Mutations in the β-globin gene (HBB) cause haemoglobinopathies where current treatments have serious limitations. Gene correction by homologous recombination (HR) is an attractive approach to gene therapy for such diseases and is stimulated by gene-specific endonucleases, including zinc finger nucleases (ZFNs). Customised nucleases targeting HBB have previously been shown to promote HR-mediated HBB modification in 0.3–60% of drug-selected cells, although frequencies among unselected cells, more relevant to the goal of correcting HBB in primary stem cells, have not been reported.
ZFNs targeting HBB were tested for HBB binding (two-hybrid assay) or HBB cleavage followed by inaccurate end joining (surveyor assay) in bacteria or human cancer cell lines, respectively. ZFN-stimulated HR was measured in cell lines by a modified fluorescence-based reporter assay or by targeted insertion of a drug-resistance marker into endogenous HBB confirmed by Southern analyses.
Although the ZFNs that we assembled in-house showed limited potential, a commercially commissioned nuclease (ZFN4) enhanced HR-mediated HBB modification in up to 95% of drug-selected cells. Among unselected cells, however, this frequency was less than 0.2%. Furthermore, ZFN4 cleaved HBB at an efficiency of 1–2% (surveyor assay) and enhanced the HR reporter assay 20-fold less efficiently than a control endonuclease.
The haemoglobinopathies β-thalassaemia major (β-TM)  and sickle cell anaemia (SCA)  are among the most common monogenic diseases worldwide and both are caused by mutations in the β-globin gene (HBB). Current treatments are blood transfusions and allogeneic haematopoietic stem cell (HSC) transplantation, although these have major disadvantages and limitations, including iron overload, expense and restricted donor availability [3, 4]. Gene addition therapy in autologous HSCs is an attractive alternative [5, 6] that has been successful for the treatment of the rare monogenic disease X-linked severe combined immune deficiency (Xl-SCID) . A major limitation of this approach, however, is the uncontrolled integration of DNA into the recipient genome. This may affect host genes located at, or close to, the chromosomal integration site, with potentially serious consequences. For example, a vector enhancer may activate a nearby proto-oncogene, as was seen in some Xl-SCID patients who developed leukaemia during gene addition therapy [8, 9]. Furthermore, any randomly integrating DNA has the potential to be oncogenic via insertional inactivation of a tumour suppressor gene. So far, gene addition therapy for β-TM has been described for only one patient where viral vector integration activated of a host gene (HMGA2) previously implicated in tumorigenesis . Apart from such safety concerns, gene addition is unlikely to fully suppress the effects of dominant disease-causing mutations and does not guarantee stable and physiological expression of the therapeutic gene. Physiological expression is particularly important for β-globin gene therapy to ensure expression in the appropriate lineage at the same time as avoiding an excess of β-globin that may cause β-thalassaemia-like syndrome .
Therapeutic gene targeting offers a way around these difficulties [12-14]. In gene targeting, a delivered DNA template has homology to a chromosomal target locus with which it undergoes homologous recombination (HR), a cellular mechanism for repairing chromosomal double-strand breaks (DSBs). In this way, defined sequence alterations engineered into the template can be introduced into the chromosomal target locus. This contrasts with gene addition where the delivered DNA integrates at unpredictable loci using the nonhomologous end-joining (NHEJ) DSB repair pathway. Unlike gene addition, gene targeting (or gene editing) can be used for gene correction [i.e. the removal of disease-causing (including dominant) mutations], thereby restoring stable physiological expression, without the risks of genotoxic side effects. Alternatively, expression cassettes can be targeted to a benign chromosomal locus that acts as a ‘safe harbour’ [15, 16].
The principle of therapeutic gene targeting of HBB has been demonstrated in mouse embryonic stem cells [17, 18] and induced pluripotent stem cells (iPSCs) [19, 20], where extremely rare targeted cells (e.g. 2/107)  can be selected and expanded in culture without losing pluripotency. Significant barriers to the clinical use of corrected iPSCs remain, however, including inefficient reprogramming of iPSCs into engraftable HSCs or red blood cells [21-23] and safety concerns relating to the methods used to generate iPSCs . These restrictions would not apply to corrected primary HSCs but, unlike corrected iPSCs, HSCs are notoriously difficult to grow and select in cell culture without losing pluripotency . Methods for increasing the absolute frequencies of gene targeting (i.e. frequencies prior any selection/enrichment procedure) are therefore especially important in this context.
