We detected significant differences in aedeagus size between species but more notably between flies reared in different cacti (Table 2a). However, in a nested ANOVA design as ours, the random factors Line in Species and Line by Cactus are the error terms of the fixed factors (Species, Cactus and their interaction). Large values of these terms may provide unreliable results in the testing of the fixed factors. Consequently we performed "a posteriori" contrasts of the Species by Cacti interaction to confirm the general results. Post hoc comparisons showed that D. buzzatii male flies reared in Opuntia had larger male genitalia than those grown in T. candicans (Tukey, p = 0.028) while in D. koepferae differences between flies grown in different cactus media were not significant (Tukey, p = 0.69). As visually observed, D. buzzatii and D. koepferae significantly differed in their genitalic shape and presented morphological variation not only among lines within species but also interacting with the breeding substrate (Table 2a).

The results of intraspecific ANOVAs also revealed important differences (Table 2b). In D. buzzatii the Cactus by Line interaction was significant and accounted for a relatively high percentage (12.1%) of phenotypic variance in aedeagus size. However, only the Line factor was significant in D. koepferae, neither the Cactus effect nor the Cactus by Line interaction were significant.

In summary, according to our experimental design, based on the isofemale line technique 30, shows that aedeagus size is not only phenotypically plastic, but also that substantial heterogeneity exists among lines in their plastic response, suggesting that plasticity has a genetic basis in D. buzzatii. In D. koepferae, in contrast, our results show that variation in aedeagus size has a genetic component, devoid of any plastic response in relation to the breeding substrate. According to the results of the MANOVAs, variability among lines in aedeagus shape was significant in both species (Table 2a, b). The proportion of total shape variation explained by the interaction Cactus by Line also differed between species. Approximately 9% of total shape variance was explained by the Cactus by Line interaction in D. koepferae. This interaction was significant for PC3 which is related with variation in thickness in both dorsal and ventral median portions of the organ (Figure 1). Conversely, in D. buzzatii, the Cactus by Line interaction was significant for PC1, which describes changes in the process of the ventral margin (Figure 1) and accounts for an important proportion (30%) of the explained morphological variance.

Correlation analysis between aedeagus size and shape also revealed important interspecific differences. On one hand, aedeagus shape was strongly correlated with size in D. buzzatii (more than 16% of the total shape variation was allometric, Table 3). Conversely, none of the 5 principal shape variables in D. koepferae were significantly associated with genital size (Table 3).

We also studied the relationship between variables describing size of male genitalia and wing length. These variables were not significantly correlated in D. koepferae (r = 0.13, p = 0.07), whereas in D. buzzatii, we detected a significant allometric relationship (r = 0.32, p < 0.001). Furthermore, aedeagus size and wing length varied isometrically in D. buzzatii as suggested by a coefficient of allometry not significantly different from 1 (slope value of linear adjusted function = 0.82; 95% confidence interval values [0.42 to 1.23]).

Four interspecific crosses, out of 25 attempted, (crosses 4855, 4853, 8832 and 3512) yielded enough hybrid progeny as to perform the present study. These results are in agreement with previous studies reporting strong premating isolation between D. buzzatii and D. koepferae 31. In order to asses hybrid male fertility, and prior to dissection, hybrid males were aged for 1 week and placed for 5 days in vials with several mature virgin females of D. buzzatii or D. koepferae. In all cases hybrid males failed to produce offspring even though copulation attempts were observed in the vials. Hybrid progeny obtained in crosses 4853 and 8832 could only be tested in vials prepared with the medium prepared with fermenting Opuntia due to low numbers of hybrid larvae, while in the other crosses the yield of hybrid progeny was high enough to be reared in both cactus media.

Regarding size, differences among genotypes (hybrids plus both parental lines) were significant in all crosses (Table 4). In Figure 2 we illustrate size differences among crosses and genotypes reared in Opuntia, the rearing substrate where all crosses were able to be tested. F1 hybrid males from crosses 4853 and 8832 reared in Opuntia vials presented intermediate values that differed significantly from both parental strains (p < 0.05 in both cases, Tukey's post hoc comparisons). In the other crosses, in which hybrids could be reared in both cacti, a significant Genotype by Cactus interaction (F_2,69 = 4, 49, p < 0.01) was only detected in cross 4855. Hybrids male progeny in this cross presented larger aedeagi than males of the parental D. buzzatii line in Opuntia, while differences between hybrids and the D. buzzatii parent were not significant in Trichocereus. In all cases, D. koepferae presented the largest genitalia in both cacti. In one cross (3512) mean genitalic size in hybrids was significantly lower than the male parental D. buzzatii line (p = 0.025, Tukey's test). Based on the correlation matrix, only PC1 scores were correlated with organ size in hybrids accounting for 50.4% of shape variation (Table 3). Unfortunately, the low number of hybrids and the high proportion of individuals with improperly unfolded wings precluded the analysis of allometric relationships between wing size and the variable describing aedeagus size.

