We unite genetic data with a robust test of niche divergence to test the hypothesis that patterns of gene flow between two lineages of the nine-banded armadillo are influenced by their climatic niches. We collected Geographical Information System (GIS) data on climate using locality information from 111 individuals from two lineages that had associated genetic material. We tested whether niches of these lineages were more conserved or divergent than the background environments of their geographic ranges and found evidence for niche conservatism on two axes and no evidence for divergence on any axis. To address the role of niche similarity in gene flow, we genotyped the 111 individuals at five microsatellite loci and tested whether admixed individuals tended to be located in parts of multidimensional environmental space (E-space) shared between the two lineages. We observed an asymmetrical pattern of overlap, in which the West lineage's E-space was almost completely included inside East lineage's E-space. Genetic admixture levels were significantly higher in the West lineage and, for both lineages, in shared portions of E-space. This suggests that niche similarity can facilitate gene flow among disjunct groups with moderate-to-good dispersal capabilities, contrasting with the prevailing view of niche conservatism as a diversifying force.

Integrating environmental and genetic data allows researchers to ask how ecology influences evolutionary patterns in the present (Storfer et al. 2006; Kozak et al. 2008) and past (Hugall et al. 2002; Richards et al. 2007; McCormack et al. 2008). With regard to divergence and speciation, this interdisciplinary approach has been strongly influenced by the concept of niche conservatism (Peterson et al. 1999; Wiens and Graham 2005), or the idea that recently derived taxa tend to be similar in their ecological niches. Although there has been conceptual debate about whether niche conservatism better describes a pattern or process (Losos 2008), it has nevertheless been linked to the process of speciation through the idea that lineages that are strongly conserved in their niches are unlikely to cross areas of unsuitable habitat, facilitating reproductive isolation (Wiens 2004) through processes that are undescribed, but would seem to involve genetic drift, mutation-order speciation (Mani and Clarke 1990), or the accumulation of Bateson–Dobzhansky–Muller incompatibilities (Gavrilets 2003). This hypothesis has found some support in poorly dispersing organisms (Kozak and Wiens 2006). Other studies incorporating genetic and environmental data have also suggested speciation via niche divergence (Graham et al. 2004; Rissler and Apodaca 2007). Although the preceding studies have tackled the role of niche divergence or conservatism from a phylogenetic perspective, population genetic studies are needed to examine exactly how niche conservatism or divergence might influence the all-important process of gene flow.

The integration of environmental data with genetics to address speciation questions is currently being driven by conceptual and methodological advances in quantifying and comparing ecological niches among species. GIS data from weather stations and remote-sensing satellites provide a trove of environmental information on the Earth's surface, but because they show strong spatial autocorrelation (Dormann et al. 2007), they pose daunting and very specific challenges for researchers attempting to compare these data between species. For example, strong spatial autocorrelation in GIS data can lead to patterns of niche divergence among allopatric lineages that are likely to be exaggerated and of no greater import than the divergence of random points from the geographic ranges of the two lineages (Godsoe 2010). Several recent studies have promoted the idea of comparing observed niche divergence between lineages to null models of divergence generated from random points from the geographic ranges of the lineages to robustly test for niche divergence and conservatism (for a method using principal components analysis, see McCormack et al. 2010; for a method using niche models, see Warren et al. 2008). Visualization of species’ available background environments has also illuminated several recent theoretical (Soberón and Nakamura 2009) and empirical (Broennimann et al. 2008; Godsoe et al. 2009) treatments of niche divergence in an evolutionary context. The next challenge is to unite these robust tests with explicit predictions of how niche conservatism or divergence influences landscape-level gene flow.

