Redesign of extensive protein–DNA interfaces of meganucleases using iterative cycles of in vitro compartmentalization

Identification of Genomic Sequences to Be Targeted by Engineered Meganucleases.

To develop a strategy to efficiently retarget multiple types of meganucleases to genomic targets of interest, we selected two genome loci from the human genome for targeted modification on the basis of our research interest in potential therapeutic applications for hemoglobinopathies and cystic fibrosis (Fig. 1B): the exon 4 of bcl11a [which encodes a transcription regulator involved in B lymphopoiesis, erythropoiesis, and regulation of globin expression (27)] and the exon 11 in cftr [which encodes a transmembrane conductance regulator which is dysfunctional in individuals with cystic fibrosis, most often as a result of deletion of codon 508 within that exon (28)].

Using a publically available bioinformatic meganuclease Web server (LAHEDES; www.homingendonuclease.net) (26), we searched each of these regions for 22 base pair sequences that displayed significant identity to DNA targets for wild-type, monomeric meganucleases that have been crystallized in complex with their DNA targets (Fig. S2). An increasing number of such structures are now available, including I-OnuI (PDB ID 3QQY), I-LtrI (3R7P), I-PanMI (4JEE), I-GzeMII (4EFJ), I-HjeI (3UVF) I-LtrWI (4LQ0), and I-SmaMI (4LOX). We excluded candidate target sites that contained multiple nucleotide mismatches within their central four base pairs compared with the wild-type endonuclease targets because base substitutions at those base positions may lead to DNA backbone distortion that significantly compromises meganuclease specificity and activity (29).

We ultimately chose three target sites within the two target loci that exhibited 50–70% identity against the target sequences recognized by different wild-type meganucleases (I-OnuI, I-PanMI, and I-HjeMI) (Fig. 1B). Two of the three sites are situated in exons, and the other is in a flanking intron region. A site for a redesigned I-PanMI meganuclease (CFTR2) spans a well-known human single-nucleotide polymorphism (SNP) (A/G in cftr exon 11, rs213950; www.ensembl.org/). We therefore also sought to create two variant endonucleases that each act respectively at only one of the SNP variants (to assess whether this approach can produce reagents to discriminate between such closely related DNA sequences).

Reprogramming of Meganuclease Specificity.

Structural studies of meganucleases have demonstrated that two to four consecutive nucleotide base pairs within a DNA target site are generally recognized by a cluster of six to nine amino acid residues (26). These protein regions are termed “contact modules” and are cataloged for each meganuclease in the LAHEDES Web server. We randomized amino acid residues within those modules to alter the specificity of each meganuclease at altered base positions and constructed variant endonuclease libraries by overlap extension PCR using degenerate primers. Oligonucleotides used for redesign of meganucleases and amino acid positions subjected to randomization are shown in Dataset S1.

Shuffling protein residues at up to nine positions gives rise to a large theoretical number of unique variant endonucleases (as large as 20⁹, or ∼5 × 10¹¹). To screen as many unique protein variants as possible from randomized libraries, we modified a previously described in vitro compartmentalization system (23, 25), where DNA fragments encoding variant endonucleases are individually expressed in compartmentalized, aqueous droplets that contain an in vitro transcription/translation reaction mixture (Fig. 1A and Fig. S1). In each droplet, a variant endonuclease has the opportunity to bind and cleave a target site that is placed in the same DNA construct as for its own gene, allowing us to couple the cleavage activity against the target site to the corresponding meganuclease gene.

To identify variant enzymes that are capable of cleaving a target site containing a cluster of altered base pairs, we carried out two to three rounds of selection under increasing levels of stringency (by reducing time periods for in vitro protein synthesis and DNA strand cleavage and/or by elevating temperatures during each round; see Materials and Methods for more detail). We continued with an iterative approach, in which variant genes enriched after the first rounds of site-directed mutagenesis were used as templates for further randomization of additional amino acids and subsequent selections against a target site containing more base pair substitution(s) (as illustrated in Fig. S3). Using this strategy, we redesigned the N- and C-terminal half domains of meganucleases to engineer variant enzymes that cleaved chimeric target sites composed of one half of human genomic target sites and one half of the original meganuclease target sites. We then shuffled together the engineered N- and C-terminal half domains, and identified a collection of variant endonucleases that recognized human genomic target sequences in the IVC selection system.

