INTRODUCTION

Top Abstract Introduction Materials & Methods Results Discussion References

Tandem repeat instability is implicated in several human genetic diseases. The best-documented examples of deleterious rearrangements in tandem repeats are the massive amplifications of microsatellite DNA, known to be responsible for a dozen diseases, including fragile X syndrome and Huntington's disease (for reviews, see references 62 and 74). Rearrangements affecting minisatellites (repeats of 10 to 50 nucleotides) can be harmful, too (4). For example, expansions of a minisatellite are associated with epilepsy (30, 31, 73). During meiosis, minisatellites can display a very high rate of modification, including intra-allele duplications and deletions, and nonreciprocal interallelic transfer of information (2, 24). Recently, a human minisatellite was also found to display massive amplification (78). Rearrangements in tandem repeats are not specific to micro- and minisatellites. Expansions and contractions of larger tandem repeats have been observed in Drosophila melanogaster and yeast (48, 49, 69, 75, 76).

While replication slippage can easily account for small changes in microsatellite copy number (63), the origin of massive amplifications remains a mystery. Since the predominant rearrangement events observed in minisatellites are nonreciprocal interallelic transfers of information, the meiotic instability affecting those sequences is thought to result from gene conversions rather than replication (24). Tandem repeat rearrangements observed in Drosophila are linked to P-M dysgenesis and have also been supposed to be the consequence of genetic recombination, because P-element excision is known to induce gene conversion (9, 29, 48, 49, 69).

Gene conversions are most often explained by the double-strand break (DSB) repair model, proposed by Resnick and Martin (54) and Szostak et al. (68) to account for recombination events in yeast and other fungi. Many of the features of this model, or of its revised version (67), have been experimentally verified in Saccharomyces cerevisiae. The initial observation that a DSB in the DNA double helix induced a gene conversion in mitotic cells (45) was corroborated by results showing that some site-specific gene conversions, such as mating-type switching and intron homing, are initiated by site-specific DSBs (3, 52, 64), as are most meiotic gene conversions and chromosomal exchanges (33, 66). DSB formation is followed by resection of the ends (Fig. 1A, step 1) (65, 67, 77). The resulting 3' ends can then invade the template (Fig. 1A, step 2) and provide a primer for new DNA synthesis, resulting in the restoration of the degraded single strands.

View larger version (12K): [in this window] [in a new window]   FIG. 1.   The SDSA model, and how it could explain rearrangements in tandem arrays. (A) Two alternative DSB repair models. The ends of the break are first resected (step 1), and the resulting 3' ends can invade a homologous template and prime DNA synthesis (steps 2 and 3). Szostak et al. (68) proposed that resolution would require cutting of two Holliday junctions, resulting in crossover or noncrossover products (step 4), but to explain the lack or low frequency of crossover events accompanying mitotic gene conversion, other authors (10, 22, 42) proposed that the newly synthesized strands could be unwound from the template and anneal together (step 5). (B) A model of tandem repeat rearrangements derived from the DSB repair model of Szostak et al. (68). Expansions and contractions could be due to slippage events during semiconservative DNA synthesis (step 1). The resulting loop could be corrected in either direction, or not be corrected, but anyway, after resolution of the Holliday junctions (step 2), the rearrangements should be found on both donor and recipient molecules. (C) Origin of tandem repeat rearrangements according to a simple SDSA model. Both 3' ends of the broken molecule invade the homologous template and prime DNA synthesis (step 1). Then, the newly synthesized strands are unwound from the template and anneal together. Two of the many possibilities of annealing are represented (step 2).

