HHS Public Access
`Author manuscript
`Annu Rev Biomed Eng. Author manuscript; available in PMC 2017 February 02.
`
`Published in final edited form as:
`Annu Rev Biomed Eng. 2015 ; 17: 191–216. doi:10.1146/annurev-bioeng-071114-040733.
`
`Advances in Antibody Design
`
`Kathryn E. Tiller and Peter M. Tessier
`Center for Biotechnology and Interdisciplinary Studies, Isermann Department of Chemical and
`Biological Engineering, Rensselaer Polytechnic Institute, Troy, New York 12180
`
`Abstract
`The use of monoclonal antibodies as therapeutics requires optimizing several of their key
`attributes. These include binding affinity and specificity, folding stability, solubility,
`pharmacokinetics, effector functions, and compatibility with the attachment of additional antibody
`domains (bispecific antibodies) and cytotoxic drugs (antibody–drug conjugates). Addressing these
`and other challenges requires the use of systematic design methods that complement powerful
`immunization and in vitro screening methods. We review advances in designing the binding loops,
`scaffolds, domain interfaces, constant regions, post-translational and chemical modifications, and
`bispecific architectures of antibodies and fragments thereof to improve their bioactivity. We also
`highlight unmet challenges in antibody design that must be overcome to generate potent antibody
`therapeutics.
`
`Keywords
`IgG; scFv; VH; Fab; CDR; complementarity-determining region
`
`1. INTRODUCTION
`
`Antibodies are affinity proteins that play a central role in humoral immunity. Their ability to
`bind to foreign invaders with high affinity and specificity is central to their function. Equally
`important is their ability to serve as adaptor molecules and recruit immune cells for various
`effector functions. There are five main classes of antibodies with diverse functions:
`immunoglobulin (Ig)A IgD, IgE, IgG, and IgM (1). IgGs are the most abundant class of
`antibodies, as they constitute approximately 75% of the serum immunoglobulin repertoire.
`There are four subclasses of IgGs, which vary in their abundance and ability to elicit specific
`effector functions.
`
`The overall architecture of IgGs is conserved across its four subclasses, and consists of two
`light chains and two heavy chains (Figure 1). The light chains contain variable (VL) and
`constant (CL) domains, and the heavy chains contain one variable (VH) and three constant
`(CH1, CH2, and CH3) domains. One notable difference between IgG subclasses is the
`location of the disulfide bonds between CH1 and CL (which link the heavy and light chains)
`
`DISCLOSURE STATEMENT
`P.M.T. has received honorariums or consulting fees, or both, from MedImmune, Eli Lilly, Bristol-Myers Squibb, Janssen Biotech,
`Merck, Genentech, Amgen, Pfizer, Adimab, Abbott, AbbVie, Bayer, Roche, DuPont, and Schrödinger.
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Lassen - Exhibit 1053, p. 1
`
`