We then filtered these populations of active meganucleases through cleavage assays in bacteria (30) to select individual clones that displayed substantial activity in living cells. In this assay, meganucleases that are expressed in individual bacterial cells can rescue the cell growth only when they recognize and cleave their target sequences on reporter plasmids that bacterial cells harbor, leading to elimination of the reporter plasmid that can induce expression of a toxic gene. If necessary, random mutagenesis was introduced to incorporate additional point mutations in the entire ORFs of variant endonuclease genes that might increase overall activity and/or stability.

The final engineered endonucleases all efficiently increased the survival rate of bacterial cells that harbored reporter plasmids containing their target sites but not those containing the original target sites recognized by their parental meganucleases (Fig. 2A). Sequencing of the retargeted meganucleases indicated that 21–28 amino acid substitutions were introduced into each wild-type endonuclease scaffold, with a small number of substitutions (0–5) found outside of the protein–DNA interface (Fig. S4). The two I-PanMI–derived meganucleases (termed “CFTRpan/A” and “CFTRpan/G”) that targeted two SNP variants within the CFTR2 site displayed excellent discrimination between those targets. Out of total of nine amino acid residues that differ between those redesigned meganucleases, six were situated on the DNA interface adjacent to the SNP base position.

Targeted Mutagenesis by Redesigned Meganucleases at Their Endogenous Target Sites in Human Cells.

We next examined if the retargeted meganucleases could induce targeted modification at their endogenous sites in human embryonic kidney (HEK) 293T cells. Sequencing of the cftr exon 11 indicated that this cell line harbored only the SNP variant A; we therefore excluded the CFTRpan/G endonuclease from this analysis. All three engineered endonucleases were expressed at similar levels (Fig. S5A). We carried out T7 endonuclease I assays (31) to estimate the accumulation of small nucleotide insertions and deletions (indels) induced by the respective meganucleases. These assays revealed that expression of each endonuclease [either alone or in concert with the Trex2 exonuclease (32)] resulted in indels at the endogenous target sites in unsorted cell populations, with frequencies ranging from 2.5 to 34% (Fig. 2B).

The range of indel frequencies induced by the three redesigned enzymes at their corresponding target sites displayed considerably greater variation than their corresponding activities in the bacterial gene elimination assay (Fig. 2A). Most notably, the two cftr-targeting meganucleases (CFTRonu and CFTRpan/A) induce indels at their respective human genomic target sites (that are only 250 base pairs apart) with at least 10-fold different efficiencies. Additional assays in the same cell line, using an Episomal Direct Repeat (DR)-GFP reporter (33), also indicated that CFTRonu was more active (Fig. S5B), suggesting that the wide range of activities displayed by these redesigned meganucleases in human cells is not caused solely by differences in chromatin structures and/or nucleosome positioning at their target sites (34). Bioinformatic search for sites that were similar to the on-targets for the engineered meganucleases indicated that the human genome contains far more closely related sites (which contain no more than three base pair differences) for CFTRpan/A (a total of 32 sequences) than for CFTRonu (only eight such sites) (Table S1). It is therefore possible that a subset of closely related off-target sites may trap meganucleases and thereby act as competitive inhibitors because meganucleases have been shown to display more promiscuous binding specificity than DNA strand cleavage specificity (22, 35). However, we cannot rule out the possibility that these two meganucleases may have significant differences in activity that are not fully recapitulated in bacterial nuclease assays.

The same assays conducted using episomal DR-GFP reporter plasmids verified that the two I-PanMI–derived endonucleases discriminated between the SNP variants of the CFTR2 site (Fig. S5B), although the signal to noise ratio in these assays is relatively low because of spontaneous homologous recombination occurring within the episomal reporters.

Impact of Tethered TAL Effectors on Targeted Mutagenesis by Retargeted Meganucleases.

We next examined whether tethering engineered meganucleases to TAL effectors (as additional DNA binding domains) would improve the efficiency of targeted genome modification in human cells as recently described (21). The two redesigned meganucleases that targeted the cftr locus were each tethered to a TAL effector DNA binding domain (termed “N∆148/C+63,” a construct that spans from residue 149 of the canonical TAL effector N-terminal sequence to residue +63 beyond the end of the TAL effector repeat sequences). The TAL effectors were individually designed to target a 10- or 12-nucleotide sequence that was six base pairs upstream of each meganuclease target (Fig. 2C). The frequencies of indels generated by both engineered meganucleases at the same endogenous target sites were increased after incorporation of the nuclease into the MegaTAL architecture, and at least a 10-fold improved efficiency was observed in the case of CFTRpan/A (Fig. 2B).

Targeted Deletion by Multiplexed Coexpression of Two MegaTALs.

