Do dental nonmetric traits actually work as proxies for neutral genomic data? Some answers from continental- and global-level analyses proxies for neutral genomic data?

Objectives: Crown and root traits, like those in the Arizona State University Dental Anthropology System (ASUDAS), are seemingly useful as genetic proxies. However, recent studies report mixed results concerning their heritability, and ability to assess variation to the level of genomic data. The aim is to test further if such traits can approximate genetic relatedness, among continental and global samples. Materials and Methods: First, for 12 African populations, Mantel correlations were calculated between mean measure of divergence (MMD) distances from up to 36 ASUDAS traits, and F ST distances from >350,000 single nucleotide polymorphisms (SNPs) among matched dental and genetic samples. Second, among 32 global samples, MMD and F ST distances were again compared. Correlations were also calculated between them and inter-sample geographic distances to further evaluate correspondence. Results: A close ASUDAS/SNP association, based on MMD and F ST correlations, is evident, with r m - values between .72 globally and .84 in Africa. The same is true concerning their association with geographic distances, from .68 for a 36-trait African MMD to .77 for F ST globally; one exception is F ST and African geographic distances, r m = 0.49. Partial MMD/ F ST correlations controlling for geographic distances are strong for Africa (.78) and moderate globally (.4). Discussion: Relative to prior studies, MMD/ F ST correlations imply greater dental and genetic correspondence; for studies allowing direct comparison, the present correlations are markedly stronger. The implication is that ASUDAS traits are reliable proxies for genetic data — a positive conclusion, meaning they can be used with or instead of genomic markers when the latter are unavailable.


| INTRODUCTION
Nonmetric traits of the human permanent dentition, as in those from the Arizona State University Dental Anthropology System (ASUDAS), have a significant genetic component in expression. At least this was suggested in earlier research (e.g., Sofaer, Niswander, MacLean, & Workman, 1972;Scott, 1973;Brewer-Carias, Le Blanc, & Neel, 1976;Scott, Yap Potter, Noss, Dahlberg, & Dahlberg, 1983;Turner II, 1985a;Sofaer, Smith, & Kaye, 1986;Scott & Turner II, 1988, 1997Turner II, Nichol, & Scott, 1991;Pinkerton, Townsend, Richards, Schwerdt, & Dempsey, 1999;Rightmire, 1999; although see Harris, 1977). Therefore, phenetic affinities based on these traits approximate genetic relatedness, at levels of comparison ranging from local to global. At least this was assumed in previous work (Irish, 1997;Scott et al., 1983;Scott & Turner II, 1988, 1997Turner II, 1985a;Turner II et al., 1991). These and other attributes, such as their ease of recording, observer replicability-especially if dichotomized (below), evolutionarily conservative nature, lack of sexual dimorphism, and a low likelihood of selection in expression (Scott & Irish, 2017;Scott & Turner II, 1997;Scott, Turner II, Townsend, & Martinón-Torres, 2018;Turner II et al., 1991), are part of a common refrain in ASUDAS studies, including most cited here. Recently, however, the degree of genetic contribution and concordance of dental and genetic data, in particular, have come under renewed scrutinyseveral examples of which are summarized. Before proceeding, it of course goes without saying that non-ASUDAS traits are also used in dental research (e.g., Bailey & Hublin, 2013;Martinón-Torres et al., 2007); however, because few have been universally accepted or formally tested (below), such traits are not considered further.
