Y-DNA haplogroup diversity is most commonly used to analyze ancestry of individual ethnic groups or linguistic families (Kayser et al. 1997; Brisighelli 2012), because Y-DNA haplogroups generally show more distinct ethnogeographic patterns than does matrilineally inherited mitochondrial DNA (Comas et al. 2000; Nasidze et al. 2003, 2004a), most likely because of higher dispersal rates of women (Seielstad et al. 1998; Oota et al. 2001; Nasidze et al. 2004b), the effects of selective pressures on the mitochondrial genome (Mishmar et al. 2003), and/or sex ratio in favor of women causing more genetic drift in males (Dupanloup et al. 2003). Moreover, there is a popular nomenclature of the haplogroups linked to a well-established phylogenetic pattern (Under-hill et al. 2001; Y Chromosome Consortium 2002; Karafet et al. 2008; Chiaroni et al. 2009).
The Caucasus is among the most linguistically and culturally diverse regions of Eurasia (Comrie 2008; Nasidze et al. 2004a; Marchani et al. 2008; Balanovsky et al. 2011; Yunusbayev et al. 2012). Currently, the region hosts dozens of languages that are grouped into three language families (Comrie 2008): Caucasian, Indo-European, and Turkic. The Caucasian language family traditionally includes Adyghean, Vainakh, Daghestanian, and Kartvelian languages (Catford 1977), although the common origin of these languages is disputed (Starostin 1989). A recent study by Pagel et al. (2013) shows that the Kartvelian and Dravidic language families are the most basal in relation to the other Eurasian language families. Diakonoff and Starostin (1988) suggest that Vainakh and Daghestanian (i.e., Northeast Caucasian languages) are related to extinct Hurro-Urartian. Armenian and Ossetian languages belong to the Indo-European language family, and Oghuz and Kipchak subgroups of the Turkic language group are spoken in the Caucasus as well (Catford 1977; Comrie 2008). Linguistic differences, along with the differences in political history, influence (but do not determine) the ethnic identities of the people inhabiting the region. Some ethnic boundaries (e.g., that of Armenians or Ossetians) coincide with the linguistic boundaries, but those of the other ethnic groups do so only partly. Some groups speaking several mutually unintelligible but related languages consider themselves to be part of a single ethnos (e.g., Georgians or Avarians). Simultaneously, language rather than religion accounts for ethnic identity in the Caucasus; for example, Ossetians, Abkhazians, and Georgians maintain ethnic integrity in spite of different religions practiced within each of these ethnic groups. Recent molecular genetic studies (Bulayeva et al. 2003; Yunusbayev et al. 2012) demonstrated that most of the Caucasian ethnic groups are more closely related to one another than to the neighboring populations of western Eurasia. Genetic differences between populations of the Caucasus and the Eastern European Plain are much greater than those between the Caucasus and the Middle East.
The human population of the region predominantly consists of descendants of the patrilinear haplogroups G2, J2, J1, R1b, and R1a (Y Chromosome Consortium 2002). Other haplogroups widespread in western Eurasia, including 12, E1b1b, and L, are present but rare, and the rest are very rare (Nasidze et al. 2004a; Battaglia et al. 2008; Balanovsky et al. 2011; Yunusbayev et al. 2012). Haplogroup G2 dominates in the western and central northern Caucasus and in ethnic Georgians (Balanovsky et al. 2011; Yunusbayev et al. 2012; Teuchezh et al. 2013). A high frequency of the haplogroup R1b is found in Armenians (Battaglia et al. 2008; Yunusbayev et al. 2012; Herrera et al. 2012) and of the haplogroup J1 in Daghestanians (Balanovsky et al. 2011; Yunusbayev et al. 2012). Relatively high frequencies of haplogroup R1a are found in Turkic-speaking peoples of the northern Caucasus (Yunusbayev et al. 2012). High frequencies of the haplogroup J2 (>20%), presumably from the early agricultural populations of the Middle East (Cinnioğlu et al. 2004; Battaglia et al. 2008; Grugni et al. 2012), are documented for the entire Western Asia and the Caucasus, peaking in Vainakh language speakers (Nasidze et al. 2004a). Despite these patterns, it is difficult to explicitly link the current distributions of paternal lineages to ethnicity or linguistics in the Caucasus, probably because of language or gene replacement.