`

`Tiller and Tessier
`
`Page 2
`
`and the number of disulfide bonds in the hinge region (which link the heavy chains). The
`multidomain nature of IgGs elegantly divides their bioactivity into different subdomains.
`The antigen-binding fragment (Fab) contains both variable domains, and mediates antigen
`recognition via six peptide loops known as the complementarity-determining regions
`(CDRs). In contrast, the crystallizable fragment (Fc) contains the constant domains (CH2
`and CH3) that mediate effector function by binding to immunological receptor molecules
`such as complement proteins and Fc receptors.
`
`The multifunctional nature of IgGs is only one of the many reasons for the widespread
`interest in using monoclonal antibodies (mAbs) as therapeutics. The availability and
`refinement of robust methods for identifying and generating human mAbs, such as
`immunization and in vitro screening methods (2), have also contributed greatly to the
`interest in antibody therapeutics. In addition, mAbs typically display excellent
`pharmacokinetics (long circulation times), low toxicity and immunogenicity (for human or
`humanized mAbs), and high stability and solubility. It is also notable that the simplicity of
`expressing and purifying many different mAbs using similar platform processes is highly
`attractive from a manufacturing perspective and has enabled the production of a staggering
`number of different mAbs that are in clinical trials (3).
`
`Nevertheless, there are many challenges in generating mAbs for therapeutic applications. At
`the discovery stage, immunization affords limited control over antibody affinity and
`specificity due to the difficulty in controlling antigen presentation to the immune system. In
`vitro methods, such as phage and yeast surface display, enable improved control over
`antigen presentation. However, these display methods are limited by their need to screen
`large libraries, their typical use of antibody fragments instead of full-length antibodies, and
`their reduced quality-control mechanisms relative to mammalian systems. Moreover,
`antibodies identified via either immunization or display methods have variable and difficult-
`to-predict solubilities and viscosities at the high concentrations required for subcutaneous
`delivery (4, 5). Antibody aggregation is particularly concerning due to the potential
`immunogenicity of such aggregates (6), and abnormally high viscosity can prevent mAbs
`from being delivered via the subcutaneous route (7). It is also challenging to optimize
`bispecific antibodies that typically combine binding domains from different parent
`antibodies, given the large number of possible molecular architectures as well as the
`complex effects that these nonstandard antibody formats can have on antibody stability.
`Moreover, developing effective antibody–drug conjugates is extremely challenging due to
`the need to optimize the linker and conjugation chemistry as well as the location and number
`of attached drug molecules. Finally, it is difficult to engineer antibodies with the specific
`types and levels of effector functions that are optimal for a given therapeutic application.
`
`Although each of these challenges can be addressed through screening a large number of
`antibody variants, it is impractical to use such screening methods alone to address many of
`the challenges encountered in developing potent therapeutic antibodies. Attempts to
`optimize each antibody property sequentially are limited by the fact that improving one
`antibody attribute (such as binding affinity) can lead to defects in other attributes (such as
`solubility). Attempts to simultaneously optimize multiple antibody properties using
`mutagenesis and screening methods require libraries that are prohibitively large.
`
`Annu Rev Biomed Eng. Author manuscript; available in PMC 2017 February 02.
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Lassen - Exhibit 1053, p. 2
`
`

`

`Tiller and Tessier
`
`Page 3
`
`2. OVERVIEW OF APPROACHES FOR DESIGNING ANTIBODIES
`
`The complexity of optimizing several different antibody attributes (summarized in Figure 2)
`using traditional immunization and screening methods has led to intense interest in
`developing antibody-design methods. The most important antibody attributes are binding
`affinity and specificity, which involve optimizing the variable domains and the CDRs in
`particular. Colloidal stability (solubility) and conformational (folding) stability are also
`critical attributes of antibodies because therapeutic mAbs must be soluble for high-
`concentration delivery and stable for long-term storage. This typically requires optimizing
`solvent-exposed residues for solubility and solvent-shielded residues for conformational
`stability. The effector functions of antibodies are also critical to their bioactivity, and can be
`tailored by manipulating the hinge and Fc regions.
`
`Another increasingly important antibody attribute—which is uncommon in natural
`antibodies—is bispecificity for either multiple antigens or multiple epitopes on the same
`antigen. Achieving bispecificity requires methods for combining multiple antibodies into a
`single one as well as optimizing the key attributes of conventional antibodies. A second
`nonconventional attribute of antibodies that continues to grow in importance is their
`bioactivity when attached to small-molecule drugs. Developing antibody–drug conjugates
`(ADCs) requires optimizing many aspects of the chemistries and linkers used to derivatize
`antibodies in addition to the other key attributes of conventional antibodies.
`
`This review highlights progress in designing and optimizing each of these key antibody
`attributes. Given the large size and complexity of antibodies, most design efforts have
`focused on redesigning or optimizing existing antibodies rather than on de novo design of
`new antibodies. These design methods vary greatly, and range from knowledge-based
`methods based on previous mutagenesis results to advanced computational methods based
`on first principles. A commonality of these diverse methods is that they attempt to guide the
`design of antibodies in a systematic manner to reduce the need for screening and
`immunization methods. We discuss these design methods and their application to improve
`the properties of antibodies that are critical for their activity and stability.
`
`3. ANTIBODY BINDING AFFINITY AND SPECIFICITY
`
`The most important property of antibodies is their ability to recognize targets with high
`affinity and specificity. This binding activity is largely mediated by the CDRs. Several
`innovative approaches have been developed for designing CDRs that range from de novo
`design methods to those that involve the redesign of existing antibodies. Some of these
`design methods have used motif-grafting approaches to mimic natural protein interactions,
`and directed evolution approaches to achieve specificities for difficult-to-target antigens.
`
`3.1. De Novo Design
`
`The holy grail of antibody design is to accurately and reliably predict the sequences of
`antibodies that will bind with high affinity and specificity based solely on the sequence or
`composition of the antigen. Toward this ambitious goal, a computational approach named
`OptCDR (Optimal Complementarity Determining Regions) has been developed for
`
`Annu Rev Biomed Eng. Author manuscript; available in PMC 2017 February 02.
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Lassen - Exhibit 1053, p. 3
`
`

