Zinc-finger Nucleases[ The Next Generation

Zinc-finger Nucleases[ The Next Generation

(Parte 1 de 2)

© The American Society of Gene Therapyreview

Zinc-finger Nucleases: The Next Generation Emerges

Toni Cathomen1 and J Keith Joung2,3Institute of Virology (CBF), Charité Medical School, Berlin, Germany; Molecular Pathology Unit and Center for Cancer Research and Center for Computational and Integrative Biology, Massachusetts General Hospital, Charlestown, Massachusetts, USA; Department of Pathology, Harvard Medical School, Boston, Massachusetts, USA

Methods of modifying the human genome precisely and efficiently hold great promise for revolutionizing the gene therapy arena. One particularly promising technology is based on the homologous recombination (HR) pathway and is known as gene targeting. Until recently, the low frequency of HR in mammalian cells, and the resulting dependence on selection to identify these rare events, has prevented gene targeting from being applied in a therapeutic context. However, recent advances in generating customized zinc-finger nucleases (ZFNs) that can create a DNA double-strand break (DSB) at preselected sites in the human genome have paved the way for HR-based strategies in gene therapy. By introducing a DSB into a target locus of interest, ZFNs stimulate gene targeting by several orders of magnitude through activation of cellular DNA repair pathways. The capability of this technology to achieve gene conversion frequencies of up to 29% in the absence of selection demonstrates its potential power. In this paper we review recent advances in, and upcoming challenges for, this emerging technology and discuss future experimental work that will be needed to bring ZFNs safely into a clinical setting.

Received 9 October 2007; accepted 29 April 2008; published online 10 June 2008. doi:10.1038/mt.2008.114

IntroductIon The ability to modify a complex genome with precision transformed biology in the late 1980s and early 1990s.1 The underlying technology is known as gene targeting and is based on the cellular homologous recombination (HR) pathway, which has evolved mainly to promote genetic recombination during meiosis and the repair of DNA double-strand breaks (DSBs) before mitosis. Until recently, the low frequency of HR in mammalian cells (~1 HR event per 106 cells), and the resulting dependence on elaborate selection strategies to identify rare recombinants, has prevented gene targeting from being applied in a therapeutic context. A recent technological breakthrough, however, has changed this dreary perspective. By fusing engineered zinc-finger (ZF) DNA-binding domains to a nonspecific nuclease domain, so-called ZF nucleases (ZFNs) were generated. These ZFNs can be designed to introduce a DSB into a desired target locus and, as a consequence, stimulate gene targeting 100- to 10,0- fold by activating cellular DNA repair pathways.2,3 Given recently published gene modification efficiencies of 29% or higher,4 it can now be envisioned that this technology could be used for correcting inborn mutations in adult stem cells derived from patients with genetic disorders. These corrected cells could subsequently be used for repopulating an affected organ, thereby reversing the disorder.

targeted genome modIfIcatIons WIth taIlored nucleases In order to exploit DSBs for therapeutic genome modifications, researchers first had to develop novel customizable nucleases.

Such engineered endonucleases basically fall into two classes: redesigned homing endonucleases, which have been recently reviewed elsewhere,5 and ZFNs.2,3 ZFNs consist of an engineered

Cys2-His2 ZF domain fused to the nuclease domain of the type IIS restriction enzyme FokI.6 In this configuration, the DNA- binding ZF domain directs the nonspecific FokI cleavage domain to a specific DNA target site. Because the FokI cleavage domain is enzymatically active only as a dimer,7 the introduction of a DSB is dependent upon dimerization of ZFNs. Accordingly, two ZFN subunits are typically designed to recognize the target sequence in a tail-to-tail conformation, with each monomer binding to “halfsites” that are separated by a “spacer” sequence8 (Figure 1a).