Cleavage of chromosomal target genes greatly stimulates gene targeting  and this can be achieved with customised gene-specific endonucleases including zinc finger nucleases (ZFNs), transcription activator-like effector nucleases (TALENs) and clustered regularly interspaced short palindromic repeats(CRISPR)-associated systems . HBB-specific ZFNs [27, 28] or TALENs [29, 30] have been developed with a view to gene correction in iPSCs and shown to promote gene targeting to frequencies as high as 60% of drug-selected cells. Absolute frequencies of HBB gene targeting have not been reported, however, and are likely to be much lower. In the present study, we have characterised the ability of an HBB-specific ZFN (ZFN4) to cleave its target sequence and promote HR. We detected targeting frequencies among drug-selected cells as high as 95%, although absolute frequencies were less than 0.2%. We also used a modified HR assay indicating that ZFN4 promotes gene conversion approximately 20-fold less efficiently than the gold standard endonuclease I-SceI. Taken together, these data suggests that improved HBB-specific endonucleases will be required to promote primary HSC-mediated gene correction therapy for haemoglobinopathies.
Materials and methods
Modular assembly of zinc fingers
The ZF Consortium Modular Assembly Zinc Finger Set, cloned into pc3XB, was obtained from Addgene (Cambridge, MA, USA). ZF modules were chosen using ZiFiT software  and assembled at the DNA level, as described previously . The order of the modules was: ZFN1-L: ZF-92-103-89-90; ZFN1-R: ZF-60-71-106-94; ZFN2-L: ZF-69-96-74-103; ZFN2-R: ZF-106-75-82-77; ZFN3-L: ZF-73-93-77-77; ZFN3-R: ZF-73-104-83-93.
Bacterial two-hybrid assays
Detailed protocols  were followed using the ZF Consortium Assembly Accessory Reagents (Addgene). Briefly, DNA for each ZF assembly was subcloned into pGP-FF, for expression as fusion with Gal11P, and introduced in Escherichia coli carrying pAC-KAN-alphaGal4 and a reporter construct (pBAC-lacZ) carrying the appropriate recognition sequence. The positive control cells carried the reporter pBAC-BA-lacZ and the Gal11P fusion plasmid pGP-FB-origBA. β-galactosidase activity was expressed as the fold-increase in the rate of substrate cleavage by test cell lysates relative to control lysates from equivalent cells expressing unmodified Gal-11.
Development of ZFN4
The CompoZr Custom ZFN Application (Sigma Life Science, St Louis, MO, USA) was used. In response to our requirements, 16 candidate ZFNs were identified using Sangamo SuperFinder algorithms (Sangamo Biosciences, Richmond, CA, USA). We chose one of these that had the maximum (n = 5) mismatches with the δ-globin gene. Two expression plasmids, one for each subunit, were supplied, as well as in vitro transcribed mRNA and initial Cel-1 assay data. ZFN4, apparently co-commissioned by another group , is currently available as an ‘off-the-peg’ product (CKOZFN1264; Sigma Life Science).
pDR-GFP-ZFN4 was made by ligating the annealed oligonucleotides 5'-CAGGGTAATATAGGTCTGCCGTTACTGCCCTGTGGGGCAAGGTGAACGTGGATGAATAA-3' and 5'-TCATCCACGTTCACCTTGCCCCACAGGGCAGTAACGGCAGACCTATATTACCCTGTTAT-3' into the I-SceI site of DR-GFP . To make pTV-TC3, Phusion high fidelity DNA polymerase (Thermo Scientific, Waltham, MA, USA) was used to amplify homology arm DNA from K562 genomic DNA. Left arm homology (2 kb) was amplified with primers 5'-CATTGTGCGGCCGCATATCAGGGATGTGAAACAGGGTC-3' and 5'-CATTGTGAATTCTCCTCAGGAGTCAGATGCAC-3', digested with NotI and EcoR1 and cloned into pBL-Puro  to make pTV-TC1. Right arm homology (2.4 kb) was amplified with primers 5'-CATTGTATCGATCCTGGGCAGGTTGGTATCAAG-3' and 5'-TGACTGGGAGAGAGGACAAGGAC-3', digested with ClaI and KpnI and cloned into ClaI/KpnI-cut pTV-TC1 to make pTV-TC2. Finally, a 5.5-kb ApaI/KpnI fragment from pTV-TC2 was cloned into ApaI/KpnI-cut pBluescript II KS + (Agilent Technologies) to generate pTV-TC3 whose arms of homology were 1.3 kb (left) and 2.4 kb (right).