`

`Tiller and Tessier
`
`Page 4
`
`designing the CDRs of antibodies to recognize specific epitopes on a target antigen (8). This
`method uses canonical structures to generate CDR backbone conformations that are
`predicted to interact favorably with the antigen. Amino acids are then chosen for each
`position in the CDRs using rotamer libraries, and this process is repeated many times to
`refine the backbone structures and amino acid sequences. This leads to the prediction of
`several sets of CDR sequences, which can be grafted onto antibody scaffolds for evaluation.
`
`This approach has been tested for developing antibody–antigen complexes involving a
`hepatitis C virus capsid peptide, fluorescein, and vascular endothelial growth factor (VEGF)
`(8). The investigators predicted mutations that are expected to increase binding affinity
`(which were not evaluated experimentally) as well as evaluated mutations that had been
`reported previously to improve binding affinity for fluorescein antibodies. There is some
`correlation between predictions by OptCDR and experimental data for fluorescein
`antibodies. It will be important to further evaluate the ability of OptCDR and closely related
`methods (9) to make de novo predictions of CDR mutations as well as entire CDR
`sequences that either generate or improve antibody binding.
`
`3.2. Design by Mimicking Natural Protein Interactions
`
`Another fruitful approach for designing antibodies with specific binding activities has been
`to mimic natural protein interactions. For example, Williamson and colleagues (10) designed
`antibodies to recognize misfolded conformers of the prion protein (PrP) by mimicking
`natural interactions between cellular PrP (PrPC) and its misfolded counterpart (PrPSc).
`Previous studies had found that PrP residues 96–104 and 133–158 govern the ability of
`PrPSc to catalyze misfolding of soluble PrPC (11, 12). This led to the hypothesis that grafting
`such PrP peptides into heavy chain CDR3 (HCDR3) of an IgG—which originally lacks PrP-
`binding activity—would create antibodies that specifically recognized PrPSc (10). Indeed,
`they found that antibodies with PrP residues 89–112 or 136–158 in HCDR3 bound to PrPSc
`with apparent affinities in the low nanomolar range (2–25 nM), and these same antibodies
`weakly interacted with PrPC. Follow-up studies also identified a third region near the N
`terminus of PrP (residues 19–33) that resulted in binding activity for grafted PrP antibodies
`(13). Interestingly, grafting peptides from other PrP regions (such as those from the C-
`terminal domain) into the same CDR loop failed to generate binding activity. Moreover, the
`antibodies grafted with PrP residues 19–33 and 89–112 appear to bind via electrostatic
`interactions because mutating positively charged residues to alanine in these grafted peptides
`eliminated binding.
`
`This exciting study raises the question of whether grafting peptides from other aggregation-
`prone proteins into the CDRs of antibodies would also lead to specific binding activity. Our
`lab recently tested this question using the Alzheimer’s Aβ42 peptide (14). There are two
`hydrophobic segments within Aβ (residues 17–21 and 30–42) that mediate amyloid
`formation and are located within the β-sheet core of Aβ fibrils (15). We posited that grafting
`these peptide segments into CDR3 of a single-domain (VH) antibody would lead to antibody
`domains with Aβ-specific binding activity. Indeed, we found that grafted VH domains
`displaying the central hydrophobic region of Aβ (residues 17–21) in CDR3 bound to Aβ
`fibrils with submicromolar affinity (300–400 nM), and they weakly bound to Aβ monomers
`
`Annu Rev Biomed Eng. Author manuscript; available in PMC 2017 February 02.
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Lassen - Exhibit 1053, p. 4
`
`

