Introduction

India is home for moderately isolated populations, which pose a stark contrast to their immediate and sub-continental neighbours due to genetic, geographical, cultural and linguistic factors. With only anecdotal accounts of their ancestry/origins, Indian population offer a unique opportunity to anthropologists and geneticists to understand their genetic history and unravel erstwhile unknown human migrations. The country continues to be popular for explorations of genetic architecture of ancient/isolated human populations with the most recent being the novel and exciting findings from the Rakhigarhi excavation sites1. The South Asian cline has emerged as a result of multiple waves of migration and genetic mixing among diverse populations over thousands of years. The Ancient Ancestral South Indians (AASI) are regarded as one of the earliest indigenous groups (65,000 ybp) inhabiting the Indian subcontinent and predominantly engaging in a hunter-gatherer lifestyle; and possess a distinct genetic lineage contributing to the genetic diversity observed in present-day South Asian populations. Notable amount of this ancestry was prevalent among people of Indus Valley Civilization (IVC) along with Iranian hunter gatherer ancestry1 but shows minimal relatedness to any specific modern-day population outside of the Indian subcontinent. Utilizing mitochondrial, Y-chromosomal and autosomal genomic markers, extant Indian populations have been suggested to be a blend of two hypothetical ancestral populations: the Ancestral North Indian (ANI) and the Ancestral South Indian (ASI)2,3. ANI was formed as a result of admixture of IVC people with Middle or Late Bronze Age Steppe (Steppe MLBA) people, while ASI was formed by admixture of IVC with AASI individuals. The contemporary Indian populations are the result of intermixing of these two ancestral populations between 1900 and 4200 years before present (ybp), with ANI and ASI contributing to varying proportions among caste and tribe groups of North and South India3. This suggests a complex history of genetic interactions and social dynamics in shaping the genetic landscape of the Indian subcontinent.

Karnataka, a state located in the southwestern part of India, is home to various ethnic groups with diverse ancestral lineages along with notable historical interactions with foreign traders and conquerors4. The Coorgs (also known as Kodavas) inhabiting Kodagu (Coorg) are one such group, nestled away in the Western Ghats, the geographically isolated, eroded slopes of the Deccan Plateau (Fig. 1A). The term Kodava is used for the language, culture and the dominant community inhabiting the region from pre-historic times. Traditionally a group of agriculturists with martial customs interlacing their day-to-day lives, the Coorgs practice family exogamy and caste endogamy till date4,5. Furthermore, the Coorgs are distinctly different from the neighbouring populations in their religious as well as socio-cultural practices. Though they presently follow Hinduism, various customs prevalent among the Coorgs notably deviate from the Hindu way of life6, resulting in two models for their origin.

Fig. 1: Sample collection location and population structure within Coorg.
figure 1

A The geographical and sampling locations of contemporary Coorg samples. B Biplot of principal component analysis of Coorgs with modern Eurasian populations with first two components; C Stacked barplot of the ADMIXTURE analysis with K = 7 with global populations ordered geographically. (IndianIEU -Indian Indo-European, IndianDRAV—Indian Dravidian, IndianAAS—Indian Austroasiatic, IndianTB—Indian Tibeto-Burman).

Full size image

The native model ascribes the Coorgs as pre-historic inhabitants of Kodagu. While Hutton described them as descendants of brachycephalic humans who entered the Indus Valley during the Mohenjodaro period7, the Kaveri Purana an inset of the ‘Skanda Purana’ (8th Century CE), classifies them as a warlike native tribe of Kodagu. Conversely, the non-native models ascribe their lineage to migratory groups such as i) The Indo-Greek army of Alexander’s India chapter; ii) pre-muslim Kurds or pre-Christian Georgians8; and iii) an off-shoot of the Indo-Scythian Sakas4,9. These being entirely anecdotal or from the early population dating, scientific evidence is lacking.

In this study, genetic architecture of the Coorg population, likely time of their origin and inter-population relatedness across contemporary and ancient global populations were investigated. We analysed a representative subset of socio-culturally homogenous present day Coorgs, using high-resolution mitochondrial DNA (mtDNA), Y-chromosomal and autosomal markers. This first report demarcates the population into three distinct early sub-groups (Coorg1, Coorg2 and Coorg3), which eventually admixed genetically and socio-culturally.

Results

Substructuring in Coorg population

The PCA biplots presented includes Coorgs (black), Indo-Europeans (blue), Dravidians (red), Austroasiatic speakers (brown), Tibeto-Burman (khaki), Pakistani groups (forest-green), and North West Indian (light-green). Interestingly, the Coorgs segregated into three groups and we named them as Coorg1, Coorg2 and Coorg3 (Fig. 1B). Coorg1 (black circles) clustered near extreme of the South Asian cline with most of the diverged Dravidian groups with highest ASI ancestry; Coorg2 (black diamonds) showed partial affinity with Indo-European and lay mid-way among Indo-European caste groups; Coorg3 (black triangles) was unique and did not cluster with any of the modern Eurasian or south Asian populations, who are seen as a single continuous cline of high West Eurasian to low West Eurasian affinity (Fig. 1B).

