CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`Engineering Aggregation-
`Resistant Antibodies
`Joseph M. Perchiacca and Peter M. Tessier
`Center for Biotechnology and Interdisciplinary Studies, Department of Chemical and Biological
`Engineering, Rensselaer Polytechnic Institute, Troy, New York 12180; email: tessier@rpi.edu
`
`Annu. Rev. Chem. Biomol. Eng. 2012. 3:263–86
`
`First published online as a Review in Advance on
`March 29, 2012
`
`The Annual Review of Chemical and Biomolecular
`Engineering is online at chembioeng.annualreviews.org
`
`This article’s doi:
`10.1146/annurev-chembioeng-062011-081052
`Copyright c(cid:2) 2012 by Annual Reviews.
`All rights reserved
`
`1947-5438/12/0715-0263$20.00
`
`Keywords
`IgG, Fab, scFv, single-chain variable fragment, VH,
`complementarity-determining region, CDR, self-association, solubility
`
`Abstract
`The ability of antibodies to bind to target molecules with high affinity and
`specificity has led to their widespread use in diagnostic and therapeutic ap-
`plications. Nevertheless, a limitation of antibodies is their propensity to self-
`associate and aggregate at high concentrations and elevated temperatures.
`The large size and multidomain architecture of full-length monoclonal an-
`tibodies have frustrated systematic analysis of how antibody sequence and
`structure regulate antibody solubility. In contrast, analysis of single and
`multidomain antibody fragments that retain the binding activity of mono-
`clonal antibodies has provided valuable insights into the determinants of
`antibody aggregation. Here we review advances in engineering antibody
`frameworks, domain interfaces, and antigen-binding loops to prevent ag-
`gregation of natively and nonnatively folded antibody fragments. We also
`highlight advances and unmet challenges in developing robust strategies for
`engineering large, multidomain antibodies to resist aggregation.
`
`263
`
`Annu. Rev. Chem. Biomol. Eng. 2012.3:263-286. Downloaded from www.annualreviews.org
`
` Access provided by University of Minnesota - Twin Cities on 03/06/15. For personal use only.
`
`Ex. 1046 - Page 1 of 27
`
`AMGEN INC.
`Exhibit 1046
`
`Further
`ANNUAL
`REVIEWS
`Click here for quick links to
`Annual Reviews content online,
`including:
`• Other articles in this volume
`• Top cited articles
`• Top downloaded articles
`• Our comprehensive search
`
`

