Monoclonal antibody therapy is a large and growing treatment modality in medicine (Glennie and Johnson, 2000). There are more than 20 FDA-approved monoclonal antibody therapies, with many more currently in clinical trials. Antibody therapy directed against soluble factors (e.g., vascular endothelial growth factor, tumor necrosis factor), aims to reduce the free ligand concentration by immunocomplex formation. In contrast, when antibody therapy is directed at cell surface antigens (as in anti-neoplastic immunotherapy), the goal is the removal of the reactive cell itself. The therapeutic antibody may induce apoptosis directly (Shan et al., 1998, 2000) but more often it must recruit the patient's immune system to destroy the target cell (Anderson et al., 1997; Clynes et al., 1998, 2000; Golay et al., 2000; Harjunpaa et al., 2000; Idusogie et al., 2000; Reff et al., 1994; Sampson et al., 2000). There are two main mechanisms by which the antibody-activated immune system can destroy offending cells: complement-dependent cytotoxicity and antibody-dependent cellular cytotoxicity (ADCC). ADCC is an immune response generated primarily by natural killer cells against antibody-coated targets (Lewis et al., 1993). In ADCC, natural killer cells recognize the constant (Fc) region of antibodies via interaction with the FcγRIII receptor. The natural killer cells then deposit perforins and granzymes on the target cell surface inducing cell lysis and apoptosis, respectively. The Fc–FcγRIII interaction is extremely sensitive to the Fc glycosylation state. Completely nonglycosylated immunoglobulins fail to bind Fc receptors (Leader et al., 1991; Leatherbarrow et al., 1985; Walker et al., 1989). Conversely, excessive Fc glycosylation also prevents Fc receptor binding. Compared to the fucose-containing version, a nonfucosylated carbohydrate chain attached to Fc at Asn297 greatly improves binding to FcγRIII (Shields et al., 2002) and increases in vitro ADCC activity 100-fold (Niwa et al., 2004; Shields et al., 2002; Shinkawa et al., 2003). The majority of mammalian immunoglobulins are fucosylated, including those produced in Chinese hamster ovary cells (CHO cells, Cricetulus griseus) (Hamako et al., 1993; Jefferis et al., 1990; Raju et al., 2000). Fucose attachment to the Fc core region is via an α1,6 linkage catalyzed by the Fut8 protein, apparently the sole α1,6-fucosyltransferase in mammalian cells (Costache et al., 1997; Oriol et al., 1999). Disruption of the FUT8 gene in CHO cells eliminated core fucosylation of antibodies and increased ADCC by around 100-fold (Yamane-Ohnuki et al., 2004). This achievement potentially allows for more efficacious and cost-effective immuno-therapy.
Conventional gene disruption by homologous recombination is typically a laborious process. This was particularly true in the case of C. griseus FUT8, as approximately 120,000 clonal cell lines were screened in two rounds of gene targeting (one for each allele) to discover three healthy FUT8−/−clones (Yamane-Ohnuki et al., 2004). Site-specific cleavage of genomic loci offers an efficient supplement and/or alternative to conventional homologous recombination. Fusion of the nuclease domain from the Type II restriction enzyme Fok I to engineered zinc-finger proteins allows for the creation of site-specific nucleases (Kim et al., 1996). The inaccurate repair of a site-specific double-strand break (DSB) by nonhomologous end-joining (NHEJ) can result in gene disruption (Doyon et al., 2008; Perez et al., 2008; Santiago et al., 2008).
We report here the creation of zinc-finger nucleases (ZFNs) that disrupt the FUT8 gene, and CHO cells with eliminated and/or reduced Fut8 activity. Use of a Lens culinaris agglutinin-based phenotypic screen allowed isolation of clones with exclusively doubly modified chromosomes in less than 3 weeks at a frequency of 5%. When the ZFNs were used to generate the FUT8 disruption in an industrial protein-production host cell line, we obtained stable, high-titer clones producing completely nonfucosylated antibody with no detrimental effects on cell growth, viability, or product quality.