Previous studies have demonstrated that introduction of site-specific double strand breaks at two genomic loci promotes targeted deletion and insertion (36–38), which expand the range of genome engineering applications. We therefore examined whether a pair of MegaTALs were capable of inducing targeted chromosomal deletions. Expression of two MegaTALs that targeted sites ∼250 base pairs apart gave rise to three DNA bands when a genomic region spanning their target sites was PCR-amplified using a single set of primers (Fig. 3A). Sequencing of the smallest DNA band revealed that approximately a half of the targeted deletions (DEL1: 6 clones out of 12) were generated through three-base microhomology-mediated end joining between 3′, four-base overhangs generated by the two engineered meganucleases without any end processing (Fig. 3B). Sequencing of the larger DNA band identified two independent events: one was a chromosomal duplication, whereas the other was a targeted inversion and insertion of a small fragment derived from an expression plasmid for the two MegaTALs (Fig. S6A). Sequencing of clones harboring the middle-sized DNA band identified one inversion event (out of 30 clones sequenced: Fig. S6B). Deletion, duplication, and insertion promoted by the MegaTALs were substantially suppressed by coexpression of Trex2 (Fig. 3A), in a good agreement with a previous study (39).

Off-Target Cleavage Activity by an Engineered Meganuclease.

To assess off-target mutagenesis by engineered meganucleases, we assayed indels in cells expressing the CFTRonu meganuclease at 8 genomic sites that were the closest to its on-target sequence (all containing 3 base substitutions from the CFTR1 target site: Fig. 4A). When Trex2 was coexpressed, both the stand-alone CFTRonu and its corresponding MegaTAL promoted low but detectable levels of indels at two of the off-target sites (Fig. 4B). Comparable frequencies of off-target mutagenesis were induced by the two types of meganuclease constructs, suggesting that the fused TAL effector had little impact on binding and cleavage activity of an engineered meganuclease at off-target sites, presumably because of high binding affinity of the stand-alone nuclease toward the noncognate DNA target sequences. These results are consistent with our earlier study of MegaTALs (21).

To reduce the level of undesired off-target mutagenesis, further introduction of mutagenesis into the engineered meganuclease and counterselection against the off-target sites might be useful. However, an alternative (and perhaps more straightforward) approach that could also improve specificity might be the simple incorporation of a point mutation in the nuclease domain that attenuates its overall cleavage activity, to increase the dependence of cleavage activity on simultaneous, cooperative recognition of the full-length target by the two DNA binding domains (i.e., the TAL effector and the meganuclease). Such point mutations can be easily identified based on crystal structures. This concept is illustrated by the results of experiments using the CFTRonu variant meganuclease that contains a single substitution (L49Q) in the vicinity of the active site: this point mutation suppressed the formation of indels at both the on- and off-target sites (Fig. 4B). Targeted modification was specifically enhanced at the on-target site by incorporating the CFTRonu L49Q meganuclease into the MegaTAL platform, although the efficiency of on-target mutagenesis was significantly compromised relative to that induced by constructs containing the original CFTRonu meganuclease.

IVC.

The ORFs of meganucleases, I-OnuI E178D, I-HjeMI, and I-PanMI, were cloned between NcoI and NotI sites of pET21d(+) (EMD Millipore). The sequences are shown in Dataset S1. To introduce site-directed saturation mutagenesis into the ORFs, sequences containing their partial ORFs with regions (∼20 base pairs) that overlapped adjacent PCR fragments at both ends were amplified in separate tubes, using primers that contained degenerate codon NNK (coding all 20 amino acids) (Fig. S1A). The positions of amino acid residues shuffled and PCR primers used are also shown in Dataset S1. PCR products were purified by extraction from agarose gels and assembled in the subsequent round of PCR with a sequence containing two copies of target sites for variant endonucleases to construct DNA libraries (Fig. S1B). A successfully assembled DNA fragment was again purified by gel extraction (Fig. S1C).

