuncertainties have also contributed to maintaining doubts regarding the effective reproductive isolation of thetwo species (Hauser 1997; Kudrna 2001). Freese and Fiedler (2002) stated that pre-mating isolation processesmay exist between L. sinapis and L. reali and Martin et al. (2003) demonstrated the lacking of gene flowbetween the two species.

It seems that much of the still existing uncertainty in classifying some individuals of the Leptidea sinapisspecies complex on the basis of morphological characters is the result of the methods used in previous studies.In many instances the sample size was too small (Lorkovic 1993; Skalski 1995; Embacher 1996; Ábrahám1996; Hauser 1997; Karsholt 1999; Mazel & Leestmans 1999; Nelson et al. 2001) and hence did not ade-

TERM OF USEThis pdf is provided by Magnolia Press for private/research use. Commercial sale or deposition in a public library or website site is prohibited.

quately depict the within-population variability (in several papers no data were provided, making comparisonimpossible). In males, characters other than aedoeagus and saccus length were seldom taken into account(Hauser 1997; Mazel & Leestmans 1999; Martin et al. 2003; Vila et al. 2003). To date a morphometricapproach has still not been used to solve the classification difficulties. In the previous surveys, the measure-ment error associated with each character examined was never evaluated, despite the fact that if the error islarge it may produce a Type-II statistical error (Yezerinac et al. 1992).

This study on the L. sinapis species complex focuses on the following aims:i) to use genitalic characters, particularly for males, which have rarely been taken into account in the previous

surveys, as well as some new ones, in an attempt to distinguish between the two species;ii) to use a “traditional” morphometric approach (Adams et al. 2004) to highlight the occurrence of two sepa-

rate morphotypes, estimate the percentage of correctly classified specimens and identify the characters thatsubstantially contribute to distinguishing between L. sinapis and L. reali;

iii) to estimate the measurement error associated with each character examined for both species.

Material and methods

Specimens examined. Four-hundred-eighty-seven specimens (320 males and 167 females) belonging to theL. sinapis species complex were examined; all specimens were collected during samplings carried out inexisting habitats located between the regions of Umbria and the Marches (Central Italy) (Fig. 1). A few of thespecimens were captured within the same area, but in different locations. The specimens examined are pre-served in the author's collection.

FIGURE 1. Map showing sampled localities in central Italy, (�) extensively sampled localities, (�) occasional cap-

tures, (1) Gubbio, (2) Umbertide, (3) Sibillini Mountains, (4) Visso, (5) S. Anatolia di Narco, (6) Poggiodomo, (7) Cor-

ciano, (8) Perugia, (9) Assisi, (10) Subasio Mount, (11) Norcia.

On the basis of the visual inspection of a bivariate scatter plot of the aedoeagus and saccus length formale specimens (Fig. 2) and a histogram of ductus bursae length for females (data not shown), the classifica-tion tools used in many previous studies, the sample (males and females separately) has proved to be made upof two morphotypes attributable, with a certain reliability, to both L. sinapis and L. reali. The fact that thesample contained both morphotypes is important because in order to discriminate between them both must bepresent. The bivariate scatter plot shows that there are several intermediate individuals among the males (Fig.2): this was also reported in some previous studies. The occurrence of several apparently intermediate individ-uals is another test of the approach used in this study.

FIGURE 2. Male specimen bivariate scatter plot using aedoeagus (≈ aedeagus ≈ phallus) and saccus length as discrimi-native characters.

Measurements. The survey is exclusively based on the measurements of some genitalic characters. Wing

pattern (Freese & Fiedler 2004) and shape of the 8th male abdominal sternite (Kristal & Nässig 1996; Freese &Fiedler 2004) were not examined because they cannot be used easily with a traditional morphometricapproach. All the characters listed in Table 1 (containing acronyms used through the text, character descrip-tions and several additional explanations) and illustrated in Fig. 3 were measured in the males and femalesexclusively by the author. In an attempt to improve the discriminative power, certain genitalic characters sel-dom used in previous papers were considered in addition to ten new characters (six in males and four infemales).

TABLE 1. Description of characters measured with additional explanations.

* Character not used in previous surveys

FIGURE 3. Schematic drawings of characters used in morphometric analysis: A) male genitalia (lateral view), B)female genitalia (lateral view) showing both papille analis and anterior apophyses.

Sex Measurement acronym

Measurement

description

Additional explanations

male PL Phallus length. Kristensen (2003) points out that it would be better to use the term phallus

instead of aedoeagus or aedeagus. From this point on phallus is used (see Fig.3A).

* PXW Phallus apex width. (see Fig. 3A)

* PBW Phallus base width. (see Fig. 3A)

SL Saccus length. (see Fig. 3A)

* SW Saccus width. (see Fig. 3A)

VW Vinculm width. Genus Leptidea species have valve (strongly reduced) and vinculum fusedtogether in the ventral part of the genital capsule (Kristensen, 2003). As thevinculum occupies a wider surface of the ventral part of genital capsule, in thisstudy this term will be used (see Fig. 3A).

* VAH Valve apex height. Measurement taken at 90° to the straight line jointing the ventral edge of the genital capsule and the saccus apex (see Fig. 3A).

UL Uncus length. Measurement taken from the genital capsule apex (see Fig. 3A).

* UAW Uncus apex width (see Fig. 3A)

* CL Genital capsule length. Measurement taken at 90° to the straight line connecting the ventral edge of the genital capsule and the saccus apex (see Fig. 3A).

female DBL Ductus bursae length. (see Fig. 3B)

DBW Ductus bursae width. The distal part of the ductus bursae represents ostium bursae and its expan-sions, lamella praevaginalis and lamella postvaginalis (see Fig. 3b).

* PAW Papille analis width. (see Fig. 3B)

* PAL Papille analis length. Within L. sinapis species complex, papillae analis and posterior apophysis

margins are not clearly marked. Hence, in this study it will only refer to the papille analis length for simplicity (see Fig. 3B).

