Multiple new site-specific recombinases for use in manipulating animal genomes

Abstract

Site-specific recombinases have been used for two decades to manipulate the structure of animal genomes in highly predictable ways and have become major research tools. However, the small number of recombinases demonstrated to have distinct specificities, low toxicity, and sufficient activity to drive reactions to completion in animals has been a limitation. In this report we show that four recombinases derived from yeast—KD, B2, B3, and R—are highly active and nontoxic in Drosophila and that KD, B2, B3, and the widely used FLP recombinase have distinct target specificities. We also show that the KD and B3 recombinases are active in mice.

Site-specific DNA recombinases are widely used in multicellular organisms to manipulate the structure of genomes and, in turn, to control gene expression (for reviews see refs. 1–4). These enzymes, derived from bacteria and fungi, catalyze directionally sensitive DNA exchange reactions between short (30–40 nucleotides) target site sequences that are specific to each recombinase (5). These reactions enable four basic functional modules—excision/insertion, inversion, translocation and cassette exchange—that have been used individually or combined in a wide range of configurations to control gene expression (Fig. 1A).

Fig. 1.

Modifying genome structure and gene expression with site-specific recombinases. (A) Four types of reaction are diagramed. The target sites recognized by the recombinases are indicated by the colored triangles, and the black lines represent genomic DNA. In the excision/insertion reaction, a segment of DNA between two tandemly arranged target sites can be excised as a circular molecule. The reverse reaction, insertion, occurs with much lower efficiency. In the inversion reaction, a segment of DNA between two oppositely oriented target sites can undergo one or more cycles of inversion. In the translocation reaction, a segment of a chromosome arm distal to the centromere is exchanged between homologs in a diploid organism. In the cassette exchange reaction, a linear segment of DNA is exchanged between two DNA molecules. (B) An example of the use of two recombinases with distinct target sites (represented by the different colored triangles) to perform a genetic intersection (2).

The use of site-specific recombination to manipulate genomes has been limited by the availability of recombinases with high activity, distinct site specificity, and low toxicity. In Drosophila, the most widely used recombinase is FLP, encoded by the Saccharomyces cerevisiae 2-μm plasmid (6). FLP was first shown to work in a heterologous, multicellular organism by Golic and Lindquist in 1989 (7) who demonstrated the excision reaction on chromosomally inserted target sites (FRTs). Since that time FLP/FRT recombination has been widely used in Drosophila in applications based on excision (8) and translocation (9–11).

Complex manipulations of genome structure can require the use of more than one of the modules diagrammed in Fig. 1A, or parallel independent implementations of the same module, in a single individual. To accomplish such manipulations, the modules must be implemented with different recombinases that do not recognize each other’s target sites. Similarly, a number of powerful methods have been developed for using the excision and inversion reactions to control expression of a transgene specifically in cells where two independent gene expression patterns overlap (2, 3, 12, 13). Such intersectional methods rely on pairs of orthogonal recombinases; see, for example, Fig. 1B. For these reasons, we sought to discover additional recombinases with distinct site-specificity.

FLP recombinase has been mutated to recognize altered FRT sites, but some cross-reaction still remains (14, 15). Cre, encoded by the bacteriophage P1, is the most widely used recombinase in mammalian cells (16–18). Cre functions in Drosophila (19), but exhibits obvious toxicity (20), a problem also observed in mammalian cells (21, 22; reviewed in ref. 23) and plants (24).

Other site-specific recombinases have been used in metazoans, but less extensively. Dre recombinase, a close relative of Cre, has been demonstrated to work in mammalian cells (25, 26). PhiC31 integrase, which catalyzes unidirectional recombination between attP and attB sequences, has been used extensively for integration (27, 28) and cassette exchange (29, 30), applications that take advantage of its nonreversibility. The R recombinase, a relative of FLP from the yeast Zygosaccharomyces rouxii (31, 32), has been shown to function in plants (33).