`

`Tiller and Tessier
`
`Page 5
`
`or oligomers (14). Interestingly, VH domains grafted with the hydrophobic C terminus of Aβ
`(residues 30–42) bound both Aβ fibrils and oligomers with submicromolar affinity (300–700
`nM) and weakly recognized Aβ monomers. We refer to these grafted antibodies as
`gammabodies (Grafted AMyloid-Motif AntiBODIES). We have also verified that this
`grafting approach can be applied to other amyloid-forming proteins, including α-synuclein
`(associated with Parkinson’s disease) and IAPP (associated with type 2 diabetes) (16).
`Nevertheless, future work will need to develop more systematic methods for selecting
`amyloidogenic peptides for grafting because we currently do not understand why some
`sequences mediate binding and others do not. Moreover, it will be important to evaluate how
`multiple CDR loops can be engineered to display amyloidogenic peptides on the surface of
`single- and multidomain antibodies to improve the affinities of these grafted antibodies.
`
`3.3. Semirational Design Combined with Directed Evolution Methods
`
`Despite these key advances in the de novo design of CDRs, it has been extremely
`challenging to use such rational approaches to generate antibodies with subnanomolar (or
`lower) dissociation constants. Nevertheless, several innovative approaches have been
`developed that involve designing some CDR residues while randomizing others, and
`screening such libraries using in vitro display methods to select variants with high binding
`affinity and specificity. One of the first examples of this hybrid approach was the design of
`antibody libraries specific for integrins (17). The RGD sequence (arginine-glycine-aspartate)
`was inserted in the middle of HCDR3, and three flanking residues were randomized on each
`side of the RGD sequence. In addition, cysteines were introduced at each edge of HCDR3 to
`constrain the loop, which was posited to be necessary to generate high affinity for antibody
`binding that is mediated primarily via a single CDR. The investigators displayed a Fab
`library with these HCDR3 sequences on the surface of phage and screened for binding to
`integrins. Impressively, several antibody variants were identified with subnanomolar binding
`affinities, and these antibodies retained the same binding epitope as natural integrin ligands.
`This and related work (18, 19) has demonstrated the potential of using natural protein
`interactions to guide the design of high quality antibody libraries.
`
`Another example of this hybrid approach is a method for generating antibodies that
`recognize post-translational modifications (20, 21). It is difficult to isolate antibodies that
`recognize chemical modifications such as phosphorylation, especially for phosphoserine and
`phosphothreonine, because of the relatively small size of their side chains. Therefore, the
`investigators sought to introduce a common phosphate-binding motif from proteins such as
`kinases into the CDRs of antibodies (Figure 3). By identifying an antibody with a CDR loop
`(HCDR2) that naturally displays a similar anion-binding motif, the investigators first
`confirmed that this antibody bound weakly to phosphorylated peptides. Next, the affinities
`and specificities of such antibodies for phosphorylated serine, threonine, and tyrosine were
`evolved by randomizing sites within the anion-binding motif. After mutants were identified
`by phage display with selective and improved affinity for each type of modification, CDR
`residues outside the phospho-binding pocket in HCDR2 as well as in LCDR3 and HCDR3
`were randomized, and antibodies were selected for binding to different phosphorylated
`targets. Impressively, this approach generated many phospho-specific antibodies for a wide
`range of target peptides with modified serine and threonine in addition to tyrosine. Although
`
`Annu Rev Biomed Eng. Author manuscript; available in PMC 2017 February 02.
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Lassen - Exhibit 1053, p. 5
`
`

