High-frequency homologous recombination in plants mediated by zinc-finger nucleases


(fax 515 294 7155; e-mail voytas@iastate.edu).


Homologous recombination offers great promise for plant genome engineering. This promise has not been realized, however, because when DNA enters plant cells homologous recombination occurs infrequently and random integration predominates. Using a tobacco test system, we demonstrate that chromosome breaks created by zinc-finger nucleases greatly enhance the frequency of localized recombination. Homologous recombination was measured by restoring function to a defective GUS:NPTII reporter gene integrated at various chromosomal sites in 10 different transgenic tobacco lines. The reporter gene carried a recognition site for a zinc-finger nuclease, and protoplasts from each tobacco line were electroporated with both DNA encoding the nuclease and donor DNA to effect repair of the reporter. Homologous recombination occurred in more than 10% of the transformed protoplasts regardless of the reporter's chromosomal position. Approximately 20% of the GUS:NPTII reporter genes were repaired solely by homologous recombination, whereas the remainder had associated DNA insertions or deletions consistent with repair by both homologous recombination and non-homologous end joining. The DNA-binding domain encoded by zinc-finger nucleases can be engineered to recognize a variety of chromosomal target sequences. This flexibility, coupled with the enhancement in homologous recombination conferred by double-strand breaks, suggests that plant genome engineering through homologous recombination can now be reliably accomplished using zinc-finger nucleases.


The ability to modify plant chromosomes through homologous recombination (gene targeting) has been a long sought after goal of plant biology (reviewed in Hanin and Paszkowski, 2003; Puchta, 2002; Reiss, 2003). The value of gene targeting for the study of gene function has been made abundantly evident in model organisms such as yeast, where homologous recombination is routinely used to create precise deletions, insertions or mutations of DNA sequences within their native chromosomal contexts. In addition to helping discern the function of plant genes, gene targeting opens up new possibilities for crop improvement. With gene targeting it is possible to carry out the genetic surgery required to reorchestrate metabolic pathways to create high-value crops, including seed with altered oil or carbohydrate profiles, food with enhanced nutritional qualities or plants with increased resistance to disease and stress.

To date no efficient method for gene targeting has been implemented in higher plants. The primary barrier is the high frequency with which DNA integrates at non-homologous sites by illegitimate recombination. When DNA is introduced into plant cells there are typically 105 to 107 illegitimate recombination events for every homologous recombination event (Halfter et al., 1992; Hrouda and Paszkowski, 1994; Lee et al., 1990; Miao and Lam, 1995; Offringa et al., 1990; Paszkowski et al., 1988; Risseeuw et al., 1995). The high ratio of illegitimate to homologous recombination makes it impractical to screen large numbers of transformed plants to identify the rare individuals that have undergone homologous recombination.

A variety of approaches have been tried to overcome the barrier imposed by illegitimate recombination. Selection strategies have been developed to identify rare homologous recombination events from the large background of illegitimate recombination events. For example, negative selectable marker genes have been incorporated in the input DNA that are lost upon homologous recombination. This makes it possible to select against cells in which the donor DNA has randomly integrated and to enrich for cells that have sustained a gene targeting event (Terada et al., 2002; Thykjaer et al., 1997). Alternatively, attempts have been made to enhance homologous recombination relative to illegitimate recombination by manipulating the plant's recombination machinery. Expression of the Escherichia coli recA or ruvC genes in plants or disruption of plant RAD50 homologs realizes some gains in the frequency of homologous recombination, but these gains are modest (two- to threefold) (Gherbi et al., 2001; Reiss et al., 2000; Shalev et al., 1999).

One of the most effective means for enhancing the frequency of homologous recombination is to create a chromosome break at the target site (reviewed in Puchta, 2005). The break stimulates the cell's DNA repair system, and in the presence of a homologous template repair proceeds through homologous recombination. In tobacco, cleavage of chromosomal targets by I-SceI endonuclease causes a 100-fold increase in recombination between a chromosomal locus and a transforming T-DNA in tobacco (Puchta et al., 1996). Transposon-induced chromosome breaks also enhance recombination; for example, excision of the Ac/Ds transposon increases somatic intrachromosomal recombination in Arabidopsis 1000-fold (Xiao and Peterson, 2000). All of these approaches, however, are of little practical value for modifying native plant genes through recombination, because each requires specific features at the target locus such as I-SceI cut sites or transposable elements.

Recently, chimeric zinc-finger nucleases (ZFNs) have been used to create site-specific chromosome breaks in the absence of pre-engineered target sites (Bibikova et al., 2003; Urnov et al., 2005). Zinc-finger nucleases have a DNA recognition domain composed of an array of Cys2–His2 zinc fingers. The zinc fingers recognize and bind to specific nucleotide triplets. Zinc fingers are available that recognize all GNN and ANN and some CNN and TNN triplets, and multiple zinc fingers can be joined together to generate DNA-binding arrays that recognize extended sequence patterns with great specificity and high affinity (Figure 1a) (Choo and Isalan, 2000; Pabo et al., 2001; Segal et al., 2003). Fused to the zinc-finger array is a nuclease, typically a non-specific cleavage domain from a type IIS restriction endonuclease such as FokI (Figure 1b) (Kim et al., 1996). FokI only functions as a dimer, and this has been capitalized upon to enhance the target specificity of the ZFN. Each FokI monomer is fused to a zinc-finger array that recognizes a different DNA sequence; only when the two recognition sites are in close proximity do the inactive monomers come together to create a functional enzyme (Figure 1c). By requiring DNA binding to activate the nuclease, a highly site-specific restriction enzyme is created.

Figure 1.

Zinc-finger nucleases.
(a) Zinc fingers are depicted that recognize nucleotide triplets of a target sequence. Multiple fingers can be joined together to create zinc-finger proteins that recognize extended sequence patterns.
(b) Zinc-finger domains are fused to a type II restriction endonuclease such as FokI.
(c) When the FokI monomers are brought into proximity by DNA binding, a functional nuclease is created that cleaves the target sequence.

