Inferring the Geographic Range Evolution

Introduction
Some methods in biogeography are based on the assumption that there is a single branching pattern among areas caused by vicariance and that this pattern is common to many different groups of organisms (Nelson, 1974; Rosen, 1976; Nelson and Platnick, 1981). Other approaches points to the reconstruction of the distribution history of individual groups (taxon biogeography) and in the search for general area relationships (area biogeography); the latter use character optimization methods which allow the reconstruction of ancestral distributions without constraining area relationships to hierarchical patterns (Bremer, 1992; Ronquist, 1994). Among this methods are the Dispersal-vicariance analysis (Ronquist, 1997) that uses a Fitch Optimization and the Dispersal-extinction-cladogenesis (DEC) model (Rei and Smith, 2008) that implements a maximum likelihood optimization. The objective of the present study was to reconstruct the ancestral distributions using both approaches and to contrast the findings.

Methods
Sequence data information for the avian genera Pipilo and Toxostoma previously published (Zink et al., 1998; 1999) were used. For Pipilo sp. a mitochondrial region control, the cytochrome b and NADH dehydrogenase subunit 2 genes were considered. For Toxostoma only the mitochondrial region control and the cytochrome b genes were Included. Each gene for each taxon was analyzed separately. The sequences were aligned with Muscle software (Edgar et al., 2004) using the default parameters. The best-fit model of nucleotide substitution was determined using a hierarchical likelihood ratio test (Posada and Crandall, 2001) as implemented in the Modeltest software (Posada and Crandall, 1998). Maximum Likelihood (ML) optimization analyses were done in phyML software (Guindon and Gascuel, 2003). The distributions of the taxa and their ancestral area were described in terms of the areas proposed by Zink et al. (2000) with minor modifications (Figure 1): California plus Baja California (area A), Sonoran desert (area B), Chihuahuan desert plus Sinaloan shrubland (area C), and the highlands of southern Mexico (area D). To reconstructs the ancestral distributions for the areagrams, a dispersal-vicariance optimization (Ronquist, 1997) was undertaken in DIVA software (Ronquist, 1996) and a Dispersal-extinction-cladogenesis (DEC) model in Lagrange (Rei and Smith, 2008).

Results and Discussion
The final data sets for each gene included six taxa for Pipilo sp. and seven taxa for Toxostoma sp. (GenBank accession numbers available upon request). The lengths of the obtained alignments with Muscle software (Edgar et al., 2004) are presented in Figure 2. The Hasegawa-Kishino-Yano plus Γ distribution model (HKY + Γ model) (Hasegawa et al., 1985) was the best fit to each data with an α (shape parameter) value of 0.3 for the mitochondrial region control gene of Pipilo sp. and 0.02 for the remainder data sets. The ML phylogenetic trees are shown in Figure 3. The same relationships were found with each gene for each genus. There were differences in the branch lengths among genes. Only one areagram resulted for each genus as show in Figure 4. In the DEC model the most likely ancestral areas for Pipilo sp. and Toxostoma sp. were the area B and area D respectively, with other areas for each genus having lower likelihoods (−ln(L) values available upon request) (Figure 5). The dispersion-vicariance optimal distribution showed as the ancestral area for Pipilo sp, the union of the areas BD and for Toxostoma the combinations AD, BD and CD. Because in the DEC model widespread ranges are the direct outcome of dispersal events, some optimization (see Figure 5 for Pipilo sp. /NADH gen) are the outcome of solely dispersion and extinction. In all the scenarios, the number of biogeographical events required for explain the actual distribution are lower in the DIVA reconstruction that in any of the DEC model reconstructions; because some cladogenesis events are explained by DIVA as a result of vicariance from a widespread ancestor, and not by dispersals followed by extinctions in the original area. Like Fitch optimization, DIVA minimizes dispersal and extinction and it is based on Allopatric speciation (vicariance) rather than on sympatric speciation. In the other hand, DEC model assumes that if an ancestor is widespread, the speciation arises either between a single area and the rest of the range (Allopatric speciation), or within a single area (sympatric speciation) (Ree et al., 2005). Nonetheless, the results presented here show that DEC model preferred the former one with one daughter species always inheriting a single-area range, and the other inheriting the remainder. To compare how, the branch lengths affect the dispersion and extinction outcome, all the branch lengths in the phylogenetic tree for the NADH gene in the genus Pipilo were set in two separately analyses to 1.000 and to 0.001. The results showed an inverse relation between branch lengths and the dispersion/extinction rates: a long branch length indicated that such taxon had less change of disperse and goes extinct. Hence, under DEC model we have to assume that the he rate of evolutionary change is equal throughout the tree and, furthermore, that we can relate such change with the potential of a taxa to expand or reduce its geographic range. Finally, seems that the restriction of one area to the root could be problematic (and maybe only could work for island scenarios where we can refer to colonization and geography range expansion to the nearest islands. A pure dispersalism approach) when trying to search for general area relationships using different hypotheses and try to fitting areagrams to them. In our data analyses, the DEC model suggested different ancestral areas for each Genus, whereas DIVA considered both possibilities in each case.