`

`Tiller and Tessier
`
`Page 6
`
`the binding affinities are modest (40–5,000 nM), they are similar to or better than those for
`previously reported phospho-specific antibodies [see (20) and references therein]. More
`importantly, this innovative approach to designing antibody libraries addresses the
`challenging problem of achieving binding specificities to subtly different antigens that are
`difficult to obtain using immunization.
`
`3.4. Antibody Redesign and Optimization
`
`Much effort has focused on using design methods to improve the binding affinity of existing
`antibodies due to the complexity of de novo design. These redesign efforts are important
`because immunization typically yields antibodies with affinities that are not high enough for
`therapeutic applications, and directed evolution approaches are limited in their ability to
`identify multiple synergistic mutations, given the unrealistically large libraries that would be
`required.
`
`One interesting study demonstrated the potential of optimizing electrostatic interactions to
`improve the binding affinity of antibodies (22). The authors used physics-based methods to
`initially evaluate the binding energy of an anti-lysozyme antibody (D1.3); every possible
`mutation was evaluated for 60 CDR positions using the crystal structure of the complex as a
`starting point. Interestingly, the mutations predicted to be most favorable were those that
`introduced large side chains into the binding interface to maximize van der Waals
`interactions, yet most of these mutations failed to improve affinity when evaluated
`experimentally. Instead, the electrostatic component of the binding energy was found to be a
`better predictor of mutations that improve affinity. This approach led to significant
`improvements in affinity (of one to two orders of magnitude) for cetuximab and an anti-
`lysozyme antibody (D44.1).
`
`The success of this study stems largely from two key types of mutations within the CDRs.
`One is the elimination of residues with unsatisfied polar groups (e.g., side chains of
`asparagine or threonine) in which desolvation is not compensated for by favorable
`interactions (e.g., hydrogen bonds) in the bound state (22). By mutating such residues to
`small hydrophobic ones, the authors observed increased binding affinity. A second key type
`of mutation is either the introduction or removal of charged residues at sites within the
`CDRs that are peripheral to the residues that contact the antigen in the nonoptimized
`complex. This finding builds on previous work demonstrating that charged residues outside
`the antibody–antigen interface but within the CDRs can increase the on-rate (and thereby the
`affinity) of an anti-VEGF antibody (23). More generally, these findings are consistent with
`other studies revealing that the likelihood of identifying beneficial mutations is higher
`outside the initial binding interface (24–27), likely due to the reduced risk of disrupting
`existing antibody–antigen interactions. This (22) and related work (24–26, 28, 29) have
`demonstrated the potential of using existing methods for calculating electrostatic
`interactions to identify mutations that improve antibody affinity.
`
`Another study elegantly showed that it is not necessary to use the crystal structures of
`antibodies or antibody–antigen complexes to guide efforts to improve affinity or alter
`binding specificity (25). The investigators sought to redesign a dengue virus antibody (4E11)
`to be broadly neutralizing. This is particularly challenging because the starting structures
`
`Annu Rev Biomed Eng. Author manuscript; available in PMC 2017 February 02.
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Lassen - Exhibit 1053, p. 6
`
`

