Hairpin formation in the template or nascent strands during replication can generate genomic instability by contraction or expansion, respectively, of the number of microsatellite sequence repeats (Figure 1). To investigate the influence of DNA replication on TNR stability in a human chromosomal context, we used Flp recombinase to generate clonal cell lines containing 12 or 102 (CTG)•(CAG) repeats alongside the wild type c-myc replication origin, or alongside a mutant c-myc origin inactivated by deletion of the DNA unwinding element (ΔDUE) 20–22 (Figure 2a). Flp recombinase targeted cells are collectively referred to as (CTG)_n•(CAG)_n cells, while individual cell lines are denoted according to the trinucleotide sequence in the lagging strand template when replicated from the flanking c-myc origin 20–22. Thus, (CTG)₁₀₂ cells contain 102 (CTG) repeats in the lagging strand template for replication from the proximal c-myc origin. The single copy integration of each (CTG)•(CAG) cassette at the Flp recombinase target (FRT) site was confirmed by PCR and Southern hybridization (Supplementary Figure 1).

Approximately 25 population doublings after clonal selection, PCR amplification with primers flanking the repeat tracts detected no instability at the (CTG)₁₀₂•(CAG)₁₀₂ TNR (Figure 2b); the lower mobility shadow bands observed above the ~450 bp amplification product of the (CTG)₁₀₂•(CAG)₁₀₂ progenitor sequence are slipped strand artifacts of reannealing during the PCR 6, as demonstrated by re-amplification of the linear ~450 bp product band (Supplementary Figure 2). Strikingly, after about 250 population doublings, the (CTG)₁₀₂ and (CAG)₁₀₂ cells showed clear patterns of instability that differed for the two TNR orientations (Figure 2c). The Flp targeted cell lines are clonally derived thus, as confirmed by PCR of early passage cells, the variation in PCR product size following prolonged culture did not arise from length heterogeneity of the progenitor allele.

The (CTG)₁₀₂ cells displayed predominantly contractions, while the (CAG)₁₀₂ cells displayed both expansions and contractions (Figure 2c). Instability was not detected in (CTG)₁₂ or (CAG)₁₂ cells after 25 or 250 population doublings (Figure 2d), or when (CTG)₁₀₂ or (CAG)₁₀₂ TNRs were integrated next to the inactivated ΔDUE c-myc origin and grown for 250 population doublings (Figure 2e). These results indicate that (CTG)•(CAG) TNR instability depends on repeat tract length at this locus. The dependence of microsatellite expansion and contraction on proximal origin activity, and the effect of TNR orientation relative to the origin argue that the pattern of instability in this model system is also replication dependent.

Treatment of DM1 patient cells in culture with the replication inhibitors emetine or aphidicolin increased (CTG)•(CAG) instability at the DMPK locus 12. To test the effect of replication inhibitors on (CTG)•(CAG) stability at the ectopic c-myc origin, (CAG)₁₀₂ or (CTG)₁₀₂ cell lines at approximately 25 population doublings were treated with aphidicolin, emetine or Fen1 siRNA. When cells were treated with low dose aphidicolin to slow replication fork movement without inducing cell cycle arrest there was a clear orientation-dependent induction of TNR instability, with (CTG)₁₀₂ cells showing primarily contractions and (CAG)₁₀₂ cells showing both expansions and contractions (Supplementary Figure 3a). The patterns of instability in (CTG)₁₀₂ cells and (CAG)₁₀₂ cells treated with aphidicolin at ~25 population doublings were similar to the patterns of instability observed when these cells were maintained in long term (~250 pd) culture, suggesting that the effects of aphidicolin are qualitatively similar to those occurring in the absence of drug treatment.

Aphidicolin inhibits both leading and lagging strand DNA polymerases, whereas emetine preferentially inhibits Okazaki fragment synthesis 12,23. Similar to the patterns of instability induced by aphidicolin treatment, emetine treatment of (CTG)₁₀₂ cells led primarily to contractions of the (CTG)₁₀₂•(CAG)₁₀₂ TNR whereas (CAG)₁₀₂ cells showed a strong tendency towards expansions (Supplementary Figure 4). These results suggest that emetine-induced instability is due to hairpins formed by the lagging strand template (CTG) repeats or leading strand template (CAG) repeats. Following this reasoning, (CTG) hairpins in the leading strand nascent DNA may account for the distinct pattern of expansions seen in emetine treated (CAG)₁₀₂ cells. Sequencing of expanded or contracted DNA bands after emetine treatment (Supplementary Figure 4) confirmed that these bands contain uninterrupted (CTG)•(CAG) sequences, indicating that emetine-induced instability occurs by an error-free mechanism. The differential effects of emetine on (CTG)₁₀₂ and (CAG)₁₀₂ cells were also evident in flow cytometry, where the drug induced a sizeable increase in the sub-G1 population of (CTG)₁₀₂ cells (Supplementary Figure 5).

