AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 151:506–517 (2013)
Cranial Morphological Variation Among Contemporary Mexicans: Regional Trends, Ancestral Affinities, and Genetic Comparisons Cris E. Hughes,1* Meredith L. Tise,2 Lindsay H. Trammell,3 and Bruce E. Anderson4 1
Department of Anthropology, University of Illinois at Urbana-Champaign, Urbana, IL 61801 Department of Anthropology, University of South Florida, Tampa, FL 33620 3 Saint Louis Office of the Medical Examiner, Saint Louis, MO 63134 4 Pima County Office of the Medical Examiner, Tucson, AZ 85714 2
KEY WORDS
craniometrics; Mexico; biogeographic variation
ABSTRACT Genetic research has documented geographical variation within Mexico that corresponds to trends in ancestry admixture from postcolonial times on. The purpose of this study is to determine whether craniometric variation among contemporary Mexicans is comparable to that reported in genetic studies. Standard osteometric measurements were taken on 82 male crania derived from forensic cases, with geographic origins of the specimens spanning over two-thirds of Mexico’s states. To study similarities in regional clustering patterns with genetic data, k-means clustering analyses were performed, followed by chi-square tests of association between cluster assignments and geographic region of origin. Normal mixtures analyses were performed, centered on three “ancestral” sample proxies to estimate classification probability to each ancestry. The results
demonstrate that the cranial morphological sample data cluster similarly to the regional groupings inferred from the genetic data. Additionally, the results indicate a gradient trend in population structure for contemporary Mexicans, with the proportion of Amerindian ancestry increasing from North to South while, conversely, European ancestry proportion estimates increase from South to North. Furthermore, the probabilities for classification of African ancestry remained low across the regions, again reflecting the results for the genetic data. Cranial morphological variation is well aligned with the genetic data for describing broad trends among Mexican populations, as well as yielding comparable estimates of general ancestry affiliations that reflect Mexico’s history of Spanish contact and colonialism. Am J Phys Anthropol 151:506–517, 2013. VC 2013 Wiley Periodicals, Inc.
Population structure in present day Mexico has been aptly described as a tri-hybrid admixture event that began with the arrival of Spanish and African individuals at the end of the 15th century and continues today as migrations within and beyond Mexico’s borders take place (Kirkwood, 2000; Garcıa, 2006). Studies examining biological variation within Mexico use a variety of data, including dermatoglyphics (Cummins, 1930), cranial morphology and nonmetric traits (Ross et al., 2002a; Slice and Ross, 2004; Martinez-Abadıas et al., 2006; Hurst, 2012), and genetic systems of mitochondrial, Ychromosome, and autosomal DNA (Gorodezky et al., 2001; Falush et al., 2003; Bonilla et al., 2005; RangelVillalobos et al., 2008; Martinez-Cortes et al., 2010). Recently, Rubi-Castellanos et al. (2009) studied regional genetic variation among contemporary Mexican mestizos using 13 combined DNA index system-short tandem repeats loci (CODIS-STR). This study was unique because it combined genetic data from contemporary individuals from all across Mexico (included in previous studies) to regionally assess the interaction of two key events in the country’s history: 1) the influence of colonial contact on the genetic variation among Mexico’s indigenous populations and 2) the significance of precontact expansion events as reflected in the genetic variation. The study’s results exhibited two regional clustering scenarios. Based on AMOVA results, RubiCastellanos et al. (2009) inferred that the most
appropriate differentiation for their Mexican sample yielded three clusters, corresponding geographically to North/West, Central, and Southeast Mexico. As a secondary analysis, MDS plots of the FST values indicated a second organization for the CODIS-STR loci data into two regional clusters: North/West and Central/Southeast, broadly corresponding with the northernmost boundary of the Mesoamerican expansion (Fig. 1a). Furthermore, the authors demonstrated a gradient approach to admixture in Mexico, with the proportion of Amerindian ancestry increasing from North to South, and the proportion of European ancestry increasing from South to North. Under models assuming neutral expectations of evolution, craniometric data mirrors genetic patterns of variation, estimating that variation among populations comprises a disproportionately low amount of the total
Ó 2013 WILEY PERIODICALS, INC.
*Correspondence to: Cris E. Hughes, Department of Anthropology, University of Illinois at Urbana-Champaign, 109 Davenport Hall, 607 S. Matthews Ave, Urbana, IL 61801, USA. E-mail:
[email protected] Received 6 September 2012; accepted 8 April 2013 DOI: 10.1002/ajpa.22288 Published online 11 June 2013 in Wiley Online Library (wileyonlinelibrary.com).