Zinc-finger nucleases have been shown to enhance homologous recombination in Xenopus oocytes, human cells and Drosophila larvae (Bibikova et al., 2001, 2003; Porteus and Baltimore, 2003; Urnov et al., 2005). In Drosophila, engineered ZFNs that recognize the yellow gene were expressed in fly larvae in the presence of donor DNA that was either integrated elsewhere in the Drosophila genome or that existed as free-floating molecules released from Drosophila chromosomes by FLP recombinase (Bibikova et al., 2003). Targeted recombination occurred with remarkable efficiency. In some cases, more than 1% of the progeny flies had sustained a gene targeting event. A study in human cells used a defective green fluorescent protein (GFP) reporter gene that was repaired by ZFN-assisted homologous recombination (Porteus and Baltimore, 2003). Cleavage of the GFP reporter by ZFNs enhanced recombination more than 2000-fold. More recently, a report has shown that engineered ZFNs can be used to correct human genetic defects through homologous recombination (Urnov et al., 2005). Mutations in the IL2RG gene that cause X-linked severe combined immune deficiency (SCID) were corrected at frequencies approximating 18%. Furthermore, about 7% of the cells acquired the desired genetic modification on both chromosomes.

Recently, ZFNs were expressed in plants to create double-strand breaks that were repaired by non-homologous end joining (Lloyd et al., 2005). Here we report the use of ZFNs to create breaks to increase the frequency of gene targeting. In our test system we observe more than one homologous recombination event per 10 illegitimate recombination events, a 104- to 106-fold enhancement over frequencies of unassisted homologous recombination. The fidelity of gene targeting was found to approximate 20%; that is, one in five characterized gene targeting events was free of DNA insertions or deletions sustained during repair of the target locus. The high efficiency and fidelity of ZFN-assisted gene targeting, coupled with the fact that ZFNs can be designed to recognize any chromosomal target, suggest that the long-standing barrier to conducting plant genome modification through homologous recombination can be overcome through the use of ZFNs.


Molecular reagents to measure homologous recombination

A target gene for measuring homologous recombination was developed that confers both selectable and screenable phenotypes to plant cells. The gene is a translational fusion between β-glucuronidase (GUS) and neomycin phosphotransferase (NPTII) (Figure 2a, pDW1273). Cells expressing the GUS:NPTII fusion are both kanamycin resistant and turn blue when incubated in appropriate substrates (Datla et al., 1991). GUS:NPTII was rendered non-functional by deleting 600 bases that encode the active site of GUS and part of the ATP-binding domain of NPTII (Figure 2a, pDW1364). Inserted at the site of the deletion was a recognition site for a previously described ZFN (Zif268:FokI, see below). After cleavage of the target gene by Zif268:FokI, homologous recombination with the appropriate donor DNA molecule should restore GUS:NPTII function (Figure 2b).

Figure 2.

Molecular reagents for measuring homologous recombination.
(a) pDW1273 encodes a functional GUS:NPTII reporter gene. AI denotes the artificial intron within the GUS coding sequence. The ColE1 replicon and Ampr gene are for recovery of the integrated construct by plasmid rescue. The GUS:NPTII coding sequence in pDW1363 has a 600 bp deletion that includes GUS and NPTII coding sequences critical for function (mutant forms of GUS and/or NPTII are indicated by lower-case letters). A Zif268 recognition site (depicted in Figure 1) is inserted at the site of the deletion. The Hygr marker functions in plants and can be used to select cells carrying the reporter construct. Filled triangles depict the left and right borders of the T-DNA. Open arrows indicate the primers used for the PCR reactions in Figure 4. The donor DNA, pDW1269, lacks sequences 5′ of the artificial intron and is used to repair the GUS:NPTII reporter in pDW1363 by homologous recombination. The donor DNA has a diagnostic XhoI restriction site.
(b) The target gene and donor DNA are shown undergoing recombination. Numbers adjacent to the open arrows indicate the size of expected PCR products. The length of homology between donor and target is given below the donor DNA.

The donor DNA molecule for homologous repair is 4887 bp long and includes the 600 bases missing from the target gene (Figure 2, pDW1269). Overall, the donor has 4279 bp of sequence identity with the target, 1286 bp upstream and 2993 bp downstream of the Zif268:FokI recognition site. In addition to carrying out homologous repair, the donor DNA molecule can integrate into plant chromosomes by illegitimate recombination. To prevent such illegitimate integration events from conferring GUS activity, the donor DNA lacks the 5′ 457 bp of the GUS coding sequence. Although the donor DNA encodes a complete NPTII gene, it lacks a promoter to minimize the likelihood that illegitimate integration events confer kanamycin resistance. The donor DNA also has a diagnostic restriction endonuclease site (XhoI) that is absent from the target GUS:NPTII construct to help in the molecular characterization of putative recombinants.

The construct encoding a Zif268:FokI fusion protein was similar to the hybrid endonuclease previously described (Kim et al., 1996). A nuclear localization signal and an AcV5 epitope tag were added to the N-terminus of the fusion to direct the protein to the nucleus and to enable detection by immunoblot analysis (Hohmann and Faulkner, 1983; Monsma and Blissard, 1995). To assess function of the fusion protein, Zif268:FokI was expressed in E. coli. Activity was measured in vitro by incubating a few microliters of E. coli lysate with the target plasmid. The target vector was cleaved efficiently (data not shown).

A tobacco test system to measure homologous recombination

The GUS:NPTII target gene was introduced into tobacco chromosomes by Agrobacterium-mediated transformation. Twelve independent transgenic plants were regenerated on media containing hygromycin, which selected for presence of the target construct. Southern hybridizations and segregation of the hygromycin marker gene linked to GUS:NPTII revealed that all plants contained a single locus with the exception of plant 12, which carried two unlinked T-DNAs (data not shown). To facilitate characterization of putative recombinants, the integrated T-DNAs and flanking chromosomal sequences were recovered from plants 4, 6 and 9 (see below and data not shown). In plants 4 and 6, the target gene had integrated into a retrotransposon integrase gene and a cellulose synthase gene, respectively. The target gene in plant 9 had integrated into highly repeated sequences. When DNAs flanking the plant 9 target gene were used as probes in Southern hybridizations, they hybridized to numerous restriction fragments (data not shown).

