Selecting for developability of polypeptide drugs in eukaryotic cell display systems

Abstract

Use of the surface presentation level of binders (e.g., antibodies, receptors) on cultured higher eukaryotic cells in vitro as a predictive indicator of developability characteristics, e.g., solubility, of the binders. Display libraries of higher eukaryotic cells, e.g., mammalian cells, adapted for use in screening surface-displayed binders for developability and affinity of target binding. High-throughput screening of display libraries with in-built selection for developability including binder solubility, capability to be formulated at high concentrations, low propensity for non-specific binding, and half-life. Enrichment of populations of binders for developability characteristics and/or other qualities such as target binding and affinity, by controlling cell surface presentation of binders from an inducible promoter operably linked to binder-encoding DNA.

Claims

1. A method of distinguishing or ranking binders according to their solubility and/or resistance to self-association in solution, and/or enriching for binders exhibiting greater solubility and/or greater resistance to self-association in solution, comprising (i) providing a library of mammalian cell clones each containing DNA encoding a binder, (ii) culturing the clones in vitro under conditions for expression of the binders, wherein the binders are presented on the cell surface, (iii) determining surface presentation levels, measured in terms of number of displayed binders per cell, of the binders on the clones, optionally by labelling the binders with an agent incorporating a detectable label, (iv) selecting one or more clones that exhibit higher surface presentation of binders compared with other clones, and (v) identifying binders encoded by the one or more selected clones as having good solubility and/or resistance to self-association in solution, and optionally providing the selected clones for use in one or more further screening steps, wherein the binders are antibodies, and wherein the antibodies comprise transmembrane domains.

2. The method according to claim 1, comprising determining surface presentation levels of the binders on the clones by labelling the binders with an agent incorporating a detectable label, wherein the agent binds to a constant region of the binders, optionally wherein the binders comprise an Fc region and the agent binds to the Fc region.

3. The method according to claim 1, comprising sorting cells into a collected fraction and a discarded fraction according to the level of surface presentation of binders on the cells, whereby cells displaying surface presentation above a pre-determined threshold are sorted into a collected fraction and cells displaying surface presentation below the pre-determined threshold are sorted into a discarded fraction.

4. The method according to claim 3, wherein the discarded fraction comprises cells expressing comparator polypeptides that have a critical concentration of at least 10 mg/ml and wherein the collected fraction comprises cells expressing binders that have a critical concentration at least 1.5-fold higher than the comparator polypeptides in the discarded fraction.

5. The method according to claim 3, wherein sorting is performed by a fluorescence activated cell sorter (FACS).

6. The method according to claim 4, wherein sorting is performed by a FACS.

7. The method according to claim 1, wherein step (ii) comprises culturing the clones of the library as a mixture in one vessel.

8. The method according to claim 1, wherein step (ii) comprises culturing each clone of the library in a separate vessel.

9. The method according to claim 1, wherein the binders are sequence variants of a parent binder.

10. The method according to claim 9, wherein the parent binder has been identified as requiring improvement in solubility or resistance to self-association in solution.

11. The method according to claim 9, wherein the method comprises generating sequence variants of the parent binder and integrating DNA encoding the sequence variants into cellular DNA of mammalian cells to provide the library of cell clones containing DNA encoding the binders, wherein the method comprises analyzing the polypeptide sequence of the parent, identifying one or more amino acid residues that are predicted to promote self-association and/or reduce solubility, and generating mutation at the one or more amino acid residues.

12. The method according to claim 10, wherein the method comprises generating sequence variants of the parent binder and integrating DNA encoding the sequence variants into cellular DNA of mammalian cells to provide the library of cell clones containing DNA encoding the binders, wherein the method comprises analyzing the polypeptide sequence of the parent, identifying one or more amino acid residues that are predicted to promote self-association and/or reduce solubility, and generating mutation at the one or more amino acid residues.

13. The method according to claim 9, wherein the parent binder has a critical concentration of less than 50 mg/ml in phosphate buffered saline solution (PBS) and/or has a solubility limit of less than 50 mg/ml in PBS, and/or wherein the method comprises identifying binders encoded by the one or more selected clones as having a critical concentration and/or a solubility limit at least 1.5-fold higher than that of the parent binder.

14. The method according to claim 10, wherein the parent binder has a critical concentration of less than 50 mg/ml in PBS and/or has a solubility limit of less than 50 mg/ml in PBS, and/or wherein the method comprises identifying binders encoded by the one or more selected clones as having a critical concentration and/or a solubility limit at least 1.5-fold higher than that of the parent binder.

15. The method according to claim 1, comprising simultaneously determining surface presentation levels of the binders and levels of target binding by the binders, and co-selecting clones displaying cognate binders exhibiting higher surface presentation compared with other clones; or simultaneously determining surface presentation levels of the binders and levels of non-specific binding to non-target molecules, and co-selecting clones displaying binders exhibiting higher surface presentation and lower non-specific binding compared with other clones; or simultaneously determining levels of target binding and levels of non-specific binding to non-target molecules by the binders, and co-selecting clones displaying cognate binders exhibiting lower non-specific binding compared with other clones.

16. The method according to claim 1, comprising determining the sequence of the DNA encoding the binder from the one or more selected clones, and providing isolated nucleic acid encoding the binder.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Embodiments of the invention will now be described in more detail, with reference to the drawings, which are as follows:

(2) FIG. 1. pINT17-BSD, a dual promoter antibody IgG expression cassette for surface expression

(3) pINT17-BSD, a schematic map is shown depicting the key features (1-7500 bp)

(4) pINT17-BSD-D1.3, a dual promoter antibody expression cassette for surface expression of the anti-lysozyme antibody D1.3. The full annotated nucleic acid sequence is shown.

(5) Features:

(6) AAVS left homology arm 9-812

(7) Blasticidin resistance gene 853-1254

(8) pEF promoter 1522-2705

(9) BM40 leader 2745-2799

(10) Humanised D1.3 VL 2799-3130

(11) Human C kappa 3138-3443

(12) BGH poly A 3468-3682

(13) CMV promoter 3701-4273

(14) Mouse VH leader with intron 4299-4426

(15) Humanised D1.3 VH 4432-4779

(16) Optimised human IgG1 CH1-CH3 4780-5775

(17) Myc tag 5776-5805

(18) PDGFR anchor 5806-5961

(19) BGH polyA 6011-6225

(20) AAVS right homology arm 6288-7124

(21) f1 replication origin 7282-7695

(22) pUC replication origin 7916-8590

(23) Kanamycin resistance gene 9310-10104

(24) FIG. 2. Expression of IgG on cell surface.

(25) Analysis was focused on viable cells using forward scatter and staining in the FL3 channel. Cells positive for staining in the FL3 channel (representing non-viable cells which took up 7-AAD) were excluded. HEK293 cells were transfected with pINT17-antibodies in presence of the AAVS TALENs. Stable populations were selected with Blasticidin. 14 days post-transfection, cells were stained with anti-Fc PE (FL2). Panels depicting fluorescence intensity (anti-Fc-PE, x-axis) plot against cell count (y-axis) and include the CNTO607 (a), MEDI1912 (b) and Ang2mAb (c) pairs. For all panels the plots include stained HEK293 cells (dotted line), parental antibody (dashed line), improved mutant (solid black line).

(26) FIG. 3. Preparative size exclusion chromatography of MEDI-1912 and MEDI-1912-STT. Antibodies were expressed by transient transfection of Expi293 cells followed by protein A affinity purification and dialysis. Purified MEDI-1912 (0.5 ml, 1.1 mg/ml) or MEDI-1912-STT (0.5 ml, 1.7 mg/ml) were loaded onto a Superdex 200 10/300 column connected to an AKTA Pure system using a PBS (pH 7.4) running buffer. The elution volume (ml) us plotted on the x axis against the absorbance at 280 nm (mAU) on the y-axis. The elution volume (Ve) for MEDI-1912 and MEDI-1912-STT was 10.3 ml and 11.7 ml respectively. MEDI-1912 shows an earlier elution volume indicating self-interaction.

(27) FIG. 4. DNA and protein sequence of MED-1912 variable heavy (VH) chain. Primers employed for library creation are labelled above the nucleic acid sequence.

(28) FIG. 5. FACS separation of mixed cell displayed antibody populations based on antibody expression.

(29) An equal mix of MEDI-1912 and MEDI-1912_STT IgG genes were targeted via nuclease-directed integration into the AAVS locus of HEK293 cells. This mixed cell population, 15 days post-transfection, was separated on the basis of antibody expression by FACS using a BD Influx sorter. Cells were stained with anti-Fc labelled with phycoerythrin (PE) and NGF-biotin/streptavidin labelled with allophycocyanin (APC). Analysis was focused on viable cells using forward scatter and staining. Cells positive for staining in the λem=450/40, λexc=355 channel (representing non-viable cells which took up 7-AAD) were excluded.
a. Histogram plot of fluorescence intensity for anti-Fc-PE against cell counts for the mixed input population (dashed line), and monoclonal HEK293 cell lines displaying MEDI-1912 (grey line) and MEDI-1912_STT (black line).
b. The dot-plot shows fluorescence intensity for anti-Fc-PE (x-axis), representing antibody expression level, plotted against fluorescence intensity for antigen binding (NGF-biotin/NGF-biotin/streptavidin-APC) on the y-axis. The gates chosen to separate the high and low antibody expression populations are labelled P5 and P6 and are shown as black boxes on the dot-plot. The total event count was 3.9×10.sup.6 cells and the number of cells sorted in the P5 and P6 gates was 2.5×10.sup.5 and 2.8×10.sup.5 cell respectively.

(30) FIG. 6. Enrichment of antibodies selected by mammalian cell display level. Cells sorted in gates 5 and 6 (FIG. 5) were expanded, genomic DNA prepared and antibody VH genes isolated by PCR. The two antibody populations were analysed by Nextgen sequencing to determine the proportion of MEDI-1912 and MEDI-1912_STT in the two gated populations. The histogram shows the percentage frequency of MEDI-1912 (hatched bars) and MEDI-1912_STT (solid black bars) in the low antibody expression population (Gate 6) and the high antibody expression population (Gate 5).

(31) FIG. 7. FACS separation of the MEDI-1912 library antibody populations based on antibody expression.

(32) A library of MEDI-1912 IgG genes, where NNS oligonucleotide-directed mutagenesis was used to randomly mutate the codon encoding W30, F31 and L56, were targeted via nuclease-directed integration into the AAVS locus of HEK293 cells. This mixed cell population, 15 days post-transfection, was separated on the basis of antibody expression and antigen binding by FACS using a BD Influx sorter. Cells were stained with anti-Fc labelled with phycoerythrin (PE) and NGF-biotin/streptavidin labelled with allophycocyanin (APC). Analysis was focused on viable cells using forward scatter and staining. Cells positive for staining in the λem=450/40, λexc=355 channel (representing non-viable cells which took up 7-AAD) were excluded. The dot-plot shows fluorescence intensity for anti-Fc-PE (x-axis), representing antibody expression level, plotted against fluorescence intensity for antigen binding (NGF-biotin/NGF-biotin/streptavidin-APC) on the y-axis for the parental MEDI-1912 displaying monoclonal cell line (a), the MEDI-1912_STT displaying monoclonal cell line (b) and the MEDI-1912 amino-acid position 30, 31 and 56 random library (c). The gates chosen for analysis are labelled P5 and P6 and are shown as boxes on the dot-plot.

(33) FIG. 8. Sequence distribution of selected MEDI-1912 variants.

(34) Histogram plots of amino acid identity frequency at MEDI-1912 VH positions 30, 31 and 56 for the MEDI-1912 VH library after mammalian display and FACS gated on Fc expression and NGF binding (P5 gate, FIG. 7). Amino acid (single letter code) is plotted on the x-axis against frequency (percentage occurrence) on the y-axis for the adjacent amino acids 30, hatched bars and 31, black bars (a) and amino acid 56 (b). Leucine in position 56 was excluded from the analysis.

(35) FIG. 9. Alignment of Bococizumab mouse parental antibody 5A10 VH (A) and VL (B) with the humanized intermediate antibody 5A10-i and Bococizumab.

(36) CDRs are indicated by bars above the sequence and residues to be mutated are highlighted in bold and underlined. Paratopic residues (amino acids that contribute to direct binding to PCSK9) are highlighted in italic and underlined. Dots indicate identity with the parental mouse mAb 5A10.

(37) FIG. 10. Antibody mammalian display expression of Bococizumab and the parental humanized intermediate antibody 5A10-i

(38) Targeting vector pINT17 encoding Bococizumab or 5A10-i) IgG were integrated into the AAVS locus of Hek293 cells via TALE nuclease. Cell (10.sup.6) were stained with anti-Fc-PE at 1, 8 or 21 days post transfection (dpt) and 106 cells analysed by flow cytometry with an iQue Intellicyte flow cytometer. Dead cells were excluded from the analysis. Histogram plots show fluorescence intensity (anti-Fc-PE) against cell count for the Bococizumab and 5A10-i cell display populations at 1, 8 and 21 days post transfection (dpt) with a staining of wild-type HEK293 cells included as a negative control.

(39) FIG. 11. Alignment of Bococizumab VH with human germ-line sequences (IMGT). Bococizumab VH (Query_1) was subject to an Ig Basic Local Alignment Search (IgBLAST) against the human VDJ database. The results are presented as an alignment to the query sequence encompassing framework region 1 (FR1), complementarity determining region 1 (CDR1), FR2, CDR2 and FR3 in order of percentage identity (second column). The human germ-line is shown in column 1. Residue identity to the Bococizumab VH sequence (.) and differences (single amino acid code) is shown in the multiple alignment.

(40) FIG. 12. DNA sequences encoding the Bococizumab VH variants and VL plus stop codon template. Variations compared with the original “wild-type” Bococizumab (row f) are highlighted in bold and underlined. The flanking 5′ and 3′ VH restriction sites (NcoI and XhoI) or VL restriction sites (NheI and NotI) are underlined. The VL stop codon are highlighted in bold and underlined.

(41) FIG. 13. Analytical flow cytometry analysis post-MACS purification of the Bococizumab libraries

(42) Hek293 cells transfected with the Bococizumab library were MACS purified 7 dpt with either anti-Fc or PCSK9i. (a) Flow cytometry dot plots are shown of anti-Fc expression (FL2, x-axis) plotted against PCSK9 binding (FL4, y-axis) for the post-MACS purified libraries, HEK293 cells and the unsorted library, Bococizumab and 5A10i transfectants, 9dpt. (b) Histogram of fluorescence intensity (anti-Fc, FL2, x-axis) plotted against cell count. Plots are (from top to bottom) the HEK293 control, Bococizumab, 5A10i, Bococizumab Library, anti-PCSK9 MACS purified Bococizumab Library, and anti-Fc MACS purified Bococizumab Library.

(43) FIG. 14. BD Influx sorter dot plots of Bococizumab libraries previously MACS purified based on antigen binding or anti-Fc.

(44) A library of Bococizumab IgG genes were targeted via nuclease-directed integration into the AAVS locus of HEK293 cells. This mixed cell population was first MACS purified based on PCSK9 binding (a) or anti-Fc (b). 16 days post-transfection the MACS enriched libraries were separated on the basis of antibody expression and antigen binding by FACS using a BD Influx sorter. Cells were stained with anti-Fc labelled with phycoerythrin (PE) and PCSK9-biotin/streptavidin labelled with allophycocyanin (APC). Analysis was focused on viable cells using forward scatter and staining. Cells positive for staining in the λem=450/40, λexc=355 channel (representing non-viable cells which took up 7-AAD) were excluded. The dot-plot shows fluorescence intensity for anti-Fc-PE (x-axis), representing antibody expression level, plotted against fluorescence intensity for antigen binding (PCSK9-biotin/streptavidin-APC) on the y-axis. The gate chosen for analysis are labelled P5 and P6 are shown as boxes on the dot-plot.

(45) FIG. 15. Bococizumab VH distribution after mammalian display selection. Random unselected input clones (84), antigen sorted (75) and Fc selected (85) were sequenced and the VH identity determined. A histogram plot shows the VH germ-line identity on the x-axis plotted against the percentage occurrence for the input (white filled bars), antigen MACS followed by antigen and anti-Fc FACS selected (black filled bars) and anti-Fc MACS followed by antigen and anti-Fc FACS selected (grey filled bars) mammalian cell display selected populations.

(46) FIG. 16. Bococizumab VL sequence analysis after mammalian display selection. Random un-selected input clones (84), antigen sorted (75) and Fc selected (85) were sequenced and the VL sequence determined. The average pI and aliphatic index was calculated for the 3 mutated codons. This showed a reduction in both pI and aliphatic index for the mammalian display selected antibodies.

(47) FIG. 17. Table listing the mammalian display Bococizumab clones, enriched for antigen binding by MACS followed by FACS enrichment for both antibody display level (anti-Fc) and antigen binding.

(48) Clones were sequenced and the VH CDR1 and CDR2 and VL CDR2 and CDR3 single letter amino acid sequences are shown with variations from the original Bococizumab sequence highlighted in bold and red. Targeted amino acids which retained the Bococizumab sequence are underlined. Binding of antibodies, including the original parental antibodies Bococizumab and 5A10-i, to antigen in a capture ELISA was performed and the binding signal in fluorescence units is shown in column 2. The AC-SINS assay was performed as described previously by Liu et al, 2014.sup.30 and column 3 shows the maximal absorbance wavelength shift compared to a no antibody PBS control (nm). The selected human VH germ-line is also indicated in column 6 by letter as detailed in Example 5:
a: VH Y33A-IGHV1-3*01
b: VH Y33D-IGHV1-8*01
c: VH S52N, F54S, R57S-IGHV1-46*01
d: VH Y33A, S52N, F54S, R57S (a. and c. mutants combined)
e: VH Y33D, S52N, F54S, R57S (b. and c. mutants combined)
f: Bococizumab “wild-type” sequence

(49) FIG. 18. HPLC-SEC of anti-PCSK9 IgG1 antibodies.

(50) Antibodies were expressed by transient transfection of Expi-293 cells, affinity purified by protein A chromatography and dialysed. Samples (2 μl at 1 mg/ml) were loaded onto an Agilent AdvancedBio SEC 300A, 2.7 um, 4.6×300 mm column (Agilent Technologies, Cat. No. PL1580-5301) at a flow rate of 0.35 ml/min using an Agilent 1100 HPLC instrument. A plot of retention time against absorbance is shown for selected antibodies. From black to progressively lighter shades of grey are: 5A10-i, 884_01_G01 (identified by mammalian cell display), Bococizumab and Alirocumab.

(51) FIG. 19. Gel filtration analysis Nivolumab (a) and Vesencumab (b).

(52) Antibodies were expressed by transient transfection of Expi293 cells followed by protein A affinity purification and dialysis. Purified Nivolumab (0.5 ml, 1.3 mg/ml) or Vesencumab (0.5 ml, 1.2 mg/ml) were loaded onto a Superdex 200 10/300 column connected to an AKTA Pure system using a PBS (pH 7.4) running buffer. The elution volume (ml) us plotted on the x axis against the absorbance at 280 nm (mAU) on the y-axis. The elution volume (Ve) for Nivolumab and Vesencumab was 12.0 ml and 13.7 ml respectively.

(53) FIG. 20. Stability determination of Nivolumab (a) and Vesencumab (b) after storage at 4° C., 2 weeks.

(54) Vesencumab and Nivolumab were purified by size exclusion chromatography (see FIG. 19) and their concentrations adjusted to 0.5 mg/ml in PBS (pH7.4). The antibodies were then stored at 4° C. for 2 weeks. Dynamic light scattering measurements were performed at 20° C. using a Zetasizer APS (Malvern Instruments, Malvern, UK) according to the manufacturer instructions. The hydrodynamic radius was evaluated with the Einstein-Stokes equation and plotted against scatter intensity. A single mono-disperse peak was observed for Nivolumab (a) in comparison to multiple aggregate peaks for Vesencumab (b).

(55) FIG. 21. Human serum binding to IgG on cell surface.

(56) Analysis was focused on viable cells using forward scatter and staining in the FL3 channel. Cells positive for staining in the FL3 channel (representing non-viable cells which took up 7-AAD) were excluded. Cells were transfected with pINT17-Nivolumab or pINT17-Vesencumab in presence of the AAVS TALENs. Stable populations were selected with Blasticidin. 20 days post-transfection, cells were stained with anti-Fc PE (FL2) and human serum (H4522, Sigma) labelled with Dylight 633 (325-0000, Innova). Panels are untransfected HEK293 cells (a), pINT17-Nivolumab (b) or pINT17-Vesencumab (c) transfected HEK293 cells.

(57) FIG. 22. Relationship between affinity, concentration of antigen and concentration of antibody

(58) a Concentration of complex using 0.1 nM antigen with differing concentrations of antibody of either K.sub.D 10 nM (dashed line) or 0.1 nM (solid line).

(59) bi Relative selectivity of binding to 0.1 nM antigen for higher affinity antibody (K.sub.D 0.1 nM) versus lower affinity (K.sub.D 10 nM) at different antibody concentrations

(60) Even with low (“stringent”) antigen concentrations, there is relatively little selectivity at high antibody concentrations but this increases as the antigen concentration drops.

(61) FIG. 23. Splice acceptor/donor variants to control antibody display level.

(62) The nucleic acid sequence from the HindIII site to the 5′ intron, including the splice donor region is shown for the pINT17-J9, J10, J29 and J30 variants. The original human “wild-type” sequence is J9 and the nucleotides varying from J9, at the splice junction, for J10, J29 and J30 are underlined.
FIGS. 24A-24D. pINT17-J30, a dual promoter antibody IgG expression cassette for reduced display surface expression. The annotated nucleic acid sequence is shown between the XhoI (4804) and SbfI (8387) restriction sites. Features: IgG1 CH1-3 4805-5808 Splice junction 5801-5802 Intron 5802-7104 M1 exon 7105-7239
BGH pA 7264-7478
AAVS right homology arm 7544-8381
3′β-globin insulator 8421-8492

(63) FIG. 25. Reduced surface expression of IgG on the cell surface using alternative transmembrane domain and splice variants.

(64) Analysis was focused on viable cells using forward scatter and staining in the FL3 channel. Cells positive for staining in the FL3 channel (representing non-viable cells which took up 7-AAD) were excluded. Cells were transfected with pINT17 targeting vectors in presence of the AAVS TALENs. Stable populations were selected with Blasticidin. 27 days post-transfection, cells were stained with anti-Fc PE (FL2). Flow cytometry dot-plot panels include pINT17-J9-Nivolumab (a), pINT17-J10-Nivolumab (b), pINT17-J29-Nivolumab (c), pINT17-J30-Nivolumab (d) and pINT17-BSD-Nivolumab (e).

(65) FIG. 26. Quantitation of IgG display level on the cell surface for antibodies expressed from the pINT17-BSD or pINT17-J30 targeting vectors

(66) Calibration beads FL2 staining was performed as described in the manufacturer instructions for the Quantum Simply Cellular anti-mouse IgG beads (catalogue number 815, Bangs Laboratories Inc) stained with mouse IgG-PE label. (a) Labelled histogram plot shows staining of the calibration bead set with peaks labelled 1, 2, 3 and 4 representing bead copy numbers of 12257, 72745, 283360, 886417 respectively. The blank peak represents bead with no capture antibody. (b) Calibration graph showing median fluorescence intensity (x-axis) plotted against copy number (y-axis). Cells-lines displaying 337_1_C08 (c) and Nivolumab (d) from either the pINT17-BSD or pINT17-J30 expression cassette were stained with anti-Fc-PE (5 μl, 0.1 mg/ml; 10.sup.5 cells). Analysis was focused on viable cells using forward scatter and staining in the FL3 channel. Cells positive for staining in the FL3 channel (representing non-viable cells which took up 7-AAD) were excluded. Histogram plots show fluorescence intensity against cell count for the pINT17-BSD with PDGFR TM (labelled and solid black line), pINT17-J30 (labelled and dotted line) and wild-type HEK293 cell lines (grey solid line) for cells displaying 337_1_C08 (c) and Nivolumab (d) respectively.

(67) FIG. 27. Separation of antibodies with different affinities for their target by mammalian display is enabled by a reduction in cell display copy number.

(68) Hek293 cells displaying Nivolumab and 337_1_C08 antibodies were labelled with 50 nM cell tracker green and 50 nM cell tracker red respectively. (a) demonstrates the display using the J30 splice variant and (b) demonstrates the display using PDGFR transmembrane domain encoded by the pINT17-BSD vector. Labelled cells were mixed equally and MACS sorted based on antigen binding. Sorted cells were analysed using the intellicyt flow cytometer. Dot plots represents Nivolumab on x-axis (FL1) and 337_1_C08 on y-axis (FL4). Panel i, ii, iii and iv represents 10 nM, 1 nM, 0.1 nM and no antigen respectively employed for MACS purification.

(69) FIG. 28. pINT18-Tet1, an inducible promoter antibody IgG expression vector for reduced display surface expression. The annotated nucleic acid sequence is shown between the AsiS1 (5) and SbfI (7672) restriction sites. The vector backbone exterior to the AsiSI and SbfI sites (7673-10922 and 1-4) encompassing the origins of replication and kanamycin resistance gene is identical to pINT17-BSD (FIG. 1).

(70) Key Features:

(71) AAVS left homology arm 9-812

(72) Blasticidin resistance gene 853-1254

(73) CMV promoter 1540-2112

(74) Reverse Tet activator (tTA) CDS 2164-3168

(75) SV40 pA 3178-3395

(76) tetO heptamer 3679-3932

(77) Minimal CMV promoter (PminCMV) 3946-4005

(78) BM40 leader 4016-4066

(79) Anti-PD1 MK3475 VL 4068-4413

(80) Human C kappa 4421-4738

(81) Furin cleavage site 4745-4756

(82) P2A peptide 4757-4816

(83) Mouse VH leader with intron 4829-4960

(84) Anti-PD1 MK3475 VH 4962-5321

(85) Optimised human IgG1 CH1-CH3 5322-6317

(86) Myc tag 6318-6347

(87) PDGFR anchor 6348-6503

(88) BGH polyA 6553-6767

(89) AAVS right homology arm 6829-7666

(90) 3′ β-globin insulator 7706-7777

(91) f1 replication origin 7824-8237

(92) pUC replication origin 8458-9132

(93) Kanamycin resistance gene 9852-10646

(94) FIG. 29. Inducible mammalian display expression

(95) Histogram representing staining results from a HEK293 cell line co-transfected with pINT18-Tet1-377_1_C08 and TALE nucleases and a stable cell population selected for 20 days in the presence of blasticidin. The sample was split into 5×10.sup.5 cells/ml in 20 mls and induced with either 20 ng/ml, 2 ng/ml and 0 ng/ml Doxycycline. 24 hours post induction, a flow staining was carried out using 1×10.sup.6 cells from each doxycycline induced sample. Cells were stained using anti-Fc-PE and TOPRO-3 viability stain. The histogram shows the fluorescence intensity on the FL2 channel (anti-Fc-PE) plotted against cell count. HEK293 WT control (grey solid line), HEK293-pINT18-Tet1-377_1_C08 stable cell line induced with 0 ng/ml doxycycline (black dashed line), 2 ng/ml doxycycline (black dotted line) and 20 ng/ml doxycycline (black solid line).

(96) FIG. 30. Binding of cell displayed antibodies to FcRn.

(97) HEK293 cells expressing Briakinumab and Ustenkinumab were stained with biotinylated FcRn (50 nM) preconjugated with streptavidin PE (11 nM) using different buffers:

(98) a. Hek293 WT

(99) b. Streptavidin PE-control

(100) c. Cells stained with buffer pH6.0

(101) d. Cells stained with buffer pH7.4

(102) FIG. 31. FACS separation of HEK293 cell displayed anti-Mesothelin IgG by display level.

(103) A population of anti-Mesothelin antibody genes were integrated into the human AAVS locus of HEK293 cells by nuclease mediated gene transfer. The polyclonal population of HEK293 cell displayed antibodies were separated by FACS according to antibody display level by staining with anti-human Fc-PE. 16 days post-transfection the MACS enriched libraries were separated on the basis of antibody expression by FACS using a BD Influx sorter. Cells were stained with anti-Fc labelled with phycoerythrin (PE). Analysis was focused on viable cells using forward scatter and staining. Cells positive for staining in the λem=450/40, λexc=355 channel (representing non-viable cells which took up DAPI) were excluded. The histogram shows fluorescence intensity for anti-Fc-PE (x-axis), representing antibody expression level, plotted against cell count on the y-axis. The gate chosen for analysis are labelled P4, P6 and P5 representing the low, medium and high display level populations respectively.

(104) FIG. 32. HPLC-SEC of two anti-mesothelin IgG1 antibody clones originating from the high display level group (solid line) and low display level group (dotted line) respectively.

(105) FIG. 33. DNA binding and depletion of DNA binders using MACS. (A) Overlay of HEK293 cells (solid grey), or HEK293 cells displaying ustekinumab (dashed), briakinumab (long dashed) and amatuximab (dotted) stained with biotinylated DNA detected with streptavidin PE; (B) Dot plot representing the mixture of three antibody cell populations displaying ustekinumab (unlabelled, Q4), amatuximab (labelled with CellTace Far red, X-axis) and briakinumab (labelled with CellTrace CFSE, Y-axis) stained with DNA before MACS sorting; (C) Dot plot of flow-through showing the depletion of DNA binders.

(106) FIG. 34. Dual staining with Heparin-FITC (x-axis) and anti-human Fc APC (y-axis). (a) Dot plot showing overlay of ustekinumab (grey) and briakinumab (black). (b) Dot plot showing overlay of ustekinumab (grey) and ganitumab (black). Gate within the overlay plots indicates the cells to be high expressers and non-binders to heparin.

(107) FIG. 35. Dual staining with chaperones conjugated with DyLight 633 (x-axis) and anti-human Fc PE (y-axis). (a) Dot plot showing overlay of ustekinumab (grey) and briakinumab (black) double-stained with Hsp70-DyLight 633 and anti-human Fc PE. (b) Dot plot showing overlay of ustekinumab (grey) and briakinumab (black) double-stained with Hsp90-DyLight 633 and anti-human Fc PE. Gate within the overlay plots indicates the cells to be high expressers and non-binders to chaperones (Hsp70 and Hsp90). (c and d) Overlay histogram plot shows lenzilumab and brentuximab binding Hsp70 and Hsp90 respectively.

(108) FIG. 36. Histogram plots for antibodies stained with anti-human Fc PE and various polyreactivity probes. Stable monoclonal HEK293 cell lines, displaying a selection of antibodies, were created by nuclease mediated gene integration. Histograms plots of cell count (y-axis) against fluorescence intensity (x-axis) are shown for different antibodies displayed on the surface of HEK293 cells with the following probes: (a) anti-human Fc-PE, (b) biotinylated DNA detected using streptavidin PE, (c) Heparin-FITC, (d) Streptavidin PE, (e) Hsp70-DyLight 633, (f) Hsp90-DyLight 633 and (g) FcRn pre-conjugated with streptavidin PE.

(109) FIG. 37. pINT17-Tet-D1.3, an inducible antibody IgG mammalian display expression vector. The full annotated nucleic acid sequence is shown between the AAVS homology arms and promoter-less blasticidin gene from the BgIII to BstZ171 restriction sites. Numbering is from the BgIII restriction site. Key features are listed below.

(110) BGH poly A 223-9 (reverse strand)

(111) Human C kappa 544-236 (reverse strand)

(112) D1.3 VL 877-549 (reverse strand)

(113) Human VL leader with intron 1168-883 (reverse strand)

(114) TRE3G promoter 1230-1618

(115) CMV promoter 1237-1809

(116) VH leader with intron 1644-1782

(117) D1.3 VH 1783-2127

(118) IgG1 CH1-3 2125-3120

(119) Myc tag 3121-3150

(120) PDGFR anchor 3151-3306

(121) BGH poly A 3356-3570

(122) pEF promoter 3621-4955

(123) rtTA-3G 5063-5809

(124) SV40 poly A 5832-6274

(125) FIG. 38. Inducible IgG mammalian display cell lines. 1549_02_D06 (1), 1535_01_E03 (2), and 337_1_C08 (3), bococizumab (4), 884_01_G01 (5), 5A10i (6) and alirocumab (7). 27 dpt the cell lines were induced by the addition of 0 (a), 2 (b), 4 (c) or 100 (d) ng/ml doxycycline. 24 hours post induction the cells were stained with anti-Fc-PE. Histograms of fluorescence intensity (anti-Fc, FL2, x-axis) plotted against cell count.

(126) FIG. 39. Inducible IgG mammalian display cell lines: cell surface IgG turn-over pINT17-Tet harbouring the VH and VL of anti-PD1 antibodies: 1549_02_D06 (1), 1535_01_E03 (2), and 337_1_C08 (3) and the anti-PCSK9 antibodies bococizumab (4), 884_01_G01 (5), 5A10i (6) and alirocumab (7) was used to create stable HEK293 cell lines by AAVS TALE nuclease mediated gene integration and blasticidin selection. 27 dpt the cell lines were induced by the addition of 100 ng/ml doxycycline. 48 hours post induction the cells were stained with anti-Fc-PE. Histograms of fluorescence intensity (anti-Fc, FL2, x-axis) are shown plotted against cell count.

(127) FIG. 40. Cell lines displaying the anti-PD1 antibodies 1549_02_D06 (K.sub.D=2.9 nM for PD-1) and 337_1_C08 (K.sub.D=74 nM for PD-1) were induced with (a) 0, (b) 2, (c) 4 and (d) 100 ng/ml doxycycline respectively. Dot plots of fluorescence in the FL1 channel (y-axis) against forward side-scatter (FSC, x-axis) are shown. Labelled cells displaying 1549_02_D06 are shown in the upper quadrant in each dot plot and unlabelled cells displaying 337_1_C08 are shown in the lower quadrant. Panels i, ii, iii and iv represent 0.1, 1, 10 nM concentration of PD-1-biotin respectively employed for MACS purification. Panel iv represents the input pre-MACS population. The percentage of each cell population is shown within each quadrant.

(128) FIG. 41. Overlay dot plot of double-stained population of ustekinumab (grey) and briakinumab (black). Dual staining with 50 nM FcRn-Avi tag pre-conjugated with streptavidin PE (x-axis) and anti-human Fc APC (y-axis). The gate within the plot represents ustekinumab as the FcRn non-binder which can be FACS sorted from the FcRn binder.

(129) FIG. 42. Germ-line analysis of the anti-mesothelin variable heavy (VH) domain antibody populations. The chart plots frequency of occurrence in the input and low, medium and high mammalian display gated populations for each VH germ-line.

(130) FIG. 43. Germ-line analysis of the anti-mesothelin variable light kappa (VLκ) domain antibody populations. The chart plots frequency of occurrence in the input and low, medium and high mammalian display gated populations for each VLκ germ-line.

(131) FIG. 44. Germ-line analysis of the anti-mesothelin variable light lambda (VLλ) domain antibody populations. The chart plots frequency of occurrence in the input and low, medium and high mammalian display gated populations for each VLλ germ-line.

(132) FIG. 45. HPLC-SEC of anti-mesothelin IgG1 antibodies. Antibodies were expressed by transient transfection of Expi-293 cells, affinity purified by protein A chromatography and dialysed. Samples (2 μl at 1 mg/ml) were loaded onto an Agilent AdvancedBio SEC 300A, 2.7 um, 4.6×300 mm column (Agilent Technologies, Cat. No. PL1580-5301) at a flow rate of 0.35 ml/min using an Agilent 1100 HPLC instrument. A plot of absorbance at 215 nm against retention time is shown for selected anti-Mesothelin antibodies: 932_01_A03 (black line), originating from the high display level group and 930_01_A12 (alternating dot and dash) 930_01_1B02 (long dash), 930_01_C12 (short dash line) originating from the low display level group.

(133) FIG. 46. Alignment of the human and CHO AAVS intron 1 TALE-nuclease (TALEN) target binding sites. The CHO AAVS intron 1 DNA sequence was obtained from the ENSEMBL annotated CHO-K1 glutamine synthetase (GS) knockout cell line, accession: CHOK1GS_HDv1:scaffold_52:2374828:2406177:1. Numbering is referenced to human PPP1R12C intron 1 start. Bold indicates the left and right arms of the human TALEN target sites. Asterisks indicate homology between the human and CHO sequence and dash (-) indicates a deletion. Underline and italic indicate the ends of the AAVS left and right homology arms within the pINT17 targeting vector. This alignment was used to design CHO AAVS homology arms within the pINT17-CHO targeting vector and, for comparison, CRISPR/Cas9 guide RNAs. Sense or anti-sense CRISPR guide RNA recognition sites numbered 1 to 3 are shown above or below the sequence respectively.

(134) FIG. 47. CHO AAVS homology arms within the vector pINT17-BSD-CHO, a dual promoter antibody IgG expression cassette for surface expression on CHO cells. An annotated DNA sequence is shown for the left and right CHO AAVS homology arms within the vector. All remaining features, including those not shown in this figure, encompassing the dual promoter antibody expression cassette, are as described for the vector pINT17-BSD (FIG. 1) and are listed below

(135) Features:

(136) CHO AAVS left homology arm 9-899

(137) Blasticidin resistance gene 942-1343

(138) pEF promoter 1611-2794

(139) BM40 leader 2834-2885

(140) Humanised D1.3 VL 2888-3219

(141) Human C kappa 3227-3532

(142) BGH poly A 3468-3682

(143) CMV promoter 3790-4362

(144) Mouse VH leader with intron 4388-4515

(145) Humanised D1.3 VH 4521-4868

(146) Optimised human IgG1 CH1-CH3 4869-5864

(147) Myc tag 5865-5894

(148) PDGFR anchor 5895-6050

(149) BGH polyA 6100-6314

(150) CHO AAVS left homology arm 6376-7266

(151) f1 replication origin 7424-7837

(152) pUC replication origin 8058-8732

(153) Kanamycin resistance gene 9452-10246

(154) FIG. 48. Display of antibodies on the surface of CHO cells by TALEN or CRISPR/Cas9 nuclease mediated gene integration. Histograms of fluorescence intensity (anti-Fc, FL2, x-axis) plotted against cell count. Plots are (from top to bottom) the CHO control, pINT17-BSD-CHO V2-Nivolumab minus nuclease, pINT17-BSD-CHO V1-Nivolumab minus nuclease, pINT17-BSD-CHO V1-Nivolumab plus CHO TALENs, pINT17-BSD-CHO V1-Nivolumab plus CRISPR3, pINT17-BSD-CHO V1-Nivolumab plus CRISPR2, pINT17-BSD-CHO V1-Nivolumab plus CRISPR1.