CRANIAL MORPHOLOGICAL VARIATION IN MEXICANS
507
Fig. 1. Maps illustrating geographical origin of samples in Mexico used in (a) Rubi-Castellanos et al. (2009) paper and (b) this study.
biological diversity (Relethford, 1994; Relethford and Harpending, 1994; von Cramon-Taubadel, 2009). The limited role of natural selection in cranial morphological variation suggests that data like craniometrics can be used to study population structure. Roseman (2012) articulated caution when using cranial morphology to estimate biological variation, suggesting sample sizes typically required for assessing concordance between
genetic and craniometric data are much larger than usually collected or realistically available. Because cranial morphological variation has a heritable component and can be used to generally estimate population structure (Relethford, 2002; Harvati and Weaver, 2006), cranial variation is a natural extension for the genetic research completed by Rubi-Castellanos et al. (2009) and others focusing on genetic variation in Mexico. American Journal of Physical Anthropology
508
C.E. HUGHES ET AL.
One goal of this study is to determine whether similar trends of ancestry proportions and regional clustering are exhibited when utilizing craniometric data. The results of this study will determine whether CODIS-STR loci and craniometric data among Mexicans demonstrate similar geographic patterns of variation, as well as estimates of regional trends of admixture. Several comparisons of genetic and craniometric estimates of biological variation have been completed for Mexican populations. Herrera and Spradley (2012) demonstrated that biological distance matrices derived from craniometric and genetic data (mtDNA haplogroup frequencies and YSTRs) among three Mexican populations were statistically significantly correlated. Slice and Ross (2004) reported that contemporary Mexicans cluster most closely with Amerindian ancestry, exhibiting minimal association with either European or African ancestries. Martinez-Abadıas et al. (2006) offer one of the most comprehensive cranial morphology admixture studies in central Mexico, testing a bihybrid (Spanish and Amerindian) model of admixture to explain phenotypic variation present in two postcontact samples. This study found that Mexican cranial morphology reflects the temporal component of gene flow related to historical events of European contact and colonialism. A second goal of this research is to provide an additional layer to these works on admixture estimates of Mexico’s population from cranial variation—a comparison of the regional nuances of admixture across Mexico. To examine geographic notions of admixture in Mexico, this study includes a sample with extensive geographic coverage. In addition, this study incorporates a trihybrid approach to admixture, the results of which are compared to the estimates of population variation and ancestral affinities from genetic data. The final component of this study is to address the applied perspective of the results. By demonstrating that there is a relationship between proportions of genetic ancestries and regions of Mexico, Rubi-Castellanos et al. (2009) provided a foundation for exploring these trends in the forensic context. Beyond the implications of historical gene flow and admixture events, patterns of regional variation can be used to increase the chances of identifications of unidentified skeletal remains by linking them to a geographic area. If similar regional trends can also be found in cranial variation, this would be particularly useful within the context of the U.S.–Mexico border crisis, where hundreds of unidentified border crossers presumed to be from Mexico are being recovered and analyzed annually with an end goal of identification. Because genetic analysis can be costly and is not always performed, an alternative, cost-effective approach is desirable. If similar regional patterns can be inferred using cranial morphological variation, this line of research could serve as the foundation for the development of methods predicting region of origin in future studies.