Cell culture and transfection
HT1080 fibrosarcoma cells were cultured, as described previously . K562 erythroleukaemia cells were cultured similarly but in RPMI 1640 medium (Gibco, Gaithersburg, MD, USA). Transient transfection of HT1080 cells was carried out using Lipofectamine 2000 (Invitrogen, Carlsbad, CA, USA) in accordance with the manufacturer's instructions. A total of 4 µg and 0.8 µg of nucleic acid/well was used for six-well and 24-well lipofections, respectively. Typical lipofection efficiencies were >90%. Stable transfection of HT1080 was by electroporation, as described previously  selecting in 0.4 µg/ml puromycin. All transfections of K562 cells were by nucleofection using an Amaxa Nucleofector I (Lonza, Basel, Switzerland) set to program T016 with compatible transfection reagents (Amaxa Nucleofector Kit V; Lonza) and selection, where necessary, was in 0.4 µg/ml puromycin. All transfection efficiencies for lipofection, electroporation and nucleofection were determined by delivery of a green fluorescent protein (GFP) reporter plasmid under identical conditions and flow cytometric analysis at 48 h post-transfection.
Genomic DNA and polymerase chain reaction (PCR)
Genomic DNA was made from one to two million HT1080 or K562 cells, resuspending in a final volume of 100 µl of Tris/ethylenediaminetetraacetic acid (pH 8). For HT1080, a plate lysis method was used , whereas, for K562, Wizard reagents (Promega, Madison, WI, USA) were used in accordance with the manufacturer's instructions. Each PCR reaction (50 µl) contained 1 µl (approximately 60 ng) of genomic DNA. Alternatively, small numbers (approximately 103) of cells were incubated at 50 °C for 60 min and then at 98 °C for 10 min in PCR buffer (25 µl) containing proteinase K (100 µg/ml), Tween 20 (0.2%) and NP40 (0.2%); PCR reagents at 2 × final concentrations were then added in 25 µl of PCR buffer. Unless stated otherwise, amplification was with Dreamtaq Polymerase (Fermentas, Glen Burnie, MD, USA).
Surveyor (Cel-I) and MwoI cleavage assays
K562 cells (106) were nucleofected, or HT1080 cells (5 × 105) lipofected (in six-well plates), with equal weights (2 µg) of plasmids or mRNAs encoding ZFN4L and ZFNR, or with control DNA (pGP-FF, 4 µg). Genomic DNA was extracted from cultures 48 h after transfection. Phusion high fidelity DNA polymerase (Thermo Scientific) was used to amplify target loci from genomic DNA with β-globin primers (5'-AGGGTTGGCCAATCTACTCC-3' and ZFN-R 5'-CAAAGAACCTCTGGGTCCAA-3') or δ-globin primers (5'-AACTGCTGAAAGAGATGCGGTGG-3' and δ-ZFN-R 5'-CCTACCTGCTCTTCTCCCACATT-3'). PCR products were purified using the Zymoclean kit (Zymo Research, Irvine, CA, USA) and analyzed using a commercially available Surveyor assay (Transgenomic, Omaha, NE, USA) in accordance with the manufacturer's instructions. Alternatively, purified PCR products were digested with MwoI (New England Biolabs) in accordance with the manufacturer's instructions. Products of Cel-I and MwoI digestions were analysed by 10% polyacrylamide gel electrophoresis, as described previously . Frequencies of heteroduplex formation were estimated by comparing the intensity of the digested products with known amounts of 100-bp ladder (New England Biolabs, Beverly, MA, USA).