As gene conversion in yeast was observed to be frequently associated with crossover, the DSB repair model postulates that resolution of the gene conversion occurs through the cutting of two Holliday junctions (Fig. 1A, step 4) (58, 68). Holliday junctions are symmetrical structures, and it was assumed that they could be resolved in two different ways, resulting either in crossover or in no crossover (23). In the original experiment supporting the DSB repair model, half of the gene conversion events were accompanied by crossover (46). However, in most subsequent studies, the rate of crossover associated with gene conversion was found to be much less than 50%, and many types of gene conversion are rarely associated with crossover, in Saccharomyces (26, 27, 51), Drosophila (10, 18), Ustilago maydis (11), bacteria (41), and humans (24). Thus, a series of models which do not require Holliday junctions have been proposed. Their basic assumption is that after strand invasion and new DNA synthesis, the newly synthesized DNA strands are unwound from the template, allowing the annealing of the two free 3' ends surrounding the DSB (Fig. 1A, step 5) (4, 10, 11, 22, 38, 41-44). Because this model of gene conversion includes an event of single-strand annealing, it was termed synthesis-dependent strand annealing (SDSA) by Nassif et al. (43).

The distinctive feature of SDSA models is that the recipient locus receives two newly synthesized strands of DNA and the donor template remains unchanged. DNA synthesis is thus conservative and fundamentally different from the semiconservative genome replication that occurs during the S phase of the cell cycle. As McGill et al. (38) noted, this could be the result of an unwinding, by a topoisomerase, of the two duplex DNAs created during gene conversion, restoring the original template strands to one duplex and the newly made strands to the other. An alternative model is that newly synthesized strands are continuously displaced from the template, analogous to the bubble migration mechanism proposed by Formosa and Alberts (14). In this view, newly synthesized DNA maintains only a short base-paired contact near the point of synthesis, as in RNA transcription, and the newly synthesized DNA is largely unpaired with its template and is free to anneal with a complementary strand.

SDSA models were initially proposed to explain a lack of crossover, but they can also explain the observation that during yeast mating-type switching and P-element excision, gene conversion events are unidirectional (18, 38), in other words, that there is no transfer of information from the repaired molecule to the template. In addition, they can explain the P-induced rearrangements in tandem repeats in Drosophila (9, 29, 48, 49, 69) and the meiotic rearrangements in human minisatellites (2, 4, 24) and in yeast CUP1 sequences (75, 76) as the consequences of DSB repair. Figure 1C shows how unwinding of newly synthesized sequences can lead to tandem repeat rearrangements. Although recombination has been implicated in these systems, there are limitations to a complete description of the process: the hypothesis of an initiating DSB cannot be checked, and the donor and recipient molecules cannot both be recovered (in the Drosophila and human systems) or cannot be identified (in the yeast CUP1 system). Moreover, to explain P-induced rearrangements, one had to suppose that the template for gene conversion was the sister chromatid, which is impossible to demonstrate.

In this paper, we present an experimental system in which both the donor and recipient molecules can be recovered and identified after recombination induced by a known DSB at a specific site. We demonstrate that when a repeated array is transferred by DSB repair to a broken molecule in yeast, the transferred sequence displays various expansions and contractions in half of the recombinants. The template itself is not modified. The expansions sometimes involve the duplication of more than 4 kb. To explain such events by "classical" replication slippage, one would have to hypothesize that the newly synthesized DNA strands are unwound from their template over several kilobases, which is the distinctive feature of conservative DNA synthesis (14). Furthermore, we formally demonstrate that DSB repair can be achieved from two different templates, which also requires the detachment of newly synthesized DNA. These results support the SDSA model for gene conversion and the idea that the DNA synthesis associated with DNA repair is fundamentally different from the semiconservative replication of the genome. We suggest that DSB repair can be a major source of amplification of tandemly repeated sequences. Thus, the massive amplifications of microsatellites responsible for many genetic diseases could have their origin in DNA repair rather than in genome replication.