Concerning genetic input, narrow-sense heritability (h 2 ) estimates in samples of Australian twins for several ASUDAS molar traits surpass 0.60, with UM1 Carabelli's reaching 0.80 and the UM1 and UM2 hypocone 0.87 and 0.93, respectively (Higgins, Hughes, James, & Townsend, 2009;Hughes & Townsend, 2011Hughes, Townsend, & Bockmann, 2016). Yet, analyses of an African American Gullah sample (Stojanowski, Paul, Seidel, Duncan, & Guatelli-Steinberg, 2018 yielded some estimates that are more modest. For eight distinct ASUDAS crown traits recorded across incisor, canine, and premolar fields and between antimeres and isomeres, the average h 2 for traits with significant p-values is ≥.33. The range is .00-.82, depending on whether the rank-scale traits were treated as continuous data or dichotomized using alternate breakpoints (defined in Turner II et al., 1991;Scott & Turner II, 1997;Scott & Irish, 2017;Scott, Turner II, et al., 2018;below). In the same Gullah sample, Stojanowski et al. (2019) then looked at 14 distinct ASUDAS premolar and molar traits in maxillary and mandibular dental fields. Heritabilities are nearer the Australian findings, as UM1 Carabelli's reached 0.85 and, on average, all statistically significant trait estimates are higher than for Gullah anterior teeth. Still, several premolar and molar values are unexpectedly low. The practice of pooling sexes, in this sample at any rate, was also questioned contra one of the abovementioned ASUDAS attributes. This h 2 range is 0.00-1.00, depending on if the traits were treated as continuous or dichotomized; the latter provided much higher estimates. Socioeconomic stress and reproductive isolation in the Gullah, along with small samples, are acknowledged that may account for the low to moderate h 2 estimates (Stojanowski et al., 2018(Stojanowski et al., , 2019. In support, prior research also suggests stress affects expression of certain crown traits (Riga, Belcastro, & Moggi-Cecchi, 2014). Lastly, several of the above authors contributed to another study analyzing Australian twins (Paul, Stojanowski, Hughes, Brook, & Townsend, 2020). Results for permanent teeth reflect those of the Gullah studies, but with higher h 2 estimates (mostly 0.4-0.8), greater trait heritability when appropriately dichotomized, and less concern for sexual dimorphism in expression. Nevertheless, the above results offer mixed signals about genetic input to expression across a range of ASUDAS traits.
With regard to correspondence of dental and neutral genetic data in appraising relatedness, four recent studies are referenced. First, in a model-free analysis of four samples of living Kenyans (n = 295 individuals), correlations were calculated between pseudo-Mahalanobis D 2 distances based on nine ASUDAS crown traits, and delta-mu squared distances from 42 short tandem repeats (STRs) (Hubbard, 2012;Hubbard, Guatelli-Steinberg, & Irish, 2015). To date, this is the only research to compare phenotypic and genomic data in the same individuals. A moderate to strong (per Cohen, 1988) positive, though nonsignificant correlation, .50, resulted from Mantel (p = 0.21) and bivariate Pearson tests (p = 0.31) (Hubbard, Guatelli-Steinberg, & Irish, 2015). Second, 19 dental and up to 19 genetic global samples were matched for comparison by population or similar provenience and ethno-linguistic affinity . Most are recentalthough the dates are not listed, two samples are from medieval to Victorian times, and one includes prehistoric material (>2000 BP). The model-bound R-matrix method, initially derived to compare allele frequencies, was used to produce pairwise population kinships from both dental (methods in Relethford, 1991;Konigsberg, 2006) and genomic data, all of which had been published previously: 12 ASUDAS crown traits (19 samples, 1872 individuals), 28 crown measurements (19 samples, 1,016 inds), 645 STR loci (13 samples, 265 inds), and 1,778 SNPs, that is, single-nucleotide polymorphisms (19 samples, 1,652 inds).
Focusing just on ASUDAS results, Mantel tests yielded higher correlations than Hubbard et al. (2015), with an r m -value for STRs in 13 samples of .55 (10,000 random permutations, p < .001) and for SNPs in 19 samples, .64 (p < .001). The dental-SNP correlation is strong (per Cohen, 1988), prompting Rathmann et al. (2017:3) to suggest dental data may be used as genetic proxies, although they "reason that a substantial portion of the variation can be explained by natural selection on dental morphology." Indeed, the presence and expression of several traits have been linked with selection (Bryk et al., 2008;Hlusko et al., 2018;Kimura et al., 2009;Park et al., 2012); this idea challenges yet another perceived attribute of the ASUDAS (see Scott & Irish, 2017;Turner II et al., 1991). Third, somewhat tangentially, Delgado et al. (2019) calculated phenetic distances from 16 ASUDAS traits in 477 living Colombians to those in dental samples of Europeans, Native Americans, and Africans (Irish, 1993(Irish, , 1997Scott & Turner II, 1997); these three are said to represent principal ancestors of admixed Latin Americans. The same Colombians had also been genotyped, and 93,328 SNPs were compared with samples of the same populations to quantity admixture proportions. Dental-based affinities revealed the closest link with Europeans, and average ancestry estimates based on genetic and dental data generally concur.
