In silico reversal of repeat-induced point mutation (RIP) identifies the origins of repeat families and uncovers obscured duplicated genes

Repeat-induced point mutation (RIP) is a genome defence mechanism found within filamentous ascomycete fungi that is purported to combat transposon invasion. RIP mutates duplicated DNA sequences during sexual reproduction, thereby inactivating genes encoded in both copies. First discovered in Neurospora crassa[1, 2], RIP was later demonstrated in the Ascomycetes Magnaporthe oryzae[3, 4], Podospora anserina[5], Leptosphaeria maculans[6] and Fusarium graminearum[7]. Putative RIP events have also been detected bioinformatically in Aspergillus fumigatus[8], Fusarium oxysporum[9–11], Aspergillus nidulans[12], Neurospora tetrasperma[13], Microbotryum violaceum[14], Aspergillus oryzae[15], Magnaporthe oryzae[16], Colletotrichum cereal[17], Aspergillus niger[18], Penicillium chysogenum[18] and most recently in Stagonospora nodorum[19, 20]. Given this broad distribution, it is reasonable to assume that RIP is widespread across, but so far restricted to, filamentous ascomycota and basidiomycota.

The consequences of RIP are that repeated DNA segments, such as would result from the transposition of a retrotransposon, or the duplication of a gene, are mutated and inactivated. RIP would be expected to operate in successive sexual cycles until the sequence identity between duplicated sequences is reduced below the minimum homology threshold required by the RIP machinery. The genome would then contain a repeat family consisting of relics of the duplication event degraded to varying degrees.

The rapid increase in the number of fungal genome assemblies has created a demand for methods to detect and quantify RIP. Two approaches have been used; RIP indices and alignment methods. RIP increases the frequency of particular di-nucleotides (TpA in most cases studied to date) in affected regions of DNA. Thus RIP can be identified by comparing ratios of di-nucleotide frequencies in pre-RIP to post-RIP sequences; these ratios are referred to as "RIP indices" [8, 12, 25, 26]. However, in reality RIP depends upon the alignment of two similar regions of double-stranded DNA [23] and therefore it is more appropriate to use alignments of repeat families to identify and quantify RIP. We have previously introduced a rapid, automated alignment-based procedure for the whole-genome analysis of RIP mutation called RIPCAL [20]. Using this procedure, we readily identified and quantified the degree of RIP in all repeated DNA families within the genome of the necrotrophic fungal wheat pathogen S. nodorum.

Stagonospora (syn. Septoria) nodorum [teleomorph: Phaeosphaeria (syn. Leptosphaeria) nodorum (Müll) Hedjar.] is a major pathogen of wheat and is a model for the fungal class Dothideomycetes, a taxon that includes many important pathogens of crops [27]. S. nodorum infects wheat crops in most wheat-growing areas of the world [28]. Infection is predominantly determined by the presence of various effectors (host-specific toxins) harboured by different strains of the fungus [29]. The fungus is heterothallic (out-crossing) and the mating types are evenly distributed [30]. The fungus over-summers as ascospores on stubble [28] and multiplies via asexual reproduction during the growing season. The pathogen displays high levels of variability as determined by genomic analyses [31, 32] and this has been exploited to determine the biogeographic history of the pathogen [33]. The pattern of micro-satellite markers is consistent with a pattern whereby the pathogen originated in the "Golden Triangle" region and spread as wheat cultivation was adopted in Eurasia and North Africa several thousand years ago and into North and South America, South Africa and Australia since European colonisation.

An initial survey of the nuclear genome sequence of a West Australian isolate (strain SN15) [19] identified 26 repeat families which comprised 6.2% of assembly. The role and origin of several repeat families could not be inferred by homology. We ascribed this to RIP mutation, after which all copies were unrecognisable. RIPCAL analysis showed that the repetitive DNA of SN15 was subject to RIP-like changes [20]. The rDNA repeat (Y1) exhibits selective susceptibility to RIP mutation (Figure 1). RIP does not affect copies located within the tandem rDNA array (also referred to as the nucleolus organiser region, or NOR, Figure 1: regions 3 & 4) [1, 34]. One exception was found in a repeat at the array terminus, which showed evidence of RIP at similar levels to those of non-rDNA array repeats. rDNA repeats were also found scattered throughout the genome. Within the non-rDNA array repeats, short repeats (defined as < 1 kb, however the majority were < 300 bp) did not show evidence of RIP whilst the long repeats (> 1 kb) were RIP-affected [20]. Due to the presence of both RIP-affected and non-RIP-affected copies, the rDNA repeat was perfectly suited to be used as a test case for the validity of bioinformatic predictions of RIP.

