The microsatellites used in this study are highly polymorphic, and thus are useful for exploring A. darlingi's population genetic structure. Anopheles darlingi is a species characterized by moderate levels of molecular variability [21–29, 67], and our microsatellite analysis is in agreement with earlier studies. The high allelic diversity and heterozygosity observed in Peru, BV and PLT, Brazil are similar to the results of previously analyzed Amazonian Brazil A. darlingi [37, 38], A. albimanus  in Latin America, and African vectors A. gambiae  and A. funestus [40, 41]. We detected significant departures from HW equilibrium due to heterozygote deficits, and no linkage disequilibrium, in loci in Peru, BV and PLT in Brazil, and in Central America. In contrast, most loci across all populations in eastern Amazonian Brazil had deficits with linkage disequilibrium, interpreted as due to either the Wahlund effect or selection . In western and central Amazonian Brazil significant deficits were detected in 50.79% of the HW equilibrium tests, with only minimal linkage disequilibrium interpreted as null alleles . An allozyme study of two Amazonian populations detected significant deviations from HW equilibrium in 7/8 loci examined , although no significant deviations were detected in earlier allozyme studies [9, 24]. The high levels of heterozygote deficits and null alleles could be the result of an accumulation of mutations in the primer binding sites which may be a consequence of the microsatellite library being constructed of A. darlingi from eastern Amazonian Brazil . The incidence of null alleles found in A. darlingi is similar to that reported from many anopheline microsatellite studies [12, 35, 39–41]; perhaps mosquitoes with large population sizes and high levels of polymorphism are more likely to have null alleles .
The eight microsatellite loci used in this study have not been physically mapped to A. darlingi polytene chromosomes. Therefore, their location with respect to polymorphic chromosome inversions is unknown, and such information may modify the interpretation of the data because neutrality cannot be assumed. Since the analyses were done in two ways, including all amplified loci (5–8) and including only the 5 loci amplified from all populations, we were able to compare the results of these two treatment methods. The differentiation and mean heterozygosity (Table 2) results were not significantly different between these two methods; both recovered very similar values. The allelic richness (Table 2) and the neutrality test estimates showed a little more variance, and the effective population size estimates a large disparity between the two treatment methods, which demonstrates that this test is more sensitive and should be interpreted with caution. Although there was variance in these estimates, the same trends were shown in both treatments.
Substantial population structure was found in Amazonia, which was undetected with more conservative nuclear markers and isozymes [9, 10, 24, 71]. Four population clusters were detected in Amazonia, three in the Brazilian Amazon (northeastern Amazonia, southeastern Amazonia and central, and western and central Amazonia) and one including the Peruvian Amazon subpopulations, attributed to an isolation-by-distance effect. There was a moderate amount of significant differentiation and reduced gene flow between these Amazonian population clusters. The considerable differences in Ne among the populations may have contributed to the observed genetic differentiation [72, 73]. The level of differentiation among the Amazonian population clusters is comparable to that detected between A. albimanus populations from Central and South America (FST = 0.114) , among A. gambiae populations in west Africa separated by > 200 km (FST = 0.034–0.167)  and those separated by the Great Rift Valley complex in Kenya (FST = 0.104) , as well as between A. funestus populations from west, central, and eastern Africa (FST = 0.110) . An earlier mtDNA study of A. darlingi , although lacking western Amazonian Brazil samples, detected considerable population structure throughout South America that is congruent with some of the Amazonian differentiation detected here; specifically, differentiation across the Amazon River  and between the NEA and SEAC Brazil population clusters was also detected with mtDNA . The main forces responsible for partitioning the genetic variation in Amazonian are most likely the result of geographic distance and/or differing demographic histories, rather than physical barriers (e.g., rivers or mountains).
Within the WCA Brazil population cluster there was little genetic structure and differentiation, and the isolation-by-distance model explained nearly all of the differentiation observed . Within the NEA Brazil population cluster there was no significant population structure or differentiation, likely because these three localities are 4–8 km apart and probably a single population. Within the SEAC Brazil cluster there was more structure and significant differentiation than observed for the other Amazonian clusters, which is explained by isolation-by-distance and also may be affected by the differing effective population sizes among these subpopulations. The two central Amazonian Brazil populations, PEX and PLT, were an admixture of the Amazonian clusters. PEX was primarily an admixture of SEAC and WCA Brazil populations, which are the two nearest population clusters. Interestingly, PLT shared identity primarily with the SEAC populations, which are in close proximity (although not the closest), and secondly shared identity with the Peruvian populations that are 1611–2044 km apart. BV, the northern Amazonian Brazil locality, was most similar to the southeastern Amazonian Brazil populations, which again were not the nearest. This demonstrates that their population identity was not solely based on proximity, and may be influenced by demographic history, migration, and/or ecology.