HT1080 cells were electroporated with pDR-GFP-ZFN4 (10 µg) and PuroR clones were selected. Expanded clones were lipofected with ZFN4L and ZFN4R expression plasmids (0.4 µg each) in 24-well plates or with 0.4 µg each of I-SceI expression plasmid pCMV3xnls-I-SceI  and a control plasmid (pGP-FF). Cells were analysed for GFP expression 48 h after transfection in a FACScalibur flow cytometer (Becton-Dickinson Biosciences, Franklin Lakes, NJ, USA) .
K562 cells (1 × 106) were nucleofected with 2, 4 or 8 µg of PvuI-linearised pTV-TC3 and 2 µg of each ZFN4L and ZFN4R mRNA or expression plasmid. Cells were expanded without selection for 48 h before setting up dilutions (96-well plates at a density of 103 or 0.5 × 103 cells/well for the experiments with 2–4 µg and 8 µg of pTV-TC3, respectively) and adding puromycin to both bulk cultures and dilutions. Genomic DNA for PCR analysis was prepared from puromycin resistant (PuroR) bulk cultures or multi-well plates after 15 days of selection. Target-specific and construct-specific PCR primers for detecting gene targeting were: 5'-AGAGCTGAAAGGAAGAAGTAGGAG-3' and 5'-GCATTCTAGTTGTGGTTTGTCC-3'. In limiting dilutions, the average number of PuroR clones per well (m) was calculated as –lnF0, where F0 is the fraction of wells with no PuroR cells. The overall frequency in the original culture = m/1000 or m/10%.
HindIII digested genomic DNA (approximately 6 µg per lane) was prepared, separated, blotted and probed as described previously . The HBB probe was a 557-bp PCR fragment generated using Phusion Hot Start II high-fidelity DNA polymerase (Fermentas), K562 genomic DNA and primers 5'-GATTTGAAACTGAGGCTCTGACC-3' and 5'-CAAGACCCTGTTTCACATCCCTG-3', and 32P-labelled by random priming.
Design of HBB-specific ZFNs
Initially, we adopted the modular assembly approach of Wright et al.  to design three ZFNs (ZFN1, ZFN2, ZFN3) with cleavage sites close to many known mutations causing βTM [39, 40], with those for ZFN1 and ZFN2 also being 60 and 107 nt downstream of the SCA mutation, respectively (Figure 1). To test for binding of each ZFN DNA binding domain to its target sequence, we used a bacterial two-hybrid assay  (Figure 2A). The results (Figure 2B) obtained, however, showed only one of the six DNA binding domains (from (ZFN1R) to bind effectively. We then used a commercial service to develop a customised ZFN specific for the human β-globin gene (see Materials and methods). The resulting protein (ZFN4) was designed to cleave 22 nt downstream of the SCA mutation (Figure 1) and close to many known βTM mutations [39, 41]. The ZFN4 subunits, ZFN4L and ZFN4R, have four and six zinc fingers, respectively, generating a ZFN4 recognition sequence of 30 nt.
Detection of ZFN4-induced HBB-specific modifications
To assess cleavage of HBB by ZFN4 in human cells, we transfected K562 cells with expression plasmids, or in vitro synthesised mRNA, encoding ZFN4L and ZFN4R. After amplifying a 394-bp region of HBB from the transfected cells, we used the surveyor nuclease (Cel-I) assay [42, 43] to detect the formation of insertions/deletions (indels) resulting from inaccurate NHEJ at the target site. Surveyor nuclease, which cleaves short regions of heteroduplex DNA, cleaved the 394-bp PCR product into fragments of 291 and 175 bp, as expected for heteroduplex formation at the ZFN4 target site (Figure 3A). Cleavage was only detected in cells transfected with ZFN4-encoding DNA or mRNA. Equivalent experiments in HT1080 cells generated similar results (not shown). Digestion of the same PCR products with the restriction enzyme MwoI, whose recognition site overlaps the ZFN4 target site in HBB, showed that only products from ZFN4-expressing cells were incompletely digested (Figure 3B), providing additional evidence for ZFN4-induced indels. The proportions of cleaved products in the Cel-I assay, or undigested DNA in the MwoI assay, were reproducibly in the order of 1–2%.
