This study extends the past finding that S. paradoxus
shows rapid evolution in five centromeres [5
] to all centromeres of this species and to a second species with point centromeres, S. cerevisiae
(Figure , Tables and , Figures S1 and S2). S. cerevisiae
is an especially useful species in which to study the cause of rapid centromere evolution because its point centromeres are easy to sequence and patterns of recombination and mutation within its genome are exceptionally well characterised [13
]. The population and comparative analyses of centromere DNA sequences presented here lead to the conclusion that rapid centromere evolution is caused by a generalised increase in the mutation rate and not by meiotic drive or recombination.
Under the meiotic drive theory of rapid centromere evolution, we expect low polymorphism within centromeres. Point centromeres, the one type of centromere where this is easily and accurately assayed, show the opposite pattern; high diversity within species or populations (Table , Table ). This is the case for at least three populations of S. cerevisiae
(Table ) and one population of S. paradoxus
(Table , [5
]). Whether levels of divergence or polymorphism are measured, CDEII seems to evolve 3 or 4 times faster than selectively unconstrained DNA (Tables and ). High levels of polymorphism suggest that rapid centromere evolution is not the result of the sweeps of natural selection predicted by meiotic drive. The exceptionally high levels of polymorphism seen here, and the constant level at which centromere polymorphism or divergence is increased relative to other parts of the genome, are expected if high mutation at centromeres causes their rapid evolution.
The proteins that bind to animal and plant centromeres also contain rapidly evolving regions, and this could be because of positive selection for mutations that suppress meiotic drive of centromeres during female meiosis [1
]. In contrast, there is no such evidence of positive selection in the centromere binding proteins of yeast [27
], and this is consistent with a high mutation rate as a mechanism for rapid centromere evolution in yeast. If there is no evidence for compensatory mutations in yeast centromere binding proteins, then perhaps this implies that the rapid divergence of CDEII sequences has no functional consequences. Experimental replacement of CDEII sequences with random sequence of the same length and base composition does however appear to increase rates of segregation distortion in S. cerevisiae
]. Therefore it is possible that the high mutation rate at yeast centromeres has functional consequences, but these could only contribute to yeast speciation under a simple Dobzhansky-Muller model [28
]: centromeres diverge so that they are no longer bound by their associated binding proteins, as opposed to a meiotic drive model for speciation [1
] in which meiotic drive at centromeres and its suppression by centromere binding proteins leads to post-zygotic reproductive isolation.
Recent evidence suggests that gene conversion at centromeres could contribute to rapid centromere evolution in maize [10
], leading to the proposal that this force could generally explain rapid centromere evolution in eukaryotes [10
]. The findings of the study in maize came as a surprise, since it has long been thought that recombination is suppressed at centromeres [29
], and this has been confirmed in yeast [30
] and other species [11
]). Using population data in yeast and some of the same approaches used for maize [10
], I find evidence for recombination very close to centromeres though not within them (Table , see Additional file 1
, Table S2, Table S3). A number of recombination events may have occurred close to CEN2 (Table ), where a double-stranded break hotspot is also predicted [25
]. High-resolution genome-wide mapping of the crossover and non-crossover events from a large number of meioses in S. cerevisiae
also shows that crossovers sometimes occur close to centromeres, but not within them, and that gene conversion does not occur close to centromeres at all [26
]. The absence of a detectable signature of recombination events within centromeres, together with the lack of an obvious mechanism by which gene conversion would increase diversity in non-repetitive point centromeres, suggests that gene conversion does not lead to rapid centromere evolution, at least in the way proposed in maize.
Gene conversion is mutagenic [13
], so even if the signatures of gene conversion have been obscured in yeast centromeres, perhaps their high mutation rate does result from high rates of gene conversion as a result of this mutagenicity, if not as a result of the products of recombination. Analysis of the mutation spectrum at S. cerevisiae
centromeres, suggests this too is not the case. The mutation spectrum in CDEII is more like that seen genome-wide and in wild-type strains in studies of spontaneous mutation than it is like the spectrum expected specifically under gene conversion (Table ). Thus the rapid evolution of yeast centromeres may not rely on the action of a specific DNA repair system like that involved in gene conversion.
In summary, it appears that a generalised increase in the mutation rate is responsible for the rapid evolution seen at point centromeres, and this is not the result of gene conversion as recently proposed [10
]. Given that rapid centromere evolution occurs in a broad range of eukaryotes [3
], it is possible that high mutation rates could also contribute to the rapid evolution of other eukaryotic centromeres.
Apart from rapid evolution, another general characteristic of centromeres is that their DNA is wound round a histone specific to centromeres, CENH3 [8
]. This leads to a nucleosome structure that is fundamentally different at centromeres compared to other parts of the genome [8
]. There is evidence in yeast, human and fish that rates of evolution are higher in DNA that is bound in canonical nucleosomes than in flanking linker regions [35
]. In addition, experimental studies on S. cerevisiae
and human show increased mutation rates at nucleosomes because DNA repair proteins have reduced access to DNA once DNA is packaged on histones [39
], so this may explain the elevated evolutionary rates observed for DNA in nucleosomes [36
]. It may be especially difficult to unwind DNA from a relatively rare histone variant, such as CENH3, with an unusual nucleosome structure, and this could lead to inefficient DNA repair at centromeres. Similarly, the subtelomeres of S. cerevisiae
show accelerated base-pair substitution and also have a non-canonical chromatin structure [41
]. The alternative conformation of chromatin at centromeres may be necessary for centromere inheritance in the case of regional centromeres or more generally for centromere function [33
], but may also expose centromere DNA to higher rates of mutation and sequence evolution.