Zinc-Finger Nucleases That Disrupt FUT8 and FUT8−/− CHO-K1 Cells
As the Chinese hamster genome is unsequenced, we cloned the C. griseus FUT8 cDNA to identify potential ZFN targets. The three motifs that constitute the catalytic core of α2- and α1,6-fucosyltransferases were readily identifiable in the hamster sequence (Fig. 1A) (Javaud et al., 2000; Oriol et al., 1999). In particular, the hamster FUT8 Fut motif II is identical to the cow and human motifs, and differs by only one amino acid from those of the pig and mouse (Javaud et al., 2000). When the FUT8 cDNA was scanned for favorable ZFN recognition sites, one such location overlapped with this second Fut motif (Fig. 1A and B). This ZFN pair was assembled from one- and two-finger modules.
The ZFN was assayed by transient transfection into CHO-K1 cells. The inaccurate DNA repair of ZFN-generated DSBs characteristic of nonhomologous end-joining results in mutations at the site of ZFN activity (Santiago et al., 2008). When the targeted portion of FUT8 was PCR-amplified from a population of ZFN-treated cells then melted and reannealed, wild-type and mutant stands reassorted to form imperfectly base-paired hybrid molecules. These molecules are a substrate for the mismatch-sensitive Surveyor enzyme (CEL-I). CEL-I digestion products from cells transfected with the pVAX vector and from cells transfected with the ZFNs are shown in Figure 1C. Treatment with the FUT8 ZFNs resulted in modification of 4.4% of chromosomes.
Lens culinaris agglutinin (LCA) is a lectin highly specific for oligosaccharides with core fucosylation (Tateno et al., 2009). Binding and endocytosis of LCA-bound membrane proteins results in cell death. The CHO cell line Lec13 contains a mutation in the fucose biosynthetic gene GMD allowing it to grow in concentrations of LCA 50-fold higher than wild-type CHO cells (Ohyama et al., 1998; Ripka and Stanley, 1986). FUT8−/− cells fail to bind fluorescently labeled LCA (Yamane-Ohnuki et al., 2004). We reasoned that FUT8−/− cells would also be capable of growth in LCA. To test this hypothesis, we exposed a population of ZFN-treated cells to LCA. Within 16 h after application of LCA the majority of cells in the culture had become round and detached from the flask (data not shown). After 2 days of growth in LCA, colonies of morphologically normal CHO cells became visible (Fig. 2A). DNA harvested from these cells was dramatically enriched for mutations at the location cleaved by the ZFNs (Fig. 2B), confirming that LCA selects for cells with mutation of both FUT8 alleles. No LCA-resistant cells were seen in the absence of ZFN treatment (data not shown).
ZFN-treated cells were cloned by limiting dilution. In order to identify double-mutant clones rapidly, we used a phenotypic screen based on the morphological difference between wild-type and FUT8−/− cells exposed to LCA. Three weeks after cell cloning, half the cells in each well were moved to medium supplemented with 50 µg/mL LCA. After 18 h, the morphology of the LCA-exposed clones was inspected. Of the 711 clones analyzed in this manner, 25 were resistant to LCA (3.5%). Cell lines were expanded from the parallel cultures not exposed to LCA. The FUT8 genotypes of 21 of these clones are shown in Table I. All 21 clones had modification of both alleles of FUT8. Consistent with the output of NHEJ, the majority of alleles were small insertions and deletions. Performing an LCA selection prior to dilution cloning resulted in isolation of almost exclusively FUT8−/− clones (data not shown).
Table I. Genotypes of LCA-resistant clones treated with the SBS 12172/SBS 12176 ZFN pair.
The region of sequence shown here is identical to that in Figure 1. A three base pair gap (- - -) has been inserted into the wild-type sequence to facilitate sequence alignment. In the wild-type sequence, the ZFN binding sites are shown in bold and the intron 10 splice donor site in italics. Alleles are designated as A and B. Only one allele was recovered from clones 3 and 13.