In the spread of anatomically modern humans, heterogeneous terrain and climate cased many groups to form; limited migration among these groups—especially those stranded within glacial refugia—triggered the formation of distinct genetic lineages and languages (i.e., parallelism between genetic and linguistic evolution). For example, the groups that survived a number of glacial cycles, expanding from and contracting into their refugia, evolved into distinct Y-chromosome populations (Underhill et al. 2001; Wei et al. 2012); over time during periods favorable for migration, genetic and cultural admixture caused full language or partial gene replacement in some of these groups, though the correlation between the current genetic lineage trees and linguistic families generally remains high (Cavalli-Sforza 1997). The formation of the current ethnic, linguistic, and paternal layers of the Caucasus involved postglacial human expansions into the region from different parts of Eurasia. Divergence among the paternal lineages found in the Caucasus happened before their expansions into the Caucasus as a result of spatial isolation (i.e., isolation by distance) between 50 kya and the early Holocene (Wei et al. 2012). Some major genetic admixture events in the region happened 1,000–1,500 years ago (Hellenthal et al. 2014), so until relatively recent historical time the correlation between the haplogroups and ethnolinguistic groups in the Caucasus may have been much higher than now.
Variation in human adaptations to different environments is rarely taken into account in attempts to explain current genetic diversity of human populations. Of the ancient populations, spatially separated from one another during glacial periods at least until the end of the Last Glacial Maximum (LGM), some were probably adapted to differential climatic and ecological conditions. Although humans easily adapt to completely new environments, it is likely that, when expanding into new areas during warm periods, they initially settled in the environments similar to those of their origin. For this reason, one can expect the composition and diversity of the human gene pool to vary with environment (i.e., climate, terrain, vegetation cover) within a geographic region so diverse in climate and geography as the Caucasus but too small to cause isolation by distance among human populations given the high levels of human mobility and gregariousness. Here, we hypothesize that geographic gradients in combination with terrain and climate might influence the current distribution of the Y-DNA lineages in the Caucasus. In addition to linguistics, we included ecological environment as a predictor for the Y-DNA haplogroup distribution. To the best of our knowledge, this study is the first of its kind to explicitly consider Y-chromosome haplogroup distributions in relation to ecology and environment.
Materials and Methods
We collected hair or blood samples from 224 ethnically Georgian men throughout Georgia. Each man represented a unique exogamous clan/surname. Georgians living in Georgia have a high variety of patrilinearly inherited family names, most of which are associated with specific geographic areas (places of origin). Thus, we used the online database on Georgian family names and their places of origin (www.geogen.ge) to link our genetic data to geography. If the data on geographic origin were not available from the online database, then we made geographic assignments based on (a) the information provided by the sampled person about his historic place of origin or, if that was unknown, (b) the area where the sample was taken. We supplied our data set with the genetic profiles of 87 individuals from the Georgian DNA Project at Family Tree DNA (www.familytreedna.com/ public/georgia/). These profiles are provided with surnames that we used to link each individual to a specific geographic area of origin.
Our data set of patrilineages was supplemented with published data on haplogroup composition in other ethnic groups of the Caucasus region, which belong to different linguistic families and maintain different ethnic identities. We used the proportions of Y-DNA haplogroups in 20 ethnic groups from the Armenian DNA Project (www.familytreedna.com/ public/armeniadnaproject/), Turkey (Cinnioğlu et al. 2004), Azerbaijan (Nasidze et al. 2004a), and the northern Caucasus and Georgia (Yunusbayev et al. 2012) (see Appendix 1).
DNA was extracted from blood (65 samples) and hair (159 samples). For extraction procedures, 200 µl whole blood and a 0.5–1 cm piece of the base of the hair (10–12 follicles) was used. Extraction was performed using Qiagen DNeasy Blood and Tissue Kit following the manufacturer's recommendations (Qiagen, Valencia, CA). All samples were genotyped for 23 loci (DYS576, DYS389I, DYS448, DYS389II, DYS19, DYS391, DYS481, DYS549, DYS533, DYS438, DYS437, DYS570, DYS635, DYS390, DYS439, DYS392, DYS643, DYS393, DYS458, DYS385a, DYS385b, DYS456, and Y-GATA-H4). PCR was conducted with the PowerPlex Y23 System amplification kit (Promega Corp., Madison, WI), recommended for similar studies by Davis et al. (2013). DNA samples were amplified in 10 µl total volume with 4–6 µl template DNA, 2 ml PowerPlex Y23 5× Master Mix, and 1 µl PowerPlex Y23 10× Primer Pair Mix. Thermal cycling was performed at 95°C for 2 min; 30 cycles of 94°C for 20 s, 61°C for 1 min, and 72°C for 45 s; and final extension at 60°C for 20 min. Amplicons were run on an ABI 3130 Genetic Analyzer with Hi-Di formamide and CC5 Internal Lane Standard 500 Y23 (Applied Bio-systems, Carlsbad, CA). Genotypes were screened using GeneMapper, version 3.5 (Perkin-Elmer, Waltham, MA). The updated recommendations of the DNA Commission of the International Society of Forensic Genetics foranalysis of Y-chromosome short tandem-repeat (Y-STR) systems were followed (Gusmão et al. 2006).