Significant shape differences among genotypes were detected in all crosses (Table 4). Post hoc comparisons showed that all genotypes differed from each other in the shape of the genitalia at least in PC1 shape scores (p < 0.001 in all cases). However, neither the Cactus factor nor the Cactus by Genotype interaction were significant in the crosses in which hybrids were raised in the two cactus media (4855 and 3512).

In figure 3 we present a plot of the first two principal components describing shape variation (PC1 and PC2). The species differentiate themselves along the first shape axis (PC1) and the hybrid scores fall within the parental values. As can be observed the mean PC1 values of D. buzzatii lines involved in successful interspecific crosses tended to be negative and those of D. koepferae positive. However, hybrids failed to present intermediate values for both shape variables simultaneously. For instance, hybrids of cross 8832 had shape scores for both PC1 and PC2 that placed them in the morphological space closer to the D. koepferae parent (Line 88) than to D. buzzatii (Line 32). On the contrary, in cross 3512 a hybrid genital morphology was more similar to D. buzzatii for PC1 (Line12) but the mean for PC2 was more extreme than any of the parental lines. Suggestively, as explained above, hybrids in 3512 also presented smaller genitalia than both parental lines.

In order to evaluate the degree of resemblance of the morphology of hybrids to each parental line, we calculated the Euclidean distance to the morphological centroid of each parental strain using the shape (PCs) scores of each individual hybrid. As a rough index of morphological dissimilarity, hybrids would show equal mean distances to the centroids of both parental clouds of points if they have intermediate aedeagus morphology. Expression dominance of one genome over the other would produce phenotypes resembling more closely one parental strain or the other. Morphological dominant expression was tested with an ANOVA in which the variable was the Euclidean distance of each hybrid male to the centroid of each parental species (mean parental shape) with Cross and Parents as fixed factors. The ANOVA revealed significant differences among crosses though it should be noted that hybrid resemblance to parental strains were not independent of the cross (F_3,110=17.59; p < 0.0001; Figure 4).

Fifteen isofemale lines (lines hereafter) of each species, derived from collections in the locality of Suyuque (San Luis province, Argentina), were employed in the experiments outlined below. In this area both species coexist in nearly equal proportions (45% of captured females were D. koepferae, 55% D. buzzatii). The advantages of the use of the isofemale line technique in quantitative evolutionary genetics have been thoroughly described in 30. Briefly, the isofemale line technique is a convenient methodology for the analysis of quantitative traits under laboratory conditions. All experiments described below were conducted 24 generations after the foundation of isofemale lines.

Thirty first-instar larvae from each line were seeded in vials containing 6 ml. of media prepared with artificially fermented cactus (see 41 for details). Two different cactus species, Opuntia sulphurea and Trichocereus candicans which are commonly used as breeding substrates by D. buzzatii and D. koepferae in the locality sampled, were employed for the preparation of the semi-natural media. For each line, 4 replicated vials were run in each cactus type.

Five lines of each species were randomly chosen to generate interspecific hybrids. All possible combinations were attempted (25 potential crosses). F1 hybrids were produced by crossing 25 virgin females of a D. koepferae line with 50 males of a D. buzzatii line. The reciprocal cross invariably failed to produce viable progeny. Crosses were identified according to the number of the female parental line (D. koepferae) preceding the number of the male parental (D. buzzatii) line (e.g. the cross between D. koepferae line 48 and D. buzzatii line 55 was designated as 4855). Batches of 30 first-instar hybrid larvae were transferred to culture vials containing one of the two 'semi-natural' media. Four replicated vials were run per every combination of cactus and cross when the number of hatched larvae allowed it. The hybrid status of the descendants was ascertained by the cytological analysis of the polytene chromosomes of progeny larvae grown in vials run in parallel.

All cultures were maintained at 25 +/- 1°C with a 12:12 light/dark photoperiod until the emergence of adults. Adult flies were simultaneously collected and sexed under light CO₂ anesthesia.

The aedeagus and both wings were dissected from 2 to 5 males emerged in each replicate. Aedeagi were mounted on slides and photographed with a digital camera mounted on a microscope at 400 × magnification. Wings were also mounted and ventral views of wing images were captured with a digital camera attached to a binocular microscope (25 ×) connected to a computer. In each image, we scored total wing length (WL) using TpsDig 55.

