Mutation is the ultimate source of all genetic variation, and has been a long-standing research focus in evolutionary genetics. Until recently, estimating mutation rates has necessitated using reporter loci (Drake 1991) and it has not been possible to obtain direct estimates of the mutation spectrum, which is the relative frequency of different nucleotide substitutions, insertions/deletions (indels), and rearrangements. However, with current methods, sequencing massive amounts of DNA to discover the relatively few newly arising, spontaneous mutations at the genome level is now affordable. This has allowed mutation accumulation (MA) studies (Lynch et al. 1999) to be used to estimate both the genome-wide mutation rate and its spectrum in a variety of organisms: Arabidopsis thaliana (Ossowski et al. 2010; Jiang et al. 2014), Bacillus subtilis (Sung et al. 2015), Burkholderia cenocepacia (Dillon et al. 2015), Caenorhabditis elegans (Denver et al. 2009; Meier et al. 2014), Chlamydomonas reinhardtii (Ness et al. 2012; Sung et al. 2012), Daphnia pulex (Keith et al. 2015), Dictyostelium dicsoideum (Saxer et al. 2012), Dienococcus radiodurans (Long et al. 2015), Drosophila melanogaster (Schrider et al. 2013; Keightley et al. 2014), Escherichia coli (Lee et al. 2012), Heliconius melpomene (Keightley et al. 2015), Mesoplasma florum (Sung et al. 2012), Paramecium tetraurelia (Sung et al. 2012), Pristionchus pacificus (Weller et al. 2014), Pseudomonas aeruginosa (Dettman et al. 2016), Pseudomonas fluorescens (Long et al. 2015), Saccharomyces cerevisiae (Lynch et al. 2008; Nishant et al. 2010; Zhu et al. 2014), Schizosaccromyces pombe (Farlow et al. 2015; Behringer and Hall 2016), and Tetrahymena thermophila (Long et al. 2015). As the list of species for which genome-wide estimates or mutation rate and spectrum are available increases, the question of repeatability becomes paramount.

Genome-wide estimates of parameters of mutation have been performed in the same species only a few times, either as controls for an experimental evolution project or as a specific effort to understand the spontaneous mutation rate, so the reproducibility of results has seldom been examined. The five species with two independent genome-wide estimates reported before 2015 are included in Table 1, which indicates that they often suffer from low power due to small sample size, making it difficult to evaluate reproducibility.

Table 1 Comparison of repeated, genome-wide estimates of single nucleotide (μ SNM) and small (≤50 bp) insertion/deletion (μ indel) mutation rates across species obtained from mutation accumulation (MA) experiments
Full size table

Our recent mutation accumulation study in Schizosaccharomyces pombe (Behringer and Hall 2016) unintentionally overlapped a concurrent study of similar scope (Farlow et al. 2015) and allowed a statistically robust comparison of the repeatability of mutation rate and spectrum estimates in this species. The conditions of the two experiments differed with respect to starting strain, culture temperature, time between line transfers and growth medium, but were otherwise similar.

Genome-wide mutation rates vary across studies. In the five species with data prior to 2015, the single nucleotide mutation rate estimates vary 1.34–3.50-fold (Table 1). In comparison, the single nucleotide mutation rate estimates in the two S. pombe studies are only 1.18-fold different. This difference, though smaller than seen in the other five species, is statistically significant (p < 0.05). In previous studies, the mutation rate for small (≤50 bp) indels varied from 1.10–3.00-fold, though none of the differences were significant because of the low number of observed mutational events (Table 1). In comparison, the indel mutation rate estimates in the two S. pombe studies are 2.83-fold different, which is a highly statistically significant difference (p < 10−5).

There are at least three possible explanations for the threefold difference observed in indel rates in S. pombe. First, indels are more challenging to accurately detect bioinformatically than base substitutions, not only because the resulting mismatches can make an indel-containing sequencing read more difficult to map, but also because indels commonly occur within microsatellites and highly repetitive regions which already have an increased PCR and sequencing error rate (Jiang et al. 2015; Narzisi et al. 2014). Two different pipelines were used for small indel detection in the two studies. In Farlow et al., sequencing reads were mapped with BWA (Li and Durbin 2009), and realigned with both Breakdancer (Chen et al. 2009) and Pindel (Ye et al. 2009). In our study, we also mapped sequencing reads with BWA, but realigned them with GATK’s IndelRealigner (Auwera et al. 2013). Both practices are not without their issues; GATK is less sensitive when it comes to larger indels, while Pindel has trouble with insertions (Narzisi et al. 2014). In our study, we estimated the false-positive and false-negative error rates and found them both to be less than 2.5 %. Even if error rates were an order of magnitude greater in the Farlow et al. study, they would be insufficient to explain the difference in indel mutation rate.

Second, the indel mutation rate may be altered by the different environmental conditions of the two experiments, with temperature being a likely candidate (Matsuba et al. 2013). Stressful temperatures have been demonstrated to affect microsatellite mutation rate in Caenorhabditis elegans (Matsuba et al. 2013), with increased temperatures leading to an increase in mutation rate. However, if S. pombe also exhibits increased indel mutation rate at stressful temperatures, we would expect the Farlow et al. study to have a higher rate, since they performed their mutation accumulation experiment at a higher, presumably more stressful, temperature (32° vs 30 °C). There is no obvious other difference that would suggest that S. pombe was more or less stressed in one versus the other experiment.