However, dental traits were not useful in assessing individual ancestries (Delgado et al., 2019). Lastly, to again explore whether dental and genetic data return similar information on admixture, Gross and Edgar (2019) employed Fisher Information in samples from West Africa, Europe, and North America. A model-bound clustering method, for multi-locus genotype data in the program STRUCTURE, was then applied to consider correspondence in estimating ancestry of individuals, that is, African, European/European American, and African American, with 53 unspecified crown traits (797 inds), up to 992,601 SNPs (271 inds), and 645 STRs (177 inds). Like most all dental studies many trait data are missing, which was suggested to affect the performance.
Still, SNPs, followed by STRs delivered superior results in "detecting differences in admixture proportions between individuals within admixed populations;" dental data were, however, deemed to be useful to investigate population-level variation (Gross & Edgar, 2019:528). As above, outcomes of these four studies offer mixed support for ASUDAS traits, in this case pertaining to use as genetic proxies for analyses of populations and, in the latter two cases, individuals.
Today, it is patent that neutral genomic markers are the definitive choice in population (and individual) studies, and the standard to which all phenotypic data are and should be compared (see Rathmann et al., 2017). That said, on the above bases expression of the latter is minimally heritable for some, while analyses based on 9, 12, 16 and 53 ASUDAS traits failed to account fully for variation among population samples and/or individuals. Two other long held attributes-lack of sexual dimorphism and minimal selection-were also called into question. Nonetheless, given its long successful run, the aim here is to give the ASUDAS another chance to demonstrate its once-posited potential, through enhanced comparative analyses at both continental and global levels; heritability is not considered directly, but rather the capacity for these traits to approximate genetic relatedness.
All of these recent studies provide a foundation on which to build. populations, and (c) these markers seemingly correspond more closely with dental nonmetric data (Gross & Edgar, 2019;Rathmann et al., 2017). Matrices calculated from the same dental traits and SNPs were then tested for correlation in an expanded analysis of 32 total global populations. Finally, it is assumed genetic and, by extension (as above), phenetic distances among populations increase exponentially as geographic distances increase (Wright, 1943;Relethford, 2004); thus, correlations between the latter distances among samples and those from dental and genetic data were calculated to explore the influence of geographic structure (e.g., extreme isolation) on the two datasets.

| MATERIALS AND METHODS
Given the large dataset and familiarity of JDI with the post-Pleistocene peopling of Africa, the latter was the clear choice for continental-level analyses; dental and genetic samples of three North and nine sub-Saharan African populations were compared. Other than one dental sample (Riet River San, Table 1) with a few earlier historic specimens, all data were recorded in recent, 19-20th century crania and hardstone casts to match close as possible the existing genomic data from living individuals (Table 2). Next, combined with Africans for global analyses were 20 dental samples from Europe, Asia, Australia, Melanesia, and the Americas. With two earlier historic exceptions (Table 1), the latter comprise similarly recent specimens to compare with 20 matched genetic samples (Table 2; Figure 1). In all following tables and figures, samples are abbreviated with a prefix of "D" (dental) or "G" (genetic), followed by sample number (1-32), and three letters for the name. For instance, the Bedouin dental sample D1_BED (Table 1) corresponds with genetic sample G1_MOR from Morocco, and so forth (Table 2). An extensive anthropological literature review facilitated sample matching based on (a) shared language and ethnic groups (e.g., Turner II, 1985a;Scott & Turner II, 1997;Irish, 1993Irish, , 1997Irish, , 2000Irish et al., 2014;Irish, 2016; see below), and (b) similar geographic locations. The average distance in km between the matched African dental and genetic samples, as determined from latitude and longitude coordinates (Tables 1 and 2), is 286.6 with a low of 64.2 and high of 505.1. The mean for all 32 samples is 347.9 km with a range, excluding the African low, of 89.6 to 1,361.3 km. The latter is the distance between the dental and genetic Aleut samples-one from the American and the other from the Russian side of the island chain. These and the two Pima samples separated by 581.9 km were included to provide some level of New World coverage. That is, recent Native North American dental data are ample in Turner's database, but matching genetic data are not Skoglund et al., 2015), with the reverse true for recent dental and genetic data in Meso-and South Americans.