With this study, we seek to unite robust tests of niche divergence with population genetic data to test the prediction that patterns of dispersal and gene flow between two lineages of the nine-banded armadillo (Dasypus novemcinctus) are influenced by their climatic niches as quantified using GIS data. The nine-banded armadillo is distributed in a wide variety of environmental conditions from northern Argentina to the southern United States (McBee and Backer 1982). Genetic study indicates that Mexican populations consist of two highly divergent mtDNA lineages (4.5% divergence) with allopatric distributions in Eastern and Western Mexico (Arteaga 2011). Given that armadillos are derived from South America (Simpson 1980; Eisenberg 1981), the occurrence of different South American haplotypes within each of the Mexican lineages suggests that they originated from two separate founding events, with the Eastern Mexican lineage later providing the ancestral stock for northern population expansion into the United States. A small number of mismatched mtDNA haplotypes between East and West (Fig. 1) and higher levels of gene flow in nuclear markers suggest long-distance, and potentially male-biased dispersal (Arteaga 2011). Regardless of the precise details of how gene flow occurs, judging from the wide distribution of armadillos in North and South America, their recent expansion into the United States (Taulman and Robbins 1996), and the aforementioned genetic results, nine-banded armadillos appear to be fairly good dispersers. Gene flow among the East and West Mexican lineages also suggests that after dispersal to the geographic range of the other lineage, individuals find their new environments reasonably suitable (i.e., there appears to be mating and introgression). For this reason, we made the prediction that genetic admixture among the two lineages might actually be facilitated by niche conservatism. We tested for niche conservatism using a new method that constructs null models of niche divergence on orthogonal environmental axes using background environments available to the two lineages (McCormack et al. 2010). Further, by visualizing genetically admixed individuals in multivariate environmental space (E-space, sensu Soberón and Nakamura 2009), we were able to test a resulting prediction that regions of shared E-space should show the highest incidence of genetic admixture compared to E-space that is unique to a particular lineage.

Figure 1.

Geographical distributions of the West (blue) and East (green) mtDNA lineages, including a region in southern Mexico between the two lineages where mtDNA affiliation is not known. Note the five geographically mismatched mtDNA haplotypes, four of which occur in the West, which suggests dispersal. The Sierra Madre Oriental, a putative vicariant barrier between the lineages is shown in gray.



We collected GIS data on climate using 85 unique D. novemcinctus localities comprising 39 individuals from the West lineage and 72 individuals from the East lineage. Locality information came from GPS devices for wild-caught individuals and from georeferenced museum specimens (Tables S1 and S2). Because our focus was on the two allopatric mtDNA lineages in Mexico with known gene flow and distinct geographic limits, we did not consider the whole geographic range of D. novemcinctus in the United States, Central, and South America. Importantly, genetic results indicate that the two Mexican lineages are evolutionary distinct from one another. Within the two considered lineages, our occurrence points were well-distributed both geographically and in the wide variety of environmental conditions in Mexico where they occur, with the exception of a small gap in southern Mexico where lineage affinity is unknown (Fig. 1).

GIS data included BioClim climate layers that have a resolution of 1 km2, describing aspects of temperature, precipitation, seasonality, as well as potentially biologically limiting extremes of these variables (e.g., Bio6, minimum temperature of coldest month). Using ArcView version 3.2, we first extracted GIS data from 19 BioClim layers (Bio1-Bio19; at armadillo occurrence points. Using JMP 5.01 software (SAS Cary, NJ), we then conducted a correlation test to remove highly correlated variables, which could bias subsequent analyses (Graham et al. 2004; Walker et al. 2009). Specifically, if two variables showed a correlation coefficient higher than 0.75, we considered them highly correlated, and for each pair of correlated variables, we selected the variable that was more temporally inclusive (e.g., preferring mean temperature over mean temperature of driest quarter) or those likely to be most relevant to armadillos (temperature variables over precipitation variables given that armadillos are poor thermoregulators (McNab 1980) and their distribution might be partially limited by environmental temperatures). The remaining six climate variables included Bio2 = mean diurnal range of temperature; Bio5 = minimum temperature of warmest month; Bio6 = minimum temperature of coldest month; Bio14 = precipitation of driest month; Bio15 = precipitation seasonality; and Bio18 = precipitation of warmest quarter.