* AAL Anterior apophysis length. (see Fig. 3B)

After Eosin Red staining, male genitalia (Fig. 3A) were measured by introducing them into a hollow pro-duced with glue on a polyethylene rectangle (0.5 mm thick), with a square hole (about 5×5 mm) in the middle,on a microscope slide. The hollow was filled with glycerol and then covered with a coverslip. This device wasdesigned to avoid the crushing of several structures which could produce distortions (particularly the saccus).Before any measurements were taken the position of each “presentation” (sensu Arnqvist & Mårtensson1998) was “arranged” by small shifts of the coverslip until the margins of the two central-upper free proximalparts of the genitalic capsule overlapped in the lateral view (in most of the published photos and drawingsthese parts do not match but are both visible). It was not possible to adopt a similar device for measuringfemale genitalia, because it was difficult to find a reproducible position. Hence, after Methylene Blue stainingthe female genitalia were measured by introducing them (in lateral view as in Fig. 3B) in a hollow asdescribed above, but this time a polyethylene square (0.1 mm thick) was used to flatten weakly sclerotizedstructures.

Three replicates were made for each specimen for the whole data set. This choice was based on the factthat it is a time-consuming task and more than four replicates did not substantially decrease the ME (Arnqvist& Mårtensson 1998). Measurements were made under a stereoscopic microscope using a micrometric eye-piece measuring to the nearest 0.01 mm. In male genitalia, with paired structures such as vinculum and uncuseach side of each “presentation” was measured by changing the side of the microscope slide (care was takennot to modify each “presentation”). Prior to any statistical analyses, the average of all the measurements takenfor each character was calculated (except for the measurement error estimate).

Morphometric approach. Taxonomists usually prefer characters that differ in shape rather than size indiscriminating between species (Adams et al. 2004). In “traditional” morphometrics, statistics cannot com-pletely separate size and shape. To overcome this limitation many approaches have been developed to removeshape from size (Jungers et al. 1995). Mosimann (Mosimann 1970; Darroch & Mosimann 1985) has devel-oped some methods that have proved to be quite efficient in removing size (Jungers et al. 1995), making itpossible to determine the extent to which differences among individuals can be attributed to a combination ofsize and shape rather than shape only. In the present paper the method proposed by Mosimann (1970) has beenfollowed. Hereafter, data sets containing both size and shape information will be designated as “size-and-shape” throughout the text, whereas those modified to create size-free variables (Mosimann 1970), will benamed “size-adjusted”.

Statistical analyses. Male and female data sets (290 and 147 specimens respectively) were first log10(x+1)

transformed to make variance independent of the mean (Sokal & Rohlf 1995); the resulting matrices corre-spond to those designated as “size-and-shape”. Size was then removed subtracting the logged geometric meanfrom each variable in the row (Mosimann 1970); the resulting matrices correspond to those designed as “size-adjusted”.

The “size-and-shape” and “size-adjusted” male and female data sets were then subjected to principal com-ponent analysis (PCA) and discriminant analysis (DA). PCA, derived from the covariance matrix, was used todetect groups of specimens that were not classified a priori. The correlation between size and PCs derivedfrom “size-and-shape” and “size-adjusted” data sets was tested by using the geometric mean of all the vari-ables of each individual as a size measure (Darroch & Mosimann 1985) and individual scores of the principalcomponents. DA was then used to test differences between morphotypes that were identified by PCA (DAneeds the a priori classification of specimens) and to identify the best discriminative variables. DA was per-formed using a stepwise procedure with variables entered in a forward manner and classification was com-puted using the jackknife method (Sokal & Rohlf 1995). An additional sub-sample of 50 specimens (30 malesand 20 females, containing both the putative morphotypes) was used as the holdout sample to estimate theexternal validity of classification functions because DA tends to overestimate the percentage of correct classi-fications. DA, based on stepwise procedure, was computed on the “size-and-shape” data sets, as everyremoval of a character in the “size-adjusted” ones would require size correction of the remaining characters

(Klimov & OConnor 2004). Comparison between “size-and-shape” and “size-adjusted” DA was done withoutusing stepwise procedure.

ME estimate was calculated separately on the correctly classified specimens (there may have been MEdifferences between species for each character used in this study). The ME was calculated using the followingformula (Yezerinac et al. 1992):%ME = [S2

within / S2within + S2

among] × 100, where %ME is the percentage of measurement error, S2among is the vari-

ance among individuals (S2among = MSamong – MSwithin / n, MSamong is the mean square among individuals, MSwithin is

the mean square within individuals and n is the number of repeated measurements) and S2within is the variance

within individuals. A model II one-way ANOVA was used to evaluate ME.All statistical tests were computed using Statistica 6.0 (Statsoft 2001), except for the DA for which Systat

11.0 (Systat Software Inc. 2004) was used.

Results

Morphometrics. In the PCA derived from the male “size-and-shape” data set two components accounted for79.74% of all the variability (Table 2). PC1 explained 66.71% of the variability and is quite strongly corre-lated with size (r = 0.827, p < 0.0001). In PC1 the loadings do not share the same sign, SL and PL have veryhigh loadings, PXW and SW have intermediate-small loadings and all the remaining characters have verysmall loadings (Table 2). PC2 explains a further 13.03% of the variability and is negatively correlated withsize (r = – 0.546, p < 0.0001). Characters having a very high loading in PC2 are, listed in decreasing order:CL, VW and UL; VAH and PBW have quite high loadings, whereas UAW, SW, PXW, PL and SL have smallor very small loadings (Table 2). The scatter plot of PC1 and PC2 highlights two separate groups of individu-als, with some individuals close enough to the two clouds, so as not to be considered as outliers (Fig. 4A). Theseparation that occurs along PC1 is mainly linked to differences in PL and SL. PC2 does not separate the twomorphotypes; there is only a slight difference in size in those characters having a high loading in PC2. Hence,the morphotype having a shorter phallus and saccus on the one side and a wider vinculum, longer uncus andgenitalic capsule on the other, can be referred to as L. sinapis. The morphotype having the opposite character-istics can be referred to as L. reali. In the PCA performed on the female “size-and-shape” data set, two com-ponents accounted for 90.37% of all the variability (Table 2). PC1 explained 80.62% of the variance and isstrongly correlated with size (r = 0.888, p < 0.0001). In PC1 all loadings share the same positive sign, DBLhas a very high loading (≈1), whereas the remaining characters have intermediate-small (PAW) or small load-ings (DBW and PAL). PC2 explains a further 9.75% of the variance and has a certain correlation with size (r =0.443, p < 0.0001). PAL, AAL and PAW have the highest loadings in PC2, whereas DBL and DBW havesmall loadings. As in the males, the scatter plot of PC1 and PC2 highlights two different morphotypes (Fig.5A) that are separated along PC1. The morphotype having a longer ductus bursae can be referred to as L.reali, while the morphotype having the opposite characteristics can be referred to as L. sinapis.