	MATERIALS AND METHODS

Top Abstract Introduction Materials & Methods Results Discussion References

Plasmids. A series of five plasmids, diagrammed in Fig. 2, were derived from Ted, a centromeric plasmid marked by the URA3 gene (provided by W. Kramer). In pFP14, a genomic XhoI/SalI fragment including the LEU2 gene was inserted into the polylinker of Ted. In the other four Ted derivatives, an extra sequence was inserted in the KpnI site of LEU2. In pFP5, the extra sequence contains two 5S arrays of seven 5S genes surrounding the white gene from D. melanogaster and was taken from pCA422 (49). When pFP5 is transformed into yeast, the repeated 5S genes undergo deletions similar to those described in this study, but the plasmid is stable after transformation (50). The two plasmids containing 5S genes and diagrammed in Fig. 2 are transformant derivatives of pFP5. In pFP23, the insert is a EcoRI fragment containing white, and in pFP24, it is a EcoRI/SmaI fragment containing a truncated white copy. pFP13, the plasmid shown in Fig. 6, is a centromeric plasmid marked by TRP1 and contains an XhoI/SalI fragment corresponding to the LEU2 gene, where the 400-bp KpnI/EcoRI fragment has been replaced by a 40-bp synthetic HO cut site.

View larger version (18K): [in this window] [in a new window]   FIG. 2.   Experimental system to study DSB repair involving tandem arrays. An HO cut site is introduced into a KpnI site of the chromosomal endogenous LEU2 gene. The HO gene can be expressed from an inducible promoter and will cut the LEU2 gene (A). The structures of the templates are shown as follows: no homologous template (B); plasmid LEU2 template (C); plasmid LEU2 templates containing eight D. melanogaster 5S genes in a tandem array (D), two 5S arrays surrounding the D. melanogaster white gene (E), the 5' part of the white gene (F), or the entire white gene (G); chromosomal LEU2 template (H); and chromosomal leu2 template with a 9.1-kb insert corresponding to two 5S arrays surrounding the white gene (I). Shaded box, LEU2 gene; open box, one D. melanogaster 5S gene (375 bp); solid box, D. melanogaster white gene; open circle, centromere. The number of repeats in each 5S array is given above the array.

Strains. The S. cerevisiae strains studied in this study all derive from JKM111 (40), which contains a GAL::HO fusion inserted into the chromosomal ADE3 locus (57), and G304 (20). All transformations were performed as described by Chen et al. (7). One-step disruption (55) and two-step replacement (59) were used to modify the genome, and the corresponding transformed strains were checked by Southern blotting. The MAT locus of JKM111 was disrupted with pFP18, and an HO cut site was introduced into its LEU2 gene with the pLS27 plasmid, resulting in the YFP17 strain. To provide a chromosomal LEU2 donor template, pJH18 or pFP20 was integrated into the ura3-52 gene, on chromosome V of YFP17. To provide a plasmid LEU2 template, pFP5, pFP14, pFP23, or pFP24 was transformed into YFP17.

DSB induction and characterization of recombinants. Yeast was grown for 24 h in yeast extract-peptone (YEP)-dextrose, or in uracil or tryptophan dropout medium if plasmid selection was required. This culture was then used to inoculate 50 ml of YEP-glycerol at an initial concentration of 10⁶ cells per ml. The YEP-glycerol culture was grown overnight, to a final concentration of 1 × 10⁷ to 5 × 10⁷ cells per ml, to prepare the cells for galactose induction. Then cells were plated on YEP-dextrose and YEP-galactose plates at a concentration of about 200 cells per plate. In the absence of any DSB, colonies appear on YEP-dextrose and YEP-galactose with the same efficiency (data not shown). For strains with an HO cut site on the chromosomal LEU2 gene, DSB repair efficiency was scored as the ratio of the number of colonies on YEP-galactose to that on YEP-dextrose. Independent colonies were then subcloned, and independent subclones were molecularly characterized as described by Pâques et al. (49). For the strains with an HO cut site on a plasmid LEU2 gene, we used a different procedure. For the strains with a single truncated leu2 template (Fig. 6B) or with two truncated leu2 templates (Fig. 6C), colonies were replicated onto leucine dropout plates, and the efficiency of gap repair was scored as the frequency of Leu⁺ colonies. Twenty Leu⁺ recombinants from the strain with two templates were checked molecularly and were shown to have gap repaired the plasmid LEU2 gene. For the strain with a single continuous LEU2 template (Fig. 6A), the LEU2 marker cannot be used to monitor homologous repair because all the cells already have a functional LEU2 copy. Therefore, colonies were replicated onto tryptophan dropout plates, and the efficiency of gap repair was scored as the frequency of Trp⁺ colonies among all colonies. Twenty Trp⁺ recombinants were checked molecularly, and in these strains, the plasmid LEU2 gene had been gap repaired.