To test the specificity of ZFN4, genomic DNA samples showing clear evidence of ZFN4-induced indels in HBB were used to amplify an 811-bp fragment of the δ-globin gene (HBD). This fragment carries the human sequence (GACTGCTGTCAATGCCCTGTGGGGCAAAGTGAACGTGGAT) most similar to the ZFN4 target site, as revealed by BLAST analysis of the human genome. Cleavage by surveyor nuclease at this site would generate products of 322 and 489 bp, although no cleavage was detected under identical conditions used to detect cleavage of the β-globin gene (Figure 3A).
Homology-directed repair assay
To assess the ability of ZFN4 to promote HR, we modified the recombination reporter construct DR-GFP [33, 44] to generate DR-GFP-ZFN4, which carries adjacent target sites for ZFN4 and the homing endonuclease I-SceI (Figure 4A). These sites disrupt a GFP cassette that is linked to a fragment of the GFP gene capable of restoring an intact GFP cassette by gene conversion.
We stably transfected DR-GFP-ZFN4 into HT1080 cells to generate clones that we then transiently transfected with expression plasmids for I-SceI or ZFN4. Results for one clone are shown in Figures 4B and 4C. Background levels of GFP-positive (GFP+) cells were below 0.01% and expression of I-SceI increased this to 1.4%. Expression of ZFN4 also increased the frequency of GFP+ cells but only to 0.07%, which is 20-fold lower than the levels induced by I-SceI. These results demonstrate the ability of ZFN4 to cleave the substrate and promote HR but suggest that it is up to 20-fold less efficient than I-SceI at binding and cleaving its target site, consistent with the relatively low levels of cleavage we detected in the Cel-1 assay (Figure 3).
Gene targeting of the β-globin endogenous locus
We next investigated whether ZFN4 could promote gene targeting at the endogenous HBB. For this, we made a targeting construct (pTV-TC3) designed to introduce a puromycin resistance (PuroR) cassette at the target site (Figure 5A). We cotransfected K562 cells with varying amounts (2, 4 and 8 µg) of pTV-TC3, with or without ZFN4-encoding mRNA or expression plasmids. Bulk cultures of transfected cells selected in puromycin for 15 days tested positively in a PCR assay for targeted integration (Figure 5A) only when ZFN4 was expressed (Figure 5B). Furthermore, transfected cells plated at limiting dilution into multi-well plates gave rise to many more wells of PuroR cells when ZFN4 was expressed, most of which also tested positively in the PCR assay (Table 1, rows 9 and 11). To confirm that the PCR results were detecting true targeting events, Southern analysis of PCR-positive pools and clones of PuroR cells was carried out. In all cases, a 3.7-kb fragment diagnostic for targeted integration was detected by a target-specific probe (Figure 5C).
The proportion of drug-resistant clones that were targeted (the relative targeting frequency), as determined by PCR analysis of individual PuroR clones, rose from one in 26 (4%) in the absence of ZFN4 to 54 in 60 (90%) when ZFN4 was expressed (Table 1, row 12). Absolute targeting frequencies were also calculated (Table 1, row 14) and rose from < 0.001% in the absence of ZFN4 to an average of 0.13% when ZFN4 was expressed. As expected, absolute frequencies were dependent on the amount of targeting plasmid used (Figure 5D), the highest frequency 0.17% occurring with the highest amount of DNA (8 µg).
The aim of the present study was to design, build and characterise HBB-specific ZFNs capable of supporting high absolute frequencies of HBB gene targeting. Methods for developing customised endonucleases evolved rapidly during the study. Initially, we chose the modular assembly approach of Wright et al. , based on its convenience and affordability. Of the six subunits that we assembled with this system, only one (ZFN2R) bound its target sequence well. Although disappointing, this was consistent with the low success rates subsequently reported by the same group , and ZFN2R may be useful for developing further HBB targeting reagents. Although more advanced in-house ZFN assembly systems were developed [46, 47], we chose instead a commercial system (CompoZr) based on its validated use of dimeric ZF libraries [48, 49], and heterodimeric FokI nuclease subunits known to increase ZFN activity and specificity . The frequencies of HBB modifications that we obtained with the resulting ZFN4 nuclease are summarised in Table 2, along with related data from other studies published during completion of our work.
