Hairpin Formation during DNA Synthesis Primer Realignmentin Vitro in Triplet Repeat Sequences from Human Hereditary Disease Genes*

Genetic expansion of DNA triplet repeat sequences (TRS) found in neurogenetic disorders may be due to abnormal DNA replication. We have previously observed strong DNA synthesis pausings at specific loci within the long tracts (>∼70 repeats) of CTG·CAG and CGG·CCG as well as GTC·GAC by primer extensions in vitro using DNA polymerases (the Klenow fragment ofEscherichia coli DNA polymerase I, the modified T7 DNA polymerase (Sequenase), and human DNA polymerase β). Herein, we have isolated and analyzed the products of stalled synthesis found at ∼30–40 triplets from the beginning of the TRS. DNA sequence analyses revealed that the stalled products contained short tracts of homogeneous TRS (6–12 repeats) in the middle of the sequence corresponding to the flanking region of the primer-template sequence. The sequence at the 3′-side terminated at the end of the primer, indicating that the primer molecule had served as a template. In addition, chemical probe and polyacrylamide gel electrophoretic analyses revealed that the stalled products existed in hairpin structures. We postulate that these products of the DNA polymerases are caused by the existence of an unusual DNA conformation(s) within the TRS, during the in vitro DNA synthesis, enhancing the DNA slippages and the hairpin formations in the TRS due to primer realignment. The consequence of these steps is DNA synthesis to the end of the primer and termination. Primer realignment including hairpin formation may play an important intermediate role in the replication of TRS in vivo to elicit genetic expansions.

Neurogenetic diseases including myotonic dystrophy, fragile X syndrome, Huntington's disease, spinobulbar muscular atrophy, spinocerebellar ataxia type 1, and Friedreich's ataxia have been found to result from expanded triplet repeat sequences (TRS) 1 CTG⅐CAG, CGG⅐CCG, and GAA⅐TTC within their genes (1,2). Also, instabilities (expansions and deletions) of TRS have been associated with limb developmental diseases including human synpolydactyly (3), hypodactyly in mice (4), and hereditary nonpolyposis colon cancer (5). The age of onset and the severity of the neurological diseases are influenced by the lengths of the TRS. Long tracts of TRS are unstable and are prone to change repeat size by expansions and deletions in successive generations. In addition to observations in humans, TRS instabilities have been demonstrated in Escherichia coli (6), and subsequent investigations revealed the influences of genetic factors including host strain (6 -9), DNA replication (6,8), methyl-directed mismatch repair (10), transcription, 2 and nucleotide excision repair. 3 Simple repetitive sequences are known to cause misalignment-mediated DNA synthesis errors that give rise to the instabilities (deletions and expansions) (11)(12)(13)(14). These instabilities are thought to be due to the formation of unusual secondary structures, which can cause frameshift mutations during DNA synthesis (15). Mono-and dinucleotide repeats including Z-DNA-forming sequences promote multiple slippages in templates (12)(13)(14)16). Inverted repeats, which have the potential to form hairpin structures that stabilize misalignments (17), produced a stem-loop DNA structure in pUC19 (18) due to template switching during replication in E. coli (11). Pause sites are hot spots for mutation caused by DNA misalignment (19 -21). Polypurine⅐polypyrimidine tracts including GAA⅐TTC repeats caused pausing (arrest) of DNA polymerases in vitro due to triplex formation (22)(23)(24).
The mechanism of TRS expansion is uncertain. However, it has been postulated that DNA replication errors may cause the expansion of the TRS due to the slippage of DNA complementary strands (6,25,26). The molecular basis of expansions versus deletions of CTG⅐CAG (6) and CGG⅐CCG (7) was explained on the basis of preferential stabilization of loop structures during replication (26). Also, the FMR1 gene containing CGG⅐CCG repeats associated with the fragile X syndrome delayed replication in vivo (27). Similar studies with CTG⅐CAG in yeast agreed with this molecular model (28).
