Review
`
`Toward aggregation-resistant
`antibodies by design
`
`Christine C. Lee, Joseph M. Perchiacca, and Peter M. Tessier
`
`Center for Biotechnology & Interdisciplinary Studies, Department of Chemical & Biological Engineering, Rensselaer Polytechnic
`Institute, Troy, NY 12180, USA
`
`Monoclonal antibodies are attractive therapeutics for
`treating a wide range of human disorders due to their
`exquisite binding specificity and high binding affinity.
`However, a limitation of antibodies is their highly vari-
`able and difficult-to-predict propensities to aggregate
`when concentrated during purification and delivery. De-
`spite the large size and complex structure of antibodies,
`recent findings suggest that antibody solubility can be
`dramatically improved using rational design methods in
`addition to conventional selection methods. Here, we
`review key advances and unmet challenges in engineer-
`ing the variable and constant regions of antibody frag-
`ments and full-length antibodies to resist aggregation
`without reducing their binding affinity. These experi-
`mental and computational discoveries should accelerate
`the development of robust algorithms for designing
`aggregation-resistant antibodies.
`
`Monoclonal antibodies (mAbs): structure, function, and
`stability
`mAbs are large, multidomain proteins used by the immune
`system to recognize and neutralize foreign invaders such
`as bacteria and viruses (see Glossary). Antibodies are
`attractive therapeutic molecules due to their high binding
`affinity, their long circulation times in the blood stream
`(>1 week), the relative ease of identifying them using well-
`established in vitro (phage and related display methods)
`[1–5] and in vivo (immunization) [6,7] discovery methods,
`and the nontoxic nature of their breakdown products
`(amino acids). These unique attributes have led to the
`successful development of therapeutic mAbs for treating
`disorders ranging from arthritis to cancer [8,9].
`Nevertheless, one of the most difficult challenges in
`developing safe and effective therapeutic mAbs is antibody
`aggregation [10–12]. This problem is particularly challeng-
`ing because the preferred delivery of mAbs (which are not
`orally active) is subcutaneous delivery. The large amount
`of mAb that must be typically delivered (>100 mg) in small
`volumes (<2 ml) necessitates concentrated mAb formula-
`tions (>50 mg/ml) that are susceptible to aggregation.
`Aggregation is not only a concern because it inactivates
`
`Corresponding author: Tessier, P.M. (tessier@rpi.edu).
`Keywords: monoclonal antibody; IgG; VH; VL; scFv; Fv; Fab; variable domain;
`complementarity-determining region (CDR); bispecific; solubility; antibody engineer-
`ing.
`
`0167-7799/$ – see front matter
`ß 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.
`tibtech.2013.07.002
`
`612
`
`Trends in Biotechnology, November 2013, Vol. 31, No. 11
`
`the therapeutic activity of mAbs, but such aggregates can
`also be immunogenic [13–15].
`A typical mAb (IgG1) shown in Figure 1 is composed of
`12 domains in four polypeptide chains – two heavy chains
`each containing four domains and two light chains each
`containing two domains – that are connected via multiple
`disulfide bonds. Most sequence differences between anti-
`bodies occur in the variable heavy (VH) and variable light
`(VL) domains. Each variable domain contains three com-
`plementarity-determining regions (CDRs) that mediate
`antibody binding. Antibody fragments such as isolated
`VH and VL domain antibodies (dAbs) or fusions thereof
`(single chain antibody fragments or scFvs) can be generat-
`ed that bind to target proteins with affinities rivaling full-
`length antibodies [16,17]. The heavy and light chains also
`contain constant domains, three in the heavy chain (CH1,
`CH2, and CH3) and one in the light chain (CL). The two
`identical antibody arms of the ‘Y’ structure – which contain
`the VH, CH1, VL and CL domains – are referred to as
`antigen-binding fragments (Fabs). The base of the Y struc-
`ture of antibodies – which contain two CH2 and two CH3
`domains – is referred to as the crystallizable (Fc) domain.
