Protein Engineering, Design & Selection, 2016, vol. 29 no. 10, pp. 419 –426 doi: 10.1093/protein/gzw024 Advance Access Publication Date: 21 June 2016 Short Communication
Short Communication
Computationally driven antibody engineering enables simultaneous humanization and thermostabilization Yoonjoo Choi1, Christian Ndong2, Karl E. Griswold2,3,4,*, and Chris Bailey-Kellogg1,* 1
Department of Computer Science, Dartmouth College, Hanover, NH 03755, USA, 2Thayer School of Engineering, Dartmouth College, Hanover, NH 03755, USA, 3Norris Cotton Cancer Center at Dartmouth, Lebanon, NH 03766, USA, and 4Department of Biological Sciences, Dartmouth, Hanover, NH 03755, USA *To whom correspondence should be addressed. E-mail: [email protected] (K.E.G.); [email protected] (C.B.-K.) Edited by Brian Kuhlman Received 8 February 2016; Revised 23 May 2016; Accepted 25 May 2016
Abstract Humanization reduces the immunogenicity risk of therapeutic antibodies of non-human origin. Thermostabilization can be critical for clinical development and application of therapeutic antibodies. Here, we show that the computational antibody redesign method Computationally Driven Antibody Humanization (CoDAH) enables these two goals to be accomplished simultaneously and seamlessly. A panel of CoDAH designs for the murine parent of cetuximab, a chimeric anti-EGFR antibody, exhibited both substantially improved thermostabilities and substantially higher levels of humanness, while retaining binding activity near the parental level. The consistently high quality of the turnkey CoDAH designs, over a whole panel of variants, suggests that the computationally directed approach encapsulates key determinants of antibody structure and function. Key words: antibody, cetuximab, computational protein design, humanization, thermostability
Introduction Antibodies have emerged as the leading class of biotherapeutics (Aggarwal, 2014), but despite a long history and large market presence, developers of new antibodies continue to face a number of challenges (Lu et al., 2012). One such issue is immunogenicity risk, wherein anti-biotherapeutic antibody responses can reduce therapeutic efficacy and manifest a range of other detrimental side effects (Schellekens, 2002; Pendley et al., 2003; Roskos et al., 2004; Swann et al., 2008; Jawa et al., 2013). This issue has been substantially, though not entirely (Hwang and Foote, 2005; Harding et al., 2010), addressed by the establishment of well-defined humanization methods (Jones et al., 1986; Roguska et al., 1994; Baca et al., 1997; Gonzales et al., 2004; Dall’Acqua et al., 2005; Khee Hwang et al., 2005; Osbourn et al., 2005; Lazar et al., 2007; Dennis, 2010) and the advent of ‘fully human’ antibodies (McCafferty et al., 1990; Feldhaus et al.,
2003; Li et al., 2006; Lonberg, 2008; Duvall et al., 2011; Lee et al., 2014). Stability is another key determinant of therapeutic antibody ‘developability’ (Jarasch et al., 2015) as loss of structural integrity can degrade binding activity, cause aggregation and generally undermine therapeutic potential (Hermeling et al., 2004; Frokjaer and Otzen, 2005). Note that biotherapeutic aggregation and immunogenicity are often correlated as protein aggregates, resulting from poor thermostability or other factors, can exacerbate classical anti-drug antibody responses and produce detrimental responses via nonclassical pathways (Rosenberg, 2006; Ratanji et al., 2013). Despite accelerating growth in fully human antibody technologies, immunization of animal models followed by antibody cloning and humanization is simple, convenient and continues to be widely used (Nelson et al., 2010). Humanization via complementarity determining region (CDR) grafting is a common strategy, but unfortunately CDR
© The Author 2016. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: [email protected]
419
Y. Choi et al.
420 grafting often leads to considerable decreases in thermostability and binding affinity, requiring back substitution of non-human residues to mitigate loss of functionality (Clark, 2000). Thus, successful engineering of humanized antibodies requires simultaneous consideration of multiple objectives: humanness, thermostabililty and binding affinity. The multi-objective nature of this protein design problem led us to develop Computationally Driven Antibody Humanization (CoDAH), a structure-based computationally driven humanization method that simultaneously optimizes both humanness and stability, while maintaining affinity (Choi et al., 2015). Here, we applied CoDAH to humanize cetuximab, a chimeric antibody targeting the epidermal growth factor receptor (EGFR) (Kirkpatrick et al., 2004). We demonstrate that our computationally designed variants consistently exhibited enhanced thermostabilities (up to 6°C) and substantially improved humanness, along with binding activities equivalent to that of the parental antibody. Notably, in this study, CoDAH humanization yielded a perfect hit rate, wherein every tested design was found to exhibit high thermostability, high binding activity and increased humanness.
on mutual sequence differences (Fig. 1B). The selected CoDAH designs share numerous common mutations, presumably due to the fact that these ‘humanizing’ mutations are the most energetically favored.