Our previous studies showed that DNA polymerases, including the Klenow fragment of E. coli DNA polymerase I, the modified T7 DNA polymerase (Sequenase), and human DNA polymerase ␤, paused in vitro within certain TRS, including CTG⅐CAG, CGG⅐CCG, and GTC⅐CAG, for lengths of Ͼϳ70 repeats (29). Herein, we report that the stalled products at the distal region of the TRS insert were termination products during DNA synthesis due to primer realignment by the formation of hairpin structures of TRS. Our results indicate that the unusual DNA conformation(s) of the TRS (30, 31) that blocks DNA polymerization may be responsible for the primer realignment and may play an important role in expansion in vivo. Table I lists the plasmids used in this study. pRW1981 contains a segment of the myotonic dystrophy gene (32) with 130 CTG⅐CAG triplet repeats along with 19 and 43 base pairs of the genomic flanking sequences, which was cloned into the HincII site of the polylinker of pUC19 (29,33). pRW3262 was a recloned product of the Sau3AI fragment containing (CTG⅐CAG) 130 of pRW1981 cloned into the BamHI site of pUC19 (29). pRW3306 containing (CGG⅐CCG) 160 was cloned using a head-to-tail dimer of (CGG⅐CCG) 80 , which was derived from the FMR1 gene (34), into the HincII site of pUC19 (7). pRW3416 containing (GTC⅐GAC) 98 was produced using a synthetic oligonucleotide and the in vivo expansion method (6,22). The lengths of the repeat sequences for pRW3306 and pRW3416 were confirmed by DNA sequence analysis. The inserts for pRW1981 and pRW3262 were partially sequenced, and the repeat lengths were estimated by agarose gel electrophoresis. The number of repeats is estimated as Ϯ5 triplets.
3 g of each plasmid DNA was dissolved in 20 l of a solution containing 0.2 M NaOH and 2 g of 5Ј-32 P-end-labeled primer. The mixture was heated for 90 s at 90°C, followed by cooling to room temperature for 4 min, and neutralized with 2 l of 3 M sodium acetate (pH 5.2). After DNA was precipitated by ethanol, the pellet was washed with 70% ethanol and dried under vacuum. The DNA was resuspended in 10 l of a buffer containing 40 mM Tris-Cl (pH 7.5), 50 mM NaCl, 20 mM MgCl 2 , 10 mM dithiothreitol, and 0.5 mM 2Ј-deoxynucleoside triphosphates. Before the extension reaction was performed, the mixture was incubated at 37°C for 10 min. 5 units of the Klenow fragment of E. coli DNA polymerase I (U. S. Biochemical Corp.) was added to the mixture, which was incubated for 10 min at 37°C. The reaction was terminated by the addition of 95% formamide and 20 mM EDTA. The mixture was heated for 4 min at 90°C and chilled on ice. A portion of the reaction mixture was run on a 12% polyacrylamide gel containing 7 M urea.
For the investigations with betaine and the E. coli single-stranded DNA-binding protein (SSB), before adding the Klenow fragment, 5.5 M betaine (Sigma) or SSB (Ͼ98% pure; U. S. Biochemical Corp.) was added to the reaction mixtures to give different concentrations in which 1 or 0.5 g of DNA was used as a template, respectively. The mixtures were incubated at 37°C for 30 min. After DNA elongation reactions, SSB was extracted with phenol prior to gel electrophoresis.
DNA Sequencing and Chemical Modifications-To sequence the stalled products, a total of 24 g of DNA was used for the primer extension reactions as described above (3 g of DNA/reaction). After the reaction mixtures were run on a 12% denaturing polyacrylamide electrophoresis gel, the stalled products (see Fig. 1) were excised and eluted by electrophoresis. The eluates were purified by Sephadex G-50 (medium) and precipitated by ethanol along with an oligomer as a carrier.