Identification of crossover events. A leu2 donor was inserted on chromosome V of YFP17 by targeted integration of pFP20 in ura3-52, so that the leu2 donor is surrounded by ura3-52 and URA3. Spontaneous "pop-out" events recombining the functional URA3 gene and the mutated ura3-52 in a single ura3-52 gene occur with a frequency of about 10⁴ and result in papillations when the strain is replicated onto 5-fluoro-orotic acid (5-FOA) medium (21). Crossover occurring between the two leu2 copies on chromosomes III and V will give rise to a reciprocal translocation. The URA3 and ura3-52 gene will then be separated, pop-out will be impossible, and the papillations will disappear (21). Recombinants were patched onto 5-FOA plates, and if no papillation was observed, the strain was analyzed molecularly by Southern blotting. Crossover structure was tested with the diagnostic SphI and EagI restriction enzymes (see Fig. 4). However, because of coconversions of an adjacent delta sequence, only 20% of total crossover events could be identified by this screen. Nevertheless, these coconversions do not seem to affect the nature and frequency of the rearrangements we observe (50).

Media and growth conditions. YEP-dextrose and synthetic dropout media used for the growth of S. cerevisiae were made according to the method of Sherman et al. (60). YEP-galactose contains 2% galactose (wt/vol) instead of glucose to replace glucose as a carbon source. YEP-glycerol contains 2% glycerol (wt/vol) instead of glucose.

	RESULTS

Top Abstract Introduction Materials & Methods Results Discussion References

P-induced rearrangements in tandem repeats in Drosophila and meiotic rearrangements in the yeast CUP1 locus and in human minisatellites led to the hypothesis that copying of a repeated array during DSB repair could occur with deletions and duplications of the repeated sequence (4, 9, 24, 29, 48, 49, 69, 75, 76). To test this hypothesis, we designed an assay with the yeast S. cerevisiae (Fig. 2). We used the site-specific HO endonuclease under the control of an inducible promoter to deliver one DSB per cell within the LEU2 gene. In S. cerevisiae, DSBs are repaired mainly by homologous recombination, so when no homologous template is available, DSB repair is very inefficient (0.3% of the cells). However, a homologous LEU2 template, either on a centromeric plasmid (Fig. 2C) or inserted on a chromosome (Fig. 2H), allows DSB healing in 28 to 36% of the cells. Various additional sequences were inserted in the template, at the exact same location as the HO cut site, so that in order to repair the break by homologous recombination, the broken molecules would have to copy the insert. Three of these inserts have a complex repeated structure: they carry a single D. melanogaster 5S RNA gene array containing eight repeats (Fig. 2D) or the D. melanogaster white gene surrounded by two 5S gene arrays (Fig. 2E and 2I). These same constructs were made and used previously to characterize P-induced rearrangement in Drosophila (49). Repair occurred in 6 to 13% of the cells, depending on the size of the insert and the location (plasmid or chromosome) of the template. When repair occurred on a template containing a repeated array, we checked the structure of a sample of repaired molecules to look for DSB-induced tandem repeat rearrangements.

Copying of a repeated array during DSB repair is accompanied by expansions and contractions of this array. After galactose induction of DSBs, independent recombinants were recovered. With a plasmid template, only noncrossover events can be recovered, because crossover would result in a dicentric chromosome. With a chromosomal template (Fig. 2I), molecular analysis showed that a crossover event created a pair of reciprocal translocations in 5 of 80 cases (6.3%). We first focused on the noncrossover events.