Repair of a DSB induced by a pair of ZFNs can occur by one of two potential pathways: nonhomologous end-joining (NHEJ) or HR. If repair is mediated by NHEJ, an error-prone pathway that can lead to insertions or deletions,9 the target gene can be disrupted by frameshift mutations which, in many instances, will lead to the expression of a truncated and/or nonfunctional protein (Figure 1b). The feasibility of such an approach was initially demonstrated in Drosophila. Endogenous expression of ZFNs targeting the y, ry, and bw loci was controlled by the hsp70 heat-shock promoter, which allowed transient expression of the ZFNs in developing larvae.10,1 Between 0 and 78% of the resulting flies carried the targeted mutation in the germline, depending on the target locus and the temperature used for inducing the hsp70 promoter. This suggests that ZFN expression levels and the chromosomal context of the target locus are crucial parameters

Correspondence: Toni Cathomen, Charité Medical School, Institute of Virology (CBF), Hindenburgdamm 27, D-12203 Berlin, Germany. E-mail: toni.cathomen@charite.de

© The American Society of Gene TherapyNext Generation ZFNs

that determine the level of NHEJ-mediated mutagenesis. Recently, ZFN-mediated gene knockout has also been achieved in mammalian cells. After transfection of a Chinese hamster ovary cell line, the transient expression of ZFNs targeting the DHFR (dihydrofolate reductase) locus resulted in ~15% of cells carrying a mutation in DHFR and ~1% of cell clones revealed a biallelic modification.12 The observed range of insertion and deletions caused by mutagenic DNA repair of the cleavage event was from +2 to –302 base pair (bp). Future therapeutic applications of such a strategy may include the targeted knockout of dominant mutant alleles or the disruption of receptors used by pathogens, e.g., the knockout of the CCR5 (chemokine receptor 5) locus to render human T cells of HIV-infected individuals resistant to infection with the virus.13

Gene targeting, on the other hand, relies on HR between the endogenous target locus and an exogenously introduced homologous DNA fragment, which we refer to as the “donor DNA.” In the absence of a DSB at the target locus, typically fewer than 1 in 105 of targeted cells will contain the desired genetic modification, a frequency too low to be useful for gene therapy.14 However, proofof-principle experiments involving the meganuclease I-SceI, which binds to an 18-bp recognition site, demonstrated that the insertion of a DSB in the target locus stimulates recombination with an exogenous donor DNA by several orders of magnitude.15,16 Importantly, stimulation of HR by I-SceI has been accomplished in various cell lines, including mouse ES cells,17 thereby indicating that DSB-induced stimulation of HR may be operative in a wide variety of different cell types. A concern is that NHEJ will compete with HR to seal the broken chromosome. Recent experiments in two different human cell lines indicated that HR-mediated gene targeting accounted for 60 to 70% of all DSB repair events at an endogenous locus,4 suggesting that, in the presence of large amounts of a donor DNA, HR is the preferred pathway to repair DSBs.

The architecture of the donor DNA determines the outcome of the gene targeting event. For example, donor DNA can be designed to either create or correct a mutation in a specific gene locus (Figure 1c). In the first study in which an endogenous mammalian locus was targeted by ZFNs, gene editing at the human IL2Rγ (interleukin-2 receptor gamma chain) locus was achieved in up to 18% of transiently transfected K562 cells, a human erythroleukemia cell line, and in ~5% of primary human T cells.18 When integrase-deficient lentiviral vectors were used for ZFN gene transfer, the gene targeting frequency at the IL2Rγ locus in K562 cells could be increased up to 29%,4 suggesting that gene transfer efficiency and ZFN expression levels are crucial factors for the success of such an approach.

Alternatively, the donor DNA can contain either an entire expression cassette or a complementary DNA fragment of the gene to be corrected. In the latter case, the main benefit of gene addition is that a single pair of ZFNs, combined with a single donor DNA, can be used for “correcting” all mutations located downstream of the DSB in a given gene, while still keeping gene expression under the control of the endogenous promoter (Figure 1d). Targeted insertions of entire expression cassettes, on the other hand, aim at restoring the cellular phenotype by integration of a therapeutic expression cassette into a yet to be defined “safe harbor.” Such a “safe harbor” should include the support of high and sustained transgene expression levels without any signs of genotoxic side effects. Therefore, at least in theory, a single pair of ZFNs combined with a customized donor DNA can be used for the safe “correction” of any given monogenetic inherited disorder that is amenable to combined gene/stem cell therapy protocols. Because a Zinc-finger nuclease bGene disruption cGene correction

Gene addition Donor DNA

Donor DNA d cDNA fragment pA pA pA pA pA pA pA


Target half-site (L)Spacer sequenceTarget half-site (R)