CDR grafting As a comparator, a panel of humanized variants was also constructed using a conventional approach to CDR grafting. The CDRs of cetuximab (defined by the Kabat numbering scheme) were grafted onto the most similar human germline antibodies from the databases. Select murine framework residues were retained based on knowledge of packing in the immunoglobulin fold (Chothia et al., 1998), the surface accessibility (Pedersen et al., 1994), the interface regions of VH/VL (Chothia et al., 1985), the Vernier zone (Foote and Winter, 1992) and other known risk factors (Studnicka et al., 1994). In total, four CDR-grafted designs were selected for each chain (Table I, G1–G4).
Materials and methods Computational design CoDAH (Choi et al., 2015) is a structure-based protein design method that optimizes variants of a parental Fv by selecting sets of allowed mutations according to in silico evaluation of their effects on humanness and stability. Humanness The human string content (HSC) score (Lazar et al., 2007) assesses humanness as the extent of identity of nonamer peptides within a variant to corresponding nonamers within human germline antibody sequences. The heavy- (VH) and light-chain (Vκ) variable regions of cetuximab were aligned to a total of 212 unique VH and 85 Vκ human germline antibody sequences extracted from publicly available databases including VBASE and IMGT (Retter et al., 2005; Poiron et al., 2010). Stability Structure-based rotameric energies assess the energetic impacts of mutations according to a molecular mechanics force field. CoDAH uses one- and two-body energy potentials defined by the AMBER force field (Pearlman et al., 1995; Qiu et al., 1997) as implemented by OSPREY (Chen et al., 2009; Gainza et al., 2012). The Fv portion from a crystal structure of cetuximab (PDB code: 1YY8) was used to parameterize these energies. Allowed mutations To facilitate comparison with traditional CDR grafting, the allowed mutations were restricted to those from a small set of the most similar germline sequences: three for VH (IGHV4-4, 34 and 59) and two for Vκ (IGKV6D-21 and 41), which were found to be the most similar sequences to cetuximab by IgBLAST (Ye et al., 2013). Mutations were not allowed within the CDRs, as defined by the Kabat numbering scheme using the AbNum program (Abhinandan and Martin, 2008). No further restrictions were imposed on allowed mutations. CoDAH generated a set of 43 Pareto optimal (Parker et al., 2010, 2013; He et al., 2012) integrated VH/Vκ cetuximab variants, those making the best tradeoffs between the HSC score and rotameric energy (Fig. 1A). Five of these designs (Table I, C1–C5) were selected based
Fig. 1 CoDAH designs. (A) CoDAH generated 43 Pareto optimal humanized variants of cetuximab. A phylogenetic tree of the designs is shown, where each design is annotated with its corresponding HSC score. The phylogenetic tree was constructed using a neighbor joining method based on the Hamming distance between design sequence pairs (Letunic and Bork, 2007). Five different designs were selected so as to broadly sample sequence diversity and degree of humanness (filled circles), though the selected designs share several common mutations. (B) For the selected CoDAH designs (filled circles), computed rotameric energies are plotted as a function of HSC score. CoDAH cross-pairs are also shown (2-tone circles, left: VL, right: VH), as are CDR-grafted designs (hatched circles) and the cetuximab parental antibody (star).
