`
`J. Mol. Biol. (2004) 335, 41–48
`
`COMMUNICATION
`Optimization of the Antibody CH3 Domain by Residue
`Frequency Analysis of IgG Sequences
`
`Stephen J. Demarest*, Jeff Rogers and Genevie` ve Hansen
`
`Torrey Mesa Research Institute
`3115 Merryfield Row, San
`Diego, CA 92130, USA
`
`In an attempt to enhance the overall assembly, yield and half-life of
`recombinant antibody proteins, we have cloned and expressed several
`IgG1 CH3 domains and examined their folding/refolding characteristics.
`We utilized a cytoplasmic bacterial expression system with a thioredoxin
`reductase knock-out strain of BL21(DE3) to produce bovine, murine and
`human CH3. Under identical conditions, expression of bovine CH3 resulted
`consistently in the highest yields of properly folded/oxidized protein.
`Circular dichroism and fluorescence experiments demonstrate that
`oxidized bovine and murine CH3 have surprisingly similar structures
`and stabilities, considering the marginal sequence conservation between
`the two molecules. Residue frequency analysis using a limited data set of
`36 unique Fc sequences originating from 19 different mammalian species
`targeted five specific sites for optimization within bovine CH3. Combi-
`nation of three of these mutants increased the thermal stability of the
`molecule to 86 8C. Comparison of this approach to similar studies using
`larger sequence databases and/or different selection criteria suggests
`sequence database design can increase the success rate for identifying
`residue sites worth optimizing. This optimized CH3 domain can be used
`as a particularly stable platform for functional design and can be grafted
`into full-length antibody sequences to enhance their thermodynamic
`parameters and shelf-life.
`
`q 2003 Elsevier Ltd. All rights reserved.
`
`*Corresponding author
`
`Keywords: antibody; immunoglobulin; CH3; protein design; protein folding
`
`antibody-based pharmaceuticals
`Monoclonal
`represent a large and growing fraction of all drug
`candidates currently entering clinical trials.1 Cur-
`rent technologies have enabled the production of
`antibodies capable of recognizing virtually any
`antigen or molecular target with extremely high
`
`Supplementary data associated with this article can be
`found at doi: 10.1016/y.jmbi.2003.10.040
`Present addresses: S. J. Demarest, J. Rogers, G. Hansen,
`Diversa Corp., 4955 Directors Place, San Diego, CA
`92121, USA.
`Abbreviations used: ASA, accessible surface area;
`bCH3, bovine CH3; Fc, antibody heavy chain CH2CH3
`dimer; Hepes, 4-(2-hydroxyethyl)-1-
`piperazineethanesulfonic acid; LB, Luria broth; hCH3,
`human CH3; mCH3, murine CH3; NCBI, National Center
`for Biotechnology Information; scFv, single-chain
`antibody variable domains; Tm, temperature midpoint of
`thermal denaturation.
`E-mail address of the corresponding author:
`stephen.demarest@diversa.com
`
`affinity. Additionally, antibodies can identify and
`neutralize pathogens (bacteria, viruses),
`toxins
`and cancerous cell types. Thus, antibodies have
`emerged as powerful diagnostic tools in all aspects
`of life sciences, as well as an attractive means of
`creating highly specific therapeutics.
`Mass production of antibodies is costly and
`requires tedious processing before the product is
`ready for market.2 For these reasons, many recom-
`binant methodologies for producing monoclonal
`antibodies are being investigated, including pro-
`duction in mammalian, plant, yeast and bacterial
`cell systems.3,4 The ability to increase the yield and
`molecular half-life of fully assembled and func-
`tional antibody products within these expression
`systems is also a major concern.5
`Here, we describe our approach to optimize the
`portion of
`the IgG Fc responsible for heavy
`chain association (i.e. the CH3 domain; Figure 1).
`Traditionally, research has focused on the study of
`antibody variable domains to understand their
`
`0022-2836/$ - see front matter q 2003 Elsevier Ltd. All rights reserved.
