Only cooccurrences of geographically spread haplo types (i.e. those that were found in at least two sites) were regarded as evidence for recent gene flow between populations, because the mere existence of a single haplotype at numerous sites could simply be the result of unique dispersal events (of sufficient individuals to found a new population), and need not imply that there was an exchange of individuals between different populations. It is important to make this differentiation, because dispersal may be uncoupled from gene flow (e.g. Bohonak and Jenkins 2003).
Morphology
Material and characters
A wide variety of morphological characters was investigat ed systematically for a representative subset of samples (established, distinctive characters and new ones were tested in a consistent manner). Only those characters that were of sufficiently low variability and showed differences among the major phylogenetic sublineages of Triops mauritanicus were subsequently studied for the remaining samples. For males, additional samples of the African T. m. mauritanicus (comprising specimens from populations 108,
109, 112, 113 and 115–119; Table A1) and T. m. simplex
(populations 104–107) were investigated in order to compare the morphology among all known sublineages in the species. In total, the morphology of 459 individuals was investigated. The size of the studied specimens varied
widely, so that several of the morphological characters needed to be standardized for size. We used the minimal width of the telson at its anterior margin (henceforth called telson width; Fig. 1a) as a surrogate for body size. Consistent measurements of total body length are impossi ble in fixed specimens because of variable degrees of body contraction during fixation (Longhurst 1955).
To validate the usefulness of telson width as a measure of body size, we compared it to the carapace length for a representative subset of samples [carapace length shows isometric growth in Triops (see Longhurst 1955), but in T. mauritanicus the long terminal carina spines are frequently broken in preserved specimens, raising the need for another character to represent body size]. Graphic presentation of the resulting data using logarithmic scales revealed a growth coefficient (k) of 1.018, which is indicative of isometric growth of carapace length and telson width (carapace length = 7.57 * telson width^{1.01}^{8}; Hartnoll 1978) and confirms the usefulness of the latter as a measure of body size. All measurements were made on digital photo graphs using PixeLINK Capture SE (Version 3.1, obtained from www.pixelink.com). Photographs of trunk limbs (mounted on microscope slides) were taken with a PL B686CU PixeLINK colour microscopy camera on a Nikon Eclipse E200 microscope, using 2–40× magnification lenses.
Telson morphology
Korn et al. (2006) established the telson length ratio (the ratio of furcal spine length to the distance between furcal spine tip and the anteriorlateral edge of the telson) to characterise the size of the furcal spines. Following Korn
et al. (2006), we used two subsidiary lines (telson subsidiary line and furcal subsidiary line; Fig. 1a) to define the anterior starting point of furcal spines as the point where both subsidiary lines meet. In the present survey, we used additional characters to describe the shape of furcal spines: (1) the furcal spine width, measured at the anterior starting point of furcal spines as defined above (to standardize for the size of investigated specimens, the ratio of furcal spine width to telson width was used to represent this character in statistical analysis); (2) the ratio of furcal spine length to furcal spine width (henceforth called furcal spine size ratio).
We call the central part of the telson (i.e., excluding furcal spines) posterior to the telson subsidiary line (Fig. 1a) the telson posterior marginal section. Its posterior margin is typically incised medially, giving it a bilobed appearance. We measured the distance from the foremost point of the margin within the medial incision to the telson subsidiary line (henceforth called minimum length of posterior marginal section). In addition, the distance from the telson subsidiary line to the posteriormost points of the lobes was determined (maximum length of posterior marginal section, expressed as the mean from measure ments of both lobes), as was the distance between these two posteriormost points (lobe distance of posterior marginal section). Furthermore, we measured the area of the telson posterior marginal section. In specimens lacking a clear incision in the posterior margin, we used the distalmost points in which the maximum length of the posterior marginal section was reached as fixpoints for measuring maximum length and lobe distance of the posterior marginal section. Measurements were made on digital photographs of the telson taken in dorsal view. Subsidiary
