Increased Negative Superhelical Density in Vivo Enhances the Genetic Instability of Triplet Repeat Sequences*

The influence of negative superhelical density on the genetic instabilities of long GAA·TTC, CGG·CCG, and CTG·CAG repeat sequences was studied in vivo in topologically constrained plasmids in Escherichia coli. These repeat tracts are involved in the etiologies of Friedreich ataxia, fragile X syndrome, and myotonic dystrophy type 1, respectively. The capacity of these DNA tracts to undergo deletions-expansions was explored with three genetic-biochemical approaches including first, the utilization of topoisomerase I and/or DNA gyrase mutants, second, the specific inhibition of DNA gyrase by novobiocin, and third, the genetic removal of the HU protein, thus lowering the negative supercoil density (–σ). All three strategies revealed that higher –σ in vivo enhanced the formation of deleted repeat sequences. The effects were most pronounced for the Friedreich ataxia and the fragile X triplet repeat sequences. Higher levels of –σ stabilize non-B DNA conformations (i.e. triplexes, sticky DNA, flexible and writhed DNA, slipped structures) at appropriate repeat tracts; also, numerous prior genetic instability investigations invoke a role for these structures in promoting the slippage of the DNA complementary strands. Thus, we propose that the in vivo modulation of the DNA structure, localized to the repeat tracts, is responsible for these behaviors. Presuming that these interrelationships are also found in humans, dynamic alterations in the chromosomal nuclear matrix may modulate the –σ of certain DNA regions and, thus, stabilize/destabilize certain non-B conformations which regulate the genetic expansions-deletions responsible for the diseases.

(Promega) and were described in detail (12,33,34). The repeat tracts in pRW3546 (12), pRW3691 (34), and pRW3246 (33) are located downstream of the functional lacZ promoter, thus allowing for the direct control of the transcription level by isopropyl-␤-D-thiogalactopyranoside (IPTG) induction and lacI Q repressor inhibition. The lacI Q repressor was encoded by the pI Q -kan plasmid (35). All plasmids used herein were characterized by restriction mapping (to determine the orientation and lengths of the cloned TRS) and by dideoxy sequencing of both strands as described earlier (36) using pUC19-and pGEM-specific primers (New England Biolabs, Inc.).
All plasmids were purified from HB101 using standard maxiprep procedures (Sigma), and plasmids were subsequently electrophoretically separated overnight on long (30 cm) 1.2% agarose gels. The bands corresponding to the full-length monomer plasmids (no TRS deletions) were excised without ethidium bromide staining, and DNA was recovered by electroelution (37) and purified by phenol/chloroform extraction and ethanol precipitation. The size of the TRS insert and the extent of instability was determined using restriction digestion and polyacrylamide gel electrophoresis followed by quantitative phosphorimaging analyses as described below. Preparations containing more than 90% of the full-length, undeleted TRS tracts were used in the further experiments. E. coli JRY880 was a generous gift from Dr. J. Rouviere-Yaniv.
Determination of the Plasmid Superhelical Density in Vivo-pUC19 was isolated from different E. coli strains grown to an A 600 of 0.9, and the topoisomer distributions were analyzed by agarose gel electrophoresis (1% w/v) in 90 mM Tris borate, 2 mM EDTA, pH 8.0, in the presence of 3.5 M chloroquine. The plasmids were separated on 30-cm-long gels (2-3 V/cm) for 36 h. After electrophoresis, the gels were rinsed with distilled water, and DNA was stained with ethidium bromide. The quantitation of the results was performed using FluorChem version 3.04 (Alpha Innotech Corp.). The average negative superhelical density was obtained as described earlier (38). 3 Determination of the Plasmid Copy Number-To evaluate the possible influence of the supercoil-dependent replicative potential of different E. coli strains, the copy number of pUC19 was determined for each strain as described earlier (36,39,40). The quantitative analyses of the plasmid and genomic DNAs separated by agarose gels were performed using FluorChem version 3.04 (Alpha Innotech Corp.).
Preparation of Bacterial Lysates and Fluorometric Quantitation of GFP Expression-pGFPuv (Clontech) was used as a reporter plasmid to characterize the lacZ promoter expression levels in different E. coli strains and under varying transcription and supercoiling conditions. The preparation of the bacterial lysates and quantitation of the GFP levels were performed precisely as described (41). Fluorescence measurements were conducted on a Turner Quantech fluorometer using an NB405 excitation filter and a SC500 emission filter. A standard curve was made by supplementing GFP-free lysates with known amounts of recombinant GFPuv (Clontech), which gave the expected linear relationship. Analyses of the GFPuv expression were performed in E. coli JTT1, RS2, SD7, and KD112 for cells grown in LB medium and in LB supplemented with 2 mM IPTG. The amount of GFPuv expressed in E. coli grown in the presence of pI Q -kan (encoding the lacI Q repressor) was undetectable in all strains. In the typical conditions for our recultivation studies (LB medium), the GFP expression level was ϳ10-, 6-, and 3-fold higher in the KD112, RS2, and SD7 strains, respectively, than in the parental E. coli JTT1. Hence, we found no direct correlation between the in vivo superhelical density in these four strains and the GFP expression level from the lacZ promoter.