Y-STR haplotypes were grouped into Y-DNA “major” haplogroups as defined by the classification of Y Chromosome Consortium (2002), using two methods: (1) Athey's Bayesian approach to estimate the probability of assignment to a particular haplogroup (Athey 2005,2006) and (2) the haplogroup predictor (YPredictor) by Vadim Urasin (version 1.5.0;http://predictor.ydna.ru/). Both of the online calculators grouped our sample into haplogroups G2, J2, J1, R1b, R1a, L, 12, T, and E1b1b. Probability of wrong assignment of the haplogroups based on the online software could be unacceptably high (Muzzio et al. 2011). Therefore, we tested the accuracy of the online calculators using individual Y-STR profiles of eight Caucasian ethnic groups, Turks, Iranians, and Russians with SNP-typed Y-DNA haplogroup assignments from the Family Tree DNA database (www.familytreedna. com/). One hundred Y-STR profiles of each major patrilineage (G, J1, J2, R1a, R1b) were downloaded from the database. Haplogroups inferred from the Y-STR profiles through both online calculators were checked against SNP-typed haplogroup assignments.
To locate Y-DNA haplogroup proportions throughout the Caucasus, first we masked out unpopulated or poorly populated areas in the Caucasus—that is, areas that are >2,200 m average sea level or areas where annual rainfall is <250 mm. We used SRTM 90 × 90 m digital elevation data (Jarvis et al. 2008) to mask terrain and climatic data from WorldClim version 1.4 (Hijmans et al. 2005) to mask rainfall. Then, we developed a data set of Y-DNA haplogroup proportions for 38 populated areas (“population units”) in the Caucasus (18 in Georgia and 20 in an area encompassing Armenia, Azerbaijan, northeast Turkey, and the northern Caucasus). In our data set, the number of population units per ethnic group was highest for Georgians because obtaining genetic samples and relatively accurate information on the places of origin of clans (see above) was possible only in Georgia at the time of our study.
The population units in Georgia were identified as the country's smallest administrative units (districts). We grouped our sample of places of origin into districts. If the sample within a district was too small (i.e., <10 places of origin), then the district was merged with the neighboring district(s). Each of the units formed this way consisted of one to five districts. Further merging neighboring population units would greatly decrease geographic resolution for environmental predictors (see below), and there was no need to further increase the minimum sample size because the Wilcoxon signed-rank test procedure via SPSS version 21.0 (IBM Corp., Armonk, NY) indicated that a sample of 10 was as representative of haplogroup frequencies within each population unit as were larger samples. For this procedure, we randomly selected 10 individuals from each of the Georgian population units, recalculated haplogroup frequencies, and compared the frequency distributions with those of the original data set via the Wilcoxon test. We repeated randomly selecting 10 individuals and the test procedure three times in order to increase the confidence of the results. At every run the Wilcoxon tests showed no significant differences in haplogroup distributions between the original data set and the data set where sample size per population unit was reduced to 10 (see Suppl. Table 3).
The population units of North Caucasus were separated based on the dominant ethnolinguistic groups, following the data of Yunusbayev et al. (2012) and an ethnic map of the Caucasus (www. zonu.com). Two population units of northeastern Turkey were separated based data from Cinnioğlu et al. (2004). Neither Armenia nor Azerbaijan was further subdivided, and northern parts of each of these countries were treated as a single population unit (Figure 1). We calculated proportions of “major” Y-DNA haplogroups within each population unit using our and existing genetic data (see above). We used this sample of 38 population units and Y-DNA haplogroups G2,J2, R1b, J1, and R1a for further analyses. We focused on paternal lineages because they overwhelmingly dominated in our data set (see Appendix 1 and Suppl. Tables 1 and 2). We managed and mapped the proportion of each of these five haplogroups in the population units using ArcGIS Desktop, version 9.3 (ESRI Inc., Redlands, CA).
Figure 1. The study area (the Caucasus) showing population units (gray polygons), no data or unpopulated areas (white polygons), and abbreviations of individual polygons that indicate linguistic groups. The upper thick line is the Greater Caucasus; the lower thick line, the Lesser Caucasus. KA: Kartvelian (Georgian, Megrelian, Svan); AD: Adyghean (Circassian, Kabardin, Abkhazian, etc.); VA: Vainakh (Chechen, Ingush); DA: Daghestanian (Avar, Lezgi, Darghin, etc.); OS: Ossetian (eastern Indo-European); AR: Armenian (basal or western Indo-European); OG: Turkic Oghuz subgroup; KY: Turkic Kipchak subgroup. All maps in this article were projected to Mollweide, false easting: 0, false northing: 0, central meridian: 45, World Geodetic System 1984.