1 1 D
Chothia Kabat Cetuximab
V V V V V V V V
V V
3 3 L
E E
E E
6 6 Q
D D A D D A
D
F F F F F F
F F
10 10 I
L
A
9 9 V
L L
12 12 V
A A
9 9 P
Q Q
Q
Q
11 11 L
K K K K K K K K K K K K
13 13 Q
T T T T T T
T T
14 14 S
E E E
E E
E
E
16 16 Q
K K
K K
K
16 16 G
T T T T T T T T T T T T
17 17 S
K K K K K K
K K
18 18 R
L L L
L L L L
20 20 I
T T T T T T T T T T T
20 20 S
P P P P P P P
40 40 S
T T T T T T
22 22 S
N N N
N
58 58 D
Y
Y
34 34 H
P P P
P
61 61 T
K K K K K K K K K K K
39 39 R
L L L
L
63 63 F
P P P
P P P P
40 40 T
K K K
K
64 64 T
D D D
D D D D D
41 41 N
T T T T T T T T T T T T
68 68 S
Q Q Q Q
Q Q Q Q Q Q
42 42 G
S S S S S S S
S S S S
70 70 N
A
A
43 43 S
V V V
V V
71 71 K
K K
K K K K
45 45 R
T T T
T T
T T T
72A 73 N
V V
V V V V
58 58 I
N N N
N N
N N N
73 76 S
T T T T T T T T T T T
74 74 S
S S S S S S S
S S S
76 79 F
A A A A A A
83 83 I
L L L
L L L
77 80 F
L L L L L
L L L
79 82 M
S S S S S S S
S S S
80 82A N
V V V V V
V V V
82 82C L
T T T T T T T
T T T
83 83 Q
A A A A A A A
84 84 S
A A A A A A A
85 85 N
V V V
V V
V V V V V
89 89 I
15 14 8 9 7 14 12 12 17 32 35
Difference (cetuximab)
17 16 17 7 6 11 21 20 14 47 45 43
Difference (cetuximab)
6 9 15 14 16 10 12 15 9
Difference (germline)
8 7 7 15 17 12 9 7 10
Difference (germline)
The last two columns are positional differences between cetuximab and the closest human antibody germline sequences (excluding CDRs). For CDR-grafted designs, knowledge-based restrictions were imposed. For G2 VH, some mutations were introduced at non-overlapping positions in CDR-H2 by the Kabat and Chothia numbering schemes (underlined italic).
E E
E E E
Q Q Q Q Q Q Q Q Q Q Q Q
Design (VH) C1 C2 C3 C4 C5 G1 G2 G3 G4 VH4-34 VH4-4 VH4-59
Design (Vκ) C1 C2 C3 C4 C5 G1 G2 G3 G4 VK6D-21 VK6D-41
5 5 K
Chothia Kabat Cetuximab
Table I. Mutations made in each humanized design (C: CoDAH and G: Grafted, see Supplementary Fig. S1)
Simultaneous humanization and thermostabilization 421
Y. Choi et al.
422
Gene construction, protein expression and antibody purification Following in-house design of both CoDAH and CDR-grafted antibodies, a commercial service provider was contracted to synthesize, clone, express via transient transfection in HEK cells, and purify all constructs (MIGS LLC, Lebanon, NH, USA). In addition to the five CoDAH designs with jointly optimized VH/Vκ pairs (Table I, C1– C5), the cross VH/Vκ pairs were also constructed, creating 20 cross pairs for a total of 25 CoDAH antibodies (also plotted in Fig. 1A). CDR grafting had no preferred pairs, so the four CDR-grafted VH were likewise paired with each of the four CDR-grafted Vκ (Table I, G1–G5), for a total of 16 pairs.
Thermostability Differential scanning fluorimetry (DSF) was performed essentially as described (Niesen et al., 2007) using an ABI 7500 Fast Real-Time PCR System from Applied Biosystems. Proteins and SYPRO Orange were diluted in phosphate-buffered saline (PBS). Final protein concentrations were 100 µg/ml, and final dye concentrations were 5×. The PCR gradient was run from 25 to 98°C with a 1 min equilibration at each degree centigrade. Fluorescence was quantified using the preset TAMRA parameters. To eliminate confounding signals from Fc denaturation, thermostability was determined using recombinant Fab fragments. Since some of the Fabs displayed multi-transition melting curves, fractional melting temperatures were computed by analyzing local transitions, ignoring those that plateaued below arbitrarily chosen
thresholds (i.e. indicating relatively minor unfolding). For example, T20* M was computed as the midpoint between the minimum intensity and the first local plateau exceeding 20% of the total fluorescence signal (ignoring transitions below 20% as well as any above the first local transition, 50* where observed). Values for T10* M and TM were likewise determined, 100* while TM indicates the midpoint between the minimum and maximum intensities. Note that for antibodies exhibiting a simple, single 20* 50* 100* are identical transition melting curve, the T10* M , TM , TM and TM (e.g. see CoDAH design C1 VH/C1 Vκ in Supplementary Table SI).
