Genomic history of the Sardinian population

Abstract

The population of the Mediterranean island of Sardinia has made important contributions to genome-wide association studies of complex disease traits and, based on ancient DNA studies of mainland Europe, Sardinia is hypothesized to be a unique refuge for early Neolithic ancestry. To provide new insights on the genetic history of this flagship population, we analyzed 3,514 whole-genome sequenced individuals from Sardinia. Sardinian samples show elevated levels of shared ancestry with Basque individuals, especially samples from the more historically isolated regions of Sardinia. Our analysis also uniquely illuminates how levels of genetic similarity with mainland ancient DNA samples varies subtly across the island. Together, our results indicate that within-island substructure and sex-biased processes have substantially impacted the genetic history of Sardinia. These results give new insight into the demography of ancestral Sardinians and help further the understanding of sharing of disease risk alleles between Sardinia and mainland populations.

Figure 1:. Geographical map of Sardinia.

The provincial boundaries are given as black lines. The provinces are abbreviated as Cag (Cagliari), Cmp (Campidano), Car (Carbonia), Ori (Oristano), Sas (Sassari), Olb (Olbia-tempio), Nuo (Nuoro), and Ogl (Ogliastra). For sampled villages within Ogliastra, the names and abbreviations are indicated in colored boxes. Color corresponds to the color used in the PCA plot (Figure 2). The Gennargentu region referred to in the main text is the mountainous area shown in brown that is centered in western Ogliastra (Ogl) and southeastern Nuoro.

Figure 2:. Within-island population structure.

(A) Top two principal components of PCA based on 1,577 unrelated Sardinians. Each individual is labeled by geographical origin as defined by grandparental birthplaces (Methods); otherwise they are labeled with x or with ? for missing information. Subpopulations with less than 8 individuals are displayed in the background in grey color. (B) Admixture result at K = 2, which had the lowest cross-validation errors from K = 2 to K = 7 (not shown). Individuals from Ogliastra and outside of Ogliastra that were not assigned to a major location or had mixed grandparental origins are grouped under Ogl and non-Ogl, respectively. Bars for locations with individuals fewer than 40 (Bar, Gai, Loc, Olb, Tor, Vgs) were expanded to a fixed minimum width to aid visualization. (C) Genetic differentiation among Ogliastra villages (left) or among Sardinia provinces (right) as measured by Weir and Cockerham’s unbiased estimator of Fst. Ogliastra appears to be the most differentiated from other provinces and within Ogliastra the level of differentiation between villages is substantial (reaching as high as 0.0075 between Villagrande and Arzana), with Arzana being consistently well differentiated from other villages. (D) Estimated effective migration surface plot within Sardinia based on 181 Sardinians with all four grandparents born in the same location. Refer to Figure 1 for abbreviations of subpopulations.

Figure 3:. Allele-sharing across the island and within Ogliastra.

As a function of allele frequencies, allele-sharing across the island (top) and within Ogliastra (bottom) are shown. Allele-Sharing between a pair of population is defined as the ratios of the probability that two randomly drawn carriers of the allele of a given minor allele frequency are from different populations, normalized by the panmictic expectation, and visualized here by the color of the lines connecting two populations. Island-wide analysis used subpopulations with at least 70 individuals, and the minor allele counts of each of 1 million randomly selected variants were down-sampled to 140 chromosomes. Within Ogliastra analysis used subpopulations with at least 17 individuals, and the minor allele counts were down-sampled to 34 chromosomes.

Figure 4:. Population structure relative to mainland Europeans.

(A) Top two principal components of PCA of the merged dataset of Sardinia and Human Origins Array data. Populations are enclosed in dashed ellipses by major subcontinents. (B) Estimated effective migration surface result for the pan-Mediterranean analysis. (C) Admixture results at K = 4, which has the lowest cross-validation error in analysis from K = 2 to K = 15. For clarity, only populations with sample size > 8 are visualized. Arzana and Cagliari contained 100% and 89% of the European-dominant, “blue”, ancestry. Populations are ordered by sub-continental regions and then by population median values in PC1. See Figure S3 for the full result. Populations labels are color coded by major sub-continental regions. Abbreviations are: (North Africa) LibJ, Jewish in Libya; Egy, Egyptian; Moz, Mozabite; (Middle East) IrnJ, Jewish in Iran; Dru, Druze; Pal, Palestinian; Jor, Jordanian; Bed, Bedouin; (Caucasus) Abk, Abkhasian; Geo, Georgian; Arm, Armenian; Che, Chechen; Lez, Lezgin; Ady, Adygei; Bak, Balkar; OsN, North Ossetian; Nog, Nogai; (Turkey and Greece) Gre, Greece; Tur, Turkey; (Europe) Arz, Arzana; HGb, HGDP Sardinian; HGa, HGDP Sardinian; Cag, Cagliari; ItB, Bergamo; ItS, Sicilians; BasS, Spanish Basque; Spa, Spanish; BasF, French Basque; Fre, French; Est, Estonian; Lit, Lithuanian; Bel, Belarusian; Rus, Russian; Mrd, Mordovian; Ukr, Ukranian; Cze, Czech Republican; Hun, Hungarian; Cro, Croatian; Chu, Chuvash; Bul, Buglarian; Nor, Norwegian; Orc, Orcadian; GBR, British Great Britain.

Figure 5:. Coalescent-based inference of demographic history using SMC++.

(A) Inference of population divergence times and (B) population size history. Analysis based on 4 focal individuals and 90 low-coverage samples from the combined dataset of Lanusei and Arzana individuals (Ogliastra) and 1000 Genomes CEU and TSI. The uncertainty and mean point estimates of population divergence time are shown using 10 bootstrap samples (the violin plot and the black dot, respectively). We also show the point estimate using all of the data by the black cross. For population size trajectories, we estimated the size until 150 generations in the past and use 10 internal spline knots when running SMC++. Uncertainty reflected through 10 bootstrap samples are also shown in the same but lighter colors. The orange shaded box denotes the Neolithic period, approximately 4500 to 8000 years ago, converted to units of generations assuming 30 years per generation.

Figure 6:. Similarity of ancient samples to populations across Europe and within Sardinia.

(A) Outgroup f3 statistics of the form f3(Mbuti; Stuttgart, X), where X is a population across the merged dataset of Sardinia and Human Origins Array data. Higher f3 values suggest larger shared drift between a pair of populations. Arrow indicates Sardinia populations. (B) Outgroup f3 statistics of the form f3(Mbuti; Ancient, Sardinian) across Sardinian samples binned in steps of 5% of Gennargentu ancestry estimated in Figure 2B. The increase of outgroup f3 statistics as function of ancestry is positive for Stuttgart and Loschbour (0.00263 and 0.00274, respectively), and slightly negative for Yamnaya (−4.4×10⁻⁴). Ancient samples used include a reference Neolithic farmer individual (Stuttgart, orange), a reference pre-Neolithic hunter-gather individual, (Loschbour, blue), and a reference Steppe population(Yamnaya, green) from the merged dataset with Haak et al. (Methods). Error bars represent the s.e. of the estimated f3 values from blocked jackknife procedure. Sizes of the circle are proportional to the number of samples per bin (max N = 281 per bin).