Fokl nuclease domain

Zinc-finger DNA-binding domain geneA geneA geneA geneA geneA geneA figure 1 Zinc-finger nuclease (Zfn)-mediated genome editing. (a) Architecture and application of ZFNs. A ZFN designed to create a DNA double-strand break (DSB) in the target locus is composed of two monomer subunits. Each subunit encompasses three zinc-fingers (orange, 1-2-3), which recognize 9 base pairs within the full target site, and the FokI endonuclease domain (green). A short linker (grey) connects the two domains. After dimerization the nuclease is activated and cuts the DNA in the spacer sequence, separating the two target halfsites (L) and (R). (b) Gene disruption. A DSB (yellow flash) introduced by the ZFN into a dominant mutant allele (geneA) is repaired by the error-prone nonhomologous end-joining pathway. Deletions and insertions that can occur disrupt the coding sequence (geneA) and render the expressed protein nonfunctional. (c) Gene correction. In order to restore a genetic defect directly in the genome (geneA), a targeting vector (donor DNA) encompassing wild-type sequences homologous to the mutant gene (grey areas) is transduced into the target cell. A ZFN- induced DSB stimulates homologous recombination (HR) between the donor DNA and the defective gene (geneA) to generate a corrected locus (geneA). (d) Gene addition. In order to restore the phenotype of a cell harboring a genetic defect (geneA), a partial cDNA flanked by sequences homologous to the mutant gene, is embedded in a targeting vector. A ZFN-induced DSB stimulates HR between the donor DNA and the mutant gene. Expression of the gene is reconstituted (geneA) and remains under the control of the endogenous promoter.

© The American Society of Gene TherapyNext Generation ZFNs individuals who harbor a homozygous deletion in the CCR5 gene are healthy,19 this locus represents a candidate site for targeted transgene insertion. In proof-of-principle studies, targeted integration of an enhanced green fluorescent protein expression cassette into the human CCR5 locus was achieved in 39% of Jurkat cells, 3.5% of embryonic stem cells, and 0.1% of hematopoietic stem cells.4 Moreover, it has been shown that ZFNs can mediate targeted insertion of a 7.8-kb-long DNA sequence into an endogenous locus in 6% of transfected K562 cells.20 Surprisingly, two homology arms of 750 bp were sufficient for efficient gene targeting. These results demonstrate that the addition of large DNA fragments is feasible, but that the HR frequency is highly cell-type specific, an observation made also in another study.21 Although these experiments in immortalized cell lines and in embryonic stem cells appear to be promising, it remains to be determined in long-term follow-up studies whether the human CCR5 locus will ultimately prove to be a “safe harbor.” engIneerIng Platforms for Zf domaIns

Cys2-His2 ZF domains are the most abundant DNA-binding motif in eukaryotes and consist of ~30 residues that fold into a ββα- structure coordinated by a zinc ion.2 A number of investigators have demonstrated that changing one or more of the six critical residues located within or adjacent to the structurally conserved α-helix (also referred to as the “recognition helix”) can alter the DNA-binding specificity of a single ZF.23–30 Many individual fingers with new specificities have been created by rational design or by selection from randomized libraries, using either phage display or a bacterial cell-based two-hybrid (B2H) system.31–39 ZFNs described in the literature to date contain three or four ZF domains arranged in tandem arrays. Because a single ZF recognizes ~3 bp of a DNA sequence,40 a ZFN subunit binds to 9- or 12-bp-long target sites, depending on the number of ZF domains present. Assuming perfect specificity by each ZFN monomer, dimers of three-finger ZFNs would therefore be expected to bind to an 18-bp target DNA site while dimers of four-finger proteins recognize a 24-bp target site. Recognition sites of 18 or 24 bp are long enough to define a statistically unique sequence in a human genome.