(155) FIG. 49. Display levels of antibodies on the surface of CHO. Histograms of fluorescence intensity (anti-Fc, FL2, x-axis) plotted against cell count for CHO cells (filled plot) (a) Bococizumab (solid line) and 884_01_G01 (dashed line). (b) MEDI-1912 (solid line) and MEDI-1912-STT (dashed line).

(156) FIG. 50. Creation of a “developability enhanced” population using mammalian display for subsequent binding selection

(157) a. Sequences of anti-PD1 337_1_C08 VH (i) and VL (ii) chains. Nucleotide sequences are shown with translation single letter amino acid code above the codons. CDRs are annotated (under-lined) and CDR3 amino-acids subject to site-directed mutagenesis highlighted in bold.
b. The anti-PD1 antibody VH and VL CDR3 mammalian display library was separated by FACS on the basis of high, medium and low antibody cell display levels and the analysed by analytical flow cytometry by staining with anti-Fc-PE. Histogram plots of cell count (y-axis) against Fc expression (x-axis) are shown (from top to bottom) the high (i), medium (ii) and low (iii) anti-PD1 populations. For reference, the starting anti-Fc MACS population (iv), the “wild-type” 337_1_C08 parental clone (v) and HEK293 cells with no displayed antibody (vi) are shown.

(158) FIG. 51. pINT17-Bi-CMV-Emicizumab, a bi-directional CMV and elongation factor (pEF) promoter containing plasmid for cell surface expression of the bi-specific “knobs-into-holes”, common light chain IgG Emicizumab. This is a tri-cistronic targeting vector with three promoters driving the expression of three genes: the anti-FIXa heavy chain, the anti-FX heavy chain and common light chain. The full annotated nucleic acid sequence is shown between the AAVS homology arms from the BgIII to BstZ171 restriction sites. Numbering is from the BgIII restriction site. Key features are listed below.

(159) BGH poly A 222-8 (reverse strand)

(160) Human C kappa 546-232 (reverse strand)

(161) Emicizumab VL 876-547 (reverse strand)

(162) Human VL leader with intron 1143-884 (reverse strand)

(163) Minimal CMV promoter 1230-1167 (reverse strand)

(164) CMV promoter 1237-1809

(165) Mouse VH leader with intron 1835-1973

(166) Emicizumab anti-FIXa VH 1974-2339

(167) Emicizumab anti-FIXa CH1-3 2340-3317

(168) Myc tag 3318-3347

(169) PDGFR anchor 3348-3503

(170) BGH poly A 3553-3767

(171) pEF promoter 3818-5152

(172) Human VH leader with intron 5260-5401

(173) Emicizumab anti-FX VH 5402-5758

(174) Emicizumab anti-FX CH1-3 5759-6733

(175) Myc tag 6734-6763

(176) PDGFR anchor 6764-6916

(177) SV40 poly A 6942-7384

(178) FIG. 52. Binding of FIXa and FX to the bi-specific antibody Emicizumab displayed on the surface of HEK293 cells

(179) pINT17-Bi-CMV-Emicizumab or pINT17-BSD-anti-FIXa was used to transfect HEK293 cells in the presence of plasmids encoding the AAVS TALENs. 24 hours post transfection the cells were analysed antibody display and the ability to bind the antigens FIXa or FX. The histogram plots depict cell count against fluorescence intensity when stained with, from left to right: anti-Fc-APC, FX-biotin or FIXa-biotin, pre-conjugated with streptavidin-PE or streptavidin-PE alone for HEK293 cells displaying (a) Bi-specific Emicizumab (black dash line), (b) anti-FIXa IgG (solid blck line), (c) HEK293 cells.

(180) FIG. 53. Alignment of Emicizumab VL with parental Emicizumab VLs.

(181) CDRs are indicated by bars above the sequence. Dots indicate identity with the final Emicizumab VL. Residues contributing to the positive charge patch are highlighted in bold.

(182) FIG. 54. Relationship between display of knotbodies on the surface of HEK293 cells and their biophysical properties

(183) (a) HEK293 cells were transfected with pINT17-knotbodies in presence of the AAVS TALENs. Stable populations were selected with Blasticidin. 7 days post-transfection, cells were stained with anti-Fc PE (FL2) and analysed by flow cytometry. The histogram depicts cell count (y-axis) plot against fluorescence intensity (anti-Fc-PE, x-axis) and include the KB_A12 EETI-II (black solid line), KB_A12 Hstx1 (dotted line) and KB_A12 ProTxIII (dashed line). The traces of KB_A12 Hstx1 (dotted line) and KB_A12 ProTxIII (dashed line) over-lap and appear merged.

(184) Knotbodies were expressed by transient transfection of Expi293 cells and purified by Protein A affinity chromatography. The knotbodies were analysed by HPLC-SEC as described above and plots of absorbance against elution volume are shown for (b) KB_A12 EETI-II, (c) Trastuzumab and (d) KB_A12 ProTx-III.

(185) FIG. 55. Mutant libraries of knotbodies contain clones with improved display levels compared with the parental knotbodies. HEK293 cells displaying knotbodies were stained with anti-Fc-PE and analysed by flow cytometry. Histogram plots of cell count against fluorescence intensity are shown for the three libraries (after anti-Fc MACS purification) compared to their relevant parental knobody control displaying cell line. (From Left to Right): (a) KB_A12 ProTxIII Library Set A (dotted line) with KB_A12 ProTxIII Control (filled line), (b) KB_A12 ProTxIII Library Set B (dotted line) with KB_A12 ProTxIII Control (filled line), (c) KB_A12 HsTx1 library (dotted line) with KB_A12 HsTx1 Control (filled line).

EXAMPLES

Example 1. Construction of Targeting Vectors for Soluble Expression and Cell Surface Displayed IgG Formatted Antibodies

(186) To enable the display of binder molecules, including antibodies, on the surface of higher eukaryotic cells and their subsequent genetic selection, vectors may be used to target the binder gene to a particular location in the host genome. The vector may encode a selectable marker, to enable selection of stable cell lines and this selectable marker may encode genes conferring resistance to blasticidin, G418/Geneticin, hygromycin, puromycin or zeocin. The targeting vector may contain an exogenous promoter to drive expression of the gene encoding the selectable marker. Alternatively, the transgene may be integrated into the cellular DNA at a location downstream of an endogenous promoter to enable the preferential selection of the correctly integrated transgenes. The targeting vector will also encode homology arms to allow homologous recombination to the relevant chromosomal locus and promoters to drive expression of the binder molecule and polyadenylation (pA) sites. The binder molecule gene will be fused to DNA encoding a leader sequence to allow secretion, via the endoplasmic reticulum (ER), to the cell surface and a membrane anchor such as a transmembrane domain or glycosylphosphatidylinositol (GPI) anchor.

(187) A schematic map of the targeting vector used here is shown in FIG. 1a and the full annotated DNA sequence is shown in FIG. 1b. The plasmid includes the AAVS homology arms, flanking the expression cassette, to allow homologous recombination of the transgene into the human AAVS site. The AAVS locus was originally identified as a common integration site of the adeno-associated virus and is a “safe harbour” locus for insertion and expression of heterologous genes in human cells.sup.93. After nuclease mediated cleavage within the AAVS site, the AAVS homology arms in the targeting vector promote the integration of the expression cassette by homologous recombination. The blasticidin gene lacks a promoter within the vector, but is preceded by a splice acceptor that creates an in-frame fusion with the upstream exon from the AAVS locus. The details of the antibody heavy and light chain expression cassette are described below and in FIG. 1.

(188) The targeting vector pINT17-BSD (FIGS. 1a and 1b) was constructed by polymerase chain reaction (PCR) amplification of selected fragments from vectors previously described (WO2015166272A2) with the addition of restriction sites to enable their subsequent assembly. The origins of the various elements of pINT17-BSD are now described. DNA encoding the AAVS-left homology arm, splice acceptor, blasticidin resistance gene, poly-adenylation site and the elongation factor 1 alpha promoter (pEF1α) originated from the pD2 plasmid (WO2015166272A2) by PCR amplification (1511 bp) with the addition of the 5′ AsiSI and 3′ BgIII restriction enzymes. DNA encoding the Myc-tag and PDGFR transmembrane domain was PCR amplified from the pD2 plasmid with the addition of a 5′ IgG1 CH3 homology sequence and 3′-HindIII site. DNA encoding the light chain BM40 leader, variable light chain (VL) of the anti-lysozyme antibody D1.3.sup.94, the human constant light (CL), bovine growth hormone (BGH) pA, the immediate early cytomegalovirus promoter (CMV promoter), a mouse heavy chain leader split by an intron, variable heavy chain (VH) of the anti-lysozyme antibody D1.3 and the IgG1 antibody constant heavy domain 1 to 3 (IgG1 CH1-3) was PCR amplified from the previously described pINT3 plasmid (WO2015166272A2) with a 5′ BgIII site and 3′ addition of DNA encoding the Myc-tag. The two fragments encoding the 4446 bp region of pINT17-BSD from BgIII to HindIII were combined by PCR assembly to add the PDGR transmembrane domain directly to the CH3 terminus. The region from HindIII to SbfI, encoding the AAVS right homology arm was PCR amplified (1168 bp) from the pD2 plasmid (WO2015166272A2) with the addition of the HindIII and SbfI restriction sites. The vector backbone encompassing the f1 and pUC origin of replications and Kanamycin resistance gene from the SbfI to AsiSI sites originated from pSF-EF1 alpha (Oxford Genetics OG43). The example shown in FIG. 1b encodes the VL and VH of a human anti-lysozyme antibody but this can be conveniently substituted for other specificities using standard molecular biology techniques (for example using the flanking restriction enzymes to replace the VL and VH genes).

Example 2. Comparison of Surface Presentation Level of Parental and Developability Enhanced Clones for 3 Pairs of Antibodies

(189) We examined three antibody pairs where the original parental antibody possesses a poor developability profile and their re-engineered daughter molecules, which were altered to improve their self-interaction and cross-interaction properties. The panel included CNTO607, a monoclonal antibody against interleukin IL-13, and its modified counterpart CNTO607 W100A.sup.14. CNTO607 is poorly soluble at neutral pH, precipitates in PBS buffer at high concentrations and displays self-interaction as measured in an affinity-capture self-interaction nanoparticle spectroscopy (AC-SINS) assay.sup.39. Structure determination of CNTO697 revealed a hydrophobic patch in the heavy chain CDR3. The VH CDR3 mutation W100A improved both its antibody solubility and cross-interaction chromatography (CIC) profile.sup.47. CIC measures binding to human serum polyclonal antibodies immobilized on a column matrix. A second example is a monoclonal antibody, named Ang2mAb, which targets Angiopoietin 2, a soluble ligand for the Tie2 receptor and regulator of pathological angiogenesis. However, Ang2mAb was reported to exhibit both poor expression and aggregation. A combination of structural modelling and experimental screening of 19 variants led to the engineering of the better expressing Ang2mAb C49T.sup.6, which mutated an unpaired cysteine residue. Finally, we included MEDI-1912, an anti-nerve growth factor (NGF) antibody that inhibits signaling via the TrkA and p75 receptors.sup.7. MEDI-1912 could potentially be used in the treatment of chronic pain, but shows precipitation and aggregation in solution and a poor pharmacokinetic profile. MEDI-1912 binds to NGF with pico-molar affinity and was affinity matured from a “grand-parental” antibody named MEDI-578 which was well-behaved in terms of self-aggregation. By hydrogen/deuterium exchange—mass spectrometry (HDX-MS) and molecular modelling, a hydrophobic patch was identified on the VH domain caused by residues within VH CDR1 and CDR2. This allowed the prediction of the amino acids responsible for self-association and consequent aggregation. This in turn enabled the design of a triple mutant (MEDI-1912_STT) with mutations W30S, F31T and L56T that interrupted the self-interaction interface whilst retaining potency and affinity for NGF.sup.7.

(190) Synthetic DNA encoding the CNTO607, CNTO607-W100A, Ang2mAb, Ang2mAb-C49T, MEDI-1912 and MEDI-1912_STT heavy and light variable domains (see Table 1) for sequences) were cloned into the mammalian display vector pINT17-BSD (see Example 1 for vector maps and sequences), DNA sequence confirmed and transfection quality plasmid DNA prepared. Suspension adapted HEK293 cells were seeded at 5×10.sup.5 cells per ml in HEK FreeStyle 293 expression media one day before transfection. PEI-transfection was performed when the cells reached a density of 1×10.sup.6 cells/ml in 10 ml. pINT17-harboring antibody genes (1 ug), left and right TALEN plasmids (5 ug each) were mixed and diluted in unsupplemented HEK FreeStyle 293 expression media (1 ml). Polyethylenimine (PEI), linear, 25000 Da MW (10 ul, 1 mg/ml, Polysciences) was added, incubated for 10 minutes at room-temperature. The plasmid DNA/PEI mix was then added to HEK293 suspension cells (1×10.sup.6 cells/ml in 10 ml HEK FreeStyle 293 expression media). Blasticidin selection was started 48 hours after transfection at a concentration of 7 μg/ml. The population was kept under selection for the duration of the experiment. After 15 days post-transfection (dpt) cells were stained with anti-human Fc PE (Biolegend). The monoclonal cell lines displaying antibodies were then stained by the following protocol. HEK293 cell lines displaying antibodies or wild-type HEK293 cells (one million cells) were pelleted (200 g, 3 minutes in an Eppendorf tube (1.5 ml). The pellet was resuspended in PBS (1 ml) and centrifuged (600 g, 2.5 min). The pellet was resuspended in 1% BSA, PBS (100 μl) containing anti-Fc PE (5 μl, Biolegend). The mix was incubated, shielded from light, at 4° C. for 30 min. 0.1% BSA, PBS (900 μl) was added and cells pelleted (600 g, 2.5 min). The cells were resuspended in 0.1% BSA, PBS (1 ml) and this wash step was repeated once. The cells were resuspended in 0.1% BSA, PBS (200 μl) with 7-AAD (5 μl per million cells). Labelled cells (50 μl) were analysed using the Intellicyte iQue screener. Flow cytometry analysis (FIG. 2) showed increased antibody display levels for the improved display levels for the improved daughter molecules for all three antibody pairs compared with the original problematic parental molecules.

(191) TABLE-US-00002 TABLE 1 Protein sequence of VH and VL genes of test antibodies. Amino acid sequences in single letter code are shown of the variable antibody heavy and light chains. The variable domains are underlined. Variant residues between antibody pairs are highlighted in bold. Chain Protein Sequence Ang2 VH QVQLVESGGGVVQPGRSLRLSCAASGFTFTNYG MHWGRQAPGKGLEWVAVISHDGNNKYYVDSVKG RFTISRDNSKNTLYLQMNSLRAEDTAVYYCARE GIDFWSGLNWFDPWGQGTLVTVSS (SEQ ID NO: 2) Ang2 VL EIVLTQSPGTLSLSPGERATLSCRASQSITGSY LAWYQQKPGQAPRLLICGASSWATGIPDRFSGS GSGTDFTLTISRLEPEDFAVYYCQQYSSSPITF GQGTRLEIK (SEQ ID NO: 3) Ang2 VL C49T EIVLTQSPGTLSLSPGERATLSCRASQSITGSY LAWYQQKPGQAPRLLITGASSWATGIPDRFSGS GSGTDFTLTISRLEPEDFAVYYCQQYSSSPITF GQGTRLEIK (SEQ ID NO: 4) CNT607 VH QVQLVESGGGLVQPGGSLRLSCAASGFTFNSYW INWVRQAPGKGLEWVSGIAYDSSNTLYADSVKG RFTISRDNSKNTLYLQMNSLRAEDTAVYYCARG LGAFHWDMQPDYWGQGTLVTVSSAS (SEQ ID NO: 5) CNT607 QVQLVESGGGLVQPGGSLRLSCAASGFTFNSYW VH W100A INWVRQAPGKGLEWVSGIAYDSSNTLYADSVKG RFTISRDNSKNTLYLQMNSLRAEDTAVYYCARG LGAFHADMQPDYWGQGTLVTVSS (SEQ ID NO: 6) CNT607 VL SYELTQPPSVSVAPGQTARISCSGDNIGGTFVS WYQQKPGQAPVLVIYDDNDRPSGIPERFSGSNS GNTATLTISGTQAEDEADYYCGTWDMVTNNVFG GGTKLTVL (SEQ ID NO: 7) MEDI-1912 VH QVQLVQSGAEVKKPGSSVKVSCKASGGTFWFGA FTWVRQAPGQGLEWMGGIIPIFGLTNLAQNFQG RVTITADESTSTVYMELSSLRSEDTAVYYCARS SRIYDLNPSLTAYYDMDVWGQGTMVTVSS (SEQ ID NO: 8) MEDI1912 VH  QVQLVQSGAEVKKPGSSVKVSCKASGGTFSTGA STT FTWVRQAPGQGLEWMGGIIPIFGTTNLAQNFQG RVTITADESTSTVYMELSSLRSEDTAVYYCARS SRIYDLNPSLTAYYDMDVWGQGTMVTVSS (SEQ ID NO: 9) MEDI-1912 VL QSVLTQPPSVSAAPGQKVTISCSGSSSDIGNNY VSWYQQLPGTAPKLLIYDNNKRPSGIPDRFSGS KSGTSATLGITGLQTGDEADYYCGTWDSSLSAW VFGGGTKLTVL (SEQ ID NO: 10) Vesencumab VH EVQLVESGGGLVQPGGSLRLSCAASGFTFSSYA MSWVRQAPGKGLEWVSQISPAGGYTNYADSVKG RFTISADTSKNTAYLQMNSLRAEDTAVYYCARG ELPYYRMSKVMDVWGQGTLVTVSS (SEQ ID NO: 11) Vesencumab VL DIQMTQSPSSLSASVGDRVTITCRASQYFSSYL AWYQQKPGKAPKLLIYGASSRASGVPSRFSGSG SGTDFTLTISSLQPEDFATYYCQQYL GSPPTFGQGTKVEIK (SEQ ID NO: 12)

Example 3a. The Relationship Between Cell Surface Presentation Level and Self-Interaction at High Concentrations

(192) To examine the properties of the antibodies described in Example 2, antibody expression and purification was performed. Synthetic DNA encoding CNTO607, CNTO607-W100A, Ang2mAb, Ang2mAb-C49T, MEDI-1912 and MEDI-1912_STT heavy and light variable domains (see Table 1 for sequences) were cloned into a dual promoter IgG soluble expression vector based on pINT3 (WO2015166272A2) and correct cloning confirmed by DNA sequencing.

(193) Plasmid DNA was prepared and this was used to transfect Expi293 cells (30 ml final culture volume scale) using the transfection reagent ExpiFectamine according to the manufacturer instructions (A14525, ThermoFisher Scientific). Cells were seeded at a density of 2×106 cells/ml in 25.5 ml of Expi293 Expression Medium 24 hours prior to transfection. Plasmid DNA (30 μg) was diluted in Opti-MEM Medium (1.5 ml) and ExpiFectamine 293 Reagent (80 μl) was diluted in Opti-MEM Medium (1.5 ml) and incubated for 5 minutes at room temperature. The diluted plasmid DNA (30 μg in 1.5 ml Opti-MEM Medium) was then added to the diluted ExpiFectamine 293 Reagent (80 μl ExpiFectamine in 1.5 ml Opti-MEM Medium) and incubated for 20 minutes at room temperature. The cells were incubated at 37° C., 5% CO2, 5% humidity and agitated at 130 rpm (25 mm orbital throw, ISF1-X, Climo-Shaker, Kuhner). Following 5 days of expression, culture supernatant was harvested by centrifugation (2000 g, 20 min).

(194) The culture supernatants in 50 ml centrifuge tubes were pH adjusted by the addition of 1/10 volume of PBS (pH7.4) and Protein A sepharose FF resin (300 μl, Generon, PC-A100) added and incubated by agitation for 1 hour at room-temperature. The 50 ml tubes were centrifuged at 2000 g for 5 mins to collect the beads and supernatant discarded leaving approximately 1 ml behind of bead slurry. This slurry was resuspended loaded onto a column with a frit (Proteus “1 step batch” midi spin column. Generon, GEN-1SB08), centrifuged (50 g, 1 min at 4° C.) and flow through discarded The column was washed with 2×PBS (2×10 ml) followed by centrifugation (50 g, 1 min at 4° C.) after each wash step. Antibody was eluted using elution buffer (900 ul, 0.2 M Glycine pH 2.6) which was added to the column matrix and the eluate immediately neutralized using neutralisation buffer (300 ul, 1 M Tris-HCl, pH 8). Antibodies were then eluted from the Protein A sepharose column by centrifugation (50 g, 1 min at 4° C.) directly into the neutralisation buffer. The antibodies were buffer exchanged by transfer to a GeBAflex maxi tube (8 kDa molecular weight cut-off, Generon, D045) and dialysed in 4 litres of PBS and incubation for at least 3-18 hours at 4° C. This dialysis step was repeated with a second 4 L PBS dialysis step. The antibody yield and concentration was determined by measurement of the absorbance at 280 nm and calculating using the Beer-Lambert Law using an estimated extinction coefficient of 1.4 to approximate the concentration.

(195) The yield of polypeptide generated by transient expression may be considered as an indicator of developability potential. Although comparison of the expression yields in transient transfection between the 3 pairs of antibodies from example 2 showed lower expression of the parental antibody, the significant difference in developability potential was not apparent simply by comparing yield in transient transfection (Table 2). For example, the expression yield of the parental CNTO607 antibody was 34 mg/L whereas the solubility improved CNTO607-W100A antibody expression yield was 55 mg/L. Similarly, the expression yields of parental antibodies MEDI-1912 was 33 compared with 53 mg/l for the improved version. The yield of Ang2mAb was 13 mg/L compared with 34 mg/L for the engineered off-spring Ang C49T.

(196) The melting temperature of a polypeptide is sometimes taken as a surrogate to predict “developability” and in some instances antibodies have been selected for improved melting temperature in the expectation of generating more developable antibodies.sup.9-11. The melting temperature (Tm) and temperature of the onset of aggregation (Tagg) were determined using Prometheus NT.4B (Nanotemper) according to the manufacturer instructions. Small capillaries were used to take up approximately 8-10 μl of antibody solution at 0.5 mg/ml. The capillaries were then clipped in place for fluorescent scanning by the Prometheus instrument for thermal melt analysis. The Prometheus fitting software was used to determine the temperature for the onset of melting and the temperature for the onset of scattering. The melting temperature (Tm) and the aggregation temperature (Tagg) of the antibodies were similar (see Table 2), both between the antibody pair sets and compared to the clinically approved positive control anti-PD1 antibody, Nivolumab. This data indicates that the melting and aggregation temperatures of this antibody set are not predictive of their biophysical profiles of self-interaction and non-specific cross-interaction.

(197) During preparative size exclusion chromatography MEDI-1912 displayed an earlier elution profile compared with MEDI-1912_STT (FIG. 3), indicating that it exists as a higher molecular weight species and is prone to self-interaction. The remaining antibodies eluted with a similar profile to Nivolumab. To enable measurement of antibody self-interaction by dynamic light scattering (DLS), the size purified antibodies were concentrated by ultra-filtration. The antibody concentration achieved for each antibody pair is shown in Table 2. This revealed that it was not possible to concentrate the parental antibodies MEDI-1912 and CNTO607 beyond 1.4 mg/ml and 1.8 mg/ml respectively before antibody precipitation occurred which blocked the ultra-filtration membrane. In contrast, it was possible to concentrate the solubility enhanced daughter molecules MEDI-1912_STT and CNTO607_W100A to 29 and 30 mg/ml respectively with no evidence of precipitation. No precipitation was observed for the concentrated Ang2mAb pair. Dynamic light scattering (DLS) detected higher order aggregated species for the parental antibodies MEDI-1912 and CNTO607 (Table 2), but not for the daughter molecules MEDI-1912_STT and CNTO607_W100A, as judged from the calculated polydispersity index (PDI) and the cumulant (or z-average) size. For example, PDI for the parental CNTO607 and MEDI-1912 were 0.22 and 0.15 respectively, whereas the PDI for the daughter molecules was 0.1 and 0.12 respectively indicating a more homogenous, mono-disperse state (Table 2). Similarly, the average particle size of the parental MEDI-1912 was 22 nm respectively, whereas the average particle size for the daughter molecule MEDI-1912-STT was 13 nm indicating a lower order aggregation state (Table 2). Thus significant self-interaction is occurring resulting in detectable self-interaction at lower concentrations and precipitation at higher concentrations.

(198) TABLE-US-00003 TABLE 2 IgG biophysical properties. Expression Clone T.sub.m T.sub.agg Yield C-Max Zav ID Antibody (° C.) (° C.) (mg/L) (mg/ml) (nm) PDI 1 CNTO607 - 62.6 75.6 33.6 1.8 16.2 0.22 parental 2 CNTO607 61.4 75.4 55.2 >30 14.8 0.1 (W100A) 3 MEDI- 71.5 73.2 33.2 1.4 21.5 0.15 1912- parental 4 MEDI-1912 70.8 74.8 52.8 >29 12.5 0.05 (STT) 5 Ang2mAb- 63.8 65.0 13.4 >21 13.1 0.12 parental 6 Ang2mAb 66.5 65.2 34.4 >18 12.7 0.09 C49T 14 Nivolumab 67.6 67.3 102.8 >50 nd nd The melting temperature (T.sub.m) and temperature of the onset of aggregation (T.sub.agg) were determined using Prometheus NT.4B (Nanotemper) according to the manufacturer instructions. The expression yield in terms of amount of antibody expressed (mg) per liter of culture volume was determined by transient transfection of Expi293 cells at 30 ml scale (ThermoFisher) followed by affinity purification (Protein A) and the yield of purified antibody determined from the absorbance at 280 nm and estimated antibody extinction coefficient of 1.4. Antibodies were further purification by size-exclusion chromatography on a Superdex 200 10/300 using the AKTA Pure system with PBS (pH 7.4) running buffer. Dynamic light scattering measurement were performed with a Nano S DLS (Malvern Instruments, Malvern, UK) on samples and polydispersity index (PDI) and the cumulant (or z-average) size (Zav) calculated using the zetasizer software (Malvern Instruments, Malvern, UK).

(199) This example demonstrates a very clear relationship between the mammalian cell display level of an antibody and its biophysical properties for three different antibody pairs. Parental antibodies specific to IL-13 (CNTO607), Angiopoietin2 (Ang2mAb) and Nerve growth factor (MEDI-1912), with documented problems regarding self-interaction, cross-interaction and poor pharmaco-kinetics (MEDI-1912) all resulted in lower cell display levels compared with their solubility enhanced daughter molecules (FIG. 2).

Example 3b. Supporting Theory on Antibody Concentration at the Cell Surface

(200) This Example presents underlying reasoning that may assist in understanding the inventors' proposal that strong polypeptide expression in a eukaryotic cell can potentially achieve high local concentrations when the polypeptide is retained on the cell surface.

(201) In the work described here, suspension adapted HEK293 cells are used for antibody display. The HEK293 cell line is of mammalian (human) origin. The cells are approximately spherical with a radius of 10 microns.sup.95. If we treat the suspension HEK293 cell as a sphere we can calculate the volume occupied by an antibody on its surface. (We assume for the sake of this example that all areas of the cell surface are equally accessible for antibody display. Higher local concentrations of antibody would be achieved if this were not the case.) The radius (r) of a sphere can be calculated from the formula 4/3πr.sup.3=4.18 r.sup.3. Taking an antibody to be of height 150 angstroms (Å) (=15 nm), it will be present in a larger sphere of volume 4.18(r+150 Å).sup.3. Thus the antibody volume is the difference between this and the volume of the cell.

(202) The volume of a cell of 10 micron radius is:
4.18×10.sup.−15 m.sup.3(4180×10.sup.−15 litre).

(203) The volume of the larger sphere including the antibody is:
4.198×10.sup.−15 m.sup.3(4198×10.sup.−15 litre).

(204) Thus the displayed antibody occupies a volume of
0.018×10.sup.−15 m.sup.3(18×10.sup.−15 litre).

(205) In a similar way we can calculate the difference in volume for different sized cells.

(206) Knowing the number of antibodies/cell, the molecular weight and the volume occupied we can then calculate the concentration achieved at the cell surface. Using Avogadro's constant we know that 6×10.sup.23 antibody molecules/l will have a concentration of 150,000 mg/ml. Thus 6×10.sup.18 antibody molecules/ml will have a concentration of 1.5 mg/ml (10 μM). By this approach the concentrations shown in Table 3 are calculated.

(207) TABLE-US-00004 TABLE 3 5 10 15 20 Cell radius micron micron micron micron Cell volume 522.5 4180 14107.5 33440 (×10.sup.−15 litres) volume occupied by 527.2 4198 14150 33515 cell plus antibody (×10.sup.−15 litres) volume occupied 4.7 18 .8 42.3 75 by antibody (×10.sup.−15 litres) No. Abs/litre (×10.sup.18) 212 53.2 23.64 13.3 if 10.sup.6 displayed/cell Concentration (mg/ml) 53 13.2 5.9 3.3 assuming 10.sup.6 displayed antibodies Concentration 354 88 39 22 (microM) assuming 10.sup.6 displayed antibodies Copy number 19,000 75,000 170,000 300,000 providing 1 mg/ml

(208) In these calculations the number of antibodies on the cell surface is taken to be 10.sup.6 copies/cell. Methods of experimentally determining copy number are detailed elsewhere herein and illustrated in Example 7.

(209) We see from Table 3 that display of 10.sup.6 antibodies/cell on a cell of 10 micron radius (such as a HEK293 cell in suspension culture) is estimated to give a concentration in excess of 10 mg/ml. At such concentrations, problems of protein self-interaction can potentially occur. Antibodies with a tendency to aggregate may thus have a reduced representation on the cell surface due to reduced passage through the endoplasmic reticulum.sup.96 and increased degradation. As a result, a lower level of display would be observed for an antibody prone to self-aggregation compared with a non self-interacting antibody.

Example 4. Enriching “Developability Enhanced” Anti-NGF Antibodies on the Basis of Surface Presentation Level

(210) In Example 3 the relationship between the biophysical properties of an antibody, particularly self-aggregation and their mammalian cell display levels was described. This was illustrated by taking three antibody pairs where the original parental antibody had poor biophysical properties in terms of self and cross-interaction and these properties were improved by changing selected amino acids to create daughter molecules with improved biophysical properties. In all three cases the daughter molecules with improved biophysical properties display on the surface of HEK293 cells at increased levels compared with the problematic parental antibodies. In this example we demonstrate that it is possible to enrich for antibodies with superior biophysical properties from a mixed population of clones by mammalian display. The parental anti-NGF MEDI-1912, which has self-interaction properties and poor pharmaco-kinetics in a mouse model.sup.7, and the improved daughter MEDI-1912_STT were chosen for this study. A model experiment was carried out where we created a mixed population of HEK293 cells displaying MEDI-1912 or MEDI-1912_STT by transfecting HEK293 cells with equal quantities of mammalian display plasmid (example 1) encoding the parental and modified antibodies along with plasmids encoding the TALE nuclease pair which directed the donor plasmid to the AAVS locus, as previously described (WO2015166272A2) Following drug selection fluorescence activated cell sorting (FACS) was carried out and cell were selected based on high antibody presentation level, isolation of the selected antibody genes and sequence analysis we demonstrate the selected enrichment of antibodies with improved biophysical properties from a mixed population.

(211) pINT17-MEDI-1912 and pINT17-MEDI-1912_STT (see Example 2 for description) were mixed at a 1:1 ratio and this mix was integrated into the genome of HEK293 cells using nuclease-mediated gene targeting into HEK293 cells. Mid-log-phase HEK293 suspension cells (grown to a cell density of 1×10.sup.6 cells/ml) were harvested by centrifugation at 200 g for 10 min and resuspended in MaxCyte electroporation buffer at a density of 1×10.sup.8 cells/ml. Plasmid DNA mix consisting of pINT17-MEDI-1912 (1 μg), pINT17-MEDI-1912_STT (1 μg), AAVS directed TALEN vector pair (10 μg each) was added to HEK293 cells (100 μl, 1×10.sup.7 cells total in MaxCyte electroporation buffer) and transferred into a OC100 electroporation cuvette and electroporated using a MaxCyte STX electroporation system. Following electroporation, cells were recovered at 37° C. for 20 min, diluted in HEK FreeStyle 293 expression media and maintained at 120 rpm, 37° C. under 5% CO2. Blasticidin selection was started 48 hours after transfection at a concentration of 7 μg/ml. The population was kept under selection for the duration of the experiment. 15 days post-transfection cells were stained as described in Example 2 except that DAPI stain replaced the 7-AAD stain, NGF-biotin/Streptavidin-APC stain was employed to detect antigen binding and the quantities scaled up to stain 10 million cells. The MEDI-1912/MEDI-1912_STT mixed HEK293 mammalian display population was analysed for antibody presentation level by flow-cytometry (FIG. 5a) and this revealed two main cell populations displaying different antibody levels. The two populations correlated with the monoclonal MEDI-1912 and MEDI-1912_STT antibody display levels as shown in the overlay plot (FIG. 5a). The mixed population was sorted by FACS into two populations: antibody presentation level low and high presentation level groups (Gates 5 and 6 respectively, FIG. 5b).

(212) Genomic DNA was prepared from the FACS sorted cell populations (Gates 5 and 6, FIG. 5b). DNA encoding the IgG insert was amplified by nested PCR using KOD Hot Start DNA polymerase (Merck Millipore). Outer PCR was performed with the following genome-specific primers: Forw: CCGGAACTCTGCCCTCTAAC (SEQ ID NO: 13) and Rev: TCCTGGGATACCCCGAAGAG (SEQ ID NO: 14). PCR product from the outer PCR was used as a template to amplify the integrated IgG insert with following primers; Forw: GAGGGCCTGGATCTTCTTTCTC (SEQ ID NO: 15) and Rev: GAAGTAGTCCTTGACCAGGCAG (SEQ ID NO: 16) using KOD polymerase (71086, Merck) according to manufacturer conditions. PCR products were bar-coded by following the manufacturer instructions (20015964, Illumina) and sequenced with the Illumina MiSeq sequencing platform. Approximately one million reads were analysed and this revealed that population sorted for high antibody presentation level (Gate 5, FIG. 5b) was enriched for the MEDI-1922_STT antibody (96%, FIG. 6). The population sorted for low antibody presentation was enriched for the parental MEDI-1912 antibody (85%, FIG. 6). This example demonstrates the selected enrichment of antibodies with improved biophysical properties by mammalian display, from a mixed population, by selecting clones on the basis of antibody cell surface display levels. This was exemplified by the enrichment of MEDI-1912_STT, with superior biophysical properties compared to its parental antibody MEDI-1912 from a mixed population of stable cell-lines.

(213) We demonstrate that it is possible to enrich for an antibody with superior biophysical properties in a mixed population and we show that it is possible to identify improved antibodies from a large library of variants based only on presentation levels by mammalian display. As discussed residues W30, F31 and L56 on the VH MEDI1912 have potential to form a hydrophobic patch on the surface of this antibody.sup.7. These residues were chosen for randomization and a VH library was constructed by PCR assembly mutagenesis from a synthetic DNA template (see FIG. 4 for amino acid and nucleic acid sequences and position of primers). Three PCR products were amplified from the VH template using KOD polymerase (71086-3, Merck) according the manufacturer instructions):

(214) TABLE-US-00005 a. VH1 (95 bp) amplified with primers MEDI-1912-F3 (SEQ ID NO: 17) (CCATGGCCCAGGTTCAGCTG) and MEDI1912_W30NNS_F31NNS (SEQ ID NO: 18) (CTGTCGGACCCATGTAAAGGCGCCSNNSNNAAAGGTGCCGCCGCTTGC TTTGCA). b. VH2 (102 bp) amplified with primers MEDI-1912-F (SEQ ID NO: 19) (GGCGCCTTTACATGGGTCCGACAG) and MEDI-1912_L56NNS (SEQ ID NO: 20) (CTGGAAGTTCTGGGCCAGATTGGTSNNGCCGAAGATAGGGATGATGCC GCC). c. VH3 (213 bp) amplified with primers MEDI-1912-F2 (SEQ ID NO: 21) (ACCAATCTGGCCCAGAACTTCCAG) and MEDI-1912-R (SEQ ID NO: 22) (ACTCGAGACGGTGACCATTGTG)

(215) The three PCR products (VH1, VH2 and VH3) listed above were combined (10 ng each) and assembled in a PCR reaction with outer primers MEDI-1912-F3 and MEDI-1912-R using KOD polymerase (71086-3, Merck) according the manufacturer instructions). The PCR product was digested with NcoI and XhoI and ligated with Nco1/Not 1 digested pINT17-MEDI-1912 (the pINT17 mammalian display vector (FIG. 1) encoding the VL of MEDI1912), (100 ng). This ligation mix was then was purified using the mini-Elute PCR purification kit (Qiagen) and purified ligation mix was transformed into 50 μl E. cloni 10G elite electrocompetent cells (60061-1, Lucigen). Cells were pulsed using a 0.1 cm cuvette, recovered with 2 ml recovery medium and grown for 1 h at 37° C., 250 rpm. In order to calculate the library size, cells were diluted 1 in 1000 and plated 10 μl and 100 μl in a 10 cm diameter 2TY-Kanamycin plates. The remaining cells were spun down and plated in 2×10 cm diameter 2TY-Kanamycin plates and incubated at 37° C. overnight. Colonies were counted from the 10 ul plate and a library size of 1.1×10.sup.6 was calculated. Since a library constructed by randomizing three residues using NNS codons encodes 32,768 variants, the experimental library size exceeded the theoretical library size by 34-fold. The transformant plates were scraped, the cell density measured by reading the absorbance at 600 nm (OD600), the equivalent of 2 OD units of culture (2×OD600) used to inoculate 50 ml Circlegrow culture, culture grown 3 to 4 hours at 37° C. in a 250 ml baffled flask, approximately 400×OD600 units harvested and midiprep plasmid DNA prepared (pINT17-MEDI-1912-library).