Geographically explicit predictions of climatic niches can often provide a good starting point for exploring niche differences and locating regions of niche similarity, so we generated environmental niche models (ENMs) for both lineages. ENMs are geographic predictions of lineage distributions based on environmental data from known sampling points (Peterson 2001). We constructed ENMs for each lineage from occurrence points and the six climate layers using MAXENT 3.2.1 (Phillips et al. 2006). We used the default convergence threshold (10−5) and 500 iterations (Pearson et al. 2007), using 25% of localities for model training and 75% for model testing. To assess model performance, we used the area under the receiver operating characteristic curve (AUC, Mertz 1978) as a measure of overall classification accuracy (capacity to discriminate between occupied and unoccupied records). The AUC can vary from 0.5, indicating no discrimination capacity, to 1, indicating perfect discrimination capacity. We produced a map of niche overlap by converting each lineage's niche model to a binary prediction of presence/absence with a prediction threshold of 10% (Pearson et al. 2007). Both binary prediction models were then summed and overlapping niche predictions were retained.

To test for niche conservatism, we employed the multivariate method introduced in McCormack et al. (2010) that compares niche divergence to a null hypothesis of divergence in available background environments on several orthologonal axes of E-space. This method does not actually involve ENMs sensu stricto, but rather uses principal components analysis (PCA) to reduce the raw GIS data to a smaller, uncorrelated set of axes. We chose this method over the method of Warren et al. (2008), which uses ENMs to test for niche divergence and conservatism, because our initial results indicated that ENMs for the different lineages were strongly influenced by differing climatic variables (see Results), which makes comparing ENMs difficult. In contrast, the multivariate method leads to readily interpretable axes, whose interpretation is the same for both lineages. Furthermore, the description of multiple niche axes in the multivariate method allows for the kind of detailed study we wished to undertake, as opposed to the broader, joint estimation of the niche afforded by ENMs.

The general idea behind the McCormack et al. (2010) method is that a pattern of divergence in GIS data could be attributable either to meaningful niche divergence between species (or lineages) or to the fact that GIS data are strongly spatially autocorrelated. Therefore, a strong test of niche divergence or conservatism must compare niche divergence between species to baseline levels of divergence drawn from the background of available habitat contained within each species’ geographic range (see next paragraph for detail). The null hypothesis is rejected when niches are more similar (niche conservatism) or more different (niche divergence) between species than the null model of background divergence. It is very important to note that if the null hypothesis is not rejected, this does not mean that there is no meaningful niche divergence between the species (i.e., failure to reject the null hypothesis does not mean that the null hypothesis is true). Rather, it means that whether divergence between species is meaningful or due to spatial autocorrelation both remain plausible alternatives.

To conduct the multivariate test for niche divergence/conservatism, we extracted raw data from our occurrence points for the two armadillo lineages and, to generate the background predictions, from 1000 random points from within the geographic ranges of each lineage. We drew the random points from within a minimum convex polygon drawn around our occurrence sites (Warren et al 2008; McCormack et al 2010) using ArcView version 3.2 and the Hawth's Tools package. Next, we conducted a PCA on these data, extracting the first three PC (niche) axes for further consideration because they comprised the bulk of the variation and were readily interpretable (see Results). Niche divergence or conservatism was evaluated on each niche axis by comparing the observed difference between the means for each lineage on that axis to the mean difference in their background environments on the same axis. A null distribution of background divergence was created by recalculating the background divergence score over 1000 jackknife replicates with 75% replacement. Significance for rejecting the null was evaluated at the 95% level. All analyses were conducted using Stata version 10 (StataCorp 2003). To provide a rough measure of how spatially autocorrelated an axis is, we assessed the correlation between our three niche axes and longitude/latitude.


To test whether genetically admixed individuals were located in parts of E-space that were more similar between the two armadillo lineages, we first visualized this E-space within the envelope of the background E-space available to each of them by making a bivariate plot using the first two PC axes. The first two axes were used because they explained the most variation and because our initial analyses did not detect niche conservatism on PC1, suggesting that this niche axis might be good for observing patterns of partially, but not wholly overlapping niches.