The effects of emetine on TNR stability are apparently not due to downregulation of Fen1 nuclease (Supplementary Figure 5), which acts during Okazaki fragment maturation and DNA repair synthesis to remove 5’ single-stranded flaps before ligation. Similar to the effects of emetine treatment however, siRNA knockdown of Fen1 led to a large increase in contractions in (CTG)₁₀₂ cells, but fewer large expansions in (CTG)₁₀₂ and (CAG)₁₀₂ cells than aphidicolin or emetine (Supplementary Figure 6). Collectively, the results of replication inhibition suggest that slowing of replication fork movement promotes TNR instability, with (CTG)₁₀₂ repeats in the lagging strand template favoring contraction, and (CTG)₁₀₂ repeats in the leading strand nascent DNA favoring expansion.

The patterns of instability induced by these replication inhibitors are consistent with the predicted stability of (CTG) hairpins vs. (CAG) hairpins 24 resulting in a greater frequency of (CTG) hairpins in the lagging strand template. Nevertheless, the stronger tendency towards contractions in (CTG)₁₀₂ cells vs. (CAG)₁₀₂ cells could be a result of hairpin formation in the (CAG) leading strand template 19. To determine whether leading or lagging strand templates could form hairpins in vivo, we decided to interrogate these proposed structures by the expression of synthetic nucleases.

We designed zinc finger nucleases (ZFNs) that could be expressed in human cells to test directly for replication-dependent formation of hairpins in vivo. The engineered ZFNs contain the sequence nonspecific Fok 1 restriction endonuclease catalytic cleavage domain (Fok 1_CD) fused to three tandem zinc finger protein domains that each recognize the (GCT) (ZF_GCT zinc finger) or (AGC) (ZF_AGC zinc finger) trinucleotides to produce, respectively, nucleases that target (CTG) hairpins (termed ZFN_CTG), or (CAG) hairpins (termed ZFN_CAG). Like Fok 1, ZFNs dimerize through their Fok 1 cleavage domains only when bound to DNA, and this dimerization is necessary for catalytic activity 25,26. Whereas co-expression of ZFN_CTG and ZFN_CAG is predicted to be necessary for cleavage of B-form (CTG)•(CAG) DNA (Figure 3a), we reasoned that the requirement for cleavage domain dimerization would preclude hydrolysis of B-form DNA by a single ZFN but allow digestion of (CTG) and (CAG) hairpin structures, respectively, by the individual ZFN_CTG and ZFN_CAG nucleases (Figure 3b).

Previous in vitro characterization of binding by individual zinc finger domains indicates that certain (GCT) zinc fingers may recognize (GCA) triplets and that some (GCA) zinc fingers may recognize (GCT) triplets 27,28. To assess the specificity of the ZFNs these nucleases were assayed in the in vitro system shown to be more permissive for nuclease digestion than in vivo 28,29. ZFN_CTG and ZFN_CAG were expressed individually in HeLa cells and immunoprecipitated through their FLAG epitope tags (Supplementary Figure 7). The linear (CTG)₁₀₂•(CAG)₁₀₂ PCR product band was reamplified and used as a substrate for cleavage by the immunoprecipitated nucleases. Consistent with the predicted specificity of these enzymes, the individual ZFNs cleaved the slow mobility, slipped strand PCR product structures 6 and a fraction of the intermediate mobility slipped strand structures, but did not cleave the (CTG)₁₀₂•(CAG)₁₀₂ B-form DNA during extended incubation (Figure 3c). Shorter time course digestions revealed intermediates in the slipped-strand digestion process (Supplementary Figure 8). In contrast, the mixture of ZFN_CTG and ZFN_CAG quantitatively cleaved the (CTG)₁₀₂•(CAG)₁₀₂ B-form DNA in addition to digesting the slipped strand structures (Figure 3d).

The resistance of a significant fraction of the template to digestion by the individual nucleases argued that the ZFNs were specific and that the extracts did not display nonspecific nuclease activity. To confirm this, extracts were prepared from HeLa cells expressing the same zinc finger proteins (ZFPs) omitting the Fok I nuclease domain. We observed that the ZFP containing extracts did not digest the (CTG)₁₀₂•(CAG)₁₀₂ DNA substrate (Supplementary Figure 8) or (CNG)₄₆ synthetic hairpin substrates (Supplementary Figure 9). Furthermore, the ZFN_CTG immunoprecipitate digested the (CTG)₄₆ hairpin substrate but not the (CAG)₄₆ hairpin substrate, while the ZFN_CAG immunoprecipitate digested the (CAG)₄₆ substrate but not the (CTG)₄₆ hairpin substrate.