`Importantly, the positions of the Fabs relative to the Fc
`domain are not fixed, which renders the overall antibody
`structure neither a perfect Y shape nor symmetric. The Fc
`domains of antibodies are glycosylated and such oligosac-
`charides are important for mediating effector functions in
`processes such as complement activation and antibody-
`dependent cell-mediated cytotoxicity [18]. Glycosylation
`
`Glossary
`
`in the VH or VL
`
`antigen-binding fragment (Fab): antibody fragment containing the VH, CH1, VL
`and CL domains
`CH1: first constant domain of the heavy chain adjacent to the VH domain
`CL: constant domain of the light chain
`complementarity-determining region (CDR): peptide loop
`domains that is involved in binding to antigens
`dAb: domain antibody fragment such as single VH or VL domains
`gammabody: Grafted AMyloid-Motif AntiBODY that binds to amyloid fibrils
`and related conformers
`Fc: crystallizable domain containing two CH2 and two CH3 domains
`mAb: monoclonal antibody
`single-chain antibody fragment (scFv): antibody fragment composed of VH and
`VL domains connected via a peptide linker
`spatial-aggregation propensity (SAP): a measure of the dynamic exposure of
`hydrophobic patches on protein surfaces
`supercharging: an approach to solubilizing proteins in which many solvent-
`exposed residues are mutated to charged residues of the same polarity
`VH: variable domain of the heavy chain
`VL: variable domain of the light chain
`
`Ex. 2030-0001
`
`

`
`Review
`
`HCDRs
`
`LCDRs
`
`mAb
`
`VH
`
`CH1
`
`CH2
`
`CH3
`
`VL
`
`CL
`
`HCDRs
`
` LCDRs
`
`Fab
`
`scFv
`
`dAb
`
`VH
`
`VL
`
`TRENDS in Biotechnology
`
`Figure 1. Molecular architecture of monoclonal antibodies (mAbs) and antibody
`fragments. The heavy chain complementarity-determining regions (HCDRs) and
`light chain CDRs (LCDRs) are highlighted in the crystal structure 1N8Z.
`
`occurs at a conserved site in the CH2 domain (Asn297) of
`diverse antibodies and at nonconserved sites in the Fabs of
`some antibodies [19–21].
`The large size of mAbs (>1000 residues) and their
`multidomain architecture make it extremely challenging
`to identify the mechanisms governing their aggregation
`using either experimental or computational methods. In
`the past, this complexity typically required investigators
`to screen for rare antibody mutants with improved solu-
`bility instead of using rational design methods. However,
`several studies in recent years using antibody fragments
`and full-length antibodies have identified key sequence
`and structural determinants that differentiate aggrega-
`tion-prone and aggregation-resistant antibodies. These
`findings are accelerating the development of rational
`methods for designing extremely soluble antibodies. Here,
`we review these important recent studies and discuss
`outstanding challenges in rationally engineering antibody
`CDRs, frameworks, and domain interfaces to generate
`antibodies with high solubility without reducing binding
`affinity.
`
`CDRs
`The simplest explanation for the highly variable solubili-
`ties of antibodies is that their most variable regions –
`namely the hypervariable CDRs – are directly involved
`in mediating antibody aggregation. One of the earliest and
`most striking studies of the impact of CDR sequences on
`the solubility of antibodies was conducted using dAbs
`[22,23]. This study showed that the solubility of an aggre-
`gation-prone VH domain (Dp47d) could be improved due to
`sequence variation only within the CDRs. The investiga-
`tors developed a powerful selection strategy in which
`multiple copies of individual dAbs were displayed on the
`surface of phage, and the phage particles were heated to
`induce unfolding and aggregation of poorly soluble dAbs.
`After cooling the phage, rare dAbs that refolded without
`aggregating were selected and characterized.
`
`Trends in Biotechnology November 2013, Vol. 31, No. 11
`
`Although conventional wisdom suggests that protein
`folding stability and solubility are closely related, the more
`soluble dAbs were found to have similar or lower folding
`stabilities than their aggregation-prone counterparts [22].
`Moreover, the aggregation-resistant variants failed to stick
`to size-exclusion columns and unfolded reversibly without
`aggregating, unlike the aggregation-prone dAbs. Inspec-
`tion of the CDR sequences did not reveal obvious explana-
`tions for the large differences in antibody solubility beyond
`a moderate increase in the number of negatively charged
`residues in the CDRs of soluble dAbs. Studies of a large
`number of dAbs selected using the same phage display
`method also found that the CDRs of aggregation-resistant
`dAbs generally have lower b-sheet propensities and are
`less hydrophobic (in addition to being more negatively
`charged) than their poorly soluble counterparts [24].