Unique ancestral component of Coorg3 in ADMIXTURE analysis

In the unsupervised model-based clustering with ADMIXTURE using K = 7 (Figs. 1C), Coorg3 which was isolated in PCA, formed a unique component (blue) different from other Eurasians. Coorg2 was similar in ancestry profile to Indo-Europeans, with similar proportions of red, teal and khaki components but with a minor Coorg3 component. Coorg1 individuals were enriched for the South Asian-specific (red) and with negligible Coorg3 components. Collectively, all the three Coorg groups share the Coorg3 component with each other.

Shared alleles among three Coorg groups and modern Eurasians

In the outgroup F3 statistics, three Coorg groups showed most allele sharing with each other followed by south Indian caste groups such as Gaud, Kunabi, Mala and Kallar caste groups (Supplementary Data 1a-c). Coorg1/GroupA shows highest outgroup F3 with Coorg2 (F3 = 0.04294; z = 135.58) followed by Coorg3 (F3 = 0.04288; z = 136.99) and Gaud (F3 = 0.04279; z = 129.72) from Telangana, India (Supplementary Data1a). Coorg2 shows highest allele sharing with Coorg1 (F3 = 0.04294; z = 135.58) followed by Coorg3 (F3 = 0.04284; z = 136.38) and Gaud (F3 = 0.04269; z = 129.72) (Supplementary Data 1b). Similarly, Coorg3 showed highest F3 value with Coorg1 (F3 = 0.04288; z = 136.99) followed by Coorg2 (F3 = 0.04284; z = 136.38) and Gaud (F3 = 0.04263; z = 132.59) (Supplementary Data 1c).

Genetic dissection of the three groups uncovers ancient ancestral lineages

In distal modelling, Andamanese Hunter-Gatherers (AHG), Iran Neolithic samples (Iran_N), Anatolia Neolithic samples (Anatolia_N) and Eastern European Hunter-Gatherers (EEHG) were used as left groups. Among all South Asian groups tested, Iran_N component was comparatively higher in Coorg3 (0.38) and Coorg2 (0.31) followed by Kamboj (0.27), Dogra (0.24) and Sikh_Jatt (0.21). Interestingly, composition of Coorg1, like Palliyar, was similar to AHG (0.7), Iran_N (0.2) and EEHG (0.04) (Fig. 2A; Supplementary Data 1g).

Fig. 2: Admixture modelling to infer ancient ancestral sources in Coorg.
figure 2

A Distal and B Proximal admixture modelling with qpAdm for three Coorg groups and other Indo-European and Dravidian populations.

Full size image

In proximal modelling with Bronze Age sources (AHG, Indus_Periphery and Steppe MLBA), Coorg3 had second highest contribution from Indus_Periphery (0.59) group after Kamboj (0.61), followed by Coorg2 (0.58), Dogra (0.57) and Sikh_Jatt (0.50) (Fig. 2B; Supplementary Data 1g).

Clues towards an ancestry from an additional admixture for Coorg3 using qpGraph

The three Coorg groups were jointly fitted into the Admixture graph topology using qpGraph implementation of Admixtools 210. Initial graph based on previous South Asian graph11 was used as a starting point with some simplification and automated graph exploration was run 30 times for each group using find_graph feature of Admixtools 2. Of the 10 best fits with score closer to zero and consistent with the known admixture history of South Asia, the top two best fitted topology are shown (Fig. S6A-B). Of note, both graph topologies required an additional source of ancestry from an unknown lineage for Coorg3 with 1% and 9% contribution for first and second graphs respectively; same held true with all the 10 best fitted graph topologies. Sample scores and bootstrap resampled scores for both fitted admixture graphs are shown in the Supplementary Data 1h.

Extreme population-specific drift and divergence in Coorg3 subgroup

In the Maximum likelihood tree constructed using TreeMix v.1.1212, although placed among Indian Indo-European caste groups, Coorg3 showed significant population-specific drift as indicated by longer branch length compared to other similar groups in South Asia except Kalash and Gujjar (Fig. S13). Though Coorg2 was placed in same clade with Coorg3, there was no significant drift. Coorg1 shared the same clade with Palliyar with exactly similar drift and both shared the clade with Ulladan, which had a much higher extent of drift (Fig. S13).

In distance matrix of WC-Fst, Coorg1 and Coorg3 showed similar drift to that of Austroasiatic group (Fig. S7). The extent of population-specific drift among Coorg2, Coorg3, Kalash and Gujjar was compared using qpGraph. Coorg3 had similar amount of drift as that of Kalash and Gujjar (Fig. S8B), while Coorg2 did not have any significant population-specific drift (Fig. S8A).