We genotyped 111 individuals (72 from East and 39 from West) from the 85 unique localities at five microsatellite loci (autosomal: Dnov1, Dnov6, Dnov7, Dnov16, Dnov24; Prodöhl et al. 1996) in the DNA was extracted from tissue using a DNeasy Kit (Qiagen®). Individuals were genotyped in a 10-μl reaction containing 1× of buffer, 0.3 μM of each primer, 0.15 mM of dNTP, 1–2 mM of MgCl2, and 1.5 U of Taq DNA polymerase. The thermal profile for amplification consisted of an initial denaturation at 94°C for 5 min, followed by 35 cycles of 15 sec at 94°C, 30 sec at 56–58°C and 1 sec at 72°C, with a final extension of 7 min at 72°C. All reaction products were run on an ABI PRISM genetic analyzer with a four-capillary system (Applied Biosystems, Foster City, CA) and alleles were scored using GENEMAPPER 4.0. Because loci showed deviations from Hardy–Weinberg equilibrium (HWE) likely due to a Wahlund effect (see Results), we also tested for deviations from HWE and linkage by locus at a smaller geographic scale (eight populations grouped using natural breaks in the sampling distribution) using Genepop On the Web (Raymond and Rousset 1995) and significance was assessed after correction for multiple tests (Rice 1989). Diversity statistics for the two lineages were calculated with ARLEQUIN 3.11 (Excoffier et al. 2005).

We assessed levels of genetic admixture using the Bayesian clustering algorithm implemented in STRUCTURE 2.3.3 (Pritchard et al. 2000), which clusters individuals into groups minimizing HW disequilibrium. Individuals are assigned a probability to one or more clusters if their genotype indicates that they are admixed (Pritchard et al. 2000). Because we were interested in levels of admixture between two geographically isolated lineages with strong structuring in mtDNA, we set the number of clusters to K = 2. We did not attempt to detect further genetic structure within the East and West lineages, which was the goal of another study describing broad-scale phylogeographic patterns in this species (Arteaga 2011). The rationale for setting K = 2 is that, although genetic structure in the nuclear microsatellites does not fall into two clear clusters, there is a statistically significant association between mtDNA haplotypes and microsatellite cluster assignment (Arteaga 2011). We ran 10 iterations of STRUCTURE at K = 2, using the admixture model and correlated frequencies. Runs had a burn-in period of 100,000 followed by 1,000,000 Markov chain Monte Carlo replicates.

To explore how genetic admixture levels were related to areas of overlap between the niches of the two lineages, we calculated an admixture score using our STRUCTURE results. We first averaged the assignment scores for individuals across the 10 runs. A value that represented the level of admixture was then created by taking the absolute value of the difference between assignment of an individual to one cluster and 0.5. Because this resulted in a counter-intuitive scaling where low admixture individuals had high values, we reversed the scale by taking the absolute value of the score minus 0.5. The final values were admixture scores ranging from 0 (very pure) to 0.5 (very admixed). We then plotted the locations of these individuals with their admixture score in two-dimensional E-space using the first two niche axes. We tested the hypothesis that there should be more admixture in regions of E-space that are shared between the two lineages. We conducted ANOVA with partial Student's t-tests on the admixture scores for the two lineages and for the four different groups that are defined by their location in E-space: (1) West individuals in E-space unique to the West lineage, (2) West individuals in E-space shared with the East lineage, (2) East individuals in E-space unique to the East lineage, (4) East individuals in E-space shared with the West lineage (see Fig. 2). Because admixture scores appeared to be nonnormally distributed (see Results), we also assessed differences with a nonparametric Wilcoxon test. The predictions were that the lineage with the most overlapping E-space would be the one with the most genetic admixture and, further, that within a lineage, there would be more genetic admixture in portions of E-space shared with the other lineage compared to individuals in E-space unique to that lineage. The statistic analyses were conducted using JMP 5.01.

Figure 2.

The climatic E-spaces of West (dark blue) and East (dark green) lineages visualized within their respective available background environments (light blue and light green).



Niche models for the two armadillo lineages showed largely disjunct distributions that in general agreed with the known distributions of these lineages (Fig. S1). However, the niche model for the East lineage showed some evidence for prediction into the West lineage's range along the Pacific coast of Mexico. The most relevant environmental variable for predicting the climatic distribution for the West lineage was rainfall seasonality (45.5%) and for the East lineage was minimum temperature in the coldest month (68.2%). The predictive power of occurrence was high for the two models (the AUC of test data were 0.90 and 0.92 for West and East lineages, respectively).

