Various anthropological analyses have documented extensive regional va
riation among populations on the subcontinent of India using morpholog
ical, protein, blood group, and nuclear DNA polymorphisms. These patte
rns are the product of complex population structure (genetic drift, ge
ne flow) and a population history noted for numerous branching events.
As a result, the interpretation of relationships among caste populati
ons of South India and between Indians and continental populations rem
ains controversial. The Hindu caste system is a general model of genet
ic differentiation among endogamous populations stratified by social f
orces (e.g., religion and occupation). The mitochondrial DNA (mtDNA) m
olecule has unique properties that facilitate the exploration of popul
ation structure. We analyzed 36 Hindu men born in Andhra Pradesh who w
ere unrelated matrilineally through at least 3 generations and who rep
resent 4 caste populations: Brahmin (9), Yadava (10), Kapu (7), and Re
lli (10), Individuals from Africa (36), Asia (36), and Europe (36) wer
e sampled for comparison, A 200-base-pair segment of hypervariable seg
ment 2 (HVS2) of the mtDNA control region was sequenced in all individ
uals. In the Indian castes 25 distinct haplotypes are identified. Asid
e from the Cambridge reference sequence, only two haplotypes are share
d between caste populations. Middle castes form a highly supported clu
ster in a neighbor-joining network, Mean nucleotide diversity within e
ach caste is 0.015, 0.012, 0.011, and 0.012 for the Brahmin, Yadava, K
apu, and Relli, respectively. mtDNA variation is highly structured bet
ween castes (G(ST) = 0.17; p < 0.002), The effects of social structure
on mtDNA variation are much greater than those on variation measured
by traditional markers. Explanations for this discordance include (1)
the higher resolving power of mtDNA, (2) sex-dependent gene flow, (3)
differences in male and female effective population sizes, and (4) ele
ments of the kinship structure. Thirty distinct haplotypes are found i
n Africans, 17 in Asians, and 13 in Europeans. Mean nucleotide diversi
ty is 0.019, 0.014, 0.009, and 0.007 for Africans, Indians, Asians, an
d Europeans, respectively. These populations are highly structured geo
graphically (G(ST) = 0.15; p < 0.001). The caste populations of Andhra
Pradesh cluster more often with Africans than with Asians or European
s. This is suggestive of admixture with African populations.