However, the nature of the domestication process forms an open question. The number of animals captured, its duration and whether it was discrete or continuous over a long period remain unknown. Whereas overall mitochondrial DNA (mtDNA) evidence does seem to point towards some restriction in place and time in domestication events (Machugh & Bradley 2001), low-frequency phylogenetically outlying haplotypes betray some degree of secondary introgression from the wild (Achilli et al. 2008, 2009; Stock et al. 2009). These must be regarded as suggesting much greater secondary autosomal contribution. Incorporation of wild matrilines must always have involved difficult captures; alternatively, input to the domestic pool via wild inseminations would have immediately resulted in semidomestic hybrids and seems clearly a more likely, and hence frequent, type of introgression to have occurred within domestic history.

The recent genomewide analysis by the bovine HapMap consortium found high levels of genetic diversity in all cattle breeds surveyed, exceeding those found in human and dog populations, and revealed more than twice as much nucleotide diversity in one indicine breed in comparison to B. taurus breeds (Marth et al. 2003; Ostrander & Wayne 2005; Gibbs et al. 2009; Lewin 2009). These data agree with previous findings from major histocompatibility complex (MHC) allele diversity by Vila et al. (2005), which strongly argue that the allele spectra in many domestic species today are not depleted in diversity, as might be expected from a traditional view of early domestic history involving a severe capture bottleneck. Additionally, it has been noted that levels of protein polymorphism in cattle, as assayed by classical methods, are typical for mammals (Lenstra & Bradley 1999), showing no dramatic indication of a restricted capture of diversity from the wild.

In contrast, two recent analyses suggest discernible population genetic signatures of the domestication process within modern cattle. Estimates of past effective population sizes derived from examining linkage disequilibrium decay with distance, using a medium density single nucleotide polymorphism (SNP) panel, indicate a large predomestication population with an approximately 50-fold decline at domestication followed by a further decline with the formation of modern breeds (Goddard & Hayes 2009). Also, Finlay et al. (2007) examined mtDNA datasets from four domestic Bovidae (cattle, mithan, yak and water buffalo) using the Bayesian approach to the coalescent process implemented in BEAST, and compared these to two wild bovid samples (bison and African buffalo). In each domestic sample, a sharp recent increase in effective population size was estimated that seems consistent with the timeframe of domestic history. That this was related to human capture and taming was suggested by the observation that neither wild bovine showed this phenomenon.

At a descriptive level, nucleotide diversity seems higher within indicus; in 11 out of 16 loci, π is greatest in B. indicus, and in eight of 16 loci θ_W is greatest in B. indicus. Haplotype diversity is also consistently higher with 11 of 16 loci giving a highest value in this sample. This strongly suggests divergence in demographic histories between the two taxa.

Our initial modelling approach (not shown) was to assume possible bottlenecks followed by population expansion in the history of each of the three sampled populations. We also allowed migration between continents. Under this approach, the B. taurus allele frequency spectra fitted a demography with a severe bottleneck but this predates by a factor of 2.75 the modelled divergence of African and European cattle which, from other evidence, probably dates to the early domestic period or possibly before (Wendorf & Schild 1994; Bradley et al. 1996; Achilli et al. 2008; Ho et al. 2008). However, the search for a maximum-likelihood fit would not allow a bottleneck within B. indicus history, and an excess of high-frequency-derived alleles in that population suggested a more complex scenario.

To account for this, and given the clear similarity between the B. taurus populations illustrated in a PCA (figure 1), we developed a second three-population model with a single B. taurus population and two B. indicus populations: one a domesticated lineage with a bottleneck and recovery, plus a second parallel wild population that was allowed to mix with the first. This allowed a closer fit of observed and modelled data.

Several features of the best-fitting model are interesting. First, the early and clearly predomestic population bottleneck persists in the ancestral demography of European and African B. taurus. When one assumes an ancestral population size calculated at 14 127 and a generation time of 5 years this maps to between 40 and 36 kyrs ago with s.d. of the order of 11 kyrs. There is uncertainty in this calibration but we note that some estimates of the ancestral population size for wild cattle are larger (Goddard & Hayes 2009), which would push these calendar estimates further into the past. It is possible that this single-modelled bottleneck represents an amalgam of a more variegated demographic history, which perhaps includes a domestication bottleneck. However, it does seem to point primarily toward an earlier episode, perhaps associated with the effects of the glacial period on the West Asian aurochs. Interestingly, ancient DNA analysis of 34 mtDNA sequences sampled from European aurochsen bones also gives a signature of population expansion that concurs with the timeframe of glacial retreat and which, by definition, cannot be a result of the domestication process (Edwards et al. 2007; Ho et al. 2008). We note that an equivalent early population constriction is missing from B. indicus ancestry, probably reflecting a more benevolent glacial ecology within South Asia. Divergence between zebu and taurine mtDNA has been calibrated repeatedly, typically being of the order of hundreds of thousands of years (most recently by Ho et al. (2008) as between 84–219 kyr ago). Our estimate of 280 kyr with s.d. 22 kyr for ancestral separation is of similar order.