View larger version (22K): [in this window] [in a new window]   View larger version (37K): [in this window] [in a new window]   FIG. 3.   Recombinants obtained by DSB repair involving tandem repeats. For each panel, the structure of the template is diagrammed in the upper box, and the numbers of recipient (R) and donor (D) molecules with each structure are given on the right. (A) Forty-four recombinants obtained after DSB repair using the plasmid template diagrammed in Fig. 2D. (B) Forty recombinants obtained after DSB repair using the plasmid template diagrammed in Fig. 2E. (C) Thirty-four noncrossover recombinants obtained after repair using the template diagrammed in Fig. 2I. (D) Forty-one noncrossover recombinant lines obtained after repair in G1 of the template diagrammed in Fig. 2I. In seven cases (bottom), two populations of cells differing in the structure of the repaired molecule, were obtained. In panels B and D, one recombinant displays tandem repeat rearrangement in both donor and recipient. In these cases, the corresponding donor and recipient molecules are labeled with stars.

Crossovers and tandem repeat rearrangements are compatible. In its simplest form, the SDSA model of DSB repair predicts that rearrangements should occur on the recipient molecule without altering the donor, and without any associated crossover (10, 22, 42). As illustrated in Fig. 1C, both ends would invade the template and prime new DNA synthesis. Rearrangements would appear during the resolution step, because of the multiple possibilities of annealing offered by the repetitive structure. In this version of SDSA, rearrangements and crossovers would be mutually exclusive.

View larger version (26K): [in this window] [in a new window]   FIG. 4.   Crossover events. The diagram at the top represents the donor and the recipient. The cutting of the donor by HO endonuclease, and genetic and molecular identification of crossover events (with SphI and EagI diagnostic restriction enzymes) allowed us to characterize the 29 events whose structures are shown here.

View larger version (15K): [in this window] [in a new window]   FIG. 5.   A model of SDSA compatible with crossovers. (A) A single 3' end invades the template and primes synthesis (step 1) and is extended by bubble migration (step 2). Most of the time, resolution can occur by annealing (step 3). However, invasion of the template by the second 3' end may sometimes stabilize the strand displaced by the first 3' end (step 4). DNA synthesis would become semiconservative (step 5), and Holliday junctions could be formed (step 6) and subsequently cut. (B) Creation of tandem repeat rearrangements by reinvasion. A single 3' end would invade the donor template and initiate DNA synthesis (step 1). It would then be unwound from the template (step 2) and invade this template a second time. Because repeated sequences have been added to the 3' end, this second invasion can be initiated on any of the tandem repeats on the template. Two different possibilities are shown here (step 3). The conversion could then be accomplished by annealing or by formation and cutting of Holliday junctions, as shown in panel A.

Use of a split template during DSB repair. In Drosophila, many complex P-induced gene conversion events have been found, where sequences were supposed to be copied from two different templates (18, 19, 43). However, in these studies, one of the two templates was presumed to be a sister chromatid, which could not be demonstrated. Similar observations have been reported with S. cerevisiae in a transformation experiment (61). The origin of such events could rely on the ability of newly synthesized DNA to unwind from its template during DNA repair: DNA synthesis could be initiated on two different templates, and the two different new DNA molecules could then be annealed together (18, 43, 61). This corresponds to the model of SDSA diagrammed in Fig. 1A, except that the two 3' ends invade sequences on different chromosomes. Another possibility is that template switching would occur during DNA synthesis (19), which corresponds to the SDSA model diagrammed in Fig. 5A, the one we favor (see above).

6

View larger version (22K): [in this window] [in a new window]   FIG. 6.   Direct evidence for the SDSA model. (A) A plasmid with a gapped leu2 gene can be repaired on an uninterrupted LEU2 template. This event minimally requires one strand invasion. The newly synthesized strand can then anneal to the other 3' end of the DSB. (B) A truncated template with no overlap with one side of the gap does not allow for efficient repair. (C and D) Two truncated templates also allow for gap repair. Each template is lacking an overlap on one side of the gap but is overlapping with the other template. On each panel, an arrow(s) represents the invading 3' end(s) after it has been extended by DNA synthesis. Panels C and D represent two different mechanisms to account for the same events: both 3' ends invade and prime DNA synthesis before the newly synthesized DNA strands are unwound from the templates and anneal together (C), or DNA synthesis from a single 3' end can switch from one template to the other (D). In both cases, two strand invasion steps and one annealing step are required.