Assay for Genetic Instability-The plasmids containing various TRS were transformed into the appropriate E. coli strains and were grown in 10 ml of LB cultures for a number of generations, as described (16,33). The efficiency of transformation was determined in every experiment by plating an aliquot of the transformation mixture onto the agar plate containing ampicillin (100 g/ml); the cultures were always initiated with the equivalent of 1000 -3000 colony forming units. Importantly, to avoid the growth advantage effect of the cells harboring plasmids with shorter TRSs (33), the cells were always subcultured into the fresh media at the logarithmic phase (A 600 of 0.75-0.9, 10 -16 h of growth). Cells were recultivated 3 times in 20-generation intervals, starting each subculture with 10 3 -10 4 cells as an inoculum. An initial population of this size allowed the detection of several individual instability events while preventing a potential "bottleneck" effect if the inoculum was too small (1-100 cells). The cultures were grown at 37°C at a shaking rate of 250 rpm in LB media supplemented with an appropriate antibiotic (ampicillin at 100 g/ml, kanamycin at 50 g/ml). The cells from each culture were harvested and stored at Ϫ20°C to the end of the experiment, and then all plasmid isolations and analyses were performed simultaneously.
In the experiments with inhibited transcription, the plasmid pI Q -kan was first transformed into the appropriate E. coli strain, and the pI Qkan-containing competent cells were prepared. When required, 2 mM IPTG was included in the growth medium to induce transcription from the lacZ promoter.
In the case of studies performed in the presence of novobiocin (Sigma) the drug was added, during all stages of the experiment beginning with the preparation of competent cells, at the concentrations equal to the strain-specific inhibitory concentration of 5% (IC 05 ϭ 4 and 16 M for SD7 and HB101 E. coli, respectively). Novobiocin inhibits the assembly of the active DNA gyrase (42,43) and at the concentrations used in this work has no influence on cell growth. IC 05 values were determined experimentally for each E. coli strain used. In contrast to the quinolones, novobiocin (as well as other coumarins) does not lead to the formation of DNA double-strand breaks that may influence the results of the TRS instability studies (44 -47).
Simultaneous Analyses of the Plasmid Superhelical Density and TRS Instability from Single Colonies-Studies were conducted with E. coli JRY880, which is a deletion mutant of the hupA gene encoding the ␣ subunit of the HU protein, to evaluate its potential role in modulating the in vivo supercoil density of pRW3546 and thereby influencing the TRS genetic instability. The supercoil density and the GAA⅐TTC instability were monitored in single colonies since E. coli JRY880 is known to rapidly acquire suppressor mutations resulting in an increase of the intracellular Ϫ (48). Competent bacterial cells were co-transformed with pRW3546 which contained (GAA⅐TTC) 150 and with pACYC184. pACYC184 was used as a reference plasmid to monitor the Ϫ. Thus, harboring both plasmids in the same cells allowed the simultaneous analysis of the repeat instability at a precisely determined Ϫ. After transformation cells were plated on an agar plate containing ampicillin (100 g/ml) for selection for pRW3546, tetracycline (20 g/ml) for selection for pACYC184, and chloramphenicol (12.5 g/ml) for selection for E. coli JRY880. Single colonies were subsequently grown in liquid LB medium supplemented with the same concentrations of the antibiotics until the cultures reached an A 600 of 0.9. Plasmids were isolated using standard techniques (Wizard Plus Miniprep DNA Purification System, Promega) and were subjected to restriction digestion. A portion of the isolated DNA (containing a mixture of both pRW3546 and pACYC184) was cleaved by a combination of five restriction endonucleases (AatII, AflIII, DraIII, PvuI, and NdeI) to completely remove any supercoiled pRW3546 while leaving pACYC184 intact. This step was necessary since the presence of undigested pRW3546 could interfere with the analyses of the pACYC184 topoisomer distributions on the chloroquine-containing agarose gels. After phenol/chloroform extraction and ethanol precipitation, the topoisomer distributions of pACYC184 were analyzed by agarose gel electrophoresis (1% w/v) in 90 mM Tris borate, 2 mM EDTA, pH 8.0, in the presence of 3.5 M chloroquine as described above. In parallel, the remaining fraction of the plasmid preparation was cleaved using BssHII/HindIII restriction endonucleases (New England Biolabs) to analyze the stability of the (GAA⅐TTC) 150 tract in pRW3546. The repeat-containing inserts were separated on 5.5% polyacrylamide gels and analyzed as described below.