The purified stalled products were sequenced by the chemical degradation method (35). For the G reaction, the DNA was incubated with 0.5% dimethyl sulfate (DMS) (Aldrich) for 2 min at 25°C in 200 l of DMS buffer (50 mM sodium cacodylate (pH 7.0) and 1 mM EDTA). For the reactions of purine bases (G and A), the DNA was incubated with 1% formic acid (Fluka) at 37°C for 15 min in 20 l of H 2 O. For the pyrimidine bases (T and C), the DNA was incubated with 60% hydrazine (Aldrich) at 25°C for 9 min in 50 l of H 2 O. The reactions were terminated by the addition of DMS stop buffer (1.5 M sodium acetate (pH 7.0), 1 M ␤-mercaptoethanol, and 250 g/ml yeast tRNA) for the DMS reactions and hydrazine stop buffer (0.3 M sodium acetate (pH 7.0), 0.1 mM EDTA, and 100 g/ml yeast tRNA) for both the formic acid and hydrazine reactions.
Chemical modifications of the stalled products were performed with bromoacetaldehyde (BAA) and diethyl pyrocarbonate (DEPC) (Aldrich). BAA was prepared as described previously (36). For the BAA reactions, the eluted stalled products described above were incubated with 2% BAA in 100 l of DMS buffer for 90 min at 25°C. For the DEPC reactions, the DNAs were incubated with 10% DEPC in 100 l of DMS buffer at 25°C for 30 min. Both reactions were terminated by chilling on ice and washed twice with cold ether.
After recovery by ethanol precipitation, the DNAs were dissolved in 100 l of 10% piperidine (Aldrich) and heated to 90°C for 30 min, and then the piperidine was removed by lyophilization. The DNAs were dissolved in 95% formamide and 20 mM EDTA, heated at 90°C for 4 min, and then chilled on ice. A portion of the sample was fractionated on a 12% denaturing polyacrylamide gel. The bands were visualized by autoradiography.
Electrophoretic Mobility-The purified stalled products described above were dissolved in 10 l of DMS buffer. The DNAs were heated at 90°C for 15 min, followed by cooling to room temperature for 12 h, and subjected to 15% polyacrylamide gel electrophoresis in a buffer containing 45 mM Tris borate (pH 8.3) and 1 mM EDTA at room temperature at 8.3 V/cm. The bands were visualized by autoradiography.

Stalled Products from CTG⅐CAG Contain Homogeneous Short Tracts of TRS as Well as Symmetric Flanking
Sequences-Our previous studies suggested that a temperature-sensitive unusual DNA conformation(s) (30, 31) might be formed within long tracts of CTG⅐CAG, CGG⅐CCG, and GTC⅐GAC repeats that arrested DNA synthesis in vitro (29,37). To further investigate this idea, we isolated some of the stalled products from the gels shown in Fig. 1 and performed chemical DNA sequence analyses.
For (CTG⅐CAG) 130 in pRW1981, three strong stalled products were observed at sites corresponding to the G residues in the 37th, 39th, and 41st CAG triplets, respectively ( Fig. 1). DNA sequence analyses ( Fig. 2A) revealed that the stalled product at the 37th repeat contained 8 repeats of CAG adjacent to the 5Ј-flanking sequence (88 nt). In addition, the complementary sequence (87 or 88 nt; described below) to the 5Ј-flanking region was contiguous on the 3Ј-side of (CAG) 8 ( Fig. 2A), and the product molecule ended with the sequence complementary to the primer. The stalled products at the 39th and 41st repeats also contained homogeneous (CAG) 10 and (CAG) 12 , respectively, that were contiguous with the 5Ј-flanking and the primer-complementary sequences ( Fig. 2A). We are not certain that the 3Ј-flanking sequence contained an A residue adjacent to G 5 . However, the A residue might not be incorporated since the subtraction of 24 nt (the length of the synthesized (CAG) 8 ) from 111 nt (the length between the first pausing site and the beginning of the CAG triplet) was 87 nt instead of 88 nt, which was the known length of the flanking region ( Fig. 1).