The PCA of raw GIS data indicated three main niche axes that together explained 87.7% of the variation. The first niche axis was associated with temperature and precipitation seasonality and had a high correlation with latitude (Table 1). The second niche axis was associated with temperature extremes, whereas the third niche axis was associated with rain seasonality. Niche axes 2 and 3 had smaller, but still significant correlations with latitude. Correlations with longitude were lower, but still significant in the first and the third axes (Table 1).

Table 1.  Loadings of the environmental variables for each recovered PC axis and Spearman correlations with longitude/latitude.
  1. *Significance level, P = 0.0001.

BIO15: Precipitation seasonality−0.41429  0.37099  0.49605
BIO14: Precipitation of driest month  0.49001−0.32944−0.08548
BIO6: Min temperature of coldest month  0.44895  0.49786  0.07745
BIO5: Max temperature of warmest month  0.23905  0.69663−0.21358
BIO18: Precipitation of warmest quarter  0.26705−0.11417  0.82953
BIO2: Mean diurnal range−0.50818  0.08744−0.08296
LongitudeR2= 0.027*R2= 0.00006R2= 0.036*
  F = 57.59 F = 0.88 F = 76.18
LatitudeR2= 0.458*R2= 0.020*R2= 0.130*
  F = 1632.55 F = 43.235 F = 299.371

Tests of niche divergence and conservatism on these three niche axes showed evidence for niche conservatism on niche axis 2 and 3 (Table 2). Niche axis 1 did not significantly differ from the null expectation of background divergence (although it was closer to the threshold for niche conservatism), indicating that it was not possible to know if the observed divergence was meaningful or resulted from spatial autocorrelation (Table 2).

Table 2.  Tests of niche divergence and conservatism. Observed differences in the climatic niche of the two armadillo lineages on each PC axis compared to the middle 95th percentile of a null distribution of the differences between their environmental backgrounds. Bold values indicate niche conservatism.
 PC1PC 2PC 3
  1. *Significance level, P < 0.05.

Observed difference2.13920.0016*0.5009*
Null distribution2.1246 - 2.23800.3551 - 0.50930.6451 - 0.7146
Percent variance explained42.79%33.05%11.84%


Visualization of climatic E-space of the West and East lineages and their background E-space showed an asymmetric pattern of overlap where the West lineage's niche and background was smaller and almost completely included inside the niche and background of the East lineage, which showed a much larger unique area (Fig. 2). This pattern was largely driven by niche axis 1. Although our earlier tests did not conclude that this pattern on niche axis 1 was consistent with either niche divergence or conservatism (the null hypothesis could not be rejected), it nonetheless presented an opportunity to test the relationship between niche similarity and genetic admixture because some parts of this E-space were overlapping between the two lineages.

When analyzed at the lineage scale, most loci fell outside HWE (Tables S3 and S4). However, when analyzed at a smaller geographic scale, there were few deviations and no consistent patterns across populations, suggesting the lineage-level HWE results were caused by a Wahlund effect. No statistically significant patterns of linkage disequilibrium were observed. Locus Dnov16 showed heterozygote deficiency in several populations, but following recent recommendations (Selkoe and Toonen 2006), we retained this locus in our analysis. Assignment probabilities were highly correlated in STRUCTURE runs with and without this locus (R = 0.9). For each lineage, allelic number by locus was higher and observed heterozygosity was similar to previous results from US populations using the same loci (Loughry et al. 2009; Table S4).

Under the assumption of K = 2, 61 of 111 individuals were assigned with probability higher than 0.9 to one of the two clusters, whereas the rest of the individuals were more admixed. Translated to our admixture scores, 61 individuals were “pure,” having admixture scores lower than 0.1, whereas 50 individuals were highly admixed. The ranges of the admixture scores for the two lineages were similar, but the frequency distributions were very different (Fig. S2). The data were obviously not normally distributed and attempts to transform them to normality were unsuccessful, so all further comparisons used both t-tests and nonparametric Wilcoxon tests. The genetic admixture level of the West lineage as a whole was significantly higher than the East lineage as a whole (Table 3).

Table 3.  Tests of differences in admixture scores for individuals in different regions of environmental space (see Fig. 2). Wilcoxon test is recorded in the last two columns. Bold values indicate significant results.
Level iLevel jMeans i − jDifferencet-test t − ratioProb < tWilcoxon Chi-squareProb < t
  1. West_unique = West individuals in E-space unique to the West lineage; West_mixed = West individuals in E-space shared with the East lineage; East_unique = East individuals in E-space unique to the East lineage; East_mixed = East individuals in E-space shared with the West lineage.