MATERIALS AND METHODS Materials The primary sample is comprised of 82 identified males from forensic cases analyzed at the Pima County Office of the Medical Examiner (PCOME) in Tucson, Arizona from 2004 to 2012. The following demographic data were collected from PCOME’s cases or case records for American Journal of Physical Anthropology
all individuals: sex, country of origin, state of origin, and city of origin (if known). The place of birth is Mexico for all individuals. This sample was selected for this study for several reasons. First, the sample is contemporary, which is the most appropriate for comparisons to genetic studies that sampled living Mexicans. Additionally, the sample is comprised of only identified remains, making it possible to link each individual to a particular region of origin. Furthermore, because the sample is derived from those who died while crossing the U.S.–Mexico border, it offers a sample that is representative of the border crosser demographic within the forensic context, making it an appropriate sample on which to determine the utility of the regional variation (if present) for future casework. Finally, because migration is not a localized event within Mexico, the PCOME sample provides a geographically broad representation of Mexican cranial variation, which is necessary for this study. It is often the case that skeletal material available for research represents singular communities from a microregion, but sampling in this manner can potentially simulate greater interpopulation variation than is actually present. Approximately 50% of Mexican immigrants in the United States come from only eight of the 31 Mexican states (Terrazas, 2010) but the PCOME sample includes individuals from 22 states. Geographically comprehensive sampling becomes even more important to this study, as the regional sample sizes themselves are rather small. One caveat of the study sample is that to address trends within the general population of Mexico, we make the assumption that the migrant population used here is representative of both migrant and nonmigrant cranial variation of Mexicans. Migrants do tend to be associated with a lower socioeconomic status and rural regions in Mexico, which is largely comprised of indigenous (and some mestizo) populations (Fox, 2006). Whether the distribution of cranial variation of the migrant-based PCOME sample is representative of the distribution of cranial variation for all Mexicans is unknown (see Discussion below regarding estimates of migrant ethnicities). Standard craniometric data were collected from these cases (based on Howells, 1973) using calipers by forensic anthropologists employed at PCOME during this 8-year period. For cases with damage preventing accurate measurement, the affected measurements were excluded. The majority of the data were collected by author BEA, while a portion of the 2011 cases were collected by authors CEH, LHT, and MLT. A small percentage of the data were derived from casework generated by other forensic anthropologists who temporarily consulted at PCOME. To study similarity in regional clustering patterns to the CODIS-STR loci results, the same regional assignments (North/West, Central, and Southeast) were given to Mexican states based on the geographic clusters established by Rubi-Castellanos et al. (2009). Table 1 provides a comparison of the Mexican states for which STR and craniometric data are available, along with their regional assignments. For those states where craniometric data are present but genetic data were not, the authors assigned the state to the region that best conformed to the Rubi-Castellanos et al. (2009) regional assignments. Figure 1a,b compare the genetic and cranial sampling distributions and the regional divisions of the states.
CRANIAL MORPHOLOGICAL VARIATION IN MEXICANS TABLE 1. Mexican states for which STR and/or craniometric data is available, along with their regional assignments
State/Entity Aguascalientes Distrito Federal Guanajuato Hidalgo Mexico Michoac an Nayarit Puebla San Luis Potosı Tlaxcala Veracruz Chihuahua Durango Jalisco Morelos Nuevo Leon Queretaro Sinaloa Sonora Campeche Chiapas Guerrero Oaxaca Tabasco Yucat an
STR data available? N N N Y N N N Y N N Y Y N Y Y Y Y N N Y N N N N Y n 5 2,389
Craniometric data available? Y Y Y Y Y Y Y Y Y Y Y Y Y Y
(n 5 1) (n 5 1) (n 5 4) (n 5 3) (n 5 3) (n 5 2) (n 5 3) (n 5 3) (n 5 1) (n 5 3) (n 5 6) (n 5 1) (n 5 3) (n 5 2) N Y (n 5 1) Y (n 5 2) Y (n 5 5) Y (n 5 10) N Y (n 5 12) Y (n 5 7) Y (n 5 6) Y (n 5 3) N n 5 82
3-Region Organizationa Central Central Central Central Central Central Central Central Central Central Central North/West North/West North/West North/West North/West North/West North/West North/West Southeast Southeast Southeast Southeast Southeast Southeast n(Central) 5 30 n(North/West) 5 24 n (Southeast) 5 28
The sample sizes of each sample and subsample are provided. If data present from the state, a “Y” is listed, if no data is present an “N” is listed. a Based on Rubi-Castellanos et al. (2009).