Since the distance between the primer-binding site and the beginning of the triplet repeat units influenced the length of the stalled product (29), we analyzed pRW3262. For pRW3262, four strong arrest sites were observed at the G residues in the 29th, 31st, 33rd, and 35th CAG triplet units from the beginning (Fig. 1). The first three of these products were identified to contain (CAG) 8 , (CAG) 10 , and (CAG) 12 , respectively, as well as the 5Ј-flanking (63 nt) and the primer-complementary (63 nt) sequences (Fig. 2B). The other stalled product (35th unit) was not sequenced due to poor recovery. The three arrest products observed for pRW1981 and pRW3262 contained (CAG) 8 , (CAG) 10 , and (CAG) 12 , indicating that our previous observations on the dependence of the distance between the location of the primer and the TRS were due to the different lengths of the flanking sequences. For pRW1981 and pRW3262, another strong arrest product was found (open arrowheads in Fig. 1). DNA sequence analyses revealed that these stalled products contained (CTG) 20 , and the reactions were extended through the 3Ј-flanking sequences of the plasmids, indicating that the products were derived from the deletion of the plasmids containing (CTG⅐CAG) 130 . In fact, these plasmid preparations were found to contain small amounts (ϳ15%) of deletions. Although the products prematurely terminated by DNA polymerase were observed when the CAG strand was the template for CTG⅐CAG (29), we could not identify the stalled products due to their weaker intensity.
Stalled Products from CGG⅐CCG-In the case of the CGG⅐CCG repeats in pRW3306, the paused products at the 30th, 31st, and 32nd CGG triplets also contained symmetric sequences in the 5Ј-flanking (73 nt) and 3Ј-flanking (72 nt) regions, along with the triplet repeat units, which were composed of (CGG) 6 , (CGG) 7 , and (CGG) 8 , respectively (Fig. 2C). The three arrest sites were identified at the 3rd nucleotide (CGG) in the CGG triplet unit (Fig. 1); hence, the 18 nt in the (CGG) 6 tract for the 30th product was the stalled site. The distance between the primer and the beginning of the triplets was 73 nt, and thus, there were 91 nt from the primer to the arrest site. However, the 3rd nucleotide (CGG) in the 6th CGG unit formed a base pair with the C residue in the 5Ј-flanking sequence. Therefore, the difference of 1 nucleotide between the 73-nt 5Ј-flanking and 72-nt 3Ј-flanking regions was due to the base pairing of the two complementary sequences in the hairpin (discussed below (Fig. 4B)).
Stalled Products from GTC⅐GAC-For pRW3416 containing (GTC⅐GAC) 98 , the arrest product at the G residue in the 28th CGA triplet unit (Fig. 1) was composed of (CGA) 7 CG, the 60-nt 5Ј-flanking sequence, and its complementary 60-nt sequence on the 3Ј-side (Fig. 2D). This result indicates that the 8th A residue was not incorporated during DNA synthesis, suggesting that DNA polymerase was arrested at the G residue in the 7th CGA unit, which was similar to our observations for the CTG⅐CAG and CGG⅐CCG repeats, where DNA polymerase arrested at the G residues ( Figs. 1 and 2). This result was consistent with the observations that the 83 nt between the beginning of the triplets and the arrest site were composed of (CGA) 7 CG (23 nt) and the 3Ј-flanking sequence that was complementary to the 60-nt 5Ј-flanking sequence. The stalled products at the 29th, 30th, and 32nd triplet units (Fig. 1) were not sequenced due to their scarcity.
In summary, these results revealed that the stalled products contain short tracts of homogeneous TRS and the symmetric flanking sequences, suggesting that they were produced by the termination of DNA synthesis due to template switching (see "Discussion").
Formation of Hairpin Structures-The formation of hairpin structures with duplex antiparallel conformations has been proposed for oligonucleotides containing single-stranded CTG, CAG, CGG, CCG, GTC, and GAC repeats with varying degrees of stability from thermodynamic, electrophoretic gel mobility, and NMR studies (38 -49). Thus, we conducted investigations to determine if the stalled products, which had the potential to exist as hairpins as revealed by DNA sequence analyses, were, in fact, hairpins or were in other conformations (i.e. linear duplexes). Polyacrylamide gel electrophoretic analyses showed that the stalled products that had been heat-denatured and then annealed migrated as single bands; the size of these products was almost identical to the duplex length formed by the 5Ј-and 3Ј-halves of the products (data not shown). These results suggested that the stalled products formed duplex hairpin structures.