EastWest0.128-0.200−0.071−2.610.01  9.4690.002
East_uniqueWest_mixed0.097-0.205−0.108  3.840.0002 15.430.0001
East_uniqueEast_mixed0.097-0.235−0.138−3.6940.0003  6.3450.011
West_mixedEast_mixed0.205-0.235−0.029−0.7540.452  0.0230.878
East_uniqueWest_unique0.097-0.199−0.081  1.04780.297  3.5150.06
West_uniqueEast_mixed0.199-0.235−0.056−0.6770.499  0.1580.69

When visualized in E-space (Fig. 3), 56 individuals from the East lineage occurred inside unique E-space for the East lineage, whereas 16 East individuals occurred in areas of E-space that were shared between the West and East lineage. Given that the E-space of the West lineage was almost completely subsumed within the E-space of the East, only three individuals from the West lineage occur inside its own unique E-space, whereas 36 West individuals occur inside E-space shared with the East lineage (Fig. 3).

Figure 3.

Visualization of the genetic admixture of individuals in regions of unique and shared E-space between the East and West lineages. Where multiple individuals fall into the same location, the individuals have been separated and expanded in the outlined boxes. Note that the blue color denotes genetically pure individuals and does not necessarily signify a relationship to the West lineage shown in blue in Figure 2.

Overall, there was less admixture in portions of E-space that were unique to the two lineages compared to those niche regions that were shared. Individuals from the East lineage that occur in E-space unique to the East lineage showed significantly lower admixture levels than both individuals from the West and East that are found in regions of overlapping E-space (Table 3). Although our statistical power to detect differences using individuals from the West in unique E-space was low due to low sample size (n = 3), the mean admixture score for these three individuals was larger than that of either lineage in the region of niche overlap (Table 3). There was not a significant difference between admixture scores for the individuals from the two lineages occurring in the overlapping regions of E-space (Table 3).


The integration of ecology with historical biogeography will require both the union of niche data with divergence processes at deep time scales (i.e., cladogenesis) in addition to contemporary studies of how ecology affects dispersal and gene flow at the population level (Wiens and Donoghue 2004). GIS data are ideal for bridging this gap because they are so versatile (Swenson 2008): from the perspective of cladogenesis, they can be combined with phylogenies to explore how niche evolution (or lack of niche evolution) correlates with speciation (Evans et al. 2009; Kozak and Wiens 2006; Graham et al. 2004), but they are also amenable to recent or contemporary studies of the ecological factors influencing dispersal and gene flow (e.g., landscape genetics; Manel et al. 2003). By considering niche evolution at both the phylogenetic level (between two mtDNA lineages) and population-genetic level (individual-based estimates of genetic admixture), our study provides a crucial link between genetic divergence and climatic niche evolution at the interface of recent and older time scales. The most important conclusion of our study is that climatic niche similarity between two armadillo lineages seems to have facilitated gene flow in regions of E-space that are shared. The finding that similarity in niches is associated with greater gene flow contrasts with the depiction of niche conservatism as a diversifying force that leads to reproductive isolation by reducing gene flow between populations or species separated by regions of inhospitable habitat (Wiens 2004). Although the latter may be true for species that disperse poorly or that are reluctant for behavioral reasons to cross regions of less-hospitable habitat (e.g., Kozak and Wiens 2006), niche similarity (or conservatism) for other more vagile species may equate to enhanced opportunities for contact and genetic admixture, thereby reducing the likelihood of speciation.