To test whether the GUS:NPTII target genes were in chromosomal environments conducive to transcription, reverse transcriptase (RT)-PCR experiments were conducted with RNA isolated from leaf tissue of the various transgenic lines. The primers flanked an artificial intron (AI) that was inserted into the GUS coding sequence. This made it possible to distinguish between RNA and contaminating genomic DNA (or unspliced mRNA). Nine of the 12 transgenic plants showed strong expression of the GUS:NPTII reporter (Figure 3). Three of the plants (2, 3 and 10) displayed bands corresponding to sizes predicted for both spliced RNA (483 bp) and genomic DNA or unspliced mRNA (671 bp). In these plants the product containing the intron probably derives from unspliced mRNA because there was no evidence of contaminating genomic DNA in controls performed with the same mRNA samples and primers flanking an intron in the gene for the small subunit of ribulose bisphosphate carboxylase (Rubisco). Because the spliced transcript was of considerably lower abundance in plants 2, 3 and 10 we concluded that the target gene in these test plants was not well-expressed. Ten of the plants were used in subsequent experiments (1, 3–6 and 8–12).

Figure 3.

Expression of the GUS:NPTII reporter as measured by RT-PCR.
Ribonucleic acid was prepared from 12 transgenic plants with the GUS:NPTII reporter (pDW1363). Primers flanking the artificial intron in GUS:NPTII were used to measure expression (top panel). The 483 bp and 671 bp fragments correspond to spliced and unspliced transcripts, respectively. The bottom panel depicts control RT-PCR experiments using the same RNA samples and primers flanking an intron in the tobacco Rubisco small subunit gene. The spliced product is 306 bp.

Introduction of the zinc-finger nuclease and donor DNA into target cells

Tobacco protoplasts were prepared from the 10 target plants. To serve as controls, aliquots of 1.25 × 106 protoplasts were electroporated with either 10 μg of DNA encoding a functional GUS:NPTII fusion (pDW1273), 10 μg of donor DNA (pDW1269) or 30 μg of DNA encoding Zif268:FokI. To measure ZFN-assisted recombination, protoplasts were electroporated with both the donor DNA (10 μg) and DNA encoding Zif268:FokI (30 μg). The Zif268:FokI-encoding plasmid was in a 5:1 molar excess relative to the donor DNA. All plasmids were linearized by restriction enzyme digestion prior to electroporation.

After electroporation, the cells were allowed to recover for 6 days on callus-inducing media. A fivefold dilution was then made, and one-fifth of the cells were cultured without selection to assess the effect of the various DNA constructs on plating efficiency; no significant effect was observed for any of the constructs, including the Zif268:FokI construct (data not shown). Four-fifths of the cells (106 protoplast-derived cells) were selected on kanamycin at 50 μg ml−1. After approximately 20 days, kanamycin-resistant colonies were counted and stained histochemically to evaluate GUS expression (Table 1).

Table 1.  ZFN-assisted gene targeting in tobacco
Plant IDGUS:NPTIIDonor DNA onlyZFN DNA onlyDonor + ZFN DNA
  1. The columns labeled Kanr and GUS+ specify the number of calli with these phenotypes that were recovered per 106 protoplasts.

Frequency × 10−5448.73100.7332.33   109.959.11

The frequency of transformation or illegitimate recombination is given in Table 1 in the column labeled ‘GUS:NPTII’. The average frequency of kanamycin resistance was 4.5 × 10−3, and the average frequency of GUS expression was 1.0 × 10−3. Because both kanamycin resistance and GUS activity are conferred by the same polypeptide, we anticipated that these two frequencies would be equivalent. The discrepancy was found to be due to the insensitivity of the GUS histochemical stain. When GUS activity was measured using fluorometric assays with proteins extracted from kanamycin-resistant (Kanr) GUS calli, the calli originally scored as GUS were found to have significant GUS activity (data not shown). The histochemical GUS assay, therefore, underestimated the frequency of illegitimate recombination when the functional gene was used.

Kanamycin resistance, but not GUS expression, was detected in target cells treated with only donor DNA. The donor molecule contains the complete coding sequence for NPTII, but has a substantial, 457 bp deletion at the 5′ end of GUS. The average frequency of kanamycin resistance was 3.2 × 10−4, about 14-fold less than the estimate of illegitimate recombination. This reduced rate is occasioned by the lack of a promoter on the donor construct. Resistance is probably due to transcription driven by chromosomal sequences at the site of donor DNA integration. The frequency of kanamycin resistance conferred by the donor DNA is germane to calculations of homologous recombination, because kanamycin resistance caused by illegitimate recombination must be distinguished from kanamycin resistance caused by homologous recombination at the target gene. It is important to note that no gene targeting events were recovered when the donor DNA was used in the absence of a ZFN. This is not unexpected, because based on the frequency of transformation (Table 1, column 1), only about 50 000 cells were transformed with the donor DNA, and unassisted frequencies of homologous recombination in tobacco cells approximate 1 event per 106 transformed cells (Hrouda and Paszkowski, 1994; Lee et al., 1990; Offringa et al., 1990; Paszkowski et al., 1988; Risseeuw et al., 1995). In cells treated with DNA encoding only Zif268:FokI, neither kanamycin resistance nor GUS expression was detected (Table 1).