As mentioned earlier, various methods of engineering multi-ZF domains have been described in the literature. In order to simplify our review we have grouped these systems into three categories: (i) “modular assembly” methods (Figure 2a), (i) “context-sensitive selection” methods (Figure 2b), and (i) “2 + 2” method of the company Sangamo BioSciences (Richmond, CA) (Figure 2c). (i) “Modular assembly” has been described by several groups,25,26,31,41,42 but the efficacy of this approach as a general method for making ZFNs is controversial. As implied by its name, the process of modular assembly involves the joining of single ZF domains of known DNA-binding specificities to create the DNA- binding domain of a ZFN (Figure 2a).43–45 Different groups of investigators have described large archives of single-finger modules31,36–38,41,46 and two laboratories provide free web-based software tools47,48 that facilitate the identification of potential ZFN sites and the design of multi-finger domains. Although modular assembly is conceptually appealing in its simplicity, and some modularly assembled domains revealed excellent DNA-binding activity in vitro49 and at endogenous loci in Drosophila and Caenorhabditis elegans cells,1,50,51 other reports have suggested that the efficacy rate for producing active three-finger proteins for use in mammalian cells may be far less than 100%.41,52 Motivated by these findings, we recently performed a systematic large-scale evaluation of 168 three-finger domains designed to recognize 104 diverse target sites using a B2H assay as a screening method. This B2H assay is rapid and accurately identifies three-finger domains that will fail to function as ZFNs in human cells. For 79 of the 104 target sites that we targeted by modular assembly, we failed to obtain even a single three-finger array that showed evidence of DNA-binding activity in the B2H assay.53 The apparently low efficacy of modular assembly is particularly problematic for the purpose of making ZFNs because of the need to engineer two different ZF domains for each target site of interest. Assuming a success rate of ~24% (25/104) for making three-finger domains by modular assembly, the theoretical success rate for generating a dimeric ZFN complex with this method would be ~6%. However, even these values are likely to be overestimates of the true success rate of modular assembly because we have found that not all ZF arrays that scored positively in the B2H will be active in human cells.53

The low success rate of modular assembly is most likely because neighboring ZF domains are not truly independent in their DNA-binding activities (i.e., that ZFs are not always truly modular in their behavior). Structural studies of ZFs bound to their cognate DNA-binding sites have demonstrated that fingers can “reach over” to make contacts in the target sites of neighboring fingers, and that residues in the recognition helix of one finger can influence the orientation of recognition helix side-chains in an adjacent finger.32,40,54–61 The fact that modular assembly disregards these context-dependent effects may explain why this method frequently produces ZF domains with suboptimal DNA-binding affinities and specificities, and thereby likely fails to produce functional ZFNs for use in human cells. (i) A number of “context-sensitive selection strategies” that attempt to account for context-dependent effects have been described in the literature, including methods known as bi-partite selection, sequential optimization, and context-sensitive parallel optimization (Figure 2b).3,62–64 These approaches attempt to take into account the relative position of an individual ZF module in the final ZF array (i.e., position 1, 2, or 3) and the impact of the neighboring finger(s); or, more simply stated, they try to identify combinations of fingers that work well together. A significant disadvantage of all of these approaches is that they remain inaccessible to most research groups because they require specialized expertise in the construction of large randomized ZF libraries and the use of labor-intensive selection methods (e.g., phage display or B2H system) for interrogating them. However, published and unpublished experience with these methods suggests that they are robust and that they each yield multi-finger domains with high DNA-binding affinities and specificities, as measured in vitro with purified proteins.62–64 Moreover, when fused to the catalytic FokI domain, the resulting ZFNs showed higher activity and lower toxicity in human cells as compared to their modularly assembled cousins.65 (i) The proprietary ZF engineering platform of Sangamo

BioSciences has been shown to yield ZFNs capable of editing endogenous mammalian genes.4,12,18 Although we do not know

© The American Society of Gene TherapyNext Generation ZFNs

Assessment of the in vivo cleavage specificity of ZFNs, i.e., the ratio of on-target versus off-target cleavage events in a complex genome, remains a significant and thus far unsolved challenge. Although a recently published bacteria-based in vivo specificity profiling system for ZF DNA-binding domains can provide information about the DNA-binding profile of monomeric ZF domains,75,76 it cannot predict actual cleavage sites of ZFN dimers in the human genome. A possible approach to identify ZFN cleavage sites directly might be to exploit the fact that DSBs in a cellular genome serve as efficient integration sites for episomal DNA, such as vectors based on adeno-associated virus.7 Sequencing of the adeno-associated virus vector integration sites after ectopic ZFN expression could offer direct information about the locations of off-target DSBs in a cell.