`

`CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`INTRODUCTION
`Antibodies are multidomain proteins used by the immune system to recognize and neutralize
`foreign antigens with remarkable specificity. This specificity has been exploited for myriad diag-
`nostic applications in vitro, including immunofluorescence, western blotting, and enzyme-linked
`immunosorbent assay (ELISA) analysis. Antibodies have also attracted intense interest as thera-
`peutic molecules, as evidenced by the large number of antibodies either approved or in clinical
`trials for treating human disorders ranging from cancer and rheumatoid arthritis to osteoporosis
`and asthma (1–3). Antibodies are attractive therapeutic molecules owing not only to their speci-
`ficity but also to their long half-life in vivo (typically two to four weeks) (4) as well as their expected
`lack of immunogenicity for fully humanized antibodies (5, 6). From a discovery point of view, an-
`tibodies are also attractive because well-established in vivo (immunization) and in vitro (phage
`display) methods exist for identifying and maturing high-affinity antibody variants against diverse
`antigens (7–14).
`Although several classes of human antibodies (e.g., IgG, IgA, and IgM) exist, all of them possess
`the same basic architecture. Full-length antibodies [herein referred to as monoclonal antibodies
`(mAbs)] are composed of four polypeptide chains—two light chains (210–220 amino acids per
`chain) and two heavy chains (450–550 amino acids per chain)—that are linked together via disul-
`fide bonds to form a Y shape (Figure 1). A typical antibody contains 12–14 folded domains,
`and each domain possesses a similar Greek key fold in which two β-sheets form a sandwich
`(Figure 1) (15). The variable domains of the heavy (VH) and light (VL) chains contain
`three antigen-binding loops each (5–20 amino acids per loop), which are also known as the
`complementarity-determining regions (CDRs; Figure 1). Each antibody arm containing VH and
`VL domains also contains two constant domains (CL in the light chain and CH1 in the heavy chain).
`Collectively, these four domains are referred to as the antigen-binding fragment (Fab; Figure 1).
`In addition, the base of the antibody is referred to as the crystallizable or Fc domain. The Fc
`domain is composed of the C-terminal domains of the two heavy chains (Figure 1), and each
`chain contains a conserved N-glycosylation site. Glycosylated Fc domains activate the immune
`system upon antibody binding, which is critical to the activity of some therapeutic antibodies
`(16).
`Because the binding activity of antibodies is localized to their variable regions, small frag-
`ments of antibodies can be generated that also retain the binding activity of their parent anti-
`bodies (17–24). Indeed, even single variable domains (VL or VH; Figure 1)—known as domain
`antibodies or nanobodies—can be engineered to bind to diverse targets with high affinity (17,
`25–27). The smallest antibody fragments that contain both VH and VL domains are Fv fragments
`(composed of two polypeptide chains) and single-chain variable fragments (scFvs) in which the
`VH and VL domains are connected via a flexible peptide linker (Figure 1). ScFvs are generally
`preferred to Fv fragments because the amino acid linker connecting the VH and VL domains
`limits the dissociation of the two variable domains (23, 24). However, Fabs are the most widely
`used antibody fragments because they contain both variable domains and stabilizing constant
`domains.
`Antibody fragments and mAbs each have unique advantages and disadvantages for diverse
`applications (see References 28–30 and references therein). However, both types of antibod-
`ies are susceptible to aggregation upon exposure to a variety of stresses (31, 32), including
`high antibody concentrations necessary for subcutaneous therapeutic delivery (50–200 mg ml−1)
`(33, 34), elevated temperatures (34–36), freeze-thaw cycles (37, 38), agitation (39–42), low pH (43,
`44), and long storage times (>2 years for therapeutic applications) (45, 46). Antibody aggregation
`is of particular concern for therapeutic applications because such aggregates can be immunogenic
`
`Perchiacca· Tessier
`
`Antigen: any
`molecule that is
`recognized specifically
`by an antibody
`mAb: monoclonal
`antibody
`VH: variable domain
`of the heavy chain
`VL: variable domain
`of the light chain
`Complementarity-
`determining regions
`(CDRs): peptide
`loops on the surface of
`VH and VL domains
`that mediate antigen
`recognition
`CL: constant domain
`of the light chain
`CH1: first constant
`domain of the heavy
`chain adjacent to the
`VH domain
`Antigen-binding
`fragment (Fab):
`antibody fragment
`containing two
`complementary
`variable (VH/VL) and
`constant (CH1/CL)
`domains
`Fc: crystallizable
`domain of the heavy
`chain containing the
`CH2 and CH3 domains
`Fv: variable domain
`containing VH and VL
`Single-chain variable
`fragment (scFv):
`antibody fragment
`containing a VH and a
`VL domain connected
`via a peptide linker
`
`264
`
`Annu. Rev. Chem. Biomol. Eng. 2012.3:263-286. Downloaded from www.annualreviews.org
`
` Access provided by University of Minnesota - Twin Cities on 03/06/15. For personal use only.
`
`Ex. 1046 - Page 2 of 27
`
`