Within Peru there was no significant population structure and low differentiation among the seven subpopulations, in agreement with an earlier RAPD-PCR analysis of A. darlingi in the Peruvian Amazon that detected high homogeneity among populations (within 60 km) irrespective of different habitat types . We detected little differentiation between the subpopulations even at distances up to 433 km and there was no indication of isolation-by-distance. Most of the significant low differentiation among the subpopulations occurred between samples greater than 120 km apart, except for between PCO-NAU (59 km apart, significant differentiation), PCO-PRT (134 km apart, no significant differentiation), and MAZ-PRT (147 km apart, no significant differentiation). There was a large amount of variability in Ne among the Peru subpopulations (93.6 – 8) that may contribute to the small significant differentiation observed among many of the localities. Anopheles darlingi appears to be panmictic in this region of Peru. There is some evidence of a population expansion in MAZ, NAU, and SAE. The expansion in NAU and SAE is reflected in a very large Ne in these localities. Prior to 1991, A. darlingi had not been reported around Iquitos, the major Peruvian Amazon city [43, 44]. This expansion may reflect the introduction of A. darlingi into the Peruvian Amazon possibly from PLT, Brazil, where there is the most genetic similarity, in the early 1990's and the resultant increase in malaria [45, 46].
Within Central America there was much less variation (mean RS = 4.3, mean HO = 0.457) as compared to within Amazonia (mean RS = 7.62, mean HO = 0.742), and there was no evidence of isolation-by-distance. Low haplotype and nucleotide diversity was also observed within Central America with mtDNA COI sequences as compared to within South America . The low diversity can be at least partially explained by low effective population sizes in this region, or perhaps these populations suffered a recent population bottleneck due to an unknown historical event. A founder effect resulting from the establishment of the Central American A. darlingi populations from only a few individuals from the Colombian population is also consistent with the data . The only significant differentiation observed among the six Central American subpopulations was between GOL, Belize and all other localities (FST range of 0.1063–0.1489, all P values < 0.05), GOL is separated from the other subpopulations by the Guatemalan Highlands and the Maya Mountains (Figure 1), which may act as a natural barrier, restricting gene flow. There was no significant differentiation between the northern Belize populations (CAV and SIB) and the Guatemalan populations, although they are separated by 257–270 km and by the mountain ranges as well. Therefore, in northern Belize and within Guatemala A. darlingi appears to be one panmictic unit. In comparison, A. albimanus populations throughout Central America displayed only minor genetic differences using microsatellites, there was weak isolation by distance, throughout Guatemala populations were genetically homogenous between Atlantic and Pacific regions and thus the Guatemalan Highlands did not appear to restrict gene flow . The level of differentiation observed between GOL and the other A. darlingi Central American populations was similar to that observed between A. albimanus populations in Central and South America .
The data suggest that the main division within A. darlingi corresponds to Amazonia (genotype 1) and Central America (genotype 2) . Earlier nuclear white, ribosomal ITS, and mitochondrial COI sequence data together established a deep divergence between genotypes 1 and 2 [10, 28], interpreted as incipient species . In the present study, there is marked differentiation between Central America and all four Amazonian population clusters. All pairs of genotype 1 and 2 populations showed a large amount of highly significant differentiation, there was little or no recurrent gene flow between them, they demonstrate different microsatellite allele frequencies and variation, and appear as separate clusters with the Bayesian analysis. The NJ trees based on genetic differentiation and distance both cluster the populations according to the two genotypes. The mixture of shared and private alleles in the Central America population cluster is consistent with shared ancestral polymorphism and a recent divergence between these two genotypes. The presence of a large amount of private alleles suggests some degree of independence between the gene pools . The independent pairwise differentiation analyses of each locus found significant differentiation across the genome between genotypes 1 and 2. The differentiation observed between the genotypes was attributed to isolation by distance, although, as the graph shows (top right portion of Figure 3), the comparisons between Central and South American populations do not fit the positive correlation trend line, and may be a consequence of comparing diverse genetic groups that are geographically separated [11, 68]. The level of differentiation observed between genotype 1 and 2 populations was similar to that observed among the closely related A. dirus, A. scanloni, and A. baimaii (former A. dirus species A, C, and D, respectively) in Thailand (mean FST = 0.263) , A. gambiae M and S forms (FST = 0.1–0.3 ; mean FST = 0.203 ), and between A. gambiae and A. arabiensis (FST = 0.12–0.27 ; mean FST = 0.349 ) using microsatellites. These microsatellite data are consistent with and substantiate the hypothesis, initially proposed based on mitochondrial and nuclear data [10, 28], that genotypes 1 and 2 represent incipient species within A. darlingi. The divergence between these genotypes was estimated to have occurred during the Pleistocene using mitochondrial data, most likely attributed to complex Pleistocene climatic changes .
With the detection of a recent population expansion or the departure from MDE in many of the populations in Amazonia and two populations in Central America, the FST values do not translate into meaningful rates of gene flow . In the expanded populations, the migration rates will be overestimated by FST, and the differentiation will be underestimated as compared to neutral equilibrium values. Therefore, the low level of differentiation measured within Peru and WCA Brazil may be an underestimation as well as an overestimation of gene flow; and, the differentiation and gene flow between the genotypes and population clusters may be underestimating the current degree of isolation. Despite possible departures from MDE, our large sample sizes and number of populations add statistical power to our study.