`
`Exhibit 2070
`Page 01 of 08
`
`
`
`42
`
`Antibody Constant Domain Folding and Optimization
`
`Figure 1. Top: A ribbon diagram of the X-ray crystal structure of human CH3.20,39 Residue positions identified as
`potential sites for optimization on the basis of residue frequency analysis are displayed in stick format. The ribbon dia-
`gram was rendered using MolMol.32 Middle: Alignment of the consensus sequence from a non-redundant database of
`36 NCBI mammalian IgG Fc sequences against the human, bovine and murine IgG1 CH3 sequences. The human CH3
`domain was cloned from Genbank source #BC026038 purchased from the ATCC. Bovine CH3 was cloned from
`in-house bovine constructs derived from a bovine spleen cDNA library purchased from Stratagene. Murine CH3 was
`cloned from an in-house synthetic murine IgG1 gene based on the murine IgG1 Kabat database consensus sequence.33
`CH3 PCR inserts and commercial pET21a plasmid (Novagen) were digested using the designed Bam HI and Xho I
`restriction sites (enzymes from Invitrogen). Inserts were ligated into the vector using the H.C. T4 DNA ligase (Invitro-
`gen). The QuickChangew Site-Directed Mutagenesis kit (Stratagene) was used to create mutations in the bovine CH3
`pET21a vector. Residue positions within the bovine CH3 domain where directed mutations were made are indicated.
`Below the alignments, (.), conservative variation; (:), strict variation; (*), no variation. All inserts and mutations were
`confirmed by DNA sequencing. BL21-trxB(DE3) (Novagen) was utilized as the host for IPTG-inducible cytoplasmic
`expression. BL21(DE3) did not yield significant amounts of oxidized protein and Rosetta-Gami (Novagen) had a
`significantly longer growth phase without providing a significant increase in CH3 oxidation over the BL21-trxB(DE3)
`
`Exhibit 2070
`Page 02 of 08
`
`
`
`Antibody Constant Domain Folding and Optimization
`
`43
`
`antigen-binding capabilities.6 However, effector
`functions
`that
`allow antibodies
`to
`involve
`additional factors within the immune system are
`embedded within the constant domains.7,8 For
`these reasons, the past few years have witnessed
`an explosion in the sequencing of antibody con-
`stant domains. We utilize the publicly available Fc
`sequences to find residue positions within the
`bovine IgG1 CH3 fold that are poorly represented
`and describe a limited set of mutations at these
`sites designed to enhance the thermal stability of
`the domain.
`The CH3 domain of bovine IgG1 was chosen as
`our model system on the basis of analysis of a con-
`densed set of 36 IgG Fc
`sequences. These
`sequences were collected by performing the Tera-
`Blaste routine on the non-redundant protein
`sequence database maintained by NCBI
`(The
`National Center for Biotechnology Information)
`using bovine, murine and human Fc sequences as
`bait. Other than primate IgG sequences, bovine
`IgG1 demonstrated the closest resemblance to the
`consensus sequence, while exhibiting a few defini-
`tive sites that, once mutated, might
`lead to
`increases in thermal stability. The sequences of
`bovine, human and murine IgG1 CH3 (denoted
`bCH3, hCH3 and mCH3, respectively) are shown in
`Figure 1. Proteins with disulfides have tradition-
`ally been expressed in the periplasmic space of
`bacteria.38 However, cytosolic yields tend to be
`much higher; therefore, bCH3 was expressed in the
`cytoplasm of several bacterial cell lines. The best
`results were obtained in BL21-trxB(DE3),
`a
`bacterial cell line carrying a mutation in the thiore-
`doxin reductase gene, trxB.10 bCH3 was expressed
`at 20 – 40 mg/l in a shaker flask in BL21-trxB(DE3).
`The protein was affinity-purified on Ni2þ
`-NTA
`resin (Qiagen) using the C-terminal hexahistidine
`tag. Oxidized protein was separated from reduced
`material using reverse-phase HPLC (Figure 1).
`Purified yields of folded/oxidized protein were
`approximately 70% of the expressed material. The
`exact mass of oxidized bCH3, 15064.9402 AMU,
`was determined by electrospray,
`time-of-flight
`(TOF) mass spectroscopy. The reduced fraction
`(analyzed in the presence of DTT) yielded a major
`peak with a molecular mass of 15066.9403 AMU,
`exactly 2 AMU higher than the oxidized form, due
`to the protonation of the free thiol groups. The
`theoretical masses of oxidized and reduced bCH3
`are 15066.6 and 15068.6, respectively, and are indis-
`tinguishable within the error of the measurement
`from the experimentally derived values. Oxidized
`mCH3 could be obtained in only small quantities
`
`(, 1 mg per 1 l shaker flask) and oxidized hCH3
`could not be separated from its reduced form
`using our purification scheme.