All SNP data, except whole genome sequences of the West African Wolof sample (Table 2), were genotyped with the Affymetrix Human Origins Array (AHOA) (in Patterson et al., 2012;Pickrell & Pritchard, 2012;Lazaridis et al., 2014;Pickrell et al., 2014;Skoglund et al., 2016; or by request from these authors via signed letters). At both geographic levels a model-free approach (Hubbard et al., 2015), standard in most phenotype affinity studies, was conducted (though see below). For example the R-matrix method, to estimate between-population kinship coefficients , may not be the best suited for SNP data-particularly the large numbers, and correlation of results (below) based on dental and genomic data is unlikely to be dependent on the distance measures (Relethford, personal communication, 2019). An advantage of the Rmatrix method is that it can correct for genetic drift among samples, but this weighting procedure is only possible when effective population sizes are known (Leigh, Relethford, Park, & Konigsberg, 2003;Relethford & Crawford, 1995), a difficult proposition with premodern peoples (Irish, 2016). So, to evaluate correspondence, common fieldspecific measures of divergence based on ASUDAS and SNP data among respective dental and genetic samples were obtained here, where low distance values indicate similitude and vice versa between samples.
For dental data, the mean measure of divergence (MMD) was chosen relative to others, for example, pseudo-Mahalanobis D 2 (Konigsberg, 1990). It is a robust statistic that yields reliable results even with problematic traits, such as those, that are highly intercorrelated or invariant across samples; it is also less affected by missing data that characterize most dental studies and, while not necessary for the present comparative analyses, has a significance test (Irish, 2010;Nikita, 2015;Sjøvold, 1973Sjøvold, , 1977. Finally, it was found that MMD values are more highly correlated with geographic distances (Irish, 2010(Irish, , 2016Schillaci, Irish, & Wood, 2009). The formula used here has a bias correction, the Freeman and Tukey angular transformation, to correct for very low or high trait frequencies and small sample sizes (Green & Suchey, 1976;Sjøvold, 1973Sjøvold, , 1977. As required by the MMD and to simplify presentation of dental trait frequencies, rank-scale ASUDAS data were dichotomized into categories of present and absent using standard breakpoints (refer to Table 3) (Irish, 1993(Irish, , 1997(Irish, , 2005(Irish, , 2006Scott & Irish, 2017;Scott & Turner II, 1997;Scott, Turner II, et al., 2018). A few workers suggest rankscale data would give better results (Gross & Edgar, 2019;Nikita, 2015;Rathmann et al., 2017)-a concept that is not novel (see Sjøvold, 1977;Turner II, 1985b). However, suitably dichotomized trait data, perhaps surprisingly, hold several advantages (a) importantly, h 2 estimates were demonstrated to be higher (see Stojanowski et al., 2019 for details), (b) weighting bias from different grade numbers across ASUDAS traits is avoided, (c) proven distance statistics like the MMD (and D 2 ) can be applied, and (d) residual intra-and inter-observer error is reduced further. That said, the latter should be negligible, at least relative to the above studies, because data were recorded by Turner-the ASUDAS designer, and JDI, who was directly instructed by and calibrated with him (Haeussler, Turner II, & Irish, 1988;Irish & Turner II, 1990).
Many measures of divergence are available for genomic data.