Gap repair efficiency decreases as the size of the gap increases. When we induced a DSB in a chromosomal leu2 gene to obtain gap repair events that would have copied 5S genes, we noticed that as the size of the repeated array inserted in the leu2 template increased, gap repair efficiency decreased. With a plasmid template, repair occurred in 28% of the cells when the LEU2 donor contained no insert (Fig. 2C), but repair efficiency dropped to 12.6% with a 2.9-kb insert of eight 5S genes (Fig. 2D) and to 6.4% with a 9.1-kb insert containing two 5S arrays surrounding the white gene (Fig. 2E). When the template was on chromosome V, repair efficiencies were quite similar, dropping from 35.7% with no insert (Fig. 2H) to 12.5% with the 9.1-kb insert (Fig. 2I). This effect on gap repair could be specific to repeated sequences or to the 5S gene; therefore, we checked the effects of two other inserts. When we introduced a truncated 2.1-kb white gene into the LEU2 gene (Fig. 2F), DSB repair efficiency was 11.3%, very similar to the 12.6% efficiency obtained with a 2.9-kb 5S array. With a complete 4.1-kb white gene as an insert, repair efficiency reproducibly dropped further, to 9.8% (Fig. 2G). These results indicate that the decrease in DSB repair efficiency observed with inserts of increasing size is not 5S specific or specific to a redundant structure: increasing amounts of sequences from the white gene have the same effect. Our conclusion is that during double-strand gap repair, the size of the DNA segment to be copied limits the efficiency of repair. The most likely difficulty in gap repair appears to be in the processivity of DNA synthesis.

	DISCUSSION

Top Abstract Introduction Materials & Methods Results Discussion References

We describe here rearrangements of a repeated array during DSB-induced gene conversion. These rearrangements are very similar to changes in tandem repeats that have previously been observed in Drosophila by using the same substrates (48, 49), in human minisatellites (24), and in the yeast CUP1 repeated locus (75, 76) and which have been explained as the consequences of gene conversion. However, this is the first study in which we know not only where but also when the DSB appears, so that we can determine how often DSB repair results in tandem repeat rearrangements. One of our most striking results is that half of the repair events are associated with a deletion or a duplication.

This is also the first study in which we could recover and identify both donor and recipient sequences. This is impossible in the Drosophila and human systems, as mentioned above. Although Welch et al. (75, 76), who characterized meiotic contractions and expansions of the CUP1 locus in yeast, could recover all the recombination products, they could not identify donor and recipient molecules, nor could they tell how often repair occurred without rearrangement. This advantage allowed us to examine the mechanism of DSB repair in detail. We made two key observations.

First, when gene conversion is not associated with crossing-over, the rearrangements are almost always at the recipient site while the donor template remains unchanged. Second, while crossing-over associated with gene conversion is rare, about half of these events also contain changes in the number of repeats, but most of the time on only one of the two participating chromosomes. As we discuss below, these events are not readily explained by the DSB gap repair model proposed by Szostak et al. (68) or by an SDSA model that does not permit crossing-over (10, 22, 42). We can account for all these events by a modified form of the SDSA model that allows crossovers to occur (11).

DSB-induced rearrangements in tandem repeats are best explained in terms of an SDSA model. The gap repair model of Szostak et al. (68) envisions that the two 3' ends of a DSB will each invade the donor template and copy a new strand of DNA. This DNA synthesis is assumed to be semiconservative, like genome replication. Thus, the recipient locus as well as the donor locus become heteroduplex, with one original and one newly synthesized strand. Therefore, in the discussion below, we will refer to the Szostak et al. (68) model for DSB gap repair as a semiconservative gene conversion (SCGC) model. It might be imagined that during this semiconservative replication process, there was replication slippage, leaving a loop on either the newly synthesized or the template strand, to add or subtract repeats (5, 36, 63, 71, 72). However, such events should be just as likely to occur at the donor locus as at the recipient locus, since each is the product of an equivalent semiconservative replication process (Fig. 1B). But we find such rearrangements almost always at the recipient site. Moreover, although replication slippage can explain small duplications, such as the 100-bp duplications observed in a rad27 mutant (70), they can hardly account for the events we observe, involving sometimes more than 4 kb. To explain these duplications by replication slippage, one has to assume that the newly synthesized strands can be unwound from their template over several thousand base pairs, so that they can reinitiate DNA synthesis far upstream from the point they reached.