Analysis of Lengths of Genetic Instability Products-Plasmids were isolated as described above and were subjected to restriction digestion. BssHII/HindIII restriction endonucleases (New England Biolabs) were utilized to analyze the stability of the (GAA⅐TTC) 150 tract in pRW3546, and EcoRI/BamHI were used in the case of pRW3691 and pRW3246. The repeat-containing inserts were radioactively labeled using with the Klenow fragment of E. coli DNA polymerase I and [␣-32 P]dATP and separated on 5.5-7% polyacrylamide gels in TAE buffer (40 mM Tris acetate, 1 mM EDTA, pH 8.0). Results were analyzed using Storm 820 and ImageQuant software (Amersham Biosciences). The statistical analyses were performed using SigmaStat Version 2.03.

RESULTS
Increased Negative Superhelical Density Enhances the Instability of GAA⅐TTC and CGG⅐CCG Repeats but Not CTG⅐CAG-Four isogenic E. coli strains (parental (JTT1) as well as cells harboring mutations in the topoisomerase I and/or DNA gyrase genes (RS2, SD7, and KD112)) were transformed with plasmids containing (GAA⅐TTC) 150 , (CGG⅐CCG) 73 , or (CTG⅐CAG) 98 (TABLE ONE) and were subjected to recultivation experiments. The recultivation assay is a powerful method for analyzing the genetic instability of repetitive sequences and has been widely used previously (12,16,18,19,33). The data from the recultivation experiments with pRW3546 containing (GAA⅐TTC) 150 in the four strains are shown in Fig. 1. The strains used in this study differ in their intracellular level of negative supercoil density, which in turn will modulate the propensity of repetitive DNA sequences to adopt any non-B DNA conformations with under-wound primary helices (6,7,49). Analysis of the topoisomer distributions of the pUC19 reference plasmid on chloroquine-containing agarose gels demonstrated that plasmids isolated from the parental JTT1 and RS2 (topA mutation) strains are highly supercoiled, whereas DNAs from SD7 and KD112 strains (both DNA gyrase mutants) are much more relaxed (Fig. 2). Calculation of the average Ϫ revealed a significant difference between the plasmids grown in JTT1 ( ϭ Ϫ0.057) or RS2 ( ϭ Ϫ0.058) and SD7 ( ϭ Ϫ0.049) or KD112 ( ϭ Ϫ0.052).
Analysis of the instability of the (GAA⅐TTC) 150 tract in E. coli RS2 revealed that the full-length insert was entirely deleted from pRW3546, and only shorter DNA inserts (Ͻ150 repeats) could be detected after the 3rd recultivation (Fig. 1). Similar results were obtained in the parental JTT1 strain, confirming that the (GAA⅐TTC) 150 tract is highly unstable genetically when subjected to high superhelical tension. However, the stability of the TRS was much increased in the SD7 and KD112 strains, which maintained a lower level of negative superhelical density. Even after the 3rd recultivation, a substantial amount (ϳ10 -25%) of the plasmids still harbored the full-length (GAA⅐TTC) 150 repeats. Similar results were obtained in the case of pRW3691 containing the (CGG⅐CCG) 73 repeats (Fig. 3A), where the effect of Ϫ was significant although not as pronounced as in the case of pRW3546 containing the (GAA⅐TTC) 150 (Fig. 3A). No full-length fragment containing the TRS remained after the 3rd subculture step in E. coli JTT1 as well as in RS2; however, 5-8% of the undeleted (CGG⅐CCG) 73 inserts were still present after identical cultivations in SD7 and KD112. As observed previously (12,50), the (GAA⅐TTC) 150 insert and especially the (CGG⅐CCG) 73 tract are very unstable and difficult to maintain due to their propensity to delete in E. coli. No difference in the average negative superhelical density between the wild type (JTT1) and the topoisomerase I mutant (RS2) of E. coli was detected. Indeed, previous reports demonstrated an increase in superhelical density in chromosomal DNA and plasmids expressing the tetracycline resistance gene in the strains harboring topoisomerase I mutation (51)(52)(53)(54), whereas plasmids lacking the tet gene (such as pUC19 derivatives) showed normal supercoiled densities (52,54).
Interpretation of these results could be seriously complicated by the differences in plasmid metabolism in the various E. coli strains even after the conditions of bacterial growth were standardized, and the cultures never reached stationary phase ("Materials and Methods"). The main concern is the replicative propensity (copy number) of the plasmids in the different E. coli strains. Results of the recultivation experiments can be properly interpreted if both the growth of the bacterial population and the plasmid copy number are monitored. Therefore, we

Plasmids and E. coli strains used in this study
The terms "orientation I" and "orientation II" refer to the orientation of the TRS relative to the origin of replication; for example, for the plasmids containing (GAA⅐TTC) n tracts in orientation I, the GAA repeat is in the leading strand template, whereas for the plasmids harboring (GAA⅐TTC) n tracts in orientation II, the GAA repeat is in the lagging strand template (16). NA, not applicable.