In the male “size-adjusted” PCA, the total variance accounted for is reduced from 0.002072 (“size-and-shape data set) to 0.001594 (Table 2). Shape accounts for 76.93% of the variance and the remaining varianceis in residual size. Eigenvalues of the first two components are slightly less (Table 2) than those of the “size-and-shape” PCA (a third component accounted for a variance similar in value to the second one). PC1accounts for 69.32% of the total variance and is quite strongly correlated with size (r = 0.688, p < 0.0001). InPC1 the highest loadings (Table 2) are those of SL and PL (very similar to the “size-and-shape” PCA), manyof the variables, such as VW, BW, CL, UAW, UL and VAH have quite highly negative loadings; two vari-ables, SW and PXW, have an intermediate loading. PC2 accounts for 7.97% of the total variance (Table 2) andhas a weak negative correlation with size (r = – 0.241, p < 0.0001). All of the characters except VAH, whichhas a high positive loading (> 0.8), have small loadings that do not exceed 0.35 and most have a negative sign(Table 2). Loadings of the first two components are noticeably different from those of the “size-and-shape”PCA in that the first component alone shares the highest loadings of PC1 and PC2 (Table 2). The scatter plotof the “size-adjusted” PCA still separates two morphotypes (Fig 4B), similar to the results of “size-and-shape”

PCA, but the main difference between them is the lack of any differentiation along PC2. In the female “size-adjusted” PC1, the total variance accounted for is reduced from 0.001909 to 0.001499 (Table 2), hence shapeaccounts for 78.52% of the variance and the remaining variance is in residual size. PC1 explains 82.74% ofthe total variance and is very strongly correlated with size (r = 0.888, p < 0.0001). PC1 highlights a very highpositive loading for DBL (≈1) and smaller negative loadings for the other characters (Table 2). PC2 explains afurther 7.05% of the variance (Table 2) and has a weak negative correlation with size (r = – 0.297, p < 0.0001).PC2 has high loadings for DBW (positive) and AAL (negative), whereas the loadings of the remaining char-acters are close to zero (Table 2). The difference between the “size-adjusted” PCA loadings and those of the“size-and-shape” PCA are obvious, even if less than the male data set. As for the males, the scatter plot of the“size-adjusted” PCA still identifies two morphotypes (Fig 5B), similar to the results of the “size-and-shape”PCA. Based on the PCA results, 227 males were assigned to L. sinapis and 63 to L. reali, whereas 103 femaleswere assigned to L. sinapis and 44 to L. reali.

TABLE 2. Eigenvalues, variance explained, cumulative variance and character loading of “size-and-shape” and “size-adjusted” PCA.

Sex Charac-ter

“Size-and-shape” PCA “Size-adjusted” PCA

PC1 PC2 PC1 PC2

male Eigenvalue 0.001382 Eigenvalue 0.018750 Eigenvalue 0.001105 Eigenvalue 0.000127

Variance explained 66.71% Variance explained 13.03% Variance explained 69.32% Variance explained 7.95%

Cumulative variance 66.71% Cumulative variance 79.74% Cumulative variance 69.32% Cumulative variance 77.27%

SL 0.9848 – 0.0504 0.9797 – 0.0236

PL 0.9615 0.0690 0.9349 0.0108

PXW 0.4155 – 0.1181 – 0.2247 0.3504

SW 0.3912 0.2699 – 0.3469 0,1880

PBW – 0.2123 0.4511 – 0.7213 0,3127

UAW 0.0745 0.3412 – 0.6729 0.2645

UL 0.0684 0.7558 – 0.6136 0.1454

CL – 0.0670 0,8874 – 0.6789 – 0.1259

VW – 0.0647 0.7665 – 0.7488 0.0845

VAH – 0.0218 0.5952 – 0.5001 – 0.8372

female Eigenvalue 0.001580 Eigenvalue 0.128472 Eigenvalue 0.001240 Eigenvalue 0.000106

Variance explained 80.62% Variance explained 9.75% Variance explained 82.74% Variance explained 7.05%

Cumulative variance 80.62% Cumulative variance 90.37% Cumulative variance 82.74% Cumulative variance 89.79%