`

`CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`Heavy
`chain CDRs
`
`Light
`chain CDRs
`
`VH
`
`CH1
`
`VL
`
`CL
`
`CH2
`
`CH3
`
`VH
`
`VL
`
`scFv
`
`Fab
`
`Figure 1
`Molecular architecture of monoclonal antibodies (mAbs) and antibody fragments. A typical mAb is composed
`of two heavy chains and two light chains that contain a total of twelve individual domains. The variable
`heavy (VH) and light (VL) domains each display three peptide loops—referred to as complementarity-
`determining regions (CDRs)—that contact antigens and mediate binding specificity. Because antibody-
`binding activity is localized to the variable domains, smaller antibody fragments containing one or more
`variable domains retain binding activity without the constant domains. The crystal structure of the variable
`domains is Protein Data Bank Number 1N8Z. Abbreviations: CH, heavy chain constant domain; CL, light
`chain constant domain; Fab, antigen-binding fragment; scFv, single-chain variable fragment.
`
`(47, 48). Nevertheless, the widespread use of antibodies and the ability to fully humanize them
`(49, 50) have motivated investigators to understand how to engineer antibodies to resist aggre-
`gation (51, 52). Here we review the most important studies aimed at elucidating how to stabilize
`antibodies against aggregation.
`
`www.annualreviews.org • Aggregation-Resistant Antibodies
`
`265
`
`Annu. Rev. Chem. Biomol. Eng. 2012.3:263-286. Downloaded from www.annualreviews.org
`
` Access provided by University of Minnesota - Twin Cities on 03/06/15. For personal use only.
`
`Ex. 1046 - Page 3 of 27
`
`

`

`CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`High nonnative
`colloidal stability
`(reversible unfolding)
`
`Stress to antibody
`(high concentration,
`elevated temperature,
`long storage time)
`
`Low
`colloidal
`stability
`
`Low domain
`interface
`stability
`
`Partially or fully
`unfolded antibodies
`
`Low kinetic or
`thermodynamic
`folding stability
`
`Low nonnative
`colloidal stability
`(irreversible unfolding)
`
`Nonnative
`antibody aggregates
`
`Native antibody
`aggregates
`
`Domain-swapped
`antibody aggregates
`
`Figure 2
`Antibody aggregation pathways. Antibodies exposed to a variety of stresses are susceptible to aggregation
`through three primary pathways. Stresses such as elevated temperature can lead to unfolding of one or more
`antibody domains because of low thermodynamic and/or kinetic folding stability. This unfolding of
`antibodies can lead to aggregation if their unfolded conformations are competent for aggregation (low
`nonnative colloidal stability). However, unfolded antibodies can also refold without aggregating (high
`nonnative colloidal stability). Stresses such as high concentration or low temperature can lead to antibody
`aggregation without unfolding for antibodies with low native colloidal stability (i.e., low native solubility).
`Finally, antibodies can also aggregate via domain swapping without unfolding if the interfaces between
`complementary antibody domains are unstable.
`
`ANTIBODY AGGREGATION PATHWAYS
`Antibodies can aggregate through multiple pathways owing to physical and chemical instabilities
`(53, 54). For the purposes of this review, we consider only aggregation pathways due to physical
`instabilities of antibodies (Figure 2). Moreover, we refer to antibody aggregation as condensation
`of folded or unfolded antibodies into reversible or irreversible antibody aggregates. Stresses such
`as elevated temperature or low pH can cause antibodies to partially or fully unfold if antibodies
`possess low thermodynamic or kinetic folding stability. In the case of multidomain antibodies, one
`or more domains may unfold without unfolding of the other domains. The propensity of partially or
`fully unfolded antibodies to aggregate is determined by a competition between refolding (governed
`by intramolecular interactions) and aggregation (governed by intermolecular interactions). We
`refer to the propensity of antibodies to aggregate when unfolded as their nonnative colloidal
`stability. If intermolecular interactions between unfolded antibodies are sufficiently attractive
`
`Perchiacca· Tessier
`
`266
`
`Annu. Rev. Chem. Biomol. Eng. 2012.3:263-286. Downloaded from www.annualreviews.org
`
` Access provided by University of Minnesota - Twin Cities on 03/06/15. For personal use only.
`
`Ex. 1046 - Page 4 of 27
`
`

