The germline mutational process in rhesus macaque and its implications for phylogenetic dating

Abstract

Background Understanding the rate and pattern of germline mutations is of fundamental importance for understanding evolutionary processes. Results Here we analyzed 19 parent-offspring trios of rhesus macaques (Macaca mulatta) at high sequencing coverage of ∼76× per individual and estimated a mean rate of 0.77 × 10−8de novo mutations per site per generation (95% CI: 0.69 × 10−8 to 0.85 × 10−8). By phasing 50% of the mutations to parental origins, we found that the mutation rate is positively correlated with the paternal age. The paternal lineage contributed a mean of 81% of the de novo mutations, with a trend of an increasing male contribution for older fathers. Approximately 3.5% of de novo mutations were shared between siblings, with no parental bias, suggesting that they arose from early development (postzygotic) stages. Finally, the divergence times between closely related primates calculated on the basis of the yearly mutation rate of rhesus macaque generally reconcile with divergence estimated with molecular clock methods, except for the Cercopithecoidea/Hominoidea molecular divergence dated at 58 Mya using our new estimate of the yearly mutation rate. Conclusions When compared to the traditional molecular clock methods, new estimated rates from pedigree samples can provide insights into the evolution of well-studied groups such as primates.

Mutationathon: towards standardization in estimates of pedigree-based germline mutation rates

“…However, it remains unclear if alternative library preparation and sequencing platforms introduce additional biases compared to standard Illumina protocols. Most pedigree-based studies of germline mutations have sequenced each individual to a depth between 30X and 50X (Besenbacher et al, 2019; Campbell et al, 2021; Jónsson et al, 2017; Kessler et al, 2020; Malinsky et al, 2018; Milholland et al, 2017; Sasani et al, 2019; Smeds et al, 2016; Thomas et al, 2018; Turner et al, 2017; Wang et al, 2020; Wu et al, 2020), three studies sequenced at a higher depth of ∼ 80X (Bergeron et al, 2021; Maretty et al, 2017) and 150X (Tatsumoto et al, 2017), while six studies sequenced at a depth lower than 25X on average (Harland et al, 2017; Koch et al, 2019; Lindsay et al, 2019; Martin et al, 2018; Pfeifer, 2017; Rahbari et al, 2016). A minimum coverage of 15X has been advised to call single nucleotide polymorphisms (SNPs) accurately (Fumagalli et al, 2013).…”

“…To find DNMs, we must first find where in the genome each of the short sequencing reads comes from. The Burrows-Wheeler Aligner (BWA; Li and Durbin, 2009) is an algorithm developed to map short reads (50-250 bp) to a reference genome and has been used in the majority of studies on direct mutation rate estimation (Bergeron et al, 2021; Besenbacher et al, 2019; Harland et al, 2017; Jónsson et al, 2017; Kessler et al, 2020; Koch et al, 2019; Malinsky et al, 2018; Maretty et al, 2017; Milholland et al, 2017; Pfeifer, 2017; Sasani et al, 2019; Smeds et al, 2016; Tatsumoto et al, 2017; Thomas et al, 2018; Turner et al, 2017; Wang et al, 2021b, 2020; Wu et al, 2020). In particular, the BWA-MEM algorithm is fast, accurate, and can be implemented with an insert size option to improve the matching of paired reads.…”

“…These parameters take into account the quality of a call at a given site (QD), the mapping quality (MQ), the strand bias (FS and SOR), the mapping quality bias between reference and alternative allele (MQRankSum), the position bias within reads (ReadPosRankSum); see Supplementary Table 2 for details on each filter). Although some studies followed these best practices (Jónsson et al, 2017; Wu et al, 2020), others implemented only a subset of filters (e.g., three studies reported the GATK filters without SOR > 3.0 (Koch et al, 2019; Thomas et al, 2018; Wang et al, 2020) and Besenbacher et al (2019) kept only four parameters -- FS, ReadPosRankSum, BaseQualityRankSum, and MQRankSum -- as they are calculated based on statistical tests following a known distribution) or readjusted the filtering thresholds based on previous results (e.g., Koch et al (2019) changed the ReadPosRankSum threshold from −8 to 15 while Bergeron et al (2021) changed the site filters by first using the advised parameters and then adjusting them to reduce the apparent false-positive calls). Several of the earlier studies implemented a different suite of site filters altogether (Pfeifer, 2017; Smeds et al, 2016; Tatsumoto et al, 2017).…”

“…Some studies only set a minimum DP of approximately 10 reads (e.g., Jónsson et al, 2017; Pfeifer, 2017; Sasani et al, 2019), while other higher coverage studies were able to set more conservative minimum and maximum thresholds, varying from a minimum of 10 to 20 to a maximum of 60 to 150 (e.g., Maretty et al, 2017: DP < 10 and DP > 150; Thomas et al, 2018: DP < 20 and DP > 60; Wang et al, 2020: DP < 20 and DP > 60). Another approach is to use a relative depth threshold for each individual (e.g., depth individual ± 3 σ , with σ being the standard deviation around the average depth (Tatsumoto et al, 2017), or a maximum threshold of 2 × depth individual (Besenbacher et al, 2019)) or, when all individuals were sequenced at a similar depth, an relative depth per trio (e.g., a DP filter of 0.5 × depth trio and 2 × depth trio (Bergeron et al, 2021)). Alternatively, Rahbari et al (2016) and Wu et al (2020) tested if the depth at each site followed a Poisson distribution under the null hypothesis that lambda was depth individual , and filtered away sites where at least one individual of the trio had a p-value higher than 2 × 10 −4 .…”

“…Of note, some studies refer to a false positive rate instead of the FDR (eg. Bergeron et al, 2021; Jónsson et al, 2017; Wang et al, 2020), yet, it also refers to the ratio of false-positive calls on the total number of candidates (i.e. true positives and false positives).…”