`The CH3 domain is known to form a specific non-
`covalent dimer.11 Oxidized bCH3 runs appro-
`priately as a dimer on a BioSpec2000 (Pharmacia)
`HPLC gel-filtration column. Molecular mass stan-
`dards included bovine serum albumin (66 kDa),
`carbonic anhydrase (28 kDa),
`lysozyme (14 kDa)
`and ubiquitin (8.5 kDa). The elution time of bCH3
`was constant over a 200 nM – 200 mM injected
`range of concentrations. Similar to bCH3, oxidized
`mCH3 appears to be a dimer, as judged by gel-
`filtration. Purified hCH3 elutes in several forms,
`predominantly trimer, dimer and monomer, an
`indication that it is not wholly folded/oxidized.
`Repeated expressions/purifications of hCH3 never
`produced viable protein.
`three CH3 domains exhibited significant
`All
`secondary structure, as judged by their circular
`dichroism (CD) spectra (Figure 2A). Far-UV CD
`spectra of bCH3 solubilized from cell pellets using
`8 M urea and subsequently Ni2þ
`-purified in urea
`and refolded were identical with the bCH3 spectra
`solubilized and purified using strictly aqueous
`buffers. bCH3, mCH3 and hCH3 all have minima at
`217 nm, a characteristic of b-sheet structures con-
`sistent with the antibody fold being composed of
`a twofold b-sheet sandwich. The hCH3 spectrum
`was significantly different in intensity and shape
`from the spectra of bCH3 and mCH3, but was very
`similar to what was observed for reduced bCH3
`the bCH3 and
`(data not shown). Surprisingly,
`mCH3 spectra were nearly superimposable, indicat-
`ing that their structures must be nearly identical.
`This was unexpected considering bCH3 and mCH3
`are only 54% identical in sequence.
`the CH3
`Temperature-induced unfolding of
`domains was monitored separately using fluor-
`escence, near-UV CD and far-UV CD. The fluor-
`escence excitation wavelength was 278 nm with a
`3 nm bandwidth and no emission cutoff filter; the
`near-UV CD signal was monitored at 278 nm
`using a 2 nm bandwidth; and the far-UV CD signal
`was monitored at 217 nm using a 1 nm bandwidth.
`bCH3 and mCH3 undergo single unfolding tran-
`sitions with midpoints (Tm) at 348.8(^ 0.5) K and
`346.8(^ 0.5) K,
`respectively. Refolding of both
`bCH3 and mCH3 was reversible. The CD spectra at
`278 K (5 8C) prior to and one hour after tempera-
`ture-denaturation were virtually unchanged.
`The Tm of both bCH3 and mCH3 were unaffected
`by concentration. Fluorescence measurements
`were performed using 0.5 mM and 4 mM protein,
`
`strain. Bottom: Purification of bCH3. Bacterial pellets containing CH3 constructs were dissolved in 8 M urea (pH 8),
`spun down and the supernatants were passed over Ni2þ
`-NTA resin (Qiagen). Protein was eluted from the resin by
`adding 8 M urea (pH 4). Reverse-phase HPLC was performed on a Dionex DX500 chromatography system using a
`Phenomenex C5-Prep column with water/acetonitrile gradients and an absorbance detector set to 280 nm. The bottom
`panel displays the chromatograph of bCH3. The peak eluting at 22.9 minutes is oxidized CH3, while the peak eluting at
`25.4 minutes is reduced CH3.
`
`Exhibit 2070
`Page 03 of 08
`
`
`
`44
`
`Antibody Constant Domain Folding and Optimization
`
`range of concentrations, no change in the Tm was
`observed, as might be expected for a folded dimer
`transition.12
`to unfolded monomer
`two-state
`Buchner and co-workers, however, observed a con-
`centration-dependent change in the guanidinium –
`HCl denaturation midpoint of the CH3 domain
`from the murine monoclonal antibody MAK33
`typical for a native dimer to unfolded monomer
`transition.13,14 Thermal unfolding transitions of the
`dimeric Arc Repressor protein and the dimeric
`four-helix bundle protein, ROP, both display con-
`centration-dependent Tm shifts.15,16 The lack of a
`temperature shift between different concentrations
`of protein could be explained if the folding rate of
`the protein is not dominated by the rate of dimer
`association, but instead dominated by the trans –
`cis isomerization of P177(374). Another possibility
`could be that the dimer dissociates before reaching
`the temperature at which the protein unfolds. Due
`to the uncertainty in the model that should be
`applied for quantitative interpretation of the free
`energy of folding, we use the Tm value of bCH3 as
`a standard to assess the relative stability changes
`induced by the designed mutants described below.