DBL 0.9995 0.0294 0.9991 0.0255

DBW 0.1812 – 0.1516 – 0.4192 0.6016

PAW 0.3543 – 0.6613 – 0.2812 – 0.0417

PAL 0.1361 – 0.8459 – 0.4360 – 0.2733

AAL 0.1181 – 0.7202 – 0.3860 – 0.8457

Even though “size-adjusted” matrices are more reliable for classifying specimens because they are lessaffected by size difference, DA cannot be carried out using the stepwise forward procedure on the “size-adjusted” data sets for the previously mentioned problems. DA computed on the “size-and-shape” data setsgave the following results. A significant discrimination was achieved for the males (Wilks' Lambda = 0.079,F(6, 283) = 548.314, p < 0.00001) and 100% of the specimens were classified correctly (227 as L. sinapis and 63

as L. reali respectively). The importance of the characters for discriminative purpose was evaluated by exam-ining the standardized canonical discriminant function coefficients (PL =1.094, VW = – 0.794, SL = 0.415,UL = – 0.343, PBW = – 0.194, SW = 0.177), where the higher the absolute value of a coefficient, the higherits load on classification. Six characters, listed in order of importance, contributed substantially to the classifi-cation: PL, VW, SL, UL, PBW and SW; PBW and SW had a rather small impact on the discrimination. A sig-nificant discrimination was achieved in the females (Wilks' Lambda = 0.079, F(3, 145) = 557.611, p < 0.00001)

and 100% of the specimens were classified correctly (103 as L. sinapis and 44 as L. reali respectively). Threecharacters contributed substantially to the classification, DBL, DBW and AAL (standardized canonical coeffi-cients: DBL = 1.108, DBW = – 0.392, AAL = – 0.245), despite DBW and AAL had a small discriminativepower. On the “size-adjusted” data sets DA (performed without stepwise procedure) gave very similar resultsin comparison to those obtained from the “size-and-shape” data sets. In the males, the same set of characterspreviously selected gave a significant contribution to classification. In females, only DBL and DBW gave acontribution to the classification.

FIGURE 4. PC1 and PC2 scatter plot of male specimens: A) “size-and-shape” PCA, B) “size-adjusted” PCA.

FIGURE 5. PC1 and PC2 scatter plot of female specimens: A) “size-and-shape” PCA, B) “size-adjusted” PCA.

Measurement error. In the male specimens assigned to either L. sinapis or L. reali, the ME was rathervariable, being low (< 5%) in some characters (PL, SL, VW, UL, CL) and high (SW, PXW, PBW, VAH,UAW) in others (Table 3). It should be noted that since VW and UL are double structures, they had twice asmany measurements as the other characters. In the females, the ME was only low for the DBL for the speci-mens assigned to both L. sinapis and L. reali, whereas it was rather high (>10%) in the remaining characters(Table 3).

TABLE 3. Measurement error (%) associated to the measured characters in L. sinapis and L. reali.

Character selection. Among the discriminative characters in males, PBW and SW had rather high ME.Considering their scarce contribution to classification, they were not considered reliable for this purpose,based on the advice given by Bailey and Byrnes (1990). Therefore, only four characters, PL, VW, SL and UL,were considered in classifying the L. sinapis species complex individuals. The DA was rerun using these fourcharacters on the “size-and-shape” matrix because unstandardized canonical functions are easier use for clas-sification purposes than “size-adjusted” measurements. A 100% correct classification was obtained for boththe data set and subsample of 30 not a priori classified specimens used as an external validation. The follow-ing unstandardized canonical coefficients were obtained which allow unknown specimens to be classified(measurements are given in mm, mean of canonical variables are: L. sinapis = – 1.736, L. reali = 6.256; criti-cal value = 1.084):100.674 × [log (PL + 1)] – 92.856 × [log (VW + 1)] + 33.426 × [log (SL +1)] – 34.593 ×[ log (UL + 1)] –20.151[constant].

For practical purposes, a good separation of morphotypes can be achieved by using a scatter plot of PLagainst VW (it works better than that of PL against SL). In females, the DA rerun on the “size-and-shape”data set using only DBL among the discriminative characters because of its very high impact on the classifica-tion and its small ME yielded 100% correct classifications. DBW and AAL cannot be considered fundamentalfor classification purposes because of the high associated ME. All of the 20 specimens used as external valida-tion were correctly classified. Given that there was only one discriminative character and due to the lack ofany value overlap in DBL, the unstandardized canonical coefficients for females are not provided (criticalvalue = 0.79 mm).

Mean values, standard deviation and minimum and maximum values for each measured character in bothspecies (using correctly classified specimens of the “size-and-shape” PCA) are given in Table 4.

Sex Species Character

male PL PXW PBW SL SW VW VAH UL UAW CL

L. sinapis (n = 227) 1.11 16.67 6.78 2.07 12.14 2.87 3.44 3.08 17.06 2.62

L. reali (n = 63) 0.43 14.86 10.51 1.43 11.39 3.09 6.73 2.54 20.05 4.81

female DBL DBW PAW PAL AAL – – – – –

L. sinapis (n = 103) 3.63 38.57 13.45 8.35 13.66 – – – – –

L. reali (n = 44) 1.04 42.37 19.21 13.85 11.56 – – – – –

TABLE 4. Descriptive statistics of characters measured (in mm) in L. sinapis and L. reali.

Discussion

Morphometrics. In males, “size-and-shape” PCA has given results that could be interpreted as evidence ofdifferences both in size and shape between groups of genitalia, whereas in females these differences shouldonly be referred to size. Many authors consider PC1s to be size-related when their loadings share the samesign and they consider a PC1 to be both size and shape relates when PC1 loadings have different signs. Mosi-mann (1970) and Sprent (1972) advise against this interpretation, considering it too arbitrary. This interpreta-tion will also be followed and the results of the “size-adjusted” data sets will be considered more reliable.

