We analyzed genotype data for 678 autosomal and 29 X-chromosome microsatellites collected in the Mestizo populations together with similar data available in samples from Europeans, Native Americans and Africans
. Bar charts summarizing the estimated ancestry proportions of the populations examined are shown in (the corresponding values and standard errors of these estimates are presented in Table S2
). The autosomal data indicate substantial variation in Native American ancestry, ranging from ~70% in Salta to ~20% in Rio Grande do Sul (RGS), the Central Valley of Costa Rica (CVCR) and Medellin. African ancestry is low (<5%) in most of the populations examined, although it approaches 10% in Medellin, RGS and Oriente. African ancestry is often accentuated in a few outlier individuals for each population (Figure S1
). The observed variation in ancestry is consistent with historical differences in Native population density and with the extent of past immigration to the regions sampled. The Mestizo with the highest Native ancestry are in areas which historically (and to the present) have had relatively large Native populations: Andean regions (Salta, Huilliche) and meso-America (Mexico City, Oriente), where major pre-Columbian civilizations developed
. By contrast, the Mestizo with highest European ancestry (CVCR, Medellin and RGS) are from areas with relatively low pre-Columbian Native population density (occupied then by heterogeneous groups of chiefdoms or hunter-gatherers) and where the current Native population is sparse
. Categorizing the Mestizo examined into three groups, based on the relative pre-Columbian Native population density in the region (Table S1
), results in a significant Spearman rank correlation with levels of Native ancestry (ρ
0.04). The highest African ancestry (~10%) occurs in Mestizo in relative proximity to circum-Caribbean areas (Oriente and Medellin) and in Southern Brazil (RGS), and thus at the periphery of regions with large past African immigration. Based on the autosomal data, estimates of the mean time since admixture in the 13 Mestizo populations range between ~6–14 generations (Table S3
), in agreement with independent estimates made in some of the populations examined here 
and consistent with the notion that most admixture in these populations is likely to have occurred in colonial times
. These age estimates are obtained based on the inferred mean frequency of transitions in ancestry along the genome, under the assumption of a single past admixture event
. These estimates should be viewed with caution, as it is doubtful that such as model applies in the populations we examined and the added complication of a non-negligible three-way admixture in some of these populations (). The observed variation in the estimated age of admixture in different populations is in fact likely to be influenced by variable levels of historical gene flow in different regions
: the relatively more isolated populations (e.g. CVCR)
tend to show older age estimates, while populations in the vicinity of large local native populations (e.g. Salta, in Northern Argentina)
or near areas of recent European immigration (e.g. RGS, in Southern Brazil)
show younger estimates, consistent with more recent gene flow and possibly ongoing admixture.
Ancestry Proportions in 13 Mestizo Populations.
A positive correlation (ρ
<0.01) is observed between autosomal heterozygosity and European ancestry (). This increase in heterozygosity with higher European ancestry agrees well with expectations, based on the difference in mean diversity of European and Native American populations and their genetic differentiation, as measured by FST
<0.01; See Methods
). No significant correlation is seen between settlement size and genetic diversity or between settlement size and Native American ancestry (results not shown). Large differences in the variation of individual admixture estimates were seen across populations, with the variance in Native American ancestry between individuals ranging from 0.005 in Quetalmahue to 0.07 in Mexico City (, Figure S1
, and Table S2
), an observation consistent with previous studies
Heterozygosity vs. Proportion of European Ancestry in Mestizo Populations.
Distribution of Admixture Estimates for Individuals from Mexico City and Quetalmahue.