ZFN_CTG or ZFN_CAG were expressed in (CTG)₁₀₂ or (CAG)₁₀₂ cells and genomic DNA was isolated for spPCR amplification 48 hours later. As shown in Figure 4, expression of either ZFN_CTG (Figure 4a) or ZFN_CAG (Figure 4b) in (CTG)₁₀₂ cells led to a decrease in the length of the TNR tracts. These results indicate that both the lagging strand (CTG) template and the leading strand (CAG) template strand can form hairpin structures that produce contractions in vivo. Treatment of (CAG)₁₀₂ cells with ZFN_CTG or ZFN_CAG also resulted in shortening of the TNR tracts, reinforcing the view that either template strand can form hairpins in vivo.

It is formally possible that the decrease in TNR repeat length following ZFN expression is due to ZFN binding to (CTG)•(CAG) hairpins that subsequently induces cleavage by cellular nucleases. We addressed this question further by expressing the zinc finger proteins ZFP_CTG and ZFP_CAG in (CTG)₁₀₂ or (CAG)₁₀₂ cells. Omission of the Fok I nuclease domain in ZFPs resulted in significantly decreased frequencies of TNR contraction (compare Figures 4a, b, d). Based on the Fok I nuclease domain-dependent difference in TNR PCR patterns between cells expressing ZFPs and ZFNs, we conclude that the ZFNs directly cleave hairpin structures in vivo.

The progenitor (CTG)₁₀₂•(CAG)₁₀₂ band was preserved when cells expressed either ZFN individually, while the (CTG)₁₀₂•(CAG)₁₀₂ band was completely digested when either (CTG)₁₀₂ or (CAG)₁₀₂ cells were co-transfected with a mixture of ZFN_CTG and ZFN_CAG expression plasmids (Figure 4c). In contrast to the deletions that were detected by spPCR after single ZFN expression, expansions were detected after combined expression of ZFN_CTG and ZFN_CAG (although these are underrepresented by the spPCR). Significantly, this suggests that digestion by a homodimeric ZFN is repaired differently than digestion by a heterodimeric ZFN. Inasmuch as there is only one copy of the ectopic origin/TNR integrant per genome, we postulate that the expansions that occur after heterodimeric ZFN digestion arise by homology directed repair of double strand breaks using the misaligned sister chromatid TNR during S or G2 phase, whereas single strand gaps are formed by homodimeric ZFN hairpin excision, and are repaired by primer extension.

The resistance of the (CTG)₁₀₂•(CAG)₁₀₂ band to digestion by a single ZFN supports the view that the ZFNs selectively target hairpin structures. To test whether the sensitivity of the (CTG)₁₀₂•(CAG)₁₀₂ TNR to ZFN digestion is dependent on cell division, ZFN_CTG or ZFN_CAG was expressed in (CTG)₁₀₂ or (CAG)₁₀₂ cells during serum starvation. Digestion by ZFN_CTG or ZFN_CAG was markedly reduced in serum deprived cells, whereas expression of the heteromeric pair of ZFNs eliminated the (CTG)₁₀₂•(CAG)₁₀₂ TNR band almost completely, and generated expansion products upon spPCR (Figure 5). These data underscore the conclusion that the individual ZFNs selectively target non-B DNA structures, and that hairpin formation is replication-dependent. We also constructed (CTG)₄₅•(CAG)₄₅ FRT targeted cell lines, and found that these cells showed similar sensitivity to the individual ZFNs and insensitivity to ZFPs as the (CTG)₁₀₂•(CAG)₁₀₂ cells (Supplementary Figure 10), indicating that premutation length TNRs can form hairpin structures in vivo.

(CTG)₁₂•(CAG)₁₂ cells did not show time-dependent instability, and did not show induction of instability when treated with aphidicolin, emetine, or Fen1 siRNA (Figure 6). (CTG)₁₂•(CAG)₁₂ cells were also tested for their sensitivity to ZFN_CTG and ZFN_CAG. Because the predicted minimum binding site for a dimeric ZFN is eight trinucleotide repeats, the stems of the TNR hairpins in these cells should be too short to allow nuclease binding and dimerization, while the B-form (CTG)₁₂•(CAG)₁₂ TNR could theoretically bind dimeric ZFNs. Indeed, the individual ZFNs did not cleave the (CTG)₁₂•(CAG)₁₂ cell TNRs, but the co-expressed ZFN pair could digest the TNRs. Presumably because there are fewer possible binding configurations that would accommodate the dimeric nuclease, digestion was less efficient on the shorter repeat tracts. Considered together these data corroborate the conclusion that spontaneous or drug-induced hairpin formation and TNR instability are length-dependent phenomena.