Two to three rounds of IVC were conducted following each round of site-directed saturation mutagenesis to enrich active variant genes. The oil–surfactant mixture [2% (vol/vol) ABIL EM 90 (gift from Evonik Industries AG Personal Care) and 0.05% Triton X-100 in light mineral oil] was thoroughly mixed with the saturation buffer [100 mM potassium glutamate (pH 7.5), 10 mM magnesium acetate (pH 7.5), 1 mM DTT, and 5 mg/mL BSA], incubated at 37 °C for 20 min, and centrifuged at 16,000 × g for 15 min at 4 °C. Five hundred microliters of the upper phase was used to emulsify 30 µL of the in vitro protein synthesis mixture [25 μL of PURExpress (New England Biolabs), 20 units of RNase inhibitor, 1 mg/mL BSA, and a DNA library] by constant stirring at 1,400 r.p.m. for 3.5 min on ice (Fig. 1A and Fig. S1D). Eight nanograms of a DNA library were added in the aqueous phase in the first round of IVC and reduced to 1 ng or 0.5 ng in the second or third round, respectively. The emulsion was incubated at 30 °C for 4 h in the first round of IVC and then heated at 75 °C for 15 min. After an addition of 170 µL of 10 mM Tris⋅HCl (pH 8.0), emulsified droplets were collected by centrifugation at 16,000 × g for 15 min at 4 °C and broken by an addition of phenol/chloroform/isoamyl alcohol. Nucleic acids were recovered by isopropanol precipitation and treated with 5 µL of RNase Mixture Enzyme Mix (Life Technologies) at 37 °C for 30 min. After purification using QIAquick PCR purification kit (Qiagen), a gene library was ligated to a more than 100-fold excess of a (unphosphorylated) DNA adaptor with a 4-base 3′ overhang sequence complementary to a sticky end of target sites generated by variant endonucleases during IVC (Fig. S1E) and added to PCR mixtures containing a pair of primers, one of which was specific to a DNA adaptor to enrich ORFs of variant genes coupled to cleaved target sites (Fig. S1F). Sequences of adaptors and adaptor-specific primers used in this study are shown in Dataset S1. PCR products were gel-purified, and sequences encoding variant endonucleases were further PCR-amplified (Fig. S1G) and assembled with a sequence containing the T7 promoter and one containing two copies of a target site (Fig. S1H).

In the second round of selection, compartmentalized reaction mixtures were incubated at 42 °C for 75 min, and the third round of IVC was carried out at the same temperature for 30 min.

Two-Plasmid Cleavage Assay in Bacteria.

The assays were carried out based on previous studies (30, 41). Briefly, the NovaXGF′ competent cells harboring a pCcdB reporter plasmid (that contained four copies of a target site) were transformed with the Endo plasmid encoding a meganuclease by electroporation. The transformants were grown in the 2 × YT medium (16 g/L tryptone, 10 g/L yeast extract, 5.0 g/L NaCl) supplemented with 0.02% L-arabinose at 37 °C for an hour to induce expression of meganuclease genes and spread on both the nonselective plates (1 × M9 salt, 1% glycerol, 0.8% tryptone, 1 mM MgSO₄, 1 mM CaCl₂, 2 µL/mL thiamine, and 100 µg/mL carbenicillin) and the selective plates (the nonselective plates supplemented with 0.02% L-arabinose and 0.4 mM isopropyl β-D-1-thiogalactopyranoside to induce expression of the toxic CcdB protein). After incubation at 37 °C for 18–24 h, colonies were counted to calculate the survival rates on the selective plates.

I-OnuI variant endonucleases that were engineered through IVC were cloned between NcoI and NotI sites of pEndo and subjected to two rounds of selection under the conditions described above. The pEndo plasmids were extracted from individual colonies and sequenced.

When engineered meganucleases derived from I-HjeMI and I-PanMI were screened, a pool of variant genes were subjected to selection under the conditions where meganuclease genes were expressed in the 2 × YT medium containing 0.02% L-arabinose and 100 µg/mL carbenicillin at 30 °C for 15 h before plating as described in our previous study (41). Surviving colonies were scraped off from the selective plates, and random mutagenesis was introduced into the variant genes using Gene Morph II Random Mutagenesis Kit (Agilent Technologies). The resultant PCR fragments were cloned into the same vector and screened with an increasing stringency through four rounds of selection. Expression of endonuclease genes was induced at 30 °C for 4 h before plating in the first two rounds (plates were incubated at 30 °C) and at 37 °C for an hour in the subsequent two rounds (plates were incubated at 37 °C). ORFs of variant endonucleases were PCR-amplified from survival clones and cloned into the pEndo vector containing the C-terminal tag (AANDENYAAAV) that facilitates proteolytic degradation in Escherichia coli as described previously (42). The sequence of the plasmid (termed “pEndo/Degron”) is shown in Dataset S1. Two more rounds of selection were carried out under the same conditions as for the last two rounds, and ORFs of retargeted meganuclease genes recovered from survival colonies were sequenced. We expected that incorporation of the C-terminal degradation tag would accelerate turnover of a meganuclease, facilitating identification of engineered enzymes with higher activity; however, relatively little improvement was observed.