At the X-chromosome level, the proportions of African and Native American ancestry estimated are usually larger than those based on autosomal markers, with a concomitant reduction in European ancestry (, Wilcoxon signed rank test P
0.02). This pattern is consistent with admixture involving predominantly European men and Native women. Such a sex bias in European-Native admixture has been inferred in Mestizo populations mainly based on mtDNA and Y-chromosome polymorphisms
and the data collected here confirm that it is a common phenomenon across Latin America. Interestingly, these data also indicate that a similar sex bias in admixture applies (even more strongly) to the African ancestry in Mestizos: in all the populations examined a higher estimate of African ancestry is observed on the X-chromosome than on autosomes (, Wilcoxon signed rank test P
<0.001). Such a sex bias in African admixture has been inferred African Americans from the US
but had not been evidenced in Mestizos. also indicates that the difference in European ancestry between X-chromosome and autosomal markers is positively correlated with the extent of European ancestry of the population (ρ
0.01). This suggests that the sex bias of admixture has been more pronounced in areas with lower Native population density, consistent with the observation that Mestizo populations from areas with low Native population density (such as Medellin and CVCR) can have a predominantly European autosomal background and at the same time an almost exclusively Native American mtDNA ancestry
. This pattern could also have been influenced by the collapse of the Native population soon after the establishment of the Mestizo in these regions, and the continuing immigration of European men over several generations
. A relatively high sex bias of European/African admixture in the regions sampled here (possibly associated with a historically low African population density) is consistent with the uniformly higher estimates of African ancestry obtained with the X-chromosome relative to autosomes ().
Admixture analyses generally face the difficulty of not knowing with certainty the specific ancestral populations that were involved in the admixture, particularly since such ancestral populations might not be available for study or they could have undergone extensive genetic drift. Admixture estimates are therefore usually obtained by pooling data from related putative ancestral population samples, as a way to approximate a “mean” ancestral gene pool. All previous reports of admixture in Latin American populations have therefore pooled population data from African, European and Native American samples into “continental” samples; as done for the analyses discussed above. However, since there is a high level of population structure amongst ancestral Native American populations 
it is conceivable that the Native component of Mestizos could be genetically differentiated across different geographic regions. We investigated whether it is possible to detect such an underlying genetic differentiation amongst Mestizos through an admixture analysis allowing for a structured ancestral Native American population sample (see Methods
). The results from this analysis are not strictly ancestry proportions reflecting an underlying admixture between multiple Native populations. This is particularly so because the proportions obtained are influenced by the variable degree of genetic relatedness amongst the various Native groups examined. Rather, these proportions reflect the relative genetic similarity of the Native American component in the Mestizo to the Native groups considered in the analyses. shows such a partitioning of the Native American ancestry in Mestizos when admixture is estimated with data from Native Americans subdivided based on linguistic grounds; using the classification of Ruhlen
(the corresponding values and errors of these estimates are presented in Table S4
). Generally, the native component in the Mestizo shows greatest genetic similarity to Native populations from the linguistic stock which is most widespread in the region where the Mestizo population sampled is located (): Central/Northern Amerind in Mexico City and Oriente; Chibchan-Paezan in CVCR, Medellin and Peque; Andean in Pasto, Salta, Catamarca and Quetalmahue. RGS shows no strong similarity to Natives from any linguistic stock but is the Mestizo population with greatest similarity to the Equatorial-Tucano, consistent with RGS being the Easternmost of the Mestizo populations examined (). Overall, these observations agree with the expectation that admixture is likely to have involved mainly Natives from the region where the Mestizo populations are located.
Regional Native American Ancestry of 13 Mestizo Populations Considering the Major Native American Linguistic Stocks.
Performing a similar admixture analysis but this time considering each individual Native American population as ancestral, revealed several instances of increased similarity between the Native component in the Mestizo and Native populations located in relative geographic proximity (, and Table S5
). Most notably, Quetalmahue (in Southern Chile) shows a strong genetic similarity to the Huilliche, a Native population from the vicinity. Also, the population of Paposo in Northern Chile is markedly more similar to the neighboring Aymara than to any other Native American population. The three populations from North West Argentina (Salta, Tucuman and Catamarca) show greatest genetic similarity to the Quechua (sampled in Southern Peru) and the Aymara (sampled in Northern Chile). The population of Pasto in Southern Colombia is most similar to the Inga, a Quechua-speaking population also from Southern Colombia. Peque in North-West Colombia shows greatest similarity to the Wayuu (sampled in Northern Colombia and genetically closest to Chibchan-speakers, although not classified as Chibchan
) and the Cabecar (from Costa Rica, in lower Central America). The Cabecar are also the Native population most similar to the Mestizo population of the Central Valley of Costa Rica (CVCR). Finally, Oriente (in Guatemala) shows greatest genetic similarity to the Maya (sampled in Southern Mexico) and the Kaqchikel (sampled in Guatemala). The populations of Mexico City, Medellin, Cundinamarca and RGS appear to have more heterogeneous Native American ancestries.