`One potential computational approach for distinguishing
`between highly and poorly soluble antibodies is to use
`for
`aggregation-scoring algorithms originally developed
`identifying aggregation hotspots within amyloid-forming
`polypeptides (e.g., Ab in Alzheimer’s disease) and related
`proteins [25–29]. These algorithms – which only use the
`amino acid sequence of the target protein – identify small
`peptides predicted to be aggregation-prone based on proper-
`ties such as hydrophobicity, charge and b-sheet propensity.
`These and related aggregation-scoring algorithms have been
`used to identify aggregation hotspots within diverse anti-
`bodies (including mAbs currently in use in the clinic) [30–32].
`An interesting finding from these studies is that aggregation
`hotspots are commonly located within CDRs (as well as the
`Fc domains) of commercial antibodies. These findings further
`suggest that the same peptide loops that mediate antigen
`recognition also mediate antibody aggregation.
`Multiple aggregation-scoring algorithms were used to
`investigate the elusive sequence determinants of an aggre-
`gation-resistant VH domain (Hel4) [22,23] that only differed
`from its aggregation-prone parent (Dp47d) in terms of its
`three CDRs [33]. Interestingly, five aggregation-scoring
`(TANGO
`[25], Waltz
`[29], PASTA
`[27],
`algorithms
`AGGRESCAN [26] and ZipperDB [28]) predicted several
`aggregation hotspots within and near the three CDRs of the
`poorly soluble dAb, but the only consensus prediction over-
`lapped with CDR1 (residues 28–32) [33]. Therefore, the
`investigators posited that CDR1 regulated the poor solubil-
`ity of Dp47d. Indeed, grafting CDR1 from Hel4 into Dp47d
`eliminated aggregation even when the dAb was heated to
`95 8C, whereas grafting CDR2 and CDR3 from Hel4 into
`Dp47d failed to prevent aggregation. Grafting the entire
`CDR1 from Hel4 into Dp47d was unnecessary because
`grafting only three negatively charged residues from Hel4
`CDR1 (31-Asp-Glu-Asp-33) into Dp47d was sufficient to
`prevent dAb aggregation. Thus, aggregation-scoring algo-
`rithms can be used to identify CDRs that mediate antibody
`aggregation. However, further development is needed to
`improve the accuracy of these algorithms because they were
`unable to differentiate between solubilizing and nonsolubi-
`lizing mutations within CDR1 of the VH domain [33].
`Although charged mutations within CDRs can endow
`antibody fragments with extreme solubility, it is possible
`that such mutations will reduce binding affinity. Thus,
`the investigators also asked whether charged mutations
`
`613
`
`Ex. 2030-0002
`
`

`
`Review
`
`Trends in Biotechnology November 2013, Vol. 31, No. 11
`
`The investigators also asked whether it is necessary to
`insert charged residues at both edges of hydrophobic
`CDR3s or if inserting charged residues only at one edge
`of CDR3 could prevent aggregation [34]. Using a panel of
`Ab dAbs with aggregation hotspots positioned at the N
`terminus, middle, and C terminus of CDR3, they found
`(Asp-Glu-Asp)
`that three negatively charged residues
`inserted near the edge of CDR3 closest to the aggregation
`hotspot eliminate aggregation (Figure 2) [34]. In contrast,
`the same insertion mutations at the edge of CDR3 opposite
`to the aggregation hotspot fail to prevent aggregation.
`Interestingly, if the aggregation hotspot is near the center
`of CDR3, then inserting charged residues at either edge of
`CDR3 eliminates aggregation. Importantly, the binding
`affinity of the charged Ab dAbs is indistinguishable from
`their uncharged counterparts, demonstrating that it is
`possible to improve the solubility of antibody fragments
`significantly by inserting charged mutations in the CDRs
`without reducing binding affinity. Nevertheless, additional
`research is needed to evaluate whether this strategy –
`which was developed for dAbs in which one CDR dominates
`both antibody binding and aggregation – can be applied to
`other single- and multidomain antibodies in which multi-
`ple CDRs contribute to binding (and potentially to aggre-
`gation as well).