Plasmids Used for Transfection of HEK 293T Cells.

Meganuclease and MegaTAL genes with the N-terminal HA epitope tag and nuclear localization signal and the mCherry fluorescent gene were individually cloned into pExodus (32) (Addgene plasmid 39991) using Gibson Assembly Master Mix (New England Biolabs) or standard molecular cloning techniques. Double-stranded oligonucleotides containing each of the target sites for meganucleases and a stop codon were inserted into the DR-GFP reporter plasmid (33). Sequences of individual plasmids used in this study are shown in Dataset S1.

T7EI Assay for Estimating Indels.

HEK 293T cells were seeded in 0.9 mL of Opti-MEM (Life Technology) supplemented with 5% (vol/vol) FBS at 1.6–1.8 × 10⁵ cells per well in 12-well plates 1 d before transfection and transfected with 0.4 μg each of the pExodus plasmid for expression of a retargeted meganuclease or MegaTAL and pExodus CMV.Trex2, using X-tremeGene 9 transfection reagent (Roche). The pUC19 plasmid was used as a control. Three days after transfection, genomic DNA was extracted using Quick-gDNA miniprep kit (Zymo Research), and genomic regions spanning a target site were PCR-amplified using Phusion DNA polymerase (New England Biolabs). Primers used are shown in Table S2. After purification using DNA clean and concentrator-5 kit (Zymo Research), 200–400 ng of double-stranded fragments were heated in 10 μL of 1 × NEBuffer 2 at 95 °C for 5 min and slowly cooled down (95 °C to 85 °C at −2 °C/s and 85 °C to 25 °C at −0.1 °C/s). Reannealed DNA was incubated with 5 units of T7EI (New England Biolabs) at 37 °C for 60 min, and the reaction was terminated by an addition of 4 × loading dye [40 mM Tris⋅HCl (pH8.0), 4 mM EDTA, 0.4% SDS, 20% (vol/vol) glycerol, 0.1% orange G, and 0.4 mg/mL proteinase K]. DNA fragments were separated on a 2% (wt/vol) agarose-Tris-borate-EDTA (TBE) gel. The frequency of indel was estimated using the following equation as described previously (43): (% indel) = 100 × (1 − (1 − fraction cleaved)^1/2).

Chromosomal Modifications Generated by Coexpression of the Two MegaTALs.

HEK 293T cells were transfected with 0.2 μg each of the two pExodus plasmids expressing MegaTALs and 0.4 μg of pExodus CMV.Trex2, as described above. Three days after transfection, genomic DNA was extracted, and the genomic target region was PCR-amplified using a pair of primers (5′-TCT GAA TCA TGT GCC CCT TCT C-3′ and 5′-AAC CAT TGA GGA CGT TTG TCT C-3′). PCR fragments were separated on a 1.5% (wt/vol) agarose-TBE gel. Gel-extracted PCR fragments were cloned using TOPO TA cloning kit (Life Technology) and sequenced.

Episomal GFP Reporter Gene Recombination Assay.

HEK 293T cells were transfected with 0.36 μg each of the pExodus plasmid for meganucleases and a DR-GFP reporter plasmid containing a target site and 0.08 μg of pExodus-mCherry, as described above. Two days after transfection, cells were analyzed by flow cytometry, and the transfection efficiency was normalized based on a fraction of mCherry-positive cells.

Western Blotting.

Transfected cells were lysed in 20 mM Tris⋅HCl (pH 6.8), 500 mM NaCl, 1 mM EDTA, and 1% Triton X-100 and centrifuged at 16,000 × g for 30 min at 4 °C following incubation on ice for 20 min. Three micrograms of each supernatant were separated on a 15% (wt/vol) polyacrylamide-SDS gel and transferred to a PVDF membrane. HA-tagged proteins and β-tubulin were detected using anti-HA polyclonal antibody (1:200; Y-11; Santa Cruz) and anti-β-tubulin monoclonal antibody (1:500; clone TUB 2.1; Sigma Aldrich) as primary antibodies, and ECL anti-rabbit IgG HRP-linked whole antibody (1:25,000) and ECL anti-mouse IgG HRP-linked whole antibody (1:4,000; GE Healthcare) as secondary antibodies. Protein–antibody complexes were detected using ECL Western Blotting Detection reagent (GE Healthcare Life Sciences) and X-ray films (Thermo Scientific).