Evidence page · lineage markers

Eighty-one per cent of Tamil male lineages are indigenous, and castes and tribes carry the same ones.

Y-chromosome haplogroups are the markers most often invoked when caste and ancestry are discussed together, and they are the most frequently misread. This page sets out which lineages the Dravidian-speaking populations actually carry, how they are distributed across caste and tribal groups, what structure exists in that distribution — and what the largest study of the question found the structure tracks, which is not rank.

The lineages.

what South Indian populations carry

Frequencies reported for Tamil castes: R1a1 27%, H 21% (predominantly H1), L 13% (predominantly L1), R2 11%, J2 11%, F* 10%. Each of F*, H1, J2, L1, R1a1 and R2 exceeds 5% in most castes — that is, the common lineages are shared across caste groups rather than sorted between them.

The finding that matters most.

1,680 Y chromosomes, 31 populations, Tamil Nadu

The largest study of the question this museum has located genotyped 1,680 Y chromosomes across 12 tribal and 19 caste endogamous populations from Tamil Nadu, taking advantage of the fact that both the Y chromosome and caste membership pass down the male line.

The groups, and what each carries.

31 endogamous populations, sorted seven ways

The Tamil Nadu study did something unusual: rather than assume varṇa rank was the right way to group people, it tested several classifications against the data and let the genetics choose. The seven groups that emerged are defined by subsistence, funerary custom and mother tongue — not by rank.

Overall frequencies — seven haplogroups carry 82% of the variation

Where each lineage peaks — the actual per-group numbers

But the study is explicit that these peaks conceal wide internal variation. F*-M89 in the forager group ranges from 75% down to 28.6% across its constituent populations. The same holds for H1-M52 within HTK and L1-M27 within DLF. Not all populations within a group share a genetic makeup — drift, fragmentation and isolation have acted differently on each.

Which classification the data actually prefers.

AMOVA, and the number that decides it

The subsistence grouping maximised difference between groups and minimised variation within them — significantly better than grouping by geography or by varṇa rank. And breaking it down further is where it becomes interesting:

The same appears in the PCA and MDS plots: HTF, HTK and BRH form three distinct, distant clusters, and every other population is interspersed in the middle — the foragers at the F-M89 pole, the Kannada-speaking tribes at H1-M52, the Brahmin-related at R1a1-M17, and everyone else between.

The language connection.

who is closest to whom, and it is not who lives nearby

The Kannada-speaking hill tribes make the point sharply. They are hunter-gatherers inside Tamil Nadu speaking a different Dravidian language, they carry the highest H1-M52 frequency of any group, and they cluster with the foragers rather than with the Tamil-speaking populations around them. Language boundary, subsistence boundary and genetic boundary coincide — and the varṇa boundary runs across all three without matching any.

When the groups separated.

coalescence dates, and what they rule out

And the R1a1 story does not hold up either

Brahmin-related populations do carry the highest R1a1-M17 frequency in the sample, at 42.2%. But three separate results argue against reading that as an incoming Indo-Aryan lineage:

• The highest R1a1 diversity is not in Brahmin populations. The highest STR variances for R1a1 were found in the Scheduled Castes and the dry-land farmers — and diversity, not frequency, indicates where a lineage has been longest.
• The R1a1 network shows no population-specific clusters — no signature of a distinct group carrying it in.
• The coalescence analysis could not produce a modal tree for R1a1 at all, which it managed for F*, L1 and H1.

Together the authors read these as arguing against introduction of these lineages through a single wave of Brahmin migration. And the Brahmin-related populations carry no significant frequency of any haplogroup with a likely origin outside India.

One more result that inverts an assumption

The maternal line tells a different story.

and the asymmetry is the point

Where Y lineages show some structure, mitochondrial DNA shows very little. Tribal groups analysed for mtDNA coalesce at Indian-specific branches of haplogroups M and N that cover populations of different social rank across the whole subcontinent. Studies of caste groups in eastern India found Brahmins with high paternal affinity to eastern Europeans through R1a1 — while maternal polymorphisms revealed primarily Indian-specific lineages.

How the data got handled when it was inconvenient.

the Chenchu case

What these markers actually measure.

the mechanics, because the findings are unreadable without them

A haplogroup traces one line out of thousands

Go back ten generations and you have 1,024 ancestors. The Y chromosome traces exactly one of them — father's father's father, unbroken. Mitochondrial DNA traces exactly one other — mother's mother's mother. Between them they capture two ancestors out of a thousand and tell you nothing about the other 1,022.