(CTG)•(CAG) trinucleotide repeats were amplified from genomic DNA or plasmid clones by PCR and inserted into the Not I restriction site at the 3’ edge of the c-myc replicator plasmids pFRT.myc or pFRT.myc. ΔDUE 21. DMPK flanking DNA in the TNR insert comprise the 5' sequence TTGTAGCCGGGAATGCTG and the 3' sequence TGGTCTGTGATCCCCC. All primer sequences used in this work are available from the authors. For cloning of TNR PCR products, bands were separated by electrophoresis on a 2% agarose gel and extracted. The purified PCR products were subcloned into the Not I site of pFRT.myc and sequenced.

HeLa/406 acceptor cells containing a single chromosomal Flp recombinase target were constructed, grown, and transfected as described 20. Transfectants were subjected to drug selection using G418 and ganciclovir 21. Cells were grown at 37 °C and 5% CO₂ in DMEM supplemented with 10% newborn calf serum. (CTG)_n•(CAG)_n cells were treated with emetine or aphidicolin as described 12. Approximately 5×10⁵ cells were grown in DMEM with 1 µM emetine (18 hr) or 0.2 µM aphidicolin (18 hr), and allowed to recover in drug-free medium for two population doublings. One cycle of treatment includes drug exposure and recovery. This treatment was repeated for a total of 5 cycles. Control experiments were performed using drug free medium. (CTG)₁₀₂•(CAG)₁₀₂ cells (30% confluent) were transfected with 50 nM Fen1 siRNA in 6-well plates every 48 hr for a total of 3 transfections. Mock experiments were carried out using non-specific siRNA. Whole-cell lysates were prepared from treated or untreated HeLa cells and probed using antibodies against human Fen1 (Santa Cruz) or Mcm7 (a gift from Dr. Aloys Schepers, GSF-Haematologikum).

The single zinc finger plasmids pc3XB-ZF72, pcXB-ZF83, and a mammalian expression vector, pST1374, were purchased from Addgene Inc. (Cambridge, MA). Proteins with three zinc fingers (ZFPs) were assembled using standard molecular cloning methods. ZFP_CTG, containing three ZF72 zinc fingers (Addgene plasmid 13206), and ZFP_CAG, containing three ZF83 zinc fingers (Addgene plasmid 13217), were designed with Zinc Finger Targeter (ZiFiT) Version 2.0 49. ZFPs cDNAs were subcloned into the mammalian expression vector pST1374 (Addgene plasmid 13426) between the coding sequences of the FLAG tag and Fok I DNA cleavage domain to generate zinc finger nucleases (ZFNs). Plasmids expressing ZFPs were constructed by introducing a stop codon at the Bam HI restriction site between zinc finger domain and Fok I cleavage domain in plasmid pST1374ZFN_CTG or pST1374ZFN_CAG to generate pST1374ZFN_CTG or pST1374ZFP_CAG respectively.

For in vitro digestion, ZFP and ZFN proteins were expressed in HeLa cells and immunoprecipitated using EZview Red Anti-FLAG M2 affinity gel (Sigma). ZFN and ZFP proteins were expressed from the same vectors at similar levels (Supplementary Figure 4). The in vitro digestion 29 substrates were produced by PCR amplification of linear (CTG) ₁₀₂•(CAG)₁₀₂ DNA or by annealing of TNR oligonucleotide 5’-TCCTTGTAGCCGGGAATG(CNG)₄₆GGGGGATCACAGACCATT-3’ to 5’-AATGGTCTGTGATCCCCCAAAACATTCCCGGCTACAAGGA-3’ (where (CNG)₄₆ stands for (CTG)₄₆ or (CAG)₄₆) to form hairpin structures. The digestion products were analyzed by electrophoresis in 7.5% polyacrylamide gels. For in vivo digestion ZFN or ZFP plasmid DNAs were transfected into the indicated cell lines at 50% confluence in 6-well plates twice every 72 hr. Cells were harvested 72 hr after the second transfection, and genomic DNA was isolated for PCR.

Standard PCR used 40 ng genomic DNA per reaction. Amplification conditions were 94 °C (5’); 35 cycles of 94 °C (30”), 55 °C (30”), 72 °C (2’); and 72 °C (7’). spPCR used 130 pg genomic DNA per reaction. Amplification conditions were 94 °C (5’); 35 cycles of 94 °C (30”), 55 °C (30”), 72 °C (30”); and 72 °C (7’). Under the short extension conditions of spPCR, the expanded templates seen by standard PCR of (CTG)₁₀₂ and (CAG)₁₀₂ genomic DNA were not detected. PCR products were resolved in 7.5% polyacrylamide gels. Images were obtained on a Fuji LAS-3000 using ImageReader and Adobe Photoshop software.

Cells were trypsinized, washed in PBS and fixed overnight at −20 °C in 70% ethanol. Cells were resuspended in PBS (pH 7.4), treated with 100 U of RNase A for 20 min at 37 °C, and stained with 50 µg of propidium iodide/ml. DNA content was detected using a Becton Dickinson FACScan flow cytometer.