Fig. 1 Triops mauritanicus, schematic drawings of morphological features and measurements. a Posterior part of abdomen, dorsal view (modified from Korn et al. 2006), with dotted lines used in measure ments concerning furcal spines and posterior marginal section of telson. b Distal part of second trunk limb, anterior view, with dashed lines used in length measurements of endopodite and fifth endite.
Abbreviations: ENP = endopodite; EN4, 5 = fourth, fifth endite; F = furcal ramus; FSL = furcal subsidiary line; MAX, MIN = maximum, minimum length of telson posterior marginal section; SMSP = submarginal spines; SP = furcal spine; SW = furcal spine width; TE = telson; TLD = telson lobe distance; TPMS = telson posterior marginal section; TSL = telson subsidiary line; TW = telson width
lines were drawn in PixeLINK Capture SE using the
‘Annotate’ function.
Trunk limb morphology (nomenclature following
Fryer 1988)
Spine counts of the tenth trunk limb Each endite bears a row of submarginal spines on the anterior face, and one row of meshwork spines each on the anterior and posterior faces of the endite. For the present survey, we counted the number of spines of the anterior row of meshwork spines on endite three, as well as the number of submarginal spines on endite four in the tenth trunk limb. In some specimens, the smallest submarginal spines are positioned at the edge of the endite, or are even displaced to its posterior face. Thus, to count the number of submarginal spines, the fourth endite was investigated in anterior and posterior views, at 100–400× magnification.
Morphology of the second trunk limb The characters of the second trunk limb were studied in males only, since they show different levels of modification between sub lineages, possibly linked to the functional role of the anterior trunk limbs in mating (for further modifications attributed to the role of anterior trunk limbs in mating, see Lynch 1972). The characters investigated in this study show different patterns of allometric growth. Formulas to standardize these characters for the size of investigated specimens were derived from bestfit curves as described for the following example: if the curve was indicated as Y = a − b * X, where X is telson width [mm], then each observed point (X_{i}, Y_{i}) was transformed into a size standardized point (X_{standar}_{d}, Y_{i}*). Standardized values of Y_{i }_{ }(i.e., Y_{i}*) were thus obtained by application of the
^{for}^{m}^{u}^{l}^{a}^{:}^{ }^{Y}_{i}_{ }^{»}^{ }^{¼}^{ }^{Y}_{i}_{ }^{þ}^{ }^{b}^{ }^{»}^{ }^{X}_{i}_{ }^{.}
Maximum length of submarginal spines on the second trunk limb The proportional length of spines gradually decreases in size towards the anteriormost trunk limbs. On the fifth endite of the second trunk limb, they are confined to the proximal region of the endite, and do not appear to play any functional role (M.K. pers. obs.). In male specimens, we measured the proportional (percent) length of the longest submarginal spine of the fifth endite in relation to the length of the fifth endite (henceforth called proportional spine length). We used the ‘polyline’ tool implemented in PixeLINK Capture SE to measure the length of the fifth endite, as it is usually curved towards its base in Triops mauritanicus (see Fig. 1b). To standardize this character prior to analysis, the following formula was applied: Y_{i}_{ }^{»}^{ }¼ Y_{i}_{ }þ 0:7 ^{»}^{ }X_{i}_{ }, where Y = proportional spine length, and X = telson width [mm].
Proportional length of the endopodite in relation to length of the fifth enditeIn male specimens, we measured the proportional (percent) length of the endopodite in relation to the length of the fifth endite on the second trunk limb (henceforth called proportional endopodite length). This character shows a nonlinear correlation to body size. Thus, prior to analysis, data were transformed by applying the formula: Y_{i}_{ }^{»}^{ }¼ Y_{i}_{ }þ 30 ^{»}^{ }LOG_{10}_{ }X_{i}_{ }, where Y = proportion al endopodite length, and X = telson width [mm]. Measure ments were made on digital microscopy photographs of the trunk limbs in anterior view. The length of the fifth endite was measured using the ‘polyline’ function as described above, whereas endopodite length was measured using the
‘caliper’ tool implemented in PixeLINK Capture SE.
Number of apodous abdominal segments