(216) The pINT17-MEDI-1912-library was used for nuclease mediated gene targeting into HEK293 cells. Mid-log-phase HEK293 suspension cells (grown to a cell density of 1×10.sup.6 cells/ml) were harvested by centrifugation at 200 g for 10 min and resuspended in MaxCyte electroporation buffer at a density of 1×10.sup.8 cells/ml. Plasmid DNA mix consisting of pINT17-MEDI-1912-library (8 μg), AAVS directed TALEN vector pair (40 μg each) was added to HEK293 cells (400 μl, 4×10.sup.8 cells total in MaxCyte electroporation buffer) and transferred into a OC400 electroporation cuvette and electroporated using a MaxCyte STX electroporation system. Following electroporation, cells were recovered at 37° C. for 20 min, diluted in HEK FreeStyle 293 expression media and maintained at 120 rpm, 37° C. under 5% CO2. Blasticidin selection was started 48 hours after transfection at a concentration of 7 μg/ml. The population was kept under selection for the duration of the experiment. 15 days post-transfection cells were analysed were stained as described in Example 3. Flow cytometry analysis of the HEK293 displayed MEDI-1912-library (FIG. 7c) indicated that the library possessed cells within the mixed population that displayed equivalent antibody display levels as the MEDI-1912_STT monoclonal cell line (FIG. 7b) and higher display levels than the parental MEDI-1912 monoclonal cell line (FIG. 7a). This suggested that clones were present in the MEDI-1912 population that were equivalent to MEDI-1912_STT in terms of display level.

(217) The library population was sorted by FACS according to antibody presentation level (FIG. 7c), antibody genes were recovered from the P5 and P6 gated populations and the VH gene sequenced by “next generation sequencing”. (NextGen) as described above. FIG. 8 shows amino acid identity histogram plots for residues 30, 31 and 56 for the mammalian display selected population. This showed an enrichment of amino acids S, T, P at position 31, enrichment of amino acids S, P, N at position 32 and an enrichment of amino acids R, S, T and P at position 56.

(218) To enable a biophysical characterization of the mammalian display selected antibodies, VH genes were PCR amplified from the genomic DNA of the selected population (Gates P5 and P6 FIG. 7), as described in Example 3. VH inserts were cloned into pINT17-MEDI-1912, NcoI and XhoI cut vector harbouring the MEDI-1912 VL, as described above and the ligation mix used to transform E. coli DH10B cells. 188 transformants were picked, plasmid DNA prepared and these were DNA sequenced to identify the identity of the codons at positions 30, 31 and 56. Selected clones, based on the frequency of occurrence by NextGen sequencing (FIG. 8), were then picked for expression by transient transfection and affinity purification as described in Example 2. Antibodies were concentrated prior to analysis by dynamic light scattering (DLS) by ultra-filtration. All the antibodies were able to be concentrated to between 8-fold and 29-fold greater than the parental MEDI-1912 antibody (Table 4), with no evidence of precipitation at these concentrations, indicating that the selected antibodies had higher solubility than the parental antibody. DLS also showed that the selected antibodies had lower average particle size (Z-Ave) and less polydispersity (PDI) than the parental antibody MEDI-1912 (Table 4). Four selected clones (P5_C06, P5_F01, P6_C08 and P6_F02) showed superior or equivalent mono-dispersity compared to the previously reported improved clone MEDI-1912_STT. The improved variants, selected by random sub-library creation and mammalian display selection, on average changed the original hydrophobic residues to hydrophilic residues.

(219) TABLE-US-00006 TABLE 4 Selected MEDI-1912 variant biophysical properties. C Z-Ave ID aa30 aa31 aa56 (mg/ml) (d.nm) PDI P5_C06 T S R 52.1 13.81 0.06 P5_C11 P P N 42.3 24.77 0.135 P5_F01 T H T 48.4 13.48 0.037 P5_F07 N T L 43.9 18.86 0.108 P5_F12 D H L 38.3 15.5 0.113 P5_G12 H S L 31.8 16.44 0.103 P6_B08 T P L 40.8 15.05 0.075 P6_C08 S T A 30.7 12.73 0.054 P6_C11 S L L 15.2 31.97 0.207 P6_E07 R P L 33.9 19.54 0.177 P6_F02 R S Y 39.1 12.96 0.036 MEDI-1912_STT S T T 53.1 14 0.048 MEDI-1912 W F L 1.8 23.2 0.148 Antibodies were expressed by transient transfection of Expi293 cells at 30 ml scale (ThermoFisher) followed by affinity purification (Protein A). Antibodies were further purification by size-exclusion chromatography on a Superdex 200 10/300 using the AKTA Pure system with PBS (pH 7.4) running buffer. Antibodies were concentrated by centrifugal filtration and the concentration obtained are shown (C) in milligrams per ml (mg/ml). Dynamic light scattering measurement were performed with a Nano S DLS (Malvern Instruments, Malvern, UK) on samples and polydispersity index (PDI) and the cumulant (or z-average) size (Zav) calculated using the zetasizer software (Malvern Instruments, Malvern, UK). Amino acid identity is shown in single letter code for positions 30, 31 and 32.

(220) This example demonstrates that it is possible transform an antibody with poor biophysical properties to one with improved properties in terms of solubility and low self-interaction by mammalian display selection. This was achieved by the random mutagenesis of selected residues and the creation of a large random antibody variant library displayed on the surface of HEK293 cells. Current state of the art techniques to assess an antibody developability profile (e.g. solubility) require large scale expression and purification at the multi-mg scale to enable complete biophysical and PK measurements. In this example using differences in polypeptide presentation level, as judged by differences in mean fluorescence intensity (MFI) we demonstrate that it is possible to create millions of variants and select for antibodies which are subsequently shown to have improved biophysical properties. This process could be applied where novel antibodies are being selected from a naïve library or a library pre-selected in another system such as phage display. Alternatively, the present invention could also be applied during the affinity maturation or humanization of an antibody where a library of variants is created and displayed on the surface of mammalian cells.

Example 5. Construction of Variant Library and Selection of Developability Enhanced Anti-PSK9 Clones

(221) Bococizumab is an anti-proprotein convertase substilisin/kexin type 9 (PCSK9) mAb that was in development by Pfizer to reduce low-density lipoprotein cholesterol (LDL-C) in serum. The mechanism of action of Bococizumab is to inhibit the PCSK9 mediated degradation of LDL receptor (LDLR) and thereby decrease serum LDL-cholesterol (LDL-C).sup.97. This antibody was withdrawn from development in November 2016 with Pfizer announcing “it was not likely to provide value for patients, physicians or shareholders”. It has been reported that the biophysical properties of bococizumab are not optimal.sup.2 and this may be a reason for its clinical failure. For example, Bococizumab displayed both self-aggregation and cross-interaction in a variety of assays.sup.2. In contrast, the FDA approved anti-PCSK9 Alirocumab (Regeneron) antibody did not display the same levels of self-aggregation and cross-interaction in the same assays.

(222) Bococizumab was originally discovered by immunization of PCSK9 knockout mice and screening monoclonal antibodies (mAbs) producing hybridoma clones for their ability to inhibit PCSK9 activity.sup.98. The mouse mAb 5A10 (U.S. Pat. No. 8,399,646 B2) was then humanized by cloning DNA encoding the complementarity determining regions (CDRs) from the variable heavy (VH) and variable light (VL) domains into a human framework.sup.99 with an amino acid substitution in VH CDR1 and VH CDR2 to give the humanized mAb 5A10-i. This humanized antibody 5A10-i was further affinity matured, as described previously.sup.100, to give Bococizimab. A sequence alignment for the VH and VL domains for the parental mouse mAb 5A10, the humanized intermediate antibody 5A10-i and Bococizumab is shown in FIG. 9. The affinities (equilibrium dissociation constants or K.sub.D) of 5A10, 5A10-i and Bococizimab for PCSK9 are 1 nM, 1.5 nM and 7 pM respectively as determined by surface plasmon resonance (SPR) or KinExA (U.S. Pat. No. 8,399,646 B2). The crystal structure of Bococizimab Fab fragment complexed with PCSK9 has been determined.sup.98 and this has shown the antibody binds to the catalytic domain of PSK9 through both light and heavy chains, with the main contribution through VH CDR3.

(223) After nuclease-mediated antibody gene integration into HEK293 cells and display on the cell surface we have observed reduced cell surface presentation of Bococizumab, compared with the humanized intermediate version 5A10-i from which it was derived (FIG. 10). The aim of this example is to demonstrate that, from a library of variants, a variant of Bococizumab can be selected by mammalian display with good presentation level indicating improved biophysical properties of stability, reduced self-aggregation and reduced cross-interaction properties or “stickiness” with retained target antigen binding. It is important to first identify regions or “patches” of the antibody which may contribute to its poor biophysical properties. For example, it is known that contiguous hydrophobic amino-acid residues within a polypeptide sequence can give rise to poor expression levels of that protein.sup.101. Also hydrophobic patches on antibodies can give rise to poor biophysical properties..sup.6,7,14 Similarly, positive charge patches on the antibody surface from clustered lysine or arginine residues can also give rise to cross-interaction by non-specific binding to the neonatal Fc receptor (FcRn).sup.22 or cell expressed negatively charged molecules such as heparin sulphate.sup.23.

(224) The process of creating an improved binder using the present invention begins with the identification of residues within a sequence as candidates for changing within a library. These positions can act as sites for randomization using more than one alternative amino acid or could be sites for substitution with a single amino acid. Mutagenesis may be carried out using approaches known to those skilled in the art, such as oligonucleotide-directed mutagenesis (.sup.102Molecular Cloning: a Laboratory Manual: 3rd edition, Russell et al., 2001, Cold Spring Harbor Laboratory Press, and references therein). In this case of Bococizumab the three dimensional structure was available, this was analysed and candidate amino acid residues were identified for mutagenesis. Structural modelling may be used as an alternative to help identify target amino acids for mutagenesis.

(225) The facility to create and screen millions of variants within the present invention means that a thorough search of sequence variants can be conducted even in the absence of any 3D structural information or model e.g. by looking at linear sequences. This could be done by analysing linear sequence for features such as hydrophobicity or charge clustering. As an alternative, mutational scans focused on individual amino acids can be carried out in order to guide larger scale, combinatorial mutagenic campaigns during affinity maturation campaigns.sup.103. By the same approach individual amino acids may be substituted with alternative sets of amino acids to identify individual residues with potential for improving biophysical properties. These may subsequently form the basis for combinatorial mutagenesis wherein multiple positions are changed simultaneously. In the case of antibody genes an alignment with germ-line sequences may help identify optimal amino acid changes for improving expression. This approach was taken for non-paratopic amino-acid residues on the VH. Residues that contributed to hydrophobic or charge patches were Y33 within VH CDR1, F54 and R57 within VH CDR2 (FIG. 9). A multiple alignment with human VH germ-line sequences is shown in FIG. 11. Based on this alignment the non-paratopic bococizumab VH amino acids contributing to hydrophobic or charge patches were reverted to germ-line sequences listed below: a. VH Y33A (reversion to germline IGHV1-3*01) b. VH Y33D (reversion to germline IGHV1-8*01) c. VH S52N, F54S, R57S (triple mutant reversion to germline IGHV1-46*01) d. VH Y33A, S52N, F54S, R57S (a. and c. mutants combined) e. VH Y33D, S52N, F54S, R57S (b. and c. mutants combined)

(226) For paratopic residues, random mutant libraries were created to enable selection for both presentation and retained antigen binding (FIG. 9). Candidate problematic residues within the VL were: Y53, L94 and W95. From the co-crystal structure of Bococizumab with PCSK9 these residues either directly interact with the target antigen or indirectly contribute to binding through allosteric interactions (for example VL CDR3 residue W95 appears to pack against VH CDR3 residues and may maintain the VH CDR3 conformation for optimal binding to PCSK9. By the construction of random libraries in these positions and selection it will be possible to explore whether there is an optimal amino acid combination for improved biophysical properties with retained antigen binding. An alternative library design could have involved the random mutagenesis of selected non-paratopic residues within the VH CDR1 and CDR2.

(227) Synthetic VH geneblocks were designed and synthesised encoding the following constructs listed (a) to (e) above and the original wild-type Bococizumab (f). The DNA sequences encoding these synthetic genes are shown FIG. 12. These geneblocks were PCR amplified with primers 3054 and 3055 (Table 5) to yield a 375 bp product. This product was spin column purified, digested with NcoI/XhoI and spin column purified. The 6 digested VH inserts were then ligated with pINT17-blasticidin NcoI/XhoI cut vector, ligations used to transform E. coli DH5α, individual colonies picked, mini-prep plasmid DNA prepared and DNA sequence confirmed.

(228) TABLE-US-00007 TABLE 5 Primer sequences. 3052 TTTTTTGCCATGGCCCAAGTG (SEQ ID NO: 23) 7A2-VH-F 3053 AAAAAAACTCGAGACGGTGACC (SEQ ID NO: 24) 7A2, 107_A07-VH-R 3054 TTTTTTGCCATGGCCCAGG (SEQ ID NO: 25) Bococizumab-VH-F 3055 AAAAAAACTCGAGACTGTCACGG (SEQ ID NO: 26) Bococizumab-VH, 7D4- intermediate-R 3056 TTTTTTGCTAGCGACATCCAGATG (SEQ ID NO: 27) Bococizumab, 7D4- intermediate-VL-F 3057 TTTTTTGCCATGGCCCAGGTTC (SEQ ID NO: 28) Bococizumab-VL-R 3052 TTTTTTGCCATGGCCCAAGTG (SEQ ID NO: 29) 7A2-VH-F 3053 AAAAAAACTCGAGACGGTGACC (SEQ ID NO: 30) 7A2, 107_A07-VH-R 3069 CTGGGCACGCCGGTGTATCTSNNGCTGGCGCTGTAGATCAGC Bococizumab-VL-R-Y53- AG (SEQ ID NO: 31) random 3070 GTGCCCTGGCCAAATGTCCGSNNSNNAGAGTACCGCTGCTGG Bococizumab-VL-R- CAGTAG (SEQ ID NO: 32) L94W95-random 3071 TTTTTTGCTAGCGACATCCAGATG (SEQ ID NO: 33) Bococizumab-VL-F1 3072 GCTGGCGCTGTAGATCAGCAG (SEQ ID NO: 34) Bococizumab-VL-R1 3073 AGATACACCGGCGTGCCCAG (SEQ ID NO: 35) Bococizumab-VL-F2 3074 AGAGTACCGCTGCTGGCAGTAG (SEQ ID NO: 36) Bococizumab-VL-R2 3075 AAAAAAGCGGCCGCGGTACGCTTGATTTCCAGCTTGGTGCCC Bococizumab-VL-R3- TGGCCAAATGTCCG (SEQ ID NO: 37) extension 3076 TTTTTTGCCATGGCCCAGGTTCAG (SEQ ID NO: 38) Bococizumab-VH-F1 3077 AAAAAAACTCGAGACTGTCACGGTGG (SEQ ID NO: 39) Bococizumab-VH-R1

(229) The VL Y53, L94, W95 codons was randomized by NNS PCR assembly mutagenesis using a VL gene template containing stop codons at the positions subject to mutagenesis (see FIG. 12 for VL gene template sequence). The following PCRs were performed: a) The Bococizumab VL plus stops geneblock (see FIG. 12 for sequence) was PCR amplified with primers 3071/3047 (Table 5) to give a 353 bp product. b) A PCR was performed with template (a) above with primers 3071/3069 (Table 5) to give 191 bp product. c) A PCR was performed with template (a) above with primers 3073/3070 to give a 146 bp product. d) A PCR assembly reaction was performed with the products of PCR reactions b and c above with outer primers 3071/3075 to give a 353 bp insert. The product was digested with NheI/NotI and purified by spin column.

(230) The 6 VH variants a to f above (see FIG. 12) were PCR assembled with a “stuffer” fragment encoding the constant kappa light chain (CL-kappa), polyA, CMV promoter and signal sequence and the VL NNS library. The stuffer fragment was amplified from pINT3 plasmid (WO2015166272A2) using primers Kappa stuffer F4 (GTACCGCGGCCGCACCTTCCG (SEQ ID NO: 40)) and Lambda stuffer R3 (CAGCCATGGCGCCTGTGGAGAGAAAGG (SEQ ID NO: 41)). The assembled inserts were digested with NheI and XhoI, spin column purified and ligated (50 ng insert per ligation reaction) with pINT17-BSD targeting vector (100 ng), pre-digested with NheI and XhoI. Ligation mixture (20 μl) was purified using the mini-Elute PCR purification kit (Qiagen) and purified ligation mix (4 μl) was transformed into E. cloni 10G elite electrocompetent cells (50 μl, 600512, Lucigen). Cells were pulsed using a 0.1 cm cuvette, recovered with 2 ml recovery medium and grown for 1 h at 37° C., 250 rpm. In order to calculate the library size, cells were diluted 1 in 1000 and plated (10 μl and 100 μl) in a 10 cm diameter 2TY-Kanamycin plates. The remaining cells were spun down and plated in 2×10 cm diameter 2TY-Kanamycin plates and incubated at 37° C. overnight. Colonies were counted from the 10 μl plate and library size was calculated to be 2×10.sup.6 transformants. In order to represent every variant the required library size 1.2×10.sup.5 clones. (32.sup.3=3.4×10.sup.4 per library×6 VH mutants) so the library generated represented a 16-fold over-representation of the required library diversity. The transformant plates were scraped, OD600 measured, 2 OD600 used to inoculate 50 ml circlegrow culture, culture grown 3 to 4 hours at 37° C. in a 250 ml baffled flask, approximately 400 OD600 units harvested and 6 midiprep plasmid DNA prepared representing the 6 VH Bococizumab variants combined with the three NNS codon VL library (FIG. 12). The 6 midiprep plasmid DNAs were quantitated by reading the absorbance at 260 nm and mixed at an equimolar ratio to give the Bococizimab targeting vector library pINT17-BSD-Boco1-library.

(231) Suspension adapted HEK293 cells were seeded at 2.5×10.sup.5 cells per ml in HEK FreeStyle 293 expression media two days before transfection. On the day of transfection cells were centrifuged and re-suspended in a final volume of 10.sup.8 cells/ml in the manufacturer's electroporation buffer (1 ml, Maxcyte Electroporation buffer, Thermo Fisher Scientific Cat. No. NC0856428) containing pINT17-BSD-Bococizumab-library (20 μg) and plasmids encoding the AAVS left and right TALE nucleases (TALENs, 100 μg each). The HEK293/plasmid DNA mix (0.4 ml) was transferred to a single OC-400 Cuvette (MaxCyte, Cat. No. OC-400R10) and pulsed on the HEK293 setting with the MaxCyte STXG2. The controls (minus TALENs and pINT17-BSD-Bococizumab and pINT17-BSD-5A10-i) were transfected using OC-100 Cuvettes (MaxCyte, Cat. No. OC-100R10) on the same setting. After transferring the electroporated cells into an Erlenmeyer flask (250 ml) the cells were allowed to rest for 30 minutes before FreeStyle 293 Expression Media (40 ml, LifeTech. Cat. No. 12338018) was added. The cells were resuspended thoroughly and placed in a orbital shaking incubator set to 130 RPM, 37° C. and 5% CO.sub.2.

(232) After 24 hours, 1×10.sup.6 cells were stained with Anti-Human Fc PE (Cambridge Bioscience, Cat. No. 409304) to confirm transient expression. Briefly, the cells were centrifuged at 600×g for 2.5 minutes. The supernatant was discarded and cells re-suspended in 0.1% BSA (Diluted from 7.5% solution: LifeTech, Cat. No. 15260037). These were centrifuged again and resuspended in 100 μl of 1% BSA, PBS with 1 μl of anti-human Fc PE added. These were incubated for 30 minutes in the dark at 4° C. The cells were washed twice with 1 ml of 0.1% BSA, PBS and resuspended in 0.5 ml of 0.1% BSA, PBS containing 5 μl of 7-AAD (eBioscience, Cat. No. 00-6993-50). 50 μl was removed and added to wells of a 96 well plate. The IntelliCyt flow cytometer was used to analyse the presentation levels and this showed transient cell surface antibody for the transfection. As transient expression was observed the cultures were taken forward for selection using the antibiotic Blasticidin S HCl (LifeTech, Cat. No. R21001) at a concentration of 7.5 μg/ml. Cells were seeded at 0.25×10.sup.6 cells per ml in Erlenmeyer flasks. Cells were also plated into 10 cm dishes (Corning, Cat. No. 353003) in DMEM (LifeTech, Cat. No. 41965039) with 10% FBS (Sigma Aldrich, Cat. No. F9665-500ML) and 1% Penicillin Streptomycin (Sigma Aldrich, Cat. No. P0781-100ML) at either 10,000 cells, or 1000 cells in 10 ml. These were allowed to attach for 24 hours before Blasticidin S HCl (LifeTech, Cat. No. R21001) was added at 7.5 μg/ml. After 12 days of transfection the plates were stained with 2% methylene blue. The percentage transfection efficiency was calculated by counting the number of blasticidin colonies achieved for a given input of total cells. The integration efficiency was calculated to be 2% and the library size achieved for a 40 million cell transfection was 800,000, 7-fold greater than the required theoretical library size (120,000). Thus a mammalian display library was constructed encoding all possible combination of variants.

(233) After 5 days of Blasticidin S HCl (LifeTech, Cat. No. R21001) selection the cells were enriched using MACS beads and columns (Miltenyi, Cat. No. 130-048-801 & Cat. No. 130-042-401). The library had been expanded to over 200 million cells. 100 million cells were centrifuged at 200×g and washed in 0.1% BSA-PBS. These were resuspended in 9.9 ml of 1% BSA-PBS and 100 μl of anti-human Fc PE antibody (Cambridge Bioscience, Cat. No. 409304) added. The remaining 100 million were also spun down, washed, and incubated with biotinylated PCSK9 antigen (10 nM, PC9-H82E7, AcroBiosystems, 10 ml, diluted in 1% BSA-PBS). Both were incubated in the dark at 4° C. for 30 minutes. From this point, autoMACS Rinsing Solution (Miltenyi, Cat. No. 130-091-221) was used. The cells were washed in autoMACS Rinsing Solution (10 ml, 1×PBS+2 mM EDTA+0.5% BSA), centrifuged at 200×g and resuspended in 800 μl of autoMACS Rinsing Solution. 200 μl of either anti-PE (Miltenyi, Cat. No. 130-048-801) microbeads or Streptavidin (Miltenyi, Cat. No. 130-048-101) microbeads were added. These were incubated for 10 minutes in the dark at 4° C. before washing in 10 ml of autoMACS Rinsing Solution and resuspended in 5 ml ready to be applied to the columns. The MACS LS Columns (Miltenyi, Cat. No. 130-042-401) were pre-washed with 3 ml of autoMACS Rinsing Solution before the cells were added. 4× columns were used for each set of cells. Once ¼ (roughly 1.25 ml) of cells were added to each column, the columns were washed 3× times with 3 ml of buffer. The LS columns were removed from the magnetic holder and 5 ml of buffer was added. This was pushed through the column using the plunger into a 15 ml Falcon tube to elute the bound cells. To further purify the population, this 5 ml was added to a fresh column (pre-washed as before) and processed as before. The cells were counted and were found to be roughly 1.5×10.sup.6 cells/ml in 5 ml. These were spun down and resuspended in 30 ml of FreeStyle media FreeStyle 293 Expression Media (LifeTech. Cat. No. 12338018) with blasticidin at 7.5 μg/ml) and incubated at 37° C., 5% CO2 until ready to passage.

(234) 48 hours post-MACS the cells were stained for Fc presentation and antigen binding (FIG. 13). Thus 2 populations were created based on a first round selection on either antigen or Fc expression (selection numbers 884 and 885 respectively). These were subsequently selected on a combination of antigen and Fc expression by flow cytometry. The procedure was the same as for the 24-hour stain described previously, with the following adjustment: the cells were incubated with 10 nM Biotinylated Human PCSK9, Avi-Tag (Cat. No. PC9-H82E7-25 μg, ACRObiosystems) for 30 minutes at 4° C. before washing and incubation with a mix of Anti-Human Fc PE (1 μl per 1×10.sup.6 cells) (Cambridge Bioscience, Cat. No. 409304) and Anti-strep APC (Invitrogen, Cat. No. SA1005) (0.5 μl per 1×10.sup.6 cells). 7-AAD (eBioscience, Cat. No. 00-6993-50) was used to assess viability as before. The cells were washed and analyzed using the IntelliCyt instrument as described previously (FIG. 10). After 14 days of selection, FACS was carried out using the BD Influx. 20×10.sup.6 cells of the MACS sorted populations sorted on Antigen binding or Fc Presentation were incubated (as previously) with the 10 nM Biotinylated Human PCSK9 (PC9-H82E7-25 μg, ACRObiosystems) before washing and incubation with a mix of anti-Human Fc PE (1 μl per 1×10.sup.6 cells) (Cambridge Bioscience, Cat. No. 409304) and anti-strep APC (0.5 μl per 1×10.sup.6, SA1005, Invitrogen). DAPI was added (1 μl/million cells) immediately before sorting. The cells were sorted into two further populations: a higher antigen binding population (gate P5) and a lower antigen binding population (gate P6) as shown in FIG. 14. These FACS purified populations were grown without blasticidin but with 1% penicillin streptomycin to avoid contamination from the cell sorting process. After 4 days in culture 1×10.sup.6 cells for each population were taken for genomic DNA extraction.

(235) DNA encoding the IgG was amplified by nested PCR using KOD Hot Start DNA polymerase (Merck Millipore) as described in Example 4. PCR products were gel purified and digested with NheI and XhoI, cloned into the pINT3 mammalian expression vector and used to transform E. coli DH10B cells.

(236) Random un-selected input clones (84), antigen sorted (75) and Fc selected (85) were sequenced and the VH identity determined. From the set of sequenced clones derived from a cycle of mammalian display selection, none had the original Bococizumab VH gene and there was a strong bias towards the variant composed of the IGHV1-46*01 germ-line (see FIG. 15). The VL sequences were determined for the same clone sets. The average pI and aliphatic index was calculated for the 3 mutated codons. This showed a reduction in both pI and aliphatic index for the mammalian display selected antibodies (FIG. 16), indicating a switch away from the original hydrophobic amino acids in CDR2 and 3.

(237) To show that the mammalian display selected Bococizumab variants had superior biophysical properties compared with the parental antibodies, selected antibodies were next expressed. Clones were picked into 96-well plates per selection (91 per population): MACS on PCSK9 (named selection 884) or MACS on anti-Fc (named selection 885). These colonies were used to prepare two plates of DNA for transfection using the Qiagen Plasmid Plus 96 Miniprep Kit (Qiagen, Cat. No. 16181) following the manufacturers instructions. This DNA was used to transfect two 96 well plates of Expi293 cells using the Expi293 transfection system (LifeTech, Cat. No. A14525) following the manufacturers instructions. After 5 days these were harvested and the supernatants kept at 4° C. To determine the propensity of the antibodies to aggregate a method called AC-SINS (Affinity-capture self interaction nanoparticle spectroscopy) was used. The method used was essentially as described by Liu et al., 2014.sup.39 with the following modifications. Once the gold nanoparticles (AuNP, citrate-stabilized 20 nm gold nanoparticles, 15705, Ted Pella Inc.) were blocked with PEG Thiol, the AuNP were stored until needed (up to one week) at 4° C. Rather than using a syringe filter to concentrate to 10×, the AuNP were centrifuged at 15,000 RPM for 10 minutes at 4° C. with 95% of the supernatant removed and further centrifuged at the same conditions. The final AuNP were resuspended in 1/10th of the starting volume. 10 μl was added to each well of a polypropylene 96 well plate (containing 100 μl of test antibody, either in supernatant or purified in PBS. The plate was incubated for 2 hours at room temperature on a shaking platform set to 700 RPM. As stated in Liu et al. (2014).sup.39 the contents were carefully transferred to a polystyrene UV transparent plate. Absorbance data are collected from 450 to 650 nm at in increment of 2 nm using a BMG Pherastar instrument. The wavelength of maximum absorbance is identified and 10 points either side are averaged with the points directly before and after to reduce error from noise. The highest point from these averages is taken as the maximum absorbance. FIG. 17 lists the results of the AC-SINS assay together with the antibody CDR sequences. The majority (86/91) of selected variant clones displayed AC-SINS wavelength shifts of 12 nm or less, equivalent to the humanized intermediate clone 5A10-i. This wavelength indicates that self-association is not occurring to any great extent in these samples. In contrast Bococizumab gave a wavelength shift of 26 nm (FIG. 17) by AC-SINS and only 5 clones in the selected set resulted in a wavelength shift of greater than 20 nm. Therefore, Bococizumab variant clones have been selected by mammalian cell display with a lower propensity to self-aggregate than the original parental clone as judged by the AC-SINS assay.

(238) The supernatants from the expressed plate were also used to compare ability of the antibodies to retain binding to PCSK9. This was performed in a capture ELISA assay with monomeric antigen, which has been shown to be an effective way to affinity rank antibodies for binding to their target. Briefly, 96-well Maxisorp plates (Nunc, Cat. No. 437111) were coated with anti-human Fc antibody (Jackson ImmunoResearch, Cat. No. 209-005-098) at 3 μg/ml in PBS overnight at 4° C. The following day the plates were washed 3 times with 1×PBS and subsequently blocked with 300 μl of 3% (w/v) dried milk (Marvel) in 1×PBS (M-PBS) for 1 hour at room temperature. These were washed 3 times with 1×PBS, and 30 μl of 6% (w/v) dried milk (Marvel) added to each well. 30 μl of each supernatant was added and incubated for 1 hour at room temperature. The plates were then washed 3 times with 1×PBS-Tween (0.1%) and then 3 times with 1×PBS. 60 μl/well of 0.1 nM Biotinylated Human PCSK9, Avi-Tag (ACRObiosystems, Cat. No. PC9-H82E7) was added to each well and the plates incubated for 1 hour at room temperature. The plates were washed as previously with 1×PBS-Tween followed by 1×PBS. 60 μl/well of Streptavidin-Europium (Perkin Elmer, Cat. No. 1244-360) in DELFIA assay buffer (Perkin Elmer, Cat. No. 1244-111) (1 in 500 dilution) was added and incubated for 1 hour at room temperature. The plates were washed a final time with 1×PBS-Tween and 1×PBS before addition of 50 μl/well of DELFIA enhancement solution (Perkin Elmer, Cat No. 4001-0010). The plates were placed on a plate shaker for 5 minutes at 300 RPM and read on a BMG Labtech PHERAStar Plate reader (Excitation 340 nm, Emission 615 nm). This showed that the majority of antibodies retained binding for PCSK9 with several displaying a capture ELISA signal equivalent to Bococizumab (K.sub.D=7 pM) and the 5A10-i (K.sub.D=1.5 nM) intermediate clone (FIG. 17). This shows that by library creation and mammalian display selection it is possible to target antibody paratope residues and simultaneously select for antibodies with improved biophysical properties and the retained ability to bind to target antigen.

(239) Clones were then selected, based on the AC-SINS culture supernatant score of a low AC-SINS wavelength shift and retention of antigen binding (FIG. 17). These clones, together with Bococizumab, 5A10-i and Alirocumab (an approved anti-PCSK9 antibody) were then expressed by transient transfection of Expi-293 cells (50 ml scale) and purified by Protein A affinity chromatography, followed by dialysis, as described in Example 3. The antibodies were then analysed by HPLC-SEC and this showed that all the selected antibodies displayed equivalent HPLC retention times and peak widths at the control Alirocumab antibody and 5A10-i (FIG. 18). In contrast, Bococizumab was retarded on the column and displayed a longer retention time. Also Bococizumab showed a non-symmetrical peak also indicating that it possesses cross-interaction properties and was non-specifically binding to the column matrix. The purified antibodies were also analysed by AC-SINS and this showed wavelength shifts equivalent to Alirocumab and 5A10-i (Δλ=8 to 12 nm), whereas Bococizumab displayed a longer AC-SINS wavelength shift indicating that it possessed self-interaction properties (Δλ=39 nm). The expression yields, AC-SINS wavelength shifts, HPLC-SEC retention times and HPLC-SEC peak widths are summarised in Table 6.

(240) This example has therefore exemplified that it is possible to use binder display on higher eukaryotic cells to select variants with an improved developability profile including reduced self-interaction and with reduced non-specific interactions while retaining binding to the target. In this example, this was achieved by first identifying hydrophobic and positive charge patches on the surface of an antibody, random or targeted mutagenesis to create a variant library and the use of nuclease mediated binder gene targeting to enable a single gene copy per cell. The cell display library was then sorted on the basis of cell display level and antigen binding to identify variants of the parental antibody with improved biophysical properties.

(241) TABLE-US-00008 TABLE 6 Comparison of biophysical properties of Bococizumab and improved variants (including wavelength shift (nm) in an AC-SINS assay, HPLC-SEC retention time, peak width and expression yield). HPLC-SEC Peak Expression Test Clone or AC-SINS Retention Width in Expi293 Control Δλ (nM) Time (min) (min) (mg/L) 884_01_G01 9 6.84 0.45 15 884_01_A01 10 6.87 0.42 14 884_01_A04 9 6.86 0.41 16 884_01_F02 12 6.91 0.42 16 884_01_E12 9 6.97 0.65 16 Bococizumab 39 7.35 1.14 11 5A10-i 10 6.91 0.56 33 Alirocumab 8 6.87 0.26 14

Example 6a. Developability Enhancement by Selection for Non-Cross-Interacting Clones

(242) Antibodies which possess the property of non-specific binding to molecules other than their target tend have poor half-life in vivo, can give rise to “off-target” binding resulting in poor pharmaco-kinetics (PK) and pharmacodynamics (PD). In addition, the properties of cross-interaction or “stickiness” can give rise to problems during the manufacture of the antibodies leading, for example, to retardation to a column matrix during purification or formulation problems.

(243) This example demonstrates that it is possible to use antibody mammalian display to differentiate between an antibody with known “stickiness” or cross-interaction problems from an antibody that is well-behaved and has been approved for clinical use. The anti-neuropilin-1 antibody Vesencumab (or MNRP1685A) was chosen as an example of a “sticky” antibody. This antibody is known to be retarded during size exclusion chromatography and non-specifically binds to the column matrix. This is thought to contribute to its poor half-life in animal models.sup.104. In addition, the clinical development of this antibody was halted after the observation of the side-effect of proteinuria.sup.105. The anti-PD1 antibody Nivolumab was chosen as an example of a well-behaved antibody that has been approved for clinical use.sup.106. Vesencumab also displayed some self-interaction in an affinity-capture self-interaction nanoparticle spectroscopy (AC-SINS) assay.sup.39.

(244) TABLE-US-00009 TABLE 7 Protein sequences of Vesencumab and Nivolumab heavy and light chains. The amino acid sequences in single letter code are shown of the complete antibody heavy and light chains. The variable domains are underlined. Chain Protein Sequence Vesen- EVQLVESGGG LVQPGGSLRL SCAASGFTFS SYAMSWVRQA cumab PGKGLEWVSQ ISPAGGYTNY ADSVKGRFTI SADTSKNTAY (heavy LQMNSLRAED TAVYYCARGE LPYYRMSKVM DVWGQGTLVT chain) VSSASTKGPS VFPLAPSSKS TSGGTAALGC LVKDYFPEPV TVSWNSGALT SGVHTFPAVL QSSGLYSLSS VVTVPSSSLG TQTYICNVNH KPSNTKVDKK VEPKSCDKTH TCPPCPAPEL LGGPSVFLFP PKPKDTLMIS RTPEVTCVVV DVSHEDPEVK FNWYVDGVEV HNAKTKPREE QYNSTYRVVS VLTVLHQDWL NGKEYKCKVS NKALPAPIEK TISKAKGQPR EPQVYTLPPS REEMTKNQVS LTCLVKGFYP SDIAVEWESN GQPENNYKTT PPVLDSDGSF FLYSKLTVDK SRWQQGNVFS CSVMHEALHN HYTQKSLSLS PGK (SEQ ID NO: 42) Vesen- DIQMTQSPSS LSASVGDRVT ITCRASQYFS SYLAWYQQKP cumab GKAPKLLIYG ASSRASGVPS RFSGSGSGTD FTLTISSLQP (light EDFATYYCQQ YLGSPPTFGQ GTKVEIKRTV AAPSVFIFPP chain) SDEQLKSGTA SVVCLLNNFY PREAKVQWKV DNALQSGNSQ ESVTEQDSKD STYSLSSTLT LSKADYEKHK VYACEVTHQG LSSPVTKSFN RGEC (SEQ ID NO: 43) Nivo- QVQLVESGGG VVQPGRSLRL DCKASGITFS NSGMHWVRQA lumab PGKGLEWVAV IWYDGSKRYY ADSVKGRFTI SRDNSKNTLF (heavy LQMNSLRAED TAVYYCATND DYWGQGTLVT VSSASTKGPS chain) VFPLAPCSRS TSESTAALGC LVKDYFPEPV TVSWNSGALT SGVHTFPAVL QSSGLYSLSS VVTVPSSSLG TKTYTCNVDH KPSNTKVDKR VESKYGPPCP PCPAPEFLGG PSVFLFPPKP KDTLMISRTP EVTCVVVDVS QEDPEVQFNW YVDGVEVHNA KTKPREEQFN STYRVVSVLT VLHQDWLNGK EYKCKVSNKG LPSSIEKTISK AKGQPREPQV YTLPPSQEEM TKNQVSLTC LVKGFYPSDI AVEWESNGQP ENNYKTTPPV LDSDGSFFLY SRLTVDKSRW QEGNVFSCSV MHEALHNHYT QKSLSLSLGK (SEQ ID NO: 44) Nivo- EIVLTQSPAT LSLSPGERAT LSCRASQSVS SYLAWYQQKP lumab GQAPRLLIYD ASNRATGIPA RFSGSGSGTD FTLTISSLEP (light EDFAVYYCQQ SSNWPRTFGQ GTKVEIKRTV AAPSVFIFPP  chain) SDEQLKSGTA SVVCLLNNFY PREAKVQWKV DNALQSGNSQ ESVTEQDSKD STYSLSSTLT LSKADYEKHK VYACEVTHQG LSSPVTKSFNRGEC (SEQ ID NO: 45)

(245) Synthetic DNA encoding the Vesencumab and Nivolumab heavy and light variable domains (see Table 7 for sequences) were cloned into a dual promoter IgG soluble expression vector based on pINT3 (WO2015166272A2) and DNA sequence confirmed. To examine the properties of the soluble antibody, plasmid DNA was prepared of pINT3-Vesencumab and pINT3-Nivolumab and this was used to transfect Expi293 cells (30 ml final culture volume scale) using the transfection reagent ExpiFectamine according to the manufacturer instructions (A14525, ThermoFisher Scientific). Cells were seeded at a density of 2×10.sup.6 cells/ml in 25.5 ml of Expi293 Expression Medium 24 hours prior to transfection. Plasmid DNA (30 μg) was diluted in Opti-MEM Medium (1.5 ml) and ExpiFectamine 293 Reagent (80 μl) was diluted in Opti-MEM Medium (1.5 ml) and incubated for 5 minutes at room temperature. The diluted plasmid DNA (30 μg in 1.5 ml Opti-MEM Medium) was then added to the diluted ExpiFectamine 293 Reagent (80 μl ExpiFectamine in 1.5 ml Opti-MEM Medium) and incubated for 20 minutes at room temperature. The cells were incubated at 37° C., 5% CO2, 5% humidity and agitated at 200 rpm (50 mm orbital throw, ISF1-X, Climo-Shaker, Kuhner). Following 5 days of expression, culture supernatant was harvested by centrifugation (2000 g, 20 min) and purified by Protein A affinity chromatography as described above (Example 3). The antibody yield and concentration was determined by measurement of the absorbance at 280 nm and calculating using the Beer-Lambert Law using an estimated extinction coefficient of 1.4 to approximate the concentration. The expression yields of Vesencumab and Nivolumab were similar (95 mg/L and 103 mg/L respectively). Again the expression yield achieved in transient transfection efficiency is no guide to impending problems when the concentration of antibodies are increased.