`Our goal was to discover residue positions
`fully optimized for protein
`within bCH3 not
`stability. We performed positional frequency and
`positional entropy analyses on our limited IgG
`data set (36 sequences from 19 mammalian species)
`utilizing a Perl program written in-house. A
`ClustalW alignment of
`the 36 sequences
`is
`provided in the Supplementary Material†.17 All
`sequences were at least 95% unique (i.e. at least
`ten residue variations out of , 220 residues from
`any other sequence within the data set). Without
`the 5% cutoff, the dataset burgeons to over 100
`sequences. The redundancy creates considerable
`bias and is therefore eliminated. The residue fre-
`quency, piðrÞ; for each position, i, in an individual
`sequence is simply the number of times that a
`(r ¼ A,C,D…V,W,Y)
`particular
`residue-type
`is
`observed within the dataset divided by the total
`number of sequences. The positional entropy, NðiÞ;
`was calculated as a measure of every residue
`position’s variability.18,19 The positional entropy is
`a function of the Shannon informational entropy,
`HðiÞ :
`
`XY
`
`r¼A
`
`NðiÞ ¼ eHðiÞ
`
`; HðiÞ ¼ 2
`
`piðrÞlnðpiðrÞÞ
`
`ð1Þ
`
`Data from the residue frequency analysis were
`used to design mutants capable of optimizing
`bCH3 stability. Mutations were selected based on
`four criteria. The first is that the bovine sequence’s
`residue frequency at a given position is much
`smaller than the most common residue frequency,
`mcpiðrÞ; (i.e. piðrÞ=mcpiðrÞ # 0:2). Next, mutations
`must be conservative based on CH3 structure
`analysis and residue type. Thus, no charge reversals
`
`† http://www.ebi.ac.uk/clustalw
`
`Figure 2. Structure and stability analyses of bCH3,
`mCH3 and hCH3. A, The CD spectra of bCH3, mCH3 and
`hCH3 at 5 8C. The CD and fluorescence measurements
`were performed using an Aviv model 215 spectrometer
`equipped with a thermoelectric cuvette-holder and the
`total fluorescence accessory. The buffer contained 2 mM
`phosphate, borate, citrate, 10 mM NaCl (pH 7.5). All
`final spectra were the average of at least four scans
`utilizing a signal averaging time of 2 s/l. Signal
`averaging times for all melting experiments were 50 s/
`deg. C. Protein concentrations were determined using
`the method of Pace and co-workers.34 B. Temperature-
`denaturation of bCH3, mCH3 and hCH3 monitored by
`the far-UV CD signal at 217 nm. All melts increased the
`temperature at 2 deg. C intervals for CD and fluor-
`escence sampling and utilized a 180 s equilibration
`period between data acquisitions. The melting tempera-
`ture of bovine and murine CH3 did not vary with concen-
`tration; therefore the far and near-UV CD curves and the
`fluorescence curves were all fit to a two-state, single-
`molecule unfolding model.35 The stability of each variant
`was qualitatively ranked using Tm values. A theoretical
`DCp8, 1727 cal/mol K (1 cal ¼ 4.184 J), was calculated
`based on an estimate of the change in accessible surface
`area (DASA) between the folded and unfolded states of
`the CH3 dimer.36,37
`
`far-UV CD measurements were performed using
`10 mM protein and near-UV CD measurements
`using 30 mM protein (see Figure 2B for far-UV
`CD temperature-denaturation). Over this 60-fold
`
`Exhibit 2070
`Page 04 of 08
`
`
`
`Antibody Constant Domain Folding and Optimization
`
`45
`
`bovine IgG2 or IgG3). This final criterion is related
`to the no-covariation criterion.