`

`CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`(low colloidal stability), then unfolded antibodies condense into nonnative aggregates. Conversely,
`if intermolecular interactions between unfolded antibodies are repulsive or insufficiently attractive
`(high colloidal stability), then unfolded antibodies fold reversibly without aggregating (assuming
`that the refolding kinetics are not limiting).
`Antibodies can also aggregate via mechanisms that do not require unfolding. Elevated anti-
`body concentrations, low temperatures, and related stresses can cause antibodies with low na-
`tive colloidal stability to condense into native protein aggregates owing to attractive antibody
`self-interactions. Multidomain antibodies can domain swap with complementary domains from
`other identical molecules, leading to aggregation in which each individual domain is folded.
`In some cases, antibody aggregation occurs as a result of multiple physical instabilities that in-
`volve more than one aggregation mechanism described above. In this review, we highlight ad-
`vances in engineering antibodies for maximal stability that prevents aggregation through each
`pathway.
`
`Antibody Thermodynamic Folding Stability
`The thermodynamic folding stability of antibodies is one of the most fundamental and widely
`studied physical properties that govern the propensity of antibodies to aggregate. Because the
`immunoglobulin (Ig) fold is conserved across diverse antibodies, much work has focused on iden-
`tifying and/or engineering antibody scaffolds with high thermodynamic stability. This is partic-
`ularly important for antibody fragments because they often possess lower folding stabilities than
`their full-length counterparts. The folding stability of antibodies is typically measured using cir-
`cular dichroism spectroscopy and/or tryptophan fluorescence (55). Below we discuss important
`advances in engineering single- and multidomain antibody fragments with high conformational
`(folding) stability.
`
`Folding stability of single-domain antibodies. It would be logical to assume that single-domain
`antibodies (e.g., VH) cannot be engineered to be as stable as larger antibody fragments (e.g., Fabs)
`and mAbs (e.g., IgG) because these small antibody domains lack complementary variable (e.g.,
`VL) and constant (e.g., CH1/CL) domains that stabilize their folded structure. However, several
`studies have convincingly demonstrated that multidomain architecture is unnecessary for indi-
`vidual antibody domains to possess high conformational stability (56–61). One fruitful approach
`for stabilizing individual VH or VL domains has been to identify mutations in the former VH/VL
`interface that increase folding stability to compensate for the loss of the stabilizing interactions
`between variable domains (19, 60–62). Sidhu and coworkers (60) elegantly demonstrated that such
`stabilizing mutations could be readily identified for a human VH variant. They randomized ap-
`proximately 20 residues at the former VH/VL interface of a VH domain, which included residues
`within β-strands, non-CDR loops, and CDR3. Interestingly, they identified four mutations in two
`β-strands, as shown in Figure 3, that significantly increase the folding stability of the wild-type
`VH domain. This increased stability was observed both in terms of an increase in the midpoint
`temperature of antibody unfolding (also known as the apparent melting temperature) from 58 to
`79◦C (60) as well as an increase in the Gibbs free energy of unfolding ( GN-U) from 28 kJ mol−1
`to 52 kJ mol−1 ( J.M. Perchiacca & P.M. Tessier, unpublished results). Notably, the four muta-
`tions are localized to two pairs of interacting residues. Two mutations in close proximity near the
`base of the stabilized antibody domain are oppositely charged (Arg39 and Glu45) and presumably
`stabilize the folded structure via complementary electrostatic interactions. The other stabilizing
`mutations (Gly35 and Ser50) are localized at the edges of CDR1 and CDR2. These mutations lead
`to significant changes in the orientation of adjacent aromatic residues that may explain the increase
`
`www.annualreviews.org • Aggregation-Resistant Antibodies
`
`267
`
`Annu. Rev. Chem. Biomol. Eng. 2012.3:263-286. Downloaded from www.annualreviews.org
`
` Access provided by University of Minnesota - Twin Cities on 03/06/15. For personal use only.
`
`Ex. 1046 - Page 5 of 27
`
`