Building upon the RIPCAL procedure, we describe here a new technique to reverse the effects of RIP mutation in silico: "deRIP". The deRIP process involves scanning a multiple alignment of a repeat family for RIP-like polymorphism and reverting the alignment consensus to the putative pre-RIP-mutated sequence. The resultant "deRIPped" sequence is a prediction of what a RIP-mutated repeat DNA may have looked like prior to RIP mutation. We have applied the deRIP process to the repetitive DNA of S. nodorum SN15 which has increased the number of recognisable repeat families from 65% (17/26) to 92% (23/25).

The deRIP algorithm was designed to reverse the affect of RIP upon repeat families in-silico and thereby help determine the evolutionary history of repeated elements. The process automates the selective alteration of bases within a consensus according to a set of rules that can be determined by considering the RIP machinery operating in the organism concerned. Despite its validation with known non-RIP-affected sequences, deRIP has some limitations and will not necessarily perfectly predict the ancestral sequence. DeRIP can only choose within options provided by the aligned set of repeats. In the example given in Figure 2 the TpA sequence at position 1900-1901 was converted to TpG. To do this, at least one of the copies of the repeat must have the presumably ancestral di-nucleotide TpG at this site. If all extant copies had been mutated, the reversion would have no support. The deRIP process is therefore critically dependent on the degree of RIP within a repeat. The success of deRIP is also dependent upon the accuracy of the alignment. A noteworthy aside is that default alignment parameters often fail to align fungal repeats correctly due to complex internal repeat structures. Finally, the diagnostic metrics of deRIP success, deRIP improvement and hit discovery, are only possible to calculate if appropriate matching sequences exist in the queried databases. If a repeat sequence is truly novel, its "homology" cannot be improved until homologs are found.

RIPCAL uses a model sequence to compare to aligned repeats for RIP-like polymorphism. Selecting the repeat with the highest total count of G and C nucleotides assumes that high G:C content is representative of the least RIP-affected repeat. The majority consensus model on the other hand could be representative of the least or most RIP-affected repeat depending on the level of RIP mutation within the repeat family. Previously, we had selected the sequence with the highest G:C content as the RIPCAL model [20]. While in most cases this rationale is sound, the G:C model has several shortcomings. If a sequence with the highest total G:C content does not span the full length of its alignment, RIP data from the un-covered regions would be lost. Alternatively, a repeat longer than the least RIP-affected repeat (e.g. resulting from a large sequence insertion into a RIP-affected repeat) may have higher total G:C content merely due to its greater length. The G:C model is also sensitive to variations in G:C content not related to RIP. Furthermore, RIP occurs between multiple combinations of repeats over time. The G:C model sequence therefore comprises of an amalgam of pre-RIP and post-RIP di-nucleotides relative to the alignment as a whole.

RIP mutations have directionality (Table 1), so the combination of pre- and post-RIP sites makes it necessary to consider RIP mutation both towards and away from the G:C model sequence. In contrast, a deRIP consensus model, being a prediction of the pre-RIP-mutated sequence, has the advantage of polarity. As such, deRIP mutation calculations can be restricted to one direction: proceeding from the deRIP consensus to the RIP-affected repeat.

The prior isolation of active copies of three transposons, Molly Pixie and Elsa, as well as the differential effect of RIP on the rDNA repeats, allowed a thorough test of the power of deRIP to reconstruct the ancestral sequence. In all four examples the predicted deRIP consensus of the RIP-affected sequences was the best match to the active copy indicating that deRIP was able to accurately revert the RIP-degraded repeats close to their original states. These analyses helped define the concepts of hit discovery number and deRIP improvements as applied more broadly in Table 3.