Interpretation of the “size-adjusted” data sets appears to be more straightforward. The lower variancerecorded in comparison to the “size-and-shape” PCA, the PC1 having almost all of the highest variable load-ings (in this case with different signs) and the bivariate scatter plot highlighting two separate morphotypes, are

Sex Species Character

male PL PXW PBW SL SW VW VAH UL UAW CL

L. sinapis (n = 227) mean 1.690 0.091 0.290 0.653 0.119 0.839 0.170 0.657 0.117 0.927

SD. 0.069 0.016 0.026 0.051 0.019 0.039 0.034 0.040 0.012 0.048

min. 1.48 0.05 0.22 0.52 0.07 0.73 0.10 0.55 0.09 0.77

max. 1.89 0.14 0.36 0.81 0.17 0.95 0.27 0.78 0.14 1.06

L. reali (n = 63) mean 2.024 0.110 0.270 0.914 0.136 0.811 0.166 0.643 0.116 0.913

SD. 0.063 0.019 0.022 0.059 0.019 0.026 0.026 0.033 0.011 0.032

min. 1.88 0.07 0.22 0.77 0.10 0.75 0.11 0.56 0.09 0.83

max. 2.16 0.15 0.32 1.04 0.17 0.87 0.23 0.72 0.14 0.99

female DBL DBW PAW PAL AAL – – – – –

L. sinapis (n = 103) mean 0.643 0.245 0.445 0.690 0.435 – – – – –

SD. 0.038 0.020 0.031 0.046 0.036 – – – – –

min. 0.56 0.20 0.36 0.58 0.35 – – – – –

max. 0.74 0.30 0.51 0.80 0.52 – – – – –

L. reali (n = 44) mean 0.980 0.248 0.461 0.694 0.439 – – – – –

SD. 0.071 0.027 0.024 0.029 0.042 – – – – –

min 0.84 0.20 0.41 0.64 0.35 – – – – –

max. 1.13 0.31 0.51 0.76 0.53 – – – – –

evidence of true differences in shape within the L. sinapis species complex, in both sexes. However, eventhese results should be interpreted with some caution. The PC1 of “size-adjusted” data sets are still quitehighly correlated with size. Moreover, a widely recognized limitation of “traditional” morphometrics is theloss of some aspects of shape (Adams et al. 2004). Hence, the differentiation in shape between L. sinapis andL. reali in both sexes can be considered related to size, although at least a part of the differences between mor-photypes is also attributable to shape. This feature has also been considered in previous papers (Simmons etal. 1991; Falsetti et al. 1993; Klimov et al. 2006). Gathering all the information associated with shape is a taskthat could be better accomplished by a “geometric” morphometric approach, that was successfully applied indescribing the genitalic morphology of some moth species (Mutanen 2005; Mutanen & Kaitala 2006;Mutanen et al. 2006; Mutanen et al. 2007). An application of “geometric” morphometrics could show that thisaspect was overlooked by the current approach in the L. sinapis species complex. However, even thisapproach could have some limitations. A few landmarks are available (this problem could be overcome usingsliding semilandmarks). Spurious results could be obtained due to the mobility of some structures such as sac-cus, uncus and phallus in males and, in particular, the difficulty to find a reproducible position of the DBW infemales. From a practical point of view (i.e. the possibility of discriminating specimens), “geometric” mor-phometrics could make a few contributions to the “traditional” morphometric approach. Despite the aboveuncertainties of interpretation and the impossibility of gathering all the information linked to shape, this studyhas stressed that L. sinapis species complex is made up of two separate genitalic morphotypes in both sexes.DA confirmed this separation.

Dealing with classification results, the first issue is that in previous papers, only PL and SL were used asdiscriminative characters in males. Two separate morphotypes could not be distinguished because there weresome intermediate individuals. This usually occurred in studies where a large sample size was used (Neumayr& Segerer 1995; Göhl & Buchsbaum 1996, 1997; Hauser 1997). When Martin et al. (2003) and Verovnik andGlogovčan (2007), only used a morphological approach based on the examination of PL and SL, they alsofound some “doubtful” specimens. In the present study, two specimens were misclassified when the DA wasbased only on PL and SL. On the contrary, the use of genitalic characters in addition to PL and SL, as well asmultivariate statistics, allowed two separate morphotypes to be detected. All of the specimens examined werecorrectly classified even though the data set showed a slight overlap within PL and SL values (in the remain-ing characters the value overlap was almost complete, Table 4). This survey has highlighted the importance ofVW and, to a much lesser extent UL, as discriminative characters of the L. sinapis species complex speci-mens. Vila et al. (2003) have already suggested the importance of VW (they called this distance valve). How-ever, these authors used VW in conjunction with SL, as the SL / VW ratio. They stated (no data wereprovided) that SL is equal to or longer than VW in L. reali, whereas SL is always shorter than VW in L. sina-pis. The data set examined in the present study showed a somewhat different situation than that reported byVila et al. (2003). Some specimens classified as L. reali had a VW that was greater than SL (in one specimenup to 0.05 mm); Hauser (1997) also reported similar data. To test the discriminative power of the SL / VWratio, a DA was rerun using the SL / VW ratio as discriminative variable. The DA gave 98% of correctly clas-sified specimens among those attributed to L. sinapis; by adding PL as an additional variable in conjunctionwith the SL / VW ratio, the percentage of correctly classified specimens increased to 100% (despite the factthat scores of posteriori probabilities were noticeably lower than in the DA computed using the four discrimi-native characters in several specimens). Hence, the use of SL and VW (together with PL and UL) allows aslightly sharper morphotype separation than the SL /VW ratio alone. It is obvious that the SL / VW ratio is abit less reliable than the two characters considered separately in classifying the L. sinapis species complexmale specimens. All of the new genitalic characters tested (PBW, PXW, SW, UAW, CL and VAH) had little orno discriminative power and were often associated with a high ME (see below for further discussions). Infemales, DBL was the only character that did not show any value overlap within the L. sinapis species com-plex, making it particularly suitable for discriminative purposes. DBW, considered by Hauser (1997) to be a

discriminative character, had very little discriminative power and a high associated ME, making it absolutelyunreliable. New genitalic characters tested in this study (PAW, PAL and AAL) showed slightly significantresults with AAL, but this character was eliminated because of its high associated ME or no results for PAWand PAL (see below for further discussions).