`

`CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`4D5
`
`Leu45Leu45
`
`Leu45
`
`His35His35
`His35
`
`Arg50
`
`Arg50Arg50
`
`4D5
`B1a
`
`40
`
`60
`T (ºC)
`
`B1a
`
`80
`
`100
`
`Gly35
`Gly35Gly35
`
`Ser50
`
`Ser50Ser50
`
`1.0
`
`0.8
`
`0.6
`
`0.4
`
`0.2
`
`Fraction folded
`
`0.0
`20
`
`Gln39Gln39
`
`Gln39
`
`Arg39Arg39
`
`Arg39
`
`
`Glu45Glu45
`Glu45
`
`Figure 3
`Mutations that enhance the thermodynamic folding stability of a human single-domain antibody. An
`unstable human VH domain (4D5; PDB: 1FVC) was converted into a highly stable variant (B1a; PDB:
`3B9V) by introducing four mutations within or near the former VH/VL interface (60). Abbreviations:
`VH, variable heavy domain; VL, variable light domain.
`
`in folding stability (60). Notably, these mutations obtained from synthetic antibody libraries are
`uncommon to natural human VH domains, which demonstrates that highly conserved antibody
`sequences can be further optimized.
`The enhanced folding stability of single-domain antibodies due to mutations at the edges of
`their CDR loops suggests that the sequence of CDRs may have a greater impact on antibody
`folding stability than previously realized (63, 64). Although small loops (<5 residues) on the
`surface of proteins generally have minimal impact on protein folding stability, the relatively large
`CDR loops (5–20 residues) on the surface of variable antibody domains can either destabilize
`or stabilize the antibody fold. For example, the close proximity between CDR loops suggests
`that direct interactions between CDR loops could be either stabilizing or destabilizing. Indeed,
`
`Perchiacca· Tessier
`
`268
`
`Annu. Rev. Chem. Biomol. Eng. 2012.3:263-286. Downloaded from www.annualreviews.org
`
` Access provided by University of Minnesota - Twin Cities on 03/06/15. For personal use only.
`
`Ex. 1046 - Page 6 of 27
`
`

`

`CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`CDR3
`
`Former
`light chain
`interface
`
`Figure 4
`Camelid variable heavy domain (VHH) antibodies are stabilized via interactions between complementarity-
`determining region 3 (CDR3) and the former variable heavy domain/light domain (VH/VL) interface. The
`crystal structure of a camelid VHH domain specific for lysozyme (PDB: 1MEL) reveals that part of CDR3
`packs against the former VH/VL interface, thereby stabilizing the folded structure (66).
`
`multiple studies have revealed that grafting a subset of the three CDR loops from one variable-
`domain antibody onto a second variable domain generally results in a lower stability for the grafted
`variants than that for the parent antibody domains (58, 63, 65). In contrast, grafting all three
`CDR loops from unstable VH scaffolds onto highly stable VH scaffolds often produces antibody
`domains with high folding stability (58, 59). These findings confirm that the ability of CDR loops
`to assume complementary conformations on the surface of antibodies is critical to stabilizing the
`fold of single-domain antibodies.
`CDR loops can also stabilize antibodies by packing against the antibody scaffold (56, 66, 67).
`The most striking example of this behavior is observed for single-domain antibodies from camelids
`(e.g., camels and llamas). Camelid antibodies are typically composed of two identical polypeptide
`chains that are analogous to the heavy chains of human antibodies, yet they recognize antigens
`with similar affinity as human antibodies with both heavy and light chains (25, 68, 69). Camelid
`VHH domains have unusually large CDR3 loops (8–24 residues, with an average of 16 residues) that
`appear necessary to confer high affinity without the assistance of the CDRs from VL domains (70,
`71). Importantly, part of CDR3 of camelid VHH domains packs against hydrophobic residues at
`the former VH/VL interface (Figure 4) (66, 67). These stabilizing interactions (which are typically
`absent in human antibody domains) enable camelid VHH domains to display long CDR3 loops
`that would be destabilizing to their human counterparts (56).
`
`Folding stability of multidomain antibody fragments. Engineering multidomain antibod-
`ies to possess high folding stability is more complex than for single domains owing to multiple
`complicating factors. First, the individual domains of a multidomain antibody often do not unfold
`cooperatively. Because unfolding of even one domain of a multidomain antibody can lead to aggre-
`gation, it is critical to engineer each antibody domain to resist unfolding. A second complication is
`that the folding stability of multidomain antibodies is determined not only by the intrinsic stability
`of the individual antibody domains (e.g., VL), but also via the interaction between complementary
`domains (e.g., VH/VL).
`
`VHH: variable domain
`of camelid antibodies
`analogous to the heavy
`chain VH domain in
`human antibodies
`
`www.annualreviews.org • Aggregation-Resistant Antibodies
`
`269
`
`Annu. Rev. Chem. Biomol. Eng. 2012.3:263-286. Downloaded from www.annualreviews.org
`
` Access provided by University of Minnesota - Twin Cities on 03/06/15. For personal use only.
`
`Ex. 1046 - Page 7 of 27
`
`