`

`Tiller and Tessier
`
`Page 7
`
`were unknown for both the 4E11 antibody alone or when bound to its antigen (domain III of
`the dengue E protein). Nevertheless, the authors used computational docking methods to
`generate structures of the 4E11 antibody bound to four variants of the dengue E protein.
`Notably, the poor interaction between 4E11 and its antigen from one serotype (type 4) stems
`from the loss of key interactions (such as salt bridges and hydrogen bonds) that are predicted
`to be present in binding interfaces of the antibody with antigens from types 1–3. These
`insights led to the identification of several mutations that improve 4E11 binding to type 4
`without reducing binding to types 1–3. By combining successful mutations, the affinity for
`type 4 was enhanced by more than two orders of magnitude without reducing affinity for
`other serotypes. Most of the successful mutations were charged or polar, and located at the
`periphery of the binding interface, which is generally consistent with findings from other
`studies (22–24, 26). This exciting study highlights the potential of using design methods to
`achieve unique binding specificities that are difficult to achieve using conventional discovery
`methods.
`
`4. ANTIBODY CONFORMATIONAL (FOLDING) STABILITY
`
`Another key attribute of antibodies is their folding stability (30, 31), which is critical for
`maintaining long-term activity. Folding stability is a particularly important consideration
`when altering the CDRs of antibodies to improve binding affinity and specificity. Several
`studies have demonstrated that either grafting CDRs from one antibody to another or
`mutating CDRs can significantly affect folding stability (32–36). More generally, mutations
`that are located outside of the CDRs can also significantly impact folding stability (37–40).
`These challenges necessitate design methods that can rationally stabilize antibodies without
`compromising binding affinity and other key attributes.
`
`The high degree of sequence similarity between different antibodies, as well as the large
`number of available antibody sequences and structures, has led to significant understanding
`in how to stabilize antibodies. These approaches can be classified as (a) knowledge-based,
`(b) statistical, and (c) structure-based methods. Knowledge-based methods are those that
`rely on previous experimental studies in which stabilizing mutations or scaffolds have been
`identified (31, 36, 37, 41). Importantly, this general approach does not depend on natural
`antibody sequences and also includes mutations that are rare or absent in conventional
`antibody repertoires. Statistical methods use consensus approaches to identify stabilizing
`mutations based on the assumption that the most common antibody sequences are optimal
`(31, 39, 42). This approach can be used in powerful ways to evaluate not only the consensus
`of individual positions but also pairwise and higher order conservation at noncontiguous
`sites, such as interfaces between antibody domains (43). Structure-based methods are
`computational approaches that use either existing or predicted antibody structures to identify
`stabilizing mutations (24, 27, 44).
`
`The combination of the three approaches is especially powerful, which was demonstrated in
`a study of an unstable single-chain variable fragment (scFv) (44). The investigators sought to
`stabilize this scFv (initial melting temperature of 51°C) to the point that it could be fused to
`an IgG to generate a bispecific antibody with both high activity and stability. They used (a)
`knowledge-based approaches (31); (b) statistical methods, such as covariation and frequency
`
`Annu Rev Biomed Eng. Author manuscript; available in PMC 2017 February 02.
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Lassen - Exhibit 1053, p. 7
`
`

`

`Tiller and Tessier
`
`Page 8
`
`analysis (37, 43); and (c) structure-based methods, such as Rosetta (45) and molecular
`simulations (46), to predict positions within the scFv that are most important to stability.
`This led to the identification of 18 stabilizing mutations at 10 different positions (44). These
`included single mutations that increased the melting temperature significantly (67°C for a
`variant with P101D in VH) as well as combinations of mutations that were even more
`stabilizing (melting temperature of 82°C for a variant with S16E, V55G, and P101D in VH,
`and S46L in VL).
`
`Four of these stabilizing mutations are featured in Figure 4. Two are located at the VH–VL
`interface (VH P101D and VL S46L), which has been shown to be critical in determining the
`overall stability of Fv and scFv antibody fragments (37, 40, 43, 47, 48). The negatively
`charged mutation (VH P101D) forms a salt bridge with a neighboring residue (VH R98), and
`the hydrophobic mutation (VL S46L) increases van der Waals contacts and induces favorable
`structural changes in neighboring residues. The other two stabilizing mutations in the VH
`domain are located far from the VH–VL interface. The VH S16E mutation appears to be
`stabilizing due to electrostatic interactions (it is located in a positively charged region) as
`well as favorable van der Waals interactions involving the aliphatic side chain (which is
`longer for glutamic acid than serine). The V55G mutation is located in a turn near HCDR2
`and appears to stabilize the scFv by eliminating strain caused by the unfavorable φ and ψ
`angles for valine at this position.
`
`Many other impactful studies have also demonstrated rational approaches for stabilizing
`antibodies. One approach is to introduce additional intramolecular disulfide bonds within
`single-domain antibodies as well as interdomain disulfide bonds within Fvs and scFvs to
`increase folding stability (49–52). These methods have yielded significant improvements in
`stability, although in some cases they have resulted in reduced expression. In addition, much
`effort has focused on optimizing the VH–VL interfaces using noncysteine mutations (40, 43,
`48). This is important to improve both thermodynamic and kinetic folding stability
`(especially for Fvs and scFvs) (30, 47), and to reduce the complexity of the resulting
`antibodies by avoiding additional disulfide bonds. These and related studies (38, 41, 53, 54)
`are improving the systematic and robust optimization of antibody conformational stability.
`
`5. ANTIBODY COLLOIDAL STABILITY (SOLUBILITY)
`
`The colloidal stability of antibodies—which is governed by solvent-exposed residues in their
`native folded structure—is not as well understood as conformational stability. Nevertheless,
`colloidal stability is also a critical attribute of antibodies, especially for those antibodies with
`high conformational stability (as observed for many IgGs) (4, 5, 55). There are three key
`elements of antibodies that impact their solubility, namely the (a) CDRs, (b) frameworks of
`the variable and constant domains, and (c) glycans. CDRs commonly contain hydrophobic
`and charged residues to mediate high-affinity binding, yet these same residues can also
`mediate antibody self-association and aggregation (35, 56–60). Therefore, mutations in the
`CDRs can significantly impact antibody solubility (24, 35, 56–59). The frameworks of
`antibodies are also important determinants of their solubility (24, 56, 58, 61, 62). These
`regions typically contain hydrophobic patches (e.g., Fc receptor-binding sites) and
`oppositely charged domains that can interact with themselves or with the CDRs, leading to
`
`Annu Rev Biomed Eng. Author manuscript; available in PMC 2017 February 02.
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Author Manuscript
`
`Lassen - Exhibit 1053, p. 8
`
`

