Materials
The V660L polymorphism sits on a 56.2-kb-long haplotype ( r 2 >0.8 in all 1000G individuals) defined by 28 private SNPs on the Neandertal lineage (i.e., the Neandertal allele is missing in 108 Yoruba individuals) with coordinates chr11:100877202–100933412 (hg19). Inserting this length into the formula derived by Huerta-Sánchez et al. (2018) yields a probability of P = 0.02 for incomplete ancestral lineage sorting (ILS), using a generation time of 25 years, the local recombination rate of 0.87 cM/Mb (deCODE; Kong et al. 2010 ), and a branch length of 200 ky for the modern human branch, and 100 ka for the Neandertal branch. We use a conservative lower estimate of the branch lengths because the formula underestimates the probability of ILS if the branch lengths are overestimated. Previous genome-wide analyses ( Sankararaman et al. 2014 ; Danneman and Kelso 2017 ; Li et al. 2018 ) have similarly indicated this locus as carrying introgressed Neandertal haplotypes.
If an insertion is not present in the reference genome, the read depth should drop (to zero if homozygous) in a symmetrical fashion around the position of insertion. This pattern is observed for the Chagyrskaya Neandertal genome. A smaller drop is seen for the Altai Neandertal genome. We realigned the sequenced fragments to a reference sequence containing the Alu insertion. A sequence 11 bp upstream to the Alu insertion and 11 bp into the Alu insertion (that is unique in the entire NCBI database) was used to identify junction fragments.
We find that V660L is not in particularly high linkage disequilibrium with the Alu insertion ( r 2 = 0.72). To rule out sequencing error, we verified this in two of the 24 high-coverage 1000G genomes, which carried V660L but not the Alu insertion (NA19625 and HG01500). Linkage disequilibrium between V660L and the Alu element for all 1000G subpopulations is given supplementary table S1 , Supplementary Material online.
We investigated phenotypic associations in the UK Biobank using the GeneAtlas tool ( Canela-Xandri et al. 2018 ). In brief, the Gene ATLAS provides 778 associations based on Mixed Linear Models using 452,264 Britons with European descent. As fixed effects, the model includes sex, array batch, UK Biobank Assessment Center, age, and 20 genomic principal components. Population structure was captured as a random effect. Both V660L (rs1042838) and S344T (rs3740753) are included in UK Biobank SNP array, so no imputation was needed for these. However, the Alu insertion is not genotyped and our analyses are thus limited to the missense variants of the haplotype. P -values were adjusted for multiple comparisons by controlling for the family-wise error rate ( supplementary table S3 , Supplementary Material online). Confidence intervals for the odds ratios were calculated using the method described in Altman and Bland (2011) . Of the 778 association in the GeneAtlas, 22 where classified as belonging to ICD chapter XV (Pregnancy, childbirth, and the puerperium, supplementary table S3 , Supplementary Material online). In addition, we identified 53 traits, which we considered to be related to the female reproductive system. For V660L we found negative correlations with “hemorrhage in early pregnancy” ( P = 0.002), miscarriage ( P = 0.01), and positive correlation with more sisters ( P = 0.0036). This result was replicated for S344T ( P = 0.0006, P = 0.028, and P = 0.00097, respectively). P -values and odds ratios for V660L are shown in figure 3 and supplementary table S3 , Supplementary Material online.
We analyzed the age of menarche and menopause using the Biobank Japan ( Horikoshi et al. 2018 ). Among 43,861 Japanese women, the Neandertal haplotype is associated with a somewhat earlier menopause (rs585447, 0.19 years per allele, P = 0.048) whereas there is no effect on menarche. P -values and effect sizes were taken directly from Biobank Japan.
We analyzed the effect of V660L (rs1042838) on PGR mRNA expression levels using the version eigth release of the Genotype-Tissues Expression (GTEx) project. Tissues which were predicted to show an effect were selected based on a posterior probability (i.e., m -value) >0.9. The effect across tissues was estimated using Han and Eskin’s Random Effects model (RE2; Han and Eskin 2012 ).
Supplementary Material
Click here for additional data file.