Estimated binding activity Binding activities of IgG antibodies were analyzed by biolayer interferometry on a ForteBio Octet Red instrument (ForteBio, Menlo Park, CA, USA). Purified IgG samples at 12.5 µg/ml in PBS (137 mM NaCl, 2.6 mM KCl, 10 mM Na2HPO4, 1.7 mM KH2PO4, pH 7.4) were immobilized on protein A tips (ForteBio) and exposed to 100 nM recombinant, soluble his-tagged human EGFR (Sino Biologicals, North Wales, PA, USA) in PBS. Association and dissociation rates and equilibrium affinity were determined using software provided with the instrument. Samples were analyzed as technical duplicates in two to three independent experiments, estimated KD values were normalized to that of the cetuximab parental antibody and normalized values were employed for relative ranking of antibody binding activity. While precise KD values cannot be determined by biolayer interferometry measurements at a single antigen concentration, the estimated KD value for cetuximab found here (5.8 ± 0.3 nM, average of
Fig. 2 Positional HSC of CoDAH and grafted designs in (A) heavy and (B) light chains. Note that CDR 3s are not labeled due to lack of germline sequences. (C) On average, both design strategies yield similar levels of overall HSC scores. (D) Compared with CDR grafting, CoDAH designs generally contain more exact nonamer matches against human germline antibody sequences.
Simultaneous humanization and thermostabilization
423
11 measurements in 4 separate experiments) matches that determined by Biacore analysis at 6 concentrations (5.2 nM) (Patel et al., 2007).
between the most similar grafted design (G3 VH and G4 Vκ) and its corresponding germline sequence.
Results
Thermostability and binding activity
Humanness analysis On average, CDR-grafted and CoDAH humanized antibody designs possessed similar levels of humanness. Their average HSC scores (CoDAH: 83.8 and grafted: 83.7) were similar across all of the designs (Figs. 1B and 2C), and superior to cetuximab (75.2). However, CoDAH designs tended to have more exact matches against human germline nonamers [CoDAH: 57.8 (VH 21.4 + Vκ 36.4) and grafted: 47.8 (20 + 27.8), compared with cetuximab: 14 (1 + 13); Fig. 2D and Supplementary Fig. S2]. Furthermore, the most optimized CoDAH designs (C2 or C3 VH, and C1 Vκ) were highly similar to human germline chains (Table I), with only 13 positional differences from the closest germline antibody chains (7 for VH + 6 for Vκ, excluding CDR loops), whereas there were 16 positional differences (7 + 9)
The relative stability of all constructs, formatted as Fab fragments, was analyzed by DSF (Niesen et al., 2007). Despite the absence of the Fc region, the majority of the CDR-grafted constructs and one of the CoDAH constructs exhibited complex melting curves consisting of two or more transitions (superposition in Fig. 3; individuals in Supplementary Figs. S3 and S4). To facilitate comparison of thermostabilities, a denaturation threshold was chosen to ignore transitions below 20% maximum fluorescence intensity (T20* M ); Supplementary cutoff, the Table SI considers other thresholds. Using the T20* M five Pareto optimal CoDAH designs exhibited melting temperatures 3.2–6.3°C higher than the parental cetuximab template, and all but two of the cross-paired CoDAH constructs likewise manifested enhanced thermostability (Fig. 4, Supplementary Table SI). The G2
Fig. 3 DSF melting curves of CoDAH (left) and CDR-grafted (right) designs. The fluorescence intensity values are normalized to corresponding peak values. Shown are replicate traces from two independent measurements (top and bottom, respectively). Grafted designs tended to exhibit multiple transitions, so the T20* M melting point was defined for subsequent comparison, wherein minor transitions below 20% maximum fluorescence intensity (horizontal dashed line) were ignored. Similar conclusions resulted from other arbitrarily chosen fractional melting cutoffs (Supplementary Table SI).
424
Y. Choi et al.
Fig. 4 Normalized EGFR binding activity and T20* M of CoDAH and CDR-grafted designs. Stability of CoDAH designs (circles) was consistently better than that of the parental cetuximab, while binding activities were near identical to cetuximab. CDR-grafted variants (triangles) exhibited a wider range of binding activities and stabilities. In general, grafted designs bearing G2 VH, which contains mutations nearby CDR-H2 (Table I), exhibited lower stability and lower binding than the parental cetuximab (star). Box plots summarize normalized binding (top) and T20* M (right) values.