To check to see if linguistics and environment accounted for geographical difference in frequency of each of the Y-DNA haplogroups, we used dominant language or linguistic group (Figure 1) and average values of three environmental predictors measured within each of the population units: Bailey's effective temperature (Bailey 1960), slope, and percent tree canopy cover. Bailey's effective temperature reflects the variability of the local climate and provides a simple comparative measure of sunlight, warmth, and length of the growing season. Data to calculate Bailey's effective temperature were downloaded from WorldClim version 1.4 with a resolution of 1 km2 (Hijmans et al. 2005). We derived slope from an SRTM elevation grid of 90 × 90 m (Jarvis et al. 2008). Percent tree canopy cover was extracted from 500-m MODIS data (MOD44B, http://reverb.echo.nasa.gov/reverb/#utf8=â&spatial_map=satellite&spatial_ type=rectangle).
Nonlinear canonical correlation analysis (OVERALS) was conducted for exploring the associations among major Y-DNA haplogroups of the Caucasus region, geographic variations in landscape and climate, and linguistic divisions within the region. OVERALS is a multivariate statistical method that helps to find the best-fit equations among more than two sets of variables that can be scaled as nominal, ordinal, or numerical; for example, Manly (2004) demonstrates this method in finding the relationship of frequencies of several allozyme alleles with multiple ecological variables. This method allows assigning scores to objects and categories of variables, which can be used to plot a geometrical representation of the dependencies in the data in a low-dimensional Euclidean space. OVERALS is equivalent to (1) principal components analysis if each set contains one variable, (2) multiple correspondence analysis if each of these variables is multiple nominal, and (3) categorical multiple regression if two sets of variables are involved and one of the sets contains only one variable.
Table 1. Accuracy of STR-Based Y-DNA Haplogroup Predictors, Checked against SNP-Typed Y-DNA Haplogroup Assignments
We used the OVERALS procedure via SPSS, version 21.0. The fit and loss values, loadings, and weights of individual variables and relative importance of the OVERALS dimensions were estimated as described in the user's manual for the software. Prior to the analyses, the variables were transformed according to the procedure's requirements. Haplogroup frequencies were treated as discrete numeric variables varying between 1 and 100; effective temperature, canopy cover, and slope were converted into ordinal variables by dividing each into three equal-size intervals and categorizing these intervals as low, medium, and high. We divided languages spoken in the Caucasus into eight linguistic groups (KA, VA, DA, AD, KY, OG, OS, AR; see Figure 1 for details). Each of these eight linguistic groups was treated as an ordinal variable by assigning 2 to a linguistic group if dominant within a population unit or 1 if otherwise.
We applied multivariate general linear modeling (MGLM) for testing significance and power of the effect of the environmental variables and linguistic differences on the haplogroup composition in the Caucasus. We performed three runs of the analysis estimating the effect of (1) environment only, (2) linguistics only, and (3) both linguistics and environment on geographical difference in frequency of each of the Y-DNA haplogroups. The software used was SPSS, version 21.
Results
Athey's (2005) calculator correctly predicted 83% (J1) to 98% (R1a) of the SNP-typed haplogroups (92.4% of the total number of the validated individuals), with the highest misidentification rates between J1 and J2 (Table 1). YPredictor correctly identified 90% (J) to 97% (R1a) of the SNP-typed haplogroups (93.4% of the total number of the validated individuals). Of individuals assigned to the same haplogroup by both calculators, 99.2% were in agreement with the SNP-typed assignments; hence, the probability of misclassification did not exceed 1%. Thus, for further analyses we used only the Georgian Y-STR profiles, whose haplogroup assignments both Athey's and Urasin's predictors were in agreement on (Suppl. Tables 1 and 2), reducing the total sample size of Georgians (including those from the Family Tree DNA database) from 311 to 295 men.
Figure 2 shows frequencies of paternal lineages G2, J2, R1b, J1, and R1a throughout the Caucasus. G2 reached the highest proportions (> 30%) in the western and central parts of the Greater Caucasus, peaking at 80% in Svan speakers of the Kartvelian linguistic group. J2 had the highest proportion in the eastern part of the Greater Caucasus, peaking at 82% in Ingush speakers of the Vainakh linguistic group, and was also common in the lowland areas both in eastern and western portions of the region. Haplogroup R1b had considerably high frequencies (>25%) in the southern part of Georgia, in Armenia, and at the Caspian Sea coast, peaking at 50% in Georgian speakers of the Kartvelian linguistic group in southern Georgia in close proximity to Armenia. At the national and ethnic level, R1b was most frequent in Armenia. R1a was relatively frequent in the northwestern Greater Caucasus (peaking at 30% in the Kipchak linguistic group of the Turkic languages), and J1 dominated in the eastern Greater Caucasus, peaking at 92% in Darghin speakers of the Daghestanian linguistic group.