(246) The biophysical properties of Nivolumab and Vesencumab were determined by several techniques. The melting temperature (T.sub.m) and temperature of the onset of aggregation (T.sub.agg) were determined using Prometheus NT.4B (Nanotemper) as described above (Example 3). The melting temperature (T.sub.m) and the aggregation temperature (T.sub.agg) of the two antibodies were similar (see Table 8a). This again demonstrates that melting temperature is not predictive of impending problems.

(247) TABLE-US-00010 TABLE 8a Vesencumab and Nivolumab IgG biophysical properties. Expression Poly- T.sub.m T.sub.agg Yield Ve Zav MW.sub.est dispersity Antibody (° C.) (° C.) (mg/L) (ml) PDI (nm) (KDa) (%) Vesencumab 69.9 70.4 94.8 13.7 0.163 7.7 401 66 Nivolumab 67.6 67.3 103 12.0 0.053 5.1 156  0 The melting temperature (Tm) and temperature of the onset of aggregation (Tagg) were determined using Prometheus NT.4B (Nanotemper) according to the manufacturer instructions. The expression yield in terms of amount of antibody expressed (mg) per liter of culture volume was determined by transient transfection of Expi293 cells at 30 ml scale (ThermoFisher) followed by affinity purification (Protein A) and the yield of purified antibody determined from the absorbance at 280 nm and estimated antibody extinction coefficient of 1.4. Antibodies were further purification by size-exclusion chromatography on a Superdex 200 10/300 using the AKTA Pure system with PBS (pH 7.4) running buffer. Dynamic light scattering measurement were performed with a Nano S DLS (Malvern Instruments, Malvern, UK) on samples and polydispersity index (PDI) and the cumulant (or z-average) size (Zav) calculated using the zetasizer software (Malvern Instruments, Malvern, UK).

(248) However, during preparative size exclusion chromatography significant column matrix binding and retardation was observed (Table 8a and FIG. 19), as described previously. The elution volume (Ve) for Nivolumab and Vesencumab was 12.0 ml and 13.7 ml respectively, indicating retardation of Vesencumab and non-specific interaction with the column matrix. An additional elution peak at 10.4 ml indicated the presence of some higher molecular weight aggregated antibody, not observed for Nivolumab. To investigate the stability of the antibodies during storage, size purified antibodies were incubated at 4° C. for 2 weeks in PBS pH7.4. Dynamic light scattering (DLS) detected higher order aggregated species for Vesencumab, but not for Nivolumab (see FIG. 20 and Figure legend for methodology). Dynamic light scattering measurement were performed with a Zetasizer APS (Malvern Instruments, Malvern, UK) on samples that had been stored at 4° C. for 2 weeks. The DLS derived biophysical parameters of calculated percentage polydispersity, the polydispersity index (PDI), the cumulant (or z-average) size and average molecular weight are shown in Table 8 and show indicate significant aggregation of Vesencumab upon storage at 4° C. for 2 weeks compared with Nivolumab which is mono-disperse when stored under the same conditions.

(249) Synthetic DNA encoding the Vesencumab and Nivolumab heavy and light variable domains were cloned into the mammalian display vector pINT17-BSD (see Example 1 for vector maps and sequences), DNA sequence confirmed and transfection quality plasmid DNA prepared. HEK293 cells were transfected with TALE nucleases and stable cell lines created as described above un Example 2. After 14 days post-transfection (dpt) cells were stained with anti-human Fc PE (409303, Biolegend) for 30 min at 4° C. to determine antibody display presentation level (see Example 2). The monoclonal cell lines displaying either Nivolumab or Vesencumab were then stained with labelled human serum by the following protocol.

(250) Heat inactivated, human AB Serum (5 μl, 40 mg/ml H4522, Sigma) was diluted in PBS (195 μl) to give a final concentration of 1 mg/ml. This diluted human serum was then labelled with Dylight 633 using the Lightning-Link® Rapid Dylight® 633 kit (325-0000, Innova) according to the manufacturer instructions. HEK293 cell lines displaying Nivolumab or Vesencumab or wild-type HEK293 cells (one million cells) were pelleted (200 g, 3 minutes in am Eppendorf tube (1.5 ml). The pellet was resuspended in PBS (1 ml) and pelleted (600 g, 2.5 min). The pellet was resuspended in 1% BSA, PBS (100 μl) containing anti-Fc PE (0.5 μl, 409303, Biolegend) and either AB serum Dylight 633 labelled (5 μl, 0.5 mg/ml). The mix was incubated, shielded from light, at 4° C. for 30 min. 0.1% BSA, PBS (900 μl) was added and cells pelleted (600 g, 2.5 min). The cells were resuspended in 0.1% BSA, PBS (1 ml) and this wash step was repeated once. The cells were resuspended in 0.1% BSA, PBS (200 μl) with 7-AAD (5 μl per million cells). Labelled cells (50 μl) were analysed using the Intellicyte iQue screener. Flow cytometry analysis (FIG. 21) showed increased binding of labelled human serum to HEK293 cells displaying Vesencumab compared with Nivolumab (12.3% and 3.7% respectively) double positive for antibody and human serum binding respectively)

(251) Vesencumab is an antibody that failed to be developed beyond Phase 1 clinical trials.sup.105 and is an antibody with known self-aggregation and cross-interaction properties.sup.39,104. We have also shown that this antibody displays non-specific interaction with a size exclusion column matrix (FIG. 19) and aggregates upon storage (FIG. 20 and Table 8) compared with the clinically approved anti-PD1 antibody Nivolumab which shows no cross-interaction during size exclusion chromatography and remains mono-disperse after storage. We here show that we can differentiate between Vesencumab and Nivolumab when they are displayed on the surface of HEK293 cells in a cross-interaction flow cytometry assay. It is likely that this assay can be optimized to allow even greater discrimination between “sticky” and well-behaved antibodies. For example, the human serum could be biotinylated and conjugated to fluorophore labelled streptavidin conjugated to increase avidity. Alternatively one could modify any number of cross-interaction assays.sup.2 for flow cytometry as described previously and including, but not exclusively labelled baculovirus.sup.16,19 or a labelled mix of proteins, DNA and heparin sulphate containing molecules.sup.107.

Example 6b. Improvements in Polyreactivity Screening

(252) Using known antibodies with known polyreactivity profiles, we further exemplify the possibility of discriminating polyreactive binders from non-polyreactive binders within a population of binder-displaying cell clones, based on differences in binding to a non-target molecule (a polyreactivity probe). We demonstrate enrichment of clones which fail to bind the polyreactivity probe.

(253) Individual populations of HEK293 cells were prepared using nuclease-directed integration to express ustekinumab, briakinumab and amatuximab. The individual populations were stained with biotinylated DNA. Binding to DNA was detected on cells expressing briakinumab and amatuximab whereas cells expressing ustekinumab were not stained. DNA binding was normalised to mode. FIG. 33A. Staining with anti-Fc antibodies revealed that the briakinumab population was a mixture of IgG expressing and non-expressing cells accounting for an approximate 50:50 mix of DNA binding and DNA non-binding cells within the population. The different cell populations were labelled with cell-tracker dyes and mixed in equal proportion. Amatuximab cells were labelled with CellTrace Far Red (shown on x axis in Q3—FIGS. 33B and C), briakinumab with CellTrace CFSE (shown on y axis in Q1—FIGS. 33B and C) and ustekinumab remained unlabelled (double negative population in Q4—FIGS. 33B and C). The mixed population was stained with biotinylated DNA (20 μg DNA per 1 million cells in 200 μl 1% BSA) and labelled with anti-biotin microbeads. The population was sorted using MiniMACS beads in combination with the MS column. The flow through fraction from the MACS sort was analysed using the Intellicyt flow cytometer to count cells. 7-AAD was used as viability stain and dead cells were excluded from analysis. Ustekinumab, which did not bind DNA, was observed to enriched compared with briakinumab and amatuximab (FIG. 33B, FIG. 33C, Table 8b).

(254) TABLE-US-00011 TABLE 8b Relative cell counts (normalised to the number of ustekinumab cells and normalised to 10,000 counts) for each antibody before and after MACS. Sample Briakinumab Mix Amatuximab Ustekinumab Pre-MACS 3120 114 3420 3350 Post-MACS 2540 0 697 6790

(255) The relative percentage of briakinumab-displaying cells was reduced to 37% and the amatuximab-displaying cells were reduced to 10% compared with the ustekinumab-displaying cells. It is likely that the enrichment factors can be even higher since the input population of briakinumab included a relatively high proportion of non-antibody expressing cells and these will be retained in the “unbound population” which is selected here. This background could be reduced further by pre-sorting, post-sorting or co-sorting the cells for IgG expression or antigen binding as described earlier.

(256) Here we successfully used MACS, but the resolution and effective enrichment would be expected to be even greater with flow sorting by FACS.

(257) Additional polyreactivity probes were also tested for their ability to discriminate between cellular clones expressing either polyreactive and non-polyreactive antibodies.

(258) We found that cells expressing ustekinumab could be distinguished and separated from cells expressing briakinumab or ganitumab based on the extent of heparin sulphate binding. Briefly, 250,000 cells were stained with 9 μM Heparin-FITC (Creative PEGWorks) using the standard staining protocol as described previously in Example 5. Briakinumab and ganitumab showed heparin binding. Overlay plots are shown in FIG. 34. This non-specific binding possibly occurs through positively charged patches in the heavy chain CDRs of briakinumab and ganitumab.

(259) Chaperone proteins represent further polyreactivity probes which may be used as non-target molecules for de-selecting polyspecific binders. Chaperones are functionally related and assist in protein folding. Heat shock proteins (Hsp) are overexpressed in stressful conditions such as high temperature. Most chaperones are also abundantly expressed in normal cells where they recognise and bind non-native proteins thus preventing aggregation.

(260) A variety of therapeutic antibodies were displayed on HEK293 cells and tested for binding to Hsp70 and Hsp90.

(261) Of the antibodies tested in our experiment, brentuximab and lenzilumab showed binding to Hsp70 and Hsp90. Brentuximab (Vedotin) is an anti-CD30 antibody-drug conjugate that failed in a clinical trial to treat Hodgkin's lymphoma. It was previously shown to exhibit self-interaction and cross-interaction.sup.2. Lenzilumab is an anti-GM-CSF (granulocyte-macrophage colony-stimulating factor) antibody that failed a phase II trial for severe asthma.

(262) FIG. 35 shows an overlay of ustekinumab and briakinumab double stained with anti-human Fc PE and heat shock proteins (Hsp70 and 90) conjugated with DyLight 633. The gate within the overlay plots indicates cells that did not show detectable interaction with chaperones (Hsp70 and Hsp90) and which can be FACS sorted to provide a selected population of clones in which binders that recognise chaperones have been depleted (and preferably eliminated).

(263) Pooled data from multiple experiments with a variety of different antibodies and different polyreactivity probes are depicted in FIG. 36. The individual ampules were tested in separate, independent experiments but the Intellicyt flow cytometer has a fixed voltage so the fluorescence intensity is expected to be consistent for all samples. Additionally, Hek293 was used an internal control for each experiment. Table 8c below summarises data from the panel of antibodies tested for binding to the polyreactivity probes.

(264) TABLE-US-00012 TABLE 8c Summary of antibody polyreactivity screening. Antibodies DNA Heparin Chaperones FcRn Alirocumab No No No No Amatuximab Yes Yes Yes Yes Brentuximab ND ND Yes No Briakinumab Yes Yes Yes Yes Ganitumab Yes Yes No Yes Lenzulimab ND Yes Yes No Ustekinumab No No No No Vesencumab ND ND No Yes Yes = antibody showed binding to polyreactivity probe. No = antibody did not show binding to polyreactivity probe. ND = binding to the polyreactivity probe was not determined.

(265) Amatuximab, brentuximab, briakinumab and lenzilumab showed binding to chaperone proteins. The polyreactivity of these antibodies may arise due to hydrophobic clusters of amino acids within the antibody variable domains, giving rise to van der Waals interactions with proteins that also possess hydrophobic regions such as the chaperone proteins. Other reasons for polyreactivity can be the presence of positively charge patches of amino acids (e.g., consisting of arginine or lysine residues) which interact with molecules that have a net negative charge such as DNA or heparan sulphate or with proteins such as FcRn that have a positive charge patch on their surface.sup.22. Amatuximab, briakinumab, ganitumab and lenzulimab all bound DNA and heparin in our experiment. Antibodies that possess both hydrophobic and positively charged patches on their surface may have increased polyreactivity. Based on our data, examples of antibodies capable of both binding chaperone proteins via hydrophobic patches and binding DNA, heparain sulphate or FcRn at neutral pH include briakinumab, amatuximab and lenzulimab. These data are consistent with earlier reports of non-specific binding shown by briakinumab.sup.2.

Example 7. Quantitation of Display Level for Parental and Improved Clones

(266) In Example 2 we observed higher eukaryotic cell display-level differences for three pairs of antibodies, after nuclease mediated transgene integration into HEK293 cells and the selection of stable cell lines. The display level, judged by staining the cells with PE labelled anti-Fc and measuring the mean fluorescence intensity by flow cytometry, correlated by the antibody self-interaction and cross-interaction properties (Example 3). In this Example 7 we quantitate the antibody display copy number on the cell surface and show that the display copy number correlates with the antibody biophysical properties. Quantitative measurement was done using bead based calibration curve where beads have a precisely defined number of Fc-specific capture antibodies.

(267) Quantum Simply Cellular (QSC) microspheres kit (815, Bangs laboratories, Inc.) has 5 bead populations—one blank and four bead populations with increasing amount of Fc-specific capture antibody (goat anti-mouse IgG). QSC beads are stained with the same flurochrome-conjugated antibody that is used to label cells and analysed on the flow cytometer according to manufacturers instructions. Briefly one drop of QSC microspheres was added to a microcentrifuge tube and 50 μL of staining buffer (1% BSA) was added and the tube was gently flicked. 5 ul of PE anti-human IgG Fc Antibody was added to the QSC microspheres, mixed gently and incubated in dark for 30 minutes. QSC microspheres was washed twice with 1 ml wash buffer (PBS containing 0.1% BSA) by centrifuging at 2500×G for 5 minutes. Bead pellet was resuspended in 150 ul wash buffer. Stained bead populations and blank population are combined (10 ul per population) in a single well and run in the Intellicyt. In parallel cell staining was performed for the HEK293 cells expressing different antibodies on the cell surface as described in the Example 2.

(268) A calibration curve is generated by plotting the median fluorescent intensity (FIG. 26) of each bead population versus its assigned antibody binding capacity. Fluorescence intensity of the antibody expressing population is compared with the antibody binding capacity of the beads and linear regression is calculated using QuickCal (Bangs laboratories, Inc.) to enable the calculation of antibody display copy number (Table 9).

(269) TABLE-US-00013 TABLE 9 Copy number of antibodies displayed on HEK293 cells calculated using Quantum Simply Cellular beads Antibody Copy Number (×10.sup.3) CNTO607 113 CNTO607-W100A 313 MEDI-1912 48 MEDI1-912_STT 433 Ang2mAb 125 Angiopoietin2-C49T 570 Briakinumab 273 Ustenikumab 706 Pembrolizumab 910

(270) The clinically approved anti-PD1 antibody Pembrolizumab, known to have good biophysical characteristics in terms of low self-interaction properties.sup.2 had the highest copy number of this test set. The most intense calibration bead had 886,000 copies/bead and this was in excess of that (approximately 910,000 copies/cell). Similarly, the anti-IL12 antibody Ustekinumab, approved for the clinical treatment of Crohns disease, displayed a higher copy number on cells (706,000) compared with the anti-IL-12 antibody Briakinumab (273,000 copies), which showed poor efficacy in a Phase III human clinical trial to treat psoriasis. Briakinumab is described as having increased self- and cross-interaction properties, as measured in a variety of assays, compared to Ustekinumab including the self-interaction assay AC-SINS and cross-interaction assays with a poly-specificity reagent and baculovirus particles.sup.2,23.

(271) For all three antibody pairs the copy number was lower than controls, the re-engineered daughter clones with improved biophysical properties had higher cell display copy numbers. For example, the improved daughter antibodies CNTO607-W100A, MED-1912_STT and Ang2mAb_C49 had display copy numbers of 313 thousand, 433 thousand and 570 thousand copies respectively representing a 2.8-fold, 9-fold and 4.6-fold increase in copy number compared to the original parental molecules with known problem of self and cross-interaction.

(272) In this example we show a clear relationship between the antibody copy number displayed on the cell surface, after nuclease mediated transgene integration into the host genome and stable cell line selection, and the biophysical properties of the antibody displayed. Antibodies with the properties of self-interaction and cross-interaction displayed a lower copy number compared with antibodies that did not score highly in assays designed to measure self and cross-interaction. The better behaved antibodies with a good biophysical profile of low self-interaction and low cross-interaction properties displayed a high copy number on the surface of higher eukaryotic cells.

Example 8a. Combining High Level Expression for Developability and Low Level Expression for Affinity Stringency

(273) High level polypeptide expression, e.g. for antibodies where the antibody heavy and light chains genes are driven by strong constitutive promoters, has been demonstrated to be useful in the enrichment from populations of antibodies with superior biophysical properties, such as low self-interaction (as described above in Examples 3, 4 and 5). The presentation of high polypeptide concentrations on a cell can help detect self-interaction and increased avidity allows sensitive detection of undesired non-specific interactions with other molecules. One adverse consequence of this however is that there may be a reduction in the achieved rate of enrichment of high affinity over low affinity even when low concentrations of antigen are used to drive stringency. An additional consequence of this “low powered” enrichment is that there may in fact be a preferential selection for surface presentation even when the goal is to enrich higher affinity binders. The rate of enrichment could be enhanced however by using a lower density of binder presentation on the cell surface. In the discussion below we will use antibodies and their antigens as an example to represent in general interactions of binders of different affinities.

(274) If we consider 10.sup.6 cells in a 100 microlitres volume displaying 6×10.sup.5 monovalent binding sites/cell then we have 6×10.sup.11 molecules/100 microlitre or 6×10.sup.15 molecules/l. This number of binding sites in this volume would be equivalent to a concentration of 10 nM. Using antigen concentrations below 10 nM in this situation means that the antibody will be in excess over antigen and the relatively high antibody concentration will help drive association even for lower affinity antibodies. For example if one is trying to separate rare cells expressing an antibody with K.sub.D 0.1 nM from an excess of cells expressing an antibody with a lower affinity of K.sub.D 10 nM, then under the conditions outlined above, (i.e., the equivalent of an antibody concentration of 10 nM) a significant proportion of the antigen will be in complex with the lower affinity antibody.

(275) The law of mass action deals with two interacting molecules forming a complex. These formulae are intended to cover interactions in free solution and we are dealing with antibodies immobilised on suspension cells. The calculations of selectivity below consider the concentration of complex formed in each case for a single antibody in solution. Nonetheless we can use this knowledge of the behaviour of molecules in solution from the law of mass action to better understand the relationship between affinity, antibody concentration, antigen concentration and the formation of complex as it might affect surface presented antibodies.

(276) If two interacting molecules e.g. monovalent antibody (A) and antigen (B) are mixed, they can form a complex (A:B) and eventually reach equilibrium. The position of the equilibrium is dependent on the concentration of antibody and antigen but can be described by the dissociation constant, K.sub.D as follows (where [A]. [B] and [AB] denote the concentration at equilibrium):

(277) $K_D=\frac{\left[A\right]\left[B\right]}{\left[A\text{:}B\right]}$

(278) This equation can be re-arranged to calculate the concentration of complex (AB) formed under different conditions of K.sub.D, concentration of A and concentration of B.

(279) If you have a binding reaction that is in equilibrium:
A+B.Math.AB  (1)
then the dissociation constant (K.sub.D) is defined as:

(280) $\begin{array}{cc}{K}_D=\frac{\left[A\right]\left[B\right]}{\left[\mathrm{AB}\right]}& \left(2\right)\end{array}$
where [A], [B], and [AB] are the concentrations of the reactants at equilibrium. The total concentrations of the reactants (A.sub.T and B.sub.T, which are the concentrations you added to the “test tube”) are as follows:
[A.sub.T]=[A]+[AB] which can be rearranged as [A]=[A.sub.T]−[AB]  (3)
[B.sub.T]=[B]+[AB] which can be rearranged as [B]=[B.sub.T]−[AB]  (4)
Substitute eq. 3 and eq. 4 into eq. 2:

(281) $\begin{array}{cc}{K}_D=\frac{\left(\left[A_T\right]-\left[\mathrm{AB}\right]\right)\left(\left[B_T\right]-\left[\mathrm{AB}\right]\right)}{\left[\mathrm{AB}\right]}& \left(9\right)\end{array}$
Rearrange the Equation:
K.sub.D[AB]=([A.sub.T]−[AB])([B.sub.T]−[AB])  (10)
Multiply it out and rearrange (concentration brackets are removed for clarity):
AB.sup.2−(A.sub.T+B.sub.T+K.sub.D)(AB)+(A.sub.TB.sub.T)=0  (11)
into the form
ax.sup.2+bx+c=0  (12)
where,
a=1  (14)
b=−(A.sub.T+B.sub.T+K.sub.D)  (15)
c=(A.sub.TB.sub.T)  (16)
which allows for solving via the quadratic equation:

(282) $\begin{array}{cc}x=\frac{-b\pm \sqrt{b^2-4\mathrm{ac}}}{2a}& \left(16\right)\\ \mathrm{AB}=\frac{\left(A_T+B_T+K_D\right)-\sqrt{{\left(A_T+B_T+K_D\right)}^2-4\left(A_T{B}_T\right)}}{2}& \left(17\right)\end{array}$

(283) Assuming an antigen concentration of 0.1 nM is used, FIG. 22a shows the effect of varying concentration of antibody (A) for 2 antibodies of different affinities (K.sub.D equal to 10 nM for Ab1 or 0.1 nM for Ab2). FIG. 22b shows the relative “selectivity” between the 2 antibodies being the ratio of concentration of complex formed when the antibodies are individually incubated with 0.1 nM antigen and varying antibody concentrations.

(284) This shows that despite using an antigen concentration well below the K.sub.D of the interaction, the concentration of complex is 50 pM representing 50% of the antigen in complex when Ab1 is present at a concentration of 10 nM. (10 nM is the concentration calculated in the example above for cell surface display at 6×10.sup.5 copies/cell). Under the same conditions using the high affinity Ab2 antibody the concentration of complex will be 99 pM (representing 99% of the antigen) so there is only a 2 fold difference in selectivity (Table 10). If however the antibody concentration is reduced to 0.1 nM (equivalent to reducing the display level by 100 fold in this example) then the concentration of complex drops to 1 pM for Ab1 while a concentration of complex of 38.2 pM will be achieved with Ab2 (Table 10) representing a selectivity of 38-fold. Reduced density also has the advantage of reducing the potential for target rebinding. The problem of rebinding in the presence of a high density of immobilised binder is particularly well recognised and documented in surface based affinity measurement such as surface plasmon resonance (BIAcore manual).

(285) With flow sorting one is measuring the relative concentration of complex by detecting the presence of fluorescently labelled antigen (either directly or indirectly labelled) on the surface of the cell. The fluorescent signal on the cell detected within the flow cytometer is therefore an indication of the concentration of complex formed on the cell under the conditions used. Thus under more limiting conditions of antibody presentation a greater separation will be achieved between clones presenting high and low affinity antibodies.

(286) TABLE-US-00014 TABLE 10 Concentration of complex formed in solution using 0.1 nM antigen and different concentrations of a high affinity (K.sub.D 0.1 nM) and a low affinity (K.sub.D 10 nM) antibody [complex in pM] for [complex in pM] for [Ab in nM] Ab1 (K.sub.D 10 nM) Ab2 (K.sub.D 0.1 nM) Selectivity 10 49.9 99.0 2 1 9.0 90.1 10 0.1 1.0 38.2 39

(287) During the process of selecting for the optimal antibody, during an antibody discovery campaign, it can be desirable to select for antibodies with higher affinity to their target. A method to increase stringency and enrich for clones with improved affinity for their target during antibody phage display selection is to reduce the concentration of the target antigen during selection.sup.57 (Fellouse F A, Sidhu S S: Making antibodies in bacteria. In: Making and Using Antibodies: A Practical Handbook. Edited by Howard G C, Kaser M R: CRC Press; 2007: 157-180.sup.108). In mammalian display selections, the labelled antigen concentration can also be reduced during FACS or MACS to enable the selective enrichment of clones with improved affinity. However, the high display level achieved, where the antibody expression is driven by strong constitutive promoters can equate to a concentration of antibody above the desired target affinity. It therefore may be desirable to reduce the antibody cell display level on HEK293 cells e.g. to below 60,000 per cell giving a 10 fold selectivity between 10 nM and 0.1 nM in the example above. This will enable superior enrichment of antibodies with improved affinity over lower affinity clones.

(288) Antibody display level could be reduced by several different methods at the transcriptional, post-transcriptional, translational or post-translational stages of antibody. For example, weak promoters could be employed to reduce the rate of production of primary mRNA transcript, non-optimal splice/acceptor sites could be incorporated to reduce the efficiency and rate of the production of mature mRNA and export from the nucleus to the cytoplasm. The stability of the mRNA could be reduced thus reducing the transcript half-life and effective concentration. Translational control of expression could be by altering the Kozak consensus sequence to affect ribosomal binding to the mRNA. An example of post-translational control could be by the use of non-optimal leader sequences to reduce the efficiency of transport to the endoplasmic reticulum.

(289) In this example we show the use of splice/acceptor site engineering to reduce the antibody display level. This reduced display system was shown to enable the more efficient separation of HEK293 cell line mixtures displaying antibodies with different affinities for their target.

(290) We have devised a way to combine surface display with antibody secretion that is based on the natural system used in B cells. During B cell maturation antibodies are expressed in a membrane bound form and this switches to a mainly secreted form as plasma cells mature. This matches the requirements for eukaryotic display where the ability to effect cell surface display relies on expression of the trans-membrane form. Alternatively, the ability to express the antibody in a secreted form once a clone has been selected would allow immediate production of free, soluble antibody for further characterization.

(291) The balance between surface display and secretion in B cells is driven largely by the balance between poly A addition (leading to secreted IgG) and splicing (leading to membrane tethering).sup.109,110. A “proximal” polyadenylation site is found 100-200 bp after the end of the CH3 domain and this generates an mRNA which stops translation at the end of the CH3 domain resulting in a secreted product. Near the end of the CH3 domain there is also a potential splice donor site which can splice to a downstream exon (M1) to create an in-frame fusion with a “hinge” and transmembrane domain. (The M1 exon in turn splices to an M2 exon encoding an intracellular domain). The balance of secreted versus membrane bound IgG presentation depends on the balance between polyadenylation at the proximal poly A site and splicing to the M1 exon. This is distinct from more recently published methods where one of 2 alternative exons are used to switch between secreted and membrane bound.sup.111.

(292) Splicing normally occurs through U1 small nuclear RNA (snRNA) which is required to initiate spliceosome assembly leading to intron removal. The splice donor site at the end of CH3 is sub-optimal compared to consensus splice donor sites. The non-optimal splice donor is conserved throughout evolution and would be expected to give non-optimal base-pairing to U1 snRNA. In fact it has been shown that mutation of the non-optimal splice donor site to a consensus splice donor sequence changes the balance from processed RNA encoding a predominantly secreted form to a predominantly membrane bound as a result of increased splicing.sup.86. This represents an early example of modifying a splice donor to alter the balance between splicing versus polyadenylation to effect a significant change in the balance between secretion and polyadenylation. Based on the work of Peterson et al.sup.109,110 it would be anticipated that the degree of optimization of the splice donor would affect the balance between splicing and polyadenylation therefore affecting the proportion of displayed antibody. In order to find the optimal balance between membrane and secreted forms a number of alternative splice donor sites were created at the end of the CH3 exon.

(293) The sequence of U1 snRNA involved in splicing initiation is shown above the mRNA sequence generated from IgG2 CH3 domain. Positions of mismatch are underlined:

(294) TABLE-US-00015 U1 snRNA uc/cauuca IgG2 CH3 splice donor gg/guaaau

(295) Four variants were designed around the splice donor including wild type (J9—GG/GTAAT), partial optimisation (J10—AG/GTAAA), partial optimisation (J29—GG/GTAAG) and fully optimized (J30—AG/GTAAG) as shown in FIG. 23. From the hybridization with the U1snRNA, it would be expected that the J30 variant would allow the most efficient splicing, resulting in a greater proportion of membrane tethered antibody at the cell surface compared with the “wild-type” sequence J9 where less efficient splicing would result in a lower level of membrane tethered antibody and a greater proportion of secretion. The J10 and J29, partially optimized variants would expect to give rise to antibody display levels intermediate between J9 and J30.

(296) Variants of the antibody display targeting vector pINT17-BSD were constructed where an embedded HindIII restriction site was added to the DNA encoding the C-terminus of the IgG1 CH3 domain and the human IgG intron and M1 exon, encoding a transmembrane domain replaced the PDGFR transmembrane domain encoded by pINT17-BSD (Example 1). The splice donor variants J9, 10, 29 and 30 were constructed by a combination of synthetic gene synthesis, PCR assembly and restriction enzyme cloning. The annotated DNA sequence of the pINT17-J30 vector is shown in FIGS. 24A-24D from the XhoI to SbfI restriction enzyme sites. The vector backbone, exterior to the XhoI—SbfI insert and not shown in FIGS. 24A-24D, is identical to pINT17-BSD (FIG. 1).

(297) DNA encoding the VH and VL chains of the anti-PD1 antibody Nivolumab was cloned into the four splice donor variant targeting vectors pINT17-J9, pINT17-J10, pINT17-J29 and pINT17-J30 and used to create stable cell lines by nuclease mediated gene integration as described in Example 2. For comparison Nivolumab was also cloned into the standard pINT17_BSD vector shown in FIG. 1. This construct fuses the antibody CH3 domain directly to the PDGFr transmembrane domain without splicing and is referred to in this example as pINT17-PDGFR. After 27 days of blasticidin selection, the cells were stained with anti-Fc labelled with PE and analysed by flow cytometry as described above (FIG. 25). This showed that the antibody display level was greater for pINT17-PDGFR, the antibody expressed with a direct fusion between the IgG1 CH3 domain and a PDGFR transmembrane, (FIG. 25e) compared with the constructs with the native IgG transmembrane domain with an intron placed between CH3 and transmembrane domains (FIG. 25a to d). J30 is completely complementary to the U1 snRNA and so would be expected to result in more efficient splicing with the M1 exon, encoding the transmembrane domain than the J9, J10 or J29 variants. FIG. 25d show that this is indeed the case with the J30 variant showing the highest level. It also shows however that the level of expression is significantly lower even for J30 variant compared with that found on the pINT17-PDGFr construct (FIG. 25d) used throughout the earlier examples.

(298) To demonstrate that a reduction in antibody display copy number can aid the differentiation and separation of antibodies with different affinities, a test pair of anti-PD1 antibodies was chosen with different affinities for their target. The antibodies chosen were Nivolumab and another PD1 antibody (337_1_C08) with equilibrium dissociation constant (K.sub.D) affinities, determined by surface plasmon resonance (SPR), of 3 nM.sup.112 and 74 nM respectively. DNA encoding the VH and VL chains of the anti-PD1 antibodies were cloned into the targeting vectors pINT17-J30 and pINT17-PDGFr (also referred to as pINT17-BSD) and used to create stable cell lines by nuclease mediated gene integration as described in Example 2. The quantitation of antibody display levels by quantitative flow cytometry analysis was performed as described previously.sup.113 and in the manufacturer instructions (Quantum Simply Cellular anti-mouse IgG beads, catalogue number 815, Bangs Laboratories Inc), (FIG. 26) for cells 21 days post blasticidin selection. The copy number on cells, calculated from measurement of the median fluorescence intensity of cells stained with anti-Fc-PE compared with a reference bead set (catalogue number 815, Bangs Laboratories Inc) (FIG. 26). As shown in Table 11A, the cell display copy number was reduced for pINT17-J30 (plus intron) expression cassettes compared with the pINT17-BSD expression cassette, with no intron between the CH3 and transmembrane domains, for both antibodies.

(299) TABLE-US-00016 TABLE 11A Copy number calculation Copy Number (×10.sup.3) Antibody pINT17-BSD pINT17-J30 Nivolumab 61 18 337_1_C08 607 11

(300) As the copy number of antibody and concentration of antigen diminishes, the intensity of signal observed by flow cytometry also diminishes (FIG. 25). It is possible to enrich labeled cells below the sensitivity limits of flow cytometry using magnetic bead sorting and this was employed here. To test if reduced copy number on the cell surface would aid the separation of antibodies with different affinities, the Nivolumab or 337_1_C08 expressing cell lines were each stained with different fluorophore dyes This will allow the relative enrichment of one cell over the other to be observed. The labelled cells were mixed, incubated with varying concentrations of biotinylated PD1 and separated by MACS. Flow cytometry analysis was then performed to determine if there was enrichment of cells expressing the antibody with a higher affinity for PD1. Four HEK293 cell lines were tested in this example, originally transfected with: a. pINT17-BSD-Nivolumab b. pINT17-J30-Nivolumab c. pINT17-BSD-337_1_C08 d. pINT17-J30-337_1_C08

(301) The pINT17-BSD expression cassette (with a direct fusion between the antibody CH3 and transmembrane domains) was shown above to express a higher antibody display level than from the pINT17-J30 expression cassette. The Nivolumab expressing cells (5×10.sup.6) were stained with Cell Tracker Green (50 nM, C7025, Thermo Fisher) and the 337_1_C08 expressing cells were stained with Cell Tracker Deep Red (50 nM, C34565, Thermo Fisher) according to the manufacturer instructions. Briefly cells were washed with PBS, incubated with PBS containing tracker dye (50 nM) and incubated at 37° C. for 10 minutes. Cells were then washed with PBS. The pre-stained cells expressing Nivolumab or 337_1_C08, derived from the pINT17-J30 expression cassette were mixed at a 1:1 ratio (5×10.sup.6 cells each in a volume of 10 ml). The mixed cells were pelleted (100 g, 3 minutes) and each cell pellet resuspended in PBS (1 ml) containing either 0, 0.1, 1, or 10 nM biotinylated PD1 (PD1-H82E4, AcroBiosystems) and this was incubated at 4° C. for 30 minutes. Cells were washed with 0.1% BSA, PBS and Streptavidin beads (10 μl) and 90 μls of 1% BSA was added to each sample and mixed. Samples were incubated in the fridge (4° C.) for 15 minutes. Cells were washed using 2 ml of 0.1% BSA and spun down at 200×G for 4 minutes. The pellet was re-suspended in 500 μls of Separation Buffer (MACS Rinsing Buffer consisting of 1×PBS+2 mM EDTA+0.5% BSA). LS columns were washed with 3 ml Separation Buffer. The cell suspension was added to columns, one sample per column. Uncaptured cells were collected by washing the columns with 3 ml separation buffer three times and the flow through was collected. The column was placed in a new collection tube and 5 ml separation buffer was added to each column and immediately flushed out using the supplied column plunger. Elution samples and flow through samples were counted using cell counter and trypan blue to determine cell recovery rate. 1×10.sup.6 cells for each sample was diluted into 500 μl 0.1% BSA, PBS. 50 μl of diluted cells were analysed by flow cytometry.