VH/G2 Vκ CDR-grafted design exhibited the single highest T20* M of any antibody in this study, but all other CDR-grafted designs exhibited T20* M values equal to or less than that of cetuximab (Fig. 4, Supplementary Table SI). Additionally, CoDAH designs overwhelmingly exhibited single transition melting curves (Fig. 3, Supplementary Fig. S3), whereas all CDR-grafted designs manifested multiple transitions (Fig. 3, Supplementary Fig. S4). Thermostabilities of CoDAH designs were statistically significantly higher than those of the CDR-grafted designs (P-value
Short Communication
Computationally driven antibody engineering enables simultaneous humanization and thermostabilization Yoonjoo Choi1, Christian Ndong2, Karl E. Griswold2,3,4,*, and Chris Bailey-Kellogg1,* 1
Department of Computer Science, Dartmouth College, Hanover, NH 03755, USA, 2Thayer School of Engineering, Dartmouth College, Hanover, NH 03755, USA, 3Norris Cotton Cancer Center at Dartmouth, Lebanon, NH 03766, USA, and 4Department of Biological Sciences, Dartmouth, Hanover, NH 03755, USA *To whom correspondence should be addressed. E-mail: [email protected] (K.E.G.); [email protected] (C.B.-K.) Edited by Brian Kuhlman Received 8 February 2016; Revised 23 May 2016; Accepted 25 May 2016
Abstract Humanization reduces the immunogenicity risk of therapeutic antibodies of non-human origin. Thermostabilization can be critical for clinical development and application of therapeutic antibodies. Here, we show that the computational antibody redesign method Computationally Driven Antibody Humanization (CoDAH) enables these two goals to be accomplished simultaneously and seamlessly. A panel of CoDAH designs for the murine parent of cetuximab, a chimeric anti-EGFR antibody, exhibited both substantially improved thermostabilities and substantially higher levels of humanness, while retaining binding activity near the parental level. The consistently high quality of the turnkey CoDAH designs, over a whole panel of variants, suggests that the computationally directed approach encapsulates key determinants of antibody structure and function. Key words: antibody, cetuximab, computational protein design, humanization, thermostability
Introduction Antibodies have emerged as the leading class of biotherapeutics (Aggarwal, 2014), but despite a long history and large market presence, developers of new antibodies continue to face a number of challenges (Lu et al., 2012). One such issue is immunogenicity risk, wherein anti-biotherapeutic antibody responses can reduce therapeutic efficacy and manifest a range of other detrimental side effects (Schellekens, 2002; Pendley et al., 2003; Roskos et al., 2004; Swann et al., 2008; Jawa et al., 2013). This issue has been substantially, though not entirely (Hwang and Foote, 2005; Harding et al., 2010), addressed by the establishment of well-defined humanization methods (Jones et al., 1986; Roguska et al., 1994; Baca et al., 1997; Gonzales et al., 2004; Dall’Acqua et al., 2005; Khee Hwang et al., 2005; Osbourn et al., 2005; Lazar et al., 2007; Dennis, 2010) and the advent of ‘fully human’ antibodies (McCafferty et al., 1990; Feldhaus et al.,
2003; Li et al., 2006; Lonberg, 2008; Duvall et al., 2011; Lee et al., 2014). Stability is another key determinant of therapeutic antibody ‘developability’ (Jarasch et al., 2015) as loss of structural integrity can degrade binding activity, cause aggregation and generally undermine therapeutic potential (Hermeling et al., 2004; Frokjaer and Otzen, 2005). Note that biotherapeutic aggregation and immunogenicity are often correlated as protein aggregates, resulting from poor thermostability or other factors, can exacerbate classical anti-drug antibody responses and produce detrimental responses via nonclassical pathways (Rosenberg, 2006; Ratanji et al., 2013). Despite accelerating growth in fully human antibody technologies, immunization of animal models followed by antibody cloning and humanization is simple, convenient and continues to be widely used (Nelson et al., 2010). Humanization via complementarity determining region (CDR) grafting is a common strategy, but unfortunately CDR
© The Author 2016. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: [email protected]
419
Y. Choi et al.
420 grafting often leads to considerable decreases in thermostability and binding affinity, requiring back substitution of non-human residues to mitigate loss of functionality (Clark, 2000). Thus, successful engineering of humanized antibodies requires simultaneous consideration of multiple objectives: humanness, thermostabililty and binding affinity. The multi-objective nature of this protein design problem led us to develop Computationally Driven Antibody Humanization (CoDAH), a structure-based computationally driven humanization method that simultaneously optimizes both humanness and stability, while maintaining affinity (Choi et al., 2015). Here, we applied CoDAH to humanize cetuximab, a chimeric antibody targeting the epidermal growth factor receptor (EGFR) (Kirkpatrick et al., 2004). We demonstrate that our computationally designed variants consistently exhibited enhanced thermostabilities (up to 6°C) and substantially improved humanness, along with binding activities equivalent to that of the parental antibody. Notably, in this study, CoDAH humanization yielded a perfect hit rate, wherein every tested design was found to exhibit high thermostability, high binding activity and increased humanness.
on mutual sequence differences (Fig. 1B). The selected CoDAH designs share numerous common mutations, presumably due to the fact that these ‘humanizing’ mutations are the most energetically favored.
