The accurate maintenance of repeated genomic regions presents a difficult problem for any cell as recombination processes naturally alter repeat number. A number of proteins are known to repress excision recombination in the yeast rDNA and hence prevent loss of repeats [such as linker histone Hho1 and topoisomerase-related protein Hpr1 (68
)]. However, mutagens such as hydroxyl radicals and ionizing radiation introduce DNA DSBs that are excellent substrates for single-strand annealing in repetitive DNA regions (70
). This pathway by definition causes repeat loss, and should therefore decrease rDNA repeat number with time, whereas repeat numbers are actually very stable, implying the existence of compensating mechanisms for repeat expansion. One candidate mechanism involves the initiation of recombination from replication forks stalled at the RFB (72
), combined with cohesin removal by ncRNA expression in the intergenic spacer (35
). This mechanism is capable of generating both expansions and contractions but there is no evidence that it is directionally controlled. In cell populations forced to transcribe ribosomal RNA using RNA pol II, colonies that carry very large rDNA expansions do arise in a replication fork-dependent manner, suggesting expansion biased unequal sister chromatid exchange (73
). However, this is actively selected for, because pol II transcription of the rDNA is relatively weak, so cells carrying more repeats have a growth advantage.
Here, we report the existence of a pathway in budding yeast that acts exclusively to increase rDNA repeat number. This pathway is constitutively active in strains lacking either Asf1 or Rtt109, which form a complex with a key role in directing broken replication forks into the HR pathway (74
). Notably, all mutations that blocked the expansion pathway also caused rDNA repeat loss, strongly suggesting that the ongoing activity of this pathway is required for rDNA maintenance. This indicates that regulated Asf1-Rtt109 activity may control rDNA expansion by directing fork recovery into either HR or the alternative non-HR expansion pathway.
Much of our data on factors required for rDNA expansion is consistent with a BIR mechanism, including dependence on Pol32 and Pol2. However, we cannot find any evidence for the involvement of Rad52, which other studies found to be vital for BIR as it is for other homologous recombination reactions in yeast. Rad52 has multiple biochemical activities important for various modes of recombination; it can stimulate Rad51-mediated removal of RPA from single-stranded DNA regions allowing Rad51 to initiate strand displacement, and it can promote annealing of complementary strands in SSA, an activity enhanced by Rad59. In the absence of Rad52, any single-stranded DNA that would initiate BIR is likely coated with RPA and so could neither initiate Rad51-dependent strand displacement nor efficiently anneal to other single-stranded regions.
Nonetheless, Rad52-independent BIR is not unknown. Microhomology-mediated BIR (MMBIR) (76
) has been invoked to explain the formation of complex non-recurrent rearrangements underlying human gene copy number disorders, where break points in repeated sequences do not show the homologous sequences that are a hallmark of HR (77
). This mechanism resembles BIR, except that it is initiated with only a few nucleotides homology to the target sequence, leading to rearrangements with only microhomology detectable at the breakpoints. As for the rDNA expansions reported here, Rad51 and Rad52 are not required for MMBIR in yeast but Pol32 is necessary (61
). Since this mechanism is Rad51-independent, it is likely that the free single strand that initiates MMBIR does so by annealing to another single-stranded region, albeit one with very limited homology. A source of single-stranded targets is other replication forks, as proposed for the fork stalling and template switching model (FoSTeS), in which the free DNA end at one broken replication fork can invade an adjacent fork, resulting in replication fork restart and DNA rearrangement (77
). We propose that a FoSTeS-type mechanism is involved in rDNA expansion, with the single-stranded end from a collapsed replication fork invading an adjacent fork (). This would form an alternative repair pathway that acts when BIR with the sister chromatid is blocked.
Figure 6. Proposed mechanism for rDNA expansion without the HR machinery. A replication fork collapses at one end of a replication bubble, leaving a single-stranded end from lagging strand synthesis. This single-stranded end anneals to the single-stranded region (more ...)
We found no evidence of partial rDNA repeats in asf1
Δ strains, which would be clearly predicted if expansion events occurred at sites of microhomology. A key feature of the rDNA is the repetitive nature of the substrate, which turned out to be crucial for rDNA expansion since expansion was abolished in strains carrying only two rDNA repeats. Even in the context of ongoing rDNA expansion, a marker embedded inside the rDNA was not amplified. Within the rDNA, there is an unusually high density of homologous sequences undergoing replication, and, given that DNA replication appears to occur in discrete factories [reviewed in (79
)], a free single-stranded end from one broken replication fork would rapidly be in very close proximity to a highly homologous single-stranded region at another fork. If higher homology leads to increased efficiency of annealing or fork restart, recombination would be strongly biased towards perfect repeat expansion. The problem remains of how a single-stranded region, which should be rapidly coated with RPA in the absence of the Rad52, can anneal to single-stranded regions at another replication fork, which should also be coated with RPA. It seems that there must be another factor capable of mediating strand annealing in the absence of Rad52, but the identity of this factor remains unknown. In summary, rDNA expansion in asf1
Δ mutants appears to occur by a BIR-type mechanism, but represents a very unusual form of homologous recombination that works in the absence of key homologous recombination proteins.
One clear prediction of our model is that it is highly dependent on the freedom of the broken DNA end to move away and find another replication fork, which would be severely curtailed by cohesin binding in the rDNA. In keeping with this, in a sir2Δ background where rDNA cohesion is lost due to elevated ncRNA transcription, the repeat expansions that occur in asf1Δ cells are massively amplified. It should be noted that the mechanism proposed in is not innately expansion biased, as it would also be possible for the single-stranded region shown in to invade a converging replication fork (i.e. one approaching from the left of this image). This would cause rDNA contraction; however, replication fork restart would lead to partial chromosome XII re-replication and aneuploidy that would most likely be lethal, so such contractions would not be observed.
Human MMBIR-type genome rearrangements are medically important in generating copy number variations in germline and somatic cells, but are difficult to study because their occurrence is both temporally and spatially unpredictable. The yeast rDNA potentially provides a useful model for this process since the recombination events occur at reproducible and predictable rates and are limited to a well-defined genomic region.