The two first dimensions of OVERALS explained 92.4% of the total variation in the data (Tables 2 and 3). Figure 3 shows ordination of the included variables along the first two axes. Variables showing the highest loadings along the first axis were haplogroup J1, having the highest frequencies in Daghestanian language speakers, and poorly forested areas. Variables showing the lowest loadings along both the first and second axes were Y-DNA haplogroup R1a, most associated with speakers of Kipchak subgroup of Turkic languages, and cold areas. Y-DNA haplogroup G2, dense forest, and Kartvelian and Ossetian speakers from the mountains of the central-western Greater Caucasus were strongly associated with one another, having the lowest loadings along the first axis and intermediate loadings along the second axis (Figure 3, left).
Removal of variables with the highest absolute ordination scores (J1, R1a, DA, and KY) from the analysis (Figure 3) distinguished some additional groupings. High values along the first axis and low values along the second axis (West Caucasus) identified a high proportion of the haplogroup G2, speakers of Ossetian and Adyghean languages, and Kartvelian speakers that inhabit well-forested mountains. Speakers of the Vainakh languages, as well as Kartvelian speakers from mountains of eastern Georgia (high values along both the first and the second axis), were associated with a high proportion of the J2 haplogroup but not with any specific environmental variable. The area with low values along the first axis (the southern Caucasus) clustered a high proportion of the haplogroup R1b and medium slope with Armenian, Oghuz, and Georgian speakers from Armenia, Azerbaijan, southern Georgia, and eastern lowland Georgia. This area is associated with high effective temperature and sparse canopy cover. There was a large cluster in the central part of the plot, not clearly associated with any of the Y-DNA haplogroups and occupied by populations with comparable proportions of the haplogroups G2, J2, and to lesser extent R1b. This cluster mostly encompassed lowland areas in Georgia.
Initially we visually checked frequencies of Y-DNA haplogroups against landscape types. This procedure suggested an obvious link of well-forested and poorly or nonforested mountains to the distribution of some paternal lineages. Consequently, we used derivative variables such as the product of canopy cover and slope, and the product of (1 — canopy cover) and slope, accounting for well-forested mountains and poorly forested mountains, respectively. Models including these variables performed much better than those considering no interaction between slope and canopy cover. The outputs of MGLM are shown in Table 4. Frequencies of the major Y-DNA haplogroups, with the exception of R1b, were significantly associated not with individual linguistic groups but with different sets of these groups. The frequency of G2 did not respond significantly to any individual linguistic group, although it was significantly linked to the Adyghean, Ossetian, and Kartvelian linguistic groups. The frequency of J2 was significantly associated with the Vainakh linguistic group. R1b, more common in Armenian than in the other linguistic groups, did not show significant association with any linguistic groups, including Armenians. R1a was significantly associated with both Kipchak and Adyghean speakers. Only J1 was clearly associated with Daghestanian linguistic group.
Association with ecological conditions was significant for G2 and J1. The frequency of G2 significantly increased in areas where the portion of well-forested mountains was dominant and decreased in poorly forested mountains and warmer lowland areas. The frequency of J1 was associated with poorly forested mountains.
Running MGLM on both linguistic and environmental predictors slightly modified the outputs. G2 was significantly associated either with well-forested mountains or with linguistic groups such as Adyghean or Ossetian. Patrilineage J2 had a significant positive response to either effective temperature or poorly forested mountains and a significant negative response to all linguistic groups but Vainakh. The multivariate effect of the predictors on the individual haplogroups remained significant for all major haplogroups except for R1b.
Figure 3. Ordination of the major Y-DNA haplogroups (black circles), environmental conditions (gray squares), and linguistic units (white circles) in the Caucasus along the first two dimensions of nonlinear canonical correlation analysis. Left: The entire data set. Right: The data set excluding outliers such as haplogroups J1 and R1a and linguistic groups DA and KY. See Table 3 notes for abbreviations.
Discussion
In the 2000s, hundreds of publications appeared that describe the Y-DNA haplogroup composition in different ethnic or linguistic groups and geographic regions. However, no studies explicitly attempted to analyze human genotype distribution in relation to specific ecological conditions; thus, our research is pioneering in this respect. The analysis provided here suggests that there is significant impact of physical environment on the spatial distribution of patrilineages in the Caucasus. Ecological environment contributes to the paternal distribution pattern no less than the ethnic orlinguistic boundaries;hence, we conclude that this pattern formed before these subdivisions appeared, probably in the Neolithic to Bronze Age. Later historical turmoil had less influence on the patrilineage composition and spatial distribution than is traditionally thought.