`

`CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`VH
`
`VL
`
`scFv
`
`Fab
`
`1
`
`0
`
`1
`
`0
`
`1
`
`0
`
`1
`
`0
`
`Fraction unfolded
`
`0
`
`1
`
`3
`2
`[GuHCl] (M)
`
`4
`
`5
`
`Figure 5
`Interactions between complementary domains stabilize multidomain antibodies. The poor stability of a
`variable light (VL) domain was enhanced via interactions between complementary domains within a
`single-chain variable fragment (scFv) and an antigen-binding fragment (Fab) (73). Abbreviations: GuHCl,
`guanidine hydrochloride; VH, variable heavy domain.
`
`To understand the origins of folding stability for multidomain antibody fragments, several
`studies have compared the stability of paired variable domains (i.e., VH/VL) relative to the indi-
`vidual domains (72–74). For example, Pl ¨uckthun and coworkers compared the stability of human
`VH and VL domains when isolated and when paired together in an scFv (Figure 5) (73). In this case,
`the VH domain is significantly more stable than the VL domain. When the two domains are paired
`together, the scFv antibody displays two unfolding transitions owing to the noncooperative un-
`folding of each variable domain. Importantly, the unfolding of the less stable variable domain (VL)
`in the scFv occurs at higher denaturant concentrations than that of the isolated VL domain owing
`
`Perchiacca· Tessier
`
`270
`
`Annu. Rev. Chem. Biomol. Eng. 2012.3:263-286. Downloaded from www.annualreviews.org
`
` Access provided by University of Minnesota - Twin Cities on 03/06/15. For personal use only.
`
`Ex. 1046 - Page 8 of 27
`
`