Plasmids
Repeating analyzed the copy numbers of pGFPuv in all four strains (Fig. 4). SD7 and KD112 (the "relaxed DNA strains") maintained 2.2 and 1.4 times, respectively, more copies of the plasmid than JTT1 and RS2, which had essentially the same amount of plasmid molecules. However, the former strains demonstrated a much higher stability of the TRS inserts. Thus, regardless of the higher copy numbers, which represent a larger number of plasmid replication events, the (CGG⅐CCG) 73 and (GAA⅐TTC) 150 tracts were substantially more stable in the SD7 and KD112 strains. If the replication properties of the TRS-containing plasmids were similar in all four strains, the supercoil-dependent difference in the instability of the repeat tracts might be even more pronounced than shown in Figs. 1 and 3 since the gyrase-impaired strains (SD7 and KD112) demonstrated the greatest TRS stability but had the highest number of plasmid replication events per cell (Fig. 4). Surprisingly, the plasmid containing the (CTG⅐CAG) 98 tract, the most stable of the three inserts tested, was not influenced by the level of intracellular Ϫ (Fig. 3B). The overall stability of the (CTG⅐CAG) 98 tract is much higher in all strains than found for pRW3546 and pRW3691. In fact, after the 3rd subculture step, at least 50 -70% of the plasmid preparations contained the full-length (CTG⅐CAG) 98 insert (Fig. 3B). The reason for the weaker effect of negative superhelical density on the genetic instability of the CTG⅐CAG repeats compared with the GAA⅐TTC and CGG⅐CCG tracts may be related to the differences in the stabilities of the slipped hairpin structures formed by CTG⅐CAG repeats (6,16,18,49,(55)(56)(57)(58) versus the triplex structures (adopted by GAA⅐TTC tracts) (30, 59 -62) and the tetraplex and/or flexible and writhed conformations (demonstrated for CGG⅐CCG repeats) (34,63) under the pressure of negative superhelical density (see "Discussion").
Transcription in Concert with High Negative Supercoil Density Facilitates TRS Instability-The influence of transcription through the repeat tracts on their genetic instability under the various Ϫ conditions was determined. Plasmids harboring (GAA⅐TTC) 150 , (CGG⅐CCG) 73 , or (CTG⅐CAG) 98 were each transformed into E. coli JTT1, RS2, SD7, and KD112, and the cells were recultivated under three different conditions; they are (i) transcription inhibition by expression of the lacI Q repressor from the pI Q -kan plasmid, (ii) background level of transcription (growth in LB medium with no inhibition or stimulation of transcription), and (iii) induced transcription from the lacZ promoter by 2 mM IPTG. The level of expression from the lacZ promoter was analyzed under all experimental conditions and in the four E. coli strains using the pGFPuv plasmid, which expressed the GFPuv protein ("Materials and Methods").
The results obtained with plasmids containing the (GAA⅐TTC) 150 and (CGG⅐CCG) 73 repeats showed that the background transcription (LB medium) from the lacZ promoter located upstream of the TRS significantly destabilized the repeat tracts in all E. coli strains analyzed (compare Figs. 1 and 5 along with Fig. 3A). This effect is in agreement gyrB226) were transformed with the purified non-deleted monomer of pRW3546 and cultured as described under "Materials and Methods." The numbers of the subculture steps are indicated above the gel lanes. M designates the 1-kilobase DNA Ladder (Invitrogen) used as size marker. The arrowhead indicates the full-length DNA fragment containing (GAA⅐TTC) 150 . Analysis of the TRS insert size was performed using BssHII/HindIII digestion followed by radioactive labeling of recessed ends and electrophoresis in 5.5% polyacrylamide gels as described under "Materials and Methods." B, results of quantitation of the amount of undeleted TRS insert (% full-length fragment). Data from three independent experiments were quantitated using phosphorimaging, and the relative stability of the TRS tract is presented as the percentage of the full-length, undeleted starting-length fragment remaining after subsequent recultivations. For control purposes, radioactively labeled DNA fragments were excised from a representative number of polyacrylamide gel lanes, and the level of radioactivity was measured using a liquid scintillation counter (Beckman Instruments). The level of transcription from the lacZ promoter was neither inhibited (by lacI Q repressor) nor stimulated (by IPTG) through the course of the experiment. topol, topoisomerase; gyr., gyrase. with previous studies conducted with the CTG⅐CAG and GTC⅐GAC repeats (18,19,58). In the case of pRW3691, transcription stimulated extensive TRS deletions; therefore, no full-length (CGG⅐CCG) 73 fragment could be detected after the third subculture step in the parental JTT1 and RS2 strains ( Fig. 3A and data not shown). On the other hand, more than 7% of the undeleted TRS insert was detected after propagation of this plasmid in the same E. coli strains but in the absence of transcription from the lacZ promoter (i.e. in the presence of the lacI Q repressor, Fig. 3A). A similar destabilizing effect of active transcription through the repeats was also observed for the plasmid containing the (GAA⅐TTC) 150 insert in all strains (Figs. 1 and 5). Further stimulation of transcription by 2 mM IPTG had no significant effect on the extent of genetic instabilities of the repeating tracts (data not shown).