However, only f 2 , outgroup-f 3 , and F ST were considered due to their ubiquitous application, the availability of online programs and, importantly, the capability of these programs to process large numbers of  Reich, Thangaraj, Patterson, Price, & Singh, 2009;Skoglund et al., 2015). Of these, F ST was selected as most appropriate for the SNP data, samples, and overall approach (Supporting information Text S1.1). Calculated under the Hardy-Weinberg model, it is theoretically not model-free; however, it is used in that capacity here, like the MMD, to "describe overall patterns of variation that can be interpreted in light of population history and structure" (Relethford & Harpending, 1994:251 (Hudson, Slatkin, & Maddison, 1992) were used, with the results of the latter detailed below. The Hudson estimator is stated to return more accurate distances with genomic data, and is less affected by very small sizes, that is, n < 4, that affect several genetic samples (see Table 2 (Ersts, 2014).

| African analyses
Next, the 36 ASUDAS traits were reduced to 25, matching those in Turner's database (refer to list in Table 7, below). However, before that, the 12 dental and 12 genetic samples were again compared to quantify any effect that reduced trait number has on the results. So as before, this trait set was submitted to the MMD. The new values are listed in Table 6 and depicted via MDS in Figure 5; for this solution the stress increased slightly (.107), and r 2 decreased (.939). Minor differences are evident between the 36-and 25-trait matrices (compare Tables 5 and 6) and in relative sample locations (Figures 2 and 5). Yet, not unexpectedly, the correlation between the two MMD matrices is almost perfect, r m = 0.977 (p = 0.000), as seen in Figure 6a

| Global analyses
The 25 ASUDAS percentages for the 20 non-African dental samples using the same breakpoints as before are listed in Table 7. To include the maximum number of traits, plus dental samples-to match with    Table S4) is strongly positive, r m = 0.710 (p = 0.000), as it is now also for

| Interpretations
Some unpublished data were included (D3_KKU and D11_YOR), but the 36-trait MMD distances (Table 4) and MDS plot of 12 African dental samples ( Figure  2) parallel prior results (Irish, 1993(Irish, , 1997(Irish, , 1998a(Irish, , 2010(Irish, , 2016Irish et al., 2014). That is, based on documented population history these ASUDAS traits provided reliable phenetic affinities among samples, and patterning indicative of geographic provenience. Reliability is why the ASUDAS has had such a long run in population studies worldwide, as evidenced by hundreds of publications (https://scholar.google.com/) and as summarized in several compendia (Scott & Irish, 2013Scott & Turner II, 1997;Scott, Turner II, et al., 2018). Now, however, what had only been assumed from earlier research about the genetic component of trait expression (Scott et al., 1983;Sofaer et al., 1972;Turner II, 1985a), but recently queried (Hughes et al., 2016;Stojanowski et al., 2019;etc.)-that dental affinities approximate genetic relatedness, is readily testable empirically with genomic data (Delgado et al., 2019;Gross & Edgar, 2019;Hubbard et al., 2015;Rathmann et al., 2017). In the present analysis, the MDS plot of F ST distances ( Figure 3) is somewhat akin to that from dental data, but the correlation between 36-trait MMD and F ST matrices (Table 4) is most telling ( Figure 4; Table 8). Stronger than r m -values in Hubbard et al. (2015) and Rathmann et al. (2017), it affords additional support for using ASUDAS traits as genetic proxies. So too does the relation between MMD and dental geographic distances (Table 5), and the partial MMD-F ST correlation controlling for geographic midpoint distances (Supplemental Information Table S1). The r m -value of the latter infers that geographic separation is not an overriding factor in the African dental-genetic correspondence. The lower, yet still moderately positive correlation between F ST and geographic distances (Table 5) may, on this continental level, indicate that F ST does not specifically detect geographic variation/isolation by distance to the same degree of some genetic distance measures (Séré et al., 2017). Conversely, it more likely indicates that F ST is better able to detect increased gene flow between geographically remote populations-particularly since the 19th-20th century dates of the present dental samples, along with extreme reproductive isolation, as in those populations who may be geographically proximate but genetically divergent (Jay, Sjödin, Jakobsson, & Blum, 2012;Ramachandran et al., 2005).