In this context it is important to mention that, at least in theory, only one donor DNA molecule will be used as a template for HR with a target allele, while the remaining donor molecules could potentially integrate into other naturally occurring or ZFN-induced off-target DSBs. Therefore an additional critical parameter for therapeutic gene targeting approaches will be to assess the ratio of targeted vs. untargeted donor integration events. Both the immediate risk of ZFN-induced mutagenesis and the untargeted integration of the donor DNA to induce unpredictable oncogenicity can be assessed by soft agar transformation studies78 or in vitro transformation assays using purified lineage-negative cells from murine bone marrow.79 Moreover, cytogenetic analyses, like spectral karyotyping,80 can provide information about whether ZFN activity induces chromosomal abnormalities and/or translocations. We emphasize however that the long-term consequences of ZFN-induced DSBs can be studied only in vivo. Assays developed earlier to evaluate the genotoxicity of retroviral vectors in gene therapy protocols81,82 should prove useful in studying the malignant potential of cells after overexpression of ZFNs.

precisely how these four-finger domains are generated, published papers and presentations at meetings suggest a two-step procedure:18,24,6–68 in the first step, four-finger domains are assembled from large pre-existing archives of two-finger units with known DNA-binding specificities; in the second step, promising “lead” proteins may be optimized using a proprietary algorithm-based approach (Figure 2c). Because it requires access to two proprietary resources (the archive and the algorithm), this method is thus far accessible only to academic researchers who are collaborating with Sangamo.6,69

Zfn-assocIated toxIcIty ZFN-induced cytotoxicity is a major potential issue, and has been reported in several studies.10,1,52,70–72 Cell death and apoptosis associated with ZFN expression are most likely the result of excessive cleavage at off-target sites, which, in turn, suggests imperfect target-site recognition by the ZF DNA-binding domains. Given that therapeutic gene targeting will strongly depend on creation of a DSB at a specific target site, the implementation of quantitative assays to assess immediate and long-term genotoxicity of artificial nucleases is of paramount importance.73 In some studies, the extent of cytotoxicity associated with ZFN expression was quantified by measuring cell survival65,70,74 or apoptosis;52 however, these assays are very coarse measures of toxicity and provide little information about the contributing mechanisms. In order to address this problem, assays that directly document the number of offtarget cleavage events have been developed. Using antibodies specific for phosphorylated histone H2AX (γ-H2AX) or p53 tumor suppressor-binding protein 1 (53BP1), we as well as others have quantified the relative number of ZFN-induced repair foci formed after the creation of a DSB.71,72 These quantitative assays can be used to characterize the specificity and immediate genotoxicity of any artificial nuclease of interest, but they do not provide information about the sites at which off-target DSBs occur.

Zinc-finger archive Modular assemblya

2c 2b 5�-ggagtagct 2a gaa gac gag gat gca gcc gcg gct gga ggc g ggt gta gtc gtg gtt

Context-sensitive selectionCustomized libraries b

Two-finger archive

Improvement 5�-gccggagtagct

2+2 strategyc

2 gaagaa gacgaa gaggaa gatgaa 1

2b 1b 2a 1a 2b 1b 2b 1a


Selections AnchorAnchorAnchor

Shuffling figure 2 Zinc-finger engineering platforms. (a) “Modular assembly” involves the joining of single zinc-finger domains of known DNA-binding specificities. Large archives of single-finger modules have been created by selection from randomized libraries using phage display. This approach is conceptually simple but neglects positional and context-dependent effects. (b) “Context-sensitive selection” strategies attempt to identify combinations of zinc-fingers that work well together. One particular strategy for performing such a selection (among several described in the literature) is shown in the figure. The first selection step takes into account the relative position of the finger (1, 2, or 3), while the second step factors in the impact of the respective neighbor(s). (c) The “2 + 2 strategy” is a proprietary platform and details are not known. It is likely that the four-finger domains are assembled from pre-existing archives of two-finger units with known DNA-binding specificities, followed by further optimization using an algorithm-based approach.

© The American Society of Gene TherapyNext Generation ZFNs

the desIgn of safer nucleases In principle, at least three general strategies could be employed to increase the specificity of engineered ZFNs: (i) improving the DNA-binding specificities of the ZF domains, (i) optimizing the linker sequence that connects the ZF domain with the FokI cleavage domain, and (i) regulating DNA-cleavage activity of the FokI nuclease domain.

