Y chromosome diversity, human expansion, drift, and cultural evolution

Results and Discussion

Haplogroups: Their Phylogenesis and Geographic Distributions.

The original nomenclature system of Y chromosome genotypes standardized in 2002 defined 18 haplogroups, labeled A through R (12). There are now 20 main haplogroups, designated A through T (13). Their basic phylogenetic relationships are shown in Fig. 1. Two deep branches, unifying haplogroups IJK and MNOPS, are shown. The geographic distributions of frequencies of each haplogroup were calculated on the basis of published data (listed in Table S1) and are shown in Fig. 2 for the 15 numerically and geographically most important haplogroups. The K, M, and S haplogroup distributions are not shown because they occupy rather unique space, mostly in Oceania. The F* and P* haplogroups are paraphyletic and often rare, except for F* in India, which codistributes with H, and are also omitted.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Current phylogenetic relationships of the 20 major haplogroups of the global Y chromosome gene tree. Star denotes the new topology formed by the M522 and M523 SNPs that now join the previously independent IJ-M429 and KT-M9 haplogroups. Diamond indicates position of M526 SNP that now unifies haplogroups KMNOPS. M522, M523, and M526 marker specifications are given in Table S3.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Y chromosome haplogroup geographic frequency distribution maps.

Haplogroups A and B are the deepest branches in the phylogeny and are essentially restricted to Africa, bolstering the evidence that modern humans first arose there (14, 15). Haplogroup A is mainly found in the Rift Valley from Ethiopia to Cape Town, mostly but not exclusively in some of the oldest hunter-gatherers who still survive and speak Khoikhoi and San languages, proposed by some to be the oldest languages. The interruption of its distribution in the middle of the Rift Valley is possibly the consequence of replacement by Bantu-speaking farmers who settled the region starting in the first millennium of the Christian era. Haplogroup B is found mainly among African Pygmies, who live in the central African forest and are still predominantly hunters-gatherers but speak Bantu languages borrowed from farmers who arrived in the area between 2,000 and 3,000 years ago. The third predominantly African haplogroup, E, diversified some time afterward, probably descending from the East African population that generated the Out of Africa expansion. The geographic distributions of the major branches of this haplogroup, given in Fig. S1b, suggest that most of the settlement outside of Africa by haplogroup E members involves the later mutant E-M35 varieties like M78, M81, and M123 that extended to Arabia and the northern Mediterranean coast.

Although the precise phylogenetic relationships among haplogroups F–H remain uncertain (Fig. 1), 3 phylogenetic unifying relationships in the interior framework of the phylogeny reveal shared patterns of common molecular heritage. First, the CF-P143 common ancestor shows an ancient split from the DE-YAP clade, clarifying the earliest diversification events outside Africa (Fig. S2a). Second, the unification of haplogroups IJK by M522 and M523 now creates evolutionary distance from F–H representatives, as well as supporting the inference that both IJ-M429 (Fig. S3a) and KT-M9 arose closer to the Middle East than central Asia or eastern Asia. Third, the very successful and widespread MNOPS-M526 (Fig. S3a) ancestor indicates that haplogroups L and T have individual ancestry consistent with their current distinctive geography (Fig. S3a). The phylogeographic patterns shown in the 4 last rows in Fig. 2 are consistent with the Out of Africa expansion approximately 60,000 years ago, characterized by rapid dispersal across Eurasia and Oceania and followed by subsequent isolation (16).

Haplogroups G–J, T, and L generally are constrained geographically near Europe, the Middle East, and parts of western Asia with various extensions, especially to Arabia and India. Haplogroups D, H, T, and L have more localized distributions and often lower frequencies. The haplogroups with the greatest spatial distribution are those associated with widespread haplogroups NO and PQR ancestry. Haplogroup R peopled central and western Asia and a great part of Europe. Haplogroup N is frequent across boreal and northwest Asia and O in southeast Asia including the islands and part of New Guinea.

Haplogroups C and Q display Asian ancestry and hold the unique privilege of having settled America. Not surprisingly their origin seems to have been in northeast Asia. The absence of haplogroup N in the Americas indicates that its spread across Asia happened after the submergence of the Bering land bridge. It is likely that haplogroup C entered America after Q, even though C originated phylogenetically earlier than Q. In fact, C is found only in the northern part of North America, limited to the area where the family of languages by American Indians called Na-Dene (17) is spoken.

Inferring the Place of Origin of Haplotypes or Haplogroups.

Haplogroup centroids with their standard deviations are given in Tables S2 and S3 and their location in Fig. S4b. Standard deviations measure the spread of each haplogroup as average distance from the mean and are calculated on straight-line distances from the means.