`

`CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`Bispecific antibody:
`an antibody, typically
`composed of two
`antibodies (e.g.,
`IgG-scFv), that binds
`to two different
`antigens
`
`to the stabilizing interactions of the VL/VH interface (Figure 5). Moreover, addition of CH1 and
`CL further increases the antibody stability by shifting the unfolding transition of the VL domain
`to even higher denaturant concentrations. The additional stability afforded by the paired constant
`domains is strongly dependent on the presence of a disulfide bond linking CH1 and CL (73).
`These findings that interactions between complementary variable (VH/VL) and constant
`(CH1/CL) domains enhance antibody folding stability have inspired the design of highly sta-
`ble multidomain antibodies (see References 51, 52, 75 and references therein). One strategy to
`engineer Fvs and scFvs with enhanced folding stability is to introduce a disulfide bond between the
`VH and VL domains that is similar to the stabilizing disulfide bond linking the constant (CL/CH1)
`domains. Indeed, this strategy provides significant stability to both Fvs (76, 77) and scFvs (78, 79),
`and also has been used to stabilize scFvs fused to mAbs (also known as bispecific antibodies) (80).
`Drawbacks of this strategy are that some antibody variants do not readily form disulfide bonds at
`the VH/VL interface (81), as well as that the expression levels of scFvs bearing additional disulfide
`bonds are typically much lower than those of the original scFvs (78, 79, 81). A complementary ap-
`proach for stabilizing multidomain antibodies is to compare their sequences with those of closely
`related antibodies to identify mutations that may enhance folding stability. The hypothesis behind
`this approach is that replacing nonconserved residues with conserved ones at positions where anti-
`bodies differ from consensus sequences will lead to increased folding stability. Several studies have
`demonstrated that introduction of conserved residues into the interfaces between domains and
`other regions within multidomain antibodies often stabilizes such antibodies against unfolding
`(75, 81–88).
`Notably, the sequence of CDR loops can significantly impact the folding stability of multido-
`main antibody fragments (85, 89), just as observed for single-domain antibodies (58, 63, 65). Two
`scFvs that differ only in the sequence of CDR3 in the VL domain provide a striking example of
`the sensitivity of antibody folding stability to the sequence of CDRs (89). Although the stabil-
`ities of the VL domains bearing either CDR3 were similar, the stabilities of the corresponding
`scFvs were strongly dependent on the CDR3 sequence. As expected, one of the scFv variants
`displayed increased folding stability relative to the individual domains owing to stabilizing VH/VL
`interactions. However, the other scFv variant was destabilized relative to the individual domains.
`Inspection of the destabilizing CDR3 sequence revealed two consecutive proline residues that may
`impose an unfavorable CDR3 conformation and impede proper pairing of VH and VL domains
`(89). These findings are also consistent with related observations that grafting CDRs onto scFv
`scaffolds produces antibody variants whose stability is dependent on the sequence of the CDR
`loops (85).
`
`Antibody Kinetic Stability
`The rate at which antibodies unfold is also a key determinant of their aggregation propensity.
`Here we refer to high kinetic stability as the ability of antibodies to unfold slowly, thereby re-
`sisting aggregation by slowly populating unfolded states competent for nonnative aggregation
`(Figure 2). It has been recognized for decades that mAbs possess high kinetic stability and un-
`fold extremely slowly (require months to unfold in denaturant; see Reference 51 and references
`therein). In fact, the high kinetic stability of mAbs is likely one of the primary reasons for their
`successful use in diverse applications that require long-term stability.
`Despite the importance of the kinetic stability of antibodies, the molecular origins of such sta-
`bility are less well understood than those of thermodynamic folding stability. Nevertheless, several
`key studies have begun to define the determinants of kinetic stability for antibody fragments and
`mAbs (72, 73, 90, 91). An important observation is that the individual domains of antibodies
`
`www.annualreviews.org • Aggregation-Resistant Antibodies
`
`271
`
`Annu. Rev. Chem. Biomol. Eng. 2012.3:263-286. Downloaded from www.annualreviews.org
`
` Access provided by University of Minnesota - Twin Cities on 03/06/15. For personal use only.
`
`Ex. 1046 - Page 9 of 27
`
`