The strongest contrast between West and South Asian domestic history lies in the modelled complexity within the domestication process. The comparatively large haplotypic diversity and distinctive allele frequency spectrum within B. indicus seem to defy a uniform domestication narrative. Whereas the best-fitting model allows a B. indicus ancestral bottleneck that lies conceivably within the domestication timeframe, this requires a majority input (approx. 80%) into the domestic indicine gene pool from wild admixture. We are restricted here from examining more complex models but it should be noted that this may not exclude alternatives such as multiple strands of wild ox ancestry arising within B. indicus ancestry through two or more independent domestications.

Some geographical and genetic complexity within B. indicus domestication does seem likely from prior work. mtDNA sequencing has described two clusters of indicine mtDNA haplotypes and these have a non-random distribution when samples from east and west are compared (Baig et al. 2005; Lai et al. 2006; Magee et al. 2007). This is argued as suggesting that more than one domestication process may have given rise to B. indicus. Domestic cattle first appear in the Baluchistan early agricultural center, Mehrgarh, 8000–7000 bp. This is in the context of other domesticates (goats, barley and wheat) that probably diffused eastwards from the Fertile Crescent (Fuller 2006). However, artistic representations and thoracic vertebrae with morphology typical of zebu have allowed Meadow (1987) to argue that these early cattle were B. indicus of local origin. A transition in the nature of Bos bone finds (abundance, size) points towards this as a plausible primary centre for domestication of zebu from the South Asian aurochs, Bos primigenius namadicus.

Additional centres for cattle domestication in the subcontinent are possible. The South Indian Neolithic features distinctive ashmounds that are thought to have been produced by the burning of cattle dung (Misra et al. 2001). These have been argued as representing sites of cattle pens that may have served as enclosures for capture and taming of aurochs 5000 years ago (Allchin & Allchin 1974). That these may have given rise to a separate strain of zebu is (tentatively) indicated by early representations that seem to depict cattle with longer horns and more delicate limbs in comparison to those represented in early seals from Baluchistan. More persuasively, finds of large Bos bones suggest the survival of wild cattle populations into the Neolithic in South India and the Ganges region further East (reviewed in Fuller 2006). Recruitment of wild oxen via crossbreeding, or less commonly by capture, in different regions of the subcontinent is thus eminently plausible. Recently, mitochondrial sequence diversity levels within B. indicus has been described widely through Asia and strongly suggest that highest sequence diversity and therefore domestic origins lie within the Indian subcontinent but that these may well not be restricted to the primary Baluchistan region (Chen et al. 2010).

Our model suggests admixture between the continental groups. This is not surprising for a domesticate that would have been a passenger or subject of human trade and migration. This mirrors earlier work that shows traces of African genes in European B. taurus and vice versa (Cymbron et al. 1999; Loftus et al. 1999) and ancestral exchange between Near Eastern and Indian breeds (Loftus et al. 1999; Kumar et al. 2003). For example, a huge secondary input of B. indicus genes into Africa has been described, probably reflecting an ancient Indian Ocean trade corridor stretching back over three millennia (Hanotte et al. 2002).

One issue with the analysis we present here is that many of the loci chosen for resequencing have importance in the immune response and therefore may have been subject to natural selection, which may have skewed the SFS. We use only synonymous and non-coding SFS in our analysis but a possibility remains that skew may persist because of linkage disequilibrium with adaptive sequence variants. However, the high frequency bump in our B. indicus SFS that drives the strong conclusion of majority wild admixture in Indian cattle ancestry is a feature that remains even when the three genes (TRIF, TLR4, ART4) showing highest linkage disequilibrium are removed from the analysis. Also, a skew because of selective effects should manifest in both B. indicus and B. taurus and our strongest conclusion is of divergent domestication histories between the two taxa. Finally, it should be noted that reasonable modelling inference has been achieved in human populations using sequence data with a strong bias towards immunologically important loci, although of course this does not guarantee that such will always be the case (Akey et al. 2004).