Electroporation of tobacco protoplasts with both the Zif268:FokI construct and donor DNA gave rise to large numbers of kanamycin-resistant colonies, at frequencies more than threefold higher than when donor DNA was used alone (1.1 × 10−3 versus 3.2 × 10−4). Because no Kanr GUS+ calli were obtained when protoplasts were transformed with either the donor DNA or Zif268:FokI constructs alone, we consider the frequency of homologous recombination to be the Kanr GUS+ frequency obtained with both the donor DNA and the Zif268:FokI (9.1 × 10−5). This value is approximately 10-fold lower than the number of Kanr GUS+ obtained when the functional GUS:NPTII reporter was electroporated into cells (1.0 × 10−3), and indicates that approximately one in 10 protoplasts that take up DNA (as determined by the frequency of illegitimate recombination) sustain a homologous recombination event (the precise value of homologous recombination as a fraction of illegitimate recombination is 0.09). Because the GUS reporter is not very sensitive, the above calculation is an underestimate of the frequency of homologous recombination. If the more sensitive kanamycin resistance reporter is considered, the homologous recombination frequency approximates 10−3 (7.8 × 10−4). This value is obtained by subtracting the frequency of Kanr observed with the donor DNA alone (3.2 × 10−4) from the frequency obtained with the donor DNA and the Zif268:FokI construct (11.0 × 10−4). Comparing this estimate of homologous recombination (7.8 × 10−4) with the frequency of Kanr obtained with the functional GUS:NPTII gene (4.5 × 10−3), a homologous recombination event is sustained by approximately one in five protoplasts that take up DNA (homologous recombination as a fraction of illegitimate recombination equals 0.17).

High-frequency homologous recombination was observed at all 10 chromosomal loci, indicating that a variety of chromosomal sites are amenable to gene targeting. Calculations based on the Kanr phenotype indicate that homologous recombination as a fraction of illegitimate recombination ranged from 0.12 (for plant 6) to 0.43 (for plant 8), a 3.6-fold difference. As noted in Figure 2, plants 3 and 10 did not express the target locus efficiently. This observation was supported by the failure to recover GUS+ calli from these plants; the level of target gene expression was apparently below that necessary to detect GUS activity by our histochemical stain but not so low as to prevent recovery of Kanr calli. Lack of GUS activity was also observed for plant 5, and subsequent RT-PCR experiments revealed that the target locus in plant 5 had become partially silenced in the time between its initial characterization and the time when the recombination experiment was performed (data not shown). Homologous recombination as a fraction of illegitimate recombination was 0.15, 0.16 and 0.18 for plants 3, 5 and 10, respectively. These values approximate the average of all experiments (0.17), indicating that despite the lack of expression, these loci were modified with high efficiency.

Molecular characterization of homologous recombination events

Tobacco plants were regenerated from Kanr GUS+ calli obtained from experiments performed with plants 4, 6 and 9. The target genes were characterized in the putative recombinants by PCR, using a primer pair designed to amplify the region spanning the deletion in the GUS:NPTII gene (see Figure 2). To prevent amplification of donor DNA molecules that may have integrated into chromosomes by illegitimate recombination, one primer was specific to the target gene and the other was complementary to sequences in both the target and donor DNA. The primers amplify a 3.3 kb fragment from the original, defective locus and a 3.9 kb fragment from a locus that has been repaired by recombination (Figure 2b).

Results of PCR amplification for 12 Kanr GUS+ plants derived from plant 9 are shown in Figure 4. These 12 recombinants were generated from protoplasts derived from two T1 plants, one of which was hemizygous and the other homozygous for the target gene. The copy number of the target genes in these plants was confirmed in subsequent generations by segregation analysis (data not shown). Polymerase chain reactions for 11 of the 12 recombinant plants produced a 3.9 kb PCR product, consistent with the target locus having sustained a homologous recombination event (Figure 4). No PCR product was obtained for plant 9-2, probably due to failure of the PCR reaction. This could occur through loss of one of the primer sites upon recombination. Three of the 12 recombinants (plants 9-1, 9-10 and 9-11) also produced a PCR product of 3.3 kb, the size predicted for an unmodified chromosomal target gene. The parental plants from which these recombinants were derived were homozygous for the target locus (see below), and the presence of the two PCR products indicates that only one allele was corrected by recombination. Plant 9-8 produced a 3.9 kb PCR product and a slightly larger fragment, suggesting that it sustained a rearrangement in one of its two copies of the target locus that gave rise to the larger product.

Figure 4.

PCR analysis of putative recombinants generated from target plant 9.
Deoxyribonucleic acid was prepared from 12 putative recombinants, and the PCR was used to amplify the GUS:NPTII target gene. The 5′ primer was specific to the target locus, whereas the 3′ primer was complementary to both the target and donor DNA (see Figure 2). As indicated for parental plant 9, the size of the PCR product generated from the defective target locus is 3.3 kb. Amplification of GUS:NPTII reporters repaired by homologous recombination generates a 3.9 kb product.

Polymerase chain reactions were also performed with DNA prepared from recombinants derived from plant 6 (data not shown). Six of the putative recombinants tested produced 3.9 kb PCR products, consistent with having sustained a homologous recombination event. For these six recombinants, an 800 bp region that encompassed the site of repair was sequenced from the 3.9 kb PCR product. The sequences of the PCR products included the 600 bp missing from the target as well as an XhoI site diagnostic of the donor DNA. The PCR products also had sequences unique to the chromosomal target gene, indicating that they were the recombinant products of the target and donor. On the basis of the collective PCR and sequence data we concluded that the Kanr GUS+ phenotypes of the putative recombinants were acquired by restoration of the chromosomal target gene through homologous recombination.

To further assess the fidelity of ZFN-assisted recombination, target loci in the 12 recombinants derived from plant 9 were subjected to Southern hybridization analyses. DNA was digested with BclI, which cuts once at the 5′ end of the GUS gene and in flanking tobacco DNA (Figure 5a). Southern hybridizations performed with a probe for the hygromycin resistance gene revealed a single band of 6 kb in the parental tobacco line and each of the recombinants (Figure 5b). This indicates that the region 5′ of the GUS:NPTII gene is not substantially altered in any of the recombinants. As mentioned above, some plant 9 recombinants are hemizygous for the target whereas others are homozygous. Because the same amount of DNA was loaded in each gel, the homozygous individuals can be identified by the twofold higher hybridization intensity relative to the hemizygous individuals (see also below). Homozygous plants include the parent, as well as plants 9-1, 9-4, 9-5, 9-8, 9-9, 9-10 and 9-11.