The number of apodous abdominal segments was counted using the methods described in Korn et al. (2006; nomenclature following Longhurst 1955; whether or not Notostraca have a true abdomen still awaits confirmation, see Schram and Koenemann 2004).
Size of resting eggs
The outer coating (comprising an alveolar layer, covered distally by an outer cortex) of the resting eggs is still smooth at the time when the eggs are deposited. [Thiéry (1987) states that the alveolar layer swells after the eggs are exposed to water, when they are released from the brood pouches. During this process, the thickness of the alveolar layer is reported to expand from approx. 20 μm to 55–100 μm. Clearly, this requires that the outer cortex and the alveolar layer are still flexible at that time.] Consequently, the eggs adapt their shape in response to the morphology of the sediment (M.K. pers. obs.). Very fine sediments typically result in a roughly ballshaped resting egg, whereas coarse sediments usually result in asymmetric shapes of the outer coating. Thus, a simple measurement of the egg diameter appeared to be inappropriate to characterize the size of the eggs. Therefore, we used digital images to measure profile area, and calculated eggdiameter values by using the standard formula for diameterarea relationships in a circle, resting egg diameter = 2 * square root (profile area / π), to get a more accurate estimate of the size of the eggs. All eggs were measured in dry condition and were extracted from natural sediments or sediments obtained from lab cultures.
Analysis of morphological data
Data for morphological characters in adult Triops had to be tested separately for males and females due to a high level
of sexual dimorphism in several characters, so that two separate datasets were formed for males and females from the Iberian populations. A third dataset included males of all known sublineages of T. mauritanicus, i.e. including additional samples of the two recognized subspecies occurring in northern Africa, T. m. mauritanicus and T. m. simplex. For each of these morphological datasets, the null hypothesis that there were no significant differences between means of statistical populations was tested with discriminant function analysis. Predetermination of statisti cal groups was based on the molecular determination of Triops populations.
Our sampling was highly asymmetrical due to the fact that levels of abundance and the sizes of distribution ranges clearly differed among the phylogenetic lineages studied. In the Iberian Peninsula, the ‘S.Iberian’ lineage clearly out numbers the other lineages (Table A1). Thus, in order to achieve a less unbalanced design, we included data from only a single randomly chosen individual from each of the populations of the ‘S.Iberian’ lineage in the set of samples used to calculate discriminant functions. The remaining samples were treated as ungrouped cases in the discriminant function analysis and were classified using the classification functions derived from the model. The set of dependent variables was chosen individually for the three datasets to meet all the assumptions of
discriminant function analysis. Some variables had to be lntransformed or square roottransformed in order to reach homogenous variances (see Table 2). Variables that did not reach homogeneity of variances were excluded from analysis. To test for homogeneity of variance, the Hartley Fmax statistic, Cochran C statistic, and the Bartlett Chi square test were calculated, and normality was checked by plotting expected normal values against observed values. A priori classification probability was set to ‘same for all groups’.
The standard method for sampling observations for post hoc classification, i.e. resubstitution, may result in under estimation of the classification error rate even at rather high sample sizes (Lance et al. 2000). To minimize this bias, we additionally used a Jackknife sampling procedure (e.g. Quinn and Keough 2003) to classify observations. For the
‘S.Iberian’ lineage, we had sufficient samples to perform a
modified, populationlevel Jackknife sampling, i.e. for each of the populations all observations were classified based on a model that contained only samples of the remaining populations. The resulting classifications thus represent a realistic, unbiased estimate of the classification success for new independent observations (i.e. individuals from hither to uninvestigated populations).
As a measure of differentiation between phylogenetic lineages, squared Mahalanobis distances (obtained by DFA)
_{Table 2 Morphological characters and character ratios included in discriminant function analysis for each of the three datasets investigated}
Dependent variable

