The Counteracting Effects of Demography on Functional Genomic Variation: The Roma Paradigm

Abstract

Demographic history plays a major role in shaping the distribution of genomic variation. Yet the interaction between different demographic forces and their effects in the genomes is not fully resolved in human populations. Here we focus on the Roma population, the largest transnational ethnic minority in Europe. They have a South Asian origin and their demographic history is characterized by recent dispersals, multiple founder events and extensive gene flow from non-Roma groups. Through the analyses of new high-coverage whole exome sequences and genome-wide array data for 89 Iberian Roma individuals together with forward simulations, we show that founder effects have reduced their genetic diversity and proportion of rare variants, gene flow has counteracted the increase in mutational load, runs of homozygosity show ancestry-specific patterns of accumulation of deleterious homozygotes, and selection signals primarily derive from pre-admixture adaptation in the Roma population sources. The present study shows how two demographic forces, bottlenecks and admixture, act in opposite directions and have long-term balancing effects on the Roma genomes. Understanding how demography and gene flow shape the genome of an admixed population provides an opportunity to elucidate how genomic variation is modelled in human populations.

Admixture Has Shaped Romani Genetic Diversity in Clinically Relevant Variants

“…Regarding linkage disequilibrium, decay patterns are not statistically different between Roma and non-Roma ( Supplementary Figure 1 , p -value > 0.95). The total number of deleterious alleles per individual is similar among Roma and non-Roma groups ( Font-Porterias et al, 2021 ), and here we further show that the genomic distribution of accumulation of deleterious mutations (i.e., number of deleterious alleles per gene per individual) has comparable patterns between populations ( Supplementary Note 2 , Supplementary Figures 2 , 3 , and Supplementary Tables 1 , 2 ). Thus, the overall allele sharing and linkage disequilibrium patterns are comparable among populations.…”

“…RFMix v1.5.4 ( Maples et al, 2013 ) was run with one expectation-maximization (EM) iteration to infer the local ancestry of the phased haplotypes, using balanced reference panels representing European (IBS and TSI populations) and South Asian (PJL, GIH, and ITU populations) ancestries. As previously explained ( Font-Porterias et al, 2021 ), the Roma individuals included in the present study show, on average, 68.4% and 31.6% of European and South Asian global ancestry proportions, with a standard deviation of 7%. Ancestry was assigned when RFMix posterior probability was higher than 0.9, resulting in 96.3% of the variants with assigned ancestry.…”

“…In order to assess the genetic portability from non-Roma to Roma, we examined the allele sharing and linkage disequilibrium patterns between populations. We have previously shown that Roma exhibit a considerable amount of private variants; however, their proportion is lower than in other populations: 15,287 population-specific variants in Roma; 25,060 in IBS; 24,158 in TSI; 21,160 in PJL; 22,040 in GIH; and 24,070 in ITU ( Font-Porterias et al, 2021 ). In addition, allele sharing is high for common variants (MAF > 5%) (over 86% of Roma variants are present in non-Roma) ( Figure 1 ).…”

“…We used WES (mean depth of 54X), and genome-wide autosomal SNP data (Affymetrix Axiom Genome-Wide Human Origins 1 array) for Spanish Roma individuals (89 and 62 samples, respectively) ( Font-Porterias et al, 2021 ), deposited at EGA (EGAS00001004599). The Spanish Roma WES were merged with previously published non-Roma WES from 1000G (mean depth of 65.7X): Iberian Population in Spain (IBS), Toscani in Italia (TSI), Punjabi from Lahore (PJL), Indian Telugu from the United Kingdom (ITU), and Gujarati Indian from Houston (GIH) ( Auton et al, 2015 ), resulting in a dataset with 512 individuals and 410,225 variants.…”

“…Both datasets were then combined to increase the covered genomic variants, building a merged WES-array dataset with 487 individuals and 878,162 SNPs. Variant annotation was performed using the Variant Effect Predictor tool (VEP) from Ensembl ( McLaren et al, 2016 ) focusing on three deleterious prediction scores: PolyPhen-2 ( Adzhubei et al, 2010 ), GERP ( Davydov et al, 2010 ) and CADD ( Rentzsch et al, 2019 ), as previously explained ( Font-Porterias et al, 2021 ). For each analysis, the corresponding dataset used is specified (WES dataset or merged WES-array dataset).…”