Measurement error. Genitalic “presentations” could be one of the main ME sources (for a thorough dis-cussion see Arnqvist & Mårtensson 1998). The measurement device adopted for both males and femalesshould have ensured uniform measurement conditions and quite flat “presentations” (only Hauser 1997 usedquite a similar measurement device, whereas in several previous studies measurement devices were notreported). Moreover, the “arrangements” of the “presentations” of the male genitalia limited the ME greatly:in a subset of thirty specimens the MEs were noticeably higher (PL 10.81%, SL 6.93%, VW 19.42%, UL 9.41%) than those on the “arranged” presentations of the same subset (PL 3.34%, SL 1.86%, VW 7.92%, UL4.31%). No similar measurement device was adopted for the females.

The ME discrepancies that were detected by comparing the same characters between species (Table 3)might have been due to sample size and character size. The difference in the number of specimens belongingto L. sinapis (males n = 227, females n = 103) and L. reali (males n = 63, females n = 44) might account forpart of the differences in ME. Bailey and Byrnes (1990) and Yezerinac et al. (1992) suggested increasing thenumber of measured specimens to help reduce ME. Yezerinac et al. (1992) also demonstrated a negative rela-tionship between ME and the absolute size of the character. The high MEs (Table 3) associated with the small-sized characters (Table 4), could be thus explained. This last point could, in part, account for the wide range ofvalues recorded for the within-species ME (Table 3), both in males and females. The high ME of PXW, PBW,SW and UAW in males and PAW, PAL and AAL in females could, be partially reduced by making stable prep-arations and by cutting and flatting the weakly sclerotized parts of the female genitalia (Mutanen 2005). How-ever, such an approach seems inadequate for measuring male genitalia, because of the saccus distortion. Thislatter effect, if not associated with a device to limit ME, could result in a biased measurement, even of dis-criminative characters, that would strongly affect the classification results.

Therefore when separating the L. sinapis species complex specimens, the subtle differences between char-acters in this group, must be considered and the ME should always be limited by using measurement devicesthat ensure uniform conditions. If necessary, a couple of measurements for each character should be averaged.

Intrapopulation variability. In the populations from central Italy, the sizes of the PL, SL, VW and UL inmales and DBL and DBW in females (Table 4) (only data on PL, SL and DBL are commonly available in theliterature, VW, UL and DBW were provided for just a few specimens) are comparable to those reported forthe L. sinapis species complex in western and central Europe (Lorkovic 1993; Laštůvka et al. 1995; Neumayr& Segerer 1995; Embacher 1996; Karsholt 1999; Gianti & Gallo 2002; Mazel & Leestmans 1999; Nelson etal. 2001; Freese & Fiedler 2004; Cupedo & Hoen, 2006). The populations from east-central Europe and Asiahave genitalic size lesser than those from western and central Europe (Lorkovic 1993; Skalski 1995; Göhl &Buchsbaum 1996, 1997; Ábrahám 1998; Gorbunov 2001).

This survey has highlighted a wider intraspecific variability and size-overlap of some characters (Table 4)in comparison with previous papers. In particular, male L. sinapis specimens have extremely long phallus andsaccus with values that are only equivalent to those recorded in an area of northern Italy (Gianti & Gallo2002). In females, the gap between L.sinapis and L. reali with respect to the DBL size is narrower than in pre-vious surveys. These results could have two, not mutually exclusive, explanations: i) there are differences ingenitalic sizes among populations even in relatively small areas and; ii) the small sample size considered inseveral previous papers could have prevented the less likely specimens to be found in a population, that is,those having extremely long phallus and saccus or the opposite in males (compare the data provided by Lork-ovic 1993 and Lelo 2002 relative to L. sinapis samples collected in Croatia and Bosnia) as well as for the DBLin females.

Conclusions

This study has shown that a traditional morphometric approach based on the measurement of PL, VW, SL andUL is a valuable tool for separating the male L. sinapis species complex specimens. Females can be easilyseparated using DBL alone. It is also important to adopt devices that reduce the ME that is associated with dis-criminative characters and, if possible, to estimate it. It should be noted that using a morphometric approachalone there may be some difficulties in classifying single male specimens collected from areas where no pre-vious studies were made due to the lack of distinctly separate values for the characters in L. sinapis and L.reali. The differences of genitalic sizes within the L. sinapis species complex range may also cause problems.

It would be of interest to test the proposed discriminative approach by taking into account a wider area oreven the whole range of species and combining it with a molecular technique (Cesaroni et al. 1989, 1994;Monti et al. 2001; Nice & Shapiro 1999; Porter et al. 1995).

Acknowledgments

I wish to thank the “Parco Nazionale dei Monti Sibillini” for allowing me to collect Leptidea sinapis speciescomplex specimens and an anonymous referee for his useful comments on earlier version of manuscript.

References

Ábrahám, L. (1996) Leptidea reali Reissinger, 1989 a new species of the Hungarian butterfly fauna (Lepidoptera:Pieridae). Dunántúli Dolgozatok (A) Természettudományi Sorozat, 9, 387–394 (in Hungarian, English abstract).

Adams, D.C., Rohlf, F.J. & Slice, D.E. (2004) Geometric morphometrics: ten years of progresses following the “revolu-tion”. Italian Journal of Zoology, 71, 5–16.

Arnqvist, G. & Mårtensson, T. (1988) Measurement error in geometric morphometrics: empirical strategies to assess andreduce its impact on measures of shape. Acta Zoologica Academiae Scientiarum Hungaricae, 44, 73–96.

Bailey, R.C. & Byrnes, J. (1990) A new, old method for assessing measurement error in both univariate and multivariatemorphometric studies. Systematic Zoology, 39,124–130.

Beneš, J., Konvička, M., Vrabec, V. & Zámečník, J. (2003) Do the sibling species of small whites, Leptidea sinapis andL. reali (Lepidoptera, Pieridae) differ in habitat preferences? Biologia (Bratislava), 58, 943–951.

Cesaroni, D., Allegrucci, G., Angelici, M.C., Racheli, T. & Sbordoni, V. (1989) Allozymic and morphometric analysis ofpopulations in the Zygaena purpuralis complex (Lepidoptera, Zygaenidae). Biological Journal of the Linnean Soci-ety, 36, 271–280.