Direct evidence of SDSA. The most compelling evidence for SDSA comes from our demonstration of gap repair when the template consists of two unlinked overlapping sequences, each of which has homology to only one end of the DSB (Fig. 6). Reconstitution of the intact LEU2 gene on a centromeric plasmid was accomplished 0.3% of the time, approximately 40 times less often than when the two ends of the DSB were both homologous to a single template strand. The difference in efficiency most likely reflects the fact that repair from two templates requires an additional strand invasion step. Gap repair from a single, intact template minimally requires a single strand invasion producing a displaced new strand of DNA that can anneal with the other 3' end of the DSB (Fig. 6A). In contrast, repair from two overlapping unlinked templates requires two independent strand invasion steps, followed by an annealing step (Fig. 6C and D). Once one end of the DSB interacts with a donor, the interaction of the second end with the same template could be greatly enhanced.

A modified SDSA model can account for crossover-associated events. About half of the gap repair events yield the same number of repeats as the donor. One explanation for this distinctive clustering of "perfect" events is that many of them arise by a mechanism (i.e., SCGC) that is different from that which produces the gene conversions with different repeat numbers (SDSA). This possibility cannot be excluded, and it led us to examine gene conversion events with crossing-over. If all SDSA occurred without crossover, then one might conjecture that the crossover-associated events might derive from SCGC. In this case, one might expect all such events to have the same number of repeats as the donor; but we showed that this was not so (Fig. 4). Moreover, 13 of the 16 crossover-associated events showed a rearrangement on only one of the two arrays. This is not consistent with an unequal reciprocal exchange event occurring between two initially identical regions, although it could occur if some repeats were looped out in heteroduplex DNA (75). Consequently, although we cannot rule out the possibility that some gap repair events occur via SCGC, the crossover outcome predicted by such a model was not found.

DNA synthesis as a rate-limiting step. If DNA repair synthesis by bubble migration is prone to be interrupted, it could also explain another feature of our results: gap repair efficiency decreases when the size of the gap increases. During genome replication, DNA polymerases are apparently very processive, but PolI, the bacterial DNA polymerase associated with DNA repair, is not (28). Nothing is known about the DNA polymerase(s) associated with DSB repair in eukaryotes, but our results suggest that they are not processive. When long sequences have to be synthesized before DNA repair can be accomplished, repair fails most of the time. An insert of 9 kb decreases repair efficiency fourfold. In addition, half of the repair events apparently require a reinitiation of DNA synthesis, resulting in a rearrangement. It is only the other half, the perfect repair events with no deletion or duplication, which may reflect uninterrupted DNA synthesis across the repeated template.

DSB repair, a major cause of tandem repeat amplification? Fu et al. (15) proposed that the massive amplifications of microsatellites could be the consequence of reinitiation of DNA synthesis. Our result show that DNA synthesis resulting from DSB repair is more likely to induce tandem repeat rearrangements than DNA synthesis associated with genome replication. Thus, the origin of the massive amplifications of microsatellites in fragile X syndrome or Huntington's disease may be linked more to DNA repair than to genome replication. Thus, in the case of fragile X syndrome, amplifications appear during oogenesis or shortly after fertilization (15, 37), possibly during meiosis, when DSBs may frequently cut the genome. Some microsatellites have been shown to have the potential to form very stable hairpins (8, 16, 39), which could pause the replication machinery (25) and be responsible for small slippage events. The effect of such structures would be much more severe on the synthesis associated with DNA repair, because any pause of the polymerase could favor dissociation of the elongating strand from its template. One can imagine that with microsatellites, several events of dissociation and reinvasion of the elongating strand could occur during a single repair event. Very large amplifications could thus appear as a consequence of repairing a single DSB.