Seemingly contrary to purpose, reducing the number of dental traits from 36 to 25 (Table 6)  ( Figure 5). Deleting the 11 traits essentially functioned to emulate the editing process typically used prior to submitting data to the MMD and other similar statistics (Irish, 2010). That is, reliable results are attainable with problematic traits as noted, but it is prudent to delete the same. If this study focused only on ASUDAS-based affinities (Irish, 2005(Irish, , 2006(Irish, , 2016etc.), then nine of these traits would have been deleted in any event as standard practice, for being: 1) mostly invariant (palatine torus, UI2 peg-reduced, mandibular torus), 2) otherwise unimportant for driving inter-sample variation based on low loadings (<.5) in principal component analysis (UC distal accessory ridge, rocker jaw), and 3) highly inter-correlated (τ b ≥ |0.5|) with other traits (UI1 labial curvature; LM1 anterior fovea, LM1 deflecting wrinkle, LM1 C1-C2 crest). Thus, while one idea was to include more traits than similar studies to maximize comparative analyses, it is apparent here that "more" is not always "better." Finally, for the global-level analyses, the MDS plot of 32 dental samples (Figure 7) based on 25-trait MMD distances (Supporting   information Table S2), and the matching plot (Figure 8) of the F ST matrix (Supporting information Table S3) are comparable. Explicitly, from sub-Saharan Africa to the Americas and Melanesia, the samples cluster by regional origin and evidence overall geographic patterning.
The latter is quantified by correlations >.7 (Table 8; Figure 9a,b) for both MMD and F ST with dental and genetic sample geographic distances (Supporting information Table S4-S5). As first suggested >20 years ago (Irish, 1997:463), the MMD variation depicted in Figure 7 may be detecting the vestiges of "an expansive dental morphological cline." Recent sub-Saharan Africans were revealed to possess high frequencies of ancestral dental nonmetric traits, while other world populations transition toward higher frequencies of derived traits with increasing geographic separation (Irish, 1998b;Irish & Guatelli-Steinberg, 2003). The idea was later revisited using other dental data (Hanihara, 2013;. Therefore, if beginning with the two Considerably expanded from continental to global in scale with an additional 20 samples, the r m -value between 25-trait MMD and F ST matrices is still >.70 (Table 8; Figure 9c)-despite the caution that results may be influenced by low counts or small samples. Thus, for these 32 dental samples and MMD distances from ASUDAS data, and these 32 genetic samples and F ST distances from SNP data, the correlation further supports use of dental traits for population affinity research, with or instead of neutral genomic data if the latter are unavailable. The higher correlation between F ST and geographic distances relative to the African results is likely linked to the expanded scale. Irrespective of whether F ST is or is not unequivocally suited to detect isolation by distance (above), it is perhaps picking up on ancient among-region affinities rather than that of recent within-region (orcontinent) population movements and genetic exchange. This possibility is also likely why the global partial MMD-F ST correlation controlling for geographic midpoint distances (Supplemental Information Table S6; Figure 9d), though positive (Table 8), is much lower than the African r mvalue; geographic separation does appear to be a contributing factor to the overall dental-genetic association in this broad-scale example.

| Implications
The results cannot be equated directly with ancestry estimates in indi- Why? In answer, it is stressed that the intent is to explore potential reasons for the enhanced dental-genetic correspondence, not to critique prior studies (Delgado et al., 2019;Gross & Edgar, 2019;Hubbard et al., 2015;Rathmann et al., 2017) that provide the foundation for this research. Multiple explanations are possible vis-à-vis differences in data, samples, and methods. First, the numbers of traits, 36 and 25, are larger than in three of the four studies, and unlike all four include root and root-related traits that are highly diagnostic in characterizing populations (Scott & Irish, 2017;Scott & Turner II, 1997;Scott, Turner II, et al., 2018;Turner II et al., 1991). Moreover, when the 11 traits were dropped from analyses, testing revealed that nine are problematic for the specific 32 samples under study, including four that are highly inter-correlated; this "editing" further enhanced the dental data for comparative purposes. And, dental data were recorded by Turner and JDI. Though standardized, the ASUDAS is not intuitive to the point where trait recording can be undertaken without requisite training, including quantification of inter-observer error (Scott & Irish, 2017); examples of suspect affinities and misidentified traits illustrate this potential issue (e.g., Irish & Morris, 1996). With regard to the >350,000 SNPs, among other attributes these genomic data have been shown to better differentiate among populations, and they appear to correspond more closely with dental traits than, for example, STRs (Gross & Edgar, 2019;Rathmann et al., 2017). These markers are also substantially greater in number than in three of the recent studies, including 1,718 SNPs in Rathmann et al. (2017).