As ZFN- and NHEJ-induced mutations are random, not all mutations will be null alleles. The propinquity of the FUT8 ZFN cleavage site to the coding sequence for the active site of the enzyme suggested that it might be possible to generate hypomorphic alleles of FUT8. Such hypomorphs might be resistant to the relatively low concentration of LCA used to perform the initial screen (50 µg/mL) but remain sensitive to higher concentrations of LCA. FUT8−/− CHO cells grow normally in greater than 1,000 µg/mL LCA. To search for potential FUT8 hypomorphs, we performed a secondary screen of the initial ZFN-modified LCA-resistant cell lines with varying concentrations of LCA. Four clones (7, 14, 19, and 24 in Table I) showed LCA resistance only up to 400 µg/mL, all of which shared an allele with a three nucleotide (ATT) insertion that adds one leucine residue to the C. griseus Fut8 protein at position 415. The other allele in each of these four clones is likely to eliminate enzyme activity. The results of the LCA titration were confirmed by assay of fluorescent LCA (F-LCA) binding to cell surface-exposed α1,6-linked fucose. All of the hypomorphic clones examined by F-LCA binding had approximately five-fold less F-LCA binding than wild-type; all clones resistant to 800 µg/mL LCA showed no ability to bind F-LCA (Fig. 2C).
No functional commercial antibody exists for Fut8. We therefore confirmed that the LCA-resistance phenotype was due solely to Fut8 removal by complementation of this phenotype with the FUT8 cDNA. Transfection of sense (but not anti-sense) FUT8 cDNA into a FUT8−/− cell line converted these LCA-resistant cells back into LCA-sensitive ones (Fig. 2D, middle panels). GFP was co-transfected in this experiment to identify transfected cells. All fluorescent cells returned to LCA-sensitivity (Fig. 2D, bottom right panel). We conclude that the LCA-resistance and LCA-nonbinding phenotypes were caused by ZFN-mediated disruption of FUT8.
FUT8−/− Cell Lines Made in a DHFR−/− CHO Cell Line Suitable for Industrial Bioproduction
To test the applicability of our FUT8 ZFNs in a protein-production context, we recapitulated the FUT8 knockout in a DHFR−/− suspension CHO cell line used to generate biopharmaceutical production cell lines. Cells were transfected with the FUT8 ZFN pair and subjected to LCA selection 48 h post-transfection. After 7 days of LCA exposure, the cells were dilution cloned and genotyped. Numerous FUT8−/− clones were isolated and tested for their ability to express an antibody during transient transfection. ZFN treatment and FUT8 deletion did not alter the cells' transfectability or transient antibody production compared to the parental cell line (Fig. 3A). The glycosylation profile of the secreted antibody was assayed by mass spectrometry. Antibody produced in these FUT8−/− cells completely lacked fucosylation but was otherwise normally glycosylated (Fig. 3B, compare the G0 + F and the G0 − F groups). The best seven FUT8−/− clones were selected based on their growth profiles in seed train, transfectability, and antibody expression levels in transient transfection. The growth properties of these cell lines were examined in a scale-down production system. Compared to the parental cell line, the majority of FUT8−/− cell lines had equal or better viable cell counts and culture viability throughout the experiment (Fig. 4A and B). The integrated viable cell count (IVCC) was calculated at the end of the experiment and similarly, the majority of FUT8−/− cell lines had an IVCC equal to or better than the precursor line (Fig. 4C). In contrast to the parental cell line, the majority of FUT8−/− lines consumed all lactate in the culture on day 14 (Fig. 4D).