Chemical and enzymatic probe analyses (38 -42) were used to identify hairpin DNA conformations at the base pair level in oligomers containing TRS, making use of the specific reactivities for the stems and for the single-stranded loops. Therefore, we performed analyses using BAA and DEPC to identify the structures formed in the homogeneous TRS stalled products: BAA reacts specifically with adenines and cytosines in singlestranded DNA (50); DEPC reacts at the N-7 positions of adenines and guanines in single-stranded DNA (50). For the stalled products containing CAG repeats, only the C base in the 5th CAG triplet at the 37th stalled site (Fig. 1), which was identified to contain (CAG) 8 (Fig. 3A), was hypersensitive to BAA, but the rest of the C and A bases in (CAG) 8 as well as the flanking sequences were not specifically modified. However, the DEPC modifications were observed at all A bases in (CAG) 8 , except for the A base in the 8th unit, as well as at the G base in the 4th unit; the A bases in the 4th and 5th repeats reacted strongly, whereas the A bases in the other repeats (1st, 2nd, 3rd, 6th, and 7th repeats) reacted less strongly. For the stalled products at the 39th and 41st repeats, which contained  Fig. 1 were eluted from the gels. The purified DNAs were modified by either BAA or DEPC as described under "Experimental Procedures." The 5Ј 3 3Ј direction is from the bottom to the top of the gel. GϩA represents chemical sequencing reactions. The numbers on the top of each gel designate the triplet units at the stalled sites shown in Fig. 1. (CAG) 10 and (CAG) 12 , respectively, BAA strongly modified the C bases in the 6th and 7th triplet units, respectively, and DEPC was hyper-reactive at the middle A bases and the G base in the 5th and 6th units, respectively. These results indicate that the hairpin loops were composed of 4 bases (AGCA) and that the hairpin stems were composed of the other residues in the CAG repeats and the symmetric flanking regions (Fig. 4A). However, the strength of the A⅐A mismatch pairings in the hairpin stems for the CAG repeats was dependent on the repeat lengths: A⅐As embedded in the stems nearest the G⅐C tracts were stronger than the A⅐As nearest the loops (Fig. 4A). Hence, the longer lengths of CAG repeats probably form more stable hairpins.
For CGG⅐CCG, BAA specifically modified the C bases in the 4th CGG units at the 30th and 31st arrested sites and the C base in the 5th unit at the 32nd stalled site, whereas neither of the other C bases in the CGG repeats nor the C and A bases in the flanking regions were modified (Fig. 3B). DEPC weakly modified the G bases in the 3rd and 4th CGG units at the 30th and 32nd stalled sites, respectively. However, at the 31st arrest site, the 2nd G base in the 3rd CGG unit (CGG) was strongly modified, and the G bases in the 4th CGG unit were weakly reactive (Fig. 3B). The rest of the G bases in the CGG repeats as well as the G and A bases in the flanking regions were not sensitive to DEPC. These results indicate that, unlike the case of the CAG repeats described above, the loop structures were different depending on the repeat lengths. In addition, G⅐G mismatch base pairings were formed to give duplexes in the stems (Fig. 4B), in which the Hoogsteen base pairing involved atoms at O-6 and N-7 of the guanine in the syn-conformation with atoms at N-1 and N-2 of the anti-paired guanine (51). In fact, the DMS sequencing reactions, which attack N-7 of guanine (50), protected the G bases in the CGG repeats of the stem regions to a certain extent (Figs. 2C, 3B, and 4B).
For GTC⅐GAC, at the 28th and 30th stalled sites, similar modifications were observed by BAA and DEPC (Fig. 3C), except for the difference in the stem structures due to the repeat lengths. BAA modifications were observed in the middle of the GTC repeats; the C base in the 5th repeat was strongly modified, and the C and A bases in the 4th repeat were slightly modified at the 28th stalled site. The A bases in the middle region (3rd, 4th, and 5th repeats) were also hypersensitive to DEPC. The rest of the A bases were weakly modified. These modifications indicate the presence of unpaired A bases in the stems, and the loop regions contained 7 bases (Fig. 4C). This larger hairpin loop structure than observed for CTG⅐CAG and CGG⅐CCG (Fig. 4) is probably a manifestation of the different stabilities for the hairpin structures. At the 29th arrest site, BAA modification was observed only at the C base in the 5th repeat, and strong DEPC modifications were observed at the A base in the 4th repeat and at the G and A bases in the 5th repeat, indicating a different loop structure from that found at the 28th and 30th stalled sites (Fig. 3C).