Fine scale population structure and haplotype sharing

PCA with the co-ancestry matrix from ChromoPainter13 clearly differentiated all three Coorg groups from other populations with distinct clustering at left corner, while others formed a cline along the diagonal at bottom-right (Fig. S9). Individuals of Coorg2 were in the same cluster with Coorg3, while Coorg1 individuals were placed away from Coorg2 and Coorg3 clusters (Fig. S9).

The fineSTRUCTURE13 tree kept all three Coorg group individuals in an isolated cluster of 2 minor clades, whereas other West Eurasian and South Asian groups were in three separate major clusters (Fig. S10). Coorg1 individuals were placed in one of the two minor clades along with Coorg2 (Fig. S10). However, none of the Coorg1/GroupA individuals were sharing clade with Coorg2 individuals. Coorg3 individuals were placed altogether separately from Coorg2 or Coorg1 individuals. In the simple coancestry matrix calculated from Chromopainter chunkcounts, Coorg1 was observed sharing chunks with Palliyar individuals (indicated by yellow ellipse in bottomright) (Fig. S11). All the three Coorg groups shared highest number of chunks with each other and also among themselves (top-right corner of the matrix) (Fig. S11).

Runs of homozygosity, relative IBD score and admixture dating

In the Runs of Homozygosity (RoH) analysis using Plink 1.914, Coorg1 showed highest mean of total length of RoH at the window size of 1000 kb, outcompeting even Palliyar (Fig. 3A) but mean total number of RoH was highest in Narikuruvar and Malaikuruvar. Conversely, Coorg3 had the least distribution in terms of both mean length and mean number of RoH among all the South Asian populations. Coorg2 was placed along with other Indo-European caste groups like Brahmin_Tiwari, Dogra and Lodhi indicating higher effective population size and low level of recent consanguinity. But for higher window sizes of 2500 kb (Fig. 3B) and 5000 kb (Fig. 3C), HakkiPikki outcompeted Coorg1. Yet, Coorg1 had highest distribution as compared to majority of castes and tribes in South Asia. We also tested the effect of Founder event/population bottleneck on RoH distribution using forward genetic simulation-based approach. We recreated three distinct demographic scenarios based on strength of founder events in Coorg1 and Coorg3 and also with no bottleneck scenario. Very similar distribution of RoH segments in both Coorg1 and Coorg3 was witnessed in comparison to empirical data (Fig S16).

Fig. 3: Runs of homozygosity and identity by descent score in Coorg compared to modern references.
figure 3

Scatter plot of the average lengths of Runs of Homozygosity (RoH) against average number of RoH segments for three Coorg groups and other South Asians using three different windows of (A) 1000 kb, (B) 2500 kb and (C) 5000 kb. D IBD score of three Coorg groups along with populations from India with significantly higher IBD score in relative to Finnish population.

Full size image

IBD score of Coorg1 relative to Finnish population was very high (2.854) compared to Coorg2 (1.447) and Coorg3 (0.6), and was very close to Vysya (3.122), Reddy_Telangana (2.19) and Panta Kapu (2.175) (Fig. 3D) (Supplementary Data 1i). In the IBD chunk sharing matrix, Coorg2 and Coorg3 shared IBD only with each other and none of the modern Eurasian populations. On the other hand, Coorg1 shared IBD only with Palliyar and not with Coorg2 and Coorg3 (Fig S12).

In the estimation of the best fitted model and date of admixture using fastGlobeTrotter15, Coorg3 was fitted twice for admixture model with modern Eurasians. In the first model, we excluded Coorg2 and Coorg1 individuals as donor/surrogtaes for Coorg3, and the best sources for admixture were Pathan (ancestry proportion = 0.42) and Mala (ancestry proportion = ) dating approximately to ~98 generations before present (Supplementary Data 1d). Another model inferred after including Coorg1 and Coorg2 as donor/surrogates suggested admixture between Sikh_Jatt (ancestry proportion = 0.10) and Coorg2/GroupB (ancestry proportion = 0.90) (Supplementary Data 1d), suggesting a more recent admixture. Best fitted model of admixture for Coorg2 was between Coorg3 (ancestry proportion = 0.86) and Coorg1 (ancestry proportion = 0.16) (Supplementary Data 1e). The admixture model was one-date and more recently 11 generations ago (Supplementary Data 1e). The best fitted admixture model for Coorg1 was between Mala (ancestry proportion=0.40) and Palliyar (ancestry proportion = 0.60) at approximately 64 generations ago (Supplementary Data 1f).