`Five potentially non-ideal residue sites were
`identified when pitting the bovine CH3 sequence
`against the Fc data set. These positions have been
`rendered onto the structure in Figure 1 and are
`indicated in the sequence alignment below the
`structure. Several CH3 domains, including bovine
`CH3, contain insertions or deletions resulting in an
`ambiguous numbering system when compared to
`the human IgG sequence. Therefore, we chose to
`number according to our Fc data set with residue
`G35 corresponding to residue G237, the first resi-
`due of the Fc structure of human IgG1.20 Numbers
`in parentheses refer to the standard full-length
`human IgG residue numbering, while the numbers
`directly to the left of the residue letter refer to num-
`bering originating from our Fc data set. The sites
`are S174(371), Y179(376), G197(392), S207(402) and
`T246(441). Each residue site is spatially distant
`from the four other sites, on the basis of the struc-
`ture of hCH3,20 and is thus expected to have an
`independent and additive effect on the stability of
`the domain. The following point mutations were
`made based on the residue frequency analysis of
`bCH3: S174G, Y179D, G197K, G197A (conservative
`mutation), S207G and T246L. Four of these five
`mutant positions are at residue sites of reasonable
`heterogeneity
`(i.e.
`the measured
`positional
`entropies are in the top 70% for all residue pos-
`itions). Mutation at S207(402) was the only position
`that was highly conserved; only four of the 36
`sequences have a residue other than Gly, with Ser
`being represented twice.
`All mutant proteins display nearly identical CD
`spectra compared to the native protein (Figure 3A),
`indicating that no major structural change was
`induced by any of
`the point mutations. Each
`mutant domain exhibits a reproducible Tm, as
`judged by both near and far-UV CD (Figure 3B).
`Near and far-UV CD denaturations were per-
`formed on all the mutants, using the same concen-
`trations that were used to study the wild-type
`protein (Table 1). G197K, S207G and T246L
`mutations had a stabilizing affect on bCH3. Combi-
`nation of these mutants into a single construct led
`to near-additive increases in the thermal stability
`of
`the domain. The experimentally determined
`
`Figure 3. Structure and stability analyses of bCH3 and
`the mutant constructs of the domain. A, The CD spectra
`of native and mutant bCH3 constructs at 5 8C, pH 7.5.
`B, Plots of the unfolded fractions of the bCH3 variants
`between 325 and 375 K.
`
`were made and no large, buried hydrophobic
`amino acids were converted to charged residues
`or Gly. The third criterion is that there must be an
`absence of co-variation at this position with other
`residue positions throughout the sequence. Lastly,
`the residue to be mutated must not be conserved
`within the individual species IgG subclass (i.e. in
`
`Table 1. Results of fitting the CD temperature melts of bCH3, the six single-mutant variants and a triple-mutant variant
`to a two-state unfolding model
`
`Construct
`
`Tm (K) fluor. Ex. l280
`
`Tm (K) CD [u]217
`
`Tm (K) [u]280
`
`Tm (K) Avg.
`
`DTm (K)
`
`bCH3
`S174G
`Y179D
`G197K
`G197A
`S207G
`T246L
`G197K/S207G/T246L
`pG197K/S207G/T246L
`
`348.0 ^ 0.3
`–
`–
`–
`–
`–
`–
`–
`–
`
`349.9 ^ 0.3
`348.7
`349.3
`354.2
`351.2
`352.6
`352.2
`–
`–
`
`349.0 ^ 0.3
`349.6
`349.9
`354.2
`352.2
`353.7
`352.3
`358.8
`–
`
`348.8 ^ 0.5
`349.1
`349.6
`354.2
`351.7
`353.2
`352.3
`358.8
`p361.1
`
`–
`0.3
`0.8
`5.4
`2.9
`4.4
`3.5
`10.0
`p13.3
`
`The asterisk ( p ) indicates a hypothetical triple mutant, assuming the individual mutant stabilizations are additive.
`
`Exhibit 2070
`Page 05 of 08
`
`
`
`46
`
`Antibody Constant Domain Folding and Optimization
`
`increase in Tm was slightly lower than the increase
`predicted when assuming full additivity; however,
`the difference between the two numbers lies at the
`edge of experimental uncertainty. Temperature-
`denaturation curves of bCH3, the six mutant bCH3
`domains, and the experimental and hypothetical
`triple mutant are shown in Figure 3B.