In summary, chemical probe analyses of the stalled products revealed that different types of loop structures formed in the hairpins as well as the stem structures depending on the TRS.
Effect of Secondary Structure-destabilizing Agents on Stalling-Betaine alters DNA stability due to reduction of the base pair composition dependence on DNA thermal melting transitions (52,53). 2 M betaine in the DNA primer extension reactions effectively suppressed pausing by DNA polymerases in G⅐C-rich regions (52). Thus, we performed polymerase studies to investigate if betaine reduces or eliminates the arrest sites. The primer-template complexes for (CGG⅐CCG) 160 in pRW3306 were incubated in the presence of 2 M betaine at 37°C for 30 min before the DNA elongation reactions by the Klenow fragment; the stallings were not significantly influenced (data not shown). Other studies with higher concentrations of betaine (3.3 and 5.2 M) were not informative due to the general inhibition of DNA polymerization. Hence, betaine did not alter the secondary structure (30,31) in the TRS, in agreement with the observation that the stallings were heat-resistant up to 90°C.
Furthermore, we also investigated the effect of SSB on the stallings. SSB is known to prevent the formation of secondary structures such as hairpins and triplexes, which cause the pausing of DNA polymerases (24,54). Also, SSB stabilizes plasmids containing long tracts of TRS in E. coli (9). Different amounts (1, 2, and 10 g) of SSB were preincubated at 37°C for 30 min with the primer-template complexes for (CGG⅐CCG) 160 in pRW3306. The addition of 10 g of SSB decreased the stallings by 70%, whereas 1 or 2 g showed an ϳ20% reduction (data not shown). These results indicate that, in contrast to betaine treatment, SSB may stabilize TRS by preventing the formation of DNA secondary structures (9).

DISCUSSION
Stalling of DNA Synthesis-This study revealed that the strongly arrested in vitro products of DNA polymerases (the Klenow fragment of E. coli DNA polymerase I and human DNA polymerase ␤), which were observed in the region distal (ϳ30 -40 triplet units) from the beginning of long tracts of CTG⅐CAG, CGG⅐CCG, and GTC⅐GAC, contain short tracts of homogeneous TRS embedded in the middle of symmetric flanking sequences. Hence, these products have the capability to form hairpin structures. Thus, we propose that the strong stallings were caused by termination of DNA synthesis due to the hairpin formation of the TRS followed by primer realignment (Fig. 5). First, the DNA polymerase might encounter a flexible and writhed TRS (30,31), which impedes the progression of DNA synthesis. Pausings in the proximal regions of the TRS (ϳ12 triplet units) (brackets in Fig. 1) could be intermediate paused products during the DNA synthesis; however, since they have not been sequenced, their identity is uncertain. Second, the idling of the impeded DNA polymerase might enable DNA slippage in the nascent strand. Third, the hairpin which was formed might allow for primer realignment, creating a functional primer 3Ј-end. Fourth, as the template is now switched, the DNA chain would be elongated using the nascent strand as a template. Finally, DNA synthesis will be terminated at the end of the primer-nascent strand molecule. Hence, the stalling of DNA polymerase in the proximal region (ϳ12 triplet units) caused by an unusual TRS conformation (30,31) is critical for template switching.
Unusual DNA structures including triplexes and hairpins cause the premature termination by DNA polymerases in vitro (23, 24, 54 -56). Our previous study showed that doublestranded (not single-stranded) DNA is required for the termi-  (30,31). The idling of the polymerase results in dissociation of the nascent strand of the TRS from the template, followed by DNA slippage to form metastable hairpin structures. After primer realignment, DNA synthesis continues using the nascent strand as a template through the 5Ј-end of the primer, which results in the termination of DNA synthesis to give a hairpin structure as product. nation (29). Long duplex tracts of CTG⅐CAG and CCG⅐CGG are more flexible and writhed than random sequence DNA (30,31), which may cause the aberrant DNA polymerization reactions. We observed that SSB reduced the stallings, probably by preventing the formation of the flexible and writhed conformation(s). Alternatively, the stallings were not influenced by betaine or heat treatment (29), which indicates the thermodynamic stability of the flexible and writhed structure.