(302) When the low density display vector pINT17-J30 was employed, it was possible to selectively separate the cell line displaying the low affinity anti-PD1 antibody 337_1_C08 from the higher affinity anti-PD1 antibody Nivolumab expressing cells. This is shown in FIG. 27a where dot-plot flow cytometry results are presented for either the MACS elution or flow through populations. Each panel shows a dot-plot where the green fluorescence intensity is plotted on the x-axis (FL1, Nivolumab expressing cells) and the red fluorescence intensity on the y-axis (FL4, 337_C08 expressing cells) for the elution and flow-through fractions pre-incubated with 0, 0.1, 1, or 10 nM biotinylated PD1. For example, upon incubation of the cell-line mix with 1 nM biotinylated PD1 followed by MACS separation, this resulted in 95% enrichment for the green dye stained high affinity anti-PD1 Nivolumab displaying cells compared with the red dye stained lower affinity anti-PD1 337_1_C08 antibody. In contrast, the unbound cells in the flow through contained predominantly the lower affinity red stained anti-PD1 337_1_C08 antibody displaying cells. Also, superior separation was observed with 1 nM PD1 compared with 10 nM PD1 incubations. However, when the same experiment was performed with the higher copy number cell lines, derived from the pINT17-BSD cassette, no preferential enrichment was observed between the Nivolumab and 337_1_C08 expressing cells (FIG. 27b).

(303) This example has therefore illustrated that reducing the antibody display level on the cell surface can increase the enrichment of high affinity antibodies over lower affinities, particularly as high affinities are achieved within the population. Optimal cell display separation of a high affinity antibody from an antibody with lower affinity for its target was achieved by a combination of a reduction in antibody display level and a reduction in the concentration of labelled target antigen employed during selection. The ability to separate antibodies, during eukaryotic cell display selection, according to affinity is important during an antibody discovery campaign depending on the required target candidate profile of the desired antibody. It is also important to enrich for antibodies with improved affinity for their target during affinity maturation. This example has shown the importance of cell display copy number in helping to discriminate antibodies with high affinity for their target.

Example 8b. Improvements Using Inducible Tet Promoter

(304) It is advantageous to create a vector that enables the inducible expression of mammalian cell displayed antibodies. This facilitates the combination of selection steps at different surface presentation levels, e.g., selection of low display levels required for stringent selection can be followed by the induction of high display levels required for the selection of antibodies with improved biophysical properties as exemplified in Examples 4 and 5.

(305) An inducible antibody display targeting vector (pINT18-Tet1) was constructed by a combination of synthetic gene synthesis, PCR assembly and restriction enzyme mediated cloning (FIG. 28). pINT18-Tet1 contains the same vector back-bone, AAVS homology arms and promoter-less blasticidin resistance gene as pINT17-BSD. FIG. 28 shows and the annotated nucleic acid sequence of pINT18-Tet1 between the AAVS homology arms. The key features of this vector include a CMV promoter driving the expression of a reverse Tet activator (rtTA) protein.sup.114, a Tet operator (TetO) tetrad followed by a minimal CMV promoter, BM40 leader fused to the light chain genes (VL and CL), a P2A peptide.sup.115 to enable ribosome “skipping”, followed by the antibody heavy chain coding sequence. The heavy and light variable genes (VH and VL) of Pembrolizumab, Nivolumab and 337_1_C03 were cloned into this inducible targeting vector and a stable HEK293 cell lines were created by nuclease mediated gene integration as described above. FIG. 29 shows low basal antibody display in the absence of doxycycline when cells are stained with anti-Fc-PE and analysed by flow cytometry. However, 24 hours after the addition of 20 ng/ml doxycycline antibody expression is observed on the cell surface. This display level can be controlled as shown by a reduction in Fc staining when the doxycycline concentration is titrated to 2 ng/ml. Thus, the exemplification of an inducible cell displayed binder is shown and the ability to control display levels by varying the concentration of inducer.

(306) An improved (“third generation”) inducible targeting vector was later constructed to enable an improved range of antibody display levels. pINT17-Tet was constructed by a combination of synthetic gene synthesis, PCR assembly and restriction enzyme mediated cloning. pINT17-Tet contains the same vector back-bone, AAVS homology arms and promoter-less blasticidin resistance gene as pINT17-BSD. FIG. 37 shows the annotated nucleic acid sequence of pINT17-Tet between the AAVS homology arms. This plasmid includes an elongation factor promoter driving the expression of the reverse Tet activator (rtTA-3G). rtTA-3G is a modified form of the original rtTA protein.sup.114 which was evolved to display higher sensitivity to the inducer doxycycline.sup.116. pINT17-Tet also contains a b-directional inducible promoter (pTRE3G) to drive the expression of the antibody heavy and light chains. pTRE3G was optimised to widen the window between low basal expression and high maximal expression after induction.sup.117. It consists of 7 repeats of a 19 bp tet operator (TetO) sequence with two flanking minimal CMV promoters.

(307) The VH and VL genes of the anti-PD-1 antibodies: 1549_02_D06, 1535_01_E03 and 337_1_C08 and the anti-PCSK9 antibodies bococizumab, 884_01_G01, 5A10i and alirocumab were cloned into pTet17-Tet. The anti-PD1 antibody 337_1_C08 was described in Example 8 with an affinity for PD1 of 74 nM. The antibodies 1549_02_D06 and 1535_01_E03 are affinity matured daughter clones of the parental antibody 337_1_C08 which have an affinity (K.sub.D) for PD-1 of 2.9 nM and 17 nM respectively. The VH and VL genes of the anti-PCSK9 antibodies bococizumab, 884_01_G01, 5A10i. and alirocumab (all described previously in Example 5) were also cloned into pINT17-Tet.

(308) The pINT17-Tet targeting vector harbouring the genes of the anti-PD1 and anti-PCSK9 antibodies were used to transfect HEK293 cells with plasmids encoding the AAVS TALE nucleases as described above. After 25 days of blasticidin drug selection, the stable cell lines were induced with 0, 2, 4 or 100 ng/ml doxycycline. 24 hours post induction the induced cell lines were stained with anti-Fc-PE and cells analysed by flow cytometry. Histogram plots of the cell-lines were generated by plotting cell number against fluorescence intensity.

(309) The cell lines showed very low basal expression in the absence of doxycycline (FIG. 38a). Full induction was observed by the addition of doxycycline at a concentration of 100 ng/ml (FIG. 38d) 24 hours post induction. Intermediate induction was achieved by the addition of doxycycline concentrations of 2 and 4 ng/ml (FIGS. 38b and 38c). This resulted in a lower average display level compared with full induction, but did result in a spread of induction resembling the bi-modal gene expression observed previously with the seven TetO repeats.sup.118.

(310) The average displayed copy number of antibodies was quantitated by the method described in Example 7. Here the average mean fluorescence intensity (MFI) of the cell lines, induced with varying amounts of doxycycline, was determined and this used to convert to copy number using a calibration plot generated using the Quantum Simply Cellular anti-mouse IgG beads (catalogue number 815, Bangs Laboratories Inc) stained with mouse IgG-PE label. Tables 11c and 11d show the calculated display copy numbers 24 or 48 hours post induction (hpi). Very low basal expression was observed for the antibodies in the absence of doxycycline, which fell below the limits of detection when staining with anti-Fc-PE in some instances. There was detectable basal expression for all the antibodies for 48 hpi and this may be because the cells would have reached a stationary growth phase compared with 24 hpi. The average display level increased for all the antibodies as the concentration of doxycycline increased from 2 ng/ml to 4 ng/ml to 100 ng/ml doxycycline. Increasing the concentration of doxycycline above 100 ng/ml did not result in an increase in display level and so maximal induction was achieved at a concentration of 100 ng/ml doxycycline.

(311) Display level differences were observed between the anti-PCSK9 IgG bococizumab, which is prone to self-interaction and polyreactivity 2, and the well behaved parental antibody 5A10i. The parental antibody 5A10i was displayed on the surface of HEK293 cells with a copy number of 235,000 whereas bococizumab displayed on the surface of HEK293 cells with a copy number of 36,000 when cells were induced with 100 ng/ml doxycycline and display levels determine 24 hours post induction (hpi). This represents a 6.5-fold reduction in display level of bococizumab compared to the parental antibody 5A10i. This observed reduction in cell displayed copy number for bococizumab compared with 5A10i was observed previously (Example 5, FIG. 10) when the antibody expression was driven by constitutive promoters. Therefore, this inducible system, at full induction, is able to differentiate between antibodies that have different developability profiles in terms of self-interaction and polyreactivity as we demonstrated previously with a constitutive promoter display system.

(312) We also displayed a variant of bococizumab which was identified by mammalian display library screening named 884_01_G01 (Example 5). This antibody was shown to be well behaved biophysically in an AC-SINS assay and by HPLC-SEC (Table 6, Example 5 and FIG. 18, Example 5). This antibody displayed to high levels on the surface of HEK293 cells 24 hpi (100 ng/ml doxycycline) with a display copy number of 768,000 (Table 11B). The well-behaved antibody alirocumab.sup.2 also displayed to high levels on the surface of HEK293 cells 24 hpi under full induction conditions (Table 11B).

(313) Doxycycline is known to degrade in culture supernatant with a half-life of approximately one day, depending on culture conditions. Examining the display level of antibodies at different time points post induction with no replenishment of doxycycline provides some evidence regarding the dynamic turn-over of antibodies on the cell surface. Alternatively, induction could occur in the presence of doxycycline for a 24-hour period followed by a complete media change so the cells are no longer exposed to doxycycline. Here we simply re-examined the display levels 48 hours post induction (Table 11C). Several well-behaved antibodies such as alirocumab maintained or increased their cell surface display levels 48 hpi. Other antibodies such as bococizumab showed a greater than 2-fold reduction in cell surface display level at 48 hpi compared with 24 hpi (FIG. 39). The rate of turn-over, degradation or internalisation of an antibody displayed on a cell, where the displayed antibody is not being continuously replenished with newly expressed antibody, is an additional means to select antibodies based on their developability characteristics in terms of self-interaction and stability (see also Example 11).

(314) TABLE-US-00017 TABLE 11B Copy number of antibodies displayed on HEK293 cells calculated using Quantum Simply Cellular beads 24 hpi with 0, 2, 4 or 100 ng/ml Doxyclicycline (Dox) induction. nd, not detectable. Copy Number (×10.sup.3) Dox Dox Dox Dox (0 ng/ml) (2 ng/ml) (4 ng/ml) (100 ng/ml) 1549_02_D06 Nd 4.5 21.9 356 1535_01_E03 Nd 3.3 16.5 137 337_1_C08 Nd 5.6 31.0 251 Bococizumab Nd 2.4 5.1 36.1 884_01_G01 Nd 26.7 115 768 5A10i 0.1 14.0 48 235 Aliricocumab 0.4 32.8 144 557

(315) TABLE-US-00018 TABLE 11C Copy number of antibodies displayed on HEK293 cells calculated using Quantum Simply Cellular beads 48 hpi with 0, 2, 4 or 100 ng/ml Doxyclicycline (Dox) induction. Copy Number (×10.sup.3) Dox Dox Dox Dox (0 ng/ml) (2 ng/ml) (4 ng/ml) (100 ng/ml) 1549_02_D06 0.4 11.1 51.5 317 1535_01_E03 0.7 6.1 46.8 238 337_1_C08 0.4 20.4 108 294 Bococizumab 0.7 2.0 5 17.5 884_01_G01 0.6 26.4 112 417 5A10i 5.2 20.7 40.2 73.5 Aliricocumab 1.7 101 339 668

(316) pINT17-Tet therefore represents an example of an inducible mammalian display targeting vector that can be used to create monoclonal cell lines by nuclease mediated gene integration that can be switched to high expression, full induction mode to enable the developability screening described above for antibody self-interaction and polyreactivity screening. Cell lines created with pINT17-Tet can also be switch to low copy display mode by either basal display levels, in the absence of inducer, or adding limiting concentrations of doxycycline. This low copy number display mode will enable the more efficient separation of antibody clones with different affinities as described in Example 8a.

(317) To demonstrate the utility of inducible mammalian display cell lines for the separation of antibodies displayed on the cell surface with different affinities for their target, HEK293 cells displaying the high affinity anti-PD1 antibody 1549_02_D06 (K.sub.D=2.9 nM for PD-1) or the low affinity anti-PD-1 antibody 337_1_C08 (K.sub.D=74 nM for PD-1) were created by nuclease mediated gene integration with the pINT17-Tet targeting vector. Hek293 cells displaying the 1549_02_D06 antibody were labelled with CellTrace CFSE Cell Proliferation Kit (Thermo cat #C34554—Excitation/emission wavelengths—492 nm/517 nm) and cells displaying the 337_1_C08 antibody were not labelled. Cells were induced with 0, 2, 4 or 100 ng/ml doxycycline. 48 hours post induction (hpi), the cells displaying 1549_02_D06 were stained with CellTrace CFSE Cell Proliferation Kit (C34554, ThermoFisher Scientific) and mixed with unlabelled cells displaying 337_1_C08. The labelled and unlabelled cells were mixed equally and MACS purification was then performed on the mixed IgG display cell populations with 0.1, 1 or 10 nM PD-1-biotin as described in detail in Example 8a. The cells were then analysed by flow cytometry (FIG. 40) to determine the relative enrichment of the high affinity anti-PD1 IgG compared to the low affinity anti-PD1 IgG.

(318) The most efficient separation of cells displaying the high and low affinity anti-PD-1 antibodies occurred when the cells were induced with a limiting doxycycline concentration of 2 ng/ml to reduce the displayed antibody level and 0.1 nM PD-1-biotin was employed in a MACS purification. Here an enrichment of the population was achieved for the high affinity clone to 96% of the population (FIG. 40bi). When the cells were induced with 2 ng/ml doxycycline, the cell copy number 48 hpi was 11,000 and 20,000 respectively for the high and low affinity anti-PD-1 clones. In contrast, when the cell-lines were fully induced, MACS with 0.1 nM PD-1 only achieved a 77% enrichment of the high affinity clone (FIG. 40di). When the cells were fully induced with 100 ng/ml doxycycline, the cell copy number 48 hpi was 317,000 and 294,000 respectively for the high and low affinity anti-PD-1 clones. This provides further evidence, in addition to the data in Example 8, that a reduction in cell surface antibody display level (copy number) will increase the efficiency of separation of displayed antibodies with different affinities for their target. As the concentration of PD-1 used for MACS was increased, this led to less efficient enrichment of the high affinity clone as expected from the theoretical analysis in Example 8 (FIGS. 40bii and 40iii).

(319) In this Example, we have demonstrated the utility of an inducible promoter system for higher eukaryotic cell display, to enable both high copy display for self-interaction and aggregation propensity screening and low copy number display for the stringent selection of high affinity antibodies. Here we used limiting concentrations of inducer to give a low display level for the efficient separation of a clone with an affinity for its target of 2.9 nM from a second clone with an affinity of 74 nM for the same target. It is envisaged that by controlling the copy number of the displayed antibody on the cell surface and the concentration of the target antigen used for MACS separation, that it would increase the efficiency of the separate clones with an affinity (K.sub.D) for its target of less than 1 nM from clones with single digit nM affinity.

Example 9. Developability Enhancement by Selection for Optimal Interaction with Fc Receptors

(320) Example 6 demonstrated that it is possible to differentiate between eukaryotic cell displayed antibodies that have different cross-interaction properties. This was achieved by incubating the cells with labelled human serum and detecting binding by flow cytometry. Vesencumab, an antibody with known cross-interaction properties.sup.39, that did not proceed further than a Phase 1 clinical trial.sup.82 resulted in greater binding to human serum compared with the clinically approved antibody Nivolumab which is considered to be a well-behaved antibody. Vesencumab has a short half-life in vivo.sup.105. “Stickiness” or cross-interaction properties of an antibody is generally correlated with poor pharmacokinetics and a short half-life in vivo is thought to be caused by systemic clearance by the non-specific binding to disseminated tissue.sup.23,122,123.

(321) Antibody half-life in vivo is also critically dependent of the interaction between the IgG, via its CH2 and CH3 domains, and the neonatal Fc receptor (FcRn).sup.124. Long antibody half-life in circulation is achieved by the cellular internalization of IgG by unspecific pinocytosis or Fc receptor mediated uptake. Once internalized, IgG binds with high affinity to FcRn within the endosome at pH5-6, thereby protecting the IgGs from lysosomal degradation. Finally IgGs are transported to the cell surface and released back into circulation because the affinity between IgG and FcRn is very weak at physiological pH 7.4. The anti-IL12 Briakinumab.sup.20 did not show efficacy in a Phase III clinical trial to treat psoriasis and is known to have a short half-life in vivo. Briakinumab has positive charge patches, consisting of arginine and lysine residues, within its variable domain that can bind to a negative charge patch on FcRn resulting in increased affinity for binding to FcRn at pH7.4. The binding of Briakinumab to FcRn at pH7.4 has been shown to correlate with poor half-life in vivo in mice. Also, the anti-IL12 antibody Ustekinumab, clinically approved to treat Crohn's disease, does not possess a positive charge patch within its variable domain, is known to bind weakly to FcRn at pH7.4 and has superior in vivo half-life compared with Briakinumab. In this example we show it is possible to differentiate between antibodies with known differing affinities for FcRn at pH7.4 by higher eukaryotic cell display and the detection of FcRn binding by flow cytometry.

(322) Synthetic DNA encoding Briakinumab and Ustekinumab heavy and light variable domains (see Table 12 for sequences) were cloned into the pINT17-BSD targeting vector (see Example 1 for vector map and sequence) and DNA sequence confirmed.

(323) TABLE-US-00019 TABLE 12 Sequences of Briakinumab and Ustekinumab VH and VL chains Briakinumab QVQLVESGGGVVQPGRSLRLSCAASGFTFSSYGMHW VH VRQAPGKGLEWVAFIRYDGSNKYYADSVKGRFTISR DNSKNTLYLQMNSLRAEDTAVYYCKTHGSHDNWGQG TMVTVS (SEQ ID NO: 46) Briakinumab QSVLTQPPSVSGAPGQRVTISCSGSRSNIGSNTVKW VL YQQLPGTAPKLLIYYNDRQPSGVPDRFSGSKSGTSA SLAITGLQAEDEADYYCQSYDRYTHPALLFGTGTKV TVLGQP (SEQ ID NO: 47) Ustekinumab EVLQVQSGAEVKKPGESLKISCKGSGYSFTTYWLGW VH VRQMPGKGLDWIDIMSPVDSDIRYSPSFQGQVTMSV DKSITTAYLQWNSLKASDTAMYYCARRRPGQGYFDF WGQGTLVTVS (SEQ ID NO: 48) Ustekinumab DIQMTQSPSSLSASVGDRVTITCRASQGISSWLAWY VL QQKPEKAPKSLIYAASSLQSGVPSRFSGSGSGTDFT LTISSLQPEDFATYYCQQYNIYPYTFGQGTKLEIKR TA (SEQ ID NO: 49)

(324) Suspension adapted HEK293 cells were seeded at 5×10.sup.5 cells per ml in HEK FreeStyle 293 expression media one day before transfection. On day of transfection cells were centrifuged and re-suspended in a final volume of 10.sup.8 cells/ml in the manufacturer's electroporation buffer (Maxcyte Electroporation buffer, Thermo Fisher Scientific Cat. No. NC0856428). 100 μl was pulsed using an OC-100 cuvette (Maxcyte). Two days after transfection cells were seeded at 2.5×10.sup.5 cells per ml and 7.5 μg/ml Blasticidin was added. 57 days after transfection cells were labeled with biotinylated FcRn (50 nM) preconjugated with streptavidin PE (11 nM). Preconjugation was done in either of the staining buffer at different pH (PBS containing 1% BSA, pH7.4 or 20 mM MES containing 140 mM NaCl and 1% BSA, pH6.0). Briefly 1×10.sup.6 cells were spun down, washed once with either PBS at pH7.4 or 20 mM MES buffer at pH 6.0. Cell pellet was resuspended in 100 ul staining buffer and washed twice with 1.0 ml wash buffer (PBS containing 0.1% BSA, pH7.4 or 20 mM MES containing 140 mM NaCl and 0.1% BSA, pH6.0). Cell pellets were resuspended in 500 ul wash buffer containing viability dye 7AAD (5 ul per million cells). Cells were analysed using the Intellicyt flow cytometer. Dead cells were excluded during the analysis. Histogram plots were generated using FlowJo software (FIG. 30).

(325) At pH6, both Briakinumab and Ustekinumab bound to FcRn when the antibodies were displayed on the surface of HEK293 cells (FIG. 30). At pH 7.4, only Briakinumab displayed significant binding to FcRn, unlike Ustekinumab where no binding of FcRn was observed. This example demonstrates that it is possible to differentiate between clones with different binding affinities to FcRn at pH7.4 by higher eukaryotic cell display and analysis of FcRn binding by flow cytometry. This will enable the prediction of an antibody pharmacokinetic (PK) profile. Also when selecting from a library of cell displayed binders, this technique could be employed to eliminate clones that bind to FcRn at pH7.4 and therefore select for clones expected to possess a favourable PK profile. For example, binding of cell displayed antibodies to MACS beads could be achieved at pH6. Washing the beads at pH7.4 should elute the clones with low affinity at pH7.4 and these clones would be expected to display a longer half-life in vivo compared to the binding clones which retain binding at pH7.4.

(326) FcRn binding to IgG is known to be partially dependent on post-translational glycosylation of the IgG. Therefore, performing the selections described with higher eukaryotic cells has the advantage that authentic glycosylation will occur to enable FcRn binding, unlike display in unmodified lower eukaryotic cell display systems such as yeast.

(327) We further demonstrated that it is possible to isolate antibody that shows reduced binding to FcRn from a mixed antibody population. Two populations of higher eukaryotic cell clones expressing different anti-IL-12 antibodies, briakinumab and ustekinumab, respectively, were co-cultured. Cells expressing briakinumab were shown to bind to FcRn at pH 7.4 (FIG. 30). The binding could be due to its positive charge patch within its variable domain binding to negatively charged side chains of FcRn (see Example 6, in which briakinumab was shown to bind negatively-charged polyreactivity probes). Ustekinumab did not show binding to FcRn at pH 7.4 (FIG. 30). The mixed antibody expressing population was dual stained and enriched by FACS for cells displaying ustekinumab. FIG. 41. We thus demonstrated that it is possible to enrich FcRn non-binders from a mixed population which also contains FcRn binders. The same technique may be applied for more diverse mixtures of clones, e.g., a library containing millions of different antibodies.

(328) The antibody Fc region is known to elicit multiple effector functions including Fc receptor.sup.139 and complement binding.sup.140. By the creation of libraries of Fc domains followed by selection with known effector molecules such as Fc gamma receptors, NK receptors, FcRn or members of the complement cascade it would be possible to use higher eukaryotic cell display to select for variants which either enhance or reduce binding to effector function molecules. This would select for Fc variants that may enhance, reduce or silence antibody-dependent cellular cytotoxicity (ADCC), complement-dependent cytotoxicity (CDC) or increase antibody half-life in vivo. Fc variants could also be selected which bind to Fc receptors to increase their agonism of target receptors and therefore their potency.sup.141,142. Fc variants can reduce the stability of antibodies and negatively affect their biophysical properties.sup.134,143. An advantage of selecting for Fc variants by higher eukaryotic cell display is that variants that display to a high level on the cell surface can be selected for and these variants would be expected to possess superior biophysical properties.

Example 10. Selection of Optimal Developability Clones within a Selected Population by Mammalian Display (Mesothelin Selections)

(329) Here we demonstrate that it is possible to use higher eukaryotic (mammalian) cell display to select for antibodies with optimal developability potential from a highly diverse input panel of antibodies. A complex mixture of antibodies of different germline sequences was separated, allowing antibodies to be grouped and selected based upon their self-interaction properties.

(330) We show this by first generating an enriched panel of anti-mesothelin antibodies by antibody phage display selection, then cloning the antibodies into a mammalian cell display library and comparing the characteristics of binders from cell clones exhibiting high, medium or low surface presentation respectively.

(331) Mesothelin is an antigen over-expressed in several human tumours and is therefore of interest as a therapeutic antibody target to treat cancer. To generate an enriched population of anti-human mesothelin, two rounds of antibody phage display selection was performed where the antigen was directly immobilized on a polystyrene surface, as described previously.sup.8. A naïve human antibody library was employed (WO2015166272A2), described previously and constructed in a similar way to a previously described human naïve antibody library.sup.8. Selections were performed with VL lambda and VL kappa germ-lines separately and the round 1 output numbers were 2×10.sup.5 (average). Conversion to IgG format and cloning into the pINT17-BSD targeting vector was performed “en masse”, as described previously (WO2015166272A2). E. coli transformants for the lambda and kappa enriched populations were plated onto agar plates as described in Examples 4 and 5 to create a combined library size of 2.4×10.sup.6 clones (a 12-fold excess of the input antibody population). Transfection quality plasmid DNA was prepared encoding the anti-Mesothelin antibodies in the pINT17-BSD vector. The library was transfected with AAVS TALE nucleases as described above (Examples 2, 4 and 5) using a single OC-400 Cuvette (10.sup.8 cells total, MaxCyte, Cat. No. OC-400R10), on the HEK293 setting on the MaxCyte STXG2. The controls were transfected using OC-100 Cuvettes (MaxCyte, Cat. No. OC-100R10) on the same setting. After transferring the electroporated cells into an appropriate sized Erlenmeyer flask the cells were allowed to rest for 30 minutes before an appropriate amount of FreeStyle 293 Expression Media (LifeTech. Cat. No. 12338018) was added. The cells were resuspended and placed in an orbital shaking incubator set to 130 rpm, 37° C. and 5% CO.sub.2. Stable transfection efficiency was calculated as described above (Examples 4 and 5) to be 4.2% to give a total library size of 4.2×10.sup.6 stable antibody display HEK293 cell lines. After 19 days of blasticidin selection, FACS was carried out according as described above (Examples 4 and 5) using a BD Influx cell sorter. 100×10.sup.6 cells were incubated with Anti-Human Fc PE (1 μl per 1×10.sup.6 cells) (Cambridge Bioscience, Cat. No. 409304) and DAPI was added (1 μl/million cells) immediately before sorting. Three gates were drawn allowing for a low display level population (P4), a medium display level population (P6), and a high display level population (P5). The Influx sorter can only sort two populations at a time, therefore 50×10.sup.6 cells were sorted for gate P4 and P5, and the other 50×10.sup.6 cells were sorted for P6 (FIG. 31). Genomic DNA was extracted from the three populations, as described above (Examples 4 and 5), antibody genes cloned into pINT3 vector (WO2015166272A2) and the VH and VL genes sequenced.

(332) TABLE-US-00020 TABLE 13 Sequence analysis of the anti- Mesothelin mammalian display selected clone populations. The populations gated in FACS for low, medium and high display levels (FIG. 30) were DNA sequence analysed and compared with the input antibody generation used to produce the starting anti-Mesothelin display library. Germ-lines and VL CDR1, 2 and 3 and VH CDR1 and 2 followed IMGT nomenclature whereas VH CDR3 was defined by the Kabat numbering system. Low Medium High Input Display Display Display Total sequenced 79 89 92 85 Unique VH 56 (71%) 38 (49%) 62 (67%) 63 (74%) CDR3 Unique VL 49 (62%) 32 (36%) 57 (62%) 48 (56%) CDR3 Unique 69 (87%) 43 (48%) 79 (89%) 71 (84%) VH + VL CDR3 VH germ-lines 14 10 10 12 VL.sub.K germ-lines 11 12 10 11 VLλ germ-lines  3  2  5  5

(333) Sequence analysis revealed great diversity within all sub-populations (Table 13). In order to ensure that this was not due to a sampling issue, we focused the analysis on clones which appeared more than once to determine whether “binning” of different clone groups into the different display groups was occurring. Thus an overlap analysis was performed for antibodies where duplicate clones had been identified (based on the VH and VL CDR3 sequence) in the low, medium and high display level populations (Table 14). There were 13 such clones in the high display level population and 28 clones which appeared multiple times in the low display level population with no overlap between them. The group of clones identified as “medium expressors” had 11 sequences that appeared more than once and again these were not found in the other groups. There were 4 clones which appeared in the “medium” group which overlapped in 3 cases with the low group and in one case with the high group. This result is evidence that sorting based on display level in mammalian cells selects for specific sub-populations of sequence that are unique to that group.

(334) TABLE-US-00021 TABLE 14 Antibody clone overlap analysis for the anti-Mesothelin antibodies pooled according to high, medium and low HEK293 cell display levels. Mammalian cell display selected anti-Mesothelin antibodies were pooled according to cell surface display level (FIG. 31). Clones from the high, medium and low display level populations were DNA sequenced and the unique clones (based on VH and VL CDR3 sequences) that appeared more than once were characterised according to the number of appearances in the high, medium or low display groups. The numbers in the Table indicate the number of appearances. No of occurrences No. of occurrences No of occurrences Clone VH VL in the High in the Medium in the Low number Germline Germline display group display group display group 1 IGHV1-69 IGKV2-28 3 0 0 2 IGHV4-34 IGLV2-8 2 0 0 3 IGHV1-46 IGKV1-5 2 0 0 4 IGHV1-46 IGKV6-21 2 0 0 5 IGHV3-9 IGLV3-19 2 0 0 6 IGHV3-9 IGLV3-19 2 0 0 7 IGHV1-46 IGKV3-15 3 0 0 8 IGHV3-9 IGLV3-19 1 1 0 9 IGHV3-9 IGLV3-19 1 1 0 10 IGHV3-9 IGLV3-19 2 0 0 11 IGHV1-2 IGKV3-15 2 0 0 12 IGHV3-9 IGLV3-19 3 0 0 13 IGHV3-9 IGLV3-19 1 1 0 14 IGHV3-9 IGLV3-19 0 0 2 15 IGHV5-51 IGKV1-39 0 0 6 16 IGHV1-3 IGKV1-39 0 0 10 17 IGHV1-2 IGKV4-1 0 0 4 18 IGHV1-18 IGKV1-33 0 0 3 19 IGHV1-46 IGKV2-30 0 0 3 20 IGHV1-69 IGKV1-39 0 0 3 21 IGHV1-69 IGKV1-39 0 0 2 22 IGHV1-69 IGKV4-1 0 0 5 23 IGHV3-30/33rn IGKV4-1 0 0 4 24 IGHV3-9 IGLV3-19 0 0 3 25 IGHV3-9 IGLV3-19 0 0 2 26 IGHV1-46 IGKV1-39 0 0 2 27 IGHV1-69 IGKV1-39 0 1 3 28 IGHV5-51 IGKV4-1 0 0 2 29 IGHV3-30/33rn IGKV2-28 0 0 2 30 IGHV1-69 IGKV4-1 0 0 6 31 IGHV3-9 IGKV1-39 0 2 0 32 IGHV3-9 IGLV3-19 0 2 0 33 IGHV3-9 IGLV3-19 0 2 0 34 IGHV1-69 IGKV2-28 0 2 0 35 IGHV3-9 IGLV3-19 0 2 0 36 IGHV3-9 IGKV3-20 0 2 0 37 IGHV3-9 IGLV3-19 0 3 0 38 IGHV3-9 IGLV3-19 0 2 0 39 IGHV3-9 IGLV3-19 0 2 0 40 IGHV3-9 IGLV3-19 0 2 0 41 IGHV3-9 IGLV3-19 0 2 0

(335) The distinctness of the cell populations selected on the basis of surface presentation was confirmed by an in-depth analysis of the antibody populations using next generation sequencing (NGS). In brief, paired variable domain genes of antibodies from the high, medium and low display level populations and from the input population were amplified, sequenced and demultiplexed using asymmetrical barcodes,.sup.125,126 and the output BAM files were then converted to FASTQ file format to enable antibody sequence analysis with Geneious Biologics software.

(336) A total of 19792 paired VL and VH gene reads resulted from this analysis. After de-multiplexing the populations, this resulted in 2998, 1516, 6180 and 9098 CCS reads for the input and high, medium and low display level populations respectively. Antibody sequences were annotated to map their framework and complementarity determining regions (CDRs) and heavy and light chain germ-lines were assigned according to the IMGT database.sup.127. A summary of the results of this analysis for the clones which had annotated VH and VL CDR3 sequences with no stop codons is in Table 15. The antibody input population, which was pre-selected by performing two rounds of antibody phage display selection against Mesothelin with a naïve antibody phage display library.sup.8, was highly diverse with 51% of the clones being VH CDR3 sequence unique and 88% of the clones being VH and VL CDR3 sequence unique. This input population was also antibody germ-line diverse with 19, 18 and 10 VH, VLκ and VLλ germ-line identified respectively per 1000 clones sequenced. This diversity both in terms of CDR3 sequences and germ-lines was reduced when analysing the high, medium and low mammalian display gated populations. The diversity of the low display level population was markedly reduced with the number of VH CDR3 unique antibodies dropping to less than 4% and the number of VH, VLκ and VLλ germ-line per 1000 clones sequenced dropping to 4, 2 and 3 respectively. This reduction in antibody diversity compared with the input population indicates that certain clones are being enriched in the display level gated antibody populations, thus increasing the redundancy in the populations.

(337) The ratio of the VLκ and VLλ germ-lines was also examined in the different populations. Although the starting input population had an almost 2-fold excess of VLλ compared with VLκ antibodies, this ratio was reversed for the low display level gated population which had an over four-fold excess of VLκ antibodies. These results indicate that the low display group is enriching for a particular sub-set of antibodies with particular light chain germ-line sequences.

(338) TABLE-US-00022 TABLE 15 Sequence analysis of the anti- Mesothelin mammalian display selected clone populations generated by PacBio sequencing. The populations gated in FACS for low, medium and high display levels (FIG. 30) were DNA sequence analysed and compared with the input antibody generation used to produce the starting anti-Mesothelin display library. Germ-lines and VL CDR1, 2 and 3 and VH CDR1 and 2 followed IMGT nomenclature whereas VH CDR3 was defined by the Kabat numbering system. Low Medium High Input Display Display Display Total sequenced and 1132  7378    3837  968  annotated Unique VH CDR3 575 (51%) 284 (3.8%) 547 (14%) 236 (24%) Unique VL CDR3 535 (47%) 196 (2.7%) 446 (11.6%) 185 (19%) Unique 995 (88%) 1170 (15.9%) 2204 (57%) 637 (66%) VH + VL CDR3 Total VH germ-lines 22 14   18 16 Total VL.sub.K germ-lines 20 19   19 17 Total VLλ germ-lines 11 4   17  9 VH germ-lines   19.4 1.9   4.7   16.5 per 1000 clones VL.sub.K germ-lines   17.7 2.6   5.0   17.6 per 1000 clones VLλ germ-lines   9.7 0.5   4.4   9.3 per 1000 clones VL.sub.K/VLλ ratio   0.54 4.4    0.82   1.3

(339) To enable a more detailed germ-line analysis of the input population and the clones that were selected using higher eukaryotic mammalian display on the basis of low, medium and high display levels, antibodies were assigned to their closest matching VH, VLκ and VLλ germ-lines and the frequency of occurrence of the germ-lines were calculated for each group. The results are shown in both tabular format (Tables 16, 17 and 18) and as histogram plots (FIGS. 42, 43 and 44). This indicated that in the populations sequenced, several germ-lines were either enriched or absent in the low or high display level groups, indicating that mammalian display selection on the basis of antibody display levels enables the selective enrichment of antibody germ-lines. For example, the IGHV3-23 and IGHV1-2 germ-lines were enriched by 105 and 9-fold respectively in the high display group compared with the low display group. The IGHV3-53, IGHV3-21 and IGHV4-34 germ-lines occurred at a frequency of 2.7%, 1.1% and 0.9% respectively in the high display group but were completely absent in the low display group. The IGHV3-23 germ-line is considered to be a well-behaved germ-line and is represented in several therapeutic antibodies that have been approved for clinical use including alirocumab, bevacizumab, certolizumab, daratumumab, denosumab, dupilumab, emicizumab, ocrelizumab, ranibizumab, and siltuximab. The IGHV3-21 and IGHV3-53 germ-lines are considered to be well-behaved have been included in human synthetic antibody library designs.sup.12. Five VH germ-lines were found to be enriched in the low display group: IGHV1-69D, IGH3-30, IGHV5-51, IGHV1-58 and IGH3-64D. Previously it has been observed that the IGH3-30 germ-line, when paired with certain light chains, is prone to self-interaction as determined by dynamic light scattering measurements.sup.12.

(340) Analysis of the frequency of occurrence of the antibody light chain germ-lines also showed enrichment of certain germ-lines in the high display group compared with the low display group. For example, IGKV1 D-13 and IGKV3-19 occurred at a frequency 3.5% and 6% respectively in the high display group but were completely absent in the low display group. The IGKV1-12, IGKV1-17, IGKV2-28, IGKV2D-29 and IGKV6-21 germ-lines were enriched by 11-, 22-, 6-, 11-, and 132-fold respectively in the high display group compared with the low display group. The IGKV2-28 and related germ-line IGKV2D-29 are considered to be well-behaved and is represented in several antibodies either approved for clinical use or under-going clinical evaluation including cantuzumab, dupilumab, lucatumumab, mogamulizumab, obinutuzumab, sevirumab, tenatumomab and zatuximab. Four VLκ germ-lines were found to be enriched in the low display group: IGKV1-16, IGKV1-33, IGKV2-30 and IGKV4-1. The VHλ germ-line analysis is less clear because of the domination of a single VHλ germ-line in the input population (IGLV3-19). Nevertheless the following VHλ germ-lines were enriched in the high display group: IGLV1-47, IGLV2-11, IGLV2-14, IGLV2-23, IGLV2-8, IGLV3-10 and IGLV6-57. The Vλ1-47 and Vλ2-14 light chains are known to be well behaved and have been included as the scaffold in synthetic library designs.sup.12,128. IGLV2-14 is the light chain present in the clinically approved anti-PCSK9 antibody Evolocumab which is a well behaved antibody.sup.2

(341) This analysis enables the “binning” of human antibody germ-lines into either the mammalian display high or low display groups. It was previously shown that high level display correlates with well-behaved biophysical properties in terms of a low propensity for protein self-interaction (Example 3). Therefore, the observation of enrichment of certain germ-lines in the mammalian display high display level group is due to their inherent biophysical properties. Therefore, mammalian display enables the enrichment and identification of antibody germ-lines with well-behaved biophysical properties in terms of a low propensity for self-interaction.