Multiple studies conducted in the Caucasus (Nasidze et al. 2004a, 2004b; Balanovsky et al. 2011; Yunusbayev et al. 2012; Teuchezh et al. 2013) showed substantial differences in Y-DNA haplogroup proportions among the ethnic groups populating the region. Balanovsky et al. (2011) showed significant association among the patrilineal differences and linguistic differences in the northern Caucasus. However, more comprehensive study of Yunusbayev et al. (2012), which covered both nonrecombinant and autosomal markers, showed high genetic similarity among all ethnic groups of the Caucasus and suggested that, in Western Asia, geography is far more important to genetic structure than linguistics. Our explicit findings are in line with this conclusion.
Currently, the Western Asian population, including Turkey, Iran, and the Caucasus, is largely composed of the following patrilineages: J2, J1, G2, R1b, R1a, E1b, based on data from Eupedia (www.eupedia. com) and Family Tree DNA (www.familytreedna. com) databases. In general, it forms a distinct genetic cluster, closest to but different from the European population (Nasidze et al. 2004a; Yunusbayev et al. 2012). An early split among the patrilineages G, JI (ancestral to J and I), and K (ancestral to R1) happened as early as 43–51 kya (Wei et al. 2012), shortly after the expansion of their common ancestral group F from Africa (Karafet et al. 2008). The genetic split must have been caused by limited or no gene flow among human refugia—that is, climatically suitable areas where humans survived during shorter 1,000- to 2,000-year-long glacial episodes periodically taking place prior to the LGM (Clement and Peterson 2006). The gene flow rate must have been at its lowest during the LGM. The final split of major patrilineages, such as the split between R1a and R1b, most likely happened in the Holocene (Wei et al. 2012) and could have been caused by dispersal rather than vicariance as it is traditionally defined (Mayr 1970).
The expansion of people from human refugia in post-LGM times (Banks et al. 2008) caused admixture among the patrilineages, which further increased after the development of early agriculture 9–10 kya (Cavalli-Sforza 1997; Pinhasi and Stock 2011; Thomas et al. 2013).The expanding tribes were probably adapted to differential climatic conditions and used different technologies. Patrilineages that our study focuses on come from different refugia. Most researchers place the origin of the haplogroup J2 in the northern part of the Fertile Crescent (Battaglia et al. 2008), the location of the earliest Neolithic agrarian cultures (Diamond 1997; Abbo et al. 2010). Both high frequencies and phylogenetic diversity of G2 in the Caucasus (Balanovsky et al. 2011) suggest that the western Caucasus, well known as a glacial refugium (Tarkhnishvili et al. 2012), is indeed the ancestral area for this patrilineage. The place of origin of people of patrilineage R1b (most likely early speakers of the Western Indo-European languages, in sense of Renfrew 1987) is either the Atlantic coast of Europe (Wilson et al. 2001; Young et al. 2011; Busby et al. 2012) or the western part of Anatolia (Balaresque et al. 2010). R1a stemmed from grassland areas north of the Black Sea and the Caucasus (Keyser et al. 2009; Underhill et al. 2009), and J1 originated somewhere near the Caspian Sea coast of Iran (Grugni et al. 2012) or in the Zagros Mountains.