Demographic history and demographic parameter estimation

We proposed two alternate demographic models for each of the Coorg groups based on known South Asian genetic admixture history using Demes16. For group specific demography, we used corresponding admixture model from fastGlobeTrotter15 runs. We used Moments’17 inference optimization function to arrive at best likelihood model and parameters. For Coorg1, we tested two alternate models, where Model1 hypothesize ASI (Ulladan and Palliyar-like) as direct ancestral source, while Model2 hypothesize putative admixture between group on ANI-ASI South Asian cline and ASI (Ulladan and Palliyar like). We selected Model2 (Log-likelihood: -165173.44976401704) over Model1 (Log-likelihood: -171341.95843789072) based on their likelihood scores, which corroborated well with admixture model inferred from fastGlobeTrotter15 run. Further, Coorg2 fits well in a relatively simple model of admixture between ANI and ASI (Log-likelihood: -150681.82191639318) compared to a more complex model of a putative source on Indian cline (ANI-ASI) and ASI (Ulladan and Palliyar) (Log-likelihood: -164313.6476346726). For Coorg3, we could obtain a best fitted model (Log-likelihood: -138845.99235543355) in which this group is derived from Indian cline (ANI-ASI admixture) with relatively recent pulse of admixture from ANI-like group. This recent admixture is also obtained in fastGlobeTrotter run (where Sikh_jatt is proxy for ANI and Coorg2 is proxy for Coorg3) (Supplementary Data 1d) (Fig. 4A-C).

Fig. 4: Demographic modelling in three Coorg groups and inferred parameters.
figure 4

A Best fitted Demographic model for Coorg1/GroupA inferred from Moments and corresponding fitted parameters. B Best fitted Demographic model for Coorg2/GroupB inferred from Moments and corresponding fitted parameters. C Best fitted Demographic model for Coorg3/GroupC inferred from Moments and corresponding fitted parameters.

Full size image

Model parameter estimates from best fitted model of Coorg1 suggest significant reduction in effective population size in this group, with much less migration rates between ANI and Coorg1 (M_French_GroupA = 7.35e-05) (Fig. 4A). This effective population size change was lesser in case of Coorg3 and least in case of Coorg2 (Fig. 4B-C). Migration rate between ANI/French and Coorg2 was higher as compared to Coorg3, which reflects that Coorg3 remained comparatively isolated.

Historical effective population separation history and Y STR-based network analysis

We first tested the presence of any founder event in the three Coorg groups using ASCEND, but we observed significant founder event only in Coorg1 (If = 2.7%; Tf=40 GBP) (Fig. 5A). In Linkage-Disequilibrium (LD)-based historical population size history using HapNe-LD18, we observed a very slight reduction in size in Coorg1 at generation 40, then continuous increase in size thereafter (Fig. 5D). This correlates with ASCEND19 based founder event estimate at 40 generation ago (Fig. 5A). On the other hand, both Coorg2 and Coorg3 show continuous decline in effective population size just after 40 GBP (Fig. 5E-F).

Fig. 5: Strength and probable timing of founder event in Coorg groups.
figure 5

ASCEND plots showing time (Tf) and strength (If) of founder events (if any) happening in the population history of: (A) Coorg1 (B) Coorg2 and (C) Coorg3. HapneLD-based historical effective population size history for Coorg1 group (D) Coorg2 group (E) and Coorg3 group (F) X-axis represents times in generations before and Y-axis represents effective population size.

Full size image

There was a major migration and population turnover in India during the Bronze age related to Steppe groups carrying R1a lineage; and R1a-M780 and R1a-Z2125 were two distinct lineages present among Bronze age cultures of Steppe11. Therefore, we explored the relative contribution and diversity of these two Y chromosomal R1a lineages among Coorgs (as our Globetrotter admixture model suggests Coorg3 lineage dating back to Bronze age). Median-Joining (MJ) network for two sister clades R1a-M780 and R1a-Z2125 was constructed using a dataset of 17 Y chromosome STRs of the three Coorg groups (Supplementary Data 2e). Samples not used for genome-wide SNP array (for whom affiliation to Coorg1, Coorg2 and Coorg3 is unknown) were also included. In the analysis with both markers (M780 and Z2125), distinct clustering of Coorg individuals in a separate branch and not with any of the Central Asian, Middle East or South Asian branches was observed (with exception of a single individual with Central Asians in case of M780) (Fig. S17). The Time to Most Recent Common Ancestor (TMRCA) analysis for R1a-M780 and R1a-Z2125 suggest more distant divergence time for Coorg haplotypes compared to any other Indian R1a haplotypes. Among the three Coorg groups, Coorg3 R1a lineages are most ancient followed by Coorg1 and then Coorg2 (which is comparatively more recent) (Supplementary Data 2f).

Mitochondrial markers enriched for South Asian haplogroups across the three Coorg groups

Coorg2 and Coorg3 were more diverse compared to Coorg1 in terms of mitochondrial haplogroup diversity, with presence of four major haplogroups (M, U, R and H or HV) (Supplementary Data 2a-c). All the three Coorg subgroups had highest frequency of South Asian-specific major haplogroup M, followed by H. Haplogroup U was observed only in Coorg2 (0.25) and Coorg3 (0.16) subgroups, while mitochondrial haplogroup R was present with highest frequency in Coorg1. Apart from these, unique haplogroups such as haplogroup G in Coorg2 (0.04), haplogroup HV in Coorg3 (0.07) and haplogroup T in Coorg1 (0.08) were also observed in low frequency (Supplementary Data 2a-c).