METHODS Regional variation R For the following statistical analyses, JMPV 7.0.1 C 13.00.05 (SAS Institute, 2002) and/or SYSTATV (SYSTAT Software, 2009) were used. Two analysis of variance (ANOVA) tests were performed to determine whether there were statistically significant differences among the craniometric means for a) the three-region grouping of North/West, Central, and Southeast and b) the Mesoamerican two-region grouping of North/West and Central/Southeast. Univariate t-tests among North/ West, Central, and Southeast samples (pooled covariance matrices and Bonferroni corrections) suggested that several mean measures of the cranial vault (GOL, XCB, FRC, and PAC) are greater for the North/West and/or Central samples versus the Southeast sample. To account for effects of size masking the shape variation, shape variables were calculated and used to generate all further analyses. Shape variables were calculated using the Darroch and Mosimann (1985) approach—for a given individual or case, the geometric mean is calculated for all measurements, and then each measurement is divided by the calculated geometric mean. To determine whether craniometric variation corresponds to geographic regions of Mexico, a k-means clustering analysis (MacQueen, 1967) was performed on the craniometric data excluding information of regional assignment for k 5 2 and k 5 3 clusters using Euclidean distances (50 iterations). K-means clustering is a
509
straightforward approach for maximizing the separation of data objects (here, individuals) into a predefined number of clusters. The approach selected here excludes the initial random selection of k “seed” cases, but instead begins with a single cluster in which all cases are included. The second cluster’s seed is the case farthest from the center of the first cluster. Each case is then reassigned to either cluster based on Euclidean distances and the procedure is repeated for the third cluster (when k 5 3), with the objective being to minimize the within-groups sum of squares while maximizing the between-groups sum of squares. Case assignment and cluster characterizations are produced using the expectation-maximization algorithm (named by Dempster et al., 1977), which first assigns each case’s probability of membership to each cluster, based on the proximity to each cluster’s mean. Once the case is assigned to a cluster, the cluster’s centroid is re-estimated, incorporating the newly clustered case. The craniometric variables used for the k-means clustering analyses included the maximum number of variables (excluding those with intertrait correlations of 0.40 or greater) that still allowed for adequate sample sizes. Therefore, 10 variables were included in the analyses (GOL, ZYB, BPL, MAB, AUB, NLH, NLB, OBH, DKB, and OCC were used). Once the cluster assignments were obtained for the sample, chi-square tests were performed to test for associations between the blind clustering assignments of each individual with their regional assignments (North/West, Central, or Southeast). Three grouping scenarios were used for the chi-square test. As above, the three-region and Mesoamerican groupings were used, and additionally a two-region grouping that combines the Central and Southeast samples while the North/West sample remains separate. Lastly, a canonical variates plot was produced to visualize the multivariate partitioning among the North/ West, Central, and Southeast samples. The canonical plot was generated from a reduced set of the craniometric variables that showed within group intertrait correlations of 0.40 or less, using the forward stepwise procedure. Because discriminant analysis is sensitive to the ratio of sample size to number of variables, only six variables were selected for the analysis (MAB, NLH, NLB, OBH, DKB, and OCC). Regularized discriminant analyses (lambda 5 0.5, gamma 5 0.5) were performed to generate the canonical variates plot. This type of discriminant analysis can be a compromised approach between the quadratic and linear discriminant analyses, used here to accommodate the number of variables to small sample size ratio (Friedman, 1989). The parameters used in governing the regularized discriminant analysis include lambda and gamma. The lambda operates as a function of the covariance matrices and is defined as any number between zero and one, with a decreasing value corresponding more closely with the quadratic discriminant assumption of different covariance matrices. Gamma values also range from zero to one, and act on the covariances of the nondiagonal elements, so that selecting a number closer to one weights a diagonal covariance matrix.
Regional ancestry trends Genetic admixture trends have been documented as regionally variable in Mexico, with the proportion of Amerindian ancestry increasing from North to South American Journal of Physical Anthropology
510
C.E. HUGHES ET AL. TABLE 2. Samples used as ancestry proxies, the data source, geographic affiliation, and sample size
Sample
Geographic origin
Source
Ancestry proxy
Sample size
Maya Dogon Norse Berg Zalavar Spanish
Guatemala Mali Norway Austria Hungary 16th century Spain
Author CEH Howells (1996) Howells (1996) Howells (1996) Howells (1996) Ross (2002b)
Amerindian African European European European European
180 47 55 56 53 55