`The stabilizing affect of the three mutations can
`be rationalized by analyzing the structure of the
`hCH3 domain.20 G197(392)K inserts a Lys at the
`interface between the CH3 dimer. The hydrophobic
`portion of the side-chain packs against a hydro-
`phobic patch created by F210(405) and V202(397)
`of the opposite dimer subunit. Adding a hydro-
`phobic moiety to this cavity also explains why Ala
`is preferred over Gly at this position. The posi-
`tively charged primary amine is exposed to solvent
`and is in close proximity to D204(399), possibly
`enhancing the favorability of Lys at this position
`over Ala. S207(402)G places glycine in a tight turn
`between b-strands. The backbone dihedral angles
`of this residue in the crystal structure are outside
`the generally allowed ranges for a-substituted
`amino acids (all amino acids other than Gly).
`Mutation to Gly at this position likely releases
`strain induced by the tight
`turn. Finally,
`the
`T246(441)L mutation completely buries the iso-
`butyl side-chain into the hydrophobic interior of
`the b-sandwich. The favorable contribution is not
`as large as might be expected for the burial of an
`additional three methylene groups, indicating that
`steric placement or hydrogen bonding of the Thr
`side-chain must compensate somewhat
`for the
`loss of buried hydrophobic surface area.21
`The expression levels of the mutant variants
`were relatively unchanged compared to that of the
`wild-type bCH3. Variants with enhanced thermal
`stability did not have
`increased expression
`compared to the wild-type protein. One way of
`rationalizing this result is that the thermal stability
`of the native protein, 76 8C, is so far above 37 8C,
`the temperature at which the bacteria were
`cultured,
`that
`stabilization does not
`affect
`expression significantly. There was one exception,
`S174G. Expression of this variant led to protein
`yields fourfold higher than what was observed for
`wild-type bCH3 and the other variants. The
`mutation had a minimal effect on the overall stab-
`ility of the domain. It is possible that the S174G
`mutation leads to increased kinetics of folding and
`therefore less in vivo degradation of the protein.
`Overall, however, increased thermal stability did
`not correlate with increased expression.
`Many approaches have been applied success-
`fully to increase the thermal stability of proteins
`or enzymes, including rational design,22 random
`mutagenesis,23 computational design,24 disulfide
`engineering25 and even the creation of circular
`proteins.26 Statistical analysis of protein sequence
`data sets provides another rapid and effective
`means of semi-rationally designing more stable
`proteins, provided a big enough data set exists
`from which to draw information. Davidson and
`
`co-workers have used residue frequency analysis
`using a database of 266 non-redundant SH3
`domain sequences to stabilize the C-terminal SH3
`domain of
`the yeast actin-regulating protein
`Abp1p.18,27 This , 60 residue protein is only
`marginally stable and contains many non-standard
`SH3-type residues. Their mutational criteria were
`similar to those reported here; however, four of
`their seven mutations had a destabilizing effect on
`the domain, whereas none of the bCH3 mutations
`was destabilizing. Bovine IgG1 matches the IgG
`consensus sequence to a much greater extent than
`Abp1p matches the SH3 consensus; approximately
`50% of the Abp1p sequence did not match the con-
`sensus. The magnitude of the stabilizing mutations
`within Abp1p led were much greater, however,
`increasing the Tm values on average by 12 deg.C.
`This may be due to the greater size of their data-
`base and its ability to statistically find the real
`winner at an individual position. Also, the low
`intrinsic stability of
`the Abp1p domain likely
`leaves room for greater enhancement. The pitfall
`of choosing residues for replacement based on the
`statistically most frequent residue type is that each
`particular
`residue position interacts with and
`directs the best residue types of residues around
`it.28 However, the limited taxonomy of the mam-
`malian Fc database described here likely does not
`allow for
`the broad evolutionary sampling
`required to generate significant linked pairs of
`mutations.
`Antibody sequencing efforts that have focused
`primarily on antibody variable domains have pre-
`viously enabled residue frequency analysis to aid
`in the design of a thermostable variable domain.29
`Steipe and co-workers utilized all the heavy-chain
`variable domain sequences of the Kabat database
`to design mutations at several positions: six of ten
`mutants were stabilizing, three had no effect on
`stability and only one was destabilizing. Their
`relatively high success rate, similar to ours, may
`be due, in part, to the limited taxonomy of their
`database allowing them to avoid significant
`divergence between sequences,
`leading to more
`instances of covariation. The number of mam-
`malian IgG constant domains that have been
`sequenced is comparatively small. Only recently,
`with the number of publicly available unique IgG
`constant domains tripling in the last five years,
`has residue frequency analysis of the IgG constant
`domains been possible.
