Mutation rate estimates for 110 Y-chromosome STRs combining population and father-son pair data

Abstract

Y-chromosome microsatellites (Y-STRs) are typically used for kinship analysis and forensic identification, as well as for inferences on population history and evolution. All applications would greatly benefit from reliable locus-specific mutation rates, to improve forensic probability calculations and interpretations of diversity data. However, estimates of mutation rate from father-son transmissions are available for few loci and have large confidence intervals, because of the small number of meiosis usually observed. By contrast, population data exist for many more Y-STRs, holding unused information about their mutation rates. To incorporate single locus diversity information into Y-STR mutation rate estimation, we performed a meta-analysis using pedigree data for 80 loci and individual haplotypes for 110 loci, from 29 and 93 published studies, respectively. By means of logistic regression we found that relative genetic diversity, motif size and repeat structure explain the variance of observed rates of mutations from meiosis. This model allowed us to predict locus-specific mutation rates (mean predicted mutation rate 2.12 × 10(-3), SD=1.58 × 10(-3)), including estimates for 30 loci lacking meiosis observations and 41 with a previous estimate of zero. These estimates are more accurate than meiosis-based estimates when a small number of meiosis is available. We argue that our methodological approach, by taking into account locus diversity, could be also adapted to estimate population or lineage-specific mutation rates. Such adjusted estimates would represent valuable information for selecting the most reliable markers for a wide range of applications.

Figure 1

Mutation rate estimates (measured in mutations per generation) from meiosis for 80 Y-STR loci (points) and prediction from logistic regression for the eight categories of loci defined by motif size and repeat structure (lines). Continuous lines represent the predictions for loci with a simple repeat structure and dashed lines for complex loci. Thick black lines are used for the predictions of tetranucleotide loci, thick gray lines for hexa- loci, thin black lines for penta- loci and thin gray lines for tri- loci. The logistic regression model (Lμ=−6.863+0.539R̂_H−1.176M_tri+0.478M_tetra−1.130M_penta+0.236S_simple+error, see Supplementary Table S3 for coefficient P-values) gives the relationship between the logit of mutation rate (Lμ) and the predictive variables R̂_H (population relative mutation rate estimated using homozigosity), M (motif size: tri-, tetra-, penta- and hexanucleotide classes) and S (repeat structure: simple or complex). The model shows that Lμ increases with R̂_H and depends on repeat size (highest for tetranucleotide loci followed by hexa-, penta- and tri- in this order) and on the complexity of the loci (higher for simple than for complex loci).

Figure 2

Root of the relative mean squared error (RrelMSE) for mutation rate estimates calculated from meiosis data (filled circles) or from regression (open circles, triangles and squares) at each of 108 simulated loci. RrelMSE for estimates for loci mutating at true mutation rates of (a) 4 × 10⁻⁴ mutations per generation, (b) 8 × 10⁻⁴, (c) 1.6 × 10⁻³ and (d) 3.2 × 10⁻³. RrelMSE for regression-based estimates also depends on the category the loci belong to: ‘tri' (triangles), ‘tetra' (squares) or ‘penta' (open circles).