`

`Tiller and Tessier
`
`Page 9
`
`poor solubility. Glycans also significantly impact solubility, typically in a positive manner
`(24, 63, 64).
`
`An excellent study of the impact of each of these factors on antibody solubility is
`summarized in Figure 5 (61). The investigators sought to improve the poor solubility of an
`antibody specific for the glycoprotein LINGO-1. Because this antibody has excellent
`(subnanomolar) binding affinity and high bioactivity, multiple strategies were pursued to
`increase its solubility without reducing binding activity. The first approach was to switch the
`antibody framework from an IgG1 framework to IgG2 and IgG4 frameworks. Surprisingly,
`this simple change resulted in dramatic increases in solubility for both frameworks (increase
`from <1 to >30 mg/mL). The origin of these improvements is not clear given that the
`isoelectric points of the three variants are all high (>pH 8.2), and the IgG2 and IgG4 variants
`display folding stabilities that are similar to or lower than that of the parent IgG1 antibody.
`
`Given the preference of specific IgG subtypes for different therapeutic applications, the
`investigators also evaluated the impact of CDR mutations and glycans on the solubility of
`the parent IgG1 (anti-LINGO-1) antibody (Figure 5). Hydrophobic residues in HCDR2
`(isoleucine 57), HCDR3 (tryptophan 104), and LCDR3 (tryptophan 94) were mutated to be
`less hydrophobic or polar residues. Multiple mutations at these sites improved solubility by
`more than an order of magnitude without reducing binding affinity. Glycosylation was also
`found to significantly impact the solubility of the wild-type antibody. For example,
`removing the glycans from the IgG2 and IgG4 variants reduced the solubility to levels
`similar to those of the parent IgG1 antibody (either with or without glycans), suggesting that
`the glycans of the non-IgG1 variants are critical to their superior solubility. The authors also
`introduced glycosylation sites within the CH1 domain of the IgG1 antibody at four different
`positions. Interestingly, the solubilizing activity of the glycosylation variants was high for
`two mutants (>50 mg/mL) and modest for the others (3–5 mg/mL), and the differences were
`not predictable based on the proximity of the glycosylation sites to the variable domains.
`
`Several additional studies have used related approaches to improve antibody solubility. For
`example, multiple studies have reported how the sequences for CDRs, both of antibody
`fragments and full-length antibodies, can be engineered to increase solubility without
`compromising binding activity (35, 46, 56–59). These studies have also revealed that the
`location of aggregation hot spots within the CDRs is variable, and identifying effective sites
`for mutation requires the use of systematic design approaches (46, 57, 59, 65–67).
`Significant progress has also been made in engineering the fram