Figure 5.

Southern hybridization analysis of recombinants generated from target plant 9.
(a) A map of the plant 9 target locus. Probes used in panels B, C and E are depicted, as are the sizes of the hybridizing BclI restriction fragments expected in the parental plant. Note that probes 1 and 2 are specific to the target locus and do not hybridize to the donor DNA; probe 3 hybridizes to both the donor and target.
(b) DNA from untransformed tobacco (Xanthi), parental plant 9, and 12 recombinants was hybridized with probe 1. Hybridization is only observed to the expected 6.0 kb fragment. The twofold variation in hybridization intensity reflects the copy number of the target locus: the parental plant and recombinants 9-1, 9-4, 9-5, 9-8, 9-9, 9-10, and 9-11 are homozygous; the remainder are hemizygous.
(c) The same filter was stripped and rehybridized with probe 2. Black dots denote 8.0 kb restriction fragments, the size predicted for repair to the target locus by homologous recombination.
(d) The filter was stripped and rehybridized with the Zif268:FokI construct (ZFN probe). White dots indicate BclI fragments that hybridized to probe 2 in (c).
(e) The filter was stripped and rehybridized with probe 3, which recognizes both the donor and target gene. Black dots identify 8.0 kb restriction fragments that are the size predicted for target genes repaired by homologous recombination. The size in kb of the molecular length markers is given to the right of (c) and (e).

The same Southern filter was stripped and rehybridized with a probe specific to the 5′ end of GUS that is unique to the target DNA (Figure 5c). In the parental line, the hybridizing fragment was 7.4 kb. The reconstituted GUS:NPTII reporter should be 600 bp larger, and this was observed in three of the 12 plants (9-1, 9-7 and 9-11). In plants 9-1 and 9-11, which were homozygous for the reporter, one of the two alleles was corrected. The remaining nine recombinants showed a significant increase in the size of the BclI fragment that encompasses the GUS:NPTII reporter. Among these were four homozygous plants (9-4, 9-5, 9-8 and 9-9) in which both copies of the reporter gene were modified. Plant 9-4 acquired an additional BclI site, as evidenced by the 1.8 kb hybridizing fragment, suggesting that it sustained a rearrangement involving the addition of an extra BclI site. Seven recombinants derived from plant 4 were also analyzed by Southern analyses (data not shown). Of these, two had hybridizing fragments 600 bases larger than the parental plant, and in five the BclI fragment encompassing the GUS:NPTII reporter gene was significantly larger.

Homology-dependent DNA repair in plants is often accompanied by non-homologous end joining (NHEJ), which frequently leads to DNA insertions or deletions (Gorbunova and Levy, 1997; Salomon and Puchta, 1998). The large, variably sized fragments observed in nine of the recombinants are consistent with DNA insertion associated with NHEJ. To test if this was the case, the Southern filter was stripped and rehybridized with a DNA fragment specific to the Zif268:FokI construct (Figure 5d) and one that recognizes both the target gene and donor DNA (Figure 5e). In the nine recombinants with a larger than expected BclI fragment, a fragment of the same size also hybridized to the Zif268:FokI probe (Figure 5d). This suggests that the Zif268:FokI construct was incorporated at the site of repair, and the variability in intensity of the hybridization signal suggests that in some cases multiple copies of the Zif268:FokI construct were incorporated. Several of the plants also had other Zif268:FokI-hybridizing fragments, consistent with integration of the ZFN DNA elsewhere in the genome. Extra bands deriving from donor DNAs integrated by illegitimate recombination were also observed (e.g. plants 9-3, 9-4, 9-6 and 9-8) (Figure 5e). Importantly, in the three recombinants that appear to have been repaired only by homologous recombination (i.e. 9-1, 9-7 and 9-11) neither Zif268:FokI-hybridizing fragments nor extraneous donor DNAs were observed.

In summary, a total of 19 recombinants were analyzed at two different target sites. Based on whether or not the plants were hemizygous or homozygous, there was a total of 26 target genes that could have been repaired by recombination. Of these, five were repaired without accompanying DNA rearrangements, indicating that the overall fidelity of recombination approximated 20%.


Zinc-finger nucleases induce high-frequency homologous recombination

Deoxyribonucleic acid introduced into plant cells most frequently integrates into chromosomes by illegitimate recombination. Only rarely does homology-dependent exchange result in specific sequence modifications at pre-selected target loci. Estimates of homologous recombination in tobacco using Agrobacterium-mediated DNA delivery range from one homologous recombination event per 8.4 × 105 to 2.2 × 106 illegitimate events (Lee et al., 1990; Offringa et al., 1990; Risseeuw et al., 1995). Comparable frequencies are observed when naked DNA is introduced into tobacco cells (Hrouda and Paszkowski, 1994; Paszkowski et al., 1988). Here we demonstrate that cleavage of a chromosomal target by ZFNs dramatically alters the ratio of homologous to illegitimate recombination events, such that more than one homologous recombination event occurs per every 10 illegitimate recombination events.

Zinc-finger nuclease-assisted gene targeting was first implemented in animal systems (Bibikova et al., 2001), and two studies in humans are particularly relevant to the work described here (Porteus and Baltimore, 2003; Urnov et al., 2005). Like our experiments in tobacco, gene targeting in humans was measured at an integrated, defective reporter gene (GFP). Cleavage of the GFP reporter by ZFNs enhanced recombination, resulting in approximately one homologous recombination event per five illegitimate recombination events—a frequency comparable to what we observed in tobacco. In one study, ZFNs were engineered to recognize the IL2RG gene, mutations in which cause X-linked SCID (Urnov et al., 2005). Experiments were performed to correct a disease-causing IL2RG mutation, and as with the GFP reporter, gene targeting frequencies approximated 18%. Importantly, about 7% of the cells acquired the desired genetic modification on both chromosomes. We observed biallelic gene modification in four of seven plants that were homozygous for the reporter gene. Endogenous genes have also been the target of ZFN-assisted recombination experiments in Drosophila (Bibikova et al., 2003), and the use of engineered ZFNs to modify native plant genes is a logical extension of the work described here.