(342) TABLE-US-00023 TABLE 16 Variable heavy (VH) germ-line analysis of the anti- Mesothelin mammalian display selected clone populations. Antibodies were assigned to the closest matching human VH germ-line sequence and the frequency of occurrence in the input and low, medium and high mammalian display gated populations is shown. The last column shows a ratio of occurrence of the germ-lines for the high and low display level gated populations where germ-lines were found in both populations. Frequency of occurrence (%) High/Low VH Input Low Medium High Ratio IGHV1-18 1.9 8.1 1.8 3.5 0.43 IGHV1-2 5.5 0.42 6.7 3.9 9.34 IGHV1-3 3.0 13 5.6 4.3 0.34 IGHV1-46 5.8 5.6 9.7 13 2.40 IGHV1-58 1.4 5.9 0 0.00 IGHV1-69 9.3 28 2.1 12 0.44 IGHV1-69D 5.4 5.8 0 0.83 0.14 IGHV2-70D 0.18 0 0 0 — IGHV3-11 0.18 0 0 0.31 — IGHV3-20 61 17 54 47 2.75 IGHV3-21 0.18 0 0 1.1 — IGHV3-23 1.1 0.04 0.57 4.2 104 IGHV3-30 1.5 4.5 3.3 0.21 0.05 IGHV3-33 0.71 1.1 2.4 4.2 3.9 IGHV3-43 0.18 0.11 0.57 0 — IGHV3-48 0.09 0 0 0 — IGHV3-53 0.80 0 2.19 2.7 — IGHV3-64D 0 0.84 0 0 — IGHV3-66 0 0 0.03 0 — IGHV3-7 0.09 0 0.03 0 — IGHV3-72 0.09 0 0 0 — IGHV4-34 0.09 0 0.16 0.93 — IGHV5-51 2.5 15 2.50 1.34 — IGHV6-1 0.18 0 2.32 0.21 — IGHV7-81 0.09 0 0.08 0 0.43

(343) TABLE-US-00024 TABLE 17 Variable heavy (VL.sub.K) germ-line analysis of the anti- Mesothelin mammalian display selected clone populations. Antibodies were assigned to the closest matching human VL.sub.K germ-line sequence and the frequency of occurrence in the input and low, medium and high mammalian display gated populations is shown. The last column shows a ratio of occurrence of the germ-lines for the high and low display level gated populations where germ-lines were found in both populations. Frequency of occurrence (%) High/Low VL.sub.K Input Low Medium High Ratio IGKV1-12 0.76 0.02 1.85 0.18 11 IGKV1-13 0.25 0 0.75 0 — IGKV1-16 1.76 3.47 1.45 0.18 0.05 IGKV1-17 0.50 0.02 0.41 0.36 22 IGKV1-27 0.76 1.16 0.87 0.73 0.6 IGKV1-33 1.76 7.14 2.72 1.28 0.18 IGKV1-39 56.2 49.2 42.2 61.2 1.2 IGKV1-5 3.02 0.70 4.05 0.18 0.26 IGKV1-6 0 0.03 0 0 — IGKV1-8 0 0 0 0.18 — IGKV1-9 2.02 0.56 0.29 1.28 2.3 IGKV1D-13 1.76 0 4.05 3.46 — IGKV1D-16 0.25 0.05 0.12 0 — IGKV2-24 0.25 0.07 0 0 — IGKV2-28 6.30 2.19 11.93 13.66 6.2 IGKV2-30 3.02 4.19 5.50 0 — IGKV2D-29 0.25 0.05 0 0.55 11 IGKV3-11 0.50 0 0.12 0 — IGKV3-15 6.30 4.39 10.19 3.64 0.83 IGKV3-19 0 0 0 6.01 — IGKV3-20 1.26 0.40 4.00 1.09 2.7 IGKV3D-15 0 0.02 0.06 0 — IGKV4-1 12.85 26.36 8.63 3.83 0.15 IGKV6-21 0.25 0.02 0.87 2.19 131

(344) TABLE-US-00025 TABLE 18 Variable heavy (VLλ) germ-line analysis of the anti- Mesothelin mammalian display selected clone populations. Antibodies were assigned to the closest matching human VLλ germ-line sequence and the frequency of occurrence in the input and low, medium and high mammalian display gated populations is shown. The last column shows a ratio of occurrence of the germ-lines for the high and low display level gated populations where germ-lines were found in both populations. Frequency of occurrence (%) High/Low VLλ Input Low Medium High Ratio IGLV1-40 0.14 0 0 0 — IGLV1-44 0.14 0 0.33 0 — IGLV1-47 0 0 0.28 0.72 — IGLV1-50 0 0 0.24 0 — IGLV2-11 1.09 0.07 0.24 2.63 36 IGLV2-14 0.41 0 0.14 0.48 — IGLV2-18 0 0 0.09 0 — IGLV2-23 0 0 0.85 0.24 — IGLV2-8 0.14 0 0.19 1.67 — IGLV3-1 1.50 0.29 0.95 0 — IGLV3-10 0.14 0 0.47 1.43 — IGLV3-19 94.55 97.64 90.57 90.21 0.9 IGLV3-21 0 0 0.09 0 — IGLV3-25 1.09 1.99 2.13 1.19 0.60 IGLV3-27 0 0 0.24 0 — IGLV3-9 0.14 0 0.71 0 — IGLV5-45 0 0 0.05 0 — IGLV6-57 0.68 0 2.42 1.43 —

(345) TABLE-US-00026 TABLE 19 Antibody clone cluster analysis for the anti-Mesothelin antibodies from the high,   medium and low HEK293 cell display levels. Mammalian cell display selected  anti-Mesothelin antibodies were pooled according to cell surface display  level (Figure 31). The high (H), medium (M) and low (L)display level populations were DNA sequenced by PacBio NGS and the top 10 most abundant   clones in each group were identified and are depicted in this table with the VH and VH CDR3 sequences shown in single letter amino acid code separated by  a dash. Antibodies which were previously identified by Sanger sequencing of   individual clones are assigned a clone name shown in the second column. The percentage  abundance of each clone in the high, medium and low display groups is shown.  The cells are shaded if an antibody is present in more than one display level group.  Selected clones were expressed, purified and tested in an AC-SINS assay.sup.40  and the wave-length shift (ΔAC-SINS) in nm is shown. HPLC-SEC  retention volume is shown in the last column (nd, not done). HPLC- SEC Ren- Medi- ten- Heavy Light High um Low tion Dis- chain  chain Dis- dis- Dis- In- ΔAC- Vol- Clone play germ- germ- play play play put SINS ume Clone  name Group VH-VL CDR3 line line (%) (%) (%) (%) (nm) (ml) 1 932_01_D10 H DSRPPYYGMDV    IGHV1- IGKV2- 0.00 0.47 1 nd (SEQ ID NO: 50)-   69 28 MQALQTPPT  (SEQ ID NO: 51) 2 932_01_C12 H DGRGGFDY   IGHV3- IGKV1- 1.69 0.00 0.00 0.00 2.5 1.53 (SEQ ID NO: 52)- 43 39  QQFNSYHLLT    (SEQ ID NO: 53) 3 932_01_A03 H GRSSVIDYGMDV    IGHV3- IGLV3- 0.00 0.00 2 1.54 (SEQ ID NO: 54)-    20 19 NSRDSSGNHVV (SEQ ID NO: 55) 4 N/A H DRVAATHYYYYGMDV    IGHV3- IGKV1- 0.00 0.00 nd nd (SEQ ID NO: 56)-     30 39 QQSYGSPFT (SEQ ID NO: 57) 5 N/A H GLLEKGAFDI    IGHV1- IGKV1- 0.99 0.00 0.00 0.00 nd nd (SEQ ID NO: 58)- 46 39  QQSYSTPQT    (SEQ ID NO: 59) 6 N/A H DSRPPYYGMDV    IGHV1- IGKV2- 0.00 0.00 nd nd (SEQ ID NO: 60)- 69 28 QQSYNSRPYT   (SEQ ID NO: 61) 7 932_01_A06 H TSPYSGSYNN    IGHV4- IGLV2- 0.79 0.00 0.00 0.00 3 1.59 (SEQ ID NO: 62)- 34 8 SSYGGNYKYL    (SEQ ID NO: 63) 8 932_01_F02 H GLSSTWAGGAFDI    IGHV3- IGLV3- 0 1.41 2 1.54 (SEQ ID NO: 64)- 20 19 NSRDSSGNHVV     (SEQ ID NO: 65) 9 N/A H GAHSGYDSDFDY    IGHV1- IGKV1- 0.00 0.00 nd nd (SEQ ID NO: 66)- 18 39 QQYNSYPLT    (SEQ ID NO: 67) 10 932_01_B01 H VSGSSNHAFDI    IGHV1- IGKV1- 0.00 0.00 1 1.55 (SEQ ID NO: 68)- 2 2 HQSSSFPLT   (SEQ ID NO: 69) 11 931_01_F01 M DTSSRYAGGAFDI    IGHV1- IGLV1- 0.00 0.71 6 1.58 (SEQ ID NO: 70)- 69D 50 NSRDSSGNHVV   (SEQ ID NO: 71) 12 932_01_C04 M DTGSSARGGDFDY    IGHV3- IGLV3- 0 0.55 2 1.54 (SEQ ID NO: 72)- 20 19  NSRDSSGNHVV    (SEQ ID NO: 73) 13 931_01_B10 M DERYYGDMDV    IGHV1- IGKV2- 0.00 0.31 nd nd (SEQ ID NO: 74)-  69 28 MQGLQTPRT    (SEQ ID NO: 75) 14 930_01_A01 M DYSSGWSGDAFDI    IGHV1- IGKV1- 0.39 0.08 nd nd (SEQ ID NO: 76)- 3 39 NSRDSSGNHVV    (SEQ ID NO: 77) 15 931_01_B02 M RGTRGYYYYGMDV    IGHV1- IGKV1- 2.36 23 1.61 (SEQ ID NO: 78)- 69D 39 QQSYNSRPYT     (SEQ ID NO: 79) 16 N/A M TNNGFIDY   IGHV6- IGKV2- 0.00 1.12 0.00 0.08 nd nd (SEQ ID NO: 80)- 1 30  SQASHWPYT     (SEQ ID NO: 81) 17 N/A M GRGRRYFDL   IGHV1- IGKV1- 0.00 0.55 0.00 0.00 nd nd (SEQ ID NO: 82)- 2 39  QQSYSTPRT    (SEQ ID NO: 83)  18 931_01_E03 M DGDEGELGAFDI    IGHV1- IGKV1- 0.00 0.42 0.00 0.00 nd nd (SEQ ID NO: 84)-     46 39 QQSYGSPFT (SEQ ID NO: 85) 19 930_01_D09 L GDNNYYFDY    IGHV1- IGKV4- 0.00 0.00 1.88 0.00 5 1.55 (SEQ ID NO: 86)- 69 1  QQYYSRPIT   (SEQ ID NO: 87) 20 930_01_A09 L ETGEGRWELLGY    IGHV1- IGKV2- 0.00 0.00 2.22 0.00 4 nd (SEQ ID NO: 88)- 46 30 MQGTHWPRT     (SEQ ID NO: 89) 21 N/A L LSHTAPLVDY    IGHVS- IGKV3- 0.00 0.63 nd nd (SEQ ID NO: 90)- 51 15 QQYSKWPLT   (SEQ ID NO: 91) 22 930_01_C02 L AIAPRRYYYGMDV     IGHV1- IGKV1- 0.00 0 1.18 nd nd (SEQ ID NO: 92)- 69 39 QQSYSTPRT     (SEQ ID NO: 93) 23 930_01_Al2 L AIAPRRYYYGMDV     IGHV1- IGKV1- 0.00 0.00 2.90 0.08 20 1.61 (SEQ ID NO: 94)- 69 39 QQYNSYPLT    (SEQ ID NO: 95) 24 930_01_A08 L DGYNSDY   IGHV1- IGKV1- 0.00 0.00 3.70 0.00 1 1.55 (SEQ ID NO: 96)- 39 33  QQYYSKPLT    (SEQ ID NO: 97) 25 930_01_C07 L RRYNWDYDYVDV    IGHV5- IGKV1- 0.00 0.00 5.79 0.08 25 1.56 (SEQ ID NO: 98)- 51 39 QQSYSTPRT    (SEQ ID NO: 99) 26 930_01_A05 L DKPVGSSGWYPFDY    IGHV1- IGKV1- 0.00 0.00 10.50 0.00 nd nd (SEQ ID NO: 100)- 3 39 QQSYSTPYT     (SEQ ID NO: 101) 27 930_01_B02 L QINWGYYFDY    IGHV1- IGKV4- 0.00 0.00 11.68 0.00 22 1.73 (SEQ ID NO: 102)- 69 1 MQALQTPPT   (SEQ ID NO: 103)

(346) Cluster analysis was performed to generate a list of sequence unique VH and VL CDR3 clones with their frequency of occurrence for each of the four anti-Mesothelin antibody populations. The top 10 clones, by frequency of occurrence, are shown in Table 19 for the high, medium and low display level anti-Mesothelin antibodies FACS gated as depicted in FIG. 31. For comparison, the frequency of occurrence of the clones in the input and alternate populations is also shown. As was previously shown with a more limited clone set, there was no over-lap between the top 7 most abundant clones in the high display level group with the low display group. The 7 most abundant clones in the high display level group were completely absent in the low display level group. The 9 most abundant clones in the low display level group were completely absent in the high display level group. For the most abundant medium display level group there was some over-lap with the high and low display level groups: three clones were only present in the medium display level group, two clones were only present in the medium and high display level groups and three clones were present in all three display level groups. For clones originating from the high display level group where there was some over-lap with the low display level group (e.g. clones 8, 9 and 10) there was an enrichment of these clones in the high display level group by a factor of over 41-fold.

(347) The binning of the antibodies into the high, medium and low display level groups according to their sequence indicates that separation based on eukaryotic cell antibody display level is due to the properties of that antibody as determined by their polypeptide sequences. The frequency of occurrence of antibodies as determined by PacBio NGS analysis agreed well with the results from sequencing a more limited set of clones by Sanger sequencing. By PacBio NGS 9, 8 and 6 of the top 10 most abundant clones in the low, medium and high display level groups had been previously identified by Sanger sequencing.

(348) Selected antibodies from the high and low display level groups were then expressed and purified and their biophysical properties determined by AC-SINS.sup.41 or HPLC-SEC. The AC-SINS results are shown in the penultimate column of Table 19. The average AC-SINS wavelength-shift of clones originating from the high or low display groups was 1.9 and 15.2 respectively. The significantly higher average AC-SINS wavelength shift observed from the clones originating from the low display level group compared with the high display level group indicates that the antibodies the low display level group are more prone to self-interaction and aggregation than the antibodies in the high display level group. Selected antibodies were also examined by HPLC-SEC. FIG. 45 shows an example of four clones: one from the high display group (932_01_A03) and three from the low display group (930_01_A12, 930_01_B02 and 930_01_C12) which were expressed and purified and analysed by HPLC-SEC. This shows that antibody 932_01_A03, isolated from the high display level group, is 96% monomeric. In contrast, the antibody 930_01_C12 is 82% monomeric with 18% dimer and also shows some retardation on the column with a later elution profile indicating some non-specific interaction with the column. This clone also included a small fragment (16 kDa molecular weight) indicating that it may have fragmented during expression or purification. Two additional clones from the low display group (930_01_A12, 930_01_B02) also displayed a delayed retention time on the column indicating that the antibodies were non-specifically absorbing to the column matrix. The HPLC-SEC retention volume (last column, Table 19) is a measure on non-specific antibody interacting with the column matrix. Here a larger retention volume can give rise to a broader HPLC-SEC peak and indicate non-specific interaction with the HPLC-SEC column matrix. The average HPLC-SEC retention volume in the high display group clones was 1.55 ml whereas the average HPLC-SEC retention volume in the low display group clones was 1.6 ml proving evidence that the clones in the low display group were more prone to non-specific HPLC-SEC column matrix interaction than the clones in the high display group. One clone in the low display level group (930_01_C07) precipitated after attempted concentration by ultra-filtration. A second clone in the low display level group failed expression and purification. None of the clones originating from the high display level group demonstrated any problems during their expression, purification and subsequent analysis. This included the ability to be concentrated by ultra-filtration to greater than 12 mg/ml in PBS (pH7.4).

(349) Therefore, this example has exemplified that it is possible to use higher eukaryotic mammalian cell display to separate antibodies present within a complex mixture, including antibodies with different germ-line sequences, based upon their self-interaction and cross-interaction properties.

Example 11. Selecting Developable Clones on the Basis of Rapid Accumulation on Cell Surface

(350) In the examples above we have noted the relationship between constitutive antibody display level on the surface of a stable population of higher eukaryotic cells and the biophysical properties of that antibody. Here we also note display level differences are observed 24 hours after transfection. In Example 5, FIG. 10 we observed display level differences between Bococizumab and the parental antibody 5A10i at the 1-day post transfection (1 dpt) time point (FIG. 10, left panel). This initial rate of accumulation difference of displayed antibody correlates with the display levels differences in the stable cell lines (FIG. 10, right panel, 21 dpt) at equilibrium and the biophysical properties of the antibodies. It was shown in Example 5 that Bococizumab has high self-interaction and cross-interaction properties compared with the parental antibody 5A10-i as measured in the AC-SINS assay and HPLC-SEC. Bococizumab displayed at a lower level both at the 24-hour post transfection transient phase and the stable cell line phase, 21 days post-transfection (FIG. 10), compared with the parental antibody 5A10-i. Therefore, the initial transient antibody cell display level is predictive of the antibody cell display level in the stable cell line, created after several days of drug selection.

(351) We have therefore shown a relationship between the rate of display of an antibody on the surface of a higher eukaryotic cell surface and the biophysical properties of the antibody. The demonstrated relationship between the initial transient display of antibodies and their biophysical properties will be useful for enriching antibodies with superior biophysical characteristics from a population or library of antibodies at an earlier stage compared with the generation of stable cell lines. It will also be useful for the screening of individual sets of antibodies by individual monoclonal transfection. Also differences in the initial rate of cell surface display of antibodies may be useful where antibody expression is regulated by an inducible promoter and selection of the population on the basis of cell surface display level is initiated post induction.

Example 12. Selection for Developability of IgG Based on Surface Presentation Level in CHO Cell Library

(352) IgG antibodies can be displayed on the surface of CHO cells, with the level of display correlating with the biophysical properties of the individual antibody including its propensity to self-interact and aggregate. In this Example we show antibody display on the surface of CHO cells after two alternative nuclease-mediated antibody gene integration methods, and we demonstrate that the level of display correlates with the biophysical properties of the displayed antibodies.

(353) A gene targeting vector was designed and used for integration of an antibody gene expression cassette into intron 1 of the CHO AAVS gene, orthologous to the human AAVS genomic locus described in Example 1.sup.93. TALEN nucleases targeting intron 1 of the CHO AAVS gene locus created a double strand break at this location. For comparison, CRISPR/Cas9 nucleases were also tested in a parallel method.

(354) Antibody display on CHO cells was successfully achieved using both the TALEN-directed and the CRISPR/Cas9-directed integration method. The best display levels were achieved using the TALEN nuclease pair or the CRISPR 1 design. The alternative CRISPR designs (CRISPR 2 and CRISPR 3) were also successful although slightly lower levels of antibody display were observed. FIG. 48.

(355) We further demonstrated that the cell surface levels of antibodies displayed on CHO cells correlated with the biophysical characteristics of the displayed antibody, mirroring the results seen with HEK293 cells. Selected antibodies were cloned into the pINT17-BSD-CHO targeting vector: MEDI-1912, MEDI-1912-STT, bococizumab, 884_01_G01.

(356) MEDI-1912 (anti-NGF) has poor biophysical properties in terms of its propensity to self aggregate.sup.7. Reduced cell surface display levels were observed on the CHO cells for MEDI-1912 compared with the solubility enhanced daughter clone MEDI1012-STT. 884_01_G01 is a derivative of bococizumab and was identified by mutagenesis and mammalian display selection on the basis of display level and retained binding to PCSK9 (Example 5). 884_01_G01 was previously shown to be “well behaved” in terms of giving a low AC-SINS score and a monomeric peak by HPLC-SEC. Reduced CHO cell surface display levels were also observed for bococizumab compared with the improved daughter clone 884_01_G01. FIG. 49.

(357) The example therefore confirms the utility of CHO cells for the display of antibodies and the differentiation of antibodies with different biophysical properties based on their cell surface display levels. The antibodies with different biophysical properties are able to be well separated by flow cytometry, enabling the isolation of antibodies with better biophysical properties by CHO cell display within complex mixtures.

(358) Materials & Methods

(359) DNA encoding the CHO AAVS left and right homology arms was PCR amplified from suspension adapted CHO cell line genomic DNA, isolated from CHO cells with primer pairs 3165/3166 and 3169/3170 (Table 20) to generate PCR products of sizes 1.2 kb and 1.35 kb respectively. These PCR products were then used as templates to generate the left and right CHO AAVS homology arms while simultaneously knocking out various restriction site required for subsequent antibody gene cloning. The CHO left AAVS homology arm was created, with mutation of the existing NcoI, NheI, NsiI and DraIII sites by PCR amplification with the PCR primer pairs 3195/3196; 3197/3198; 3199/3200 using the CHO AAVS left arm template described above to create three PCR products with sizes of 241 bp, 576 bp and 142 bp respectively. The fragments were then assembled using primers 3199 and 3196 (Table 20) to generate a product of 893 bp in size. An additional assembly PCR reaction was performed except that primer 3123 was employed in place of 3196 to give a slightly shorter AAVS-left homology arm (880 bp). The reason for generating this slightly shorter left homology arm is to avoid the CRISPR 2 and 3 design recognition sites within the targeting vector (FIG. 46). The assembled AAVS left homology arms were digested with AsiSI and NsiI restriction enzymes and cloned into the pINT17-BSD vector pre-cut with AsiSI and NsiI. To PCR amplify the CHO AAVS homology arm and knock-out the BcII restriction two separate PCR reactions were performed with primer pairs 3201/3202 and 3203/3204 to create PCR products of sizes 707 bp and 208 bp respectively which were then PCR assembled with primers 3201 and 3204 to generate a PCR product of size 900 bp. The assembled AAVS right homology arms were digested with BstZ171 and SbfI restriction enzymes and cloned into the pINT17-BSD vector pre-cut with BstZ171 and SbfI and then cloned into BstZ171 and SbfI cut pINT17-BSD cut vector which already harboured the left CHO AAVS homology arm. This then created the CHO targeting vector pINT17-BSD-CHO, which is identical to pINT17-BSD (FIG. 1) except that the human AAVS homology arms have been replaced by CHO AAVS homology arms.

(360) TABLE-US-00027 TABLE 20 Primer sequences to enable amplification of the CHO AAVS left and right homology arms and knock-out the restriction sites and CRISPR guide RNA primers. Restriction sites are underlined and mutations to knock out NheI, NcoI, BclI and DraIII restriction sites shown in bold. Name Primer sequence Description 3165 GGTGCTCGACTCCACCAA CHO-AAVS-Left-F (SEQ ID NO: 104) 3166 GATGGAAGTTGCCATGAAAGA CHO-AAVS-Left-R (SEQ ID NO: 105) 3169 TCTTGTATTGCCGGGATCCTTC CHO-AAVS-Right-F (SEQ ID NO: 106) 3170 TAACTCCCAGCCCTACCTACTC CHO-AAVS-Right-R (SEQ ID NO: 107) 3195 CTCCACCTACCACCTCATGGACTATATTTG CHO-AAVS-left-F1-exNcoI (SEQ ID NO: 108) 3196 TTTTTTATGCATCTTATGCCAGCTTTTGGATGACGG CHO-AAVS-left-R1-exNsiI (SEQ ID NO: 109) 3197 CTCCTCTGAGTCTAGCCAGGCC CHO-AAVS-left-F2-exNheI (SEQ ID NO: 110) 3198 CAAATATAGTCCATGAGGTGGTAGGTGGAG CHO-AAVS-left-R2-exNcoI (SEQ ID NO: 111) 3199 TTTTTTGCGATCGCGATGGCTTACATCCCGTGCCTTTC CHO-AAVS-left-F3 + AsiS1-exDraIII (SEQ ID NO: 112) 3200 GGCCTGGCTAGACTCAGAGGAG CHO-AAVS-left-R3-exNheI (SEQ ID NO: 113) 3201 TATATTGTATACGGCGCGCCTGTCAGGGACAAGATTAGTCACAG CHO-AAVS-right-F4-exBstZ171 + AscI (SEQ ID NO: 114) 3202 GACTTTGGTGATAATGTGAGCAGC CHO-AAVS-right-R4-ex-BclI (SEQ ID NO: 115) 3203 GCTGCTCACATTATCACCAAAGTC CHO-AAVS-right-F5-ex-BclI (SEQ ID NO: 116) 3204 TATATTCCTGCAGGCTCCTGCAAAGGCCTGAAGAG CHO-AAVS-right-R5 + SbfI (SEQ ID NO: 117) 3213 TTTTTTATGCATCTTGATGACGGGGAGATAAAAGCATC CHO-AAVS-left-R1 + NsiI-CRISPR2 + 3 (SEQ ID NO: 118) 3201 TATATTGTATACGGCGCGCCTGTCAGGGACAAGATTAGTCACAG CHO-AAVS-right-F4 + BstZ171 + AscI (SEQ ID NO: 119) 3202 GACTTTGGTGATAATGTGAGCAGC CHO-AAVS-right-R4-ex-BclI (SEQ ID NO: 120) 3203 GCTGCTCACATTATCACCAAAGTC CHO-AAVS-right-F5-ex-BclI (SEQ ID NO: 121) 3204 TATATTCCTGCAGGCTCCTGCAAAGGCCTGAAGAG CHO-AAVS-right-R5 + SbfI (SEQ ID NO: 122) 3213 TTTTTTATGCATCTTGATGACGGGGAGATAAAAGCATC CHO-AAVS-left-R1 + NsiI-CRISPR2 + 3 (SEQ ID NO: 123) 3220 GGAATCATGGGAAATAGGCCCT CRISPR1-R-vector (SEQ ID NO: 124) 3221 CGCTCACAATTCCACACAACAT CRISPR1-F-vector (SEQ ID NO: 125) 3222 AGGGCCTATTTCCCATGATTCC CRISPR-seqR (SEQ ID NO: 126) 3223 ATGTTGTGTGGAATTGTGAGCG CRISPR-seqR (SEQ ID NO: 127)

(361) A TALEN pair was designed to recognise the CHO AAVS locus which recognised the TAL target sequences: TCCCCGTCATCCAAAAGC (SEQ ID NO: 128) and TCTGCTGTGACTAATCTT (SEQ ID NO: 129) as shown in FIG. 46. For comparison against site-specific DNA cleavage by TALEN, we also tested the performance of an alternative nuclease—the nucleic acid guided nuclease CRISPR/Cas9. Three CRISPR/Cas9 guide RNAs were designed to target the CHO AAVS locus, and the recognition sequences are shown in FIG. 46.

(362) The nucleic acid sequence of the CHO AAVS homology arms cloned into the targeting vector is shown in FIG. 47. Synthetic geneblocks were designed encoding the U6 RNA polymerase promoter guide RNA and tracrRNA and PCR amplified with primers 3222 and 3223 (Table 21). The CRISPR/Cas9 vector (A21177, Geneart) was PCR amplified with primers 3220/3221 to generate a 9375 bp product. This was then assembled with the three CRISPR geneblocks (Table 20) using the NEB Builder (New England Biolabs), to generate the three CRISPR/Cas9 vectors encoding the three guide RNAs depicted in FIG. 46.

(363) TABLE-US-00028 TABLE 21 Geneblock sequences encoding the CHO CRISPR sequences. The CHO CRISPR guide and tracrRNA sequences are shown in bold and italics respectively. The U6 RNA polymerase promoter is shown underlined. CHO CRISPR Geneblock sequence 1 AGGGCCTATTTCCCATGATTCCTTCATATTTGCATATACGAT ACAAGGCTGTTAGAGAGATAATTAGAATTAATTTGACTGTAA ACACAAAGATATTAGTACAAAATACGTGACGTAGAAAGTAAT AATTTCTTGGGTAGTTTGCAGTTTTAAAATTATGTTTTAAAA TGGACTATCATATGCTTACCGTAACTTGAAAGTATTTCGATT TCTTGGCTTTATATATCTTGTGGAAAGGACGAAACACCGATC CAAAAGCTGGCATTGTCGTTTTAGAGCTAGAAATAGCAAGTT AAAATAAGGCTAGTCCGTTATCAACTTGAAAAAGTGGCACCG AGTCGGTGCTTTTTTCTAGTATACCGTCGACCTCTAGCTAGA GCTTGGCGTAATCATGGTCATAGCTGTTTCCTGTGTGAAATT GTTATCCGCTCACAATTCCACACAACAT (SEQ ID NO: 130) 2 AGGGCCTATTTCCCATGATTCCTTCATATTTGCATATACGAT ACAAGGCTGTTAGAGAGATAATTAGAATTAATTTGACTGTAA ACACAAAGATATTAGTACAAAATACGTGACGTAGAAAGTAAT AATTTCTTGGGTAGTTTGCAGTTTTAAAATTATGTTTTAAAA TGGACTATCATATGCTTACCGTAACTTGAAAGTATTTCGATT TCTTGGCTTTATATATCTTGTGGAAAGGACGAAACACCGTCT CCCCGTCATCCAAAAGCGTTTTAGAGCTAGAAATAGCAAGTT AAAATAAGGCTAGTCCGTTATCAACTTGAAAAAGTGGCACCG AGTCGGTGCTTTTTTCTAGTATACCGTCGACCTCTAGCTAGA GCTTGGCGTAATCATGGTCATAGCTGTTTCCTGTGTGAAATT GTTATCCGCTCACAATTCCACACAACAT (SEQ ID NO: 131) 3 AGGGCCTATTTCCCATGATTCCTTCATATTTGCATATACGAT ACAAGGCTGTTAGAGAGATAATTAGAATTAATTTGACTGTAA ACACAAAGATATTAGTACAAAATACGTGACGTAGAAAGTAAT AATTTCTTGGGTAGTTTGCAGTTTTAAAATTATGTTTTAAAA TGGACTATCATATGCTTACCGTAACTTGAAAGTATTTCGATT TCTTGGCTTTATATATCTTGTGGAAAGGACGAAACACCGATG CCAGCTTTTGGATGACGGTTTTAGAGCTAGAAATAGCAAGTT AAAATAAGGCTAGTCCGTTATCAACTTGAAAAAGTGGCACCG AGTCGGTGCTTTTTTCTAGTATACCGTCGACCTCTAGCTAGA GCTTGGCGTAATCATGGTCATAGCTGTTTCCTGTGTGAAATT GTTATCCGCTCACAATTCCACACAACAT (SEQ ID NO: 132)

(364) The targeting vector pINT17-BSD-CHO with either the 893 bp or 880 bp AAVS left homology arms (named V1 or V2 respectively) harbouring the Nivolumab antibody heavy and light chain genes was used to transfect Freestyle CHO—S cells (Thermo Fisher Scientific) in the absence or presence of plasmids encoding either the CHO AAVS TAL-1F and CHO AAVS TAL-1R TALEN pair (M3770, Thermo Fisher Scientific) or the CRISPR/Cas9 with CHO AAVS specific guide RNAs and subject to blasticidin drug selection (7.5 μg/ml). The transfection method was as described in Example 2 for the TALEN transfection. For the CRISPR/Cas9 transfection a CRISPR/Cas9, pINT17-BSD-CHO plasmid DNA ratio of 10:1 was employed. Cells were stained with anti-Fc PE 14 days post transfection (dpt). Fluorescence intensity (anti-Fc) was plotted against cell count.

(365) The heavy and light chain genes of two antibody pairs were cloned into the CHO targeting vector pINT17-BSD-CHO with the 893 bp AAVS left homology arms (FIG. 47): the anti-NGF antibodies MEDI-1912 or MED-1912-STT.sup.7 or anti-PCSK9 Bococizumab or the Bococizumab derivative clone 884_01_G01 (see Example 5). The resultant targeting vectors were used to transfect Freestyle CHO-S cells (Thermo Fisher Scientific) in the presence of plasmids encoding the CHO AAVS TAL-1F and CHO AAVS TAL-1R TALEN pair (M3770, Thermo Fisher Scientific) and subject to blasticidin drug selection (7.5 μg/ml). 15 dpt cells were stained with anti-Fc-PE as described in Example 2. 15 dpt CHO cells displaying the antibodies were purified by anti-Fc-PE MACS and stained with anti-Fc PE. Fluorescence intensity (anti-Fc) was plotted against CHO cell count.

Example 13. Creation of a “Developability Enhanced” Population Using Mammalian Display for Subsequent Binding Selection

(366) Example 10 demonstrates that a population of clones pre-selected on antigen (in this example from phage display selection) can be further resolved based on different presentation levels which in turn correlate with biophysical properties. As described elsewhere binding to polyreactivity probes could also be used to identify and remove clones with polyreactivity properties. Thus, problematic clones that might otherwise be selected for further characterisation or development can be eliminated.

(367) In example 10 the antibody population was initially selected based on binding properties (using phage display in this example) followed by subsequent selection based on biophysical properties. It is also possible to reverse the order of selection and generate a population of clones that have been selected for optimal biophysical properties based on presentation level in higher eukaryotes or binding to polyreactive probes. For example, a display library of binders could be made in mammalian cells or other higher eukaryotes and selected for presentation level using an agent that binds to a constant region of the binders so that all binders can be equally labelled independent of their sequence. For example, where the binders are IgG antibodies, cells may be contacted with a detectable agent that binds to the IgG Fc region, e.g., a labelled anti-IgG antibody. A fraction of the population showing presentation level higher than the mode or median can then be selected e.g. by flow cytometry. For example, the top 5%, 10% or 25% of clones based on presentation level can be selected in one or more rounds of sorting using methods described herein. This will create a “developability enhanced” population of binders which can be used for subsequent selection based on binding to different antigens.

(368) Using the approach of example 10, a large phage display library of 4×10.sup.10 clones was selected to generate a sub-population of binders which were then incorporated into a mammalian display library for selection based on presentation level. One potential disadvantage of reversing the order i.e. preselecting for biophysical properties in higher eukaryotes prior to selection of binders is that the creation of a large library in mammalian cells is more laborious and costly than creating a library for use with in vitro systems such as ribosome display or bacterial systems (e.g. phage display). Thus, if a starting mammalian display library of 10.sup.7 clones for example is used and the top 10% selected based on presentation level then the potential diversity of the library to be used for subsequent selection for antigen binding will be reduced e.g. to 10.sup.6 clones reducing the likelihood of selecting high affinity binders. An alternative strategy is to pre-select components of the library in mammalian display e.g. in the case of an antibody separately selected for optimal VH and optimal VL components and then and then recombining them to benefit from the combinatorial diversity.

(369) Thus in one branch a mammalian display library may be created with antibodies in an IgG or scFv format containing a single or limited selection of VLs and a large diversity of VH chains to select optimal VH genes. The single or limited partner VL gene(s) may be randomly chosen or chosen based on poor biophysical properties with a view to selecting VH genes which rescue the poor biophysical properties of the partner chain. Alternatively, a VL gene with good biophysical properties may be chosen to identify and remove those VH genes which compromise presentation levels. Using the same numbers as the example above a library of 10.sup.7 clones with VH diversity may be created and the top 10% of VH genes selected based on presentation level potentially generating 10.sup.6 selected VH clones. In a parallel branch a single or limited number of VH genes may be combined with a diversity of VL genes to select optimal VL genes. The combinatorial diversity of bring the different selected VH and VL genes together will be significantly higher than the number of individual selected chains. Thus selecting the top 10.sup.6 VH and the top 10.sup.6 VL genes in each branch creates a potential combinatorial diversity of 10.sup.12 variants. The selected VH and VL domains may be presented within any display system allowing selection for target binding including phage display, ribosome display, yeast display and display on higher eukaryotes. In this way selection for optimal biophysical properties using the present invention may be used to generate a “developability enhanced” library within other display systems. For example a phage display library which is pre-disposed to yield clones with optimal biophysical properties.

(370) In a further example of the above approach the selection for optimal biophysical properties may be directed to regions within an individual domain. For example, a VH domain includes 3 different complementary determining region (CDR1, CDR2 and CDR3) and biophysical liabilities may be caused by sequences within individual CDR or combination of CDRs. For example the CDR1 and 2 regions of VH genes a non-immunised antibody repertoire may be recovered combined with a single or limited number of VH CDR3 regions and VL partner chains and the resulting antibody population, with diversity focused within CDR1 and 2 selected for optimal biophysical property. The selected CDR1 and 2 regions can then be combined with a diversity of CDR3 segments and VL segments which may be unselected or may have been selected for optimal biophysical properties in a similar way. The potential combinatorial diversity increases by this “shuffling” approach. The example above describes selection for optimal biophysical properties on CDR 1 and CDR2 segments but it will be obvious how the same approach can be used with different CDR regions alone or in combination. In some case the input library to be selected for optimal biophysical properties may be a naïve library from a non-immunised source or from diversification of a starting scaffold. Alternatively, the input library may be diversified library based on an input starting clone or selection of clones with a view to improving biophysical properties potentially alongside affinity.

(371) In each case the approaches described above will use standard molecular biology techniques known to those skilled in the art. In addition, methods for the construction and use of libraries, including chain shuffled libraries are also set out in WO2015/166272 (Iontas Limited), the content of which is incorporated herein by reference. Combining different regions within a VH or VL domain will benefit from the use of optimal PCR primers to amplify individual regions within VH and VL domains. Germline sequences of antibody VH and VL genes are readily available e.g. the IMGT database.sup.132 allowing design of such primers. Furthermore examples in WO2017/118761 (Iontas Limited) describe methods for introducing diversity and combining different CDR regions to re-constitute an intact VH:VL combination.