This is why haplogroup findings and genome-wide findings can look contradictory without conflicting. They are answering different questions. "81% of male lineages are indigenous" is a statement about one inherited line in each man. It is not a statement that these populations are 81% indigenous in overall ancestry — the genome-wide estimates, which integrate all those thousands of ancestors, give different proportions. Anyone quoting a haplogroup percentage as though it were an ancestry percentage has made a category error, and the error is extremely common.

Why the paternal and maternal records diverge

The consistent finding — external input visible in the Y, largely absent from mtDNA — follows from three mechanisms working together:

What endogamy does to a genome, and how fast

Sealing a community has predictable consequences that compound over generations:

• Drift accelerates. In a small closed population, allele frequencies wander randomly and fast. Variants that were rare can become common for no reason but chance — which is how the Vysya reached a hundredfold elevated rate of a metabolic disorder with no selective advantage involved.
• Founder effects fix whatever the founders happened to carry. If a hundred people seal, the entire subsequent population inherits their particular sample of variation, including its accidents.
• Shared segments lengthen. Members of an endogamous group share long identical stretches of chromosome, because their common ancestors are recent. This is what makes founder events detectable at all.
• And differentiation appears between groups that were recently identical. Two communities sealed side by side for two thousand years will diverge measurably without either having come from anywhere else.

The unit problem: varṇa is not the endogamous unit

This is the flaw that runs through much of the literature, and it maps directly onto a distinction this platform enforces everywhere.

Endogamy operates at jāti level. A study that recruits samples labelled "Brahmin" or "Kshatriya" is aggregating dozens of distinct jātis that have not intermarried with each other for two thousand years — pooling populations that are, genetically, no more one group than they are one group with their neighbours.

Two timescales that get collapsed

Arguments about caste and genetics almost always collapse these. A claim that caste groups are "genetically distinct" is usually true at the third timescale, marginally true at the second, and false at the first — and the three get stated as one. The deepest layer is shared, the middle layer varies, and the most recent layer is the one that produced today's boundaries. The boundaries are the youngest thing in the data.

What these markers cannot do.

stated because this is where genetics gets weaponised

• A haplogroup is one locus out of a genome. The Y chromosome traces a single unbroken father-to-son line and says nothing about the thousands of other ancestors a person has in the same generation. It is a thread, not a portrait.
• Frequency is not identity. No haplogroup is exclusive to any caste. All the major lineages appear across castes and tribes; what varies is proportion.
• Antiquity is not rank. Carrying a deeper indigenous lineage is not a lower status and carrying a steppe-associated one is not a higher one. The markers have no ranking built into them; the ranking was applied to people and then read backwards onto their chromosomes.
• The categories are constructed. "Caste" in these studies means the group a sample was recruited under. Those categories were shaped by colonial enumeration and by the communities themselves, and they are not natural biological units.
• And the direction of inference runs one way. Genetics can show that populations mixed and stopped mixing. It cannot show why, who enforced it, or what it cost — and the strongest finding here is that the structure predates the system that is usually invoked to explain it.

The sources, audited.

run before relying on them

The honest limits.

• Sources: per-group frequencies, significance tests and coalescence estimates from ArunKumar et al. 2012 (PLOS One, doi, 1,680 Y chromosomes, 31 populations); shared clades across castes and tribes from Kivisild et al. (AJHG); the haplogroup-package critique and the Chenchu case from Sahoo et al. 2006 (PNAS) — snippet level; the quoted phrase “an aberrant phenomenon” has not been checked against the paper; rank-correlation results from BMC Genetics 9:86 — snippet level; the reading of those results as historical female mobility is an interpretation, not a finding of that study. Kivisild et al. (AJHG) — snippet level; the citation lacks a year and volume, and is recorded as incomplete.
• A haplogroup percentage is not an ancestry percentage. The Y traces one ancestral line out of more than a thousand in the same generation. Genome-wide estimates are a different measurement and give different numbers.
• Endogamy operates at jāti level. Studies recruiting under varṇa labels pool dozens of jātis that have not intermarried for two millennia; their figures are averages over heterogeneous populations.
• Frequencies vary widely within every group — F*-M89 ranges from 75% to 28.6% across the forager populations alone — and by study. They are not a profile of any community.
• Commercial ancestry-testing figures are not used here.