`The significant contribution of CDRs to antibody solu-
`bility and the large sequence diversity of CDRs within
`natural and synthetic antibody repertoires suggest that
`it is unlikely any single set of charged mutations within the
`CDRs will prevent aggregation of all antibodies. However,
`it is possible that specific CDRs are generally more impor-
`tant in mediating aggregation, and that charged mutations
`introduced into such CDRs will reduce aggregation of most
`antibodies. For example, a recent study demonstrated that
`introducing negatively charged substitution mutations in-
`to CDR1 of a VH dAb (HCDR1) prevented aggregation,
`whereas introducing the same mutations into HCDR2
`failed to prevent aggregation (Figure 3A) [37]. Strikingly,
`this is the opposite for VL dAbs, as charged mutations
`within CDR2 in the light chain (LCDR2) inhibited aggre-
`gation but such mutations within LCDR1 failed to inhibit
`(Figure 3B). Charged mutations within
`aggregation
`HCDR1 and LCDR2 not only solubilize VH and VL dAbs
`with specific CDR sequences, but the same mutations
`solubilize most antibody variants within a synthetic dAb
`repertoire with diverse CDR sequences [37].
`An obvious concern with mutating CDRs to improve
`antibody solubility – especially with substitution muta-
`tions that eliminate CDR residues – is that the binding
`affinity will be reduced. This could not be tested for the VH
`and VL dAbs described above because the dAbs were germ-
`line sequences without known binding partners [37]. Thus,
`the authors investigated whether the charged HCDR1 and
`LCDR2 mutations could increase the solubility of a thera-
`peutic antibody (trastuzumab; Herceptin) without reduc-
`ing binding affinity. An scFv composed of the trastuzumab
`VH and VL domains bound to its antigen (Her2) with
`nanomolar binding affinity. Although some of the solubi-
`lizing CDR mutations dramatically decreased binding
`affinity of the trastuzumab scFv, others had little effect
`on binding. These exciting findings for a multidomain
`
`adjacent to CDR1 within the poorly soluble dAb (Dp47d)
`could be identified that would prevent aggregation as
`potently as those within CDR1 [33]. Strikingly, a novel
`mutation (Phe to Asp at position 29) – which is immedi-
`ately adjacent to CDR1 and not from the highly soluble
`Hel4 dAb – completely inhibited aggregation of Dp47d.
`This is interesting because it raises the attractive possibil-
`ity that charged mutations could be introduced at the edges
`of CDRs containing aggregation hotspots to increase anti-
`body solubility without eliminating residues that contrib-
`ute to antibody binding.
`This possibility was examined using a panel of novel VH
`dAbs that display hydrophobic peptide segments from the
`[34–36]. These
`Alzheimer’s Ab peptide within CDR3
`Grafted AMyloid-Motif AntiBODIES (gammabodies) bind
`to toxic Ab oligomers and fibrils with nanomolar binding
`affinity, whereas they bind weakly to Ab monomers. Their
`hydrophobic CDRs cause Ab dAbs to aggregate within
`minutes when heated above 70 8C and within days at
`25 8C. In contrast, the wild type dAb (which is identical
`except for CDR3) fails to aggregate at either condition.
`Therefore, the investigators evaluated whether inserting
`one (Asp), two (Asp–Asp) or three (Asp-Glu-Asp) negatively
`charged residues at the edges of the hydrophobic CDR3
`loops – without removing any CDR3 residues – prevented
`aggregation while maintaining binding activity [34]. Strik-
`ingly, the solubility of Ab dAbs improved dramatically with
`an increasing number of negatively charged residues, and
`three negative charges inserted at each edge of CDR3
`completely suppressed aggregation (Figure 2).