`

`CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`Fab
`
`Fab
`
`VL
`
`CL
`
`VL
`
`CL
`
`VH
`
`CH1
`
`VH
`
`CH1
`
`Figure 6
`Structural hypotheses for the high kinetic stability of antigen-binding fragments (Fabs). The slow unfolding
`of Fabs is due to interdomain interactions between two complementary constant heavy and light domains
`(CH1/CL) (73, 91). Because paired variable heavy and light domains (VH/VL) unfold much more rapidly
`than the corresponding constant domains, the larger hydrophobic interface and/or unique interdomain
`orientation of β-strands encode the high kinetic stability of Fabs (91). The antibody crystal structure is for a
`Fab against HER2 (PDB: 1N8Z).
`
`(e.g., VL, CL) unfold rapidly in denaturant (minutes or less) (72, 73), which confirms that the
`origin of the kinetic stability of large, multidomain antibodies is not due to high kinetic stability of
`their individual domains. Moreover, two antibody domains within a single polypeptide chain that
`are noncomplementary—such as VL and CL—also unfold rapidly (73, 90). However, multidomain
`antibody fragments containing complementary domains (e.g., VH/VL in a Fab) unfold much more
`slowly than their individual domains (72, 73, 91). These findings suggest that stabilizing interac-
`tions between complementary antibody domains are linked to the kinetic stability of multidomain
`antibodies.
`To further elucidate the contribution of interactions between complementary domains to
`the kinetic stability of multidomain antibody fragments, Pl ¨uckthun and coworkers (73) elegantly
`dissected the origins of the kinetic stability of a Fab. They first confirmed that the individual
`antibody domains (e.g., VL and CL) unfold rapidly in denaturant, consistent with previous findings
`(72). Moreover, they found that the VH and VL domains linked together in an scFv also unfold
`rapidly, revealing that scFvs are not necessarily kinetically stable (51, 73). Most importantly,
`the paired constant domains (CH1/CL) without the variable domains required weeks to unfold
`in denaturant (73). This unfolding rate of the paired constant domains (but not the individual
`constant domains) was similar to that of the entire Fab, revealing that the interface between the
`constant domains is the most important determinant of the kinetic stability of Fabs. The origin of
`the kinetic stability afforded by the constant domains remains unknown. However, the interface
`between constant domains is larger, more hydrophobic, and oriented differently relative to the
`interface between variable domains (Figure 6). The unique orientation of β-strands at the CH1/CL
`interface may block the unfolding of each domain, leading to the high kinetic stability of these
`multidomain antibody fragments (73).
`
`Perchiacca· Tessier
`
`272
`
`Annu. Rev. Chem. Biomol. Eng. 2012.3:263-286. Downloaded from www.annualreviews.org
`
` Access provided by University of Minnesota - Twin Cities on 03/06/15. For personal use only.
`
`Ex. 1046 - Page 10 of 27
`
`

`

`CH03CH12-Tessier
`
`ARI
`
`14 May 2012
`
`15:15
`
`Antibody Nonnative Colloidal Stability
`Most antibodies aggregate when unfolded due to attractive intermolecular interactions between
`solvent-exposed hydrophobic residues that are normally solvent shielded within the folded anti-
`body core. Nevertheless, antibodies that are unfolded may refold instead of aggregating (referred
`to as reversible unfolding) if the intermolecular interactions between unfolded antibodies are
`insufficiently attractive to mediate aggregation and the refolding kinetics are not limiting. The
`multidomain architecture of most antibody fragments (scFv and Fab) and mAbs greatly reduces the
`likelihood that these antibodies will unfold reversibly without aggregating. However, one would
`expect that single antibody domains (e.g., VH or VL) could be engineered to unfold reversibly
`given that several single-domain, globular proteins display such reversible unfolding behavior
`(92, 93).
`Unfortunately, most variable domains (e.g., VH) from human antibodies are poorly soluble and
`readily aggregate when unfolded (94–96), which had suggested that these domains do not unfold
`reversibly. The discovery of heavy chain antibodies in camels and related species challenged this
`initial conclusion (68). The fact that these antibodies lack light chains suggests that their isolated
`VH domains (typically referred to as VHH) may have superior biophysical properties relative to
`their human counterparts because their folding and stability are independent of complementary
`VL domains. Indeed, several camelid VHH domains fail to aggregate when unfolded at elevated
`temperatures (25, 58, 59, 97) even though their folding stability is similar to that of aggregation-
`prone human VH domains (98). Sequence comparison of human and camelid variable-domain
`antibodies reveals four key amino acid differences (known as the VHH tetrad) at the former VH/VL
`interface (71, 99, 100). Three of the sequence differences increase the hydrophilicity of camelid
`VHH domains by replacing solvent-exposed hydrophobic or nonpolar residues in human VH do-
`mains with charged or less hydrophobic residues in camelid VHH domains. The fourth sequence
`difference involves replacement of a small hydrophobic residue (valine) in human domains with a
`large, aromatic residue (phenylalanine or tyrosine) in camelid domains. This mutation increases
`the hydrophilicity of the former VH/VL interface in camelid antibodies by packing against a
`portion of CDR3, as shown in Figure 4.
`These findings led several investigators to attempt to transfer the desirable properties of camelid
`VHH domains to human VH domains (59, 82, 101). Unf