CDR grafting As a comparator, a panel of humanized variants was also constructed using a conventional approach to CDR grafting. The CDRs of cetuximab (defined by the Kabat numbering scheme) were grafted onto the most similar human germline antibodies from the databases. Select murine framework residues were retained based on knowledge of packing in the immunoglobulin fold (Chothia et al., 1998), the surface accessibility (Pedersen et al., 1994), the interface regions of VH/VL (Chothia et al., 1985), the Vernier zone (Foote and Winter, 1992) and other known risk factors (Studnicka et al., 1994). In total, four CDR-grafted designs were selected for each chain (Table I, G1–G4).
Materials and methods Computational design CoDAH (Choi et al., 2015) is a structure-based protein design method that optimizes variants of a parental Fv by selecting sets of allowed mutations according to in silico evaluation of their effects on humanness and stability. Humanness The human string content (HSC) score (Lazar et al., 2007) assesses humanness as the extent of identity of nonamer peptides within a variant to corresponding nonamers within human germline antibody sequences. The heavy- (VH) and light-chain (Vκ) variable regions of cetuximab were aligned to a total of 212 unique VH and 85 Vκ human germline antibody sequences extracted from publicly available databases including VBASE and IMGT (Retter et al., 2005; Poiron et al., 2010). Stability Structure-based rotameric energies assess the energetic impacts of mutations according to a molecular mechanics force field. CoDAH uses one- and two-body energy potentials defined by the AMBER force field (Pearlman et al., 1995; Qiu et al., 1997) as implemented by OSPREY (Chen et al., 2009; Gainza et al., 2012). The Fv portion from a crystal structure of cetuximab (PDB code: 1YY8) was used to parameterize these energies. Allowed mutations To facilitate comparison with traditional CDR grafting, the allowed mutations were restricted to those from a small set of the most similar germline sequences: three for VH (IGHV4-4, 34 and 59) and two for Vκ (IGKV6D-21 and 41), which were found to be the most similar sequences to cetuximab by IgBLAST (Ye et al., 2013). Mutations were not allowed within the CDRs, as defined by the Kabat numbering scheme using the AbNum program (Abhinandan and Martin, 2008). No further restrictions were imposed on allowed mutations. CoDAH generated a set of 43 Pareto optimal (Parker et al., 2010, 2013; He et al., 2012) integrated VH/Vκ cetuximab variants, those making the best tradeoffs between the HSC score and rotameric energy (Fig. 1A). Five of these designs (Table I, C1–C5) were selected based
Fig. 1 CoDAH designs. (A) CoDAH generated 43 Pareto optimal humanized variants of cetuximab. A phylogenetic tree of the designs is shown, where each design is annotated with its corresponding HSC score. The phylogenetic tree was constructed using a neighbor joining method based on the Hamming distance between design sequence pairs (Letunic and Bork, 2007). Five different designs were selected so as to broadly sample sequence diversity and degree of humanness (filled circles), though the selected designs share several common mutations. (B) For the selected CoDAH designs (filled circles), computed rotameric energies are plotted as a function of HSC score. CoDAH cross-pairs are also shown (2-tone circles, left: VL, right: VH), as are CDR-grafted designs (hatched circles) and the cetuximab parental antibody (star).