We originally predicted that the two armadillo lineages would show niche conservatism because of evidence from mtDNA and nuclear genes that individuals disperse between the geographic ranges of the two lineages and apparently survive to mate (as documented by mtDNA mismatch and coalescent-based estimates of migration rates described in Arteaga 2011), long-distance feats that would be facilitated if their niches were similar. Results from a test that compared the amount of climatic divergence between armadillo lineages to the null expectation of background climatic divergence validated this prediction by showing niche conservatism on two axes of E-space related to temperature extremes (PC2) and rain seasonality (especially rain in the warmest month, PC3). Armadillos are poor thermoregulators (McNab 1980), so it makes sense that their distribution would be partially limited by temperature extremes. Rainfall seasonality could be related to change in insect populations, which are an important food resource for armadillos (Humphrey 1974; Taulman and Robbins 1996). Although we could not reject the null hypothesis in favor of niche conservatism on the first niche axis, which was related both to temperature and rain seasonality, the amount of observed divergence fell toward the conserved side of the null distribution. It is thus unlikely that the observed difference on niche axis 1 translates to meaningful climatic niche divergence, and spatial autocorrelation—which is expected to be acute with climatic data (Soberón 2007)—cannot be ruled out as the driving force behind this pattern.

The results of niche conservatism from tests using null models stand in contrast to the map-based projections of the climatic niche models themselves (Fig. S1). Despite some projection of the East lineage into the range of the West lineage, based simply on the disjunct patterns, one might be tempted to draw the conclusion that these armadillo lineages are strongly divergent in their climatic niches. This contrast further serves to highlight the problems with drawing conclusions about niche divergence between species from ENM projections on a map or in multivariate E-space in the absence of controls for spatial autocorrelation or, in the case of multivariate E-space, without simultaneous visualization of available background environments. The question is not whether the two lineages live in places that differ in some environmental attributes (one can know that simply by looking at a map), but rather if these differences are large enough to reject a model created by drawing background points that, because they are random, are dissociated from any specific biological relevance to armadillos. Recent conceptual work (Soberón and Nakamura 2009; Soberón 2007; Soberón and Peterson 2005), simulation studies (Godsoe 2010), and methodological advances (McCormack et al. 2010; Warren et al. 2008) are beginning to address the problem of spatial autocorrelation of GIS data for comparing species’ niches, but further advances and refinements of the methods will undoubtedly be necessary before speciation studies can reap the full benefits afforded by GIS data.


Given our prediction that niche conservatism would facilitate dispersal and gene flow between armadillo lineages, a subsequent hypothesis is that greater genetic admixture should be observed in regions of E-space that are especially similar between the two lineages. In thinking about these patterns it is important to note that disjunct and overlapping regions of multivariate E-space (e.g., Figs. 2 and 3) do not directly correspond to geographic space, which is entirely alloparapatric between the two lineages (we discuss the geographic position of admixed individuals in greater detail below). Our hypothesis was supported by higher estimates of genetic admixture from nuclear microsatellites in regions of overlapping E-space (Fig. 3 and Table 3). Additionally, we found an intriguing pattern of niche overlap that could explain the genetic admixture patterns documented in this study and a previous study. The fact that the E-space of the West lineage was a subset of the E-space of the East lineage (Fig. 2) suggests that individuals from the East lineage might find dispersal to the West's geographic range especially tolerable. This pattern of asymmetrical overlap might explain three genetic patterns in this system, the first documented in this study and the last two from a previous study: (1) why the West lineage as a whole showed more genetic admixture (because it would receive more individuals from the East than it sent to the East), (2) why, of the five individuals found with mtDNA haplotypes that did not match their geographic location, four of these indicated dispersal from East to West, and (3) why coalescent-based estimates of migration rates from microsatellites suggested higher gene flow from East to West (although the 95% confidence intervals were overlapping; Arteaga 2011). These results show how more detailed information on the link between niche characteristics and dispersal and gene flow can be obtained when broad tests of niche conservatism/divergence between lineages are coupled with individual-based genetic information.