Several geographic distributions have multiple modes. These may be due to geographic barriers and/or secondary migrations or expansions, but also to later arrivals of other haplogroups that replaced/displaced earlier settlements in the regions between 2 widely separated modes. This may have happened specifically, for instance for haplogroup A, which probably at the beginning occupied continuously the Ethiopia–South Africa region, but in the first millennium after Christ much of it was subsequently replaced because of occupation by various waves of expanding Bantu farmers from Cameroon across all southern Africa. Another phenomenon is the presence of small propagules far away from the rest of the distribution, as in haplogroups B, G, I, J, L, H, and Q. In part this phenomenon is due to the small size of many samples. Sometimes there is complete geographic separation between areas showing fairly important presence of the haplogroup, as in G, H, Q, and T. For example, haplogroup H in Europe is often confined to ethnic Roma. Another phenomenon possibly responsible for some of these irregularities, especially in the early part of the expansion, is the surfing effect described later.

If migrations were random, the geographic distribution of individuals with a specific haplogroup would be approximately normal (Gaussian) around the place of origin of the oldest mutation defining the haplogroup, apart from irregularities due to vagaries of the environment: obstacles, like mountains and deserts, or favored routes, like coasts and rivers. Below we list the most prominent 16 haplogroups by the number of major modes (i.e., the local geographic maxima of the gene frequencies of each haplogroup), which are easily observable on the maps as the darkest blue areas (Fig. 2): haplogroups B, D, G, J, L, M, and O are largely unimodal geographically, apart from isolated propagules (e.g., in B). The presence of haplogroup O in remote Madagascar is consistent with a well-known part of the Polynesian expansion. Most of these haplogroups have a clear, major mode and a relatively symmetric geographic distribution around it. Some like G, L, S, M, and T do not reach very high frequencies and/or have a very limited geographic distribution. Haplogroups A, E, and H are sharply bimodal, whereas C, I, N, Q, R, and T are clearly multimodal and several of them have a wide and abundant world representation. By subdividing each haplogroup into a tree showing the geographic distributions of its subclades, one can distinguish within-haplogroup paths of expansions and their mutational hierarchy. Examples are shown in Figs. S1–S3, S4a, and S5 for the most interesting cases of complicated distributions: A, B, E, I, J, O, and R. Even for superficially unimodal haplogroups like J and R this analysis reveals interesting complexity. The similarity of patterns of different mutants indicates some secondary expansions. It is also interesting to sum the distributions of different haplogroups descending from the same mutation, as for example D and E, which both descend from DE-YAP, the first mutation that split into the E branch that perhaps returned to Africa (or arose there), whereas the other branch, D, is found today mainly in the Himalayas and Japan.

For unimodal haplogroups, centroids are not far from the modes of the geographic distribution of the population represented by the haplogroup. Their approximately spatial (bidimensional) Gaussian distribution suggests that the populations carrying the haplogroup may have expanded relatively slowly and relatively homogenously in all directions around their place of origin. This is more likely to happen if the local environment has no major barriers or irregularities. Under these assumptions, the centroid or the mode or some intermediate position are reasonable estimates of the place of origin of the earliest mutation(s) marking the haplogroup. Such a geographic assignment of the place of origin of the mutation leading each haplogroup allows superimposing a phylogenetic tree on a geographic map of the origin of each haplogroup, producing an approximate space and time display of major migrations of the human species in the past.

The Surfing Effect.

Fisher (18) studied the rate of expansion of an advantageous mutant and showed that a “wave of advance” forms and proceeds at a constant rate in which its spread pattern is dictated by migration rate and the degree of selective advantage. The same formula is valid also for the expansion of a demographically growing population that enters new territory under a constant rate of migration, and a constant population density at saturation during the whole advance: thus, populations will move away from their origin by a “wave of advance” that soon reaches a constant shape from beginning to end of the expansion.

Mutants that arise in the wave front of an expanding population have an interesting advantage over mutants arising behind the expansion front, in the fully saturated portions of the expansion. This advantage is due to random genetic drift, because, as is well known, the probability of success (final fixation) by drift of a just-arisen mutant is equal to 1/N, where N is the population size. Therefore a mutant arisen within the expansion front finds itself in a position of advantage over mutants arisen back of the expansion front, where populations have reached saturation levels and therefore higher Ns. This is due to the fact that these mutants find themselves for some time in a locally small population, which is smaller the closer the new mutation is located to the extreme line of the advancing front. The smaller the local population in the immediate neighborhood of the mutant's origin within the advancing front, the higher will be the relative local frequency of the mutant, and therefore the mutant's probability of final success by drift alone is greater. The result is that they have a greater chance of final success, even if they are not favored by natural selection (i.e., are selectively neutral). The advantage increases the closer the just-born mutant is to the extreme of the advancing front, where the population density is thinnest. The faster the population expansion, the greater the probability of success of a mutant that arises in the wave front, because then the wave front is longer.