The inferred pattern linking individual languages of the Caucasus and Y-DNA haplogroups with their geographic areas of origin is shown in Figure 4. Ancestral areas of languages currently spoken in the Caucasus do not necessarily coincide with those of the paternal lineages. Paternal lineage J2 originates from the area where Hurro-Urartian languages were spoken in Bronze Age. These languages are related to Vainakh and Daghestanian languages (Diakonoff and Starostin 1988) spoken by people of predominantly J (J2 + J1) origin. Lineages R1b and R1a are usually associated with the western and eastern Indo-European languages, respectively (Underhill et al. 2009; Balaresque et al. 2010; Thomas et al. 2013). Archaeologists suggest that the first Indo-Europeans expanded to the southern Caucasus from the southwest in the third millennium BC (Melikishvili 1959; Melaart 1970). Currently, relatively high proportion of the haplogroup R1b is found in Indo-European-speaking Armenians but also in Georgians from the areas south of the Lesser Caucasus (this study) and in some Daghestanian ethnic groups (Yunusbayev et al. 2012). The presence of the haplogroup R1a in the northern Caucasus is associated with Turkic speakers that most likely descend from the inhabitants of the Eastern European Plain. Ancient DNA research suggests that this lineage was dominating throughout the Eurasian steppe in the Bronze Age (Keyser et al. 2009), and probably people of this lineage spoke Scytho-Sarmatian that was later replaced by Turkic and Slavic languages from Central Asia and Eastern Europe, respectively. Although Ossetian is a language closely related to Scytho-Sarmatian (Lubotsky 2002; Nasidze et al. 2003, 2004b), speakers of this language have a very high frequency of G2, not R1a. Adyghean languages spoken by G2-dominant people are linked to extinct languages of Anatolia (Ivanov 1985; Kassian 2010), spoken by ancient people of paternal lineage J2 that is rare in Adygheans. Thus, the languages spoken by present-day Ossetians and Adygheans, who genetically descend from the glacial-period population (patrilineage G2) of the Caucasus, have probably been adopted from paternally unrelated populations of the Eastern Europe and the Middle East, respectively. The Kartvelian and Dravidic language families hold the most basal position in a tree of Euroasiatic languages (Bomhard and Kerns 1994; Pagel et al. 2013). Y-DNA haplogroups G and H dominate in speakers of these two linguistic groups: Kartvelian (present results; Yunusbayev et al. 2012) and Dravidic (Sengupta et al. 2006), respectively, similarly hold the most basal position in a tree of patrilineages descending from superhaplogroup F widespread in Eurasia (Karafet et al. 2008). This fact may indicate correlated evolution of the G and H patrilineages and the Kartvelian and Dravidic languages, respectively. This logic suggests that the Kartvelian languages originate from people dominated by G lineage.
Figure 4. Ancestral geographic areas, linguistic groups, and Y-DNA haplogroups found in the Caucasus. Ancestral geographic areas (gray rectangles) are as follows. W: west (West Anatolia or Europe) associated with haplogroup R1b and Proto-Indo-European (IE) linguistic group, PCS: Ponto-Caspian Steppe associated with haplogroup R1 a and the Scytho-Sarmatian (SS) linguistic group, CA: Central Asia associated with an undefined Y-DNA haplogroup (?) and the Proto-Turkic (TR) linguistic group, CAUCASUS: the Caucasus associated with haplogroup G2, FC: Fertile Crescent associated with haplogroup J2 and the Hurro-Urartian (HU) linguistic group, S: south associated with the Zagros or the Alborz, haplogroup J1, and an undefined (?) linguistic group. For current ethnolinguistic units (black rectangles), see Figure 1. Black arrows show inferred genetic ancestry, as follows. Solid line: the most frequent Y-DNA haplogroup, dashed line: the second most frequent haplogroup with frequency exceeding 0.2. Gray arrows show inferred linguistic ancestry.
The association of patrilineage G2 with forested mountains may be a result of both similarity of this landscape to the refugial area where they survived the LGM, and competition with invading lineages from distant human refugia in post-LGM times. Grasslands or sparsely forested areas provide a greater percentage of production accessible to hunter-gatherers (Kelly 1983) and better starting conditions for agriculture than do forests (Diamond 1997). Expanding tribes (most likely dominated by haplogroups J1 and J2) that had survived the Ice Age south of the Caucasus probably started settling in the Caucasus both before emerging early agricultural settlements in the Fertile Crescent about 9.5 kya (Allaby et al. 2008) and after that. The earliest invaders could have been settled in mountain areas of the eastern Caucasus, which were relatively distant from the LGM refugial area of the West Caucasus and probably less populated. The Neolithic or later invaders, already familiar with agricultural technologies, would prefer lowland, less forested areas. They could have been more successful (compared with the older settlers dominated by Y-DNA lineage G2) in populating the most productive lowlands of both the eastern and the western Caucasus. Less productive forested mountain areas remained dominated by the local tribes. The invaders were more successful in the most productive areas probably because they outnumbered the locals, possessed better weaponry, orintroduced contagious diseases to which the locals were less resistant. Later invaders (e.g., R1b from Western Europe or western Anatolia and R1a from the Eastern European Plain) further increased competition in productive and agriculturally suitable areas, forcing J2 and J1 to concentrate in poorly forested or nonforested mountains currently inhabited by Vainakh and Daghestanian speakers in the eastern part of the Greater Caucasus.
Our results and reasoning are in line with those of Yunusbayev et al. (2012), who suggest that the core of the genetic structure of the Caucasian populations formed long before its present-day linguistic pattern. This particularly applies to the patrilineages of the earliest settlers, G2 and J2, which still dominate in most of the region. It appears that historical religious, linguistic, and political expansions have had less influence on the current geographic distribution of the dominant paternal lineages in the Caucasus than does territoriality established in prehistoric times in the context of local ecological adaptations.