1 1 D
Chothia Kabat Cetuximab
V V V V V V V V
V V
3 3 L
E E
E E
6 6 Q
D D A D D A
D
F F F F F F
F F
10 10 I
L
A
9 9 V
L L
12 12 V
A A
9 9 P
Q Q
Q
Q
11 11 L
K K K K K K K K K K K K
13 13 Q
T T T T T T
T T
14 14 S
E E E
E E
E
E
16 16 Q
K K
K K
K
16 16 G
T T T T T T T T T T T T
17 17 S
K K K K K K
K K
18 18 R
L L L
L L L L
20 20 I
T T T T T T T T T T T
20 20 S
P P P P P P P
40 40 S
T T T T T T
22 22 S
N N N
N
58 58 D
Y
Y
34 34 H
P P P
P
61 61 T
K K K K K K K K K K K
39 39 R
L L L
L
63 63 F
P P P
P P P P
40 40 T
K K K
K
64 64 T
D D D
D D D D D
41 41 N
T T T T T T T T T T T T
68 68 S
Q Q Q Q
Q Q Q Q Q Q
42 42 G
S S S S S S S
S S S S
70 70 N
A
A
43 43 S
V V V
V V
71 71 K
K K
K K K K
45 45 R
T T T
T T
T T T
72A 73 N
V V
V V V V
58 58 I
N N N
N N
N N N
73 76 S
T T T T T T T T T T T
74 74 S
S S S S S S S
S S S
76 79 F
A A A A A A
83 83 I
L L L
L L L
77 80 F
L L L L L
L L L
79 82 M
S S S S S S S
S S S
80 82A N
V V V V V
V V V
82 82C L
T T T T T T T
T T T
83 83 Q
A A A A A A A
84 84 S
A A A A A A A
85 85 N
V V V
V V
V V V V V
89 89 I
15 14 8 9 7 14 12 12 17 32 35
Difference (cetuximab)
17 16 17 7 6 11 21 20 14 47 45 43
Difference (cetuximab)
6 9 15 14 16 10 12 15 9
Difference (germline)
8 7 7 15 17 12 9 7 10
Difference (germline)
The last two columns are positional differences between cetuximab and the closest human antibody germline sequences (excluding CDRs). For CDR-grafted designs, knowledge-based restrictions were imposed. For G2 VH, some mutations were introduced at non-overlapping positions in CDR-H2 by the Kabat and Chothia numbering schemes (underlined italic).
E E
E E E
Q Q Q Q Q Q Q Q Q Q Q Q
Design (VH) C1 C2 C3 C4 C5 G1 G2 G3 G4 VH4-34 VH4-4 VH4-59
Design (Vκ) C1 C2 C3 C4 C5 G1 G2 G3 G4 VK6D-21 VK6D-41
5 5 K
Chothia Kabat Cetuximab
Table I. Mutations made in each humanized design (C: CoDAH and G: Grafted, see Supplementary Fig. S1)
Simultaneous humanization and thermostabilization 421
Y. Choi et al.
422
Gene construction, protein expression and antibody purification Following in-house design of both CoDAH and CDR-grafted antibodies, a commercial service provider was contracted to synthesize, clone, express via transient transfection in HEK cells, and purify all constructs (MIGS LLC, Lebanon, NH, USA). In addition to the five CoDAH designs with jointly optimized VH/Vκ pairs (Table I, C1– C5), the cross VH/Vκ pairs were also constructed, creating 20 cross pairs for a total of 25 CoDAH antibodies (also plotted in Fig. 1A). CDR grafting had no preferred pairs, so the four CDR-grafted VH were likewise paired with each of the four CDR-grafted Vκ (Table I, G1–G5), for a total of 16 pairs.
Thermostability Differential scanning fluorimetry (DSF) was performed essentially as described (Niesen et al., 2007) using an ABI 7500 Fast Real-Time PCR System from Applied Biosystems. Proteins and SYPRO Orange were diluted in phosphate-buffered saline (PBS). Final protein concentrations were 100 µg/ml, and final dye concentrations were 5×. The PCR gradient was run from 25 to 98°C with a 1 min equilibration at each degree centigrade. Fluorescence was quantified using the preset TAMRA parameters. To eliminate confounding signals from Fc denaturation, thermostability was determined using recombinant Fab fragments. Since some of the Fabs displayed multi-transition melting curves, fractional melting temperatures were computed by analyzing local transitions, ignoring those that plateaued below arbitrarily chosen
thresholds (i.e. indicating relatively minor unfolding). For example, T20* M was computed as the midpoint between the minimum intensity and the first local plateau exceeding 20% of the total fluorescence signal (ignoring transitions below 20% as well as any above the first local transition, 50* where observed). Values for T10* M and TM were likewise determined, 100* while TM indicates the midpoint between the minimum and maximum intensities. Note that for antibodies exhibiting a simple, single 20* 50* 100* are identical transition melting curve, the T10* M , TM , TM and TM (e.g. see CoDAH design C1 VH/C1 Vκ in Supplementary Table SI).