Interestingly, the influence of transcription on the genetic instability of the TRS was more apparent in E. coli RS2 and JTT1 than in the SD7 and KD112 strains (Figs. 1, 3, and 5). This difference is particularly evident when the amount of undeleted progenitor (GAA⅐TTC) 150 and (CGG⅐CCG) 73 inserts are compared after the second subculture step; a comparison after the third growth step is less appropriate due to the complete loss of the full-length inserts in the experiments with active transcription (Figs. 1 and 3). In the case of the plasmid containing the (GAA⅐TTC) 150 repeats, only ϳ2% of the progenitor insert was detected in E. coli RS2 (high Ϫ) after the 2nd round of growth in LB medium (Fig. 1). Inhibition of transcription increased this amount ϳ10-fold (Fig.  5). In contrast, identical experiments performed in E. coli SD7 (low Ϫ) resulted in ϳ2.5-fold increase in TRS instability from the transcribed to the non-transcribed conditions. For the plasmid harboring the (CGG⅐CCG) 73 insert, the corresponding values were 3 versus 12% (in E. coli RS2) and 12 versus 17% (in E. coli SD7). These differences were statistically significant (p Ͻ 0.05). Perhaps, the initial high level of negative superhelical density in E. coli RS2 and JTT1, influenced by local changes of DNA topology associated with transcription, further accelerated TRS instability, whereas the transcription through the relaxed DNA template in E. coli SD7 and KD112 had a significantly less deleterious effect on the stability of the repeats. These results show a strong link between transcription and global superhelical density and their synergistic effect on the TRS instability.
Hence, we conclude that transcription through the repeats strongly induces the instability of the CGG⅐CCG and GAA⅐TTC repeats. The dominant effect of transcription might mask the actual impact of the negative superhelical density changes. To evaluate the influence of global Ϫ on the TRS instability unbiased by changes in DNA topology associated with RNA synthesis, we conducted recultivation experiments in the presence of the lacI Q repressor. A comparison of the results between the strains, which differed in their average global Ϫ, is shown in Figs. 3 and 5. Importantly, in the absence of transcription, the instability of the TRS was greatly enhanced by the increase of the global Ϫ. After the 3rd subculture step, the amount of the full-length (CGG⅐CCG) 73 as well as the (GAA⅐TTC) 150 fragments was ϳ2 times greater in the SD7 and KD112 strains when compared with the E. coli RS2 and JTT1 strains (Figs. 3 and 5). Thus, regardless of the transcrip-

Influence of transcription and superhelical density on the instability of (CGG⅐CCG) 73 (pRW3691) (A) and (CTG⅐CAG) 98 (pRW3246) triplet repeats (B).
Results of the quantitation are presented. The experiments and data analyses were conducted as described under "Materials and Methods." ϩ indicates active transcription from the lacZ promoter (LB medium), whereas Ϫ designates the inhibition of transcription through the repeat tract by overexpression of the lacI Q repressor from the separate replicon (pI Q -kan). Note the difference in the scale of the y axes between A and B reflecting the general level of instability of the CGG⅐CCG versus the CTG⅐CAG tracts of similar length. topol, topoisomerase; gyr., gyrase. tional state of the plasmid, the global intracellular Ϫ is an important factor in modulating TRS instability.
In summary, these data show that global superhelical tension per se or in concert with transcription has a profound effect on CGG⅐CCG as well as GAA⅐TTC instability. Also, the stability of pRW3246 containing the (CTG⅐CAG) 98 tract was not significantly affected by these factors (data not shown).
Effect of DNA Gyrase Inhibitor Novobiocin on TRS Instability-DNA gyrase is responsible for introducing negative supercoils into DNA (64,65). Because we demonstrated above that lower levels of superhelical density facilitate more stable propagation of plasmids harboring TRS tracts, we analyzed the effect of the specific inhibition of this enzyme on TRS instability. Novobiocin, a coumarin antibiotic, inhibits the assembly of the active gyrase (42,43) and, in contrast to the quinolones, does not lead to the formation of DNA double-strand breaks (44 -47). Because an increase in the global Ϫ stimulates TRS instability, novobiocin should stabilize the repeats through the inhibition of its primary target DNA gyrase.