Cesaroni, D., Lucarelli, M., Allori, P., Russo, F. & Sbordoni, V. (1994) Patterns of evolution and multidimensional sys-tematics in graylings (Lepidoptera: Hipparchia). Biological Journal of the Linnean Society, 52, 101–119.

Cupedo, F. & Hoen, F.W. (2006) Leptidea sinapis and Leptidea reali (Lepioptera: Pieridae) in The Netherlands. Entomol-ogische berichten, 66(4),118–123.

Darroch, J.N. & Mosimann, J.E. (1985) Canonical and principal components of shape. Biometrika, 72, 241–259.Embacher, G. (1996) Beitrag zur Verbreitung und Biologie von Leptidea sinapis L. und L. reali Reiss. Zeitschrift der

Arbeitsgemeinschaft Österreichischer Entomologen, 48, 107–112.Falsetti, A.B., Jungers, W.L. & Cole, T.M. III (1993) Morphometrics of the Callitrichid Forelimb: A case Study in Size

and Shape. International Journal of Primatology, 14, 551–572.Freese, A. & Fiedler, K. (2002) Experimental evidence for specific distinctness of the two wood white butterfly taxa,

Leptidea sinapis and L. reali (Pieridae). Nota lepidoperologica, 25, 39–59.Freese, A. & Fiedler, K. (2004) Unterscheidungsmerkmale von Leptidea sinapis (Linnaeus, 1758) und Leptidea reali

Reissinger, 1989 (Lepidoptera, Pieridae) und ihre Eignung zur Artbestimmung. Nachrichten des entomologishenVereins Apollo N.F., 25, 65–77.

Gianti, M. & Gallo, E. (2002) Nuovi dati sulla presenza in Italia settentrionale di Leptidea reali (Lepidoptera Pieridae).Bollettino della Società entomologica italiana, 134, 67–72.

Göhl, K. & Buchsbaum, W. (1996) Leptidea reali Reissinger, 1989 auch in Thüringen (Lepidoptera, Pieridae). Mitteilun-gen des Thüringer Entomologenverbandes E.V., 3, 46–55.

Göhl, K. & Buchsbaum, W. (1997) Leptidea sinapis L. und L. reali REISS. in Thüringen Ergebnisse der bisher durch-

geführten Untersuchungen Lepidoptera, Pieridae). Mitteilungen des Thüringer Entomologenverbandes E.V., 4, 58–70.

Gorbunov, P. (2001) The butterflies of Russia: classification, genitalia, keys for identifications. Thesis, Ekaterimburg,320 pp.

Hauser, E. (1997) Leptidea sinapis (Linnaeus 1758) und Leptidea reali Reissinger 1989: zwei verschiedene Arten? (Lep-idoptera, Pieridae). Beiträge zur Naturkunde Oberösterreichs, 5, 65–75.

Karsholt, O. (1999) Dagsommerfugleslæten Leptidea Billberg i Danmark. Har — eller rettere havde — vi to Leptideaarter i Danmark? Lepidoptera, VII, 237–249.

Jungers, W.L., Falsetti, A.B. & Wall, C.E. (1995) Shape, Relative Size, and Size-Adjustments in Morphometrics. Year-book of Physical Anthropology, 38, 137–161.

Klimov, P.B. & OConnor, B.M. (2004) Multivariate discrimination among cryptic species of the mite genus Chaetodac-tylus (Acari: Chaetodactylidae) associated with the bees of the genus Lithurgus (Hymenoptera: Megachilidae) inNorth America. Experimental and Applied Acarology, 33, 157–182.

Klimov, P.B., Bochkov, A.V. & OConnor, B.M (2006) Host specificity and multivariate diagnostic of cryptic species inpredacious cheyletid mites of the genus Cheletophyes (Acari: Cheyletidae) associated with large carpenter bees. Bio-logical Journal of the Linnean Society, 87, 45–58.

Kristal, P.M. & Nässig, W.A. (1996) Leptidea reali Reissinger, 1989 auch in Deutschland und einigen anderen europäis-chen Ländern (Lepidoptera: Pieridae). Nachrichten des entomologishen Vereins Apollo N.F., 16, 345–361.

Kristensen, N.P. (ed.) 2003. Lepidoptera, Moths and Butterflies. Vol. 2: Morphology, Physiology, and Development. In:Fisher M.: Handbook of Zoology 4. Arthopoda, part 36.

Kudrna, O. (2001) Miscellaneous notes on the taxonomy of four European butterflies (Lepidoptera: Rhopalocera). Ento-mologist’s Gazette, 52, 253–261.

Laštůvka, Z., Králíćek, M., Jakeš, O. & Štěrba, V. (1995) Leptidea reali – nový druh běláska v České republice a naSlovensku (Lepidoptera: Pieridae). Klapalekiana, 31, 35–39.

Lelo, S. (2002) Variation in exogenous and endogenous (genitalia) characteristics of butterflies of the species Leptideasinapis Linnaeus, 1758 (Pieridae, Dismorphiinae) within populations from the area around Sarajevo. Natura Croat-ica, 11, 293–319.

Lorković, Z. (1993) Leptidea reali Reissinger, 1989 (= lorkovici Réal 1988), a new European species (Lepid., Pieridae).Natura Croatica, 2, 1–26.

Martin, J., Gilles, A. & Descimon, H. (2003) Species concepts and sibling species: the case of Leptidea sinapis and Lep-tidea reali. In: Boggs, C.L., Watt, W.B. & Ehrlich, P.R. (Eds), Butterflies: ecology and evolution taking flight. Chi-cago University Press, Chicago, pp. 459–476.

Mazel, R. & Leestmans, R. (1999) Séparation biométrique des Leptidea sinapis L., morsei Fenton et reali Reissinger.Linneana Belgica, XVII, 46–52

Monti, L., Baylac, M. & Lalanne-Cassou, B. (2001) Elliptic Fourier analysis of the form of genitalia in two Spodopteraspecies and their hybrids (Lepidoptera: Noctuidae). Biological Journal of the Linnean Society, 72, 391–400.