Table 2. Comparison of HBB-specific ZFN activities
Judged by its ability to support an average relative targeting frequency of 90%, ZFN4 is highly effective, at least in K562 cells. The equivalent figure in iPSCs was only 0.3%, however, even after selecting against nontargeted integration with ganciclovir . This difference is difficult to explain in terms of HBB transcription that is silent both in iPSCs  and in K562 cells [51-53] but may reflect an apparent tendency for the targeted drug-resistance gene in iPSCs cells to undergo silencing . Alternatively, it may reflect a favourable HR/NHEJ ratio in K562 compared to iPSCs. In other studies of nuclease-induced HBB targeting in iPSCs, however, relative targeting frequencies ranged from 8% to 60% (Table 2), suggesting either that the nucleases used were more efficient than ZFN4 or that there is considerable variation in HR/NHEJ between the different iPSCs used. Cel-I assays provide a measure of nuclease efficiency and, where measured, ranged between 1% and 13%, although these were measured in different cell types making differences difficult to interpret. The 1–2% cleavage by ZFN4 in K562 cells was nevertheless disappointing, given that values for other ZFNs in excess of 45% in K562 cells [43, 54], and 80% in osteosarcoma cells , have been reported.
In another assay for ZFN4 nuclease function using a reporter carrying adjacent recognition sites for ZFN4 and I-SceI, we found that ZFN4 promoted HR 20-fold less efficiently than I-SceI, providing further evidence that ZFN4 cleavage may be suboptimal. Similar reporter assays (but without I-SceI sites) were used to demonstrate the ability of other HBB-specific nuclease to promote HR [28, 29], although it is difficult to use these results to compare the efficiencies of the different nucleases because of variations between studies in the host cells used, as well as in the design of the reporter. Use of I-SceI-induced HR as in internal control in the manner that we have described here may help in future to allow such comparisons.
Off-target cleavage events are a major concern for the safety of any therapeutic approach based on customised nucleases . Off-target ZFN activity can cause cell toxicity , although we did not observe overt toxicity in ZFN4-expressing cells. Furthermore, neither we (present study) nor Zhou et al.  found any evidence for cleavage by ZFN4 of the HBD gene, which carries the genomic sequence most related to the ZFN4 cleavage site, using the Cel-I assay and a fluorescent HR reporter assay, respectively. Nevertheless, low level off-target activity that might be detectable by more sensitive analyses  cannot be ruled out.
Absolute frequencies of unstimulated gene targeting in human cells are typically less than 10  and, from this perspective, the 0.2% absolute frequency of HBB targeting stimulated by ZFN4 is encouraging. With ZFNs targeting other loci in K562 cells, however, frequencies as high as 29% have been reported [48, 57]. Furthermore, even if 0.2% correction can be achieved in HSC, where HR tends to be less frequent, this is probably too low a frequency to confer any therapeutic benefit after transplantation or to be overcome by any natural or engineered in vivo selective advantage for corrected cells [5, 58]. Efficiencies might be enhanced by use of viral vectors  instead of nucleofection, although nucleofection reduces the potential for immunological reactions and has been used to achieve efficient ZFN-mediated gene targeting in human CD34+ cells .
From the above arguments, we conclude that further efforts to develop efficient HBB-specific nucleases remain highly desirable for accompanying other efforts aiming to promote therapeutic gene targeting, such as enhancing HR in HSCs or conferring an in vivo selective advantage to corrected cells . Indeed, an increase in the number, as well as the efficiency, of HBB-specific nucleases is desirable for HBB gene correction in β-TM patients where the offending mutation can occur in or around any exon. Thus, a choice of nucleases is required because the frequency of gene correction if favoured by minimising the distance between the mutation and the DSB . An alternative to correcting each different disease mutation is to target a functional cDNA downstream of the endogenous promoter using the same nuclease for all patients , although this cannot be guaranteed to restore physiological expression. Fortunately, the methods available for developing gene-specific nucleases continue to expand and improve so that the prospects for developing many efficient HBB-specific nucleases are better than ever before.
We are grateful to the UK Thalassaemia Society for their support (T.V. PhD studentship); to Leuka and CHAMPS charities for financing the development of ZFN4; to Leuka for additional support for T.V.; and to Maria Jasin (Memorial Sloane Kettering Institute, New York) for supplying the plasmids pCMV3xnls-I-SceI and DR-GFP. The authors declare that there are no conflicts of interest and that no ethical approval was required for this work.