The genome wide SNP data of 456 Indians was obtained from the following publicly available sources: 102 samples from Telugu (South Asian Ancestry) ethnic group living in United Kingdom and 103 samples from Gujarati (South Asian Ancestry) ethnic group currently residing in Houston, Texas (The 1000 Genomes Project Consortium et al., 2015); 140 samples from 29 ethnic groups from Metspalu et al., 2011 study (Metspalu et al., 2011); 22 samples from 6 ethnic groups from Tätte et al. (2019) study (Tätte et al., 2019); 44 samples from 3 ethnic groups from Pathak et al. (2018) study (Pathak et al., 2018); 19 samples from Parsi ethnoreligious group from Chaubey et al. (2017) study (Chaubey et al., 2017); 26 samples from 10 ethnic groups from Chaubey et al. (2010) study (Supplementary Table S1.2). In total, we generated a combined dataset of 343,295 SNPs from a total of 1,159 subjects from Lithuania (N = 416), Latvia (N = 287), and India (N = 456).

For further analysis, to see a broader view in the context of other world populations, the generated pooled dataset was merged with 36 samples from Estonian (Tambets et al., 2018) and with 17 populations from 1000 Genomes project (The 1000 Genomes Project Consortium et al., 2015) (Supplementary Table S1.3). In total this dataset consisted of 2,883 individuals and 343,295 SNP.

Lithuanian, Latvian, and Indian dataset was also merged with a dataset of ancient and recent samples from Eurasia. Besides Lithuanians, Latvians and Indians, this dataset consisted of 406 ancient and recent samples from Lazaridis et al. (2016) study (Lazaridis et al., 2016), 17 recent samples from Behar et al. (2013) study (Behar et al., 2013), 54 ancient samples from Mathieson et al., (2018) study (Mathieson et al., 2018) and 11 ancient samples from Mittnik et al. (2018) study (Mittnik et al., 2018) (Supplementary Table S1.4). In total this dataset consisted of 1,647 individuals and 13,522 SNPs.

Datasets were merged and processed using plink v1.90, VCFtools v.01.16 and bcftools v.1.14 programs (Purcell et al., 2007; Danecek et al., 2011; Danecek et al., 2021).

Principal component analysis (PCA) was performed with SmartPCA package from EIGENSOFT v7.2.1 (Patterson et al., 2006) on the pruned SNPs. SNPs in linkage disequilibrium were removed with the indep-pairwise option of PLINK (v1.07) using a window size of 50 SNPs, a step size of 5, and a r2 threshold of 0.5.

Inbreeding coefficients (F) among individuals from Lithuanian, Latvian and Indian populations was measured using plink v1.90 program (Purcell et al., 2007). Negative F values were changed to zeros as they probably represent sampling errors. Kinship was measured using KING v.2.2.5 (Manichaikul et al., 2010). PCA outliers, individuals having F than that expected for second cousin mating offspring (F ≥ 0.0156) and second-degree relatives (kinship coefficient >0.0884) were omitted from all further analyses (in total 17 samples), except for Indian samples (Supplementary Table S2.1). PCA results were plotted using R v4.3.1 software ggplot2 package (Wickham, 2016).

Assessment of each individual genetic ancestry was performed with ADMIXTURE v1.3.0 (Alexander et al., 2009). In analysis we included both present-day and ancient samples with K from 2 to 12 with 10 iterations. The results were plotted using the PONG tool (Behr et al., 2016).

Outgroup f3-statistics were computed using qp3Pop program from ADMIXTOOLS V4.1 (Patterson et al., 2012). Mbuti was considered as outgroup in the analysis and calculated the shared drift between each putative ancient group and all the modern groups in the dataset in the form (Mbuti; Ancient, Modern).

The long-term Ne and divergence time for the populations under study were estimated using R Package NeON (Neon, 2015) based on the genetic distance between SNPs. For the analysis we used binary PLINK files and updated genetic map information of the markers to calculate the Ne over time. First, the squared correlation coefficient of linkage disequilibrium (r2LD) within predefined recombination distance categories between markers is estimated. We used a function that generates 250 overlapping recombination distance categories with a step of 0.001 centiMorgan (cM) from 0.005 to 0.25. Then, Ne with a confidence interval 95% for each recombination distance category applying the formula Ne ≈ 1/(4c) * [(1/r2)-2] is calculated (c is the distance between markers in Morgan). The long-term Ne was estimated as the harmonic mean of the effective population size along the generations in the past for each population (Wright, 1931). Knowing the values of Ne and having the matrix of the calculated pairwise F _ST values with 4P software (Benazzo et al., 2015), we could estimate the time of divergence between populations using the Tdverg function of the NeON R package. Divergence time between pairs of study populations in generations was estimated as follows: T = ln(1 − F _ST )/ln(1–1/2Ne), where T represents divergence time. A generation is assumed to be 25 years long. The evolutionary history based on estimated divergence times was inferred using the UPGMA method implemented in MEGA X software v.10.2.5 (Kumar et al., 2018).