In this report we describe the characterization of R and three other recombinases encoded by 2-μm circle-like plasmids of other yeasts. Although such plasmids are rare among yeasts, those characterized so far are very similar in structure to the 2-μm plasmid of S. cerevisiae (reviewed in ref. 34). In particular, they appear to encode site-specific recombinases related to FLP and undergo high-frequency intramolecular recombination between inverted repeat domains both in their native species and in S. cerevisiae (32, 34–39). However, only the R recombinase has previously been shown to work in a multicellular organism (33), and none has been used in an animal genome. We compare the properties of these four yeast recombinases to FLP, Cre, and Dre in Drosophila. We demonstrate four non–cross-reacting pairs of recombinases and target sites that have low toxicity and high activity. We also show that two of the new recombinases function in mammalian cells.

Results and Discussion

KD, R, B2, and B3 Yeast Recombinases Are Active in Drosophila.

The ORFs predicted to encode the recombinases KD (from Kluyveromyces drosophilarum; ref. 35), R (from Z. rouxii; ref. 31), B2 (from Zygosaccharomyces bailii; ref. 37) and B3 (from Zygosaccharomyces bisporus; ref. 40) were synthesized with Drosophila codon usage. We wanted to decrease the half-lives of the recombinases to achieve tighter temporal control of activity and reduce potential toxicity. Therefore, a PEST sequence corresponding to residues 422–461 of the mouse ornithine decarboxylase gene (41) was appended at their C termini. The synthesized genes were cloned downstream of upstream activation sites (UAS) that bind the GAL4 transcription factor, thereby placing recombinase expression under the control of GAL4 (Methods; ref. 42).

Recombinase activity was assayed by using the excision reaction (Fig. 1A). We constructed a set of “stop cassettes” consisting of tandem hsp70 and SV40 transcriptional terminators flanked by target sites specific for one of the recombinases. To create recombinase-specific reporter constructs, the stop cassettes were placed between 20 UAS sites and the coding sequence for a membrane-targeted, codon-optimized myristoylated Red Fluorescent Protein (pJFRC72; refs. 43 and 44). RFP expression from these constructs would be expected only if the KD, R, B2, or B3 recombinases were able to “kick-out,” “rip-out,” “bail-out,” or “blow-out” their respective stop cassettes by excisional recombination.

Coexpression of each recombinase, under the control of 20XUAS sites (pJFRC7; ref. 43), and its cognate stop-cassette reporter using the pan-neuronal elav-GAL4 driver resulted in broad RFP expression in neurons; no RFP was observed in the absence of the construct encoding the recombinase. These results demonstrated that all four recombinases were active in Drosophila. We next asked if recombination went to completion, that is, whether the stop cassette excised in all recombinase-expressing cells. For this purpose, we used a more specific GAL4 driver line, R31F10 (constructed as described in ref. 45), which expresses in a single cell, the T1 neuron, in each of the 800 columns that comprise the medulla of the adult optic lobe. In a tangential section through the M2 layer of the medulla, a grid of T1 terminals is observed. The processes of different T1 neurons in this array do not overlap with each other, allowing us to score individual cells and determine the percentage in which an excision event had occurred (Fig. 2). High levels of recombinase expression resulted in the excision reaction going to completion for each of the four recombinases (Table 1). We also found that 3XUAS recombinase constructs, which we estimate express less than one-fifth the recombinase of the 20XUAS constructs used above (43), still result in complete, or nearly complete, excision for the three recombinases tested (B2, B3, KD).

Fig. 2.

Measuring recombinase activity and cross-reactivity. The images show projections of series of optical cross-sections through layer M2 of the medulla, a part of the optic lobes of Drosophila that is organized in regular vertical columns. The tangential orientation of the sections was chosen to reveal an array of columns. Examples from animals of differing genotypes are shown; each section is shown twice, GFP in green in A, C, E, and G or RFP in magenta in B, D, F, and H. Expression is driven in a single T1 cell in each column by the GAL4 driver line R31F10 and can be visualized by expression of GFP from a 10XUAS-mCD8::GFP construct (pJFRC2; ref. 43). To assay for the activity of a recombinase, a second reporter construct is included from which expression of RFP is observed only when a transcriptional stop cassette flanked by the target sites for a specific recombinase is excised. In A and B, the activity of the KD recombinase on its cognate target sites (“KD>” in the diagram shown in B, below the micrograph) is demonstrated. Note that RFP expression is observed in each of the ≈50 cells present in this field of view, indicating that the excision of the stop cassette by the recombinase has gone to completion; the neuronal processes of the same single cell are delineated in A and B by the white circle; the cell bodies of this and the other T1 cells are located outside the optical sections shown. In C and D, the ability of the B2 recombinase to recognize its cognate target sites (B2>) is shown; again the excision reaction appears to have gone to completion because the same cells can be seen in C, where expression of GFP does not require recombinase action, and in D, where expression of RFP is seen only in cells where recombinase action has led to removal of the stop cassette. The experiments presented in E–H show that the two recombinases show no detectable activity on each other’s target sites. The KD recombinase is unable to remove a stop cassette flanked by B2 recombinase target sites (E and F), and the B2 recombinase is unable to remove a stop cassette flanked by KD recombinase target sites (G and H).