`
`HCDR3 sequence
`DEDVHHQKLVFFADED
` VHHQKLVFFADED
`
`DEDVHHQKLVFFA
` VHHQKLVFFA
`
`HCDR3 sequence
`DEDVFFAEDVGSNDED
`DEDVFFAEDVGSN
`
`0
`
`
`1
`
`
`
`2
`
`
`
`3
`4
`5
`
`
`Time (day)
`
`
`
`6
`
`
`
`7
`
` VFFAEDVGSNDED
` VFFAEDVGSN
`
`0
`
`
`1
`
`
`
`2
`
`
`
`3
`4
`5
`
`
`Time (day)
`
`
`
`6
`
`
`7
`
`TRENDS in Biotechnology
`
`60
`
`50
`
`40
`
`30
`
`20
`
`10
`
`0
`
`60
`
`50
`
`40
`
`30
`
`20
`
`10
`
`0
`
`Soluble dAb (µM)
`
`Soluble dAb (µM)
`
`Figure 2. Solubility of Ab VH domain antibody fragments (dAbs) is governed by the
`location of charged insertion mutations within hydrophobic HCDR3 loops. Three
`negatively charged residues (Asp-Glu-Asp) were inserted at one or both edges of
`HCDR3 of Alzheimer’s Ab dAbs. The solubilizing activity of charged mutations
`inserted at one edge of HCDR3 was maximal when
`located near the most
`text). The solubility at 25 8C was
`hydrophobic residues (highlighted
`in red
`monitored by sedimenting aggregated protein and evaluating the concentration
`of soluble dAb. Adapted from [34].
`
`614
`
`Ex. 2030-0003
`
`

`
`Review
`
`Trends in Biotechnology November 2013, Vol. 31, No. 11
`
`of hydrophobic CDRs without eliminating residues in-
`volved in binding.
`An alternative strategy to substituting hydrophobic
`CDR residues for more hydrophilic ones is to introduce
`solubilizing glycans near such hydrophobic residues. In-
`deed, the poor solubility of the wild type mAb (with Phe-
`His-Trp in HCDR3) was improved by introducing a glyco-
`sylation site in a neighboring CDR (HCDR2) not involved
`in binding [38]. Strikingly, the glycan in HCDR2 dramati-
`cally increased the solubility of the wild type mAb (>8-fold)
`– without removing any of the hydrophobic residues in
`HCDR3 – and did not reduce the binding affinity. Glycans
`in CH1 domains of poorly soluble mAbs can also increase
`antibody solubility without reducing binding affinity [40].
`However, the solubilizing activity of glycans is highly
`dependent on their location within CH1 in a manner that
`cannot be predicted using the structure of the parent
`antibody. Interestingly, the type of glycan also significant-
`ly impacts antibody solubility, as an IgG expressed in yeast
`that has mannose-rich glycans is much more resistant to
`aggregation than the same IgG expressed in mammalian
`cells that has more complex human glycans [42]. These and
`related findings [43,44] demonstrate the importance of
`further evaluating the potential of glycans as enhancers
`of antibody solubility and the need for methods capable of
`predicting the impact of glycans on antibody solubility (see
`[45] for recent progress with non-antibody proteins).
`A second seminal study that demonstrated the impact of
`CDRs on mAb aggregation used a novel approach for
`identifying aggregation hotspots within full-length anti-
`bodies [41]. Accurate determination of aggregation hot-
`spots within antibodies requires not only consideration
`of static antibody structures (e.g., X-ray crystal structures)
`but consideration of the dynamics as well. Therefore, the
`investigators used atomistic molecular simulations of IgGs
`to identify aggregation-prone structural motifs that are
`solvent exposed and expected to mediate aggregation,
`including those regions whose solvent exposure is dynamic
`and would be missed using static antibody structures.
`The solvent exposure of every atom in the antibody was
`quantified during the molecular simulations, and the rela-
`tive aggregation propensity (referred to as the spatial
`aggregation propensity or SAP) of each amino acid was
`calculated. Interestingly, these structure-based simula-
`tions identified aggregation hotspots (which are typically
`not contiguous sequences like those calculated by se-
`quence-based algorithms [25–29]) with high SAP scores
`in both the CDRs and constant domains.