Cell lines stably expressing a model antibody were developed from the parental CHO cells, the ZFN-derived FUT8−/− cells, and homologous recombination-derived FUT8−/− CHO DG44 cells (Yamane-Ohnuki et al., 2004). Several stably expressing clones from each line were assayed in a scale-down shake flask model of bioreactor production; data on the best four such clones is shown in Figure 5. Clones of both the ZFN-derived FUT8 knockout line and the parental CHO cell line achieved a ∼2 g/L antibody titer after 2 weeks of culture (2.0 ± 0.1 and 1.5 ± 0.6 g/L, respectively; Fig. 5A). In contrast, clones of the HR-derived FUT8 knockout cell line reached only 0.5 ± 0.1 g/L (Fig. 5A). The ZFN-mediated FUT8 knockout clones had integrated viable cell counts equivalent to or better than clones from the parental line and the HR-derived FUT8 knockout line (114 ± 25, 79 ± 34, and 59 ± 33 billion cell-days/L, respectively; Fig. 5B). Gel filtration chromatography of the antibody produced in this experiment revealed that all but one parental-derived clone had acceptably low levels of aggregation (Fig. 5C). As expected, antibody produced in both types of FUT8−/− cells was completely nonfucosylated (Fig. 5D). We conclude that ZFN-mediated FUT8 deletion in CHO cells results in cell lines compatible with industrial bioprocess and produces antibodies without detectable fucosylation.
We present zinc-finger nuclease reagents for the knockout of the C. griseus FUT8 gene and ZFN-treated cells with inactivated Fut8. Many CHO cell lines exist, often with custom-made genetic or phenotypic changes. A major advantage of FUT8 gene disruption using ZFNs is the ability to reproduce this disruption rapidly in a wide variety of these CHO cell subtypes. We recreated the FUT8 disruption in a CHO cell line used for stable protein production and demonstrated its suitability for the manufacture of nonfucosylated antibodies.
The three-dimensional structure of human Fut8 revealed that three α2/α6 fucosyltransferase motifs form the catalytic core of the enzyme (Ihara et al., 2007). In this region, point mutation of many single residues to alanine results in complete inactivation of the enzyme (Ihara et al., 2007; Takahashi et al., 2000). NHEJ-derived mutations are small relative to those made by conventional gene disruption, but our ability to target these mutations to the DNA coding for the critical catalytic region of FUT8 ensured that even short, in-frame deletions would generally result in elimination of Fut8 activity. Targeting of the catalytic site also allowed NHEJ to generate hypomorphic mutations in FUT8. While these mutations occurred randomly, homology-directed repair could be used with this ZFN pair to create similar, defined FUT8 alleles (Urnov et al., 2005).
Cell lines with reduced FUT8 expression have been made using RNAi (Imai-Nishiya et al., 2007). Simultaneous knock-down of GMD, a second gene in the fucose biosynthetic pathway was required to reduce antibody fucosylation to background levels. While this strategy effectively prevented fucosylation of an expressed antibody, repression of gene expression by RNAi can be difficult to maintain in large-scale culture (Lim et al., 2006). In contrast, genetic knockouts of FUT8 should not revert and require no selection pressure to maintain.
ZFN-mediated gene deletion also has advantages relative to traditional gene knockout technology. Conventional homologous recombination also produces true knockouts but is inefficient, laborious, and requires the use of positive/negative selection strategies to isolate the targeted event. In contrast, we show here that ZFN action produces rapid, highly efficient, biallelic knockouts without the need for a homologous recombination donor construct or drug selection. Additionally, unlike conventional homologous recombination-based knockout constructs, ZFNs require only a small amount of conserved exonic sequence to work across species.
We performed both an LCA-based screen and an LCA-based selection to isolate FUT8−/− CHO cell clones. While screening clones with LCA avoids exposure of the cells to lectin, it is a labor-intensive procedure compared to LCA selection. There are more than 30 years' experience growing CHO cells in LCA without report of deleterious effects from lectin exposure (Stanley et al., 1975). We suspect transient exposure to this plant-derived material is innocuous and did not notice a difference in growth rate between lectin-exposed and lectin-naïve cells.
In addition to the absence of fucosylation, Fut8 deletion resulted in a slight increase in Man5 and Man6 glycoforms on antibodies produced via transient transfection (Fig. 3B). Stably expressing FUT8−/− cell lines gave antibodies with Man6 levels similar to transiently transfected cells (Fig. 5D). While not resolvable by capillary electrophoresis, Man5 levels were likely also elevated in these stably transfected cells as stably- and transiently transfected cells typically have near-identical glycoform profiles (Galbraith et al., 2006; Ye et al., 2009). We conclude that removal of Fut8 results in a very modest perturbation of the gylcosylation machinery.