Table 1.

Specificity of target site recognition by recombinases measured as percent of cells showing an excision event

	Recombinase
Target	B2	B3	KD	R	FLP
B2RT	99.9	<0.1	<0.1	36	<0.1
B3RT	<0.1	99.9	<0.1	<0.1	<0.1
KDRT	<0.1	<0.1	99	<0.1	<0.1
RSRT	0.9	<0.1	<0.1	96	<0.1
FRT	<0.1	<0.1	<0.1	<0.1	91

Excision events were measured by using the assay shown in Fig. 2 (Methods). The values presented are based on scoring >1,000 cells for each combination of recombinase and target site. The stronger D5 version of FLP was used in these experiments.

Three additional yeast recombinases have been described; we did not try TD1 (46), and our initial attempts to use SM (34) and KW (34) were unsuccessful. In addition to the yeast recombinases, we also tested the ability of Dre, a recombinase closely related to Cre, to work in Drosophila. Dre showed only moderate efficiency in Drosophila; we were only able to get the excision reaction to go to ≈70% (179/269) completion. However, we found that Dre did not demonstrate the toxicity associated with Cre (see below). Thus, Dre would be a useful recombinase for applications where low efficiency is desired, such as stochastic cell labeling. We did not evaluate other members of the Cre family of recombinases, such as those recently described in Vibrio and Shewanella species (47).

During the course of this work, we used two different versions of FLP that differ in the amino acid residue at position 5, having either a glycine (G; ref. 6) or an aspartic acid (D; ref. 48). Although these two versions have not been distinguished in published reports describing the use of FLP in Drosophila, we found that the version containing an aspartic acid was >10 times more active based on the ability of hsFLP constructs encoding each version to induce excision of an FRT-flanked stop cassette under parallel conditions in the T1 medulla neuron. We observed an average of 6.5 excisions per optic lobe with the G5 FLP and 289 with the D5 FLP.

Together with FLP, the Recombinases We Analyzed Comprise Four Non–Cross-Reacting Recombinase-Target Site Pairs.

We next assayed the ability of the recombinases to recognize each other’s target sites and those of the FLP recombinase. We constructed Drosophila lines that paired all combinations of the FLP, KD, R, B2, and B3 recombinases and target sites. B2 and R showed partial cross-reaction, but all other noncognate pairs of recombinases and target sites were inactive (Fig. 2 and Table 1), even when the recombinases were expressed at several times the level needed to go to completion on their cognate sites.

Unlike Cre, FLP and the Other Yeast Recombinases Do Not Show Toxicity.

Flies that expressed Cre-recombinase at high levels (20XUAS, pJFRC7; ref. 43) under the control of the pan-neural driver elav-GAL4 either did not reach the adult stage or were extremely uncoordinated and short-lived. By contrast, no obvious morphological or motility phenotypes were observed in otherwise identical crosses with 20XUAS KD, R, B2, B3, FLP, or Dre, or lower levels of Cre (3XUAS, pJFRC4; ref. 43). Similarly when flies homozygous for a 20XUAS recombinase were crossed to flies heterozygous for the ubiquitous driver tubulin-GAL4 (11) and a balancer chromosome (TM6B), we recovered tubulin-GAL4/20XUAS recombinase and balancer/20XUAS recombinase flies in roughly the expected 1:1 ratio for Dre (68:68), Flp (55:73), R (66:46), B2 (96:87), B3 (54:38), and KD (65:35). However, for Cre, no tubulin-GAL4/20XUAS recombinase progeny were recovered (0:79). Consistent with these results, high-level expression of Cre—but not the other recombinases—in the developing and adult eye using pGMR-GAL4 (49) resulted in strong defects in both external eye morphology and the arrangement of cells seen in sections cut through the retina. The lack of detectable toxicity resulting from expression of the yeast recombinases is more striking when we consider that the 20XUAS recombinase constructs used in these experiments are estimated to produce approximately fivefold more recombinase than needed to drive an excision reaction to apparent completion.