Ethnogenetic processes that include major linguistic and cultural expansion occurred in the relatively recent past (Geary 2002) and rarely caused full or substantial displacement of rural communities by invaders. Rural communities typically changed their identities and spoken languages as a result of political circumstances, without changing preferred environments and paternal genetic structure. Our results are in line with such a vision. There is a clear geographic pattern for patrilineages G2, J2, and R1b, which together make up 75% of the ethnic Georgian population. This pattern is not associated with historical provinces of Georgia speaking different dialects and even different languages such as Megrelian and Svan, or with ethnic boundaries between Georgians and Ossetians or between Georgians and Armenians. The frequency of patrilineage G2 significantly correlates with forested mountains and declines in both eastern and western parts of what is usually called the Transcaucasian depression (Zimina 1978), that is, plains in both the Black and Caspian Sea basins, where agriculturally productive lands are concentrated and human population has been dense since early postglacial times (Murtskhvaladze et al. 2010). The proportion of G2 also declines in forestless mountains of the southern and western Caucasus. The population of the Transcaucasian depression is the most diverse patrilineally and comprises comparable frequencies of G2 and J2 and to lesser extent R1b and J1. Even though inhabitants of lowlands of western and eastern Georgia speak mutually unintelligible languages of the Kartvelian linguistic group (Megrelian and Georgian, respectively), they are paternally identical and are dominated by the J2 lineage.
Away from the Caucasus, our modeled relationship between G2 frequency and environment seems to be applicable to at least two areas covered with forested mountains and characterized by a reasonably high rainfall level: in the Alps (Berger et al. 2013) and Iranian provinces of Gilan and Mazenderan at the southern Caspian coast (Grugni et al. 2012). Regions where haplogroup J2 is frequent (Battaglia et al. 2008; King et al. 2008; Myres et al. 2011) are dominated by either poorly forested mountains or lowlands, which also agrees with our model. Patrilineage J1, which is found in dry mountain areas of the Caucasus, is common in even drier parts of the Middle East. Our models obtained for R1b and R1a are not transferable outside the Caucasus, which might suggest that these paternal lineages were the last of the five studied haplogroups to reach the Caucasus and had to adapt to new niches different from their places of origin.
It is likely that the formation of the major ethnolinguistic groups of the southern Caucasus followed shaping of the current pattern of the major haplogroup distribution rather than preceded it. Historical records suggest that the first Georgian political state, comprising parts of the ancient states of Colchis and Iberia, emerged in the third century BC at the latest (Suny 1994; Rapp 2003). This state probably expanded over several adjacent geographic areas with different proportions of the major paternal lineages J2, G2, and R1b. Therefore, proto-Georgian ethnos has incorporated at least three genetically and ecologically distinct units with long-established economic and cultural interactions. This might have been reflected in writings of Strabo, who says, “The plain of the Iberians is inhabited by people who are rather inclined to farming and to peace…but the major, or warlike, portion occupy the mountainous territory” (qtd. in Suny 1994). It is likely that the segregation of the mountain and lowland rural populations marked by different Y-DNA haplogroups was stronger in Strabo's time than now. It was recently shown that gene pool of present-day Georgians is a result of a major admixture event in eleventh century that involved, on one side, a population genetically similar to the present-day inhabitants of the West Greater Caucasus and, and on the other side, a population genetically similar to the rest of West Asia (Hellenthal et al. 2014). This was the time of consolidation of the Georgian state in medieval times under the rule of kings of the Bagratid dynasty (Suny 1994).
Most likely, similar ethnogenetic processes took place in Armenia and current Azerbaijan, although the spatial-genetic structure of these countries was different because of different proportions of major landscape types.
ACKNOWLEDGMENTS
The research was financed from the budget of Ilia State University and implemented at the Center of Biodiversity studies, established within the CoRE framework of a Georgian Research and Developmental Foundation/Georgian National Science Foundation joint project. MSc and BSc students Ardashel Latsusbaia, Giorgi Iankoshvili, and Levan Kalatozishvili assisted in processing the genetic data. We appreciate editorial work of the Human Biology editorial team on the text of the manuscript.
Author contributions: D.T. and A.G. analyzed the data and wrote the text. M.M. coordinated molecular genetic work and provided primary data analysis; M.G. did much of genetic analysis and identification of the haplogroups; G.T. initiated human genetic studies at Ilia State University, discussed the results from the standpoint of history and social sciences, and made much effort to provide the research with all necessary resources.
LITERATURE CITED
Three supplemental tables are available at the following web address: http://digitalcommons.wayne.edu/cgi/viewcontent.cgi?filename=0&article=1053&context=hum biol_preprints&type=additional
Tables
Print Friendly
Article Impact
RSS Feeds