How the asymmetrical pattern of overlap in multivariate E-space between the two armadillo lineages has influenced their dispersal and evolutionary history is further elucidated by visualizing the geographic locations of individuals from shared and unique niche space (Fig. 4A) and their admixture levels (Fig. 4B). East individuals in niche space shared with the West are numerous near a well-known lowland corridor linking the Mexican coasts, the Isthmus of Tehuantepec. This suggests that niches for the two lineages, while broadly conserved throughout their distributions, are especially similar in this geographic location. However, geographic proximity, while undoubtedly important, does not seem to be the sole driver of admixture patterns because in this region gene flow appears largely unidirectional from East to West (or more accurately near the Isthmus, north to south) into regions of niche overlap (Fig. 4B). Meanwhile, regions near the Isthmus that comprise part of E-space unique to the East (i.e., north of the Isthmus) show low levels of admixture (Fig. 4A, B; also see Table 3 for quantitative validation of these patterns). Additionally, within the geographic range of the West lineage (whose E-space is almost entirely subsumed within the E-space of the East lineage), highly admixed individuals (red dots in Fig. 4B) do not occur exclusively near the contact zone with the East near the Isthmus, but rather occur mostly along the West coast, almost exclusively in geographic regions where both lineages are predicted according to their ENMs (Fig. 4B). In fact, of the 23 individuals from both East and West that show the most genetic admixture (red dots in Fig. 4B), 16 occur in the relatively narrow geographic zone where ENMs for both lineages overlap.

Figure 4.

(A) Visualization of the geographic locations of individuals from the four groups created by considering their positions in climatic E-space (see Figure 3). East individuals in shared E-space occur primarily south of the Isthmus of Tehuantepec, whereas most West individuals occur in E-space shared with the East lineage. The biogeograhical region known as the Mexican Gulf Province is shown in gray. (B) Genetic admixture of individuals visualized in geographic space. Regions where ENMs for East and West overlap are shown in black. A strong role for niche similarity in promoting gene flow, as opposed to geographic distance alone, is suggested by the large proportion of admixed East individuals to the south of the Isthmus of Tehuantepec, where East and West niches overlap, but not to the north of the Isthmus, where niches do not overlap. Additionally, highly admixed West individuals occur primarily in regions of niche overlap that are not necessarily in close proximity to geographical areas of contact. This results in a pattern of gene flow from East to West, toward regions of shared E-space, as validated quantitatively by the t-test results comparing admixture levels between East and West (Table 3).

Niche similarities seem to play an outsized role in genetic admixture among armadillo lineages, but the geography of dispersal should not be neglected. Our results suggest that admixture between armadillo lineages has likely not occurred by long-distance dispersal through the high mountain chains of Central Mexico, but rather via a habitat corridor through the Isthmus of Tehuantepec farther to the south. It is also possible that the East lineage originally occupied regions both north and south of the Isthmus after their original colonization, with the southern population becoming more admixed by influx of West individuals unimpeded by a physical barrier into suitable habitat. Meanwhile, the part of the niche most unique to the East lineage, which contains a high proportion of the genetically pure East individuals, lies in the coastal biogeographical unit known as the Mexican Gulf Province (Fig. 4A). This region has been very important in the biogeographical history of angiosperms (Méndez-Larios and Villaseñor 1995), insects (Hamilton 1994; Llorente-Bousquets et al. 1997), fish (Lydeard et al. 1995), amphibians (Arriaga et al. 1997), reptiles (Campbell and Lamar 1989), birds (Arriaga et al. 1997), and mammals (Müller 1973). Connectivity between these largely genetically pure East populations and those more admixed East populations on the other side of the Isthmus will require more detailed population-level study.

In conclusion, we have presented a novel combination of lineage-level tests of niche conservatism with more detailed individual-based estimates of the role of niche similarity in genetic admixture that could be applied to any study seeking to evaluate the role of niche evolution in diversification. The surprising conclusion that niche conservatism is facilitating gene flow is not so surprising when the dispersal capabilities of armadillos are considered, in addition to the fact that some of the regions of E-space that are especially similar occur close to a major lowland corridor linking the Mexican coasts. None of these conclusions would have been possible without a conceptual framework and methodology that considers the duality of the niche (sensu Soberón and Nakamura 2009) as an abstract environmental space in addition to a geographically explicit realization of this environmental space on a map.

Associate Editor: L. Kubatko


We wish to thank all the persons and institutions that provided biological samples for this study; they are listed in Table S1. We are grateful to D. Piñero, O. Gaona, E. Aguirre, L. Espinoza, J. Gasca, and A. Correa for their help with laboratory analyses, technical support, and advice in the data analysis. This work is part of MCA's Ph.D. dissertation in biological sciences at the Universidad Nacional Autónoma de México (UNAM). MCA is grateful to the graduate program Doctorado en Ciencias Biológicas and to UNAM for the scholarship granted for doctoral studies.