B3 and KD Recombinases Show Activity in Mammalian Cells.

We prescreened the KD, R, B2, and B3 recombinases for activity in cultured Chinese hamster ovary (CHO) cells at 37 °C by cotransfecting CMV-GAL4::p65 with a UAS recombinase construct and its cognate stop-cassette reporter construct. These preliminary experiments suggested that B3 and KD were considerably more active under these conditions than R and B2. To further confirm B3 and KD activity, CHO cells were transfected with four plasmids: CMV-GAL4::p65 to drive transcription of the genes encoded by the other plasmids; 10XUAS-myr::GFP (pJFRC12; ref. 43) to provide a measure of transfection efficiency; 20XUAS B3 or 20XUAS KD recombinase; and a reporter construct with the appropriate stop cassette between the UAS sites and the RFP gene (Fig. 3). Recombinase activity was estimated as the fraction of GFP-expressing cells that also express RFP after correction for the recombination-independent background expression of RFP observed in some cells (Fig. 3B). Although the need to subtract this background (< 10% of transfected cells for both KD and B3) limited the precision with which we could estimate recombinase efficiency, these data demonstrate that both B3 (65% of transfected cells showing excision of the stop cassette) and KD (32%) were functional in mammalian cells at 37 °C, with B3 appearing more active than KD.

Fig. 3.

Activity of the B3 recombinase in cultured Chinese hamster ovary cells. Images of cultured Chinese hamster ovary cells are shown. A and C have been imaged to reveal GFP fluorescence, and B and D show RFP fluorescence. Cells in all images have been transfected with a plasmid that expresses the transcriptional activator GAL4, a plasmid that expresses membrane localized GFP in response to GAL4 (which serves to identify those cells that have been successfully transfected) and a UAS–B3RT–stop cassette–B3RT–RFP construct. RFP expression is only expected if B3 recombinase is expressed and is able to remove the stop cassette, which is flanked by B3 recombination targets, allowing transcription of the RFP gene. A construct encoding the B3 recombinase (UAS-B3 recombinase) was included in the mixture of transfected DNAs in C and D. The observation that a high proportion of the cells marked by GFP in C show RFP fluorescence in D indicates that the B3 recombinase is active in CHO cells. A background of cells that express RFP weakly is seen in B; presumably, this expression results from either transcription reading through the stop cassette or internal transcriptional initiation.

Based on these initial results, we tested the B3 and KD recombinases in cortical neurons in the mouse brain. Layer 2/3 progenitor cells were transduced by in utero electroporation (refs. 50 and 51; embryonic day 16) with two separate plasmids. One plasmid drove expression of B3 or KD under control of the synapsin promoter (refs. 52 and 53; syn-B3 and syn-KD, respectively). The other plasmid contained the tdTomato gene (44) driven by the CAG promoter (54–56), but separated from the promoter by either the B3 or KD-specific stop cassette (CAG_BlownOUT_tdTomato and CAG_KickedOUT_tdTomato, respectively).