In terms of Y chromosome haplogroup distribution, Middle Eastern haplogroup J and Indus civilisation-related haplogroup L were present at comparable frequency in all three Coorg groups (Supplementary Data 2a-c). However, South Asian-specific Y chromosome haplogroup H was observed with highest frequency in Coorg 3 (0.45), lesser in Coorg 2 (0.14) and with a complete absence in Coorg 1 (see discussion). Steppe related haplogroup R1 was highly prevalent in Coorg1 (0.5), followed by Coorg2 (0.32) and Coorg3 (0.07). South Asian-specific haplogroup R2 was observed in all three groups with moderate frequency. A few unique haplogroups such as haplogroups O in Coorg2 (0.04) and Coorg3 (0.03) and haplogroup K in Coorg2 (0.04) were also observed.

Discussion

The population history of India is rich, varied, interesting as well as intriguing. The extensive diversity in this rather old subcontinent is due to multiple waves of migration into the region from foreign lands, over millennia and eventual geographical and linguistic isolation and practicing of endogamy. This in-depth study on one such small, isolated and socio-culturally homogenous unique sub-population of Coorgs revealed their ancient origin, immense population drift due to isolation and a notable contrast with the neighbouring populations.

Multifaceted analyses revealed: i) a clear delineation of the present-day Coorg population into three distinct groups: Coorg1, Coorg2 and Coorg3 (Fig. 1B); and ii) their distinctness from neighbouring populations (Fig. 1B) warrants further investigation. Diverse ethnolinguistic groups such as the Kannadigas, Konkanis, and Tuluvas populating Karnataka are a mixture of ANI and ASI ancestry but Coorg1 and Coorg2 though falling on this cline, remain distinct entities. Furthermore, other martial ethno-linguistic groups similar to the Coorgs including Nairs, Bunts, Ezhava, and Hoysala Brahmins have shown an early separation from the Indo-Gangetic Indo-Europeans and the Dravidians and with enhanced middle-eastern component20, Coorg2 being the most similar to them (Fig. 1B). Coorg1 with the highest similarity to tribal populations such as the Palliyar, is also distinct from local tribes such as the Kurchas, Kurubas, Kurumans, Ezhavas (Fig. 1B). This distinction can be attributed to a significant population size change in this group (Fig. 5D-F) which correlates with the occurrence of a founder event ~40 GBP (Fig. 5A-C). Conversely, Coorg3 stands out as a starkly distinct group (Fig. 1B). The original individualities of these three groups are further cemented by the differing divergence times observed in Coorg1, Coorg2, and Coorg3 (~64 GBP, ~11 GBP, ~98 GBP, respectively; Supplementary Data 1d-f). Coorg2 arose from admixture events between comparatively older Coorg1 and Coorg3 (Supplementary Data 1e). The subsequent practice of family exogamy and caste endogamy among contemporary Coorgs (and throughout the Indian subcontinent)3 might be responsible for the substructure retained in the study cohort. However, all these groups eventually admixed genetically (Supplementary Data 1a-c, Figs. S10-12) which is in line with the socio-cultural history of the contemporary Coorgs (see Introduction). These admixture events are corroborated by chromopainter co-ancestry matrix and IBD sharing where Coorg2 showed some affinity to Coorg3 (Figs. S11-S12).

Recent consanguinity or high-level inbreeding as the cause of divergence patterns observed among Coorg1 and Coorg 3 was ruled out by RoH analysis (Fig. 3A-C). Higher RoH measures seen in Coorg1 comparable to those of old and divergent populations like Palliyar and Kallar and higher relative IBD scores of Coorg1 suggest founder event and endogamy and suggest its antiquity, compared to Coorg2 and Coorg3 (Fig. 3D) (Supplementary Data 1f). Analyses of uni-parental markers lend to this observation as well, wherein higher mtDNA diversity observed among Coorg2 and Coorg3 reflects the past expanding nature of these populations, whereas Coorg1, with its homogenous haplogroup distribution indicates a population bottleneck. High frequency of Y-chromosome haplogroup R1 in Coorg1 (Supplementary Data 2a-c) may also be a consequence of an identical founder event, suggesting that this group likely suffered similar drift like Palliyar due to long-term isolation (discussed below).