Estimated binding activity Binding activities of IgG antibodies were analyzed by biolayer interferometry on a ForteBio Octet Red instrument (ForteBio, Menlo Park, CA, USA). Purified IgG samples at 12.5 µg/ml in PBS (137 mM NaCl, 2.6 mM KCl, 10 mM Na2HPO4, 1.7 mM KH2PO4, pH 7.4) were immobilized on protein A tips (ForteBio) and exposed to 100 nM recombinant, soluble his-tagged human EGFR (Sino Biologicals, North Wales, PA, USA) in PBS. Association and dissociation rates and equilibrium affinity were determined using software provided with the instrument. Samples were analyzed as technical duplicates in two to three independent experiments, estimated KD values were normalized to that of the cetuximab parental antibody and normalized values were employed for relative ranking of antibody binding activity. While precise KD values cannot be determined by biolayer interferometry measurements at a single antigen concentration, the estimated KD value for cetuximab found here (5.8 ± 0.3 nM, average of
Fig. 2 Positional HSC of CoDAH and grafted designs in (A) heavy and (B) light chains. Note that CDR 3s are not labeled due to lack of germline sequences. (C) On average, both design strategies yield similar levels of overall HSC scores. (D) Compared with CDR grafting, CoDAH designs generally contain more exact nonamer matches against human germline antibody sequences.
Simultaneous humanization and thermostabilization
423
11 measurements in 4 separate experiments) matches that determined by Biacore analysis at 6 concentrations (5.2 nM) (Patel et al., 2007).
between the most similar grafted design (G3 VH and G4 Vκ) and its corresponding germline sequence.
Results
Thermostability and binding activity
Humanness analysis On average, CDR-grafted and CoDAH humanized antibody designs possessed similar levels of humanness. Their average HSC scores (CoDAH: 83.8 and grafted: 83.7) were similar across all of the designs (Figs. 1B and 2C), and superior to cetuximab (75.2). However, CoDAH designs tended to have more exact matches against human germline nonamers [CoDAH: 57.8 (VH 21.4 + Vκ 36.4) and grafted: 47.8 (20 + 27.8), compared with cetuximab: 14 (1 + 13); Fig. 2D and Supplementary Fig. S2]. Furthermore, the most optimized CoDAH designs (C2 or C3 VH, and C1 Vκ) were highly similar to human germline chains (Table I), with only 13 positional differences from the closest germline antibody chains (7 for VH + 6 for Vκ, excluding CDR loops), whereas there were 16 positional differences (7 + 9)
The relative stability of all constructs, formatted as Fab fragments, was analyzed by DSF (Niesen et al., 2007). Despite the absence of the Fc region, the majority of the CDR-grafted constructs and one of the CoDAH constructs exhibited complex melting curves consisting of two or more transitions (superposition in Fig. 3; individuals in Supplementary Figs. S3 and S4). To facilitate comparison of thermostabilities, a denaturation threshold was chosen to ignore transitions below 20% maximum fluorescence intensity (T20* M ); Supplementary cutoff, the Table SI considers other thresholds. Using the T20* M five Pareto optimal CoDAH designs exhibited melting temperatures 3.2–6.3°C higher than the parental cetuximab template, and all but two of the cross-paired CoDAH constructs likewise manifested enhanced thermostability (Fig. 4, Supplementary Table SI). The G2
Fig. 3 DSF melting curves of CoDAH (left) and CDR-grafted (right) designs. The fluorescence intensity values are normalized to corresponding peak values. Shown are replicate traces from two independent measurements (top and bottom, respectively). Grafted designs tended to exhibit multiple transitions, so the T20* M melting point was defined for subsequent comparison, wherein minor transitions below 20% maximum fluorescence intensity (horizontal dashed line) were ignored. Similar conclusions resulted from other arbitrarily chosen fractional melting cutoffs (Supplementary Table SI).
424
Y. Choi et al.
Fig. 4 Normalized EGFR binding activity and T20* M of CoDAH and CDR-grafted designs. Stability of CoDAH designs (circles) was consistently better than that of the parental cetuximab, while binding activities were near identical to cetuximab. CDR-grafted variants (triangles) exhibited a wider range of binding activities and stabilities. In general, grafted designs bearing G2 VH, which contains mutations nearby CDR-H2 (Table I), exhibited lower stability and lower binding than the parental cetuximab (star). Box plots summarize normalized binding (top) and T20* M (right) values.
VH/G2 Vκ CDR-grafted design exhibited the single highest T20* M of any antibody in this study, but all other CDR-grafted designs exhibited T20* M values equal to or less than that of cetuximab (Fig. 4, Supplementary Table SI). Additionally, CoDAH designs overwhelmingly exhibited single transition melting curves (Fig. 3, Supplementary Fig. S3), whereas all CDR-grafted designs manifested multiple transitions (Fig. 3, Supplementary Fig. S4). Thermostabilities of CoDAH designs were statistically significantly higher than those of the CDR-grafted designs (P-value