Chromosome breaks are key to the efficiency of ZFN-assisted recombination. Double-strand DNA breaks stimulate repair by multiple mechanisms, and a considerable body of literature exists regarding break-enhanced DNA repair in plants (reviewed in (Puchta, 2005)). Of particular relevance to our study are experiments performed in tobacco using I-SceI to introduce chromosome breaks at integrated, defective reporter genes which, upon correction by homologous recombination, confer a selectable phenotype (Puchta, 1998; Puchta et al., 1996). Gene targeting frequencies using I-SceI approximated one event per (2.2–18.3) × 103 illegitimate recombination events (Puchta et al., 1996). Although our frequencies of ZFN-assisted homologous repair were at least one order of magnitude higher, we do not believe this is due to any inherent difference in ZFNs versus restriction enzymes for creating chromosome breaks. Rather, the I-SceI experiments used a fundamentally different test system: donor DNAs were delivered by Agrobacterium to seedling explants. Furthermore, the donor DNA and I-SceI-encoding DNA were delivered on separate T-DNAs in separate Agrobacterium strains, and therefore cells had to be infected by both Agrobacterium strains to effect homologous recombination. In future experiments, we can directly compare methods of DNA delivery as well as DNA-breaking reagents; the GUS:NPTII target gene used in our experiments has an I-CeuI site adjacent to the Zif268:FokI recognition site.

One important aspect of our study was that frequencies of recombination were assessed at 10 different chromosomal loci. When considering the Kanr phenotype, variation in recombination frequency across the targets ranged between three- and four-fold (one homologous recombination event per 2.3 to 8.3 illegitimate recombination events with the average being 5.9). Surprisingly, the differences observed between target loci were not correlated with target gene expression. At least three of the targets were transcriptionally quiescent, as determined by RT-PCR experiments, yet they yielded Kanr cells at frequencies comparable to other loci (one homologous recombination event per 5.6 to 6.7 illegitimate recombination events). Although chromosome position has been shown to influence interchromatid and interhomolog recombination in Arabidopsis (Molinier et al., 2004), the uniformity in recombination frequencies observed here suggests that a variety of chromosomal sites are conducive to recombinational repair by ZFNs.

Fidelity of zinc-finger nuclease-assisted recombination

The repair of double-strand breaks in plants is best described by the synthesis-dependent strand-annealing (SDSA) model (Nassif et al., 1994; Puchta, 1998; Rubin and Levy, 1997). Recombination products predicted by SDSA are depicted in Figure 6. After induction of a double-strand break in the target (Figure 6a), a 3′ single-strand overhang is released by exonuclease digestion (Figure 6b). The 3′ end invades the double-stranded donor to form a D-loop, which can be resolved either through homologous recombination or a combination of homologous recombination and NHEJ (Figure 6c). Products resulting strictly from homologous recombination are generated when the 3′ end of the invading strand is elongated, and homology to the second 3′ end of the double-strand break allows the two single strands to anneal and repair the break (Figure 6d,e). Non-homologous end-joining comes into play if the 3′ end of the invading strand cannot find complementary sequences at the broken target (Figure 6f,g). Lack of complementarity can occur in a variety of ways; for example, exonuclease digestion can remove complementary sequences from the target, or the invading 3′ end may copy non-complementary sequences from a donor that lacks complete target homology.

Figure 6.

Products of recombination predicted by the synthesis-dependent strand-annealing (SDSA) model.
(a) A double-strand break is introduced into the target gene by the ZFN.
(b) A 3′ single-strand overhang is released by exonuclease digestion.
(c) The 3′ end invades the double-stranded donor to form a D-loop. The blue colored sequences in the donor depict the 600 bp missing from the target locus that are required to restore GUS:NPTII function.
(d, e) Products resulting strictly from homologous recombination are generated when the 3′ end of the invading strand is elongated, and homology to the second 3′ end of the double-strand break allows the two single strands to anneal and repair the break.
(f, g) If the 3′ end of the invading strand cannot find complementary sequences at the broken target, the break is repaired by a combination of homologous recombination and NHEJ. The red sequences denote insertions or deletions that can occur through NHEJ.

Our analysis of ZFN-induced chromosome breaks supports the SDSA model: repair was either carried out by homologous recombination or a combination of homologous recombination and NHEJ. Because cells were selected for GUS and NPTII activity, all characterized target sites had sustained a homologous recombination event. This was documented by PCR amplification of the target gene from putative recombinants; the size of the PCR products was consistent with having obtained 600 bp from the donor DNA. A portion of six PCR products was sequenced, and all were identical, indicating that the fidelity of recombination at the site of the deletion was 100%. If repair were carried out solely by homologous recombination, then the structure of the recombinant loci should be identical to the target (with the exception of the additional 600 bp). Southern blot analysis, however, indicated that this was only the case for five of the 26 target sites analyzed from plants 4 and 9 (20% of the total). Although the five recombinant targets did not appear to differ from the parental target, it is possible that they sustained small insertions or deletions that could not be detected by Southern analysis. Complete DNA sequencing of the target loci in these plants will be required to fully assess the fidelity of homologous recombination.