Principal component analysis (PCA) was performed on genome wide 343,295 SNP data of 1,142 individuals, which were distinguished by the country of origin (Lithuania, Latvia, and India). We revealed 2 Latvian and 5 Indian populations outliers (Supplementary Figure S2.1), which were subsequently removed from further analyses.

The Indian population was classified based on the families of languages they mostly speak (Indo-European, Dravidian, Austroasiatic, and Sino-Tibetan). This classification was chosen because it provides information about geographical location of the samples (Supplementary Table S1.2). Samples from the Lithuanian and Latvian populations were distinguished solely by their country of origin, as they exhibited a high level of homogeneity and formed a single cluster. This approach was adopted in order to accurately represent the diversity within the Indian population and to ensure that meaningful conclusions can be drawn from the analysis (Supplementary Figure S2.1).

The first two principal components explained 45,49% of genetic variance among populations and showed that individuals from Lithuanian and Latvian populations form one cluster, while individuals from the Indian population form a distinct group. Also, as quite expected, the Indian population had a much bigger data dispersion in the plot when compared to the Balts. This shows a higher genetic diversity of the Indian population due to a complex interplay of historical migrations, long-term population structure, and social factors like the caste system and linguistic diversity. The samples of Indian speakers of Indo-European languages showed a closer allocation to the Balts compared to other linguistic groups of Indians. The most distant from the Balts are samples which belong to the speakers of Sino-Tibetan languages (Figure 1).

Further PCA was performed on the Lithuanian, Latvian, Indian dataset merged with other world populations from 1000 Genomes project. In total it included 2,859 individuals and 343,295 SNP. Individuals from the Indian population were again distinguished based on the family of their main language to get a more accurate view and see which linguistic group from this large diverse population has a greater resemblance to the Balts. The first principal component explained 57,2% of genetic variance among populations, the second principal component – 26,97% (Figure 2). The first principal component distinguishes ESN, GWD, LWK, YRI and MSL populations (Africa) from other analyzed populations. The second principal component distinguishes KHV, JPT, CHB, CHS populations (East Asia) from populations of Europe and South Asia. Lithuania and Latvia together form a distinct cluster but are closer to the Indian populations, GIH, PJL, STU, BEB (South Asia) and Estonians (Europe). Close to these clusters another group is found of TSI, FIN, CEU and GBR populations (Europe). The Indian population has the biggest data dispersion in the plot when compared to other populations. Indo-European and Dravidian speaking individuals cluster closer to the Baltic populations than Sino-Tibetan or Austroasiatic speaking individuals.

Principal component analysis was conducted on the dataset by merging the samples from the Baltic and Indian populations with those from other ancient and modern populations across the Eurasian continent (as shown in Supplementary Figure S1.4). The purpose of this analysis was to determine whether ancient populations, particularly those from Anatolian and Steppe regions, exhibit a closer genetic relationship to Indo-European speaking populations. This investigation also aimed to evaluate the hypotheses of the homeland of the Indo-European languages discussed in the Introduction. Some modern populations from the European continent were included in the analysis with the objective to investigate whether the speakers of the Baltic languages share a stronger genetic connection with the Indian population compared to other neighboring European populations. This dataset consisted of 1,628 samples and 13,522 SNPs. Indian ethnolinguistic groups which were assigned to Austroasiatic or Sino-Tibetan languages families were removed from analysis since they have been shown as clearly distant from the Baltic populations in the previous analysis (Figure 1). After filtering, the dataset consisted of 1,569 samples and 10,208 SNPs. First principal component analysis revealed 1 Levantian Neolithic and 2 Iranian Neolithic outliers which were removed from the further analysis (Supplementary Figure S2.4). Ancient samples were assigned to populations as in Supplementary section table 1.4., and individuals from modern populations were distinguished based on the language branches and subbranches, the Indo-European family is represented by Baltic (Lithuanian and Latvian populations), East Slavic (Belarusian, Russian, Ukrainian populations), West Slavic (Polish population), North Germanic (Icelandic, Norwegian populations), West Germanic (English and Scottish populations), the Uralic family is represented by Finnic (Estonian and Finnish populations), and Mordvin (Mordovian population). The first principal component explained 2.6% of genetic variance and distinguished the Indian populations from the analyzed modern Eurasian populations. However, Indo-European speaking Indians clustered closer to the Eurasian populations compared to the Dravidian speaking Indians. Balts display a tight cluster partly overlapping the West Slavic, and North Germanic groups with close proximity towards one of the extremes of the Finnic population clusters (Figure 3). Compared to ancient samples, modern Balts and Finns positioned as the closest present-day European populations to ancient hunter-gatherer groups (Latvian HG, Western HG, Eastern HG, Scandinavian HG), neolithic ancient samples from Lithuania, Estonia and Latvia, also Ukrainian Mesolithic and neolithic samples, and showed high proximity to the Steppe and late Neolithic Bronze Age European samples (Figure 3). The Balts did not show a closer genetic relationship with Indians compared to other Eurasian populations (Figure 3).