Genetic architecture of Coorg2 and Coorg3 in the light of their neighbouring, globally contemporary, and ancient populations draws attention. The mtDNA diversity of the three Coorg groups largely reflects their South Asian-specific maternal lineage along with some West Eurasian admixture (as evident by the presence of mtDNA haplogroups HV, H, U). However, Coorg2 showed moderate affinity to the sub-populations with ANI ancestry such as Nair, Bunt, Thiyya and Hoysala (Fig. 1B) reported to have comparably higher Middle Eastern components20; comparatively higher frequency of haplogroup R1 in Coorg2 (Supplementary Data 2a-c) suggesting their local Indo-European affinities, also reflected in the allele frequency-based analysis (Fig. S8). The presence of South Asian-specific haplogroup H in Coorg3 (frequency 0.45) is surprising considering their Indo-European genetic composition (Fig. 1C). A much earlier migration and settlement of Coorg3 (supported by earliest coalescent time of the R1a-M780 and R1a-Z2125 haplotypes) in South Asia compared to all other Indo-European migrations; and assimilation of local group(s) followed by typical South Asian ANI-ASI admixture event much earlier is plausible. After this major admixture event, Coorg3 witnessed long-term isolation followed by only very recent (~16 GBP), but minimal admixture with a northwest Indian group (Sikh_Jatt as a surrogate)—a representative of ANI-like population, corroborated both via admixture and demographic modelling. (Supplementary Data 1f). Northwest Indian groups had shown additional Middle-Eastern component along with Southwest Indian warrior clans (Nairs, Bunt and Thiyya)20. This second admixture event seems to be putative source for enhanced Middle-Eastern component in Coorg3 and later in Coorg2. Coorgs historically being a martial race lend further support to these events. Furthermore, all the qpAdm rotation models for Coorg3 best fitted only with additional source of Middle Eastern ancestry (Supplementary Data 1j). More ancient lineage and earlier separation history of Coorgs is further supported by M780 and Z2125 marker Y-STR networks and also earliest coalescent times for M780 and Z2125 haplotypes for Coorg3 (Fig S17) (Supplementary Data 2f), which were of independent Y chromosome lineage. They were not sharing Y chromosome haplotypes with any of the South Asian groups under M780 or Z2125 affiliations. These may have arrived in India much earlier and with no admixture event with any R1a lineage of India thereafter (Fig. S17).

At this juncture, despite inhabiting distinct and isolated regions of South India, it may be relevant to discuss the shared genetic ancestry of Coorg1 with Palliyar, a population group with negligible Steppe contribution11, evident in their sharing of clade in TreeMix analysis with similar amount of drift (Fig. S13) and also in fineSTRUCTURE tree (Fig. S10). Admixture modelling further indicates that admixture between Palliyar and Mala has given rise to C1 ~ 64 GBP (Supplementary Data 1d).

Population genomic studies with such a strong and unique population-specific drift as witnessed in Coorg3 have not been reported till date in the Indian context, except for the Kalash of Pakistan being reported as a genetic isolate21. Another such group is Gujjar, a small tribe from Kashmir, India, which clusters in PCA and admixture analysis with main Indian cline along with northwest India and Pakistan groups and has higher IBD scores22 and also reflected in our study. They also have higher RoH distribution exhibiting high level of endogamy and likely explaining their higher drift length in TreeMix. However, Coorg3 shows a notable genetic drift in PCA (Fig. 1B) and Admixture (Fig. 1C) but not in IBD (Fig. 5D) and RoH distribution (Fig. 5A-C). Additionally, there is no clue of any significant founder event or population bottleneck in this group (Fig. 3A-F), indicating that only possible cause of drift may be long-term isolation. This phenomenon is further validated by the very low migration rate parameter in this group as revealed by demographic modelling (Fig. 4). A population demonstrating such a drift may serve as a good resource to study evolutionary aspects in natural populations. Furthermore, Coorg1 captured here to be a sister clade of Palliyar, which is proposed as a proxy for ASI11 may also be useful.

Finally, this study presents the insightful evidence for an ancient origin and unique genetic architecture of the Coorg population. However, a recent study could not corroborate their oral history with the genetic architecture along with many other groups from South West India23. The three distinct genetically heterogeneous clusters date back to late Bronze Age. However, it is evident that a) this group had a much higher contribution of ancient Bronze-Age Middle Eastern ancestry; b) they had diverged and separated much before which stands in stark contrast to their Indian subcontinental neighbours; and c) an additional wave of admixture more recently from a source group related to contemporary northwest Indian groups. Coorg1 is an old population and an integral part of the present-day Coorgs. Considerable genetic drift observed in Coorg3 was not found among any Indian population on ANI-ASI cline till date, except Onge, who had history of long-time isolation and migration prior to major ANI-ASI admixture event. Furthermore, the negligible load of autosomal recessive disorders in this small (<0.3 million) population practising family exogamy but caste endogamy, even in the light of high IBD scores of Coorg1 similar to Finnish or Ashkenazi Jews, lends further support to a contribution of diverse gene pools from the three Coorg groups through an eventual admixture among themselves.