`Although the strict criteria used here strength-
`ened our chances of choosing stabilizing mutations
`successfully, they almost certainly exclude other
`potentially stabilizing mutants that do not fall
`within the limits.
`In two independent studies,
`(using GroEL)30
`and
`Wang and co-workers
`Nikolova and co-workers (using the p53 DNA-
`binding domain)9 utilized a different
`residue
`frequency strategy to find residue positions worth
`optimization. Their criterion for mutation was to
`create libraries at residue sites with high positional
`entropies rather than adopting our strategy of
`
`Exhibit 2070
`Page 06 of 08
`
`
`
`Antibody Constant Domain Folding and Optimization
`
`47
`
`mutating residue positions where the wild-type
`residue frequency was extremely low.
`In both
`reports, residue sites of improvement were found
`where the wild-type residue frequency was as
`high as 20 – 91% of all the residues observed for
`those individual positions. Our strategy ignores
`potentially stabilizing mutations such as these.
`The positive aspect of mutating at positions of low
`residue frequency was that five of six residue
`replacements stabilized the domain and none was
`destabilizing. The success rate for finding posi-
`tively stabilizing mutants was much lower in the
`designed mutants of GroEL and p53.
`In both
`studies, approximately 29% of the mutations were
`stabilizing, 42% had relatively little affect on the
`stability and 29% were destabilizing.
`Biophysical studies of the CH3 domain have
`focused mainly on the human and murine
`proteins.13,31,32 The Tm values of bovine CH3 and
`murine CH3 measured here were both close to the
`reported Tm of the human domain, 80.4 8C.32 Both
`the bovine and murine domains refolded to their
`native structures within an hour of lowering the
`temperature to favor the native state. This is in
`fair agreement with the work of Buchner and
`co-workers. At 4 8C, they observe approximately
`80% refolding of their murine CH3 construct over
`the period of an hour, with the accumulation of a
`folding intermediate accounting for the other 20%
`of the population. However, we refolded while
`cooling from 101 8C over a period of an hour. The
`native state becomes stable at temperatures below
`75 8C, where folding/refolding is likely to be
`accelerated greatly and it is not surprising that we
`observe nearly full recovery of folded protein over
`the course of an hour.
`Creating a stabilized CH3 or Fc domain as
`described here would be useful as a platform for
`functional design. Potential reasons for function-
`ally modifying constant domains include modify-
`ing the interaction strengths of Fc domains for
`individual Fc receptors8 or for the creation of
`bispecific
`antibodies. Carter
`and co-workers
`designed a heterodimeric CH3 system with the
`goal of producing antibodies capable of binding
`two targets.32 They achieved remarkable success;
`however,
`their
`initial heterodimeric constructs
`were
`significantly
`compromised in stability
`compared to the wild-type protein. Functional
`mutations often lead to decreases in the stability
`of a protein. Combining these mutations can
`destabilize the fold to a point where most of the
`protein remains unfolded and functionless. Start-
`ing from an ultra-stable platform would provide
`more room for compromise when mutating for
`functional purposes.
`In conclusion, we demonstrated that the publicly
`available IgG sequences, although limited in
`number,
`provide
`enough
`information
`for
`sequence-based optimization of CH3 stability. The
`limited taxonomy of the database appeared to
`enhance
`our
`ability
`to pinpoint
`stabilizing
`mutations, suggesting that database design may
`
`increase the success rate for identifying residue
`positions that can be optimized. Future studies
`may include the engineering of thermostable, full-
`length antibody constructs with increased half-life
`and potentially increased molecular yield.
`
`Acknowledgements
`
`The authors thank G. Chen for critically reading
`the manuscript and for insightful discussions.
`
`References
`
`1. van Dijk, M. A. & van de Winkel, J. G. J. (2001).
`Human antibodies as next generation therapeutics.
`Curr. Opin. Chem. Biol. 5, 368 – 374.
`2. Gura, T. (2002). Therapeutic antibodies: magic bullets
`hit the target. Nature, 417, 584 – 586.
`3. Verma, R., Boleti, E. & George, A. J. T. (1998). Anti-
`body engineering: comparison of bacterial, yeast,
`insect
`and mammalian
`expression
`systems.
`J. Immunol. Methods, 216, 165 – 181.
`4. Smith, M. D. (1996). Antibody production in plants.
`Biotechnol. Advan. 14, 267 –281.