The highest producing clones derived from ZFN-mediated FUT8 deletion gave fourfold higher antibody titers than the most productive clone made in an alternate, DG44 CHO FUT8−/− cell line. This difference in titer likely reflects underlying properties of the parental cell lines rather than any difference in the FUT8 alleles, highlighting the advantage of genotypic portability driven by ZFN-mediated gene disruption. Furthermore, ZFN treatment and FUT8 deletion did meaningfully not impair production cell line development and did not change antibody glycoform profiles (except for the expected absence of fucosylation). Given the extensive optimization typical of protein production cell lines, retention of embedded beneficial traits throughout cell engineering is quite advantageous. Similarly, ZFN-mediated FUT8 knockout should work equally well in existing high-titer production cell lines. We conclude that ZFN treatment can be broadly compatible with bioprocess cell line development and look forward to the large-scale manufacture of nonfucosylated monoclonal antibodies.
Growth and Transfection of CHO-K1 Cells
CHO-K1 cells were obtained from the American Type Culture Collection and grown as recommended in F-12 medium (Invitrogen, Carlsbad, CA) supplemented with 10% qualified fetal calf serum (FCS, Hyclone, Logan, UT). Cells were removed from plasticware using TrypLE Select protease (Invitrogen). For transfection, one million CHO-K1 cells were mixed with 1 µg each ZFN and 100 µL Amaxa Solution T. Cells were transfected in an Amaxa Nucleofector II using program U-23 and recovered into 1.4 mL warm F-12 medium + 10% FCS. The DHFR−/− CHO cell line was transfected with Lipofectamine 2000 according to the manufacturer's instructions (Invitrogen) and grown as described (Chaderjian et al., 2005).
Cloning of C. griseus FUT8
Ten nanograms of a cDNA library derived from CHO-S cells was PCR amplified using GJC 119F (5′-aacagaaacttattttcctgtgt-3′) and GJC 106R (5′-ggtcttctgcttggctgaga-3′), Topo-Cloned (Invitrogen) and sequenced. The cDNA sequence was used for the design of ZFNs. FUT8 intron 9 was PCR amplified from C. griseus genomic DNA using EasyA polymerase (Stratagene, Santa Clara, CA) and the oligonucleotides GJC 71F (5′-gcttggcttcaaacatccag-3′) and GJC 72R (5′-cactttgtcagtgcgtctga-3′). The PCR product was cloned and sequenced. The partial sequence of intron 10 was obtained by PCR amplification of C. griseus genomic DNA using EasyA polymerase (Stratagene) and the oligonucleotides GJC 75F (5′-agtccatgtcagacgcactg-3′) and GJC 77R (5′-cagaactgtgagacataaactg-3′).
Zinc-Finger Nuclease Design and Production
ZFNs (Catalog # CKOCHOFUT8) were obtained from Sigma-Aldrich (St. Louis, MO). All ZFNs used here are of the high-fidelity type (Miller et al., 2007).
Analysis of ZFN Activity
Cells were harvested 2 days post-transfection and chromosomal DNA prepared using a Masterpure DNA Purification Kit (Epicentre, Madison, WI). The appropriate region of the FUT8 locus was PCR amplified using Accuprime High-fidelity DNA polymerase (Invitrogen) and the oligos GJC90 (5′-ctgttgattccaggttccca-3′) and GJC91 (5′-tgttacttaagccccaggc-3′). PCR reactions were heated to 94°, then gradually cooled to room temperature. Approximately 200 ng of the annealed DNA was mixed with 0.33 µL CEL-I enzyme (Transgenomic, Omaha, NE) and incubated for 20 min at 42°. Reaction products were analyzed by polyacrylamide gel electrophoresis in 1X Tris–borate–EDTA buffer.