Taken together, the results of this study corroborate a model wherein Coorg1 (native) and Coorg3 (neighbouring, with recent local contribution from ANI-like Sikh Jatt population) are all late Bronze Age and were initially isolated spatio-temporally but eventually converged geographically and admixed genetically to give rise to Coorg2. This model, upheld by the extensive tools utilised in the study, explain the contemporary socio-cultural homogeneity of the current day Coorgs. Further, in case of Coorgs, cultural assimilation occurred much later than their genetic delineation into distinct groups.

Methods

Study subjects

Coorgs are a very small population group comprising <0.3 million individuals, belonging to approximately 1200 extended families (each identified with a unique family name) residing throughout Kodagu (Fig. 1A). As contemporary Coorgs (individuals recruited for this study) practice family exogamy and caste endogamy, only one member each from approximately 12% (n = 144) of the total families from across Kodagu were recruited. Furthermore, considering an overall historical prevalence of male-centric migrations and to capture both maternal and paternal lineages, only males (25–70 years of age) were recruited with informed written consent. Institutional ethical committee clearance was obtained prior to sample collection. DNA was isolated from ~5.0 ml of venous blood from each participant.

A geographical map showing the sampling location (Fig. 1A) in the Kodagu district in Karnataka, India was produced in R24 using the package ggplot225 and sf26.

Genotyping of samples and statistical analysis

Autosomal markers

A subset of the samples (n = 70) was genotyped using Affymetrix Axiom GW Human Origin Array for 633,994 SNPs as per the manufacturer’s specifications through a commercial facility (Imperial Life Sciences, Gurugram, Haryana).

Quality Control: The dataset was merged with published DNA dataset of contemporary Indian populations (Supplementary Table 1) after filtering for missingness using Plink 1.914, and only autosomal markers on 22 chromosomes having genotyping call rate >99% and minor allele frequency >1% were included. Dataset was further pruned by removing individuals with first-degree and second-degree relatedness utilizing KING-robust27 feature implemented in Plink214. After all filtering, final merged dataset comprised of 968 modern individuals genotyped at 405,962 SNPs.

In order to minimize the effect of background LD in PCA28 and ADMIXTURE29 like analysis, the markers were further thinned by removing SNPs in strong LD (r2 > 0.4, window of 200 SNPs, sliding window of 25 SNPs at a time) using Plink 1.914. For all the analysis with ancient DNA, Coorg samples were merged with west Eurasian autosomal DNA published datasets of 765 individuals with relevance to the incumbent sample set11,30,31,32,33,34,35,36,37,38,39. In this merged dataset, missingness criteria of geno >0.7 was applied to include only those individuals covered at at least 70% of sites resulting into 968 individuals covered at 442230 sites.

mtDNA markers

Mitochondrial DNA of all samples was PCR amplified using a set of 24 sets of primer40 followed by Sanger sequencing.

Y-chromosome markers

Genotyping of all samples for a total of 18 Y-chromosome binary markers to determine haplogroups was performed. PCR-amplified amplicons were sequenced using ABI 3730 automated Genetic Analyzer. Y-STR typing for 17 markers was done using ampFLSTRTM YfilerTM PCR amplification kit.

Statistics and reproducibility

Principal component analysis

Principal Component analysis (PCA) was performed on the merged dataset of modern Eurasian using the smartpca package implemented in EIGENSOFT 7.2.128 with default settings. The first two components were plotted to infer genetic variability.

ADMIXTURE analysis

Model-based clustering algorithm ADMIXTURE29 was run to infer ancestral genomic components in Coorg population inferred from the PCA performed. Cross validation was run 25 times for 12 ancestral clusters (K = 3 to K = 14) (Fig S1). Lowest CV error parameter was obtained at K = 7 and was used for downstream analysis.

Maximum likelihood tree construction

A maximum likelihood (ML) tree was constructed for the merged dataset comprising of modern South Asian populations and the Coorgs with TreeMix v.1.1212 using LD blocks of 500 SNPs grouped together and Onge as an outgroup.

Runs of homozygosity

Runs of Homozygosity (RoH) analysis was performed using PLINK v1.914 with three homozygous windows of 1000 kb, 2500 kb and 5000 kb with minimum 50 consecutive SNPs.

IBD score calculation and IBD sharing

IBD scores for the three Coorg groups relative to Finnish population was calculated with same pipeline as used in our earlier study22. SHAPEIT version 4.2.241 for phasing the genotype data and Refined-IBD tool42 for IBD detection were used. Then slightly modified R script from Nathan et al. 2017 was used for IBD score calculation. IBD sharing matrix was plotted using custom script in R. Historical effective population size history was determined using HapNe-LD method using genotype data.

F3-statistics

qp3Pop implementation of ADMIXTOOLS43 package was utilised to calculate outgroup F3 statistics. To infer gene flow from modern Eurasians in three groups of Coorg populations F3 statistics were used in the form of F3 (Yoruba; GroupA/GroupB/GroupC, X), where X is any modern west Eurasian or south Asian population. (GroupA=Coorg1; GroupB=Coorg2 and GroupC=Coorg3).