Lineages: Sex:

All
Male

Iberian
Male

Iberian
Female

Number of anterior meshwork spines on 3rd endite of 10th trunk limb


Included

Included

Included

Number of submarginal spines on 4th endite of 10th trunk limb
Proportional spine length on 5th endite of 2nd trunk limb [%]


Included
Included^{a}

Included
Included

Included
Unavailable

Proportional endopodite length of 2nd trunk limb [%]


–

Included

Unavailable

Telson length ratio
Furcal spine width / telson width
Furcal spine size ratio


Included Included^{b }Included^{a}

Included
–
Included^{a}

Included Included Included

Number of apodous abdominal segments


–

Included

Included

Minimum length of TPMS / area of TPMS


–

–

Included

Maximum length of TPMS / minimum length of TPMS


–

Included

Included

Length of telson posterior incision / telson width


Included

Included

Included

Length of telson posterior incision / maximum length of TPMS


Included

Included

Included

Length of telson posterior incision / area of TPMS


Included^{b}

Included

Included^{a}

Area of TPMS / telson width


Included

Included

Included

Telson lobe distance / maximum length of TPMS Telson lobe distance / area of TPMS


Included^{a}
Included^{a}

–
Included

–
Included

Telson lobe distance / telson width


Included

Included

Included

– excluded from analysis, as variable did not reach homogeneity of variances even after data transformations were applied
TPMS telson posterior marginal section
^{a}^{ }data lntransformed
^{b}^{ }data square roottransformed
were calculated between the group centroids (multivariate means) of the phylogenetic lineages. For comparison with molecular phylogenetic reconstructions, a NJ tree based on squared Mahalanobis distances between the group centroids of males of all phylogenetic lineages was calculated using PAUP* (Swofford 2003).
For the size of resting eggs, the null hypothesis that there were no significant differences between means of populations was tested with a singlefactor analysis of variance (ANOVA). To test for homogeneity of variance Levene´s test was used, and normality was checked by plotting expected normal values against observed values. A logarithmic transformation was used, which greatly improved the approximation to a normal distribution and homogeneity of variances within this dataset. However, since the assumption of homogeneity of variances was still clearly violated, only a data subset that met all the assumptions of ANOVA was used for calculating statis tics. This subset excluded some populations of Triops c. cancriformis with unusually low variability (populations
121–123 and 130, see Table A1), but retained all populations that were important in evaluating the useful ness of this morphological character for discriminating among phylogenetic lineages, as the mean values of excluded populations were within the range of those observed in the other populations of this species. As the null hypothesis was rejected, differences among single populations were investigated using a Tukey posthoc test. All statistics on morphological data were undertaken with STATISTICA 6.0 (StatSoft, Inc.).
Results
Nucleotide composition, substitution patterns and sequence variability
The nucleotide composition in the 12S rDNA gene segment sequenced showed a pronounced ATbias (33.0% T, 38.9% A, 17.9% C, 10.2% G) for the ingroup (Triops cancriformis + T. mauritanicus). The alignment consisted of 552 sites, of which 446 (80.8%) were constant within the ingroup. Within this lineage 101 sites were variable, and 78 of these (14.1% of the total sequence) were parsimony informative. The mean 12S sequence divergences between the sublineages of the ingroup, and the maximum sequence divergences within the sublineages are presented in Table 3.
The dataset of 16S sequences also showed the AT bias (33.1% T, 31.9% A, 12.7% C, 22.3% G); it consisted of 432 sites, of which 389 were constant, 40 were variable, and 24 were parsimony informative. The combined 12S and 16S dataset consisted of 984 sites, of
which 847 were constant, 130 were variable, and 97 were parsimony informative (outgroups not included, but calculated using the alignments that were used for phylogenetic reconstructions).
Phylogenetic analysis
The molecular analysis of a high number of Triops populations from southwestern Iberia revealed the pres ence of a sixth, previously undiscovered main lineage within T. mauritanicus, occurring in western parts of Cádiz province (Fig. 2; ‘Cádiz’ haplotypes in Table A1). The ML calculation based on the 16S dataset (not shown; results available from http://purl.org/phylo/treebase/phylows/ study/TB2:S10349) indicates the newly discovered lineage in a sistergroup position to the remaining ingroup taxa, thus rendering T. mauritanicus paraphyletic with respect to T. c. cancriformis. The MP calculation (16S) could not resolve the relationships among the ingroup lineages. In contrast, all calculations based on the 12S haplotypes, as well as the dataset with 12S and 16S sequences combined, indicated T. mauritanicus and T. c. cancriformis as monophyletic sister taxa (Fig. 2). The ML calculations resulted in similar topologies for the 12S and the combined datasets (Fig. 2a, b). For the combined dataset, the first of two ML trees is presented (Fig. 2b; topologies of the second ML phylogram and the Bayesian inference majority rule tree were identical with respect to relationships among the main lineages). The T. mauritanicus samples form three separate monophyletic clusters (Fig. 2a, b): (1) a clade consisting of the western Cádiz samples belonging to the
‘Cádiz’ haplotype group; (2) a clade formed by the south central Portuguese samples and Spanish samples from Extremadura, Sevilla, Huelva and northern Cádiz provinces (‘S.Iberia’ haplotype group); (3) a clade including the southwest Portuguese samples (‘Portugal’ haplotype group), the samples from ‘Gitanilla’ lineage, as well as the samples of T. m. simplex and T. m. mauritanicus. Within that third cluster, southwest Portuguese samples and samples from the ‘Gitanilla’ lineage form a monophylum either in an unresolved trichotomy with the African subspecies T. m. simplex and T. m. mauritanicus (dataset with 12S and 16S sequences combined) or forming the sister group to the latter two (12S dataset). MP calculations could not resolve relationships among the T. mauritanicus clades using 12S sequences alone, but resolved all T. mauritanicus clades in the calculation based on the combined dataset (strict consensus tree presented in Fig. 2c): in this phylogeny reconstruction, the ‘S.Iberian’ haplotype group forms the sister group to the remaining samples within T. mauritanicus. Among these remaining samples, the ‘Cádiz’ haplotype group is in a sistergroup relationship with a clade comprising two monophyletic 