Although recent work in the field has highlighted the tremendous power and broad applicability of ZFNs for biological studies and gene therapy, the low numbers of natural loci successfully targeted by ZFNs in mammalian genomes testifies how difficult it is to develop effective ZFNs. To our knowledge, the sole published examples of endogenous mammalian loci that were successfully altered using ZFNs are the IL2Rγ and CCR5 genes in human cells4,18 and the DHFR locus in Chinese hamster ovary cells.12 The Zinc Finger Consortium (http://w.zincfingers. org), an international group of 14 academic laboratories, is dedicated to the development of a robust and effective “open-source” ZF engineering platform. Our current method of choice for engineering multi-finger domains is based on a previously described context-sensitive optimization approach.64 Although laborious, it is a robust method known to yield three-finger ZFNs that function with higher efficiencies and lower toxicities in human cells when compared with analogous ZFNs made by modular assembly.65 This finding is consistent with the hypothesis that suboptimal DNA- binding activity—frequently observed with modularly assembled ZF domains—can be a primary component of ZFN-associated toxicity (Figure 3a). The development of a robust and publicly available ZF engineering platform remains a high priority for the Zinc Finger Consortium and such a platform will become available in the near future.

Although interdomain ZFN linkers of various lengths have been described and tested,8 to our knowledge no linker has yet been described that provides perfect specificity for a “spacer” DNA of a single length. The consequence of this ambiguity is that a pair of ZFNs can cut not only at sites with a spacer length of e.g., 6 bp, but also at sites with 5-bp and 7-bp spacers (Händel and Cathomen, unpublished results), thereby increasing the number of potential off-target sites and cleavage events (Figure 3b). Systematic analysis of candidate-based linkers or, alternatively, a combinatorial library approach in which both linker composition and length are randomized, followed by iterative selection and counter-selections, may help to identify linker variants that enforce ZFN-mediated DNA cleavage at target sites of a single spacer length.

In contrast with many natural endonucleases, ZFNs do not contain an allosteric mechanism that regulates DNA-cleavage. In vitro experiments with the natural FokI restriction enzyme or ZFNs showed that a FokI monomer or a ZFN subunit bound to its target site can associate through protein–protein interaction with a second monomer/subunit that remains detached from the recognition sequence, in order to catalyze DNA cleavage.7,83 It is therefore conceivable that at high intranuclear concentrations a ZFN subunit binds to its canonical 9-bp target site as a monomer, but then becomes incorrectly activated once it forms a dimer with a second ZFN subunit, which is not properly bound to DNA (Figure 3c). Given that a 9-bp half-site statistically occurs >10,0 times in the human genome, such a scenario would generate many additional potential cleavage sites. Applying an in silico protein engineering technology, we have recently shown that the attenuation of dimerization by reducing the number of hydrophobic interactions in the dimer interface decreased ZFN-associated toxicity significantly.72 We speculate that weakening the dimer interface prevents ZFN dimers from forming in solution, thereby making endonuclease activity more dependent on DNA binding. This helps to ensure that two ZFN subunits can dimerize only after both subunits are properly bound to the DNA target site.

Cleavage of a target locus requires that two different ZFN subunits bind as a heterodimer at the desired cleavage site. However, symmetry at the FokI dimerization interface also permits homodimers to form, thus enabling cleavage at additional sites (Figure 3d). By altering interacting residues in the protein–protein interface of the FokI dimerization domain, we as well as others have recently engineered asymmetric ZFN variants that prevent the undesirable homodimerization of ZFN subunits.71,72 Although the actual mechanism by which the dimerization variants overcome toxicity is open to speculation because of lack of biochemical in vitro data, these studies show that ZFN dimerization variants harboring an asymmetric dimer interface revealed significantly reduced toxicity without compromising on performance.71,72 An ideal ZFN architecture figure 3 Various sources for zinc-finger nuclease (Zfn) off-target activity: (a) insufficient specificity of DNA-binding, permitting ZFN binding to unintended DNA sites, (b) ambiguity of the interdomain linker, allowing cleavage at noncanonical spacer lengths (e.g., at a 7-base-pair (bp) spacer instead of the intended 6-bp spacer), (c) cleavage at isolated target half-sites, and (d) cleavage by homodimeric ZFNs. (e) An ideal ZFN architecture consists of an affinity-matured DNA-binding domain, an optimized linker sequence, and a destabilized and asymmetric dimer interface that regulates the FokI cleavage activity.

b c d e

(Parte 1 de 2)