The dataset of ancient and modern samples from Eurasian continent was used, with previously mentioned outliers and Indian populations speaking Austroasiatic or Sino-Tibetan languages removed. This dataset consisted of 1,569 samples and 10,208 SNPs. We included 17 modern populations (Lithuanian, Latvian, Belarusian, Ukrainian, Russian, Polish, English, Scottish, Orcadian, Icelandic, Norwegian, Iranian, Indian (Indo-European), Indian (Dravidian), Mordovian, Estonian, Finnish (Uralic)) and a group of 34 ancient populations. The best CV value of admixture analysis was determined when populations were distributed into 4 clusters (Supplementary Table S2.2).

Admixture analysis revealed that the Baltic population is characterized by three main genetic components reflecting Hunter Gatherers ancestry (in green), other components found in the European Neolithic/Anatolian Neolithic (in purple) and in the Iranian, Lithuanian Mesolithic, Lithuanian EMN Narva (red). Indians are also characterized by three genetic components–blue found maximized in Lithuanian and Estonian EMN Narva, Lithuanian Late Neolithic, and Estonian Middle Neolithic; red found in the Iranian, and purple which is observed only in one part of Indo-European speaking Indians and found in the European Neolithic/Anatolian Neolithic samples. All modern European populations are mainly characterized by three color compounds as well as the Balts. Some of them, however, have a small part of purple color (West Slavs). Out of modern European populations, the Polish population has the biggest part of the blue color component, which is characteristic to Indians. From the ancient populations, Lithuanian Early-Middle Neolithic Narva samples have the biggest amount of blue color (Supplementary Figures S3, S4). The full admixture analysis is provided in Supplementary Figure S2.8, S2.9.

Subsequently we used outgroup-f3 statistics (Patterson et al., 2012) to measure shared ancestry between Baltic and ancient populations from different geographic regions. Among the modern populations of the Lazaridis et al., (2016) dataset (Lazaridis et al., 2016), Balts showed the highest f3 values when testing Western HG, Scandinavian and Eastern HG, Latvian HG, Latvia middle Neolithic, Lithuania early middle Neolithic and late Neolithic, Ukraine Mesolithic and Neolithic, Steppe Eneolithic, and EMBA steppe pastoralists, but not for the Lithuania Mesolithic, and European Neolithic farmer component (Figure 4). No significant allele sharing was detected between the Balts and the Anatolian Neolithic farmers.

To reconstruct the evolutionary relationships among the Baltic, Indian (speaking Indo-European and Dravidian languages), and modern populations from Europe, effective population size (Ne) of each population (Supplementary Table S2, S3; Supplementary Figures S2.10, S2.11) and divergence times (Supplementary Table S2.5) among them were estimated. The estimated long-term Ne, calculated as the harmonic mean, was about 5,000 for Balts, with confidence intervals (CI) [4,672; 5,258]. Comparing the Ne values for populations analyzed we observed a variation from ≈10,000 in the Indian populations to 3,987 in Finnish population (Supplementary Figure S2.3). The constructed phylogenetic tree displays two main long branches which distinguish Indians from other populations in the European continent. This indicates an earlier separation that occurred many years ago. Divergence time analysis showed that Indian population separated from other European continent populations approximately 11,569 years ago on average (Supplementary Table S2.5). The branch representing populations from the European continent further divides into two smaller main branches, which can be partly explained by the language families and their branches and partly by the geographic proximity of the populations. One branch comprises Indo-European speakers of the Baltic (Lithuanian and Latvian), the West Slavic (Polish), and the East Slavic (Russian, Belarusian, Ukrainian), which belong to the Balto-Slavic branch, but it also includes speakers of Uralic languages from the Finnic (Estonian and Finnish) and the Mordvin (Mordovian) branches. The speakers of different subbranches and even families are notably intermixed here due to the areal proximity: Polish (West Slavic) shares a branch with Ukrainian (East Slavic), Lithuanian and Latvian (Indo-European) share a branch with Estonian (Uralic), and Russian (Indo-European) shares a branch with Mordovian (Uralic). The other branch forms a group of Germanic-speaking populations, where speakers of West Germanic (English, Scottish, Orcadian) and North Germanic (Icelandic, Norwegian) are intermixed. Within the first branch group, the Finnish population is initially separated from the remaining populations. Subsequently, the Mordovian and Russian populations are distinguished from the rest. Further smaller branches then separate into a group consisting of Estonian, Latvian, and Lithuanian populations and another group comprising Polish, Belarusian, and Ukrainian populations. The most recent separation occurred between Lithuanian and Latvian populations and between Polish and Ukrainian populations. The Scottish population splits from the rest in the Germanic branch first. Following that, the Orcadian and Norwegian populations are distinguished, and the most recent separation occurred between English and Icelandic populations (Figure 5).