The existence of this phenomenon was detected by simulations, which showed that it affects in a characteristic way the shape of the geographic distribution of selectively neutral mutants that originate in the wave front. It has been called the surfing effect because it superficially resembles the fact that a swimmer surfing on the wave crest travels with the speed of the wave (19, 20). Observations on the geographic distribution of several Y chromosome mutants show that some of the patterns strongly suggest that the surfing effect is at play. The ocean coasts and other barriers will stop expansions, and several of the observed geographic distributions show specific increases of the frequency of the mutant in the direction of migration that can reach highest values up to 100%, at or before barriers most distant from their likely place of origin. Such examples include (i) the haplogroup A-M6 mutant that reaches a very high frequency near the Atlantic Ocean between Angola and Namibia; (ii) the E-PN2 variant that probably arose near the Red Sea and has its highest frequencies near the Atlantic coast of West Africa; (iii) I-M170 on the Atlantic coast of Scandinavia; (iv) J1-M267 in southern Arabia; (v) O-M122 in China; (vi) N near the Arctic; (vii) Q in South America; and (viii) R in western Europe.

One consequence of the surfing effect is that for mutants showing such geographic patterns, the place of origin of the mutants is neither at the spatial mode of the mutant frequency nor at the spatial average frequency, which seem reasonable for most other mutants. This is also intuitively suggested by the observed spatial distributions of the frequencies of the mutants just listed, which are highly asymmetric: the shape of the distribution is such that the mode would often be extrapolated on the coast line or close to it, whereas the arithmetic average is inside land but close to the coast.

A deterministic approach (21) would place the origin of the mutant midway between the average frequency of the mutant and the place of origin of the expansion, assuming it proceeds at a constant rate. But drift tends to shift the place of origin of a mutant even closer to the origin of the expansion, by 10% in the simulations shown in ref. 19. Other sources of uncertainty are due to lack of knowledge of the exact routes followed by the expansion, considering the vagaries of the Earth's surface; the estimates used in our graphic presentations of expansion paths are tentative. Another source of uncertainty is the possibility that the evolutionary success of some of the mutants presented above is due at least in part to natural selection for the mutants. The absence of recombination within most of the Y chromosome (and especially for the haplogroups that we study) means that the particular mutant used to label the haplogroups is not necessarily responsible for the natural selection of any other mutant(s) that appeared after the one used for labeling the haplogroup.

One might surmise that the geographic distributions observed and imputed to surfing are entirely due to natural selection. Although we have not made specific computations, which would demand extensive simulations, it seems likely that the high gradients of gene frequencies observed would request unusually high selection coefficients. One cannot exclude that mutations affecting male fertility (or perhaps less probably, survival) might be involved and determine relatively high selection, but they seem unlikely. One indication in favor of surfing is that most of the examples of potential surfing mentioned above are in regions where gradients of genetic diversity have already been noted. These are indications of important partial expansion routes during the Out of Africa one, or secondary ones, connected with agro-pastoral transitions or other less well known major movements. Fast expansions favor surfing.

Genetic Diversities Among Populations for Autosomes, X Chromosome, and Y Chromosome, and the 1:4/3:4 Expected Ratio of Diversities.

As mentioned, the estimates of relative genetic diversity for the average autosome (henceforth called A), for the X chromosome, and for the Y chromosome under drift alone are expected to vary inversely to the number of chromosomes per couple of parents, which are 1 for the Y chromosome, 3 for the X chromosome, and 4 for autosomes. An earlier article (11) showed that the variations among autosomes A and X chromosome are in a ratio very close to the expected: 4/3 (= 1.33). Another article (22)* has partially confirmed the rule, with the exception that West Africans showed a slightly but significantly greater ratio than Europeans or Asians.

There are several reasons that can explain the difference between these 2 articles. The first article (11) examined 948 individuals from 51 world populations from all 5 continents, whose DNA is available at the Human Genome Diversity Project (HGDP)–Foundation Jean Dausset collection (23), for 650,000 SNPs (1 of the 52 original HGDP populations was excluded because the number of individuals is too small). The second article (22) examined 3 populations: 1 from West Africa (Nigeria), 1 from northern Europe, and 1 made of equal numbers of Chinese and Japanese for a total of 240 individuals, and for a variable number of SNPs. Because Li et al. (11) examined many populations from all continents with a larger number of SNPs, and the average data show no significant deviation from the expected 4/3 ratio, it is likely that the anomaly observed by Keinan et al. (22) has a simple explanation, which they themselves suggest: one African population they examined is slightly biased. The Li et al. (11) data included 6 African populations certainly different from theirs, and the anomaly was absent. A likely explanation is that the southern half of West Africa used by Keinan et al. underwent a recent agricultural expansion, some 4,000–5,000 years ago, whereby a new, strong founder effect probably altered considerably the local genetic diversity (18, 23).