`5. Wo¨ rn, A. & Plu¨ ckthun, A. (2001). Stability engineer-
`ing of antibody single-chain Fv fragments. J. Mol.
`Biol. 305, 989 –1010.
`6. Muyldermans, S. (2001). Single domain camel anti-
`bodies: the current status. J. Biotechnol. 74, 277 –302.
`7. Presta, L. G.
`(2002). Engineering antibodies for
`therapy. Curr. Pharm. Biotechnol. 3, 237 – 256.
`8. Ravetch, J. V. & Bolland, S. (2001). IgG Fc receptors.
`Annu. Rev. Immunol. 19, 275 – 290.
`9. Nikolova, P. V., Henckel, J., Lane, D. P. & Fersht, A. R.
`(1998). Semirational design of active tumor suppres-
`sor p53 DNA binding domain with enhanced
`stability. Proc. Natl Acad. Sci. USA, 95, 14675 – 14680.
`10. Proba, K., Ge, L. & Plu¨ ckthun, A. (1995). Functional
`antibody single-chain fragments from the cytoplasm
`of Escherichia coli: influence of thioredoxin reductase
`(TrxB). Gene, 159, 203 – 207.
`11. Isenman, D. E., Lancet, D. & Pecht, I. (1979). Folding
`pathways of immunoglobulin domains. The folding
`the Cg3 domain of human IgG1.
`kinetics of
`Biochemistry, 18, 3327 – 3336.
`12. Neet, K. E. & Tim, D. E. (1994). Conformational
`stability of dimeric proteins: quantitative studies by
`equilibrium denaturation. Protein Sci. 3, 2167 – 2174.
`13. Thies, M. J. W., Mayer, J., Augustine, J. G., Frederick,
`C. A., Lilie, H. & Buchner, J. (1999). Folding and
`association of
`the antibody domain CH3: prolyl
`isomerization preceeds dimerization. J. Mol. Biol.
`293, 67 –79.
`14. Lilie, H., Lang, K., Rudolph, R. & Buchner, J. (1995).
`Association of antibody chains at different stages of
`folding: prolyl
`isomerization occurs
`after
`the
`formation of quarternary structure. J. Mol. Biol. 248,
`190 –201.
`(1989). Equilibrium
`15. Bowie,
`J. U. & Sauer, R. T.
`the Arc repressor
`dissociation and unfolding of
`dimmer. Biochemistry, 28, 7139 – 7143.
`16. Steif, C., Weber, P., Hinz, H-J., Flossdorf, J., Cesareni,
`G. et al.
`(1993). Subunit
`interactions provide a
`significant contribution to the stability of the dimeric
`
`Exhibit 2070
`Page 07 of 08
`
`
`
`48
`
`Antibody Constant Domain Folding and Optimization
`
`four-a-helical-bundle protein ROP. Biochemistry, 32,
`3867 – 3876.
`17. Higgins, D., Thompson, J., Gibson, T., Thompson,
`J. D., Higgins, D. G. & Gibson, T. J. (1994). CLUSTAL
`W: improving the sensitivity of progressive multiple
`sequence alignment
`through sequence weighting,
`position-specific gap penalties and weight matrix
`choice. Nucl. Acids Res. 22, 4673 – 4680.
`18. Larson, S. M. & Davidson, A. R. (2000). The identifi-
`cation of conserved interactions within the SH3
`domain by alignment of sequences and structures.
`Protein Sci. 9, 2170 – 2180.
`19. Shenkin, P. S., Erman, B. & Mastrandrea, L. D. (1991).
`Information-theoretical entropy as a measure of
`sequence variability. Proteins: Struct. Funct. Genet. 11,
`297 – 313.
`20. DeLano, W. L., Ultsch, M. H., de Vos, A. M. & Wells,
`J. A. (2000). Convergent solutions to binding at a
`protein – protein interface. Science, 287, 1279 – 1283.
`21. Privalov, P. L. & Gill, S. A. (1988). Stability of protein
`structure
`and hydrophobic
`interaction. Advan.
`Protein Chem. 39, 191 – 234.
`(1997). Stabilization of
`22. Lee, B. & Vasmatzis, G.
`protein structures. Curr. Opin. Biotechnol. 8, 423 – 428.
`23. Gershenson, A. & Arnold, F. H. (2000). Enzyme
`stabilization by directed evolution. Genet. E