`These predictions were tested with hydrophobic-to-
`charged substitution mutations introduced at the pre-
`dicted aggregation hotspots (Figure 4A) [41]. Of the muta-
`tions studied, the only ones that completely suppressed
`aggregation during the accelerated stability study (52 8C
`for 36 h) were those in the CDRs (Figure 4B). Despite that
`the binding activities of the most soluble mutants (e.g., the
`CDR mutants) were reduced significantly, other partially
`solubilizing mutations in the constant regions (not shown
`in Figure 4) did not reduce binding activity. Further devel-
`opment of this and other simulation methods [32,46–50] –
`which could be used to evaluate entire antigen–antibody
`complexes – should improve the systematic identification
`
`615
`
`VH
`
`G54D
`S53D
`G52aD
`S52D
`A50D
`A40D
`Q39D
`S35D
`A33D
`Y32D
`S31D
`S30D
`T28D
`G26D
`WT
`
`HCDR1 (H1)
`
`HCDR2 (H2)
`
`VL
`
`TRENDS in Biotechnology
`
`S56D
`Q55D
`S53D
`S52D
`A51D
`A50D
`Y49D
`N34D
`Y32D
`S31D
`S30D
`I29D
`S28D
`Q27D
`S26D
`R24D
`WT
`
`LCDR1 (L1)
`
`
`LCDR2 (L2)
`
`100
`
`80
`
`60
`
`40
`
`20
`
`0
`
`(A)
`
`RelaƟve binding aŌer
`
`heaƟng (%)
`
`100
`
`80
`
`60
`
`40
`
`20
`
`0
`
`(B)
`
`RelaƟve binding aŌer
`
`heaƟng (%)
`
`Figure 3. Aggregation propensity of human domain antibody fragments (dAbs) is
`strongly influenced by the sequences of CDR1 for VH domains and CDR2 for VL
`domains. (A and B) Single charged substitution mutations were introduced into
`the heavy and light chain CDR1 and CDR2 loops (H1, H2, L1, and L2) of (A) VH and
`(B) VL dAbs that were displayed on phage. The phage particles were heated (80 8C,
`10 min), cooled (4 8C, 10 min) and the retained binding activity to Protein A was
`evaluated. Adapted from [37].
`
`antibody fragment demonstrate that it is possible to mu-
`tate the CDRs in a systematic manner to reduce aggrega-
`tion without reducing binding affinity. Nevertheless,
`additional work is needed to evaluate if these mutations
`in HCDR1 and LCDR2 improve the solubility of full-length
`antibodies (including trastuzumab) and are compatible
`with high-affinity binding for antibodies other than tras-
`tuzumab.
`These studies using single- and multidomain antibody
`fragments to investigate the impact of CDRs on antibody
`aggregation have inspired related studies using full-length
`antibodies [38–41]. Given the extremely large size of IgGs
`(>1000 residues), it would seem unlikely that a small
`number of mutations in the CDRs could significantly im-
`pact the solubility of full-length antibodies. However, sev-
`eral studies have convincingly disproven this notion [38–
`41]. One study investigated the origins of a poorly soluble
`IgG1 antibody that has a triad of hydrophobic residues
`(Phe-His-Trp) in HCDR3 [38]. The investigators reasoned
`that the hydrophobic triad governed the poor antibody
`solubility, and eliminating only these three residues would
`significantly increase solubility. Strikingly, mutating these
`residues to alanine increased the solubility of the wild type
`mAb (10 mg/ml) by over an order of magnitude (>160 mg/
`ml), demonstrating the dramatic impact of hydrophobic
`CDRs on antibody solubility. However, the binding affinity
`of this highly soluble mAb was reduced significantly
`(>1000-fold) due to the elimination of hydrophobic resi-
`dues involved in binding. This finding reinforces the need
`for mutational strategies that increase the hydrophilicity
`
`Ex. 2030-0004
`
`

`
`Review
`
`Trends in Biotechnology November 2013, Vol. 31, No. 11
`
`B3
`
`L235K
`
`B4
`
`K
`
`4
`
`W 9
`
`B5
`
`L
`
`2
`
`3
`
`5
`
`K
`
`W 94K
`
`W100K, F101K
`
`B1
`
`W 94K
`
`W100KF101K
`
`(A)
`
`B2
`
`Fab
`
`Fc
`
`(B)
`
`100
`
`95
`
`90
`
`85
`
`80
`
`IgG monomer (%)
`
`Wild type
`
`Key:
`
`B1
`
`0 h
`
`B2
`
`B3
`
`B4
`
`12 h
`
`24 h
`
`B5
`
`36 h
`
`TRENDS in Biotechnology
`
`Figure 4. Molecular simulations reveal aggregation hotspots within Fab and Fc
`domains that differentially impact monoclonal antibody (mAb) aggregation. (A)
`Structures of Fab and Fc domains with the predicted aggregation hotspots
`highlighted in red [positive SAP scores] relative to the regions predicted not to be
`involved in aggregation highlighted in blue (negative SAP scores). (B) Kinetic
`aggregation analysis of the wild type and mutant antibodies at 52 8C monitored via
`size-exclusion chromatography. The mAb mutants B1–B5 were generated by
`introducing single- or double-charged substitution mutations into CDR3 of the VH
`and VL domains (B1, B2, B4, and B5) as well as the Fc domains (B3 and B5).