Hairpin Formation in TRS-The discovery of hairpins formed in vitro is important for confirming, at least partially, our in vivo model of genetic instabilities (6,11,28,38,43,57). Other investigations with relatively short oligonucleotides postulated models for hairpin structures of CTG⅐CAG, CGG⅐CCG, and GTC⅐GAC repeats on the basis of NMR, thermodynamic, and polyacrylamide gel electrophoretic studies (38 -49). Moreover, Mitas et al. (38 -41) have used enzymatic and chemical probe analyses to identify the single-stranded regions in the hairpins. However, our structures (Fig. 4) are somewhat different from those proposed previously (38 -41), which may be due to the conditions investigated (physical properties versus DNA polymerase reactions). Also, our investigations provided information on the minimum length appropriate to adopt hairpin structures during DNA synthesis. In the case of CAG repeats in the nascent strand when CTG was the template, (CAG) 8 , (CAG) 10 , and (CAG) 12 were the terminated products, indicating their stability as hairpin structures. The other repeat lengths, such as (CAG) 6 , (CAG) 7 , (CAG) 9 , (CAG) 11 , and (CAG) 13 , might be too unstable to form hairpin structures. This could explain why the CAG repeat units differed in length by 2 repeats.
For CGG⅐CCG, the terminated products contained (CGG) 6 , (CGG) 7 , and (CGG) 8 . The distance between the two stallings was 1 CGG unit, instead of 2 as observed for CTG⅐CAG, indicating that the capacity to form hairpins for CGG repeats was not influenced by the different repeat lengths even though two different loop structures were found (Fig. 4B). However, the hairpin structures might be influenced by the flanking sequences since the last incorporated G nucleotide was paired with the C nucleotide from the 5Ј-flanking sequence. Therefore, we cannot rule out the formation of another type of hairpin structure that is composed of G⅐G mismatches between the 2nd G bases (39,43) in the CGG unit instead of the 3rd G bases.
Processivity of DNA Polymerases-The termination of DNA synthesis was dependent on the processivity of the DNA polymerases as well as temperature. The Klenow fragment of E. coli DNA polymerase I and human DNA polymerase ␤ produced strong terminated products in the distal region of the TRS insert, whereas Sequenase showed stallings principally in the proximal region (29). It is plausible that since the Klenow fragment and human DNA polymerase ␤ have low processivity (58,59), they stalled upon encountering the flexible and writhed conformation (30,31), resulting in idling of the DNA polymerase, which allows the template switching observed in this study. On the other hand, even though Sequenase was arrested in the proximal TRS region (29), it could pass by the unusual DNA conformation due to its high processivity (58,59). As expected, when we studied the processive Taq polymerase at 70°C in a parallel experiment, no distal stalling was observed (data not shown).
Possible Stalling In Vivo-Plasmids containing inverted repeats produced a stem-loop DNA structure during DNA replication in E. coli due to template switching (11,18). We have shown that certain TRS produced similar stem-loop structures during DNA synthesis in vitro; hence, replication may be terminated even in vivo. In fact, the growth kinetics of E. coli harboring plasmids containing CTG⅐CAG repeats are influenced by the repeat length; as the length of CTG⅐CAG repeats is increased, the growth rate is reduced (8). 2 Thus, replication of the TRS in vivo may be subject to the stalling of DNA synthesis, which could result in the slower growth.
The stalling by DNA polymerases may play an important role in the expansion process in vivo. As an alternative to the third line in Fig. 5, if the template was not switched but the DNA polymerase continued to copy the original template strand, expansion of the product molecule would ensue. Hence, certain types and lengths of TRS in human chromosomes may cause the stalling of DNA polymerases, which promotes genetic expansion.