Mosimann, J.E. (1970) Size Allometry: Size and Shape Variables with Characterizations of the Lognormal and General-ized Gamma Distributions. Journal of the American Statistical Association, 65, 930–945.

Mutanen, M. (2005) Delimitation difficulties in species split: a morphometric case study on the Euxoa tritici complex(Lepidoptera, Noctuidae). Systematic Entomology, 30, 632–643.

Mutanen, M. & Kaitala, A. (2006) Genital variation in a dimorphic moth Selenia tetralunaria (Lepidoptera,Geometridae). Biological Journal of the Linnean Society 87, 297–307.

Mutanen, M, Kaitala A. & Mönkkönen, M. (2006) Genital variation within and between three related Euxoa moth spe-cies: testing the lock and key hypotesis. Journal of Zoology, 268, 109–119.

Mutanen, M., Rytkönen, S., Lindén, J. & Sinkkonen, J. (2007) Male genital variation in a moth Pammene luedersiana(Lepidoptera: Tortricidae). European Journal of Entomology, 104, 259–265.

Nelson, B., Hughes, M., Nash, R. & Warren, M. (2001) Leptidea reali Reissinger, 1989 (Lepid., Pieridae): a butterflynew to Britain and Ireland. Entomologist’s Record and Journal of Variation, 113, 94–101.

Neumayr, L. & Segerer, A. (1995) Leptidea reali Reissinger, 1989 in Bayern (Lepidoptera: Pieridae). Galathea, Berichtedes Kreises Nürnberger Entomologen, 11, 139–150.

Nice, C.C. & Shapiro, A.M. (1999) Molecular and morphological divergence in the butterfly genus Lycaeides (Lepi-doprera: Lycaenidae) in North America: evidence for recent speciation. Journal of Evolutionary Biology, 12, 936–950.

Porter, A.H., Schneider, R.W. & Price, B.A. (1995) Wing pattern and allozyme relationships in the Coenonympha arca-nia group, emphasising the C. gardetta-darwiniana contact area at Bellwald, Switzerland (Lepidoptera, Satyridae).Nota lepidopterologica, 17, 155–174.

Reissinger, E., (1990) Checkliste Pieridae Duponchel, 1835 (Lepidoptera) der Westpalaearctis (Europa, Nordwestafrika,Kaukasus, Kleinasien). Atalanta, 20, 149–185.

Segerer, A. (2001) Beitrag zur Genitaldiagnose einiger bayerischer Tagfalterarten unter besonderer Berücksichtigung der

&&. Beiträge zur bayerischen Entomofaunistik, 4, 5–25. Simmons, T., Falsetti, A.B. & Smith, F.H. (1991) Frontal bone morphometrics of southwest Asian Pleistocene hominids.

Journal of Human Evolution, 20, 249–269.Skalski, A.W. (1996) Genus Leptidea Billberg, 1820 in Poland (Lepidoptera: Pieridae). Acta Entomologica Silesiana, 3,

8–12 (in Polish, English abstr.).Sokal, R.R. & Rohlf, F.J. (1995) Biometry. Freeman & Co, San Francisco, 887 pp.Sprent, P. (1972) The mathematics of size and shape. Biometrics, 25–37.StatSoft Inc. (2001) Statistica for Windows 6.0. Tulsa.Systat Software Inc. (2004) Systat 11.0. Richmond, CA.Vila, R., Viader, S. & Jubany, J. (2003) Leptidea sinapis (Linnaeus, 1758) i L. reali (Reissinger, 1989): dues species “bes-

sones” a Catalunya i Andorra (Lepidoptera: Pieridae). Butlletí de la Societad Catalana de Lepidopterologia, 90, 25–47.

Verovnik, R. & Glogovčan, P. (2007) Morphological and molecular evidence of a possible hybrid zone of Leptidea sina-pis and L. reali (Lepidoptera: Pieridae). European Journal of Entomology, 104, 667–674.

Yezerinac, S.M., Lougheed, S.C. & Handford, P. (1992) Measurement error and morphometric studies: statistical powerobserver experience. Systematic Biology, 41, 471–482.

Distinguishing between Leptidea sinapis and L. reali (Lepidoptera: Pieridae) using a morphometric...

Documents

Transcript of Distinguishing between Leptidea sinapis and L. reali (Lepidoptera: Pieridae) using a morphometric...

Discriminative Keyword Spotting

Distinguishing among dark matter annihilation channels with neutrino telescopes

Adaptive local dissimilarity measures for discriminative dimension reduction of labeled data

Discriminative Color Descriptors

The Productivity Advantages of Large Cities: Distinguishing Agglomeration From Firm Selection

No Word is an Island: Distinguishing “Nerd” and “Geek”

Mechanisms of microtunneling in rock substrates: distinguishing endolithic biosignatures from abiotic microtunnels

I "Reali" dell'Altissimo. Un ciclo di cantari fra oralità e scrittura

Sulla convergenza delle serie reali a termini non negativi

"Flexibility and solution-driven work – two of our distinguishing ...

Learning Discriminative Space–Time Action Parts from Weakly Labelled Videos

Distinguishing psychogenic from neurogenic dysfluency when neurologic and psychologic factors coexist

Sulci Detection in Photos of the Human Cortex Based on Learned Discriminative Dictionaries

Fuzzy Pattern Recognition per il modellamento di sistemi complessi in contesti reali

Mining Low-Support Discriminative Patterns from Dense and High-Dimensional Data

2006 GROTTA REALI

A Discriminative Model for Tree-to-Tree Translation

vector representations of graphs and distinguishing quantum ...

Distinguishing Service Quality and Customer Satisfaction: The Voice of the Consumer

Distinguishing Xanthogranulomatous Cholecystitis from the ...