We analyzed the effect of novobiocin on the instability of the repeats in E. coli JTT1, RS2, and SD7 as well as in the HB101 and DH5␣ strains routinely used for cloning and the stable propagation of plasmids containing the TRS tracts (32, 36, 66). Fig. 6 shows the influence of novo-biocin on the instability of the (CGG⅐CCG) 73 repeat and on the global DNA Ϫ in E. coli HB101 and SD7. Propagation of the plasmids containing (GAA⅐TTC) 150 or (CGG⅐CCG) 73 tracts in medium supplemented with 16 M novobiocin (which is the IC 05 for E. coli HB101) resulted in a statistically significant increase of the TRS stability (Fig. 6, A  and B, and data not shown). Furthermore, this concentration of novobiocin reduced the intracellular Ϫ by more than 5% (from Ϫ0.056 to Ϫ0.054; Fig.  6C, left panel) but had negligible consequences for the viability and growth of the cells. Similar results were obtained using the IC 05 concentration of novobiocin for E. coli JTT1, RS2, and DH5␣ (data not shown).  150 . The restriction fragment derived from the pI Q -kan plasmid expressing the lacI Q repressor is indicated. Experiments were conducted as described under "Materials and Methods" and the legend to Fig. 1. B, the results of quantitation of the amount of undeleted TRS insert (% full-length fragment). topol, topoisomerase; gyr., gyrase. Alternatively, the instabilities of neither CGG⅐CCG nor GAA⅐TTC repeats were affected by the IC 05 concentration of novobiocin (4 M) in E. coli SD7. Likewise, this concentration of the drug did not change the topoisomer distributions in the plasmid isolated from SD7 cells (Fig. 6C,  right panel). This result was expected since the product of the gyrB gene, which is targeted by novobiocin, is mutated in E. coli SD7. Even increasing the novobiocin concentration to 40 M (inhibitory concentration of 80%, IC 80 ) did not influence the TRS instability but caused a significant lengthening of the doubling time of the cells (by ϳ35%). This effect is presumably due to the inhibition of topoisomerase IV, a secondary target for novobiocin (47). Such high concentrations of novobiocin were indeed shown to inhibit the activity of this enzyme, which is involved primarily in decatenation of DNA and removal of positive supercoils in front of a moving replication fork. However, it seems unlikely that inhibition of topoisomerase IV could have influenced our results on TRS instability. Hence, blocking the activity of DNA gyrase by novobiocin decreased the intracellular Ϫ, which then increased the stability of the repeats.
DNA Relaxation Associated with HU Deficiency Increases TRS Stability-The binding of architectural proteins to DNA is an important factor in modulating DNA topology in vivo (29,67). The histone-like HU protein contributes to the plasmid superhelical density in E. coli in two distinct ways; first, it constrains DNA through the physical binding process, and second, it interferes with topoisomerase I and DNA gyrase reactions (29,48,68,69). The removal of HU protein results in partial relaxation of chromosomal and plasmid DNA (48,68). To evaluate the influence of changes in the DNA topology caused by HU deficiency on TRS instability, E. coli JRY880 lacking the hupA gene transformed with pRW3546 containing (GAA⅐TTC) 150 were used. The hupA gene encodes the ␣ subunit of the HU protein, and its deletion dramatically decreases the level of HU in the cells to 5-10% of the wild type amount (68). Because the HU mutants readily acquire suppressor mutations, most likely in the DNA gyrase gene, leading to an increase in the average Ϫ (48), population studies would reflect the effect of both low Ϫ (in the hupA mutants) and high Ϫ (in the hupA mutants containing an additional suppressor mutation). Therefore, single colony analyses instead of the population assays (recultivation assays) were performed ("Materials and Methods"). The JRY880 cells were co-transformed with pRW3546 and a pACYC184 reporter plasmid followed by the simultaneous analysis of the Ϫ (as monitored with pACYC184 topoisomer distributions, Fig. 7A) and TRS instability (as monitored by the deletions of the GAA⅐TTC repeat tract in pRW3546, Fig. 7B) using DNA isolated from single colonies as described under "Materials and Methods." These analyses allow for a direct correlation between Ϫ and TRS instability. Plasmids were isolated from a total of ϳ60 colonies of E. coli JRY880; 30% of the colonies maintained a low Ϫ (see Fig. 7A, lane 1, for a typical example), whereas the remaining 70% of the plasmids demonstrated high Ϫ (JRY880 revertants, Fig. 7A, lane 2). Fig. 7B shows the results of the GAA⅐TTC instability analyses performed on plasmids isolated from 20 colonies of E. coli JRY880; 10 colonies demonstrated low supercoil densities (left side), whereas the other 10 colonies showed high levels of Ϫ (right side), as determined by the analyses of the pACYC184 reference plasmid on the chloroquine-containing agarose gels. The average amount of the full-length progenitor insert was 14 and 2% for cells maintaining low and high levels of Ϫ, respectively. Hence, the Ϫ, rather than the HU protein per se, was the dominant factor for determining the instability of the GAA⅐TTC tract, since the suppressor gyrase mutation increased the Ϫ and destabilized the repeats despite the absence of the HU protein.