`Adapted from [41].
`
`antibody aggregation [42]. This charged peptide – which is
`a fragment of a yeast signal sequence – rendered the IgG
`much more soluble than the same IgG expressed in mam-
`malian cells (which lacked such N-terminal charged pep-
`tides; Figure 6). Additional analysis revealed that the
`negatively charged peptides at the N terminus of either
`heavy or light chains increase antibody solubility, but the
`maximum solubility is achieved when both antibody chains
`contain such N-terminal peptides. This simple and under-
`utilized mutational strategy has significant potential for
`increasing antibody solubility without altering antibody-
`binding affinity. Nevertheless, more work is needed to
`evaluate the impact of the size and polarity of such pep-
`tides on their solubilizing activity, as well as whether
`charged peptides at both N and C termini would yield
`superior antibody solubility.
`The frameworks of isolated VH and VL domains are also
`a critical determinant of the solubility of dAbs. Most
`human dAbs are poorly soluble – even when obtained from
`
`of solubilizing mutations that are compatible with high-
`affinity binding.
`
`Antibody frameworks and domain interfaces
`Although CDRs are key determinants of antibody aggre-
`gation, the frameworks of antibodies are also important
`contributors to their solubility. One attractive approach to
`increase antibody solubility without mutating the CDRs is
`to increase the net charge of the antibody scaffold signifi-
`cantly [33,34,51]. A recent study evaluated whether the
`scaffold of an scFv could be redesigned to increase antibody
`solubility without reducing binding affinity using a method
`known as supercharging [51]. This approach – originally
`demonstrated by others for model proteins such as GFP
`and streptavidin [52] – involves mutating many solvent-
`exposed residues to charged residues of the same polarity
`to achieve highly charged antibodies.
`The investigators used a powerful computational pack-
`age (Rosetta [53]) to evaluate the impact of charged muta-
`tions on folding stability. By selecting charged mutations
`in both the VH and VL domains of an scFv that were
`predicted not to destabilize the antibody fold, they gener-
`ated a panel of antibody variants with different numbers of
`solvent-exposed charged residues (Figure 5) [51]. Although
`none of the mutations were in the CDRs, the binding
`activities of most (three out of four) of the negatively
`charged scFvs were either significantly reduced or elimi-
`nated. In contrast, most of the positively charged scFvs
`(three out of five) retained significant binding activity after
`heating. This difference may be due to the fact that the wild
`type scFv is positively charged and is more compatible with
`positively charged mutations.
`the
`investigators evaluated whether
`Nevertheless,
`the supercharged scFvs remain soluble when heated
`(70 8C for 1 h) [51]. Interestingly, the positively charged
`scFvs were much more resistant to aggregation than the
`wild type or negatively charged variants (Figure 5). This
`finding is interesting because the opposite has been gen-
`erally observed for dAbs (e.g., VH), as aggregation-resistant
`VH dAbs typically have acidic isoelectric points [22,24,33,
`34,37,54]. The folding stabilities of the negatively charged
`scFvs were not reported. Therefore, it may be that these
`variants were destabilized relative to the positively
`charged ones (despite that the Rosetta calculations suggest
`otherwise) [51]. These findings demonstrate the need to
`better understand the impact of polarity and location of
`charged residues within antibody frameworks on their
`solubilizing activity for multidomain antibody fragments
`(scFvs and Fabs) and full-length antibodies.