One of our interests in collecting Y chromosome data was to test their usefulness in evolutionary studies. An analysis of early data indicated that the Y chromosome showed considerable world variation (24). Many later data accumulated since that early beginning gave stronger proof of the correctness of this hypothesis (25). The published data we collected for this article allowed us to calculate an Fst among populations (Table 1), which we compare with autosome and X chromosome estimates by Li et al. (11).

The expectations for X and Y are calculated with respect to the autosome value A, which being an average of all of the autosomes whose individual Fsts vary very little, has a much smaller statistical error. The X chromosome expectation (1.33 A) is practically identical to the observed value, and that of the expected Y chromosome value is significantly higher (P ca. 0.001) than the observed one, but the difference in value is only 17%. There is no indication that Y chromosome SNPs used for population analysis are subject to natural selection. It is practically certain that many autosomal or X genes do show some natural selection differences among populations. Taken at face value, these data suggest that there is somewhat more natural selection on average in A+X than in the Y chromosome, but the difference from the drift expectation based on the 1:4/3:4 rule the difference for the Fst value is modest.

Slopes of the Serial Founder Effect in Different Situations and with Different Markers.

Fst calculates the genetic difference between 2 or more populations. The total diversity among individuals is usually partitioned into 2 variances, among populations and within populations. A simple approach to studying the within-population fraction is to calculate as genetic variance among individuals forming a given population, the “average heterozygosity,” which is the frequency of heterozygotes for each gene averaged over all genes. This is obtained using the average heterozygote frequency for every gene in each population expected under random mating. Although this quantity has full genetic meaning only in autosomal genes, it can also be calculated from the frequencies of haplogroups of the Y chromosome and has the same meaning for studying genetic diversity within populations. Between-populations variance g regularly increases proportionately to their geographic distance from the population of origin. Conversely diversity within populations decreases regularly because it is 1-g in relative values. In fact, the within-population variance does fall linearly with the distance from the population of origin, an observed phenomenon called serial founder effect (10) that is explained almost entirely by drift and migration. In Table 2 we compare data available for the slope of the decrease of genetic diversity within populations for autosomes, X, and Y chromosome for 3 different types of markers (SNPs, microsatellites, or haplogroups), as a function of the geographic distance of the population from that of origin of modern humans.

Although there is no general theory about the expected slope of g under the serial founder effect, it seams reasonable to assume that also for different chromosomes the slope of the within-population variance will vary proportionately to the expectations under drift, which are given for A:X:Y as 1:4/3:4. But there may be other factors affecting the slope that differ according to marker type or for other reasons. Simulations conducted for microsatellites and SNPs (11, 25) indicate that the serial founder effect slope depends on mutation rates, decreasing in absolute value for increases in certain mutation rates. Because no satisfactory theory of the dependence of the slope on the mutation with different types of markers is available, accurate comparisons can be made only for marker types that are subject to the same mutation rates.

Table 2 shows that present data do not afford making comparisons using the same type of markers, but at least qualitatively some expectations of the simple drift model are satisfied. In particular the highest value of the slope is obtained with the Y chromosome (25.7), and it is near the expectation of 4 times the autosome value, but only when it is compared with the microsatellite value (25.7/6.5 = 3.9). The comparison between the slopes obtained for A with 783 microsatellites (6.5) and 650,000 SNPs (3.8) shows that microsatellites have significantly flatter slope than SNPs, as expected because of their much higher mutation rate. The only comparison among different types of chromosomes using the same type of SNPs is from Li et al. between X and A, but the 2 values differ significantly from the 4/3 ratio, X being much higher than A in absolute value. There is a good explanation for this disagreement: SNPs of A are a random choice, whereas the X data refer to selected haplotype frequencies and are likely to be closer to 50% than those of random SNPs. This will undoubtedly increase the expected absolute value for the Xm and may explain the discrepancy. In conclusion, it is worth pursuing further the approach of the serial founder effect slope. In general these results from the analysis of the serial founder effect are in general agreement with the idea that drift is responsible for a very high proportion of the observed genetic variation among human populations, but presently available data can offer only qualitative support for it. Further work aimed more directly to it can supply stronger evidence.