(372) The example below is used for illustration. Diversity was introduced into the VH CDR3 and VL CDR3 of an initial PD1 binding antibody clone and the resulting population selected for optimal presentation on mammalian cells. The same approach could be taken to introduce diversity into other starting frameworks and other CDRs including the CDR 1 and 2 of germline encoded genes. For example, Table 16 identifies germline genes that frequently appear in populations selected for high levels of presentation on mammalian cells e.g. IGHV1-2, IGHV3-23 and IGKV6-21 and these may be chosen as starting frameworks for further diversification and selection for developability. Similarly sub-regions for binding domains other than antibodies may be identified for separate selection for optimal biophysical properties.

(373) By way of example an anti-PD1 antibody (337_1_C08) which blocked the interaction of PD-1 with PD-L1 was identified by phage display (FIG. 50a). The affinity (K.sub.D) of 337_1_C08 for PD-1 is 74 nM. This was used to create a mutagenised library with diversification focused on VH CDR3 and VL CDR3. A variety of methods for diversification are known to those skilled in the art. Diversification for example may involve creation of a mutagenic library by saturation mutagenesis where randomizing codons such as NNS (encoding all 20 amino acids within 32 codons) are used. An alternative is to retain sequence information while exploring the surrounding sequence space. To achieve this oligonucleotides were designed to a continuous stretch of 8 amino acids in the CDR3 of VH and VLs such that every possible 2-mer and 1-mer variant, using all amino acids (with the exception of cysteine) were included (FIG. 50a).

(374) Thus, the library was designed to retain at least 6 of the original 8 amino acids in each CDR3 region. It was possible to accomplish this by synthesis of 9216 oligonucleotides directed to each of VH CDR3 or VL CDR3. Geneblocks of the VH and VL where every 2-mer amino acid variation was encompassed within an 8-amino acid window (9216 variants) were synthesised by TWIST Biosciences. The presence of all 9216 oligonucleotides in each set was confirmed by high throughput sequencing (Twist Bioscience). The VH gene was amplified using the primer pairs primers Forward: 5′-CTTTCTCTCCACAGGCGCCCATGGCCGAAGTGCAGCC-3′ (SEQ ID NO: 133) and Reverse: 5′-TTTTTTCTCGAGACGGTGACCAGGGTTC-3′ (SEQ ID NO: 134). The VL gene was amplified using the primer pairs primers Forward: 5′-TTTTTTGCTAGCTCCTATGAGCTGACTC-3′ (SEQ ID NO: 135) and Reverse: 5′-GTCACGCTTGGTGCGGCCGCGGGCTGACCTAG-3′ (SEQ ID NO: 136). A “stuffer” fragment encoding the constant light (CL) and CMV promote was PCR amplified from the pINT17-BSD vector using the primer pair Forward: 5′-GGCCGCACCAAGCGTGAC-3′ (SEQ ID NO: 137) and Reverse: 5′-GGCGCCTGTGGAGAGAAAG-3′ (SEQ ID NO: 138). The three gene fragments were first assembled in a “mock” PCR with no amplification primers to ensure no bias introduced by partial assembly and subsequent amplification. The mock PCR was performed by mixing the three PCR products above at an equimolar ratio (90 nM each) and performing a PCR reaction with the KOD Hot Start Master Mix (Sigma, 71842) according to the manufacturer's instructions with an annealing temperature of 60° C. and elongation temperature of 68° C. (45 s). The assembled product was subsequently amplified using the primer pairs primers Forward: 5′-TTTTTTGCTAGCTCCTATGAGCTGACTC-3′ (SEQ ID NO: 139) and Reverse: 5′-TTTTTTCTCGAGACGGTGACCAGGGTTC-3′ (SEQ ID NO: 140). The assembled anti-PD-1 VH and VL CDR3 library was digested with NheI and XhoI restriction enzymes and cloned into the pINT17-BSD targeting vector and by electroporation of E. cloni 10G SUPREME Electrocompetent Cells (Lucigen cat #60081-1), a library size of 1.1×10.sup.8 was created, as determined by counting individual kanamycin resistant colonies on agar plates plated with dilutions of the transformation mix. Transfection quality plasmid DNA was prepared and used to co-transfect HEK293 suspension cells (1.35×10.sup.9 cells) by Maxcyte electroporation with TALE AAVS left and right arm nucleases to enable single copy antibody gene integration. The efficiency of gene targeting was 0.8%, as determined by counting blasticidin resistant colony forming units (CFUs) in dilution plates post transfection, to yield a mammalian display library size of 10.8 million. The library was propagated for 7 days under blasticidin selection and then selected by anti-Fc MACS to remove clones which were not expressing IgG.

(375) As described in example 10, the population of Fc positive cells were separated by flow cytometry on the basis of expression level using binding of a fluorescently labelled anti-Fc antibody (FIG. 50b). The separate populations were cultured for several days and re-analysed for Fc expression. FIG. 50b shows that distinct sub-populations with different modal values for expression have been generated. This including a population with inferior presentation levels as well as a population with presentation levels equivalent to the parental. As described in example 10, antibody genes can be recovered from these cells to generate a sub-population with optimal biophysical properties. In this example cells were sorted on the basis of biophysical properties only but it is also possible to conduct a selection based on antigen binding together or in sequence.

Example 14. Display of Bi-Specific Antibodies on the Surface of Mammalian Cells

(376) Bi-specific antibodies or alternative formats (reviewed by Spiess et al, 2015.sup.129) have enabled new therapeutic mechanisms of action (MOA), not previously possible with mono-specific antibodies or proteins. Examples of the use of bi-specific therapeutics include redirecting the cytotoxic activity of T-cells as cancer therapeutics, enabling the crossing of the blood-brain barrier, blocking two signalling pathways simultaneously and the tissue specific delivery or activity of antibodies.sup.130. Bi-specific formats can face the same developability hurdles as traditional mono-specific antibodies in terms of their propensity for self or cross-interaction properties. It is an advantage to screen the final format of a bi-specific moiety because each binding region can have different properties which may either compensate or compound its self-interaction or cross-interaction properties. These properties may not be apparent when screening the individual binding “arms” of a bi-specific because avidity can increase the affinity for off-target molecules such as heparin sulphate by several orders of magnitude. The property of self-interaction may also be different for a bi-specific molecule compared to its individual binding regions because the overall surface properties such as hydrophobicity may change. Bi-specific antibody (bsAb) formats can exist in many different forms.sup.129, but can be broadly grouped into: IgG-like bsAbs such as κ/λ-bodies.sup.131, common light chain, knobs-into-holes.sup.132, charge pair and crossmab format.sup.130; fragment based bsAbs such as BiTE format, appended IgGs such as DVD-IgGs or antibodies engineered to possess additional binding regions in their constant domains such as Fcab or mAb2 format.sup.133. However, the alteration of the structural framework of the CH3 domain of Fcabs was found to reduce their thermal stability.sup.134 and additional engineering of the Fcab molecule was required to increase their developability.sup.134. In this example, we demonstrate that higher eukaryotic mammalian display can be applied to the display of bi-specific antibodies by displaying the bi-specific antibody Emicizumab on the surface of HEK293 cells and showing that this can bind the antigens FIXa and FX.

(377) Emicizumab is a bi-specific antibody generated using the heavy chain “knobs into holes” technology with a common light chain described previously.sup.132 which was developed to treat haemophilia and acts as a Factor VIII mimetic.sup.135. One arm of Emicizumab is specific to Factor IXa (FIXa) and the second arm is specific to Factor X (FX). A tri-cistronic targeting vector was constructed by cloning the Emicizumab anti-FIXa heavy chain, anti-FX heavy chain and common light chain genes (Table 22) into the inducible targeting vector pINT17-Tet. The common light chain gene including the downstream poly-adenylation (pA) site was cloned into the BgIII and NheI restriction sites of pINT17-Tet. The anti-FIXa heavy chain gene including the PDGFR transmembrane domain was cloned between the NcoI and HindIII restriction sites of pINT17-Tet. The anti-FX heavy chain gene including the signal peptide, PDGFR transmembrane domain and SV40 pA was cloned between the EcoRI and BstZ171 restriction sites of pINT17-Tet. The final vector (pINT17-Bi-CMV-Emicizumab) contained coding sequences between the AAVS homology arms. FIG. 51. The Emicizumab anti-FIXa and VL genes were also cloned into the pINT17-BSD targeting vector to enable display of a monospecific anti-FIXa IgG antibody.

(378) To demonstrate HEK293 cell surface display of the bi-specific antibody Emicizumab pINT17-Bi-CMV-Emicizumab or pINT17-BSD-anti-FIXa was used to transfect HEK293 cells in the presence of plasmids encoding the AAVS TALENs as described above. 24 hours post transfection the cells were analysed for antibody display with anti-Fc-APC (FIG. 52). This showed that HEK293 cells transfected with the mono-specific anti-FIXa arm in IgG format or the bi-specific antibody Emicizumab had detectable antibody expression on the cell surface. However, the expression of the bi-specific Emicizumab with “knobs-in-holes” heavy chains was reduced compared to the standard mono-specific antibody format. It is envisaged that through the creation of a library of Fc variants followed by selection for high level expression by mammalian display that Fc variants could be selected to enable more efficient heterologous heavy chain pairing and increased cell display levels. In this way, new antibody CH2 or CH3 variants or a combination of CH2 and CH3 variants could be selected that are both more efficient in heterologous heavy chain pairing to create bi-specific antibodies but also would have superior developability properties including a low propensity to self-aggregate. The Fc variant libraries could also be screened for a low propensity to cross-interact with other molecules.

(379) The ability of the displayed bi-specific Emicizumab to bind its target antigens was also demonstrated by flow cytometry. FIXa and FX (Complement Technology Inc) were chemically biotinylated (EZ-Link Sulfo-NHS-Biotin, ThermoFisher Scientific). Since the affinity of Emicizumab for FIXa and FX is relative low (K.sub.D in the micro-molar range.sup.136). The antigens were pre-conjugated with tetrameric streptavidin-PE, to increase the binding avidity, prior to cell staining at an antigen complex concentration of 100 nM. HEK293 cells displaying the bi-specific Emicizumab were shown to bind both FIXa and FX with no binding to unconjugated streptavidin-PE (FIG. 52a). This demonstrates functional display of a bi-specific antibody on the surface of HEK293 cells. Display of the mono-specific anti-FIXa arm in IgG format on the surface of HEK293 cells also bound to FIXa (FIG. 52b). The anti-FIXa antibody also showed some binding to FX, although to a lower level than Emicizumab and this may be due either to the common light chain or the anti-FIXa heavy chain cross-reacting with FX.

(380) Example 6b described the differential binding of heparin sulphate to the anti-IL12 antibodies briakinumab and ustekinumab displayed on the surface of HEK293 cells. Briakinumab has a positive charge patch within its variable domain, which is likely to contribute to its binding a negative charge patch on FcRn.sup.22. The positive charge patch on briakinumab is also likely to be the cause of it its cross-interaction with negatively charge heparin sulphate. Binding of antibodies to heparin sulphate can lead to increased non-specific clearance in vivo resulting in a decreased half-life.sup.137,138. Therefore, the binding of antibodies or therapeutic proteins, including bi-specific molecules, to heparin is an undesired property. Example 6b demonstrated that it is possible to differentiate antibodies that bind heparin sulphate and so it will be possible to separate and eliminate heparin sulphate binding proteins by higher eukaryotic mammalian display selection. From the data presented in this example, it would be possible to envisage the demonstration of differentiation a series of anti-FIXa/anti-FX bi-specific antibodies on the basis of their ability to bind heparin sulphate.

(381) During the course of development of Emicizumab, a precursor humanised antibody was discovered named hBS106.sup.136. This molecule was found to have poor pharmacokinetics in mice with rapid clearance and short in vivo half-life compared with human IgG4. This rapid clearance in vivo was attributable to a positive charge patch on the VH and VL anti-FIXa arm. The amino acids on the common light chain contributing to the positive charge patch included K24, R27 and R31 within VL CDR1 and R53 and R54 within VL CDR2 and R61 within FW3 (FIG. 53). VH amino acids contributing to the positive charge patch included R60 and R95. The lysine and arginine residues contributing to the positive charge patch were paratopic residues indispensable for FIXa binding. Therefore, negatively charged amino acid residues (glutamate or aspartate) were introduced to disrupt the positive charge patches and also lower the isoelectric point (pI) of the antibody. The introduction of negatively charged amino acid changes in the common light chain resulted in an 8-fold reduction in clearance rate in mice and an almost 6-fold increase in maximal in vivo plasma concentration (Cmax). FIG. 53 shows an alignment of the common VL chain of Emicizumab with a series of precursor VLs (US 2016/0222129) which have three to one less negative charge amino acids compared with the final Emicizumab VL. The naming of the Emicizumab parental antibodies are relative to Emicizumab. For example, the clone: E30Y_E55Y_D93S has tyrosine, tyrosine and serine residue at positions 30, 55 and 93 in place of glutamate, glutamate and aspartate respectively relative to Emicizumab.

(382) Emicizumab pre-cursor VL anti-FIXa genes (Table 23) could be cloned into the pINT17-Bi-CMV-Emicizumab targeting vector nd this used to co-transfect HEK293 cells with the human AAVS TALEN pair as described above to enable nuclease mediated antibody gene integration. After 14 dpt, HEK293 cells could be stained with heparin sulphate (FITC labelled) and anti-Fc-PE as described in Example B. Flow cytometry analysis would show that the mammalian cell displayed anti-FIXa Emicizumab clones E30Y_K54R_E55Y_D93S and E30Y_E55Y_D93S would display enhanced binding to heparin sulphate compared with anti-FIXa Emicizumab and the E30Y VL clone. The E30Y_K54R_E55Y_D93S and E30Y_E55Y_D93S clones possess the most intact positive charge patch and these clones would result in detectable heparin sulphate binding to cells displaying these antibodies. This example has shown how it would be possible to use higher eukaryotic mammalian display to differentiate between clones with point mutation differences in their ability to bind to heparin sulphate. This will enable the elimination of bi-specific clones within a complex library that bind heparin sulphate and therefore remove clones that possess positive charge patches within their variable domains which may contribute to poor pharmacokinetics in vivo.

(383) This example has illustrated the potential of bi-specific antibody or alternative bi-specific format screening by mammalian display. The number of variants required to be screened in a bispecific discovery campaign is multiplied by several orders of magnitude compared to a single target antibody discovery project. For example, to identify the lead starting molecule in the Emicizumab discovery campaign 200 anti-FIXa antibodies were crossed with 200 anti-FX antibodies to create 40,000 bi-specific molecules that were then screened by laborious plate based screening. The lead bi-specific molecule then had to proceed through several iterations including humanisation, affinity and specificity optimisation and developability enhancement regarding reducing cross-interaction with heparin sulphate by the disruption of variable domain positive charge patches.sup.136. The ability to perform multi-dimensional FACS on millions of bi-specific antibody clones by mammalian display including affinity for the target(s), specificity, display level (to screen for a low propensity to self-aggregate) and cross-interaction would enable a faster, more efficient bi-specific antibody screening process on a greater number of clones with a greater screening depth.

(384) TABLE-US-00029 TABLE 22 Emicizumab heavy and light chain genes. Variable and PDGFR transmembrane domain encoding regions are highlighted in italic and bold respectively. Restriction sites are underlined. Gene Gene sequence Emicizumab TAATAAGCTAGCAGAGGAGACATCCAGATGACACAGA VL-CL GCCCTAGCAGCCTGTCTGCCAGCGTGGGAGACAGAGT GACCATCACATGCAAGGCCAGCCGGAACATCGAGAGA CAGCTGGCCTGGTATCAGCAGAAGCCTGGACAGGCTC CTGAGCTGCTGATCTATCAGGCCAGCAGAAAAGAAAG CGGCGTGCCCGATAGATTCAGCGGCAGCAGATACGGC ACCGACTTCACCCTGACAATATCCAGCCTCCAGCCTG AGGATATCGCCACCTACTACTGCCAGCAGTACAGCGA CCCTCCACTGACATTTGGCGGAGGCACCAAGGTGGAA ATCAAGCGGACAGCGGCCGCCCCTAGCGTGTTCATCT TTCCACCTAGCGACGAGCAGCTGAAGTCTGGCACAGC CTCTGTCGTGTGCCTGCTGAACAACTTCTACCCCAGA GAAGCCAAGGTGCAGTGGAAGGTGGACAACGCCCTCC AGAGCGGCAATAGCCAAGAGAGCGTGACCGAGCAGGA CAGCAAGGACTCTACCTACAGCCTGAGCAGCACACTG ACCCTGAGCAAGGCCGACTACGAGAAGCACAAAGTGT ACGCCTGCGAAGTGACCCACCAGGGCCTTTCTAGCCC TGTGACCAAGAGCTTCAACCGGGGCGAATGTTAATAA TCTAGAGCCTCGACTGTGCCTTCTAGTTGCCAGCCAT CTGTTGTTTGCCCCTCCCCCGTGCCTTCCTTGACCCT GGAAGGTGCCACTCCCACTGTCCTTTCCTAATAAAAT GAGGAAATTGCATCGCATTGTCTGAGTAGGTGTCATT CTATTCTGGGGGGTGGGGTGGGGCAGGACAGCAAGGG GGAGGATTGGGAAGACAATAGCAGGCATGCTGGGGAA GATCTTAATAA (SEQ ID NO: 141) Emicizumab TTATTAGCCATGGCCCAGGTGCAGCTGGTTGAATCTG anti-FIXa GCGGAGGACTGGTTCAGCCTGGCGGATCTCTGAGACT VH-CH1- GTCTTGTGCCGCCAGCGGCTTCACCTTCAGCTACTAC CH2-CH3 GATATCCAGTGGGTCCGACAGGCCCCTGGCAAAGGAC TTGAATGGGTGTCCAGCATCAGCCCCTCTGGCCAGTC CACCTACTACCGGCGAGAAGTGAAGGGCAGATTCACC ATCAGCCGGGACAACAGCAAGAACACCCTGTACCTGC AGATGAACAGCCTGAGAGCCGAGGACACCGCCGTGTA CTACTGCGCCAGAAGAACCGGCAGAGAGTACGGCGGA GGCTGGTACTTTGATTACTGGGGCCAGGGCACCCTGG TCACAGTCTCGAGCGCCTCTACAAAGGGCCCCAGCGT TTTCCCACTGGCTCCCTGTAGCAGAAGCACCAGCGAA TCTACAGCCGCTCTGGGCTGCCTGGTCAAGGACTACT TTCCTGAGCCTGTGACCGTGTCCTGGAACTCTGGCGC TCTGACATCTGGCGTGCACACCTTTCCAGCCGTGCTG CAAAGCAGCGGCCTGTACAGTCTGAGCAGCGTCGTGA CAGTGCCTAGCAGCTCTCTGGGCACCCAGACCTACAC CTGTAATGTGGACCACAAGCCTAGCAACACCAAGGTG GACAAGCGCGTGGAATCTAAGTACGGCCCTCCTTGTC CTCCATGTCCTGCACCTGAGTTTCTCGGCGGACCCTC CGTGTTCCTGTTTCCTCCAAAGCCTAAGGACACCCTG ATGATCTCCAGAACACCCGAAGTGACCTGCGTGGTGG TGGACGTTTCACAAGAGGACCCCGAGGTGCAGTTTAA TTGGTACGTGGACGGCGTGGAAGTGCACAACGCCAAG ACCAAGCCTAGAGAGGAACAGTACAACAGCACCTACA GAGTGGTGTCCGTGCTGACAGTGCTGCACCAGGATTG GCTGAACGGCAAAGAGTACAAGTGCAAGGTGTCCAAC AAGGGCCTGCCAAGCAGCATCGAGAAAACCATCAGCA AGGCCAAGGGCCAGCCTAGGGAACCCCAGGTTTACAC ACTGCCTCCAAGCCAGAAAGAGATGACCAAGAACCAG GTGTCCCTGACCTGCCTCGTGAAGGGCTTCTACCCTT CCGATATCGCCGTGGAATGGGAGAGCAATGGCCAGCC AGAGAACAACTACAAGACCACACCTCCTGTGCTGGAC AGCGACGGCTCATTCTTCCTGTACAGCAAGCTGACCG TGGACAAGAGCAGATGGCAAGAGGGCAACGTGTTCAG CTGCAGCGTGATGCACGAGGCCCTGCACAACAGATAC ACCCAGAAGTCCCTGTCTCTGAGCCCCGAACAAAAAC TCATCTCAGAAGAGGATCTGAATGCTGTGGGCCAGGA CACGCAGGAGGTCATCGTGGTGCCACACTCCTTGCCC TTTAAGGTGGTGGTGATCTCAGCCATCCTGGCCCTGG TGGTGCTCACCATCATCTCCCTTATCATCCTCATCAT GCTTTGGCAGAAGAAGCCACGTTAGTAAAAGCTTTTA TTA (SEQ ID NO: 142) Emicizumab TTATTAGAATTCAACATGGACTGGACCTGGAGGGTCT anti-FX VH- TCTGCTTGCTGGCTGTAGCTCCAGGTAAAGGGCCAAC CH1-CH2-CH3 TGGTTCCAGGGCTGAGGAAGGGATTTTTTCCAGTTTA GAGGACTGTCATTCTCTACTGTGTCCTCTCCGCAGGT GCTCACTCCCAGGTTCAGCTGGTGCAGTCTGGCAGCG AGCTGAAAAAACCTGGCGCCTCCGTGAAGGTGTCCTG CAAGGCTTCTGGCTACACCTTTACCGACAACAACATG GACTGGGTCCGACAGGCCCCTGGACAAGGACTTGAGT GGATGGGCGACATCAACACCAGAAGCGGCGGCAGCAT CTACAACGAAGAGTTCCAGGACAGAGTCATCATGACC GTGGACAAGAGCACCGACACCGCCTACATGGAACTGA GCAGCCTGAGAAGCGAGGACACCGCCACCTATCACTG CGCCAGAAGAAAGAGCTACGGCTACTACCTGGACGAG TGGGGCGAGGGAACACTGGTCACAGTGTCTAGCGCCA GCACAAAGGGCCCTAGCGTTTTCCCACTGGCTCCCTG TAGCAGAAGCACCAGCGAATCTACAGCCGCTCTGGGC TGCCTCGTGAAGGACTACTTTCCTGAGCCTGTGACCG TTAGCTGGAACAGCGGAGCACTGACAAGCGGCGTGCA CACATTTCCAGCCGTGCTGCAAAGCAGCGGCCTGTAC TCTCTGAGCAGCGTCGTGACAGTGCCTAGCAGCTCTC TGGGCACCCAGACCTACACCTGTAATGTGGACCACAA GCCTAGCAACACCAAGGTGGACAAGCGCGTGGAATCT AAGTACGGCCCTCCTTGTCCTCCATGTCCTGCTCCAG AGTTTCTCGGCGGACCCTCCGTGTTCCTGTTTCCTCC AAAGCCTAAGGACACCCTGATGATCTCCAGAACACCC GAAGTGACCTGCGTGGTGGTGGACGTTTCACAAGAGG ACCCCGAGGTGCAGTTCAATTGGTACGTGGACGGCGT GGAAGTGCACAACGCCAAGACCAAGCCTAGAGAGGAA CAGTACAACAGCACCTACAGAGTGGTGTCCGTGCTGA CAGTGCTGCACCAGGATTGGCTGAACGGCAAAGAGTA CAAGTGCAAGGTGTCCAACAAGGGCCTGCCAAGCAGC ATCGAGAAAACCATCAGCAAGGCCAAGGGCCAGCCTA GGGAACCCCAGGTTTACACACTGCCTCCAAGCCAAGA GGAAATGACCAAGAACCAGGTGTCCCTGACCTGCCTG GTCAAGGGCTTCTACCCTTCCGATATCGCCGTGGAAT GGGAGAGCAATGGCCAGCCAGAGAACAACTACAAGAC CACACCTCCTGTGCTGGACAGCGACGGCTCATTCTTC CTGTACAGCAAGCTGACTGTGGATAAGAGCCGGTGGC AAGAGGGCAACGTGTTCAGCTGTAGCGTGATGCACGA GGCCCTGCACAACCACTACACCCAAGAGAGCCTGTCT CTGAGCCCTGAACAAAAACTCATCTCAGAAGAGGATC TGAATGCTGTGGGCCAGGACACGCAGGAGGTCATCGT GGTGCCACACTCCTTGCCCTTTAAGGTGGTGGTGATC TCAGCCATCCTGGCCCTGGTGGTGCTCACCATCATCT CCCTTATCATCCTCATCATGCTTTGGCAGAAGAAGCC ACGTTAGTAACTAAGTCGACATCCAGACATGATAAGA TACATTGATGAGTTTGGACAAACCACAACTAGAATGC AGTGAAAAAAATGCTTTATTTGTGAAATTTGTGATGC TATTGCTTTATTTGTAACCATTATAAGCTGCAATAAA CAAGTTAACAACAACAATTGCATTCATTTTATGTTTC AGGTTCAGGGGGAGGTGTGGGAGGTTTTTTAAAGCAA GTAAAACCTCTACAAATGTGGTATGGCTGATTATGAT CCTGCAAGCCTCGTCGTCCTGGCCGGACCACGCTATC TGTGCAAGGTCCCCGGCCCCGGACGCGCGCTCCATGA GCAGAGCGCCCGCCGCCGAGGCGAAGACTCGGGCGGC GCCCTGCCCGTCCCACCAGGTCAACAGGCGGTAACCG GCCTCTTCATCGGGAATGCGCGCGACCTTCAGCATCG CCGGCATGTCCCCCTGGCGGACGGGAAGTATGTATAC TTATTA (SEQ ID NO: 143)

(385) TABLE-US-00030 TABLE 23 VL and VH synthetic geneblock DNA sequences of the anti-FIXa Emicizumab and Emicizumab parental antibodies. Emicizumab parental antibodies are named by the single letter amino acid substitutions relative to Emicicuzumab. For example, parental antibody Emicizumab VL E30Y indicates that Emicizumab glutamate is replaced by tyrosine at position 30. Changes relative to the final Emicizumab VL are highlighted in bold. Flanking restriction sites are underlined. Gene Gene sequence Emicizumab TTATTAGCCATGGCCCAGGTGCAGCTGGTTGAATCT anti-FIXa VH GGCGGAGGACTGGTTCAGCCTGGCGGATCTCTGAGA CTGTCTTGTGCCGCCAGCGGCTTCACCTTCAGCTAC TACGATATCCAGTGGGTCCGACAGGCCCCTGGCAAA GGACTTGAATGGGTGTCCAGCATCAGCCCCTCTGGC CAGTCCACCTACTACCGGCGAGAAGTGAAGGGCAGA TTCACCATCAGCCGGGACAACAGCAAGAACACCCTG TACCTGCAGATGAACAGCCTGAGAGCCGAGGACACC GCCGTGTACTACTGCGCCAGAAGAACCGGCAGAGAG TACGGCGGAGGCTGGTACTTTGATTACTGGGGCCAG GGCACCCTGGTCACAGTCTCGAGTTATTA (SEQ ID NO: 144) Emicizumab TTATTAGCTAGCGACATCCAGATGACACAGAGCCCT VL AGCAGCCTGTCTGCCAGCGTGGGAGACAGAGTGACC ATCACATGCAAGGCCAGCCGGAACATCGAGAGACAG CTGGCCTGGTATCAGCAGAAGCCTGGACAGGCTCCT GAGCTGCTGATCTATCAGGCCAGCAGAAAAGAAAGC GGCGTGCCCGATAGATTCAGCGGCAGCAGATACGGC ACCGACTTCACCCTGACAATATCCAGCCTCCAGCCT GAGGATATCGCCACCTACTACTGCCAGCAGTACAGC GACCCTCCACTGACATTTGGCGGAGGCACCAAGGTG GAAATCAAGCGGACAGCGGCCGCTTATTA (SEQ ID NO: 145) Emicizumab TTATTAGCTAGCGACATCCAGATGACACAGAGCCCT VL E30Y AGCAGCCTGTCTGCCAGCGTGGGAGACAGAGTGACC ATCACATGCAAGGCCAGCCGGAACATCTATAGACAG CTGGCCTGGTATCAGCAGAAGCCTGGACAGGCTCCT GAGCTGCTGATCTATCAGGCCAGCAGAAAATATAGC GGCGTGCCCGATAGATTCAGCGGCAGCAGATACGGC ACCGACTTCACCCTGACAATATCCAGCCTCCAGCCT GAGGATATCGCCACCTACTACTGCCAGCAGTACAGC GACCCTCCACTGACATTTGGCGGAGGCACCAAGGTG GAAATCAAGCGGACAGCGGCCGCTTATTA (SEQ ID NO: 146) Emicizumab TTATTAGCTAGCGACATCCAGATGACACAGAGCCCT VL E30Y AGCAGCCTGTCTGCCAGCGTGGGAGACAGAGTGACC E55Y ATCACATGCAAGGCCAGCCGGAACATCTATAGACAG CTGGCCTGGTATCAGCAGAAGCCTGGACAGGCTCCT GAGCTGCTGATCTATCAGGCCAGCAGAAAAGAAAGC GGCGTGCCCGATAGATTCAGCGGCAGCAGATACGGC ACCGACTTCACCCTGACAATATCCAGCCTCCAGCCT GAGGATATCGCCACCTACTACTGCCAGCAGTACAGC GACCCTCCACTGACATTTGGCGGAGGCACCAAGGTG GAAATCAAGCGGACAGCGGCCGCTTATTA (SEQ ID NO: 147) Emicizumab TTATTAGCTAGCGACATCCAGATGACACAGAGCCCT VL E30Y AGCAGCCTGTCTGCCAGCGTGGGAGACAGAGTGACC E55Y D935 ATCACATGCAAGGCCAGCCGGAACATCTATAGACAG CTGGCCTGGTATCAGCAGAAGCCTGGACAGGCTCCT GAGCTGCTGATCTATCAGGCCAGCAGAAAATATAGC GGCGTGCCCGATAGATTCAGCGGCAGCAGATACGGC ACCGACTTCACCCTGACAATATCCAGCCTCCAGCCT GAGGATATCGCCACCTACTACTGCCAGCAGTACAGC AGCCCTCCACTGACATTTGGCGGAGGCACCAAGGTG GAAATCAAGCGGACAGCGGCCGCTTATTA (SEQ ID NO: 148) Emicizumab TTATTAGCTAGCGACATCCAGATGACACAGAGCCCT VL E30Y AGCAGCCTGTCTGCCAGCGTGGGAGACAGAGTGACC K54R E55Y ATCACATGCAAGGCCAGCCGGAACATCTATAGACAG D935 CTGGCCTGGTATCAGCAGAAGCCTGGACAGGCTCCT GAGCTGCTGATCTATCAGGCCAGCAGAAGATATAGC GGCGTGCCCGATAGATTCAGCGGCAGCAGATACGGC ACCGACTTCACCCTGACAATATCCAGCCTCCAGCCT GAGGATATCGCCACCTACTACTGCCAGCAGTACAGC AGCCCTCCACTGACATTTGGCGGAGGCACCAAGGTG GAAATCAAGCGGACAGCGGCCGCTTATTA (SEQ ID NO: 149)

Example 15. Selection for Developability of KnotBodies Based on Surface Presentation Level

(386) Example 14 demonstrated that it is possible to display the “knobs-in holes” bi-specific antibody Emicizumab on the surface of HEK293 cells. Bi-specific molecules can also be created by the fusion of polypeptides to antibody heavy or light chains or engineering of the heavy or light chains to confer novel binding specificities.sup.133. The ability to display bi-specific molecules on the surface of mammalian cells will enable screening for binding to both targets combined with selecting for bi-specific molecules with well-behaved biophysical properties as described above. A KnotBody is a novel antibody fusion format where a cysteine rich peptide (knottin) is incorporated into a peripheral CDR loop of an antibody domain (WO2017/118761) and the VH or VL domain is available to bind a second epitope on the same antigen or on a different antigen. The aim of this example was to demonstrate the ability of mammalian display technology to differentiate and separate KnotBodies that have different properties regarding their self-interaction and polyreactivity properties.

(387) The KnotBody patent WO2017/118761, described the generation of two trypsin binding KnotBodies (KB_A07 and KB_A12) by inserting the trypsin binding knottin EETI-II into the VL CDR2 position of antibodies. The KnotBodies tested in this example were KB_A12 ProTx-III 2 M (hereafter referred as KB_A12 ProTx-III) and KB_A12 HsTx1. These KnotBodies were created by replacing the EETI-II knottin at the VL CDR2 position of KB_A12 KnotBody with ion channel blocking knottins or toxin peptides ProTx-III 2 M (PCT/EP2018/068855 filed 11 Jul. 2018) and HsTxI (PCT/EP2018/068856 filed 11 Jul. 2018). The VL sequences of KB_A12 EETI-II, KB_A12 ProTx-III and KB_A12 HsTx1 are shown in

(388) TABLE-US-00031 TABLE 24 KnotBody VL sequences. The knottin with linker amino acids inserted in the CDR2 is highlighted in bold KnotBody KnotBody VL sequence KB_A12 QSVLTQPPSVSEAPRQRVTITCSGSSSNIGNNAVNWYQ EETI-II QLPGKAPKLLIYAAGRCPRILMRCKQDSDCLAGCVCGP NGFCGANSGVSDRFSAAKSGTSASLAINGLRSEDEADY YCAAWDDSLNGYVFGTGTKLTVLG (SEQ ID NO: 150) KB_A12 QSVLTQPPSVSEAPRQRVTITCSGSSSNIGNNAVNWYQ ProTx-III QLPGKAPKLLIYAAGRGCLKFGWKCNPRNDKCCSGLKC GSNHNWCKWHIGANSGVSDRFSAAKSGTSASLAINGLR SEDEADYYCAAWDDSLNGYVFGTGTKLTVLG (SEQ ID NO: 151) KB_A12 QSVLTQPPSVSEAPRQRVTITCSGSSSNIGNNAVNWYQ HsTxI QLPGKAPKLLIYAAGRASCRTPKDCADPCRKETGCPYG KCMNRKCKCNRCANSGVSDRFSAAKSGTSASLAINGLR SEDEADYYCAAWDDSLNGYVFGTGTKLTVLG (SEQ ID NO: 152)
Genes encoding the knotBodies KB_A12 EETI-II, ProTx-III and HsTxI were cloned into the targeting vector pINT17-BSD and this was used to create stable cell lines by nuclease mediated gene integration into HEK293 cells as described above. FIG. 54a shows that the display level of KB_A12 EETI-II on the surface of HEK293 cells was higher than KB_A12 ProTx-III or KB_A12 HsTxI. KB_A12 ProTx-III or KB_A12 HsTxI displayed on the surface of HEK293 cells at a relatively low level and the flow cytometry histogram plots of cell count against Fc expression (FIG. 54a) are equivalent. The knotBodies were expressed by transient transfection of Expi293 cells, purified by Protein A affinity chromatography and analysed by HPLC-SEC. FIG. 54b shows that KB_A12 EETI-II displayed a monomeric peak with an equivalent retention time and volume as the well behaved anti-HER2 antibody Trastuzumab (FIG. 54c). In contrast, KB_A12 ProTx-III showed evidence of dimer and multimer formation as illustrated by an earlier retention time (FIG. 54d). The larger elution volume of KB_A12 ProTx-II compared with Trastuzumab and KB_A12 EETI-II is evidence of KB_A12 ProTx-III heterogeneous multimer and aggregate formation. KB_A12 ProTx-III and KB_A12 HsTxI also displayed an increased propensity to self-interact compared to KB_A12 EETI-II in an AC-SINS assay (Table 25).sup.39 where KB_A12 ProTx-III and KB_A12 HsTxI displayed longer AC-SINS wavelength shifts compared to KB_A12 EETI-II. Therefore, as seen previously with IgG display, there is a relationship between the higher eukaryotic cell display levels of the knotBody and their biophysical properties of propensity to self-interact and aggregate. The knotBody is an example of a bi-specific antibody and therefore this relationship between higher eukaryote cell display level and biophysical properties of the displayed molecule will likely transfer to alternative bi-specific modalities.sup.129.

(389) TABLE-US-00032 TABLE 25 AC-SINS scores of purified knotBodies.sup.39. KnotBody AC-SINS score KB_A12 EETI-II 3 KB_A12 ProTx III 10 KB_A12 HsTx I 20

(390) In order to improve the biophysical properties of these KnotBodies, three KnotBody mammalian libraries were created by targeted mutagenesis of hydrophobic or positively charged residues in the knottin sequences. Hydrophobic or positively charged residues for targeted mutagenesis on each knottin is highlighted in bold and underlined (see below). These residues were mutated using primers encoding VNS codons (represented as X in the amino acid sequence) or NSG codons (represented as Z in the amino acid sequence). VNS codons (V=A/C/G, N=A/G/CT and S=G/C) encode 16 amino acids (excludes cysteine, tyrosine, tryptophan, phenylalanine and the stop codons) from 23 codon combinations, whilst NSG codons (N=A/G/C/T and S=G/C) encode 7 amino acids (arginine, tryptophan, glycine, threonine, serine, alanine and proline) from 8 codon combinations. NSG codons are used in positions where the wild-type tryptophan residue may be involved in binding contacts to the ion channel.