Second, 32 dental samples consisting of 2,844 individuals were compared with 32 genetic samples comprised of 530 individuals-all considerably more than the previous studies. But most importantly, though not the same individuals (Hubbard et al., 2015), creation of the dental samples and matching them with their genetic counterparts were less subjective. To illustrate, given the focus of one admixture study, recent African Americans, the authors were obliged to use casts of individuals whose "race was assigned by the orthodontists who  Tables 1, 3-7 Pedi, Sotho, and Tswana in the Sotho Branch of the Bantu Language Family, and (c) dental Haya of Niger-Congo Superfamily vs. genetic sample with Hadza who, if not a language isolate belong to the Khoisan Superfamily (Greenberg, 1963). Further, the overall lack of matches for the Americas necessitated their pairing of a recent dental sample of varied Mexican ethnicities with a genetic sample of archeological specimens, including 2,500 year-old Zapotec .
Third, methodological differences are likely a key factor in correspondence of dental and genomic data, especially their performance in estimating individual ancestry (Delgado et al., 2019;Gross & Edgar, 2019) relative to population affinities (Hubbard et al., 2015;Rathmann et al., 2017;this study). Again, comparing results between these two types of study is impractical, but another matter is the nature of ASUDAS data. While progress has been made using dental nonmetric traits to estimate individual ancestry and affinities (Edgar, 2013(Edgar, , 2015Irish, 2015;Scott, Pilloud, et al., 2018;Stojanowski & Paul, 2015;Stojanowski & Schillaci, 2006), the ASUDAS was designed to analyze samples and variation therein (Scott & Turner II, 1997;Turner II et al., 1991); thus, the "low predictive power for genetic ancestry of individuals" compared with SNPs is not surprising (Delgado et al., 2019:439). In the Gross and Edgar (2019) study, as recognized, a contributing factor is the genetic program used to estimate ancestry, which is affected by missing data; thus, the poorer performance of dental traits is also related to this issue relative to more complete genomic datasets. Concerning population studies (Hubbard et al., 2015;Rathmann et al., 2017), the justification for instead using MMD and F ST distances in a model-free  (Jernvall & Jung, 2000). That said, prior dental nonmetric studies found few or no statistically significant differences for cusp number or other trait by sex (Bermudez de Castro, 1989;Hanihara, 1992;Irish, 1993;Smith & Shegev, 1988). Significant differences that may occur appear random, in that different traits are affected among studies depending on the population (Irish, 2016); for example, it was a factor in the Gullah  Tables 2, 4-5 heritability paper (Stojanowski et al., 2019), though not so much in the most recent Australian study (Paul et al., 2020). Concerning selection, it was established to impact some traits (Bryk et al., 2008;Kimura et al., 2009;Park et al., 2012;Hlusko et al., 2018), but indirectly as a consequence of pleiotropy. Recent trait variation, at any rate, is seemingly more "a product of random processes (i.e., genetic drift and founder effect) rather than genetic adaptation" (Scott, Turner II, et al., 2018: 223). Lastly, trait heritability was shown to vary across studies and dental fields. The effects of various stressors and other issues on the populations under study may play a role, along with methodological factors like appropriate dichotomization of traits (Higgins et al., 2009;Hughes et al., 2016;Hughes & Townsend, 2011Paul et al., 2020;Riga et al., 2014;Stojanowski et al., 2018Stojanowski et al., , 2019. However, the exact answers will require additional research, which is beyond the present scope of study. Of course, dental and genetic data correspondence was addressed, and they do correspond to a much greater degree than before, based on comparing distances calculated from them. The r m -values, however, do not approach 1.0 (Table 8). Consequently, under the assumption that neutral genomic markers are indeed the definitive choice in affinity study, dental nonmetric traits are not about to supplant them. That said, these correlations may be considered minimum values (also see Rathmann et al., 2017). The data were not collected from the same individuals, and though able to match at a high level of