Generation of ZFN-Modified Clones
Between 6 and 30 days post-transfection, CHO-K1 cells transfected with SBS 12172 and 12176 were dilution cloned in 96-well plates at approximately 0.4 cells/well. After 2 weeks of growth the number of clones per well was scored, the cells washed in 1X PBS, and 20 µL TrypLE Select added. Ten microliters of the dissociated cells were transferred to replicate 96-well plates containing F-12 medium + 10% FCS + 50 µg/mL LCA (Vector Laboratories, Burlingame, CA). One hundred microliters of F-12 medium + 10% FCS was added to the remaining 10 µL of cells in the original 96-well plate. The morphology of cells in the LCA-containing plates was scored 18 h later. Clones retaining a wild-type CHO-K1 morphology in the presence of LCA were noted and the corresponding colony from the non-LCA-treated plate was expanded. If the original well was found to contain more than one clone (and also therefore produce a mixture of rounded and wild-type-appearing cells when grown in LCA), the contents of the well were re-dilution cloned as above. Disruption of FUT8 in the DHFR−/− CHO cell line was done as described above, but with cell cloning done from an LCA-preselected population rather than an LCA-based screen.
Screen for CHO-K1 Clones With Reduced Fut8 Activity
Cell lines resistant to 50 µg/mL LCA were assayed for growth in 100, 200, 400, and 800 µg/mL LCA. Eleven of the 16 cell lines tested in this manner exhibited wild-type growth and cell morphology at 800 µg/mL LCA. Five of the 16 cell lines tested exhibited wild-type growth and cell morphology only at LCA concentrations below 800 µg/mL. Twelve of the 16 cell lines resistant to 50 µg/mL LCA were analyzed by fluorescent LCA binding. Approximately 100,000 cells were trypsinized, washed in 1X PBS, and mixed with 2 µg/mL fluorescein–LCA binding was assayed by flow cytometery (Guava Technologies, Billerica, MA).
Analysis of ZFN-Modified Clones
Genomic DNA was harvested from candidate clones and a portion of the FUT8 locus PCR amplified using the oligos GJC 75F and GJC 91R. Half of the PCR product (∼200 ng) was analyzed using the CEL-I assay, the other half was gel purified. Purified bands that were CEL-I-negative (homozygotes) were sequenced directly. CEL-I-positive bands were Topo-Cloned and clones sequenced until two alleles were recovered.
Culture Viability and Viable Cell Count Measurements
DHFR−/− CHO and DHFR−/− CHO FUT8−/− cells were assayed in shake-flask scale-down systems designed to closely mimic large-scale bioreactor performance. Cells were seeded at high-density in production medium and grown in 50 mL volume in shake-flasks. Culture viability and viable cell counts were measured by Trypan blue dye exclusion on days 3, 7, 10, and 14.
Glycan Distribution by Capillary Electrophoresis and MALDI
HCCF samples and reference material are purified using a PhyTip protein A column. The purified protein is directly transferred for enzyme digestion with PNGase F overnight to remove asparagine-linked carbohydrates that are then labeled with APTS, a fluorescent chromophore. The relative amounts of glycans are determined by capillary electrophoresis with fluorescence detection. Glycans are separated in a buffer of 40 mM ε-amino-n-caproic acid/0.2% hydroxy–propylmethyl–cellulose, pH 4.5. The assay is performed on a Beckman instrument, utilizing capillary cartridge temperature control (20 ± 2°C), a coated, fused silica capillary (50 µm i.d.), and fluorescence detection (Arg-ion laser, 488 nm excitation, 520 nm (DF 20 nm) emission). Matrix-assisted laser desorption–ionization (MALDI) was performed as previously described (Papac et al., 1998).
We thank Ed Rebar, Lei Zhang, Sarah Hinkley, George Katibah, Gladys Dulay, Anna Vincent, and Rainier Amora for ZFN design and construction. We thank Brad Snedecor for his leadership and support throughout the project, Laura Simmons for her supervision of cell culture work, Genentech's internal reviewers for their feedback on the manuscript, and the Genentech analytical operations group for their support with glycan analysis and titer assays. Special thanks go to Rod Keck at Genentech for his expertise with the determination and analysis of mammalian glycosylation patterns.