Third, differences in genetic background may explain the different indel mutation rates. To examine this hypothesis, we compared the ancestral strains in the two studies, specifically to ask whether ours exhibited a higher relative number of indels when compared to the reference. In our MA study, we found 315 indel and single nucleotide substitution differences between the ancestor and reference genome, which was a surprisingly large number given that both isolates have the same strain designation (972 h-). A comparison of our ancestor with the one used by Farlow et al., identified 208 shared differences. This high number of shared differences between the ancestors strongly suggests that they represent errors in the reference assembly. The reference genome is thought to have at least 190 errors (Hu et al. 2015; Wood et al. 2011), 183 of which are among the 208 shared mutations, and thus confirmed by our analysis. The remaining 25 have not been previously inferred (Supplemental Table). Of the remaining 107 differences between our ancestor and the reference, there are approximately 3.5-fold more indels than single nucleotide mutations. A similar analysis of Farlow et al.’s ancestor indicates 0.95-fold more indels than single nucleotide mutations, relative to the reference. Thus, our ancestor shows a 3.7-fold higher number of indels, relative to single nucleotide changes, compared to the Farlow et al. ancestor. This suggests that there may indeed be a genetic background difference between the two strains that is causing a relatively higher indel mutation rate in our ancestral strain. We note that there is no evidence for selection having played a major role in the mutational differences between either ancestor: the effects of mutational differences between the ancestors and the reference are not significantly different from those that arose during MA, when selection is known to be ineffective (Fig. 1).

Fig. 1
figure 1

Comparison of mutational effects in MA ancestor and MA lines. There is no difference between any of the three spectra (Chi-squared test: p value = 0.61)

Full size image

When examining differences between the ancestors for their potential to cause differences in indel rate, two mutations were found in genes associated with DNA repair in the Farlow et al. ancestor. One of these is in rev7 (SPBC12D12.09), which is a subunit of DNA polymerase zeta with inferred involvement in translesion synthesis (Kosarek et al. 2008), but the mutation is synonymous and thus not likely to have an effect, unless it alters mRNA stability and thus protein levels in the cell. The second is a missense mutation in cdc6 (SPBC336.04), also known as POLD3, which is a subunit of DNA polymerase delta. A mutation in cdc6, specifically cdc6-121 has known mutator qualities, while another variant, cdc6-23, may reduce the mutation rate relative to wild type (Liu et al. 1999). It is possible that the cdc6 missense mutation in the Farlow ancestor has anti-mutator qualities, which could account for the indel differences between the two ancestors and the estimated rate differences in the two MA experiments.

Regardless of whether it is environmental or due to genetic background, the substantial variation seen in indel rate has serious implications for the use of microsatellite repeats as genetic markers. In nature, populations that appear to be significantly different in their microsatellite genotypes and are thus inferred to be genetically isolated from others may simply have higher mutation rates. Microsatellite mutation models, such as the stepwise-mutation model (Slatkin 1995), assume uniformity in the indel rate within species. While warnings have been issued about the robustness of microsatellites due to differing mutation rates amongst loci (Putman and Carbone 2014), microsatellites may also miscalculate genetic distance because of differing mutation rates amongst populations.

Other parameters of the mutational process, which require large numbers of mutations to estimate with precision, including the spectrum of single nucleotide mutations, the insertion to deletion ratio for small indels, and the location of mutational hotspots do not differ between the two S. pombe studies. Three of the previous studies in other species have sufficient numbers of single nucleotide mutations to estimate the spectrum for this class of mutations (Fig. 2). In these three species, and in S. pombe, the spectra are remarkably similar to one another. The relative frequency of insertions versus deletions in small indels across studies within species can only be compared within S. pombe; there are insufficient numbers of indels in other studies. The ratio of insertions to deletions is not significantly different in the two S. pombe studies; it is 5.88 in Farlow et al. and 6.12 in our study. Further, the ratio of insertions to deletions is similar, 6.4, for those indels that differ between our ancestor and the reference, after removing those that are shared with the Farlow et al. ancestor. It is only in the Farlow ancestor, in which the ratio is 0.82, based on 20 indels, that we find a significant difference (Fisher’s exact test, p ≤ 0.0003) from the others we have observed.

Fig. 2
figure 2

Comparison of mutation accumulation spectra for the four species in which there are sufficient numbers of mutations to make a comparison. In all four species, spectra are not significantly different across studies (Chi-squared test: a A. thaliana, p value = 0.75, b D. melanogaster, p value = 0.58, c C. elegans, p value = 0.164, d S. pombe, p value = 0.66)

Full size image

In conclusion, the inadvertent overlap of our S. pombe MA experiment with that of Farlow et al. allowed one of the first statistically robust comparisons of estimates of parameters of mutation within a species. Generally, estimates revealed remarkable repeatability. The single nucleotide mutation rates, though statistically significantly different, were within 20 % of one another, and the mutational spectrum for these mutations was not different. Further the relative occurrence of insertions to deletion was also not different across the two studies. The only substantial difference was the indel mutation rate, which varies by threefold across the two studies and is highly statistically significant. This suggests that the mutation rate for indels may be more sensitive to genetic background, environment, or both.