Although the remaining 21 targets (80%) had repaired the GUS:NPTII reporter by homologous recombination, they also appear to have sustained a rearrangement in the vicinity of the target gene. We predict that in these plants, after extension of the invading 3′ end through the GUS:NPTII coding sequences on the donor, the D-loop was resolved by NHEJ. Repair by NHEJ in plants is frequently associated with the addition of extra DNA or deletions (Gorbunova and Levy, 1997; Salomon and Puchta, 1998). The hybridizing BclI restriction fragment that encompasses the target gene and the Zif268 recognition site was typically increased in size, suggesting that additional DNA was incorporated in the vicinity of the double-strand break. When the Southern filters were rehybridized with probes for the Zif268:FokI construct, in all cases hybridization was observed to the larger BclI fragments. This indicates that additional DNA was, in fact, incorporated at the site of repair and that at least some of it was derived from the Zif268:FokI construct.

All of the observed DNA rearrangements occurred downstream of the GUS:NPTII reporter. In contrast, the BclI fragment encompassing the 5′ end of the target locus appeared to be the same in the parental plants and the recombinants. Because either 3′ end of the broken target can invade the donor DNA, repair by NHEJ would be expected to occur at comparable frequencies on both sides of the break. The absence of observed insertions or deletions near the 5′ end of the GUS:NPTII reporter is probably a consequence of bias due to selection; imprecise repair by NHEJ at the 5′ end would compromise GUS:NPTII expression, and such events would have escaped detection. We predict that if selection was not invoked, higher frequencies of NHEJ may be observed. This is germane to gene targeting experiments, because it is desirable to minimize the unwanted sequence changes associated with NHEJ. It may be possible to lower the frequency of repair by NHEJ by altering the form of the donor DNA (e.g. circular versus linear), by reducing the amount of donor and ZFN-encoding DNA delivered to plant cells, or by using different methods of DNA delivery such as Agrobacterium. Alternatively, strategies could be employed to enrich for precise gene targeting events by incorporating negative selectable marker genes in the donor DNA that are lost upon homologous recombination (Terada et al., 2002; Thykjaer et al., 1997).

One class of products not observed in our experiments are those referred to as ‘ectopic targeting’ events (Offringa et al., 1990). In ectopic targeting, sequences are copied from the target locus onto the donor DNA, which then integrates elsewhere in the genome. Whereas ectopic targeting events have been found repeatedly in plant gene targeting experiments (Hanin et al., 2001; Reiss et al., 2000), the Southern data suggest that in all cases described here the repaired reporter genes are at the same chromosomal position as the original target locus: the target-specific probe that recognizes the 5′ portion of the GUS gene locus did not hybridize to any additional bands in any of the recombinants.

Zinc-finger nuclease design and gene targeting

The primary advantage in using ZFNs to enhance homologous recombination is that they can be engineered to recognize any chromosomal site (Carroll, 2004). By contrast, all existing methods to create specific chromosome breaks require targets with pre-engineered restriction enzyme recognition sites or transposable element insertions. Zinc-finger design has become increasingly modular; that is, coding sequences for individual ZFs that recognize specific triplets can be joined together to create ZF arrays that recognize longer target sites with high specificity and affinity (Choo and Isalan, 2000; Pabo et al., 2001; Segal et al., 2003). Individual ZFs are available that recognize all 16 GNN triplets and ANN triplets, and the design of fingers that recognize many CNN and TNN triplets is currently in progress. Engineered ZFs are increasingly being employed for plant genome modification: Zinc fingers have been fused to transcription activation and repressor domains to modulate transcription (Guan et al., 2002; Ordiz et al., 2002; Stege et al., 2002), and ZFNs have been shown to work in plants as gene-specific mutagens to create chromosome breaks, which upon repair by NHEJ alter the sequence of the target (Lloyd et al., 2005).

Zinc-finger nucleases for gene targeting are typically composed of two ZF arrays, each of which recognizes a nine-base target sequence separated by a six-base spacer. With the available libraries for GNN and ANN triplets, DNA sequences amenable to ZFN design can be found in plant genes every 25–100 bases (data not shown). This indicates that with existing reagents for ZF design, it will be possible to engineer ZFNs that cleave in close proximity to desired sites of modification on plant chromosomes. The next step is clearly to employ engineered ZFs to recognize and modify endogenous plant genes. In addition, we can further exploit our tobacco model to optimize parameters for high-efficiency ZFN-assisted gene modification. For example, ZFNs can be designed to recognize sites elsewhere within the GUS:NPTII reporter. Such experiments will not only assess the facility of ZF design, but they will test other parameters that can affect recombination, such as changes in homologous recombination frequency that might occur when cut sites are at variable distances from the desired site of modification. The tobacco model can also be exploited to test other methods to deliver the donor DNA and ZFN, such as Agrobacterium, and to develop strategies to detect recombinants that do not confer positive selection.

The enhancement in homologous recombination we report here, coupled with the fact that ZFNs can be engineered to recognize any target loci, suggests that gene targeting can be employed to study plant gene function. Gene targeting also offers great promise for harnessing plant genes to create crops with new and valuable traits. Because gene targeting introduces changes in plant genomes in a highly specific and controlled manner, crops created through gene targeting may meet greater public acceptance than those generated by traditional transformation methods. At the very least, the use of gene targeting to create new cultivars is likely to generate new debate on the use and regulation of genetically modified crops.

Experimental procedures

DNA constructs

The GUS:NPTII reporter gene was similar to the one described by (Datla et al., 1991), except that the sequence of the linker between the two coding regions includes an XhoI site. In addition, an artificial intron (Goodall and Filipowicz, 1989) was introduced into the GUS coding region at the SnaBI site. The GUS:NPTII coding sequence was placed between the cauliflower mosaic virus 35S promoter and the nopaline synthase transcriptional terminator. The complete fusion construct is designated pDW1273.