CHROMOPAINTER and FineStructure

Haplotype-based approach implemented in CHROMOPAINTER13 and FineStructure13 was used to derive co-ancestry matrix and fine scale population clustering, respectively. Data was first phased with SHAPEIT544 using default parameters, followed by CHROMOPAINTER run to infer co-ancestry matrix, first by performing 10 Expectation-Maximization (EM) iteration with 5 randomly selected chromosomes with a subset of individuals to infer global mutation rate (µ) and switch rate parameters (Ne). Then the main algorithm was run with 22 chromosomes with all the individuals to derive the co-ancestry matrix. This matrix was used by FineStructure to derive clustering using a probability model by applying Markov chain Monte Carlo (MCMC) procedure and then inferring hierarchical tree by merging all clusters with least change in posterior probability. For the run 500,000 burn-in iterations and 1,000,000 subsequent iterations were used, and the results stored from every 10,000th iteration. Admixture dates estimates and best admixture models were inferred using fastGlobeTrotter15 using Chromopainter chunklength files.

Proximal and distal modelling with ancient DNA

qpAdm in the ADMIXTOOLS 210 package in R was used to estimate proportions of ancient ancestral components in a test population (Coorg1/Coorg2/Coorg3) derived from a set of N source population groups having shared drift with a set of reference populations. Distal and Proximal modelling of admixture was performed using pre-Bronze Age aDNA source groups and Bronze Age proximal sources, respectively (Supplementary Table 2). In distal modelling AHG, Iran_N, Anatolia_N and EEHG were used as source groups, while Ethiopia_4500BP_published.SG, ANE, Shamanka_EN, PPNB, Dai.DG, Natufian, WSHG and WEHG as references. In proximal modelling AHG, Indus_Periphery and Steppe_MLBA were used as source groups and Ethiopia_4500BP_published.SG, Ganj_Dareh_N, EEHG, PPNB, Dai.DG, Anatolia_N, WEHG and Shamanka_EN were taken as references. Fitted admixture graph topology were obtained with qpGraph function of ADMIXTOOLS 210 using automated graph exploration with find_graph for three Coorg groups using modern and ancient Eurasians as reference. qpGraph was further used to model Coorg2 and Coorg3 along with Kalash and Gujjar as a mixture of ANI and ASI ancestry, using the model (YRI, (Coorg2/Coorg3/Kalash/Gujjar, (Georgians, ANI)), [(ASI, Onge])) proposed by Moorjani et al (17). This method was used earlier in Nakatsuka et al. (2017)22 to estimate the strength of founder effect in Indian populations by measuring post-admixture drift.

Demographic history of three Coorg groups using Demes and moments

To infer best fitted demographic model and model parameters we used parameter optimization method implemented in Moments17. For each of three groups (Coorg1, Coorg2 and Coorg3), we used preliminary model based on known south Asian genetic history and fastGlobeTrotter15 admixture models of Coorg groups. For model construction, we used Demes16 Python package. Parameter files were prepared based on respective Demes models. Two alternative models were used for comparison of demographic scenario of each of three Coorg groups. Site Frequency Spectrum were calculated from empirical data in VCF format as well as from Demes model specifications using Moments. Model parameter optimizations were performed using 300 iterations and using “lbfgsb” method. Confidence intervals for inferred parameters were calculated using moments.Demes.Inference.uncerts function of Moments.

Forward genetic simulation to test the effect of bottleneck history on the RoH distribution

Slim v3 was used to simulate three distinct population history models. Using forward genetic simulation we tested the effects of (A) no Bottleneck, (B) Coorg1 group-specific population bottleneck and (C) Coorg3 group-specific population bottleneck on the genome-wide RoH distribution. The time and strength of Coorg1 and Coorg3 models were replicated from the results of ASCEND based and IBDNe based methods.

mtDNA

Sequences were assembled with the reference sequence rCRS45 using AutoAssembler. Variations observed were used to assign the haplogroup using phylotree build 1746 and Haplogrep247.

Y-chromosome

Sequences were compared with reference to mark the variations and assign the haplogroups. In order to estimate population divergence, Weir-Cockerham’s Fst was measured using R package SambaR48 utilising the same pruned dataset as used in PCA and Admixture. Basic Quality Checks of data were done in the same R package. For the Y STR data analysis and TMRCA, the Y-LineageTracker49 tool was used.

Ethical approval

Informed written consent was taken from each participant. The project was carried out in agreement with the guidelines approved by the Institutional Ethical Committees of Centre for Cellular and Molecular Biology, Hyderabad, India. All experimental protocols were also approved by the Committee of the Centre for Cellular and Molecular Biology-CSIR, Hyderabad, India. All ethical regulations relevant to human research participants were followed and all the experimental methods comply with the Helsinki Declaration.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.