The deRIP process serves to highlight the effectiveness of RIP as a transposon-silencing mechanism. In most observed cases, the resemblance between RIP-degraded repeats and their non-RIP-affected, functional counterparts is minimal (Additional file 1). In the case of the Molly, Elsa and Pixie transposons, functional sequences of transposon proteins were available for comparison. No viable open-reading frames could be found in any of their respective genomic matches in S. nodorum SN15 (Additional file 1). Some repeat families could not even be classified by homology prior to deRIP (Table 4). The deRIP process is therefore an essential tool which facilitates the identification and understanding of the role and origin of fungal repetitive DNA. The effectiveness of deRIP was such that functional assignments were improved quantitatively or qualitatively in nearly all cases. This was most clearly the case when repeat families were most clearly affected by RIP (Table 3, Table 4).

Conversely when the repeat family was not RIP-affected, the deRIP process was not able to improve the homology assignment. An example is the transposon repeat family R37 which had low RIP dominance (by majority consensus) of 0.25, indicating that R37 is not greatly affected by RIP mutation.

The R10 repeat contained regions corresponding to three S. nodorum genes (SNOG_15997, SNOG_11270 and SNOG_16585 [NCBI: EAT76576.1, EAT81769.1, EAT76052.1]) (Table 4) which are located in separate regions of the genome assembly. The sub-telomeric repeat X26, which contained telomere-associated RecQ helicase sequence and was subject to relatively high levels of RIP mutation (Table 3). RecQ helicase plays a critical role in genome maintenance and is essential for DNA replication in eukaryotes [41]. Twenty six putative functional copies (i.e annotated gene models) of RecQ are present within the S. nodorum genome (Additional file 6). It is currently unclear by what mechanism function is preserved in certain copies of this highly repeated gene family, but not in others.

The repeat family X3X3R8, which replaced the previously defined repeat families X3 and R8, matched to a cluster of endogenous S. nodorum SN15 genes (Table 4) - some of these coding for a DNA repair helicase and ubiquitin conjugating enzyme (Figure 3). The helicase and ubiquitin conjugating enzyme genes within X3X3R8 were homologous to Saccharomyces cerevisiae proteins Rad5 [SGD: YLR032W] and Rad6 [SGD: YGL058W] respectively (Additional file 2). These proteins are involved in the post-replication repair of UV-damaged DNA, the epigenetic silencing of telomeres and sporulation in yeast [42–44].

The absence of active copies of the Elsa, Molly and Pixie families in the Australian SN15 isolate contrasts with the situation in the UK. Rawson screened several isolates for active transposons and used a trapping process to isolate active copies [35]. Using these transposons as probes showed great variation in copy number and band intensity amongst a collection of UK isolates. We have only looked at one Australian isolate, but it appears to be devoid of active transposons (Additional file 1). Consideration of the properties of RIP, the need for sexual reproduction by the fungus in Mediterranean climates and the biogeography of S. nodorum, could explain the absence of transposons in the Australian isolate. Repeated elements over a threshold size and above a threshold identity would be subject to RIP. This would inactivate all copies of a transposon during a meiotic event that appears to be necessary for survival over the hot summer [30]. It would not be conceivable for an active copy to be reconstituted in an asexual population derived from such an event. The survival of a transposon in a population of a sexually reproducing fungus would require mating with an isolate with an active (and presumably single) copy of the transposon. The invasion of S. nodorum into Australia most likely occurred via the propagation of a small founder population consistent with the reduced polymorphism of populations found here [33]. We speculate that no active transposons have survived within any Australian isolate capable of RIP. Screening of a larger population of Australian and Eurasian isolates to determine differences in frequency and distribution of active copies between the founder and derived populations would be required to confirm this.

In summary, we present a facile and rapid method to assist the annotation of repetitive elements of ascomycete genomes. The deRIP process can predict ancestral functional sequence from degraded repeat elements. Analysis of the repeat families of the fungal phytopathogen Stagonospora nodorum (strain SN15) using deRIP-converted sequences increased the number of recognisable repeat families from 65% (17/26) to 92% (23/25). This has enabled the characterization of many repeat families and has advanced our progress towards the goal of understanding and accounting for the evolutionary history of all regions of a genome.