(391) TABLE-US-00033 Wild type ProTx-III knottin sequence, with residues targeted for mutagenesis in bold and under-lined: (SEQ ID NO: 153) GCLKFGWKCNPRNDKCCSGLKCGSNHNWCKWHI ProTx-III sequence in KB A12 ProTx-III Set-A Library: (SEQ ID NO: 154) GCXXXXZKCNPRNDKCCSGLKCGSNHNWCKWHI ProTx-III sequence in KB A12 ProTx-III Set-B Library: (SEQ ID NO: 155) GCXXXXZKCNPRNDKCCSGLXCGSNHNWCKZXX Wild type HsTx1 knottin sequence, with residues targeted for mutagenesis in bold and under-lined: (SEQ ID NO: 156) ASCRTPKDCADPCRKETGCPYGKCMNRKCKCNRC HsTx1 sequence in KB A12 HsTx-I library: (SEQ ID NO: 157) ASCRTPKDCADPCXXETGCPYGKCXNRXCKCNRC

(392) The libraries were generated by a two-fragment assembly PCR of KnotBody VL genes. Individual fragments for each library was amplified using primers and templates described in Table 26. The amplified fragments corresponding to each library were assembled using primers pINT BM40 Lead Fwd and pINT CLambda Not Rev. All primer sequences are given in Table 27. Assembled PCR fragments were digested using NheI and NotI restriction enzymes and ligated into pINT17-blasticidin vector encoding D1A12 or D12 VH (the heavy chain used for the KnotBody constructs, described in WO2017/118761). The ligation products were purified using MinElute PCR purification kit (Qiagen, Cat. no. 28004). 3×2.5 μl of the purified ligation mix was electroporated into 3×50 μl of E. coli cells (E. cloni 10 G Supreme, Lucigen, Cat. No. 60080-2). Briefly, 50 μl of cells were pulsed using a 0.1 cm cuvette, recovered with 2 ml recovery medium and grown for 1 h at 37° C., 250 rpm and the library seizes were estimated by dilution plating. The library sizes obtained for KB_A12 ProTx-III Set-A, KB_A12 ProTx-III Set-B and KB_A12 HsTx1 libraries were 9×10.sup.7, 7×10.sup.7 and 2×10.sup.7 respectively. In order to create mammalian display libraries, transfection quality DNA were prepared from these library stocks using Macherey Nagel Midi prep kit (Macherey Nagel, Cat. No. 740410.10), following manufactures instructions.

(393) TABLE-US-00034 TABLE 26 Primers used for amplifying Fragment 1 and Fragment 2 for the libraries. Fragment 1 Fragment 1 Fragment 2 Fragment 2 forward reverse forward reverse Library primer primer primer primer KB_A12 pINT BM40 HsTx1 HsTx1 pINT HsTx1 Lead Fwd Part 1 Part 2 CLambda Rev Fwd Not rev KB_A12 pINT BM40 ProTx-III ProTx-III pINT ProTx-III Lead Fwd Part1 Part 2 CLambda Set-A Rev Fwd Not rev KB_A12 pINT BM40 ProTx-III ProTx-III pINT ProTx-III Lead Fwd setB setB CLambda Set-B Part 1 Rev Part 2 fwd Not rev

(394) TABLE-US-00035 TABLE 27 Sequences of the primers used. Primer Name Primer Sequence (5′-3′) HsTx1 Part 1 ACAGGGGTCCGCGCAGTCTTTAGGAGTTCG Rev (SEQ ID NO: 158) HsTx1 Part 2 CCTAAAGACTGCGCGGACCCCTGTVNSVNSGAGACT Fwd GGATGTCCATACGGTAAGTGCVNSAATAGAVNSTGC AAATGTAACCGATGCGCAAACAGT (SEQ ID NO: 159) ProTx-III CATCCCCTTCCCGCTGCGTAAATGAGAAG Part1 Rev (SEQ ID NO: 160) ProTx-III CTCATTTACGCAGCGGGAAGGGGATGCVNSVNSVNS Part 2 Fwd VNSNSGAAATGCAACCCAAGAAACGATAAA (SEQ ID NO: 161) ProTx-III GAGTCCTGAGCAGCATTTATCGTTTCTTGGGTTGCA setB Part 1 TTTCSNSNBSNBSNBSNBGCATCCCCTTCCCGCTGC Rev GTAAATGAG (SEQ ID NO: 162) ProTx-III CCAAGAAACGATAAATGCTGCTCAGGACTCVNSTGC setB Part 2 GGCAGCAACCACAACTGGTGCAAANSGVNSVNSGGC fwd GCAAACAGTGGCGTCAGTGAC (SEQ ID NO: 163) pINT BM40 GTTTGCCTGGCCGGGAGGGCTCTGGC Lead Fwd (SEQ ID NO: 164) pINT Clambda AGTCACGCTTGGTGCGGCCGC Not Rev (SEQ ID NO: 165)

(395) The DNA for each library was electroporated into suspension adapted HEK293F cells using MaxCyte STXG2 (see Example 5). Similarly, wild type KB_A12 ProTx-III and KB_A12 HsTx1 constructs were cloned and electroporated into HEK293F cells as controls. 100 million cells were used each library and 10 million cells were used for each control construct. Two days post transfection, the antibiotic Blasticidin S HCl (LifeTech, Cat. No. R21001) was added at a concentration of 7.5 μg/ml. Cells were seeded at 0.25×10.sup.6 cells per ml in Erlenmeyer flasks. To calculate gene integration efficiency, cells were also plated into 10 cm dishes as described in Example 5. The integration efficiency was calculated to be 3%, 2.5% and 2% for the KB_A12 ProTx-III Library Set A, KB_A12 ProTx-III Library Set B, and KB_A12 HsTx1 Library, respectively. Therefore, the library sizes achieved for the 100 million cell transfections were 3×10.sup.6, 2.5×10.sup.6, and 2×10.sup.6, respectively. After 5 days of Blasticidin S HCl (LifeTech, Cat. No. R21001) selection the cells were enriched using anti-Fc MACS beads as described in Example 5. After 14 days of selection the libraries were analysed by flow cytometry for knotbody expression level. As shown in FIG. 55, all three knotbody libraries displayed an improved average expression level compared with the parental knotbodies displayed on HEK293 cells. This provides evidence that knotbody variants have been generated that will possess improved biophysical properties compared with the parental knotbody molecules.

(396) Next, FACS was carried out on the knotbody libraries using the BioRad S3e Cell Sorter. 30×10.sup.6 cells of the MACS sorted populations were incubated (as previously) with anti-Human Fc PE (1 μl per 1×10.sup.6 cells) (Cambridge Bioscience, Cat. No. 409304). A gate was drawn on a histogram plot for high Fc Expression for the three libraries. This took 8.16% of the gated cells for KB_A12 ProTxIII Set A Library, whilst 0.15% of the control KB_A12 ProTxIII was found in the region. 7.25% of the gated cells for KB_A12 ProTxIII Set B Library was taken, whilst 0.10% of the control was found in this region. 12.6% of the KB_A12 HsTxI Library was taken, whilst 0.03% of the KB_A12 HsTxI control was found in this region. 0.5×10.sup.6 cells for each population were taken for genomic DNA extraction. DNA encoding the IgG was amplified by nested PCR using KOD Hot Start DNA polymerase (Merck Millipore) as described in Example 4. PCR products were gel purified and digested with NheI and XhoI, cloned into the pINT3 mammalian expression vector and used to transform E. coli DH10B cells.

(397) By following the same methods as described in Examples 4 and 5, where mammalian cell display mutant libraries were created which were then selected on the basis of surface display level, individual knotbody clones will be identified with improved biophysical properties compared with their parental molecules. This example has therefore described the utility of higher eukaryotic mammalian display to improve the biophysical properties of knotbodies or any alternative bi-specific format. The methods described here could also be performed during a knotbody or bi-specific discovery project where multi-parameter FACS can be employed to select both on high display level and specific binding to a target molecule.

REFERENCES

(398) 1 Jarasch, A. et al. Developability assessment during the selection of novel therapeutic antibodies. J Pharm Sci 104, 1885-1898, doi:10.1002/jps.24430 (2015). 2 Jain, T. et al. Biophysical properties of the clinical-stage antibody landscape. Proceedings of the National Academy of Sciences of the United States of America 114, 944-949, doi:10.1073/pnas.1616408114 (2017). 3 Yang, X. et al. Developability studies before initiation of process development: improving manufacturability of monoclonal antibodies. mAbs 5, 787-794, doi:10.4161/mabs.25269 (2013). 4 Ratanji, K. D., Derrick, J. P., Dearman, R. J. & Kimber, I. Immunogenicity of therapeutic proteins: influence of aggregation. J Immunotoxicol 11, 99-109, doi:10.3109/1547691X.2013.821564 (2014). 5 Moussa, E. M. et al. Immunogenicity of Therapeutic Protein Aggregates. J Pharm Sci 105, 417-430, doi:10.1016/j.xphs.2015.11.002 (2016). 6 Buchanan, A. et al. Engineering a therapeutic IgG molecule to address cysteinylation, aggregation and enhance thermal stability and expression. mAbs 5, 255-262, doi:10.4161/mabs.23392 (2013). 7 Dobson, C. L. et al. Engineering the surface properties of a human monoclonal antibody prevents self-association and rapid clearance in vivo. Scientific Reports 6, 38644, doi:10.1038/srep38644 (2016). 8 Schofield, D. J. et al. Application of phage display to high throughput antibody generation and characterization. Genome biology 8, R254, doi:10.1186/gb-2007-8-11-r254 (2007). 9 Jespers, L., Schon, O., Famm, K. & Winter, G. Aggregation-resistant domain antibodies selected on phage by heat denaturation. Nat Biotechnol 22, 1161-1165, doi:10.1038/nbt1000 (2004). 10 Fennell, B. J. et al. CDR-restricted engineering of native human scFvs creates highly stable and soluble bifunctional antibodies for subcutaneous delivery. mAbs 5, 882-895, doi:10.4161/mabs.26201 (2013). 11 Christ, D., Famm, K. & Winter, G. Repertoires of aggregation-resistant human antibody domains. Protein engineering, design & selection: PEDS 20, 413-416, doi:10.1093/protein/gzm037 (2007). 12 Tiller, T. et al. A fully synthetic human Fab antibody library based on fixed VH/VL framework pairings with favorable biophysical properties. mAbs 5, 445-470, doi:10.4161/mabs.24218 (2013). 13 Dudgeon, K. et al. General strategy for the generation of human antibody variable domains with increased aggregation resistance. Proceedings of the National Academy of Sciences of the United States of America 109, 10879-10884, doi:10.1073/pnas.1202866109 (2012). 14 Bethea, D. et al. Mechanisms of self-association of a human monoclonal antibody CNTO607. Protein Engineering Design and Selection 25, 531-538 (2012). 15 Kelly, R. L. et al. High throughput cross-interaction measures for human IgG1 antibodies correlate with clearance rates in mice. mAbs 7, 770-777, doi:10.1080/19420862.2015.1043503 (2015). 16 Hötzel, I. et al. A strategy for risk mitigation of antibodies with fast clearance. mAbs 4, 753-760, doi:10.4161/mabs.22189 (2012). 17 Pries, A. R., Secomb, T. W. & Gaehtgens, P. The endothelial surface layer. Pflugers Arch 440, 653-666, doi:10.1007/s004240000307 (2000). 18 Nieuwdorp, M. et al. The endothelial glycocalyx: a potential barrier between health and vascular disease. Curr Opin Lipidol 16, 507-511 (2005). 19 Xu, Y. et al. Addressing polyspecificity of antibodies selected from an in vitro yeast presentation system: a FACS-based, high-throughput selection and analytical tool. Protein engineering, design & selection: PEDS 26, 663-670, doi:10.1093/protein/gzt047 (2013). 20 Martin, W. L., West, A. P., Jr., Gan, L. & Bjorkman, P. J. Crystal structure at 2.8 A of an FcRn/heterodimeric Fc complex: mechanism of pH-dependent binding. Mol Cell 7, 867-877 (2001). 21 Suzuki, T. et al. Importance of neonatal FcR in regulating the serum half-life of therapeutic proteins containing the Fc domain of human IgG1: a comparative study of the affinity of monoclonal antibodies and Fc-fusion proteins to human neonatal FcR. Journal of immunology (Baltimore, Md.: 1950) 184, 1968-1976, doi:10.4049/jimmunol.0903296 (2010). 22 Schoch, A. et al. Charge-mediated influence of the antibody variable domain on FcRn-dependent pharmacokinetics. Proceedings of the National Academy of Sciences of the United States of America 112, 5997-6002, doi:10.1073/pnas.1408766112 (2015). 23 Kelly, R. L. et al. Target-independent variable region mediated effects on antibody clearance can be FcRn independent. mAbs 8, 1269-1275, doi:10.1080/19420862.2016.1208330 (2016). 24 Boder, E. T. & Wittrup, K. D. Yeast surface display for screening combinatorial polypeptide libraries. Nat Biotechnol 15, 553-557, doi:10.1038/nbt0697-553 (1997). 25 Chao, G. et al. Isolating and engineering human antibodies using yeast surface display. Nat. Protocols 1, 755-768, (2006). 26 Shusta, E. V., Kieke, M. C., Parke, E., Kranz, D. M. & Wittrup, K. D. Yeast polypeptide fusion surface display levels predict thermal stability and soluble secretion efficiency. Journal of molecular biology 292, 949-956, (1999). 27 Kowalski, J. M., Parekh, R. N., Mao, J. & Wittrup, K. D. Protein Folding Stability Can Determine the Efficiency of Escape from Endoplasmic Reticulum Quality Control. J. Biol. Chem. 273, 19453-19458 (1998). 28 Kowalski, J. M., Parekh, R. N. & Wittrup, K. D. Secretion Efficiency in Saccharomyces cerevisiae of Bovine Pancreatic Trypsin Inhibitor Mutants Lacking Disulfide Bonds Is Correlated with Thermodynamic Stability. Biochemistry 37, 1264-1273, doi:10.1021/bi9722397 (1998). 29 Hackel, B. J., Ackerman, M. E., Howland, S. W. & Wittrup, K. D. Stability and CDR Composition Biases Enrich Binder Functionality Landscapes. Journal of molecular biology 401, 84-96, (2010). 30 Julian, M. C. et al. Co-evolution of affinity and stability of grafted amyloid-motif domain antibodies. Protein Eng. Des. Sel. 28, 339-350, doi:10.1093/protein/gzv050 (2015). 31 Rabia, L. A., Desai, A. A., Jhajj, H. S. & Tessier, P. M. Understanding and overcoming trade-offs between antibody affinity, specificity, stability and solubility. Biochemical Engineering Journal 137, 365-374, (2018). 32 Igawa, T., Haraya, K. & Hattori, K. Sweeping antibody as a novel therapeutic antibody modality capable of eliminating soluble antigens from circulation. Immunol Rev 270, 132-151, doi:10.1111/imr.12392 (2016). 33 Qin, J. Y. et al. Systematic comparison of constitutive promoters and the doxycycline-inducible promoter. PloS one 5, e10611, doi:10.1371/journal.pone.0010611 (2010). 34 Walker, J. M. The protein protocols handbook. 3rd edn, (Humana Press, 2009). 35 Janson, J. C. Protein purification: principles, high resolution methods, and applications. 3rd edn, (John Wiley & Sons, 2011). 36 Schiel, J. E., Davis, D. L. & Borisov, O. State-of-the-art and emerging technologies for therapeutic monoclonal antibody characterization. (American Chemical Society, 2014). 37 Saro, D. et al. in State-of-the-Art and Emerging Technologies for Therapeutic Monoclonal Antibody Characterization Volume 2. Biopharmaceutical Characterization: The NISTmAb Case Study Vol. 1201 ACS Symposium Series Ch. 7, 329-355 (American Chemical Society, 2015). 38 Sun, T. et al. High throughput detection of antibody self-interaction by bio-layer interferometry. mAbs 5, 838-841, doi:10.4161/mabs.26186 (2013). 39 Liu, Y. et al. High-throughput screening for developability during early-stage antibody discovery using self-interaction nanoparticle spectroscopy. mAbs 6, 483-492, doi:10.4161/mabs.27431 (2014). 40 Geng, S. B., Wittekind, M., Vigil, A. & Tessier, P. M. Measurements of Monoclonal Antibody Self-Association Are Correlated with Complex Biophysical Properties. Molecular Pharmaceutics 13, 1636-1645, doi:10.1021/acs.molpharmaceut.6b00071 (2016). 41 Wu, J. et al. Discovery of highly soluble antibodies prior to purification using affinity-capture self-interaction nanoparticle spectroscopy. Protein engineering, design & selection: PEDS 28, 403-414, doi:10.1093/protein/gzv045 (2015). 42 Nobbmann, U. et al. Dynamic light scattering as a relative tool for assessing the molecular integrity and stability of monoclonal antibodies. Biotechnology and Genetic Engineering Reviews 24, 117-128 (2007). 43 Geng, S. B., Cheung, J. K., Narasimhan, C., Shameem, M. & Tessier, P. M. Improving monoclonal antibody selection and engineering using measurements of colloidal protein interactions. J Pharm Sci 103, 3356-3363, doi:10.1002/jps.24130 (2014). 44 Howlett, G. J., Minton, A. P. & Rivas, G. Analytical ultracentrifugation for the study of protein association and assembly. Curr Opin Chem Biol 10, 430-436, doi:10.1016/j.cbpa.2006.08.017 (2006). 45 Salinas, B. A. et al. Understanding and modulating opalescence and viscosity in a monoclonal antibody formulation. J Pharm Sci 99, 82-93, doi:10.1002/jps.21797 (2010). 46 Gabel, F. et al. Protein dynamics studied by neutron scattering. Q Rev Biophys 35, 327-367 (2002). 47 Jacobs, S. A., Wu, S. J., Feng, Y., Bethea, D. & O'Neil, K. T. Cross-interaction chromatography: A rapid method to identify highly soluble monoclonal antibody candidates. Pharm. Res. 27, 65-71, doi:10.1007/s11095-009-0007-z (2010). 48 Josic, D., Bal, F. & Schwinn, H. Isolation of plasma proteins from the clotting cascade by heparin affinity chromatography. J Chromatogr 632, 1-10 (1993). 49 Fekete, S., Veuthey, J. L., Beck, A. & Guillarme, D. Hydrophobic interaction chromatography for the characterization of monoclonal antibodies and related products. J Pharm Biomed Anal 130, 3-18, doi:10.1016/j.jpba.2016.04.004 (2016). 50 Eriksson, K. O. & Belew, M. Hydrophobic interaction chromatography. Methods Biochem Anal 54, 165-181 (2011). 51 Bakalova, R. & Ohba, H. Interaction of soybean agglutinin with leukemic T-cells and its use for their in vitro separation from normal lymphocytes by lectin-affinity chromatography. Biomed Chromatogr 17, 239-249, doi:10.1002/bmc.218 (2003). 52 Melidoni, A. N., Dyson, M. R., Wormald, S. & McCafferty, J. Selecting antagonistic antibodies that control differentiation through inducible expression in embryonic stem cells. Proceedings of the National Academy of Sciences of the United States of America 110, 17802-17807, doi:10.1073/pnas.1312062110 (2013). 53 Xie, J., Zhang, H., Yea, K. & Lerner, R. A. Autocrine signaling based selection of combinatorial antibodies that transdifferentiate human stem cells. Proceedings of the National Academy of Sciences 110, 8099-8104, doi:10.1073/pnas.1306263110 (2013). 54 Zhang, H. et al. Selecting Agonists from Single Cells Infected with Combinatorial Antibody Libraries. Chemistry & Biology 20, 734-741, doi:10.1016/j.chembiol.2013.04.012 (2013). 55 Igoucheva, O., Alexeev, V. & Yoon, K. Targeted gene correction by small single-stranded oligonucleotides in mammalian cells. Gene Therapy 8, 391, doi:10.1038/sj.gt.3301414 (2001). 56 Liang, X., Potter, J., Kumar, S., Ravinder, N. & Chesnut, J. D. Enhanced CRISPR/Cas9-mediated precise genome editing by improved design and delivery of gRNA, Cas9 nuclease, and donor DNA. J. Biotechnol. 241, 136-146, (2017). 57 Dyson, M. R. et al. Mapping protein interactions by combining antibody affinity maturation and mass spectrometry. Analytical biochemistry 417, 25-35, doi:10.1016/j.ab.2011.05.005 (2011). 58 Gronwald, R. G. et al. Cloning and expression of a cDNA coding for the human platelet-derived growth factor receptor: evidence for more than one receptor class. Proceedings of the National Academy of Sciences of the United States of America 85, 3435-3439 (1988). 59 Dux, R., Kindler-Rohrborn, A., Lennartz, K. & Rajewsky, M. F. Calibration of fluorescence intensities to quantify antibody binding surface determinants of cell subpopulations by flow cytometry. Cytometry 12, 422-428, doi:10.1002/cyto.990120507 (1991). 60 Brockhoff, G., Hofstaedter, F. & Knuechel, R. Flow cytometric detection and quantitation of the epidermal growth factor receptor in comparison to Scatchard analysis in human bladder carcinoma cell lines. Cytometry 17, 75-83, doi:10.1002/cyto.990170110 (1994). 61 Gordon, I. L. Scatchard analysis of fluorescent concanavalin A binding to lymphocytes. Cytometry 20, 238-244, doi:10.1002/cyto.990200307 (1995). 62 Sheehan, J. & Marasco, W. A. Phage and Yeast Display. Microbiol Spectr 3, AID-0028-2014, doi:10.1128/microbiolspec.AID-0028-2014 (2015). 63 Puck, T. T. The genetics of somatic mammalian cells. Adv Biol Med Phys 5, 75-101 (1957). 64 Wurm, F. CHO Quasispecies—Implications for Manufacturing Processes. Processes 1, 296-311 (2013). 65 Goh, J. B. & Ng, S. K. Impact of host cell line choice on glycan profile. Critical Reviews in Biotechnology 38, 851-867, doi:10.1080/07388551.2017.1416577 (2018). 66 Skerra, A. Alternative non-antibody scaffolds for molecular recognition. Curr Opin Biotechnol 18, 295-304, doi:10.1016/j.copbio.2007.04.010 (2007). 67 Gebauer, M. & Skerra, A. Engineered protein scaffolds as next-generation antibody therapeutics. Curr Opin Chem Biol 13, 245-255, doi:10.1016/j.cbpa.2009.04.627 (2009). 68 Tiede, C. et al. Adhiron: a stable and versatile peptide display scaffold for molecular recognition applications. Protein engineering, design & selection: PEDS 27, 145-155, doi:10.1093/protein/gzu007 (2014). 69 Koide, A., Bailey, C. W., Huang, X. & Koide, S. The fibronectin type III domain as a scaffold for novel binding proteins. Journal of molecular biology 284, 1141-1151, doi:10.1006/jmbi.1998.2238 (1998). 70 Nygren, P. A. & Skerra, A. Binding proteins from alternative scaffolds. J Immunol Methods 290, 3-28, doi:10.1016/j.jim.2004.04.006 (2004). 71 Chang, H. J. et al. Molecular evolution of cystine-stabilized miniproteins as stable proteinaceous binders. Structure 17, 620-631, doi:10.1016/j.str.2009.01.011 (2009). 72 Ward, E. S., Güssow, D., Griffiths, A. D., Jones, P. T. & Winter, G. Binding activities of a repertoire of single immunoglobulin variable domains secreted from Escherichia coli. Nature 341, 544, doi:10.1038/341544a0 (1989). 73 McCafferty, J., Griffiths, A. D., Winter, G. & Chiswell, D. J. Phage antibodies: filamentous phage displaying antibody variable domains. Nature 348, 552-554, doi:10.1038/348552a0 (1990). 74 Holt, L. J., Herring, C., Jespers, L. S., Woolven, B. P. & Tomlinson, I. M. Domain antibodies: proteins for therapy. Trends in Biotechnology 21, 484-490, doi:10.1016/j.tibtech.2003.08.007 (2003). 75 Huston, J. S. et al. Protein engineering of antibody binding sites: recovery of specific activity in an anti-digoxin single-chain Fv analogue produced in Escherichia coli. Proceedings of the National Academy of Sciences 85, 5879 (1988). 76 Bird, R. E. et al. Single-chain antigen-binding proteins. Science 242, 423-426 (1988). 77 Holliger, P., Prospero, T. & Winter, G. “Diabodies”: Small bivalent and bispecific antibody fragments. Proceedings of the National Academy of Sciences of the United States of America 90, 6444-6448 (1993). 78 Reiter, Y., Brinkmann, U., Lee, B. & Pastan, I. Engineering antibody Fv fragments for cancer detection and therapy: Bisulfide-stabilized Fv fragments. Nat. Biotechnol. 14, 1239, doi:10.1038/nbt1096-1239 (1996). 79 Holliger, P. & Hudson, P. J. Engineered antibody fragments and the rise of single domains. Nat Biotechnol 23, 1126-1136, doi:10.1038/nbt1142 (2005). 80 Knappik, A. et al. Fully synthetic human combinatorial antibody libraries (HuCAL) based on modular consensus frameworks and CDRs randomized with trinucleotides. Journal of molecular biology 296, 57-86, doi:10.1006/jmbi.1999.3444 (2000). 81 Krebs, B. et al. High-throughput generation and engineering of recombinant human antibodies. J Immunol Methods 254, 67-84 (2001). 82 Holliger, P. & Bohlen, H. Engineering antibodies for the clinic. Cancer Metastasis Rev 18, 411-419 (1999). 83 Glennie, M. J., McBride, H. M., Worth, A. T. & Stevenson, G. T. Preparation and performance of bispecific F(ab′ gamma)2 antibody containing thioether-linked Fab′ gamma fragments. Journal of immunology (Baltimore, Md.: 1950) 139, 2367-2375 (1987). 84 Repp, R. et al. G-CSF-stimulated PMN in immunotherapy of breast cancer with a bispecific antibody to Fc gamma RI and to HER-2/neu (MDX-210). J Hematother 4, 415-421, doi:10.1089/scd.1.1995.4.415 (1995). 85 Suresh, M. R., Cuello, A. C. & Milstein, C. Bispecific monoclonal antibodies from hybrid hybridomas. Methods in enzymology 121, 210-228 (1986). 86 Staerz, U. D. & Bevan, M. J. Hybrid hybridoma producing a bispecific monoclonal antibody that can focus effector T-cell activity. Proceedings of the National Academy of Sciences 83, 1453 (1986). 87 Merchant, A. M. et al. An efficient route to human bispecific IgG. Nat. Biotechnol. 16, 677, doi:10.1038/nbt0798-677 (1998). 88 Von Kreudenstein, T. S. et al. Improving biophysical properties of a bispecific antibody scaffold to aid developability: quality by molecular design. mAbs 5, 646-654, doi:10.4161/mabs.25632 (2013). 89 Ghetie, V. et al. Increasing the serum persistence of an IgG fragment by random mutagenesis. Nat Biotechnol 15, 637-640, doi:10.1038/nbt0797-637 (1997). 90 Borrok, M. J. et al. An “Fc-Silenced” IgG1 Format With Extended Half-Life Designed for Improved Stability. J Pharm Sci 106, 1008-1017, doi:10.1016/j.xphs.2016.12.023 (2017). 91 Schlothauer, T. et al. Analytical FcRn affinity chromatography for functional characterization of monoclonal antibodies. mAbs 5, 576-586, doi:10.4161/mabs.24981 (2013). 92 Gossen, M. et al. Transcriptional activation by tetracyclines in mammalian cells. Science 268, doi:10.1126/science.7792603 (1995). 93 Sadelain, M., Papapetrou, E. P. & Bushman, F. D. Safe harbours for the integration of new DNA in the human genome. Nature reviews. Cancer 12, 51-58, doi:10.1038/nrc3179 (2011). 94 Foote, J. & Winter, G. Antibody framework residues affecting the conformation of the hypervariable loops. Journal of molecular biology 224, 487-499 (1992). 95 Haghparast, S. M., Kihara, T. & Miyake, J. Distinct mechanical behavior of HEK293 cells in adherent and suspended states. Peer J 3, e1131, doi:10.7717/peerj.1131 (2015). 96 Rivera, V. M. et al. Regulation of protein secretion through controlled aggregation in the endoplasmic reticulum. Science 287, 826-830 (2000). 97 Ferri, N., Corsini, A., Sirtori, C. R. & Ruscica, M. Bococizumab for the treatment of hypercholesterolaemia. Expert Opin Biol Ther 17, 237-243, doi:10.1080/14712598.2017.1279602 (2017). 98 Liang, H. et al. Proprotein Convertase Substilisin/Kexin Type 9 Antagonism Reduces Low-Density Lipoprotein Cholesterol in Statin-Treated Hypercholesterolemic Nonhuman Primates. Journal of Pharmacology and Experimental Therapeutics 340, 228 (2012). 99 Jones, T. D., Crompton, L. J., Carr, F. J. & Baker, M. P. in Therapeutic Antibodies: Methods and Protocols (ed Antony S. Dimitrov) 405-423 (Humana Press, 2009). 100 Bostrom, J., Lee, C. V., Haber, L. & Fuh, G. in Therapeutic Antibodies: Methods and Protocols (ed Antony S. Dimitrov) 353-376 (Humana Press, 2009). 101 Dyson, M. R., Shadbolt, S. P., Vincent, K. J., Perera, R. L. & McCafferty, J. Production of soluble mammalian proteins in Escherichia coli: identification of protein features that correlate with successful expression. BMC biotechnology 4, 32, doi:10.1186/1472-6750-4-32 (2004). 102 Neylon, C. Chemical and biochemical strategies for the randomization of protein encoding DNA sequences: library construction methods for directed evolution. Nucleic acids research 32, 1448-1459, doi:10.1093/nar/gkh315 (2004). 103 Fujino, Y. et al. Robust in vitro affinity maturation strategy based on interface-focused high-throughput mutational scanning. Biochemical and biophysical research communications 428, 395-400, doi:10.1016/j.bbrc.2012.10.066 (2012). 104 Xin, Y. et al. Anti-neuropilin-1 (MNRP1685A): unexpected pharmacokinetic differences across species, from preclinical models to humans. Pharm Res 29, 2512-2521, doi:10.1007/s11095-012-0781-x (2012). 105 Patnaik, A. et al. A Phase Ib study evaluating MNRP1685A, a fully human anti-NRP1 monoclonal antibody, in combination with bevacizumab and paclitaxel in patients with advanced solid tumors. Cancer Chemother Pharmacol 73, 951-960, doi:10.1007/s00280-014-2426-8 (2014). 106 Shin, D. S. & Ribas, A. The evolution of checkpoint blockade as a cancer therapy: What's here, what's next? Curr. Opin. Immunol. 33, 23-35, doi:10.1016/j.coi.2015.01.006 (2015). 107 Mouquet, H. et al. Polyreactivity increases the apparent affinity of anti-HIV antibodies by heteroligation. Nature 467, 591-595, doi:10.1038/nature09385 (2010). 108 Fellouse, F. A. & Sidhu, S. S. in Making and Using Antibodies: A Practical Handbook (eds G. C. Howard & M. R. Kaser) Ch. 8, 157-180 (CRC Press, 2007). 109 Peterson, M. L. Mechanisms controlling production of membrane and secreted immunoglobulin during B cell development. Immunol Res 37, 33-46 (2007). 110 Peterson, M. L. & Perry, R. P. The regulated production of mu m and mu s mRNA is dependent on the relative efficiencies of mu s poly(A) site usage and the c mu 4-to-M1 splice. Mol Cell Biol 9, 726-738 (1989). 111 Yu, B., Wages, J. M. & Larrick, J. W. Antibody-membrane switch (AMS) technology for facile cell line development. Protein Engineering Design and Selection 27, 309-315, doi:10.1093/protein/gzu039 (2014). 112 Wang, C. et al. In vitro characterization of the anti-PD-1 antibody nivolumab, BMS-936558, and in vivo toxicology in non-human primates. Cancer Immunol Res 2, 846-856, doi:10.1158/2326-6066.CIR-14-0040 (2014). 113 D'Hautcourt, J. L. in Current Protocols in Cytometry (John Wiley & Sons, Inc., 2001). 114 Gossen, M. & Bujard, H. Tight control of gene expression in mammalian cells by tetracycline-responsive promoters. Proc Natl Acad Sci USA 89, doi:10.1073/pnas.89.12.5547 (1992). 115 Ryan, M. D. & Drew, J. Foot-and-mouth disease virus 2A oligopeptide mediated cleavage of an artificial polyprotein. The EMBO Journal 13, 928-933 (1994). 116 Zhou, X., Vink, M., Klaver, B., Berkhout, B. & Das, A. T. Optimization of the Tet-On system for regulated gene expression through viral evolution. Gene Ther 13, 1382-1390, doi:10.1038/sj.gt.3302780 (2006). 117 Loew, R., Heinz, N., Hampf, M., Bujard, H. & Gossen, M. Improved Tet-responsive promoters with minimized background expression. BMC biotechnology 10, 81, doi:10.1186/1472-6750-10-81 (2010). 118 To, T. L. & Maheshri, N. Noise can induce bimodality in positive transcriptional feedback loops without bistability. Science 327, 1142-1145, doi:10.1126/science.1178962 (2010). 119 Mullick, A. et al. The cumate gene-switch: a system for regulated expression in mammalian cells. BMC biotechnology 6, 1-18, doi:10.1186/1472-6750-6-43 (2006). 120 Yao, F. et al. Tetracycline repressor, tetR, rather than the tetR-mammalian cell transcription factor fusion derivatives, regulates inducible gene expression in mammalian cells. Hum Gene Ther 9, doi:10.1089/hum.1998.9.13-1939 (1998). 121 Wang, Y., O'Malley, B. W., Jr., Tsai, S. Y. & O'Malley, B. W. A regulatory system for use in gene transfer. Proceedings of the National Academy of Sciences of the United States of America 91, 8180-8184 (1994). 122 Datta-Mannan, A. et al. The interplay of non-specific binding, target-mediated clearance and FcRn interactions on the pharmacokinetics of humanized antibodies. mAbs 7, 1084-1093, doi:10.1080/19420862.2015.1075109 (2015). 123 Sigounas, G., Harindranath, N., Donadel, G. & Notkins, A. L. Half-life of polyreactive antibodies. J Clin Immunol 14, 134-140 (1994). 124 Roopenian, D. C. & Akilesh, S. FcRn: the neonatal Fc receptor comes of age. Nat Rev Immunol 7, 715-725, doi:10.1038/nri2155 (2007). 125 Eid, J. et al. Real-time DNA sequencing from single polymerase molecules. Science 323, 133-138, doi:10.1126/science.1162986 (2009). 126 Giordano, F. et al. De novo yeast genome assemblies from MinION, PacBio and MiSeq platforms. Scientific Reports 7, 3935, doi:10.1038/s41598-017-03996-z (2017). 127 Ehrenmann, F., Duroux, P., Giudicelli, V. & Lefranc, M. P. Standardized Sequence and Structure Analysis of Antibody Using IMGT®. 11-31, doi:10.1007/978-3-642-01147-4_2 (2010). 128 Van Blarcom, T. et al. Productive common light chain libraries yield diverse panels of high affinity bispecific antibodies. mAbs 10, 256-268, doi:10.1080/19420862.2017.1406570 (2018). 129 Spiess, C., Zhai, Q. & Carter, P. J. Alternative molecular formats and therapeutic applications for bispecific antibodies. Molecular Immunology 67, 95-106, (2015). 130 Husain, B. & Ellerman, D. Expanding the Boundaries of Biotherapeutics with Bispecific Antibodies. BioDrugs 32, 441-464, doi:10.1007/s40259-018-0299-9 (2018). 131 Fischer, N. et al. Exploiting light chains for the scalable generation and platform purification of native human bispecific IgG. Nat Commun 6, 6113, doi:10.1038/ncomms7113 (2015). 132 Ridgway, J. B. B., Presta, L. G. & Carter, P. ‘Knobs-into-holes’ engineering of antibody CH3 domains for heavy chain heterodimerization. Protein Eng. Des. Sel. 9, 617-621, doi:10.1093/protein/9.7.617 (1996). 133 Wozniak-Knopp, G. et al. Introducing antigen-binding sites in structural loops of immunoglobulin constant domains: Fc fragments with engineered HER2/neu-binding sites and antibody properties. Protein engineering, design & selection: PEDS 23, 289-297, doi:10.1093/protein/gzq005 (2010). 134 Traxlmayr, M. W. et al. Directed evolution of Her2/neu-binding IgG1-Fc for improved stability and resistance to aggregation by using yeast surface display. Protein engineering, design & selection: PEDS 26, 255-265, doi:10.1093/protein/gzs102 (2013). 135 Mahlangu, J. et al. Emicizumab Prophylaxis in Patients Who Have Hemophilia A without Inhibitors. N Engl J Med 379, 811-822, doi:10.1056/NEJMoa1803550 (2018). 136 Sampei, Z. et al. Identification and Multidimensional Optimization of an Asymmetric Bispecific IgG Antibody Mimicking the Function of Factor VIII Cofactor Activity. PloS one 8, e57479, doi:10.1371/journal.pone.0057479 (2013). 137 Tabrizi, M. A., Tseng, C. M. & Roskos, L. K. Elimination mechanisms of therapeutic monoclonal antibodies. Drug Discov Today 11, 81-88, doi:10.1016/S1359-6446(05)03638-X (2006). 138 Dostalek, M., Gardner, I., Gurbaxani, B. M., Rose, R. H. & Chetty, M. Pharmacokinetics, pharmacodynamics and physiologically-based pharmacokinetic modelling of monoclonal antibodies. Clin Pharmacokinet 52, 83-124, doi:10.1007/s40262-012-0027-4 (2013). 139. Sondermann, P. & Szymkowski, D. E. Harnessing Fc receptor biology in the design of therapeutic antibodies. Curr Opin Immunol 40, 78-87 (2016). 140. Park, H. I., Yoon, H. W. & Jung, S. T. The Highly Evolvable Antibody Fc Domain. Trends in Biotechnology 34, 895-908 (2016). 141. Hogarth, P. M. & Pietersz, G. A. Fc receptor-targeted therapies for the treatment of inflammation, cancer and beyond. Nature reviews. Drug discovery 11, 311-331 (2012). 142. Mayes, P. A., Hance, K. W. & Hoos, A. The promise and challenges of immune agonist antibody development in cancer. Nature reviews. Drug discovery 17, 509-527 (2018). 143. Jacobsen, F. W. et al. Engineering an IgG scaffold lacking effector function with optimized developability. J. Biol. Chem. (2016).