Regional Native American Ancestry of 13 Mestizo Populations Considering each Individual Native Population.
The congruence between pre-Columbian genetic structure and the genetic differentiation of the Mestizo is also evidenced in the correlation of the logarithm of the geographic distance between Mestizo and Native populations, and the size of the corresponding ancestry components (as shown in ). These correlations are negative for all Mestizo populations (), in agreement with a stronger genetic affinity of the Mestizo to Native populations that are geographically closer. The correlations are usually higher when considering an effective geographic distance (a distance considering preferential migration along the coastal outline, see Methods
), consistent with the influence of the coasts on Paleolithic Native population dispersals
. The mean weighted Pearson correlation coefficient over all mestizo populations is −0.481 (R2
0.232) for Euclidean distances and −0.570 (R2
0.325) for effective distances.
Correlation between Geographic Distance and Estimated Ancestry Components.
The analysis of admixture considering a structured ancestral Native American population also suggests that a stronger regional ancestry is present in the smaller urban centers sampled. The variance of the estimated Native ancestry components is negatively correlated with the logarithm of population size (Table S1
), both in the linguistic-based analysis (, ρ
0.05) and in the population-based analysis (, ρ
0.05). The more heterogeneous Native American ancestry of larger urban agglomerations is consistent with them having attracted immigrants from relatively distant areas, thus potentially tracing their ancestry to various, differentiated Native groups. By contrast, smaller urban centers appear to trace their ancestry to fairly defined Native groups, with subsequent maintenance of greater isolation than larger agglomerations. This genetic pattern agrees with demographic data showing that the expansion of major cities in Latin America has often been driven by regional immigration rather than by internal growth
The large variation amongst the Mestizo examined in the mean Native American ancestry and in individual admixture proportions, and the regionally differentiated Native American ancestry, testify to the marked genetic heterogeneity of Latin American populations. These observations have a number of implications for the application of admixture mapping in the region. The large variation in mean Native American ancestry between populations implies that the power of admixture mapping will vary considerably in studies targeting different geographic areas
. The differentiated Native Ancestry of Mestizos will affect the informativeness of admixture maps across populations and could result in an increase of false positives when admixture mapping is attempted populations other than the one from which informative markers were selected. Ideally, admixture maps should therefore be developed for each Mestizo population studied. An alternative would be to select markers for mapping based on their lack of differentiation across Native American populations. Our results also show that mean African ancestry in Mestizo populations is typically low (<10%). This reduces the potential complexity of an extensive three way admixture and confirms that admixture mapping in these populations should be feasible within the two-population admixture framework usually considered
. Mapping in Mestizos should thus be practical with marker maps that mainly distinguish Native from European ancestry (or Native from non-Native ancestry), possibly supplemented by the exclusion of outlier individuals showing a marked increase in African ancestry. It is likely, however, that areas where historically there has been substantial African immigration (e.g. circum-Caribbean areas) will show higher levels of African admixture and represent additional challenges for admixture mapping. Finally, individual admixture estimates can vary markedly in certain Mestizo populations, particularly in large urban agglomerations such as Mexico City (). These populations could be particularly useful for evaluating the effect of ancestry on phenotype, an important initial step prior to admixture mapping of genes influencing such phenotypic variation.
In conclusion, this initial genome-wide analysis of admixture across Latin America has revealed a hitherto undetected differentiation of the Native American ancestry in Mestizos. This fact, together with the extensive variation observed in rates of admixture across populations, and sometimes also between individuals within populations, needs to be considered when designing admixture mapping studies in specific Latin American populations. Despite these complications, we anticipate that admixture mapping in Mestizos should prove a fruitful strategy for analyzing the genetic basis of phenotypic traits, including disease, differing between Native Americans and Europeans.