The global Ϫ in E. coli JRY880 also had an influence on the extent of deletions. The average size of the GAA⅐TTC tract in the plasmids iso-lated from low Ϫ colonies was 28 repeats longer than for the plasmids propagated in the conditions of high Ϫ (Fig. 7B). Thus, a higher level of Ϫ facilitates the TRS instability by promoting large deletions, spanning several repeat units. This behavior may be related to the propensities of the GAA⅐TTC sequences to adopt more stable triplex structures under higher superhelical tension conditions (Refs. 30, 59, and 62; see "Discussion").
In summary, lowering the superhelical density by eliminating one of the most abundant proteins involved in constraining negative supercoils in E. coli reduces the instability of the GAA⅐TTC repetitive tracts. However, this effect is caused by the change in Ϫ rather than by the absence of the HU protein. These results provide further evidence for the strong association between Ϫ and TRS instability.

DISCUSSION
The role of negative supercoil density in the genetic instabilities of certain disease-associated repeating sequences has been postulated (58,70,71), but the studies described herein present the first proof of this relationship, as transduced by non-B DNA conformations. Statistical mechanical calculations of DNA molecules showed that CGG⅐CCG and CTG⅐CAG repeats begin to writhe at lesser free energy of supercoiling than random, nonrepeating DNA (34,70). Hence, these TRSs were proposed to act as sinks for the accumulation of negative superhelical density. In addition, transcription is also accompanied by the accumulation of positive supercoils in front of the RNA polymerase complex and negative supercoils behind it (72)(73)(74). These waves of supercoiling associated with transcription through the TRS were postulated to influence TRS instability (58).
The effects of the changes in DNA topology on the genetic instabilities of GAA⅐TTC, CGG⅐CCG, and CTG⅐CAG repeats were determined. Three different strategies were utilized: first, we showed that high Ϫ facilitates repeat instabilities by using topoisomerase I and/or DNA gyrase mutants of E. coli where various levels of intracellular Ϫ were maintained. This effect was even more pronounced when transcription throughout the TRS was blocked. Second, specific inhibition of DNA gyrase using novobiocin, which resulted in the relaxation of the DNA, increased the stability of the TRS tracts in the plasmids. Third, we lowered the Ϫ of the target plasmid by eliminating the HU protein, which is involved in constraining the negative supercoiling in E. coli as well as in the regulation of the topoisomerase I and DNA gyrase activity. This resulted in a significant stabilization of the GAA⅐TTC repeats in E. coli hupA mutants. Thus, all three strategies revealed that an increase in the negative superhelical density promotes the instability of triplet repeats.
Hence, we postulate that Ϫ plays an important role in the stimulation of the genetic instability of the TRS tracts by favoring formation of the non-B DNA structures. The extent of the repeat instability can be influenced by both the level of Ϫ and the different propensities of the TRS tracts to adopt the unusual structures. In addition to the global Ϫ, local or transient supercoiling, which is induced (for example by tran-scription), accelerates the formation of non-B DNA structures at the TRS (Fig. 8). It is likely that these changes in the Ϫ affect TRS tracts more than other DNA regions since these sequences (at least CGG⅐CCG and CTG⅐CAG repeats) can locally accumulate a high level of negative supercoiling (34,70).
The formation of stable non-B DNA structures is a critical element in practically all models of microsatellite instability (1,5). Supercoil-dependent DNA conformations can arrest replication fork progression, leading to nicks and/or double-strand break formation (83)(84)(85). These structures can also be recognized and cleaved by repair proteins (Fig. 8, left site) (86 -88). Moreover, stable structures existing on the leading or lagging strands during DNA replication can be bypassed by DNA polymerases (Fig. 8, right site) (16). Hence, all the mechanisms described above would lead to the instability of the TRS tracts.