As shown in Figure 5 there are conspicuous differences in aedeagus morphology between D. buzzatii and D. koepferae that preclude the possibility of determining an adequate number of reliable homologous landmarks. However, the aedeagus is a flat quitinous organ that can be effectively described in shape and size in two dimensions when flattened under a cover slip. Consequently, we decided to employ an approach based on elliptic Fourier descriptors (EFDs) 56 as a proper resolution to the problem of shape quantification 47. This is a type of analysis in which differences in x and y coordinates of the outline of the studied organ are fit separately as functions of arc length by Fourier analysis, so that the outline can be decomposed into a weighted sum of sine and cosine functions designated as harmonics. Outlines from digital images were used to obtain Fourier coefficients for a polynomial function of 30th degree which were computed with SHAPE v1.2 package, 57 using Elliptic Fourier analysis 565859. For the quantification of organs' shape we only considered the distal part of the aedeagus excluding the apodeme and the gonopods (Shaded areas in Figure 1. Thus, we simplified the studied contour by taking into account only the portion of the organ effectively involved in the penetration of female genitalia.

The area of each aedeagus (in pixels) was calculated from the digital images and considered as an estimator of the size of the organ. We performed a normalization of the descriptors based on the first harmonic ellipse that corresponds to the first Fourier approximation to the contour information (reviewed in 60). Thus, size, orientation and starting position of the contours were standardized in accordance with the size and alignment of the major axis of the first ellipse, leading to representations of the organs that are only based on internal properties of the outlines (i.e. shape) 56.

The variance-covariance matrix of the 120 (4 per harmonic) estimated EFDs coefficients was used as input in a principal components analysis. This procedure allowed us to summarize the information assessed in the coefficients and reduce the dimensionality of the variables 61 in a lower number of principal components (PC). Thus, the resulting scores of each PC of each specimen could be considered as reorganized uncorrelated morphological traits representing different aspects of total shape variation 57 that were used as shape descriptor variables in subsequent analyses. For the sake of simplicity and in order to avoid components accounting for possible biologically meaningless morphological variation only the first 5 components of both intraspecific and interspecific analyses were considered as shape variables in the ANOVAs.

We worked with two sets of PC scores. The first set of principal components were calculated from the matrix of coefficients derived form the analysis of the outlines of the genitalia of males of the parental lines employed in successful interspecific crosses and the hybrids. The second set was obtained separately for each species improving the assessment of intraspecific variation in aedeagus morphology by avoiding the noise resulting from conspicuous interspecific morphological differences in the estimation of the PCs. This set was used in the evaluation of intraspecific sources of shape variation. Preliminary analyses with this technique showed that it is repeatable and reliable in species discrimination 62.

Both interspecific and intraspecific size differences were investigated by means of ANOVAs, with Species (2 levels, fixed factor), Cactus (2 levels, fixed factor) and Line (nested in Species random factor) as main sources of variation. The variable was log-transformed to ensure homoscedasticity. According to our experimental design, in the ANOVAs for species, a significant cactus effect (C) may be interpreted as phenotypic plasticity, while significant differences among isofemale lines as due to genetic differences (since all lines were reared under the same conditions). Finally, a significant L × C interaction may be construed as an estimation of genotype by environment interaction (GEI) or more explicitly, that the response of isofemale lines is not independent of the rearing cactus.

Shape variation was assessed by means of MANOVAs using the scores of each PC as dependent variables and Cactus and Line as main sources of variation.

In the presence of allometry, a fraction of the changes in shape might be explained by changes in the size of the organ. We were interested to comparatively evaluate the amount of allometric change in genital morphology in these species. Thus, we calculated Pearson's product-moment correlation matrix among shape scores and aedeagus size in each species separately to explore the relationship between aedeagus shape and general size of the aedeagus (within organ allometry).

We also evaluated the allometric relationship between aedeagus size and wing length, a trait correlated with overall body size, which in the studied species are known to be affected by the rearing substrate 41. Size data were log-transformed prior to all analysis.

The principal components scores obtained by means of the general assessment of interspecific shape variation of parental lines and F1 hybrids were used in the examination of morphological patterns of variation in the offspring of interspecific crosses. Size was analyzed by means of an ANOVA and shape with a MANOVA. In both cases the principal factor was Genotype (both parents and the hybrids) and in those crosses in which hybrid larvae could be reared in both cactus hosts, Cactus was also considered as a fixed factor.

Statistica 6.0 63 was used for statistical analyses and in all cases the corresponding assumptions were properly tested.