concordance, the paired dental and genetic samples are of different ages and several come from similar, not identical, populations. Some sample size issues, inter-observer error in the ASUDAS data, and other stochastic and nonstochastic factors , also cannot be ruled out totally. In any event, the results speak for themselves. If sufficient attention is paid to the data, samples, and methods, it does appear ASUDAS traits can serve as highly reliable proxies for neutral genomic markers-regardless of potential issues mentioned, including sexual dimorphism, selection, and heritability. If the latter could be identified on individual trait and/or sample bases, which was not the case here, data correspondence may conceivably be even higher.        Table 1. b ASUDAS rank-scale trait breakpoint information in Irish (1993Irish ( , 1997Irish ( , 2005Irish ( , 2006, Scott & Turner II (1997), Scott & Irish (2017) and Scott, Turner II, et al., 2018. The capability of using dental traits as proxies is not insignificant given the cost, destructive sampling, and processing time of genetic analyses. More critically, the traits can be substituted if DNA and, more likely, ancient DNA is not recoverable. Specifically, degradation is of particular concern in tropical and sub-tropical environments, like Africa, Southeast Asia, and other equatorial regions; in more temperate climates time is also a factor, though to a lesser degree, for example, negatively affecting recovery in specimens of Middle Pleistocene age and older (Kistler, Ware, Smith, Collins, & Allaby, 2017;Pinhasi et al., 2015;Smith, Chamberlain, Riley, Stringer, & Collins, 2003). In the latter instance, "paleoanthropologists [currently do] consider teeth the "safe box" of the genetic code" (Martinón-Torres et al., 2007:7), as evidenced by many studies using dental nonmetric traits (Bailey & Hublin, 2013;Bailey, Weaver, & Hublin, 2017;Irish & Guatelli-Steinberg, 2003;Irish, Guatelli-Steinberg, Legge, de Ruiter, & Berger, 2013;Martinón-Torres et al., 2007among others). So, while there is no assurance heritability estimates in recent humans apply to our Plio-Pleistocene ancestors, these dental traits are likely as close as possible to genomic data for determining hominin origins and affinities (also see Irish, Bailey, Guatelli-Steinberg, Delezene, & Berger, 2018).
Lastly, while genomic markers are largely seen as the definitive data for population studies some caution, like that with the dental traits, should be exercised in choice and interpretation. For example, one limitation with the present SNP data is that each locus can only be represented by two alleles, as genetic software programs (see above) generally do not permit input of multi-allelic states. The result is a decrease in among-population variability, especially on a broad global basis. Another potential limitation is bias introduced when using coding positions potentially influenced by selection, that is, not all markers are neutral. Such is the case for rare variant positions like those, that code for diseases. In the present study, such bias is limited because of the large SNP number. Specifically, drift, unlike selection, influences the whole genome so selection effects are relatively few (Kimura, 1968). Moreover, the Human Origins ascertainment used in this study, unlike some other arrays, is not biased by the inclusion of SNPs with medical/clinical interest.

| CONCLUSION
In sum, the present study is the most comprehensive to date comparing dental nonmetric traits and neutral genomic markers, in selection, etc.), is that dental nonmetric traits actually can and do work well as proxies for neutral genomic data.

ACKNOWLEDGMENTS
Gareth Weedall, Liverpool John Moores University, provided invaluable assistance to AM during the bioinformatics processing for the Museum of Natural History.

DATA AVAILABILITY STATEMENT
The ASUDAS frequency data used in the analyses are available in Tables 3 and 7 of this article, as well as the relevant references cited in the main text. Many ASUDAS nondichotomized data are available in Scott & Irish (2017). The remaining ASUDAS data from the study are in preparation for publication, and/or are available upon reasonable request from the first or fifth authors. As noted, all SNP data, except the whole genome sequences of the West African Wolof sample, were genotyped with the Affymetrix Human Origins