The GUS:NPTII gene was rendered defective by deleting 600 bases of the coding region to produce pDW1364. This removed the active site for GUS (Islam et al., 1999; Matsumura and Ellington, 2001) and the ATP-binding site of NPTII (Burk et al., 2001; Thompson et al., 2002). At the site of deletion, a sequence with recognition sites for I-CeuI, XmaI and Zif268:FokI was added. This sequence is as follows, where small letters indicate the enzyme recognition sites and capital letters indicate GUS:NPTII coding sequences: AAGAACTTCT GGCCTGGCAG aataactata acggtcctaa ggtagcgacc cgggacgccc acgcattaaa gcgtgggcga ACCGACCTGT CCGGTGCCCT. The defective gene was cloned into pDW700, a binary vector for plant transformation by Agrobacterium (D. Wright, unpublished data). pDW700 is based on the BIBAC vector (Hamilton, 1997), and contains between the left and right borders a hygromycin resistance gene modified for expression in plants, a ColE1 replication origin and a beta-lactamase gene. The latter two features allow plasmid rescue of chromosomally integrated target genes. The donor DNA (pDW1269) was created as a precursor to pDW1273. pDW1269 differs from pDW1273 in that the 35S promoter and 5′ end of the GUS gene up to the SnaBI site are missing.

Sequences encoding the Zif268 DNA-binding domain and the FokI nuclease domain were individually assembled using PCR and overlapping oligonucleotides. Both were modified to match codon bias for dicot plants (http://www.kazusa.or.jp/codon/cgi-bin/showcodon.cgi?species=Arabidopsis + thaliana + %5Bgbpln%5D). A MASS translation signal (Sawant et al., 2001), an SV40 nuclear localization signal (Hicks and Raikhel, 1993; Lassner et al., 1991; Varagona et al., 1992) and an AcV5 epitope tag (Hohmann and Faulkner, 1983; Monsma and Blissard, 1995) were added to the N-terminus of the Zif268. The Zif268 DNA-binding domain and the FokI nuclease domain were then joined and placed downstream of an artificial promoter, which mediates high levels of gene expression in tobacco protoplasts (D. Wright, unpublished data), and upstream of the nopaline synthase transcriptional terminator. The gene was then cloned into a derivative of pSP72 (Promega, Madison, WI, USA) giving rise to pDW1345. A version of the construct (pDW1344) was made for expression in E. coli using the expression vector pQE80L (Qiagen, Valencia, CA, USA). Escherichia coli harboring the plasmid were induced for protein expression, and cell lysates were tested for their ability to cleave a plasmid with a Zif268 recognition site.

Transgenic tobacco target plants

pDW1364 was introduced into Agrobacterium strain C58CI (pMP90) (Koncz and Schell, 1986). Tobacco (cultivar Xanthi) leaf explants were co-cultivated, and 12 transgenic plants were regenerated by leaf disk transformation (Horsch et al., 1985). The expression of the target gene in the T0 plants was measured by RT-PCR using primers flanking the intron in the GUS coding sequence (PDIO259 5′-GAAAGCCGGGCAATTGCTGT and PDIO298 5′-CAGACGCGTGGTTACAGTCTT). As a control, primers were used that flank an intron of the tobacco Rubisco small subunit (DVO1473 5′-CAGCAATGTTGCTCAAGCTAACATG and DVO1513 5′-GGTGACTTGTTGTTTTCACGGTAG).

Measuring recombination in tobacco protoplasts

Target plants were grown aseptically as shoot cultures. To prepare leaf mesophyll protoplasts, leaf tissue was abraded with a soft paintbrush in a suspension of Celite in K3 media (20% w/v) (Nagy and Maliga, 1976). The abraded leaves were treated with cellulase R-10 (0.25% w/v) and macerozyme (0.05%) in K3 media. Protoplasts were recovered by floatation in K3 media and washed once. Other conditions for the maintenance of source tissue and the isolation of protoplasts are as described in Adams and Townsend (1983).

Electroporation was conducted using a Bio-Rad GenePulser (Hercules, CA, USA) with 0.4 cm cuvettes and a capacitor discharge of 170 V, 1200 μF. A pulse time constant of 25–30 msec was realized by the addition of 10–20 μl of 2 m KCl per cuvette. The donor DNA was linearized by digesting with BamHI prior to transformation; the BamHI site immediately precedes the artificial intron in GUS. The amounts of DNA used in the electroporation experiments are described in Results. Culture and selection of protoplast-derived calli were as described (Van den Elzen et al., 1985). Numbers of Kanr resistant calli were scored after 30 days. GUS activity was measured by overnight incubation at 37°C in McCabe's histochemical stain (McCabe et al., 1988). A vital stain was used to recover Kanr GUS+ (Swoboda et al., 1994). Plants were regenerated from representative Kanr GUS+ calli.

Molecular characterization of recombinants

Chromosomal sites of target gene integration were obtained for plants 4, 6 and 9. The target site was recovered for plant 6 by plasmid rescue (pDW1528) using genomic DNA digested with SphI. Target sites for plants 4 and 9 were recovered by inverse and adapter-mediated PCR. All PCR products and pDW1528 were sequenced.

To assess the structure of the target gene in the putative recombinants, PCR reactions were performed with primers specific to the hygromycin resistance gene and to GUS:NPTII (PDIO256 5′-CTTCCCGCTTCAGTGACAAC and PDIO535 5′-GGTTTCAGGCAGGTCTTGCAAC). The hygromycin gene is not present in the donor DNA so only the target locus is amplified. The expected sizes of the PCR products are 3.3 kb from the defective target gene and 3.9 kb from the repaired locus. A commercial PCR system was used to ensure fidelity of the PCR products and to minimize strand transfers during amplification (Expand Long Template PCR System; Roche, Indianapolis, IN, USA). An 800 bp region was sequenced from the amplification products obtained from six plant 6 recombinants. The sequenced region spans the site of deletion within the GUS:NPTII target and lies between the following primer sequences: PDIO256 and PDIO540 5′-CCTGAACCGTTATTACGGATG. Genomic DNA was prepared from recombinants derived from plant 9 and digested with BclI. Southern hybridizations were conducted by DNA Landmarks, Inc., St-Jean-sur-Richelieu, Quebec, Canada using probes described in Figure 5.


We thank Dana Carroll for advice on experimental design and Erica Unger-Wallace for helpful comments on the manuscript. This work was supported by NSF-SBIR Phase I award 0319602 to Phytodyne, Inc.