The effect of the changes in the intracellular Ϫ, regardless of the methods used to promote them, was different depending on the sequence studied. The (GAA⅐TTC) 150 and the (CGG⅐CCG) 73 tracts were sensitive to the changes in the level of Ϫ. On the other hand, the genetic stability of the (CTG⅐CAG) 98 tract was unaffected by both the increase of Ϫ and the stimulation of transcription. These variations in the extent of the Ϫ influence on TRS instability may arise from the different structural properties of the repeat tracts. Relatively short, single tracts of the (GAA⅐TTC) 9 -23 repeats can adopt an intramolecular triplex conformation, and its stability greatly depends on the superhelical density (30). At the high Ϫ, tracts containing only 42 GAA⅐TTC repeats adopt a very stable bi-triplex structure (30). The CTG⅐CAG as well as CGG⅐CCG repeats can adopt hairpin/cruciform conformations (slipped structures) (25,26,56,82). The intrinsic thermodynamic stability of the CGG hairpins is much greater than CTG and especially CAG hairpins (for the structures composed of identical number of repeat units), which may elicit the differences in the Ϫ dependent instability between the CGG⅐CCG and CTG⅐CAG sequences (82,89). Additionally, the CGG hairpins can be fur- FIGURE 7. DNA relaxation caused by HU protein deficiency increases the stability of (GAA⅐TTC) 150 triplet repeats. A, determination of the average superhelical density of pACYC184 isolated from single colonies of the JRY880 cells (lane 1) as well as the JRY880 mutants containing an additional suppressor mutation, resulting in an increase of the intracellular Ϫ (lane 2). Topoisomers were separated in a 1% agarose gel in the presence of 3.5 M chloroquine as described in the legend to Fig. 2. The DNA fragments resulting from the restriction digestion of the pRW3546 with AatII, AflIII, DraIII, PvuI, and NdeI are indicated at the bottom of the gel. The average Ϫ and the size of the GAA⅐TTC tracts were simultaneously analyzed. B, analysis of the genetic instability of the (GAA⅐TTC) 150 triplet repeat tract in 10 JRY880 colonies maintaining a low Ϫ (left side) and a high Ϫ (right side). The single colony analyses were performed as described under "Materials and Methods." M designates the 1-kilobase DNA Ladder (Invitrogen) used as size marker. The arrowheads indicate the full-length DNA containing (GAA⅐TTC) 150 and the approximate gel migration of the insert lacking the entire repeat tract (0 repeats). Analysis of the TRS insert size was performed using BssHII/HindIII digestion followed by radioactive labeling of the recessed ends and electrophoresis of the fragments in 5.5% polyacrylamide gels as described under "Materials and Methods." ther stabilized by Hoogsteen interactions between the G residues in the "folded back" tetraplex structures (63,90).
In summary, we postulate that the formation of stable non-B DNA structures at the TRS tracts driven by the energy of supercoiling stimulates the genetic instability of these sequences. Thus, the results presented in this study are yet another example of the DNA structureinduced DNA mutagenesis.
Biological Significance-Expansions of microsatellites can reach thousands of repeats in some extreme cases (as many as 11,000 tet-ranucleotide repeats, more than 40 kbp of DNA in DM2) (91). How these sequences expand to such a magnitude within a relative small number of cell divisions is still unknown. What is responsible for the differences in instability at the various stages of development? Why do repeats at only one particular locus expand, whereas identical sequences of similar length located on different chromosomes do not? What determines the tissue specificity of the somatic instability events? These and a number of other questions related to microsatellite instabilities remain unsolved (1,5). We propose that the topol- ogy of the chromatin at a particular site might be a predisposing factor.
Chromosomal domains (loop domains) are the basic structural and topological units in eukaryotic chromatin and have been detected in many organisms, including humans (92)(93)(94)(95)(96)(97)(98). These loops range in size from 5 to 200 kbp with an average size of 80 -100 kbp for somatic cells and ϳ30 kbp for male gametes (92)(93)(94)99). The DNA in these domains is attached to the nuclear matrix, or scaffold, to form topologically closed segments (similar to the closed circular plasmids (TABLE ONE)) that have a deficit in linking numbers (superhelical turns) (98,100,101). Domain boundaries are defined by specific DNA regions, scaffold/matrix attachment regions (S/MAR) (97,98). The sizes and organization of these chromosomal loops can undergo dramatic changes during various stages of development and cell differentiation upon induction of gene expression and in different tissues. These changes in the size and organization of the chromosomal loops may modulate the unconstrained superhelical density of a TRS tract. Consequently, these differences in Ϫ may result in very distinct patterns of locus-specific genetic instability observed in the repeat expansion disorders and, hence, account for somatic, tissue-specific TRS instabilities.
The analysis of topoisomerase/gyrase mutants and the use of specific inhibitors of the enzymes involved in regulating DNA topology in vivo together with reliable methods of monitoring the changes in DNA supercoiling made E. coli an appropriate model to study the influence of DNA topology on genome instability. Parallel assays in eukaryotic cells remain to be developed to conduct similar studies in mammalian systems.