while, conversely, the amount of European ancestry increases from South to North (Rubi-Castellanos et al., 2009). To visualize the relationship of cranial variation among the North/West, Central, and Southeast regions in relation to estimated ancestry proportions, a canonical variates plot was created using the same reduced set of craniometric variables and parameters as above, with the addition of “ancestral” reference samples. Six comparative samples were included in the canonical variates analysis to broadly represent the Amerindian, African, and European ancestral samples. Table 2 provides the information on these samples and their ancestral proxy labels for the proceeding analyses of this study. Howells’ (1996) craniometric data are used to represent European and African ancestral samples. Because Howells’ data does not include a European sample from Spain (the origin of Mexico’s colonizers), a 16th century Spanish sample was also included (Ross et al., 2002b; Ross et al., 2011). Amerindian ancestry is represented by a sample of contemporary indigenous Maya craniometric data from Guatemala collected by author CEH. For all ancestral samples, only male craniometric data were used in the analyses and were converted to shape variables using the Darroch and Mosimann (1985) calculation. While Howells’ samples are the most accessible and documented data available to use as proxies for ancestral samples, the authors acknowledge that they (as well as the Maya and Spanish samples) are still only proxies, and not intended to be conflated as representing parental populations or as parallels to the genetic reference populations. The Dogon sample was chosen to represent the African ancestry sample because of the West African location of the Dogon people, and thus is the most comparable option from Howells (1996) for the West African slave trade’s source population. Because the samples selected are not the actual ancestral lineages of Europeans and Africans that came to Mexico, individualized analyses will not reach the maximum potential for ancestry affinity estimations. The canonical variates plot only provides a visual representation of three Mexican regional samples in relation to the ancestral samples, and the statistical approach of this analysis considers all six samples, Mexican and ancestral, as discrete groups. An additional statistical test was necessary to estimate ancestral admixture among the Mexican regional samples. To mimic the CODIS-STR admixture estimates generated by the program LEADMIX (Wang, 2003), a normal mixtures analysis was performed to estimate admixture R 7.0.1 (SAS using craniometric shape variables in JMPV Institute, 2007). Normal mixtures analysis assumes that a finite number of clusters (k) of unique multivariate Gaussian distributions exist within the complete dataset (Wolfe, 1970). Normal mixtures analysis is similar to the k-means clustering approach to an extent, but utilizes covariance structure information and allows for cluster American Journal of Physical Anthropology
overlap, as it assumes that all sampled individuals are a mixture of varying proportions of k groups that sum to 1. Mixture-based statistics have been recently used in anthropological studies observing population structure and/or admixture within the context of race (Konigsberg et al., 2009). Most notably, Algee-Hewitt (2011) used a finite mixture model approach to test for worldwide trends of structure using craniometric data of several thousand individuals, as well as to compare estimations of biogeographic ancestry with biosocial racial identities. The robust results of Algee-Hewitt’s (2011) work will undoubtedly serve as a catalyst for implementing finite mixture methods in continued anthropological work on race, ancestry, and population structure in biological anthropology. For this study, three cluster centers were predefined using the European, African, and Amerindian samples to estimate mixture proportions for the Mexican regional samples. JMP 7.0.1 provides a “centers” option in the normal mixtures platform that allows the user to treat sample data as cluster identifiers, fixing the three clusters’ initial means and standard deviations as those observed for each sample, similar to a discriminant analysis. The statistical software uses a BFGS Quasi-Newton algorithm, the estimates of which greatly depend on the initially selected seeds. Because the purpose here is to look at classification of cases into clusters based on predefined and unchanging seeds (the African, Amerindian, and European samples), the estimates of the algorithm will be more stable. Several variations of the normal mixtures analysis were run to assess parameter influence on the cluster probabilities. The initial test included the maximum number of variables (GOL, XCB, BBH, ZYB, NLH, NLB, OBH, AUB, PAC, and OCC) whose correlations were less than 0.40, which keeps the covariance matrices for each cluster close to zero. Each subsequent test removed one or more variables from the mixture analysis until only six of the original 10 variables were included. Selection and order of variables for model removal were informed from a backward stepwise variable selection in a discriminant analysis. The mixtures analyses were performed using a trihybrid and bihybrid (excluding the African sample) approach. For all mixture runs, covariance matrices were unconstrained. However, to assess how constrained covariance matrices would alter the clustering dynamics, a series of mixture analyses were completed using a diagonal matrix, which proved to have limited influence on the cluster proportions. Using the cluster assignment probabilities generated for the PCOME sample, the average probability of African, Amerindian, and European cluster memberships were calculated for the three regions, North/West, Central, and Southeast. Previous work (Martinez-Abadıas et al., 2006) found that lower and upper facial morphology of admixed
511
CRANIAL MORPHOLOGICAL VARIATION IN MEXICANS TABLE 3. ANOVA tests for significant craniometric variation among the two organizations of the Mexico regions 3-Region ANOVA Craniometric variable and abbreviation Maximum cranial length GOL Maximum cranial breadth XCB Bizygomatic breadth ZYB Basion-bregma height BBH Basion-nasion length BNL Basion-prosthion length BPL Palate breadth MAB Palate width MAL Biauricular breadth AUB Minimum frontal breadth WFB Nasal height NLH Nasal breadth NLB Orbit height OBB Orbit breadth OBH Bi-orbital breadth EKB Interorbital breadth DKB Frontal chord FRC Parietal chord PAC Occipital chord OCC
Mesoamerican ANOVA
F
P-value
Adjusted R2
F
P-value
Adjusted R2
9.51 2.49 0.12 1.63 2.89 2.04 2.33 0.23 1.61 1.84 1.56 1.74 1.22 1.12 0.19 1.01 7.60 4.37 0.25
chi-square
Pearson chi-square
Probability > chi-square
17.283 2.938
0.002* 0.230
15.450 0.012