Brains were harvested from mature mice (28–35 d postnatal) and inspected for tdTomato expression. Mouse brains that expressed matching recombinases and reporter constructs typically harbored numerous red fluorescent neurons in layer 2/3 (syn-B3 plus CAG_BlownOUT_tdTomato, 9/10 mice; syn-KD plus CAG_KickedOUT_tdTomato, 5/7 mice). Neurons were strongly labeled with tdTomato without obvious signs of toxicity. The fluorescent neurons appeared to have normal dendritic morphology, and local and long-range axonal projections (Fig. 4). The absence of fluorescence in some brains is likely explained by the failure rate, ≈20%, of the in utero electroporation technique. These experiments show that KD and B3 are active in the mouse brain. Fluorescent neurons were never detected in brains transduced with unmatched recombinases and reporter constructs (syn-B3 plus CAG_KickedOUT_tdTomato, 0/5 mice; syn-KD plus CAG_BlownOUT_tdTomato, 0/3 mice), suggesting that, as in flies, B3 and KD are not cross-reactive.

Fig. 4.

Activity of the B3 and KD recombinases in mice. Plasmids encoding the recombinase and a reporter of recombinase activity were introduced into embryonic day 16 embryos by in utero electroporation. Removal of the transcriptional stop cassette flanked by recombinase target sites causes tdTomato expression. Imaging of tdTomato fluorescence was in brain section (thickness, 50 microns) from an adult mouse. (A) KD recombinase and a reporter for KD activity were introduced into layer 2/3 cells. Imaging was with a macroscope. Arrows (from top to bottom) correspond to layer 2/3 pyramidal cells, their local axons in layer 5, and their long-range axons in the corpus callosum. (Scale bar: 1 mm.) (B) Same as A for the B3 recombinase. (C–E) Different regions of interest in one brain section imaged with confocal microscopy (B3 recombinase): (C) dendritic arbors of layer 2/3 pyramidal neurons; (D) local axonal projections from layer 2/3 neurons in layer 5; (E) long-range projections in the corpus callosum. (Scale bar: 100 microns.)

Further experiments will be required to compare the activity and toxicity of these recombinases with that of FLP and Cre. In the case of FLP, engineering the protein to be more thermostable significantly increased activity (57, 58) and similar efforts could be applied here if necessary. However, based on the cell culture assays, it appears that B3 is already highly active at 37 °C.

Concluding Remarks

A variety of methods that are based on site-specific recombination have been used in Drosophila (3, 7–11), but the ability to apply multiple methods, or methods that require multiple recombinases, in the same animal has been severely limited by the lack of a set of non–cross-reacting, nontoxic recombinases. The results we present here remove this limitation and establish a robust toolkit of four distinct recombinases—KD, B2, B3, and FLP—for use in Drosophila. Each has sufficient activity to drive an excision reaction to completion without displaying detectable cross-reactivity or obvious toxicity. We also show that at least two of the recombinases, KD and B3, are active in mammalian tissues.

Methods

Molecular Biology and Drosophila Genetics.

Constructs for use in Drosophila and transgenic fly lines were generated by standard methods as described in refs. 43 and 45; see SI Methods for details. Drosophila codon-optimized recombinases and their cognate recognition sites were synthesized by DNA2.0, Inc.; complete DNA sequences are given in Fig. S1, Fig. S2, and Fig. S3. Plasmid constructs are available from Addgene.

Imaging of Optic Lobes.

Optic lobes from 5- to 10-d adult flies, mounted in an appropriate orientation, were imaged on a Zeiss LSM 710 confocal microscope by using a 20× 0.8 NA objective. For further details, see SI Methods.

CHO assays.

CHO-K1 cells (ATCC CCL-61) were plated at a density of 4 × 10⁵ in 35 mm MatTEk culture plates (Mattek). The next day, a total of 4 μg of plasmid DNA in 250 μL of Opti-MEM I Reduced Serum Medium (Invitrogen) were combined with Lipofectamine 2000 (Invitrogen) and used to transfect CHO-K1 cells according to manufacturer recommendations. GFP and RFP fluorescence were visualized ≈24 h after transfection by using a Zeiss LSM 510 confocal microscope.

In Vivo Mouse Assays.

Approximately 1 μL of DNA solution consisting of a mixture of plasmids in a 1:1 ratio at a final DNA concentration of 2 μg/μL was injected into the right lateral ventricle of each embryo (embryonic day 16 C57BL/6J mice) and then electroporated as described in SI Methods. The plasmids used for mouse in utero electroporation were based on an Adeno-associated virus (AAV) backbone (56); see SI Methods for details.