`A particularly simple method of increasing the net
`charge of antibodies without perturbing their structures
`is to add charged peptides to their termini. Interestingly,
`multiple studies have demonstrated that this simple ap-
`proach is surprisingly effective [42,55]. The poor solubility
`of an scFv (<0.1 mg/ml) could be increased by adding five
`glutamic acid residues to its C terminus [55]. Surprisingly,
`this relatively modest decrease in isoelectric point (reduced
`from pH 7.5 to pH 6.1) dramatically increased scFv solu-
`bility (>150-fold). Similarly, a negatively charged peptide
`(Glu-Ala-Glu-Ala) at the N terminus of the heavy and light
`chains of an IgG expressed in yeast significantly reduced
`
`616
`
`Ex. 2030-0005
`
`

`
`Review
`
`Trends in Biotechnology November 2013, Vol. 31, No. 11
`
`Binding acƟvity aŌerheaƟng (AU)
`0.0
`0.5
`1.0
`1.5
`2.0
`2.5
`
`H-E/L
`
`H/L-E
`
`H-E/L-E
`
`Key:
`
`H/L
`1.2
`
`1.0
`
`0.8
`
`0.6
`
`0.4
`
`0.2
`
`0.0
`
`Normalized light scaƩering
`
`70
`
`75
`
`80
`Temperature (°C)
`
`85
`
`90
`
`TRENDS in Biotechnology
`
`Figure 6. Negatively charged residues added to the N termini of heavy and light
`chains increase IgG solubility. The amino acid sequence Glu-Ala-Glu-Ala reduces
`antibody aggregation when present at the N termini of the heavy and light chains.
`Aggregation was monitored via light scattering at 500 nm. Adapted from [42].
`
`of camelid antibodies pack against the solvent-exposed
`phenylalanine and thereby increase the hydrophilicity of
`the former VH/VL interface [64–66]. These findings highlight
`the need for methods that accurately assess the interactions
`between CDRs and framework residues to design aggrega-
`tion-resistant antibodies.
`Although the propensity of antibodies to aggregate is
`directly
`related
`to
`their
`folding
`stability
`not
`[22,23,33,34,37], more stably folded antibodies typically
`have lower aggregation propensities [56–59,67,68]. One
`systematic approach for stabilizing dAbs is to introduce
`a second disulfide bond between their two opposing b-
`sheets (Figure 7A). This strategy results in significant
`(including
`the one
`in
`stabilization of diverse dAbs
`Figure 7B), and these stabilized variants typically have
`low aggregation propensities (as evidenced by the im-
`in Figure 7C)
`proved size-exclusion behavior seen
`[67,69–73]. A related approach that has been used to
`stabilize scFvs and other multidomain antibodies is to
`introduce a disulfide bond between the VH and VL domains
`[74–78]. ScFvs are particularly aggregation-prone because
`the two variable domains can dissociate and aggregate via
`domain swapping [79]. Adding an interdomain disulfide
`bond – which is absent in normal VH/VL interfaces –
`improves both the folding stability and solubility of scFvs
`[76,77], although drawbacks of this approach are that not
`all scFvs readily form such interdomain disulfides and the
`expression levels of such scFv mutants are relatively low
`[76,77,80]. The VH/VL interface of scFvs and other antibody
`formats can also be stabilized by replacing nonconserved
`residues with conserved ones, which are identified using
`statistical methods that harness the large number of
`known antibody sequences [80–84].
`
`617
`
`k-neg-3
`(-16)
`
`k-neg-2
`(-11)
`
`Wild-type
`scFv
`(+5)
`
`k-pos-1
`(+13)
`
`k-pos-3
`(+20)
`
`TRENDS in Biotechnology
`
`resist
`(scFvs)
`fragments
`antibody
`single-chain
`5. Supercharged
`Figure
`aggregation. The residual binding activity of scFvs (specific for a bacteriophage
`coat protein) after being heated (70 8C, 1 h). The Rosetta computational design
`package was used to identify solvent-exposed amino acids that could be mutated
`to charged residues without destabilizing the scFv. The mutations are highlighted
`with space-filled amino acids, and the net charge is reported as the difference
`between the number of positively (Arg and Lys) and negatively (Glu and Asp)
`charged residues. Adapted from [51].
`
`soluble parent antibodies – suggesting that exposure of
`hydrophobic residues at the former VH/VL interface is linked
`to aggregation. Indeed, several studies have demonstrated
`that hydrophilic mutations within and near the former VH/
`VL interface increase the solubility (and, in some cases, the
`folding stability as well) of poorly soluble dAbs [23,56–59].
`Camelid antibodies have been valuable for guiding the
`selection of such solubilizing mutations because these un-
`usual antibodies typically only possess heavy chains [60,61].
`Despite that there are several sequence differences between
`the variable domains of camelid (VHH) and human (VH)
`antibodies, there are four notable differences in the former
`VH/VL interface (referred to as the VHH tetrad) that distin-
`guish camelid